#include "llvm/Transforms/Vectorize.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/EquivalenceClasses.h"
+#include "llvm/ADT/Hashing.h"
#include "llvm/ADT/MapVector.h"
+#include "llvm/ADT/SetVector.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/SmallVector.h"
/// Maximum vectorization unroll count.
static const unsigned MaxUnrollFactor = 16;
+/// The cost of a loop that is considered 'small' by the unroller.
+static const unsigned SmallLoopCost = 20;
+
namespace {
// Forward declarations.
updateAnalysis();
}
-private:
+ virtual ~InnerLoopVectorizer() {}
+
+protected:
/// A small list of PHINodes.
typedef SmallVector<PHINode*, 4> PhiVector;
/// When we unroll loops we have multiple vector values for each scalar.
/// originated from one scalar instruction.
typedef SmallVector<Value*, 2> VectorParts;
+ // When we if-convert we need create edge masks. We have to cache values so
+ // that we don't end up with exponential recursion/IR.
+ typedef DenseMap<std::pair<BasicBlock*, BasicBlock*>,
+ VectorParts> EdgeMaskCache;
+
/// Add code that checks at runtime if the accessed arrays overlap.
/// Returns the comparator value or NULL if no check is needed.
Instruction *addRuntimeCheck(LoopVectorizationLegality *Legal,
/// Create an empty loop, based on the loop ranges of the old loop.
void createEmptyLoop(LoopVectorizationLegality *Legal);
/// Copy and widen the instructions from the old loop.
- void vectorizeLoop(LoopVectorizationLegality *Legal);
+ virtual void vectorizeLoop(LoopVectorizationLegality *Legal);
+
+ /// \brief The Loop exit block may have single value PHI nodes where the
+ /// incoming value is 'Undef'. While vectorizing we only handled real values
+ /// that were defined inside the loop. Here we fix the 'undef case'.
+ /// See PR14725.
+ void fixLCSSAPHIs();
/// A helper function that computes the predicate of the block BB, assuming
/// that the header block of the loop is set to True. It returns the *entry*
void vectorizeBlockInLoop(LoopVectorizationLegality *Legal, BasicBlock *BB,
PhiVector *PV);
+ /// Vectorize a single PHINode in a block. This method handles the induction
+ /// variable canonicalization. It supports both VF = 1 for unrolled loops and
+ /// arbitrary length vectors.
+ void widenPHIInstruction(Instruction *PN, VectorParts &Entry,
+ LoopVectorizationLegality *Legal,
+ unsigned UF, unsigned VF, PhiVector *PV);
+
/// Insert the new loop to the loop hierarchy and pass manager
/// and update the analysis passes.
void updateAnalysis();
/// This instruction is un-vectorizable. Implement it as a sequence
/// of scalars.
- void scalarizeInstruction(Instruction *Instr);
+ virtual void scalarizeInstruction(Instruction *Instr);
/// Vectorize Load and Store instructions,
- void vectorizeMemoryInstruction(Instruction *Instr,
+ virtual void vectorizeMemoryInstruction(Instruction *Instr,
LoopVectorizationLegality *Legal);
/// Create a broadcast instruction. This method generates a broadcast
/// value. If this is the induction variable then we extend it to N, N+1, ...
/// this is needed because each iteration in the loop corresponds to a SIMD
/// element.
- Value *getBroadcastInstrs(Value *V);
+ virtual Value *getBroadcastInstrs(Value *V);
/// This function adds 0, 1, 2 ... to each vector element, starting at zero.
/// If Negate is set then negative numbers are added e.g. (0, -1, -2, ...).
/// The sequence starts at StartIndex.
- Value *getConsecutiveVector(Value* Val, int StartIdx, bool Negate);
+ virtual Value *getConsecutiveVector(Value* Val, int StartIdx, bool Negate);
/// When we go over instructions in the basic block we rely on previous
/// values within the current basic block or on loop invariant values.
VectorParts &getVectorValue(Value *V);
/// Generate a shuffle sequence that will reverse the vector Vec.
- Value *reverseVector(Value *Vec);
+ virtual Value *reverseVector(Value *Vec);
/// This is a helper class that holds the vectorizer state. It maps scalar
/// instructions to vector instructions. When the code is 'unrolled' then
/// The vectorization SIMD factor to use. Each vector will have this many
/// vector elements.
unsigned VF;
+
+protected:
/// The vectorization unroll factor to use. Each scalar is vectorized to this
/// many different vector instructions.
unsigned UF;
Value *ExtendedIdx;
/// Maps scalars to widened vectors.
ValueMap WidenMap;
+ EdgeMaskCache MaskCache;
};
-/// \brief Check if conditionally executed loads are hoistable.
-///
-/// This class has two functions: isHoistableLoad and canHoistAllLoads.
-/// isHoistableLoad should be called on all load instructions that are executed
-/// conditionally. After all conditional loads are processed, the client should
-/// call canHoistAllLoads to determine if all of the conditional executed loads
-/// have an unconditional memory access to the same memory address in the loop.
-class LoadHoisting {
- typedef SmallPtrSet<Value *, 8> MemorySet;
-
- Loop *TheLoop;
- DominatorTree *DT;
- MemorySet CondLoadAddrSet;
-
+class InnerLoopUnroller : public InnerLoopVectorizer {
public:
- LoadHoisting(Loop *L, DominatorTree *D) : TheLoop(L), DT(D) {}
-
- /// \brief Check if the instruction is a load with a identifiable address.
- bool isHoistableLoad(Instruction *L);
+ InnerLoopUnroller(Loop *OrigLoop, ScalarEvolution *SE, LoopInfo *LI,
+ DominatorTree *DT, DataLayout *DL,
+ const TargetLibraryInfo *TLI, unsigned UnrollFactor) :
+ InnerLoopVectorizer(OrigLoop, SE, LI, DT, DL, TLI, 1, UnrollFactor) { }
- /// \brief Check if all of the conditional loads are hoistable because there
- /// exists an unconditional memory access to the same address in the loop.
- bool canHoistAllLoads();
+private:
+ virtual void scalarizeInstruction(Instruction *Instr);
+ virtual void vectorizeMemoryInstruction(Instruction *Instr,
+ LoopVectorizationLegality *Legal);
+ virtual Value *getBroadcastInstrs(Value *V);
+ virtual Value *getConsecutiveVector(Value* Val, int StartIdx, bool Negate);
+ virtual Value *reverseVector(Value *Vec);
};
-bool LoadHoisting::isHoistableLoad(Instruction *L) {
- LoadInst *LI = dyn_cast<LoadInst>(L);
- if (!LI)
- return false;
+/// \brief Look for a meaningful debug location on the instruction or it's
+/// operands.
+static Instruction *getDebugLocFromInstOrOperands(Instruction *I) {
+ if (!I)
+ return I;
- CondLoadAddrSet.insert(LI->getPointerOperand());
- return true;
-}
+ DebugLoc Empty;
+ if (I->getDebugLoc() != Empty)
+ return I;
-static void addMemAccesses(BasicBlock *BB, SmallPtrSet<Value *, 8> &Set) {
- for (BasicBlock::iterator BI = BB->begin(), BE = BB->end(); BI != BE; ++BI) {
- if (LoadInst *LI = dyn_cast<LoadInst>(BI)) // Try a load.
- Set.insert(LI->getPointerOperand());
- else if (StoreInst *SI = dyn_cast<StoreInst>(BI)) // Try a store.
- Set.insert(SI->getPointerOperand());
+ for (User::op_iterator OI = I->op_begin(), OE = I->op_end(); OI != OE; ++OI) {
+ if (Instruction *OpInst = dyn_cast<Instruction>(*OI))
+ if (OpInst->getDebugLoc() != Empty)
+ return OpInst;
}
-}
-
-bool LoadHoisting::canHoistAllLoads() {
- // No conditional loads.
- if (CondLoadAddrSet.empty())
- return true;
-
- MemorySet UncondMemAccesses;
- std::vector<BasicBlock*> &LoopBlocks = TheLoop->getBlocksVector();
- BasicBlock *LoopLatch = TheLoop->getLoopLatch();
-
- // Iterate over the unconditional blocks and collect memory access addresses.
- for (unsigned i = 0, e = LoopBlocks.size(); i < e; ++i) {
- BasicBlock *BB = LoopBlocks[i];
- // Ignore conditional blocks.
- if (BB != LoopLatch && !DT->dominates(BB, LoopLatch))
- continue;
-
- addMemAccesses(BB, UncondMemAccesses);
- }
-
- // And make sure there is a matching unconditional access for every
- // conditional load.
- for (MemorySet::iterator MI = CondLoadAddrSet.begin(),
- ME = CondLoadAddrSet.end(); MI != ME; ++MI)
- if (!UncondMemAccesses.count(*MI))
- return false;
+ return I;
+}
- return true;
+/// \brief Set the debug location in the builder using the debug location in the
+/// instruction.
+static void setDebugLocFromInst(IRBuilder<> &B, const Value *Ptr) {
+ if (const Instruction *Inst = dyn_cast_or_null<Instruction>(Ptr))
+ B.SetCurrentDebugLocation(Inst->getDebugLoc());
+ else
+ B.SetCurrentDebugLocation(DebugLoc());
}
/// LoopVectorizationLegality checks if it is legal to vectorize a loop, and
DominatorTree *DT, TargetLibraryInfo *TLI)
: TheLoop(L), SE(SE), DL(DL), DT(DT), TLI(TLI),
Induction(0), WidestIndTy(0), HasFunNoNaNAttr(false),
- MaxSafeDepDistBytes(-1U), LoadSpeculation(L, DT) {}
+ MaxSafeDepDistBytes(-1U) {}
/// This enum represents the kinds of reductions that we support.
enum ReductionKind {
MRK_FloatMax
};
- /// This POD struct holds information about reduction variables.
+ /// This struct holds information about reduction variables.
struct ReductionDescriptor {
ReductionDescriptor() : StartValue(0), LoopExitInstr(0),
Kind(RK_NoReduction), MinMaxKind(MRK_Invalid) {}
MinMaxReductionKind MinMaxKind;
};
- // This POD struct holds information about the memory runtime legality
- // check that a group of pointers do not overlap.
+ /// This struct holds information about the memory runtime legality
+ /// check that a group of pointers do not overlap.
struct RuntimePointerCheck {
RuntimePointerCheck() : Need(false) {}
Pointers.clear();
Starts.clear();
Ends.clear();
+ IsWritePtr.clear();
+ DependencySetId.clear();
}
/// Insert a pointer and calculate the start and end SCEVs.
SmallVector<unsigned, 2> DependencySetId;
};
- /// A POD for saving information about induction variables.
+ /// A struct for saving information about induction variables.
struct InductionInfo {
InductionInfo(Value *Start, InductionKind K) : StartValue(Start), IK(K) {}
InductionInfo() : StartValue(0), IK(IK_NoInduction) {}
/// pointer itself is an induction variable.
/// This check allows us to vectorize A[idx] into a wide load/store.
/// Returns:
- /// 0 - Stride is unknown or non consecutive.
+ /// 0 - Stride is unknown or non-consecutive.
/// 1 - Address is consecutive.
/// -1 - Address is consecutive, and decreasing.
int isConsecutivePtr(Value *Ptr);
void collectLoopUniforms();
/// Return true if all of the instructions in the block can be speculatively
- /// executed.
- bool blockCanBePredicated(BasicBlock *BB);
+ /// executed. \p SafePtrs is a list of addresses that are known to be legal
+ /// and we know that we can read from them without segfault.
+ bool blockCanBePredicated(BasicBlock *BB, SmallPtrSet<Value *, 8>& SafePtrs);
/// Returns True, if 'Phi' is the kind of reduction variable for type
/// 'Kind'. If this is a reduction variable, it adds it to ReductionList.
bool HasFunNoNaNAttr;
unsigned MaxSafeDepDistBytes;
-
- /// Utility to determine whether loads can be speculated.
- LoadHoisting LoadSpeculation;
};
/// LoopVectorizationCostModel - estimates the expected speedups due to
unsigned Width;
/// Vectorization unroll factor.
unsigned Unroll;
+ /// Vectorization forced (-1 not selected, 0 force disabled, 1 force enabled)
+ int Force;
- LoopVectorizeHints(const Loop *L)
+ LoopVectorizeHints(const Loop *L, bool DisableUnrolling)
: Width(VectorizationFactor)
- , Unroll(VectorizationUnroll)
+ , Unroll(DisableUnrolling ? 1 : VectorizationUnroll)
+ , Force(-1)
, LoopID(L->getLoopID()) {
getHints(L);
// The command line options override any loop metadata except for when
Width = VectorizationFactor;
if (VectorizationUnroll.getNumOccurrences() > 0)
Unroll = VectorizationUnroll;
+
+ DEBUG(if (DisableUnrolling && Unroll == 1)
+ dbgs() << "LV: Unrolling disabled by the pass manager\n");
}
/// Return the loop vectorizer metadata prefix.
Vals.push_back(LoopID->getOperand(i));
Vals.push_back(createHint(Context, Twine(Prefix(), "width").str(), Width));
+ Vals.push_back(createHint(Context, Twine(Prefix(), "unroll").str(), 1));
MDNode *NewLoopID = MDNode::get(Context, Vals);
// Set operand 0 to refer to the loop id itself.
unsigned Val = C->getZExtValue();
if (Hint == "width") {
- assert(isPowerOf2_32(Val) && Val <= MaxVectorWidth &&
- "Invalid width metadata");
- Width = Val;
+ if (isPowerOf2_32(Val) && Val <= MaxVectorWidth)
+ Width = Val;
+ else
+ DEBUG(dbgs() << "LV: ignoring invalid width hint metadata\n");
} else if (Hint == "unroll") {
- assert(isPowerOf2_32(Val) && Val <= MaxUnrollFactor &&
- "Invalid unroll metadata");
- Unroll = Val;
- } else
- DEBUG(dbgs() << "LV: ignoring unknown hint " << Hint);
+ if (isPowerOf2_32(Val) && Val <= MaxUnrollFactor)
+ Unroll = Val;
+ else
+ DEBUG(dbgs() << "LV: ignoring invalid unroll hint metadata\n");
+ } else if (Hint == "enable") {
+ if (C->getBitWidth() == 1)
+ Force = Val;
+ else
+ DEBUG(dbgs() << "LV: ignoring invalid enable hint metadata\n");
+ } else {
+ DEBUG(dbgs() << "LV: ignoring unknown hint " << Hint << '\n');
+ }
}
};
/// Pass identification, replacement for typeid
static char ID;
- explicit LoopVectorize() : LoopPass(ID) {
+ explicit LoopVectorize(bool NoUnrolling = false, bool AlwaysVectorize = true)
+ : LoopPass(ID),
+ DisableUnrolling(NoUnrolling),
+ AlwaysVectorize(AlwaysVectorize) {
initializeLoopVectorizePass(*PassRegistry::getPassRegistry());
}
TargetTransformInfo *TTI;
DominatorTree *DT;
TargetLibraryInfo *TLI;
+ bool DisableUnrolling;
+ bool AlwaysVectorize;
virtual bool runOnLoop(Loop *L, LPPassManager &LPM) {
// We only vectorize innermost loops.
DT = &getAnalysis<DominatorTree>();
TLI = getAnalysisIfAvailable<TargetLibraryInfo>();
+ // If the target claims to have no vector registers don't attempt
+ // vectorization.
+ if (!TTI->getNumberOfRegisters(true))
+ return false;
+
if (DL == NULL) {
- DEBUG(dbgs() << "LV: Not vectorizing because of missing data layout");
+ DEBUG(dbgs() << "LV: Not vectorizing: Missing data layout\n");
return false;
}
DEBUG(dbgs() << "LV: Checking a loop in \"" <<
L->getHeader()->getParent()->getName() << "\"\n");
- LoopVectorizeHints Hints(L);
+ LoopVectorizeHints Hints(L, DisableUnrolling);
+
+ if (Hints.Force == 0) {
+ DEBUG(dbgs() << "LV: Not vectorizing: #pragma vectorize disable.\n");
+ return false;
+ }
+
+ if (!AlwaysVectorize && Hints.Force != 1) {
+ DEBUG(dbgs() << "LV: Not vectorizing: No #pragma vectorize enable.\n");
+ return false;
+ }
- if (Hints.Width == 1) {
- DEBUG(dbgs() << "LV: Not vectorizing.\n");
+ if (Hints.Width == 1 && Hints.Unroll == 1) {
+ DEBUG(dbgs() << "LV: Not vectorizing: Disabled/already vectorized.\n");
return false;
}
// Check if it is legal to vectorize the loop.
LoopVectorizationLegality LVL(L, SE, DL, DT, TLI);
if (!LVL.canVectorize()) {
- DEBUG(dbgs() << "LV: Not vectorizing.\n");
+ DEBUG(dbgs() << "LV: Not vectorizing: Cannot prove legality.\n");
return false;
}
Attribute::AttrKind SzAttr = Attribute::OptimizeForSize;
Attribute::AttrKind FlAttr = Attribute::NoImplicitFloat;
unsigned FnIndex = AttributeSet::FunctionIndex;
- bool OptForSize = F->getAttributes().hasAttribute(FnIndex, SzAttr);
+ bool OptForSize = Hints.Force != 1 &&
+ F->getAttributes().hasAttribute(FnIndex, SzAttr);
bool NoFloat = F->getAttributes().hasAttribute(FnIndex, FlAttr);
if (NoFloat) {
unsigned UF = CM.selectUnrollFactor(OptForSize, Hints.Unroll, VF.Width,
VF.Cost);
+ DEBUG(dbgs() << "LV: Found a vectorizable loop ("<< VF.Width << ") in "<<
+ F->getParent()->getModuleIdentifier() << '\n');
+ DEBUG(dbgs() << "LV: Unroll Factor is " << UF << '\n');
+
if (VF.Width == 1) {
DEBUG(dbgs() << "LV: Vectorization is possible but not beneficial.\n");
- return false;
+ if (UF == 1)
+ return false;
+ DEBUG(dbgs() << "LV: Trying to at least unroll the loops.\n");
+ // We decided not to vectorize, but we may want to unroll.
+ InnerLoopUnroller Unroller(L, SE, LI, DT, DL, TLI, UF);
+ Unroller.vectorize(&LVL);
+ } else {
+ // If we decided that it is *legal* to vectorize the loop then do it.
+ InnerLoopVectorizer LB(L, SE, LI, DT, DL, TLI, VF.Width, UF);
+ LB.vectorize(&LVL);
}
- DEBUG(dbgs() << "LV: Found a vectorizable loop ("<< VF.Width << ") in "<<
- F->getParent()->getModuleIdentifier()<<"\n");
- DEBUG(dbgs() << "LV: Unroll Factor is " << UF << "\n");
-
- // If we decided that it is *legal* to vectorize the loop then do it.
- InnerLoopVectorizer LB(L, SE, LI, DT, DL, TLI, VF.Width, UF);
- LB.vectorize(&LVL);
-
// Mark the loop as already vectorized to avoid vectorizing again.
Hints.setAlreadyVectorized(L);
}
Value *InnerLoopVectorizer::getBroadcastInstrs(Value *V) {
- // Save the current insertion location.
- Instruction *Loc = Builder.GetInsertPoint();
-
// We need to place the broadcast of invariant variables outside the loop.
Instruction *Instr = dyn_cast<Instruction>(V);
bool NewInstr = (Instr && Instr->getParent() == LoopVectorBody);
bool Invariant = OrigLoop->isLoopInvariant(V) && !NewInstr;
// Place the code for broadcasting invariant variables in the new preheader.
+ IRBuilder<>::InsertPointGuard Guard(Builder);
if (Invariant)
Builder.SetInsertPoint(LoopVectorPreHeader->getTerminator());
// Broadcast the scalar into all locations in the vector.
Value *Shuf = Builder.CreateVectorSplat(VF, V, "broadcast");
- // Restore the builder insertion point.
- if (Invariant)
- Builder.SetInsertPoint(Loc);
-
return Shuf;
}
return Builder.CreateAdd(Val, Cv, "induction");
}
+/// \brief Find the operand of the GEP that should be checked for consecutive
+/// stores. This ignores trailing indices that have no effect on the final
+/// pointer.
+static unsigned getGEPInductionOperand(DataLayout *DL,
+ const GetElementPtrInst *Gep) {
+ unsigned LastOperand = Gep->getNumOperands() - 1;
+ unsigned GEPAllocSize = DL->getTypeAllocSize(
+ cast<PointerType>(Gep->getType()->getScalarType())->getElementType());
+
+ // Walk backwards and try to peel off zeros.
+ while (LastOperand > 1 && match(Gep->getOperand(LastOperand), m_Zero())) {
+ // Find the type we're currently indexing into.
+ gep_type_iterator GEPTI = gep_type_begin(Gep);
+ std::advance(GEPTI, LastOperand - 1);
+
+ // If it's a type with the same allocation size as the result of the GEP we
+ // can peel off the zero index.
+ if (DL->getTypeAllocSize(*GEPTI) != GEPAllocSize)
+ break;
+ --LastOperand;
+ }
+
+ return LastOperand;
+}
+
int LoopVectorizationLegality::isConsecutivePtr(Value *Ptr) {
- assert(Ptr->getType()->isPointerTy() && "Unexpected non ptr");
+ assert(Ptr->getType()->isPointerTy() && "Unexpected non-ptr");
// Make sure that the pointer does not point to structs.
- if (cast<PointerType>(Ptr->getType())->getElementType()->isAggregateType())
+ if (Ptr->getType()->getPointerElementType()->isAggregateType())
return 0;
// If this value is a pointer induction variable we know it is consecutive.
return 0;
unsigned NumOperands = Gep->getNumOperands();
- Value *LastIndex = Gep->getOperand(NumOperands - 1);
-
Value *GpPtr = Gep->getPointerOperand();
// If this GEP value is a consecutive pointer induction variable and all of
// the indices are constant then we know it is consecutive. We can
return -1;
}
- // Check that all of the gep indices are uniform except for the last.
- for (unsigned i = 0; i < NumOperands - 1; ++i)
- if (!SE->isLoopInvariant(SE->getSCEV(Gep->getOperand(i)), TheLoop))
+ unsigned InductionOperand = getGEPInductionOperand(DL, Gep);
+
+ // Check that all of the gep indices are uniform except for our induction
+ // operand.
+ for (unsigned i = 0; i != NumOperands; ++i)
+ if (i != InductionOperand &&
+ !SE->isLoopInvariant(SE->getSCEV(Gep->getOperand(i)), TheLoop))
return 0;
- // We can emit wide load/stores only if the last index is the induction
- // variable.
- const SCEV *Last = SE->getSCEV(LastIndex);
+ // We can emit wide load/stores only if the last non-zero index is the
+ // induction variable.
+ const SCEV *Last = SE->getSCEV(Gep->getOperand(InductionOperand));
if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(Last)) {
const SCEV *Step = AR->getStepRecurrence(*SE);
Type *DataTy = VectorType::get(ScalarDataTy, VF);
Value *Ptr = LI ? LI->getPointerOperand() : SI->getPointerOperand();
unsigned Alignment = LI ? LI->getAlignment() : SI->getAlignment();
+ // An alignment of 0 means target abi alignment. We need to use the scalar's
+ // target abi alignment in such a case.
+ if (!Alignment)
+ Alignment = DL->getABITypeAlignment(ScalarDataTy);
unsigned AddressSpace = Ptr->getType()->getPointerAddressSpace();
unsigned ScalarAllocatedSize = DL->getTypeAllocSize(ScalarDataTy);
unsigned VectorElementSize = DL->getTypeStoreSize(DataTy)/VF;
if (ScalarAllocatedSize != VectorElementSize)
return scalarizeInstruction(Instr);
- // If the pointer is loop invariant or if it is non consecutive,
+ // If the pointer is loop invariant or if it is non-consecutive,
// scalarize the load.
int ConsecutiveStride = Legal->isConsecutivePtr(Ptr);
bool Reverse = ConsecutiveStride < 0;
// Handle consecutive loads/stores.
GetElementPtrInst *Gep = dyn_cast<GetElementPtrInst>(Ptr);
if (Gep && Legal->isInductionVariable(Gep->getPointerOperand())) {
+ setDebugLocFromInst(Builder, Gep);
Value *PtrOperand = Gep->getPointerOperand();
Value *FirstBasePtr = getVectorValue(PtrOperand)[0];
FirstBasePtr = Builder.CreateExtractElement(FirstBasePtr, Zero);
Gep2->setName("gep.indvar.base");
Ptr = Builder.Insert(Gep2);
} else if (Gep) {
+ setDebugLocFromInst(Builder, Gep);
assert(SE->isLoopInvariant(SE->getSCEV(Gep->getPointerOperand()),
OrigLoop) && "Base ptr must be invariant");
// The last index does not have to be the induction. It can be
// consecutive and be a function of the index. For example A[I+1];
unsigned NumOperands = Gep->getNumOperands();
-
- Value *LastGepOperand = Gep->getOperand(NumOperands - 1);
- VectorParts &GEPParts = getVectorValue(LastGepOperand);
- Value *LastIndex = GEPParts[0];
- LastIndex = Builder.CreateExtractElement(LastIndex, Zero);
-
+ unsigned InductionOperand = getGEPInductionOperand(DL, Gep);
// Create the new GEP with the new induction variable.
GetElementPtrInst *Gep2 = cast<GetElementPtrInst>(Gep->clone());
- Gep2->setOperand(NumOperands - 1, LastIndex);
- Gep2->setName("gep.indvar.idx");
+
+ for (unsigned i = 0; i < NumOperands; ++i) {
+ Value *GepOperand = Gep->getOperand(i);
+ Instruction *GepOperandInst = dyn_cast<Instruction>(GepOperand);
+
+ // Update last index or loop invariant instruction anchored in loop.
+ if (i == InductionOperand ||
+ (GepOperandInst && OrigLoop->contains(GepOperandInst))) {
+ assert((i == InductionOperand ||
+ SE->isLoopInvariant(SE->getSCEV(GepOperandInst), OrigLoop)) &&
+ "Must be last index or loop invariant");
+
+ VectorParts &GEPParts = getVectorValue(GepOperand);
+ Value *Index = GEPParts[0];
+ Index = Builder.CreateExtractElement(Index, Zero);
+ Gep2->setOperand(i, Index);
+ Gep2->setName("gep.indvar.idx");
+ }
+ }
Ptr = Builder.Insert(Gep2);
} else {
// Use the induction element ptr.
assert(isa<PHINode>(Ptr) && "Invalid induction ptr");
+ setDebugLocFromInst(Builder, Ptr);
VectorParts &PtrVal = getVectorValue(Ptr);
Ptr = Builder.CreateExtractElement(PtrVal[0], Zero);
}
if (SI) {
assert(!Legal->isUniform(SI->getPointerOperand()) &&
"We do not allow storing to uniform addresses");
+ setDebugLocFromInst(Builder, SI);
+ // We don't want to update the value in the map as it might be used in
+ // another expression. So don't use a reference type for "StoredVal".
+ VectorParts StoredVal = getVectorValue(SI->getValueOperand());
- VectorParts &StoredVal = getVectorValue(SI->getValueOperand());
for (unsigned Part = 0; Part < UF; ++Part) {
// Calculate the pointer for the specific unroll-part.
Value *PartPtr = Builder.CreateGEP(Ptr, Builder.getInt32(Part * VF));
PartPtr = Builder.CreateGEP(PartPtr, Builder.getInt32(1 - VF));
}
- Value *VecPtr = Builder.CreateBitCast(PartPtr, DataTy->getPointerTo(AddressSpace));
+ Value *VecPtr = Builder.CreateBitCast(PartPtr,
+ DataTy->getPointerTo(AddressSpace));
Builder.CreateStore(StoredVal[Part], VecPtr)->setAlignment(Alignment);
}
+ return;
}
+ // Handle loads.
+ assert(LI && "Must have a load instruction");
+ setDebugLocFromInst(Builder, LI);
for (unsigned Part = 0; Part < UF; ++Part) {
// Calculate the pointer for the specific unroll-part.
Value *PartPtr = Builder.CreateGEP(Ptr, Builder.getInt32(Part * VF));
PartPtr = Builder.CreateGEP(PartPtr, Builder.getInt32(1 - VF));
}
- Value *VecPtr = Builder.CreateBitCast(PartPtr, DataTy->getPointerTo(AddressSpace));
+ Value *VecPtr = Builder.CreateBitCast(PartPtr,
+ DataTy->getPointerTo(AddressSpace));
Value *LI = Builder.CreateLoad(VecPtr, "wide.load");
cast<LoadInst>(LI)->setAlignment(Alignment);
Entry[Part] = Reverse ? reverseVector(LI) : LI;
// Holds vector parameters or scalars, in case of uniform vals.
SmallVector<VectorParts, 4> Params;
+ setDebugLocFromInst(Builder, Instr);
+
// Find all of the vectorized parameters.
for (unsigned op = 0, e = Instr->getNumOperands(); op != e; ++op) {
Value *SrcOp = Instr->getOperand(op);
Instruction *Cloned = Instr->clone();
if (!IsVoidRetTy)
Cloned->setName(Instr->getName() + ".cloned");
- // Replace the operands of the cloned instrucions with extracted scalars.
+ // Replace the operands of the cloned instructions with extracted scalars.
for (unsigned op = 0, e = Instr->getNumOperands(); op != e; ++op) {
Value *Op = Params[op][Part];
// Param is a vector. Need to extract the right lane.
SmallVector<TrackingVH<Value> , 2> Starts;
SmallVector<TrackingVH<Value> , 2> Ends;
+ LLVMContext &Ctx = Loc->getContext();
SCEVExpander Exp(*SE, "induction");
- // Use this type for pointer arithmetic.
- Type* PtrArithTy = Type::getInt8PtrTy(Loc->getContext(), 0);
-
for (unsigned i = 0; i < NumPointers; ++i) {
Value *Ptr = PtrRtCheck->Pointers[i];
const SCEV *Sc = SE->getSCEV(Ptr);
Starts.push_back(Ptr);
Ends.push_back(Ptr);
} else {
- DEBUG(dbgs() << "LV: Adding RT check for range:" << *Ptr <<"\n");
+ DEBUG(dbgs() << "LV: Adding RT check for range:" << *Ptr << '\n');
+ unsigned AS = Ptr->getType()->getPointerAddressSpace();
+
+ // Use this type for pointer arithmetic.
+ Type *PtrArithTy = Type::getInt8PtrTy(Ctx, AS);
Value *Start = Exp.expandCodeFor(PtrRtCheck->Starts[i], PtrArithTy, Loc);
Value *End = Exp.expandCodeFor(PtrRtCheck->Ends[i], PtrArithTy, Loc);
if (PtrRtCheck->DependencySetId[i] == PtrRtCheck->DependencySetId[j])
continue;
- Value *Start0 = ChkBuilder.CreateBitCast(Starts[i], PtrArithTy, "bc");
- Value *Start1 = ChkBuilder.CreateBitCast(Starts[j], PtrArithTy, "bc");
- Value *End0 = ChkBuilder.CreateBitCast(Ends[i], PtrArithTy, "bc");
- Value *End1 = ChkBuilder.CreateBitCast(Ends[j], PtrArithTy, "bc");
+ unsigned AS0 = Starts[i]->getType()->getPointerAddressSpace();
+ unsigned AS1 = Starts[j]->getType()->getPointerAddressSpace();
+
+ assert((AS0 == Ends[j]->getType()->getPointerAddressSpace()) &&
+ (AS1 == Ends[i]->getType()->getPointerAddressSpace()) &&
+ "Trying to bounds check pointers with different address spaces");
+
+ Type *PtrArithTy0 = Type::getInt8PtrTy(Ctx, AS0);
+ Type *PtrArithTy1 = Type::getInt8PtrTy(Ctx, AS1);
+
+ Value *Start0 = ChkBuilder.CreateBitCast(Starts[i], PtrArithTy0, "bc");
+ Value *Start1 = ChkBuilder.CreateBitCast(Starts[j], PtrArithTy1, "bc");
+ Value *End0 = ChkBuilder.CreateBitCast(Ends[i], PtrArithTy1, "bc");
+ Value *End1 = ChkBuilder.CreateBitCast(Ends[j], PtrArithTy0, "bc");
Value *Cmp0 = ChkBuilder.CreateICmpULE(Start0, End1, "bound0");
Value *Cmp1 = ChkBuilder.CreateICmpULE(Start1, End0, "bound1");
// We have to do this trickery because the IRBuilder might fold the check to a
// constant expression in which case there is no Instruction anchored in a
// the block.
- LLVMContext &Ctx = Loc->getContext();
- Instruction * Check = BinaryOperator::CreateAnd(MemoryRuntimeCheck,
- ConstantInt::getTrue(Ctx));
+ Instruction *Check = BinaryOperator::CreateAnd(MemoryRuntimeCheck,
+ ConstantInt::getTrue(Ctx));
ChkBuilder.Insert(Check, "memcheck.conflict");
return Check;
}
const SCEV *ExitCount = SE->getBackedgeTakenCount(OrigLoop);
assert(ExitCount != SE->getCouldNotCompute() && "Invalid loop count");
+ // The exit count might have the type of i64 while the phi is i32. This can
+ // happen if we have an induction variable that is sign extended before the
+ // compare. The only way that we get a backedge taken count is that the
+ // induction variable was signed and as such will not overflow. In such a case
+ // truncation is legal.
+ if (ExitCount->getType()->getPrimitiveSizeInBits() >
+ IdxTy->getPrimitiveSizeInBits())
+ ExitCount = SE->getTruncateOrNoop(ExitCount, IdxTy);
+
+ ExitCount = SE->getNoopOrZeroExtend(ExitCount, IdxTy);
// Get the total trip count from the count by adding 1.
ExitCount = SE->getAddExpr(ExitCount,
SE->getConstant(ExitCount->getType(), 1));
BasicBlock *ScalarPH =
MiddleBlock->splitBasicBlock(MiddleBlock->getTerminator(), "scalar.ph");
+ // Create and register the new vector loop.
+ Loop* Lp = new Loop();
+ Loop *ParentLoop = OrigLoop->getParentLoop();
+
+ // Insert the new loop into the loop nest and register the new basic blocks
+ // before calling any utilities such as SCEV that require valid LoopInfo.
+ if (ParentLoop) {
+ ParentLoop->addChildLoop(Lp);
+ ParentLoop->addBasicBlockToLoop(ScalarPH, LI->getBase());
+ ParentLoop->addBasicBlockToLoop(VectorPH, LI->getBase());
+ ParentLoop->addBasicBlockToLoop(MiddleBlock, LI->getBase());
+ } else {
+ LI->addTopLevelLoop(Lp);
+ }
+ Lp->addBasicBlockToLoop(VecBody, LI->getBase());
+
// Use this IR builder to create the loop instructions (Phi, Br, Cmp)
// inside the loop.
- Builder.SetInsertPoint(VecBody->getFirstInsertionPt());
+ Builder.SetInsertPoint(VecBody->getFirstNonPHI());
// Generate the induction variable.
+ setDebugLocFromInst(Builder, getDebugLocFromInstOrOperands(OldInduction));
Induction = Builder.CreatePHI(IdxTy, 2, "index");
// The loop step is equal to the vectorization factor (num of SIMD elements)
// times the unroll factor (num of SIMD instructions).
// This is the IR builder that we use to add all of the logic for bypassing
// the new vector loop.
IRBuilder<> BypassBuilder(BypassBlock->getTerminator());
+ setDebugLocFromInst(BypassBuilder,
+ getDebugLocFromInstOrOperands(OldInduction));
// We may need to extend the index in case there is a type mismatch.
// We know that the count starts at zero and does not overflow.
// Create a new block containing the memory check.
BasicBlock *CheckBlock = BypassBlock->splitBasicBlock(MemRuntimeCheck,
"vector.memcheck");
+ if (ParentLoop)
+ ParentLoop->addBasicBlockToLoop(CheckBlock, LI->getBase());
LoopBypassBlocks.push_back(CheckBlock);
// Replace the branch into the memory check block with a conditional branch
// Get ready to start creating new instructions into the vectorized body.
Builder.SetInsertPoint(VecBody->getFirstInsertionPt());
- // Create and register the new vector loop.
- Loop* Lp = new Loop();
- Loop *ParentLoop = OrigLoop->getParentLoop();
-
- // Insert the new loop into the loop nest and register the new basic blocks.
- if (ParentLoop) {
- ParentLoop->addChildLoop(Lp);
- for (unsigned I = 1, E = LoopBypassBlocks.size(); I != E; ++I)
- ParentLoop->addBasicBlockToLoop(LoopBypassBlocks[I], LI->getBase());
- ParentLoop->addBasicBlockToLoop(ScalarPH, LI->getBase());
- ParentLoop->addBasicBlockToLoop(VectorPH, LI->getBase());
- ParentLoop->addBasicBlockToLoop(MiddleBlock, LI->getBase());
- } else {
- LI->addTopLevelLoop(Lp);
- }
-
- Lp->addBasicBlockToLoop(VecBody, LI->getBase());
-
// Save the state.
LoopVectorPreHeader = VectorPH;
LoopScalarPreHeader = ScalarPH;
LoopExitBlock = ExitBlock;
LoopVectorBody = VecBody;
LoopScalarBody = OldBasicBlock;
+
+ LoopVectorizeHints Hints(Lp, true);
+ Hints.setAlreadyVectorized(Lp);
}
/// This function returns the identity element (or neutral element) for
}
}
+static Intrinsic::ID checkUnaryFloatSignature(const CallInst &I,
+ Intrinsic::ID ValidIntrinsicID) {
+ if (I.getNumArgOperands() != 1 ||
+ !I.getArgOperand(0)->getType()->isFloatingPointTy() ||
+ I.getType() != I.getArgOperand(0)->getType() ||
+ !I.onlyReadsMemory())
+ return Intrinsic::not_intrinsic;
+
+ return ValidIntrinsicID;
+}
+
+static Intrinsic::ID checkBinaryFloatSignature(const CallInst &I,
+ Intrinsic::ID ValidIntrinsicID) {
+ if (I.getNumArgOperands() != 2 ||
+ !I.getArgOperand(0)->getType()->isFloatingPointTy() ||
+ !I.getArgOperand(1)->getType()->isFloatingPointTy() ||
+ I.getType() != I.getArgOperand(0)->getType() ||
+ I.getType() != I.getArgOperand(1)->getType() ||
+ !I.onlyReadsMemory())
+ return Intrinsic::not_intrinsic;
+
+ return ValidIntrinsicID;
+}
+
+
static Intrinsic::ID
getIntrinsicIDForCall(CallInst *CI, const TargetLibraryInfo *TLI) {
// If we have an intrinsic call, check if it is trivially vectorizable.
case Intrinsic::log10:
case Intrinsic::log2:
case Intrinsic::fabs:
+ case Intrinsic::copysign:
case Intrinsic::floor:
case Intrinsic::ceil:
case Intrinsic::trunc:
case Intrinsic::rint:
case Intrinsic::nearbyint:
+ case Intrinsic::round:
case Intrinsic::pow:
case Intrinsic::fma:
case Intrinsic::fmuladd:
+ case Intrinsic::lifetime_start:
+ case Intrinsic::lifetime_end:
return II->getIntrinsicID();
default:
return Intrinsic::not_intrinsic;
LibFunc::Func Func;
Function *F = CI->getCalledFunction();
// We're going to make assumptions on the semantics of the functions, check
- // that the target knows that it's available in this environment.
- if (!F || !TLI->getLibFunc(F->getName(), Func))
+ // that the target knows that it's available in this environment and it does
+ // not have local linkage.
+ if (!F || F->hasLocalLinkage() || !TLI->getLibFunc(F->getName(), Func))
return Intrinsic::not_intrinsic;
// Otherwise check if we have a call to a function that can be turned into a
case LibFunc::sin:
case LibFunc::sinf:
case LibFunc::sinl:
- return Intrinsic::sin;
+ return checkUnaryFloatSignature(*CI, Intrinsic::sin);
case LibFunc::cos:
case LibFunc::cosf:
case LibFunc::cosl:
- return Intrinsic::cos;
+ return checkUnaryFloatSignature(*CI, Intrinsic::cos);
case LibFunc::exp:
case LibFunc::expf:
case LibFunc::expl:
- return Intrinsic::exp;
+ return checkUnaryFloatSignature(*CI, Intrinsic::exp);
case LibFunc::exp2:
case LibFunc::exp2f:
case LibFunc::exp2l:
- return Intrinsic::exp2;
+ return checkUnaryFloatSignature(*CI, Intrinsic::exp2);
case LibFunc::log:
case LibFunc::logf:
case LibFunc::logl:
- return Intrinsic::log;
+ return checkUnaryFloatSignature(*CI, Intrinsic::log);
case LibFunc::log10:
case LibFunc::log10f:
case LibFunc::log10l:
- return Intrinsic::log10;
+ return checkUnaryFloatSignature(*CI, Intrinsic::log10);
case LibFunc::log2:
case LibFunc::log2f:
case LibFunc::log2l:
- return Intrinsic::log2;
+ return checkUnaryFloatSignature(*CI, Intrinsic::log2);
case LibFunc::fabs:
case LibFunc::fabsf:
case LibFunc::fabsl:
- return Intrinsic::fabs;
+ return checkUnaryFloatSignature(*CI, Intrinsic::fabs);
+ case LibFunc::copysign:
+ case LibFunc::copysignf:
+ case LibFunc::copysignl:
+ return checkBinaryFloatSignature(*CI, Intrinsic::copysign);
case LibFunc::floor:
case LibFunc::floorf:
case LibFunc::floorl:
- return Intrinsic::floor;
+ return checkUnaryFloatSignature(*CI, Intrinsic::floor);
case LibFunc::ceil:
case LibFunc::ceilf:
case LibFunc::ceill:
- return Intrinsic::ceil;
+ return checkUnaryFloatSignature(*CI, Intrinsic::ceil);
case LibFunc::trunc:
case LibFunc::truncf:
case LibFunc::truncl:
- return Intrinsic::trunc;
+ return checkUnaryFloatSignature(*CI, Intrinsic::trunc);
case LibFunc::rint:
case LibFunc::rintf:
case LibFunc::rintl:
- return Intrinsic::rint;
+ return checkUnaryFloatSignature(*CI, Intrinsic::rint);
case LibFunc::nearbyint:
case LibFunc::nearbyintf:
case LibFunc::nearbyintl:
- return Intrinsic::nearbyint;
+ return checkUnaryFloatSignature(*CI, Intrinsic::nearbyint);
+ case LibFunc::round:
+ case LibFunc::roundf:
+ case LibFunc::roundl:
+ return checkUnaryFloatSignature(*CI, Intrinsic::round);
case LibFunc::pow:
case LibFunc::powf:
case LibFunc::powl:
- return Intrinsic::pow;
+ return checkBinaryFloatSignature(*CI, Intrinsic::pow);
}
return Intrinsic::not_intrinsic;
}
Value *Cmp;
- if (RK == LoopVectorizationLegality::MRK_FloatMin || RK == LoopVectorizationLegality::MRK_FloatMax)
+ if (RK == LoopVectorizationLegality::MRK_FloatMin ||
+ RK == LoopVectorizationLegality::MRK_FloatMax)
Cmp = Builder.CreateFCmp(P, Left, Right, "rdx.minmax.cmp");
else
Cmp = Builder.CreateICmp(P, Left, Right, "rdx.minmax.cmp");
return Select;
}
+namespace {
+struct CSEDenseMapInfo {
+ static bool canHandle(Instruction *I) {
+ return isa<InsertElementInst>(I) || isa<ExtractElementInst>(I) ||
+ isa<ShuffleVectorInst>(I) || isa<GetElementPtrInst>(I);
+ }
+ static inline Instruction *getEmptyKey() {
+ return DenseMapInfo<Instruction *>::getEmptyKey();
+ }
+ static inline Instruction *getTombstoneKey() {
+ return DenseMapInfo<Instruction *>::getTombstoneKey();
+ }
+ static unsigned getHashValue(Instruction *I) {
+ assert(canHandle(I) && "Unknown instruction!");
+ return hash_combine(I->getOpcode(), hash_combine_range(I->value_op_begin(),
+ I->value_op_end()));
+ }
+ static bool isEqual(Instruction *LHS, Instruction *RHS) {
+ if (LHS == getEmptyKey() || RHS == getEmptyKey() ||
+ LHS == getTombstoneKey() || RHS == getTombstoneKey())
+ return LHS == RHS;
+ return LHS->isIdenticalTo(RHS);
+ }
+};
+}
+
+///\brief Perform cse of induction variable instructions.
+static void cse(BasicBlock *BB) {
+ // Perform simple cse.
+ SmallDenseMap<Instruction *, Instruction *, 4, CSEDenseMapInfo> CSEMap;
+ for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E;) {
+ Instruction *In = I++;
+
+ if (!CSEDenseMapInfo::canHandle(In))
+ continue;
+
+ // Check if we can replace this instruction with any of the
+ // visited instructions.
+ if (Instruction *V = CSEMap.lookup(In)) {
+ In->replaceAllUsesWith(V);
+ In->eraseFromParent();
+ continue;
+ }
+
+ CSEMap[In] = In;
+ }
+}
+
void
InnerLoopVectorizer::vectorizeLoop(LoopVectorizationLegality *Legal) {
//===------------------------------------------------===//
LoopVectorizationLegality::ReductionDescriptor RdxDesc =
(*Legal->getReductionVars())[RdxPhi];
+ setDebugLocFromInst(Builder, RdxDesc.StartValue);
+
// We need to generate a reduction vector from the incoming scalar.
// To do so, we need to generate the 'identity' vector and overide
// one of the elements with the incoming scalar reduction. We need
if (RdxDesc.Kind == LoopVectorizationLegality::RK_IntegerMinMax ||
RdxDesc.Kind == LoopVectorizationLegality::RK_FloatMinMax) {
// MinMax reduction have the start value as their identify.
- VectorStart = Identity = Builder.CreateVectorSplat(VF, RdxDesc.StartValue,
- "minmax.ident");
+ if (VF == 1) {
+ VectorStart = Identity = RdxDesc.StartValue;
+ } else {
+ VectorStart = Identity = Builder.CreateVectorSplat(VF,
+ RdxDesc.StartValue,
+ "minmax.ident");
+ }
} else {
+ // Handle other reduction kinds:
Constant *Iden =
- LoopVectorizationLegality::getReductionIdentity(RdxDesc.Kind,
- VecTy->getScalarType());
- Identity = ConstantVector::getSplat(VF, Iden);
-
- // This vector is the Identity vector where the first element is the
- // incoming scalar reduction.
- VectorStart = Builder.CreateInsertElement(Identity,
- RdxDesc.StartValue, Zero);
+ LoopVectorizationLegality::getReductionIdentity(RdxDesc.Kind,
+ VecTy->getScalarType());
+ if (VF == 1) {
+ Identity = Iden;
+ // This vector is the Identity vector where the first element is the
+ // incoming scalar reduction.
+ VectorStart = RdxDesc.StartValue;
+ } else {
+ Identity = ConstantVector::getSplat(VF, Iden);
+
+ // This vector is the Identity vector where the first element is the
+ // incoming scalar reduction.
+ VectorStart = Builder.CreateInsertElement(Identity,
+ RdxDesc.StartValue, Zero);
+ }
}
// Fix the vector-loop phi.
Builder.SetInsertPoint(LoopMiddleBlock->getFirstInsertionPt());
VectorParts RdxParts;
+ setDebugLocFromInst(Builder, RdxDesc.LoopExitInstr);
for (unsigned part = 0; part < UF; ++part) {
// This PHINode contains the vectorized reduction variable, or
// the initial value vector, if we bypass the vector loop.
// Reduce all of the unrolled parts into a single vector.
Value *ReducedPartRdx = RdxParts[0];
unsigned Op = getReductionBinOp(RdxDesc.Kind);
+ setDebugLocFromInst(Builder, ReducedPartRdx);
for (unsigned part = 1; part < UF; ++part) {
if (Op != Instruction::ICmp && Op != Instruction::FCmp)
ReducedPartRdx = Builder.CreateBinOp((Instruction::BinaryOps)Op,
ReducedPartRdx, RdxParts[part]);
}
- // VF is a power of 2 so we can emit the reduction using log2(VF) shuffles
- // and vector ops, reducing the set of values being computed by half each
- // round.
- assert(isPowerOf2_32(VF) &&
- "Reduction emission only supported for pow2 vectors!");
- Value *TmpVec = ReducedPartRdx;
- SmallVector<Constant*, 32> ShuffleMask(VF, 0);
- for (unsigned i = VF; i != 1; i >>= 1) {
- // Move the upper half of the vector to the lower half.
- for (unsigned j = 0; j != i/2; ++j)
- ShuffleMask[j] = Builder.getInt32(i/2 + j);
-
- // Fill the rest of the mask with undef.
- std::fill(&ShuffleMask[i/2], ShuffleMask.end(),
- UndefValue::get(Builder.getInt32Ty()));
-
- Value *Shuf =
+ if (VF > 1) {
+ // VF is a power of 2 so we can emit the reduction using log2(VF) shuffles
+ // and vector ops, reducing the set of values being computed by half each
+ // round.
+ assert(isPowerOf2_32(VF) &&
+ "Reduction emission only supported for pow2 vectors!");
+ Value *TmpVec = ReducedPartRdx;
+ SmallVector<Constant*, 32> ShuffleMask(VF, 0);
+ for (unsigned i = VF; i != 1; i >>= 1) {
+ // Move the upper half of the vector to the lower half.
+ for (unsigned j = 0; j != i/2; ++j)
+ ShuffleMask[j] = Builder.getInt32(i/2 + j);
+
+ // Fill the rest of the mask with undef.
+ std::fill(&ShuffleMask[i/2], ShuffleMask.end(),
+ UndefValue::get(Builder.getInt32Ty()));
+
+ Value *Shuf =
Builder.CreateShuffleVector(TmpVec,
UndefValue::get(TmpVec->getType()),
ConstantVector::get(ShuffleMask),
"rdx.shuf");
- if (Op != Instruction::ICmp && Op != Instruction::FCmp)
- TmpVec = Builder.CreateBinOp((Instruction::BinaryOps)Op, TmpVec, Shuf,
- "bin.rdx");
- else
- TmpVec = createMinMaxOp(Builder, RdxDesc.MinMaxKind, TmpVec, Shuf);
- }
+ if (Op != Instruction::ICmp && Op != Instruction::FCmp)
+ TmpVec = Builder.CreateBinOp((Instruction::BinaryOps)Op, TmpVec, Shuf,
+ "bin.rdx");
+ else
+ TmpVec = createMinMaxOp(Builder, RdxDesc.MinMaxKind, TmpVec, Shuf);
+ }
- // The result is in the first element of the vector.
- Value *Scalar0 = Builder.CreateExtractElement(TmpVec, Builder.getInt32(0));
+ // The result is in the first element of the vector.
+ ReducedPartRdx = Builder.CreateExtractElement(TmpVec,
+ Builder.getInt32(0));
+ }
// Now, we need to fix the users of the reduction variable
// inside and outside of the scalar remainder loop.
for (BasicBlock::iterator LEI = LoopExitBlock->begin(),
LEE = LoopExitBlock->end(); LEI != LEE; ++LEI) {
PHINode *LCSSAPhi = dyn_cast<PHINode>(LEI);
- if (!LCSSAPhi) continue;
+ if (!LCSSAPhi) break;
// All PHINodes need to have a single entry edge, or two if
// we already fixed them.
// incoming bypass edge.
if (LCSSAPhi->getIncomingValue(0) == RdxDesc.LoopExitInstr) {
// Add an edge coming from the bypass.
- LCSSAPhi->addIncoming(Scalar0, LoopMiddleBlock);
+ LCSSAPhi->addIncoming(ReducedPartRdx, LoopMiddleBlock);
break;
}
}// end of the LCSSA phi scan.
assert(IncomingEdgeBlockIdx >= 0 && "Invalid block index");
// Pick the other block.
int SelfEdgeBlockIdx = (IncomingEdgeBlockIdx ? 0 : 1);
- (RdxPhi)->setIncomingValue(SelfEdgeBlockIdx, Scalar0);
+ (RdxPhi)->setIncomingValue(SelfEdgeBlockIdx, ReducedPartRdx);
(RdxPhi)->setIncomingValue(IncomingEdgeBlockIdx, RdxDesc.LoopExitInstr);
}// end of for each redux variable.
- // The Loop exit block may have single value PHI nodes where the incoming
- // value is 'undef'. While vectorizing we only handled real values that
- // were defined inside the loop. Here we handle the 'undef case'.
- // See PR14725.
+ fixLCSSAPHIs();
+
+ // Remove redundant induction instructions.
+ cse(LoopVectorBody);
+}
+
+void InnerLoopVectorizer::fixLCSSAPHIs() {
for (BasicBlock::iterator LEI = LoopExitBlock->begin(),
LEE = LoopExitBlock->end(); LEI != LEE; ++LEI) {
PHINode *LCSSAPhi = dyn_cast<PHINode>(LEI);
- if (!LCSSAPhi) continue;
+ if (!LCSSAPhi) break;
if (LCSSAPhi->getNumIncomingValues() == 1)
LCSSAPhi->addIncoming(UndefValue::get(LCSSAPhi->getType()),
LoopMiddleBlock);
}
-}
+}
InnerLoopVectorizer::VectorParts
InnerLoopVectorizer::createEdgeMask(BasicBlock *Src, BasicBlock *Dst) {
assert(std::find(pred_begin(Dst), pred_end(Dst), Src) != pred_end(Dst) &&
"Invalid edge");
+ // Look for cached value.
+ std::pair<BasicBlock*, BasicBlock*> Edge(Src, Dst);
+ EdgeMaskCache::iterator ECEntryIt = MaskCache.find(Edge);
+ if (ECEntryIt != MaskCache.end())
+ return ECEntryIt->second;
+
VectorParts SrcMask = createBlockInMask(Src);
// The terminator has to be a branch inst!
for (unsigned part = 0; part < UF; ++part)
EdgeMask[part] = Builder.CreateAnd(EdgeMask[part], SrcMask[part]);
+
+ MaskCache[Edge] = EdgeMask;
return EdgeMask;
}
+ MaskCache[Edge] = SrcMask;
return SrcMask;
}
return BlockMask;
}
-void
-InnerLoopVectorizer::vectorizeBlockInLoop(LoopVectorizationLegality *Legal,
- BasicBlock *BB, PhiVector *PV) {
- // For each instruction in the old loop.
- for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; ++it) {
- VectorParts &Entry = WidenMap.get(it);
- switch (it->getOpcode()) {
- case Instruction::Br:
- // Nothing to do for PHIs and BR, since we already took care of the
- // loop control flow instructions.
- continue;
- case Instruction::PHI:{
- PHINode* P = cast<PHINode>(it);
- // Handle reduction variables:
- if (Legal->getReductionVars()->count(P)) {
- for (unsigned part = 0; part < UF; ++part) {
- // This is phase one of vectorizing PHIs.
- Type *VecTy = VectorType::get(it->getType(), VF);
- Entry[part] = PHINode::Create(VecTy, 2, "vec.phi",
- LoopVectorBody-> getFirstInsertionPt());
- }
- PV->push_back(P);
- continue;
- }
+void InnerLoopVectorizer::widenPHIInstruction(Instruction *PN,
+ InnerLoopVectorizer::VectorParts &Entry,
+ LoopVectorizationLegality *Legal,
+ unsigned UF, unsigned VF, PhiVector *PV) {
+ PHINode* P = cast<PHINode>(PN);
+ // Handle reduction variables:
+ if (Legal->getReductionVars()->count(P)) {
+ for (unsigned part = 0; part < UF; ++part) {
+ // This is phase one of vectorizing PHIs.
+ Type *VecTy = (VF == 1) ? PN->getType() :
+ VectorType::get(PN->getType(), VF);
+ Entry[part] = PHINode::Create(VecTy, 2, "vec.phi",
+ LoopVectorBody-> getFirstInsertionPt());
+ }
+ PV->push_back(P);
+ return;
+ }
- // Check for PHI nodes that are lowered to vector selects.
- if (P->getParent() != OrigLoop->getHeader()) {
- // We know that all PHIs in non header blocks are converted into
- // selects, so we don't have to worry about the insertion order and we
- // can just use the builder.
- // At this point we generate the predication tree. There may be
- // duplications since this is a simple recursive scan, but future
- // optimizations will clean it up.
-
- unsigned NumIncoming = P->getNumIncomingValues();
-
- // Generate a sequence of selects of the form:
- // SELECT(Mask3, In3,
- // SELECT(Mask2, In2,
- // ( ...)))
- for (unsigned In = 0; In < NumIncoming; In++) {
- VectorParts Cond = createEdgeMask(P->getIncomingBlock(In),
- P->getParent());
- VectorParts &In0 = getVectorValue(P->getIncomingValue(In));
-
- for (unsigned part = 0; part < UF; ++part) {
- // We might have single edge PHIs (blocks) - use an identity
- // 'select' for the first PHI operand.
- if (In == 0)
- Entry[part] = Builder.CreateSelect(Cond[part], In0[part],
- In0[part]);
- else
- // Select between the current value and the previous incoming edge
- // based on the incoming mask.
- Entry[part] = Builder.CreateSelect(Cond[part], In0[part],
- Entry[part], "predphi");
- }
- }
- continue;
+ setDebugLocFromInst(Builder, P);
+ // Check for PHI nodes that are lowered to vector selects.
+ if (P->getParent() != OrigLoop->getHeader()) {
+ // We know that all PHIs in non-header blocks are converted into
+ // selects, so we don't have to worry about the insertion order and we
+ // can just use the builder.
+ // At this point we generate the predication tree. There may be
+ // duplications since this is a simple recursive scan, but future
+ // optimizations will clean it up.
+
+ unsigned NumIncoming = P->getNumIncomingValues();
+
+ // Generate a sequence of selects of the form:
+ // SELECT(Mask3, In3,
+ // SELECT(Mask2, In2,
+ // ( ...)))
+ for (unsigned In = 0; In < NumIncoming; In++) {
+ VectorParts Cond = createEdgeMask(P->getIncomingBlock(In),
+ P->getParent());
+ VectorParts &In0 = getVectorValue(P->getIncomingValue(In));
+
+ for (unsigned part = 0; part < UF; ++part) {
+ // We might have single edge PHIs (blocks) - use an identity
+ // 'select' for the first PHI operand.
+ if (In == 0)
+ Entry[part] = Builder.CreateSelect(Cond[part], In0[part],
+ In0[part]);
+ else
+ // Select between the current value and the previous incoming edge
+ // based on the incoming mask.
+ Entry[part] = Builder.CreateSelect(Cond[part], In0[part],
+ Entry[part], "predphi");
}
+ }
+ return;
+ }
- // This PHINode must be an induction variable.
- // Make sure that we know about it.
- assert(Legal->getInductionVars()->count(P) &&
- "Not an induction variable");
-
- LoopVectorizationLegality::InductionInfo II =
- Legal->getInductionVars()->lookup(P);
-
- switch (II.IK) {
- case LoopVectorizationLegality::IK_NoInduction:
- llvm_unreachable("Unknown induction");
- case LoopVectorizationLegality::IK_IntInduction: {
- assert(P->getType() == II.StartValue->getType() && "Types must match");
- Type *PhiTy = P->getType();
- Value *Broadcasted;
- if (P == OldInduction) {
- // Handle the canonical induction variable. We might have had to
- // extend the type.
- Broadcasted = Builder.CreateTrunc(Induction, PhiTy);
- } else {
- // Handle other induction variables that are now based on the
- // canonical one.
- Value *NormalizedIdx = Builder.CreateSub(Induction, ExtendedIdx,
- "normalized.idx");
- NormalizedIdx = Builder.CreateSExtOrTrunc(NormalizedIdx, PhiTy);
- Broadcasted = Builder.CreateAdd(II.StartValue, NormalizedIdx,
- "offset.idx");
- }
- Broadcasted = getBroadcastInstrs(Broadcasted);
- // After broadcasting the induction variable we need to make the vector
- // consecutive by adding 0, 1, 2, etc.
+ // This PHINode must be an induction variable.
+ // Make sure that we know about it.
+ assert(Legal->getInductionVars()->count(P) &&
+ "Not an induction variable");
+
+ LoopVectorizationLegality::InductionInfo II =
+ Legal->getInductionVars()->lookup(P);
+
+ switch (II.IK) {
+ case LoopVectorizationLegality::IK_NoInduction:
+ llvm_unreachable("Unknown induction");
+ case LoopVectorizationLegality::IK_IntInduction: {
+ assert(P->getType() == II.StartValue->getType() && "Types must match");
+ Type *PhiTy = P->getType();
+ Value *Broadcasted;
+ if (P == OldInduction) {
+ // Handle the canonical induction variable. We might have had to
+ // extend the type.
+ Broadcasted = Builder.CreateTrunc(Induction, PhiTy);
+ } else {
+ // Handle other induction variables that are now based on the
+ // canonical one.
+ Value *NormalizedIdx = Builder.CreateSub(Induction, ExtendedIdx,
+ "normalized.idx");
+ NormalizedIdx = Builder.CreateSExtOrTrunc(NormalizedIdx, PhiTy);
+ Broadcasted = Builder.CreateAdd(II.StartValue, NormalizedIdx,
+ "offset.idx");
+ }
+ Broadcasted = getBroadcastInstrs(Broadcasted);
+ // After broadcasting the induction variable we need to make the vector
+ // consecutive by adding 0, 1, 2, etc.
+ for (unsigned part = 0; part < UF; ++part)
+ Entry[part] = getConsecutiveVector(Broadcasted, VF * part, false);
+ return;
+ }
+ case LoopVectorizationLegality::IK_ReverseIntInduction:
+ case LoopVectorizationLegality::IK_PtrInduction:
+ case LoopVectorizationLegality::IK_ReversePtrInduction:
+ // Handle reverse integer and pointer inductions.
+ Value *StartIdx = ExtendedIdx;
+ // This is the normalized GEP that starts counting at zero.
+ Value *NormalizedIdx = Builder.CreateSub(Induction, StartIdx,
+ "normalized.idx");
+
+ // Handle the reverse integer induction variable case.
+ if (LoopVectorizationLegality::IK_ReverseIntInduction == II.IK) {
+ IntegerType *DstTy = cast<IntegerType>(II.StartValue->getType());
+ Value *CNI = Builder.CreateSExtOrTrunc(NormalizedIdx, DstTy,
+ "resize.norm.idx");
+ Value *ReverseInd = Builder.CreateSub(II.StartValue, CNI,
+ "reverse.idx");
+
+ // This is a new value so do not hoist it out.
+ Value *Broadcasted = getBroadcastInstrs(ReverseInd);
+ // After broadcasting the induction variable we need to make the
+ // vector consecutive by adding ... -3, -2, -1, 0.
for (unsigned part = 0; part < UF; ++part)
- Entry[part] = getConsecutiveVector(Broadcasted, VF * part, false);
- continue;
+ Entry[part] = getConsecutiveVector(Broadcasted, -(int)VF * part,
+ true);
+ return;
}
- case LoopVectorizationLegality::IK_ReverseIntInduction:
- case LoopVectorizationLegality::IK_PtrInduction:
- case LoopVectorizationLegality::IK_ReversePtrInduction:
- // Handle reverse integer and pointer inductions.
- Value *StartIdx = ExtendedIdx;
- // This is the normalized GEP that starts counting at zero.
- Value *NormalizedIdx = Builder.CreateSub(Induction, StartIdx,
- "normalized.idx");
- // Handle the reverse integer induction variable case.
- if (LoopVectorizationLegality::IK_ReverseIntInduction == II.IK) {
- IntegerType *DstTy = cast<IntegerType>(II.StartValue->getType());
- Value *CNI = Builder.CreateSExtOrTrunc(NormalizedIdx, DstTy,
- "resize.norm.idx");
- Value *ReverseInd = Builder.CreateSub(II.StartValue, CNI,
- "reverse.idx");
-
- // This is a new value so do not hoist it out.
- Value *Broadcasted = getBroadcastInstrs(ReverseInd);
- // After broadcasting the induction variable we need to make the
- // vector consecutive by adding ... -3, -2, -1, 0.
- for (unsigned part = 0; part < UF; ++part)
- Entry[part] = getConsecutiveVector(Broadcasted, -(int)VF * part,
- true);
+ // Handle the pointer induction variable case.
+ assert(P->getType()->isPointerTy() && "Unexpected type.");
+
+ // Is this a reverse induction ptr or a consecutive induction ptr.
+ bool Reverse = (LoopVectorizationLegality::IK_ReversePtrInduction ==
+ II.IK);
+
+ // This is the vector of results. Notice that we don't generate
+ // vector geps because scalar geps result in better code.
+ for (unsigned part = 0; part < UF; ++part) {
+ if (VF == 1) {
+ int EltIndex = (part) * (Reverse ? -1 : 1);
+ Constant *Idx = ConstantInt::get(Induction->getType(), EltIndex);
+ Value *GlobalIdx;
+ if (Reverse)
+ GlobalIdx = Builder.CreateSub(Idx, NormalizedIdx, "gep.ridx");
+ else
+ GlobalIdx = Builder.CreateAdd(NormalizedIdx, Idx, "gep.idx");
+
+ Value *SclrGep = Builder.CreateGEP(II.StartValue, GlobalIdx,
+ "next.gep");
+ Entry[part] = SclrGep;
continue;
}
- // Handle the pointer induction variable case.
- assert(P->getType()->isPointerTy() && "Unexpected type.");
-
- // Is this a reverse induction ptr or a consecutive induction ptr.
- bool Reverse = (LoopVectorizationLegality::IK_ReversePtrInduction ==
- II.IK);
-
- // This is the vector of results. Notice that we don't generate
- // vector geps because scalar geps result in better code.
- for (unsigned part = 0; part < UF; ++part) {
- Value *VecVal = UndefValue::get(VectorType::get(P->getType(), VF));
- for (unsigned int i = 0; i < VF; ++i) {
- int EltIndex = (i + part * VF) * (Reverse ? -1 : 1);
- Constant *Idx = ConstantInt::get(Induction->getType(), EltIndex);
- Value *GlobalIdx;
- if (!Reverse)
- GlobalIdx = Builder.CreateAdd(NormalizedIdx, Idx, "gep.idx");
- else
- GlobalIdx = Builder.CreateSub(Idx, NormalizedIdx, "gep.ridx");
-
- Value *SclrGep = Builder.CreateGEP(II.StartValue, GlobalIdx,
- "next.gep");
- VecVal = Builder.CreateInsertElement(VecVal, SclrGep,
- Builder.getInt32(i),
- "insert.gep");
- }
- Entry[part] = VecVal;
+ Value *VecVal = UndefValue::get(VectorType::get(P->getType(), VF));
+ for (unsigned int i = 0; i < VF; ++i) {
+ int EltIndex = (i + part * VF) * (Reverse ? -1 : 1);
+ Constant *Idx = ConstantInt::get(Induction->getType(), EltIndex);
+ Value *GlobalIdx;
+ if (!Reverse)
+ GlobalIdx = Builder.CreateAdd(NormalizedIdx, Idx, "gep.idx");
+ else
+ GlobalIdx = Builder.CreateSub(Idx, NormalizedIdx, "gep.ridx");
+
+ Value *SclrGep = Builder.CreateGEP(II.StartValue, GlobalIdx,
+ "next.gep");
+ VecVal = Builder.CreateInsertElement(VecVal, SclrGep,
+ Builder.getInt32(i),
+ "insert.gep");
}
- continue;
+ Entry[part] = VecVal;
}
+ return;
+ }
+}
+void
+InnerLoopVectorizer::vectorizeBlockInLoop(LoopVectorizationLegality *Legal,
+ BasicBlock *BB, PhiVector *PV) {
+ // For each instruction in the old loop.
+ for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; ++it) {
+ VectorParts &Entry = WidenMap.get(it);
+ switch (it->getOpcode()) {
+ case Instruction::Br:
+ // Nothing to do for PHIs and BR, since we already took care of the
+ // loop control flow instructions.
+ continue;
+ case Instruction::PHI:{
+ // Vectorize PHINodes.
+ widenPHIInstruction(it, Entry, Legal, UF, VF, PV);
+ continue;
}// End of PHI.
case Instruction::Add:
case Instruction::Xor: {
// Just widen binops.
BinaryOperator *BinOp = dyn_cast<BinaryOperator>(it);
+ setDebugLocFromInst(Builder, BinOp);
VectorParts &A = getVectorValue(it->getOperand(0));
VectorParts &B = getVectorValue(it->getOperand(1));
// instruction with a scalar condition. Otherwise, use vector-select.
bool InvariantCond = SE->isLoopInvariant(SE->getSCEV(it->getOperand(0)),
OrigLoop);
+ setDebugLocFromInst(Builder, it);
// The condition can be loop invariant but still defined inside the
// loop. This means that we can't just use the original 'cond' value.
VectorParts &Cond = getVectorValue(it->getOperand(0));
VectorParts &Op0 = getVectorValue(it->getOperand(1));
VectorParts &Op1 = getVectorValue(it->getOperand(2));
- Value *ScalarCond = Builder.CreateExtractElement(Cond[0],
- Builder.getInt32(0));
+
+ Value *ScalarCond = (VF == 1) ? Cond[0] :
+ Builder.CreateExtractElement(Cond[0], Builder.getInt32(0));
+
for (unsigned Part = 0; Part < UF; ++Part) {
Entry[Part] = Builder.CreateSelect(
InvariantCond ? ScalarCond : Cond[Part],
// Widen compares. Generate vector compares.
bool FCmp = (it->getOpcode() == Instruction::FCmp);
CmpInst *Cmp = dyn_cast<CmpInst>(it);
+ setDebugLocFromInst(Builder, it);
VectorParts &A = getVectorValue(it->getOperand(0));
VectorParts &B = getVectorValue(it->getOperand(1));
for (unsigned Part = 0; Part < UF; ++Part) {
case Instruction::FPTrunc:
case Instruction::BitCast: {
CastInst *CI = dyn_cast<CastInst>(it);
+ setDebugLocFromInst(Builder, it);
/// Optimize the special case where the source is the induction
/// variable. Notice that we can only optimize the 'trunc' case
/// because: a. FP conversions lose precision, b. sext/zext may wrap,
break;
}
/// Vectorize casts.
- Type *DestTy = VectorType::get(CI->getType()->getScalarType(), VF);
+ Type *DestTy = (VF == 1) ? CI->getType() :
+ VectorType::get(CI->getType(), VF);
VectorParts &A = getVectorValue(it->getOperand(0));
for (unsigned Part = 0; Part < UF; ++Part)
// Ignore dbg intrinsics.
if (isa<DbgInfoIntrinsic>(it))
break;
+ setDebugLocFromInst(Builder, it);
Module *M = BB->getParent()->getParent();
CallInst *CI = cast<CallInst>(it);
Intrinsic::ID ID = getIntrinsicIDForCall(CI, TLI);
assert(ID && "Not an intrinsic call!");
- for (unsigned Part = 0; Part < UF; ++Part) {
- SmallVector<Value*, 4> Args;
- for (unsigned i = 0, ie = CI->getNumArgOperands(); i != ie; ++i) {
- VectorParts &Arg = getVectorValue(CI->getArgOperand(i));
- Args.push_back(Arg[Part]);
+ switch (ID) {
+ case Intrinsic::lifetime_end:
+ case Intrinsic::lifetime_start:
+ scalarizeInstruction(it);
+ break;
+ default:
+ for (unsigned Part = 0; Part < UF; ++Part) {
+ SmallVector<Value *, 4> Args;
+ for (unsigned i = 0, ie = CI->getNumArgOperands(); i != ie; ++i) {
+ VectorParts &Arg = getVectorValue(CI->getArgOperand(i));
+ Args.push_back(Arg[Part]);
+ }
+ Type *Tys[] = {CI->getType()};
+ if (VF > 1)
+ Tys[0] = VectorType::get(CI->getType()->getScalarType(), VF);
+
+ Function *F = Intrinsic::getDeclaration(M, ID, Tys);
+ Entry[Part] = Builder.CreateCall(F, Args);
}
- Type *Tys[] = { VectorType::get(CI->getType()->getScalarType(), VF) };
- Function *F = Intrinsic::getDeclaration(M, ID, Tys);
- Entry[Part] = Builder.CreateCall(F, Args);
+ break;
}
break;
}
return false;
assert(TheLoop->getNumBlocks() > 1 && "Single block loops are vectorizable");
- std::vector<BasicBlock*> &LoopBlocks = TheLoop->getBlocksVector();
+
+ // A list of pointers that we can safely read and write to.
+ SmallPtrSet<Value *, 8> SafePointes;
+
+ // Collect safe addresses.
+ for (Loop::block_iterator BI = TheLoop->block_begin(),
+ BE = TheLoop->block_end(); BI != BE; ++BI) {
+ BasicBlock *BB = *BI;
+
+ if (blockNeedsPredication(BB))
+ continue;
+
+ for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
+ if (LoadInst *LI = dyn_cast<LoadInst>(I))
+ SafePointes.insert(LI->getPointerOperand());
+ else if (StoreInst *SI = dyn_cast<StoreInst>(I))
+ SafePointes.insert(SI->getPointerOperand());
+ }
+ }
// Collect the blocks that need predication.
- for (unsigned i = 0, e = LoopBlocks.size(); i < e; ++i) {
- BasicBlock *BB = LoopBlocks[i];
+ for (Loop::block_iterator BI = TheLoop->block_begin(),
+ BE = TheLoop->block_end(); BI != BE; ++BI) {
+ BasicBlock *BB = *BI;
// We don't support switch statements inside loops.
if (!isa<BranchInst>(BB->getTerminator()))
return false;
// We must be able to predicate all blocks that need to be predicated.
- if (blockNeedsPredication(BB) && !blockCanBePredicated(BB))
+ if (blockNeedsPredication(BB) && !blockCanBePredicated(BB, SafePointes))
return false;
}
- // Check that we can actually speculate the hoistable loads.
- if (!LoadSpeculation.canHoistAllLoads())
- return false;
-
// We can if-convert this loop.
return true;
}
if (!TheLoop->getExitingBlock())
return false;
- unsigned NumBlocks = TheLoop->getNumBlocks();
+ // We need to have a loop header.
+ DEBUG(dbgs() << "LV: Found a loop: " <<
+ TheLoop->getHeader()->getName() << '\n');
- // Check if we can if-convert non single-bb loops.
+ // Check if we can if-convert non-single-bb loops.
+ unsigned NumBlocks = TheLoop->getNumBlocks();
if (NumBlocks != 1 && !canVectorizeWithIfConvert()) {
DEBUG(dbgs() << "LV: Can't if-convert the loop.\n");
return false;
}
- // We need to have a loop header.
- BasicBlock *Latch = TheLoop->getLoopLatch();
- DEBUG(dbgs() << "LV: Found a loop: " <<
- TheLoop->getHeader()->getName() << "\n");
-
// ScalarEvolution needs to be able to find the exit count.
const SCEV *ExitCount = SE->getBackedgeTakenCount(TheLoop);
if (ExitCount == SE->getCouldNotCompute()) {
}
// Do not loop-vectorize loops with a tiny trip count.
+ BasicBlock *Latch = TheLoop->getLoopLatch();
unsigned TC = SE->getSmallConstantTripCount(TheLoop, Latch);
if (TC > 0u && TC < TinyTripCountVectorThreshold) {
DEBUG(dbgs() << "LV: Found a loop with a very small trip count. " <<
static Type *convertPointerToIntegerType(DataLayout &DL, Type *Ty) {
if (Ty->isPointerTy())
- return DL.getIntPtrType(Ty->getContext());
+ return DL.getIntPtrType(Ty);
+
+ // It is possible that char's or short's overflow when we ask for the loop's
+ // trip count, work around this by changing the type size.
+ if (Ty->getScalarSizeInBits() < 32)
+ return Type::getInt32Ty(Ty->getContext());
+
return Ty;
}
Instruction *U = cast<Instruction>(*I);
// This user may be a reduction exit value.
if (!TheLoop->contains(U)) {
- DEBUG(dbgs() << "LV: Found an outside user for : "<< *U << "\n");
+ DEBUG(dbgs() << "LV: Found an outside user for : " << *U << '\n');
return true;
}
}
DEBUG(dbgs() << "LV: Found an induction variable.\n");
Inductions[Phi] = InductionInfo(StartValue, IK);
+
+ // Until we explicitly handle the case of an induction variable with
+ // an outside loop user we have to give up vectorizing this loop.
+ if (hasOutsideLoopUser(TheLoop, it, AllowedExit))
+ return false;
+
continue;
}
continue;
}
if (AddReductionVar(Phi, RK_FloatMinMax)) {
- DEBUG(dbgs() << "LV: Found an float MINMAX reduction PHI."<< *Phi <<"\n");
+ DEBUG(dbgs() << "LV: Found an float MINMAX reduction PHI."<< *Phi <<
+ "\n");
continue;
}
}
// Check that the instruction return type is vectorizable.
- if (!VectorType::isValidElementType(it->getType()) &&
- !it->getType()->isVoidTy()) {
- DEBUG(dbgs() << "LV: Found unvectorizable type." << "\n");
+ // Also, we can't vectorize extractelement instructions.
+ if ((!VectorType::isValidElementType(it->getType()) &&
+ !it->getType()->isVoidTy()) || isa<ExtractElementInst>(it)) {
+ DEBUG(dbgs() << "LV: Found unvectorizable type.\n");
return false;
}
}
}
+namespace {
/// \brief Analyses memory accesses in a loop.
///
/// Checks whether run time pointer checks are needed and builds sets for data
class AccessAnalysis {
public:
/// \brief Read or write access location.
- typedef std::pair<Value*, char> MemAccessInfo;
+ typedef PointerIntPair<Value *, 1, bool> MemAccessInfo;
+ typedef SmallPtrSet<MemAccessInfo, 8> MemAccessInfoSet;
/// \brief Set of potential dependent memory accesses.
typedef EquivalenceClasses<MemAccessInfo> DepCandidates;
/// \brief Register a load and whether it is only read from.
void addLoad(Value *Ptr, bool IsReadOnly) {
- Accesses.insert(std::make_pair(Ptr, false));
+ Accesses.insert(MemAccessInfo(Ptr, false));
if (IsReadOnly)
ReadOnlyPtr.insert(Ptr);
}
/// \brief Register a store.
void addStore(Value *Ptr) {
- Accesses.insert(std::make_pair(Ptr, true));
+ Accesses.insert(MemAccessInfo(Ptr, true));
}
/// \brief Check whether we can check the pointers at runtime for
/// non-intersection.
bool canCheckPtrAtRT(LoopVectorizationLegality::RuntimePointerCheck &RtCheck,
unsigned &NumComparisons, ScalarEvolution *SE,
- Loop *TheLoop);
+ Loop *TheLoop, bool ShouldCheckStride = false);
/// \brief Goes over all memory accesses, checks whether a RT check is needed
/// and builds sets of dependent accesses.
bool isRTCheckNeeded() { return IsRTCheckNeeded; }
bool isDependencyCheckNeeded() { return !CheckDeps.empty(); }
+ void resetDepChecks() { CheckDeps.clear(); }
- DenseSet<MemAccessInfo> &getDependenciesToCheck() { return CheckDeps; }
+ MemAccessInfoSet &getDependenciesToCheck() { return CheckDeps; }
private:
- typedef DenseSet<MemAccessInfo> PtrAccessSet;
+ typedef SetVector<MemAccessInfo> PtrAccessSet;
typedef DenseMap<Value*, MemAccessInfo> UnderlyingObjToAccessMap;
/// \brief Go over all memory access or only the deferred ones if
UnderlyingObjToAccessMap ObjToLastAccess;
/// Set of accesses that need a further dependence check.
- DenseSet<MemAccessInfo> CheckDeps;
+ MemAccessInfoSet CheckDeps;
/// Set of pointers that are read only.
SmallPtrSet<Value*, 16> ReadOnlyPtr;
bool IsRTCheckNeeded;
};
+} // end anonymous namespace
+
/// \brief Check whether a pointer can participate in a runtime bounds check.
static bool hasComputableBounds(ScalarEvolution *SE, Value *Ptr) {
const SCEV *PtrScev = SE->getSCEV(Ptr);
return AR->isAffine();
}
+/// \brief Check the stride of the pointer and ensure that it does not wrap in
+/// the address space.
+static int isStridedPtr(ScalarEvolution *SE, DataLayout *DL, Value *Ptr,
+ const Loop *Lp);
+
bool AccessAnalysis::canCheckPtrAtRT(
LoopVectorizationLegality::RuntimePointerCheck &RtCheck,
unsigned &NumComparisons, ScalarEvolution *SE,
- Loop *TheLoop) {
+ Loop *TheLoop, bool ShouldCheckStride) {
// Find pointers with computable bounds. We are going to use this information
// to place a runtime bound check.
unsigned NumReadPtrChecks = 0;
for (PtrAccessSet::iterator AI = Accesses.begin(), AE = Accesses.end();
AI != AE; ++AI) {
const MemAccessInfo &Access = *AI;
- Value *Ptr = Access.first;
- bool IsWrite = Access.second;
+ Value *Ptr = Access.getPointer();
+ bool IsWrite = Access.getInt();
// Just add write checks if we have both.
- if (!IsWrite && Accesses.count(std::make_pair(Ptr, true)))
+ if (!IsWrite && Accesses.count(MemAccessInfo(Ptr, true)))
continue;
if (IsWrite)
else
++NumReadPtrChecks;
- if (hasComputableBounds(SE, Ptr)) {
+ if (hasComputableBounds(SE, Ptr) &&
+ // When we run after a failing dependency check we have to make sure we
+ // don't have wrapping pointers.
+ (!ShouldCheckStride || isStridedPtr(SE, DL, Ptr, TheLoop) == 1)) {
// The id of the dependence set.
unsigned DepId;
if (IsDepCheckNeeded) {
- Value *Leader = DepCands.getLeaderValue(Access).first;
+ Value *Leader = DepCands.getLeaderValue(Access).getPointer();
unsigned &LeaderId = DepSetId[Leader];
if (!LeaderId)
LeaderId = RunningDepId++;
RtCheck.insert(SE, TheLoop, Ptr, IsWrite, DepId);
- DEBUG(dbgs() << "LV: Found a runtime check ptr:" << *Ptr <<"\n");
+ DEBUG(dbgs() << "LV: Found a runtime check ptr:" << *Ptr << '\n');
} else {
CanDoRT = false;
}
if (IsDepCheckNeeded && CanDoRT && RunningDepId == 2)
NumComparisons = 0; // Only one dependence set.
- else
+ else {
NumComparisons = (NumWritePtrChecks * (NumReadPtrChecks +
NumWritePtrChecks - 1));
+ }
+
+ // If the pointers that we would use for the bounds comparison have different
+ // address spaces, assume the values aren't directly comparable, so we can't
+ // use them for the runtime check. We also have to assume they could
+ // overlap. In the future there should be metadata for whether address spaces
+ // are disjoint.
+ unsigned NumPointers = RtCheck.Pointers.size();
+ for (unsigned i = 0; i < NumPointers; ++i) {
+ for (unsigned j = i + 1; j < NumPointers; ++j) {
+ // Only need to check pointers between two different dependency sets.
+ if (RtCheck.DependencySetId[i] == RtCheck.DependencySetId[j])
+ continue;
+
+ Value *PtrI = RtCheck.Pointers[i];
+ Value *PtrJ = RtCheck.Pointers[j];
+
+ unsigned ASi = PtrI->getType()->getPointerAddressSpace();
+ unsigned ASj = PtrJ->getType()->getPointerAddressSpace();
+ if (ASi != ASj) {
+ DEBUG(dbgs() << "LV: Runtime check would require comparison between"
+ " different address spaces\n");
+ return false;
+ }
+ }
+ }
+
return CanDoRT;
}
PtrAccessSet &S = UseDeferred ? DeferredAccesses : Accesses;
for (PtrAccessSet::iterator AI = S.begin(), AE = S.end(); AI != AE; ++AI) {
const MemAccessInfo &Access = *AI;
- Value *Ptr = Access.first;
- bool IsWrite = Access.second;
+ Value *Ptr = Access.getPointer();
+ bool IsWrite = Access.getInt();
DepCands.insert(Access);
!isa<Argument>(UnderlyingObj)) &&
!isIdentifiedObject(UnderlyingObj))) {
DEBUG(dbgs() << "LV: Found an unidentified " <<
- (IsWrite ? "write" : "read" ) << " ptr:" << *UnderlyingObj <<
+ (IsWrite ? "write" : "read" ) << " ptr: " << *UnderlyingObj <<
"\n");
IsRTCheckNeeded = (IsRTCheckNeeded ||
!isIdentifiedObject(UnderlyingObj) ||
}
}
+namespace {
/// \brief Checks memory dependences among accesses to the same underlying
/// object to determine whether there vectorization is legal or not (and at
/// which vectorization factor).
///
class MemoryDepChecker {
public:
- typedef std::pair<Value*, char> MemAccessInfo;
+ typedef PointerIntPair<Value *, 1, bool> MemAccessInfo;
+ typedef SmallPtrSet<MemAccessInfo, 8> MemAccessInfoSet;
- MemoryDepChecker(ScalarEvolution *Se, DataLayout *Dl, const Loop *L) :
- SE(Se), DL(Dl), InnermostLoop(L), AccessIdx(0) {}
+ MemoryDepChecker(ScalarEvolution *Se, DataLayout *Dl, const Loop *L)
+ : SE(Se), DL(Dl), InnermostLoop(L), AccessIdx(0),
+ ShouldRetryWithRuntimeCheck(false) {}
/// \brief Register the location (instructions are given increasing numbers)
/// of a write access.
void addAccess(StoreInst *SI) {
Value *Ptr = SI->getPointerOperand();
- Accesses[std::make_pair(Ptr, true)].push_back(AccessIdx);
+ Accesses[MemAccessInfo(Ptr, true)].push_back(AccessIdx);
InstMap.push_back(SI);
++AccessIdx;
}
/// of a write access.
void addAccess(LoadInst *LI) {
Value *Ptr = LI->getPointerOperand();
- Accesses[std::make_pair(Ptr, false)].push_back(AccessIdx);
+ Accesses[MemAccessInfo(Ptr, false)].push_back(AccessIdx);
InstMap.push_back(LI);
++AccessIdx;
}
///
/// Only checks sets with elements in \p CheckDeps.
bool areDepsSafe(AccessAnalysis::DepCandidates &AccessSets,
- DenseSet<MemAccessInfo> &CheckDeps);
+ MemAccessInfoSet &CheckDeps);
/// \brief The maximum number of bytes of a vector register we can vectorize
/// the accesses safely with.
unsigned getMaxSafeDepDistBytes() { return MaxSafeDepDistBytes; }
+ /// \brief In same cases when the dependency check fails we can still
+ /// vectorize the loop with a dynamic array access check.
+ bool shouldRetryWithRuntimeCheck() { return ShouldRetryWithRuntimeCheck; }
+
private:
ScalarEvolution *SE;
DataLayout *DL;
// We can access this many bytes in parallel safely.
unsigned MaxSafeDepDistBytes;
+ /// \brief If we see a non-constant dependence distance we can still try to
+ /// vectorize this loop with runtime checks.
+ bool ShouldRetryWithRuntimeCheck;
+
/// \brief Check whether there is a plausible dependence between the two
/// accesses.
///
bool couldPreventStoreLoadForward(unsigned Distance, unsigned TypeByteSize);
};
+} // end anonymous namespace
+
static bool isInBoundsGep(Value *Ptr) {
if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Ptr))
return GEP->isInBounds();
/// \brief Check whether the access through \p Ptr has a constant stride.
static int isStridedPtr(ScalarEvolution *SE, DataLayout *DL, Value *Ptr,
const Loop *Lp) {
- const Type *PtrTy = Ptr->getType();
- assert(PtrTy->isPointerTy() && "Unexpected non ptr");
+ const Type *Ty = Ptr->getType();
+ assert(Ty->isPointerTy() && "Unexpected non-ptr");
// Make sure that the pointer does not point to aggregate types.
- if (cast<PointerType>(Ptr->getType())->getElementType()->isAggregateType()) {
- DEBUG(dbgs() << "LV: Bad stride - Not a pointer to a scalar type" << *Ptr
- << "\n");
+ const PointerType *PtrTy = cast<PointerType>(Ty);
+ if (PtrTy->getElementType()->isAggregateType()) {
+ DEBUG(dbgs() << "LV: Bad stride - Not a pointer to a scalar type" << *Ptr <<
+ "\n");
return 0;
}
// The accesss function must stride over the innermost loop.
if (Lp != AR->getLoop()) {
- DEBUG(dbgs() << "LV: Bad stride - Not striding over innermost loop " << *Ptr
- << " SCEV: " << *PtrScev << "\n");
+ DEBUG(dbgs() << "LV: Bad stride - Not striding over innermost loop " <<
+ *Ptr << " SCEV: " << *PtrScev << "\n");
}
// The address calculation must not wrap. Otherwise, a dependence could be
- // inverted. An inbounds getelementptr that is a AddRec with a unit stride
+ // inverted.
+ // An inbounds getelementptr that is a AddRec with a unit stride
// cannot wrap per definition. The unit stride requirement is checked later.
+ // An getelementptr without an inbounds attribute and unit stride would have
+ // to access the pointer value "0" which is undefined behavior in address
+ // space 0, therefore we can also vectorize this case.
bool IsInBoundsGEP = isInBoundsGep(Ptr);
bool IsNoWrapAddRec = AR->getNoWrapFlags(SCEV::NoWrapMask);
- if (!IsNoWrapAddRec && !IsInBoundsGEP) {
+ bool IsInAddressSpaceZero = PtrTy->getAddressSpace() == 0;
+ if (!IsNoWrapAddRec && !IsInBoundsGEP && !IsInAddressSpaceZero) {
DEBUG(dbgs() << "LV: Bad stride - Pointer may wrap in the address space "
<< *Ptr << " SCEV: " << *PtrScev << "\n");
return 0;
return 0;
}
- int64_t Size = DL->getTypeAllocSize(PtrTy->getPointerElementType());
+ int64_t Size = DL->getTypeAllocSize(PtrTy->getElementType());
const APInt &APStepVal = C->getValue()->getValue();
// Huge step value - give up.
return 0;
// If the SCEV could wrap but we have an inbounds gep with a unit stride we
- // know we can't "wrap around the address space".
- if (!IsNoWrapAddRec && IsInBoundsGEP && Stride != 1 && Stride != -1)
+ // know we can't "wrap around the address space". In case of address space
+ // zero we know that this won't happen without triggering undefined behavior.
+ if (!IsNoWrapAddRec && (IsInBoundsGEP || IsInAddressSpaceZero) &&
+ Stride != 1 && Stride != -1)
return 0;
return Stride;
const MemAccessInfo &B, unsigned BIdx) {
assert (AIdx < BIdx && "Must pass arguments in program order");
- Value *APtr = A.first;
- Value *BPtr = B.first;
- bool AIsWrite = A.second;
- bool BIsWrite = B.second;
+ Value *APtr = A.getPointer();
+ Value *BPtr = B.getPointer();
+ bool AIsWrite = A.getInt();
+ bool BIsWrite = B.getInt();
// Two reads are independent.
if (!AIsWrite && !BIsWrite)
const SCEVConstant *C = dyn_cast<SCEVConstant>(Dist);
if (!C) {
- DEBUG(dbgs() << "LV: Dependence because of non constant distance\n");
+ DEBUG(dbgs() << "LV: Dependence because of non-constant distance\n");
+ ShouldRetryWithRuntimeCheck = true;
return true;
}
if (Val == 0) {
if (ATy == BTy)
return false;
- DEBUG(dbgs() << "LV: Zero dependence difference but different types");
+ DEBUG(dbgs() << "LV: Zero dependence difference but different types\n");
return true;
}
// Positive distance bigger than max vectorization factor.
if (ATy != BTy) {
DEBUG(dbgs() <<
- "LV: ReadWrite-Write positive dependency with different types");
+ "LV: ReadWrite-Write positive dependency with different types\n");
return false;
}
2*TypeByteSize > MaxSafeDepDistBytes ||
Distance < TypeByteSize * ForcedUnroll * ForcedFactor) {
DEBUG(dbgs() << "LV: Failure because of Positive distance "
- << Val.getSExtValue() << "\n");
+ << Val.getSExtValue() << '\n');
return true;
}
return true;
DEBUG(dbgs() << "LV: Positive distance " << Val.getSExtValue() <<
- " with max VF=" << MaxSafeDepDistBytes/TypeByteSize << "\n");
+ " with max VF = " << MaxSafeDepDistBytes / TypeByteSize << '\n');
return false;
}
bool
MemoryDepChecker::areDepsSafe(AccessAnalysis::DepCandidates &AccessSets,
- DenseSet<MemAccessInfo> &CheckDeps) {
+ MemAccessInfoSet &CheckDeps) {
MaxSafeDepDistBytes = -1U;
while (!CheckDeps.empty()) {
typedef SmallVector<Value*, 16> ValueVector;
typedef SmallPtrSet<Value*, 16> ValueSet;
- // Stores a pair of memory access location and whether the access is a store
- // (true) or a load (false).
- typedef std::pair<Value*, char> MemAccessInfo;
- typedef DenseSet<MemAccessInfo> PtrAccessSet;
-
// Holds the Load and Store *instructions*.
ValueVector Loads;
ValueVector Stores;
// but is not a load, then we quit. Notice that we don't handle function
// calls that read or write.
if (it->mayReadFromMemory()) {
+ // Many math library functions read the rounding mode. We will only
+ // vectorize a loop if it contains known function calls that don't set
+ // the flag. Therefore, it is safe to ignore this read from memory.
+ CallInst *Call = dyn_cast<CallInst>(it);
+ if (Call && getIntrinsicIDForCall(Call, TLI))
+ continue;
+
LoadInst *Ld = dyn_cast<LoadInst>(it);
if (!Ld) return false;
if (!Ld->isSimple() && !IsAnnotatedParallel) {
Stores.push_back(St);
DepChecker.addAccess(St);
}
- } // next instr.
- } // next block.
+ } // Next instr.
+ } // Next block.
// Now we have two lists that hold the loads and the stores.
// Next, we find the pointers that they use.
return true;
}
- SmallPtrSet<Value *, 16> ReadOnlyPtr;
for (I = Loads.begin(), IE = Loads.end(); I != IE; ++I) {
LoadInst *LD = cast<LoadInst>(*I);
Value* Ptr = LD->getPointerOperand();
if (NumComparisons == 0 && NeedRTCheck)
NeedRTCheck = false;
- // Check that we did not collect too many pointers or found a unsizeable
+ // Check that we did not collect too many pointers or found an unsizeable
// pointer.
if (!CanDoRT || NumComparisons > RuntimeMemoryCheckThreshold) {
PtrRtCheck.reset();
CanVecMem = DepChecker.areDepsSafe(DependentAccesses,
Accesses.getDependenciesToCheck());
MaxSafeDepDistBytes = DepChecker.getMaxSafeDepDistBytes();
+
+ if (!CanVecMem && DepChecker.shouldRetryWithRuntimeCheck()) {
+ DEBUG(dbgs() << "LV: Retrying with memory checks\n");
+ NeedRTCheck = true;
+
+ // Clear the dependency checks. We assume they are not needed.
+ Accesses.resetDepChecks();
+
+ PtrRtCheck.reset();
+ PtrRtCheck.Need = true;
+
+ CanDoRT = Accesses.canCheckPtrAtRT(PtrRtCheck, NumComparisons, SE,
+ TheLoop, true);
+ // Check that we did not collect too many pointers or found an unsizeable
+ // pointer.
+ if (!CanDoRT || NumComparisons > RuntimeMemoryCheckThreshold) {
+ DEBUG(dbgs() << "LV: Can't vectorize with memory checks\n");
+ PtrRtCheck.reset();
+ return false;
+ }
+
+ CanVecMem = true;
+ }
}
- DEBUG(dbgs() << "LV: We "<< (NeedRTCheck ? "" : "don't") <<
+ DEBUG(dbgs() << "LV: We" << (NeedRTCheck ? "" : " don't") <<
" need a runtime memory check.\n");
return CanVecMem;
// Check whether we found a reduction operator.
FoundReduxOp |= !IsAPhi;
- // Process users of current instruction. Push non PHI nodes after PHI nodes
+ // Process users of current instruction. Push non-PHI nodes after PHI nodes
// onto the stack. This way we are going to have seen all inputs to PHI
// nodes once we get to them.
SmallVector<Instruction *, 8> NonPHIs;
// Check if we found the exit user.
BasicBlock *Parent = Usr->getParent();
if (!TheLoop->contains(Parent)) {
- // Exit if you find multiple outside users.
- if (ExitInstruction != 0)
+ // Exit if you find multiple outside users or if the header phi node is
+ // being used. In this case the user uses the value of the previous
+ // iteration, in which case we would loose "VF-1" iterations of the
+ // reduction operation if we vectorize.
+ if (ExitInstruction != 0 || Cur == Phi)
return false;
+
+ // The instruction used by an outside user must be the last instruction
+ // before we feed back to the reduction phi. Otherwise, we loose VF-1
+ // operations on the value.
+ if (std::find(Phi->op_begin(), Phi->op_end(), Cur) == Phi->op_end())
+ return false;
+
ExitInstruction = Cur;
continue;
}
return !DT->dominates(BB, Latch);
}
-bool LoopVectorizationLegality::blockCanBePredicated(BasicBlock *BB) {
+bool LoopVectorizationLegality::blockCanBePredicated(BasicBlock *BB,
+ SmallPtrSet<Value *, 8>& SafePtrs) {
for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; ++it) {
// We might be able to hoist the load.
- if (it->mayReadFromMemory() && !LoadSpeculation.isHoistableLoad(it))
- return false;
+ if (it->mayReadFromMemory()) {
+ LoadInst *LI = dyn_cast<LoadInst>(it);
+ if (!LI || !SafePtrs.count(LI->getPointerOperand()))
+ return false;
+ }
// We don't predicate stores at the moment.
if (it->mayWriteToMemory() || it->mayThrow())
// Find the trip count.
unsigned TC = SE->getSmallConstantTripCount(TheLoop, TheLoop->getLoopLatch());
- DEBUG(dbgs() << "LV: Found trip count:"<<TC<<"\n");
+ DEBUG(dbgs() << "LV: Found trip count: " << TC << '\n');
unsigned WidestType = getWidestType();
unsigned WidestRegister = TTI.getRegisterBitWidth(true);
unsigned MaxSafeDepDist = -1U;
if (Legal->getMaxSafeDepDistBytes() != -1U)
MaxSafeDepDist = Legal->getMaxSafeDepDistBytes() * 8;
- WidestRegister = WidestRegister < MaxSafeDepDist ? WidestRegister : MaxSafeDepDist;
+ WidestRegister = ((WidestRegister < MaxSafeDepDist) ?
+ WidestRegister : MaxSafeDepDist);
unsigned MaxVectorSize = WidestRegister / WidestType;
DEBUG(dbgs() << "LV: The Widest type: " << WidestType << " bits.\n");
- DEBUG(dbgs() << "LV: The Widest register is:" << WidestRegister << "bits.\n");
+ DEBUG(dbgs() << "LV: The Widest register is: "
+ << WidestRegister << " bits.\n");
if (MaxVectorSize == 0) {
DEBUG(dbgs() << "LV: The target has no vector registers.\n");
if (UserVF != 0) {
assert(isPowerOf2_32(UserVF) && "VF needs to be a power of two");
- DEBUG(dbgs() << "LV: Using user VF "<<UserVF<<".\n");
+ DEBUG(dbgs() << "LV: Using user VF " << UserVF << ".\n");
Factor.Width = UserVF;
return Factor;
float Cost = expectedCost(1);
unsigned Width = 1;
- DEBUG(dbgs() << "LV: Scalar loop costs: "<< (int)Cost << ".\n");
+ DEBUG(dbgs() << "LV: Scalar loop costs: " << (int)Cost << ".\n");
for (unsigned i=2; i <= VF; i*=2) {
// Notice that the vector loop needs to be executed less times, so
// we need to divide the cost of the vector loops by the width of
// the vector elements.
float VectorCost = expectedCost(i) / (float)i;
- DEBUG(dbgs() << "LV: Vector loop of width "<< i << " costs: " <<
+ DEBUG(dbgs() << "LV: Vector loop of width " << i << " costs: " <<
(int)VectorCost << ".\n");
if (VectorCost < Cost) {
Cost = VectorCost;
else if (UF < 1)
UF = 1;
- if (Legal->getReductionVars()->size()) {
- DEBUG(dbgs() << "LV: Unrolling because of reductions. \n");
+ bool HasReductions = Legal->getReductionVars()->size();
+
+ // Decide if we want to unroll if we decided that it is legal to vectorize
+ // but not profitable.
+ if (VF == 1) {
+ if (TheLoop->getNumBlocks() > 1 || !HasReductions ||
+ LoopCost > SmallLoopCost)
+ return 1;
+
+ return UF;
+ }
+
+ if (HasReductions) {
+ DEBUG(dbgs() << "LV: Unrolling because of reductions.\n");
return UF;
}
// We assume that the cost overhead is 1 and we use the cost model
// to estimate the cost of the loop and unroll until the cost of the
// loop overhead is about 5% of the cost of the loop.
- DEBUG(dbgs() << "LV: Loop cost is "<< LoopCost <<" \n");
- if (LoopCost < 20) {
- DEBUG(dbgs() << "LV: Unrolling to reduce branch cost. \n");
- unsigned NewUF = 20/LoopCost + 1;
+ DEBUG(dbgs() << "LV: Loop cost is " << LoopCost << '\n');
+ if (LoopCost < SmallLoopCost) {
+ DEBUG(dbgs() << "LV: Unrolling to reduce branch cost.\n");
+ unsigned NewUF = SmallLoopCost / (LoopCost + 1);
return std::min(NewUF, UF);
}
- DEBUG(dbgs() << "LV: Not Unrolling. \n");
+ DEBUG(dbgs() << "LV: Not Unrolling.\n");
return 1;
}
MaxUsage = std::max(MaxUsage, OpenIntervals.size());
DEBUG(dbgs() << "LV(REG): At #" << i << " Interval # " <<
- OpenIntervals.size() <<"\n");
+ OpenIntervals.size() << '\n');
// Add the current instruction to the list of open intervals.
OpenIntervals.insert(I);
}
unsigned Invariant = LoopInvariants.size();
- DEBUG(dbgs() << "LV(REG): Found max usage: " << MaxUsage << " \n");
- DEBUG(dbgs() << "LV(REG): Found invariant usage: " << Invariant << " \n");
- DEBUG(dbgs() << "LV(REG): LoopSize: " << R.NumInstructions << " \n");
+ DEBUG(dbgs() << "LV(REG): Found max usage: " << MaxUsage << '\n');
+ DEBUG(dbgs() << "LV(REG): Found invariant usage: " << Invariant << '\n');
+ DEBUG(dbgs() << "LV(REG): LoopSize: " << R.NumInstructions << '\n');
R.LoopInvariantRegs = Invariant;
R.MaxLocalUsers = MaxUsage;
continue;
unsigned C = getInstructionCost(it, VF);
- Cost += C;
- DEBUG(dbgs() << "LV: Found an estimated cost of "<< C <<" for VF " <<
- VF << " For instruction: "<< *it << "\n");
+ BlockCost += C;
+ DEBUG(dbgs() << "LV: Found an estimated cost of " << C << " for VF " <<
+ VF << " For instruction: " << *it << '\n');
}
// We assume that if-converted blocks have a 50% chance of being executed.
// When the code is scalar then some of the blocks are avoided due to CF.
// When the code is vectorized we execute all code paths.
- if (Legal->blockNeedsPredication(*bb) && VF == 1)
+ if (VF == 1 && Legal->blockNeedsPredication(*bb))
BlockCost /= 2;
Cost += BlockCost;
return Cost;
}
+/// \brief Check whether the address computation for a non-consecutive memory
+/// access looks like an unlikely candidate for being merged into the indexing
+/// mode.
+///
+/// We look for a GEP which has one index that is an induction variable and all
+/// other indices are loop invariant. If the stride of this access is also
+/// within a small bound we decide that this address computation can likely be
+/// merged into the addressing mode.
+/// In all other cases, we identify the address computation as complex.
+static bool isLikelyComplexAddressComputation(Value *Ptr,
+ LoopVectorizationLegality *Legal,
+ ScalarEvolution *SE,
+ const Loop *TheLoop) {
+ GetElementPtrInst *Gep = dyn_cast<GetElementPtrInst>(Ptr);
+ if (!Gep)
+ return true;
+
+ // We are looking for a gep with all loop invariant indices except for one
+ // which should be an induction variable.
+ unsigned NumOperands = Gep->getNumOperands();
+ for (unsigned i = 1; i < NumOperands; ++i) {
+ Value *Opd = Gep->getOperand(i);
+ if (!SE->isLoopInvariant(SE->getSCEV(Opd), TheLoop) &&
+ !Legal->isInductionVariable(Opd))
+ return true;
+ }
+
+ // Now we know we have a GEP ptr, %inv, %ind, %inv. Make sure that the step
+ // can likely be merged into the address computation.
+ unsigned MaxMergeDistance = 64;
+
+ const SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(SE->getSCEV(Ptr));
+ if (!AddRec)
+ return true;
+
+ // Check the step is constant.
+ const SCEV *Step = AddRec->getStepRecurrence(*SE);
+ // Calculate the pointer stride and check if it is consecutive.
+ const SCEVConstant *C = dyn_cast<SCEVConstant>(Step);
+ if (!C)
+ return true;
+
+ const APInt &APStepVal = C->getValue()->getValue();
+
+ // Huge step value - give up.
+ if (APStepVal.getBitWidth() > 64)
+ return true;
+
+ int64_t StepVal = APStepVal.getSExtValue();
+
+ return StepVal > MaxMergeDistance;
+}
+
unsigned
LoopVectorizationCostModel::getInstructionCost(Instruction *I, unsigned VF) {
// If we know that this instruction will remain uniform, check the cost of
unsigned ScalarAllocatedSize = DL->getTypeAllocSize(ValTy);
unsigned VectorElementSize = DL->getTypeStoreSize(VectorTy)/VF;
if (!ConsecutiveStride || ScalarAllocatedSize != VectorElementSize) {
+ bool IsComplexComputation =
+ isLikelyComplexAddressComputation(Ptr, Legal, SE, TheLoop);
unsigned Cost = 0;
// The cost of extracting from the value vector and pointer vector.
Type *PtrTy = ToVectorTy(Ptr->getType(), VF);
}
// The cost of the scalar loads/stores.
- Cost += VF * TTI.getAddressComputationCost(ValTy->getScalarType());
+ Cost += VF * TTI.getAddressComputationCost(PtrTy, IsComplexComputation);
Cost += VF * TTI.getMemoryOpCost(I->getOpcode(), ValTy->getScalarType(),
Alignment, AS);
return Cost;
static const char lv_name[] = "Loop Vectorization";
INITIALIZE_PASS_BEGIN(LoopVectorize, LV_NAME, lv_name, false, false)
INITIALIZE_AG_DEPENDENCY(TargetTransformInfo)
+INITIALIZE_PASS_DEPENDENCY(DominatorTree)
INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)
+INITIALIZE_PASS_DEPENDENCY(LCSSA)
+INITIALIZE_PASS_DEPENDENCY(LoopInfo)
INITIALIZE_PASS_DEPENDENCY(LoopSimplify)
INITIALIZE_PASS_END(LoopVectorize, LV_NAME, lv_name, false, false)
namespace llvm {
- Pass *createLoopVectorizePass() {
- return new LoopVectorize();
+ Pass *createLoopVectorizePass(bool NoUnrolling, bool AlwaysVectorize) {
+ return new LoopVectorize(NoUnrolling, AlwaysVectorize);
}
}
return false;
}
+
+
+void InnerLoopUnroller::scalarizeInstruction(Instruction *Instr) {
+ assert(!Instr->getType()->isAggregateType() && "Can't handle vectors");
+ // Holds vector parameters or scalars, in case of uniform vals.
+ SmallVector<VectorParts, 4> Params;
+
+ setDebugLocFromInst(Builder, Instr);
+
+ // Find all of the vectorized parameters.
+ for (unsigned op = 0, e = Instr->getNumOperands(); op != e; ++op) {
+ Value *SrcOp = Instr->getOperand(op);
+
+ // If we are accessing the old induction variable, use the new one.
+ if (SrcOp == OldInduction) {
+ Params.push_back(getVectorValue(SrcOp));
+ continue;
+ }
+
+ // Try using previously calculated values.
+ Instruction *SrcInst = dyn_cast<Instruction>(SrcOp);
+
+ // If the src is an instruction that appeared earlier in the basic block
+ // then it should already be vectorized.
+ if (SrcInst && OrigLoop->contains(SrcInst)) {
+ assert(WidenMap.has(SrcInst) && "Source operand is unavailable");
+ // The parameter is a vector value from earlier.
+ Params.push_back(WidenMap.get(SrcInst));
+ } else {
+ // The parameter is a scalar from outside the loop. Maybe even a constant.
+ VectorParts Scalars;
+ Scalars.append(UF, SrcOp);
+ Params.push_back(Scalars);
+ }
+ }
+
+ assert(Params.size() == Instr->getNumOperands() &&
+ "Invalid number of operands");
+
+ // Does this instruction return a value ?
+ bool IsVoidRetTy = Instr->getType()->isVoidTy();
+
+ Value *UndefVec = IsVoidRetTy ? 0 :
+ UndefValue::get(Instr->getType());
+ // Create a new entry in the WidenMap and initialize it to Undef or Null.
+ VectorParts &VecResults = WidenMap.splat(Instr, UndefVec);
+
+ // For each vector unroll 'part':
+ for (unsigned Part = 0; Part < UF; ++Part) {
+ // For each scalar that we create:
+
+ Instruction *Cloned = Instr->clone();
+ if (!IsVoidRetTy)
+ Cloned->setName(Instr->getName() + ".cloned");
+ // Replace the operands of the cloned instructions with extracted scalars.
+ for (unsigned op = 0, e = Instr->getNumOperands(); op != e; ++op) {
+ Value *Op = Params[op][Part];
+ Cloned->setOperand(op, Op);
+ }
+
+ // Place the cloned scalar in the new loop.
+ Builder.Insert(Cloned);
+
+ // If the original scalar returns a value we need to place it in a vector
+ // so that future users will be able to use it.
+ if (!IsVoidRetTy)
+ VecResults[Part] = Cloned;
+ }
+}
+
+void
+InnerLoopUnroller::vectorizeMemoryInstruction(Instruction *Instr,
+ LoopVectorizationLegality*) {
+ return scalarizeInstruction(Instr);
+}
+
+Value *InnerLoopUnroller::reverseVector(Value *Vec) {
+ return Vec;
+}
+
+Value *InnerLoopUnroller::getBroadcastInstrs(Value *V) {
+ return V;
+}
+
+Value *InnerLoopUnroller::getConsecutiveVector(Value* Val, int StartIdx,
+ bool Negate) {
+ // When unrolling and the VF is 1, we only need to add a simple scalar.
+ Type *ITy = Val->getType();
+ assert(!ITy->isVectorTy() && "Val must be a scalar");
+ Constant *C = ConstantInt::get(ITy, StartIdx, Negate);
+ return Builder.CreateAdd(Val, C, "induction");
+}
+