X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FAnalysis%2FMemoryDependenceAnalysis.cpp;h=716e3e65e4db369fd457c5c0b2419f7a77685be3;hb=c50bf46a2290d5131e9379a7499f5451b2fbe584;hp=9a1edc763f18f62302ceab86eb37aa9cacf323e6;hpb=7bce462c15356229e13d78d14560feaac30a0f5f;p=oota-llvm.git diff --git a/lib/Analysis/MemoryDependenceAnalysis.cpp b/lib/Analysis/MemoryDependenceAnalysis.cpp index 9a1edc763f1..716e3e65e4d 100644 --- a/lib/Analysis/MemoryDependenceAnalysis.cpp +++ b/lib/Analysis/MemoryDependenceAnalysis.cpp @@ -1,4 +1,4 @@ -//===- MemoryDependenceAnalysis.cpp - Mem Deps Implementation --*- C++ -*-===// +//===- MemoryDependenceAnalysis.cpp - Mem Deps Implementation -------------===// // // The LLVM Compiler Infrastructure // @@ -8,31 +8,33 @@ //===----------------------------------------------------------------------===// // // This file implements an analysis that determines, for a given memory -// operation, what preceding memory operations it depends on. It builds on +// operation, what preceding memory operations it depends on. It builds on // alias analysis information, and tries to provide a lazy, caching interface to // a common kind of alias information query. // //===----------------------------------------------------------------------===// -#define DEBUG_TYPE "memdep" #include "llvm/Analysis/MemoryDependenceAnalysis.h" #include "llvm/ADT/STLExtras.h" #include "llvm/ADT/Statistic.h" #include "llvm/Analysis/AliasAnalysis.h" -#include "llvm/Analysis/Dominators.h" +#include "llvm/Analysis/AssumptionCache.h" #include "llvm/Analysis/InstructionSimplify.h" #include "llvm/Analysis/MemoryBuiltins.h" #include "llvm/Analysis/PHITransAddr.h" #include "llvm/Analysis/ValueTracking.h" #include "llvm/IR/DataLayout.h" +#include "llvm/IR/Dominators.h" #include "llvm/IR/Function.h" #include "llvm/IR/Instructions.h" #include "llvm/IR/IntrinsicInst.h" #include "llvm/IR/LLVMContext.h" +#include "llvm/IR/PredIteratorCache.h" #include "llvm/Support/Debug.h" -#include "llvm/Support/PredIteratorCache.h" using namespace llvm; +#define DEBUG_TYPE "memdep" + STATISTIC(NumCacheNonLocal, "Number of fully cached non-local responses"); STATISTIC(NumCacheDirtyNonLocal, "Number of dirty cached non-local responses"); STATISTIC(NumUncacheNonLocal, "Number of uncached non-local responses"); @@ -47,21 +49,23 @@ STATISTIC(NumCacheCompleteNonLocalPtr, "Number of block queries that were completely cached"); // Limit for the number of instructions to scan in a block. -// FIXME: Figure out what a sane value is for this. -// (500 is relatively insane.) -static const int BlockScanLimit = 500; +static const unsigned int BlockScanLimit = 100; + +// Limit on the number of memdep results to process. +static const unsigned int NumResultsLimit = 100; char MemoryDependenceAnalysis::ID = 0; - + // Register this pass... INITIALIZE_PASS_BEGIN(MemoryDependenceAnalysis, "memdep", "Memory Dependence Analysis", false, true) +INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker) INITIALIZE_AG_DEPENDENCY(AliasAnalysis) INITIALIZE_PASS_END(MemoryDependenceAnalysis, "memdep", "Memory Dependence Analysis", false, true) MemoryDependenceAnalysis::MemoryDependenceAnalysis() -: FunctionPass(ID), PredCache(0) { + : FunctionPass(ID), PredCache() { initializeMemoryDependenceAnalysisPass(*PassRegistry::getPassRegistry()); } MemoryDependenceAnalysis::~MemoryDependenceAnalysis() { @@ -78,20 +82,21 @@ void MemoryDependenceAnalysis::releaseMemory() { PredCache->clear(); } - - /// getAnalysisUsage - Does not modify anything. It uses Alias Analysis. /// void MemoryDependenceAnalysis::getAnalysisUsage(AnalysisUsage &AU) const { AU.setPreservesAll(); + AU.addRequired(); AU.addRequiredTransitive(); } -bool MemoryDependenceAnalysis::runOnFunction(Function &) { +bool MemoryDependenceAnalysis::runOnFunction(Function &F) { AA = &getAnalysis(); - TD = getAnalysisIfAvailable(); - DT = getAnalysisIfAvailable(); - if (PredCache == 0) + AC = &getAnalysis().getAssumptionCache(F); + DominatorTreeWrapperPass *DTWP = + getAnalysisIfAvailable(); + DT = DTWP ? &DTWP->getDomTree() : nullptr; + if (!PredCache) PredCache.reset(new PredIteratorCache()); return false; } @@ -99,7 +104,7 @@ bool MemoryDependenceAnalysis::runOnFunction(Function &) { /// RemoveFromReverseMap - This is a helper function that removes Val from /// 'Inst's set in ReverseMap. If the set becomes empty, remove Inst's entry. template -static void RemoveFromReverseMap(DenseMap > &ReverseMap, Instruction *Inst, KeyTy Val) { typename DenseMap >::iterator @@ -123,7 +128,8 @@ AliasAnalysis::ModRefResult GetLocation(const Instruction *Inst, if (LI->isUnordered()) { Loc = AA->getLocation(LI); return AliasAnalysis::Ref; - } else if (LI->getOrdering() == Monotonic) { + } + if (LI->getOrdering() == Monotonic) { Loc = AA->getLocation(LI); return AliasAnalysis::ModRef; } @@ -135,7 +141,8 @@ AliasAnalysis::ModRefResult GetLocation(const Instruction *Inst, if (SI->isUnordered()) { Loc = AA->getLocation(SI); return AliasAnalysis::Mod; - } else if (SI->getOrdering() == Monotonic) { + } + if (SI->getOrdering() == Monotonic) { Loc = AA->getLocation(SI); return AliasAnalysis::ModRef; } @@ -154,29 +161,32 @@ AliasAnalysis::ModRefResult GetLocation(const Instruction *Inst, return AliasAnalysis::Mod; } - if (const IntrinsicInst *II = dyn_cast(Inst)) + if (const IntrinsicInst *II = dyn_cast(Inst)) { + AAMDNodes AAInfo; + switch (II->getIntrinsicID()) { case Intrinsic::lifetime_start: case Intrinsic::lifetime_end: case Intrinsic::invariant_start: + II->getAAMetadata(AAInfo); Loc = AliasAnalysis::Location(II->getArgOperand(1), cast(II->getArgOperand(0)) - ->getZExtValue(), - II->getMetadata(LLVMContext::MD_tbaa)); + ->getZExtValue(), AAInfo); // These intrinsics don't really modify the memory, but returning Mod // will allow them to be handled conservatively. return AliasAnalysis::Mod; case Intrinsic::invariant_end: + II->getAAMetadata(AAInfo); Loc = AliasAnalysis::Location(II->getArgOperand(2), cast(II->getArgOperand(1)) - ->getZExtValue(), - II->getMetadata(LLVMContext::MD_tbaa)); + ->getZExtValue(), AAInfo); // These intrinsics don't really modify the memory, but returning Mod // will allow them to be handled conservatively. return AliasAnalysis::Mod; default: break; } + } // Otherwise, just do the coarse-grained thing that always works. if (Inst->mayWriteToMemory()) @@ -196,13 +206,13 @@ getCallSiteDependencyFrom(CallSite CS, bool isReadOnlyCall, // Walk backwards through the block, looking for dependencies while (ScanIt != BB->begin()) { // Limit the amount of scanning we do so we don't end up with quadratic - // running time on extreme testcases. + // running time on extreme testcases. --Limit; if (!Limit) return MemDepResult::getUnknown(); Instruction *Inst = --ScanIt; - + // If this inst is a memory op, get the pointer it accessed AliasAnalysis::Location Loc; AliasAnalysis::ModRefResult MR = GetLocation(Inst, Loc, AA); @@ -251,22 +261,17 @@ getCallSiteDependencyFrom(CallSite CS, bool isReadOnlyCall, /// /// MemLocBase, MemLocOffset are lazily computed here the first time the /// base/offs of memloc is needed. -static bool -isLoadLoadClobberIfExtendedToFullWidth(const AliasAnalysis::Location &MemLoc, - const Value *&MemLocBase, - int64_t &MemLocOffs, - const LoadInst *LI, - const DataLayout *TD) { - // If we have no target data, we can't do this. - if (TD == 0) return false; +static bool isLoadLoadClobberIfExtendedToFullWidth( + const AliasAnalysis::Location &MemLoc, const Value *&MemLocBase, + int64_t &MemLocOffs, const LoadInst *LI) { + const DataLayout &DL = LI->getModule()->getDataLayout(); // If we haven't already computed the base/offset of MemLoc, do so now. - if (MemLocBase == 0) - MemLocBase = GetPointerBaseWithConstantOffset(MemLoc.Ptr, MemLocOffs, TD); + if (!MemLocBase) + MemLocBase = GetPointerBaseWithConstantOffset(MemLoc.Ptr, MemLocOffs, DL); - unsigned Size = MemoryDependenceAnalysis:: - getLoadLoadClobberFullWidthSize(MemLocBase, MemLocOffs, MemLoc.Size, - LI, *TD); + unsigned Size = MemoryDependenceAnalysis::getLoadLoadClobberFullWidthSize( + MemLocBase, MemLocOffs, MemLoc.Size, LI); return Size != 0; } @@ -277,37 +282,37 @@ isLoadLoadClobberIfExtendedToFullWidth(const AliasAnalysis::Location &MemLoc, /// 2) safe for the target, and 3) would provide the specified memory /// location value, then this function returns the size in bytes of the /// load width to use. If not, this returns zero. -unsigned MemoryDependenceAnalysis:: -getLoadLoadClobberFullWidthSize(const Value *MemLocBase, int64_t MemLocOffs, - unsigned MemLocSize, const LoadInst *LI, - const DataLayout &TD) { +unsigned MemoryDependenceAnalysis::getLoadLoadClobberFullWidthSize( + const Value *MemLocBase, int64_t MemLocOffs, unsigned MemLocSize, + const LoadInst *LI) { // We can only extend simple integer loads. if (!isa(LI->getType()) || !LI->isSimple()) return 0; // Load widening is hostile to ThreadSanitizer: it may cause false positives // or make the reports more cryptic (access sizes are wrong). - if (LI->getParent()->getParent()->getAttributes(). - hasAttribute(AttributeSet::FunctionIndex, Attribute::ThreadSafety)) + if (LI->getParent()->getParent()->hasFnAttribute(Attribute::SanitizeThread)) return 0; - + + const DataLayout &DL = LI->getModule()->getDataLayout(); + // Get the base of this load. int64_t LIOffs = 0; - const Value *LIBase = - GetPointerBaseWithConstantOffset(LI->getPointerOperand(), LIOffs, &TD); - + const Value *LIBase = + GetPointerBaseWithConstantOffset(LI->getPointerOperand(), LIOffs, DL); + // If the two pointers are not based on the same pointer, we can't tell that // they are related. if (LIBase != MemLocBase) return 0; - + // Okay, the two values are based on the same pointer, but returned as // no-alias. This happens when we have things like two byte loads at "P+1" // and "P+3". Check to see if increasing the size of the "LI" load up to its // alignment (or the largest native integer type) will allow us to load all // the bits required by MemLoc. - + // If MemLoc is before LI, then no widening of LI will help us out. if (MemLocOffs < LIOffs) return 0; - + // Get the alignment of the load in bytes. We assume that it is safe to load // any legal integer up to this size without a problem. For example, if we're // looking at an i8 load on x86-32 that is known 1024 byte aligned, we can @@ -316,25 +321,25 @@ getLoadLoadClobberFullWidthSize(const Value *MemLocBase, int64_t MemLocOffs, unsigned LoadAlign = LI->getAlignment(); int64_t MemLocEnd = MemLocOffs+MemLocSize; - + // If no amount of rounding up will let MemLoc fit into LI, then bail out. if (LIOffs+LoadAlign < MemLocEnd) return 0; - + // This is the size of the load to try. Start with the next larger power of // two. unsigned NewLoadByteSize = LI->getType()->getPrimitiveSizeInBits()/8U; NewLoadByteSize = NextPowerOf2(NewLoadByteSize); - + while (1) { // If this load size is bigger than our known alignment or would not fit // into a native integer register, then we fail. if (NewLoadByteSize > LoadAlign || - !TD.fitsInLegalInteger(NewLoadByteSize*8)) + !DL.fitsInLegalInteger(NewLoadByteSize*8)) return 0; - if (LIOffs+NewLoadByteSize > MemLocEnd && - LI->getParent()->getParent()->getAttributes(). - hasAttribute(AttributeSet::FunctionIndex, Attribute::AddressSafety)) + if (LIOffs + NewLoadByteSize > MemLocEnd && + LI->getParent()->getParent()->hasFnAttribute( + Attribute::SanitizeAddress)) // We will be reading past the location accessed by the original program. // While this is safe in a regular build, Address Safety analysis tools // may start reporting false warnings. So, don't do widening. @@ -343,38 +348,93 @@ getLoadLoadClobberFullWidthSize(const Value *MemLocBase, int64_t MemLocOffs, // If a load of this width would include all of MemLoc, then we succeed. if (LIOffs+NewLoadByteSize >= MemLocEnd) return NewLoadByteSize; - + NewLoadByteSize <<= 1; } } +static bool isVolatile(Instruction *Inst) { + if (LoadInst *LI = dyn_cast(Inst)) + return LI->isVolatile(); + else if (StoreInst *SI = dyn_cast(Inst)) + return SI->isVolatile(); + else if (AtomicCmpXchgInst *AI = dyn_cast(Inst)) + return AI->isVolatile(); + return false; +} + + /// getPointerDependencyFrom - Return the instruction on which a memory /// location depends. If isLoad is true, this routine ignores may-aliases with /// read-only operations. If isLoad is false, this routine ignores may-aliases -/// with reads from read-only locations. +/// with reads from read-only locations. If possible, pass the query +/// instruction as well; this function may take advantage of the metadata +/// annotated to the query instruction to refine the result. MemDepResult MemoryDependenceAnalysis:: -getPointerDependencyFrom(const AliasAnalysis::Location &MemLoc, bool isLoad, - BasicBlock::iterator ScanIt, BasicBlock *BB) { +getPointerDependencyFrom(const AliasAnalysis::Location &MemLoc, bool isLoad, + BasicBlock::iterator ScanIt, BasicBlock *BB, + Instruction *QueryInst) { - const Value *MemLocBase = 0; + const Value *MemLocBase = nullptr; int64_t MemLocOffset = 0; - unsigned Limit = BlockScanLimit; + bool isInvariantLoad = false; + + // We must be careful with atomic accesses, as they may allow another thread + // to touch this location, cloberring it. We are conservative: if the + // QueryInst is not a simple (non-atomic) memory access, we automatically + // return getClobber. + // If it is simple, we know based on the results of + // "Compiler testing via a theory of sound optimisations in the C11/C++11 + // memory model" in PLDI 2013, that a non-atomic location can only be + // clobbered between a pair of a release and an acquire action, with no + // access to the location in between. + // Here is an example for giving the general intuition behind this rule. + // In the following code: + // store x 0; + // release action; [1] + // acquire action; [4] + // %val = load x; + // It is unsafe to replace %val by 0 because another thread may be running: + // acquire action; [2] + // store x 42; + // release action; [3] + // with synchronization from 1 to 2 and from 3 to 4, resulting in %val + // being 42. A key property of this program however is that if either + // 1 or 4 were missing, there would be a race between the store of 42 + // either the store of 0 or the load (making the whole progam racy). + // The paper mentionned above shows that the same property is respected + // by every program that can detect any optimisation of that kind: either + // it is racy (undefined) or there is a release followed by an acquire + // between the pair of accesses under consideration. + + // If the load is invariant, we "know" that it doesn't alias *any* write. We + // do want to respect mustalias results since defs are useful for value + // forwarding, but any mayalias write can be assumed to be noalias. + // Arguably, this logic should be pushed inside AliasAnalysis itself. + if (isLoad && QueryInst) { + LoadInst *LI = dyn_cast(QueryInst); + if (LI && LI->getMetadata(LLVMContext::MD_invariant_load) != nullptr) + isInvariantLoad = true; + } + + const DataLayout &DL = BB->getModule()->getDataLayout(); // Walk backwards through the basic block, looking for dependencies. while (ScanIt != BB->begin()) { + Instruction *Inst = --ScanIt; + + if (IntrinsicInst *II = dyn_cast(Inst)) + // Debug intrinsics don't (and can't) cause dependencies. + if (isa(II)) continue; + // Limit the amount of scanning we do so we don't end up with quadratic // running time on extreme testcases. --Limit; if (!Limit) return MemDepResult::getUnknown(); - Instruction *Inst = --ScanIt; - if (IntrinsicInst *II = dyn_cast(Inst)) { - // Debug intrinsics don't (and can't) cause dependences. - if (isa(II)) continue; - // If we reach a lifetime begin or end marker, then the query ends here // because the value is undefined. if (II->getIntrinsicID() == Intrinsic::lifetime_start) { @@ -391,17 +451,48 @@ getPointerDependencyFrom(const AliasAnalysis::Location &MemLoc, bool isLoad, // Values depend on loads if the pointers are must aliased. This means that // a load depends on another must aliased load from the same value. + // One exception is atomic loads: a value can depend on an atomic load that it + // does not alias with when this atomic load indicates that another thread may + // be accessing the location. if (LoadInst *LI = dyn_cast(Inst)) { + + // While volatile access cannot be eliminated, they do not have to clobber + // non-aliasing locations, as normal accesses, for example, can be safely + // reordered with volatile accesses. + if (LI->isVolatile()) { + if (!QueryInst) + // Original QueryInst *may* be volatile + return MemDepResult::getClobber(LI); + if (isVolatile(QueryInst)) + // Ordering required if QueryInst is itself volatile + return MemDepResult::getClobber(LI); + // Otherwise, volatile doesn't imply any special ordering + } + // Atomic loads have complications involved. + // A Monotonic (or higher) load is OK if the query inst is itself not atomic. // FIXME: This is overly conservative. - if (!LI->isUnordered()) - return MemDepResult::getClobber(LI); + if (LI->isAtomic() && LI->getOrdering() > Unordered) { + if (!QueryInst) + return MemDepResult::getClobber(LI); + if (LI->getOrdering() != Monotonic) + return MemDepResult::getClobber(LI); + if (auto *QueryLI = dyn_cast(QueryInst)) { + if (!QueryLI->isSimple()) + return MemDepResult::getClobber(LI); + } else if (auto *QuerySI = dyn_cast(QueryInst)) { + if (!QuerySI->isSimple()) + return MemDepResult::getClobber(LI); + } else if (QueryInst->mayReadOrWriteMemory()) { + return MemDepResult::getClobber(LI); + } + } AliasAnalysis::Location LoadLoc = AA->getLocation(LI); - + // If we found a pointer, check if it could be the same as our pointer. AliasAnalysis::AliasResult R = AA->alias(LoadLoc, MemLoc); - + if (isLoad) { if (R == AliasAnalysis::NoAlias) { // If this is an over-aligned integer load (for example, @@ -410,15 +501,15 @@ getPointerDependencyFrom(const AliasAnalysis::Location &MemLoc, bool isLoad, // location is 1 byte at P+1). If so, return it as a load/load // clobber result, allowing the client to decide to widen the load if // it wants to. - if (IntegerType *ITy = dyn_cast(LI->getType())) - if (LI->getAlignment()*8 > ITy->getPrimitiveSizeInBits() && + if (IntegerType *ITy = dyn_cast(LI->getType())) { + if (LI->getAlignment() * 8 > ITy->getPrimitiveSizeInBits() && isLoadLoadClobberIfExtendedToFullWidth(MemLoc, MemLocBase, - MemLocOffset, LI, TD)) + MemLocOffset, LI)) return MemDepResult::getClobber(Inst); - + } continue; } - + // Must aliased loads are defs of each other. if (R == AliasAnalysis::MustAlias) return MemDepResult::getDef(Inst); @@ -433,7 +524,7 @@ getPointerDependencyFrom(const AliasAnalysis::Location &MemLoc, bool isLoad, if (R == AliasAnalysis::PartialAlias) return MemDepResult::getClobber(Inst); #endif - + // Random may-alias loads don't depend on each other without a // dependence. continue; @@ -450,11 +541,32 @@ getPointerDependencyFrom(const AliasAnalysis::Location &MemLoc, bool isLoad, // Stores depend on may/must aliased loads. return MemDepResult::getDef(Inst); } - + if (StoreInst *SI = dyn_cast(Inst)) { // Atomic stores have complications involved. + // A Monotonic store is OK if the query inst is itself not atomic. // FIXME: This is overly conservative. - if (!SI->isUnordered()) + if (!SI->isUnordered()) { + if (!QueryInst) + return MemDepResult::getClobber(SI); + if (SI->getOrdering() != Monotonic) + return MemDepResult::getClobber(SI); + if (auto *QueryLI = dyn_cast(QueryInst)) { + if (!QueryLI->isSimple()) + return MemDepResult::getClobber(SI); + } else if (auto *QuerySI = dyn_cast(QueryInst)) { + if (!QuerySI->isSimple()) + return MemDepResult::getClobber(SI); + } else if (QueryInst->mayReadOrWriteMemory()) { + return MemDepResult::getClobber(SI); + } + } + + // FIXME: this is overly conservative. + // While volatile access cannot be eliminated, they do not have to clobber + // non-aliasing locations, as normal accesses can for example be reordered + // with volatile accesses. + if (SI->isVolatile()) return MemDepResult::getClobber(SI); // If alias analysis can tell that this store is guaranteed to not modify @@ -466,14 +578,16 @@ getPointerDependencyFrom(const AliasAnalysis::Location &MemLoc, bool isLoad, // Ok, this store might clobber the query pointer. Check to see if it is // a must alias: in this case, we want to return this as a def. AliasAnalysis::Location StoreLoc = AA->getLocation(SI); - + // If we found a pointer, check if it could be the same as our pointer. AliasAnalysis::AliasResult R = AA->alias(StoreLoc, MemLoc); - + if (R == AliasAnalysis::NoAlias) continue; if (R == AliasAnalysis::MustAlias) return MemDepResult::getDef(Inst); + if (isInvariantLoad) + continue; return MemDepResult::getClobber(Inst); } @@ -487,10 +601,12 @@ getPointerDependencyFrom(const AliasAnalysis::Location &MemLoc, bool isLoad, // need to continue scanning until the malloc call. const TargetLibraryInfo *TLI = AA->getTargetLibraryInfo(); if (isa(Inst) || isNoAliasFn(Inst, TLI)) { - const Value *AccessPtr = GetUnderlyingObject(MemLoc.Ptr, TD); - + const Value *AccessPtr = GetUnderlyingObject(MemLoc.Ptr, DL); + if (AccessPtr == Inst || AA->isMustAlias(Inst, AccessPtr)) return MemDepResult::getDef(Inst); + if (isInvariantLoad) + continue; // Be conservative if the accessed pointer may alias the allocation. if (AA->alias(Inst, AccessPtr) != AliasAnalysis::NoAlias) return MemDepResult::getClobber(Inst); @@ -501,6 +617,9 @@ getPointerDependencyFrom(const AliasAnalysis::Location &MemLoc, bool isLoad, continue; } + if (isInvariantLoad) + continue; + // See if this instruction (e.g. a call or vaarg) mod/ref's the pointer. AliasAnalysis::ModRefResult MR = AA->getModRefInfo(Inst, MemLoc); // If necessary, perform additional analysis. @@ -522,7 +641,7 @@ getPointerDependencyFrom(const AliasAnalysis::Location &MemLoc, bool isLoad, return MemDepResult::getClobber(Inst); } } - + // No dependence found. If this is the entry block of the function, it is // unknown, otherwise it is non-local. if (BB != &BB->getParent()->getEntryBlock()) @@ -534,25 +653,25 @@ getPointerDependencyFrom(const AliasAnalysis::Location &MemLoc, bool isLoad, /// depends. MemDepResult MemoryDependenceAnalysis::getDependency(Instruction *QueryInst) { Instruction *ScanPos = QueryInst; - + // Check for a cached result MemDepResult &LocalCache = LocalDeps[QueryInst]; - + // If the cached entry is non-dirty, just return it. Note that this depends // on MemDepResult's default constructing to 'dirty'. if (!LocalCache.isDirty()) return LocalCache; - + // Otherwise, if we have a dirty entry, we know we can start the scan at that // instruction, which may save us some work. if (Instruction *Inst = LocalCache.getInst()) { ScanPos = Inst; - + RemoveFromReverseMap(ReverseLocalDeps, Inst, QueryInst); } - + BasicBlock *QueryParent = QueryInst->getParent(); - + // Do the scan. if (BasicBlock::iterator(QueryInst) == QueryParent->begin()) { // No dependence found. If this is the entry block of the function, it is @@ -571,7 +690,7 @@ MemDepResult MemoryDependenceAnalysis::getDependency(Instruction *QueryInst) { isLoad |= II->getIntrinsicID() == Intrinsic::lifetime_start; LocalCache = getPointerDependencyFrom(MemLoc, isLoad, ScanPos, - QueryParent); + QueryParent, QueryInst); } else if (isa(QueryInst) || isa(QueryInst)) { CallSite QueryCS(QueryInst); bool isReadOnly = AA->onlyReadsMemory(QueryCS); @@ -581,11 +700,11 @@ MemDepResult MemoryDependenceAnalysis::getDependency(Instruction *QueryInst) { // Non-memory instruction. LocalCache = MemDepResult::getUnknown(); } - + // Remember the result! if (Instruction *I = LocalCache.getInst()) ReverseLocalDeps[I].insert(QueryInst); - + return LocalCache; } @@ -626,7 +745,7 @@ MemoryDependenceAnalysis::getNonLocalCallDependency(CallSite QueryCS) { /// the uncached case, this starts out as the set of predecessors we care /// about. SmallVector DirtyBlocks; - + if (!Cache.empty()) { // Okay, we have a cache entry. If we know it is not dirty, just return it // with no computation. @@ -634,17 +753,17 @@ MemoryDependenceAnalysis::getNonLocalCallDependency(CallSite QueryCS) { ++NumCacheNonLocal; return Cache; } - + // If we already have a partially computed set of results, scan them to // determine what is dirty, seeding our initial DirtyBlocks worklist. for (NonLocalDepInfo::iterator I = Cache.begin(), E = Cache.end(); I != E; ++I) if (I->getResult().isDirty()) DirtyBlocks.push_back(I->getBB()); - + // Sort the cache so that we can do fast binary search lookups below. std::sort(Cache.begin(), Cache.end()); - + ++NumCacheDirtyNonLocal; //cerr << "CACHED CASE: " << DirtyBlocks.size() << " dirty: " // << Cache.size() << " cached: " << *QueryInst; @@ -655,45 +774,45 @@ MemoryDependenceAnalysis::getNonLocalCallDependency(CallSite QueryCS) { DirtyBlocks.push_back(*PI); ++NumUncacheNonLocal; } - + // isReadonlyCall - If this is a read-only call, we can be more aggressive. bool isReadonlyCall = AA->onlyReadsMemory(QueryCS); SmallPtrSet Visited; - + unsigned NumSortedEntries = Cache.size(); DEBUG(AssertSorted(Cache)); - + // Iterate while we still have blocks to update. while (!DirtyBlocks.empty()) { BasicBlock *DirtyBB = DirtyBlocks.back(); DirtyBlocks.pop_back(); - + // Already processed this block? - if (!Visited.insert(DirtyBB)) + if (!Visited.insert(DirtyBB).second) continue; - + // Do a binary search to see if we already have an entry for this block in // the cache set. If so, find it. DEBUG(AssertSorted(Cache, NumSortedEntries)); - NonLocalDepInfo::iterator Entry = + NonLocalDepInfo::iterator Entry = std::upper_bound(Cache.begin(), Cache.begin()+NumSortedEntries, NonLocalDepEntry(DirtyBB)); - if (Entry != Cache.begin() && prior(Entry)->getBB() == DirtyBB) + if (Entry != Cache.begin() && std::prev(Entry)->getBB() == DirtyBB) --Entry; - - NonLocalDepEntry *ExistingResult = 0; - if (Entry != Cache.begin()+NumSortedEntries && + + NonLocalDepEntry *ExistingResult = nullptr; + if (Entry != Cache.begin()+NumSortedEntries && Entry->getBB() == DirtyBB) { // If we already have an entry, and if it isn't already dirty, the block // is done. if (!Entry->getResult().isDirty()) continue; - + // Otherwise, remember this slot so we can update the value. ExistingResult = &*Entry; } - + // If the dirty entry has a pointer, start scanning from it so we don't have // to rescan the entire block. BasicBlock::iterator ScanPos = DirtyBB->end(); @@ -705,10 +824,10 @@ MemoryDependenceAnalysis::getNonLocalCallDependency(CallSite QueryCS) { QueryCS.getInstruction()); } } - + // Find out if this block has a local dependency for QueryInst. MemDepResult Dep; - + if (ScanPos != DirtyBB->begin()) { Dep = getCallSiteDependencyFrom(QueryCS, isReadonlyCall,ScanPos, DirtyBB); } else if (DirtyBB != &DirtyBB->getParent()->getEntryBlock()) { @@ -718,14 +837,14 @@ MemoryDependenceAnalysis::getNonLocalCallDependency(CallSite QueryCS) { } else { Dep = MemDepResult::getNonFuncLocal(); } - + // If we had a dirty entry for the block, update it. Otherwise, just add // a new entry. if (ExistingResult) ExistingResult->setResult(Dep); else Cache.push_back(NonLocalDepEntry(DirtyBB, Dep)); - + // If the block has a dependency (i.e. it isn't completely transparent to // the value), remember the association! if (!Dep.isNonLocal()) { @@ -734,14 +853,14 @@ MemoryDependenceAnalysis::getNonLocalCallDependency(CallSite QueryCS) { if (Instruction *Inst = Dep.getInst()) ReverseNonLocalDeps[Inst].insert(QueryCS.getInstruction()); } else { - + // If the block *is* completely transparent to the load, we need to check // the predecessors of this block. Add them to our worklist. for (BasicBlock **PI = PredCache->GetPreds(DirtyBB); *PI; ++PI) DirtyBlocks.push_back(*PI); } } - + return Cache; } @@ -753,21 +872,64 @@ MemoryDependenceAnalysis::getNonLocalCallDependency(CallSite QueryCS) { /// own block. /// void MemoryDependenceAnalysis:: -getNonLocalPointerDependency(const AliasAnalysis::Location &Loc, bool isLoad, - BasicBlock *FromBB, +getNonLocalPointerDependency(Instruction *QueryInst, SmallVectorImpl &Result) { + + auto getLocation = [](AliasAnalysis *AA, Instruction *Inst) { + if (auto *I = dyn_cast(Inst)) + return AA->getLocation(I); + else if (auto *I = dyn_cast(Inst)) + return AA->getLocation(I); + else if (auto *I = dyn_cast(Inst)) + return AA->getLocation(I); + else if (auto *I = dyn_cast(Inst)) + return AA->getLocation(I); + else if (auto *I = dyn_cast(Inst)) + return AA->getLocation(I); + else + llvm_unreachable("unsupported memory instruction"); + }; + + const AliasAnalysis::Location Loc = getLocation(AA, QueryInst); + bool isLoad = isa(QueryInst); + BasicBlock *FromBB = QueryInst->getParent(); + assert(FromBB); + assert(Loc.Ptr->getType()->isPointerTy() && "Can't get pointer deps of a non-pointer!"); Result.clear(); - PHITransAddr Address(const_cast(Loc.Ptr), TD); - + // This routine does not expect to deal with volatile instructions. + // Doing so would require piping through the QueryInst all the way through. + // TODO: volatiles can't be elided, but they can be reordered with other + // non-volatile accesses. + + // We currently give up on any instruction which is ordered, but we do handle + // atomic instructions which are unordered. + // TODO: Handle ordered instructions + auto isOrdered = [](Instruction *Inst) { + if (LoadInst *LI = dyn_cast(Inst)) { + return !LI->isUnordered(); + } else if (StoreInst *SI = dyn_cast(Inst)) { + return !SI->isUnordered(); + } + return false; + }; + if (isVolatile(QueryInst) || isOrdered(QueryInst)) { + Result.push_back(NonLocalDepResult(FromBB, + MemDepResult::getUnknown(), + const_cast(Loc.Ptr))); + return; + } + const DataLayout &DL = FromBB->getModule()->getDataLayout(); + PHITransAddr Address(const_cast(Loc.Ptr), DL, AC); + // This is the set of blocks we've inspected, and the pointer we consider in // each block. Because of critical edges, we currently bail out if querying // a block with multiple different pointers. This can happen during PHI // translation. DenseMap Visited; - if (!getNonLocalPointerDepFromBB(Address, Loc, isLoad, FromBB, + if (!getNonLocalPointerDepFromBB(QueryInst, Address, Loc, isLoad, FromBB, Result, Visited, true)) return; Result.clear(); @@ -781,10 +943,11 @@ getNonLocalPointerDependency(const AliasAnalysis::Location &Loc, bool isLoad, /// lookup (which may use dirty cache info if available). If we do a lookup, /// add the result to the cache. MemDepResult MemoryDependenceAnalysis:: -GetNonLocalInfoForBlock(const AliasAnalysis::Location &Loc, +GetNonLocalInfoForBlock(Instruction *QueryInst, + const AliasAnalysis::Location &Loc, bool isLoad, BasicBlock *BB, NonLocalDepInfo *Cache, unsigned NumSortedEntries) { - + // Do a binary search to see if we already have an entry for this block in // the cache set. If so, find it. NonLocalDepInfo::iterator Entry = @@ -792,18 +955,18 @@ GetNonLocalInfoForBlock(const AliasAnalysis::Location &Loc, NonLocalDepEntry(BB)); if (Entry != Cache->begin() && (Entry-1)->getBB() == BB) --Entry; - - NonLocalDepEntry *ExistingResult = 0; + + NonLocalDepEntry *ExistingResult = nullptr; if (Entry != Cache->begin()+NumSortedEntries && Entry->getBB() == BB) ExistingResult = &*Entry; - + // If we have a cached entry, and it is non-dirty, use it as the value for // this dependency. if (ExistingResult && !ExistingResult->getResult().isDirty()) { ++NumCacheNonLocalPtr; return ExistingResult->getResult(); - } - + } + // Otherwise, we have to scan for the value. If we have a dirty cache // entry, start scanning from its position, otherwise we scan from the end // of the block. @@ -813,30 +976,31 @@ GetNonLocalInfoForBlock(const AliasAnalysis::Location &Loc, "Instruction invalidated?"); ++NumCacheDirtyNonLocalPtr; ScanPos = ExistingResult->getResult().getInst(); - + // Eliminating the dirty entry from 'Cache', so update the reverse info. ValueIsLoadPair CacheKey(Loc.Ptr, isLoad); RemoveFromReverseMap(ReverseNonLocalPtrDeps, ScanPos, CacheKey); } else { ++NumUncacheNonLocalPtr; } - + // Scan the block for the dependency. - MemDepResult Dep = getPointerDependencyFrom(Loc, isLoad, ScanPos, BB); - + MemDepResult Dep = getPointerDependencyFrom(Loc, isLoad, ScanPos, BB, + QueryInst); + // If we had a dirty entry for the block, update it. Otherwise, just add // a new entry. if (ExistingResult) ExistingResult->setResult(Dep); else Cache->push_back(NonLocalDepEntry(BB, Dep)); - + // If the block has a dependency (i.e. it isn't completely transparent to // the value), remember the reverse association because we just added it // to Cache! if (!Dep.isDef() && !Dep.isClobber()) return Dep; - + // Keep the ReverseNonLocalPtrDeps map up to date so we can efficiently // update MemDep when we remove instructions. Instruction *Inst = Dep.getInst(); @@ -846,10 +1010,10 @@ GetNonLocalInfoForBlock(const AliasAnalysis::Location &Loc, return Dep; } -/// SortNonLocalDepInfoCache - Sort the a NonLocalDepInfo cache, given a certain +/// SortNonLocalDepInfoCache - Sort the NonLocalDepInfo cache, given a certain /// number of elements in the array that are already properly ordered. This is /// optimized for the case when only a few entries are added. -static void +static void SortNonLocalDepInfoCache(MemoryDependenceAnalysis::NonLocalDepInfo &Cache, unsigned NumSortedEntries) { switch (Cache.size() - NumSortedEntries) { @@ -895,27 +1059,27 @@ SortNonLocalDepInfoCache(MemoryDependenceAnalysis::NonLocalDepInfo &Cache, /// not compute dependence information for some reason. This should be treated /// as a clobber dependence on the first instruction in the predecessor block. bool MemoryDependenceAnalysis:: -getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, +getNonLocalPointerDepFromBB(Instruction *QueryInst, + const PHITransAddr &Pointer, const AliasAnalysis::Location &Loc, bool isLoad, BasicBlock *StartBB, SmallVectorImpl &Result, DenseMap &Visited, bool SkipFirstBlock) { - // Look up the cached info for Pointer. ValueIsLoadPair CacheKey(Pointer.getAddr(), isLoad); // Set up a temporary NLPI value. If the map doesn't yet have an entry for // CacheKey, this value will be inserted as the associated value. Otherwise, // it'll be ignored, and we'll have to check to see if the cached size and - // tbaa tag are consistent with the current query. + // aa tags are consistent with the current query. NonLocalPointerInfo InitialNLPI; InitialNLPI.Size = Loc.Size; - InitialNLPI.TBAATag = Loc.TBAATag; + InitialNLPI.AATags = Loc.AATags; // Get the NLPI for CacheKey, inserting one into the map if it doesn't // already have one. - std::pair Pair = + std::pair Pair = NonLocalPointerDeps.insert(std::make_pair(CacheKey, InitialNLPI)); NonLocalPointerInfo *CacheInfo = &Pair.first->second; @@ -935,27 +1099,28 @@ getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, } else if (CacheInfo->Size > Loc.Size) { // This query's Size is less than the cached one. Conservatively restart // the query using the greater size. - return getNonLocalPointerDepFromBB(Pointer, + return getNonLocalPointerDepFromBB(QueryInst, Pointer, Loc.getWithNewSize(CacheInfo->Size), isLoad, StartBB, Result, Visited, SkipFirstBlock); } - // If the query's TBAATag is inconsistent with the cached one, + // If the query's AATags are inconsistent with the cached one, // conservatively throw out the cached data and restart the query with // no tag if needed. - if (CacheInfo->TBAATag != Loc.TBAATag) { - if (CacheInfo->TBAATag) { + if (CacheInfo->AATags != Loc.AATags) { + if (CacheInfo->AATags) { CacheInfo->Pair = BBSkipFirstBlockPair(); - CacheInfo->TBAATag = 0; + CacheInfo->AATags = AAMDNodes(); for (NonLocalDepInfo::iterator DI = CacheInfo->NonLocalDeps.begin(), DE = CacheInfo->NonLocalDeps.end(); DI != DE; ++DI) if (Instruction *Inst = DI->getResult().getInst()) RemoveFromReverseMap(ReverseNonLocalPtrDeps, Inst, CacheKey); CacheInfo->NonLocalDeps.clear(); } - if (Loc.TBAATag) - return getNonLocalPointerDepFromBB(Pointer, Loc.getWithoutTBAATag(), + if (Loc.AATags) + return getNonLocalPointerDepFromBB(QueryInst, + Pointer, Loc.getWithoutAATags(), isLoad, StartBB, Result, Visited, SkipFirstBlock); } @@ -977,25 +1142,34 @@ getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, DenseMap::iterator VI = Visited.find(I->getBB()); if (VI == Visited.end() || VI->second == Pointer.getAddr()) continue; - + // We have a pointer mismatch in a block. Just return clobber, saying // that something was clobbered in this result. We could also do a // non-fully cached query, but there is little point in doing this. return true; } } - + Value *Addr = Pointer.getAddr(); for (NonLocalDepInfo::iterator I = Cache->begin(), E = Cache->end(); I != E; ++I) { Visited.insert(std::make_pair(I->getBB(), Addr)); - if (!I->getResult().isNonLocal() && DT->isReachableFromEntry(I->getBB())) + if (I->getResult().isNonLocal()) { + continue; + } + + if (!DT) { + Result.push_back(NonLocalDepResult(I->getBB(), + MemDepResult::getUnknown(), + Addr)); + } else if (DT->isReachableFromEntry(I->getBB())) { Result.push_back(NonLocalDepResult(I->getBB(), I->getResult(), Addr)); + } } ++NumCacheCompleteNonLocalPtr; return false; } - + // Otherwise, either this is a new block, a block with an invalid cache // pointer or one that we're about to invalidate by putting more info into it // than its valid cache info. If empty, the result will be valid cache info, @@ -1004,10 +1178,10 @@ getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, CacheInfo->Pair = BBSkipFirstBlockPair(StartBB, SkipFirstBlock); else CacheInfo->Pair = BBSkipFirstBlockPair(); - + SmallVector Worklist; Worklist.push_back(StartBB); - + // PredList used inside loop. SmallVector, 16> PredList; @@ -1018,10 +1192,28 @@ getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, // revisit blocks after we insert info for them. unsigned NumSortedEntries = Cache->size(); DEBUG(AssertSorted(*Cache)); - + while (!Worklist.empty()) { BasicBlock *BB = Worklist.pop_back_val(); - + + // If we do process a large number of blocks it becomes very expensive and + // likely it isn't worth worrying about + if (Result.size() > NumResultsLimit) { + Worklist.clear(); + // Sort it now (if needed) so that recursive invocations of + // getNonLocalPointerDepFromBB and other routines that could reuse the + // cache value will only see properly sorted cache arrays. + if (Cache && NumSortedEntries != Cache->size()) { + SortNonLocalDepInfoCache(*Cache, NumSortedEntries); + } + // Since we bail out, the "Cache" set won't contain all of the + // results for the query. This is ok (we can still use it to accelerate + // specific block queries) but we can't do the fastpath "return all + // results from the set". Clear out the indicator for this. + CacheInfo->Pair = BBSkipFirstBlockPair(); + return true; + } + // Skip the first block if we have it. if (!SkipFirstBlock) { // Analyze the dependency of *Pointer in FromBB. See if we already have @@ -1031,16 +1223,24 @@ getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, // Get the dependency info for Pointer in BB. If we have cached // information, we will use it, otherwise we compute it. DEBUG(AssertSorted(*Cache, NumSortedEntries)); - MemDepResult Dep = GetNonLocalInfoForBlock(Loc, isLoad, BB, Cache, + MemDepResult Dep = GetNonLocalInfoForBlock(QueryInst, + Loc, isLoad, BB, Cache, NumSortedEntries); - + // If we got a Def or Clobber, add this to the list of results. - if (!Dep.isNonLocal() && DT->isReachableFromEntry(BB)) { - Result.push_back(NonLocalDepResult(BB, Dep, Pointer.getAddr())); - continue; + if (!Dep.isNonLocal()) { + if (!DT) { + Result.push_back(NonLocalDepResult(BB, + MemDepResult::getUnknown(), + Pointer.getAddr())); + continue; + } else if (DT->isReachableFromEntry(BB)) { + Result.push_back(NonLocalDepResult(BB, Dep, Pointer.getAddr())); + continue; + } } } - + // If 'Pointer' is an instruction defined in this block, then we need to do // phi translation to change it into a value live in the predecessor block. // If not, we just add the predecessors to the worklist and scan them with @@ -1057,7 +1257,7 @@ getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, NewBlocks.push_back(*PI); continue; } - + // If we have seen this block before, but it was with a different // pointer then we have a phi translation failure and we have to treat // this as a clobber. @@ -1072,12 +1272,12 @@ getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, Worklist.append(NewBlocks.begin(), NewBlocks.end()); continue; } - + // We do need to do phi translation, if we know ahead of time we can't phi // translate this value, don't even try. if (!Pointer.IsPotentiallyPHITranslatable()) goto PredTranslationFailure; - + // We may have added values to the cache list before this PHI translation. // If so, we haven't done anything to ensure that the cache remains sorted. // Sort it now (if needed) so that recursive invocations of @@ -1087,7 +1287,7 @@ getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, SortNonLocalDepInfoCache(*Cache, NumSortedEntries); NumSortedEntries = Cache->size(); } - Cache = 0; + Cache = nullptr; PredList.clear(); for (BasicBlock **PI = PredCache->GetPreds(BB); *PI; ++PI) { @@ -1097,10 +1297,10 @@ getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, // Get the PHI translated pointer in this predecessor. This can fail if // not translatable, in which case the getAddr() returns null. PHITransAddr &PredPointer = PredList.back().second; - PredPointer.PHITranslateValue(BB, Pred, 0); + PredPointer.PHITranslateValue(BB, Pred, nullptr); Value *PredPtrVal = PredPointer.getAddr(); - + // Check to see if we have already visited this pred block with another // pointer. If so, we can't do this lookup. This failure can occur // with PHI translation when a critical edge exists and the PHI node in @@ -1117,14 +1317,14 @@ getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, // the analysis and can ignore it. if (InsertRes.first->second == PredPtrVal) continue; - + // Otherwise, the block was previously analyzed with a different // pointer. We can't represent the result of this case, so we just // treat this as a phi translation failure. // Make sure to clean up the Visited map before continuing on to // PredTranslationFailure. - for (unsigned i = 0; i < PredList.size(); i++) + for (unsigned i = 0, n = PredList.size(); i < n; ++i) Visited.erase(PredList[i].first); goto PredTranslationFailure; @@ -1133,10 +1333,10 @@ getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, // Actually process results here; this need to be a separate loop to avoid // calling getNonLocalPointerDepFromBB for blocks we don't want to return - // any results for. (getNonLocalPointerDepFromBB will modify our + // any results for. (getNonLocalPointerDepFromBB will modify our // datastructures in ways the code after the PredTranslationFailure label // doesn't expect.) - for (unsigned i = 0; i < PredList.size(); i++) { + for (unsigned i = 0, n = PredList.size(); i < n; ++i) { BasicBlock *Pred = PredList[i].first; PHITransAddr &PredPointer = PredList[i].second; Value *PredPtrVal = PredPointer.getAddr(); @@ -1146,7 +1346,7 @@ getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, // predecessor, then we have to assume that the pointer is clobbered in // that predecessor. We can still do PRE of the load, which would insert // a computation of the pointer in this predecessor. - if (PredPtrVal == 0) + if (!PredPtrVal) CanTranslate = false; // FIXME: it is entirely possible that PHI translating will end up with @@ -1158,7 +1358,7 @@ getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, // result conflicted with the Visited list; we have to conservatively // assume it is unknown, but this also does not block PRE of the load. if (!CanTranslate || - getNonLocalPointerDepFromBB(PredPointer, + getNonLocalPointerDepFromBB(QueryInst, PredPointer, Loc.getWithNewPtr(PredPtrVal), isLoad, Pred, Result, Visited)) { @@ -1176,12 +1376,12 @@ getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, continue; } } - + // Refresh the CacheInfo/Cache pointer so that it isn't invalidated. CacheInfo = &NonLocalPointerDeps[CacheKey]; Cache = &CacheInfo->NonLocalDeps; NumSortedEntries = Cache->size(); - + // Since we did phi translation, the "Cache" set won't contain all of the // results for the query. This is ok (we can still use it to accelerate // specific block queries) but we can't do the fastpath "return all @@ -1194,20 +1394,20 @@ getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, // The following code is "failure"; we can't produce a sane translation // for the given block. It assumes that we haven't modified any of // our datastructures while processing the current block. - - if (Cache == 0) { + + if (!Cache) { // Refresh the CacheInfo/Cache pointer if it got invalidated. CacheInfo = &NonLocalPointerDeps[CacheKey]; Cache = &CacheInfo->NonLocalDeps; NumSortedEntries = Cache->size(); } - + // Since we failed phi translation, the "Cache" set won't contain all of the // results for the query. This is ok (we can still use it to accelerate // specific block queries) but we can't do the fastpath "return all // results from the set". Clear out the indicator for this. CacheInfo->Pair = BBSkipFirstBlockPair(); - + // If *nothing* works, mark the pointer as unknown. // // If this is the magic first block, return this as a clobber of the whole @@ -1215,13 +1415,13 @@ getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, // we have to bail out. if (SkipFirstBlock) return true; - + for (NonLocalDepInfo::reverse_iterator I = Cache->rbegin(); ; ++I) { assert(I != Cache->rend() && "Didn't find current block??"); if (I->getBB() != BB) continue; - - assert(I->getResult().isNonLocal() && + + assert((I->getResult().isNonLocal() || !DT->isReachableFromEntry(BB)) && "Should only be here with transparent block"); I->setResult(MemDepResult::getUnknown()); Result.push_back(NonLocalDepResult(I->getBB(), I->getResult(), @@ -1240,23 +1440,23 @@ getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, /// CachedNonLocalPointerInfo, remove it. void MemoryDependenceAnalysis:: RemoveCachedNonLocalPointerDependencies(ValueIsLoadPair P) { - CachedNonLocalPointerInfo::iterator It = + CachedNonLocalPointerInfo::iterator It = NonLocalPointerDeps.find(P); if (It == NonLocalPointerDeps.end()) return; - + // Remove all of the entries in the BB->val map. This involves removing // instructions from the reverse map. NonLocalDepInfo &PInfo = It->second.NonLocalDeps; - + for (unsigned i = 0, e = PInfo.size(); i != e; ++i) { Instruction *Target = PInfo[i].getResult().getInst(); - if (Target == 0) continue; // Ignore non-local dep results. + if (!Target) continue; // Ignore non-local dep results. assert(Target->getParent() == PInfo[i].getBB()); - + // Eliminating the dirty entry from 'Cache', so update the reverse info. RemoveFromReverseMap(ReverseNonLocalPtrDeps, Target, P); } - + // Remove P from NonLocalPointerDeps (which deletes NonLocalDepInfo). NonLocalPointerDeps.erase(It); } @@ -1311,20 +1511,20 @@ void MemoryDependenceAnalysis::removeInstruction(Instruction *RemInst) { // Remove this local dependency info. LocalDeps.erase(LocalDepEntry); } - + // If we have any cached pointer dependencies on this instruction, remove // them. If the instruction has non-pointer type, then it can't be a pointer // base. - + // Remove it from both the load info and the store info. The instruction // can't be in either of these maps if it is non-pointer. if (RemInst->getType()->isPointerTy()) { RemoveCachedNonLocalPointerDependencies(ValueIsLoadPair(RemInst, false)); RemoveCachedNonLocalPointerDependencies(ValueIsLoadPair(RemInst, true)); } - + // Loop over all of the things that depend on the instruction we're removing. - // + // SmallVector, 8> ReverseDepsToAdd; // If we find RemInst as a clobber or Def in any of the maps for other values, @@ -1336,29 +1536,26 @@ void MemoryDependenceAnalysis::removeInstruction(Instruction *RemInst) { MemDepResult NewDirtyVal; if (!RemInst->isTerminator()) NewDirtyVal = MemDepResult::getDirty(++BasicBlock::iterator(RemInst)); - + ReverseDepMapType::iterator ReverseDepIt = ReverseLocalDeps.find(RemInst); if (ReverseDepIt != ReverseLocalDeps.end()) { - SmallPtrSet &ReverseDeps = ReverseDepIt->second; // RemInst can't be the terminator if it has local stuff depending on it. - assert(!ReverseDeps.empty() && !isa(RemInst) && + assert(!ReverseDepIt->second.empty() && !isa(RemInst) && "Nothing can locally depend on a terminator"); - - for (SmallPtrSet::iterator I = ReverseDeps.begin(), - E = ReverseDeps.end(); I != E; ++I) { - Instruction *InstDependingOnRemInst = *I; + + for (Instruction *InstDependingOnRemInst : ReverseDepIt->second) { assert(InstDependingOnRemInst != RemInst && "Already removed our local dep info"); - + LocalDeps[InstDependingOnRemInst] = NewDirtyVal; - + // Make sure to remember that new things depend on NewDepInst. assert(NewDirtyVal.getInst() && "There is no way something else can have " "a local dep on this if it is a terminator!"); - ReverseDepsToAdd.push_back(std::make_pair(NewDirtyVal.getInst(), + ReverseDepsToAdd.push_back(std::make_pair(NewDirtyVal.getInst(), InstDependingOnRemInst)); } - + ReverseLocalDeps.erase(ReverseDepIt); // Add new reverse deps after scanning the set, to avoid invalidating the @@ -1369,27 +1566,25 @@ void MemoryDependenceAnalysis::removeInstruction(Instruction *RemInst) { ReverseDepsToAdd.pop_back(); } } - + ReverseDepIt = ReverseNonLocalDeps.find(RemInst); if (ReverseDepIt != ReverseNonLocalDeps.end()) { - SmallPtrSet &Set = ReverseDepIt->second; - for (SmallPtrSet::iterator I = Set.begin(), E = Set.end(); - I != E; ++I) { - assert(*I != RemInst && "Already removed NonLocalDep info for RemInst"); - - PerInstNLInfo &INLD = NonLocalDeps[*I]; + for (Instruction *I : ReverseDepIt->second) { + assert(I != RemInst && "Already removed NonLocalDep info for RemInst"); + + PerInstNLInfo &INLD = NonLocalDeps[I]; // The information is now dirty! INLD.second = true; - - for (NonLocalDepInfo::iterator DI = INLD.first.begin(), + + for (NonLocalDepInfo::iterator DI = INLD.first.begin(), DE = INLD.first.end(); DI != DE; ++DI) { if (DI->getResult().getInst() != RemInst) continue; - + // Convert to a dirty entry for the subsequent instruction. DI->setResult(NewDirtyVal); - + if (Instruction *NextI = NewDirtyVal.getInst()) - ReverseDepsToAdd.push_back(std::make_pair(NextI, *I)); + ReverseDepsToAdd.push_back(std::make_pair(NextI, I)); } } @@ -1402,67 +1597,66 @@ void MemoryDependenceAnalysis::removeInstruction(Instruction *RemInst) { ReverseDepsToAdd.pop_back(); } } - + // If the instruction is in ReverseNonLocalPtrDeps then it appears as a // value in the NonLocalPointerDeps info. ReverseNonLocalPtrDepTy::iterator ReversePtrDepIt = ReverseNonLocalPtrDeps.find(RemInst); if (ReversePtrDepIt != ReverseNonLocalPtrDeps.end()) { - SmallPtrSet &Set = ReversePtrDepIt->second; SmallVector,8> ReversePtrDepsToAdd; - - for (SmallPtrSet::iterator I = Set.begin(), - E = Set.end(); I != E; ++I) { - ValueIsLoadPair P = *I; + + for (ValueIsLoadPair P : ReversePtrDepIt->second) { assert(P.getPointer() != RemInst && "Already removed NonLocalPointerDeps info for RemInst"); - + NonLocalDepInfo &NLPDI = NonLocalPointerDeps[P].NonLocalDeps; - + // The cache is not valid for any specific block anymore. NonLocalPointerDeps[P].Pair = BBSkipFirstBlockPair(); - + // Update any entries for RemInst to use the instruction after it. for (NonLocalDepInfo::iterator DI = NLPDI.begin(), DE = NLPDI.end(); DI != DE; ++DI) { if (DI->getResult().getInst() != RemInst) continue; - + // Convert to a dirty entry for the subsequent instruction. DI->setResult(NewDirtyVal); - + if (Instruction *NewDirtyInst = NewDirtyVal.getInst()) ReversePtrDepsToAdd.push_back(std::make_pair(NewDirtyInst, P)); } - + // Re-sort the NonLocalDepInfo. Changing the dirty entry to its // subsequent value may invalidate the sortedness. std::sort(NLPDI.begin(), NLPDI.end()); } - + ReverseNonLocalPtrDeps.erase(ReversePtrDepIt); - + while (!ReversePtrDepsToAdd.empty()) { ReverseNonLocalPtrDeps[ReversePtrDepsToAdd.back().first] .insert(ReversePtrDepsToAdd.back().second); ReversePtrDepsToAdd.pop_back(); } } - - + + assert(!NonLocalDeps.count(RemInst) && "RemInst got reinserted?"); AA->deleteValue(RemInst); DEBUG(verifyRemoved(RemInst)); } /// verifyRemoved - Verify that the specified instruction does not occur -/// in our internal data structures. +/// in our internal data structures. This function verifies by asserting in +/// debug builds. void MemoryDependenceAnalysis::verifyRemoved(Instruction *D) const { +#ifndef NDEBUG for (LocalDepMapType::const_iterator I = LocalDeps.begin(), E = LocalDeps.end(); I != E; ++I) { assert(I->first != D && "Inst occurs in data structures"); assert(I->second.getInst() != D && "Inst occurs in data structures"); } - + for (CachedNonLocalPointerInfo::const_iterator I =NonLocalPointerDeps.begin(), E = NonLocalPointerDeps.end(); I != E; ++I) { assert(I->first.getPointer() != D && "Inst occurs in NLPD map key"); @@ -1471,7 +1665,7 @@ void MemoryDependenceAnalysis::verifyRemoved(Instruction *D) const { II != E; ++II) assert(II->getResult().getInst() != D && "Inst occurs as NLPD value"); } - + for (NonLocalDepMapType::const_iterator I = NonLocalDeps.begin(), E = NonLocalDeps.end(); I != E; ++I) { assert(I->first != D && "Inst occurs in data structures"); @@ -1480,34 +1674,31 @@ void MemoryDependenceAnalysis::verifyRemoved(Instruction *D) const { EE = INLD.first.end(); II != EE; ++II) assert(II->getResult().getInst() != D && "Inst occurs in data structures"); } - + for (ReverseDepMapType::const_iterator I = ReverseLocalDeps.begin(), E = ReverseLocalDeps.end(); I != E; ++I) { assert(I->first != D && "Inst occurs in data structures"); - for (SmallPtrSet::const_iterator II = I->second.begin(), - EE = I->second.end(); II != EE; ++II) - assert(*II != D && "Inst occurs in data structures"); + for (Instruction *Inst : I->second) + assert(Inst != D && "Inst occurs in data structures"); } - + for (ReverseDepMapType::const_iterator I = ReverseNonLocalDeps.begin(), E = ReverseNonLocalDeps.end(); I != E; ++I) { assert(I->first != D && "Inst occurs in data structures"); - for (SmallPtrSet::const_iterator II = I->second.begin(), - EE = I->second.end(); II != EE; ++II) - assert(*II != D && "Inst occurs in data structures"); + for (Instruction *Inst : I->second) + assert(Inst != D && "Inst occurs in data structures"); } - + for (ReverseNonLocalPtrDepTy::const_iterator I = ReverseNonLocalPtrDeps.begin(), E = ReverseNonLocalPtrDeps.end(); I != E; ++I) { assert(I->first != D && "Inst occurs in rev NLPD map"); - - for (SmallPtrSet::const_iterator II = I->second.begin(), - E = I->second.end(); II != E; ++II) - assert(*II != ValueIsLoadPair(D, false) && - *II != ValueIsLoadPair(D, true) && + + for (ValueIsLoadPair P : I->second) + assert(P != ValueIsLoadPair(D, false) && + P != ValueIsLoadPair(D, true) && "Inst occurs in ReverseNonLocalPtrDeps map"); } - +#endif }