/*
* Main args are always input
* Functions currently considered as input functions:
* scanf
* fscanf
* gets
* fgets
* fread
* fgetc
* getc
* getchar
* recv
* recvmsg
* read
* recvfrom
* fread
*/
bool InputDep::runOnModule(Module &M) {
// DEBUG (errs() << "Function " << F.getName() << "\n";);
NumInputValues = 0;
bool inserted;
Function* main = M.getFunction("main");
if (main) {
MDNode *mdn = main->begin()->begin()->getMetadata("dbg");
for (Function::arg_iterator Arg = main->arg_begin(), aEnd =
main->arg_end(); Arg != aEnd; Arg++) {
inputDepValues.insert(Arg);
NumInputValues++;
if (mdn) {
DILocation Loc(mdn);
unsigned Line = Loc.getLineNumber();
lineNo[Arg] = Line-1; //educated guess (can only get line numbers from insts, suppose main is declared one line before 1st inst
}
}
}
for (Module::iterator F = M.begin(), eM = M.end(); F != eM; ++F) {
for (Function::iterator BB = F->begin(), e = F->end(); BB != e; ++BB) {
for (BasicBlock::iterator I = BB->begin(), ee = BB->end(); I != ee; ++I) {
if (CallInst *CI = dyn_cast<CallInst>(I)) {
Function *Callee = CI->getCalledFunction();
if (Callee) {
Value* V;
inserted = false;
StringRef Name = Callee->getName();
if (Name.equals("main")) {
errs() << "main\n";
V = CI->getArgOperand(1); //char* argv[]
inputDepValues.insert(V);
inserted = true;
//errs() << "Input " << *V << "\n";
}
if (Name.equals("__isoc99_scanf") || Name.equals(
"scanf")) {
for (unsigned i = 1, eee = CI->getNumArgOperands(); i
!= eee; ++i) { // skip format string (i=1)
V = CI->getArgOperand(i);
if (V->getType()->isPointerTy()) {
inputDepValues.insert(V);
inserted = true;
//errs() << "Input " << *V << "\n";
}
}
} else if (Name.equals("__isoc99_fscanf")
|| Name.equals("fscanf")) {
for (unsigned i = 2, eee = CI->getNumArgOperands(); i
!= eee; ++i) { // skip file pointer and format string (i=1)
V = CI->getArgOperand(i);
if (V->getType()->isPointerTy()) {
inputDepValues.insert(V);
inserted = true;
//errs() << "Input " << *V << "\n";
}
}
} else if ((Name.equals("gets") || Name.equals("fgets")
|| Name.equals("fread"))
|| Name.equals("getwd")
|| Name.equals("getcwd")) {
V = CI->getArgOperand(0); //the first argument receives the input for these functions
if (V->getType()->isPointerTy()) {
inputDepValues.insert(V);
inserted = true;
//errs() << "Input " << *V << "\n";
}
} else if ((Name.equals("fgetc") || Name.equals("getc")
|| Name.equals("getchar"))) {
inputDepValues.insert(CI);
inserted = true;
//errs() << "Input " << *CI << "\n";
} else if (Name.equals("recv")
|| Name.equals("recvmsg")
|| Name.equals("read")) {
Value* V = CI->getArgOperand(1);
if (V->getType()->isPointerTy()) {
inputDepValues.insert(V);
inserted = true;
//errs() << "Input " << *V << "\n";
}
} else if (Name.equals("recvfrom")) {
V = CI->getArgOperand(1);
if (V->getType()->isPointerTy()) {
inputDepValues.insert(V);
inserted = true;
//errs() << "Input " << *V << "\n";
}
V = CI->getArgOperand(4);
if (V->getType()->isPointerTy()) {
inputDepValues.insert(V);
inserted = true;
//errs() << "Input " << *V << "\n";
}
}
if (inserted) {
if (MDNode *mdn = I->getMetadata("dbg")) {
NumInputValues++;
DILocation Loc(mdn);
unsigned Line = Loc.getLineNumber();
lineNo[V] = Line;
}
}
}
}
}
}
}
DEBUG(printer());
return false;
}
void StatsTracker::computeReachableUncovered() {
KModule *km = executor.kmodule;
Module *m = km->module;
static bool init = true;
const InstructionInfoTable &infos = *km->infos;
StatisticManager &sm = *theStatisticManager;
if (init) {
init = false;
// Compute call targets. It would be nice to use alias information
// instead of assuming all indirect calls hit all escaping
// functions, eh?
for (Module::iterator fnIt = m->begin(), fn_ie = m->end();
fnIt != fn_ie; ++fnIt) {
for (Function::iterator bbIt = fnIt->begin(), bb_ie = fnIt->end();
bbIt != bb_ie; ++bbIt) {
for (BasicBlock::iterator it = bbIt->begin(), ie = bbIt->end();
it != ie; ++it) {
if (isa<CallInst>(it) || isa<InvokeInst>(it)) {
if (isa<InlineAsm>(it->getOperand(0))) {
// We can never call through here so assume no targets
// (which should be correct anyhow).
callTargets.insert(std::make_pair(it,
std::vector<Function*>()));
} else if (Function *target = getDirectCallTarget(it)) {
callTargets[it].push_back(target);
} else {
callTargets[it] =
std::vector<Function*>(km->escapingFunctions.begin(),
km->escapingFunctions.end());
}
}
}
}
}
// Compute function callers as reflexion of callTargets.
for (calltargets_ty::iterator it = callTargets.begin(),
ie = callTargets.end(); it != ie; ++it)
for (std::vector<Function*>::iterator fit = it->second.begin(),
fie = it->second.end(); fit != fie; ++fit)
functionCallers[*fit].push_back(it->first);
// Initialize minDistToReturn to shortest paths through
// functions. 0 is unreachable.
std::vector<Instruction *> instructions;
for (Module::iterator fnIt = m->begin(), fn_ie = m->end();
fnIt != fn_ie; ++fnIt) {
if (fnIt->isDeclaration()) {
if (fnIt->doesNotReturn()) {
functionShortestPath[fnIt] = 0;
} else {
functionShortestPath[fnIt] = 1; // whatever
}
continue;
} else {
functionShortestPath[fnIt] = 0;
}
KFunction *kf = km->functionMap[fnIt];
for (unsigned i = 0; i < kf->numInstructions; ++i) {
Instruction *inst = kf->instrPostOrder[i]->inst;
instructions.push_back(inst);
sm.setIndexedValue(stats::minDistToReturn,
kf->instrPostOrder[i]->info->id,
isa<ReturnInst>(inst));
}
}
// I'm so lazy it's not even worklisted.
bool changed;
do {
changed = false;
for (std::vector<Instruction*>::iterator it = instructions.begin(),
ie = instructions.end(); it != ie; ++it) {
Instruction *inst = *it;
unsigned bestThrough = 0;
if (isa<CallInst>(inst) || isa<InvokeInst>(inst)) {
std::vector<Function*> &targets = callTargets[inst];
for (std::vector<Function*>::iterator fnIt = targets.begin(),
ie = targets.end(); fnIt != ie; ++fnIt) {
uint64_t dist = functionShortestPath[*fnIt];
if (dist) {
dist = 1+dist; // count instruction itself
if (bestThrough==0 || dist<bestThrough)
bestThrough = dist;
}
}
} else {
bestThrough = 1;
}
if (bestThrough) {
unsigned id = infos.getInfo(*it).id;
uint64_t best, cur = best = sm.getIndexedValue(stats::minDistToReturn, id);
std::vector<Instruction*> succs = getSuccs(*it);
for (std::vector<Instruction*>::iterator it2 = succs.begin(),
ie = succs.end(); it2 != ie; ++it2) {
uint64_t dist = sm.getIndexedValue(stats::minDistToReturn,
infos.getInfo(*it2).id);
if (dist) {
uint64_t val = bestThrough + dist;
if (best==0 || val<best)
best = val;
}
}
if (best != cur) {
sm.setIndexedValue(stats::minDistToReturn, id, best);
changed = true;
// Update shortest path if this is the entry point.
Function *f = inst->getParent()->getParent();
if (inst==f->begin()->begin())
functionShortestPath[f] = best;
}
}
}
} while (changed);
}
// compute minDistToUncovered, 0 is unreachable
std::vector<Instruction *> instructions;
std::vector<unsigned> ids;
for (Module::iterator fnIt = m->begin(), fn_ie = m->end();
fnIt != fn_ie; ++fnIt) {
if (fnIt->isDeclaration())
continue;
KFunction *kf = km->functionMap[fnIt];
for (unsigned i = 0; i < kf->numInstructions; ++i) {
Instruction *inst = kf->instrPostOrder[i]->inst;
unsigned id = kf->instrPostOrder[i]->info->id;
instructions.push_back(inst);
ids.push_back(id);
sm.setIndexedValue(stats::minDistToGloballyUncovered,
id,
sm.getIndexedValue(stats::globallyUncoveredInstructions, id));
}
}
// I'm so lazy it's not even worklisted.
bool changed;
do {
changed = false;
for (unsigned i = 0; i < instructions.size(); ++i) {
Instruction *inst = instructions[i];
unsigned id = ids[i];
uint64_t best, cur = best = sm.getIndexedValue(stats::minDistToGloballyUncovered,
id);
unsigned bestThrough = 0;
if (isa<CallInst>(inst) || isa<InvokeInst>(inst)) {
std::vector<Function*> &targets = callTargets[inst];
for (std::vector<Function*>::iterator fnIt = targets.begin(),
ie = targets.end(); fnIt != ie; ++fnIt) {
uint64_t dist = functionShortestPath[*fnIt];
if (dist) {
dist = 1+dist; // count instruction itself
if (bestThrough==0 || dist<bestThrough)
bestThrough = dist;
}
if (!(*fnIt)->isDeclaration()) {
uint64_t calleeDist = sm.getIndexedValue(stats::minDistToGloballyUncovered,
infos.getFunctionInfo(*fnIt).id);
if (calleeDist) {
calleeDist = 1+calleeDist; // count instruction itself
if (best==0 || calleeDist<best)
best = calleeDist;
}
}
}
} else {
bestThrough = 1;
}
if (bestThrough) {
std::vector<Instruction*> succs = getSuccs(inst);
for (std::vector<Instruction*>::iterator it2 = succs.begin(),
ie = succs.end(); it2 != ie; ++it2) {
uint64_t dist = sm.getIndexedValue(stats::minDistToGloballyUncovered,
infos.getInfo(*it2).id);
if (dist) {
uint64_t val = bestThrough + dist;
if (best==0 || val<best)
best = val;
}
}
}
if (best != cur) {
sm.setIndexedValue(stats::minDistToGloballyUncovered,
infos.getInfo(inst).id,
best);
changed = true;
}
}
} while (changed);
for (std::set<ExecutionState*>::iterator it = executor.states.begin(),
ie = executor.states.end(); it != ie; ++it) {
ExecutionState *es = *it;
uint64_t currentFrameMinDist = 0;
for (ExecutionState::stack_ty::iterator sfIt = es->stack().begin(),
sf_ie = es->stack().end(); sfIt != sf_ie; ++sfIt) {
ExecutionState::stack_ty::iterator next = sfIt + 1;
KInstIterator kii;
if (next==es->stack().end()) {
kii = es->pc();
} else {
kii = next->caller;
++kii;
}
sfIt->minDistToUncoveredOnReturn = currentFrameMinDist;
currentFrameMinDist = computeMinDistToUncovered(kii, currentFrameMinDist);
}
}
LOG(INFO) << "Processed " << instructions.size() << " instructions in static analysis";
}
/// 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());
}
}
void CallTargetFinder<dsa>::findIndTargets(Module &M)
{
dsa* T = &getAnalysis<dsa>();
const DSCallGraph & callgraph = T->getCallGraph();
DSGraph* G = T->getGlobalsGraph();
DSGraph::ScalarMapTy& SM = G->getScalarMap();
for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
if (!I->isDeclaration())
for (Function::iterator F = I->begin(), FE = I->end(); F != FE; ++F)
for (BasicBlock::iterator B = F->begin(), BE = F->end(); B != BE; ++B)
if (isa<CallInst>(B) || isa<InvokeInst>(B)) {
CallSite cs(B);
AllSites.push_back(cs);
Function* CF = cs.getCalledFunction();
//
// If the called function is casted from one function type to
// another, peer into the cast instruction and pull out the actual
// function being called.
//
if (!CF)
CF = dyn_cast<Function>(cs.getCalledValue()->stripPointerCasts());
if (!CF) {
Value * calledValue = cs.getCalledValue()->stripPointerCasts();
if (isa<ConstantPointerNull>(calledValue)) {
++DirCall;
CompleteSites.insert(cs);
} else {
IndCall++;
DSCallGraph::callee_iterator csi = callgraph.callee_begin(cs),
cse = callgraph.callee_end(cs);
while(csi != cse) {
const Function *F = *csi;
DSCallGraph::scc_iterator sccii = callgraph.scc_begin(F),
sccee = callgraph.scc_end(F);
for(;sccii != sccee; ++sccii) {
DSGraph::ScalarMapTy::const_iterator I = SM.find(SM.getLeaderForGlobal(*sccii));
if (I != SM.end()) {
IndMap[cs].push_back (*sccii);
}
}
++csi;
}
const Function *F1 = (cs).getInstruction()->getParent()->getParent();
F1 = callgraph.sccLeader(&*F1);
DSCallGraph::scc_iterator sccii = callgraph.scc_begin(F1),
sccee = callgraph.scc_end(F1);
for(;sccii != sccee; ++sccii) {
DSGraph::ScalarMapTy::const_iterator I = SM.find(SM.getLeaderForGlobal(*sccii));
if (I != SM.end()) {
IndMap[cs].push_back (*sccii);
}
}
DSNode* N = T->getDSGraph(*cs.getCaller())
->getNodeForValue(cs.getCalledValue()).getNode();
assert (N && "CallTarget: findIndTargets: No DSNode!\n");
if (!N->isIncompleteNode() && !N->isExternalNode() && IndMap[cs].size()) {
CompleteSites.insert(cs);
++CompleteInd;
}
if (!N->isIncompleteNode() && !N->isExternalNode() && !IndMap[cs].size()) {
++CompleteEmpty;
DEBUG(errs() << "Call site empty: '"
<< cs.getInstruction()->getName()
<< "' In '"
<< cs.getInstruction()->getParent()->getParent()->getName()
<< "'\n");
}
}
} else {
++DirCall;
IndMap[cs].push_back(CF);
CompleteSites.insert(cs);
}
}
}
// TODO(sbucur): Break this into multiple methods
void StatsTracker::getCallgraphProfile(data::GlobalProfile &globalProfile) {
Module *m = executor.kmodule->module;
uint64_t istatsMask = 0;
StatisticManager &sm = *theStatisticManager;
unsigned nStats = sm.getNumStatistics();
istatsMask |= 1<<sm.getStatisticID("Queries");
istatsMask |= 1<<sm.getStatisticID("QueriesValid");
istatsMask |= 1<<sm.getStatisticID("QueriesInvalid");
istatsMask |= 1<<sm.getStatisticID("QueryTime");
istatsMask |= 1<<sm.getStatisticID("ResolveTime");
istatsMask |= 1<<sm.getStatisticID("Instructions");
istatsMask |= 1<<sm.getStatisticID("InstructionTimes");
istatsMask |= 1<<sm.getStatisticID("InstructionRealTimes");
istatsMask |= 1<<sm.getStatisticID("Forks");
istatsMask |= 1<<sm.getStatisticID("GloballyCoveredInstructions");
istatsMask |= 1<<sm.getStatisticID("GloballyUncoveredInstructions");
istatsMask |= 1<<sm.getStatisticID("States");
istatsMask |= 1<<sm.getStatisticID("MinDistToUncovered");
for (unsigned i=0; i<nStats; i++) {
if (istatsMask & (1<<i)) {
Statistic &s = sm.getStatistic(i);
globalProfile.add_cost_label(s.getName());
}
}
globalProfile.set_time_stamp(::time(NULL));
// set state counts, decremented after we process so that we don't
// have to zero all records each time.
if (istatsMask & (1<<stats::states.getID()))
updateStateStatistics(1);
CallSiteSummaryTable callSiteStats;
if (UseCallPaths)
callPathManager.getSummaryStatistics(callSiteStats);
for (Module::iterator fnIt = m->begin(), fn_ie = m->end();
fnIt != fn_ie; ++fnIt) {
if (fnIt->isDeclaration())
continue;
data::FunctionProfile *functionProfile = globalProfile.add_function_profile();
functionProfile->set_function_id(executor.kmodule->functionMap[&(*fnIt)]->nameID);
for (Function::iterator bbIt = fnIt->begin(), bb_ie = fnIt->end();
bbIt != bb_ie; ++bbIt) {
for (BasicBlock::iterator it = bbIt->begin(), ie = bbIt->end();
it != ie; ++it) {
Instruction *instr = &*it;
const InstructionInfo &ii = executor.kmodule->infos->getInfo(instr);
unsigned index = ii.id;
data::LineProfile *lineProfile = functionProfile->add_line_profile();
executor.kmodule->fillInstructionDebugInfo(
instr, *lineProfile->mutable_debug_info());
for (unsigned i=0; i<nStats; i++) {
if (istatsMask&(1<<i)) {
lineProfile->add_cost_value(
sm.getIndexedValue(sm.getStatistic(i), index));
}
}
if (UseCallPaths &&
(isa<CallInst>(instr) || isa<InvokeInst>(instr))) {
CallSiteSummaryTable::iterator it = callSiteStats.find(instr);
if (it!=callSiteStats.end()) {
for (std::map<llvm::Function*, CallSiteInfo>::iterator
fit = it->second.begin(), fie = it->second.end();
fit != fie; ++fit) {
Function *f = fit->first;
CallSiteInfo &csi = fit->second;
data::CallSiteProfile *callsiteProfile = lineProfile->add_call_site_profile();
executor.kmodule->fillFunctionDebugInfo(
f, *callsiteProfile->mutable_debug_info());
callsiteProfile->set_call_count(csi.count);
for (unsigned i=0; i<nStats; i++) {
if (istatsMask&(1<<i)) {
Statistic &s = sm.getStatistic(i);
uint64_t value;
// Hack, ignore things that don't make sense on
// call paths.
if (&s == &stats::globallyUncoveredInstructions) {
value = 0;
} else {
value = csi.statistics.getValue(s);
}
callsiteProfile->add_cost_value(value);
}
}
}
}
}
}
}
}
if (istatsMask & (1<<stats::states.getID()))
updateStateStatistics((uint64_t)-1);
}
/// InlineFunction - This function inlines the called function into the basic
/// block of the caller. This returns false if it is not possible to inline
/// this call. The program is still in a well defined state if this occurs
/// though.
///
/// Note that this only does one level of inlining. For example, if the
/// instruction 'call B' is inlined, and 'B' calls 'C', then the call to 'C' now
/// exists in the instruction stream. Similarly this will inline a recursive
/// function by one level.
bool llvm::InlineFunction(CallSite CS, InlineFunctionInfo &IFI,
bool InsertLifetime) {
Instruction *TheCall = CS.getInstruction();
assert(TheCall->getParent() && TheCall->getParent()->getParent() &&
"Instruction not in function!");
// If IFI has any state in it, zap it before we fill it in.
IFI.reset();
const Function *CalledFunc = CS.getCalledFunction();
if (CalledFunc == 0 || // Can't inline external function or indirect
CalledFunc->isDeclaration() || // call, or call to a vararg function!
CalledFunc->getFunctionType()->isVarArg()) return false;
// If the call to the callee is not a tail call, we must clear the 'tail'
// flags on any calls that we inline.
bool MustClearTailCallFlags =
!(isa<CallInst>(TheCall) && cast<CallInst>(TheCall)->isTailCall());
// If the call to the callee cannot throw, set the 'nounwind' flag on any
// calls that we inline.
bool MarkNoUnwind = CS.doesNotThrow();
BasicBlock *OrigBB = TheCall->getParent();
Function *Caller = OrigBB->getParent();
// GC poses two hazards to inlining, which only occur when the callee has GC:
// 1. If the caller has no GC, then the callee's GC must be propagated to the
// caller.
// 2. If the caller has a differing GC, it is invalid to inline.
if (CalledFunc->hasGC()) {
if (!Caller->hasGC())
Caller->setGC(CalledFunc->getGC());
else if (CalledFunc->getGC() != Caller->getGC())
return false;
}
// Get the personality function from the callee if it contains a landing pad.
Value *CalleePersonality = 0;
for (Function::const_iterator I = CalledFunc->begin(), E = CalledFunc->end();
I != E; ++I)
if (const InvokeInst *II = dyn_cast<InvokeInst>(I->getTerminator())) {
const BasicBlock *BB = II->getUnwindDest();
const LandingPadInst *LP = BB->getLandingPadInst();
CalleePersonality = LP->getPersonalityFn();
break;
}
// Find the personality function used by the landing pads of the caller. If it
// exists, then check to see that it matches the personality function used in
// the callee.
if (CalleePersonality) {
for (Function::const_iterator I = Caller->begin(), E = Caller->end();
I != E; ++I)
if (const InvokeInst *II = dyn_cast<InvokeInst>(I->getTerminator())) {
const BasicBlock *BB = II->getUnwindDest();
const LandingPadInst *LP = BB->getLandingPadInst();
// If the personality functions match, then we can perform the
// inlining. Otherwise, we can't inline.
// TODO: This isn't 100% true. Some personality functions are proper
// supersets of others and can be used in place of the other.
if (LP->getPersonalityFn() != CalleePersonality)
return false;
break;
}
}
// Get an iterator to the last basic block in the function, which will have
// the new function inlined after it.
Function::iterator LastBlock = &Caller->back();
// Make sure to capture all of the return instructions from the cloned
// function.
SmallVector<ReturnInst*, 8> Returns;
ClonedCodeInfo InlinedFunctionInfo;
Function::iterator FirstNewBlock;
{ // Scope to destroy VMap after cloning.
ValueToValueMapTy VMap;
assert(CalledFunc->arg_size() == CS.arg_size() &&
"No varargs calls can be inlined!");
// Calculate the vector of arguments to pass into the function cloner, which
// matches up the formal to the actual argument values.
CallSite::arg_iterator AI = CS.arg_begin();
unsigned ArgNo = 0;
for (Function::const_arg_iterator I = CalledFunc->arg_begin(),
E = CalledFunc->arg_end(); I != E; ++I, ++AI, ++ArgNo) {
Value *ActualArg = *AI;
// When byval arguments actually inlined, we need to make the copy implied
// by them explicit. However, we don't do this if the callee is readonly
// or readnone, because the copy would be unneeded: the callee doesn't
// modify the struct.
if (CS.isByValArgument(ArgNo)) {
ActualArg = HandleByValArgument(ActualArg, TheCall, CalledFunc, IFI,
CalledFunc->getParamAlignment(ArgNo+1));
// Calls that we inline may use the new alloca, so we need to clear
// their 'tail' flags if HandleByValArgument introduced a new alloca and
// the callee has calls.
MustClearTailCallFlags |= ActualArg != *AI;
}
VMap[I] = ActualArg;
}
// We want the inliner to prune the code as it copies. We would LOVE to
// have no dead or constant instructions leftover after inlining occurs
// (which can happen, e.g., because an argument was constant), but we'll be
// happy with whatever the cloner can do.
CloneAndPruneFunctionInto(Caller, CalledFunc, VMap,
/*ModuleLevelChanges=*/false, Returns, ".i",
&InlinedFunctionInfo, IFI.TD, TheCall);
// Remember the first block that is newly cloned over.
FirstNewBlock = LastBlock; ++FirstNewBlock;
// Update the callgraph if requested.
if (IFI.CG)
UpdateCallGraphAfterInlining(CS, FirstNewBlock, VMap, IFI);
// Update inlined instructions' line number information.
fixupLineNumbers(Caller, FirstNewBlock, TheCall);
}
// If there are any alloca instructions in the block that used to be the entry
// block for the callee, move them to the entry block of the caller. First
// calculate which instruction they should be inserted before. We insert the
// instructions at the end of the current alloca list.
{
BasicBlock::iterator InsertPoint = Caller->begin()->begin();
for (BasicBlock::iterator I = FirstNewBlock->begin(),
E = FirstNewBlock->end(); I != E; ) {
AllocaInst *AI = dyn_cast<AllocaInst>(I++);
if (AI == 0) continue;
// If the alloca is now dead, remove it. This often occurs due to code
// specialization.
if (AI->use_empty()) {
AI->eraseFromParent();
continue;
}
if (!isa<Constant>(AI->getArraySize()))
continue;
// Keep track of the static allocas that we inline into the caller.
IFI.StaticAllocas.push_back(AI);
// Scan for the block of allocas that we can move over, and move them
// all at once.
while (isa<AllocaInst>(I) &&
isa<Constant>(cast<AllocaInst>(I)->getArraySize())) {
IFI.StaticAllocas.push_back(cast<AllocaInst>(I));
++I;
}
// Transfer all of the allocas over in a block. Using splice means
// that the instructions aren't removed from the symbol table, then
// reinserted.
Caller->getEntryBlock().getInstList().splice(InsertPoint,
FirstNewBlock->getInstList(),
AI, I);
}
}
// Leave lifetime markers for the static alloca's, scoping them to the
// function we just inlined.
if (InsertLifetime && !IFI.StaticAllocas.empty()) {
IRBuilder<> builder(FirstNewBlock->begin());
for (unsigned ai = 0, ae = IFI.StaticAllocas.size(); ai != ae; ++ai) {
AllocaInst *AI = IFI.StaticAllocas[ai];
// If the alloca is already scoped to something smaller than the whole
// function then there's no need to add redundant, less accurate markers.
if (hasLifetimeMarkers(AI))
continue;
builder.CreateLifetimeStart(AI);
for (unsigned ri = 0, re = Returns.size(); ri != re; ++ri) {
IRBuilder<> builder(Returns[ri]);
builder.CreateLifetimeEnd(AI);
}
}
}
// If the inlined code contained dynamic alloca instructions, wrap the inlined
// code with llvm.stacksave/llvm.stackrestore intrinsics.
if (InlinedFunctionInfo.ContainsDynamicAllocas) {
Module *M = Caller->getParent();
// Get the two intrinsics we care about.
Function *StackSave = Intrinsic::getDeclaration(M, Intrinsic::stacksave);
Function *StackRestore=Intrinsic::getDeclaration(M,Intrinsic::stackrestore);
// Insert the llvm.stacksave.
CallInst *SavedPtr = IRBuilder<>(FirstNewBlock, FirstNewBlock->begin())
.CreateCall(StackSave, "savedstack");
// Insert a call to llvm.stackrestore before any return instructions in the
// inlined function.
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
IRBuilder<>(Returns[i]).CreateCall(StackRestore, SavedPtr);
}
}
// If we are inlining tail call instruction through a call site that isn't
// marked 'tail', we must remove the tail marker for any calls in the inlined
// code. Also, calls inlined through a 'nounwind' call site should be marked
// 'nounwind'.
if (InlinedFunctionInfo.ContainsCalls &&
(MustClearTailCallFlags || MarkNoUnwind)) {
for (Function::iterator BB = FirstNewBlock, E = Caller->end();
BB != E; ++BB)
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
if (CallInst *CI = dyn_cast<CallInst>(I)) {
if (MustClearTailCallFlags)
CI->setTailCall(false);
if (MarkNoUnwind)
CI->setDoesNotThrow();
}
}
// If we are inlining for an invoke instruction, we must make sure to rewrite
// any call instructions into invoke instructions.
if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall))
HandleInlinedInvoke(II, FirstNewBlock, InlinedFunctionInfo);
// If we cloned in _exactly one_ basic block, and if that block ends in a
// return instruction, we splice the body of the inlined callee directly into
// the calling basic block.
if (Returns.size() == 1 && std::distance(FirstNewBlock, Caller->end()) == 1) {
// Move all of the instructions right before the call.
OrigBB->getInstList().splice(TheCall, FirstNewBlock->getInstList(),
FirstNewBlock->begin(), FirstNewBlock->end());
// Remove the cloned basic block.
Caller->getBasicBlockList().pop_back();
// If the call site was an invoke instruction, add a branch to the normal
// destination.
if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall))
BranchInst::Create(II->getNormalDest(), TheCall);
// If the return instruction returned a value, replace uses of the call with
// uses of the returned value.
if (!TheCall->use_empty()) {
ReturnInst *R = Returns[0];
if (TheCall == R->getReturnValue())
TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
else
TheCall->replaceAllUsesWith(R->getReturnValue());
}
// Since we are now done with the Call/Invoke, we can delete it.
TheCall->eraseFromParent();
// Since we are now done with the return instruction, delete it also.
Returns[0]->eraseFromParent();
// We are now done with the inlining.
return true;
}
// Otherwise, we have the normal case, of more than one block to inline or
// multiple return sites.
// We want to clone the entire callee function into the hole between the
// "starter" and "ender" blocks. How we accomplish this depends on whether
// this is an invoke instruction or a call instruction.
BasicBlock *AfterCallBB;
if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) {
// Add an unconditional branch to make this look like the CallInst case...
BranchInst *NewBr = BranchInst::Create(II->getNormalDest(), TheCall);
// Split the basic block. This guarantees that no PHI nodes will have to be
// updated due to new incoming edges, and make the invoke case more
// symmetric to the call case.
AfterCallBB = OrigBB->splitBasicBlock(NewBr,
CalledFunc->getName()+".exit");
} else { // It's a call
// If this is a call instruction, we need to split the basic block that
// the call lives in.
//
AfterCallBB = OrigBB->splitBasicBlock(TheCall,
CalledFunc->getName()+".exit");
}
// Change the branch that used to go to AfterCallBB to branch to the first
// basic block of the inlined function.
//
TerminatorInst *Br = OrigBB->getTerminator();
assert(Br && Br->getOpcode() == Instruction::Br &&
"splitBasicBlock broken!");
Br->setOperand(0, FirstNewBlock);
// Now that the function is correct, make it a little bit nicer. In
// particular, move the basic blocks inserted from the end of the function
// into the space made by splitting the source basic block.
Caller->getBasicBlockList().splice(AfterCallBB, Caller->getBasicBlockList(),
FirstNewBlock, Caller->end());
// Handle all of the return instructions that we just cloned in, and eliminate
// any users of the original call/invoke instruction.
Type *RTy = CalledFunc->getReturnType();
PHINode *PHI = 0;
if (Returns.size() > 1) {
// The PHI node should go at the front of the new basic block to merge all
// possible incoming values.
if (!TheCall->use_empty()) {
PHI = PHINode::Create(RTy, Returns.size(), TheCall->getName(),
AfterCallBB->begin());
// Anything that used the result of the function call should now use the
// PHI node as their operand.
TheCall->replaceAllUsesWith(PHI);
}
// Loop over all of the return instructions adding entries to the PHI node
// as appropriate.
if (PHI) {
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
ReturnInst *RI = Returns[i];
assert(RI->getReturnValue()->getType() == PHI->getType() &&
"Ret value not consistent in function!");
PHI->addIncoming(RI->getReturnValue(), RI->getParent());
}
}
// Add a branch to the merge points and remove return instructions.
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
ReturnInst *RI = Returns[i];
BranchInst::Create(AfterCallBB, RI);
RI->eraseFromParent();
}
} else if (!Returns.empty()) {
// Otherwise, if there is exactly one return value, just replace anything
// using the return value of the call with the computed value.
if (!TheCall->use_empty()) {
if (TheCall == Returns[0]->getReturnValue())
TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
else
TheCall->replaceAllUsesWith(Returns[0]->getReturnValue());
}
// Update PHI nodes that use the ReturnBB to use the AfterCallBB.
BasicBlock *ReturnBB = Returns[0]->getParent();
ReturnBB->replaceAllUsesWith(AfterCallBB);
// Splice the code from the return block into the block that it will return
// to, which contains the code that was after the call.
AfterCallBB->getInstList().splice(AfterCallBB->begin(),
ReturnBB->getInstList());
// Delete the return instruction now and empty ReturnBB now.
Returns[0]->eraseFromParent();
ReturnBB->eraseFromParent();
} else if (!TheCall->use_empty()) {
// No returns, but something is using the return value of the call. Just
// nuke the result.
TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
}
// Since we are now done with the Call/Invoke, we can delete it.
TheCall->eraseFromParent();
// We should always be able to fold the entry block of the function into the
// single predecessor of the block...
assert(cast<BranchInst>(Br)->isUnconditional() && "splitBasicBlock broken!");
BasicBlock *CalleeEntry = cast<BranchInst>(Br)->getSuccessor(0);
// Splice the code entry block into calling block, right before the
// unconditional branch.
CalleeEntry->replaceAllUsesWith(OrigBB); // Update PHI nodes
OrigBB->getInstList().splice(Br, CalleeEntry->getInstList());
// Remove the unconditional branch.
OrigBB->getInstList().erase(Br);
// Now we can remove the CalleeEntry block, which is now empty.
Caller->getBasicBlockList().erase(CalleeEntry);
// If we inserted a phi node, check to see if it has a single value (e.g. all
// the entries are the same or undef). If so, remove the PHI so it doesn't
// block other optimizations.
if (PHI) {
if (Value *V = SimplifyInstruction(PHI, IFI.TD)) {
PHI->replaceAllUsesWith(V);
PHI->eraseFromParent();
}
}
return true;
}
bool GenericToNVVM::runOnModule(Module &M) {
// Create a clone of each global variable that has the default address space.
// The clone is created with the global address space specifier, and the pair
// of original global variable and its clone is placed in the GVMap for later
// use.
for (Module::global_iterator I = M.global_begin(), E = M.global_end();
I != E;) {
GlobalVariable *GV = I++;
if (GV->getType()->getAddressSpace() == llvm::ADDRESS_SPACE_GENERIC &&
!llvm::isTexture(*GV) && !llvm::isSurface(*GV) &&
!llvm::isSampler(*GV) && !GV->getName().startswith("llvm.")) {
GlobalVariable *NewGV = new GlobalVariable(
M, GV->getType()->getElementType(), GV->isConstant(),
GV->getLinkage(),
GV->hasInitializer() ? GV->getInitializer() : nullptr,
"", GV, GV->getThreadLocalMode(), llvm::ADDRESS_SPACE_GLOBAL);
NewGV->copyAttributesFrom(GV);
GVMap[GV] = NewGV;
}
}
// Return immediately, if every global variable has a specific address space
// specifier.
if (GVMap.empty()) {
return false;
}
// Walk through the instructions in function defitinions, and replace any use
// of original global variables in GVMap with a use of the corresponding
// copies in GVMap. If necessary, promote constants to instructions.
for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) {
if (I->isDeclaration()) {
continue;
}
IRBuilder<> Builder(I->getEntryBlock().getFirstNonPHIOrDbg());
for (Function::iterator BBI = I->begin(), BBE = I->end(); BBI != BBE;
++BBI) {
for (BasicBlock::iterator II = BBI->begin(), IE = BBI->end(); II != IE;
++II) {
for (unsigned i = 0, e = II->getNumOperands(); i < e; ++i) {
Value *Operand = II->getOperand(i);
if (isa<Constant>(Operand)) {
II->setOperand(
i, remapConstant(&M, I, cast<Constant>(Operand), Builder));
}
}
}
}
ConstantToValueMap.clear();
}
// Walk through the metadata section and update the debug information
// associated with the global variables in the default address space.
for (Module::named_metadata_iterator I = M.named_metadata_begin(),
E = M.named_metadata_end();
I != E; I++) {
remapNamedMDNode(&M, I);
}
// Walk through the global variable initializers, and replace any use of
// original global variables in GVMap with a use of the corresponding copies
// in GVMap. The copies need to be bitcast to the original global variable
// types, as we cannot use cvta in global variable initializers.
for (GVMapTy::iterator I = GVMap.begin(), E = GVMap.end(); I != E;) {
GlobalVariable *GV = I->first;
GlobalVariable *NewGV = I->second;
// Remove GV from the map so that it can be RAUWed. Note that
// DenseMap::erase() won't invalidate any iterators but this one.
auto Next = std::next(I);
GVMap.erase(I);
I = Next;
Constant *BitCastNewGV = ConstantExpr::getPointerCast(NewGV, GV->getType());
// At this point, the remaining uses of GV should be found only in global
// variable initializers, as other uses have been already been removed
// while walking through the instructions in function definitions.
GV->replaceAllUsesWith(BitCastNewGV);
std::string Name = GV->getName();
GV->eraseFromParent();
NewGV->setName(Name);
}
assert(GVMap.empty() && "Expected it to be empty by now");
return true;
}
bool PruneEH::runOnSCC(const std::vector<CallGraphNode *> &SCC) {
CallGraph &CG = getAnalysis<CallGraph>();
bool MadeChange = false;
// First pass, scan all of the functions in the SCC, simplifying them
// according to what we know.
for (unsigned i = 0, e = SCC.size(); i != e; ++i)
if (Function *F = SCC[i]->getFunction())
MadeChange |= SimplifyFunction(F);
// Next, check to see if any callees might throw or if there are any external
// functions in this SCC: if so, we cannot prune any functions in this SCC.
// If this SCC includes the unwind instruction, we KNOW it throws, so
// obviously the SCC might throw.
//
bool SCCMightUnwind = false, SCCMightReturn = false;
for (unsigned i = 0, e = SCC.size();
(!SCCMightUnwind || !SCCMightReturn) && i != e; ++i) {
Function *F = SCC[i]->getFunction();
if (F == 0 || (F->isDeclaration() && !F->getIntrinsicID())) {
SCCMightUnwind = true;
SCCMightReturn = true;
} else {
if (F->isDeclaration())
SCCMightReturn = true;
// Check to see if this function performs an unwind or calls an
// unwinding function.
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
if (isa<UnwindInst>(BB->getTerminator())) { // Uses unwind!
SCCMightUnwind = true;
} else if (isa<ReturnInst>(BB->getTerminator())) {
SCCMightReturn = true;
}
// Invoke instructions don't allow unwinding to continue, so we are
// only interested in call instructions.
if (!SCCMightUnwind)
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
if (CallInst *CI = dyn_cast<CallInst>(I)) {
if (Function *Callee = CI->getCalledFunction()) {
CallGraphNode *CalleeNode = CG[Callee];
// If the callee is outside our current SCC, or if it is not
// known to throw, then we might throw also.
if (std::find(SCC.begin(), SCC.end(), CalleeNode) == SCC.end()&&
!DoesNotUnwind.count(CalleeNode)) {
SCCMightUnwind = true;
break;
}
} else {
// Indirect call, it might throw.
SCCMightUnwind = true;
break;
}
}
if (SCCMightUnwind && SCCMightReturn) break;
}
}
}
// If the SCC doesn't unwind or doesn't throw, note this fact.
if (!SCCMightUnwind)
for (unsigned i = 0, e = SCC.size(); i != e; ++i)
DoesNotUnwind.insert(SCC[i]);
if (!SCCMightReturn)
for (unsigned i = 0, e = SCC.size(); i != e; ++i)
DoesNotReturn.insert(SCC[i]);
for (unsigned i = 0, e = SCC.size(); i != e; ++i) {
// Convert any invoke instructions to non-throwing functions in this node
// into call instructions with a branch. This makes the exception blocks
// dead.
if (Function *F = SCC[i]->getFunction())
MadeChange |= SimplifyFunction(F);
}
return MadeChange;
}
/*
* Rewrite OpenMP call sites and their associated kernel functions -- the folloiwng pattern
call void @GOMP_parallel_start(void (i8*)* @_Z20initialize_variablesiPfS_.omp_fn.4, i8* %.omp_data_o.5571, i32 0) nounwind
call void @_Z20initialize_variablesiPfS_.omp_fn.4(i8* %.omp_data_o.5571) nounwind
call void @GOMP_parallel_end() nounwind
*/
void HeteroOMPTransform::rewrite_omp_call_sites(Module &M) {
SmallVector<Instruction *, 16> toDelete;
DenseMap<Value *, Value *> ValueMap;
for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I){
if (!I->isDeclaration()) {
for (Function::iterator BBI = I->begin(), BBE = I->end(); BBI != BBE; ++BBI) {
bool match = false;
for (BasicBlock::iterator INSNI = BBI->begin(), INSNE = BBI->end(); INSNI != INSNE; ++INSNI) {
if (isa<CallInst>(INSNI)) {
CallSite CI(cast<Instruction>(INSNI));
if (CI.getCalledFunction() != NULL){
string called_func_name = CI.getCalledFunction()->getName();
if (called_func_name == OMP_PARALLEL_START_NAME && CI.arg_size() == 3) {
// change alloc to malloc_shared
// %5 = call i8* @_Z13malloc_sharedm(i64 20) ; <i8*> [#uses=5]
// %6 = bitcast i8* %5 to float* ; <float*> [#uses=2]
AllocaInst *AllocCall;
Value *arg_0 = CI.getArgument(0); // function
Value *arg_1 = CI.getArgument(1); // context
Value *loop_ub = NULL;
Function *function;
BitCastInst* BCI;
Function *kernel_function;
BasicBlock::iterator iI(*INSNI);
//BasicBlock::iterator iJ = iI+1;
iI++; iI++;
//BasicBlock::iterator iK = iI;
CallInst /**next,*/ *next_next;
if (arg_0 != NULL && arg_1 != NULL /*&& (next = dyn_cast<CallInst>(*iJ))*/
&& (next_next = dyn_cast<CallInst>(iI)) && (next_next->getCalledFunction() != NULL)
&& (next_next->getCalledFunction()->getName() == OMP_PARALLEL_END_NAME)
&& (BCI = dyn_cast<BitCastInst>(arg_1)) && (AllocCall = dyn_cast<AllocaInst>(BCI->getOperand(0)))
&& (function = dyn_cast<Function>(arg_0)) && (loop_ub = find_loop_upper_bound (AllocCall))
&& (kernel_function=convert_to_kernel_function (M, function))){
SmallVector<Value*, 16> Args;
Args.push_back(AllocCall->getArraySize());
Instruction *MallocCall = CallInst::Create(mallocFnTy, Args, "", AllocCall);
CastInst *MallocCast = CastInst::Create(Instruction::BitCast, MallocCall, AllocCall->getType(), "", AllocCall);
ValueMap[AllocCall] = MallocCast;
//AllocCall->replaceAllUsesWith(MallocCall);
// Add offload function
Args.clear();
Args.push_back(loop_ub);
Args.push_back(BCI);
Args.push_back(kernel_function);
if (offloadFnTy == NULL) {
init_offload_type(M, kernel_function);
}
Instruction *call = CallInst::Create(offloadFnTy, Args, "", INSNI);
if (find(toDelete.begin(), toDelete.end(), AllocCall) == toDelete.end()){
toDelete.push_back(AllocCall);
}
toDelete.push_back(&(*INSNI));
match = true;
}
}
else if (called_func_name == OMP_PARALLEL_END_NAME && CI.arg_size() == 0 && match) {
toDelete.push_back(&(*INSNI));
match = false;
}
else if (match) {
toDelete.push_back(&(*INSNI));
}
}
}
}
}
}
}
/* Replace AllocCalls by MallocCalls */
for(DenseMap<Value *, Value *>::iterator I = ValueMap.begin(), E = ValueMap.end(); I != E; I++) {
I->first->replaceAllUsesWith(I->second);
}
/* delete the instructions for get_omp_num_thread and get_omp_thread_num */
while(!toDelete.empty()) {
Instruction *g = toDelete.back();
toDelete.pop_back();
g->eraseFromParent();
}
}
/// splitLiveRangesAcrossInvokes - Each value that is live across an unwind edge
/// we spill into a stack location, guaranteeing that there is nothing live
/// across the unwind edge. This process also splits all critical edges
/// coming out of invoke's.
/// FIXME: Move this function to a common utility file (Local.cpp?) so
/// both SjLj and LowerInvoke can use it.
void SjLjEHPass::
splitLiveRangesAcrossInvokes(SmallVector<InvokeInst*,16> &Invokes) {
// First step, split all critical edges from invoke instructions.
for (unsigned i = 0, e = Invokes.size(); i != e; ++i) {
InvokeInst *II = Invokes[i];
SplitCriticalEdge(II, 0, this);
// FIXME: New EH - This if-condition will be always true in the new scheme.
if (II->getUnwindDest()->isLandingPad()) {
SmallVector<BasicBlock*, 2> NewBBs;
SplitLandingPadPredecessors(II->getUnwindDest(), II->getParent(),
".1", ".2", this, NewBBs);
LPadSuccMap[II] = *succ_begin(NewBBs[0]);
} else {
SplitCriticalEdge(II, 1, this);
}
assert(!isa<PHINode>(II->getNormalDest()) &&
!isa<PHINode>(II->getUnwindDest()) &&
"Critical edge splitting left single entry phi nodes?");
}
Function *F = Invokes.back()->getParent()->getParent();
// To avoid having to handle incoming arguments specially, we lower each arg
// to a copy instruction in the entry block. This ensures that the argument
// value itself cannot be live across the entry block.
BasicBlock::iterator AfterAllocaInsertPt = F->begin()->begin();
while (isa<AllocaInst>(AfterAllocaInsertPt) &&
isa<ConstantInt>(cast<AllocaInst>(AfterAllocaInsertPt)->getArraySize()))
++AfterAllocaInsertPt;
for (Function::arg_iterator AI = F->arg_begin(), E = F->arg_end();
AI != E; ++AI) {
Type *Ty = AI->getType();
// Aggregate types can't be cast, but are legal argument types, so we have
// to handle them differently. We use an extract/insert pair as a
// lightweight method to achieve the same goal.
if (isa<StructType>(Ty) || isa<ArrayType>(Ty) || isa<VectorType>(Ty)) {
Instruction *EI = ExtractValueInst::Create(AI, 0, "",AfterAllocaInsertPt);
Instruction *NI = InsertValueInst::Create(AI, EI, 0);
NI->insertAfter(EI);
AI->replaceAllUsesWith(NI);
// Set the operand of the instructions back to the AllocaInst.
EI->setOperand(0, AI);
NI->setOperand(0, AI);
} else {
// This is always a no-op cast because we're casting AI to AI->getType()
// so src and destination types are identical. BitCast is the only
// possibility.
CastInst *NC = new BitCastInst(
AI, AI->getType(), AI->getName()+".tmp", AfterAllocaInsertPt);
AI->replaceAllUsesWith(NC);
// Set the operand of the cast instruction back to the AllocaInst.
// Normally it's forbidden to replace a CastInst's operand because it
// could cause the opcode to reflect an illegal conversion. However,
// we're replacing it here with the same value it was constructed with.
// We do this because the above replaceAllUsesWith() clobbered the
// operand, but we want this one to remain.
NC->setOperand(0, AI);
}
}
// Finally, scan the code looking for instructions with bad live ranges.
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++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.
std::set<BasicBlock*> 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)) {
NeedsSpill = true;
}
}
// 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) {
++NumSpilled;
DemoteRegToStack(*Inst, true);
}
}
}
bool SjLjEHPass::insertSjLjEHSupport(Function &F) {
SmallVector<ReturnInst*,16> Returns;
SmallVector<UnwindInst*,16> Unwinds;
SmallVector<InvokeInst*,16> Invokes;
// Look through the terminators of the basic blocks to find invokes, returns
// and unwinds.
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
if (ReturnInst *RI = dyn_cast<ReturnInst>(BB->getTerminator())) {
// Remember all return instructions in case we insert an invoke into this
// function.
Returns.push_back(RI);
} else if (InvokeInst *II = dyn_cast<InvokeInst>(BB->getTerminator())) {
Invokes.push_back(II);
} else if (UnwindInst *UI = dyn_cast<UnwindInst>(BB->getTerminator())) {
Unwinds.push_back(UI);
}
}
NumInvokes += Invokes.size();
NumUnwinds += Unwinds.size();
// If we don't have any invokes, there's nothing to do.
if (Invokes.empty()) return false;
// Find the eh.selector.*, eh.exception and alloca calls.
//
// Remember any allocas() that aren't in the entry block, as the
// jmpbuf saved SP will need to be updated for them.
//
// We'll use the first eh.selector to determine the right personality
// function to use. For SJLJ, we always use the same personality for the
// whole function, not on a per-selector basis.
// FIXME: That's a bit ugly. Better way?
SmallVector<CallInst*,16> EH_Selectors;
SmallVector<CallInst*,16> EH_Exceptions;
SmallVector<Instruction*,16> JmpbufUpdatePoints;
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
// Note: Skip the entry block since there's nothing there that interests
// us. eh.selector and eh.exception shouldn't ever be there, and we
// want to disregard any allocas that are there.
//
// FIXME: This is awkward. The new EH scheme won't need to skip the entry
// block.
if (BB == F.begin()) {
if (InvokeInst *II = dyn_cast<InvokeInst>(F.begin()->getTerminator())) {
// FIXME: This will be always non-NULL in the new EH.
if (LandingPadInst *LPI = II->getUnwindDest()->getLandingPadInst())
if (!PersonalityFn) PersonalityFn = LPI->getPersonalityFn();
}
continue;
}
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
if (CallInst *CI = dyn_cast<CallInst>(I)) {
if (CI->getCalledFunction() == SelectorFn) {
if (!PersonalityFn) PersonalityFn = CI->getArgOperand(1);
EH_Selectors.push_back(CI);
} else if (CI->getCalledFunction() == ExceptionFn) {
EH_Exceptions.push_back(CI);
} else if (CI->getCalledFunction() == StackRestoreFn) {
JmpbufUpdatePoints.push_back(CI);
}
} else if (AllocaInst *AI = dyn_cast<AllocaInst>(I)) {
JmpbufUpdatePoints.push_back(AI);
} else if (InvokeInst *II = dyn_cast<InvokeInst>(I)) {
// FIXME: This will be always non-NULL in the new EH.
if (LandingPadInst *LPI = II->getUnwindDest()->getLandingPadInst())
if (!PersonalityFn) PersonalityFn = LPI->getPersonalityFn();
}
}
}
// If we don't have any eh.selector calls, we can't determine the personality
// function. Without a personality function, we can't process exceptions.
if (!PersonalityFn) return false;
// We have invokes, so we need to add register/unregister calls to get this
// function onto the global unwind stack.
//
// First thing we need to do is scan the whole function for values that are
// live across unwind edges. Each value that is live across an unwind edge we
// spill into a stack location, guaranteeing that there is nothing live across
// the unwind edge. This process also splits all critical edges coming out of
// invoke's.
splitLiveRangesAcrossInvokes(Invokes);
SmallVector<LandingPadInst*, 16> LandingPads;
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
if (InvokeInst *II = dyn_cast<InvokeInst>(BB->getTerminator()))
// FIXME: This will be always non-NULL in the new EH.
if (LandingPadInst *LPI = II->getUnwindDest()->getLandingPadInst())
LandingPads.push_back(LPI);
}
BasicBlock *EntryBB = F.begin();
// Create an alloca for the incoming jump buffer ptr and the new jump buffer
// that needs to be restored on all exits from the function. This is an
// alloca because the value needs to be added to the global context list.
unsigned Align = 4; // FIXME: Should be a TLI check?
AllocaInst *FunctionContext =
new AllocaInst(FunctionContextTy, 0, Align,
"fcn_context", F.begin()->begin());
Value *Idxs[2];
Type *Int32Ty = Type::getInt32Ty(F.getContext());
Value *Zero = ConstantInt::get(Int32Ty, 0);
// We need to also keep around a reference to the call_site field
Idxs[0] = Zero;
Idxs[1] = ConstantInt::get(Int32Ty, 1);
CallSite = GetElementPtrInst::Create(FunctionContext, Idxs, "call_site",
EntryBB->getTerminator());
// The exception selector comes back in context->data[1]
Idxs[1] = ConstantInt::get(Int32Ty, 2);
Value *FCData = GetElementPtrInst::Create(FunctionContext, Idxs, "fc_data",
EntryBB->getTerminator());
Idxs[1] = ConstantInt::get(Int32Ty, 1);
Value *SelectorAddr = GetElementPtrInst::Create(FCData, Idxs,
"exc_selector_gep",
EntryBB->getTerminator());
// The exception value comes back in context->data[0]
Idxs[1] = Zero;
Value *ExceptionAddr = GetElementPtrInst::Create(FCData, Idxs,
"exception_gep",
EntryBB->getTerminator());
// The result of the eh.selector call will be replaced with a a reference to
// the selector value returned in the function context. We leave the selector
// itself so the EH analysis later can use it.
for (int i = 0, e = EH_Selectors.size(); i < e; ++i) {
CallInst *I = EH_Selectors[i];
Value *SelectorVal = new LoadInst(SelectorAddr, "select_val", true, I);
I->replaceAllUsesWith(SelectorVal);
}
// eh.exception calls are replaced with references to the proper location in
// the context. Unlike eh.selector, the eh.exception calls are removed
// entirely.
for (int i = 0, e = EH_Exceptions.size(); i < e; ++i) {
CallInst *I = EH_Exceptions[i];
// Possible for there to be duplicates, so check to make sure the
// instruction hasn't already been removed.
if (!I->getParent()) continue;
Value *Val = new LoadInst(ExceptionAddr, "exception", true, I);
Type *Ty = Type::getInt8PtrTy(F.getContext());
Val = CastInst::Create(Instruction::IntToPtr, Val, Ty, "", I);
I->replaceAllUsesWith(Val);
I->eraseFromParent();
}
for (unsigned i = 0, e = LandingPads.size(); i != e; ++i)
ReplaceLandingPadVal(F, LandingPads[i], ExceptionAddr, SelectorAddr);
// The entry block changes to have the eh.sjlj.setjmp, with a conditional
// branch to a dispatch block for non-zero returns. If we return normally,
// we're not handling an exception and just register the function context and
// continue.
// Create the dispatch block. The dispatch block is basically a big switch
// statement that goes to all of the invoke landing pads.
BasicBlock *DispatchBlock =
BasicBlock::Create(F.getContext(), "eh.sjlj.setjmp.catch", &F);
// Insert a load of the callsite in the dispatch block, and a switch on its
// value. By default, we issue a trap statement.
BasicBlock *TrapBlock =
BasicBlock::Create(F.getContext(), "trapbb", &F);
CallInst::Create(Intrinsic::getDeclaration(F.getParent(), Intrinsic::trap),
"", TrapBlock);
new UnreachableInst(F.getContext(), TrapBlock);
Value *DispatchLoad = new LoadInst(CallSite, "invoke.num", true,
DispatchBlock);
SwitchInst *DispatchSwitch =
SwitchInst::Create(DispatchLoad, TrapBlock, Invokes.size(),
DispatchBlock);
// Split the entry block to insert the conditional branch for the setjmp.
BasicBlock *ContBlock = EntryBB->splitBasicBlock(EntryBB->getTerminator(),
"eh.sjlj.setjmp.cont");
// Populate the Function Context
// 1. LSDA address
// 2. Personality function address
// 3. jmpbuf (save SP, FP and call eh.sjlj.setjmp)
// LSDA address
Idxs[0] = Zero;
Idxs[1] = ConstantInt::get(Int32Ty, 4);
Value *LSDAFieldPtr =
GetElementPtrInst::Create(FunctionContext, Idxs, "lsda_gep",
EntryBB->getTerminator());
Value *LSDA = CallInst::Create(LSDAAddrFn, "lsda_addr",
EntryBB->getTerminator());
new StoreInst(LSDA, LSDAFieldPtr, true, EntryBB->getTerminator());
Idxs[1] = ConstantInt::get(Int32Ty, 3);
Value *PersonalityFieldPtr =
GetElementPtrInst::Create(FunctionContext, Idxs, "lsda_gep",
EntryBB->getTerminator());
new StoreInst(PersonalityFn, PersonalityFieldPtr, true,
EntryBB->getTerminator());
// Save the frame pointer.
Idxs[1] = ConstantInt::get(Int32Ty, 5);
Value *JBufPtr
= GetElementPtrInst::Create(FunctionContext, Idxs, "jbuf_gep",
EntryBB->getTerminator());
Idxs[1] = ConstantInt::get(Int32Ty, 0);
Value *FramePtr =
GetElementPtrInst::Create(JBufPtr, Idxs, "jbuf_fp_gep",
EntryBB->getTerminator());
Value *Val = CallInst::Create(FrameAddrFn,
ConstantInt::get(Int32Ty, 0),
"fp",
EntryBB->getTerminator());
new StoreInst(Val, FramePtr, true, EntryBB->getTerminator());
// Save the stack pointer.
Idxs[1] = ConstantInt::get(Int32Ty, 2);
Value *StackPtr =
GetElementPtrInst::Create(JBufPtr, Idxs, "jbuf_sp_gep",
EntryBB->getTerminator());
Val = CallInst::Create(StackAddrFn, "sp", EntryBB->getTerminator());
new StoreInst(Val, StackPtr, true, EntryBB->getTerminator());
// Call the setjmp instrinsic. It fills in the rest of the jmpbuf.
Value *SetjmpArg =
CastInst::Create(Instruction::BitCast, JBufPtr,
Type::getInt8PtrTy(F.getContext()), "",
EntryBB->getTerminator());
Value *DispatchVal = CallInst::Create(BuiltinSetjmpFn, SetjmpArg,
"dispatch",
EntryBB->getTerminator());
// Add a call to dispatch_setup after the setjmp call. This is expanded to any
// target-specific setup that needs to be done.
CallInst::Create(DispatchSetupFn, DispatchVal, "", EntryBB->getTerminator());
// check the return value of the setjmp. non-zero goes to dispatcher.
Value *IsNormal = new ICmpInst(EntryBB->getTerminator(),
ICmpInst::ICMP_EQ, DispatchVal, Zero,
"notunwind");
// Nuke the uncond branch.
EntryBB->getTerminator()->eraseFromParent();
// Put in a new condbranch in its place.
BranchInst::Create(ContBlock, DispatchBlock, IsNormal, EntryBB);
// Register the function context and make sure it's known to not throw
CallInst *Register =
CallInst::Create(RegisterFn, FunctionContext, "",
ContBlock->getTerminator());
Register->setDoesNotThrow();
// At this point, we are all set up, update the invoke instructions to mark
// their call_site values, and fill in the dispatch switch accordingly.
for (unsigned i = 0, e = Invokes.size(); i != e; ++i)
markInvokeCallSite(Invokes[i], i+1, CallSite, DispatchSwitch);
// Mark call instructions that aren't nounwind as no-action (call_site ==
// -1). Skip the entry block, as prior to then, no function context has been
// created for this function and any unexpected exceptions thrown will go
// directly to the caller's context, which is what we want anyway, so no need
// to do anything here.
for (Function::iterator BB = F.begin(), E = F.end(); ++BB != E;) {
for (BasicBlock::iterator I = BB->begin(), end = BB->end(); I != end; ++I)
if (CallInst *CI = dyn_cast<CallInst>(I)) {
// Ignore calls to the EH builtins (eh.selector, eh.exception)
Constant *Callee = CI->getCalledFunction();
if (Callee != SelectorFn && Callee != ExceptionFn
&& !CI->doesNotThrow())
insertCallSiteStore(CI, -1, CallSite);
} else if (ResumeInst *RI = dyn_cast<ResumeInst>(I)) {
insertCallSiteStore(RI, -1, CallSite);
}
}
// Replace all unwinds with a branch to the unwind handler.
// ??? Should this ever happen with sjlj exceptions?
for (unsigned i = 0, e = Unwinds.size(); i != e; ++i) {
BranchInst::Create(TrapBlock, Unwinds[i]);
Unwinds[i]->eraseFromParent();
}
// Following any allocas not in the entry block, update the saved SP in the
// jmpbuf to the new value.
for (unsigned i = 0, e = JmpbufUpdatePoints.size(); i != e; ++i) {
Instruction *AI = JmpbufUpdatePoints[i];
Instruction *StackAddr = CallInst::Create(StackAddrFn, "sp");
StackAddr->insertAfter(AI);
Instruction *StoreStackAddr = new StoreInst(StackAddr, StackPtr, true);
StoreStackAddr->insertAfter(StackAddr);
}
// Finally, for any returns from this function, if this function contains an
// invoke, add a call to unregister the function context.
for (unsigned i = 0, e = Returns.size(); i != e; ++i)
CallInst::Create(UnregisterFn, FunctionContext, "", Returns[i]);
return true;
}
/// InputFilename is a LLVM bitcode file. Read it using bitcode reader.
/// Collect global functions and symbol names in symbols vector.
/// Collect external references in references vector.
/// Return LTO_READ_SUCCESS if there is no error.
enum LTOStatus
LTO::readLLVMObjectFile(const std::string &InputFilename,
NameToSymbolMap &symbols,
std::set<std::string> &references)
{
Module *m = getModule(InputFilename);
if (!m)
return LTO_READ_FAILURE;
// Collect Target info
getTarget(m);
if (!Target)
return LTO_READ_FAILURE;
// Use mangler to add GlobalPrefix to names to match linker names.
// FIXME : Instead of hard coding "-" use GlobalPrefix.
Mangler mangler(*m, Target->getTargetAsmInfo()->getGlobalPrefix());
modules.push_back(m);
for (Module::iterator f = m->begin(), e = m->end(); f != e; ++f) {
LTOLinkageTypes lt = getLTOLinkageType(f);
LTOVisibilityTypes vis = getLTOVisibilityType(f);
if (!f->isDeclaration() && lt != LTOInternalLinkage
&& strncmp (f->getName().c_str(), "llvm.", 5)) {
int alignment = ( 16 > f->getAlignment() ? 16 : f->getAlignment());
LLVMSymbol *newSymbol = new LLVMSymbol(lt, vis, f, f->getName(),
mangler.getValueName(f),
Log2_32(alignment));
symbols[newSymbol->getMangledName()] = newSymbol;
allSymbols[newSymbol->getMangledName()] = newSymbol;
}
// Collect external symbols referenced by this function.
for (Function::iterator b = f->begin(), fe = f->end(); b != fe; ++b)
for (BasicBlock::iterator i = b->begin(), be = b->end();
i != be; ++i) {
for (unsigned count = 0, total = i->getNumOperands();
count != total; ++count)
findExternalRefs(i->getOperand(count), references, mangler);
}
}
for (Module::global_iterator v = m->global_begin(), e = m->global_end();
v != e; ++v) {
LTOLinkageTypes lt = getLTOLinkageType(v);
LTOVisibilityTypes vis = getLTOVisibilityType(v);
if (!v->isDeclaration() && lt != LTOInternalLinkage
&& strncmp (v->getName().c_str(), "llvm.", 5)) {
const TargetData *TD = Target->getTargetData();
LLVMSymbol *newSymbol = new LLVMSymbol(lt, vis, v, v->getName(),
mangler.getValueName(v),
TD->getPreferredAlignmentLog(v));
symbols[newSymbol->getMangledName()] = newSymbol;
allSymbols[newSymbol->getMangledName()] = newSymbol;
for (unsigned count = 0, total = v->getNumOperands();
count != total; ++count)
findExternalRefs(v->getOperand(count), references, mangler);
}
}
return LTO_READ_SUCCESS;
}
bool PruneEH::runOnSCC(CallGraphSCC &SCC) {
SmallPtrSet<CallGraphNode *, 8> SCCNodes;
CallGraph &CG = getAnalysis<CallGraph>();
bool MadeChange = false;
// Fill SCCNodes with the elements of the SCC. Used for quickly
// looking up whether a given CallGraphNode is in this SCC.
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I)
SCCNodes.insert(*I);
// First pass, scan all of the functions in the SCC, simplifying them
// according to what we know.
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I)
if (Function *F = (*I)->getFunction())
MadeChange |= SimplifyFunction(F);
// Next, check to see if any callees might throw or if there are any external
// functions in this SCC: if so, we cannot prune any functions in this SCC.
// Definitions that are weak and not declared non-throwing might be
// overridden at linktime with something that throws, so assume that.
// If this SCC includes the unwind instruction, we KNOW it throws, so
// obviously the SCC might throw.
//
bool SCCMightUnwind = false, SCCMightReturn = false;
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end();
(!SCCMightUnwind || !SCCMightReturn) && I != E; ++I) {
Function *F = (*I)->getFunction();
if (F == 0) {
SCCMightUnwind = true;
SCCMightReturn = true;
} else if (F->isDeclaration() || F->mayBeOverridden()) {
SCCMightUnwind |= !F->doesNotThrow();
SCCMightReturn |= !F->doesNotReturn();
} else {
bool CheckUnwind = !SCCMightUnwind && !F->doesNotThrow();
bool CheckReturn = !SCCMightReturn && !F->doesNotReturn();
if (!CheckUnwind && !CheckReturn)
continue;
// Check to see if this function performs an unwind or calls an
// unwinding function.
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
if (CheckUnwind && isa<UnwindInst>(BB->getTerminator())) {
// Uses unwind!
SCCMightUnwind = true;
} else if (CheckReturn && isa<ReturnInst>(BB->getTerminator())) {
SCCMightReturn = true;
}
// Invoke instructions don't allow unwinding to continue, so we are
// only interested in call instructions.
if (CheckUnwind && !SCCMightUnwind)
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
if (CallInst *CI = dyn_cast<CallInst>(I)) {
if (CI->doesNotThrow()) {
// This call cannot throw.
} else if (Function *Callee = CI->getCalledFunction()) {
CallGraphNode *CalleeNode = CG[Callee];
// If the callee is outside our current SCC then we may
// throw because it might.
if (!SCCNodes.count(CalleeNode)) {
SCCMightUnwind = true;
break;
}
} else {
// Indirect call, it might throw.
SCCMightUnwind = true;
break;
}
}
if (SCCMightUnwind && SCCMightReturn) break;
}
}
}
// If the SCC doesn't unwind or doesn't throw, note this fact.
if (!SCCMightUnwind || !SCCMightReturn)
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) {
Attributes NewAttributes = Attribute::None;
if (!SCCMightUnwind)
NewAttributes |= Attribute::NoUnwind;
if (!SCCMightReturn)
NewAttributes |= Attribute::NoReturn;
Function *F = (*I)->getFunction();
const AttrListPtr &PAL = F->getAttributes();
const AttrListPtr &NPAL = PAL.addAttr(~0, NewAttributes);
if (PAL != NPAL) {
MadeChange = true;
F->setAttributes(NPAL);
}
}
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) {
// Convert any invoke instructions to non-throwing functions in this node
// into call instructions with a branch. This makes the exception blocks
// dead.
if (Function *F = (*I)->getFunction())
MadeChange |= SimplifyFunction(F);
}
return MadeChange;
}
/*
* Clone a given function removing dead stores
*/
Function* DeadStoreEliminationPass::cloneFunctionWithoutDeadStore(Function *Fn,
Instruction* caller, std::string suffix) {
Function *NF = Function::Create(Fn->getFunctionType(), Fn->getLinkage());
NF->copyAttributesFrom(Fn);
// Copy the parameter names, to ease function inspection afterwards.
Function::arg_iterator NFArg = NF->arg_begin();
for (Function::arg_iterator Arg = Fn->arg_begin(), ArgEnd = Fn->arg_end();
Arg != ArgEnd; ++Arg, ++NFArg) {
NFArg->setName(Arg->getName());
}
// To avoid name collision, we should select another name.
NF->setName(Fn->getName() + suffix);
// Fill clone content
ValueToValueMapTy VMap;
SmallVector<ReturnInst*, 8> Returns;
Function::arg_iterator NI = NF->arg_begin();
for (Function::arg_iterator I = Fn->arg_begin();
NI != NF->arg_end(); ++I, ++NI) {
VMap[I] = NI;
}
CloneAndPruneFunctionInto(NF, Fn, VMap, false, Returns);
// Remove dead stores
std::set<Value*> deadArgs = deadArguments[caller];
std::set<Value*> removeStoresTo;
Function::arg_iterator NFArgIter = NF->arg_begin();
for (Function::arg_iterator FnArgIter = Fn->arg_begin(); FnArgIter !=
Fn->arg_end(); ++FnArgIter, ++NFArgIter) {
Value *FnArg = FnArgIter;
if (deadArgs.count(FnArg)) {
removeStoresTo.insert(NFArgIter);
}
}
std::vector<Instruction*> toRemove;
for (Function::iterator BB = NF->begin(); BB != NF->end(); ++BB) {
for (BasicBlock::iterator I = BB->begin(); I != BB->end(); ++I) {
Instruction *inst = I;
if (!isa<StoreInst>(inst)) continue;
StoreInst *SI = dyn_cast<StoreInst>(inst);
Value *ptrOp = SI->getPointerOperand();
if (removeStoresTo.count(ptrOp)) {
DEBUG(errs() << "will remove this store: " << *inst << "\n");
toRemove.push_back(inst);
}
}
}
for (std::vector<Instruction*>::iterator it = toRemove.begin();
it != toRemove.end(); ++it) {
Instruction* inst = *it;
inst->eraseFromParent();
RemovedStores++;
}
// Insert the clone function before the original
Fn->getParent()->getFunctionList().insert(Fn, NF);
return NF;
}
bool TailCallElim::runOnFunction(Function &F) {
// If this function is a varargs function, we won't be able to PHI the args
// right, so don't even try to convert it...
if (F.getFunctionType()->isVarArg()) return false;
TTI = &getAnalysis<TargetTransformInfo>();
BasicBlock *OldEntry = 0;
bool TailCallsAreMarkedTail = false;
SmallVector<PHINode*, 8> ArgumentPHIs;
bool MadeChange = false;
// CanTRETailMarkedCall - If false, we cannot perform TRE on tail calls
// marked with the 'tail' attribute, because doing so would cause the stack
// size to increase (real TRE would deallocate variable sized allocas, TRE
// doesn't).
bool CanTRETailMarkedCall = true;
// Find calls that can be marked tail.
AllocaCaptureTracker ACT;
for (Function::iterator BB = F.begin(), EE = F.end(); BB != EE; ++BB) {
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
if (AllocaInst *AI = dyn_cast<AllocaInst>(I)) {
CanTRETailMarkedCall &= CanTRE(AI);
PointerMayBeCaptured(AI, &ACT);
// If any allocas are captured, exit.
if (ACT.Captured)
return false;
}
}
}
// Second pass, change any tail recursive calls to loops.
//
// FIXME: The code generator produces really bad code when an 'escaping
// alloca' is changed from being a static alloca to being a dynamic alloca.
// Until this is resolved, disable this transformation if that would ever
// happen. This bug is PR962.
if (ACT.UsesAlloca.empty()) {
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
if (ReturnInst *Ret = dyn_cast<ReturnInst>(BB->getTerminator())) {
bool Change = ProcessReturningBlock(Ret, OldEntry, TailCallsAreMarkedTail,
ArgumentPHIs, !CanTRETailMarkedCall);
if (!Change && BB->getFirstNonPHIOrDbg() == Ret)
Change = FoldReturnAndProcessPred(BB, Ret, OldEntry,
TailCallsAreMarkedTail, ArgumentPHIs,
!CanTRETailMarkedCall);
MadeChange |= Change;
}
}
}
// If we eliminated any tail recursions, it's possible that we inserted some
// silly PHI nodes which just merge an initial value (the incoming operand)
// with themselves. Check to see if we did and clean up our mess if so. This
// occurs when a function passes an argument straight through to its tail
// call.
if (!ArgumentPHIs.empty()) {
for (unsigned i = 0, e = ArgumentPHIs.size(); i != e; ++i) {
PHINode *PN = ArgumentPHIs[i];
// If the PHI Node is a dynamic constant, replace it with the value it is.
if (Value *PNV = SimplifyInstruction(PN)) {
PN->replaceAllUsesWith(PNV);
PN->eraseFromParent();
}
}
}
// At this point, we know that the function does not have any captured
// allocas. If additionally the function does not call setjmp, mark all calls
// in the function that do not access stack memory with the tail keyword. This
// implies ensuring that there does not exist any path from a call that takes
// in an alloca but does not capture it and the call which we wish to mark
// with "tail".
if (!F.callsFunctionThatReturnsTwice()) {
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
if (CallInst *CI = dyn_cast<CallInst>(I)) {
if (!ACT.UsesAlloca.count(CI)) {
CI->setTailCall();
MadeChange = true;
}
}
}
}
}
return MadeChange;
}
/**
* Generate code for
*/
void HeteroOMPTransform::gen_code_per_f (Function* NF, Function* F, Instruction *max_threads){
Function::arg_iterator FI = F->arg_begin();
Argument *ctxname = &*FI;
Function::arg_iterator DestI = NF->arg_begin();
DestI->setName(ctxname->getName());
Argument *ctx_name = &(*DestI);
DestI++;
DestI->setName("tid");
Argument *num_iters = &(*DestI);
#ifdef EXPLICIT_REWRITE
DenseMap<const Value*, Value *> ValueMap;
#else
ValueToValueMapTy ValueMap;
#endif
//get the old basic block and create a new one
Function::const_iterator BI = F->begin();
const BasicBlock &FB = *BI;
BasicBlock *NFBB = BasicBlock::Create(FB.getContext(), "", NF);
if (FB.hasName()){
NFBB->setName(FB.getName());
}
ValueMap[&FB] = NFBB;
//ValueMap[numiters] = num_iters;
ValueMap[ctxname] = ctx_name;
#if EXPLICIT_REWRITE
for (BasicBlock::const_iterator II = FB.begin(), IE = FB.end(); II != IE; ++II) {
Instruction *NFInst = II->clone(/*F->getContext()*/);
// DEBUG(dbgs()<<*II<<"\n");
if (II->hasName()) NFInst->setName(II->getName());
const Instruction *FInst = &(*II);
rewrite_instruction((Instruction *)FInst, NFInst, ValueMap);
NFBB->getInstList().push_back(NFInst);
ValueMap[II] = NFInst;
}
BI++;
for (Function::const_iterator /*BI=F->begin(),*/BE = F->end();BI != BE; ++BI) {
const BasicBlock &FBB = *BI;
BasicBlock *NFBB = BasicBlock::Create(FBB.getContext(), "", NF);
ValueMap[&FBB] = NFBB;
if (FBB.hasName()){
NFBB->setName(FBB.getName());
//DEBUG(dbgs()<<NFBB->getName()<<"\n");
}
for (BasicBlock::const_iterator II = FBB.begin(), IE = FBB.end(); II != IE; ++II) {
Instruction *NFInst = II->clone(/*F->getContext()*/);
if (II->hasName()) NFInst->setName(II->getName());
const Instruction *FInst = &(*II);
rewrite_instruction((Instruction *)FInst, NFInst, ValueMap);
NFBB->getInstList().push_back(NFInst);
ValueMap[II] = NFInst;
}
}
// Remap the instructions again to take care of forward jumps
for (Function::iterator BB = NF->begin(), BE=NF->end(); BB != BE; ++ BB) {
for (BasicBlock::iterator II = BB->begin(); II != BB->end(); ++II){
int opIdx = 0;
//DEBUG(dbgs()<<*II<<"\n");
for (User::op_iterator i = II->op_begin(), e = II->op_end(); i != e; ++i, opIdx++) {
Value *V = *i;
if (ValueMap[V] != NULL) {
II->setOperand(opIdx, ValueMap[V]);
}
}
}
}
#else
SmallVector<ReturnInst*, 8> Returns; // Ignore returns cloned.
CloneFunctionInto(NF, F, ValueMap, false, Returns, "");
#endif
//max_threads->dump();
/* Remap openmp omp_num_threads() and omp_thread_num() */
/*
* define internal void @_Z20initialize_variablesiPfS_.omp_fn.4(i8* nocapture %.omp_data_i) nounwind ssp {
* entry:
* %0 = bitcast i8* %.omp_data_i to i32* ; <i32*> [#uses=1]
* %1 = load i32* %0, align 8 ; <i32> [#uses=2]
* %2 = tail call i32 @omp_get_num_threads() nounwind readnone ; <i32> [#uses=2]
* %3 = tail call i32 @omp_get_thread_num() nounwind readnone ; <i32> [#uses=2]
%4 = sdiv i32 %1, %2
%5 = mul nsw i32 %4, %2
%6 = icmp ne i32 %5, %1
%7 = zext i1 %6 to i32
*/
vector<Instruction *> toDelete;
for (Function::iterator BB = NF->begin(), BE=NF->end(); BB != BE; ++ BB) {
for (BasicBlock::iterator II = BB->begin(); II != BB->end(); ++II){
if (isa<CallInst>(II)) {
CallSite CI(cast<Instruction>(II));
if (CI.getCalledFunction() != NULL){
string called_func_name = CI.getCalledFunction()->getName();
if (called_func_name == OMP_GET_NUM_THREADS_NAME && CI.arg_size() == 0) {
II->replaceAllUsesWith(ValueMap[max_threads]);
toDelete.push_back(II);
}
else if (called_func_name == OMP_GET_THREAD_NUM_NAME && CI.arg_size() == 0) {
II->replaceAllUsesWith(num_iters);
toDelete.push_back(II);
}
}
}
}
}
/* Delete the last branch instruction of the first basic block -- Assuming it is safe */
Function::iterator nfBB = NF->begin();
TerminatorInst *lastI = nfBB->getTerminator();
BranchInst *bI;
BasicBlock *returnBlock;
if ((bI = dyn_cast<BranchInst>(lastI)) && bI->isConditional() &&
(returnBlock = bI->getSuccessor(1)) &&
(returnBlock->getName() == "return")) {
/* modify to a unconditional branch to next basic block and not return */
Instruction *bbI = BranchInst::Create(bI->getSuccessor(0),lastI);
bbI->dump();
toDelete.push_back(lastI);
}
//NF->dump();
while(!toDelete.empty()) {
Instruction *g = toDelete.back();
//g->replaceAllUsesWith(UndefValue::get(g->getType()));
toDelete.pop_back();
g->eraseFromParent();
}
//NF->dump();
}
//
// Method: runOnModule()
//
// Description:
// Entry point for this LLVM pass. Search for insert/extractvalue instructions
// that can be simplified.
//
// Inputs:
// M - A reference to the LLVM module to transform.
//
// Outputs:
// M - The transformed LLVM module.
//
// Return value:
// true - The module was modified.
// false - The module was not modified.
//
bool SimplifyEV::runOnModule(Module& M) {
// Repeat till no change
bool changed;
do {
changed = false;
for (Module::iterator F = M.begin(); F != M.end(); ++F) {
for (Function::iterator B = F->begin(), FE = F->end(); B != FE; ++B) {
for (BasicBlock::iterator I = B->begin(), BE = B->end(); I != BE;) {
ExtractValueInst *EV = dyn_cast<ExtractValueInst>(I++);
if(!EV)
continue;
Value *Agg = EV->getAggregateOperand();
if (!EV->hasIndices()) {
EV->replaceAllUsesWith(Agg);
DEBUG(errs() << "EV:");
DEBUG(errs() << "ERASE:");
DEBUG(EV->dump());
EV->eraseFromParent();
numErased++;
changed = true;
continue;
}
if (Constant *C = dyn_cast<Constant>(Agg)) {
if (isa<UndefValue>(C)) {
EV->replaceAllUsesWith(UndefValue::get(EV->getType()));
DEBUG(errs() << "EV:");
DEBUG(errs() << "ERASE:");
DEBUG(EV->dump());
EV->eraseFromParent();
numErased++;
changed = true;
continue;
}
if (isa<ConstantAggregateZero>(C)) {
EV->replaceAllUsesWith(Constant::getNullValue(EV->getType()));
DEBUG(errs() << "EV:");
DEBUG(errs() << "ERASE:");
DEBUG(EV->dump());
EV->eraseFromParent();
numErased++;
changed = true;
continue;
}
if (isa<ConstantArray>(C) || isa<ConstantStruct>(C)) {
// Extract the element indexed by the first index out of the constant
Value *V = C->getOperand(*EV->idx_begin());
if (EV->getNumIndices() > 1) {
// Extract the remaining indices out of the constant indexed by the
// first index
ExtractValueInst *EV_new = ExtractValueInst::Create(V,
EV->getIndices().slice(1),
"", EV);
EV->replaceAllUsesWith(EV_new);
DEBUG(errs() << "EV:");
DEBUG(errs() << "ERASE:");
DEBUG(EV->dump());
EV->eraseFromParent();
numErased++;
changed = true;
continue;
} else {
EV->replaceAllUsesWith(V);
DEBUG(errs() << "EV:");
DEBUG(errs() << "ERASE:");
DEBUG(EV->dump());
EV->eraseFromParent();
numErased++;
changed = true;
continue;
}
}
continue;
}
if (LoadInst * LI = dyn_cast<LoadInst>(Agg)) {
// if the Agg value came from a load instruction
// replace the extract value intruction with
// a gep and a load.
SmallVector<Value*, 8> Indices;
Type *Int32Ty = Type::getInt32Ty(M.getContext());
Indices.push_back(Constant::getNullValue(Int32Ty));
for (ExtractValueInst::idx_iterator I = EV->idx_begin(), E = EV->idx_end();
I != E; ++I) {
Indices.push_back(ConstantInt::get(Int32Ty, *I));
}
GetElementPtrInst *GEP = GetElementPtrInst::CreateInBounds(LI->getOperand(0), Indices,
LI->getName(), LI) ;
LoadInst *LINew = new LoadInst(GEP, "", LI);
EV->replaceAllUsesWith(LINew);
EV->eraseFromParent();
changed = true;
numErased++;
continue;
}
if (InsertValueInst *IV = dyn_cast<InsertValueInst>(Agg)) {
bool done = false;
// We're extracting from an insertvalue instruction, compare the indices
const unsigned *exti, *exte, *insi, *inse;
for (exti = EV->idx_begin(), insi = IV->idx_begin(),
exte = EV->idx_end(), inse = IV->idx_end();
exti != exte && insi != inse;
++exti, ++insi) {
if (*insi != *exti) {
// The insert and extract both reference distinctly different elements.
// This means the extract is not influenced by the insert, and we can
// replace the aggregate operand of the extract with the aggregate
// operand of the insert. i.e., replace
// %I = insertvalue { i32, { i32 } } %A, { i32 } { i32 42 }, 1
// %E = extractvalue { i32, { i32 } } %I, 0
// with
// %E = extractvalue { i32, { i32 } } %A, 0
ExtractValueInst *EV_new = ExtractValueInst::Create(IV->getAggregateOperand(),
EV->getIndices(), "", EV);
EV->replaceAllUsesWith(EV_new);
DEBUG(errs() << "EV:");
DEBUG(errs() << "ERASE:");
DEBUG(EV->dump());
EV->eraseFromParent();
numErased++;
done = true;
changed = true;
break;
}
}
if(done)
continue;
if (exti == exte && insi == inse) {
// Both iterators are at the end: Index lists are identical. Replace
// %B = insertvalue { i32, { i32 } } %A, i32 42, 1, 0
// %C = extractvalue { i32, { i32 } } %B, 1, 0
// with "i32 42"
EV->replaceAllUsesWith(IV->getInsertedValueOperand());
DEBUG(errs() << "EV:");
DEBUG(errs() << "ERASE:");
DEBUG(EV->dump());
EV->eraseFromParent();
numErased++;
changed = true;
continue;
}
if (exti == exte) {
// The extract list is a prefix of the insert list. i.e. replace
// %I = insertvalue { i32, { i32 } } %A, i32 42, 1, 0
// %E = extractvalue { i32, { i32 } } %I, 1
// with
// %X = extractvalue { i32, { i32 } } %A, 1
// %E = insertvalue { i32 } %X, i32 42, 0
// by switching the order of the insert and extract (though the
// insertvalue should be left in, since it may have other uses).
Value *NewEV = ExtractValueInst::Create(IV->getAggregateOperand(),
EV->getIndices(), "", EV);
Value *NewIV = InsertValueInst::Create(NewEV, IV->getInsertedValueOperand(),
makeArrayRef(insi, inse), "", EV);
EV->replaceAllUsesWith(NewIV);
DEBUG(errs() << "EV:");
DEBUG(errs() << "ERASE:");
DEBUG(EV->dump());
EV->eraseFromParent();
numErased++;
changed = true;
continue;
}
if (insi == inse) {
// The insert list is a prefix of the extract list
// We can simply remove the common indices from the extract and make it
// operate on the inserted value instead of the insertvalue result.
// i.e., replace
// %I = insertvalue { i32, { i32 } } %A, { i32 } { i32 42 }, 1
// %E = extractvalue { i32, { i32 } } %I, 1, 0
// with
// %E extractvalue { i32 } { i32 42 }, 0
ExtractValueInst *EV_new = ExtractValueInst::Create(IV->getInsertedValueOperand(),
makeArrayRef(exti, exte), "", EV);
EV->replaceAllUsesWith(EV_new);
DEBUG(errs() << "EV:");
DEBUG(errs() << "ERASE:");
DEBUG(EV->dump());
EV->eraseFromParent();
numErased++;
changed = true;
continue;
}
}
}
}
}
} while(changed);
return (numErased > 0);
}
// insertFastDiv - Substitutes the div/rem instruction with code that checks the
// value of the operands and uses a shorter-faster div/rem instruction when
// possible and the longer-slower div/rem instruction otherwise.
static bool insertFastDiv(Function &F,
Function::iterator &I,
BasicBlock::iterator &J,
IntegerType *BypassType,
bool UseDivOp,
bool UseSignedOp,
DivCacheTy &PerBBDivCache) {
// Get instruction operands
Instruction *Instr = &*J;
Value *Dividend = Instr->getOperand(0);
Value *Divisor = Instr->getOperand(1);
if (isa<ConstantInt>(Divisor) ||
(isa<ConstantInt>(Dividend) && isa<ConstantInt>(Divisor))) {
// Operations with immediate values should have
// been solved and replaced during compile time.
return false;
}
// Basic Block is split before divide
BasicBlock *MainBB = &*I;
BasicBlock *SuccessorBB = I->splitBasicBlock(J);
++I; //advance iterator I to successorBB
// Add new basic block for slow divide operation
BasicBlock *SlowBB = BasicBlock::Create(F.getContext(), "",
MainBB->getParent(), SuccessorBB);
SlowBB->moveBefore(SuccessorBB);
IRBuilder<> SlowBuilder(SlowBB, SlowBB->begin());
Value *SlowQuotientV;
Value *SlowRemainderV;
if (UseSignedOp) {
SlowQuotientV = SlowBuilder.CreateSDiv(Dividend, Divisor);
SlowRemainderV = SlowBuilder.CreateSRem(Dividend, Divisor);
} else {
SlowQuotientV = SlowBuilder.CreateUDiv(Dividend, Divisor);
SlowRemainderV = SlowBuilder.CreateURem(Dividend, Divisor);
}
SlowBuilder.CreateBr(SuccessorBB);
// Add new basic block for fast divide operation
BasicBlock *FastBB = BasicBlock::Create(F.getContext(), "",
MainBB->getParent(), SuccessorBB);
FastBB->moveBefore(SlowBB);
IRBuilder<> FastBuilder(FastBB, FastBB->begin());
Value *ShortDivisorV = FastBuilder.CreateCast(Instruction::Trunc, Divisor,
BypassType);
Value *ShortDividendV = FastBuilder.CreateCast(Instruction::Trunc, Dividend,
BypassType);
// udiv/urem because optimization only handles positive numbers
Value *ShortQuotientV = FastBuilder.CreateExactUDiv(ShortDividendV,
ShortDivisorV);
Value *ShortRemainderV = FastBuilder.CreateURem(ShortDividendV,
ShortDivisorV);
Value *FastQuotientV = FastBuilder.CreateCast(Instruction::ZExt,
ShortQuotientV,
Dividend->getType());
Value *FastRemainderV = FastBuilder.CreateCast(Instruction::ZExt,
ShortRemainderV,
Dividend->getType());
FastBuilder.CreateBr(SuccessorBB);
// Phi nodes for result of div and rem
IRBuilder<> SuccessorBuilder(SuccessorBB, SuccessorBB->begin());
PHINode *QuoPhi = SuccessorBuilder.CreatePHI(Instr->getType(), 2);
QuoPhi->addIncoming(SlowQuotientV, SlowBB);
QuoPhi->addIncoming(FastQuotientV, FastBB);
PHINode *RemPhi = SuccessorBuilder.CreatePHI(Instr->getType(), 2);
RemPhi->addIncoming(SlowRemainderV, SlowBB);
RemPhi->addIncoming(FastRemainderV, FastBB);
// Replace Instr with appropriate phi node
if (UseDivOp)
Instr->replaceAllUsesWith(QuoPhi);
else
Instr->replaceAllUsesWith(RemPhi);
Instr->eraseFromParent();
// Combine operands into a single value with OR for value testing below
MainBB->getInstList().back().eraseFromParent();
IRBuilder<> MainBuilder(MainBB, MainBB->end());
Value *OrV = MainBuilder.CreateOr(Dividend, Divisor);
// BitMask is inverted to check if the operands are
// larger than the bypass type
uint64_t BitMask = ~BypassType->getBitMask();
Value *AndV = MainBuilder.CreateAnd(OrV, BitMask);
// Compare operand values and branch
Value *ZeroV = ConstantInt::getSigned(Dividend->getType(), 0);
Value *CmpV = MainBuilder.CreateICmpEQ(AndV, ZeroV);
MainBuilder.CreateCondBr(CmpV, FastBB, SlowBB);
// point iterator J at first instruction of successorBB
J = I->begin();
// Cache phi nodes to be used later in place of other instances
// of div or rem with the same sign, dividend, and divisor
DivOpInfo Key(UseSignedOp, Dividend, Divisor);
DivPhiNodes Value(QuoPhi, RemPhi);
PerBBDivCache.insert(std::pair<DivOpInfo, DivPhiNodes>(Key, Value));
return true;
}
void StatsTracker::computeReachableUncovered() {
KModule *km = executor.kmodule;
Module *m = km->module;
static bool init = true;
const InstructionInfoTable &infos = *km->infos;
StatisticManager &sm = *theStatisticManager;
if (init) {
init = false;
// Compute call targets. It would be nice to use alias information
// instead of assuming all indirect calls hit all escaping
// functions, eh?
for (Module::iterator fnIt = m->begin(), fn_ie = m->end();
fnIt != fn_ie; ++fnIt) {
for (Function::iterator bbIt = fnIt->begin(), bb_ie = fnIt->end();
bbIt != bb_ie; ++bbIt) {
for (BasicBlock::iterator it = bbIt->begin(), ie = bbIt->end();
it != ie; ++it) {
if (isa<CallInst>(it) || isa<InvokeInst>(it)) {
CallSite cs(it);
if (isa<InlineAsm>(cs.getCalledValue())) {
// We can never call through here so assume no targets
// (which should be correct anyhow).
callTargets.insert(std::make_pair(it,
std::vector<Function*>()));
} else if (Function *target = getDirectCallTarget(cs)) {
callTargets[it].push_back(target);
} else {
/* disable the escape analysis since it overapproximates
* too much the possible aliases
callTargets[it] =
std::vector<Function*>(km->escapingFunctions.begin(),
km->escapingFunctions.end());
*/
}
}
}
}
}
// Compute function callers as reflexion of callTargets.
for (calltargets_ty::iterator it = callTargets.begin(),
ie = callTargets.end(); it != ie; ++it)
for (std::vector<Function*>::iterator fit = it->second.begin(),
fie = it->second.end(); fit != fie; ++fit)
functionCallers[*fit].push_back(it->first);
// Initialize minDistToReturn to shortest paths through
// functions. 0 is unreachable.
std::vector<Instruction *> instructions;
for (Module::iterator fnIt = m->begin(), fn_ie = m->end();
fnIt != fn_ie; ++fnIt) {
if (fnIt->isDeclaration()) {
if (fnIt->doesNotReturn()) {
functionShortestPath[fnIt] = 0;
} else {
functionShortestPath[fnIt] = 1; // whatever
}
} else {
functionShortestPath[fnIt] = 0;
}
// Not sure if I should bother to preorder here. XXX I should.
for (Function::iterator bbIt = fnIt->begin(), bb_ie = fnIt->end();
bbIt != bb_ie; ++bbIt) {
for (BasicBlock::iterator it = bbIt->begin(), ie = bbIt->end();
it != ie; ++it) {
instructions.push_back(it);
unsigned id = infos.getInfo(it).id;
sm.setIndexedValue(stats::minDistToReturn,
id,
isa<ReturnInst>(it)
#if LLVM_VERSION_CODE < LLVM_VERSION(3, 1)
|| isa<UnwindInst>(it)
#endif
);
}
}
}
std::reverse(instructions.begin(), instructions.end());
// I'm so lazy it's not even worklisted.
bool changed;
do {
changed = false;
for (std::vector<Instruction*>::iterator it = instructions.begin(),
ie = instructions.end(); it != ie; ++it) {
Instruction *inst = *it;
unsigned bestThrough = 0;
if (isa<CallInst>(inst) || isa<InvokeInst>(inst)) {
std::vector<Function*> &targets = callTargets[inst];
for (std::vector<Function*>::iterator fnIt = targets.begin(),
ie = targets.end(); fnIt != ie; ++fnIt) {
uint64_t dist = functionShortestPath[*fnIt];
if (dist) {
dist = 1+dist; // count instruction itself
if (bestThrough==0 || dist<bestThrough)
bestThrough = dist;
}
}
} else {
bestThrough = 1;
}
if (bestThrough) {
unsigned id = infos.getInfo(*it).id;
uint64_t best, cur = best = sm.getIndexedValue(stats::minDistToReturn, id);
std::vector<Instruction*> succs = getSuccs(*it);
for (std::vector<Instruction*>::iterator it2 = succs.begin(),
ie = succs.end(); it2 != ie; ++it2) {
uint64_t dist = sm.getIndexedValue(stats::minDistToReturn,
infos.getInfo(*it2).id);
if (dist) {
uint64_t val = bestThrough + dist;
if (best==0 || val<best)
best = val;
}
}
// there's a corner case here when a function only includes a single
// instruction (a ret). in that case, we MUST update
// functionShortestPath, or it will remain 0 (erroneously indicating
// that no return instructions are reachable)
Function *f = inst->getParent()->getParent();
if (best != cur
|| (inst == f->begin()->begin()
&& functionShortestPath[f] != best)) {
sm.setIndexedValue(stats::minDistToReturn, id, best);
changed = true;
// Update shortest path if this is the entry point.
if (inst==f->begin()->begin())
functionShortestPath[f] = best;
}
}
}
} while (changed);
}
// compute minDistToUncovered, 0 is unreachable
std::vector<Instruction *> instructions;
for (Module::iterator fnIt = m->begin(), fn_ie = m->end();
fnIt != fn_ie; ++fnIt) {
// Not sure if I should bother to preorder here.
for (Function::iterator bbIt = fnIt->begin(), bb_ie = fnIt->end();
bbIt != bb_ie; ++bbIt) {
for (BasicBlock::iterator it = bbIt->begin(), ie = bbIt->end();
it != ie; ++it) {
unsigned id = infos.getInfo(it).id;
instructions.push_back(&*it);
const InstructionInfo& ii = km->infos->getInfo(&*it);
uint64_t initial = executor.patchContainer.inPatch(ii.file, ii.line, ii.assemblyLine) &&
sm.getIndexedValue(stats::uncoveredInstructions, id);
sm.setIndexedValue(stats::minDistToUncovered,
id,
initial);
}
}
}
std::reverse(instructions.begin(), instructions.end());
// I'm so lazy it's not even worklisted.
bool changed;
do {
changed = false;
for (std::vector<Instruction*>::iterator it = instructions.begin(),
ie = instructions.end(); it != ie; ++it) {
Instruction *inst = *it;
uint64_t best, cur = best = sm.getIndexedValue(stats::minDistToUncovered,
infos.getInfo(inst).id);
unsigned bestThrough = 0;
/* KATCH: don't 'step into' function calls when computing
* the minimum distance.
* This heuristic seems to perform better overall but
* can hurt us if the result of the function call is relevant
* for reaching the target.
* A better solution would be to first determine whether the
* call is relevant.
*/
#ifdef KATCH_USE_FUNCTION_CALLS
if (isa<CallInst>(inst) || isa<InvokeInst>(inst)) {
std::vector<Function*> &targets = callTargets[inst];
for (std::vector<Function*>::iterator fnIt = targets.begin(),
ie = targets.end(); fnIt != ie; ++fnIt) {
uint64_t dist = functionShortestPath[*fnIt];
if (dist) {
dist = 1+dist; // count instruction itself
if (bestThrough==0 || dist<bestThrough)
bestThrough = dist;
}
if (!(*fnIt)->isDeclaration()) {
uint64_t calleeDist = sm.getIndexedValue(stats::minDistToUncovered,
infos.getFunctionInfo(*fnIt).id);
if (calleeDist) {
calleeDist = 1+calleeDist; // count instruction itself
if (best==0 || calleeDist<best)
best = calleeDist;
}
}
}
} else
#endif
{
TerminatorInst* tinst;
if ((tinst = dyn_cast<TerminatorInst>(inst))) {
if (tinst->getNumSuccessors() > 1)
bestThrough = 1;
}
}
std::vector<Instruction*> succs = getSuccsAndCalls(m, inst);
for (std::vector<Instruction*>::iterator it2 = succs.begin(),
ie = succs.end(); it2 != ie; ++it2) {
uint64_t dist = sm.getIndexedValue(stats::minDistToUncovered,
infos.getInfo(*it2).id);
if (dist) {
uint64_t val = bestThrough + dist;
if (best==0 || val<best)
best = val;
}
}
if (best != cur) {
sm.setIndexedValue(stats::minDistToUncovered,
infos.getInfo(inst).id,
best);
changed = true;
}
}
} while (changed);
for (std::set<ExecutionState*>::iterator it = executor.states.begin(),
ie = executor.states.end(); it != ie; ++it) {
ExecutionState *es = *it;
uint64_t currentFrameMinDist = 0;
for (ExecutionState::stack_ty::iterator sfIt = es->stack.begin(),
sf_ie = es->stack.end(); sfIt != sf_ie; ++sfIt) {
ExecutionState::stack_ty::iterator next = sfIt + 1;
KInstIterator kii;
if (next==es->stack.end()) {
kii = es->pc;
} else {
kii = next->caller;
++kii;
}
sfIt->minDistToUncoveredOnReturn = currentFrameMinDist;
currentFrameMinDist = computeMinDistToUncovered(kii, currentFrameMinDist);
}
}
}
///
/// runOnFunction
///
bool LongLongMemAccessLowering::runOnFunction(Function &F) {
std::vector<StoreInst *> storeInstV;
std::vector<LoadInst *> loadInstV;
for (Function::iterator I = F.begin(), E = F.end(); I != E; ++I) {
for (BasicBlock::iterator BI = I->begin(), BE = I->end(); BI != BE; ++BI) {
Instruction *inst = BI;
if (StoreInst *storeInst = dyn_cast<StoreInst>(inst)) {
if (!storeInst->getValueOperand()->getType()->isIntegerTy(64)) {
continue;
}
assert(cast<PointerType>(storeInst->getPointerOperand()->getType())
->getElementType()->isIntegerTy(64));
storeInstV.push_back(storeInst);
}
else if (LoadInst *loadInst = dyn_cast<LoadInst>(inst)) {
if (!loadInst->getType()->isIntegerTy(64)) {
continue;
}
assert(cast<PointerType>(loadInst->getPointerOperand()->getType())
->getElementType()->isIntegerTy(64));
loadInstV.push_back(loadInst);
}
}
}
for (unsigned i = 0; i < storeInstV.size(); ++i) {
StoreInst *inst = storeInstV.at(i);
Value *storeVal = inst->getValueOperand();
Value *storePointer = inst->getPointerOperand();
// Insert new instructions
BitCastInst *storeAddrLo = new BitCastInst(storePointer, Type::getInt32PtrTy(F.getContext()), "", inst);
ConstantInt *offsetConst = ConstantInt::getSigned(Type::getInt32Ty(F.getContext()), 1);
std::vector<Value *> gepIdxList(1, offsetConst);
GetElementPtrInst *storeAddrHi = GetElementPtrInst::Create(storeAddrLo, gepIdxList, "", inst);
TruncInst *storeValLo = new TruncInst(storeVal, Type::getInt32Ty(F.getContext()), "", inst);
Value *aShrOffset = ConstantInt::getSigned(Type::getInt64Ty(F.getContext()), 32);
BinaryOperator *storeValAShr = BinaryOperator::Create(Instruction::AShr, storeVal,
aShrOffset, "", inst);
TruncInst *storeValHi = new TruncInst(storeValAShr, Type::getInt32Ty(F.getContext()), "", inst);
StoreInst *storeLo = new StoreInst(storeValLo, storeAddrLo, inst);
StoreInst *storeHi = new StoreInst(storeValHi, storeAddrHi, inst);
storeLo->setAlignment(4);
storeHi->setAlignment(4);
// Remove inst
inst->eraseFromParent();
}
for (unsigned i = 0; i < loadInstV.size(); ++i) {
LoadInst *inst = loadInstV.at(i);
Value *loadPointer = inst->getPointerOperand();
// Insert new instructions
BitCastInst *loadAddrLo = new BitCastInst(loadPointer, Type::getInt32PtrTy(F.getContext()), "", inst);
ConstantInt *offsetConst = ConstantInt::getSigned(Type::getInt32Ty(F.getContext()), 1);
std::vector<Value *> gepIdxList(1, offsetConst);
GetElementPtrInst *loadAddrHi = GetElementPtrInst::Create(loadAddrLo, gepIdxList, "", inst);
LoadInst *loadLo = new LoadInst(loadAddrLo, "", inst);
LoadInst *loadHi = new LoadInst(loadAddrHi, "", inst);
ZExtInst *loadLoLL = new ZExtInst(loadLo, Type::getInt64Ty(F.getContext()), "", inst);
ZExtInst *loadHiLL = new ZExtInst(loadHi, Type::getInt64Ty(F.getContext()), "", inst);
Value *shlOffset = ConstantInt::getSigned(Type::getInt64Ty(F.getContext()), 32);
BinaryOperator *loadHiLLShl = BinaryOperator::Create(Instruction::Shl, loadHiLL, shlOffset, "", inst);
BinaryOperator *loadValue = BinaryOperator::Create(Instruction::Or, loadLoLL, loadHiLLShl, "");
// Replace inst with new "loaded" value, the old value is deleted
ReplaceInstWithInst(inst, loadValue);
}
return true; // function is modified
}
//
// Returns of float, double and complex need to be handled with a helper
// function.
//
static bool fixupFPReturnAndCall
(Function &F, Module *M, const MipsSubtarget &Subtarget) {
bool Modified = false;
LLVMContext &C = M->getContext();
Type *MyVoid = Type::getVoidTy(C);
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
for (BasicBlock::iterator I = BB->begin(), E = BB->end();
I != E; ++I) {
Instruction &Inst = *I;
if (const ReturnInst *RI = dyn_cast<ReturnInst>(I)) {
Value *RVal = RI->getReturnValue();
if (!RVal) continue;
//
// If there is a return value and it needs a helper function,
// figure out which one and add a call before the actual
// return to this helper. The purpose of the helper is to move
// floating point values from their soft float return mapping to
// where they would have been mapped to in floating point registers.
//
Type *T = RVal->getType();
FPReturnVariant RV = whichFPReturnVariant(T);
if (RV == NoFPRet) continue;
static const char* Helper[NoFPRet] =
{"__mips16_ret_sf", "__mips16_ret_df", "__mips16_ret_sc",
"__mips16_ret_dc"};
const char *Name = Helper[RV];
AttributeSet A;
Value *Params[] = {RVal};
Modified = true;
//
// These helper functions have a different calling ABI so
// this __Mips16RetHelper indicates that so that later
// during call setup, the proper call lowering to the helper
// functions will take place.
//
A = A.addAttribute(C, AttributeSet::FunctionIndex,
"__Mips16RetHelper");
A = A.addAttribute(C, AttributeSet::FunctionIndex,
Attribute::ReadNone);
A = A.addAttribute(C, AttributeSet::FunctionIndex,
Attribute::NoInline);
Value *F = (M->getOrInsertFunction(Name, A, MyVoid, T, NULL));
CallInst::Create(F, Params, "", &Inst );
} else if (const CallInst *CI = dyn_cast<CallInst>(I)) {
const Value* V = CI->getCalledValue();
const Type* T = 0;
if (V) T = V->getType();
const PointerType *PFT=0;
if (T) PFT = dyn_cast<PointerType>(T);
const FunctionType *FT=0;
if (PFT) FT = dyn_cast<FunctionType>(PFT->getElementType());
Function *F_ = CI->getCalledFunction();
if (FT && needsFPReturnHelper(*FT) &&
!(F_ && isIntrinsicInline(F_))) {
Modified=true;
F.addFnAttr("saveS2");
}
if (F_ && !isIntrinsicInline(F_)) {
// pic mode calls are handled by already defined
// helper functions
if (needsFPReturnHelper(*F_)) {
Modified=true;
F.addFnAttr("saveS2");
}
if (Subtarget.getRelocationModel() != Reloc::PIC_ ) {
if (needsFPHelperFromSig(*F_)) {
assureFPCallStub(*F_, M, Subtarget);
Modified=true;
}
}
}
}
}
return Modified;
}
int compile(list<string> args, list<string> kgen_args,
string merge, list<string> merge_args,
string input, string output, int arch,
string host_compiler, string fileprefix)
{
//
// The LLVM compiler to emit IR.
//
const char* llvm_compiler = "kernelgen-gfortran";
//
// Interpret kernelgen compile options.
//
for (list<string>::iterator iarg = kgen_args.begin(),
iearg = kgen_args.end(); iarg != iearg; iarg++)
{
const char* arg = (*iarg).c_str();
if (!strncmp(arg, "-Wk,--llvm-compiler=", 20))
llvm_compiler = arg + 20;
}
//
// Generate temporary output file.
// Check if output file is specified in the command line.
// Replace or add output to the temporary file.
//
cfiledesc tmp_output = cfiledesc::mktemp(fileprefix);
bool output_specified = false;
for (list<string>::iterator iarg = args.begin(),
iearg = args.end(); iarg != iearg; iarg++)
{
const char* arg = (*iarg).c_str();
if (!strcmp(arg, "-o"))
{
iarg++;
*iarg = tmp_output.getFilename();
output_specified = true;
break;
}
}
if (!output_specified)
{
args.push_back("-o");
args.push_back(tmp_output.getFilename());
}
//
// 1) Compile source code using regular host compiler.
//
{
if (verbose)
{
cout << host_compiler;
for (list<string>::iterator iarg = args.begin(),
iearg = args.end(); iarg != iearg; iarg++)
cout << " " << *iarg;
cout << endl;
}
int status = execute(host_compiler, args, "", NULL, NULL);
if (status) return status;
}
//
// 2) Emit LLVM IR.
//
string out = "";
{
list<string> emit_ir_args;
for (list<string>::iterator iarg = args.begin(),
iearg = args.end(); iarg != iearg; iarg++)
{
const char* arg = (*iarg).c_str();
if (!strcmp(arg, "-c") || !strcmp(arg, "-o"))
{
iarg++;
continue;
}
if (!strcmp(arg, "-g"))
{
continue;
}
emit_ir_args.push_back(*iarg);
}
emit_ir_args.push_back("-fplugin=/opt/kernelgen/lib/dragonegg.so");
emit_ir_args.push_back("-fplugin-arg-dragonegg-emit-ir");
emit_ir_args.push_back("-S");
emit_ir_args.push_back(input);
emit_ir_args.push_back("-o");
emit_ir_args.push_back("-");
if (verbose)
{
cout << llvm_compiler;
for (list<string>::iterator iarg = emit_ir_args.begin(),
iearg = emit_ir_args.end(); iarg != iearg; iarg++)
cout << " " << *iarg;
cout << endl;
}
int status = execute(llvm_compiler, emit_ir_args, "", &out, NULL);
if (status) return status;
}
//
// 3) Record existing module functions.
//
LLVMContext &context = getGlobalContext();
SMDiagnostic diag;
MemoryBuffer* buffer1 = MemoryBuffer::getMemBuffer(out);
auto_ptr<Module> m1;
m1.reset(ParseIR(buffer1, diag, context));
//m1.get()->dump();
//
// 4) Inline calls and extract loops into new functions.
//
MemoryBuffer* buffer2 = MemoryBuffer::getMemBuffer(out);
auto_ptr<Module> m2;
m2.reset(ParseIR(buffer2, diag, context));
{
PassManager manager;
manager.add(createInstructionCombiningPass());
manager.run(*m2.get());
}
std::vector<CallInst *> LoopFuctionCalls;
{
PassManager manager;
manager.add(createBranchedLoopExtractorPass(LoopFuctionCalls));
manager.run(*m2.get());
}
//m2.get()->dump();
//
// 5) Replace call to loop functions with call to launcher.
// Append "always inline" attribute to all other functions.
//
Type* int32Ty = Type::getInt32Ty(context);
Function* launch = Function::Create(
TypeBuilder<types::i<32>(types::i<8>*, types::i<64>, types::i<32>*), true>::get(context),
GlobalValue::ExternalLinkage, "kernelgen_launch", m2.get());
for (Module::iterator f1 = m2.get()->begin(), fe1 = m2.get()->end(); f1 != fe1; f1++)
{
Function* func = f1;
if (func->isDeclaration()) continue;
// Search for the current function in original module
// functions list.
// If function is not in list of original module, then
// it is generated by the loop extractor.
// Append "always inline" attribute to all other functions.
if (m1.get()->getFunction(func->getName()))
{
const AttrListPtr attr = func->getAttributes();
const AttrListPtr attr_new = attr.addAttr(~0U, Attribute::AlwaysInline);
func->setAttributes(attr_new);
continue;
}
// Each such function must be extracted to the
// standalone module and packed into resulting
// object file data section.
if (verbose)
cout << "Preparing loop function " << func->getName().data() <<
" ..." << endl;
// Reset to default visibility.
func->setVisibility(GlobalValue::DefaultVisibility);
// Reset to default linkage.
func->setLinkage(GlobalValue::ExternalLinkage);
// Replace call to this function in module with call to launcher.
bool found = false;
for (Module::iterator f2 = m2->begin(), fe2 = m2->end(); (f2 != fe2) && !found; f2++)
for (Function::iterator bb = f2->begin(); (bb != f2->end()) && !found; bb++)
for (BasicBlock::iterator i = bb->begin(); i != bb->end(); i++)
{
// Check if instruction in focus is a call.
CallInst* call = dyn_cast<CallInst>(cast<Value>(i));
if (!call) continue;
// Check if function is called (needs -instcombine pass).
Function* callee = call->getCalledFunction();
if (!callee) continue;
if (callee->isDeclaration()) continue;
if (callee->getName() != func->getName()) continue;
// Create a constant array holding original called
// function name.
Constant* name = ConstantArray::get(
context, callee->getName(), true);
// Create and initialize the memory buffer for name.
ArrayType* nameTy = cast<ArrayType>(name->getType());
AllocaInst* nameAlloc = new AllocaInst(nameTy, "", call);
StoreInst* nameInit = new StoreInst(name, nameAlloc, "", call);
Value* Idx[2];
Idx[0] = Constant::getNullValue(Type::getInt32Ty(context));
Idx[1] = ConstantInt::get(Type::getInt32Ty(context), 0);
GetElementPtrInst* namePtr = GetElementPtrInst::Create(nameAlloc, Idx, "", call);
// Add pointer to the original function string name.
SmallVector<Value*, 16> call_args;
call_args.push_back(namePtr);
// Add size of the aggregated arguments structure.
{
BitCastInst* BC = new BitCastInst(
call->getArgOperand(0), Type::getInt64PtrTy(context),
"", call);
LoadInst* LI = new LoadInst(BC, "", call);
call_args.push_back(LI);
}
// Add original aggregated structure argument.
call_args.push_back(call->getArgOperand(0));
// Create new function call with new call arguments
// and copy old call properties.
CallInst* newcall = CallInst::Create(launch, call_args, "", call);
//newcall->takeName(call);
newcall->setCallingConv(call->getCallingConv());
newcall->setAttributes(call->getAttributes());
newcall->setDebugLoc(call->getDebugLoc());
// Replace old call with new one.
call->replaceAllUsesWith(newcall);
call->eraseFromParent();
found = true;
break;
}
}
//m2.get()->dump();
//
// 6) Apply optimization passes to the resulting common
// module.
//
{
PassManager manager;
manager.add(createLowerSetJmpPass());
PassManagerBuilder builder;
builder.Inliner = createFunctionInliningPass();
builder.OptLevel = 3;
builder.DisableSimplifyLibCalls = true;
builder.populateModulePassManager(manager);
manager.run(*m2.get());
}
//m2.get()->dump();
//
// 7) Embed the resulting module into object file.
//
{
string ir_string;
raw_string_ostream ir(ir_string);
ir << (*m2.get());
celf e(tmp_output.getFilename(), output);
e.getSection(".data")->addSymbol(
"__kernelgen_" + string(input),
ir_string.c_str(), ir_string.size() + 1);
}
return 0;
}
bool NVPTXLowerAggrCopies::runOnFunction(Function &F) {
SmallVector<LoadInst *, 4> aggrLoads;
SmallVector<MemTransferInst *, 4> aggrMemcpys;
SmallVector<MemSetInst *, 4> aggrMemsets;
const DataLayout &DL = F.getParent()->getDataLayout();
LLVMContext &Context = F.getParent()->getContext();
//
// Collect all the aggrLoads, aggrMemcpys and addrMemsets.
//
//const BasicBlock *firstBB = &F.front(); // first BB in F
for (Function::iterator BI = F.begin(), BE = F.end(); BI != BE; ++BI) {
//BasicBlock *bb = BI;
for (BasicBlock::iterator II = BI->begin(), IE = BI->end(); II != IE;
++II) {
if (LoadInst *load = dyn_cast<LoadInst>(II)) {
if (!load->hasOneUse())
continue;
if (DL.getTypeStoreSize(load->getType()) < MaxAggrCopySize)
continue;
User *use = load->user_back();
if (StoreInst *store = dyn_cast<StoreInst>(use)) {
if (store->getOperand(0) != load) //getValueOperand
continue;
aggrLoads.push_back(load);
}
} else if (MemTransferInst *intr = dyn_cast<MemTransferInst>(II)) {
Value *len = intr->getLength();
// If the number of elements being copied is greater
// than MaxAggrCopySize, lower it to a loop
if (ConstantInt *len_int = dyn_cast<ConstantInt>(len)) {
if (len_int->getZExtValue() >= MaxAggrCopySize) {
aggrMemcpys.push_back(intr);
}
} else {
// turn variable length memcpy/memmov into loop
aggrMemcpys.push_back(intr);
}
} else if (MemSetInst *memsetintr = dyn_cast<MemSetInst>(II)) {
Value *len = memsetintr->getLength();
if (ConstantInt *len_int = dyn_cast<ConstantInt>(len)) {
if (len_int->getZExtValue() >= MaxAggrCopySize) {
aggrMemsets.push_back(memsetintr);
}
} else {
// turn variable length memset into loop
aggrMemsets.push_back(memsetintr);
}
}
}
}
if ((aggrLoads.size() == 0) && (aggrMemcpys.size() == 0) &&
(aggrMemsets.size() == 0))
return false;
//
// Do the transformation of an aggr load/copy/set to a loop
//
for (unsigned i = 0, e = aggrLoads.size(); i != e; ++i) {
LoadInst *load = aggrLoads[i];
StoreInst *store = dyn_cast<StoreInst>(*load->user_begin());
Value *srcAddr = load->getOperand(0);
Value *dstAddr = store->getOperand(1);
unsigned numLoads = DL.getTypeStoreSize(load->getType());
Value *len = ConstantInt::get(Type::getInt32Ty(Context), numLoads);
convertTransferToLoop(store, srcAddr, dstAddr, len, load->isVolatile(),
store->isVolatile(), Context, F);
store->eraseFromParent();
load->eraseFromParent();
}
for (unsigned i = 0, e = aggrMemcpys.size(); i != e; ++i) {
MemTransferInst *cpy = aggrMemcpys[i];
Value *len = cpy->getLength();
// llvm 2.7 version of memcpy does not have volatile
// operand yet. So always making it non-volatile
// optimistically, so that we don't see unnecessary
// st.volatile in ptx
convertTransferToLoop(cpy, cpy->getSource(), cpy->getDest(), len, false,
false, Context, F);
cpy->eraseFromParent();
}
for (unsigned i = 0, e = aggrMemsets.size(); i != e; ++i) {
MemSetInst *memsetinst = aggrMemsets[i];
Value *len = memsetinst->getLength();
Value *val = memsetinst->getValue();
convertMemSetToLoop(memsetinst, memsetinst->getDest(), len, val, Context,
F);
memsetinst->eraseFromParent();
}
return true;
}
//
// Method: runOnFunction()
//
// Description:
// This is the entry point for this LLVM function pass. The pass manager will
// call this method for every function in the Module that will be transformed.
//
// Inputs:
// F - A reference to the function to transform.
//
// Outputs:
// F - The function will be modified to register and unregister stack objects.
//
// Return value:
// true - The function was modified.
// false - The function was not modified.
//
bool
RegisterStackObjPass::runOnFunction(Function & F) {
//
// Get prerequisite analysis information.
//
TD = &F.getParent()->getDataLayout();
LI = &getAnalysis<LoopInfoWrapperPass>().getLoopInfo();
DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
DF = &getAnalysis<DominanceFrontier>();
//
// Get pointers to the functions for registering and unregistering pointers.
//
PoolRegister = F.getParent()->getFunction ("pool_register_stack");
StackFree = F.getParent()->getFunction ("pool_unregister_stack");
assert (PoolRegister);
assert (StackFree);
// The set of registered stack objects
std::vector<CallInst *> PoolRegisters;
// The set of stack objects within the function.
std::vector<AllocaInst *> AllocaList;
// The set of instructions that can cause the function to return to its
// caller.
std::vector<Instruction *> ExitPoints;
//
// Scan the function to register allocas and find locations where registered
// allocas to be deregistered.
//
for (Function::iterator BI = F.begin(); BI != F.end(); ++BI) {
//
// Create a list of alloca instructions to register. Note that we create
// the list ahead of time because registerAllocaInst() will create new
// alloca instructions.
//
for (BasicBlock::iterator I = BI->begin(); I != BI->end(); ++I) {
if (AllocaInst * AI = dyn_cast<AllocaInst>(I)) {
//
// Ensure that the alloca is not within a loop; we don't support yet.
//
#if 0
if (LI->getLoopFor (BI)) {
assert (0 &&
"Register Stack Objects: No support for alloca in loop!\n");
abort();
}
AllocaList.push_back (AI);
#else
if (!(LI->getLoopFor (BI))) {
AllocaList.push_back (AI);
}
#endif
}
}
//
// Add calls to register the allocated stack objects.
//
while (AllocaList.size()) {
AllocaInst * AI = AllocaList.back();
AllocaList.pop_back();
if (CallInst * CI = registerAllocaInst (AI))
PoolRegisters.push_back(CI);
}
//
// If the terminator instruction of this basic block can return control
// flow back to the caller, mark it as a place where a deregistration
// is needed.
//
Instruction * Terminator = BI->getTerminator();
if ((isa<ReturnInst>(Terminator)) || (isa<ResumeInst>(Terminator))) {
ExitPoints.push_back (Terminator);
}
}
//
// Insert poolunregister calls for all of the registered allocas.
//
insertPoolFrees (PoolRegisters, ExitPoints, &F.getContext());
//
// Conservatively assume that we've changed the function.
//
return true;
}
void StatsTracker::writeIStats() {
Module *m = executor.kmodule->module;
uint64_t istatsMask = 0;
std::ostream &of = *istatsFile;
of.seekp(0, std::ios::end);
unsigned istatsSize = of.tellp();
of.seekp(0);
of << "version: 1\n";
of << "creator: klee\n";
of << "pid: " << sys::Process::GetCurrentUserId() << "\n";
of << "cmd: " << m->getModuleIdentifier() << "\n\n";
of << "\n";
StatisticManager &sm = *theStatisticManager;
unsigned nStats = sm.getNumStatistics();
// Max is 13, sadly
istatsMask |= 1<<sm.getStatisticID("Queries");
istatsMask |= 1<<sm.getStatisticID("QueriesValid");
istatsMask |= 1<<sm.getStatisticID("QueriesInvalid");
istatsMask |= 1<<sm.getStatisticID("QueryTime");
istatsMask |= 1<<sm.getStatisticID("ResolveTime");
istatsMask |= 1<<sm.getStatisticID("Instructions");
istatsMask |= 1<<sm.getStatisticID("InstructionTimes");
istatsMask |= 1<<sm.getStatisticID("InstructionRealTimes");
istatsMask |= 1<<sm.getStatisticID("Forks");
istatsMask |= 1<<sm.getStatisticID("CoveredInstructions");
istatsMask |= 1<<sm.getStatisticID("UncoveredInstructions");
istatsMask |= 1<<sm.getStatisticID("States");
istatsMask |= 1<<sm.getStatisticID("MinDistToUncovered");
of << "positions: instr line\n";
for (unsigned i=0; i<nStats; i++) {
if (istatsMask & (1<<i)) {
Statistic &s = sm.getStatistic(i);
of << "event: " << s.getShortName() << " : "
<< s.getName() << "\n";
}
}
of << "events: ";
for (unsigned i=0; i<nStats; i++) {
if (istatsMask & (1<<i))
of << sm.getStatistic(i).getShortName() << " ";
}
of << "\n";
// set state counts, decremented after we process so that we don't
// have to zero all records each time.
if (istatsMask & (1<<stats::states.getID()))
updateStateStatistics(1);
std::string sourceFile = "";
CallSiteSummaryTable callSiteStats;
if (UseCallPaths)
callPathManager.getSummaryStatistics(callSiteStats);
of << "ob=" << objectFilename << "\n";
for (Module::iterator fnIt = m->begin(), fn_ie = m->end();
fnIt != fn_ie; ++fnIt) {
if (!fnIt->isDeclaration()) {
of << "fn=" << CXXDemangle(fnIt->getName().str()) << "\n";
for (Function::iterator bbIt = fnIt->begin(), bb_ie = fnIt->end();
bbIt != bb_ie; ++bbIt) {
for (BasicBlock::iterator it = bbIt->begin(), ie = bbIt->end();
it != ie; ++it) {
Instruction *instr = &*it;
const InstructionInfo &ii = executor.kmodule->infos->getInfo(instr);
unsigned index = ii.id;
if (ii.file!=sourceFile) {
of << "fl=" << ii.file << "\n";
sourceFile = ii.file;
}
of << ii.assemblyLine << " ";
of << ii.line << " ";
for (unsigned i=0; i<nStats; i++)
if (istatsMask&(1<<i))
of << sm.getIndexedValue(sm.getStatistic(i), index) << " ";
of << "\n";
if (UseCallPaths &&
(isa<CallInst>(instr) || isa<InvokeInst>(instr))) {
CallSiteSummaryTable::iterator it = callSiteStats.find(instr);
if (it!=callSiteStats.end()) {
for (std::map<llvm::Function*, CallSiteInfo>::iterator
fit = it->second.begin(), fie = it->second.end();
fit != fie; ++fit) {
Function *f = fit->first;
CallSiteInfo &csi = fit->second;
const InstructionInfo &fii =
executor.kmodule->infos->getFunctionInfo(f);
if (fii.file!="" && fii.file!=sourceFile)
of << "cfl=" << fii.file << "\n";
of << "cfn=" << CXXDemangle(f->getName().str()) << "\n";
of << "calls=" << csi.count << " ";
of << fii.assemblyLine << " ";
of << fii.line << "\n";
of << ii.assemblyLine << " ";
of << ii.line << " ";
for (unsigned i=0; i<nStats; i++) {
if (istatsMask&(1<<i)) {
Statistic &s = sm.getStatistic(i);
uint64_t value;
// Hack, ignore things that don't make sense on
// call paths.
if (&s == &stats::locallyUncoveredInstructions) {
value = 0;
} else {
value = csi.statistics.getValue(s);
}
of << value << " ";
}
}
of << "\n";
}
}
}
}
}
}
}
if (istatsMask & (1<<stats::states.getID()))
updateStateStatistics((uint64_t)-1);
// Clear then end of the file if necessary (no truncate op?).
unsigned pos = of.tellp();
for (unsigned i=pos; i<istatsSize; ++i)
of << '\n';
of.flush();
}
//
// Method: runOnFunction()
//
// Description:
// Entry point for this LLVM pass.
//
// Return value:
// true - The function was modified.
// false - The function was not modified.
//
bool
BreakConstantGEPs::runOnFunction (Function & F) {
if (!pocl::Workgroup::isKernelToProcess(F)) return false;
bool modified = false;
// Worklist of values to check for constant GEP expressions
std::vector<Instruction *> Worklist;
//
// Initialize the worklist by finding all instructions that have one or more
// operands containing a constant GEP expression.
//
for (Function::iterator BB = F.begin(); BB != F.end(); ++BB) {
for (BasicBlock::iterator i = BB->begin(); i != BB->end(); ++i) {
//
// Scan through the operands of this instruction. If it is a constant
// expression GEP, insert an instruction GEP before the instruction.
//
Instruction * I = i;
for (unsigned index = 0; index < I->getNumOperands(); ++index) {
if (hasConstantGEP (I->getOperand(index))) {
Worklist.push_back (I);
}
}
}
}
//
// Determine whether we will modify anything.
//
if (Worklist.size()) modified = true;
//
// While the worklist is not empty, take an item from it, convert the
// operands into instructions if necessary, and determine if the newly
// added instructions need to be processed as well.
//
while (Worklist.size()) {
Instruction * I = Worklist.back();
Worklist.pop_back();
//
// Scan through the operands of this instruction and convert each into an
// instruction. Note that this works a little differently for phi
// instructions because the new instruction must be added to the
// appropriate predecessor block.
//
if (PHINode * PHI = dyn_cast<PHINode>(I)) {
for (unsigned index = 0; index < PHI->getNumIncomingValues(); ++index) {
//
// For PHI Nodes, if an operand is a constant expression with a GEP, we
// want to insert the new instructions in the predecessor basic block.
//
// Note: It seems that it's possible for a phi to have the same
// incoming basic block listed multiple times; this seems okay as long
// the same value is listed for the incoming block.
//
Instruction * InsertPt = PHI->getIncomingBlock(index)->getTerminator();
if (ConstantExpr * CE = hasConstantGEP (PHI->getIncomingValue(index))) {
Instruction * NewInst = convertExpression (CE, InsertPt);
for (unsigned i2 = index; i2 < PHI->getNumIncomingValues(); ++i2) {
if ((PHI->getIncomingBlock (i2)) == PHI->getIncomingBlock (index))
PHI->setIncomingValue (i2, NewInst);
}
Worklist.push_back (NewInst);
}
}
} else {
for (unsigned index = 0; index < I->getNumOperands(); ++index) {
//
// For other instructions, we want to insert instructions replacing
// constant expressions immediently before the instruction using the
// constant expression.
//
if (ConstantExpr * CE = hasConstantGEP (I->getOperand(index))) {
Instruction * NewInst = convertExpression (CE, I);
I->replaceUsesOfWith (CE, NewInst);
Worklist.push_back (NewInst);
}
}
}
}
return modified;
}
//
// Method: runOnModule()
//
// Description:
// Entry point for this LLVM pass.
// Find all GEPs, and simplify them.
//
// Inputs:
// M - A reference to the LLVM module to transform
//
// Outputs:
// M - The transformed LLVM module.
//
// Return value:
// true - The module was modified.
// false - The module was not modified.
//
bool SimplifyGEP::runOnModule(Module& M) {
TD = &getAnalysis<DataLayout>();
preprocess(M);
for (Module::iterator F = M.begin(); F != M.end(); ++F){
for (Function::iterator B = F->begin(), FE = F->end(); B != FE; ++B) {
for (BasicBlock::iterator I = B->begin(), BE = B->end(); I != BE; I++) {
if(!(isa<GetElementPtrInst>(I)))
continue;
GetElementPtrInst *GEP = cast<GetElementPtrInst>(I);
Value *PtrOp = GEP->getOperand(0);
Value *StrippedPtr = PtrOp->stripPointerCasts();
// Check if the GEP base pointer is enclosed in a cast
if (StrippedPtr != PtrOp) {
const PointerType *StrippedPtrTy =cast<PointerType>(StrippedPtr->getType());
bool HasZeroPointerIndex = false;
if (ConstantInt *C = dyn_cast<ConstantInt>(GEP->getOperand(1)))
HasZeroPointerIndex = C->isZero();
// Transform: GEP (bitcast [10 x i8]* X to [0 x i8]*), i32 0, ...
// into : GEP [10 x i8]* X, i32 0, ...
//
// Likewise, transform: GEP (bitcast i8* X to [0 x i8]*), i32 0, ...
// into : GEP i8* X, ...
//
// This occurs when the program declares an array extern like "int X[];"
if (HasZeroPointerIndex) {
const PointerType *CPTy = cast<PointerType>(PtrOp->getType());
if (const ArrayType *CATy =
dyn_cast<ArrayType>(CPTy->getElementType())) {
// GEP (bitcast i8* X to [0 x i8]*), i32 0, ... ?
if (CATy->getElementType() == StrippedPtrTy->getElementType()) {
// -> GEP i8* X, ...
SmallVector<Value*, 8> Idx(GEP->idx_begin()+1, GEP->idx_end());
GetElementPtrInst *Res =
GetElementPtrInst::Create(StrippedPtr, Idx, GEP->getName(), GEP);
Res->setIsInBounds(GEP->isInBounds());
GEP->replaceAllUsesWith(Res);
continue;
}
if (const ArrayType *XATy =
dyn_cast<ArrayType>(StrippedPtrTy->getElementType())){
// GEP (bitcast [10 x i8]* X to [0 x i8]*), i32 0, ... ?
if (CATy->getElementType() == XATy->getElementType()) {
// -> GEP [10 x i8]* X, i32 0, ...
// At this point, we know that the cast source type is a pointer
// to an array of the same type as the destination pointer
// array. Because the array type is never stepped over (there
// is a leading zero) we can fold the cast into this GEP.
GEP->setOperand(0, StrippedPtr);
continue;
}
}
}
} else if (GEP->getNumOperands() == 2) {
// Transform things like:
// %t = getelementptr i32* bitcast ([2 x i32]* %str to i32*), i32 %V
// into: %t1 = getelementptr [2 x i32]* %str, i32 0, i32 %V; bitcast
Type *SrcElTy = StrippedPtrTy->getElementType();
Type *ResElTy=cast<PointerType>(PtrOp->getType())->getElementType();
if (TD && SrcElTy->isArrayTy() &&
TD->getTypeAllocSize(cast<ArrayType>(SrcElTy)->getElementType()) ==
TD->getTypeAllocSize(ResElTy)) {
Value *Idx[2];
Idx[0] = Constant::getNullValue(Type::getInt32Ty(GEP->getContext()));
Idx[1] = GEP->getOperand(1);
Value *NewGEP = GetElementPtrInst::Create(StrippedPtr, Idx,
GEP->getName(), GEP);
// V and GEP are both pointer types --> BitCast
GEP->replaceAllUsesWith(new BitCastInst(NewGEP, GEP->getType(), GEP->getName(), GEP));
continue;
}
// Transform things like:
// getelementptr i8* bitcast ([100 x double]* X to i8*), i32 %tmp
// (where tmp = 8*tmp2) into:
// getelementptr [100 x double]* %arr, i32 0, i32 %tmp2; bitcast
if (TD && SrcElTy->isArrayTy() && ResElTy->isIntegerTy(8)) {
uint64_t ArrayEltSize =
TD->getTypeAllocSize(cast<ArrayType>(SrcElTy)->getElementType());
// Check to see if "tmp" is a scale by a multiple of ArrayEltSize. We
// allow either a mul, shift, or constant here.
Value *NewIdx = 0;
ConstantInt *Scale = 0;
if (ArrayEltSize == 1) {
NewIdx = GEP->getOperand(1);
Scale = ConstantInt::get(cast<IntegerType>(NewIdx->getType()), 1);
} else if (ConstantInt *CI = dyn_cast<ConstantInt>(GEP->getOperand(1))) {
NewIdx = ConstantInt::get(CI->getType(), 1);
Scale = CI;
} else if (Instruction *Inst =dyn_cast<Instruction>(GEP->getOperand(1))){
if (Inst->getOpcode() == Instruction::Shl &&
isa<ConstantInt>(Inst->getOperand(1))) {
ConstantInt *ShAmt = cast<ConstantInt>(Inst->getOperand(1));
uint32_t ShAmtVal = ShAmt->getLimitedValue(64);
Scale = ConstantInt::get(cast<IntegerType>(Inst->getType()),
1ULL << ShAmtVal);
NewIdx = Inst->getOperand(0);
} else if (Inst->getOpcode() == Instruction::Mul &&
isa<ConstantInt>(Inst->getOperand(1))) {
Scale = cast<ConstantInt>(Inst->getOperand(1));
NewIdx = Inst->getOperand(0);
}
}
// If the index will be to exactly the right offset with the scale taken
// out, perform the transformation. Note, we don't know whether Scale is
// signed or not. We'll use unsigned version of division/modulo
// operation after making sure Scale doesn't have the sign bit set.
if (ArrayEltSize && Scale && Scale->getSExtValue() >= 0LL &&
Scale->getZExtValue() % ArrayEltSize == 0) {
Scale = ConstantInt::get(Scale->getType(),
Scale->getZExtValue() / ArrayEltSize);
if (Scale->getZExtValue() != 1) {
Constant *C = ConstantExpr::getIntegerCast(Scale, NewIdx->getType(),
false /*ZExt*/);
NewIdx = BinaryOperator::Create(BinaryOperator::Mul, NewIdx, C, "idxscale");
}
// Insert the new GEP instruction.
Value *Idx[2];
Idx[0] = Constant::getNullValue(Type::getInt32Ty(GEP->getContext()));
Idx[1] = NewIdx;
Value *NewGEP = GetElementPtrInst::Create(StrippedPtr, Idx,
GEP->getName(), GEP);
GEP->replaceAllUsesWith(new BitCastInst(NewGEP, GEP->getType(), GEP->getName(), GEP));
continue;
}
}
}
}
}
}
}
return true;
}
bool AddressSanitizer::handleFunction(Module &M, Function &F) {
if (BL->isIn(F)) return false;
if (&F == AsanCtorFunction) return false;
// If needed, insert __asan_init before checking for AddressSafety attr.
maybeInsertAsanInitAtFunctionEntry(F);
if (!F.hasFnAttr(Attribute::AddressSafety)) return false;
if (!ClDebugFunc.empty() && ClDebugFunc != F.getName())
return false;
// We want to instrument every address only once per basic block
// (unless there are calls between uses).
SmallSet<Value*, 16> TempsToInstrument;
SmallVector<Instruction*, 16> ToInstrument;
SmallVector<Instruction*, 8> NoReturnCalls;
bool IsWrite;
// Fill the set of memory operations to instrument.
for (Function::iterator FI = F.begin(), FE = F.end();
FI != FE; ++FI) {
TempsToInstrument.clear();
int NumInsnsPerBB = 0;
for (BasicBlock::iterator BI = FI->begin(), BE = FI->end();
BI != BE; ++BI) {
if (LooksLikeCodeInBug11395(BI)) return false;
if (Value *Addr = isInterestingMemoryAccess(BI, &IsWrite)) {
if (ClOpt && ClOptSameTemp) {
if (!TempsToInstrument.insert(Addr))
continue; // We've seen this temp in the current BB.
}
} else if (isa<MemIntrinsic>(BI) && ClMemIntrin) {
// ok, take it.
} else {
if (CallInst *CI = dyn_cast<CallInst>(BI)) {
// A call inside BB.
TempsToInstrument.clear();
if (CI->doesNotReturn()) {
NoReturnCalls.push_back(CI);
}
}
continue;
}
ToInstrument.push_back(BI);
NumInsnsPerBB++;
if (NumInsnsPerBB >= ClMaxInsnsToInstrumentPerBB)
break;
}
}
AsanFunctionContext AFC(F);
// Instrument.
int NumInstrumented = 0;
for (size_t i = 0, n = ToInstrument.size(); i != n; i++) {
Instruction *Inst = ToInstrument[i];
if (ClDebugMin < 0 || ClDebugMax < 0 ||
(NumInstrumented >= ClDebugMin && NumInstrumented <= ClDebugMax)) {
if (isInterestingMemoryAccess(Inst, &IsWrite))
instrumentMop(AFC, Inst);
else
instrumentMemIntrinsic(AFC, cast<MemIntrinsic>(Inst));
}
NumInstrumented++;
}
DEBUG(dbgs() << F);
bool ChangedStack = poisonStackInFunction(M, F);
// We must unpoison the stack before every NoReturn call (throw, _exit, etc).
// See e.g. http://code.google.com/p/address-sanitizer/issues/detail?id=37
for (size_t i = 0, n = NoReturnCalls.size(); i != n; i++) {
Instruction *CI = NoReturnCalls[i];
IRBuilder<> IRB(CI);
IRB.CreateCall(M.getOrInsertFunction(kAsanHandleNoReturnName,
IRB.getVoidTy(), NULL));
}
return NumInstrumented > 0 || ChangedStack || !NoReturnCalls.empty();
}
/// setupEntryBlockAndCallSites - Setup the entry block by creating and filling
/// the function context and marking the call sites with the appropriate
/// values. These values are used by the DWARF EH emitter.
bool SjLjEHPass::setupEntryBlockAndCallSites(Function &F) {
SmallVector<ReturnInst*, 16> Returns;
SmallVector<InvokeInst*, 16> Invokes;
SmallSetVector<LandingPadInst*, 16> LPads;
// Look through the terminators of the basic blocks to find invokes.
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
if (InvokeInst *II = dyn_cast<InvokeInst>(BB->getTerminator())) {
Invokes.push_back(II);
LPads.insert(II->getUnwindDest()->getLandingPadInst());
} else if (ReturnInst *RI = dyn_cast<ReturnInst>(BB->getTerminator())) {
Returns.push_back(RI);
}
if (Invokes.empty()) return false;
NumInvokes += Invokes.size();
lowerIncomingArguments(F);
lowerAcrossUnwindEdges(F, Invokes);
Value *FuncCtx =
setupFunctionContext(F, makeArrayRef(LPads.begin(), LPads.end()));
BasicBlock *EntryBB = F.begin();
Type *Int32Ty = Type::getInt32Ty(F.getContext());
Value *Idxs[2] = {
ConstantInt::get(Int32Ty, 0), 0
};
// Get a reference to the jump buffer.
Idxs[1] = ConstantInt::get(Int32Ty, 5);
Value *JBufPtr = GetElementPtrInst::Create(FuncCtx, Idxs, "jbuf_gep",
EntryBB->getTerminator());
// Save the frame pointer.
Idxs[1] = ConstantInt::get(Int32Ty, 0);
Value *FramePtr = GetElementPtrInst::Create(JBufPtr, Idxs, "jbuf_fp_gep",
EntryBB->getTerminator());
Value *Val = CallInst::Create(FrameAddrFn,
ConstantInt::get(Int32Ty, 0),
"fp",
EntryBB->getTerminator());
new StoreInst(Val, FramePtr, true, EntryBB->getTerminator());
// Save the stack pointer.
Idxs[1] = ConstantInt::get(Int32Ty, 2);
Value *StackPtr = GetElementPtrInst::Create(JBufPtr, Idxs, "jbuf_sp_gep",
EntryBB->getTerminator());
Val = CallInst::Create(StackAddrFn, "sp", EntryBB->getTerminator());
new StoreInst(Val, StackPtr, true, EntryBB->getTerminator());
// Call the setjmp instrinsic. It fills in the rest of the jmpbuf.
Value *SetjmpArg = CastInst::Create(Instruction::BitCast, JBufPtr,
Type::getInt8PtrTy(F.getContext()), "",
EntryBB->getTerminator());
CallInst::Create(BuiltinSetjmpFn, SetjmpArg, "", EntryBB->getTerminator());
// Store a pointer to the function context so that the back-end will know
// where to look for it.
Value *FuncCtxArg = CastInst::Create(Instruction::BitCast, FuncCtx,
Type::getInt8PtrTy(F.getContext()), "",
EntryBB->getTerminator());
CallInst::Create(FuncCtxFn, FuncCtxArg, "", EntryBB->getTerminator());
// At this point, we are all set up, update the invoke instructions to mark
// their call_site values.
for (unsigned I = 0, E = Invokes.size(); I != E; ++I) {
insertCallSiteStore(Invokes[I], I + 1);
ConstantInt *CallSiteNum =
ConstantInt::get(Type::getInt32Ty(F.getContext()), I + 1);
// Record the call site value for the back end so it stays associated with
// the invoke.
CallInst::Create(CallSiteFn, CallSiteNum, "", Invokes[I]);
}
// Mark call instructions that aren't nounwind as no-action (call_site ==
// -1). Skip the entry block, as prior to then, no function context has been
// created for this function and any unexpected exceptions thrown will go
// directly to the caller's context, which is what we want anyway, so no need
// to do anything here.
for (Function::iterator BB = F.begin(), E = F.end(); ++BB != E;)
for (BasicBlock::iterator I = BB->begin(), end = BB->end(); I != end; ++I)
if (CallInst *CI = dyn_cast<CallInst>(I)) {
if (!CI->doesNotThrow())
insertCallSiteStore(CI, -1);
} else if (ResumeInst *RI = dyn_cast<ResumeInst>(I)) {
insertCallSiteStore(RI, -1);
}
// Register the function context and make sure it's known to not throw
CallInst *Register = CallInst::Create(RegisterFn, FuncCtx, "",
EntryBB->getTerminator());
Register->setDoesNotThrow();
// Following any allocas not in the entry block, update the saved SP in the
// jmpbuf to the new value.
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
if (BB == F.begin())
continue;
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
if (CallInst *CI = dyn_cast<CallInst>(I)) {
if (CI->getCalledFunction() != StackRestoreFn)
continue;
} else if (!isa<AllocaInst>(I)) {
continue;
}
Instruction *StackAddr = CallInst::Create(StackAddrFn, "sp");
StackAddr->insertAfter(I);
Instruction *StoreStackAddr = new StoreInst(StackAddr, StackPtr, true);
StoreStackAddr->insertAfter(StackAddr);
}
}
// Finally, for any returns from this function, if this function contains an
// invoke, add a call to unregister the function context.
for (unsigned I = 0, E = Returns.size(); I != E; ++I)
CallInst::Create(UnregisterFn, FuncCtx, "", Returns[I]);
return true;
}
bool Inliner::runOnSCC(CallGraphSCC &SCC) {
CallGraph &CG = getAnalysis<CallGraph>();
const DataLayout *TD = getAnalysisIfAvailable<DataLayout>();
const TargetLibraryInfo *TLI = getAnalysisIfAvailable<TargetLibraryInfo>();
SmallPtrSet<Function*, 8> SCCFunctions;
DEBUG(dbgs() << "Inliner visiting SCC:");
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) {
Function *F = (*I)->getFunction();
if (F) SCCFunctions.insert(F);
DEBUG(dbgs() << " " << (F ? F->getName() : "INDIRECTNODE"));
}
// 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.
SmallVector<std::pair<CallSite, int>, 16> CallSites;
// When inlining a callee produces new call sites, we want to keep track of
// the fact that they were inlined from the callee. This allows us to avoid
// infinite inlining in some obscure cases. To represent this, we use an
// index into the InlineHistory vector.
SmallVector<std::pair<Function*, int>, 8> InlineHistory;
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) {
Function *F = (*I)->getFunction();
if (!F) continue;
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
CallSite CS(cast<Value>(I));
// If this isn't a call, or it is a call to an intrinsic, it can
// never be inlined.
if (!CS || isa<IntrinsicInst>(I))
continue;
// If this is a direct call to an external function, we can never inline
// it. If it is an indirect call, inlining may resolve it to be a
// direct call, so we keep it.
if (CS.getCalledFunction() && CS.getCalledFunction()->isDeclaration())
continue;
CallSites.push_back(std::make_pair(CS, -1));
}
}
DEBUG(dbgs() << ": " << CallSites.size() << " call sites.\n");
// If there are no calls in this function, exit early.
if (CallSites.empty())
return false;
// Now that we have all of the call sites, move the ones to functions in the
// current SCC to the end of the list.
unsigned FirstCallInSCC = CallSites.size();
for (unsigned i = 0; i < FirstCallInSCC; ++i)
if (Function *F = CallSites[i].first.getCalledFunction())
if (SCCFunctions.count(F))
std::swap(CallSites[i--], CallSites[--FirstCallInSCC]);
InlinedArrayAllocasTy InlinedArrayAllocas;
InlineFunctionInfo InlineInfo(&CG, TD);
// Now that we have all of the call sites, loop over them and inline them if
// it looks profitable to do so.
bool Changed = false;
bool LocalChange;
do {
LocalChange = false;
// Iterate over the outer loop because inlining functions can cause indirect
// calls to become direct calls.
for (unsigned CSi = 0; CSi != CallSites.size(); ++CSi) {
CallSite CS = CallSites[CSi].first;
Function *Caller = CS.getCaller();
Function *Callee = CS.getCalledFunction();
// If this call site is dead and it is to a readonly function, we should
// just delete the call instead of trying to inline it, regardless of
// size. This happens because IPSCCP propagates the result out of the
// call and then we're left with the dead call.
if (isInstructionTriviallyDead(CS.getInstruction(), TLI)) {
DEBUG(dbgs() << " -> Deleting dead call: "
<< *CS.getInstruction() << "\n");
// Update the call graph by deleting the edge from Callee to Caller.
CG[Caller]->removeCallEdgeFor(CS);
CS.getInstruction()->eraseFromParent();
++NumCallsDeleted;
} else {
// We can only inline direct calls to non-declarations.
if (Callee == 0 || Callee->isDeclaration()) continue;
// If this call site was obtained by inlining another function, verify
// that the include path for the function did not include the callee
// itself. If so, we'd be recursively inlining the same function,
// which would provide the same callsites, which would cause us to
// infinitely inline.
int InlineHistoryID = CallSites[CSi].second;
if (InlineHistoryID != -1 &&
InlineHistoryIncludes(Callee, InlineHistoryID, InlineHistory))
continue;
// If the policy determines that we should inline this function,
// try to do so.
if (!shouldInline(CS))
continue;
// Attempt to inline the function.
if (!InlineCallIfPossible(CS, InlineInfo, InlinedArrayAllocas,
InlineHistoryID, InsertLifetime))
continue;
++NumInlined;
// If inlining this function gave us any new call sites, throw them
// onto our worklist to process. They are useful inline candidates.
if (!InlineInfo.InlinedCalls.empty()) {
// Create a new inline history entry for this, so that we remember
// that these new callsites came about due to inlining Callee.
int NewHistoryID = InlineHistory.size();
InlineHistory.push_back(std::make_pair(Callee, InlineHistoryID));
for (unsigned i = 0, e = InlineInfo.InlinedCalls.size();
i != e; ++i) {
Value *Ptr = InlineInfo.InlinedCalls[i];
CallSites.push_back(std::make_pair(CallSite(Ptr), NewHistoryID));
}
}
}
// If we inlined or deleted the last possible call site to the function,
// delete the function body now.
if (Callee && Callee->use_empty() && Callee->hasLocalLinkage() &&
// TODO: Can remove if in SCC now.
!SCCFunctions.count(Callee) &&
// The function may be apparently dead, but if there are indirect
// callgraph references to the node, we cannot delete it yet, this
// could invalidate the CGSCC iterator.
CG[Callee]->getNumReferences() == 0) {
DEBUG(dbgs() << " -> Deleting dead function: "
<< Callee->getName() << "\n");
CallGraphNode *CalleeNode = CG[Callee];
// Remove any call graph edges from the callee to its callees.
CalleeNode->removeAllCalledFunctions();
// Removing the node for callee from the call graph and delete it.
delete CG.removeFunctionFromModule(CalleeNode);
++NumDeleted;
}
// Remove this call site from the list. If possible, use
// swap/pop_back for efficiency, but do not use it if doing so would
// move a call site to a function in this SCC before the
// 'FirstCallInSCC' barrier.
if (SCC.isSingular()) {
CallSites[CSi] = CallSites.back();
CallSites.pop_back();
} else {
CallSites.erase(CallSites.begin()+CSi);
}
--CSi;
Changed = true;
LocalChange = true;
}
} while (LocalChange);
return Changed;
}