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) {}
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.
if (isPowerOf2_32(Val) && Val <= MaxVectorWidth)
Width = Val;
else
- DEBUG(dbgs() << "LV: ignoring invalid width hint metadata");
+ DEBUG(dbgs() << "LV: ignoring invalid width hint metadata\n");
} else if (Hint == "unroll") {
if (isPowerOf2_32(Val) && Val <= MaxUnrollFactor)
Unroll = Val;
else
- DEBUG(dbgs() << "LV: ignoring invalid unroll hint metadata");
+ DEBUG(dbgs() << "LV: ignoring invalid unroll hint metadata\n");
} else {
- DEBUG(dbgs() << "LV: ignoring unknown hint " << Hint);
+ DEBUG(dbgs() << "LV: ignoring unknown hint " << Hint << '\n');
}
}
};
return false;
if (DL == NULL) {
- DEBUG(dbgs() << "LV: Not vectorizing because of missing data layout");
+ DEBUG(dbgs() << "LV: Not vectorizing because of missing data layout\n");
return false;
}
}
DEBUG(dbgs() << "LV: Found a vectorizable loop ("<< VF.Width << ") in "<<
- F->getParent()->getModuleIdentifier()<<"\n");
- DEBUG(dbgs() << "LV: Unroll Factor is " << UF << "\n");
+ F->getParent()->getModuleIdentifier() << '\n');
+ DEBUG(dbgs() << "LV: Unroll Factor is " << UF << '\n');
if (VF.Width == 1) {
if (UF == 1)
int LoopVectorizationLegality::isConsecutivePtr(Value *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.
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;
}
LoopExitBlock = ExitBlock;
LoopVectorBody = VecBody;
LoopScalarBody = OldBasicBlock;
+
+ LoopVectorizeHints Hints(Lp, true);
+ Hints.setAlreadyVectorized(Lp);
}
/// This function returns the identity element (or neutral element) for
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 (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;
if (blockNeedsPredication(BB))
continue;
}
// 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()))
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.
+ 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. " <<
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;
}
}
}
// 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;
}
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;
}
!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) ||
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;
}
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.
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();
MaxSafeDepDistBytes = DepChecker.getMaxSafeDepDistBytes();
}
- DEBUG(dbgs() << "LV: We "<< (NeedRTCheck ? "" : "don't") <<
+ DEBUG(dbgs() << "LV: We" << (NeedRTCheck ? "" : " don't") <<
" need a runtime memory check.\n");
return CanVecMem;
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;
}
// 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);
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;
}
if (HasReductions) {
- DEBUG(dbgs() << "LV: Unrolling because of reductions. \n");
+ 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");
+ DEBUG(dbgs() << "LV: Loop cost is " << LoopCost << '\n');
if (LoopCost < SmallLoopCost) {
- DEBUG(dbgs() << "LV: Unrolling to reduce branch cost. \n");
+ 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;
unsigned C = getInstructionCost(it, VF);
BlockCost += C;
- DEBUG(dbgs() << "LV: Found an estimated cost of "<< C <<" for VF " <<
- VF << " For instruction: "<< *it << "\n");
+ 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.
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)