#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Analysis/MemoryBuiltins.h"
#include "llvm/Analysis/PHITransAddr.h"
+#include "llvm/Analysis/OrderedBasicBlock.h"
#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/Analysis/TargetLibraryInfo.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/Dominators.h"
#include "llvm/IR/Function.h"
"Number of block queries that were completely cached");
// Limit for the number of instructions to scan in a block.
-static const unsigned int BlockScanLimit = 100;
+
+static cl::opt<unsigned> BlockScanLimit(
+ "memdep-block-scan-limit", cl::Hidden, cl::init(100),
+ cl::desc("The number of instructions to scan in a block in memory "
+ "dependency analysis (default = 100)"));
// Limit on the number of memdep results to process.
static const unsigned int NumResultsLimit = 100;
// Register this pass...
INITIALIZE_PASS_BEGIN(MemoryDependenceAnalysis, "memdep",
"Memory Dependence Analysis", false, true)
-INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)
INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
+INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)
+INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass)
INITIALIZE_PASS_END(MemoryDependenceAnalysis, "memdep",
"Memory Dependence Analysis", false, true)
MemoryDependenceAnalysis::MemoryDependenceAnalysis()
- : FunctionPass(ID), PredCache() {
+ : FunctionPass(ID) {
initializeMemoryDependenceAnalysisPass(*PassRegistry::getPassRegistry());
}
MemoryDependenceAnalysis::~MemoryDependenceAnalysis() {
ReverseLocalDeps.clear();
ReverseNonLocalDeps.clear();
ReverseNonLocalPtrDeps.clear();
- PredCache->clear();
+ PredCache.clear();
}
/// getAnalysisUsage - Does not modify anything. It uses Alias Analysis.
AU.setPreservesAll();
AU.addRequired<AssumptionCacheTracker>();
AU.addRequiredTransitive<AliasAnalysis>();
+ AU.addRequiredTransitive<TargetLibraryInfoWrapperPass>();
}
bool MemoryDependenceAnalysis::runOnFunction(Function &F) {
DominatorTreeWrapperPass *DTWP =
getAnalysisIfAvailable<DominatorTreeWrapperPass>();
DT = DTWP ? &DTWP->getDomTree() : nullptr;
- if (!PredCache)
- PredCache.reset(new PredIteratorCache());
+ TLI = &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI();
return false;
}
/// location, fill in Loc with the details, otherwise set Loc.Ptr to null.
/// Return a ModRefInfo value describing the general behavior of the
/// instruction.
-static
-AliasAnalysis::ModRefResult GetLocation(const Instruction *Inst,
- AliasAnalysis::Location &Loc,
- AliasAnalysis *AA) {
+static ModRefInfo GetLocation(const Instruction *Inst, MemoryLocation &Loc,
+ const TargetLibraryInfo &TLI) {
if (const LoadInst *LI = dyn_cast<LoadInst>(Inst)) {
if (LI->isUnordered()) {
- Loc = AA->getLocation(LI);
- return AliasAnalysis::Ref;
+ Loc = MemoryLocation::get(LI);
+ return MRI_Ref;
}
if (LI->getOrdering() == Monotonic) {
- Loc = AA->getLocation(LI);
- return AliasAnalysis::ModRef;
+ Loc = MemoryLocation::get(LI);
+ return MRI_ModRef;
}
- Loc = AliasAnalysis::Location();
- return AliasAnalysis::ModRef;
+ Loc = MemoryLocation();
+ return MRI_ModRef;
}
if (const StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
if (SI->isUnordered()) {
- Loc = AA->getLocation(SI);
- return AliasAnalysis::Mod;
+ Loc = MemoryLocation::get(SI);
+ return MRI_Mod;
}
if (SI->getOrdering() == Monotonic) {
- Loc = AA->getLocation(SI);
- return AliasAnalysis::ModRef;
+ Loc = MemoryLocation::get(SI);
+ return MRI_ModRef;
}
- Loc = AliasAnalysis::Location();
- return AliasAnalysis::ModRef;
+ Loc = MemoryLocation();
+ return MRI_ModRef;
}
if (const VAArgInst *V = dyn_cast<VAArgInst>(Inst)) {
- Loc = AA->getLocation(V);
- return AliasAnalysis::ModRef;
+ Loc = MemoryLocation::get(V);
+ return MRI_ModRef;
}
- if (const CallInst *CI = isFreeCall(Inst, AA->getTargetLibraryInfo())) {
+ if (const CallInst *CI = isFreeCall(Inst, &TLI)) {
// calls to free() deallocate the entire structure
- Loc = AliasAnalysis::Location(CI->getArgOperand(0));
- return AliasAnalysis::Mod;
+ Loc = MemoryLocation(CI->getArgOperand(0));
+ return MRI_Mod;
}
if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(Inst)) {
case Intrinsic::lifetime_end:
case Intrinsic::invariant_start:
II->getAAMetadata(AAInfo);
- Loc = AliasAnalysis::Location(II->getArgOperand(1),
- cast<ConstantInt>(II->getArgOperand(0))
- ->getZExtValue(), AAInfo);
+ Loc = MemoryLocation(
+ II->getArgOperand(1),
+ cast<ConstantInt>(II->getArgOperand(0))->getZExtValue(), AAInfo);
// These intrinsics don't really modify the memory, but returning Mod
// will allow them to be handled conservatively.
- return AliasAnalysis::Mod;
+ return MRI_Mod;
case Intrinsic::invariant_end:
II->getAAMetadata(AAInfo);
- Loc = AliasAnalysis::Location(II->getArgOperand(2),
- cast<ConstantInt>(II->getArgOperand(1))
- ->getZExtValue(), AAInfo);
+ Loc = MemoryLocation(
+ II->getArgOperand(2),
+ cast<ConstantInt>(II->getArgOperand(1))->getZExtValue(), AAInfo);
// These intrinsics don't really modify the memory, but returning Mod
// will allow them to be handled conservatively.
- return AliasAnalysis::Mod;
+ return MRI_Mod;
default:
break;
}
// Otherwise, just do the coarse-grained thing that always works.
if (Inst->mayWriteToMemory())
- return AliasAnalysis::ModRef;
+ return MRI_ModRef;
if (Inst->mayReadFromMemory())
- return AliasAnalysis::Ref;
- return AliasAnalysis::NoModRef;
+ return MRI_Ref;
+ return MRI_NoModRef;
}
/// getCallSiteDependencyFrom - Private helper for finding the local
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);
+ MemoryLocation Loc;
+ ModRefInfo MR = GetLocation(Inst, Loc, *TLI);
if (Loc.Ptr) {
// A simple instruction.
- if (AA->getModRefInfo(CS, Loc) != AliasAnalysis::NoModRef)
+ if (AA->getModRefInfo(CS, Loc) != MRI_NoModRef)
return MemDepResult::getClobber(Inst);
continue;
}
- if (CallSite InstCS = cast<Value>(Inst)) {
+ if (auto InstCS = CallSite(Inst)) {
// Debug intrinsics don't cause dependences.
if (isa<DbgInfoIntrinsic>(Inst)) continue;
// If these two calls do not interfere, look past it.
switch (AA->getModRefInfo(CS, InstCS)) {
- case AliasAnalysis::NoModRef:
+ case MRI_NoModRef:
// If the two calls are the same, return InstCS as a Def, so that
// CS can be found redundant and eliminated.
- if (isReadOnlyCall && !(MR & AliasAnalysis::Mod) &&
+ if (isReadOnlyCall && !(MR & MRI_Mod) &&
CS.getInstruction()->isIdenticalToWhenDefined(Inst))
return MemDepResult::getDef(Inst);
// If we could not obtain a pointer for the instruction and the instruction
// touches memory then assume that this is a dependency.
- if (MR != AliasAnalysis::NoModRef)
+ if (MR != MRI_NoModRef)
return MemDepResult::getClobber(Inst);
}
///
/// 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) {
+static bool isLoadLoadClobberIfExtendedToFullWidth(const MemoryLocation &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.
/// 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,
- Instruction *QueryInst) {
+MemDepResult MemoryDependenceAnalysis::getPointerDependencyFrom(
+ const MemoryLocation &MemLoc, bool isLoad, BasicBlock::iterator ScanIt,
+ BasicBlock *BB, Instruction *QueryInst) {
const Value *MemLocBase = nullptr;
int64_t MemLocOffset = 0;
// 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.
- bool HasSeenAcquire = false;
+ // 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<LoadInst>(QueryInst);
if (LI && LI->getMetadata(LLVMContext::MD_invariant_load) != nullptr)
const DataLayout &DL = BB->getModule()->getDataLayout();
+ // Create a numbered basic block to lazily compute and cache instruction
+ // positions inside a BB. This is used to provide fast queries for relative
+ // position between two instructions in a BB and can be used by
+ // AliasAnalysis::callCapturesBefore.
+ OrderedBasicBlock OBB(BB);
+
// Walk backwards through the basic block, looking for dependencies.
while (ScanIt != BB->begin()) {
Instruction *Inst = --ScanIt;
// pointer, not on query pointers that are indexed off of them. It'd
// be nice to handle that at some point (the right approach is to use
// GetPointerBaseWithConstantOffset).
- if (AA->isMustAlias(AliasAnalysis::Location(II->getArgOperand(1)),
- MemLoc))
+ if (AA->isMustAlias(MemoryLocation(II->getArgOperand(1)), MemLoc))
return MemDepResult::getDef(II);
continue;
}
// Atomic loads have complications involved.
// A Monotonic (or higher) load is OK if the query inst is itself not atomic.
- // An Acquire (or higher) load sets the HasSeenAcquire flag, so that any
- // release store will know to return getClobber.
// FIXME: This is overly conservative.
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<LoadInst>(QueryInst)) {
if (!QueryLI->isSimple())
return MemDepResult::getClobber(LI);
} else if (QueryInst->mayReadOrWriteMemory()) {
return MemDepResult::getClobber(LI);
}
-
- if (isAtLeastAcquire(LI->getOrdering()))
- HasSeenAcquire = true;
}
- AliasAnalysis::Location LoadLoc = AA->getLocation(LI);
+ MemoryLocation LoadLoc = MemoryLocation::get(LI);
// If we found a pointer, check if it could be the same as our pointer.
- AliasAnalysis::AliasResult R = AA->alias(LoadLoc, MemLoc);
+ AliasResult R = AA->alias(LoadLoc, MemLoc);
if (isLoad) {
- if (R == AliasAnalysis::NoAlias) {
+ if (R == NoAlias) {
// If this is an over-aligned integer load (for example,
// "load i8* %P, align 4") see if it would obviously overlap with the
// queried location if widened to a larger load (e.g. if the queried
}
// Must aliased loads are defs of each other.
- if (R == AliasAnalysis::MustAlias)
+ if (R == MustAlias)
return MemDepResult::getDef(Inst);
#if 0 // FIXME: Temporarily disabled. GVN is cleverly rewriting loads
// If we have a partial alias, then return this as a clobber for the
// client to handle.
- if (R == AliasAnalysis::PartialAlias)
+ if (R == PartialAlias)
return MemDepResult::getClobber(Inst);
#endif
}
// Stores don't depend on other no-aliased accesses.
- if (R == AliasAnalysis::NoAlias)
+ if (R == NoAlias)
continue;
// Stores don't alias loads from read-only memory.
if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
// Atomic stores have complications involved.
// A Monotonic store is OK if the query inst is itself not atomic.
- // A Release (or higher) store further requires that no acquire load
- // has been seen.
// FIXME: This is overly conservative.
if (!SI->isUnordered()) {
if (!QueryInst)
return MemDepResult::getClobber(SI);
+ if (SI->getOrdering() != Monotonic)
+ return MemDepResult::getClobber(SI);
if (auto *QueryLI = dyn_cast<LoadInst>(QueryInst)) {
if (!QueryLI->isSimple())
return MemDepResult::getClobber(SI);
} else if (QueryInst->mayReadOrWriteMemory()) {
return MemDepResult::getClobber(SI);
}
-
- if (HasSeenAcquire && isAtLeastRelease(SI->getOrdering()))
- return MemDepResult::getClobber(SI);
}
// FIXME: this is overly conservative.
// If alias analysis can tell that this store is guaranteed to not modify
// the query pointer, ignore it. Use getModRefInfo to handle cases where
// the query pointer points to constant memory etc.
- if (AA->getModRefInfo(SI, MemLoc) == AliasAnalysis::NoModRef)
+ if (AA->getModRefInfo(SI, MemLoc) == MRI_NoModRef)
continue;
// 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);
+ MemoryLocation StoreLoc = MemoryLocation::get(SI);
// If we found a pointer, check if it could be the same as our pointer.
- AliasAnalysis::AliasResult R = AA->alias(StoreLoc, MemLoc);
+ AliasResult R = AA->alias(StoreLoc, MemLoc);
- if (R == AliasAnalysis::NoAlias)
+ if (R == NoAlias)
continue;
- if (R == AliasAnalysis::MustAlias)
+ if (R == MustAlias)
return MemDepResult::getDef(Inst);
if (isInvariantLoad)
continue;
// 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.
- const TargetLibraryInfo *TLI = AA->getTargetLibraryInfo();
if (isa<AllocaInst>(Inst) || isNoAliasFn(Inst, TLI)) {
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)
+ if (AA->alias(Inst, AccessPtr) != NoAlias)
return MemDepResult::getClobber(Inst);
// If the allocation is not aliased and does not read memory (like
// strdup), it is safe to ignore.
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);
+ ModRefInfo MR = AA->getModRefInfo(Inst, MemLoc);
// If necessary, perform additional analysis.
- if (MR == AliasAnalysis::ModRef)
- MR = AA->callCapturesBefore(Inst, MemLoc, DT);
+ if (MR == MRI_ModRef)
+ MR = AA->callCapturesBefore(Inst, MemLoc, DT, &OBB);
switch (MR) {
- case AliasAnalysis::NoModRef:
+ case MRI_NoModRef:
// If the call has no effect on the queried pointer, just ignore it.
continue;
- case AliasAnalysis::Mod:
+ case MRI_Mod:
return MemDepResult::getClobber(Inst);
- case AliasAnalysis::Ref:
+ case MRI_Ref:
// If the call is known to never store to the pointer, and if this is a
// load query, we can safely ignore it (scan past it).
if (isLoad)
else
LocalCache = MemDepResult::getNonFuncLocal();
} else {
- AliasAnalysis::Location MemLoc;
- AliasAnalysis::ModRefResult MR = GetLocation(QueryInst, MemLoc, AA);
+ MemoryLocation MemLoc;
+ ModRefInfo MR = GetLocation(QueryInst, MemLoc, *TLI);
if (MemLoc.Ptr) {
// If we can do a pointer scan, make it happen.
- bool isLoad = !(MR & AliasAnalysis::Mod);
+ bool isLoad = !(MR & MRI_Mod);
if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(QueryInst))
isLoad |= II->getIntrinsicID() == Intrinsic::lifetime_start;
} else {
// Seed DirtyBlocks with each of the preds of QueryInst's block.
BasicBlock *QueryBB = QueryCS.getInstruction()->getParent();
- for (BasicBlock **PI = PredCache->GetPreds(QueryBB); *PI; ++PI)
- DirtyBlocks.push_back(*PI);
+ for (BasicBlock *Pred : PredCache.get(QueryBB))
+ DirtyBlocks.push_back(Pred);
++NumUncacheNonLocal;
}
// 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);
+ for (BasicBlock *Pred : PredCache.get(DirtyBB))
+ DirtyBlocks.push_back(Pred);
}
}
void MemoryDependenceAnalysis::
getNonLocalPointerDependency(Instruction *QueryInst,
SmallVectorImpl<NonLocalDepResult> &Result) {
-
- auto getLocation = [](AliasAnalysis *AA, Instruction *Inst) {
- if (auto *I = dyn_cast<LoadInst>(Inst))
- return AA->getLocation(I);
- else if (auto *I = dyn_cast<StoreInst>(Inst))
- return AA->getLocation(I);
- else if (auto *I = dyn_cast<VAArgInst>(Inst))
- return AA->getLocation(I);
- else if (auto *I = dyn_cast<AtomicCmpXchgInst>(Inst))
- return AA->getLocation(I);
- else if (auto *I = dyn_cast<AtomicRMWInst>(Inst))
- return AA->getLocation(I);
- else
- llvm_unreachable("unsupported memory instruction");
- };
-
- const AliasAnalysis::Location Loc = getLocation(AA, QueryInst);
+ const MemoryLocation Loc = MemoryLocation::get(QueryInst);
bool isLoad = isa<LoadInst>(QueryInst);
BasicBlock *FromBB = QueryInst->getParent();
assert(FromBB);
/// Pointer/PointeeSize using either cached information in Cache or by doing a
/// lookup (which may use dirty cache info if available). If we do a lookup,
/// add the result to the cache.
-MemDepResult MemoryDependenceAnalysis::
-GetNonLocalInfoForBlock(Instruction *QueryInst,
- const AliasAnalysis::Location &Loc,
- bool isLoad, BasicBlock *BB,
- NonLocalDepInfo *Cache, unsigned NumSortedEntries) {
+MemDepResult MemoryDependenceAnalysis::GetNonLocalInfoForBlock(
+ Instruction *QueryInst, const MemoryLocation &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.
/// This function returns false on success, or true to indicate that it could
/// 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(Instruction *QueryInst,
- const PHITransAddr &Pointer,
- const AliasAnalysis::Location &Loc,
- bool isLoad, BasicBlock *StartBB,
- SmallVectorImpl<NonLocalDepResult> &Result,
- DenseMap<BasicBlock*, Value*> &Visited,
- bool SkipFirstBlock) {
+bool MemoryDependenceAnalysis::getNonLocalPointerDepFromBB(
+ Instruction *QueryInst, const PHITransAddr &Pointer,
+ const MemoryLocation &Loc, bool isLoad, BasicBlock *StartBB,
+ SmallVectorImpl<NonLocalDepResult> &Result,
+ DenseMap<BasicBlock *, Value *> &Visited, bool SkipFirstBlock) {
// Look up the cached info for Pointer.
ValueIsLoadPair CacheKey(Pointer.getAddr(), isLoad);
if (!Pointer.NeedsPHITranslationFromBlock(BB)) {
SkipFirstBlock = false;
SmallVector<BasicBlock*, 16> NewBlocks;
- for (BasicBlock **PI = PredCache->GetPreds(BB); *PI; ++PI) {
+ for (BasicBlock *Pred : PredCache.get(BB)) {
// Verify that we haven't looked at this block yet.
std::pair<DenseMap<BasicBlock*,Value*>::iterator, bool>
- InsertRes = Visited.insert(std::make_pair(*PI, Pointer.getAddr()));
+ InsertRes = Visited.insert(std::make_pair(Pred, Pointer.getAddr()));
if (InsertRes.second) {
// First time we've looked at *PI.
- NewBlocks.push_back(*PI);
+ NewBlocks.push_back(Pred);
continue;
}
Cache = nullptr;
PredList.clear();
- for (BasicBlock **PI = PredCache->GetPreds(BB); *PI; ++PI) {
- BasicBlock *Pred = *PI;
+ for (BasicBlock *Pred : PredCache.get(BB)) {
PredList.push_back(std::make_pair(Pred, 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, nullptr);
-
+ PredPointer.PHITranslateValue(BB, Pred, DT, /*MustDominate=*/false);
Value *PredPtrVal = PredPointer.getAddr();
// Check to see if we have already visited this pred block with another
/// This needs to be done when the CFG changes, e.g., due to splitting
/// critical edges.
void MemoryDependenceAnalysis::invalidateCachedPredecessors() {
- PredCache->clear();
+ PredCache.clear();
}
/// removeInstruction - Remove an instruction from the dependence analysis,
assert(!NonLocalDeps.count(RemInst) && "RemInst got reinserted?");
- AA->deleteValue(RemInst);
DEBUG(verifyRemoved(RemInst));
}
/// verifyRemoved - Verify that the specified instruction does not occur
projects / oota-llvm.git / blobdiff