#include "llvm/ADT/StringExtras.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/AliasSetTracker.h"
+#include "llvm/Analysis/AssumptionTracker.h"
#include "llvm/Analysis/BlockFrequencyInfo.h"
+#include "llvm/Analysis/CodeMetrics.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/LoopIterator.h"
#include "llvm/Analysis/LoopPass.h"
cl::desc("Sets the SIMD width. Zero is autoselect."));
static cl::opt<unsigned>
-VectorizationUnroll("force-vector-unroll", cl::init(0), cl::Hidden,
- cl::desc("Sets the vectorization unroll count. "
+VectorizationInterleave("force-vector-interleave", cl::init(0), cl::Hidden,
+ cl::desc("Sets the vectorization interleave count. "
"Zero is autoselect."));
static cl::opt<bool>
"force-target-num-vector-regs", cl::init(0), cl::Hidden,
cl::desc("A flag that overrides the target's number of vector registers."));
-/// Maximum vectorization unroll count.
-static const unsigned MaxUnrollFactor = 16;
+/// Maximum vectorization interleave count.
+static const unsigned MaxInterleaveFactor = 16;
-static cl::opt<unsigned> ForceTargetMaxScalarUnrollFactor(
- "force-target-max-scalar-unroll", cl::init(0), cl::Hidden,
- cl::desc("A flag that overrides the target's max unroll factor for scalar "
- "loops."));
+static cl::opt<unsigned> ForceTargetMaxScalarInterleaveFactor(
+ "force-target-max-scalar-interleave", cl::init(0), cl::Hidden,
+ cl::desc("A flag that overrides the target's max interleave factor for "
+ "scalar loops."));
-static cl::opt<unsigned> ForceTargetMaxVectorUnrollFactor(
- "force-target-max-vector-unroll", cl::init(0), cl::Hidden,
- cl::desc("A flag that overrides the target's max unroll factor for "
+static cl::opt<unsigned> ForceTargetMaxVectorInterleaveFactor(
+ "force-target-max-vector-interleave", cl::init(0), cl::Hidden,
+ cl::desc("A flag that overrides the target's max interleave factor for "
"vectorized loops."));
static cl::opt<unsigned> ForceTargetInstructionCost(
"enable-cond-stores-vec", cl::init(false), cl::Hidden,
cl::desc("Enable if predication of stores during vectorization."));
+static cl::opt<unsigned> MaxNestedScalarReductionUF(
+ "max-nested-scalar-reduction-unroll", cl::init(2), cl::Hidden,
+ cl::desc("The maximum unroll factor to use when unrolling a scalar "
+ "reduction in a nested loop."));
+
namespace {
// Forward declarations.
/// Return true if all of the instructions in the block can be speculatively
/// 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);
+ bool blockCanBePredicated(BasicBlock *BB, SmallPtrSetImpl<Value *> &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.
LoopVectorizationLegality *Legal,
const TargetTransformInfo &TTI,
const DataLayout *DL, const TargetLibraryInfo *TLI,
- const Function *F, const LoopVectorizeHints *Hints)
- : TheLoop(L), SE(SE), LI(LI), Legal(Legal), TTI(TTI), DL(DL), TLI(TLI), TheFunction(F), Hints(Hints) {}
+ AssumptionTracker *AT, const Function *F,
+ const LoopVectorizeHints *Hints)
+ : TheLoop(L), SE(SE), LI(LI), Legal(Legal), TTI(TTI), DL(DL), TLI(TLI),
+ TheFunction(F), Hints(Hints) {
+ CodeMetrics::collectEphemeralValues(L, AT, EphValues);
+ }
/// Information about vectorization costs
struct VectorizationFactor {
*TheFunction, DL, Message.str());
}
+ /// Values used only by @llvm.assume calls.
+ SmallPtrSet<const Value *, 32> EphValues;
+
/// The loop that we evaluate.
Loop *TheLoop;
/// Scev analysis.
case HK_WIDTH:
return isPowerOf2_32(Val) && Val <= MaxVectorWidth;
case HK_UNROLL:
- return isPowerOf2_32(Val) && Val <= MaxUnrollFactor;
+ return isPowerOf2_32(Val) && Val <= MaxInterleaveFactor;
case HK_FORCE:
return (Val <= 1);
}
/// Vectorization width.
Hint Width;
- /// Vectorization unroll factor.
- Hint Unroll;
+ /// Vectorization interleave factor.
+ Hint Interleave;
/// Vectorization forced
Hint Force;
- /// Array to help iterating through all hints.
- Hint *Hints[3] = { &Width, &Unroll, &Force };
/// Return the loop metadata prefix.
static StringRef Prefix() { return "llvm.loop."; }
FK_Enabled = 1, ///< Forcing enabled.
};
- LoopVectorizeHints(const Loop *L, bool DisableUnrolling)
+ LoopVectorizeHints(const Loop *L, bool DisableInterleaving)
: Width("vectorize.width", VectorizationFactor, HK_WIDTH),
- Unroll("interleave.count", DisableUnrolling, HK_UNROLL),
+ Interleave("interleave.count", DisableInterleaving, HK_UNROLL),
Force("vectorize.enable", FK_Undefined, HK_FORCE),
TheLoop(L) {
// Populate values with existing loop metadata.
getHintsFromMetadata();
- // force-vector-unroll overrides DisableUnrolling.
- if (VectorizationUnroll.getNumOccurrences() > 0)
- Unroll.Value = VectorizationUnroll;
+ // force-vector-interleave overrides DisableInterleaving.
+ if (VectorizationInterleave.getNumOccurrences() > 0)
+ Interleave.Value = VectorizationInterleave;
- DEBUG(if (DisableUnrolling && Unroll.Value == 1) dbgs()
- << "LV: Unrolling disabled by the pass manager\n");
+ DEBUG(if (DisableInterleaving && Interleave.Value == 1) dbgs()
+ << "LV: Interleaving disabled by the pass manager\n");
}
/// Mark the loop L as already vectorized by setting the width to 1.
void setAlreadyVectorized() {
- Width.Value = Unroll.Value = 1;
- writeHintsToMetadata({ Width, Unroll });
+ Width.Value = Interleave.Value = 1;
+ Hint Hints[] = {Width, Interleave};
+ writeHintsToMetadata(Hints);
}
/// Dumps all the hint information.
R << " (Force=true";
if (Width.Value != 0)
R << ", Vector Width=" << Width.Value;
- if (Unroll.Value != 0)
- R << ", Interleave Count=" << Unroll.Value;
+ if (Interleave.Value != 0)
+ R << ", Interleave Count=" << Interleave.Value;
R << ")";
}
}
}
unsigned getWidth() const { return Width.Value; }
- unsigned getUnroll() const { return Unroll.Value; }
+ unsigned getInterleave() const { return Interleave.Value; }
enum ForceKind getForce() const { return (ForceKind)Force.Value; }
private:
if (!C) return;
unsigned Val = C->getZExtValue();
+ Hint *Hints[] = {&Width, &Interleave, &Force};
for (auto H : Hints) {
if (Name == H->Name) {
if (H->validate(Val))
/// Create a new hint from name / value pair.
MDNode *createHintMetadata(StringRef Name, unsigned V) const {
LLVMContext &Context = TheLoop->getHeader()->getContext();
- SmallVector<Value*, 2> Vals;
- Vals.push_back(MDString::get(Context, Name));
- Vals.push_back(ConstantInt::get(Type::getInt32Ty(Context), V));
+ Value *Vals[] = {MDString::get(Context, Name),
+ ConstantInt::get(Type::getInt32Ty(Context), V)};
return MDNode::get(Context, Vals);
}
/// Matches metadata with hint name.
- bool matchesHintMetadataName(MDNode *Node, std::vector<Hint> &HintTypes) {
+ bool matchesHintMetadataName(MDNode *Node, ArrayRef<Hint> HintTypes) {
MDString* Name = dyn_cast<MDString>(Node->getOperand(0));
if (!Name)
return false;
}
/// Sets current hints into loop metadata, keeping other values intact.
- void writeHintsToMetadata(std::vector<Hint> HintTypes) {
+ void writeHintsToMetadata(ArrayRef<Hint> HintTypes) {
if (HintTypes.size() == 0)
return;
emitLoopVectorizeWarning(
F->getContext(), *F, L->getStartLoc(),
"failed explicitly specified loop vectorization");
- else if (LH.getUnroll() != 1)
+ else if (LH.getInterleave() != 1)
emitLoopInterleaveWarning(
F->getContext(), *F, L->getStartLoc(),
"failed explicitly specified loop interleaving");
BlockFrequencyInfo *BFI;
TargetLibraryInfo *TLI;
AliasAnalysis *AA;
+ AssumptionTracker *AT;
bool DisableUnrolling;
bool AlwaysVectorize;
BFI = &getAnalysis<BlockFrequencyInfo>();
TLI = getAnalysisIfAvailable<TargetLibraryInfo>();
AA = &getAnalysis<AliasAnalysis>();
+ AT = &getAnalysis<AssumptionTracker>();
// Compute some weights outside of the loop over the loops. Compute this
// using a BranchProbability to re-use its scaling math.
: (Hints.getForce() == LoopVectorizeHints::FK_Enabled
? "enabled"
: "?")) << " width=" << Hints.getWidth()
- << " unroll=" << Hints.getUnroll() << "\n");
+ << " unroll=" << Hints.getInterleave() << "\n");
// Function containing loop
Function *F = L->getHeader()->getParent();
return false;
}
- if (Hints.getWidth() == 1 && Hints.getUnroll() == 1) {
+ if (Hints.getWidth() == 1 && Hints.getInterleave() == 1) {
DEBUG(dbgs() << "LV: Not vectorizing: Disabled/already vectorized.\n");
emitOptimizationRemarkAnalysis(
F->getContext(), DEBUG_TYPE, *F, L->getStartLoc(),
// Check the loop for a trip count threshold:
// do not vectorize loops with a tiny trip count.
- BasicBlock *Latch = L->getLoopLatch();
- const unsigned TC = SE->getSmallConstantTripCount(L, Latch);
+ const unsigned TC = SE->getSmallConstantTripCount(L);
if (TC > 0u && TC < TinyTripCountVectorThreshold) {
DEBUG(dbgs() << "LV: Found a loop with a very small trip count. "
<< "This loop is not worth vectorizing.");
}
// Use the cost model.
- LoopVectorizationCostModel CM(L, SE, LI, &LVL, *TTI, DL, TLI, F, &Hints);
+ LoopVectorizationCostModel CM(L, SE, LI, &LVL, *TTI, DL, TLI, AT, F,
+ &Hints);
// Check the function attributes to find out if this function should be
// optimized for size.
}
void getAnalysisUsage(AnalysisUsage &AU) const override {
+ AU.addRequired<AssumptionTracker>();
AU.addRequiredID(LoopSimplifyID);
AU.addRequiredID(LCSSAID);
AU.addRequired<BlockFrequencyInfo>();
for (unsigned Part = 0; Part < UF; ++Part) {
Value *V = Builder.CreateBinOp(BinOp->getOpcode(), A[Part], B[Part]);
- // Update the NSW, NUW and Exact flags. Notice: V can be an Undef.
- BinaryOperator *VecOp = dyn_cast<BinaryOperator>(V);
- if (VecOp && isa<OverflowingBinaryOperator>(BinOp)) {
- VecOp->setHasNoSignedWrap(BinOp->hasNoSignedWrap());
- VecOp->setHasNoUnsignedWrap(BinOp->hasNoUnsignedWrap());
- }
- if (VecOp && isa<PossiblyExactOperator>(VecOp))
- VecOp->setIsExact(BinOp->isExact());
-
- // Copy the fast-math flags.
- if (VecOp && isa<FPMathOperator>(V))
- VecOp->setFastMathFlags(it->getFastMathFlags());
-
+ if (BinaryOperator *VecOp = dyn_cast<BinaryOperator>(V))
+ VecOp->copyIRFlags(BinOp);
+
Entry[Part] = V;
}
Intrinsic::ID ID = getIntrinsicIDForCall(CI, TLI);
assert(ID && "Not an intrinsic call!");
switch (ID) {
+ case Intrinsic::assume:
case Intrinsic::lifetime_end:
case Intrinsic::lifetime_start:
scalarizeInstruction(it);
/// \brief Check that the instruction has outside loop users and is not an
/// identified reduction variable.
static bool hasOutsideLoopUser(const Loop *TheLoop, Instruction *Inst,
- SmallPtrSet<Value *, 4> &Reductions) {
+ SmallPtrSetImpl<Value *> &Reductions) {
// Reduction instructions are allowed to have exit users. All other
// instructions must not have external users.
if (!Reductions.count(Inst))
// Bail out early if passed-in parameters make vectorization not feasible.
unsigned ForcedFactor = VectorizationFactor ? VectorizationFactor : 1;
- unsigned ForcedUnroll = VectorizationUnroll ? VectorizationUnroll : 1;
+ unsigned ForcedUnroll = VectorizationInterleave ? VectorizationInterleave : 1;
// The distance must be bigger than the size needed for a vectorized version
// of the operation and the size of the vectorized operation must not be
}
static bool hasMultipleUsesOf(Instruction *I,
- SmallPtrSet<Instruction *, 8> &Insts) {
+ SmallPtrSetImpl<Instruction *> &Insts) {
unsigned NumUses = 0;
for(User::op_iterator Use = I->op_begin(), E = I->op_end(); Use != E; ++Use) {
if (Insts.count(dyn_cast<Instruction>(*Use)))
return false;
}
-static bool areAllUsesIn(Instruction *I, SmallPtrSet<Instruction *, 8> &Set) {
+static bool areAllUsesIn(Instruction *I, SmallPtrSetImpl<Instruction *> &Set) {
for(User::op_iterator Use = I->op_begin(), E = I->op_end(); Use != E; ++Use)
if (!Set.count(dyn_cast<Instruction>(*Use)))
return false;
}
bool LoopVectorizationLegality::blockCanBePredicated(BasicBlock *BB,
- SmallPtrSet<Value *, 8>& SafePtrs) {
+ SmallPtrSetImpl<Value *> &SafePtrs) {
for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; ++it) {
// We might be able to hoist the load.
if (it->mayReadFromMemory()) {
}
// Find the trip count.
- unsigned TC = SE->getSmallConstantTripCount(TheLoop, TheLoop->getLoopLatch());
+ unsigned TC = SE->getSmallConstantTripCount(TheLoop);
DEBUG(dbgs() << "LV: Found trip count: " << TC << '\n');
unsigned WidestType = getWidestType();
for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; ++it) {
Type *T = it->getType();
+ // Ignore ephemeral values.
+ if (EphValues.count(it))
+ continue;
+
// Only examine Loads, Stores and PHINodes.
if (!isa<LoadInst>(it) && !isa<StoreInst>(it) && !isa<PHINode>(it))
continue;
// -- The unroll heuristics --
// We unroll the loop in order to expose ILP and reduce the loop overhead.
// There are many micro-architectural considerations that we can't predict
- // at this level. For example frontend pressure (on decode or fetch) due to
+ // at this level. For example, frontend pressure (on decode or fetch) due to
// code size, or the number and capabilities of the execution ports.
//
// We use the following heuristics to select the unroll factor:
- // 1. If the code has reductions the we unroll in order to break the cross
+ // 1. If the code has reductions, then we unroll in order to break the cross
// iteration dependency.
- // 2. If the loop is really small then we unroll in order to reduce the loop
+ // 2. If the loop is really small, then we unroll in order to reduce the loop
// overhead.
// 3. We don't unroll if we think that we will spill registers to memory due
// to the increased register pressure.
// Use the user preference, unless 'auto' is selected.
- int UserUF = Hints->getUnroll();
+ int UserUF = Hints->getInterleave();
if (UserUF != 0)
return UserUF;
- // When we optimize for size we don't unroll.
+ // When we optimize for size, we don't unroll.
if (OptForSize)
return 1;
return 1;
// Do not unroll loops with a relatively small trip count.
- unsigned TC = SE->getSmallConstantTripCount(TheLoop,
- TheLoop->getLoopLatch());
+ unsigned TC = SE->getSmallConstantTripCount(TheLoop);
if (TC > 1 && TC < TinyTripCountUnrollThreshold)
return 1;
std::max(1U, (R.MaxLocalUsers - 1)));
// Clamp the unroll factor ranges to reasonable factors.
- unsigned MaxUnrollSize = TTI.getMaximumUnrollFactor();
+ unsigned MaxInterleaveSize = TTI.getMaxInterleaveFactor();
// Check if the user has overridden the unroll max.
if (VF == 1) {
- if (ForceTargetMaxScalarUnrollFactor.getNumOccurrences() > 0)
- MaxUnrollSize = ForceTargetMaxScalarUnrollFactor;
+ if (ForceTargetMaxScalarInterleaveFactor.getNumOccurrences() > 0)
+ MaxInterleaveSize = ForceTargetMaxScalarInterleaveFactor;
} else {
- if (ForceTargetMaxVectorUnrollFactor.getNumOccurrences() > 0)
- MaxUnrollSize = ForceTargetMaxVectorUnrollFactor;
+ if (ForceTargetMaxVectorInterleaveFactor.getNumOccurrences() > 0)
+ MaxInterleaveSize = ForceTargetMaxVectorInterleaveFactor;
}
// If we did not calculate the cost for VF (because the user selected the VF)
// Clamp the calculated UF to be between the 1 and the max unroll factor
// that the target allows.
- if (UF > MaxUnrollSize)
- UF = MaxUnrollSize;
+ if (UF > MaxInterleaveSize)
+ UF = MaxInterleaveSize;
else if (UF < 1)
UF = 1;
unsigned StoresUF = UF / (Legal->NumStores ? Legal->NumStores : 1);
unsigned LoadsUF = UF / (Legal->NumLoads ? Legal->NumLoads : 1);
+ // If we have a scalar reduction (vector reductions are already dealt with
+ // by this point), we can increase the critical path length if the loop
+ // we're unrolling is inside another loop. Limit, by default to 2, so the
+ // critical path only gets increased by one reduction operation.
+ if (Legal->getReductionVars()->size() &&
+ TheLoop->getLoopDepth() > 1) {
+ unsigned F = static_cast<unsigned>(MaxNestedScalarReductionUF);
+ SmallUF = std::min(SmallUF, F);
+ StoresUF = std::min(StoresUF, F);
+ LoadsUF = std::min(LoadsUF, F);
+ }
+
if (EnableLoadStoreRuntimeUnroll && std::max(StoresUF, LoadsUF) > SmallUF) {
DEBUG(dbgs() << "LV: Unrolling to saturate store or load ports.\n");
return std::max(StoresUF, LoadsUF);
// Ignore instructions that are never used within the loop.
if (!Ends.count(I)) continue;
+ // Ignore ephemeral values.
+ if (EphValues.count(I))
+ continue;
+
// Remove all of the instructions that end at this location.
InstrList &List = TransposeEnds[i];
for (unsigned int j=0, e = List.size(); j < e; ++j)
if (isa<DbgInfoIntrinsic>(it))
continue;
+ // Ignore ephemeral values.
+ if (EphValues.count(it))
+ continue;
+
unsigned C = getInstructionCost(it, VF);
// Check if we should override the cost.
TargetTransformInfo::OK_AnyValue;
TargetTransformInfo::OperandValueKind Op2VK =
TargetTransformInfo::OK_AnyValue;
+ TargetTransformInfo::OperandValueProperties Op1VP =
+ TargetTransformInfo::OP_None;
+ TargetTransformInfo::OperandValueProperties Op2VP =
+ TargetTransformInfo::OP_None;
Value *Op2 = I->getOperand(1);
// Check for a splat of a constant or for a non uniform vector of constants.
- if (isa<ConstantInt>(Op2))
+ if (isa<ConstantInt>(Op2)) {
+ ConstantInt *CInt = cast<ConstantInt>(Op2);
+ if (CInt && CInt->getValue().isPowerOf2())
+ Op2VP = TargetTransformInfo::OP_PowerOf2;
Op2VK = TargetTransformInfo::OK_UniformConstantValue;
- else if (isa<ConstantVector>(Op2) || isa<ConstantDataVector>(Op2)) {
+ } else if (isa<ConstantVector>(Op2) || isa<ConstantDataVector>(Op2)) {
Op2VK = TargetTransformInfo::OK_NonUniformConstantValue;
- if (cast<Constant>(Op2)->getSplatValue() != nullptr)
+ Constant *SplatValue = cast<Constant>(Op2)->getSplatValue();
+ if (SplatValue) {
+ ConstantInt *CInt = dyn_cast<ConstantInt>(SplatValue);
+ if (CInt && CInt->getValue().isPowerOf2())
+ Op2VP = TargetTransformInfo::OP_PowerOf2;
Op2VK = TargetTransformInfo::OK_UniformConstantValue;
+ }
}
- return TTI.getArithmeticInstrCost(I->getOpcode(), VectorTy, Op1VK, Op2VK);
+ return TTI.getArithmeticInstrCost(I->getOpcode(), VectorTy, Op1VK, Op2VK,
+ Op1VP, Op2VP);
}
case Instruction::Select: {
SelectInst *SI = cast<SelectInst>(I);
INITIALIZE_PASS_BEGIN(LoopVectorize, LV_NAME, lv_name, false, false)
INITIALIZE_AG_DEPENDENCY(TargetTransformInfo)
INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
+INITIALIZE_PASS_DEPENDENCY(AssumptionTracker)
INITIALIZE_PASS_DEPENDENCY(BlockFrequencyInfo)
INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)