#include "llvm/ADT/STLExtras.h"
using namespace llvm;
+/// ReuseOrCreateCast - Arrange for there to be a cast of V to Ty at IP,
+/// reusing an existing cast if a suitable one exists, moving an existing
+/// cast if a suitable one exists but isn't in the right place, or
+/// creating a new one.
+Value *SCEVExpander::ReuseOrCreateCast(Value *V, const Type *Ty,
+ Instruction::CastOps Op,
+ BasicBlock::iterator IP) {
+ // Check to see if there is already a cast!
+ for (Value::use_iterator UI = V->use_begin(), E = V->use_end();
+ UI != E; ++UI) {
+ User *U = *UI;
+ if (U->getType() == Ty)
+ if (CastInst *CI = dyn_cast<CastInst>(U))
+ if (CI->getOpcode() == Op) {
+ // If the cast isn't where we want it, fix it.
+ if (BasicBlock::iterator(CI) != IP) {
+ // Create a new cast, and leave the old cast in place in case
+ // it is being used as an insert point. Clear its operand
+ // so that it doesn't hold anything live.
+ Instruction *NewCI = CastInst::Create(Op, V, Ty, "", IP);
+ NewCI->takeName(CI);
+ CI->replaceAllUsesWith(NewCI);
+ CI->setOperand(0, UndefValue::get(V->getType()));
+ rememberInstruction(NewCI);
+ return NewCI;
+ }
+ rememberInstruction(CI);
+ return CI;
+ }
+ }
+
+ // Create a new cast.
+ Instruction *I = CastInst::Create(Op, V, Ty, V->getName(), IP);
+ rememberInstruction(I);
+ return I;
+}
+
/// InsertNoopCastOfTo - Insert a cast of V to the specified type,
/// which must be possible with a noop cast, doing what we can to share
/// the casts.
return CE->getOperand(0);
}
+ // Fold a cast of a constant.
if (Constant *C = dyn_cast<Constant>(V))
return ConstantExpr::getCast(Op, C, Ty);
+ // Cast the argument at the beginning of the entry block, after
+ // any bitcasts of other arguments.
if (Argument *A = dyn_cast<Argument>(V)) {
- // Check to see if there is already a cast!
- for (Value::use_iterator UI = A->use_begin(), E = A->use_end();
- UI != E; ++UI)
- if ((*UI)->getType() == Ty)
- if (CastInst *CI = dyn_cast<CastInst>(cast<Instruction>(*UI)))
- if (CI->getOpcode() == Op) {
- // If the cast isn't the first instruction of the function, move it.
- if (BasicBlock::iterator(CI) !=
- A->getParent()->getEntryBlock().begin()) {
- // Recreate the cast at the beginning of the entry block.
- // The old cast is left in place in case it is being used
- // as an insert point.
- Instruction *NewCI =
- CastInst::Create(Op, V, Ty, "",
- A->getParent()->getEntryBlock().begin());
- NewCI->takeName(CI);
- CI->replaceAllUsesWith(NewCI);
- return NewCI;
- }
- return CI;
- }
-
- Instruction *I = CastInst::Create(Op, V, Ty, V->getName(),
- A->getParent()->getEntryBlock().begin());
- rememberInstruction(I);
- return I;
+ BasicBlock::iterator IP = A->getParent()->getEntryBlock().begin();
+ while ((isa<BitCastInst>(IP) &&
+ isa<Argument>(cast<BitCastInst>(IP)->getOperand(0)) &&
+ cast<BitCastInst>(IP)->getOperand(0) != A) ||
+ isa<DbgInfoIntrinsic>(IP))
+ ++IP;
+ return ReuseOrCreateCast(A, Ty, Op, IP);
}
+ // Cast the instruction immediately after the instruction.
Instruction *I = cast<Instruction>(V);
-
- // Check to see if there is already a cast. If there is, use it.
- for (Value::use_iterator UI = I->use_begin(), E = I->use_end();
- UI != E; ++UI) {
- if ((*UI)->getType() == Ty)
- if (CastInst *CI = dyn_cast<CastInst>(cast<Instruction>(*UI)))
- if (CI->getOpcode() == Op) {
- BasicBlock::iterator It = I; ++It;
- if (isa<InvokeInst>(I))
- It = cast<InvokeInst>(I)->getNormalDest()->begin();
- while (isa<PHINode>(It)) ++It;
- if (It != BasicBlock::iterator(CI)) {
- // Recreate the cast after the user.
- // The old cast is left in place in case it is being used
- // as an insert point.
- Instruction *NewCI = CastInst::Create(Op, V, Ty, "", It);
- NewCI->takeName(CI);
- CI->replaceAllUsesWith(NewCI);
- rememberInstruction(NewCI);
- return NewCI;
- }
- rememberInstruction(CI);
- return CI;
- }
- }
BasicBlock::iterator IP = I; ++IP;
if (InvokeInst *II = dyn_cast<InvokeInst>(I))
IP = II->getNormalDest()->begin();
- while (isa<PHINode>(IP)) ++IP;
- Instruction *CI = CastInst::Create(Op, V, Ty, V->getName(), IP);
- rememberInstruction(CI);
- return CI;
+ while (isa<PHINode>(IP) || isa<DbgInfoIntrinsic>(IP)) ++IP;
+ return ReuseOrCreateCast(I, Ty, Op, IP);
}
/// InsertBinop - Insert the specified binary operator, doing a small amount
// the sum into a single value, so just use that.
Ops.clear();
if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(Sum))
- Ops.insert(Ops.end(), Add->op_begin(), Add->op_end());
+ Ops.append(Add->op_begin(), Add->op_end());
else if (!Sum->isZero())
Ops.push_back(Sum);
// Then append the addrecs.
- Ops.insert(Ops.end(), AddRecs.begin(), AddRecs.end());
+ Ops.append(AddRecs.begin(), AddRecs.end());
}
/// SplitAddRecs - Flatten a list of add operands, moving addrec start values
A->getLoop()));
if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(Start)) {
Ops[i] = Zero;
- Ops.insert(Ops.end(), Add->op_begin(), Add->op_end());
+ Ops.append(Add->op_begin(), Add->op_end());
e += Add->getNumOperands();
} else {
Ops[i] = Start;
}
if (!AddRecs.empty()) {
// Add the addrecs onto the end of the list.
- Ops.insert(Ops.end(), AddRecs.begin(), AddRecs.end());
+ Ops.append(AddRecs.begin(), AddRecs.end());
// Resort the operand list, moving any constants to the front.
SimplifyAddOperands(Ops, Ty, SE);
}
return A; // Arbitrarily break the tie.
}
-/// GetRelevantLoop - Get the most relevant loop associated with the given
+/// getRelevantLoop - Get the most relevant loop associated with the given
/// expression, according to PickMostRelevantLoop.
-static const Loop *GetRelevantLoop(const SCEV *S, LoopInfo &LI,
- DominatorTree &DT) {
+const Loop *SCEVExpander::getRelevantLoop(const SCEV *S) {
+ // Test whether we've already computed the most relevant loop for this SCEV.
+ std::pair<DenseMap<const SCEV *, const Loop *>::iterator, bool> Pair =
+ RelevantLoops.insert(std::make_pair(S, static_cast<const Loop *>(0)));
+ if (!Pair.second)
+ return Pair.first->second;
+
if (isa<SCEVConstant>(S))
+ // A constant has no relevant loops.
return 0;
if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(S)) {
if (const Instruction *I = dyn_cast<Instruction>(U->getValue()))
- return LI.getLoopFor(I->getParent());
+ return Pair.first->second = SE.LI->getLoopFor(I->getParent());
+ // A non-instruction has no relevant loops.
return 0;
}
if (const SCEVNAryExpr *N = dyn_cast<SCEVNAryExpr>(S)) {
L = AR->getLoop();
for (SCEVNAryExpr::op_iterator I = N->op_begin(), E = N->op_end();
I != E; ++I)
- L = PickMostRelevantLoop(L, GetRelevantLoop(*I, LI, DT), DT);
- return L;
+ L = PickMostRelevantLoop(L, getRelevantLoop(*I), *SE.DT);
+ return RelevantLoops[N] = L;
+ }
+ if (const SCEVCastExpr *C = dyn_cast<SCEVCastExpr>(S)) {
+ const Loop *Result = getRelevantLoop(C->getOperand());
+ return RelevantLoops[C] = Result;
+ }
+ if (const SCEVUDivExpr *D = dyn_cast<SCEVUDivExpr>(S)) {
+ const Loop *Result =
+ PickMostRelevantLoop(getRelevantLoop(D->getLHS()),
+ getRelevantLoop(D->getRHS()),
+ *SE.DT);
+ return RelevantLoops[D] = Result;
}
- if (const SCEVCastExpr *C = dyn_cast<SCEVCastExpr>(S))
- return GetRelevantLoop(C->getOperand(), LI, DT);
- if (const SCEVUDivExpr *D = dyn_cast<SCEVUDivExpr>(S))
- return PickMostRelevantLoop(GetRelevantLoop(D->getLHS(), LI, DT),
- GetRelevantLoop(D->getRHS(), LI, DT),
- DT);
llvm_unreachable("Unexpected SCEV type!");
+ return 0;
}
namespace {
bool operator()(std::pair<const Loop *, const SCEV *> LHS,
std::pair<const Loop *, const SCEV *> RHS) const {
+ // Keep pointer operands sorted at the end.
+ if (LHS.second->getType()->isPointerTy() !=
+ RHS.second->getType()->isPointerTy())
+ return LHS.second->getType()->isPointerTy();
+
// Compare loops with PickMostRelevantLoop.
if (LHS.first != RHS.first)
return PickMostRelevantLoop(LHS.first, RHS.first, DT) != LHS.first;
SmallVector<std::pair<const Loop *, const SCEV *>, 8> OpsAndLoops;
for (std::reverse_iterator<SCEVAddExpr::op_iterator> I(S->op_end()),
E(S->op_begin()); I != E; ++I)
- OpsAndLoops.push_back(std::make_pair(GetRelevantLoop(*I, *SE.LI, *SE.DT),
- *I));
+ OpsAndLoops.push_back(std::make_pair(getRelevantLoop(*I), *I));
// Sort by loop. Use a stable sort so that constants follow non-constants and
// pointer operands precede non-pointer operands.
// The running sum expression is a pointer. Try to form a getelementptr
// at this level with that as the base.
SmallVector<const SCEV *, 4> NewOps;
- for (; I != E && I->first == CurLoop; ++I)
- NewOps.push_back(I->second);
+ for (; I != E && I->first == CurLoop; ++I) {
+ // If the operand is SCEVUnknown and not instructions, peek through
+ // it, to enable more of it to be folded into the GEP.
+ const SCEV *X = I->second;
+ if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(X))
+ if (!isa<Instruction>(U->getValue()))
+ X = SE.getSCEV(U->getValue());
+ NewOps.push_back(X);
+ }
Sum = expandAddToGEP(NewOps.begin(), NewOps.end(), PTy, Ty, Sum);
} else if (const PointerType *PTy = dyn_cast<PointerType>(Op->getType())) {
// The running sum is an integer, and there's a pointer at this level.
SmallVector<std::pair<const Loop *, const SCEV *>, 8> OpsAndLoops;
for (std::reverse_iterator<SCEVMulExpr::op_iterator> I(S->op_end()),
E(S->op_begin()); I != E; ++I)
- OpsAndLoops.push_back(std::make_pair(GetRelevantLoop(*I, *SE.LI, *SE.DT),
- *I));
+ OpsAndLoops.push_back(std::make_pair(getRelevantLoop(*I), *I));
// Sort by loop. Use a stable sort so that constants follow non-constants.
std::stable_sort(OpsAndLoops.begin(), OpsAndLoops.end(), LoopCompare(*SE.DT));
// Strip off any non-loop-dominating component from the addrec start.
const SCEV *Start = Normalized->getStart();
const SCEV *PostLoopOffset = 0;
- if (!Start->properlyDominates(L->getHeader(), SE.DT)) {
+ if (!SE.properlyDominates(Start, L->getHeader())) {
PostLoopOffset = Start;
Start = SE.getConstant(Normalized->getType(), 0);
Normalized =
// Strip off any non-loop-dominating component from the addrec step.
const SCEV *Step = Normalized->getStepRecurrence(SE);
const SCEV *PostLoopScale = 0;
- if (!Step->dominates(L->getHeader(), SE.DT)) {
+ if (!SE.dominates(Step, L->getHeader())) {
PostLoopScale = Step;
Step = SE.getConstant(Normalized->getType(), 1);
Normalized =
// First check for an existing canonical IV in a suitable type.
PHINode *CanonicalIV = 0;
if (PHINode *PN = L->getCanonicalInductionVariable())
- if (SE.isSCEVable(PN->getType()) &&
- SE.getEffectiveSCEVType(PN->getType())->isIntegerTy() &&
- SE.getTypeSizeInBits(PN->getType()) >= SE.getTypeSizeInBits(Ty))
+ if (SE.getTypeSizeInBits(PN->getType()) >= SE.getTypeSizeInBits(Ty))
CanonicalIV = PN;
// Rewrite an AddRec in terms of the canonical induction variable, if
BasicBlock::iterator SaveInsertPt = Builder.GetInsertPoint();
BasicBlock::iterator NewInsertPt =
llvm::next(BasicBlock::iterator(cast<Instruction>(V)));
- while (isa<PHINode>(NewInsertPt)) ++NewInsertPt;
+ while (isa<PHINode>(NewInsertPt) || isa<DbgInfoIntrinsic>(NewInsertPt))
+ ++NewInsertPt;
V = expandCodeFor(SE.getTruncateExpr(SE.getUnknown(V), Ty), 0,
NewInsertPt);
restoreInsertPoint(SaveInsertBB, SaveInsertPt);
SE.getUnknown(expand(Rest))));
}
- // {0,+,1} --> Insert a canonical induction variable into the loop!
- if (S->isAffine() &&
- S->getOperand(1) == SE.getConstant(Ty, 1)) {
- // If there's a canonical IV, just use it.
- if (CanonicalIV) {
- assert(Ty == SE.getEffectiveSCEVType(CanonicalIV->getType()) &&
- "IVs with types different from the canonical IV should "
- "already have been handled!");
- return CanonicalIV;
- }
-
+ // If we don't yet have a canonical IV, create one.
+ if (!CanonicalIV) {
// Create and insert the PHI node for the induction variable in the
// specified loop.
BasicBlock *Header = L->getHeader();
- PHINode *PN = PHINode::Create(Ty, "indvar", Header->begin());
- rememberInstruction(PN);
+ CanonicalIV = PHINode::Create(Ty, "indvar", Header->begin());
+ rememberInstruction(CanonicalIV);
Constant *One = ConstantInt::get(Ty, 1);
for (pred_iterator HPI = pred_begin(Header), HPE = pred_end(Header);
- HPI != HPE; ++HPI)
- if (L->contains(*HPI)) {
+ HPI != HPE; ++HPI) {
+ BasicBlock *HP = *HPI;
+ if (L->contains(HP)) {
// Insert a unit add instruction right before the terminator
// corresponding to the back-edge.
- Instruction *Add = BinaryOperator::CreateAdd(PN, One, "indvar.next",
- (*HPI)->getTerminator());
+ Instruction *Add = BinaryOperator::CreateAdd(CanonicalIV, One,
+ "indvar.next",
+ HP->getTerminator());
rememberInstruction(Add);
- PN->addIncoming(Add, *HPI);
+ CanonicalIV->addIncoming(Add, HP);
} else {
- PN->addIncoming(Constant::getNullValue(Ty), *HPI);
+ CanonicalIV->addIncoming(Constant::getNullValue(Ty), HP);
}
+ }
+ }
+
+ // {0,+,1} --> Insert a canonical induction variable into the loop!
+ if (S->isAffine() && S->getOperand(1)->isOne()) {
+ assert(Ty == SE.getEffectiveSCEVType(CanonicalIV->getType()) &&
+ "IVs with types different from the canonical IV should "
+ "already have been handled!");
+ return CanonicalIV;
}
// {0,+,F} --> {0,+,1} * F
- // Get the canonical induction variable I for this loop.
- Value *I = CanonicalIV ?
- CanonicalIV :
- getOrInsertCanonicalInductionVariable(L, Ty);
// If this is a simple linear addrec, emit it now as a special case.
if (S->isAffine()) // {0,+,F} --> i*F
return
expand(SE.getTruncateOrNoop(
- SE.getMulExpr(SE.getUnknown(I),
+ SE.getMulExpr(SE.getUnknown(CanonicalIV),
SE.getNoopOrAnyExtend(S->getOperand(1),
- I->getType())),
+ CanonicalIV->getType())),
Ty));
// If this is a chain of recurrences, turn it into a closed form, using the
// folders, then expandCodeFor the closed form. This allows the folders to
// simplify the expression without having to build a bunch of special code
// into this folder.
- const SCEV *IH = SE.getUnknown(I); // Get I as a "symbolic" SCEV.
+ const SCEV *IH = SE.getUnknown(CanonicalIV); // Get I as a "symbolic" SCEV.
// Promote S up to the canonical IV type, if the cast is foldable.
const SCEV *NewS = S;
- const SCEV *Ext = SE.getNoopOrAnyExtend(S, I->getType());
+ const SCEV *Ext = SE.getNoopOrAnyExtend(S, CanonicalIV->getType());
if (isa<SCEVAddRecExpr>(Ext))
NewS = Ext;
Instruction *InsertPt = Builder.GetInsertPoint();
for (Loop *L = SE.LI->getLoopFor(Builder.GetInsertBlock()); ;
L = L->getParentLoop())
- if (S->isLoopInvariant(L)) {
+ if (SE.isLoopInvariant(S, L)) {
if (!L) break;
if (BasicBlock *Preheader = L->getLoopPreheader())
InsertPt = Preheader->getTerminator();
// If the SCEV is computable at this level, insert it into the header
// after the PHIs (and after any other instructions that we've inserted
// there) so that it is guaranteed to dominate any user inside the loop.
- if (L && S->hasComputableLoopEvolution(L) && !PostIncLoops.count(L))
+ if (L && SE.hasComputableLoopEvolution(S, L) && !PostIncLoops.count(L))
InsertPt = L->getHeader()->getFirstNonPHI();
while (isInsertedInstruction(InsertPt) || isa<DbgInfoIntrinsic>(InsertPt))
InsertPt = llvm::next(BasicBlock::iterator(InsertPt));
}
void SCEVExpander::rememberInstruction(Value *I) {
- if (PostIncLoops.empty())
+ if (!PostIncLoops.empty())
+ InsertedPostIncValues.insert(I);
+ else
InsertedValues.insert(I);
// If we just claimed an existing instruction and that instruction had
/// canonical induction variable of the specified type for the specified
/// loop (inserting one if there is none). A canonical induction variable
/// starts at zero and steps by one on each iteration.
-Value *
+PHINode *
SCEVExpander::getOrInsertCanonicalInductionVariable(const Loop *L,
const Type *Ty) {
assert(Ty->isIntegerTy() && "Can only insert integer induction variables!");
+
+ // Build a SCEV for {0,+,1}<L>.
const SCEV *H = SE.getAddRecExpr(SE.getConstant(Ty, 0),
SE.getConstant(Ty, 1), L);
+
+ // Emit code for it.
BasicBlock *SaveInsertBB = Builder.GetInsertBlock();
BasicBlock::iterator SaveInsertPt = Builder.GetInsertPoint();
- Value *V = expandCodeFor(H, 0, L->getHeader()->begin());
+ PHINode *V = cast<PHINode>(expandCodeFor(H, 0, L->getHeader()->begin()));
if (SaveInsertBB)
restoreInsertPoint(SaveInsertBB, SaveInsertPt);
+
return V;
}
projects / oota-llvm.git / blobdiff