#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Metadata.h"
+#include "llvm/IR/Module.h"
#include "llvm/IR/Type.h"
#include "llvm/IR/ValueHandle.h"
#include "llvm/Pass.h"
NoMath("bb-vectorize-no-math", cl::init(false), cl::Hidden,
cl::desc("Don't try to vectorize floating-point math intrinsics"));
+static cl::opt<bool>
+ NoBitManipulation("bb-vectorize-no-bitmanip", cl::init(false), cl::Hidden,
+ cl::desc("Don't try to vectorize BitManipulation intrinsics"));
+
static cl::opt<bool>
NoFMA("bb-vectorize-no-fma", cl::init(false), cl::Hidden,
cl::desc("Don't try to vectorize the fused-multiply-add intrinsic"));
initializeBBVectorizePass(*PassRegistry::getPassRegistry());
}
- BBVectorize(Pass *P, const VectorizeConfig &C)
+ BBVectorize(Pass *P, Function &F, const VectorizeConfig &C)
: BasicBlockPass(ID), Config(C) {
AA = &P->getAnalysis<AliasAnalysis>();
DT = &P->getAnalysis<DominatorTreeWrapperPass>().getDomTree();
SE = &P->getAnalysis<ScalarEvolution>();
- DataLayoutPass *DLP = P->getAnalysisIfAvailable<DataLayoutPass>();
- DL = DLP ? &DLP->getDataLayout() : 0;
- TTI = IgnoreTargetInfo ? 0 : &P->getAnalysis<TargetTransformInfo>();
+ TTI = IgnoreTargetInfo
+ ? nullptr
+ : &P->getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F);
}
typedef std::pair<Value *, Value *> ValuePair;
AliasAnalysis *AA;
DominatorTree *DT;
ScalarEvolution *SE;
- const DataLayout *DL;
const TargetTransformInfo *TTI;
// FIXME: const correct?
bool trackUsesOfI(DenseSet<Value *> &Users,
AliasSetTracker &WriteSet, Instruction *I,
Instruction *J, bool UpdateUsers = true,
- DenseSet<ValuePair> *LoadMoveSetPairs = 0);
+ DenseSet<ValuePair> *LoadMoveSetPairs = nullptr);
void computePairsConnectedTo(
DenseMap<Value *, std::vector<Value *> > &CandidatePairs,
bool pairsConflict(ValuePair P, ValuePair Q,
DenseSet<ValuePair> &PairableInstUsers,
DenseMap<ValuePair, std::vector<ValuePair> >
- *PairableInstUserMap = 0,
- DenseSet<VPPair> *PairableInstUserPairSet = 0);
+ *PairableInstUserMap = nullptr,
+ DenseSet<VPPair> *PairableInstUserPairSet = nullptr);
bool pairWillFormCycle(ValuePair P,
DenseMap<ValuePair, std::vector<ValuePair> > &PairableInstUsers,
Instruction *&InsertionPt,
Instruction *I, Instruction *J);
- void combineMetadata(Instruction *K, const Instruction *J);
-
bool vectorizeBB(BasicBlock &BB) {
if (skipOptnoneFunction(BB))
return false;
AA = &getAnalysis<AliasAnalysis>();
DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
SE = &getAnalysis<ScalarEvolution>();
- DataLayoutPass *DLP = getAnalysisIfAvailable<DataLayoutPass>();
- DL = DLP ? &DLP->getDataLayout() : 0;
- TTI = IgnoreTargetInfo ? 0 : &getAnalysis<TargetTransformInfo>();
+ TTI = IgnoreTargetInfo
+ ? nullptr
+ : &getAnalysis<TargetTransformInfoWrapperPass>().getTTI(
+ *BB.getParent());
return vectorizeBB(BB);
}
AU.addRequired<AliasAnalysis>();
AU.addRequired<DominatorTreeWrapperPass>();
AU.addRequired<ScalarEvolution>();
- AU.addRequired<TargetTransformInfo>();
+ AU.addRequired<TargetTransformInfoWrapperPass>();
AU.addPreserved<AliasAnalysis>();
AU.addPreserved<DominatorTreeWrapperPass>();
AU.addPreserved<ScalarEvolution>();
dyn_cast<SCEVConstant>(OffsetSCEV)) {
ConstantInt *IntOff = ConstOffSCEV->getValue();
int64_t Offset = IntOff->getSExtValue();
-
+ const DataLayout &DL = I->getModule()->getDataLayout();
Type *VTy = IPtr->getType()->getPointerElementType();
- int64_t VTyTSS = (int64_t) DL->getTypeStoreSize(VTy);
+ int64_t VTyTSS = (int64_t)DL.getTypeStoreSize(VTy);
Type *VTy2 = JPtr->getType()->getPointerElementType();
if (VTy != VTy2 && Offset < 0) {
- int64_t VTy2TSS = (int64_t) DL->getTypeStoreSize(VTy2);
+ int64_t VTy2TSS = (int64_t)DL.getTypeStoreSize(VTy2);
OffsetInElmts = Offset/VTy2TSS;
- return (abs64(Offset) % VTy2TSS) == 0;
+ return (std::abs(Offset) % VTy2TSS) == 0;
}
OffsetInElmts = Offset/VTyTSS;
- return (abs64(Offset) % VTyTSS) == 0;
+ return (std::abs(Offset) % VTyTSS) == 0;
}
return false;
Function *F = I->getCalledFunction();
if (!F) return false;
- Intrinsic::ID IID = (Intrinsic::ID) F->getIntrinsicID();
+ Intrinsic::ID IID = F->getIntrinsicID();
if (!IID) return false;
switch(IID) {
case Intrinsic::trunc:
case Intrinsic::floor:
case Intrinsic::fabs:
+ case Intrinsic::minnum:
+ case Intrinsic::maxnum:
return Config.VectorizeMath;
+ case Intrinsic::bswap:
+ case Intrinsic::ctpop:
+ case Intrinsic::ctlz:
+ case Intrinsic::cttz:
+ return Config.VectorizeBitManipulations;
case Intrinsic::fma:
case Intrinsic::fmuladd:
return Config.VectorizeFMA;
// It is important to cleanup here so that future iterations of this
// function have less work to do.
- (void) SimplifyInstructionsInBlock(&BB, DL, AA->getTargetLibraryInfo());
+ (void)SimplifyInstructionsInBlock(&BB, AA->getTargetLibraryInfo());
return true;
}
return false;
}
- // We can't vectorize memory operations without target data
- if (DL == 0 && IsSimpleLoadStore)
- return false;
-
Type *T1, *T2;
getInstructionTypes(I, T1, T2);
if (T2->isX86_FP80Ty() || T2->isPPC_FP128Ty() || T2->isX86_MMXTy())
return false;
- if ((!Config.VectorizePointers || DL == 0) &&
- (T1->getScalarType()->isPointerTy() ||
- T2->getScalarType()->isPointerTy()))
+ if (!Config.VectorizePointers && (T1->getScalarType()->isPointerTy() ||
+ T2->getScalarType()->isPointerTy()))
return false;
if (!TTI && (T1->getPrimitiveSizeInBits() >= Config.VectorBits ||
unsigned IAlignment, JAlignment, IAddressSpace, JAddressSpace;
int64_t OffsetInElmts = 0;
if (getPairPtrInfo(I, J, IPtr, JPtr, IAlignment, JAlignment,
- IAddressSpace, JAddressSpace,
- OffsetInElmts) && abs64(OffsetInElmts) == 1) {
+ IAddressSpace, JAddressSpace, OffsetInElmts) &&
+ std::abs(OffsetInElmts) == 1) {
FixedOrder = (int) OffsetInElmts;
unsigned BottomAlignment = IAlignment;
if (OffsetInElmts < 0) BottomAlignment = JAlignment;
// An aligned load or store is possible only if the instruction
// with the lower offset has an alignment suitable for the
// vector type.
-
- unsigned VecAlignment = DL->getPrefTypeAlignment(VType);
+ const DataLayout &DL = I->getModule()->getDataLayout();
+ unsigned VecAlignment = DL.getPrefTypeAlignment(VType);
if (BottomAlignment < VecAlignment)
return false;
}
(isa<ConstantVector>(JOp) || isa<ConstantDataVector>(JOp))) {
Op2VK = TargetTransformInfo::OK_NonUniformConstantValue;
Constant *SplatValue = cast<Constant>(IOp)->getSplatValue();
- if (SplatValue != NULL &&
+ if (SplatValue != nullptr &&
SplatValue == cast<Constant>(JOp)->getSplatValue())
Op2VK = TargetTransformInfo::OK_UniformConstantValue;
}
CostSavings = ICost + JCost - VCost;
}
- // The powi intrinsic is special because only the first argument is
- // vectorized, the second arguments must be equal.
+ // The powi,ctlz,cttz intrinsics are special because only the first
+ // argument is vectorized, the second arguments must be equal.
CallInst *CI = dyn_cast<CallInst>(I);
Function *FI;
if (CI && (FI = CI->getCalledFunction())) {
- Intrinsic::ID IID = (Intrinsic::ID) FI->getIntrinsicID();
- if (IID == Intrinsic::powi) {
+ Intrinsic::ID IID = FI->getIntrinsicID();
+ if (IID == Intrinsic::powi || IID == Intrinsic::ctlz ||
+ IID == Intrinsic::cttz) {
Value *A1I = CI->getArgOperand(1),
*A1J = cast<CallInst>(J)->getArgOperand(1);
const SCEV *A1ISCEV = SE->getSCEV(A1I),
assert(CI->getNumArgOperands() == CJ->getNumArgOperands() &&
"Intrinsic argument counts differ");
for (unsigned i = 0, ie = CI->getNumArgOperands(); i != ie; ++i) {
- if (IID == Intrinsic::powi && i == 1)
+ if ((IID == Intrinsic::powi || IID == Intrinsic::ctlz ||
+ IID == Intrinsic::cttz) && i == 1)
Tys.push_back(CI->getArgOperand(i)->getType());
else
Tys.push_back(getVecTypeForPair(CI->getArgOperand(i)->getType(),
CostSavings, FixedOrder)) continue;
// J is a candidate for merging with I.
- if (!PairableInsts.size() ||
+ if (PairableInsts.empty() ||
PairableInsts[PairableInsts.size()-1] != I) {
PairableInsts.push_back(I);
}
C2->first.second == C->first.first ||
C2->first.second == C->first.second ||
pairsConflict(C2->first, C->first, PairableInstUsers,
- UseCycleCheck ? &PairableInstUserMap : 0,
- UseCycleCheck ? &PairableInstUserPairSet : 0)) {
+ UseCycleCheck ? &PairableInstUserMap : nullptr,
+ UseCycleCheck ? &PairableInstUserPairSet
+ : nullptr)) {
if (C2->second >= C->second) {
CanAdd = false;
break;
T->second == C->first.first ||
T->second == C->first.second ||
pairsConflict(*T, C->first, PairableInstUsers,
- UseCycleCheck ? &PairableInstUserMap : 0,
- UseCycleCheck ? &PairableInstUserPairSet : 0)) {
+ UseCycleCheck ? &PairableInstUserMap : nullptr,
+ UseCycleCheck ? &PairableInstUserPairSet
+ : nullptr)) {
CanAdd = false;
break;
}
C2->first.second == C->first.first ||
C2->first.second == C->first.second ||
pairsConflict(C2->first, C->first, PairableInstUsers,
- UseCycleCheck ? &PairableInstUserMap : 0,
- UseCycleCheck ? &PairableInstUserPairSet : 0)) {
+ UseCycleCheck ? &PairableInstUserMap : nullptr,
+ UseCycleCheck ? &PairableInstUserPairSet
+ : nullptr)) {
CanAdd = false;
break;
}
ChosenPairs.begin(), E2 = ChosenPairs.end();
C2 != E2; ++C2) {
if (pairsConflict(*C2, C->first, PairableInstUsers,
- UseCycleCheck ? &PairableInstUserMap : 0,
- UseCycleCheck ? &PairableInstUserPairSet : 0)) {
+ UseCycleCheck ? &PairableInstUserMap : nullptr,
+ UseCycleCheck ? &PairableInstUserPairSet
+ : nullptr)) {
CanAdd = false;
break;
}
for (DenseMap<Value *, Value *>::iterator C = ChosenPairs.begin(),
E = ChosenPairs.end(); C != E; ++C) {
if (pairsConflict(*C, IJ, PairableInstUsers,
- UseCycleCheck ? &PairableInstUserMap : 0,
- UseCycleCheck ? &PairableInstUserPairSet : 0)) {
+ UseCycleCheck ? &PairableInstUserMap : nullptr,
+ UseCycleCheck ? &PairableInstUserPairSet : nullptr)) {
DoesConflict = true;
break;
}
} while ((LIENext =
dyn_cast<InsertElementInst>(LIENext->getOperand(0))));
- LIENext = 0;
+ LIENext = nullptr;
Value *LIEPrev = UndefValue::get(ArgTypeH);
for (unsigned i = 0; i < numElemL; ++i) {
if (isa<UndefValue>(VectElemts[i])) continue;
if ((LEE || LSV) && (HEE || HSV) && !IsSizeChangeShuffle) {
// We can have at most two unique vector inputs.
bool CanUseInputs = true;
- Value *I1, *I2 = 0;
+ Value *I1, *I2 = nullptr;
if (LEE) {
I1 = LEE->getOperand(0);
} else {
I1 = LSV->getOperand(0);
I2 = LSV->getOperand(1);
if (I2 == I1 || isa<UndefValue>(I2))
- I2 = 0;
+ I2 = nullptr;
}
if (HEE) {
true, o, 1));
NewI1->insertBefore(IBeforeJ ? J : I);
I1 = NewI1;
- I1T = I2T;
I1Elem = I2Elem;
} else if (I1Elem > I2Elem) {
std::vector<Constant *> Mask(I1Elem);
true, o, 1));
NewI2->insertBefore(IBeforeJ ? J : I);
I2 = NewI2;
- I2T = I1T;
- I2Elem = I1Elem;
}
// Now that both I1 and I2 are the same length we can shuffle them
continue;
} else if (isa<CallInst>(I)) {
Function *F = cast<CallInst>(I)->getCalledFunction();
- Intrinsic::ID IID = (Intrinsic::ID) F->getIntrinsicID();
+ Intrinsic::ID IID = F->getIntrinsicID();
if (o == NumOperands-1) {
BasicBlock &BB = *I->getParent();
ReplacedOperands[o] = Intrinsic::getDeclaration(M, IID, VArgType);
continue;
- } else if (IID == Intrinsic::powi && o == 1) {
- // The second argument of powi is a single integer and we've already
- // checked that both arguments are equal. As a result, we just keep
- // I's second argument.
+ } else if ((IID == Intrinsic::powi || IID == Intrinsic::ctlz ||
+ IID == Intrinsic::cttz) && o == 1) {
+ // The second argument of powi/ctlz/cttz is a single integer/constant
+ // and we've already checked that both arguments are equal.
+ // As a result, we just keep I's second argument.
ReplacedOperands[o] = I->getOperand(o);
continue;
}
}
}
- // When the first instruction in each pair is cloned, it will inherit its
- // parent's metadata. This metadata must be combined with that of the other
- // instruction in a safe way.
- void BBVectorize::combineMetadata(Instruction *K, const Instruction *J) {
- SmallVector<std::pair<unsigned, MDNode*>, 4> Metadata;
- K->getAllMetadataOtherThanDebugLoc(Metadata);
- for (unsigned i = 0, n = Metadata.size(); i < n; ++i) {
- unsigned Kind = Metadata[i].first;
- MDNode *JMD = J->getMetadata(Kind);
- MDNode *KMD = Metadata[i].second;
-
- switch (Kind) {
- default:
- K->setMetadata(Kind, 0); // Remove unknown metadata
- break;
- case LLVMContext::MD_tbaa:
- K->setMetadata(Kind, MDNode::getMostGenericTBAA(JMD, KMD));
- break;
- case LLVMContext::MD_fpmath:
- K->setMetadata(Kind, MDNode::getMostGenericFPMath(JMD, KMD));
- break;
- }
- }
- }
-
// This function fuses the chosen instruction pairs into vector instructions,
// taking care preserve any needed scalar outputs and, then, it reorders the
// remaining instructions as needed (users of the first member of the pair
else if (H->hasName())
K->takeName(H);
- if (!isa<StoreInst>(K))
+ if (auto CS = CallSite(K)) {
+ SmallVector<Type *, 3> Tys;
+ FunctionType *Old = CS.getFunctionType();
+ unsigned NumOld = Old->getNumParams();
+ assert(NumOld <= ReplacedOperands.size());
+ for (unsigned i = 0; i != NumOld; ++i)
+ Tys.push_back(ReplacedOperands[i]->getType());
+ CS.mutateFunctionType(
+ FunctionType::get(getVecTypeForPair(L->getType(), H->getType()),
+ Tys, Old->isVarArg()));
+ } else if (!isa<StoreInst>(K))
K->mutateType(getVecTypeForPair(L->getType(), H->getType()));
- combineMetadata(K, H);
+ unsigned KnownIDs[] = {
+ LLVMContext::MD_tbaa,
+ LLVMContext::MD_alias_scope,
+ LLVMContext::MD_noalias,
+ LLVMContext::MD_fpmath
+ };
+ combineMetadata(K, H, KnownIDs);
K->intersectOptionalDataWith(H);
for (unsigned o = 0; o < NumOperands; ++o)
// Instruction insertion point:
Instruction *InsertionPt = K;
- Instruction *K1 = 0, *K2 = 0;
+ Instruction *K1 = nullptr, *K2 = nullptr;
replaceOutputsOfPair(Context, L, H, K, InsertionPt, K1, K2);
// The use dag of the first original instruction must be moved to after
static const char bb_vectorize_name[] = "Basic-Block Vectorization";
INITIALIZE_PASS_BEGIN(BBVectorize, BBV_NAME, bb_vectorize_name, false, false)
INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
-INITIALIZE_AG_DEPENDENCY(TargetTransformInfo)
+INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)
INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)
INITIALIZE_PASS_END(BBVectorize, BBV_NAME, bb_vectorize_name, false, false)
bool
llvm::vectorizeBasicBlock(Pass *P, BasicBlock &BB, const VectorizeConfig &C) {
- BBVectorize BBVectorizer(P, C);
+ BBVectorize BBVectorizer(P, *BB.getParent(), C);
return BBVectorizer.vectorizeBB(BB);
}
VectorizePointers = !::NoPointers;
VectorizeCasts = !::NoCasts;
VectorizeMath = !::NoMath;
+ VectorizeBitManipulations = !::NoBitManipulation;
VectorizeFMA = !::NoFMA;
VectorizeSelect = !::NoSelect;
VectorizeCmp = !::NoCmp;