STATISTIC(NumMoveToCpy, "Number of memmoves converted to memcpy");
STATISTIC(NumCpyToSet, "Number of memcpys converted to memset");
-static int64_t GetOffsetFromIndex(const GetElementPtrInst *GEP, unsigned Idx,
+static int64_t GetOffsetFromIndex(const GEPOperator *GEP, unsigned Idx,
bool &VariableIdxFound, const TargetData &TD){
// Skip over the first indices.
gep_type_iterator GTI = gep_type_begin(GEP);
const TargetData &TD) {
Ptr1 = Ptr1->stripPointerCasts();
Ptr2 = Ptr2->stripPointerCasts();
- GetElementPtrInst *GEP1 = dyn_cast<GetElementPtrInst>(Ptr1);
- GetElementPtrInst *GEP2 = dyn_cast<GetElementPtrInst>(Ptr2);
+ GEPOperator *GEP1 = dyn_cast<GEPOperator>(Ptr1);
+ GEPOperator *GEP2 = dyn_cast<GEPOperator>(Ptr2);
bool VariableIdxFound = false;
bool processMemCpy(MemCpyInst *M);
bool processMemMove(MemMoveInst *M);
bool performCallSlotOptzn(Instruction *cpy, Value *cpyDst, Value *cpySrc,
- uint64_t cpyLen, CallInst *C);
+ uint64_t cpyLen, unsigned cpyAlign, CallInst *C);
bool processMemCpyMemCpyDependence(MemCpyInst *M, MemCpyInst *MDep,
uint64_t MSize);
bool processByValArgument(CallSite CS, unsigned ArgNo);
}
if (C) {
+ unsigned storeAlign = SI->getAlignment();
+ if (!storeAlign)
+ storeAlign = TD->getABITypeAlignment(SI->getOperand(0)->getType());
+ unsigned loadAlign = LI->getAlignment();
+ if (!loadAlign)
+ loadAlign = TD->getABITypeAlignment(LI->getType());
+
bool changed = performCallSlotOptzn(LI,
SI->getPointerOperand()->stripPointerCasts(),
LI->getPointerOperand()->stripPointerCasts(),
- TD->getTypeStoreSize(SI->getOperand(0)->getType()), C);
+ TD->getTypeStoreSize(SI->getOperand(0)->getType()),
+ std::min(storeAlign, loadAlign), C);
if (changed) {
MD->removeInstruction(SI);
SI->eraseFromParent();
/// the call write its result directly into the destination of the memcpy.
bool MemCpyOpt::performCallSlotOptzn(Instruction *cpy,
Value *cpyDest, Value *cpySrc,
- uint64_t cpyLen, CallInst *C) {
+ uint64_t cpyLen, unsigned cpyAlign,
+ CallInst *C) {
// The general transformation to keep in mind is
//
// call @func(..., src, ...)
if (cpyLen < srcSize)
return false;
+ // Check that dest points to memory that is at least as aligned as src.
+ unsigned srcAlign = srcAlloca->getAlignment();
+ if (!srcAlign)
+ srcAlign = TD->getABITypeAlignment(srcAlloca->getAllocatedType());
+ bool isDestSufficientlyAligned = srcAlign <= cpyAlign;
+ // If dest is not aligned enough and we can't increase its alignment then
+ // bail out.
+ if (!isDestSufficientlyAligned && !isa<AllocaInst>(cpyDest))
+ return false;
+
// Check that accessing the first srcSize bytes of dest will not cause a
// trap. Otherwise the transform is invalid since it might cause a trap
// to occur earlier than it otherwise would.
bool changedArgument = false;
for (unsigned i = 0; i < CS.arg_size(); ++i)
if (CS.getArgument(i)->stripPointerCasts() == cpySrc) {
- if (cpySrc->getType() != cpyDest->getType())
- cpyDest = CastInst::CreatePointerCast(cpyDest, cpySrc->getType(),
- cpyDest->getName(), C);
+ Value *Dest = cpySrc->getType() == cpyDest->getType() ? cpyDest
+ : CastInst::CreatePointerCast(cpyDest, cpySrc->getType(),
+ cpyDest->getName(), C);
changedArgument = true;
- if (CS.getArgument(i)->getType() == cpyDest->getType())
- CS.setArgument(i, cpyDest);
+ if (CS.getArgument(i)->getType() == Dest->getType())
+ CS.setArgument(i, Dest);
else
- CS.setArgument(i, CastInst::CreatePointerCast(cpyDest,
- CS.getArgument(i)->getType(), cpyDest->getName(), C));
+ CS.setArgument(i, CastInst::CreatePointerCast(Dest,
+ CS.getArgument(i)->getType(), Dest->getName(), C));
}
if (!changedArgument)
return false;
+ // If the destination wasn't sufficiently aligned then increase its alignment.
+ if (!isDestSufficientlyAligned) {
+ assert(isa<AllocaInst>(cpyDest) && "Can only increase alloca alignment!");
+ cast<AllocaInst>(cpyDest)->setAlignment(srcAlign);
+ }
+
// Drop any cached information about the call, because we may have changed
// its dependence information by changing its parameter.
MD->removeInstruction(C);
if (DepInfo.isClobber()) {
if (CallInst *C = dyn_cast<CallInst>(DepInfo.getInst())) {
if (performCallSlotOptzn(M, M->getDest(), M->getSource(),
- CopySize->getZExtValue(), C)) {
+ CopySize->getZExtValue(), M->getAlignment(),
+ C)) {
MD->removeInstruction(M);
M->eraseFromParent();
return true;