bool Scalarizer::visitStoreInst(StoreInst &SI) {
if (!ScalarizeLoadStore)
return false;
if (!SI.isSimple())
return false;
VectorLayout Layout;
Value *FullValue = SI.getValueOperand();
if (!getVectorLayout(FullValue->getType(), SI.getAlignment(), Layout))
return false;
unsigned NumElems = Layout.VecTy->getNumElements();
IRBuilder<> Builder(SI.getParent(), &SI);
Scatterer Ptr = scatter(&SI, SI.getPointerOperand());
Scatterer Val = scatter(&SI, FullValue);
ValueVector Stores;
Stores.resize(NumElems);
for (unsigned I = 0; I < NumElems; ++I) {
unsigned Align = Layout.getElemAlign(I);
Stores[I] = Builder.CreateAlignedStore(Val[I], Ptr[I], Align);
}
transferMetadata(&SI, Stores);
return true;
}
/// \brief Combine stores to match the type of value being stored.
///
/// The core idea here is that the memory does not have any intrinsic type and
/// where we can we should match the type of a store to the type of value being
/// stored.
///
/// However, this routine must never change the width of a store or the number of
/// stores as that would introduce a semantic change. This combine is expected to
/// be a semantic no-op which just allows stores to more closely model the types
/// of their incoming values.
///
/// Currently, we also refuse to change the precise type used for an atomic or
/// volatile store. This is debatable, and might be reasonable to change later.
/// However, it is risky in case some backend or other part of LLVM is relying
/// on the exact type stored to select appropriate atomic operations.
///
/// \returns true if the store was successfully combined away. This indicates
/// the caller must erase the store instruction. We have to let the caller erase
/// the store instruction sas otherwise there is no way to signal whether it was
/// combined or not: IC.EraseInstFromFunction returns a null pointer.
static bool combineStoreToValueType(InstCombiner &IC, StoreInst &SI) {
// FIXME: We could probably with some care handle both volatile and atomic
// stores here but it isn't clear that this is important.
if (!SI.isSimple())
return false;
Value *Ptr = SI.getPointerOperand();
Value *V = SI.getValueOperand();
unsigned AS = SI.getPointerAddressSpace();
SmallVector<std::pair<unsigned, MDNode *>, 8> MD;
SI.getAllMetadata(MD);
// Fold away bit casts of the stored value by storing the original type.
if (auto *BC = dyn_cast<BitCastInst>(V)) {
V = BC->getOperand(0);
StoreInst *NewStore = IC.Builder->CreateAlignedStore(
V, IC.Builder->CreateBitCast(Ptr, V->getType()->getPointerTo(AS)),
SI.getAlignment());
for (const auto &MDPair : MD) {
unsigned ID = MDPair.first;
MDNode *N = MDPair.second;
// Note, essentially every kind of metadata should be preserved here! This
// routine is supposed to clone a store instruction changing *only its
// type*. The only metadata it makes sense to drop is metadata which is
// invalidated when the pointer type changes. This should essentially
// never be the case in LLVM, but we explicitly switch over only known
// metadata to be conservatively correct. If you are adding metadata to
// LLVM which pertains to stores, you almost certainly want to add it
// here.
switch (ID) {
case LLVMContext::MD_dbg:
case LLVMContext::MD_tbaa:
case LLVMContext::MD_prof:
case LLVMContext::MD_fpmath:
case LLVMContext::MD_tbaa_struct:
case LLVMContext::MD_alias_scope:
case LLVMContext::MD_noalias:
case LLVMContext::MD_nontemporal:
case LLVMContext::MD_mem_parallel_loop_access:
case LLVMContext::MD_nonnull:
// All of these directly apply.
NewStore->setMetadata(ID, N);
break;
case LLVMContext::MD_invariant_load:
case LLVMContext::MD_range:
break;
}
}
return true;
}
// FIXME: We should also canonicalize loads of vectors when their elements are
// cast to other types.
return false;
}
/// store {atomic|volatile} T %val, T* %ptr memory_order, align sizeof(T)
/// becomes:
/// call void @llvm.nacl.atomic.store.i<size>(%val, %ptr, memory_order)
void AtomicVisitor::visitStoreInst(StoreInst &I) {
return; // XXX EMSCRIPTEN
if (I.isSimple())
return;
PointerHelper<StoreInst> PH(*this, I);
const NaCl::AtomicIntrinsics::AtomicIntrinsic *Intrinsic =
findAtomicIntrinsic(I, Intrinsic::nacl_atomic_store, PH.PET);
checkAlignment(I, I.getAlignment(), PH.BitSize / CHAR_BIT);
Value *V = I.getValueOperand();
if (!V->getType()->isIntegerTy()) {
// The store isn't of an integer type. We define atomics in terms of
// integers, so bitcast the value to store to an integer of the
// proper width.
CastInst *Cast = createCast(I, V, Type::getIntNTy(C, PH.BitSize),
V->getName() + ".cast");
Cast->setDebugLoc(I.getDebugLoc());
V = Cast;
}
checkSizeMatchesType(I, PH.BitSize, V->getType());
Value *Args[] = {V, PH.P, freezeMemoryOrder(I, I.getOrdering())};
replaceInstructionWithIntrinsicCall(I, Intrinsic, PH.OriginalPET, PH.PET,
Args);
}
void Lint::visitStoreInst(StoreInst &I) {
visitMemoryReference(I, I.getPointerOperand(),
DL->getTypeStoreSize(I.getOperand(0)->getType()),
I.getAlignment(),
I.getOperand(0)->getType(), MemRef::Write);
}
// Lowers this interleaved access group into X86-specific
// instructions/intrinsics.
bool X86InterleavedAccessGroup::lowerIntoOptimizedSequence() {
SmallVector<Instruction *, 4> DecomposedVectors;
SmallVector<Value *, 4> TransposedVectors;
VectorType *ShuffleTy = Shuffles[0]->getType();
if (isa<LoadInst>(Inst)) {
// Try to generate target-sized register(/instruction).
decompose(Inst, Factor, ShuffleTy, DecomposedVectors);
Type *ShuffleEltTy = Inst->getType();
unsigned NumSubVecElems = ShuffleEltTy->getVectorNumElements() / Factor;
// Perform matrix-transposition in order to compute interleaved
// results by generating some sort of (optimized) target-specific
// instructions.
switch (NumSubVecElems) {
default:
return false;
case 4:
transpose_4x4(DecomposedVectors, TransposedVectors);
break;
case 8:
case 16:
case 32:
deinterleave8bitStride3(DecomposedVectors, TransposedVectors,
NumSubVecElems);
break;
}
// Now replace the unoptimized-interleaved-vectors with the
// transposed-interleaved vectors.
for (unsigned i = 0, e = Shuffles.size(); i < e; ++i)
Shuffles[i]->replaceAllUsesWith(TransposedVectors[Indices[i]]);
return true;
}
Type *ShuffleEltTy = ShuffleTy->getVectorElementType();
unsigned NumSubVecElems = ShuffleTy->getVectorNumElements() / Factor;
// Lower the interleaved stores:
// 1. Decompose the interleaved wide shuffle into individual shuffle
// vectors.
decompose(Shuffles[0], Factor, VectorType::get(ShuffleEltTy, NumSubVecElems),
DecomposedVectors);
// 2. Transpose the interleaved-vectors into vectors of contiguous
// elements.
switch (NumSubVecElems) {
case 4:
transpose_4x4(DecomposedVectors, TransposedVectors);
break;
case 16:
case 32:
interleave8bitStride4(DecomposedVectors, TransposedVectors, NumSubVecElems);
break;
default:
return false;
}
// 3. Concatenate the contiguous-vectors back into a wide vector.
Value *WideVec = concatenateVectors(Builder, TransposedVectors);
// 4. Generate a store instruction for wide-vec.
StoreInst *SI = cast<StoreInst>(Inst);
Builder.CreateAlignedStore(WideVec, SI->getPointerOperand(),
SI->getAlignment());
return true;
}
/// SimplifyStoreAtEndOfBlock - Turn things like:
/// if () { *P = v1; } else { *P = v2 }
/// into a phi node with a store in the successor.
///
/// Simplify things like:
/// *P = v1; if () { *P = v2; }
/// into a phi node with a store in the successor.
///
bool InstCombiner::SimplifyStoreAtEndOfBlock(StoreInst &SI) {
BasicBlock *StoreBB = SI.getParent();
// Check to see if the successor block has exactly two incoming edges. If
// so, see if the other predecessor contains a store to the same location.
// if so, insert a PHI node (if needed) and move the stores down.
BasicBlock *DestBB = StoreBB->getTerminator()->getSuccessor(0);
// Determine whether Dest has exactly two predecessors and, if so, compute
// the other predecessor.
pred_iterator PI = pred_begin(DestBB);
BasicBlock *P = *PI;
BasicBlock *OtherBB = nullptr;
if (P != StoreBB)
OtherBB = P;
if (++PI == pred_end(DestBB))
return false;
P = *PI;
if (P != StoreBB) {
if (OtherBB)
return false;
OtherBB = P;
}
if (++PI != pred_end(DestBB))
return false;
// Bail out if all the relevant blocks aren't distinct (this can happen,
// for example, if SI is in an infinite loop)
if (StoreBB == DestBB || OtherBB == DestBB)
return false;
// Verify that the other block ends in a branch and is not otherwise empty.
BasicBlock::iterator BBI(OtherBB->getTerminator());
BranchInst *OtherBr = dyn_cast<BranchInst>(BBI);
if (!OtherBr || BBI == OtherBB->begin())
return false;
// If the other block ends in an unconditional branch, check for the 'if then
// else' case. there is an instruction before the branch.
StoreInst *OtherStore = nullptr;
if (OtherBr->isUnconditional()) {
--BBI;
// Skip over debugging info.
while (isa<DbgInfoIntrinsic>(BBI) ||
(isa<BitCastInst>(BBI) && BBI->getType()->isPointerTy())) {
if (BBI==OtherBB->begin())
return false;
--BBI;
}
// If this isn't a store, isn't a store to the same location, or is not the
// right kind of store, bail out.
OtherStore = dyn_cast<StoreInst>(BBI);
if (!OtherStore || OtherStore->getOperand(1) != SI.getOperand(1) ||
!SI.isSameOperationAs(OtherStore))
return false;
} else {
// Otherwise, the other block ended with a conditional branch. If one of the
// destinations is StoreBB, then we have the if/then case.
if (OtherBr->getSuccessor(0) != StoreBB &&
OtherBr->getSuccessor(1) != StoreBB)
return false;
// Okay, we know that OtherBr now goes to Dest and StoreBB, so this is an
// if/then triangle. See if there is a store to the same ptr as SI that
// lives in OtherBB.
for (;; --BBI) {
// Check to see if we find the matching store.
if ((OtherStore = dyn_cast<StoreInst>(BBI))) {
if (OtherStore->getOperand(1) != SI.getOperand(1) ||
!SI.isSameOperationAs(OtherStore))
return false;
break;
}
// If we find something that may be using or overwriting the stored
// value, or if we run out of instructions, we can't do the xform.
if (BBI->mayReadFromMemory() || BBI->mayWriteToMemory() ||
BBI == OtherBB->begin())
return false;
}
// In order to eliminate the store in OtherBr, we have to
// make sure nothing reads or overwrites the stored value in
// StoreBB.
for (BasicBlock::iterator I = StoreBB->begin(); &*I != &SI; ++I) {
// FIXME: This should really be AA driven.
if (I->mayReadFromMemory() || I->mayWriteToMemory())
return false;
}
}
// Insert a PHI node now if we need it.
Value *MergedVal = OtherStore->getOperand(0);
if (MergedVal != SI.getOperand(0)) {
PHINode *PN = PHINode::Create(MergedVal->getType(), 2, "storemerge");
PN->addIncoming(SI.getOperand(0), SI.getParent());
PN->addIncoming(OtherStore->getOperand(0), OtherBB);
MergedVal = InsertNewInstBefore(PN, DestBB->front());
}
// Advance to a place where it is safe to insert the new store and
// insert it.
BBI = DestBB->getFirstInsertionPt();
StoreInst *NewSI = new StoreInst(MergedVal, SI.getOperand(1),
SI.isVolatile(),
SI.getAlignment(),
SI.getOrdering(),
SI.getSynchScope());
InsertNewInstBefore(NewSI, *BBI);
NewSI->setDebugLoc(OtherStore->getDebugLoc());
// If the two stores had AA tags, merge them.
AAMDNodes AATags;
SI.getAAMetadata(AATags);
if (AATags) {
OtherStore->getAAMetadata(AATags, /* Merge = */ true);
NewSI->setAAMetadata(AATags);
}
// Nuke the old stores.
EraseInstFromFunction(SI);
EraseInstFromFunction(*OtherStore);
return true;
}
unsigned CostModelAnalysis::getInstructionCost(Instruction *I) const {
if (!VTTI)
return -1;
switch (I->getOpcode()) {
case Instruction::Ret:
case Instruction::PHI:
case Instruction::Br: {
return VTTI->getCFInstrCost(I->getOpcode());
}
case Instruction::Add:
case Instruction::FAdd:
case Instruction::Sub:
case Instruction::FSub:
case Instruction::Mul:
case Instruction::FMul:
case Instruction::UDiv:
case Instruction::SDiv:
case Instruction::FDiv:
case Instruction::URem:
case Instruction::SRem:
case Instruction::FRem:
case Instruction::Shl:
case Instruction::LShr:
case Instruction::AShr:
case Instruction::And:
case Instruction::Or:
case Instruction::Xor: {
return VTTI->getArithmeticInstrCost(I->getOpcode(), I->getType());
}
case Instruction::Select: {
SelectInst *SI = cast<SelectInst>(I);
Type *CondTy = SI->getCondition()->getType();
return VTTI->getCmpSelInstrCost(I->getOpcode(), I->getType(), CondTy);
}
case Instruction::ICmp:
case Instruction::FCmp: {
Type *ValTy = I->getOperand(0)->getType();
return VTTI->getCmpSelInstrCost(I->getOpcode(), ValTy);
}
case Instruction::Store: {
StoreInst *SI = cast<StoreInst>(I);
Type *ValTy = SI->getValueOperand()->getType();
return VTTI->getMemoryOpCost(I->getOpcode(), ValTy,
SI->getAlignment(),
SI->getPointerAddressSpace());
}
case Instruction::Load: {
LoadInst *LI = cast<LoadInst>(I);
return VTTI->getMemoryOpCost(I->getOpcode(), I->getType(),
LI->getAlignment(),
LI->getPointerAddressSpace());
}
case Instruction::ZExt:
case Instruction::SExt:
case Instruction::FPToUI:
case Instruction::FPToSI:
case Instruction::FPExt:
case Instruction::PtrToInt:
case Instruction::IntToPtr:
case Instruction::SIToFP:
case Instruction::UIToFP:
case Instruction::Trunc:
case Instruction::FPTrunc:
case Instruction::BitCast: {
Type *SrcTy = I->getOperand(0)->getType();
return VTTI->getCastInstrCost(I->getOpcode(), I->getType(), SrcTy);
}
case Instruction::ExtractElement: {
ExtractElementInst * EEI = cast<ExtractElementInst>(I);
ConstantInt *CI = dyn_cast<ConstantInt>(I->getOperand(1));
unsigned Idx = -1;
if (CI)
Idx = CI->getZExtValue();
return VTTI->getVectorInstrCost(I->getOpcode(),
EEI->getOperand(0)->getType(), Idx);
}
case Instruction::InsertElement: {
InsertElementInst * IE = cast<InsertElementInst>(I);
ConstantInt *CI = dyn_cast<ConstantInt>(IE->getOperand(2));
unsigned Idx = -1;
if (CI)
Idx = CI->getZExtValue();
return VTTI->getVectorInstrCost(I->getOpcode(),
IE->getType(), Idx);
}
default:
// We don't have any information on this instruction.
return -1;
}
}
/// tryAggregating - When scanning forward over instructions, we look for
/// other loads or stores that could be aggregated with this one.
/// Returns the last instruction added (if one was added) since we might have
/// removed some loads or stores and that might invalidate an iterator.
Instruction *AggregateGlobalOpsOpt::tryAggregating(Instruction *StartInst, Value *StartPtr,
bool DebugThis) {
if (TD == 0) return 0;
Module* M = StartInst->getParent()->getParent()->getParent();
LLVMContext& Context = StartInst->getContext();
Type* int8Ty = Type::getInt8Ty(Context);
Type* sizeTy = Type::getInt64Ty(Context);
Type* globalInt8PtrTy = int8Ty->getPointerTo(globalSpace);
bool isLoad = isa<LoadInst>(StartInst);
bool isStore = isa<StoreInst>(StartInst);
Instruction *lastAddedInsn = NULL;
Instruction *LastLoadOrStore = NULL;
SmallVector<Instruction*, 8> toRemove;
// Okay, so we now have a single global load/store. 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.
MemOpRanges Ranges(*TD);
// Put the first store in since we want to preserve the order.
Ranges.addInst(0, StartInst);
BasicBlock::iterator BI = StartInst;
for (++BI; !isa<TerminatorInst>(BI); ++BI) {
if( isGlobalLoadOrStore(BI, globalSpace, isLoad, isStore) ) {
// OK!
} else {
// If the instruction is readnone, ignore it, otherwise bail out. We
// don't even allow readonly here because we don't want something like:
// A[1] = 2; strlen(A); A[2] = 2; -> memcpy(A, ...); strlen(A).
if (BI->mayWriteToMemory())
break;
if (isStore && BI->mayReadFromMemory())
break;
continue;
}
if ( isStore && isa<StoreInst>(BI) ) {
StoreInst *NextStore = cast<StoreInst>(BI);
// If this is a store, see if we can merge it in.
if (!NextStore->isSimple()) break;
// Check to see if this store is to a constant offset from the start ptr.
int64_t Offset;
if (!IsPointerOffset(StartPtr, NextStore->getPointerOperand(), Offset, *TD))
break;
Ranges.addStore(Offset, NextStore);
LastLoadOrStore = NextStore;
} else {
LoadInst *NextLoad = cast<LoadInst>(BI);
if (!NextLoad->isSimple()) break;
// Check to see if this load is to a constant offset from the start ptr.
int64_t Offset;
if (!IsPointerOffset(StartPtr, NextLoad->getPointerOperand(), Offset, *TD))
break;
Ranges.addLoad(Offset, NextLoad);
LastLoadOrStore = NextLoad;
}
}
// 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.moreThanOneOp())
return 0;
// Divide the instructions between StartInst and LastLoadOrStore into
// addressing, memops, and uses of memops (uses of loads)
reorderAddressingMemopsUses(StartInst, LastLoadOrStore, DebugThis);
Instruction* insertBefore = StartInst;
IRBuilder<> builder(insertBefore);
// Now that we have full information about ranges, loop over the ranges and
// emit memcpy's for anything big enough to be worthwhile.
for (MemOpRanges::const_iterator I = Ranges.begin(), E = Ranges.end();
I != E; ++I) {
const MemOpRange &Range = *I;
Value* oldBaseI = NULL;
Value* newBaseI = NULL;
if (Range.TheStores.size() == 1) continue; // Don't bother if there's only one thing...
builder.SetInsertPoint(insertBefore);
// Otherwise, we do want to transform this! Create a new memcpy.
// Get the starting pointer of the block.
StartPtr = Range.StartPtr;
if( DebugThis ) {
errs() << "base is:";
StartPtr->dump();
}
// Determine alignment
unsigned Alignment = Range.Alignment;
if (Alignment == 0) {
Type *EltType =
cast<PointerType>(StartPtr->getType())->getElementType();
Alignment = TD->getABITypeAlignment(EltType);
}
Instruction *alloc = NULL;
Value *globalPtr = NULL;
// create temporary alloca space to communicate to/from.
alloc = makeAlloca(int8Ty, "agg.tmp", insertBefore,
Range.End-Range.Start, Alignment);
// Generate the old and new base pointers before we output
// anything else.
{
Type* iPtrTy = TD->getIntPtrType(alloc->getType());
Type* iNewBaseTy = TD->getIntPtrType(alloc->getType());
oldBaseI = builder.CreatePtrToInt(StartPtr, iPtrTy, "agg.tmp.oldb.i");
newBaseI = builder.CreatePtrToInt(alloc, iNewBaseTy, "agg.tmp.newb.i");
}
// If storing, do the stores we had into our alloca'd region.
if( isStore ) {
for (SmallVector<Instruction*, 16>::const_iterator
SI = Range.TheStores.begin(),
SE = Range.TheStores.end(); SI != SE; ++SI) {
StoreInst* oldStore = cast<StoreInst>(*SI);
if( DebugThis ) {
errs() << "have store in range:";
oldStore->dump();
}
Value* ptrToAlloc = rebasePointer(oldStore->getPointerOperand(),
StartPtr, alloc, "agg.tmp",
&builder, *TD, oldBaseI, newBaseI);
// Old load must not be volatile or atomic... or we shouldn't have put
// it in ranges
assert(!(oldStore->isVolatile() || oldStore->isAtomic()));
StoreInst* newStore =
builder.CreateStore(oldStore->getValueOperand(), ptrToAlloc);
newStore->setAlignment(oldStore->getAlignment());
newStore->takeName(oldStore);
}
}
// cast the pointer that was load/stored to i8 if necessary.
if( StartPtr->getType()->getPointerElementType() == int8Ty ) {
globalPtr = StartPtr;
} else {
globalPtr = builder.CreatePointerCast(StartPtr, globalInt8PtrTy, "agg.cast");
}
// Get a Constant* for the length.
Constant* len = ConstantInt::get(sizeTy, Range.End-Range.Start, false);
// Now add the memcpy instruction
unsigned addrSpaceDst,addrSpaceSrc;
addrSpaceDst = addrSpaceSrc = 0;
if( isStore ) addrSpaceDst = globalSpace;
if( isLoad ) addrSpaceSrc = globalSpace;
Type *types[3];
types[0] = PointerType::get(int8Ty, addrSpaceDst);
types[1] = PointerType::get(int8Ty, addrSpaceSrc);
types[2] = sizeTy;
Function *func = Intrinsic::getDeclaration(M, Intrinsic::memcpy, types);
Value* args[5]; // dst src len alignment isvolatile
if( isStore ) {
// it's a store (ie put)
args[0] = globalPtr;
args[1] = alloc;
} else {
// it's a load (ie get)
args[0] = alloc;
args[1] = globalPtr;
}
args[2] = len;
// alignment
args[3] = ConstantInt::get(Type::getInt32Ty(Context), 0, false);
// isvolatile
args[4] = ConstantInt::get(Type::getInt1Ty(Context), 0, false);
Instruction* aMemCpy = builder.CreateCall(func, args);
/*
DEBUG(dbgs() << "Replace ops:\n";
for (unsigned i = 0, e = Range.TheStores.size(); i != e; ++i)
dbgs() << *Range.TheStores[i] << '\n';
dbgs() << "With: " << *AMemSet << '\n');
*/
if (!Range.TheStores.empty())
aMemCpy->setDebugLoc(Range.TheStores[0]->getDebugLoc());
lastAddedInsn = aMemCpy;
// If loading, load from the memcpy'd region
if( isLoad ) {
for (SmallVector<Instruction*, 16>::const_iterator
SI = Range.TheStores.begin(),
SE = Range.TheStores.end(); SI != SE; ++SI) {
LoadInst* oldLoad = cast<LoadInst>(*SI);
if( DebugThis ) {
errs() << "have load in range:";
oldLoad->dump();
}
Value* ptrToAlloc = rebasePointer(oldLoad->getPointerOperand(),
StartPtr, alloc, "agg.tmp",
&builder, *TD, oldBaseI, newBaseI);
// Old load must not be volatile or atomic... or we shouldn't have put
// it in ranges
assert(!(oldLoad->isVolatile() || oldLoad->isAtomic()));
LoadInst* newLoad = builder.CreateLoad(ptrToAlloc);
newLoad->setAlignment(oldLoad->getAlignment());
oldLoad->replaceAllUsesWith(newLoad);
newLoad->takeName(oldLoad);
lastAddedInsn = newLoad;
}
}
// Save old loads/stores for removal
for (SmallVector<Instruction*, 16>::const_iterator
SI = Range.TheStores.begin(),
SE = Range.TheStores.end(); SI != SE; ++SI) {
Instruction* insn = *SI;
toRemove.push_back(insn);
}
}
// Zap all the old loads/stores
for (SmallVector<Instruction*, 16>::const_iterator
SI = toRemove.begin(),
SE = toRemove.end(); SI != SE; ++SI) {
(*SI)->eraseFromParent();
}
return lastAddedInsn;
}
Value *BoUpSLP::vectorizeTree(ValueList &VL, int VF) {
Type *ScalarTy = VL[0]->getType();
if (StoreInst *SI = dyn_cast<StoreInst>(VL[0]))
ScalarTy = SI->getValueOperand()->getType();
VectorType *VecTy = VectorType::get(ScalarTy, VF);
// Check if all of the operands are constants or identical.
bool AllConst = true;
bool AllSameScalar = true;
for (unsigned i = 0, e = VF; i < e; ++i) {
AllConst &= !!dyn_cast<Constant>(VL[i]);
AllSameScalar &= (VL[0] == VL[i]);
// Must have a single use.
Instruction *I = dyn_cast<Instruction>(VL[i]);
if (I && (I->getNumUses() > 1 || I->getParent() != BB))
return Scalarize(VL, VecTy);
}
// Is this a simple vector constant.
if (AllConst || AllSameScalar) return Scalarize(VL, VecTy);
// Scalarize unknown structures.
Instruction *VL0 = dyn_cast<Instruction>(VL[0]);
if (!VL0) return Scalarize(VL, VecTy);
unsigned Opcode = VL0->getOpcode();
for (unsigned i = 0, e = VF; i < e; ++i) {
Instruction *I = dyn_cast<Instruction>(VL[i]);
// If not all of the instructions are identical then we have to scalarize.
if (!I || Opcode != I->getOpcode()) return Scalarize(VL, VecTy);
}
switch (Opcode) {
case Instruction::Add:
case Instruction::FAdd:
case Instruction::Sub:
case Instruction::FSub:
case Instruction::Mul:
case Instruction::FMul:
case Instruction::UDiv:
case Instruction::SDiv:
case Instruction::FDiv:
case Instruction::URem:
case Instruction::SRem:
case Instruction::FRem:
case Instruction::Shl:
case Instruction::LShr:
case Instruction::AShr:
case Instruction::And:
case Instruction::Or:
case Instruction::Xor: {
ValueList LHSVL, RHSVL;
for (int i = 0; i < VF; ++i) {
RHSVL.push_back(cast<Instruction>(VL[i])->getOperand(0));
LHSVL.push_back(cast<Instruction>(VL[i])->getOperand(1));
}
Value *RHS = vectorizeTree(RHSVL, VF);
Value *LHS = vectorizeTree(LHSVL, VF);
IRBuilder<> Builder(GetLastInstr(VL, VF));
BinaryOperator *BinOp = dyn_cast<BinaryOperator>(VL0);
return Builder.CreateBinOp(BinOp->getOpcode(), RHS,LHS);
}
case Instruction::Load: {
LoadInst *LI = dyn_cast<LoadInst>(VL0);
unsigned Alignment = LI->getAlignment();
// Check if all of the loads are consecutive.
for (unsigned i = 1, e = VF; i < e; ++i)
if (!isConsecutiveAccess(VL[i-1], VL[i]))
return Scalarize(VL, VecTy);
IRBuilder<> Builder(GetLastInstr(VL, VF));
Value *VecPtr = Builder.CreateBitCast(LI->getPointerOperand(),
VecTy->getPointerTo());
LI = Builder.CreateLoad(VecPtr);
LI->setAlignment(Alignment);
return LI;
}
case Instruction::Store: {
StoreInst *SI = dyn_cast<StoreInst>(VL0);
unsigned Alignment = SI->getAlignment();
ValueList ValueOp;
for (int i = 0; i < VF; ++i)
ValueOp.push_back(cast<StoreInst>(VL[i])->getValueOperand());
Value *VecValue = vectorizeTree(ValueOp, VF);
IRBuilder<> Builder(GetLastInstr(VL, VF));
Value *VecPtr = Builder.CreateBitCast(SI->getPointerOperand(),
VecTy->getPointerTo());
Builder.CreateStore(VecValue, VecPtr)->setAlignment(Alignment);
for (int i = 0; i < VF; ++i)
cast<Instruction>(VL[i])->eraseFromParent();
return 0;
}
default:
return Scalarize(VL, VecTy);
}
}
void Lint::visitStoreInst(StoreInst &I) {
visitMemoryReference(I, I.getPointerOperand(), I.getAlignment(),
I.getOperand(0)->getType());
}
