#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
using namespace llvm;
/// TODO: When ScalarEvolution gets a SCEVSDivExpr, this can be made
/// unnecessary; in its place, just signed-divide Ops[i] by the scale and
/// check to see if the divide was folded.
-static bool FactorOutConstant(const SCEV *&S,
- const SCEV *&Remainder,
- const SCEV *Factor,
- ScalarEvolution &SE,
- const DataLayout *DL) {
+static bool FactorOutConstant(const SCEV *&S, const SCEV *&Remainder,
+ const SCEV *Factor, ScalarEvolution &SE,
+ const DataLayout &DL) {
// Everything is divisible by one.
if (Factor->isOne())
return true;
// In a Mul, check if there is a constant operand which is a multiple
// of the given factor.
if (const SCEVMulExpr *M = dyn_cast<SCEVMulExpr>(S)) {
- if (DL) {
- // With DataLayout, the size is known. Check if there is a constant
- // operand which is a multiple of the given factor. If so, we can
- // factor it.
- const SCEVConstant *FC = cast<SCEVConstant>(Factor);
- if (const SCEVConstant *C = dyn_cast<SCEVConstant>(M->getOperand(0)))
- if (!C->getValue()->getValue().srem(FC->getValue()->getValue())) {
- SmallVector<const SCEV *, 4> NewMulOps(M->op_begin(), M->op_end());
- NewMulOps[0] =
- SE.getConstant(C->getValue()->getValue().sdiv(
- FC->getValue()->getValue()));
- S = SE.getMulExpr(NewMulOps);
- return true;
- }
- } else {
- // Without DataLayout, check if Factor can be factored out of any of the
- // Mul's operands. If so, we can just remove it.
- for (unsigned i = 0, e = M->getNumOperands(); i != e; ++i) {
- const SCEV *SOp = M->getOperand(i);
- const SCEV *Remainder = SE.getConstant(SOp->getType(), 0);
- if (FactorOutConstant(SOp, Remainder, Factor, SE, DL) &&
- Remainder->isZero()) {
- SmallVector<const SCEV *, 4> NewMulOps(M->op_begin(), M->op_end());
- NewMulOps[i] = SOp;
- S = SE.getMulExpr(NewMulOps);
- return true;
- }
+ // Size is known, check if there is a constant operand which is a multiple
+ // of the given factor. If so, we can factor it.
+ const SCEVConstant *FC = cast<SCEVConstant>(Factor);
+ if (const SCEVConstant *C = dyn_cast<SCEVConstant>(M->getOperand(0)))
+ if (!C->getValue()->getValue().srem(FC->getValue()->getValue())) {
+ SmallVector<const SCEV *, 4> NewMulOps(M->op_begin(), M->op_end());
+ NewMulOps[0] = SE.getConstant(
+ C->getValue()->getValue().sdiv(FC->getValue()->getValue()));
+ S = SE.getMulExpr(NewMulOps);
+ return true;
}
- }
}
// In an AddRec, check if both start and step are divisible.
PointerType *PTy,
Type *Ty,
Value *V) {
- Type *ElTy = PTy->getElementType();
+ Type *OriginalElTy = PTy->getElementType();
+ Type *ElTy = OriginalElTy;
SmallVector<Value *, 4> GepIndices;
SmallVector<const SCEV *, 8> Ops(op_begin, op_end);
bool AnyNonZeroIndices = false;
// without the other.
SplitAddRecs(Ops, Ty, SE);
- Type *IntPtrTy = SE.DL
- ? SE.DL->getIntPtrType(PTy)
- : Type::getInt64Ty(PTy->getContext());
+ Type *IntPtrTy = DL.getIntPtrType(PTy);
// Descend down the pointer's type and attempt to convert the other
// operands into GEP indices, at each level. The first index in a GEP
for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
const SCEV *Op = Ops[i];
const SCEV *Remainder = SE.getConstant(Ty, 0);
- if (FactorOutConstant(Op, Remainder, ElSize, SE, SE.DL)) {
+ if (FactorOutConstant(Op, Remainder, ElSize, SE, DL)) {
// Op now has ElSize factored out.
ScaledOps.push_back(Op);
if (!Remainder->isZero())
bool FoundFieldNo = false;
// An empty struct has no fields.
if (STy->getNumElements() == 0) break;
- if (SE.DL) {
- // With DataLayout, field offsets are known. See if a constant offset
- // falls within any of the struct fields.
- if (Ops.empty()) break;
- if (const SCEVConstant *C = dyn_cast<SCEVConstant>(Ops[0]))
- if (SE.getTypeSizeInBits(C->getType()) <= 64) {
- const StructLayout &SL = *SE.DL->getStructLayout(STy);
- uint64_t FullOffset = C->getValue()->getZExtValue();
- if (FullOffset < SL.getSizeInBytes()) {
- unsigned ElIdx = SL.getElementContainingOffset(FullOffset);
- GepIndices.push_back(
- ConstantInt::get(Type::getInt32Ty(Ty->getContext()), ElIdx));
- ElTy = STy->getTypeAtIndex(ElIdx);
- Ops[0] =
+ // Field offsets are known. See if a constant offset falls within any of
+ // the struct fields.
+ if (Ops.empty())
+ break;
+ if (const SCEVConstant *C = dyn_cast<SCEVConstant>(Ops[0]))
+ if (SE.getTypeSizeInBits(C->getType()) <= 64) {
+ const StructLayout &SL = *DL.getStructLayout(STy);
+ uint64_t FullOffset = C->getValue()->getZExtValue();
+ if (FullOffset < SL.getSizeInBytes()) {
+ unsigned ElIdx = SL.getElementContainingOffset(FullOffset);
+ GepIndices.push_back(
+ ConstantInt::get(Type::getInt32Ty(Ty->getContext()), ElIdx));
+ ElTy = STy->getTypeAtIndex(ElIdx);
+ Ops[0] =
SE.getConstant(Ty, FullOffset - SL.getElementOffset(ElIdx));
- AnyNonZeroIndices = true;
- FoundFieldNo = true;
- }
- }
- } else {
- // Without DataLayout, just check for an offsetof expression of the
- // appropriate struct type.
- for (unsigned i = 0, e = Ops.size(); i != e; ++i)
- if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(Ops[i])) {
- Type *CTy;
- Constant *FieldNo;
- if (U->isOffsetOf(CTy, FieldNo) && CTy == STy) {
- GepIndices.push_back(FieldNo);
- ElTy =
- STy->getTypeAtIndex(cast<ConstantInt>(FieldNo)->getZExtValue());
- Ops[i] = SE.getConstant(Ty, 0);
- AnyNonZeroIndices = true;
- FoundFieldNo = true;
- break;
- }
+ AnyNonZeroIndices = true;
+ FoundFieldNo = true;
}
- }
+ }
// If no struct field offsets were found, tentatively assume that
// field zero was selected (since the zero offset would obviously
// be folded away).
// Fold a GEP with constant operands.
if (Constant *CLHS = dyn_cast<Constant>(V))
if (Constant *CRHS = dyn_cast<Constant>(Idx))
- return ConstantExpr::getGetElementPtr(CLHS, CRHS);
+ return ConstantExpr::getGetElementPtr(Type::getInt8Ty(Ty->getContext()),
+ CLHS, CRHS);
// Do a quick scan to see if we have this GEP nearby. If so, reuse it.
unsigned ScanLimit = 6;
}
// Emit a GEP.
- Value *GEP = Builder.CreateGEP(V, Idx, "uglygep");
+ Value *GEP = Builder.CreateGEP(Builder.getInt8Ty(), V, Idx, "uglygep");
rememberInstruction(GEP);
return GEP;
Value *Casted = V;
if (V->getType() != PTy)
Casted = InsertNoopCastOfTo(Casted, PTy);
- Value *GEP = Builder.CreateGEP(Casted,
+ Value *GEP = Builder.CreateGEP(OriginalElTy, Casted,
GepIndices,
"scevgep");
Ops.push_back(SE.getUnknown(GEP));
}
}
- bool IncrementIsNUW = IsIncrementNUW(SE, Normalized);
- bool IncrementIsNSW = IsIncrementNSW(SE, Normalized);
-
// Save the original insertion point so we can restore it when we're done.
BuilderType::InsertPointGuard Guard(Builder);
// Expand the step somewhere that dominates the loop header.
Value *StepV = expandCodeFor(Step, IntTy, L->getHeader()->begin());
+ // The no-wrap behavior proved by IsIncrement(NUW|NSW) is only applicable if
+ // we actually do emit an addition. It does not apply if we emit a
+ // subtraction.
+ bool IncrementIsNUW = !useSubtract && IsIncrementNUW(SE, Normalized);
+ bool IncrementIsNSW = !useSubtract && IsIncrementNSW(SE, Normalized);
+
// Create the PHI.
BasicBlock *Header = L->getHeader();
Builder.SetInsertPoint(Header, Header->begin());
// Fold constant phis. They may be congruent to other constant phis and
// would confuse the logic below that expects proper IVs.
- if (Value *V = SimplifyInstruction(Phi, SE.DL, SE.TLI, SE.DT, SE.AC)) {
+ if (Value *V = SimplifyInstruction(Phi, DL, SE.TLI, SE.DT, SE.AC)) {
Phi->replaceAllUsesWith(V);
DeadInsts.push_back(Phi);
++NumElim;
<< *IsomorphicInc << '\n');
Value *NewInc = OrigInc;
if (OrigInc->getType() != IsomorphicInc->getType()) {
- Instruction *IP = isa<PHINode>(OrigInc)
- ? (Instruction*)L->getHeader()->getFirstInsertionPt()
- : OrigInc->getNextNode();
+ Instruction *IP = nullptr;
+ if (PHINode *PN = dyn_cast<PHINode>(OrigInc))
+ IP = PN->getParent()->getFirstInsertionPt();
+ else
+ IP = OrigInc->getNextNode();
+
IRBuilder<> Builder(IP);
Builder.SetCurrentDebugLocation(IsomorphicInc->getDebugLoc());
NewInc = Builder.
projects / oota-llvm.git / blobdiff