/// HandleFree - Handle frees of entire structures whose dependency is a store
/// to a field of that structure.
bool DSE::HandleFree(CallInst *F) {
MemDepResult Dep = MD->getDependency(F);
do {
if (Dep.isNonLocal()) return false;
Instruction *Dependency = Dep.getInst();
if (!hasMemoryWrite(Dependency) || !isRemovable(Dependency))
return false;
Value *DepPointer =
getStoredPointerOperand(Dependency)->getUnderlyingObject();
// Check for aliasing.
if (!AA->isMustAlias(F->getArgOperand(0), DepPointer))
return false;
// DCE instructions only used to calculate that store
DeleteDeadInstruction(Dependency, *MD);
++NumFastStores;
// Inst's old Dependency is now deleted. Compute the next dependency,
// which may also be dead, as in
// s[0] = 0;
// s[1] = 0; // This has just been deleted.
// free(s);
Dep = MD->getDependency(F);
} while (!Dep.isNonLocal());
return true;
}
/// Handle frees of entire structures whose dependency is a store
/// to a field of that structure.
static bool handleFree(CallInst *F, AliasAnalysis *AA,
MemoryDependenceResults *MD, DominatorTree *DT,
const TargetLibraryInfo *TLI,
InstOverlapIntervalsTy &IOL,
DenseMap<Instruction*, size_t> *InstrOrdering) {
bool MadeChange = false;
MemoryLocation Loc = MemoryLocation(F->getOperand(0));
SmallVector<BasicBlock *, 16> Blocks;
Blocks.push_back(F->getParent());
const DataLayout &DL = F->getModule()->getDataLayout();
while (!Blocks.empty()) {
BasicBlock *BB = Blocks.pop_back_val();
Instruction *InstPt = BB->getTerminator();
if (BB == F->getParent()) InstPt = F;
MemDepResult Dep =
MD->getPointerDependencyFrom(Loc, false, InstPt->getIterator(), BB);
while (Dep.isDef() || Dep.isClobber()) {
Instruction *Dependency = Dep.getInst();
if (!hasMemoryWrite(Dependency, *TLI) || !isRemovable(Dependency))
break;
Value *DepPointer =
GetUnderlyingObject(getStoredPointerOperand(Dependency), DL);
// Check for aliasing.
if (!AA->isMustAlias(F->getArgOperand(0), DepPointer))
break;
DEBUG(dbgs() << "DSE: Dead Store to soon to be freed memory:\n DEAD: "
<< *Dependency << '\n');
// DCE instructions only used to calculate that store.
BasicBlock::iterator BBI(Dependency);
deleteDeadInstruction(Dependency, &BBI, *MD, *TLI, IOL, InstrOrdering);
++NumFastStores;
MadeChange = true;
// Inst's old Dependency is now deleted. Compute the next dependency,
// which may also be dead, as in
// s[0] = 0;
// s[1] = 0; // This has just been deleted.
// free(s);
Dep = MD->getPointerDependencyFrom(Loc, false, BBI, BB);
}
if (Dep.isNonLocal())
findUnconditionalPreds(Blocks, BB, DT);
}
return MadeChange;
}
DataDependence::DepInfo DataDependence::getDepInfo(MemDepResult dep) {
if (dep.isClobber())
return DepInfo(dep.getInst(), Clobber);
if (dep.isDef())
return DepInfo(dep.getInst(), Def);
if (dep.isNonFuncLocal())
return DepInfo(dep.getInst(), NonFuncLocal);
if (dep.isUnknown())
return DepInfo(dep.getInst(), Unknown);
if (dep.isNonLocal())
return DepInfo(dep.getInst(), NonLocal);
llvm_unreachable("unknown dependence type");
}
/// HandleFree - Handle frees of entire structures whose dependency is a store
/// to a field of that structure.
bool DSE::HandleFree(CallInst *F) {
bool MadeChange = false;
MemoryLocation Loc = MemoryLocation(F->getOperand(0));
SmallVector<BasicBlock *, 16> Blocks;
Blocks.push_back(F->getParent());
const DataLayout &DL = F->getModule()->getDataLayout();
while (!Blocks.empty()) {
BasicBlock *BB = Blocks.pop_back_val();
Instruction *InstPt = BB->getTerminator();
if (BB == F->getParent()) InstPt = F;
MemDepResult Dep = MD->getPointerDependencyFrom(Loc, false, InstPt, BB);
while (Dep.isDef() || Dep.isClobber()) {
Instruction *Dependency = Dep.getInst();
if (!hasMemoryWrite(Dependency, *TLI) || !isRemovable(Dependency))
break;
Value *DepPointer =
GetUnderlyingObject(getStoredPointerOperand(Dependency), DL);
// Check for aliasing.
if (!AA->isMustAlias(F->getArgOperand(0), DepPointer))
break;
Instruction *Next = std::next(BasicBlock::iterator(Dependency));
// DCE instructions only used to calculate that store
DeleteDeadInstruction(Dependency, *MD, *TLI);
++NumFastStores;
MadeChange = true;
// Inst's old Dependency is now deleted. Compute the next dependency,
// which may also be dead, as in
// s[0] = 0;
// s[1] = 0; // This has just been deleted.
// free(s);
Dep = MD->getPointerDependencyFrom(Loc, false, Next, BB);
}
if (Dep.isNonLocal())
FindUnconditionalPreds(Blocks, BB, DT);
}
return MadeChange;
}
bool MemDepPrinter::runOnFunction(Function &F) {
this->F = &F;
MemoryDependenceAnalysis &MDA = getAnalysis<MemoryDependenceAnalysis>();
// All this code uses non-const interfaces because MemDep is not
// const-friendly, though nothing is actually modified.
for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I) {
Instruction *Inst = &*I;
if (!Inst->mayReadFromMemory() && !Inst->mayWriteToMemory())
continue;
MemDepResult Res = MDA.getDependency(Inst);
if (!Res.isNonLocal()) {
Deps[Inst].insert(std::make_pair(getInstTypePair(Res),
static_cast<BasicBlock *>(nullptr)));
} else if (CallSite CS = cast<Value>(Inst)) {
const MemoryDependenceAnalysis::NonLocalDepInfo &NLDI =
MDA.getNonLocalCallDependency(CS);
DepSet &InstDeps = Deps[Inst];
for (MemoryDependenceAnalysis::NonLocalDepInfo::const_iterator
I = NLDI.begin(), E = NLDI.end(); I != E; ++I) {
const MemDepResult &Res = I->getResult();
InstDeps.insert(std::make_pair(getInstTypePair(Res), I->getBB()));
}
} else {
SmallVector<NonLocalDepResult, 4> NLDI;
assert( (isa<LoadInst>(Inst) || isa<StoreInst>(Inst) ||
isa<VAArgInst>(Inst)) && "Unknown memory instruction!");
MDA.getNonLocalPointerDependency(Inst, NLDI);
DepSet &InstDeps = Deps[Inst];
for (SmallVectorImpl<NonLocalDepResult>::const_iterator
I = NLDI.begin(), E = NLDI.end(); I != E; ++I) {
const MemDepResult &Res = I->getResult();
InstDeps.insert(std::make_pair(getInstTypePair(Res), I->getBB()));
}
}
}
return false;
}
bool MemDepPrinter::runOnFunction(Function &F) {
this->F = &F;
MemoryDependenceResults &MDA = getAnalysis<MemoryDependenceWrapperPass>().getMemDep();
// All this code uses non-const interfaces because MemDep is not
// const-friendly, though nothing is actually modified.
for (auto &I : instructions(F)) {
Instruction *Inst = &I;
if (!Inst->mayReadFromMemory() && !Inst->mayWriteToMemory())
continue;
MemDepResult Res = MDA.getDependency(Inst);
if (!Res.isNonLocal()) {
Deps[Inst].insert(std::make_pair(getInstTypePair(Res),
static_cast<BasicBlock *>(nullptr)));
} else if (auto CS = CallSite(Inst)) {
const MemoryDependenceResults::NonLocalDepInfo &NLDI =
MDA.getNonLocalCallDependency(CS);
DepSet &InstDeps = Deps[Inst];
for (const NonLocalDepEntry &I : NLDI) {
const MemDepResult &Res = I.getResult();
InstDeps.insert(std::make_pair(getInstTypePair(Res), I.getBB()));
}
} else {
SmallVector<NonLocalDepResult, 4> NLDI;
assert( (isa<LoadInst>(Inst) || isa<StoreInst>(Inst) ||
isa<VAArgInst>(Inst)) && "Unknown memory instruction!");
MDA.getNonLocalPointerDependency(Inst, NLDI);
DepSet &InstDeps = Deps[Inst];
for (const NonLocalDepResult &I : NLDI) {
const MemDepResult &Res = I.getResult();
InstDeps.insert(std::make_pair(getInstTypePair(Res), I.getBB()));
}
}
}
return false;
}
/// getNonLocalPointerDepFromBB - Perform a dependency query based on
/// pointer/pointeesize starting at the end of StartBB. Add any clobber/def
/// results to the results vector and keep track of which blocks are visited in
/// 'Visited'.
///
/// This has special behavior for the first block queries (when SkipFirstBlock
/// is true). In this special case, it ignores the contents of the specified
/// block and starts returning dependence info for its predecessors.
///
/// 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(Value *Pointer, uint64_t PointeeSize,
bool isLoad, BasicBlock *StartBB,
SmallVectorImpl<NonLocalDepEntry> &Result,
DenseMap<BasicBlock*, Value*> &Visited,
bool SkipFirstBlock) {
// Look up the cached info for Pointer.
ValueIsLoadPair CacheKey(Pointer, isLoad);
std::pair<BBSkipFirstBlockPair, NonLocalDepInfo> *CacheInfo =
&NonLocalPointerDeps[CacheKey];
NonLocalDepInfo *Cache = &CacheInfo->second;
// If we have valid cached information for exactly the block we are
// investigating, just return it with no recomputation.
if (CacheInfo->first == BBSkipFirstBlockPair(StartBB, SkipFirstBlock)) {
// We have a fully cached result for this query then we can just return the
// cached results and populate the visited set. However, we have to verify
// that we don't already have conflicting results for these blocks. Check
// to ensure that if a block in the results set is in the visited set that
// it was for the same pointer query.
if (!Visited.empty()) {
for (NonLocalDepInfo::iterator I = Cache->begin(), E = Cache->end();
I != E; ++I) {
DenseMap<BasicBlock*, Value*>::iterator VI = Visited.find(I->first);
if (VI == Visited.end() || VI->second == Pointer) 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;
}
}
for (NonLocalDepInfo::iterator I = Cache->begin(), E = Cache->end();
I != E; ++I) {
Visited.insert(std::make_pair(I->first, Pointer));
if (!I->second.isNonLocal())
Result.push_back(*I);
}
++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,
// otherwise it isn't.
if (Cache->empty())
CacheInfo->first = BBSkipFirstBlockPair(StartBB, SkipFirstBlock);
else
CacheInfo->first = BBSkipFirstBlockPair();
SmallVector<BasicBlock*, 32> Worklist;
Worklist.push_back(StartBB);
// Keep track of the entries that we know are sorted. Previously cached
// entries will all be sorted. The entries we add we only sort on demand (we
// don't insert every element into its sorted position). We know that we
// won't get any reuse from currently inserted values, because we don't
// revisit blocks after we insert info for them.
unsigned NumSortedEntries = Cache->size();
DEBUG(AssertSorted(*Cache));
while (!Worklist.empty()) {
BasicBlock *BB = Worklist.pop_back_val();
// Skip the first block if we have it.
if (!SkipFirstBlock) {
// Analyze the dependency of *Pointer in FromBB. See if we already have
// been here.
assert(Visited.count(BB) && "Should check 'visited' before adding to WL");
// 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(Pointer, PointeeSize, isLoad,
BB, Cache, NumSortedEntries);
// If we got a Def or Clobber, add this to the list of results.
if (!Dep.isNonLocal()) {
Result.push_back(NonLocalDepEntry(BB, Dep));
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 phi translation fails, then we can't continue dependence analysis.
Instruction *PtrInst = dyn_cast<Instruction>(Pointer);
bool NeedsPHITranslation = PtrInst && PtrInst->getParent() == BB;
// If no PHI translation is needed, just add all the predecessors of this
// block to scan them as well.
if (!NeedsPHITranslation) {
SkipFirstBlock = false;
for (BasicBlock **PI = PredCache->GetPreds(BB); *PI; ++PI) {
// 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));
if (InsertRes.second) {
// First time we've looked at *PI.
Worklist.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.
if (InsertRes.first->second != Pointer)
goto PredTranslationFailure;
}
continue;
}
// If we do need to do phi translation, then there are a bunch of different
// cases, because we have to find a Value* live in the predecessor block. We
// know that PtrInst is defined in this block at least.
// If this is directly a PHI node, just use the incoming values for each
// pred as the phi translated version.
if (PHINode *PtrPHI = dyn_cast<PHINode>(PtrInst)) {
for (BasicBlock **PI = PredCache->GetPreds(BB); *PI; ++PI) {
BasicBlock *Pred = *PI;
Value *PredPtr = PtrPHI->getIncomingValueForBlock(Pred);
// 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
// the successor translates to a pointer value different than the
// pointer the block was first analyzed with.
std::pair<DenseMap<BasicBlock*,Value*>::iterator, bool>
InsertRes = Visited.insert(std::make_pair(Pred, PredPtr));
if (!InsertRes.second) {
// If the predecessor was visited with PredPtr, then we already did
// the analysis and can ignore it.
if (InsertRes.first->second == PredPtr)
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.
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 getNonLocalPointerDepFromBB that could reuse the cache
// value will only see properly sorted cache arrays.
if (Cache && NumSortedEntries != Cache->size())
std::sort(Cache->begin(), Cache->end());
Cache = 0;
// FIXME: it is entirely possible that PHI translating will end up with
// the same value. Consider PHI translating something like:
// X = phi [x, bb1], [y, bb2]. PHI translating for bb1 doesn't *need*
// to recurse here, pedantically speaking.
// If we have a problem phi translating, fall through to the code below
// to handle the failure condition.
if (getNonLocalPointerDepFromBB(PredPtr, PointeeSize, isLoad, Pred,
Result, Visited))
goto PredTranslationFailure;
}
// Refresh the CacheInfo/Cache pointer so that it isn't invalidated.
CacheInfo = &NonLocalPointerDeps[CacheKey];
Cache = &CacheInfo->second;
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
// results from the set" Clear out the indicator for this.
CacheInfo->first = BBSkipFirstBlockPair();
SkipFirstBlock = false;
continue;
}
// TODO: BITCAST, GEP.
// cerr << "MEMDEP: Could not PHI translate: " << *Pointer;
// if (isa<BitCastInst>(PtrInst) || isa<GetElementPtrInst>(PtrInst))
// cerr << "OP:\t\t\t\t" << *PtrInst->getOperand(0);
PredTranslationFailure:
if (Cache == 0) {
// Refresh the CacheInfo/Cache pointer if it got invalidated.
CacheInfo = &NonLocalPointerDeps[CacheKey];
Cache = &CacheInfo->second;
NumSortedEntries = Cache->size();
} else if (NumSortedEntries != Cache->size()) {
std::sort(Cache->begin(), Cache->end());
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
// results from the set" Clear out the indicator for this.
CacheInfo->first = BBSkipFirstBlockPair();
// If *nothing* works, mark the pointer as being clobbered by the first
// instruction in this block.
//
// If this is the magic first block, return this as a clobber of the whole
// incoming value. Since we can't phi translate to one of the predecessors,
// 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->first != BB)
continue;
assert(I->second.isNonLocal() &&
"Should only be here with transparent block");
I->second = MemDepResult::getClobber(BB->begin());
ReverseNonLocalPtrDeps[BB->begin()].insert(CacheKey.getOpaqueValue());
Result.push_back(*I);
break;
}
}
// Okay, we're done now. If we added new values to the cache, re-sort it.
switch (Cache->size()-NumSortedEntries) {
case 0:
// done, no new entries.
break;
case 2: {
// Two new entries, insert the last one into place.
NonLocalDepEntry Val = Cache->back();
Cache->pop_back();
NonLocalDepInfo::iterator Entry =
std::upper_bound(Cache->begin(), Cache->end()-1, Val);
Cache->insert(Entry, Val);
// FALL THROUGH.
}
case 1:
// One new entry, Just insert the new value at the appropriate position.
if (Cache->size() != 1) {
NonLocalDepEntry Val = Cache->back();
Cache->pop_back();
NonLocalDepInfo::iterator Entry =
std::upper_bound(Cache->begin(), Cache->end(), Val);
Cache->insert(Entry, Val);
}
break;
default:
// Added many values, do a full scale sort.
std::sort(Cache->begin(), Cache->end());
}
DEBUG(AssertSorted(*Cache));
return false;
}
/// GetNonLocalInfoForBlock - Compute the memdep value for BB with
/// 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(Value *Pointer, uint64_t PointeeSize,
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 =
std::upper_bound(Cache->begin(), Cache->begin()+NumSortedEntries,
std::make_pair(BB, MemDepResult()));
if (Entry != Cache->begin() && prior(Entry)->first == BB)
--Entry;
MemDepResult *ExistingResult = 0;
if (Entry != Cache->begin()+NumSortedEntries && Entry->first == BB)
ExistingResult = &Entry->second;
// If we have a cached entry, and it is non-dirty, use it as the value for
// this dependency.
if (ExistingResult && !ExistingResult->isDirty()) {
++NumCacheNonLocalPtr;
return *ExistingResult;
}
// 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.
BasicBlock::iterator ScanPos = BB->end();
if (ExistingResult && ExistingResult->getInst()) {
assert(ExistingResult->getInst()->getParent() == BB &&
"Instruction invalidated?");
++NumCacheDirtyNonLocalPtr;
ScanPos = ExistingResult->getInst();
// Eliminating the dirty entry from 'Cache', so update the reverse info.
ValueIsLoadPair CacheKey(Pointer, isLoad);
RemoveFromReverseMap(ReverseNonLocalPtrDeps, ScanPos,
CacheKey.getOpaqueValue());
} else {
++NumUncacheNonLocalPtr;
}
// Scan the block for the dependency.
MemDepResult Dep = getPointerDependencyFrom(Pointer, PointeeSize, isLoad,
ScanPos, BB);
// If we had a dirty entry for the block, update it. Otherwise, just add
// a new entry.
if (ExistingResult)
*ExistingResult = Dep;
else
Cache->push_back(std::make_pair(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.isNonLocal())
return Dep;
// Keep the ReverseNonLocalPtrDeps map up to date so we can efficiently
// update MemDep when we remove instructions.
Instruction *Inst = Dep.getInst();
assert(Inst && "Didn't depend on anything?");
ValueIsLoadPair CacheKey(Pointer, isLoad);
ReverseNonLocalPtrDeps[Inst].insert(CacheKey.getOpaqueValue());
return Dep;
}
/// getNonLocalCallDependency - Perform a full dependency query for the
/// specified call, returning the set of blocks that the value is
/// potentially live across. The returned set of results will include a
/// "NonLocal" result for all blocks where the value is live across.
///
/// This method assumes the instruction returns a "NonLocal" dependency
/// within its own block.
///
/// This returns a reference to an internal data structure that may be
/// invalidated on the next non-local query or when an instruction is
/// removed. Clients must copy this data if they want it around longer than
/// that.
const MemoryDependenceAnalysis::NonLocalDepInfo &
MemoryDependenceAnalysis::getNonLocalCallDependency(CallSite QueryCS) {
assert(getDependency(QueryCS.getInstruction()).isNonLocal() &&
"getNonLocalCallDependency should only be used on calls with non-local deps!");
PerInstNLInfo &CacheP = NonLocalDeps[QueryCS.getInstruction()];
NonLocalDepInfo &Cache = CacheP.first;
/// DirtyBlocks - This is the set of blocks that need to be recomputed. In
/// the cached case, this can happen due to instructions being deleted etc. In
/// the uncached case, this starts out as the set of predecessors we care
/// about.
SmallVector<BasicBlock*, 32> DirtyBlocks;
if (!Cache.empty()) {
// Okay, we have a cache entry. If we know it is not dirty, just return it
// with no computation.
if (!CacheP.second) {
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->second.isDirty())
DirtyBlocks.push_back(I->first);
// 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;
} 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);
NumUncacheNonLocal++;
}
// isReadonlyCall - If this is a read-only call, we can be more aggressive.
bool isReadonlyCall = AA->onlyReadsMemory(QueryCS);
SmallPtrSet<BasicBlock*, 64> 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))
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 =
std::upper_bound(Cache.begin(), Cache.begin()+NumSortedEntries,
std::make_pair(DirtyBB, MemDepResult()));
if (Entry != Cache.begin() && prior(Entry)->first == DirtyBB)
--Entry;
MemDepResult *ExistingResult = 0;
if (Entry != Cache.begin()+NumSortedEntries &&
Entry->first == DirtyBB) {
// If we already have an entry, and if it isn't already dirty, the block
// is done.
if (!Entry->second.isDirty())
continue;
// Otherwise, remember this slot so we can update the value.
ExistingResult = &Entry->second;
}
// 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();
if (ExistingResult) {
if (Instruction *Inst = ExistingResult->getInst()) {
ScanPos = Inst;
// We're removing QueryInst's use of Inst.
RemoveFromReverseMap(ReverseNonLocalDeps, Inst,
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()) {
// No dependence found. If this is the entry block of the function, it is
// a clobber, otherwise it is non-local.
Dep = MemDepResult::getNonLocal();
} else {
Dep = MemDepResult::getClobber(ScanPos);
}
// If we had a dirty entry for the block, update it. Otherwise, just add
// a new entry.
if (ExistingResult)
*ExistingResult = Dep;
else
Cache.push_back(std::make_pair(DirtyBB, Dep));
// If the block has a dependency (i.e. it isn't completely transparent to
// the value), remember the association!
if (!Dep.isNonLocal()) {
// Keep the ReverseNonLocalDeps map up to date so we can efficiently
// update this when we remove instructions.
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;
}
void DataDependence::processDepResult(Instruction *inst,
MemoryDependenceAnalysis &MDA, AliasAnalysis &AA) {
// TODO: This is probably a good place to check of the dependency
// information is calculated on-demand
MemDepResult Res = MDA.getDependency(inst);
if (!Res.isNonLocal()) {
// local results (not-non-local) can be simply handled. They are just
// a pair of insturctions and a dependency type
// Get dependency information
DepInfo newInfo;
newInfo = getDepInfo(Res);
#ifdef MK_DEBUG
//errs() << "[DEBUG] newInfo depInst == " << Res.getInst() << '\n';
if (Res.getInst() == NULL) {
errs() << "[DEBUG] NULL dependency found, dep type: "
<< depTypeToString(newInfo.Type_) << '\n';
}
#endif
// Save into map
assert(newInfo.valid());
LocalDeps_[inst] = newInfo;
}
else {
// Handle NonLocal dependencies. The function call
// getNonLocalPointerDependency() assumes that a result of NonLocal
// has already been encountered
// Get dependency information
DepInfo newInfo;
newInfo = getDepInfo(Res);
assert(newInfo.Type_ == NonLocal);
assert(Res.isNonLocal());
SmallVector<NonLocalDepResult, 4> NLDep;
if (LoadInst *LI = dyn_cast<LoadInst>(inst)) {
if (!LI->isUnordered()) {
// FIXME: Handle atomic/volatile loads.
errs() << "[WARNING] atomic/volatile loads are not handled\n";
assert(false && "atomic/volatile loads not handled");
//Deps[Inst].insert(std::make_pair(getInstTypePair(0, Unknown),
//static_cast<BasicBlock *>(0)));
return;
}
AliasAnalysis::Location Loc = AA.getLocation(LI);
MDA.getNonLocalPointerDependency(Loc, true, LI->getParent(), NLDep);
}
else if (StoreInst *SI = dyn_cast<StoreInst>(inst)) {
if (!SI->isUnordered()) {
// FIXME: Handle atomic/volatile stores.
errs() << "[WARNING] atomic/volatile stores are not handled\n";
assert(false && "atomic/volatile stores not handled");
//Deps[Inst].insert(std::make_pair(getInstTypePair(0, Unknown),
//static_cast<BasicBlock *>(0)));
return;
}
AliasAnalysis::Location Loc = AA.getLocation(SI);
MDA.getNonLocalPointerDependency(Loc, false, SI->getParent(),
NLDep);
}
else if (VAArgInst *VI = dyn_cast<VAArgInst>(inst)) {
AliasAnalysis::Location Loc = AA.getLocation(VI);
MDA.getNonLocalPointerDependency(Loc, false, VI->getParent(),
NLDep);
}
else {
llvm_unreachable("Unknown memory instruction!");
}
#ifdef MK_DEBUG
errs() << "[DEBUG] NLDep.size() == " << NLDep.size() << '\n';
#endif
for (auto I = NLDep.begin(), E = NLDep.end(); I != E; ++I) {
NonLocalDeps_[inst].push_back(*I);
}
} // end else
}
bool DSE::runOnBasicBlock(BasicBlock &BB) {
bool MadeChange = false;
// Do a top-down walk on the BB.
for (BasicBlock::iterator BBI = BB.begin(), BBE = BB.end(); BBI != BBE; ) {
Instruction *Inst = BBI++;
// Handle 'free' calls specially.
if (CallInst *F = isFreeCall(Inst)) {
MadeChange |= HandleFree(F);
continue;
}
// If we find something that writes memory, get its memory dependence.
if (!hasMemoryWrite(Inst))
continue;
MemDepResult InstDep = MD->getDependency(Inst);
// Ignore non-local store liveness.
// FIXME: cross-block DSE would be fun. :)
if (InstDep.isNonLocal() ||
// Ignore self dependence, which happens in the entry block of the
// function.
InstDep.getInst() == Inst)
continue;
// If we're storing the same value back to a pointer that we just
// loaded from, then the store can be removed.
if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
if (LoadInst *DepLoad = dyn_cast<LoadInst>(InstDep.getInst())) {
if (SI->getPointerOperand() == DepLoad->getPointerOperand() &&
SI->getOperand(0) == DepLoad && !SI->isVolatile()) {
DEBUG(dbgs() << "DSE: Remove Store Of Load from same pointer:\n "
<< "LOAD: " << *DepLoad << "\n STORE: " << *SI << '\n');
// DeleteDeadInstruction can delete the current instruction. Save BBI
// in case we need it.
WeakVH NextInst(BBI);
DeleteDeadInstruction(SI, *MD);
if (NextInst == 0) // Next instruction deleted.
BBI = BB.begin();
else if (BBI != BB.begin()) // Revisit this instruction if possible.
--BBI;
++NumFastStores;
MadeChange = true;
continue;
}
}
}
// Figure out what location is being stored to.
AliasAnalysis::Location Loc = getLocForWrite(Inst, *AA);
// If we didn't get a useful location, fail.
if (Loc.Ptr == 0)
continue;
while (!InstDep.isNonLocal()) {
// Get the memory clobbered by the instruction we depend on. MemDep will
// skip any instructions that 'Loc' clearly doesn't interact with. If we
// end up depending on a may- or must-aliased load, then we can't optimize
// away the store and we bail out. However, if we depend on on something
// that overwrites the memory location we *can* potentially optimize it.
//
// Find out what memory location the dependant instruction stores.
Instruction *DepWrite = InstDep.getInst();
AliasAnalysis::Location DepLoc = getLocForWrite(DepWrite, *AA);
// If we didn't get a useful location, or if it isn't a size, bail out.
if (DepLoc.Ptr == 0)
break;
// If we find a write that is a) removable (i.e., non-volatile), b) is
// completely obliterated by the store to 'Loc', and c) which we know that
// 'Inst' doesn't load from, then we can remove it.
if (isRemovable(DepWrite) && isCompleteOverwrite(Loc, DepLoc, *AA) &&
!isPossibleSelfRead(Inst, Loc, DepWrite, *AA)) {
DEBUG(dbgs() << "DSE: Remove Dead Store:\n DEAD: "
<< *DepWrite << "\n KILLER: " << *Inst << '\n');
// Delete the store and now-dead instructions that feed it.
DeleteDeadInstruction(DepWrite, *MD);
++NumFastStores;
MadeChange = true;
// DeleteDeadInstruction can delete the current instruction in loop
// cases, reset BBI.
BBI = Inst;
if (BBI != BB.begin())
--BBI;
break;
}
// If this is a may-aliased store that is clobbering the store value, we
// can keep searching past it for another must-aliased pointer that stores
// to the same location. For example, in:
// store -> P
// store -> Q
// store -> P
// we can remove the first store to P even though we don't know if P and Q
// alias.
if (DepWrite == &BB.front()) break;
// Can't look past this instruction if it might read 'Loc'.
if (AA->getModRefInfo(DepWrite, Loc) & AliasAnalysis::Ref)
break;
InstDep = MD->getPointerDependencyFrom(Loc, false, DepWrite, &BB);
}
}
// If this block ends in a return, unwind, or unreachable, all allocas are
// dead at its end, which means stores to them are also dead.
if (BB.getTerminator()->getNumSuccessors() == 0)
MadeChange |= handleEndBlock(BB);
return MadeChange;
}
bool MemDepPrinter::runOnFunction(Function &F) {
this->F = &F;
AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
MemoryDependenceAnalysis &MDA = getAnalysis<MemoryDependenceAnalysis>();
// All this code uses non-const interfaces because MemDep is not
// const-friendly, though nothing is actually modified.
for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I) {
Instruction *Inst = &*I;
if (!Inst->mayReadFromMemory() && !Inst->mayWriteToMemory())
continue;
MemDepResult Res = MDA.getDependency(Inst);
if (!Res.isNonLocal()) {
Deps[Inst].insert(std::make_pair(getInstTypePair(Res),
static_cast<BasicBlock *>(nullptr)));
} else if (CallSite CS = cast<Value>(Inst)) {
const MemoryDependenceAnalysis::NonLocalDepInfo &NLDI =
MDA.getNonLocalCallDependency(CS);
DepSet &InstDeps = Deps[Inst];
for (MemoryDependenceAnalysis::NonLocalDepInfo::const_iterator
I = NLDI.begin(), E = NLDI.end(); I != E; ++I) {
const MemDepResult &Res = I->getResult();
InstDeps.insert(std::make_pair(getInstTypePair(Res), I->getBB()));
}
} else {
SmallVector<NonLocalDepResult, 4> NLDI;
if (LoadInst *LI = dyn_cast<LoadInst>(Inst)) {
if (!LI->isUnordered()) {
// FIXME: Handle atomic/volatile loads.
Deps[Inst].insert(std::make_pair(getInstTypePair(nullptr, Unknown),
static_cast<BasicBlock *>(nullptr)));
continue;
}
AliasAnalysis::Location Loc = AA.getLocation(LI);
MDA.getNonLocalPointerDependency(Loc, true, LI->getParent(), NLDI);
} else if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
if (!SI->isUnordered()) {
// FIXME: Handle atomic/volatile stores.
Deps[Inst].insert(std::make_pair(getInstTypePair(nullptr, Unknown),
static_cast<BasicBlock *>(nullptr)));
continue;
}
AliasAnalysis::Location Loc = AA.getLocation(SI);
MDA.getNonLocalPointerDependency(Loc, false, SI->getParent(), NLDI);
} else if (VAArgInst *VI = dyn_cast<VAArgInst>(Inst)) {
AliasAnalysis::Location Loc = AA.getLocation(VI);
MDA.getNonLocalPointerDependency(Loc, false, VI->getParent(), NLDI);
} else {
llvm_unreachable("Unknown memory instruction!");
}
DepSet &InstDeps = Deps[Inst];
for (SmallVectorImpl<NonLocalDepResult>::const_iterator
I = NLDI.begin(), E = NLDI.end(); I != E; ++I) {
const MemDepResult &Res = I->getResult();
InstDeps.insert(std::make_pair(getInstTypePair(Res), I->getBB()));
}
}
}
return false;
}
bool MemCpyOpt::processStore(StoreInst *SI, BasicBlock::iterator &BBI) {
if (SI->isVolatile()) return false;
if (TD == 0) return false;
// Detect cases where we're performing call slot forwarding, but
// happen to be using a load-store pair to implement it, rather than
// a memcpy.
if (LoadInst *LI = dyn_cast<LoadInst>(SI->getOperand(0))) {
if (!LI->isVolatile() && LI->hasOneUse()) {
MemDepResult ldep = MD->getDependency(LI);
CallInst *C = 0;
if (ldep.isClobber() && !isa<MemCpyInst>(ldep.getInst()))
C = dyn_cast<CallInst>(ldep.getInst());
if (C) {
// Check that nothing touches the dest of the "copy" between
// the call and the store.
MemDepResult sdep = MD->getDependency(SI);
if (!sdep.isNonLocal()) {
bool FoundCall = false;
for (BasicBlock::iterator I = SI, E = sdep.getInst(); I != E; --I) {
if (&*I == C) {
FoundCall = true;
break;
}
}
if (!FoundCall)
C = 0;
}
}
if (C) {
bool changed = performCallSlotOptzn(LI,
SI->getPointerOperand()->stripPointerCasts(),
LI->getPointerOperand()->stripPointerCasts(),
TD->getTypeStoreSize(SI->getOperand(0)->getType()), C);
if (changed) {
MD->removeInstruction(SI);
SI->eraseFromParent();
MD->removeInstruction(LI);
LI->eraseFromParent();
++NumMemCpyInstr;
return true;
}
}
}
}
// There are two cases that are interesting for this code to handle: memcpy
// and memset. Right now we only handle memset.
// Ensure that the value being stored is something that can be memset'able a
// byte at a time like "0" or "-1" or any width, as well as things like
// 0xA0A0A0A0 and 0.0.
if (Value *ByteVal = isBytewiseValue(SI->getOperand(0)))
if (Instruction *I = tryMergingIntoMemset(SI, SI->getPointerOperand(),
ByteVal)) {
BBI = I; // Don't invalidate iterator.
return true;
}
return false;
}