/// Convert the instruction operands from referencing the current values into
/// those specified by VMap.
static inline void remapInstruction(Instruction *I,
ValueToValueMapTy &VMap) {
for (unsigned op = 0, E = I->getNumOperands(); op != E; ++op) {
Value *Op = I->getOperand(op);
// Unwrap arguments of dbg.value intrinsics.
bool Wrapped = false;
if (auto *V = dyn_cast<MetadataAsValue>(Op))
if (auto *Unwrapped = dyn_cast<ValueAsMetadata>(V->getMetadata())) {
Op = Unwrapped->getValue();
Wrapped = true;
}
auto wrap = [&](Value *V) {
auto &C = I->getContext();
return Wrapped ? MetadataAsValue::get(C, ValueAsMetadata::get(V)) : V;
};
ValueToValueMapTy::iterator It = VMap.find(Op);
if (It != VMap.end())
I->setOperand(op, wrap(It->second));
}
if (PHINode *PN = dyn_cast<PHINode>(I)) {
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
ValueToValueMapTy::iterator It = VMap.find(PN->getIncomingBlock(i));
if (It != VMap.end())
PN->setIncomingBlock(i, cast<BasicBlock>(It->second));
}
}
}
/// Propagate dbg.value intrinsics through the newly inserted Phis.
static void insertDebugValues(BasicBlock *OrigHeader,
SmallVectorImpl<PHINode*> &InsertedPHIs) {
ValueToValueMapTy DbgValueMap;
// Map existing PHI nodes to their dbg.values.
for (auto &I : *OrigHeader) {
if (auto DbgII = dyn_cast<DbgInfoIntrinsic>(&I)) {
if (auto *Loc = dyn_cast_or_null<PHINode>(DbgII->getVariableLocation()))
DbgValueMap.insert({Loc, DbgII});
}
}
// Then iterate through the new PHIs and look to see if they use one of the
// previously mapped PHIs. If so, insert a new dbg.value intrinsic that will
// propagate the info through the new PHI.
LLVMContext &C = OrigHeader->getContext();
for (auto PHI : InsertedPHIs) {
for (auto VI : PHI->operand_values()) {
auto V = DbgValueMap.find(VI);
if (V != DbgValueMap.end()) {
auto *DbgII = cast<DbgInfoIntrinsic>(V->second);
Instruction *NewDbgII = DbgII->clone();
auto PhiMAV = MetadataAsValue::get(C, ValueAsMetadata::get(PHI));
NewDbgII->setOperand(0, PhiMAV);
BasicBlock *Parent = PHI->getParent();
NewDbgII->insertBefore(Parent->getFirstNonPHIOrDbgOrLifetime());
}
}
}
}
ValueToValueMapTy* MCJITHelper::generateIdentityMapping(Function* F) {
ValueToValueMapTy* VMap = new ValueToValueMapTy();
// arguments
for (Function::arg_iterator argIt = F->arg_begin(), argEnd = F->arg_end(); argIt != argEnd; ++argIt) {
(*VMap)[argIt] = argIt;
}
// TODO: metadata (also in the body?)
for (Function::iterator bbIt = F->begin(), bbEnd = F->end(); bbIt != bbEnd; ++bbIt) {
// basic blocks
(*VMap)[bbIt] = bbIt;
// TODO: BB.hasAddressTaken() ?
for (BasicBlock::iterator insIt = bbIt->begin(), insEnd = bbIt->end(); insIt != insEnd; ++insIt) {
// instructions
(*VMap)[insIt] = insIt;
/** Code adapted from MapValue() called by RemapInstruction() **/
// operators
for (User::op_iterator opIt = insIt->op_begin(), opEnd = insIt->op_end(); opIt != opEnd; ++opIt) {
Value* v = *opIt;
ValueToValueMapTy::iterator I = VMap->find(v);
if (I != VMap->end() && I->second) continue; // value already exists in VMap
(*VMap)[v] = v;
}
}
}
return VMap;
}
/// UpdateCallGraphAfterInlining - Once we have cloned code over from a callee
/// into the caller, update the specified callgraph to reflect the changes we
/// made. Note that it's possible that not all code was copied over, so only
/// some edges of the callgraph may remain.
static void UpdateCallGraphAfterInlining(CallSite CS,
Function::iterator FirstNewBlock,
ValueToValueMapTy &VMap,
InlineFunctionInfo &IFI) {
CallGraph &CG = *IFI.CG;
const Function *Caller = CS.getInstruction()->getParent()->getParent();
const Function *Callee = CS.getCalledFunction();
CallGraphNode *CalleeNode = CG[Callee];
CallGraphNode *CallerNode = CG[Caller];
// Since we inlined some uninlined call sites in the callee into the caller,
// add edges from the caller to all of the callees of the callee.
CallGraphNode::iterator I = CalleeNode->begin(), E = CalleeNode->end();
// Consider the case where CalleeNode == CallerNode.
CallGraphNode::CalledFunctionsVector CallCache;
if (CalleeNode == CallerNode) {
CallCache.assign(I, E);
I = CallCache.begin();
E = CallCache.end();
}
for (; I != E; ++I) {
const Value *OrigCall = I->first;
ValueToValueMapTy::iterator VMI = VMap.find(OrigCall);
// Only copy the edge if the call was inlined!
if (VMI == VMap.end() || VMI->second == 0)
continue;
// If the call was inlined, but then constant folded, there is no edge to
// add. Check for this case.
Instruction *NewCall = dyn_cast<Instruction>(VMI->second);
if (NewCall == 0) continue;
// Remember that this call site got inlined for the client of
// InlineFunction.
IFI.InlinedCalls.push_back(NewCall);
// It's possible that inlining the callsite will cause it to go from an
// indirect to a direct call by resolving a function pointer. If this
// happens, set the callee of the new call site to a more precise
// destination. This can also happen if the call graph node of the caller
// was just unnecessarily imprecise.
if (I->second->getFunction() == 0)
if (Function *F = CallSite(NewCall).getCalledFunction()) {
// Indirect call site resolved to direct call.
CallerNode->addCalledFunction(CallSite(NewCall), CG[F]);
continue;
}
CallerNode->addCalledFunction(CallSite(NewCall), I->second);
}
// Update the call graph by deleting the edge from Callee to Caller. We must
// do this after the loop above in case Caller and Callee are the same.
CallerNode->removeCallEdgeFor(CS);
}
/// RemapInstruction - Convert the instruction operands from referencing the
/// current values into those specified by VMap.
static inline void RemapInstruction(Instruction *I,
ValueToValueMapTy &VMap) {
for (unsigned op = 0, E = I->getNumOperands(); op != E; ++op) {
Value *Op = I->getOperand(op);
ValueToValueMapTy::iterator It = VMap.find(Op);
if (It != VMap.end())
I->setOperand(op, It->second);
}
if (PHINode *PN = dyn_cast<PHINode>(I)) {
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
ValueToValueMapTy::iterator It = VMap.find(PN->getIncomingBlock(i));
if (It != VMap.end())
PN->setIncomingBlock(i, cast<BasicBlock>(It->second));
}
}
}
/// Unroll the given loop by Count. The loop must be in LCSSA form. Returns true
/// if unrolling was successful, or false if the loop was unmodified. Unrolling
/// can only fail when the loop's latch block is not terminated by a conditional
/// branch instruction. However, if the trip count (and multiple) are not known,
/// loop unrolling will mostly produce more code that is no faster.
///
/// TripCount is generally defined as the number of times the loop header
/// executes. UnrollLoop relaxes the definition to permit early exits: here
/// TripCount is the iteration on which control exits LatchBlock if no early
/// exits were taken. Note that UnrollLoop assumes that the loop counter test
/// terminates LatchBlock in order to remove unnecesssary instances of the
/// test. In other words, control may exit the loop prior to TripCount
/// iterations via an early branch, but control may not exit the loop from the
/// LatchBlock's terminator prior to TripCount iterations.
///
/// Similarly, TripMultiple divides the number of times that the LatchBlock may
/// execute without exiting the loop.
///
/// The LoopInfo Analysis that is passed will be kept consistent.
///
/// If a LoopPassManager is passed in, and the loop is fully removed, it will be
/// removed from the LoopPassManager as well. LPM can also be NULL.
///
/// This utility preserves LoopInfo. If DominatorTree or ScalarEvolution are
/// available from the Pass it must also preserve those analyses.
bool llvm::UnrollLoop(Loop *L, unsigned Count, unsigned TripCount,
bool AllowRuntime, unsigned TripMultiple,
LoopInfo *LI, Pass *PP, LPPassManager *LPM) {
BasicBlock *Preheader = L->getLoopPreheader();
if (!Preheader) {
DEBUG(dbgs() << " Can't unroll; loop preheader-insertion failed.\n");
return false;
}
BasicBlock *LatchBlock = L->getLoopLatch();
if (!LatchBlock) {
DEBUG(dbgs() << " Can't unroll; loop exit-block-insertion failed.\n");
return false;
}
// Loops with indirectbr cannot be cloned.
if (!L->isSafeToClone()) {
DEBUG(dbgs() << " Can't unroll; Loop body cannot be cloned.\n");
return false;
}
BasicBlock *Header = L->getHeader();
BranchInst *BI = dyn_cast<BranchInst>(LatchBlock->getTerminator());
if (!BI || BI->isUnconditional()) {
// The loop-rotate pass can be helpful to avoid this in many cases.
DEBUG(dbgs() <<
" Can't unroll; loop not terminated by a conditional branch.\n");
return false;
}
if (Header->hasAddressTaken()) {
// The loop-rotate pass can be helpful to avoid this in many cases.
DEBUG(dbgs() <<
" Won't unroll loop: address of header block is taken.\n");
return false;
}
if (TripCount != 0)
DEBUG(dbgs() << " Trip Count = " << TripCount << "\n");
if (TripMultiple != 1)
DEBUG(dbgs() << " Trip Multiple = " << TripMultiple << "\n");
// Effectively "DCE" unrolled iterations that are beyond the tripcount
// and will never be executed.
if (TripCount != 0 && Count > TripCount)
Count = TripCount;
// Don't enter the unroll code if there is nothing to do. This way we don't
// need to support "partial unrolling by 1".
if (TripCount == 0 && Count < 2)
return false;
assert(Count > 0);
assert(TripMultiple > 0);
assert(TripCount == 0 || TripCount % TripMultiple == 0);
// Are we eliminating the loop control altogether?
bool CompletelyUnroll = Count == TripCount;
// We assume a run-time trip count if the compiler cannot
// figure out the loop trip count and the unroll-runtime
// flag is specified.
bool RuntimeTripCount = (TripCount == 0 && Count > 0 && AllowRuntime);
if (RuntimeTripCount && !UnrollRuntimeLoopProlog(L, Count, LI, LPM))
return false;
// Notify ScalarEvolution that the loop will be substantially changed,
// if not outright eliminated.
if (PP) {
ScalarEvolution *SE = PP->getAnalysisIfAvailable<ScalarEvolution>();
if (SE)
SE->forgetLoop(L);
}
// If we know the trip count, we know the multiple...
unsigned BreakoutTrip = 0;
if (TripCount != 0) {
BreakoutTrip = TripCount % Count;
TripMultiple = 0;
} else {
// Figure out what multiple to use.
BreakoutTrip = TripMultiple =
(unsigned)GreatestCommonDivisor64(Count, TripMultiple);
}
// Report the unrolling decision.
DebugLoc LoopLoc = L->getStartLoc();
Function *F = Header->getParent();
LLVMContext &Ctx = F->getContext();
if (CompletelyUnroll) {
DEBUG(dbgs() << "COMPLETELY UNROLLING loop %" << Header->getName()
<< " with trip count " << TripCount << "!\n");
emitOptimizationRemark(Ctx, DEBUG_TYPE, *F, LoopLoc,
Twine("completely unrolled loop with ") +
Twine(TripCount) + " iterations");
} else {
DEBUG(dbgs() << "UNROLLING loop %" << Header->getName()
<< " by " << Count);
Twine DiagMsg("unrolled loop by a factor of " + Twine(Count));
if (TripMultiple == 0 || BreakoutTrip != TripMultiple) {
DEBUG(dbgs() << " with a breakout at trip " << BreakoutTrip);
DiagMsg.concat(" with a breakout at trip " + Twine(BreakoutTrip));
} else if (TripMultiple != 1) {
DEBUG(dbgs() << " with " << TripMultiple << " trips per branch");
DiagMsg.concat(" with " + Twine(TripMultiple) + " trips per branch");
} else if (RuntimeTripCount) {
DEBUG(dbgs() << " with run-time trip count");
DiagMsg.concat(" with run-time trip count");
}
DEBUG(dbgs() << "!\n");
emitOptimizationRemark(Ctx, DEBUG_TYPE, *F, LoopLoc, DiagMsg);
}
bool ContinueOnTrue = L->contains(BI->getSuccessor(0));
BasicBlock *LoopExit = BI->getSuccessor(ContinueOnTrue);
// For the first iteration of the loop, we should use the precloned values for
// PHI nodes. Insert associations now.
ValueToValueMapTy LastValueMap;
std::vector<PHINode*> OrigPHINode;
for (BasicBlock::iterator I = Header->begin(); isa<PHINode>(I); ++I) {
OrigPHINode.push_back(cast<PHINode>(I));
}
std::vector<BasicBlock*> Headers;
std::vector<BasicBlock*> Latches;
Headers.push_back(Header);
Latches.push_back(LatchBlock);
// The current on-the-fly SSA update requires blocks to be processed in
// reverse postorder so that LastValueMap contains the correct value at each
// exit.
LoopBlocksDFS DFS(L);
DFS.perform(LI);
// Stash the DFS iterators before adding blocks to the loop.
LoopBlocksDFS::RPOIterator BlockBegin = DFS.beginRPO();
LoopBlocksDFS::RPOIterator BlockEnd = DFS.endRPO();
for (unsigned It = 1; It != Count; ++It) {
std::vector<BasicBlock*> NewBlocks;
for (LoopBlocksDFS::RPOIterator BB = BlockBegin; BB != BlockEnd; ++BB) {
ValueToValueMapTy VMap;
BasicBlock *New = CloneBasicBlock(*BB, VMap, "." + Twine(It));
Header->getParent()->getBasicBlockList().push_back(New);
// Loop over all of the PHI nodes in the block, changing them to use the
// incoming values from the previous block.
if (*BB == Header)
for (unsigned i = 0, e = OrigPHINode.size(); i != e; ++i) {
PHINode *NewPHI = cast<PHINode>(VMap[OrigPHINode[i]]);
Value *InVal = NewPHI->getIncomingValueForBlock(LatchBlock);
if (Instruction *InValI = dyn_cast<Instruction>(InVal))
if (It > 1 && L->contains(InValI))
InVal = LastValueMap[InValI];
VMap[OrigPHINode[i]] = InVal;
New->getInstList().erase(NewPHI);
}
// Update our running map of newest clones
LastValueMap[*BB] = New;
for (ValueToValueMapTy::iterator VI = VMap.begin(), VE = VMap.end();
VI != VE; ++VI)
LastValueMap[VI->first] = VI->second;
L->addBasicBlockToLoop(New, LI->getBase());
// Add phi entries for newly created values to all exit blocks.
for (succ_iterator SI = succ_begin(*BB), SE = succ_end(*BB);
SI != SE; ++SI) {
if (L->contains(*SI))
continue;
for (BasicBlock::iterator BBI = (*SI)->begin();
PHINode *phi = dyn_cast<PHINode>(BBI); ++BBI) {
Value *Incoming = phi->getIncomingValueForBlock(*BB);
ValueToValueMapTy::iterator It = LastValueMap.find(Incoming);
if (It != LastValueMap.end())
Incoming = It->second;
phi->addIncoming(Incoming, New);
}
}
// Keep track of new headers and latches as we create them, so that
// we can insert the proper branches later.
if (*BB == Header)
Headers.push_back(New);
if (*BB == LatchBlock)
Latches.push_back(New);
NewBlocks.push_back(New);
}
// Remap all instructions in the most recent iteration
for (unsigned i = 0; i < NewBlocks.size(); ++i)
for (BasicBlock::iterator I = NewBlocks[i]->begin(),
E = NewBlocks[i]->end(); I != E; ++I)
::RemapInstruction(I, LastValueMap);
}
// Loop over the PHI nodes in the original block, setting incoming values.
for (unsigned i = 0, e = OrigPHINode.size(); i != e; ++i) {
PHINode *PN = OrigPHINode[i];
if (CompletelyUnroll) {
PN->replaceAllUsesWith(PN->getIncomingValueForBlock(Preheader));
Header->getInstList().erase(PN);
}
else if (Count > 1) {
Value *InVal = PN->removeIncomingValue(LatchBlock, false);
// If this value was defined in the loop, take the value defined by the
// last iteration of the loop.
if (Instruction *InValI = dyn_cast<Instruction>(InVal)) {
if (L->contains(InValI))
InVal = LastValueMap[InVal];
}
assert(Latches.back() == LastValueMap[LatchBlock] && "bad last latch");
PN->addIncoming(InVal, Latches.back());
}
}
// Now that all the basic blocks for the unrolled iterations are in place,
// set up the branches to connect them.
for (unsigned i = 0, e = Latches.size(); i != e; ++i) {
// The original branch was replicated in each unrolled iteration.
BranchInst *Term = cast<BranchInst>(Latches[i]->getTerminator());
// The branch destination.
unsigned j = (i + 1) % e;
BasicBlock *Dest = Headers[j];
bool NeedConditional = true;
if (RuntimeTripCount && j != 0) {
NeedConditional = false;
}
// For a complete unroll, make the last iteration end with a branch
// to the exit block.
if (CompletelyUnroll && j == 0) {
Dest = LoopExit;
NeedConditional = false;
}
// If we know the trip count or a multiple of it, we can safely use an
// unconditional branch for some iterations.
if (j != BreakoutTrip && (TripMultiple == 0 || j % TripMultiple != 0)) {
NeedConditional = false;
}
if (NeedConditional) {
// Update the conditional branch's successor for the following
// iteration.
Term->setSuccessor(!ContinueOnTrue, Dest);
} else {
// Remove phi operands at this loop exit
if (Dest != LoopExit) {
BasicBlock *BB = Latches[i];
for (succ_iterator SI = succ_begin(BB), SE = succ_end(BB);
SI != SE; ++SI) {
if (*SI == Headers[i])
continue;
for (BasicBlock::iterator BBI = (*SI)->begin();
PHINode *Phi = dyn_cast<PHINode>(BBI); ++BBI) {
Phi->removeIncomingValue(BB, false);
}
}
}
// Replace the conditional branch with an unconditional one.
BranchInst::Create(Dest, Term);
Term->eraseFromParent();
}
}
// Merge adjacent basic blocks, if possible.
for (unsigned i = 0, e = Latches.size(); i != e; ++i) {
BranchInst *Term = cast<BranchInst>(Latches[i]->getTerminator());
if (Term->isUnconditional()) {
BasicBlock *Dest = Term->getSuccessor(0);
if (BasicBlock *Fold = FoldBlockIntoPredecessor(Dest, LI, LPM))
std::replace(Latches.begin(), Latches.end(), Dest, Fold);
}
}
DominatorTree *DT = nullptr;
if (PP) {
// FIXME: Reconstruct dom info, because it is not preserved properly.
// Incrementally updating domtree after loop unrolling would be easy.
if (DominatorTreeWrapperPass *DTWP =
PP->getAnalysisIfAvailable<DominatorTreeWrapperPass>()) {
DT = &DTWP->getDomTree();
DT->recalculate(*L->getHeader()->getParent());
}
// Simplify any new induction variables in the partially unrolled loop.
ScalarEvolution *SE = PP->getAnalysisIfAvailable<ScalarEvolution>();
if (SE && !CompletelyUnroll) {
SmallVector<WeakVH, 16> DeadInsts;
simplifyLoopIVs(L, SE, LPM, DeadInsts);
// Aggressively clean up dead instructions that simplifyLoopIVs already
// identified. Any remaining should be cleaned up below.
while (!DeadInsts.empty())
if (Instruction *Inst =
dyn_cast_or_null<Instruction>(&*DeadInsts.pop_back_val()))
RecursivelyDeleteTriviallyDeadInstructions(Inst);
}
}
// At this point, the code is well formed. We now do a quick sweep over the
// inserted code, doing constant propagation and dead code elimination as we
// go.
const std::vector<BasicBlock*> &NewLoopBlocks = L->getBlocks();
for (std::vector<BasicBlock*>::const_iterator BB = NewLoopBlocks.begin(),
BBE = NewLoopBlocks.end(); BB != BBE; ++BB)
for (BasicBlock::iterator I = (*BB)->begin(), E = (*BB)->end(); I != E; ) {
Instruction *Inst = I++;
if (isInstructionTriviallyDead(Inst))
(*BB)->getInstList().erase(Inst);
else if (Value *V = SimplifyInstruction(Inst))
if (LI->replacementPreservesLCSSAForm(Inst, V)) {
Inst->replaceAllUsesWith(V);
(*BB)->getInstList().erase(Inst);
}
}
NumCompletelyUnrolled += CompletelyUnroll;
++NumUnrolled;
Loop *OuterL = L->getParentLoop();
// Remove the loop from the LoopPassManager if it's completely removed.
if (CompletelyUnroll && LPM != nullptr)
LPM->deleteLoopFromQueue(L);
// If we have a pass and a DominatorTree we should re-simplify impacted loops
// to ensure subsequent analyses can rely on this form. We want to simplify
// at least one layer outside of the loop that was unrolled so that any
// changes to the parent loop exposed by the unrolling are considered.
if (PP && DT) {
if (!OuterL && !CompletelyUnroll)
OuterL = L;
if (OuterL) {
ScalarEvolution *SE = PP->getAnalysisIfAvailable<ScalarEvolution>();
simplifyLoop(OuterL, DT, LI, PP, /*AliasAnalysis*/ nullptr, SE);
// LCSSA must be performed on the outermost affected loop. The unrolled
// loop's last loop latch is guaranteed to be in the outermost loop after
// deleteLoopFromQueue updates LoopInfo.
Loop *LatchLoop = LI->getLoopFor(Latches.back());
if (!OuterL->contains(LatchLoop))
while (OuterL->getParentLoop() != LatchLoop)
OuterL = OuterL->getParentLoop();
formLCSSARecursively(*OuterL, *DT, SE);
}
}
return true;
}
/// 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);
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.
PHINode *PN;
for (BasicBlock::iterator I = ExitSucc->begin(); isa<PHINode>(I); ++I) {
PN = cast<PHINode>(I);
Value *V = PN->getIncomingValueForBlock(ExitBlocks[i]);
ValueToValueMapTy::iterator It = VMap.find(V);
if (It != VMap.end()) V = It->second;
PN->addIncoming(V, NewExit);
}
}
// 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 SuperBlock::fixSideEntrances() {
// due to merging of BBs, some superblocks may have 1 BB remaining
list<map<BasicBlock*, list<BasicBlock*> >::iterator > delSuperBlocks;
for (map<BasicBlock*, list<BasicBlock*> >::iterator sp = superBlocks.begin(),
sp_e = superBlocks.end(); sp != sp_e; ++sp) {
// we need to keep track of the predecessor of the current basic block
// being checked
BasicBlock* prev = sp->first;
// don't clone basic blocks if the code size threshold is achieved
if (currCodeSize/originalCodeSize > CODE_EXPANSION_THRESHOLD) {
break;
}
// the first basic block for a superblock need not be duplicated
for (list<BasicBlock*>::iterator bb = sp->second.begin(),
bb_e = sp->second.end(); bb != bb_e; ++bb) {
// first, collect all predecessors for this BB
// (note: we could not just iterate through as the predecessor set may
// change
list<BasicBlock*> predBBs;
for (pred_iterator pred = pred_begin(*bb), pred_e = pred_end(*bb);
pred != pred_e; ++pred) {
predBBs.push_back(*pred);
}
// now, walk through all predecessors of this current basic block
BasicBlock* clonedBB = NULL;
for (list<BasicBlock*>::iterator pred = predBBs.begin(),
pred_e = predBBs.end(); pred != pred_e; ++pred) {
// if it is not the predecessor of this current basic block present in
// the superblock, duplicate!
if (*pred != prev) {
// there is no need to clone this BB multiple times
if (clonedBB == NULL) {
ValueToValueMapTy vmap;
// clone this basic block, and place the corresponding code after
// the last BB of this superblock
clonedBB = CloneBasicBlock(*bb, vmap, ".cloned",
(*bb)->getParent());
vmap[*bb] = clonedBB;
/*
errs() << "@@ BEFORE: " << *clonedBB << "\n";
// fix phi nodes in the cloned BB
for (BasicBlock::iterator I = clonedBB->begin(); isa<PHINode>(I); ++I) {
PHINode* PN = dyn_cast<PHINode>(I);
int bbIdx = PN->getBasicBlockIndex(prev);
if (bbIdx != -1) {
PN->removeIncomingValue(bbIdx, false);
}
}
*/
// add size of duplicated BBs to total code size count
currCodeSize += clonedBB->size();
// modify operands in this basic block
for (BasicBlock::iterator instr = clonedBB->begin(),
instr_e = clonedBB->end(); instr != instr_e; ++instr) {
for (unsigned idx = 0, num_ops = instr->getNumOperands();
idx < num_ops; ++idx) {
Value* op = instr->getOperand(idx);
ValueToValueMapTy::iterator op_it = vmap.find(op);
if (op_it != vmap.end()) {
instr->setOperand(idx, op_it->second);
}
}
}
}
// remove phi nodes into this BB in the trace
/*
for (BasicBlock::iterator I = (*bb)->begin(); isa<PHINode>(I); ++I) {
PHINode* PN = dyn_cast<PHINode>(I);
int bbIdx = PN->getBasicBlockIndex(*pred);
if (bbIdx != -1) {
PN->removeIncomingValue(bbIdx, false);
}
}
*/
// modify the branch instruction of the predecessor not in the superblock to
// branch to the cloned basic block
Instruction* br_instr = (*pred)->getTerminator();
br_instr->replaceUsesOfWith((Value*)*bb, (Value*)clonedBB);
}
}
// determine if we can merge the BB (definitely can be merged), and its clone
// with theirpredecessors
if (clonedBB != NULL) {
if (MergeBlockIntoPredecessor(*bb, this)) {
// since we have merged this BB, delete from our superblock mappings
partOfSuperBlock.erase(*bb);
bb = sp->second.erase(bb);
--bb;
if (sp->second.empty()) {
delSuperBlocks.push_back(sp);
}
}
MergeBlockIntoPredecessor(clonedBB, this);
}
prev = *bb;
}
}
// erase some superblocks (which only have 1 BB remaining)
for (list<map<BasicBlock*, list<BasicBlock*> >::iterator >::iterator del =
delSuperBlocks.begin(), del_e = delSuperBlocks.end(); del != del_e;
++del) {
superBlocks.erase(*del);
}
}
void WinEHPrepare::cloneCommonBlocks(Function &F) {
// We need to clone all blocks which belong to multiple funclets. Values are
// remapped throughout the funclet to propagate both the new instructions
// *and* the new basic blocks themselves.
for (auto &Funclets : FuncletBlocks) {
BasicBlock *FuncletPadBB = Funclets.first;
std::vector<BasicBlock *> &BlocksInFunclet = Funclets.second;
Value *FuncletToken;
if (FuncletPadBB == &F.getEntryBlock())
FuncletToken = ConstantTokenNone::get(F.getContext());
else
FuncletToken = FuncletPadBB->getFirstNonPHI();
std::vector<std::pair<BasicBlock *, BasicBlock *>> Orig2Clone;
ValueToValueMapTy VMap;
for (BasicBlock *BB : BlocksInFunclet) {
ColorVector &ColorsForBB = BlockColors[BB];
// We don't need to do anything if the block is monochromatic.
size_t NumColorsForBB = ColorsForBB.size();
if (NumColorsForBB == 1)
continue;
DEBUG_WITH_TYPE("winehprepare-coloring",
dbgs() << " Cloning block \'" << BB->getName()
<< "\' for funclet \'" << FuncletPadBB->getName()
<< "\'.\n");
// Create a new basic block and copy instructions into it!
BasicBlock *CBB =
CloneBasicBlock(BB, VMap, Twine(".for.", FuncletPadBB->getName()));
// Insert the clone immediately after the original to ensure determinism
// and to keep the same relative ordering of any funclet's blocks.
CBB->insertInto(&F, BB->getNextNode());
// Add basic block mapping.
VMap[BB] = CBB;
// Record delta operations that we need to perform to our color mappings.
Orig2Clone.emplace_back(BB, CBB);
}
// If nothing was cloned, we're done cloning in this funclet.
if (Orig2Clone.empty())
continue;
// Update our color mappings to reflect that one block has lost a color and
// another has gained a color.
for (auto &BBMapping : Orig2Clone) {
BasicBlock *OldBlock = BBMapping.first;
BasicBlock *NewBlock = BBMapping.second;
BlocksInFunclet.push_back(NewBlock);
ColorVector &NewColors = BlockColors[NewBlock];
assert(NewColors.empty() && "A new block should only have one color!");
NewColors.push_back(FuncletPadBB);
DEBUG_WITH_TYPE("winehprepare-coloring",
dbgs() << " Assigned color \'" << FuncletPadBB->getName()
<< "\' to block \'" << NewBlock->getName()
<< "\'.\n");
BlocksInFunclet.erase(
std::remove(BlocksInFunclet.begin(), BlocksInFunclet.end(), OldBlock),
BlocksInFunclet.end());
ColorVector &OldColors = BlockColors[OldBlock];
OldColors.erase(
std::remove(OldColors.begin(), OldColors.end(), FuncletPadBB),
OldColors.end());
DEBUG_WITH_TYPE("winehprepare-coloring",
dbgs() << " Removed color \'" << FuncletPadBB->getName()
<< "\' from block \'" << OldBlock->getName()
<< "\'.\n");
}
// Loop over all of the instructions in this funclet, fixing up operand
// references as we go. This uses VMap to do all the hard work.
for (BasicBlock *BB : BlocksInFunclet)
// Loop over all instructions, fixing each one as we find it...
for (Instruction &I : *BB)
RemapInstruction(&I, VMap,
RF_IgnoreMissingLocals | RF_NoModuleLevelChanges);
// Catchrets targeting cloned blocks need to be updated separately from
// the loop above because they are not in the current funclet.
SmallVector<CatchReturnInst *, 2> FixupCatchrets;
for (auto &BBMapping : Orig2Clone) {
BasicBlock *OldBlock = BBMapping.first;
BasicBlock *NewBlock = BBMapping.second;
FixupCatchrets.clear();
for (BasicBlock *Pred : predecessors(OldBlock))
if (auto *CatchRet = dyn_cast<CatchReturnInst>(Pred->getTerminator()))
if (CatchRet->getCatchSwitchParentPad() == FuncletToken)
FixupCatchrets.push_back(CatchRet);
for (CatchReturnInst *CatchRet : FixupCatchrets)
CatchRet->setSuccessor(NewBlock);
}
auto UpdatePHIOnClonedBlock = [&](PHINode *PN, bool IsForOldBlock) {
unsigned NumPreds = PN->getNumIncomingValues();
for (unsigned PredIdx = 0, PredEnd = NumPreds; PredIdx != PredEnd;
++PredIdx) {
BasicBlock *IncomingBlock = PN->getIncomingBlock(PredIdx);
bool EdgeTargetsFunclet;
if (auto *CRI =
dyn_cast<CatchReturnInst>(IncomingBlock->getTerminator())) {
EdgeTargetsFunclet = (CRI->getCatchSwitchParentPad() == FuncletToken);
} else {
ColorVector &IncomingColors = BlockColors[IncomingBlock];
assert(!IncomingColors.empty() && "Block not colored!");
assert((IncomingColors.size() == 1 ||
llvm::all_of(IncomingColors,
[&](BasicBlock *Color) {
return Color != FuncletPadBB;
})) &&
"Cloning should leave this funclet's blocks monochromatic");
EdgeTargetsFunclet = (IncomingColors.front() == FuncletPadBB);
}
if (IsForOldBlock != EdgeTargetsFunclet)
continue;
PN->removeIncomingValue(IncomingBlock, /*DeletePHIIfEmpty=*/false);
// Revisit the next entry.
--PredIdx;
--PredEnd;
}
};
for (auto &BBMapping : Orig2Clone) {
BasicBlock *OldBlock = BBMapping.first;
BasicBlock *NewBlock = BBMapping.second;
for (PHINode &OldPN : OldBlock->phis()) {
UpdatePHIOnClonedBlock(&OldPN, /*IsForOldBlock=*/true);
}
for (PHINode &NewPN : NewBlock->phis()) {
UpdatePHIOnClonedBlock(&NewPN, /*IsForOldBlock=*/false);
}
}
// Check to see if SuccBB has PHI nodes. If so, we need to add entries to
// the PHI nodes for NewBB now.
for (auto &BBMapping : Orig2Clone) {
BasicBlock *OldBlock = BBMapping.first;
BasicBlock *NewBlock = BBMapping.second;
for (BasicBlock *SuccBB : successors(NewBlock)) {
for (PHINode &SuccPN : SuccBB->phis()) {
// Ok, we have a PHI node. Figure out what the incoming value was for
// the OldBlock.
int OldBlockIdx = SuccPN.getBasicBlockIndex(OldBlock);
if (OldBlockIdx == -1)
break;
Value *IV = SuccPN.getIncomingValue(OldBlockIdx);
// Remap the value if necessary.
if (auto *Inst = dyn_cast<Instruction>(IV)) {
ValueToValueMapTy::iterator I = VMap.find(Inst);
if (I != VMap.end())
IV = I->second;
}
SuccPN.addIncoming(IV, NewBlock);
}
}
}
for (ValueToValueMapTy::value_type VT : VMap) {
// If there were values defined in BB that are used outside the funclet,
// then we now have to update all uses of the value to use either the
// original value, the cloned value, or some PHI derived value. This can
// require arbitrary PHI insertion, of which we are prepared to do, clean
// these up now.
SmallVector<Use *, 16> UsesToRename;
auto *OldI = dyn_cast<Instruction>(const_cast<Value *>(VT.first));
if (!OldI)
continue;
auto *NewI = cast<Instruction>(VT.second);
// Scan all uses of this instruction to see if it is used outside of its
// funclet, and if so, record them in UsesToRename.
for (Use &U : OldI->uses()) {
Instruction *UserI = cast<Instruction>(U.getUser());
BasicBlock *UserBB = UserI->getParent();
ColorVector &ColorsForUserBB = BlockColors[UserBB];
assert(!ColorsForUserBB.empty());
if (ColorsForUserBB.size() > 1 ||
*ColorsForUserBB.begin() != FuncletPadBB)
UsesToRename.push_back(&U);
}
// If there are no uses outside the block, we're done with this
// instruction.
if (UsesToRename.empty())
continue;
// We found a use of OldI outside of the funclet. Rename all uses of OldI
// that are outside its funclet to be uses of the appropriate PHI node
// etc.
SSAUpdater SSAUpdate;
SSAUpdate.Initialize(OldI->getType(), OldI->getName());
SSAUpdate.AddAvailableValue(OldI->getParent(), OldI);
SSAUpdate.AddAvailableValue(NewI->getParent(), NewI);
while (!UsesToRename.empty())
SSAUpdate.RewriteUseAfterInsertions(*UsesToRename.pop_back_val());
}
}
}
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]));
}
}
}
}
}
}