bool AMDGPUCodeGenPrepare::canWidenScalarExtLoad(LoadInst &I) const {
Type *Ty = I.getType();
const DataLayout &DL = Mod->getDataLayout();
int TySize = DL.getTypeSizeInBits(Ty);
unsigned Align = I.getAlignment() ?
I.getAlignment() : DL.getABITypeAlignment(Ty);
return I.isSimple() && TySize < 32 && Align >= 4 && DA->isUniform(&I);
}
/// %res = load {atomic|volatile} T* %ptr memory_order, align sizeof(T)
/// becomes:
/// %res = call T @llvm.nacl.atomic.load.i<size>(%ptr, memory_order)
void AtomicVisitor::visitLoadInst(LoadInst &I) {
return; // XXX EMSCRIPTEN
if (I.isSimple())
return;
PointerHelper<LoadInst> PH(*this, I);
const NaCl::AtomicIntrinsics::AtomicIntrinsic *Intrinsic =
findAtomicIntrinsic(I, Intrinsic::nacl_atomic_load, PH.PET);
checkAlignment(I, I.getAlignment(), PH.BitSize / CHAR_BIT);
Value *Args[] = {PH.P, freezeMemoryOrder(I, I.getOrdering())};
replaceInstructionWithIntrinsicCall(I, Intrinsic, PH.OriginalPET, PH.PET,
Args);
}
bool CallAnalyzer::visitLoad(LoadInst &I) {
Value *SROAArg;
DenseMap<Value *, int>::iterator CostIt;
if (lookupSROAArgAndCost(I.getOperand(0), SROAArg, CostIt)) {
if (I.isSimple()) {
accumulateSROACost(CostIt, InlineConstants::InstrCost);
return true;
}
disableSROA(CostIt);
}
return false;
}
bool IRTranslator::translateLoad(const LoadInst &LI) {
assert(LI.isSimple() && "only simple loads are supported at the moment");
MachineFunction &MF = MIRBuilder.getMF();
unsigned Res = getOrCreateVReg(LI);
unsigned Addr = getOrCreateVReg(*LI.getPointerOperand());
LLT VTy{*LI.getType()}, PTy{*LI.getPointerOperand()->getType()};
MIRBuilder.buildLoad(
VTy, PTy, Res, Addr,
*MF.getMachineMemOperand(MachinePointerInfo(LI.getPointerOperand()),
MachineMemOperand::MOLoad,
VTy.getSizeInBits() / 8, getMemOpAlignment(LI)));
return true;
}
/// \brief Check loop instructions safe for Loop versioning.
/// It returns true if it's safe else returns false.
/// Consider following:
/// 1) Check all load store in loop body are non atomic & non volatile.
/// 2) Check function call safety, by ensuring its not accessing memory.
/// 3) Loop body shouldn't have any may throw instruction.
bool LoopVersioningLICM::instructionSafeForVersioning(Instruction *I) {
assert(I != nullptr && "Null instruction found!");
// Check function call safety
if (isa<CallInst>(I) && !AA->doesNotAccessMemory(CallSite(I))) {
DEBUG(dbgs() << " Unsafe call site found.\n");
return false;
}
// Avoid loops with possiblity of throw
if (I->mayThrow()) {
DEBUG(dbgs() << " May throw instruction found in loop body\n");
return false;
}
// If current instruction is load instructions
// make sure it's a simple load (non atomic & non volatile)
if (I->mayReadFromMemory()) {
LoadInst *Ld = dyn_cast<LoadInst>(I);
if (!Ld || !Ld->isSimple()) {
DEBUG(dbgs() << " Found a non-simple load.\n");
return false;
}
LoadAndStoreCounter++;
collectStridedAccess(Ld);
Value *Ptr = Ld->getPointerOperand();
// Check loop invariant.
if (SE->isLoopInvariant(SE->getSCEV(Ptr), CurLoop))
InvariantCounter++;
}
// If current instruction is store instruction
// make sure it's a simple store (non atomic & non volatile)
else if (I->mayWriteToMemory()) {
StoreInst *St = dyn_cast<StoreInst>(I);
if (!St || !St->isSimple()) {
DEBUG(dbgs() << " Found a non-simple store.\n");
return false;
}
LoadAndStoreCounter++;
collectStridedAccess(St);
Value *Ptr = St->getPointerOperand();
// Check loop invariant.
if (SE->isLoopInvariant(SE->getSCEV(Ptr), CurLoop))
InvariantCounter++;
IsReadOnlyLoop = false;
}
return true;
}
bool Scalarizer::visitLoadInst(LoadInst &LI) {
if (!ScalarizeLoadStore)
return false;
if (!LI.isSimple())
return false;
VectorLayout Layout;
if (!getVectorLayout(LI.getType(), LI.getAlignment(), Layout))
return false;
unsigned NumElems = Layout.VecTy->getNumElements();
IRBuilder<> Builder(LI.getParent(), &LI);
Scatterer Ptr = scatter(&LI, LI.getPointerOperand());
ValueVector Res;
Res.resize(NumElems);
for (unsigned I = 0; I < NumElems; ++I)
Res[I] = Builder.CreateAlignedLoad(Ptr[I], Layout.getElemAlign(I),
LI.getName() + ".i" + Twine(I));
gather(&LI, Res);
return true;
}
/// 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;
}
/// \brief Compute the reduction possible for a given instruction if we are able
/// to SROA an alloca.
///
/// The reduction for this instruction is added to the SROAReduction output
/// parameter. Returns false if this instruction is expected to defeat SROA in
/// general.
static bool countCodeReductionForSROAInst(Instruction *I,
SmallVectorImpl<Value *> &Worklist,
unsigned &SROAReduction) {
if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
if (!LI->isSimple())
return false;
SROAReduction += InlineConstants::InstrCost;
return true;
}
if (StoreInst *SI = dyn_cast<StoreInst>(I)) {
if (!SI->isSimple())
return false;
SROAReduction += InlineConstants::InstrCost;
return true;
}
if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(I)) {
// If the GEP has variable indices, we won't be able to do much with it.
if (!GEP->hasAllConstantIndices())
return false;
// A non-zero GEP will likely become a mask operation after SROA.
if (GEP->hasAllZeroIndices())
SROAReduction += InlineConstants::InstrCost;
Worklist.push_back(GEP);
return true;
}
if (BitCastInst *BCI = dyn_cast<BitCastInst>(I)) {
// Track pointer through bitcasts.
Worklist.push_back(BCI);
SROAReduction += InlineConstants::InstrCost;
return true;
}
// We just look for non-constant operands to ICmp instructions as those will
// defeat SROA. The actual reduction for these happens even without SROA.
if (ICmpInst *ICI = dyn_cast<ICmpInst>(I))
return isa<Constant>(ICI->getOperand(1));
if (SelectInst *SI = dyn_cast<SelectInst>(I)) {
// SROA can handle a select of alloca iff all uses of the alloca are
// loads, and dereferenceable. We assume it's dereferenceable since
// we're told the input is an alloca.
for (Value::use_iterator UI = SI->use_begin(), UE = SI->use_end();
UI != UE; ++UI) {
LoadInst *LI = dyn_cast<LoadInst>(*UI);
if (LI == 0 || !LI->isSimple())
return false;
}
// We don't know whether we'll be deleting the rest of the chain of
// instructions from the SelectInst on, because we don't know whether
// the other side of the select is also an alloca or not.
return true;
}
if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
switch (II->getIntrinsicID()) {
default:
return false;
case Intrinsic::memset:
case Intrinsic::memcpy:
case Intrinsic::memmove:
case Intrinsic::lifetime_start:
case Intrinsic::lifetime_end:
// SROA can usually chew through these intrinsics.
SROAReduction += InlineConstants::InstrCost;
return true;
}
}
// If there is some other strange instruction, we're not going to be
// able to do much if we inline this.
return false;
}
/// Attempt to merge an objc_release with a store, load, and objc_retain to form
/// an objc_storeStrong. This can be a little tricky because the instructions
/// don't always appear in order, and there may be unrelated intervening
/// instructions.
void ObjCARCContract::ContractRelease(Instruction *Release,
inst_iterator &Iter) {
LoadInst *Load = dyn_cast<LoadInst>(GetObjCArg(Release));
if (!Load || !Load->isSimple()) return;
// For now, require everything to be in one basic block.
BasicBlock *BB = Release->getParent();
if (Load->getParent() != BB) return;
// Walk down to find the store and the release, which may be in either order.
BasicBlock::iterator I = Load, End = BB->end();
++I;
AliasAnalysis::Location Loc = AA->getLocation(Load);
StoreInst *Store = 0;
bool SawRelease = false;
for (; !Store || !SawRelease; ++I) {
if (I == End)
return;
Instruction *Inst = I;
if (Inst == Release) {
SawRelease = true;
continue;
}
InstructionClass Class = GetBasicInstructionClass(Inst);
// Unrelated retains are harmless.
if (IsRetain(Class))
continue;
if (Store) {
// The store is the point where we're going to put the objc_storeStrong,
// so make sure there are no uses after it.
if (CanUse(Inst, Load, PA, Class))
return;
} else if (AA->getModRefInfo(Inst, Loc) & AliasAnalysis::Mod) {
// We are moving the load down to the store, so check for anything
// else which writes to the memory between the load and the store.
Store = dyn_cast<StoreInst>(Inst);
if (!Store || !Store->isSimple()) return;
if (Store->getPointerOperand() != Loc.Ptr) return;
}
}
Value *New = StripPointerCastsAndObjCCalls(Store->getValueOperand());
// Walk up to find the retain.
I = Store;
BasicBlock::iterator Begin = BB->begin();
while (I != Begin && GetBasicInstructionClass(I) != IC_Retain)
--I;
Instruction *Retain = I;
if (GetBasicInstructionClass(Retain) != IC_Retain) return;
if (GetObjCArg(Retain) != New) return;
Changed = true;
++NumStoreStrongs;
LLVMContext &C = Release->getContext();
Type *I8X = PointerType::getUnqual(Type::getInt8Ty(C));
Type *I8XX = PointerType::getUnqual(I8X);
Value *Args[] = { Load->getPointerOperand(), New };
if (Args[0]->getType() != I8XX)
Args[0] = new BitCastInst(Args[0], I8XX, "", Store);
if (Args[1]->getType() != I8X)
Args[1] = new BitCastInst(Args[1], I8X, "", Store);
CallInst *StoreStrong =
CallInst::Create(getStoreStrongCallee(BB->getParent()->getParent()),
Args, "", Store);
StoreStrong->setDoesNotThrow();
StoreStrong->setDebugLoc(Store->getDebugLoc());
// We can't set the tail flag yet, because we haven't yet determined
// whether there are any escaping allocas. Remember this call, so that
// we can set the tail flag once we know it's safe.
StoreStrongCalls.insert(StoreStrong);
if (&*Iter == Store) ++Iter;
Store->eraseFromParent();
Release->eraseFromParent();
EraseInstruction(Retain);
if (Load->use_empty())
Load->eraseFromParent();
}
bool runOnFunction(Function &F) override {
AliasAnalysis AA = getAnalysis<AliasAnalysis>();
DependenceAnalysis *DA = &(getAnalysis<DependenceAnalysis>());
// iterate over basic blocks
Function *func = &F;
unsigned bb_num = 0;
for (Function::iterator BB = func->begin(), BE = func->end();
BB != BE; ++BB) {
errs() << "BB-" << bb_num << "\n";
bb_num++;
// iterator over instructions
unsigned inst_num = 0;
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E;++I) {
Instruction *Ins = dyn_cast<Instruction>(I);
if (!Ins)
return false;
LoadInst *Ld = dyn_cast<LoadInst>(I);
StoreInst *St = dyn_cast<StoreInst>(I);
if (!St && !Ld)
continue;
if (Ld && !Ld->isSimple())
return false;
if (St && !St->isSimple())
return false;
inst_num++;
MemInstr.push_back(&*I);
errs() << "MemInst-" << inst_num << ":" << *I << "\n";
}
ValueVector::iterator I, IE, J, JE;
for (I = MemInstr.begin(), IE = MemInstr.end(); I != IE; ++I) {
for (J = I, JE = MemInstr.end(); J != JE; ++J) {
std::vector<char> Dep;
Instruction *Src = dyn_cast<Instruction>(*I);
Instruction *Des = dyn_cast<Instruction>(*J);
if (Src == Des)
continue;
if (isa<LoadInst>(Src) && isa<LoadInst>(Des))
continue;
if (auto D = DA->depends(Src, Des, true)) {
errs() << "Found Dependency between:\nSrc:" << *Src << "\nDes:" << *Des
<< "\n";
if (D->isFlow()) {
errs () << "Flow dependence not handled";
return false;
}
if (D->isAnti()) {
errs() << "Found Anti dependence \n";
AliasAnalysis::AliasResult AA_dep = AA.alias(Src, Des);
AliasAnalysis::AliasResult AA_dep_1 = AA.alias(Des, Src);
errs() << "The Ld->St alias result is " << AA_dep << "\n";
errs() << "The St->Ld alias result is " << AA_dep_1 << "\n";
unsigned Levels = D->getLevels();
errs() << "levels = " << Levels << "\n";
char Direction;
for (unsigned II = 1; II <= Levels; ++II) {
const SCEV *Distance = D->getDistance(II);
const SCEVConstant *SCEVConst = dyn_cast_or_null<SCEVConstant>(Distance);
if (SCEVConst) {
const ConstantInt *CI = SCEVConst->getValue();
//int64_t it_dist = CI->getUniqueInteger().getSExtValue();
//int it_dist = CI->getUniqueInteger().getSExtValue();
unsigned it_dist = abs(CI->getUniqueInteger().getSExtValue());
errs() << "distance is not null\n";
//errs() << "distance = "<< *CI << "\n";
errs() << "distance = "<< it_dist << "\n";
if (CI->isNegative())
Direction = '<';
else if (CI->isZero())
Direction = '=';
else
Direction = '>';
Dep.push_back(Direction);
}
else if (D->isScalar(II)) {
Direction = 'S';
Dep.push_back(Direction);
}
else {
unsigned Dir = D->getDirection(II);
if (Dir == Dependence::DVEntry::LT || Dir == Dependence::DVEntry::LE)
Direction = '<';
else if (Dir == Dependence::DVEntry::GT || Dir == Dependence::DVEntry::GE)
Direction = '>';
else if (Dir == Dependence::DVEntry::EQ)
Direction = '=';
else
Direction = '*';
Dep.push_back(Direction);
}
}
}
}
}
}
}
errs() << "------Hello World!--------\n";
return false;
}
