//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "memcpyopt"
#include "llvm/Transforms/Scalar.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
#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/Target/TargetLibraryInfo.h"
#include "llvm/Transforms/Utils/Local.h"
#include <list>
using namespace llvm;
+#define DEBUG_TYPE "memcpyopt"
+
STATISTIC(NumMemCpyInstr, "Number of memcpy instructions deleted");
STATISTIC(NumMemSetInfer, "Number of memsets inferred");
STATISTIC(NumMoveToCpy, "Number of memmoves converted to memcpy");
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)
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)
+ if (!OpC)
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());
+ 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();
+
+ // Handle the trivial case first.
+ if (Ptr1 == Ptr2) {
+ Offset = 0;
+ return true;
+ }
+
GEPOperator *GEP1 = dyn_cast<GEPOperator>(Ptr1);
GEPOperator *GEP2 = dyn_cast<GEPOperator>(Ptr2);
// 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 == 0 && GEP1->getOperand(0)->stripPointerCasts() == Ptr2) {
- Offset = -GetOffsetFromIndex(GEP1, 1, VariableIdxFound, TD);
+ if (GEP1 && !GEP2 && GEP1->getOperand(0)->stripPointerCasts() == Ptr2) {
+ Offset = -GetOffsetFromIndex(GEP1, 1, VariableIdxFound, DL);
return !VariableIdxFound;
}
- if (GEP2 && GEP1 == 0 && GEP2->getOperand(0)->stripPointerCasts() == Ptr1) {
- Offset = GetOffsetFromIndex(GEP2, 1, VariableIdxFound, TD);
+ if (GEP2 && !GEP1 && GEP2->getOperand(0)->stripPointerCasts() == Ptr1) {
+ 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 = 0;
- TLI = 0;
- DL = 0;
+ MD = nullptr;
+ TLI = nullptr;
}
- bool runOnFunction(Function &F);
+ bool runOnFunction(Function &F) override;
private:
// This transformation requires dominator postdominator info
- virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.setPreservesCFG();
+ AU.addRequired<AssumptionCacheTracker>();
AU.addRequired<DominatorTreeWrapperPass>();
AU.addRequired<MemoryDependenceAnalysis>();
AU.addRequired<AliasAnalysis>();
- AU.addRequired<TargetLibraryInfo>();
+ AU.addRequired<TargetLibraryInfoWrapperPass>();
AU.addPreserved<AliasAnalysis>();
AU.addPreserved<MemoryDependenceAnalysis>();
}
INITIALIZE_PASS_BEGIN(MemCpyOpt, "memcpyopt", "MemCpy Optimization",
false, false)
+INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)
INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
INITIALIZE_PASS_DEPENDENCY(MemoryDependenceAnalysis)
-INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfo)
+INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass)
INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
INITIALIZE_PASS_END(MemCpyOpt, "memcpyopt", "MemCpy Optimization",
false, false)
/// attempts to merge them together into a memcpy/memset.
Instruction *MemCpyOpt::tryMergingIntoMemset(Instruction *StartInst,
Value *StartPtr, Value *ByteVal) {
- if (DL == 0) return 0;
+ 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 we have no ranges, then we just had a single store with nothing that
// could be merged in. This is a very common case of course.
if (Ranges.empty())
- return 0;
+ return nullptr;
// If we had at least one store that could be merged in, add the starting
// store as well. We try to avoid this unless there is at least something
// Now that we have full information about ranges, loop over the ranges and
// emit memset's for anything big enough to be worthwhile.
- Instruction *AMemSet = 0;
+ Instruction *AMemSet = nullptr;
for (MemsetRanges::const_iterator I = Ranges.begin(), E = Ranges.end();
I != E; ++I) {
const MemsetRange &Range = *I;
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 == 0) 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
if (LI->isSimple() && LI->hasOneUse() &&
LI->getParent() == SI->getParent()) {
MemDepResult ldep = MD->getDependency(LI);
- CallInst *C = 0;
+ CallInst *C = nullptr;
if (ldep.isClobber() && !isa<MemCpyInst>(ldep.getInst()))
C = dyn_cast<CallInst>(ldep.getInst());
for (BasicBlock::iterator I = --BasicBlock::iterator(SI),
E = C; I != E; --I) {
if (AA.getModRefInfo(&*I, StoreLoc) != AliasAnalysis::NoModRef) {
- C = 0;
+ C = nullptr;
break;
}
}
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 == 0) 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;
} 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;
+ if (A->getDereferenceableBytes() < srcSize) {
+ // 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;
- Type *StructTy = cast<PointerType>(A->getType())->getElementType();
- if (!StructTy->isSized()) {
- // The call may never return and hence the copy-instruction may never
- // be executed, and therefore it's not safe to say "the destination
- // has at least <cpyLen> bytes, as implied by the copy-instruction",
- return false;
- }
+ Type *StructTy = cast<PointerType>(A->getType())->getElementType();
+ if (!StructTy->isSized()) {
+ // The call may never return and hence the copy-instruction may never
+ // be executed, and therefore it's not safe to say "the destination
+ // has at least <cpyLen> bytes, as implied by the copy-instruction",
+ return false;
+ }
- uint64_t destSize = DL->getTypeAllocSize(StructTy);
- if (destSize < srcSize)
- return false;
+ uint64_t destSize = DL.getTypeAllocSize(StructTy);
+ if (destSize < srcSize)
+ return false;
+ }
} else {
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.
// 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());
+ SmallVector<User*, 8> srcUseList(srcAlloca->user_begin(),
+ srcAlloca->user_end());
while (!srcUseList.empty()) {
- User *UI = srcUseList.pop_back_val();
-
- if (isa<BitCastInst>(UI) || isa<AddrSpaceCastInst>(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
+ User *U = srcUseList.pop_back_val();
+
+ if (isa<BitCastInst>(U) || isa<AddrSpaceCastInst>(U)) {
+ for (User *UU : U->users())
+ srcUseList.push_back(UU);
+ continue;
+ }
+ if (GetElementPtrInst *G = dyn_cast<GetElementPtrInst>(U)) {
+ if (!G->hasAllZeroIndices())
return false;
- } else if (UI != C && UI != cpy) {
- return false;
+
+ for (User *UU : U->users())
+ srcUseList.push_back(UU);
+ continue;
}
+ if (const IntrinsicInst *IT = dyn_cast<IntrinsicInst>(U))
+ if (IT->getIntrinsicID() == Intrinsic::lifetime_start ||
+ IT->getIntrinsicID() == Intrinsic::lifetime_end)
+ continue;
+
+ if (U != C && U != cpy)
+ return false;
}
+ // Check that src isn't captured by the called function since the
+ // transformation can cause aliasing issues in that case.
+ for (unsigned i = 0, e = CS.arg_size(); i != e; ++i)
+ if (CS.getArgument(i) == cpySrc && !CS.doesNotCapture(i))
+ 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<DominatorTreeWrapperPass>().getDomTree();
// 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;
// The optimizations after this point require the memcpy size.
ConstantInt *CopySize = dyn_cast<ConstantInt>(M->getLength());
- if (CopySize == 0) return false;
+ if (!CopySize) return false;
// The are three 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 copies undefined data, and we can
- // therefore eliminate the memcpy in favor of the data that was already
- // at the destination.
+ // 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);
if (DepInfo.isClobber()) {
if (CallInst *C = dyn_cast<CallInst>(DepInfo.getInst())) {
if (MemCpyInst *MDep = dyn_cast<MemCpyInst>(SrcDepInfo.getInst()))
return processMemCpyMemCpyDependence(M, MDep, CopySize->getZExtValue());
} else if (SrcDepInfo.isDef()) {
- if (isa<AllocaInst>(SrcDepInfo.getInst())) {
+ Instruction *I = SrcDepInfo.getInst();
+ bool hasUndefContents = false;
+
+ if (isa<AllocaInst>(I)) {
+ hasUndefContents = true;
+ } else if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
+ if (II->getIntrinsicID() == Intrinsic::lifetime_start)
+ if (ConstantInt *LTSize = dyn_cast<ConstantInt>(II->getArgOperand(0)))
+ if (LTSize->getZExtValue() >= CopySize->getZExtValue())
+ hasUndefContents = true;
+ }
+
+ if (hasUndefContents) {
MD->removeInstruction(M);
M->eraseFromParent();
++NumMemCpyInstr;
/// processByValArgument - This is called on every byval argument in call sites.
bool MemCpyOpt::processByValArgument(CallSite CS, unsigned ArgNo) {
- if (DL == 0) 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(),
// 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() ||
+ if (!MDep || 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)
+ if (!C1 || C1->getValue().getZExtValue() < ByValSize)
return false;
// Get the alignment of the byval. If the call doesn't specify the alignment,
// If it is greater than the memcpy, then we check to see if we can force the
// source of the memcpy to the alignment we need. If we fail, we bail out.
+ AssumptionCache &AC =
+ getAnalysis<AssumptionCacheTracker>().getAssumptionCache(
+ *CS->getParent()->getParent());
+ DominatorTree &DT = getAnalysis<DominatorTreeWrapperPass>().getDomTree();
if (MDep->getAlignment() < ByValAlign &&
- getOrEnforceKnownAlignment(MDep->getSource(),ByValAlign, DL) < 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
bool MadeChange = false;
MD = &getAnalysis<MemoryDependenceAnalysis>();
- DataLayoutPass *DLP = getAnalysisIfAvailable<DataLayoutPass>();
- DL = DLP ? &DLP->getDataLayout() : 0;
- TLI = &getAnalysis<TargetLibraryInfo>();
+ TLI = &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI();
// If we don't have at least memset and memcpy, there is little point of doing
// anything here. These are required by a freestanding implementation, so if
MadeChange = true;
}
- MD = 0;
+ MD = nullptr;
return MadeChange;
}