/// isObjectSmallerThan - Return true if we can prove that the object specified
/// by V is smaller than Size.
static bool isObjectSmallerThan(const Value *V, unsigned Size,
const TargetData &TD) {
const Type *AccessTy;
if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(V)) {
AccessTy = GV->getType()->getElementType();
} else if (const AllocaInst *AI = dyn_cast<AllocaInst>(V)) {
if (!AI->isArrayAllocation())
AccessTy = AI->getType()->getElementType();
else
return false;
} else if (const CallInst* CI = extractMallocCall(V)) {
if (!isArrayMalloc(V, &TD))
// The size is the argument to the malloc call.
if (const ConstantInt* C = dyn_cast<ConstantInt>(CI->getOperand(1)))
return (C->getZExtValue() < Size);
return false;
} else if (const Argument *A = dyn_cast<Argument>(V)) {
if (A->hasByValAttr())
AccessTy = cast<PointerType>(A->getType())->getElementType();
else
return false;
} else {
return false;
}
if (AccessTy->isSized())
return TD.getTypeAllocSize(AccessTy) < Size;
return false;
}
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);
for (unsigned i = 1; i != Idx; ++i, ++GTI)
/*skip along*/;
// Compute the offset implied by the rest of the indices.
int64_t Offset = 0;
for (unsigned i = Idx, e = GEP->getNumOperands(); i != e; ++i, ++GTI) {
ConstantInt *OpC = dyn_cast<ConstantInt>(GEP->getOperand(i));
if (OpC == 0)
return VariableIdxFound = true;
if (OpC->isZero()) continue; // No offset.
// Handle struct indices, which add their field offset to the pointer.
if (StructType *STy = dyn_cast<StructType>(*GTI)) {
Offset += TD.getStructLayout(STy)->getElementOffset(OpC->getZExtValue());
continue;
}
// Otherwise, we have a sequential type like an array or vector. Multiply
// the index by the ElementSize.
uint64_t Size = TD.getTypeAllocSize(GTI.getIndexedType());
Offset += Size*OpC->getSExtValue();
}
return Offset;
}
StructLayout::StructLayout(const StructType *ST, const TargetData &TD) {
assert(!ST->isOpaque() && "Cannot get layout of opaque structs");
StructAlignment = 0;
StructSize = 0;
NumElements = ST->getNumElements();
// Loop over each of the elements, placing them in memory.
for (unsigned i = 0, e = NumElements; i != e; ++i) {
const Type *Ty = ST->getElementType(i);
unsigned TyAlign = ST->isPacked() ? 1 : TD.getABITypeAlignment(Ty);
// Add padding if necessary to align the data element properly.
if ((StructSize & (TyAlign-1)) != 0)
StructSize = TargetData::RoundUpAlignment(StructSize, TyAlign);
// Keep track of maximum alignment constraint.
StructAlignment = std::max(TyAlign, StructAlignment);
MemberOffsets[i] = StructSize;
StructSize += TD.getTypeAllocSize(Ty); // Consume space for this data item
}
// Empty structures have alignment of 1 byte.
if (StructAlignment == 0) StructAlignment = 1;
// Add padding to the end of the struct so that it could be put in an array
// and all array elements would be aligned correctly.
if ((StructSize & (StructAlignment-1)) != 0)
StructSize = TargetData::RoundUpAlignment(StructSize, StructAlignment);
}
static unsigned getTypeSize(TargetData &TD, Type *type) {
if (type->isFunctionTy()) /* it is not sized, weird */
return TD.getPointerSize();
if (!type->isSized())
return 100; /* FIXME */
if (StructType *ST = dyn_cast<StructType>(type))
return TD.getStructLayout(ST)->getSizeInBytes();
return TD.getTypeAllocSize(type);
}
/// performCallSlotOptzn - takes a memcpy and a call that it depends on,
/// and checks for the possibility of a call slot optimization by having
/// the call write its result directly into the destination of the memcpy.
bool MemCpyOpt::performCallSlotOptzn(MemCpyInst *cpy, CallInst *C) {
// The general transformation to keep in mind is
//
// call @func(..., src, ...)
// memcpy(dest, src, ...)
//
// ->
//
// memcpy(dest, src, ...)
// call @func(..., dest, ...)
//
// Since moving the memcpy is technically awkward, we additionally check that
// src only holds uninitialized values at the moment of the call, meaning that
// the memcpy can be discarded rather than moved.
// Deliberately get the source and destination with bitcasts stripped away,
// because we'll need to do type comparisons based on the underlying type.
Value *cpyDest = cpy->getDest();
Value *cpySrc = cpy->getSource();
CallSite CS = CallSite::get(C);
// We need to be able to reason about the size of the memcpy, so we require
// that it be a constant.
ConstantInt *cpyLength = dyn_cast<ConstantInt>(cpy->getLength());
if (!cpyLength)
return false;
// Require that src be an alloca. This simplifies the reasoning considerably.
AllocaInst *srcAlloca = dyn_cast<AllocaInst>(cpySrc);
if (!srcAlloca)
return false;
// Check that all of src is copied to dest.
TargetData *TD = getAnalysisIfAvailable<TargetData>();
if (!TD) return false;
ConstantInt *srcArraySize = dyn_cast<ConstantInt>(srcAlloca->getArraySize());
if (!srcArraySize)
return false;
uint64_t srcSize = TD->getTypeAllocSize(srcAlloca->getAllocatedType()) *
srcArraySize->getZExtValue();
if (cpyLength->getZExtValue() < srcSize)
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.
if (AllocaInst *A = dyn_cast<AllocaInst>(cpyDest)) {
// The destination is an alloca. Check it is larger than srcSize.
ConstantInt *destArraySize = dyn_cast<ConstantInt>(A->getArraySize());
if (!destArraySize)
return false;
uint64_t destSize = TD->getTypeAllocSize(A->getAllocatedType()) *
destArraySize->getZExtValue();
if (destSize < srcSize)
return false;
} else if (Argument *A = dyn_cast<Argument>(cpyDest)) {
// If the destination is an sret parameter then only accesses that are
// outside of the returned struct type can trap.
if (!A->hasStructRetAttr())
return false;
const Type *StructTy = cast<PointerType>(A->getType())->getElementType();
uint64_t destSize = TD->getTypeAllocSize(StructTy);
if (destSize < srcSize)
return false;
} else {
return false;
}
// Check that src is not accessed except via the call and the memcpy. This
// guarantees that it holds only undefined values when passed in (so the final
// memcpy can be dropped), that it is not read or written between the call and
// the memcpy, and that writing beyond the end of it is undefined.
SmallVector<User*, 8> srcUseList(srcAlloca->use_begin(),
srcAlloca->use_end());
while (!srcUseList.empty()) {
User *UI = srcUseList.pop_back_val();
if (isa<BitCastInst>(UI)) {
for (User::use_iterator I = UI->use_begin(), E = UI->use_end();
I != E; ++I)
srcUseList.push_back(*I);
} else if (GetElementPtrInst *G = dyn_cast<GetElementPtrInst>(UI)) {
if (G->hasAllZeroIndices())
for (User::use_iterator I = UI->use_begin(), E = UI->use_end();
I != E; ++I)
srcUseList.push_back(*I);
else
return false;
} else if (UI != C && UI != cpy) {
return false;
}
}
// Since we're changing the parameter to the callsite, we need to make sure
// that what would be the new parameter dominates the callsite.
DominatorTree &DT = getAnalysis<DominatorTree>();
if (Instruction *cpyDestInst = dyn_cast<Instruction>(cpyDest))
if (!DT.dominates(cpyDestInst, C))
return false;
// In addition to knowing that the call does not access src in some
// unexpected manner, for example via a global, which we deduce from
// the use analysis, we also need to know that it does not sneakily
// access dest. We rely on AA to figure this out for us.
AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
if (AA.getModRefInfo(C, cpy->getRawDest(), srcSize) !=
AliasAnalysis::NoModRef)
return false;
// All the checks have passed, so do the transformation.
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);
changedArgument = true;
if (CS.getArgument(i)->getType() == cpyDest->getType())
CS.setArgument(i, cpyDest);
else
CS.setArgument(i, CastInst::CreatePointerCast(cpyDest,
CS.getArgument(i)->getType(), cpyDest->getName(), C));
}
if (!changedArgument)
return false;
// Drop any cached information about the call, because we may have changed
// its dependence information by changing its parameter.
MemoryDependenceAnalysis &MD = getAnalysis<MemoryDependenceAnalysis>();
MD.removeInstruction(C);
// Remove the memcpy
MD.removeInstruction(cpy);
cpy->eraseFromParent();
++NumMemCpyInstr;
return true;
}
void SVMBlockSizeAccumulator::AddConstant(const TargetData &TD,
const MachineConstantPoolEntry &CPE)
{
AddConstant(TD.getTypeAllocSize(CPE.getType()), CPE.getAlignment());
}
/// processByValArgument - This is called on every byval argument in call sites.
bool MemCpyOpt::processByValArgument(CallSite CS, unsigned ArgNo) {
TargetData *TD = getAnalysisIfAvailable<TargetData>();
if (!TD) return false;
// Find out what feeds this byval argument.
Value *ByValArg = CS.getArgument(ArgNo);
const Type *ByValTy =cast<PointerType>(ByValArg->getType())->getElementType();
uint64_t ByValSize = TD->getTypeAllocSize(ByValTy);
MemDepResult DepInfo =
MD->getPointerDependencyFrom(AliasAnalysis::Location(ByValArg, ByValSize),
true, CS.getInstruction(),
CS.getInstruction()->getParent());
if (!DepInfo.isClobber())
return false;
// If the byval argument isn't fed by a memcpy, ignore it. If it is fed by
// a memcpy, see if we can byval from the source of the memcpy instead of the
// result.
MemCpyInst *MDep = dyn_cast<MemCpyInst>(DepInfo.getInst());
if (MDep == 0 || MDep->isVolatile() ||
ByValArg->stripPointerCasts() != MDep->getDest())
return false;
// The length of the memcpy must be larger or equal to the size of the byval.
ConstantInt *C1 = dyn_cast<ConstantInt>(MDep->getLength());
if (C1 == 0 || C1->getValue().getZExtValue() < ByValSize)
return false;
// Get the alignment of the byval. If it is greater than the memcpy, then we
// can't do the substitution. If the call doesn't specify the alignment, then
// it is some target specific value that we can't know.
unsigned ByValAlign = CS.getParamAlignment(ArgNo+1);
if (ByValAlign == 0 || MDep->getAlignment() < ByValAlign)
return false;
// Verify that the copied-from memory doesn't change in between the memcpy and
// the byval call.
// memcpy(a <- b)
// *b = 42;
// foo(*a)
// It would be invalid to transform the second memcpy into foo(*b).
//
// NOTE: This is conservative, it will stop on any read from the source loc,
// not just the defining memcpy.
MemDepResult SourceDep =
MD->getPointerDependencyFrom(AliasAnalysis::getLocationForSource(MDep),
false, CS.getInstruction(), MDep->getParent());
if (!SourceDep.isClobber() || SourceDep.getInst() != MDep)
return false;
Value *TmpCast = MDep->getSource();
if (MDep->getSource()->getType() != ByValArg->getType())
TmpCast = new BitCastInst(MDep->getSource(), ByValArg->getType(),
"tmpcast", CS.getInstruction());
DEBUG(dbgs() << "MemCpyOpt: Forwarding memcpy to byval:\n"
<< " " << *MDep << "\n"
<< " " << *CS.getInstruction() << "\n");
// Otherwise we're good! Update the byval argument.
CS.setArgument(ArgNo, TmpCast);
++NumMemCpyInstr;
return true;
}
