6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2025-06-22 10:24:26 +00:00
Code Issues Projects Releases Wiki Activity

whitespace

Browse Source 
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@127589 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Andrew Trick
2011-03-14 16:48:10 +00:00
parent ae3ce17bac
commit d99b39e43b
1 changed files with 66 additions and 66 deletions
Download Patch File Download Diff File Expand all files Collapse all files

132
lib/Transforms/Scalar/LoopIdiomRecognize.cpp
View File

@ -81,7 +81,7 @@ namespace {
bool processLoopStore(StoreInst *SI, const SCEV *BECount); bool processLoopStore(StoreInst *SI, const SCEV *BECount);
bool processLoopMemSet(MemSetInst *MSI, const SCEV *BECount); bool processLoopMemSet(MemSetInst *MSI, const SCEV *BECount);
bool processLoopStridedStore(Value *DestPtr, unsigned StoreSize, bool processLoopStridedStore(Value *DestPtr, unsigned StoreSize,
unsigned StoreAlignment, unsigned StoreAlignment,
Value *SplatValue, Instruction *TheStore, Value *SplatValue, Instruction *TheStore,
@ -91,7 +91,7 @@ namespace {
const SCEVAddRecExpr *StoreEv, const SCEVAddRecExpr *StoreEv,
const SCEVAddRecExpr *LoadEv, const SCEVAddRecExpr *LoadEv,
const SCEV *BECount); const SCEV *BECount);
/// This transformation requires natural loop information & requires that /// This transformation requires natural loop information & requires that
/// loop preheaders be inserted into the CFG. /// loop preheaders be inserted into the CFG.
/// ///
@ -134,50 +134,50 @@ Pass *llvm::createLoopIdiomPass() { return new LoopIdiomRecognize(); }
/// ///
static void DeleteDeadInstruction(Instruction *I, ScalarEvolution &SE) { static void DeleteDeadInstruction(Instruction *I, ScalarEvolution &SE) {
SmallVector<Instruction*, 32> NowDeadInsts; SmallVector<Instruction*, 32> NowDeadInsts;
NowDeadInsts.push_back(I); NowDeadInsts.push_back(I);
// Before we touch this instruction, remove it from SE! // Before we touch this instruction, remove it from SE!
do { do {
Instruction *DeadInst = NowDeadInsts.pop_back_val(); Instruction *DeadInst = NowDeadInsts.pop_back_val();
// This instruction is dead, zap it, in stages. Start by removing it from // This instruction is dead, zap it, in stages. Start by removing it from
// SCEV. // SCEV.
SE.forgetValue(DeadInst); SE.forgetValue(DeadInst);
for (unsigned op = 0, e = DeadInst->getNumOperands(); op != e; ++op) { for (unsigned op = 0, e = DeadInst->getNumOperands(); op != e; ++op) {
Value *Op = DeadInst->getOperand(op); Value *Op = DeadInst->getOperand(op);
DeadInst->setOperand(op, 0); DeadInst->setOperand(op, 0);
// If this operand just became dead, add it to the NowDeadInsts list. // If this operand just became dead, add it to the NowDeadInsts list.
if (!Op->use_empty()) continue; if (!Op->use_empty()) continue;
if (Instruction *OpI = dyn_cast<Instruction>(Op)) if (Instruction *OpI = dyn_cast<Instruction>(Op))
if (isInstructionTriviallyDead(OpI)) if (isInstructionTriviallyDead(OpI))
NowDeadInsts.push_back(OpI); NowDeadInsts.push_back(OpI);
} }
DeadInst->eraseFromParent(); DeadInst->eraseFromParent();
} while (!NowDeadInsts.empty()); } while (!NowDeadInsts.empty());
} }
bool LoopIdiomRecognize::runOnLoop(Loop *L, LPPassManager &LPM) { bool LoopIdiomRecognize::runOnLoop(Loop *L, LPPassManager &LPM) {
CurLoop = L; CurLoop = L;
// The trip count of the loop must be analyzable. // The trip count of the loop must be analyzable.
SE = &getAnalysis<ScalarEvolution>(); SE = &getAnalysis<ScalarEvolution>();
if (!SE->hasLoopInvariantBackedgeTakenCount(L)) if (!SE->hasLoopInvariantBackedgeTakenCount(L))
return false; return false;
const SCEV *BECount = SE->getBackedgeTakenCount(L); const SCEV *BECount = SE->getBackedgeTakenCount(L);
if (isa<SCEVCouldNotCompute>(BECount)) return false; if (isa<SCEVCouldNotCompute>(BECount)) return false;
// If this loop executes exactly one time, then it should be peeled, not // If this loop executes exactly one time, then it should be peeled, not
// optimized by this pass. // optimized by this pass.
if (const SCEVConstant *BECst = dyn_cast<SCEVConstant>(BECount)) if (const SCEVConstant *BECst = dyn_cast<SCEVConstant>(BECount))
if (BECst->getValue()->getValue() == 0) if (BECst->getValue()->getValue() == 0)
return false; return false;
// We require target data for now. // We require target data for now.
TD = getAnalysisIfAvailable<TargetData>(); TD = getAnalysisIfAvailable<TargetData>();
if (TD == 0) return false; if (TD == 0) return false;
@ -185,14 +185,14 @@ bool LoopIdiomRecognize::runOnLoop(Loop *L, LPPassManager &LPM) {
DT = &getAnalysis<DominatorTree>(); DT = &getAnalysis<DominatorTree>();
LoopInfo &LI = getAnalysis<LoopInfo>(); LoopInfo &LI = getAnalysis<LoopInfo>();
TLI = &getAnalysis<TargetLibraryInfo>(); TLI = &getAnalysis<TargetLibraryInfo>();
SmallVector<BasicBlock*, 8> ExitBlocks; SmallVector<BasicBlock*, 8> ExitBlocks;
CurLoop->getUniqueExitBlocks(ExitBlocks); CurLoop->getUniqueExitBlocks(ExitBlocks);
DEBUG(dbgs() << "loop-idiom Scanning: F[" DEBUG(dbgs() << "loop-idiom Scanning: F["
<< L->getHeader()->getParent()->getName() << L->getHeader()->getParent()->getName()
<< "] Loop %" << L->getHeader()->getName() << "\n"); << "] Loop %" << L->getHeader()->getName() << "\n");
bool MadeChange = false; bool MadeChange = false;
// Scan all the blocks in the loop that are not in subloops. // Scan all the blocks in the loop that are not in subloops.
for (Loop::block_iterator BI = L->block_begin(), E = L->block_end(); BI != E; for (Loop::block_iterator BI = L->block_begin(), E = L->block_end(); BI != E;
@ -200,7 +200,7 @@ bool LoopIdiomRecognize::runOnLoop(Loop *L, LPPassManager &LPM) {
// Ignore blocks in subloops. // Ignore blocks in subloops.
if (LI.getLoopFor(*BI) != CurLoop) if (LI.getLoopFor(*BI) != CurLoop)
continue; continue;
MadeChange |= runOnLoopBlock(*BI, BECount, ExitBlocks); MadeChange |= runOnLoopBlock(*BI, BECount, ExitBlocks);
} }
return MadeChange; return MadeChange;
@ -217,7 +217,7 @@ bool LoopIdiomRecognize::runOnLoopBlock(BasicBlock *BB, const SCEV *BECount,
for (unsigned i = 0, e = ExitBlocks.size(); i != e; ++i) for (unsigned i = 0, e = ExitBlocks.size(); i != e; ++i)
if (!DT->dominates(BB, ExitBlocks[i])) if (!DT->dominates(BB, ExitBlocks[i]))
return false; return false;
bool MadeChange = false; bool MadeChange = false;
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ) { for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ) {
Instruction *Inst = I++; Instruction *Inst = I++;
@ -226,20 +226,20 @@ bool LoopIdiomRecognize::runOnLoopBlock(BasicBlock *BB, const SCEV *BECount,
WeakVH InstPtr(I); WeakVH InstPtr(I);
if (!processLoopStore(SI, BECount)) continue; if (!processLoopStore(SI, BECount)) continue;
MadeChange = true; MadeChange = true;
// If processing the store invalidated our iterator, start over from the // If processing the store invalidated our iterator, start over from the
// top of the block. // top of the block.
if (InstPtr == 0) if (InstPtr == 0)
I = BB->begin(); I = BB->begin();
continue; continue;
} }
// Look for memset instructions, which may be optimized to a larger memset. // Look for memset instructions, which may be optimized to a larger memset.
if (MemSetInst *MSI = dyn_cast<MemSetInst>(Inst)) { if (MemSetInst *MSI = dyn_cast<MemSetInst>(Inst)) {
WeakVH InstPtr(I); WeakVH InstPtr(I);
if (!processLoopMemSet(MSI, BECount)) continue; if (!processLoopMemSet(MSI, BECount)) continue;
MadeChange = true; MadeChange = true;
// If processing the memset invalidated our iterator, start over from the // If processing the memset invalidated our iterator, start over from the
// top of the block. // top of the block.
if (InstPtr == 0) if (InstPtr == 0)
@ -247,7 +247,7 @@ bool LoopIdiomRecognize::runOnLoopBlock(BasicBlock *BB, const SCEV *BECount,
continue; continue;
} }
} }
return MadeChange; return MadeChange;
} }
@ -258,12 +258,12 @@ bool LoopIdiomRecognize::processLoopStore(StoreInst *SI, const SCEV *BECount) {
Value *StoredVal = SI->getValueOperand(); Value *StoredVal = SI->getValueOperand();
Value *StorePtr = SI->getPointerOperand(); Value *StorePtr = SI->getPointerOperand();
// Reject stores that are so large that they overflow an unsigned. // Reject stores that are so large that they overflow an unsigned.
uint64_t SizeInBits = TD->getTypeSizeInBits(StoredVal->getType()); uint64_t SizeInBits = TD->getTypeSizeInBits(StoredVal->getType());
if ((SizeInBits & 7) || (SizeInBits >> 32) != 0) if ((SizeInBits & 7) || (SizeInBits >> 32) != 0)
return false; return false;
// See if the pointer expression is an AddRec like {base,+,1} on the current // See if the pointer expression is an AddRec like {base,+,1} on the current
// loop, which indicates a strided store. If we have something else, it's a // loop, which indicates a strided store. If we have something else, it's a
// random store we can't handle. // random store we can't handle.
@ -274,9 +274,9 @@ bool LoopIdiomRecognize::processLoopStore(StoreInst *SI, const SCEV *BECount) {
// Check to see if the stride matches the size of the store. If so, then we // Check to see if the stride matches the size of the store. If so, then we
// know that every byte is touched in the loop. // know that every byte is touched in the loop.
unsigned StoreSize = (unsigned)SizeInBits >> 3; unsigned StoreSize = (unsigned)SizeInBits >> 3;
const SCEVConstant *Stride = dyn_cast<SCEVConstant>(StoreEv->getOperand(1)); const SCEVConstant *Stride = dyn_cast<SCEVConstant>(StoreEv->getOperand(1));
if (Stride == 0 || StoreSize != Stride->getValue()->getValue()) { if (Stride == 0 || StoreSize != Stride->getValue()->getValue()) {
// TODO: Could also handle negative stride here someday, that will require // TODO: Could also handle negative stride here someday, that will require
// the validity check in mayLoopAccessLocation to be updated though. // the validity check in mayLoopAccessLocation to be updated though.
@ -285,7 +285,7 @@ bool LoopIdiomRecognize::processLoopStore(StoreInst *SI, const SCEV *BECount) {
dbgs() << "NEGATIVE STRIDE: " << *SI << "\n"; dbgs() << "NEGATIVE STRIDE: " << *SI << "\n";
dbgs() << "BB: " << *SI->getParent(); dbgs() << "BB: " << *SI->getParent();
} }
return false; return false;
} }
@ -319,9 +319,9 @@ processLoopMemSet(MemSetInst *MSI, const SCEV *BECount) {
// If we're not allowed to hack on memset, we fail. // If we're not allowed to hack on memset, we fail.
if (!TLI->has(LibFunc::memset)) if (!TLI->has(LibFunc::memset))
return false; return false;
Value *Pointer = MSI->getDest(); Value *Pointer = MSI->getDest();
// See if the pointer expression is an AddRec like {base,+,1} on the current // See if the pointer expression is an AddRec like {base,+,1} on the current
// loop, which indicates a strided store. If we have something else, it's a // loop, which indicates a strided store. If we have something else, it's a
// random store we can't handle. // random store we can't handle.
@ -333,16 +333,16 @@ processLoopMemSet(MemSetInst *MSI, const SCEV *BECount) {
uint64_t SizeInBytes = cast<ConstantInt>(MSI->getLength())->getZExtValue(); uint64_t SizeInBytes = cast<ConstantInt>(MSI->getLength())->getZExtValue();
if ((SizeInBytes >> 32) != 0) if ((SizeInBytes >> 32) != 0)
return false; return false;
// Check to see if the stride matches the size of the memset. If so, then we // Check to see if the stride matches the size of the memset. If so, then we
// know that every byte is touched in the loop. // know that every byte is touched in the loop.
const SCEVConstant *Stride = dyn_cast<SCEVConstant>(Ev->getOperand(1)); const SCEVConstant *Stride = dyn_cast<SCEVConstant>(Ev->getOperand(1));
// TODO: Could also handle negative stride here someday, that will require the // TODO: Could also handle negative stride here someday, that will require the
// validity check in mayLoopAccessLocation to be updated though. // validity check in mayLoopAccessLocation to be updated though.
if (Stride == 0 || MSI->getLength() != Stride->getValue()) if (Stride == 0 || MSI->getLength() != Stride->getValue())
return false; return false;
return processLoopStridedStore(Pointer, (unsigned)SizeInBytes, return processLoopStridedStore(Pointer, (unsigned)SizeInBytes,
MSI->getAlignment(), MSI->getValue(), MSI->getAlignment(), MSI->getValue(),
MSI, Ev, BECount); MSI, Ev, BECount);
@ -365,7 +365,7 @@ static bool mayLoopAccessLocation(Value *Ptr,AliasAnalysis::ModRefResult Access,
// to be exactly the size of the memset, which is (BECount+1)*StoreSize // to be exactly the size of the memset, which is (BECount+1)*StoreSize
if (const SCEVConstant *BECst = dyn_cast<SCEVConstant>(BECount)) if (const SCEVConstant *BECst = dyn_cast<SCEVConstant>(BECount))
AccessSize = (BECst->getValue()->getZExtValue()+1)*StoreSize; AccessSize = (BECst->getValue()->getZExtValue()+1)*StoreSize;
// TODO: For this to be really effective, we have to dive into the pointer // TODO: For this to be really effective, we have to dive into the pointer
// operand in the store. Store to &A[i] of 100 will always return may alias // operand in the store. Store to &A[i] of 100 will always return may alias
// with store of &A[100], we need to StoreLoc to be "A" with size of 100, // with store of &A[100], we need to StoreLoc to be "A" with size of 100,
@ -394,12 +394,12 @@ static Constant *getMemSetPatternValue(Value *V, const TargetData &TD) {
// that doesn't seem worthwhile. // that doesn't seem worthwhile.
Constant *C = dyn_cast<Constant>(V); Constant *C = dyn_cast<Constant>(V);
if (C == 0) return 0; if (C == 0) return 0;
// Only handle simple values that are a power of two bytes in size. // Only handle simple values that are a power of two bytes in size.
uint64_t Size = TD.getTypeSizeInBits(V->getType()); uint64_t Size = TD.getTypeSizeInBits(V->getType());
if (Size == 0 || (Size & 7) || (Size & (Size-1))) if (Size == 0 || (Size & 7) || (Size & (Size-1)))
return 0; return 0;
// Don't care enough about darwin/ppc to implement this. // Don't care enough about darwin/ppc to implement this.
if (TD.isBigEndian()) if (TD.isBigEndian())
return 0; return 0;
@ -410,7 +410,7 @@ static Constant *getMemSetPatternValue(Value *V, const TargetData &TD) {
// TODO: If CI is larger than 16-bytes, we can try slicing it in half to see // TODO: If CI is larger than 16-bytes, we can try slicing it in half to see
// if the top and bottom are the same (e.g. for vectors and large integers). // if the top and bottom are the same (e.g. for vectors and large integers).
if (Size > 16) return 0; if (Size > 16) return 0;
// If the constant is exactly 16 bytes, just use it. // If the constant is exactly 16 bytes, just use it.
if (Size == 16) return C; if (Size == 16) return C;
@ -428,14 +428,14 @@ processLoopStridedStore(Value *DestPtr, unsigned StoreSize,
unsigned StoreAlignment, Value *StoredVal, unsigned StoreAlignment, Value *StoredVal,
Instruction *TheStore, const SCEVAddRecExpr *Ev, Instruction *TheStore, const SCEVAddRecExpr *Ev,
const SCEV *BECount) { const SCEV *BECount) {
// If the stored value is a byte-wise value (like i32 -1), then it may be // If the stored value is a byte-wise value (like i32 -1), then it may be
// turned into a memset of i8 -1, assuming that all the consecutive bytes // turned into a memset of i8 -1, assuming that all the consecutive bytes
// are stored. A store of i32 0x01020304 can never be turned into a memset, // are stored. A store of i32 0x01020304 can never be turned into a memset,
// but it can be turned into memset_pattern if the target supports it. // but it can be turned into memset_pattern if the target supports it.
Value *SplatValue = isBytewiseValue(StoredVal); Value *SplatValue = isBytewiseValue(StoredVal);
Constant *PatternValue = 0; Constant *PatternValue = 0;
// If we're allowed to form a memset, and the stored value would be acceptable // If we're allowed to form a memset, and the stored value would be acceptable
// for memset, use it. // for memset, use it.
if (SplatValue && TLI->has(LibFunc::memset) && if (SplatValue && TLI->has(LibFunc::memset) &&
@ -453,8 +453,8 @@ processLoopStridedStore(Value *DestPtr, unsigned StoreSize,
// do anything with a 3-byte store, for example. // do anything with a 3-byte store, for example.
return false; return false;
} }
// Okay, we have a strided store "p[i]" of a splattable value. We can turn // Okay, we have a strided store "p[i]" of a splattable value. We can turn
// this into a memset in the loop preheader now if we want. However, this // this into a memset in the loop preheader now if we want. However, this
// would be unsafe to do if there is anything else in the loop that may read // would be unsafe to do if there is anything else in the loop that may read
@ -463,36 +463,36 @@ processLoopStridedStore(Value *DestPtr, unsigned StoreSize,
CurLoop, BECount, CurLoop, BECount,
StoreSize, getAnalysis<AliasAnalysis>(), TheStore)) StoreSize, getAnalysis<AliasAnalysis>(), TheStore))
return false; return false;
// Okay, everything looks good, insert the memset. // Okay, everything looks good, insert the memset.
BasicBlock *Preheader = CurLoop->getLoopPreheader(); BasicBlock *Preheader = CurLoop->getLoopPreheader();
IRBuilder<> Builder(Preheader->getTerminator()); IRBuilder<> Builder(Preheader->getTerminator());
// The trip count of the loop and the base pointer of the addrec SCEV is // The trip count of the loop and the base pointer of the addrec SCEV is
// guaranteed to be loop invariant, which means that it should dominate the // guaranteed to be loop invariant, which means that it should dominate the
// header. Just insert code for it in the preheader. // header. Just insert code for it in the preheader.
SCEVExpander Expander(*SE); SCEVExpander Expander(*SE);
unsigned AddrSpace = cast<PointerType>(DestPtr->getType())->getAddressSpace(); unsigned AddrSpace = cast<PointerType>(DestPtr->getType())->getAddressSpace();
Value *BasePtr = Value *BasePtr =
Expander.expandCodeFor(Ev->getStart(), Builder.getInt8PtrTy(AddrSpace), Expander.expandCodeFor(Ev->getStart(), Builder.getInt8PtrTy(AddrSpace),
Preheader->getTerminator()); Preheader->getTerminator());
// The # stored bytes is (BECount+1)*Size. Expand the trip count out to // The # stored bytes is (BECount+1)*Size. Expand the trip count out to
// pointer size if it isn't already. // pointer size if it isn't already.
const Type *IntPtr = TD->getIntPtrType(DestPtr->getContext()); const Type *IntPtr = TD->getIntPtrType(DestPtr->getContext());
BECount = SE->getTruncateOrZeroExtend(BECount, IntPtr); BECount = SE->getTruncateOrZeroExtend(BECount, IntPtr);
const SCEV *NumBytesS = SE->getAddExpr(BECount, SE->getConstant(IntPtr, 1), const SCEV *NumBytesS = SE->getAddExpr(BECount, SE->getConstant(IntPtr, 1),
true /*no unsigned overflow*/); true /*no unsigned overflow*/);
if (StoreSize != 1) if (StoreSize != 1)
NumBytesS = SE->getMulExpr(NumBytesS, SE->getConstant(IntPtr, StoreSize), NumBytesS = SE->getMulExpr(NumBytesS, SE->getConstant(IntPtr, StoreSize),
true /*no unsigned overflow*/); true /*no unsigned overflow*/);
Value *NumBytes = Value *NumBytes =
Expander.expandCodeFor(NumBytesS, IntPtr, Preheader->getTerminator()); Expander.expandCodeFor(NumBytesS, IntPtr, Preheader->getTerminator());
CallInst *NewCall; CallInst *NewCall;
if (SplatValue) if (SplatValue)
NewCall = Builder.CreateMemSet(BasePtr, SplatValue,NumBytes,StoreAlignment); NewCall = Builder.CreateMemSet(BasePtr, SplatValue,NumBytes,StoreAlignment);
@ -500,10 +500,10 @@ processLoopStridedStore(Value *DestPtr, unsigned StoreSize,
Module *M = TheStore->getParent()->getParent()->getParent(); Module *M = TheStore->getParent()->getParent()->getParent();
Value *MSP = M->getOrInsertFunction("memset_pattern16", Value *MSP = M->getOrInsertFunction("memset_pattern16",
Builder.getVoidTy(), Builder.getVoidTy(),
Builder.getInt8PtrTy(), Builder.getInt8PtrTy(),
Builder.getInt8PtrTy(), IntPtr, Builder.getInt8PtrTy(), IntPtr,
(void*)0); (void*)0);
// Otherwise we should form a memset_pattern16. PatternValue is known to be // Otherwise we should form a memset_pattern16. PatternValue is known to be
// an constant array of 16-bytes. Plop the value into a mergable global. // an constant array of 16-bytes. Plop the value into a mergable global.
GlobalVariable *GV = new GlobalVariable(*M, PatternValue->getType(), true, GlobalVariable *GV = new GlobalVariable(*M, PatternValue->getType(), true,
@ -514,11 +514,11 @@ processLoopStridedStore(Value *DestPtr, unsigned StoreSize,
Value *PatternPtr = ConstantExpr::getBitCast(GV, Builder.getInt8PtrTy()); Value *PatternPtr = ConstantExpr::getBitCast(GV, Builder.getInt8PtrTy());
NewCall = Builder.CreateCall3(MSP, BasePtr, PatternPtr, NumBytes); NewCall = Builder.CreateCall3(MSP, BasePtr, PatternPtr, NumBytes);
} }
DEBUG(dbgs() << " Formed memset: " << *NewCall << "\n" DEBUG(dbgs() << " Formed memset: " << *NewCall << "\n"
<< " from store to: " << *Ev << " at: " << *TheStore << "\n"); << " from store to: " << *Ev << " at: " << *TheStore << "\n");
NewCall->setDebugLoc(TheStore->getDebugLoc()); NewCall->setDebugLoc(TheStore->getDebugLoc());
// Okay, the memset has been formed. Zap the original store and anything that // Okay, the memset has been formed. Zap the original store and anything that
// feeds into it. // feeds into it.
DeleteDeadInstruction(TheStore, *SE); DeleteDeadInstruction(TheStore, *SE);
@ -536,9 +536,9 @@ processLoopStoreOfLoopLoad(StoreInst *SI, unsigned StoreSize,
// If we're not allowed to form memcpy, we fail. // If we're not allowed to form memcpy, we fail.
if (!TLI->has(LibFunc::memcpy)) if (!TLI->has(LibFunc::memcpy))
return false; return false;
LoadInst *LI = cast<LoadInst>(SI->getValueOperand()); LoadInst *LI = cast<LoadInst>(SI->getValueOperand());
// Okay, we have a strided store "p[i]" of a loaded value. We can turn // Okay, we have a strided store "p[i]" of a loaded value. We can turn
// this into a memcpy in the loop preheader now if we want. However, this // this into a memcpy in the loop preheader now if we want. However, this
// would be unsafe to do if there is anything else in the loop that may read // would be unsafe to do if there is anything else in the loop that may read
@ -555,49 +555,49 @@ processLoopStoreOfLoopLoad(StoreInst *SI, unsigned StoreSize,
CurLoop, BECount, StoreSize, CurLoop, BECount, StoreSize,
getAnalysis<AliasAnalysis>(), SI)) getAnalysis<AliasAnalysis>(), SI))
return false; return false;
// Okay, everything looks good, insert the memcpy. // Okay, everything looks good, insert the memcpy.
BasicBlock *Preheader = CurLoop->getLoopPreheader(); BasicBlock *Preheader = CurLoop->getLoopPreheader();
IRBuilder<> Builder(Preheader->getTerminator()); IRBuilder<> Builder(Preheader->getTerminator());
// The trip count of the loop and the base pointer of the addrec SCEV is // The trip count of the loop and the base pointer of the addrec SCEV is
// guaranteed to be loop invariant, which means that it should dominate the // guaranteed to be loop invariant, which means that it should dominate the
// header. Just insert code for it in the preheader. // header. Just insert code for it in the preheader.
SCEVExpander Expander(*SE); SCEVExpander Expander(*SE);
Value *LoadBasePtr = Value *LoadBasePtr =
Expander.expandCodeFor(LoadEv->getStart(), Expander.expandCodeFor(LoadEv->getStart(),
Builder.getInt8PtrTy(LI->getPointerAddressSpace()), Builder.getInt8PtrTy(LI->getPointerAddressSpace()),
Preheader->getTerminator()); Preheader->getTerminator());
Value *StoreBasePtr = Value *StoreBasePtr =
Expander.expandCodeFor(StoreEv->getStart(), Expander.expandCodeFor(StoreEv->getStart(),
Builder.getInt8PtrTy(SI->getPointerAddressSpace()), Builder.getInt8PtrTy(SI->getPointerAddressSpace()),
Preheader->getTerminator()); Preheader->getTerminator());
// The # stored bytes is (BECount+1)*Size. Expand the trip count out to // The # stored bytes is (BECount+1)*Size. Expand the trip count out to
// pointer size if it isn't already. // pointer size if it isn't already.
const Type *IntPtr = TD->getIntPtrType(SI->getContext()); const Type *IntPtr = TD->getIntPtrType(SI->getContext());
BECount = SE->getTruncateOrZeroExtend(BECount, IntPtr); BECount = SE->getTruncateOrZeroExtend(BECount, IntPtr);
const SCEV *NumBytesS = SE->getAddExpr(BECount, SE->getConstant(IntPtr, 1), const SCEV *NumBytesS = SE->getAddExpr(BECount, SE->getConstant(IntPtr, 1),
true /*no unsigned overflow*/); true /*no unsigned overflow*/);
if (StoreSize != 1) if (StoreSize != 1)
NumBytesS = SE->getMulExpr(NumBytesS, SE->getConstant(IntPtr, StoreSize), NumBytesS = SE->getMulExpr(NumBytesS, SE->getConstant(IntPtr, StoreSize),
true /*no unsigned overflow*/); true /*no unsigned overflow*/);
Value *NumBytes = Value *NumBytes =
Expander.expandCodeFor(NumBytesS, IntPtr, Preheader->getTerminator()); Expander.expandCodeFor(NumBytesS, IntPtr, Preheader->getTerminator());
Value *NewCall = Value *NewCall =
Builder.CreateMemCpy(StoreBasePtr, LoadBasePtr, NumBytes, Builder.CreateMemCpy(StoreBasePtr, LoadBasePtr, NumBytes,
std::min(SI->getAlignment(), LI->getAlignment())); std::min(SI->getAlignment(), LI->getAlignment()));
DEBUG(dbgs() << " Formed memcpy: " << *NewCall << "\n" DEBUG(dbgs() << " Formed memcpy: " << *NewCall << "\n"
<< " from load ptr=" << *LoadEv << " at: " << *LI << "\n" << " from load ptr=" << *LoadEv << " at: " << *LI << "\n"
<< " from store ptr=" << *StoreEv << " at: " << *SI << "\n"); << " from store ptr=" << *StoreEv << " at: " << *SI << "\n");
(void)NewCall; (void)NewCall;
// Okay, the memset has been formed. Zap the original store and anything that // Okay, the memset has been formed. Zap the original store and anything that
// feeds into it. // feeds into it.
DeleteDeadInstruction(SI, *SE); DeleteDeadInstruction(SI, *SE);