/// Connect the unrolling prolog code to the original loop.
/// The unrolling prolog code contains code to execute the
/// 'extra' iterations if the run-time trip count modulo the
/// unroll count is non-zero.
///
/// This function performs the following:
/// - Create PHI nodes at prolog end block to combine values
/// that exit the prolog code and jump around the prolog.
/// - Add a PHI operand to a PHI node at the loop exit block
/// for values that exit the prolog and go around the loop.
/// - Branch around the original loop if the trip count is less
/// than the unroll factor.
///
static void ConnectProlog(Loop *L, Value *BECount, unsigned Count,
BasicBlock *PrologExit,
BasicBlock *OriginalLoopLatchExit,
BasicBlock *PreHeader, BasicBlock *NewPreHeader,
ValueToValueMapTy &VMap, DominatorTree *DT,
LoopInfo *LI, bool PreserveLCSSA) {
// Loop structure should be the following:
// Preheader
// PrologHeader
// ...
// PrologLatch
// PrologExit
// NewPreheader
// Header
// ...
// Latch
// LatchExit
BasicBlock *Latch = L->getLoopLatch();
assert(Latch && "Loop must have a latch");
BasicBlock *PrologLatch = cast<BasicBlock>(VMap[Latch]);
// Create a PHI node for each outgoing value from the original loop
// (which means it is an outgoing value from the prolog code too).
// The new PHI node is inserted in the prolog end basic block.
// The new PHI node value is added as an operand of a PHI node in either
// the loop header or the loop exit block.
for (BasicBlock *Succ : successors(Latch)) {
for (PHINode &PN : Succ->phis()) {
// Add a new PHI node to the prolog end block and add the
// appropriate incoming values.
// TODO: This code assumes that the PrologExit (or the LatchExit block for
// prolog loop) contains only one predecessor from the loop, i.e. the
// PrologLatch. When supporting multiple-exiting block loops, we can have
// two or more blocks that have the LatchExit as the target in the
// original loop.
PHINode *NewPN = PHINode::Create(PN.getType(), 2, PN.getName() + ".unr",
PrologExit->getFirstNonPHI());
// Adding a value to the new PHI node from the original loop preheader.
// This is the value that skips all the prolog code.
if (L->contains(&PN)) {
// Succ is loop header.
NewPN->addIncoming(PN.getIncomingValueForBlock(NewPreHeader),
PreHeader);
} else {
// Succ is LatchExit.
NewPN->addIncoming(UndefValue::get(PN.getType()), PreHeader);
}
Value *V = PN.getIncomingValueForBlock(Latch);
if (Instruction *I = dyn_cast<Instruction>(V)) {
if (L->contains(I)) {
V = VMap.lookup(I);
}
}
// Adding a value to the new PHI node from the last prolog block
// that was created.
NewPN->addIncoming(V, PrologLatch);
// Update the existing PHI node operand with the value from the
// new PHI node. How this is done depends on if the existing
// PHI node is in the original loop block, or the exit block.
if (L->contains(&PN)) {
PN.setIncomingValue(PN.getBasicBlockIndex(NewPreHeader), NewPN);
} else {
PN.addIncoming(NewPN, PrologExit);
}
}
}
// Make sure that created prolog loop is in simplified form
SmallVector<BasicBlock *, 4> PrologExitPreds;
Loop *PrologLoop = LI->getLoopFor(PrologLatch);
if (PrologLoop) {
for (BasicBlock *PredBB : predecessors(PrologExit))
if (PrologLoop->contains(PredBB))
PrologExitPreds.push_back(PredBB);
SplitBlockPredecessors(PrologExit, PrologExitPreds, ".unr-lcssa", DT, LI,
nullptr, PreserveLCSSA);
}
// Create a branch around the original loop, which is taken if there are no
// iterations remaining to be executed after running the prologue.
Instruction *InsertPt = PrologExit->getTerminator();
IRBuilder<> B(InsertPt);
assert(Count != 0 && "nonsensical Count!");
// If BECount <u (Count - 1) then (BECount + 1) % Count == (BECount + 1)
// This means %xtraiter is (BECount + 1) and all of the iterations of this
// loop were executed by the prologue. Note that if BECount <u (Count - 1)
// then (BECount + 1) cannot unsigned-overflow.
Value *BrLoopExit =
B.CreateICmpULT(BECount, ConstantInt::get(BECount->getType(), Count - 1));
// Split the exit to maintain loop canonicalization guarantees
SmallVector<BasicBlock *, 4> Preds(predecessors(OriginalLoopLatchExit));
SplitBlockPredecessors(OriginalLoopLatchExit, Preds, ".unr-lcssa", DT, LI,
nullptr, PreserveLCSSA);
// Add the branch to the exit block (around the unrolled loop)
B.CreateCondBr(BrLoopExit, OriginalLoopLatchExit, NewPreHeader);
InsertPt->eraseFromParent();
if (DT)
DT->changeImmediateDominator(OriginalLoopLatchExit, PrologExit);
}
/// CloneAndPruneFunctionInto - This works exactly like CloneFunctionInto,
/// except that it does some simple constant prop and DCE on the fly. The
/// effect of this is to copy significantly less code in cases where (for
/// example) a function call with constant arguments is inlined, and those
/// constant arguments cause a significant amount of code in the callee to be
/// dead. Since this doesn't produce an exact copy of the input, it can't be
/// used for things like CloneFunction or CloneModule.
void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc,
ValueToValueMapTy &VMap,
bool ModuleLevelChanges,
SmallVectorImpl<ReturnInst*> &Returns,
const char *NameSuffix,
ClonedCodeInfo *CodeInfo,
const DataLayout *DL,
Instruction *TheCall) {
assert(NameSuffix && "NameSuffix cannot be null!");
#ifndef NDEBUG
for (Function::const_arg_iterator II = OldFunc->arg_begin(),
E = OldFunc->arg_end(); II != E; ++II)
assert(VMap.count(II) && "No mapping from source argument specified!");
#endif
PruningFunctionCloner PFC(NewFunc, OldFunc, VMap, ModuleLevelChanges,
NameSuffix, CodeInfo, DL);
// Clone the entry block, and anything recursively reachable from it.
std::vector<const BasicBlock*> CloneWorklist;
CloneWorklist.push_back(&OldFunc->getEntryBlock());
while (!CloneWorklist.empty()) {
const BasicBlock *BB = CloneWorklist.back();
CloneWorklist.pop_back();
PFC.CloneBlock(BB, CloneWorklist);
}
// Loop over all of the basic blocks in the old function. If the block was
// reachable, we have cloned it and the old block is now in the value map:
// insert it into the new function in the right order. If not, ignore it.
//
// Defer PHI resolution until rest of function is resolved.
SmallVector<const PHINode*, 16> PHIToResolve;
for (Function::const_iterator BI = OldFunc->begin(), BE = OldFunc->end();
BI != BE; ++BI) {
Value *V = VMap[BI];
BasicBlock *NewBB = cast_or_null<BasicBlock>(V);
if (NewBB == 0) continue; // Dead block.
// Add the new block to the new function.
NewFunc->getBasicBlockList().push_back(NewBB);
// Handle PHI nodes specially, as we have to remove references to dead
// blocks.
for (BasicBlock::const_iterator I = BI->begin(), E = BI->end(); I != E; ++I)
if (const PHINode *PN = dyn_cast<PHINode>(I))
PHIToResolve.push_back(PN);
else
break;
// Finally, remap the terminator instructions, as those can't be remapped
// until all BBs are mapped.
RemapInstruction(NewBB->getTerminator(), VMap,
ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges);
}
// Defer PHI resolution until rest of function is resolved, PHI resolution
// requires the CFG to be up-to-date.
for (unsigned phino = 0, e = PHIToResolve.size(); phino != e; ) {
const PHINode *OPN = PHIToResolve[phino];
unsigned NumPreds = OPN->getNumIncomingValues();
const BasicBlock *OldBB = OPN->getParent();
BasicBlock *NewBB = cast<BasicBlock>(VMap[OldBB]);
// Map operands for blocks that are live and remove operands for blocks
// that are dead.
for (; phino != PHIToResolve.size() &&
PHIToResolve[phino]->getParent() == OldBB; ++phino) {
OPN = PHIToResolve[phino];
PHINode *PN = cast<PHINode>(VMap[OPN]);
for (unsigned pred = 0, e = NumPreds; pred != e; ++pred) {
Value *V = VMap[PN->getIncomingBlock(pred)];
if (BasicBlock *MappedBlock = cast_or_null<BasicBlock>(V)) {
Value *InVal = MapValue(PN->getIncomingValue(pred),
VMap,
ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges);
assert(InVal && "Unknown input value?");
PN->setIncomingValue(pred, InVal);
PN->setIncomingBlock(pred, MappedBlock);
} else {
PN->removeIncomingValue(pred, false);
--pred, --e; // Revisit the next entry.
}
}
}
// The loop above has removed PHI entries for those blocks that are dead
// and has updated others. However, if a block is live (i.e. copied over)
// but its terminator has been changed to not go to this block, then our
// phi nodes will have invalid entries. Update the PHI nodes in this
// case.
PHINode *PN = cast<PHINode>(NewBB->begin());
NumPreds = std::distance(pred_begin(NewBB), pred_end(NewBB));
if (NumPreds != PN->getNumIncomingValues()) {
assert(NumPreds < PN->getNumIncomingValues());
// Count how many times each predecessor comes to this block.
std::map<BasicBlock*, unsigned> PredCount;
for (pred_iterator PI = pred_begin(NewBB), E = pred_end(NewBB);
PI != E; ++PI)
--PredCount[*PI];
// Figure out how many entries to remove from each PHI.
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
++PredCount[PN->getIncomingBlock(i)];
// At this point, the excess predecessor entries are positive in the
// map. Loop over all of the PHIs and remove excess predecessor
// entries.
BasicBlock::iterator I = NewBB->begin();
for (; (PN = dyn_cast<PHINode>(I)); ++I) {
for (std::map<BasicBlock*, unsigned>::iterator PCI =PredCount.begin(),
E = PredCount.end(); PCI != E; ++PCI) {
BasicBlock *Pred = PCI->first;
for (unsigned NumToRemove = PCI->second; NumToRemove; --NumToRemove)
PN->removeIncomingValue(Pred, false);
}
}
}
// If the loops above have made these phi nodes have 0 or 1 operand,
// replace them with undef or the input value. We must do this for
// correctness, because 0-operand phis are not valid.
PN = cast<PHINode>(NewBB->begin());
if (PN->getNumIncomingValues() == 0) {
BasicBlock::iterator I = NewBB->begin();
BasicBlock::const_iterator OldI = OldBB->begin();
while ((PN = dyn_cast<PHINode>(I++))) {
Value *NV = UndefValue::get(PN->getType());
PN->replaceAllUsesWith(NV);
assert(VMap[OldI] == PN && "VMap mismatch");
VMap[OldI] = NV;
PN->eraseFromParent();
++OldI;
}
}
}
// Make a second pass over the PHINodes now that all of them have been
// remapped into the new function, simplifying the PHINode and performing any
// recursive simplifications exposed. This will transparently update the
// WeakVH in the VMap. Notably, we rely on that so that if we coalesce
// two PHINodes, the iteration over the old PHIs remains valid, and the
// mapping will just map us to the new node (which may not even be a PHI
// node).
for (unsigned Idx = 0, Size = PHIToResolve.size(); Idx != Size; ++Idx)
if (PHINode *PN = dyn_cast<PHINode>(VMap[PHIToResolve[Idx]]))
recursivelySimplifyInstruction(PN, DL);
// Now that the inlined function body has been fully constructed, go through
// and zap unconditional fall-through branches. This happen all the time when
// specializing code: code specialization turns conditional branches into
// uncond branches, and this code folds them.
Function::iterator Begin = cast<BasicBlock>(VMap[&OldFunc->getEntryBlock()]);
Function::iterator I = Begin;
while (I != NewFunc->end()) {
// Check if this block has become dead during inlining or other
// simplifications. Note that the first block will appear dead, as it has
// not yet been wired up properly.
if (I != Begin && (pred_begin(I) == pred_end(I) ||
I->getSinglePredecessor() == I)) {
BasicBlock *DeadBB = I++;
DeleteDeadBlock(DeadBB);
continue;
}
// We need to simplify conditional branches and switches with a constant
// operand. We try to prune these out when cloning, but if the
// simplification required looking through PHI nodes, those are only
// available after forming the full basic block. That may leave some here,
// and we still want to prune the dead code as early as possible.
ConstantFoldTerminator(I);
BranchInst *BI = dyn_cast<BranchInst>(I->getTerminator());
if (!BI || BI->isConditional()) { ++I; continue; }
BasicBlock *Dest = BI->getSuccessor(0);
if (!Dest->getSinglePredecessor()) {
++I; continue;
}
// We shouldn't be able to get single-entry PHI nodes here, as instsimplify
// above should have zapped all of them..
assert(!isa<PHINode>(Dest->begin()));
// We know all single-entry PHI nodes in the inlined function have been
// removed, so we just need to splice the blocks.
BI->eraseFromParent();
// Make all PHI nodes that referred to Dest now refer to I as their source.
Dest->replaceAllUsesWith(I);
// Move all the instructions in the succ to the pred.
I->getInstList().splice(I->end(), Dest->getInstList());
// Remove the dest block.
Dest->eraseFromParent();
// Do not increment I, iteratively merge all things this block branches to.
}
// Make a final pass over the basic blocks from theh old function to gather
// any return instructions which survived folding. We have to do this here
// because we can iteratively remove and merge returns above.
for (Function::iterator I = cast<BasicBlock>(VMap[&OldFunc->getEntryBlock()]),
E = NewFunc->end();
I != E; ++I)
if (ReturnInst *RI = dyn_cast<ReturnInst>(I->getTerminator()))
Returns.push_back(RI);
}
/// Connect the unrolling epilog code to the original loop.
/// The unrolling epilog code contains code to execute the
/// 'extra' iterations if the run-time trip count modulo the
/// unroll count is non-zero.
///
/// This function performs the following:
/// - Update PHI nodes at the unrolling loop exit and epilog loop exit
/// - Create PHI nodes at the unrolling loop exit to combine
/// values that exit the unrolling loop code and jump around it.
/// - Update PHI operands in the epilog loop by the new PHI nodes
/// - Branch around the epilog loop if extra iters (ModVal) is zero.
///
static void ConnectEpilog(Loop *L, Value *ModVal, BasicBlock *NewExit,
BasicBlock *Exit, BasicBlock *PreHeader,
BasicBlock *EpilogPreHeader, BasicBlock *NewPreHeader,
ValueToValueMapTy &VMap, DominatorTree *DT,
LoopInfo *LI, bool PreserveLCSSA) {
BasicBlock *Latch = L->getLoopLatch();
assert(Latch && "Loop must have a latch");
BasicBlock *EpilogLatch = cast<BasicBlock>(VMap[Latch]);
// Loop structure should be the following:
//
// PreHeader
// NewPreHeader
// Header
// ...
// Latch
// NewExit (PN)
// EpilogPreHeader
// EpilogHeader
// ...
// EpilogLatch
// Exit (EpilogPN)
// Update PHI nodes at NewExit and Exit.
for (PHINode &PN : NewExit->phis()) {
// PN should be used in another PHI located in Exit block as
// Exit was split by SplitBlockPredecessors into Exit and NewExit
// Basicaly it should look like:
// NewExit:
// PN = PHI [I, Latch]
// ...
// Exit:
// EpilogPN = PHI [PN, EpilogPreHeader]
//
// There is EpilogPreHeader incoming block instead of NewExit as
// NewExit was spilt 1 more time to get EpilogPreHeader.
assert(PN.hasOneUse() && "The phi should have 1 use");
PHINode *EpilogPN = cast<PHINode>(PN.use_begin()->getUser());
assert(EpilogPN->getParent() == Exit && "EpilogPN should be in Exit block");
// Add incoming PreHeader from branch around the Loop
PN.addIncoming(UndefValue::get(PN.getType()), PreHeader);
Value *V = PN.getIncomingValueForBlock(Latch);
Instruction *I = dyn_cast<Instruction>(V);
if (I && L->contains(I))
// If value comes from an instruction in the loop add VMap value.
V = VMap.lookup(I);
// For the instruction out of the loop, constant or undefined value
// insert value itself.
EpilogPN->addIncoming(V, EpilogLatch);
assert(EpilogPN->getBasicBlockIndex(EpilogPreHeader) >= 0 &&
"EpilogPN should have EpilogPreHeader incoming block");
// Change EpilogPreHeader incoming block to NewExit.
EpilogPN->setIncomingBlock(EpilogPN->getBasicBlockIndex(EpilogPreHeader),
NewExit);
// Now PHIs should look like:
// NewExit:
// PN = PHI [I, Latch], [undef, PreHeader]
// ...
// Exit:
// EpilogPN = PHI [PN, NewExit], [VMap[I], EpilogLatch]
}
// Create PHI nodes at NewExit (from the unrolling loop Latch and PreHeader).
// Update corresponding PHI nodes in epilog loop.
for (BasicBlock *Succ : successors(Latch)) {
// Skip this as we already updated phis in exit blocks.
if (!L->contains(Succ))
continue;
for (PHINode &PN : Succ->phis()) {
// Add new PHI nodes to the loop exit block and update epilog
// PHIs with the new PHI values.
PHINode *NewPN = PHINode::Create(PN.getType(), 2, PN.getName() + ".unr",
NewExit->getFirstNonPHI());
// Adding a value to the new PHI node from the unrolling loop preheader.
NewPN->addIncoming(PN.getIncomingValueForBlock(NewPreHeader), PreHeader);
// Adding a value to the new PHI node from the unrolling loop latch.
NewPN->addIncoming(PN.getIncomingValueForBlock(Latch), Latch);
// Update the existing PHI node operand with the value from the new PHI
// node. Corresponding instruction in epilog loop should be PHI.
PHINode *VPN = cast<PHINode>(VMap[&PN]);
VPN->setIncomingValue(VPN->getBasicBlockIndex(EpilogPreHeader), NewPN);
}
}
Instruction *InsertPt = NewExit->getTerminator();
IRBuilder<> B(InsertPt);
Value *BrLoopExit = B.CreateIsNotNull(ModVal, "lcmp.mod");
assert(Exit && "Loop must have a single exit block only");
// Split the epilogue exit to maintain loop canonicalization guarantees
SmallVector<BasicBlock*, 4> Preds(predecessors(Exit));
SplitBlockPredecessors(Exit, Preds, ".epilog-lcssa", DT, LI, nullptr,
PreserveLCSSA);
// Add the branch to the exit block (around the unrolling loop)
B.CreateCondBr(BrLoopExit, EpilogPreHeader, Exit);
InsertPt->eraseFromParent();
if (DT)
DT->changeImmediateDominator(Exit, NewExit);
// Split the main loop exit to maintain canonicalization guarantees.
SmallVector<BasicBlock*, 4> NewExitPreds{Latch};
SplitBlockPredecessors(NewExit, NewExitPreds, ".loopexit", DT, LI, nullptr,
PreserveLCSSA);
}
bool llvm::UnrollRuntimeLoopRemainder(Loop *L, unsigned Count,
bool AllowExpensiveTripCount,
bool UseEpilogRemainder,
bool UnrollRemainder, LoopInfo *LI,
ScalarEvolution *SE, DominatorTree *DT,
AssumptionCache *AC, bool PreserveLCSSA,
Loop **ResultLoop) {
LLVM_DEBUG(dbgs() << "Trying runtime unrolling on Loop: \n");
LLVM_DEBUG(L->dump());
LLVM_DEBUG(UseEpilogRemainder ? dbgs() << "Using epilog remainder.\n"
: dbgs() << "Using prolog remainder.\n");
// Make sure the loop is in canonical form.
if (!L->isLoopSimplifyForm()) {
LLVM_DEBUG(dbgs() << "Not in simplify form!\n");
return false;
}
// Guaranteed by LoopSimplifyForm.
BasicBlock *Latch = L->getLoopLatch();
BasicBlock *Header = L->getHeader();
BranchInst *LatchBR = cast<BranchInst>(Latch->getTerminator());
if (!LatchBR || LatchBR->isUnconditional()) {
// The loop-rotate pass can be helpful to avoid this in many cases.
LLVM_DEBUG(
dbgs()
<< "Loop latch not terminated by a conditional branch.\n");
return false;
}
unsigned ExitIndex = LatchBR->getSuccessor(0) == Header ? 1 : 0;
BasicBlock *LatchExit = LatchBR->getSuccessor(ExitIndex);
if (L->contains(LatchExit)) {
// Cloning the loop basic blocks (`CloneLoopBlocks`) requires that one of the
// targets of the Latch be an exit block out of the loop.
LLVM_DEBUG(
dbgs()
<< "One of the loop latch successors must be the exit block.\n");
return false;
}
// These are exit blocks other than the target of the latch exiting block.
SmallVector<BasicBlock *, 4> OtherExits;
bool isMultiExitUnrollingEnabled =
canSafelyUnrollMultiExitLoop(L, OtherExits, LatchExit, PreserveLCSSA,
UseEpilogRemainder) &&
canProfitablyUnrollMultiExitLoop(L, OtherExits, LatchExit, PreserveLCSSA,
UseEpilogRemainder);
// Support only single exit and exiting block unless multi-exit loop unrolling is enabled.
if (!isMultiExitUnrollingEnabled &&
(!L->getExitingBlock() || OtherExits.size())) {
LLVM_DEBUG(
dbgs()
<< "Multiple exit/exiting blocks in loop and multi-exit unrolling not "
"enabled!\n");
return false;
}
// Use Scalar Evolution to compute the trip count. This allows more loops to
// be unrolled than relying on induction var simplification.
if (!SE)
return false;
// Only unroll loops with a computable trip count, and the trip count needs
// to be an int value (allowing a pointer type is a TODO item).
// We calculate the backedge count by using getExitCount on the Latch block,
// which is proven to be the only exiting block in this loop. This is same as
// calculating getBackedgeTakenCount on the loop (which computes SCEV for all
// exiting blocks).
const SCEV *BECountSC = SE->getExitCount(L, Latch);
if (isa<SCEVCouldNotCompute>(BECountSC) ||
!BECountSC->getType()->isIntegerTy()) {
LLVM_DEBUG(dbgs() << "Could not compute exit block SCEV\n");
return false;
}
unsigned BEWidth = cast<IntegerType>(BECountSC->getType())->getBitWidth();
// Add 1 since the backedge count doesn't include the first loop iteration.
const SCEV *TripCountSC =
SE->getAddExpr(BECountSC, SE->getConstant(BECountSC->getType(), 1));
if (isa<SCEVCouldNotCompute>(TripCountSC)) {
LLVM_DEBUG(dbgs() << "Could not compute trip count SCEV.\n");
return false;
}
BasicBlock *PreHeader = L->getLoopPreheader();
BranchInst *PreHeaderBR = cast<BranchInst>(PreHeader->getTerminator());
const DataLayout &DL = Header->getModule()->getDataLayout();
SCEVExpander Expander(*SE, DL, "loop-unroll");
if (!AllowExpensiveTripCount &&
Expander.isHighCostExpansion(TripCountSC, L, PreHeaderBR)) {
LLVM_DEBUG(dbgs() << "High cost for expanding trip count scev!\n");
return false;
}
// This constraint lets us deal with an overflowing trip count easily; see the
// comment on ModVal below.
if (Log2_32(Count) > BEWidth) {
LLVM_DEBUG(
dbgs()
<< "Count failed constraint on overflow trip count calculation.\n");
return false;
}
// Loop structure is the following:
//
// PreHeader
// Header
// ...
// Latch
// LatchExit
BasicBlock *NewPreHeader;
BasicBlock *NewExit = nullptr;
BasicBlock *PrologExit = nullptr;
BasicBlock *EpilogPreHeader = nullptr;
BasicBlock *PrologPreHeader = nullptr;
if (UseEpilogRemainder) {
// If epilog remainder
// Split PreHeader to insert a branch around loop for unrolling.
NewPreHeader = SplitBlock(PreHeader, PreHeader->getTerminator(), DT, LI);
NewPreHeader->setName(PreHeader->getName() + ".new");
// Split LatchExit to create phi nodes from branch above.
SmallVector<BasicBlock*, 4> Preds(predecessors(LatchExit));
NewExit = SplitBlockPredecessors(LatchExit, Preds, ".unr-lcssa", DT, LI,
nullptr, PreserveLCSSA);
// NewExit gets its DebugLoc from LatchExit, which is not part of the
// original Loop.
// Fix this by setting Loop's DebugLoc to NewExit.
auto *NewExitTerminator = NewExit->getTerminator();
NewExitTerminator->setDebugLoc(Header->getTerminator()->getDebugLoc());
// Split NewExit to insert epilog remainder loop.
EpilogPreHeader = SplitBlock(NewExit, NewExitTerminator, DT, LI);
EpilogPreHeader->setName(Header->getName() + ".epil.preheader");
} else {
// If prolog remainder
// Split the original preheader twice to insert prolog remainder loop
PrologPreHeader = SplitEdge(PreHeader, Header, DT, LI);
PrologPreHeader->setName(Header->getName() + ".prol.preheader");
PrologExit = SplitBlock(PrologPreHeader, PrologPreHeader->getTerminator(),
DT, LI);
PrologExit->setName(Header->getName() + ".prol.loopexit");
// Split PrologExit to get NewPreHeader.
NewPreHeader = SplitBlock(PrologExit, PrologExit->getTerminator(), DT, LI);
NewPreHeader->setName(PreHeader->getName() + ".new");
}
// Loop structure should be the following:
// Epilog Prolog
//
// PreHeader PreHeader
// *NewPreHeader *PrologPreHeader
// Header *PrologExit
// ... *NewPreHeader
// Latch Header
// *NewExit ...
// *EpilogPreHeader Latch
// LatchExit LatchExit
// Calculate conditions for branch around loop for unrolling
// in epilog case and around prolog remainder loop in prolog case.
// Compute the number of extra iterations required, which is:
// extra iterations = run-time trip count % loop unroll factor
PreHeaderBR = cast<BranchInst>(PreHeader->getTerminator());
Value *TripCount = Expander.expandCodeFor(TripCountSC, TripCountSC->getType(),
PreHeaderBR);
Value *BECount = Expander.expandCodeFor(BECountSC, BECountSC->getType(),
PreHeaderBR);
IRBuilder<> B(PreHeaderBR);
Value *ModVal;
// Calculate ModVal = (BECount + 1) % Count.
// Note that TripCount is BECount + 1.
if (isPowerOf2_32(Count)) {
// When Count is power of 2 we don't BECount for epilog case, however we'll
// need it for a branch around unrolling loop for prolog case.
ModVal = B.CreateAnd(TripCount, Count - 1, "xtraiter");
// 1. There are no iterations to be run in the prolog/epilog loop.
// OR
// 2. The addition computing TripCount overflowed.
//
// If (2) is true, we know that TripCount really is (1 << BEWidth) and so
// the number of iterations that remain to be run in the original loop is a
// multiple Count == (1 << Log2(Count)) because Log2(Count) <= BEWidth (we
// explicitly check this above).
} else {
// As (BECount + 1) can potentially unsigned overflow we count
// (BECount % Count) + 1 which is overflow safe as BECount % Count < Count.
Value *ModValTmp = B.CreateURem(BECount,
ConstantInt::get(BECount->getType(),
Count));
Value *ModValAdd = B.CreateAdd(ModValTmp,
ConstantInt::get(ModValTmp->getType(), 1));
// At that point (BECount % Count) + 1 could be equal to Count.
// To handle this case we need to take mod by Count one more time.
ModVal = B.CreateURem(ModValAdd,
ConstantInt::get(BECount->getType(), Count),
"xtraiter");
}
Value *BranchVal =
UseEpilogRemainder ? B.CreateICmpULT(BECount,
ConstantInt::get(BECount->getType(),
Count - 1)) :
B.CreateIsNotNull(ModVal, "lcmp.mod");
BasicBlock *RemainderLoop = UseEpilogRemainder ? NewExit : PrologPreHeader;
BasicBlock *UnrollingLoop = UseEpilogRemainder ? NewPreHeader : PrologExit;
// Branch to either remainder (extra iterations) loop or unrolling loop.
B.CreateCondBr(BranchVal, RemainderLoop, UnrollingLoop);
PreHeaderBR->eraseFromParent();
if (DT) {
if (UseEpilogRemainder)
DT->changeImmediateDominator(NewExit, PreHeader);
else
DT->changeImmediateDominator(PrologExit, PreHeader);
}
Function *F = Header->getParent();
// Get an ordered list of blocks in the loop to help with the ordering of the
// cloned blocks in the prolog/epilog code
LoopBlocksDFS LoopBlocks(L);
LoopBlocks.perform(LI);
//
// For each extra loop iteration, create a copy of the loop's basic blocks
// and generate a condition that branches to the copy depending on the
// number of 'left over' iterations.
//
std::vector<BasicBlock *> NewBlocks;
ValueToValueMapTy VMap;
// For unroll factor 2 remainder loop will have 1 iterations.
// Do not create 1 iteration loop.
bool CreateRemainderLoop = (Count != 2);
// Clone all the basic blocks in the loop. If Count is 2, we don't clone
// the loop, otherwise we create a cloned loop to execute the extra
// iterations. This function adds the appropriate CFG connections.
BasicBlock *InsertBot = UseEpilogRemainder ? LatchExit : PrologExit;
BasicBlock *InsertTop = UseEpilogRemainder ? EpilogPreHeader : PrologPreHeader;
Loop *remainderLoop = CloneLoopBlocks(
L, ModVal, CreateRemainderLoop, UseEpilogRemainder, UnrollRemainder,
InsertTop, InsertBot,
NewPreHeader, NewBlocks, LoopBlocks, VMap, DT, LI);
// Insert the cloned blocks into the function.
F->getBasicBlockList().splice(InsertBot->getIterator(),
F->getBasicBlockList(),
NewBlocks[0]->getIterator(),
F->end());
// Now the loop blocks are cloned and the other exiting blocks from the
// remainder are connected to the original Loop's exit blocks. The remaining
// work is to update the phi nodes in the original loop, and take in the
// values from the cloned region.
for (auto *BB : OtherExits) {
for (auto &II : *BB) {
// Given we preserve LCSSA form, we know that the values used outside the
// loop will be used through these phi nodes at the exit blocks that are
// transformed below.
if (!isa<PHINode>(II))
break;
PHINode *Phi = cast<PHINode>(&II);
unsigned oldNumOperands = Phi->getNumIncomingValues();
// Add the incoming values from the remainder code to the end of the phi
// node.
for (unsigned i =0; i < oldNumOperands; i++){
Value *newVal = VMap.lookup(Phi->getIncomingValue(i));
// newVal can be a constant or derived from values outside the loop, and
// hence need not have a VMap value. Also, since lookup already generated
// a default "null" VMap entry for this value, we need to populate that
// VMap entry correctly, with the mapped entry being itself.
if (!newVal) {
newVal = Phi->getIncomingValue(i);
VMap[Phi->getIncomingValue(i)] = Phi->getIncomingValue(i);
}
Phi->addIncoming(newVal,
cast<BasicBlock>(VMap[Phi->getIncomingBlock(i)]));
}
}
#if defined(EXPENSIVE_CHECKS) && !defined(NDEBUG)
for (BasicBlock *SuccBB : successors(BB)) {
assert(!(any_of(OtherExits,
[SuccBB](BasicBlock *EB) { return EB == SuccBB; }) ||
SuccBB == LatchExit) &&
"Breaks the definition of dedicated exits!");
}
#endif
}
// Update the immediate dominator of the exit blocks and blocks that are
// reachable from the exit blocks. This is needed because we now have paths
// from both the original loop and the remainder code reaching the exit
// blocks. While the IDom of these exit blocks were from the original loop,
// now the IDom is the preheader (which decides whether the original loop or
// remainder code should run).
if (DT && !L->getExitingBlock()) {
SmallVector<BasicBlock *, 16> ChildrenToUpdate;
// NB! We have to examine the dom children of all loop blocks, not just
// those which are the IDom of the exit blocks. This is because blocks
// reachable from the exit blocks can have their IDom as the nearest common
// dominator of the exit blocks.
for (auto *BB : L->blocks()) {
auto *DomNodeBB = DT->getNode(BB);
for (auto *DomChild : DomNodeBB->getChildren()) {
auto *DomChildBB = DomChild->getBlock();
if (!L->contains(LI->getLoopFor(DomChildBB)))
ChildrenToUpdate.push_back(DomChildBB);
}
}
for (auto *BB : ChildrenToUpdate)
DT->changeImmediateDominator(BB, PreHeader);
}
// Loop structure should be the following:
// Epilog Prolog
//
// PreHeader PreHeader
// NewPreHeader PrologPreHeader
// Header PrologHeader
// ... ...
// Latch PrologLatch
// NewExit PrologExit
// EpilogPreHeader NewPreHeader
// EpilogHeader Header
// ... ...
// EpilogLatch Latch
// LatchExit LatchExit
// Rewrite the cloned instruction operands to use the values created when the
// clone is created.
for (BasicBlock *BB : NewBlocks) {
for (Instruction &I : *BB) {
RemapInstruction(&I, VMap,
RF_NoModuleLevelChanges | RF_IgnoreMissingLocals);
}
}
if (UseEpilogRemainder) {
// Connect the epilog code to the original loop and update the
// PHI functions.
ConnectEpilog(L, ModVal, NewExit, LatchExit, PreHeader,
EpilogPreHeader, NewPreHeader, VMap, DT, LI,
PreserveLCSSA);
// Update counter in loop for unrolling.
// I should be multiply of Count.
IRBuilder<> B2(NewPreHeader->getTerminator());
Value *TestVal = B2.CreateSub(TripCount, ModVal, "unroll_iter");
BranchInst *LatchBR = cast<BranchInst>(Latch->getTerminator());
B2.SetInsertPoint(LatchBR);
PHINode *NewIdx = PHINode::Create(TestVal->getType(), 2, "niter",
Header->getFirstNonPHI());
Value *IdxSub =
B2.CreateSub(NewIdx, ConstantInt::get(NewIdx->getType(), 1),
NewIdx->getName() + ".nsub");
Value *IdxCmp;
if (LatchBR->getSuccessor(0) == Header)
IdxCmp = B2.CreateIsNotNull(IdxSub, NewIdx->getName() + ".ncmp");
else
IdxCmp = B2.CreateIsNull(IdxSub, NewIdx->getName() + ".ncmp");
NewIdx->addIncoming(TestVal, NewPreHeader);
NewIdx->addIncoming(IdxSub, Latch);
LatchBR->setCondition(IdxCmp);
} else {
// Connect the prolog code to the original loop and update the
// PHI functions.
ConnectProlog(L, BECount, Count, PrologExit, LatchExit, PreHeader,
NewPreHeader, VMap, DT, LI, PreserveLCSSA);
}
// If this loop is nested, then the loop unroller changes the code in the any
// of its parent loops, so the Scalar Evolution pass needs to be run again.
SE->forgetTopmostLoop(L);
// Verify that the Dom Tree is correct.
#if defined(EXPENSIVE_CHECKS) && !defined(NDEBUG)
if (DT)
assert(DT->verify(DominatorTree::VerificationLevel::Full));
#endif
// Canonicalize to LoopSimplifyForm both original and remainder loops. We
// cannot rely on the LoopUnrollPass to do this because it only does
// canonicalization for parent/subloops and not the sibling loops.
if (OtherExits.size() > 0) {
// Generate dedicated exit blocks for the original loop, to preserve
// LoopSimplifyForm.
formDedicatedExitBlocks(L, DT, LI, nullptr, PreserveLCSSA);
// Generate dedicated exit blocks for the remainder loop if one exists, to
// preserve LoopSimplifyForm.
if (remainderLoop)
formDedicatedExitBlocks(remainderLoop, DT, LI, nullptr, PreserveLCSSA);
}
auto UnrollResult = LoopUnrollResult::Unmodified;
if (remainderLoop && UnrollRemainder) {
LLVM_DEBUG(dbgs() << "Unrolling remainder loop\n");
UnrollResult =
UnrollLoop(remainderLoop, /*Count*/ Count - 1, /*TripCount*/ Count - 1,
/*Force*/ false, /*AllowRuntime*/ false,
/*AllowExpensiveTripCount*/ false, /*PreserveCondBr*/ true,
/*PreserveOnlyFirst*/ false, /*TripMultiple*/ 1,
/*PeelCount*/ 0, /*UnrollRemainder*/ false, LI, SE, DT, AC,
/*ORE*/ nullptr, PreserveLCSSA);
}
if (ResultLoop && UnrollResult != LoopUnrollResult::FullyUnrolled)
*ResultLoop = remainderLoop;
NumRuntimeUnrolled++;
return true;
}
// Clone OldFunc into NewFunc, transforming the old arguments into references to
// VMap values.
//
void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc,
ValueToValueMapTy &VMap,
bool ModuleLevelChanges,
SmallVectorImpl<ReturnInst*> &Returns,
const char *NameSuffix, ClonedCodeInfo *CodeInfo,
ValueMapTypeRemapper *TypeMapper,
ValueMaterializer *Materializer) {
assert(NameSuffix && "NameSuffix cannot be null!");
#ifndef NDEBUG
for (Function::const_arg_iterator I = OldFunc->arg_begin(),
E = OldFunc->arg_end(); I != E; ++I)
assert(VMap.count(I) && "No mapping from source argument specified!");
#endif
// Copy all attributes other than those stored in the AttributeSet. We need
// to remap the parameter indices of the AttributeSet.
AttributeSet NewAttrs = NewFunc->getAttributes();
NewFunc->copyAttributesFrom(OldFunc);
NewFunc->setAttributes(NewAttrs);
AttributeSet OldAttrs = OldFunc->getAttributes();
// Clone any argument attributes that are present in the VMap.
for (const Argument &OldArg : OldFunc->args())
if (Argument *NewArg = dyn_cast<Argument>(VMap[&OldArg])) {
AttributeSet attrs =
OldAttrs.getParamAttributes(OldArg.getArgNo() + 1);
if (attrs.getNumSlots() > 0)
NewArg->addAttr(attrs);
}
NewFunc->setAttributes(
NewFunc->getAttributes()
.addAttributes(NewFunc->getContext(), AttributeSet::ReturnIndex,
OldAttrs.getRetAttributes())
.addAttributes(NewFunc->getContext(), AttributeSet::FunctionIndex,
OldAttrs.getFnAttributes()));
// Loop over all of the basic blocks in the function, cloning them as
// appropriate. Note that we save BE this way in order to handle cloning of
// recursive functions into themselves.
//
for (Function::const_iterator BI = OldFunc->begin(), BE = OldFunc->end();
BI != BE; ++BI) {
const BasicBlock &BB = *BI;
// Create a new basic block and copy instructions into it!
BasicBlock *CBB = CloneBasicBlock(&BB, VMap, NameSuffix, NewFunc, CodeInfo);
// Add basic block mapping.
VMap[&BB] = CBB;
// It is only legal to clone a function if a block address within that
// function is never referenced outside of the function. Given that, we
// want to map block addresses from the old function to block addresses in
// the clone. (This is different from the generic ValueMapper
// implementation, which generates an invalid blockaddress when
// cloning a function.)
if (BB.hasAddressTaken()) {
Constant *OldBBAddr = BlockAddress::get(const_cast<Function*>(OldFunc),
const_cast<BasicBlock*>(&BB));
VMap[OldBBAddr] = BlockAddress::get(NewFunc, CBB);
}
// Note return instructions for the caller.
if (ReturnInst *RI = dyn_cast<ReturnInst>(CBB->getTerminator()))
Returns.push_back(RI);
}
// Loop over all of the instructions in the function, fixing up operand
// references as we go. This uses VMap to do all the hard work.
for (Function::iterator BB = cast<BasicBlock>(VMap[OldFunc->begin()]),
BE = NewFunc->end(); BB != BE; ++BB)
// Loop over all instructions, fixing each one as we find it...
for (BasicBlock::iterator II = BB->begin(); II != BB->end(); ++II)
RemapInstruction(II, VMap,
ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges,
TypeMapper, Materializer);
}
Function*
specializeFunction(Function *f, Value*const* args)
// make a copy of f
// specialize on the arguments, a null means that that argument isn't known
{
assert(!f->isDeclaration());
ValueToValueMapTy vmap;
unsigned int i = 0;
unsigned int j = 0;
std::vector<std::string> argNames;
std::string baseName = specializeName(f, argNames);
for (Function::arg_iterator itr = f->arg_begin(); itr != f->arg_end(); itr++, i++) {
while (argNames[j] != "?") j++;
if (args[i] != NULL) {
Value* arg = (Value*) &(*itr);
assert(arg->getType() == args[i]->getType()
&& "Specializing argument with concrete value of wrong type!");
vmap.insert(std::pair<Value*, WeakVH>(arg, args[i]));
PrevirtType pt = PrevirtType::abstract(args[i]);
argNames[j] = pt.to_string();
/*
if (const ConstantInt* ci = dyn_cast<const ConstantInt> (args[i])) {
argNames[j] = ci->getValue().toString(10, true);
} else if (const Constant* c = dyn_cast<const Constant>(args[i])) {
if (c->isNullValue()) {
argNames[j] = "null";
} else {
assert(false);
}
} else if (const )
} else {
assert(false);
}
*/
}
j++;
}
assert (i == f->getArgumentList().size());
baseName += "(";
for (std::vector<std::string>::const_iterator it = argNames.begin(), be = argNames.begin(), en = argNames.end(); it != en; ++it) {
if (it != be) baseName += ",";
baseName += *it;
}
baseName += ")";
Function *result = f->getParent()->getFunction(baseName);
// If specialized function already exists, no reason
// to create another one. In fact, can cause the process
// to diverge. XXX This needs to be a more sophisticated
// check
if (!result) {
ClonedCodeInfo info;
result = llvm::CloneFunction(f, vmap, true, &info);
result->setName(baseName);
}
return result;
}
/// UnswitchNontrivialCondition - We determined that the loop is profitable
/// to unswitch when LIC equal Val. Split it into loop versions and test the
/// condition outside of either loop. Return the loops created as Out1/Out2.
void LoopUnswitch::UnswitchNontrivialCondition(Value *LIC, Constant *Val,
Loop *L) {
Function *F = loopHeader->getParent();
DEBUG(dbgs() << "loop-unswitch: Unswitching loop %"
<< loopHeader->getName() << " [" << L->getBlocks().size()
<< " blocks] in Function " << F->getName()
<< " when '" << *Val << "' == " << *LIC << "\n");
if (ScalarEvolution *SE = getAnalysisIfAvailable<ScalarEvolution>())
SE->forgetLoop(L);
LoopBlocks.clear();
NewBlocks.clear();
// First step, split the preheader and exit blocks, and add these blocks to
// the LoopBlocks list.
BasicBlock *NewPreheader = SplitEdge(loopPreheader, loopHeader, this);
LoopBlocks.push_back(NewPreheader);
// We want the loop to come after the preheader, but before the exit blocks.
LoopBlocks.insert(LoopBlocks.end(), L->block_begin(), L->block_end());
SmallVector<BasicBlock*, 8> ExitBlocks;
L->getUniqueExitBlocks(ExitBlocks);
// Split all of the edges from inside the loop to their exit blocks. Update
// the appropriate Phi nodes as we do so.
SplitExitEdges(L, ExitBlocks);
// The exit blocks may have been changed due to edge splitting, recompute.
ExitBlocks.clear();
L->getUniqueExitBlocks(ExitBlocks);
// Add exit blocks to the loop blocks.
LoopBlocks.insert(LoopBlocks.end(), ExitBlocks.begin(), ExitBlocks.end());
// Next step, clone all of the basic blocks that make up the loop (including
// the loop preheader and exit blocks), keeping track of the mapping between
// the instructions and blocks.
NewBlocks.reserve(LoopBlocks.size());
ValueToValueMapTy VMap;
for (unsigned i = 0, e = LoopBlocks.size(); i != e; ++i) {
BasicBlock *NewBB = CloneBasicBlock(LoopBlocks[i], VMap, ".us", F);
NewBlocks.push_back(NewBB);
VMap[LoopBlocks[i]] = NewBB; // Keep the BB mapping.
LPM->cloneBasicBlockSimpleAnalysis(LoopBlocks[i], NewBB, L);
}
// Splice the newly inserted blocks into the function right before the
// original preheader.
F->getBasicBlockList().splice(NewPreheader, F->getBasicBlockList(),
NewBlocks[0], F->end());
// Now we create the new Loop object for the versioned loop.
Loop *NewLoop = CloneLoop(L, L->getParentLoop(), VMap, LI, LPM);
// Recalculate unswitching quota, inherit simplified switches info for NewBB,
// Probably clone more loop-unswitch related loop properties.
BranchesInfo.cloneData(NewLoop, L, VMap);
Loop *ParentLoop = L->getParentLoop();
if (ParentLoop) {
// Make sure to add the cloned preheader and exit blocks to the parent loop
// as well.
ParentLoop->addBasicBlockToLoop(NewBlocks[0], LI->getBase());
}
for (unsigned i = 0, e = ExitBlocks.size(); i != e; ++i) {
BasicBlock *NewExit = cast<BasicBlock>(VMap[ExitBlocks[i]]);
// The new exit block should be in the same loop as the old one.
if (Loop *ExitBBLoop = LI->getLoopFor(ExitBlocks[i]))
ExitBBLoop->addBasicBlockToLoop(NewExit, LI->getBase());
assert(NewExit->getTerminator()->getNumSuccessors() == 1 &&
"Exit block should have been split to have one successor!");
BasicBlock *ExitSucc = NewExit->getTerminator()->getSuccessor(0);
// If the successor of the exit block had PHI nodes, add an entry for
// NewExit.
for (BasicBlock::iterator I = ExitSucc->begin();
PHINode *PN = dyn_cast<PHINode>(I); ++I) {
Value *V = PN->getIncomingValueForBlock(ExitBlocks[i]);
ValueToValueMapTy::iterator It = VMap.find(V);
if (It != VMap.end()) V = It->second;
PN->addIncoming(V, NewExit);
}
if (LandingPadInst *LPad = NewExit->getLandingPadInst()) {
PHINode *PN = PHINode::Create(LPad->getType(), 0, "",
ExitSucc->getFirstInsertionPt());
for (pred_iterator I = pred_begin(ExitSucc), E = pred_end(ExitSucc);
I != E; ++I) {
BasicBlock *BB = *I;
LandingPadInst *LPI = BB->getLandingPadInst();
LPI->replaceAllUsesWith(PN);
PN->addIncoming(LPI, BB);
}
}
}
// Rewrite the code to refer to itself.
for (unsigned i = 0, e = NewBlocks.size(); i != e; ++i)
for (BasicBlock::iterator I = NewBlocks[i]->begin(),
E = NewBlocks[i]->end(); I != E; ++I)
RemapInstruction(I, VMap,RF_NoModuleLevelChanges|RF_IgnoreMissingEntries);
// Rewrite the original preheader to select between versions of the loop.
BranchInst *OldBR = cast<BranchInst>(loopPreheader->getTerminator());
assert(OldBR->isUnconditional() && OldBR->getSuccessor(0) == LoopBlocks[0] &&
"Preheader splitting did not work correctly!");
// Emit the new branch that selects between the two versions of this loop.
EmitPreheaderBranchOnCondition(LIC, Val, NewBlocks[0], LoopBlocks[0], OldBR);
LPM->deleteSimpleAnalysisValue(OldBR, L);
OldBR->eraseFromParent();
LoopProcessWorklist.push_back(NewLoop);
redoLoop = true;
// Keep a WeakVH holding onto LIC. If the first call to RewriteLoopBody
// deletes the instruction (for example by simplifying a PHI that feeds into
// the condition that we're unswitching on), we don't rewrite the second
// iteration.
WeakVH LICHandle(LIC);
// Now we rewrite the original code to know that the condition is true and the
// new code to know that the condition is false.
RewriteLoopBodyWithConditionConstant(L, LIC, Val, false);
// It's possible that simplifying one loop could cause the other to be
// changed to another value or a constant. If its a constant, don't simplify
// it.
if (!LoopProcessWorklist.empty() && LoopProcessWorklist.back() == NewLoop &&
LICHandle && !isa<Constant>(LICHandle))
RewriteLoopBodyWithConditionConstant(NewLoop, LICHandle, Val, true);
}
void WorklessInstrument::CloneInnerLoop(Loop * pLoop, vector<BasicBlock *> & vecAdd, ValueToValueMapTy & VMap, set<BasicBlock *> & setCloned)
{
Function * pFunction = pLoop->getHeader()->getParent();
BasicBlock * pPreHeader = vecAdd[0];
SmallVector<BasicBlock *, 4> ExitBlocks;
pLoop->getExitBlocks(ExitBlocks);
set<BasicBlock *> setExitBlocks;
for(unsigned long i = 0; i < ExitBlocks.size(); i++)
{
setExitBlocks.insert(ExitBlocks[i]);
}
for(unsigned long i = 0; i < ExitBlocks.size(); i++ )
{
VMap[ExitBlocks[i]] = ExitBlocks[i];
}
vector<BasicBlock *> ToClone;
vector<BasicBlock *> BeenCloned;
//clone loop
ToClone.push_back(pLoop->getHeader());
while(ToClone.size()>0)
{
BasicBlock * pCurrent = ToClone.back();
ToClone.pop_back();
WeakVH & BBEntry = VMap[pCurrent];
if (BBEntry)
{
continue;
}
BasicBlock * NewBB;
BBEntry = NewBB = BasicBlock::Create(pCurrent->getContext(), "", pFunction);
if(pCurrent->hasName())
{
NewBB->setName(pCurrent->getName() + ".CPI");
}
if(pCurrent->hasAddressTaken())
{
errs() << "hasAddressTaken branch\n" ;
exit(0);
}
for(BasicBlock::const_iterator II = pCurrent->begin(); II != pCurrent->end(); ++II )
{
Instruction * NewInst = II->clone();
if(II->hasName())
{
NewInst->setName(II->getName() + ".CPI");
}
VMap[II] = NewInst;
NewBB->getInstList().push_back(NewInst);
}
const TerminatorInst *TI = pCurrent->getTerminator();
for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
{
ToClone.push_back(TI->getSuccessor(i));
}
setCloned.insert(NewBB);
BeenCloned.push_back(NewBB);
}
//remap value used inside loop
vector<BasicBlock *>::iterator itVecBegin = BeenCloned.begin();
vector<BasicBlock *>::iterator itVecEnd = BeenCloned.end();
for(; itVecBegin != itVecEnd; itVecBegin ++)
{
for(BasicBlock::iterator II = (*itVecBegin)->begin(); II != (*itVecBegin)->end(); II ++ )
{
//II->dump();
RemapInstruction(II, VMap);
}
}
//add to the else if body
BasicBlock * pElseBody = vecAdd[1];
BasicBlock * pClonedHeader = cast<BasicBlock>(VMap[pLoop->getHeader()]);
BranchInst::Create(pClonedHeader, pElseBody);
//errs() << pPreHeader->getName() << "\n";
for(BasicBlock::iterator II = pClonedHeader->begin(); II != pClonedHeader->end(); II ++ )
{
if(PHINode * pPHI = dyn_cast<PHINode>(II))
{
vector<int> vecToRemoved;
for (unsigned i = 0, e = pPHI->getNumIncomingValues(); i != e; ++i)
{
if(pPHI->getIncomingBlock(i) == pPreHeader)
{
pPHI->setIncomingBlock(i, pElseBody);
}
}
}
}
set<BasicBlock *> setProcessedBlock;
for(unsigned long i = 0; i < ExitBlocks.size(); i++ )
{
if(setProcessedBlock.find(ExitBlocks[i]) != setProcessedBlock.end() )
{
continue;
}
else
{
setProcessedBlock.insert(ExitBlocks[i]);
}
for(BasicBlock::iterator II = ExitBlocks[i]->begin(); II != ExitBlocks[i]->end(); II ++ )
{
if(PHINode * pPHI = dyn_cast<PHINode>(II))
{
unsigned numIncomming = pPHI->getNumIncomingValues();
for(unsigned i = 0; i<numIncomming; i++)
{
BasicBlock * incommingBlock = pPHI->getIncomingBlock(i);
if(VMap.find(incommingBlock) != VMap.end() )
{
Value * incommingValue = pPHI->getIncomingValue(i);
if(VMap.find(incommingValue) != VMap.end() )
{
incommingValue = VMap[incommingValue];
}
pPHI->addIncoming(incommingValue, cast<BasicBlock>(VMap[incommingBlock]));
}
}
}
}
}
}
/// CloneAndPruneFunctionInto - This works exactly like CloneFunctionInto,
/// except that it does some simple constant prop and DCE on the fly. The
/// effect of this is to copy significantly less code in cases where (for
/// example) a function call with constant arguments is inlined, and those
/// constant arguments cause a significant amount of code in the callee to be
/// dead. Since this doesn't produce an exact copy of the input, it can't be
/// used for things like CloneFunction or CloneModule.
void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc,
ValueToValueMapTy &VMap,
bool ModuleLevelChanges,
SmallVectorImpl<ReturnInst*> &Returns,
const char *NameSuffix,
ClonedCodeInfo *CodeInfo,
const TargetData *TD,
Instruction *TheCall) {
assert(NameSuffix && "NameSuffix cannot be null!");
#ifndef NDEBUG
for (Function::const_arg_iterator II = OldFunc->arg_begin(),
E = OldFunc->arg_end(); II != E; ++II)
assert(VMap.count(II) && "No mapping from source argument specified!");
#endif
PruningFunctionCloner PFC(NewFunc, OldFunc, VMap, ModuleLevelChanges,
Returns, NameSuffix, CodeInfo, TD);
// Clone the entry block, and anything recursively reachable from it.
std::vector<const BasicBlock*> CloneWorklist;
CloneWorklist.push_back(&OldFunc->getEntryBlock());
while (!CloneWorklist.empty()) {
const BasicBlock *BB = CloneWorklist.back();
CloneWorklist.pop_back();
PFC.CloneBlock(BB, CloneWorklist);
}
// Loop over all of the basic blocks in the old function. If the block was
// reachable, we have cloned it and the old block is now in the value map:
// insert it into the new function in the right order. If not, ignore it.
//
// Defer PHI resolution until rest of function is resolved.
SmallVector<const PHINode*, 16> PHIToResolve;
for (Function::const_iterator BI = OldFunc->begin(), BE = OldFunc->end();
BI != BE; ++BI) {
Value *V = VMap[BI];
BasicBlock *NewBB = cast_or_null<BasicBlock>(V);
if (NewBB == 0) continue; // Dead block.
// Add the new block to the new function.
NewFunc->getBasicBlockList().push_back(NewBB);
// Loop over all of the instructions in the block, fixing up operand
// references as we go. This uses VMap to do all the hard work.
//
BasicBlock::iterator I = NewBB->begin();
DebugLoc TheCallDL;
if (TheCall)
TheCallDL = TheCall->getDebugLoc();
// Handle PHI nodes specially, as we have to remove references to dead
// blocks.
if (PHINode *PN = dyn_cast<PHINode>(I)) {
// Skip over all PHI nodes, remembering them for later.
BasicBlock::const_iterator OldI = BI->begin();
for (; (PN = dyn_cast<PHINode>(I)); ++I, ++OldI)
PHIToResolve.push_back(cast<PHINode>(OldI));
}
// Otherwise, remap the rest of the instructions normally.
for (; I != NewBB->end(); ++I)
RemapInstruction(I, VMap,
ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges);
}
// Defer PHI resolution until rest of function is resolved, PHI resolution
// requires the CFG to be up-to-date.
for (unsigned phino = 0, e = PHIToResolve.size(); phino != e; ) {
const PHINode *OPN = PHIToResolve[phino];
unsigned NumPreds = OPN->getNumIncomingValues();
const BasicBlock *OldBB = OPN->getParent();
BasicBlock *NewBB = cast<BasicBlock>(VMap[OldBB]);
// Map operands for blocks that are live and remove operands for blocks
// that are dead.
for (; phino != PHIToResolve.size() &&
PHIToResolve[phino]->getParent() == OldBB; ++phino) {
OPN = PHIToResolve[phino];
PHINode *PN = cast<PHINode>(VMap[OPN]);
for (unsigned pred = 0, e = NumPreds; pred != e; ++pred) {
Value *V = VMap[PN->getIncomingBlock(pred)];
if (BasicBlock *MappedBlock = cast_or_null<BasicBlock>(V)) {
Value *InVal = MapValue(PN->getIncomingValue(pred),
VMap,
ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges);
assert(InVal && "Unknown input value?");
PN->setIncomingValue(pred, InVal);
PN->setIncomingBlock(pred, MappedBlock);
} else {
PN->removeIncomingValue(pred, false);
--pred, --e; // Revisit the next entry.
}
}
}
// The loop above has removed PHI entries for those blocks that are dead
// and has updated others. However, if a block is live (i.e. copied over)
// but its terminator has been changed to not go to this block, then our
// phi nodes will have invalid entries. Update the PHI nodes in this
// case.
PHINode *PN = cast<PHINode>(NewBB->begin());
NumPreds = std::distance(pred_begin(NewBB), pred_end(NewBB));
if (NumPreds != PN->getNumIncomingValues()) {
assert(NumPreds < PN->getNumIncomingValues());
// Count how many times each predecessor comes to this block.
std::map<BasicBlock*, unsigned> PredCount;
for (pred_iterator PI = pred_begin(NewBB), E = pred_end(NewBB);
PI != E; ++PI)
--PredCount[*PI];
// Figure out how many entries to remove from each PHI.
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
++PredCount[PN->getIncomingBlock(i)];
// At this point, the excess predecessor entries are positive in the
// map. Loop over all of the PHIs and remove excess predecessor
// entries.
BasicBlock::iterator I = NewBB->begin();
for (; (PN = dyn_cast<PHINode>(I)); ++I) {
for (std::map<BasicBlock*, unsigned>::iterator PCI =PredCount.begin(),
E = PredCount.end(); PCI != E; ++PCI) {
BasicBlock *Pred = PCI->first;
for (unsigned NumToRemove = PCI->second; NumToRemove; --NumToRemove)
PN->removeIncomingValue(Pred, false);
}
}
}
// If the loops above have made these phi nodes have 0 or 1 operand,
// replace them with undef or the input value. We must do this for
// correctness, because 0-operand phis are not valid.
PN = cast<PHINode>(NewBB->begin());
if (PN->getNumIncomingValues() == 0) {
BasicBlock::iterator I = NewBB->begin();
BasicBlock::const_iterator OldI = OldBB->begin();
while ((PN = dyn_cast<PHINode>(I++))) {
Value *NV = UndefValue::get(PN->getType());
PN->replaceAllUsesWith(NV);
assert(VMap[OldI] == PN && "VMap mismatch");
VMap[OldI] = NV;
PN->eraseFromParent();
++OldI;
}
}
// NOTE: We cannot eliminate single entry phi nodes here, because of
// VMap. Single entry phi nodes can have multiple VMap entries
// pointing at them. Thus, deleting one would require scanning the VMap
// to update any entries in it that would require that. This would be
// really slow.
}
// Now that the inlined function body has been fully constructed, go through
// and zap unconditional fall-through branches. This happen all the time when
// specializing code: code specialization turns conditional branches into
// uncond branches, and this code folds them.
Function::iterator I = cast<BasicBlock>(VMap[&OldFunc->getEntryBlock()]);
while (I != NewFunc->end()) {
BranchInst *BI = dyn_cast<BranchInst>(I->getTerminator());
if (!BI || BI->isConditional()) {
++I;
continue;
}
// Note that we can't eliminate uncond branches if the destination has
// single-entry PHI nodes. Eliminating the single-entry phi nodes would
// require scanning the VMap to update any entries that point to the phi
// node.
BasicBlock *Dest = BI->getSuccessor(0);
if (!Dest->getSinglePredecessor() || isa<PHINode>(Dest->begin())) {
++I;
continue;
}
// We know all single-entry PHI nodes in the inlined function have been
// removed, so we just need to splice the blocks.
BI->eraseFromParent();
// Make all PHI nodes that referred to Dest now refer to I as their source.
Dest->replaceAllUsesWith(I);
// Move all the instructions in the succ to the pred.
I->getInstList().splice(I->end(), Dest->getInstList());
// Remove the dest block.
Dest->eraseFromParent();
// Do not increment I, iteratively merge all things this block branches to.
}
}