void LTOCodeGenerator::applyScopeRestrictions() {
if (ScopeRestrictionsDone || !shouldInternalize())
return;
Module *mergedModule = Linker.getModule();
// Start off with a verification pass.
PassManager passes;
passes.add(createVerifierPass());
// mark which symbols can not be internalized
Mangler Mangler(TargetMach->getDataLayout());
std::vector<const char*> MustPreserveList;
SmallPtrSet<GlobalValue*, 8> AsmUsed;
std::vector<StringRef> Libcalls;
TargetLibraryInfo TLI(Triple(TargetMach->getTargetTriple()));
accumulateAndSortLibcalls(Libcalls, TLI, TargetMach->getTargetLowering());
for (Module::iterator f = mergedModule->begin(),
e = mergedModule->end(); f != e; ++f)
applyRestriction(*f, Libcalls, MustPreserveList, AsmUsed, Mangler);
for (Module::global_iterator v = mergedModule->global_begin(),
e = mergedModule->global_end(); v != e; ++v)
applyRestriction(*v, Libcalls, MustPreserveList, AsmUsed, Mangler);
for (Module::alias_iterator a = mergedModule->alias_begin(),
e = mergedModule->alias_end(); a != e; ++a)
applyRestriction(*a, Libcalls, MustPreserveList, AsmUsed, Mangler);
GlobalVariable *LLVMCompilerUsed =
mergedModule->getGlobalVariable("llvm.compiler.used");
findUsedValues(LLVMCompilerUsed, AsmUsed);
if (LLVMCompilerUsed)
LLVMCompilerUsed->eraseFromParent();
if (!AsmUsed.empty()) {
llvm::Type *i8PTy = llvm::Type::getInt8PtrTy(Context);
std::vector<Constant*> asmUsed2;
for (SmallPtrSet<GlobalValue*, 16>::const_iterator i = AsmUsed.begin(),
e = AsmUsed.end(); i !=e; ++i) {
GlobalValue *GV = *i;
Constant *c = ConstantExpr::getBitCast(GV, i8PTy);
asmUsed2.push_back(c);
}
llvm::ArrayType *ATy = llvm::ArrayType::get(i8PTy, asmUsed2.size());
LLVMCompilerUsed =
new llvm::GlobalVariable(*mergedModule, ATy, false,
llvm::GlobalValue::AppendingLinkage,
llvm::ConstantArray::get(ATy, asmUsed2),
"llvm.compiler.used");
LLVMCompilerUsed->setSection("llvm.metadata");
}
passes.add(
createInternalizePass(MustPreserveList, shouldOnlyInternalizeHidden()));
// apply scope restrictions
passes.run(*mergedModule);
ScopeRestrictionsDone = true;
}
/// lowerAcrossUnwindEdges - Find all variables which are alive across an unwind
/// edge and spill them.
void SjLjEHPass::lowerAcrossUnwindEdges(Function &F,
ArrayRef<InvokeInst*> Invokes) {
// Finally, scan the code looking for instructions with bad live ranges.
for (Function::iterator
BB = F.begin(), BBE = F.end(); BB != BBE; ++BB) {
for (BasicBlock::iterator
II = BB->begin(), IIE = BB->end(); II != IIE; ++II) {
// Ignore obvious cases we don't have to handle. In particular, most
// instructions either have no uses or only have a single use inside the
// current block. Ignore them quickly.
Instruction *Inst = II;
if (Inst->use_empty()) continue;
if (Inst->hasOneUse() &&
cast<Instruction>(Inst->use_back())->getParent() == BB &&
!isa<PHINode>(Inst->use_back())) continue;
// If this is an alloca in the entry block, it's not a real register
// value.
if (AllocaInst *AI = dyn_cast<AllocaInst>(Inst))
if (isa<ConstantInt>(AI->getArraySize()) && BB == F.begin())
continue;
// Avoid iterator invalidation by copying users to a temporary vector.
SmallVector<Instruction*, 16> Users;
for (Value::use_iterator
UI = Inst->use_begin(), E = Inst->use_end(); UI != E; ++UI) {
Instruction *User = cast<Instruction>(*UI);
if (User->getParent() != BB || isa<PHINode>(User))
Users.push_back(User);
}
// Find all of the blocks that this value is live in.
SmallPtrSet<BasicBlock*, 64> LiveBBs;
LiveBBs.insert(Inst->getParent());
while (!Users.empty()) {
Instruction *U = Users.back();
Users.pop_back();
if (!isa<PHINode>(U)) {
MarkBlocksLiveIn(U->getParent(), LiveBBs);
} else {
// Uses for a PHI node occur in their predecessor block.
PHINode *PN = cast<PHINode>(U);
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
if (PN->getIncomingValue(i) == Inst)
MarkBlocksLiveIn(PN->getIncomingBlock(i), LiveBBs);
}
}
// Now that we know all of the blocks that this thing is live in, see if
// it includes any of the unwind locations.
bool NeedsSpill = false;
for (unsigned i = 0, e = Invokes.size(); i != e; ++i) {
BasicBlock *UnwindBlock = Invokes[i]->getUnwindDest();
if (UnwindBlock != BB && LiveBBs.count(UnwindBlock)) {
DEBUG(dbgs() << "SJLJ Spill: " << *Inst << " around "
<< UnwindBlock->getName() << "\n");
NeedsSpill = true;
break;
}
}
// If we decided we need a spill, do it.
// FIXME: Spilling this way is overkill, as it forces all uses of
// the value to be reloaded from the stack slot, even those that aren't
// in the unwind blocks. We should be more selective.
if (NeedsSpill) {
DemoteRegToStack(*Inst, true);
++NumSpilled;
}
}
}
// Go through the landing pads and remove any PHIs there.
for (unsigned i = 0, e = Invokes.size(); i != e; ++i) {
BasicBlock *UnwindBlock = Invokes[i]->getUnwindDest();
LandingPadInst *LPI = UnwindBlock->getLandingPadInst();
// Place PHIs into a set to avoid invalidating the iterator.
SmallPtrSet<PHINode*, 8> PHIsToDemote;
for (BasicBlock::iterator
PN = UnwindBlock->begin(); isa<PHINode>(PN); ++PN)
PHIsToDemote.insert(cast<PHINode>(PN));
if (PHIsToDemote.empty()) continue;
// Demote the PHIs to the stack.
for (SmallPtrSet<PHINode*, 8>::iterator
I = PHIsToDemote.begin(), E = PHIsToDemote.end(); I != E; ++I)
DemotePHIToStack(*I);
// Move the landingpad instruction back to the top of the landing pad block.
LPI->moveBefore(UnwindBlock->begin());
}
}
/// ProcessLoop - Walk the loop structure in depth first order, ensuring that
/// all loops have preheaders.
///
bool LoopSimplify::ProcessLoop(Loop *L, LPPassManager &LPM) {
bool Changed = false;
ReprocessLoop:
// Check to see that no blocks (other than the header) in this loop have
// predecessors that are not in the loop. This is not valid for natural
// loops, but can occur if the blocks are unreachable. Since they are
// unreachable we can just shamelessly delete those CFG edges!
for (Loop::block_iterator BB = L->block_begin(), E = L->block_end();
BB != E; ++BB) {
if (*BB == L->getHeader()) continue;
SmallPtrSet<BasicBlock*, 4> BadPreds;
for (pred_iterator PI = pred_begin(*BB),
PE = pred_end(*BB); PI != PE; ++PI) {
BasicBlock *P = *PI;
if (!L->contains(P))
BadPreds.insert(P);
}
// Delete each unique out-of-loop (and thus dead) predecessor.
for (SmallPtrSet<BasicBlock*, 4>::iterator I = BadPreds.begin(),
E = BadPreds.end(); I != E; ++I) {
DEBUG(dbgs() << "LoopSimplify: Deleting edge from dead predecessor "
<< (*I)->getName() << "\n");
// Inform each successor of each dead pred.
for (succ_iterator SI = succ_begin(*I), SE = succ_end(*I); SI != SE; ++SI)
(*SI)->removePredecessor(*I);
// Zap the dead pred's terminator and replace it with unreachable.
TerminatorInst *TI = (*I)->getTerminator();
TI->replaceAllUsesWith(UndefValue::get(TI->getType()));
(*I)->getTerminator()->eraseFromParent();
new UnreachableInst((*I)->getContext(), *I);
Changed = true;
}
}
// If there are exiting blocks with branches on undef, resolve the undef in
// the direction which will exit the loop. This will help simplify loop
// trip count computations.
SmallVector<BasicBlock*, 8> ExitingBlocks;
L->getExitingBlocks(ExitingBlocks);
for (SmallVectorImpl<BasicBlock *>::iterator I = ExitingBlocks.begin(),
E = ExitingBlocks.end(); I != E; ++I)
if (BranchInst *BI = dyn_cast<BranchInst>((*I)->getTerminator()))
if (BI->isConditional()) {
if (UndefValue *Cond = dyn_cast<UndefValue>(BI->getCondition())) {
DEBUG(dbgs() << "LoopSimplify: Resolving \"br i1 undef\" to exit in "
<< (*I)->getName() << "\n");
BI->setCondition(ConstantInt::get(Cond->getType(),
!L->contains(BI->getSuccessor(0))));
Changed = true;
}
}
// Does the loop already have a preheader? If so, don't insert one.
BasicBlock *Preheader = L->getLoopPreheader();
if (!Preheader) {
Preheader = InsertPreheaderForLoop(L);
if (Preheader) {
++NumInserted;
Changed = true;
}
}
// Next, check to make sure that all exit nodes of the loop only have
// predecessors that are inside of the loop. This check guarantees that the
// loop preheader/header will dominate the exit blocks. If the exit block has
// predecessors from outside of the loop, split the edge now.
SmallVector<BasicBlock*, 8> ExitBlocks;
L->getExitBlocks(ExitBlocks);
SmallSetVector<BasicBlock *, 8> ExitBlockSet(ExitBlocks.begin(),
ExitBlocks.end());
for (SmallSetVector<BasicBlock *, 8>::iterator I = ExitBlockSet.begin(),
E = ExitBlockSet.end(); I != E; ++I) {
BasicBlock *ExitBlock = *I;
for (pred_iterator PI = pred_begin(ExitBlock), PE = pred_end(ExitBlock);
PI != PE; ++PI)
// Must be exactly this loop: no subloops, parent loops, or non-loop preds
// allowed.
if (!L->contains(*PI)) {
if (RewriteLoopExitBlock(L, ExitBlock)) {
++NumInserted;
Changed = true;
}
break;
}
}
// If the header has more than two predecessors at this point (from the
// preheader and from multiple backedges), we must adjust the loop.
BasicBlock *LoopLatch = L->getLoopLatch();
if (!LoopLatch) {
// If this is really a nested loop, rip it out into a child loop. Don't do
// this for loops with a giant number of backedges, just factor them into a
// common backedge instead.
if (L->getNumBackEdges() < 8) {
if (SeparateNestedLoop(L, LPM)) {
++NumNested;
// This is a big restructuring change, reprocess the whole loop.
Changed = true;
// GCC doesn't tail recursion eliminate this.
goto ReprocessLoop;
}
}
// If we either couldn't, or didn't want to, identify nesting of the loops,
// insert a new block that all backedges target, then make it jump to the
// loop header.
LoopLatch = InsertUniqueBackedgeBlock(L, Preheader);
if (LoopLatch) {
++NumInserted;
Changed = true;
}
}
// Scan over the PHI nodes in the loop header. Since they now have only two
// incoming values (the loop is canonicalized), we may have simplified the PHI
// down to 'X = phi [X, Y]', which should be replaced with 'Y'.
PHINode *PN;
for (BasicBlock::iterator I = L->getHeader()->begin();
(PN = dyn_cast<PHINode>(I++)); )
if (Value *V = SimplifyInstruction(PN, 0, DT)) {
if (AA) AA->deleteValue(PN);
if (SE) SE->forgetValue(PN);
PN->replaceAllUsesWith(V);
PN->eraseFromParent();
}
// If this loop has multiple exits and the exits all go to the same
// block, attempt to merge the exits. This helps several passes, such
// as LoopRotation, which do not support loops with multiple exits.
// SimplifyCFG also does this (and this code uses the same utility
// function), however this code is loop-aware, where SimplifyCFG is
// not. That gives it the advantage of being able to hoist
// loop-invariant instructions out of the way to open up more
// opportunities, and the disadvantage of having the responsibility
// to preserve dominator information.
bool UniqueExit = true;
if (!ExitBlocks.empty())
for (unsigned i = 1, e = ExitBlocks.size(); i != e; ++i)
if (ExitBlocks[i] != ExitBlocks[0]) {
UniqueExit = false;
break;
}
if (UniqueExit) {
for (unsigned i = 0, e = ExitingBlocks.size(); i != e; ++i) {
BasicBlock *ExitingBlock = ExitingBlocks[i];
if (!ExitingBlock->getSinglePredecessor()) continue;
BranchInst *BI = dyn_cast<BranchInst>(ExitingBlock->getTerminator());
if (!BI || !BI->isConditional()) continue;
CmpInst *CI = dyn_cast<CmpInst>(BI->getCondition());
if (!CI || CI->getParent() != ExitingBlock) continue;
// Attempt to hoist out all instructions except for the
// comparison and the branch.
bool AllInvariant = true;
for (BasicBlock::iterator I = ExitingBlock->begin(); &*I != BI; ) {
Instruction *Inst = I++;
// Skip debug info intrinsics.
if (isa<DbgInfoIntrinsic>(Inst))
continue;
if (Inst == CI)
continue;
if (!L->makeLoopInvariant(Inst, Changed,
Preheader ? Preheader->getTerminator() : 0)) {
AllInvariant = false;
break;
}
}
if (!AllInvariant) continue;
// The block has now been cleared of all instructions except for
// a comparison and a conditional branch. SimplifyCFG may be able
// to fold it now.
if (!FoldBranchToCommonDest(BI)) continue;
// Success. The block is now dead, so remove it from the loop,
// update the dominator tree and delete it.
DEBUG(dbgs() << "LoopSimplify: Eliminating exiting block "
<< ExitingBlock->getName() << "\n");
// If any reachable control flow within this loop has changed, notify
// ScalarEvolution. Currently assume the parent loop doesn't change
// (spliting edges doesn't count). If blocks, CFG edges, or other values
// in the parent loop change, then we need call to forgetLoop() for the
// parent instead.
if (SE)
SE->forgetLoop(L);
assert(pred_begin(ExitingBlock) == pred_end(ExitingBlock));
Changed = true;
LI->removeBlock(ExitingBlock);
DomTreeNode *Node = DT->getNode(ExitingBlock);
const std::vector<DomTreeNodeBase<BasicBlock> *> &Children =
Node->getChildren();
while (!Children.empty()) {
DomTreeNode *Child = Children.front();
DT->changeImmediateDominator(Child, Node->getIDom());
}
DT->eraseNode(ExitingBlock);
BI->getSuccessor(0)->removePredecessor(ExitingBlock);
BI->getSuccessor(1)->removePredecessor(ExitingBlock);
ExitingBlock->eraseFromParent();
}
}
return Changed;
}
bool CodeGenPrepare::runOnFunction(Function &F) {
bool EverMadeChange = false;
ModifiedDT = false;
TLInfo = &getAnalysis<TargetLibraryInfo>();
DT = getAnalysisIfAvailable<DominatorTree>();
PFI = getAnalysisIfAvailable<ProfileInfo>();
OptSize = F.hasFnAttr(Attribute::OptimizeForSize);
// First pass, eliminate blocks that contain only PHI nodes and an
// unconditional branch.
EverMadeChange |= EliminateMostlyEmptyBlocks(F);
// llvm.dbg.value is far away from the value then iSel may not be able
// handle it properly. iSel will drop llvm.dbg.value if it can not
// find a node corresponding to the value.
EverMadeChange |= PlaceDbgValues(F);
bool MadeChange = true;
while (MadeChange) {
MadeChange = false;
for (Function::iterator I = F.begin(), E = F.end(); I != E; ) {
BasicBlock *BB = I++;
MadeChange |= OptimizeBlock(*BB);
}
EverMadeChange |= MadeChange;
}
SunkAddrs.clear();
if (!DisableBranchOpts) {
MadeChange = false;
SmallPtrSet<BasicBlock*, 8> WorkList;
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
SmallVector<BasicBlock*, 2> Successors(succ_begin(BB), succ_end(BB));
MadeChange |= ConstantFoldTerminator(BB, true);
if (!MadeChange) continue;
for (SmallVectorImpl<BasicBlock*>::iterator
II = Successors.begin(), IE = Successors.end(); II != IE; ++II)
if (pred_begin(*II) == pred_end(*II))
WorkList.insert(*II);
}
for (SmallPtrSet<BasicBlock*, 8>::iterator
I = WorkList.begin(), E = WorkList.end(); I != E; ++I)
DeleteDeadBlock(*I);
// Merge pairs of basic blocks with unconditional branches, connected by
// a single edge.
if (EverMadeChange || MadeChange)
MadeChange |= EliminateFallThrough(F);
if (MadeChange)
ModifiedDT = true;
EverMadeChange |= MadeChange;
}
if (ModifiedDT && DT)
DT->DT->recalculate(F);
return EverMadeChange;
}
// Sinks \p I from the loop \p L's preheader to its uses. Returns true if
// sinking is successful.
// \p LoopBlockNumber is used to sort the insertion blocks to ensure
// determinism.
static bool sinkInstruction(Loop &L, Instruction &I,
const SmallVectorImpl<BasicBlock *> &ColdLoopBBs,
const SmallDenseMap<BasicBlock *, int, 16> &LoopBlockNumber,
LoopInfo &LI, DominatorTree &DT,
BlockFrequencyInfo &BFI) {
// Compute the set of blocks in loop L which contain a use of I.
SmallPtrSet<BasicBlock *, 2> BBs;
for (auto &U : I.uses()) {
Instruction *UI = cast<Instruction>(U.getUser());
// We cannot sink I to PHI-uses.
if (dyn_cast<PHINode>(UI))
return false;
// We cannot sink I if it has uses outside of the loop.
if (!L.contains(LI.getLoopFor(UI->getParent())))
return false;
BBs.insert(UI->getParent());
}
// findBBsToSinkInto is O(BBs.size() * ColdLoopBBs.size()). We cap the max
// BBs.size() to avoid expensive computation.
// FIXME: Handle code size growth for min_size and opt_size.
if (BBs.size() > MaxNumberOfUseBBsForSinking)
return false;
// Find the set of BBs that we should insert a copy of I.
SmallPtrSet<BasicBlock *, 2> BBsToSinkInto =
findBBsToSinkInto(L, BBs, ColdLoopBBs, DT, BFI);
if (BBsToSinkInto.empty())
return false;
// Copy the final BBs into a vector and sort them using the total ordering
// of the loop block numbers as iterating the set doesn't give a useful
// order. No need to stable sort as the block numbers are a total ordering.
SmallVector<BasicBlock *, 2> SortedBBsToSinkInto;
SortedBBsToSinkInto.insert(SortedBBsToSinkInto.begin(), BBsToSinkInto.begin(),
BBsToSinkInto.end());
std::sort(SortedBBsToSinkInto.begin(), SortedBBsToSinkInto.end(),
[&](BasicBlock *A, BasicBlock *B) {
return *LoopBlockNumber.find(A) < *LoopBlockNumber.find(B);
});
BasicBlock *MoveBB = *SortedBBsToSinkInto.begin();
// FIXME: Optimize the efficiency for cloned value replacement. The current
// implementation is O(SortedBBsToSinkInto.size() * I.num_uses()).
for (BasicBlock *N : SortedBBsToSinkInto) {
if (N == MoveBB)
continue;
// Clone I and replace its uses.
Instruction *IC = I.clone();
IC->setName(I.getName());
IC->insertBefore(&*N->getFirstInsertionPt());
// Replaces uses of I with IC in N
for (Value::use_iterator UI = I.use_begin(), UE = I.use_end(); UI != UE;) {
Use &U = *UI++;
auto *I = cast<Instruction>(U.getUser());
if (I->getParent() == N)
U.set(IC);
}
// Replaces uses of I with IC in blocks dominated by N
replaceDominatedUsesWith(&I, IC, DT, N);
DEBUG(dbgs() << "Sinking a clone of " << I << " To: " << N->getName()
<< '\n');
NumLoopSunkCloned++;
}
DEBUG(dbgs() << "Sinking " << I << " To: " << MoveBB->getName() << '\n');
NumLoopSunk++;
I.moveBefore(&*MoveBB->getFirstInsertionPt());
return true;
}
/// runOnLoop - Remove dead loops, by which we mean loops that do not impact the
/// observable behavior of the program other than finite running time. Note
/// we do ensure that this never remove a loop that might be infinite, as doing
/// so could change the halting/non-halting nature of a program.
/// NOTE: This entire process relies pretty heavily on LoopSimplify and LCSSA
/// in order to make various safety checks work.
bool LoopDeletion::runOnLoop(Loop* L, LPPassManager& LPM) {
// We can only remove the loop if there is a preheader that we can
// branch from after removing it.
BasicBlock* preheader = L->getLoopPreheader();
if (!preheader)
return false;
// If LoopSimplify form is not available, stay out of trouble.
if (!L->hasDedicatedExits())
return false;
// We can't remove loops that contain subloops. If the subloops were dead,
// they would already have been removed in earlier executions of this pass.
if (L->begin() != L->end())
return false;
SmallVector<BasicBlock*, 4> exitingBlocks;
L->getExitingBlocks(exitingBlocks);
SmallVector<BasicBlock*, 4> exitBlocks;
L->getUniqueExitBlocks(exitBlocks);
// We require that the loop only have a single exit block. Otherwise, we'd
// be in the situation of needing to be able to solve statically which exit
// block will be branched to, or trying to preserve the branching logic in
// a loop invariant manner.
if (exitBlocks.size() != 1)
return false;
// Finally, we have to check that the loop really is dead.
bool Changed = false;
if (!IsLoopDead(L, exitingBlocks, exitBlocks, Changed, preheader))
return Changed;
// Don't remove loops for which we can't solve the trip count.
// They could be infinite, in which case we'd be changing program behavior.
ScalarEvolution& SE = getAnalysis<ScalarEvolution>();
const SCEV *S = SE.getMaxBackedgeTakenCount(L);
if (isa<SCEVCouldNotCompute>(S))
return Changed;
// Now that we know the removal is safe, remove the loop by changing the
// branch from the preheader to go to the single exit block.
BasicBlock* exitBlock = exitBlocks[0];
// Because we're deleting a large chunk of code at once, the sequence in which
// we remove things is very important to avoid invalidation issues. Don't
// mess with this unless you have good reason and know what you're doing.
// Tell ScalarEvolution that the loop is deleted. Do this before
// deleting the loop so that ScalarEvolution can look at the loop
// to determine what it needs to clean up.
SE.forgetLoop(L);
// Connect the preheader directly to the exit block.
TerminatorInst* TI = preheader->getTerminator();
TI->replaceUsesOfWith(L->getHeader(), exitBlock);
// Rewrite phis in the exit block to get their inputs from
// the preheader instead of the exiting block.
BasicBlock* exitingBlock = exitingBlocks[0];
BasicBlock::iterator BI = exitBlock->begin();
while (PHINode* P = dyn_cast<PHINode>(BI)) {
int j = P->getBasicBlockIndex(exitingBlock);
assert(j >= 0 && "Can't find exiting block in exit block's phi node!");
P->setIncomingBlock(j, preheader);
for (unsigned i = 1; i < exitingBlocks.size(); ++i)
P->removeIncomingValue(exitingBlocks[i]);
++BI;
}
// Update the dominator tree and remove the instructions and blocks that will
// be deleted from the reference counting scheme.
DominatorTree& DT = getAnalysis<DominatorTree>();
SmallVector<DomTreeNode*, 8> ChildNodes;
for (Loop::block_iterator LI = L->block_begin(), LE = L->block_end();
LI != LE; ++LI) {
// Move all of the block's children to be children of the preheader, which
// allows us to remove the domtree entry for the block.
ChildNodes.insert(ChildNodes.begin(), DT[*LI]->begin(), DT[*LI]->end());
for (SmallVector<DomTreeNode*, 8>::iterator DI = ChildNodes.begin(),
DE = ChildNodes.end(); DI != DE; ++DI) {
DT.changeImmediateDominator(*DI, DT[preheader]);
}
ChildNodes.clear();
DT.eraseNode(*LI);
// Remove the block from the reference counting scheme, so that we can
// delete it freely later.
(*LI)->dropAllReferences();
}
// Erase the instructions and the blocks without having to worry
// about ordering because we already dropped the references.
// NOTE: This iteration is safe because erasing the block does not remove its
// entry from the loop's block list. We do that in the next section.
for (Loop::block_iterator LI = L->block_begin(), LE = L->block_end();
LI != LE; ++LI)
(*LI)->eraseFromParent();
// Finally, the blocks from loopinfo. This has to happen late because
// otherwise our loop iterators won't work.
LoopInfo& loopInfo = getAnalysis<LoopInfo>();
SmallPtrSet<BasicBlock*, 8> blocks;
blocks.insert(L->block_begin(), L->block_end());
for (SmallPtrSet<BasicBlock*,8>::iterator I = blocks.begin(),
E = blocks.end(); I != E; ++I)
loopInfo.removeBlock(*I);
// The last step is to inform the loop pass manager that we've
// eliminated this loop.
LPM.deleteLoopFromQueue(L);
Changed = true;
++NumDeleted;
return Changed;
}
// Calculate Edge Weights using "Loop Branch Heuristics". Predict backedges
// as taken, exiting edges as not-taken.
bool BranchProbabilityAnalysis::calcLoopBranchHeuristics(BasicBlock *BB) {
uint32_t numSuccs = BB->getTerminator()->getNumSuccessors();
Loop *L = LI->getLoopFor(BB);
if (!L)
return false;
SmallPtrSet<BasicBlock *, 8> BackEdges;
SmallPtrSet<BasicBlock *, 8> ExitingEdges;
SmallPtrSet<BasicBlock *, 8> InEdges; // Edges from header to the loop.
bool isHeader = BB == L->getHeader();
for (succ_iterator I = succ_begin(BB), E = succ_end(BB); I != E; ++I) {
BasicBlock *Succ = *I;
Loop *SuccL = LI->getLoopFor(Succ);
if (SuccL != L)
ExitingEdges.insert(Succ);
else if (Succ == L->getHeader())
BackEdges.insert(Succ);
else if (isHeader)
InEdges.insert(Succ);
}
if (uint32_t numBackEdges = BackEdges.size()) {
uint32_t backWeight = LBH_TAKEN_WEIGHT / numBackEdges;
if (backWeight < NORMAL_WEIGHT)
backWeight = NORMAL_WEIGHT;
for (SmallPtrSet<BasicBlock *, 8>::iterator EI = BackEdges.begin(),
EE = BackEdges.end(); EI != EE; ++EI) {
BasicBlock *Back = *EI;
BP->setEdgeWeight(BB, Back, backWeight);
}
}
if (uint32_t numInEdges = InEdges.size()) {
uint32_t inWeight = LBH_TAKEN_WEIGHT / numInEdges;
if (inWeight < NORMAL_WEIGHT)
inWeight = NORMAL_WEIGHT;
for (SmallPtrSet<BasicBlock *, 8>::iterator EI = InEdges.begin(),
EE = InEdges.end(); EI != EE; ++EI) {
BasicBlock *Back = *EI;
BP->setEdgeWeight(BB, Back, inWeight);
}
}
uint32_t numExitingEdges = ExitingEdges.size();
if (uint32_t numNonExitingEdges = numSuccs - numExitingEdges) {
uint32_t exitWeight = LBH_NONTAKEN_WEIGHT / numNonExitingEdges;
if (exitWeight < MIN_WEIGHT)
exitWeight = MIN_WEIGHT;
for (SmallPtrSet<BasicBlock *, 8>::iterator EI = ExitingEdges.begin(),
EE = ExitingEdges.end(); EI != EE; ++EI) {
BasicBlock *Exiting = *EI;
BP->setEdgeWeight(BB, Exiting, exitWeight);
}
}
return true;
}
/// handleEndBlock - Remove dead stores to stack-allocated locations in the
/// function end block. Ex:
/// %A = alloca i32
/// ...
/// store i32 1, i32* %A
/// ret void
bool DSE::handleEndBlock(BasicBlock &BB) {
bool MadeChange = false;
// Keep track of all of the stack objects that are dead at the end of the
// function.
SmallPtrSet<Value*, 16> DeadStackObjects;
// Find all of the alloca'd pointers in the entry block.
BasicBlock *Entry = BB.getParent()->begin();
for (BasicBlock::iterator I = Entry->begin(), E = Entry->end(); I != E; ++I)
if (AllocaInst *AI = dyn_cast<AllocaInst>(I))
DeadStackObjects.insert(AI);
// Treat byval arguments the same, stores to them are dead at the end of the
// function.
for (Function::arg_iterator AI = BB.getParent()->arg_begin(),
AE = BB.getParent()->arg_end(); AI != AE; ++AI)
if (AI->hasByValAttr())
DeadStackObjects.insert(AI);
// Scan the basic block backwards
for (BasicBlock::iterator BBI = BB.end(); BBI != BB.begin(); ){
--BBI;
// If we find a store, check to see if it points into a dead stack value.
if (hasMemoryWrite(BBI) && isRemovable(BBI)) {
// See through pointer-to-pointer bitcasts
Value *Pointer = GetUnderlyingObject(getStoredPointerOperand(BBI));
// Stores to stack values are valid candidates for removal.
if (DeadStackObjects.count(Pointer)) {
Instruction *Dead = BBI++;
DEBUG(dbgs() << "DSE: Dead Store at End of Block:\n DEAD: "
<< *Dead << "\n Object: " << *Pointer << '\n');
// DCE instructions only used to calculate that store.
DeleteDeadInstruction(Dead, *MD, &DeadStackObjects);
++NumFastStores;
MadeChange = true;
continue;
}
}
// Remove any dead non-memory-mutating instructions.
if (isInstructionTriviallyDead(BBI)) {
Instruction *Inst = BBI++;
DeleteDeadInstruction(Inst, *MD, &DeadStackObjects);
++NumFastOther;
MadeChange = true;
continue;
}
if (AllocaInst *A = dyn_cast<AllocaInst>(BBI)) {
DeadStackObjects.erase(A);
continue;
}
if (CallSite CS = cast<Value>(BBI)) {
// If this call does not access memory, it can't be loading any of our
// pointers.
if (AA->doesNotAccessMemory(CS))
continue;
unsigned NumModRef = 0, NumOther = 0;
// If the call might load from any of our allocas, then any store above
// the call is live.
SmallVector<Value*, 8> LiveAllocas;
for (SmallPtrSet<Value*, 16>::iterator I = DeadStackObjects.begin(),
E = DeadStackObjects.end(); I != E; ++I) {
// If we detect that our AA is imprecise, it's not worth it to scan the
// rest of the DeadPointers set. Just assume that the AA will return
// ModRef for everything, and go ahead and bail out.
if (NumModRef >= 16 && NumOther == 0)
return MadeChange;
// See if the call site touches it.
AliasAnalysis::ModRefResult A =
AA->getModRefInfo(CS, *I, getPointerSize(*I, *AA));
if (A == AliasAnalysis::ModRef)
++NumModRef;
else
++NumOther;
if (A == AliasAnalysis::ModRef || A == AliasAnalysis::Ref)
LiveAllocas.push_back(*I);
}
for (SmallVector<Value*, 8>::iterator I = LiveAllocas.begin(),
E = LiveAllocas.end(); I != E; ++I)
DeadStackObjects.erase(*I);
// If all of the allocas were clobbered by the call then we're not going
// to find anything else to process.
if (DeadStackObjects.empty())
return MadeChange;
continue;
}
AliasAnalysis::Location LoadedLoc;
// If we encounter a use of the pointer, it is no longer considered dead
if (LoadInst *L = dyn_cast<LoadInst>(BBI)) {
LoadedLoc = AA->getLocation(L);
} else if (VAArgInst *V = dyn_cast<VAArgInst>(BBI)) {
LoadedLoc = AA->getLocation(V);
} else if (MemTransferInst *MTI = dyn_cast<MemTransferInst>(BBI)) {
LoadedLoc = AA->getLocationForSource(MTI);
} else {
// Not a loading instruction.
continue;
}
// Remove any allocas from the DeadPointer set that are loaded, as this
// makes any stores above the access live.
RemoveAccessedObjects(LoadedLoc, DeadStackObjects);
// If all of the allocas were clobbered by the access then we're not going
// to find anything else to process.
if (DeadStackObjects.empty())
break;
}
return MadeChange;
}
/*
This method performs Unroll and Jam. For a simple loop like:
for (i = ..)
Fore(i)
for (j = ..)
SubLoop(i, j)
Aft(i)
Instead of doing normal inner or outer unrolling, we do:
for (i = .., i+=2)
Fore(i)
Fore(i+1)
for (j = ..)
SubLoop(i, j)
SubLoop(i+1, j)
Aft(i)
Aft(i+1)
So the outer loop is essetially unrolled and then the inner loops are fused
("jammed") together into a single loop. This can increase speed when there
are loads in SubLoop that are invariant to i, as they become shared between
the now jammed inner loops.
We do this by spliting the blocks in the loop into Fore, Subloop and Aft.
Fore blocks are those before the inner loop, Aft are those after. Normal
Unroll code is used to copy each of these sets of blocks and the results are
combined together into the final form above.
isSafeToUnrollAndJam should be used prior to calling this to make sure the
unrolling will be valid. Checking profitablility is also advisable.
*/
LoopUnrollResult
llvm::UnrollAndJamLoop(Loop *L, unsigned Count, unsigned TripCount,
unsigned TripMultiple, bool UnrollRemainder,
LoopInfo *LI, ScalarEvolution *SE, DominatorTree *DT,
AssumptionCache *AC, OptimizationRemarkEmitter *ORE) {
// When we enter here we should have already checked that it is safe
BasicBlock *Header = L->getHeader();
assert(L->getSubLoops().size() == 1);
Loop *SubLoop = *L->begin();
// Don't enter the unroll code if there is nothing to do.
if (TripCount == 0 && Count < 2) {
LLVM_DEBUG(dbgs() << "Won't unroll; almost nothing to do\n");
return LoopUnrollResult::Unmodified;
}
assert(Count > 0);
assert(TripMultiple > 0);
assert(TripCount == 0 || TripCount % TripMultiple == 0);
// Are we eliminating the loop control altogether?
bool CompletelyUnroll = (Count == TripCount);
// We use the runtime remainder in cases where we don't know trip multiple
if (TripMultiple == 1 || TripMultiple % Count != 0) {
if (!UnrollRuntimeLoopRemainder(L, Count, /*AllowExpensiveTripCount*/ false,
/*UseEpilogRemainder*/ true,
UnrollRemainder, LI, SE, DT, AC, true)) {
LLVM_DEBUG(dbgs() << "Won't unroll-and-jam; remainder loop could not be "
"generated when assuming runtime trip count\n");
return LoopUnrollResult::Unmodified;
}
}
// Notify ScalarEvolution that the loop will be substantially changed,
// if not outright eliminated.
if (SE) {
SE->forgetLoop(L);
SE->forgetLoop(SubLoop);
}
using namespace ore;
// Report the unrolling decision.
if (CompletelyUnroll) {
LLVM_DEBUG(dbgs() << "COMPLETELY UNROLL AND JAMMING loop %"
<< Header->getName() << " with trip count " << TripCount
<< "!\n");
ORE->emit(OptimizationRemark(DEBUG_TYPE, "FullyUnrolled", L->getStartLoc(),
L->getHeader())
<< "completely unroll and jammed loop with "
<< NV("UnrollCount", TripCount) << " iterations");
} else {
auto DiagBuilder = [&]() {
OptimizationRemark Diag(DEBUG_TYPE, "PartialUnrolled", L->getStartLoc(),
L->getHeader());
return Diag << "unroll and jammed loop by a factor of "
<< NV("UnrollCount", Count);
};
LLVM_DEBUG(dbgs() << "UNROLL AND JAMMING loop %" << Header->getName()
<< " by " << Count);
if (TripMultiple != 1) {
LLVM_DEBUG(dbgs() << " with " << TripMultiple << " trips per branch");
ORE->emit([&]() {
return DiagBuilder() << " with " << NV("TripMultiple", TripMultiple)
<< " trips per branch";
});
} else {
LLVM_DEBUG(dbgs() << " with run-time trip count");
ORE->emit([&]() { return DiagBuilder() << " with run-time trip count"; });
}
LLVM_DEBUG(dbgs() << "!\n");
}
BasicBlock *Preheader = L->getLoopPreheader();
BasicBlock *LatchBlock = L->getLoopLatch();
BranchInst *BI = dyn_cast<BranchInst>(LatchBlock->getTerminator());
assert(Preheader && LatchBlock && Header);
assert(BI && !BI->isUnconditional());
bool ContinueOnTrue = L->contains(BI->getSuccessor(0));
BasicBlock *LoopExit = BI->getSuccessor(ContinueOnTrue);
bool SubLoopContinueOnTrue = SubLoop->contains(
SubLoop->getLoopLatch()->getTerminator()->getSuccessor(0));
// Partition blocks in an outer/inner loop pair into blocks before and after
// the loop
BasicBlockSet SubLoopBlocks;
BasicBlockSet ForeBlocks;
BasicBlockSet AftBlocks;
partitionOuterLoopBlocks(L, SubLoop, ForeBlocks, SubLoopBlocks, AftBlocks,
DT);
// We keep track of the entering/first and exiting/last block of each of
// Fore/SubLoop/Aft in each iteration. This helps make the stapling up of
// blocks easier.
std::vector<BasicBlock *> ForeBlocksFirst;
std::vector<BasicBlock *> ForeBlocksLast;
std::vector<BasicBlock *> SubLoopBlocksFirst;
std::vector<BasicBlock *> SubLoopBlocksLast;
std::vector<BasicBlock *> AftBlocksFirst;
std::vector<BasicBlock *> AftBlocksLast;
ForeBlocksFirst.push_back(Header);
ForeBlocksLast.push_back(SubLoop->getLoopPreheader());
SubLoopBlocksFirst.push_back(SubLoop->getHeader());
SubLoopBlocksLast.push_back(SubLoop->getExitingBlock());
AftBlocksFirst.push_back(SubLoop->getExitBlock());
AftBlocksLast.push_back(L->getExitingBlock());
// Maps Blocks[0] -> Blocks[It]
ValueToValueMapTy LastValueMap;
// Move any instructions from fore phi operands from AftBlocks into Fore.
moveHeaderPhiOperandsToForeBlocks(
Header, LatchBlock, SubLoop->getLoopPreheader()->getTerminator(),
AftBlocks);
// 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();
if (Header->getParent()->isDebugInfoForProfiling())
for (BasicBlock *BB : L->getBlocks())
for (Instruction &I : *BB)
if (!isa<DbgInfoIntrinsic>(&I))
if (const DILocation *DIL = I.getDebugLoc())
I.setDebugLoc(DIL->cloneWithDuplicationFactor(Count));
// Copy all blocks
for (unsigned It = 1; It != Count; ++It) {
std::vector<BasicBlock *> NewBlocks;
// Maps Blocks[It] -> Blocks[It-1]
DenseMap<Value *, Value *> PrevItValueMap;
for (LoopBlocksDFS::RPOIterator BB = BlockBegin; BB != BlockEnd; ++BB) {
ValueToValueMapTy VMap;
BasicBlock *New = CloneBasicBlock(*BB, VMap, "." + Twine(It));
Header->getParent()->getBasicBlockList().push_back(New);
if (ForeBlocks.count(*BB)) {
L->addBasicBlockToLoop(New, *LI);
if (*BB == ForeBlocksFirst[0])
ForeBlocksFirst.push_back(New);
if (*BB == ForeBlocksLast[0])
ForeBlocksLast.push_back(New);
} else if (SubLoopBlocks.count(*BB)) {
SubLoop->addBasicBlockToLoop(New, *LI);
if (*BB == SubLoopBlocksFirst[0])
SubLoopBlocksFirst.push_back(New);
if (*BB == SubLoopBlocksLast[0])
SubLoopBlocksLast.push_back(New);
} else if (AftBlocks.count(*BB)) {
L->addBasicBlockToLoop(New, *LI);
if (*BB == AftBlocksFirst[0])
AftBlocksFirst.push_back(New);
if (*BB == AftBlocksLast[0])
AftBlocksLast.push_back(New);
} else {
llvm_unreachable("BB being cloned should be in Fore/Sub/Aft");
}
// Update our running maps of newest clones
PrevItValueMap[New] = (It == 1 ? *BB : LastValueMap[*BB]);
LastValueMap[*BB] = New;
for (ValueToValueMapTy::iterator VI = VMap.begin(), VE = VMap.end();
VI != VE; ++VI) {
PrevItValueMap[VI->second] =
const_cast<Value *>(It == 1 ? VI->first : LastValueMap[VI->first]);
LastValueMap[VI->first] = VI->second;
}
NewBlocks.push_back(New);
// Update DomTree:
if (*BB == ForeBlocksFirst[0])
DT->addNewBlock(New, ForeBlocksLast[It - 1]);
else if (*BB == SubLoopBlocksFirst[0])
DT->addNewBlock(New, SubLoopBlocksLast[It - 1]);
else if (*BB == AftBlocksFirst[0])
DT->addNewBlock(New, AftBlocksLast[It - 1]);
else {
// Each set of blocks (Fore/Sub/Aft) will have the same internal domtree
// structure.
auto BBDomNode = DT->getNode(*BB);
auto BBIDom = BBDomNode->getIDom();
BasicBlock *OriginalBBIDom = BBIDom->getBlock();
assert(OriginalBBIDom);
assert(LastValueMap[cast<Value>(OriginalBBIDom)]);
DT->addNewBlock(
New, cast<BasicBlock>(LastValueMap[cast<Value>(OriginalBBIDom)]));
}
}
// Remap all instructions in the most recent iteration
for (BasicBlock *NewBlock : NewBlocks) {
for (Instruction &I : *NewBlock) {
::remapInstruction(&I, LastValueMap);
if (auto *II = dyn_cast<IntrinsicInst>(&I))
if (II->getIntrinsicID() == Intrinsic::assume)
AC->registerAssumption(II);
}
}
// Alter the ForeBlocks phi's, pointing them at the latest version of the
// value from the previous iteration's phis
for (PHINode &Phi : ForeBlocksFirst[It]->phis()) {
Value *OldValue = Phi.getIncomingValueForBlock(AftBlocksLast[It]);
assert(OldValue && "should have incoming edge from Aft[It]");
Value *NewValue = OldValue;
if (Value *PrevValue = PrevItValueMap[OldValue])
NewValue = PrevValue;
assert(Phi.getNumOperands() == 2);
Phi.setIncomingBlock(0, ForeBlocksLast[It - 1]);
Phi.setIncomingValue(0, NewValue);
Phi.removeIncomingValue(1);
}
}
// Now that all the basic blocks for the unrolled iterations are in place,
// finish up connecting the blocks and phi nodes. At this point LastValueMap
// is the last unrolled iterations values.
// Update Phis in BB from OldBB to point to NewBB
auto updatePHIBlocks = [](BasicBlock *BB, BasicBlock *OldBB,
BasicBlock *NewBB) {
for (PHINode &Phi : BB->phis()) {
int I = Phi.getBasicBlockIndex(OldBB);
Phi.setIncomingBlock(I, NewBB);
}
};
// Update Phis in BB from OldBB to point to NewBB and use the latest value
// from LastValueMap
auto updatePHIBlocksAndValues = [](BasicBlock *BB, BasicBlock *OldBB,
BasicBlock *NewBB,
ValueToValueMapTy &LastValueMap) {
for (PHINode &Phi : BB->phis()) {
for (unsigned b = 0; b < Phi.getNumIncomingValues(); ++b) {
if (Phi.getIncomingBlock(b) == OldBB) {
Value *OldValue = Phi.getIncomingValue(b);
if (Value *LastValue = LastValueMap[OldValue])
Phi.setIncomingValue(b, LastValue);
Phi.setIncomingBlock(b, NewBB);
break;
}
}
}
};
// Move all the phis from Src into Dest
auto movePHIs = [](BasicBlock *Src, BasicBlock *Dest) {
Instruction *insertPoint = Dest->getFirstNonPHI();
while (PHINode *Phi = dyn_cast<PHINode>(Src->begin()))
Phi->moveBefore(insertPoint);
};
// Update the PHI values outside the loop to point to the last block
updatePHIBlocksAndValues(LoopExit, AftBlocksLast[0], AftBlocksLast.back(),
LastValueMap);
// Update ForeBlocks successors and phi nodes
BranchInst *ForeTerm =
cast<BranchInst>(ForeBlocksLast.back()->getTerminator());
BasicBlock *Dest = SubLoopBlocksFirst[0];
ForeTerm->setSuccessor(0, Dest);
if (CompletelyUnroll) {
while (PHINode *Phi = dyn_cast<PHINode>(ForeBlocksFirst[0]->begin())) {
Phi->replaceAllUsesWith(Phi->getIncomingValueForBlock(Preheader));
Phi->getParent()->getInstList().erase(Phi);
}
} else {
// Update the PHI values to point to the last aft block
updatePHIBlocksAndValues(ForeBlocksFirst[0], AftBlocksLast[0],
AftBlocksLast.back(), LastValueMap);
}
for (unsigned It = 1; It != Count; It++) {
// Remap ForeBlock successors from previous iteration to this
BranchInst *ForeTerm =
cast<BranchInst>(ForeBlocksLast[It - 1]->getTerminator());
BasicBlock *Dest = ForeBlocksFirst[It];
ForeTerm->setSuccessor(0, Dest);
}
// Subloop successors and phis
BranchInst *SubTerm =
cast<BranchInst>(SubLoopBlocksLast.back()->getTerminator());
SubTerm->setSuccessor(!SubLoopContinueOnTrue, SubLoopBlocksFirst[0]);
SubTerm->setSuccessor(SubLoopContinueOnTrue, AftBlocksFirst[0]);
updatePHIBlocks(SubLoopBlocksFirst[0], ForeBlocksLast[0],
ForeBlocksLast.back());
updatePHIBlocks(SubLoopBlocksFirst[0], SubLoopBlocksLast[0],
SubLoopBlocksLast.back());
for (unsigned It = 1; It != Count; It++) {
// Replace the conditional branch of the previous iteration subloop with an
// unconditional one to this one
BranchInst *SubTerm =
cast<BranchInst>(SubLoopBlocksLast[It - 1]->getTerminator());
BranchInst::Create(SubLoopBlocksFirst[It], SubTerm);
SubTerm->eraseFromParent();
updatePHIBlocks(SubLoopBlocksFirst[It], ForeBlocksLast[It],
ForeBlocksLast.back());
updatePHIBlocks(SubLoopBlocksFirst[It], SubLoopBlocksLast[It],
SubLoopBlocksLast.back());
movePHIs(SubLoopBlocksFirst[It], SubLoopBlocksFirst[0]);
}
// Aft blocks successors and phis
BranchInst *Term = cast<BranchInst>(AftBlocksLast.back()->getTerminator());
if (CompletelyUnroll) {
BranchInst::Create(LoopExit, Term);
Term->eraseFromParent();
} else {
Term->setSuccessor(!ContinueOnTrue, ForeBlocksFirst[0]);
}
updatePHIBlocks(AftBlocksFirst[0], SubLoopBlocksLast[0],
SubLoopBlocksLast.back());
for (unsigned It = 1; It != Count; It++) {
// Replace the conditional branch of the previous iteration subloop with an
// unconditional one to this one
BranchInst *AftTerm =
cast<BranchInst>(AftBlocksLast[It - 1]->getTerminator());
BranchInst::Create(AftBlocksFirst[It], AftTerm);
AftTerm->eraseFromParent();
updatePHIBlocks(AftBlocksFirst[It], SubLoopBlocksLast[It],
SubLoopBlocksLast.back());
movePHIs(AftBlocksFirst[It], AftBlocksFirst[0]);
}
// Dominator Tree. Remove the old links between Fore, Sub and Aft, adding the
// new ones required.
if (Count != 1) {
SmallVector<DominatorTree::UpdateType, 4> DTUpdates;
DTUpdates.emplace_back(DominatorTree::UpdateKind::Delete, ForeBlocksLast[0],
SubLoopBlocksFirst[0]);
DTUpdates.emplace_back(DominatorTree::UpdateKind::Delete,
SubLoopBlocksLast[0], AftBlocksFirst[0]);
DTUpdates.emplace_back(DominatorTree::UpdateKind::Insert,
ForeBlocksLast.back(), SubLoopBlocksFirst[0]);
DTUpdates.emplace_back(DominatorTree::UpdateKind::Insert,
SubLoopBlocksLast.back(), AftBlocksFirst[0]);
DT->applyUpdates(DTUpdates);
}
// Merge adjacent basic blocks, if possible.
SmallPtrSet<BasicBlock *, 16> MergeBlocks;
MergeBlocks.insert(ForeBlocksLast.begin(), ForeBlocksLast.end());
MergeBlocks.insert(SubLoopBlocksLast.begin(), SubLoopBlocksLast.end());
MergeBlocks.insert(AftBlocksLast.begin(), AftBlocksLast.end());
while (!MergeBlocks.empty()) {
BasicBlock *BB = *MergeBlocks.begin();
BranchInst *Term = dyn_cast<BranchInst>(BB->getTerminator());
if (Term && Term->isUnconditional() && L->contains(Term->getSuccessor(0))) {
BasicBlock *Dest = Term->getSuccessor(0);
if (BasicBlock *Fold = foldBlockIntoPredecessor(Dest, LI, SE, DT)) {
// Don't remove BB and add Fold as they are the same BB
assert(Fold == BB);
(void)Fold;
MergeBlocks.erase(Dest);
} else
MergeBlocks.erase(BB);
} else
MergeBlocks.erase(BB);
}
// 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.
simplifyLoopAfterUnroll(SubLoop, true, LI, SE, DT, AC);
simplifyLoopAfterUnroll(L, !CompletelyUnroll && Count > 1, LI, SE, DT, AC);
NumCompletelyUnrolledAndJammed += CompletelyUnroll;
++NumUnrolledAndJammed;
#ifndef NDEBUG
// We shouldn't have done anything to break loop simplify form or LCSSA.
Loop *OuterL = L->getParentLoop();
Loop *OutestLoop = OuterL ? OuterL : (!CompletelyUnroll ? L : SubLoop);
assert(OutestLoop->isRecursivelyLCSSAForm(*DT, *LI));
if (!CompletelyUnroll)
assert(L->isLoopSimplifyForm());
assert(SubLoop->isLoopSimplifyForm());
assert(DT->verify());
#endif
// Update LoopInfo if the loop is completely removed.
if (CompletelyUnroll)
LI->erase(L);
return CompletelyUnroll ? LoopUnrollResult::FullyUnrolled
: LoopUnrollResult::PartiallyUnrolled;
}
/// HandleURoRInvokes - Handle invokes of "_Unwind_Resume_or_Rethrow" calls. The
/// "unwind" part of these invokes jump to a landing pad within the current
/// function. This is a candidate to merge the selector associated with the URoR
/// invoke with the one from the URoR's landing pad.
bool DwarfEHPrepare::HandleURoRInvokes() {
if (!EHCatchAllValue) {
EHCatchAllValue =
F->getParent()->getNamedGlobal("llvm.eh.catch.all.value");
if (!EHCatchAllValue) return false;
}
if (!SelectorIntrinsic) {
SelectorIntrinsic =
Intrinsic::getDeclaration(F->getParent(), Intrinsic::eh_selector);
if (!SelectorIntrinsic) return false;
}
SmallPtrSet<IntrinsicInst*, 32> Sels;
SmallPtrSet<IntrinsicInst*, 32> CatchAllSels;
FindAllCleanupSelectors(Sels, CatchAllSels);
if (!DT)
// We require DominatorTree information.
return CleanupSelectors(CatchAllSels);
if (!URoR) {
URoR = F->getParent()->getFunction("_Unwind_Resume_or_Rethrow");
if (!URoR) return CleanupSelectors(CatchAllSels);
}
SmallPtrSet<InvokeInst*, 32> URoRInvokes;
FindAllURoRInvokes(URoRInvokes);
SmallPtrSet<IntrinsicInst*, 32> SelsToConvert;
for (SmallPtrSet<IntrinsicInst*, 32>::iterator
SI = Sels.begin(), SE = Sels.end(); SI != SE; ++SI) {
const BasicBlock *SelBB = (*SI)->getParent();
for (SmallPtrSet<InvokeInst*, 32>::iterator
UI = URoRInvokes.begin(), UE = URoRInvokes.end(); UI != UE; ++UI) {
const BasicBlock *URoRBB = (*UI)->getParent();
if (DT->dominates(SelBB, URoRBB)) {
SelsToConvert.insert(*SI);
break;
}
}
}
bool Changed = false;
if (Sels.size() != SelsToConvert.size()) {
// If we haven't been able to convert all of the clean-up selectors, then
// loop through the slow way to see if they still need to be converted.
if (!ExceptionValueIntrinsic) {
ExceptionValueIntrinsic =
Intrinsic::getDeclaration(F->getParent(), Intrinsic::eh_exception);
if (!ExceptionValueIntrinsic)
return CleanupSelectors(CatchAllSels);
}
for (Value::use_iterator
I = ExceptionValueIntrinsic->use_begin(),
E = ExceptionValueIntrinsic->use_end(); I != E; ++I) {
IntrinsicInst *EHPtr = dyn_cast<IntrinsicInst>(*I);
if (!EHPtr || EHPtr->getParent()->getParent() != F) continue;
Changed |= PromoteEHPtrStore(EHPtr);
bool URoRInvoke = false;
SmallPtrSet<IntrinsicInst*, 8> SelCalls;
Changed |= FindSelectorAndURoR(EHPtr, URoRInvoke, SelCalls);
if (URoRInvoke) {
// This EH pointer is being used by an invoke of an URoR instruction and
// an eh.selector intrinsic call. If the eh.selector is a 'clean-up', we
// need to convert it to a 'catch-all'.
for (SmallPtrSet<IntrinsicInst*, 8>::iterator
SI = SelCalls.begin(), SE = SelCalls.end(); SI != SE; ++SI)
if (!HasCatchAllInSelector(*SI))
SelsToConvert.insert(*SI);
}
}
}
if (!SelsToConvert.empty()) {
// Convert all clean-up eh.selectors, which are associated with "invokes" of
// URoR calls, into catch-all eh.selectors.
Changed = true;
for (SmallPtrSet<IntrinsicInst*, 8>::iterator
SI = SelsToConvert.begin(), SE = SelsToConvert.end();
SI != SE; ++SI) {
IntrinsicInst *II = *SI;
// Use the exception object pointer and the personality function
// from the original selector.
CallSite CS(II);
IntrinsicInst::op_iterator I = CS.arg_begin();
IntrinsicInst::op_iterator E = CS.arg_end();
IntrinsicInst::op_iterator B = prior(E);
// Exclude last argument if it is an integer.
if (isa<ConstantInt>(B)) E = B;
// Add exception object pointer (front).
// Add personality function (next).
// Add in any filter IDs (rest).
SmallVector<Value*, 8> Args(I, E);
Args.push_back(EHCatchAllValue->getInitializer()); // Catch-all indicator.
CallInst *NewSelector =
CallInst::Create(SelectorIntrinsic, Args.begin(), Args.end(),
"eh.sel.catch.all", II);
NewSelector->setTailCall(II->isTailCall());
NewSelector->setAttributes(II->getAttributes());
NewSelector->setCallingConv(II->getCallingConv());
II->replaceAllUsesWith(NewSelector);
II->eraseFromParent();
}
}
Changed |= CleanupSelectors(CatchAllSels);
return Changed;
}
void moduleDepGraph::matchParametersAndReturnValues(Function &F) {
// Only do the matching if F has any use
if (F.isVarArg() || !F.hasNUsesOrMore(1)) {
return;
}
// Data structure which contains the matches between formal and real parameters
// First: formal parameter
// Second: real parameter
SmallVector<std::pair<GraphNode*, GraphNode*>, 4> Parameters(F.arg_size());
// Fetch the function arguments (formal parameters) into the data structure
Function::arg_iterator argptr;
Function::arg_iterator e;
unsigned i;
//Create the PHI nodes for the formal parameters
for (i = 0, argptr = F.arg_begin(), e = F.arg_end(); argptr != e; ++i, ++argptr) {
OpNode* argPHI = new OpNode(Instruction::PHI);
GraphNode* argNode = NULL;
argNode = depGraph->addInst(argptr);
if (argNode != NULL)
depGraph->addEdge(argPHI, argNode);
Parameters[i].first = argPHI;
}
// Check if the function returns a supported value type. If not, no return value matching is done
bool noReturn = F.getReturnType()->isVoidTy();
// Creates the data structure which receives the return values of the function, if there is any
SmallPtrSet<llvm::Value*, 8> ReturnValues;
if (!noReturn) {
// Iterate over the basic blocks to fetch all possible return values
for (Function::iterator bb = F.begin(), bbend = F.end(); bb != bbend; ++bb) {
// Get the terminator instruction of the basic block and check if it's
// a return instruction: if it's not, continue to next basic block
Instruction *terminator = bb->getTerminator();
ReturnInst *RI = dyn_cast<ReturnInst> (terminator);
if (!RI)
continue;
// Get the return value and insert in the data structure
ReturnValues.insert(RI->getReturnValue());
}
}
for (Value::use_iterator UI = F.use_begin(), E = F.use_end(); UI != E; ++UI) {
User *U = *UI;
// Ignore blockaddress uses
if (isa<BlockAddress> (U))
continue;
// Used by a non-instruction, or not the callee of a function, do not
// match.
if (!isa<CallInst> (U) && !isa<InvokeInst> (U))
continue;
Instruction *caller = cast<Instruction> (U);
CallSite CS(caller);
if (!CS.isCallee(UI))
continue;
// Iterate over the real parameters and put them in the data structure
CallSite::arg_iterator AI;
CallSite::arg_iterator EI;
for (i = 0, AI = CS.arg_begin(), EI = CS.arg_end(); AI != EI; ++i, ++AI) {
Parameters[i].second = depGraph->addInst(*AI);
}
// Match formal and real parameters
for (i = 0; i < Parameters.size(); ++i) {
depGraph->addEdge(Parameters[i].second, Parameters[i].first);
}
// Match return values
if (!noReturn) {
OpNode* retPHI = new OpNode(Instruction::PHI);
GraphNode* callerNode = depGraph->addInst(caller);
depGraph->addEdge(retPHI, callerNode);
for (SmallPtrSetIterator<llvm::Value*> ri = ReturnValues.begin(),
re = ReturnValues.end(); ri != re; ++ri) {
GraphNode* retNode = depGraph->addInst(*ri);
depGraph->addEdge(retNode, retPHI);
}
}
// Real parameters are cleaned before moving to the next use (for safety's sake)
for (i = 0; i < Parameters.size(); ++i)
Parameters[i].second = NULL;
}
depGraph->deleteCallNodes(&F);
}
/// Given \p BBs as input, find another set of BBs which collectively
/// dominates \p BBs and have the minimal sum of frequencies. Return the BB
/// set found in \p BBs.
static void findBestInsertionSet(DominatorTree &DT, BlockFrequencyInfo &BFI,
BasicBlock *Entry,
SmallPtrSet<BasicBlock *, 8> &BBs) {
assert(!BBs.count(Entry) && "Assume Entry is not in BBs");
// Nodes on the current path to the root.
SmallPtrSet<BasicBlock *, 8> Path;
// Candidates includes any block 'BB' in set 'BBs' that is not strictly
// dominated by any other blocks in set 'BBs', and all nodes in the path
// in the dominator tree from Entry to 'BB'.
SmallPtrSet<BasicBlock *, 16> Candidates;
for (auto BB : BBs) {
// Ignore unreachable basic blocks.
if (!DT.isReachableFromEntry(BB))
continue;
Path.clear();
// Walk up the dominator tree until Entry or another BB in BBs
// is reached. Insert the nodes on the way to the Path.
BasicBlock *Node = BB;
// The "Path" is a candidate path to be added into Candidates set.
bool isCandidate = false;
do {
Path.insert(Node);
if (Node == Entry || Candidates.count(Node)) {
isCandidate = true;
break;
}
assert(DT.getNode(Node)->getIDom() &&
"Entry doens't dominate current Node");
Node = DT.getNode(Node)->getIDom()->getBlock();
} while (!BBs.count(Node));
// If isCandidate is false, Node is another Block in BBs dominating
// current 'BB'. Drop the nodes on the Path.
if (!isCandidate)
continue;
// Add nodes on the Path into Candidates.
Candidates.insert(Path.begin(), Path.end());
}
// Sort the nodes in Candidates in top-down order and save the nodes
// in Orders.
unsigned Idx = 0;
SmallVector<BasicBlock *, 16> Orders;
Orders.push_back(Entry);
while (Idx != Orders.size()) {
BasicBlock *Node = Orders[Idx++];
for (auto ChildDomNode : DT.getNode(Node)->getChildren()) {
if (Candidates.count(ChildDomNode->getBlock()))
Orders.push_back(ChildDomNode->getBlock());
}
}
// Visit Orders in bottom-up order.
using InsertPtsCostPair =
std::pair<SmallPtrSet<BasicBlock *, 16>, BlockFrequency>;
// InsertPtsMap is a map from a BB to the best insertion points for the
// subtree of BB (subtree not including the BB itself).
DenseMap<BasicBlock *, InsertPtsCostPair> InsertPtsMap;
InsertPtsMap.reserve(Orders.size() + 1);
for (auto RIt = Orders.rbegin(); RIt != Orders.rend(); RIt++) {
BasicBlock *Node = *RIt;
bool NodeInBBs = BBs.count(Node);
SmallPtrSet<BasicBlock *, 16> &InsertPts = InsertPtsMap[Node].first;
BlockFrequency &InsertPtsFreq = InsertPtsMap[Node].second;
// Return the optimal insert points in BBs.
if (Node == Entry) {
BBs.clear();
if (InsertPtsFreq > BFI.getBlockFreq(Node) ||
(InsertPtsFreq == BFI.getBlockFreq(Node) && InsertPts.size() > 1))
BBs.insert(Entry);
else
BBs.insert(InsertPts.begin(), InsertPts.end());
break;
}
BasicBlock *Parent = DT.getNode(Node)->getIDom()->getBlock();
// Initially, ParentInsertPts is empty and ParentPtsFreq is 0. Every child
// will update its parent's ParentInsertPts and ParentPtsFreq.
SmallPtrSet<BasicBlock *, 16> &ParentInsertPts = InsertPtsMap[Parent].first;
BlockFrequency &ParentPtsFreq = InsertPtsMap[Parent].second;
// Choose to insert in Node or in subtree of Node.
// Don't hoist to EHPad because we may not find a proper place to insert
// in EHPad.
// If the total frequency of InsertPts is the same as the frequency of the
// target Node, and InsertPts contains more than one nodes, choose hoisting
// to reduce code size.
if (NodeInBBs ||
(!Node->isEHPad() &&
(InsertPtsFreq > BFI.getBlockFreq(Node) ||
(InsertPtsFreq == BFI.getBlockFreq(Node) && InsertPts.size() > 1)))) {
ParentInsertPts.insert(Node);
ParentPtsFreq += BFI.getBlockFreq(Node);
} else {
ParentInsertPts.insert(InsertPts.begin(), InsertPts.end());
ParentPtsFreq += InsertPtsFreq;
}
}
}
/// inlineFuctions - Walk all call sites in all functions supplied by
/// client. Inline as many call sites as possible. Delete completely
/// inlined functions.
void BasicInlinerImpl::inlineFunctions() {
// Scan through and identify all call sites ahead of time so that we only
// inline call sites in the original functions, not call sites that result
// from inlining other functions.
std::vector<CallSite> CallSites;
for (std::vector<Function *>::iterator FI = Functions.begin(),
FE = Functions.end(); FI != FE; ++FI) {
Function *F = *FI;
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
for (BasicBlock::iterator I = BB->begin(); I != BB->end(); ++I) {
CallSite CS(cast<Value>(I));
if (CS && CS.getCalledFunction()
&& !CS.getCalledFunction()->isDeclaration())
CallSites.push_back(CS);
}
}
DEBUG(dbgs() << ": " << CallSites.size() << " call sites.\n");
// Inline call sites.
bool Changed = false;
do {
Changed = false;
for (unsigned index = 0; index != CallSites.size() && !CallSites.empty();
++index) {
CallSite CS = CallSites[index];
if (Function *Callee = CS.getCalledFunction()) {
// Eliminate calls that are never inlinable.
if (Callee->isDeclaration() ||
CS.getInstruction()->getParent()->getParent() == Callee) {
CallSites.erase(CallSites.begin() + index);
--index;
continue;
}
InlineCost IC = CA.getInlineCost(CS, NeverInline);
if (IC.isAlways()) {
DEBUG(dbgs() << " Inlining: cost=always"
<<", call: " << *CS.getInstruction());
} else if (IC.isNever()) {
DEBUG(dbgs() << " NOT Inlining: cost=never"
<<", call: " << *CS.getInstruction());
continue;
} else {
int Cost = IC.getValue();
if (Cost >= (int) BasicInlineThreshold) {
DEBUG(dbgs() << " NOT Inlining: cost = " << Cost
<< ", call: " << *CS.getInstruction());
continue;
} else {
DEBUG(dbgs() << " Inlining: cost = " << Cost
<< ", call: " << *CS.getInstruction());
}
}
// Inline
InlineFunctionInfo IFI(0, TD);
if (InlineFunction(CS, IFI)) {
Callee->removeDeadConstantUsers();
if (Callee->isDefTriviallyDead())
DeadFunctions.insert(Callee);
Changed = true;
CallSites.erase(CallSites.begin() + index);
--index;
}
}
}
} while (Changed);
// Remove completely inlined functions from module.
for(SmallPtrSet<Function *, 8>::iterator I = DeadFunctions.begin(),
E = DeadFunctions.end(); I != E; ++I) {
Function *D = *I;
Module *M = D->getParent();
M->getFunctionList().remove(D);
}
}
/// Some initialization for the rename part
///
void SSI::renameInit(SmallPtrSet<Instruction*, 4> &value) {
for (SmallPtrSet<Instruction*, 4>::iterator I = value.begin(),
E = value.end(); I != E; ++I)
value_stack[*I].push_back(*I);
}
bool ReduceCrashingBlocks::TestBlocks(std::vector<const BasicBlock*> &BBs) {
// Clone the program to try hacking it apart...
ValueToValueMapTy VMap;
Module *M = CloneModule(BD.getProgram(), VMap);
// Convert list to set for fast lookup...
SmallPtrSet<BasicBlock*, 8> Blocks;
for (unsigned i = 0, e = BBs.size(); i != e; ++i)
Blocks.insert(cast<BasicBlock>(VMap[BBs[i]]));
outs() << "Checking for crash with only these blocks:";
unsigned NumPrint = Blocks.size();
if (NumPrint > 10) NumPrint = 10;
for (unsigned i = 0, e = NumPrint; i != e; ++i)
outs() << " " << BBs[i]->getName();
if (NumPrint < Blocks.size())
outs() << "... <" << Blocks.size() << " total>";
outs() << ": ";
// Loop over and delete any hack up any blocks that are not listed...
for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
for (Function::iterator BB = I->begin(), E = I->end(); BB != E; ++BB)
if (!Blocks.count(BB) && BB->getTerminator()->getNumSuccessors()) {
// Loop over all of the successors of this block, deleting any PHI nodes
// that might include it.
for (succ_iterator SI = succ_begin(BB), E = succ_end(BB); SI != E; ++SI)
(*SI)->removePredecessor(BB);
TerminatorInst *BBTerm = BB->getTerminator();
if (!BB->getTerminator()->getType()->isVoidTy())
BBTerm->replaceAllUsesWith(Constant::getNullValue(BBTerm->getType()));
// Replace the old terminator instruction.
BB->getInstList().pop_back();
new UnreachableInst(BB->getContext(), BB);
}
// The CFG Simplifier pass may delete one of the basic blocks we are
// interested in. If it does we need to take the block out of the list. Make
// a "persistent mapping" by turning basic blocks into <function, name> pairs.
// This won't work well if blocks are unnamed, but that is just the risk we
// have to take.
std::vector<std::pair<std::string, std::string> > BlockInfo;
for (SmallPtrSet<BasicBlock*, 8>::iterator I = Blocks.begin(),
E = Blocks.end(); I != E; ++I)
BlockInfo.push_back(std::make_pair((*I)->getParent()->getName(),
(*I)->getName()));
// Now run the CFG simplify pass on the function...
std::vector<std::string> Passes;
Passes.push_back("simplifycfg");
Passes.push_back("verify");
Module *New = BD.runPassesOn(M, Passes);
delete M;
if (!New) {
errs() << "simplifycfg failed!\n";
exit(1);
}
M = New;
// Try running on the hacked up program...
if (TestFn(BD, M)) {
BD.setNewProgram(M); // It crashed, keep the trimmed version...
// Make sure to use basic block pointers that point into the now-current
// module, and that they don't include any deleted blocks.
BBs.clear();
const ValueSymbolTable &GST = M->getValueSymbolTable();
for (unsigned i = 0, e = BlockInfo.size(); i != e; ++i) {
Function *F = cast<Function>(GST.lookup(BlockInfo[i].first));
ValueSymbolTable &ST = F->getValueSymbolTable();
Value* V = ST.lookup(BlockInfo[i].second);
if (V && V->getType() == Type::getLabelTy(V->getContext()))
BBs.push_back(cast<BasicBlock>(V));
}
return true;
}
delete M; // It didn't crash, try something else.
return false;
}
/// Renames all variables in the specified BasicBlock.
/// Only variables that need to be rename will be.
///
void SSI::rename(BasicBlock *BB) {
SmallPtrSet<Instruction*, 8> defined;
// Iterate through instructions and make appropriate renaming.
// For SSI_PHI (b = PHI()), store b at value_stack as a new
// definition of the variable it represents.
// For SSI_SIG (b = PHI(a)), substitute a with the current
// value of a, present in the value_stack.
// Then store bin the value_stack as the new definition of a.
// For all other instructions (b = OP(a, c, d, ...)), we need to substitute
// all operands with its current value, present in value_stack.
for (BasicBlock::iterator begin = BB->begin(), end = BB->end();
begin != end; ++begin) {
Instruction *I = begin;
if (PHINode *PN = dyn_cast<PHINode>(I)) { // Treat PHI functions
Instruction* position;
// Treat SSI_PHI
if ((position = getPositionPhi(PN))) {
value_stack[position].push_back(PN);
defined.insert(position);
// Treat SSI_SIG
} else if ((position = getPositionSigma(PN))) {
substituteUse(I);
value_stack[position].push_back(PN);
defined.insert(position);
}
// Treat all other PHI functions
else {
substituteUse(I);
}
}
// Treat all other functions
else {
substituteUse(I);
}
}
// This loop iterates in all BasicBlocks that are successors of the current
// BasicBlock. For each SSI_PHI instruction found, insert an operand.
// This operand is the current operand in value_stack for the variable
// in "position". And the BasicBlock this operand represents is the current
// BasicBlock.
for (succ_iterator SI = succ_begin(BB), SE = succ_end(BB); SI != SE; ++SI) {
BasicBlock *BB_succ = *SI;
for (BasicBlock::iterator begin = BB_succ->begin(),
notPhi = BB_succ->getFirstNonPHI(); begin != *notPhi; ++begin) {
Instruction *I = begin;
PHINode *PN = dyn_cast<PHINode>(I);
Instruction* position;
if (PN && ((position = getPositionPhi(PN)))) {
PN->addIncoming(value_stack[position].back(), BB);
}
}
}
// This loop calls rename on all children from this block. This time children
// refers to a successor block in the dominance tree.
DomTreeNode *DTN = DT_->getNode(BB);
for (DomTreeNode::iterator begin = DTN->begin(), end = DTN->end();
begin != end; ++begin) {
DomTreeNodeBase<BasicBlock> *DTN_children = *begin;
BasicBlock *BB_children = DTN_children->getBlock();
rename(BB_children);
}
// Now we remove all inserted definitions of a variable from the top of
// the stack leaving the previous one as the top.
for (SmallPtrSet<Instruction*, 8>::iterator DI = defined.begin(),
DE = defined.end(); DI != DE; ++DI)
value_stack[*DI].pop_back();
}
void fixTerminators(MachineBasicBlock *MBB) {
SmallPtrSet<MachineBasicBlock*, 2> MissedSuccs;
MissedSuccs.insert(MBB->succ_begin(), MBB->succ_end());
MachineInstr *FirstTerminator = 0;
for (MachineBasicBlock::iterator II = MBB->getFirstTerminator(),
IE = MBB->end(); II != IE; ++II) {
MachineInstr *Inst = II;
if (!VInstrInfo::isBrCndLike(Inst->getOpcode())) continue;
MachineBasicBlock *TargetBB = Inst->getOperand(1).getMBB();
MachineOperand Cnd = Inst->getOperand(0);
//bool inserted;
//jt_it at;
//tie(at, inserted) = Table.insert(std::make_pair(TargetBB, Cnd));
// BranchFolding may generate code that jumping to same bb with multiple
// instruction, merge the condition.
//if (!inserted) {
// at->second = VInstrInfo::MergePred(Cnd, at->second, *MBB,
// MBB->getFirstTerminator(), &MRI,
// TII, VTM::VOpOr);
//}
// Change the unconditional branch after conditional branch to
// conditional branch.
if (FirstTerminator && VInstrInfo::isUnConditionalBranch(Inst)){
MachineOperand &TrueCnd = FirstTerminator->getOperand(0);
MachineOperand &FalseCnd = Inst->getOperand(0);
TrueCnd.setIsKill(false);
FalseCnd.setReg(TrueCnd.getReg());
FalseCnd.setTargetFlags(TrueCnd.getTargetFlags());
VInstrInfo::ReversePredicateCondition(FalseCnd);
}
FirstTerminator = Inst;
MissedSuccs.erase(TargetBB);
}
// Make sure each basic block have a terminator.
if (!MissedSuccs.empty()) {
assert(MissedSuccs.size() == 1 && "Fall through to multiple blocks?");
++UnconditionalBranches;
MachineOperand Cnd = VInstrInfo::CreatePredicate();
if (FirstTerminator) {
MachineOperand &TrueCnd = FirstTerminator->getOperand(0);
assert(TrueCnd.getReg() != 0 && "Two unconditional branch?");
// We will use the register somewhere else
TrueCnd.setIsKill(false);
Cnd = TrueCnd;
VInstrInfo::ReversePredicateCondition(Cnd);
}
BuildMI(MBB, DebugLoc(), VInstrInfo::getDesc(VTM::VOpToStateb))
.addOperand(Cnd).addMBB(*MissedSuccs.begin())
.addOperand(VInstrInfo::CreatePredicate())
.addOperand(VInstrInfo::CreateTrace());
}
//else if (Table.size() != MBB->succ_size()) {
// // Also fix the CFG.
// while (!MBB->succ_empty())
// MBB->removeSuccessor(MBB->succ_end() - 1);
// for (jt_it JI = Table.begin(), JE = Table.end(); JI != JE; ++JI)
// MBB->addSuccessor(JI->first);
// // Try to correct the CFG.
// TII->RemoveBranch(*MBB);
// VInstrInfo::insertJumpTable(*MBB, Table, DebugLoc());
//}
//Table.clear();
if (MBB->succ_size() == 0 && MBB->getFirstTerminator() == MBB->end()) {
++Unreachables;
BuildMI(MBB, DebugLoc(), VInstrInfo::getDesc(VTM::VOpUnreachable))
.addOperand(VInstrInfo::CreatePredicate())
.addOperand(VInstrInfo::CreateTrace());
}
}
/// At this point, we're committed to promoting the alloca using IDF's, and the
/// standard SSA construction algorithm. Determine which blocks need phi nodes
/// and see if we can optimize out some work by avoiding insertion of dead phi
/// nodes.
void PromoteMem2Reg::DetermineInsertionPoint(AllocaInst *AI, unsigned AllocaNum,
AllocaInfo &Info) {
// Unique the set of defining blocks for efficient lookup.
SmallPtrSet<BasicBlock *, 32> DefBlocks;
DefBlocks.insert(Info.DefiningBlocks.begin(), Info.DefiningBlocks.end());
// Determine which blocks the value is live in. These are blocks which lead
// to uses.
SmallPtrSet<BasicBlock *, 32> LiveInBlocks;
ComputeLiveInBlocks(AI, Info, DefBlocks, LiveInBlocks);
// Use a priority queue keyed on dominator tree level so that inserted nodes
// are handled from the bottom of the dominator tree upwards.
typedef std::pair<DomTreeNode *, unsigned> DomTreeNodePair;
typedef std::priority_queue<DomTreeNodePair, SmallVector<DomTreeNodePair, 32>,
less_second> IDFPriorityQueue;
IDFPriorityQueue PQ;
for (SmallPtrSet<BasicBlock *, 32>::const_iterator I = DefBlocks.begin(),
E = DefBlocks.end();
I != E; ++I) {
if (DomTreeNode *Node = DT.getNode(*I))
PQ.push(std::make_pair(Node, DomLevels[Node]));
}
SmallVector<std::pair<unsigned, BasicBlock *>, 32> DFBlocks;
SmallPtrSet<DomTreeNode *, 32> Visited;
SmallVector<DomTreeNode *, 32> Worklist;
while (!PQ.empty()) {
DomTreeNodePair RootPair = PQ.top();
PQ.pop();
DomTreeNode *Root = RootPair.first;
unsigned RootLevel = RootPair.second;
// Walk all dominator tree children of Root, inspecting their CFG edges with
// targets elsewhere on the dominator tree. Only targets whose level is at
// most Root's level are added to the iterated dominance frontier of the
// definition set.
Worklist.clear();
Worklist.push_back(Root);
while (!Worklist.empty()) {
DomTreeNode *Node = Worklist.pop_back_val();
BasicBlock *BB = Node->getBlock();
for (succ_iterator SI = succ_begin(BB), SE = succ_end(BB); SI != SE;
++SI) {
DomTreeNode *SuccNode = DT.getNode(*SI);
// Quickly skip all CFG edges that are also dominator tree edges instead
// of catching them below.
if (SuccNode->getIDom() == Node)
continue;
unsigned SuccLevel = DomLevels[SuccNode];
if (SuccLevel > RootLevel)
continue;
if (!Visited.insert(SuccNode))
continue;
BasicBlock *SuccBB = SuccNode->getBlock();
if (!LiveInBlocks.count(SuccBB))
continue;
DFBlocks.push_back(std::make_pair(BBNumbers[SuccBB], SuccBB));
if (!DefBlocks.count(SuccBB))
PQ.push(std::make_pair(SuccNode, SuccLevel));
}
for (DomTreeNode::iterator CI = Node->begin(), CE = Node->end(); CI != CE;
++CI) {
if (!Visited.count(*CI))
Worklist.push_back(*CI);
}
}
}
if (DFBlocks.size() > 1)
std::sort(DFBlocks.begin(), DFBlocks.end());
unsigned CurrentVersion = 0;
for (unsigned i = 0, e = DFBlocks.size(); i != e; ++i)
QueuePhiNode(DFBlocks[i].second, AllocaNum, CurrentVersion);
}
void SanityCheckInstructionsPass::findInstructions(Function *F) {
// A list of instructions that are used by sanity checks. They become sanity
// check instructions if it turns out they're not used by anything else.
SmallPtrSet<Instruction*, 128> Worklist;
// A list of basic blocks that contain sanity check instructions. They
// become sanity check blocks if it turns out they don't contain anything
// else.
SmallPtrSet<BasicBlock*, 64> BlockWorklist;
// A map from instructions to the checks that use them.
std::map<Instruction*, SmallPtrSet<Instruction*, 4> > ChecksByInstruction;
for (BasicBlock &BB: *F) {
if (findSanityCheckCall(&BB)) {
SanityCheckBlocks[F].insert(&BB);
// All instructions inside sanity check blocks are sanity check instructions
for (Instruction &I: BB) {
Worklist.insert(&I);
}
// All branches to sanity check blocks are sanity check branches
for (User *U: BB.users()) {
if (Instruction *Inst = dyn_cast<Instruction>(U)) {
Worklist.insert(Inst);
}
BranchInst *BI = dyn_cast<BranchInst>(U);
if (BI && BI->isConditional()) {
SanityCheckBranches[F].insert(BI);
ChecksByInstruction[BI].insert(BI);
}
}
}
}
while (!Worklist.empty()) {
// Alternate between emptying the worklist...
while (!Worklist.empty()) {
Instruction *Inst = *Worklist.begin();
Worklist.erase(Inst);
if (onlyUsedInSanityChecks(Inst)) {
if (SanityCheckInstructions[F].insert(Inst).second) {
for (Use &U: Inst->operands()) {
if (Instruction *Op = dyn_cast<Instruction>(U.get())) {
Worklist.insert(Op);
// Copy ChecksByInstruction from Inst to Op
auto CBI = ChecksByInstruction.find(Inst);
if (CBI != ChecksByInstruction.end()) {
ChecksByInstruction[Op].insert(CBI->second.begin(), CBI->second.end());
}
}
}
BlockWorklist.insert(Inst->getParent());
// Fill InstructionsBySanityCheck from the inverse ChecksByInstruction
auto CBI = ChecksByInstruction.find(Inst);
if (CBI != ChecksByInstruction.end()) {
for (Instruction *CI : CBI->second) {
InstructionsBySanityCheck[CI].insert(Inst);
}
}
}
}
}
// ... and checking whether this causes basic blocks to contain only
// sanity checks. This would in turn cause terminators to be added to
// the worklist.
while (!BlockWorklist.empty()) {
BasicBlock *BB = *BlockWorklist.begin();
BlockWorklist.erase(BB);
bool allInstructionsAreSanityChecks = true;
for (Instruction &I: *BB) {
if (!SanityCheckInstructions.at(BB->getParent()).count(&I)) {
allInstructionsAreSanityChecks = false;
break;
}
}
if (allInstructionsAreSanityChecks) {
for (User *U: BB->users()) {
if (Instruction *Inst = dyn_cast<Instruction>(U)) {
Worklist.insert(Inst);
}
}
}
}
}
}
