#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/AssumptionCache.h"
#include "llvm/Analysis/MemoryDependenceAnalysis.h"
+#include "llvm/Analysis/TargetLibraryInfo.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/Dominators.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
-#include "llvm/Analysis/TargetLibraryInfo.h"
#include "llvm/Transforms/Utils/Local.h"
#include <list>
using namespace llvm;
STATISTIC(NumCpyToSet, "Number of memcpys converted to memset");
static int64_t GetOffsetFromIndex(const GEPOperator *GEP, unsigned Idx,
- bool &VariableIdxFound, const DataLayout &TD){
+ bool &VariableIdxFound,
+ const DataLayout &DL) {
// Skip over the first indices.
gep_type_iterator GTI = gep_type_begin(GEP);
for (unsigned i = 1; i != Idx; ++i, ++GTI)
// 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());
+ Offset += DL.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());
+ uint64_t Size = DL.getTypeAllocSize(GTI.getIndexedType());
Offset += Size*OpC->getSExtValue();
}
/// constant offset, and return that constant offset. For example, Ptr1 might
/// be &A[42], and Ptr2 might be &A[40]. In this case offset would be -8.
static bool IsPointerOffset(Value *Ptr1, Value *Ptr2, int64_t &Offset,
- const DataLayout &TD) {
+ const DataLayout &DL) {
Ptr1 = Ptr1->stripPointerCasts();
Ptr2 = Ptr2->stripPointerCasts();
// If one pointer is a GEP and the other isn't, then see if the GEP is a
// constant offset from the base, as in "P" and "gep P, 1".
if (GEP1 && !GEP2 && GEP1->getOperand(0)->stripPointerCasts() == Ptr2) {
- Offset = -GetOffsetFromIndex(GEP1, 1, VariableIdxFound, TD);
+ Offset = -GetOffsetFromIndex(GEP1, 1, VariableIdxFound, DL);
return !VariableIdxFound;
}
if (GEP2 && !GEP1 && GEP2->getOperand(0)->stripPointerCasts() == Ptr1) {
- Offset = GetOffsetFromIndex(GEP2, 1, VariableIdxFound, TD);
+ Offset = GetOffsetFromIndex(GEP2, 1, VariableIdxFound, DL);
return !VariableIdxFound;
}
if (GEP1->getOperand(Idx) != GEP2->getOperand(Idx))
break;
- int64_t Offset1 = GetOffsetFromIndex(GEP1, Idx, VariableIdxFound, TD);
- int64_t Offset2 = GetOffsetFromIndex(GEP2, Idx, VariableIdxFound, TD);
+ int64_t Offset1 = GetOffsetFromIndex(GEP1, Idx, VariableIdxFound, DL);
+ int64_t Offset2 = GetOffsetFromIndex(GEP2, Idx, VariableIdxFound, DL);
if (VariableIdxFound) return false;
Offset = Offset2-Offset1;
/// TheStores - The actual stores that make up this range.
SmallVector<Instruction*, 16> TheStores;
- bool isProfitableToUseMemset(const DataLayout &TD) const;
-
+ bool isProfitableToUseMemset(const DataLayout &DL) const;
};
} // end anon namespace
-bool MemsetRange::isProfitableToUseMemset(const DataLayout &TD) const {
+bool MemsetRange::isProfitableToUseMemset(const DataLayout &DL) const {
// If we found more than 4 stores to merge or 16 bytes, use memset.
if (TheStores.size() >= 4 || End-Start >= 16) return true;
// size. If so, check to see whether we will end up actually reducing the
// number of stores used.
unsigned Bytes = unsigned(End-Start);
- unsigned MaxIntSize = TD.getLargestLegalIntTypeSize();
+ unsigned MaxIntSize = DL.getLargestLegalIntTypeSize();
if (MaxIntSize == 0)
MaxIntSize = 1;
unsigned NumPointerStores = Bytes / MaxIntSize;
class MemCpyOpt : public FunctionPass {
MemoryDependenceAnalysis *MD;
TargetLibraryInfo *TLI;
- const DataLayout *DL;
public:
static char ID; // Pass identification, replacement for typeid
MemCpyOpt() : FunctionPass(ID) {
initializeMemCpyOptPass(*PassRegistry::getPassRegistry());
MD = nullptr;
TLI = nullptr;
- DL = nullptr;
}
bool runOnFunction(Function &F) override;
bool processMemMove(MemMoveInst *M);
bool performCallSlotOptzn(Instruction *cpy, Value *cpyDst, Value *cpySrc,
uint64_t cpyLen, unsigned cpyAlign, CallInst *C);
- bool processMemCpyMemCpyDependence(MemCpyInst *M, MemCpyInst *MDep,
- uint64_t MSize);
+ bool processMemCpyMemCpyDependence(MemCpyInst *M, MemCpyInst *MDep);
+ bool processMemSetMemCpyDependence(MemCpyInst *M, MemSetInst *MDep);
+ bool performMemCpyToMemSetOptzn(MemCpyInst *M, MemSetInst *MDep);
bool processByValArgument(CallSite CS, unsigned ArgNo);
Instruction *tryMergingIntoMemset(Instruction *I, Value *StartPtr,
Value *ByteVal);
/// attempts to merge them together into a memcpy/memset.
Instruction *MemCpyOpt::tryMergingIntoMemset(Instruction *StartInst,
Value *StartPtr, Value *ByteVal) {
- if (!DL) return nullptr;
+ const DataLayout &DL = StartInst->getModule()->getDataLayout();
// Okay, so we now have a single store that can be splatable. Scan to find
// all subsequent stores of the same value to offset from the same pointer.
// Join these together into ranges, so we can decide whether contiguous blocks
// are stored.
- MemsetRanges Ranges(*DL);
+ MemsetRanges Ranges(DL);
BasicBlock::iterator BI = StartInst;
for (++BI; !isa<TerminatorInst>(BI); ++BI) {
// Check to see if this store is to a constant offset from the start ptr.
int64_t Offset;
- if (!IsPointerOffset(StartPtr, NextStore->getPointerOperand(),
- Offset, *DL))
+ if (!IsPointerOffset(StartPtr, NextStore->getPointerOperand(), Offset,
+ DL))
break;
Ranges.addStore(Offset, NextStore);
// Check to see if this store is to a constant offset from the start ptr.
int64_t Offset;
- if (!IsPointerOffset(StartPtr, MSI->getDest(), Offset, *DL))
+ if (!IsPointerOffset(StartPtr, MSI->getDest(), Offset, DL))
break;
Ranges.addMemSet(Offset, MSI);
if (Range.TheStores.size() == 1) continue;
// If it is profitable to lower this range to memset, do so now.
- if (!Range.isProfitableToUseMemset(*DL))
+ if (!Range.isProfitableToUseMemset(DL))
continue;
// Otherwise, we do want to transform this! Create a new memset.
if (Alignment == 0) {
Type *EltType =
cast<PointerType>(StartPtr->getType())->getElementType();
- Alignment = DL->getABITypeAlignment(EltType);
+ Alignment = DL.getABITypeAlignment(EltType);
}
AMemSet =
bool MemCpyOpt::processStore(StoreInst *SI, BasicBlock::iterator &BBI) {
if (!SI->isSimple()) return false;
-
- if (!DL) return false;
+ const DataLayout &DL = SI->getModule()->getDataLayout();
// Detect cases where we're performing call slot forwarding, but
// happen to be using a load-store pair to implement it, rather than
// Check that nothing touches the dest of the "copy" between
// the call and the store.
AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
- AliasAnalysis::Location StoreLoc = AA.getLocation(SI);
+ AliasAnalysis::Location StoreLoc = MemoryLocation::get(SI);
for (BasicBlock::iterator I = --BasicBlock::iterator(SI),
E = C; I != E; --I) {
if (AA.getModRefInfo(&*I, StoreLoc) != AliasAnalysis::NoModRef) {
if (C) {
unsigned storeAlign = SI->getAlignment();
if (!storeAlign)
- storeAlign = DL->getABITypeAlignment(SI->getOperand(0)->getType());
+ storeAlign = DL.getABITypeAlignment(SI->getOperand(0)->getType());
unsigned loadAlign = LI->getAlignment();
if (!loadAlign)
- loadAlign = DL->getABITypeAlignment(LI->getType());
+ loadAlign = DL.getABITypeAlignment(LI->getType());
- bool changed = performCallSlotOptzn(LI,
- SI->getPointerOperand()->stripPointerCasts(),
- LI->getPointerOperand()->stripPointerCasts(),
- DL->getTypeStoreSize(SI->getOperand(0)->getType()),
- std::min(storeAlign, loadAlign), C);
+ bool changed = performCallSlotOptzn(
+ LI, SI->getPointerOperand()->stripPointerCasts(),
+ LI->getPointerOperand()->stripPointerCasts(),
+ DL.getTypeStoreSize(SI->getOperand(0)->getType()),
+ std::min(storeAlign, loadAlign), C);
if (changed) {
MD->removeInstruction(SI);
SI->eraseFromParent();
if (!srcAlloca)
return false;
- // Check that all of src is copied to dest.
- if (!DL) return false;
-
ConstantInt *srcArraySize = dyn_cast<ConstantInt>(srcAlloca->getArraySize());
if (!srcArraySize)
return false;
- uint64_t srcSize = DL->getTypeAllocSize(srcAlloca->getAllocatedType()) *
- srcArraySize->getZExtValue();
+ const DataLayout &DL = cpy->getModule()->getDataLayout();
+ uint64_t srcSize = DL.getTypeAllocSize(srcAlloca->getAllocatedType()) *
+ srcArraySize->getZExtValue();
if (cpyLen < srcSize)
return false;
if (!destArraySize)
return false;
- uint64_t destSize = DL->getTypeAllocSize(A->getAllocatedType()) *
- destArraySize->getZExtValue();
+ uint64_t destSize = DL.getTypeAllocSize(A->getAllocatedType()) *
+ destArraySize->getZExtValue();
if (destSize < srcSize)
return false;
return false;
}
- uint64_t destSize = DL->getTypeAllocSize(StructTy);
+ uint64_t destSize = DL.getTypeAllocSize(StructTy);
if (destSize < srcSize)
return false;
}
// Check that dest points to memory that is at least as aligned as src.
unsigned srcAlign = srcAlloca->getAlignment();
if (!srcAlign)
- srcAlign = DL->getABITypeAlignment(srcAlloca->getAllocatedType());
+ srcAlign = DL.getABITypeAlignment(srcAlloca->getAllocatedType());
bool isDestSufficientlyAligned = srcAlign <= cpyAlign;
// If dest is not aligned enough and we can't increase its alignment then
// bail out.
// its dependence information by changing its parameter.
MD->removeInstruction(C);
+ // Update AA metadata
+ // FIXME: MD_tbaa_struct and MD_mem_parallel_loop_access should also be
+ // handled here, but combineMetadata doesn't support them yet
+ unsigned KnownIDs[] = {
+ LLVMContext::MD_tbaa,
+ LLVMContext::MD_alias_scope,
+ LLVMContext::MD_noalias,
+ };
+ combineMetadata(C, cpy, KnownIDs);
+
// Remove the memcpy.
MD->removeInstruction(cpy);
++NumMemCpyInstr;
/// processMemCpyMemCpyDependence - We've found that the (upward scanning)
/// memory dependence of memcpy 'M' is the memcpy 'MDep'. Try to simplify M to
-/// copy from MDep's input if we can. MSize is the size of M's copy.
+/// copy from MDep's input if we can.
///
-bool MemCpyOpt::processMemCpyMemCpyDependence(MemCpyInst *M, MemCpyInst *MDep,
- uint64_t MSize) {
+bool MemCpyOpt::processMemCpyMemCpyDependence(MemCpyInst *M, MemCpyInst *MDep) {
// We can only transforms memcpy's where the dest of one is the source of the
// other.
if (M->getSource() != MDep->getDest() || MDep->isVolatile())
//
// NOTE: This is conservative, it will stop on any read from the source loc,
// not just the defining memcpy.
- MemDepResult SourceDep =
- MD->getPointerDependencyFrom(AA.getLocationForSource(MDep),
- false, M, M->getParent());
+ MemDepResult SourceDep = MD->getPointerDependencyFrom(
+ MemoryLocation::getForSource(MDep), false, M, M->getParent());
if (!SourceDep.isClobber() || SourceDep.getInst() != MDep)
return false;
// source and dest might overlap. We still want to eliminate the intermediate
// value, but we have to generate a memmove instead of memcpy.
bool UseMemMove = false;
- if (!AA.isNoAlias(AA.getLocationForDest(M), AA.getLocationForSource(MDep)))
+ if (!AA.isNoAlias(MemoryLocation::getForDest(M),
+ MemoryLocation::getForSource(MDep)))
UseMemMove = true;
// If all checks passed, then we can transform M.
return true;
}
+/// We've found that the (upward scanning) memory dependence of \p MemCpy is
+/// \p MemSet. Try to simplify \p MemSet to only set the trailing bytes that
+/// weren't copied over by \p MemCpy.
+///
+/// In other words, transform:
+/// \code
+/// memset(dst, c, dst_size);
+/// memcpy(dst, src, src_size);
+/// \endcode
+/// into:
+/// \code
+/// memcpy(dst, src, src_size);
+/// memset(dst + src_size, c, dst_size <= src_size ? 0 : dst_size - src_size);
+/// \endcode
+bool MemCpyOpt::processMemSetMemCpyDependence(MemCpyInst *MemCpy,
+ MemSetInst *MemSet) {
+ // We can only transform memset/memcpy with the same destination.
+ if (MemSet->getDest() != MemCpy->getDest())
+ return false;
+
+ // Check that there are no other dependencies on the memset destination.
+ MemDepResult DstDepInfo = MD->getPointerDependencyFrom(
+ MemoryLocation::getForDest(MemSet), false, MemCpy, MemCpy->getParent());
+ if (DstDepInfo.getInst() != MemSet)
+ return false;
+
+ // Use the same i8* dest as the memcpy, killing the memset dest if different.
+ Value *Dest = MemCpy->getRawDest();
+ Value *DestSize = MemSet->getLength();
+ Value *SrcSize = MemCpy->getLength();
+
+ // By default, create an unaligned memset.
+ unsigned Align = 1;
+ // If Dest is aligned, and SrcSize is constant, use the minimum alignment
+ // of the sum.
+ const unsigned DestAlign =
+ std::max(MemSet->getAlignment(), MemCpy->getAlignment());
+ if (DestAlign > 1)
+ if (ConstantInt *SrcSizeC = dyn_cast<ConstantInt>(SrcSize))
+ Align = MinAlign(SrcSizeC->getZExtValue(), DestAlign);
+
+ IRBuilder<> Builder(MemCpy);
+
+ // If the sizes have different types, zext the smaller one.
+ if (DestSize->getType() != SrcSize->getType()) {
+ if (DestSize->getType()->getIntegerBitWidth() >
+ SrcSize->getType()->getIntegerBitWidth())
+ SrcSize = Builder.CreateZExt(SrcSize, DestSize->getType());
+ else
+ DestSize = Builder.CreateZExt(DestSize, SrcSize->getType());
+ }
+
+ Value *MemsetLen =
+ Builder.CreateSelect(Builder.CreateICmpULE(DestSize, SrcSize),
+ ConstantInt::getNullValue(DestSize->getType()),
+ Builder.CreateSub(DestSize, SrcSize));
+ Builder.CreateMemSet(Builder.CreateGEP(Dest, SrcSize), MemSet->getOperand(1),
+ MemsetLen, Align);
+
+ MD->removeInstruction(MemSet);
+ MemSet->eraseFromParent();
+ return true;
+}
+
+/// Transform memcpy to memset when its source was just memset.
+/// In other words, turn:
+/// \code
+/// memset(dst1, c, dst1_size);
+/// memcpy(dst2, dst1, dst2_size);
+/// \endcode
+/// into:
+/// \code
+/// memset(dst1, c, dst1_size);
+/// memset(dst2, c, dst2_size);
+/// \endcode
+/// When dst2_size <= dst1_size.
+///
+/// The \p MemCpy must have a Constant length.
+bool MemCpyOpt::performMemCpyToMemSetOptzn(MemCpyInst *MemCpy,
+ MemSetInst *MemSet) {
+ // This only makes sense on memcpy(..., memset(...), ...).
+ if (MemSet->getRawDest() != MemCpy->getRawSource())
+ return false;
+
+ ConstantInt *CopySize = cast<ConstantInt>(MemCpy->getLength());
+ ConstantInt *MemSetSize = dyn_cast<ConstantInt>(MemSet->getLength());
+ // Make sure the memcpy doesn't read any more than what the memset wrote.
+ // Don't worry about sizes larger than i64.
+ if (!MemSetSize || CopySize->getZExtValue() > MemSetSize->getZExtValue())
+ return false;
+
+ IRBuilder<> Builder(MemCpy);
+ Builder.CreateMemSet(MemCpy->getRawDest(), MemSet->getOperand(1),
+ CopySize, MemCpy->getAlignment());
+ return true;
+}
/// processMemCpy - perform simplification of memcpy's. If we have memcpy A
/// which copies X to Y, and memcpy B which copies Y to Z, then we can rewrite
return true;
}
+ MemDepResult DepInfo = MD->getDependency(M);
+
+ // Try to turn a partially redundant memset + memcpy into
+ // memcpy + smaller memset. We don't need the memcpy size for this.
+ if (DepInfo.isClobber())
+ if (MemSetInst *MDep = dyn_cast<MemSetInst>(DepInfo.getInst()))
+ if (processMemSetMemCpyDependence(M, MDep))
+ return true;
+
// The optimizations after this point require the memcpy size.
ConstantInt *CopySize = dyn_cast<ConstantInt>(M->getLength());
if (!CopySize) return false;
- // The are three possible optimizations we can do for memcpy:
+ // There are four possible optimizations we can do for memcpy:
// a) memcpy-memcpy xform which exposes redundance for DSE.
// b) call-memcpy xform for return slot optimization.
// c) memcpy from freshly alloca'd space or space that has just started its
// lifetime copies undefined data, and we can therefore eliminate the
// memcpy in favor of the data that was already at the destination.
- MemDepResult DepInfo = MD->getDependency(M);
+ // d) memcpy from a just-memset'd source can be turned into memset.
if (DepInfo.isClobber()) {
if (CallInst *C = dyn_cast<CallInst>(DepInfo.getInst())) {
if (performCallSlotOptzn(M, M->getDest(), M->getSource(),
}
}
- AliasAnalysis::Location SrcLoc = AliasAnalysis::getLocationForSource(M);
+ AliasAnalysis::Location SrcLoc = MemoryLocation::getForSource(M);
MemDepResult SrcDepInfo = MD->getPointerDependencyFrom(SrcLoc, true,
M, M->getParent());
+
if (SrcDepInfo.isClobber()) {
if (MemCpyInst *MDep = dyn_cast<MemCpyInst>(SrcDepInfo.getInst()))
- return processMemCpyMemCpyDependence(M, MDep, CopySize->getZExtValue());
+ return processMemCpyMemCpyDependence(M, MDep);
} else if (SrcDepInfo.isDef()) {
Instruction *I = SrcDepInfo.getInst();
bool hasUndefContents = false;
}
}
+ if (SrcDepInfo.isClobber())
+ if (MemSetInst *MDep = dyn_cast<MemSetInst>(SrcDepInfo.getInst()))
+ if (performMemCpyToMemSetOptzn(M, MDep)) {
+ MD->removeInstruction(M);
+ M->eraseFromParent();
+ ++NumCpyToSet;
+ return true;
+ }
+
return false;
}
return false;
// See if the pointers alias.
- if (!AA.isNoAlias(AA.getLocationForDest(M), AA.getLocationForSource(M)))
+ if (!AA.isNoAlias(MemoryLocation::getForDest(M),
+ MemoryLocation::getForSource(M)))
return false;
DEBUG(dbgs() << "MemCpyOpt: Optimizing memmove -> memcpy: " << *M << "\n");
/// processByValArgument - This is called on every byval argument in call sites.
bool MemCpyOpt::processByValArgument(CallSite CS, unsigned ArgNo) {
- if (!DL) return false;
-
+ const DataLayout &DL = CS.getCaller()->getParent()->getDataLayout();
// Find out what feeds this byval argument.
Value *ByValArg = CS.getArgument(ArgNo);
Type *ByValTy = cast<PointerType>(ByValArg->getType())->getElementType();
- uint64_t ByValSize = DL->getTypeAllocSize(ByValTy);
+ uint64_t ByValSize = DL.getTypeAllocSize(ByValTy);
MemDepResult DepInfo =
MD->getPointerDependencyFrom(AliasAnalysis::Location(ByValArg, ByValSize),
true, CS.getInstruction(),
*CS->getParent()->getParent());
DominatorTree &DT = getAnalysis<DominatorTreeWrapperPass>().getDomTree();
if (MDep->getAlignment() < ByValAlign &&
- getOrEnforceKnownAlignment(MDep->getSource(), ByValAlign, DL, &AC,
- CS.getInstruction(), &DT) < ByValAlign)
+ getOrEnforceKnownAlignment(MDep->getSource(), ByValAlign, DL,
+ CS.getInstruction(), &AC, &DT) < ByValAlign)
return false;
// Verify that the copied-from memory doesn't change in between the memcpy and
// 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());
+ MD->getPointerDependencyFrom(MemoryLocation::getForSource(MDep), false,
+ CS.getInstruction(), MDep->getParent());
if (!SourceDep.isClobber() || SourceDep.getInst() != MDep)
return false;
RepeatInstruction = processMemCpy(M);
else if (MemMoveInst *M = dyn_cast<MemMoveInst>(I))
RepeatInstruction = processMemMove(M);
- else if (CallSite CS = (Value*)I) {
+ else if (auto CS = CallSite(I)) {
for (unsigned i = 0, e = CS.arg_size(); i != e; ++i)
if (CS.isByValArgument(i))
MadeChange |= processByValArgument(CS, i);
bool MadeChange = false;
MD = &getAnalysis<MemoryDependenceAnalysis>();
- DataLayoutPass *DLP = getAnalysisIfAvailable<DataLayoutPass>();
- DL = DLP ? &DLP->getDataLayout() : nullptr;
TLI = &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI();
// If we don't have at least memset and memcpy, there is little point of doing
projects / oota-llvm.git / blobdiff