/// findFunctionScopedAllocas - store all allocas that are known to be valid
/// to the end of their function in a set. The current algorithm does this by
/// finding all the allocas in the entry block that are before the first
/// llvm.stacksave call (if any).
///
/// FIXME: There can also be allocas elsewhere that get deallocated at the end
/// of the function but they are pessimistically ignored for now.
///
void ExactCheckOpt::findFunctionScopedAllocas(Module &M) {
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->empty())
continue;
BasicBlock &BB = F->getEntryBlock();
for (BasicBlock::iterator I = BB.begin(), E = BB.end(); I != E; ++I) {
if (AllocaInst *AI = dyn_cast<AllocaInst>(I)) {
FunctionScopedAllocas.insert(AI);
} else if(CallInst *CI = dyn_cast<CallInst>(I)) {
Function *CalledFunction = CI->getCalledFunction();
if (CalledFunction && CalledFunction->getName() == "llvm.stacksave")
break;
}
}
}
}
bool TraceBasicBlocks::runOnModule(Module &M) {
Context =&M.getContext();
Function *Main = M.getFunction("main");
if (Main == 0) {
errs() << "WARNING: cannot insert basic-block trace instrumentation"
<< " into a module with no main function!\n";
return false; // No main, no instrumentation!
}
const char* FnName="llvm_trace_basic_block";
Constant *InstrFn = M.getOrInsertFunction (FnName, Type::getVoidTy(*Context),
Type::getInt64Ty(*Context), NULL);
unsigned BBNumber = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if(F->empty()) { continue; }
//We insert instrumentation calls in reverse order, because insertion puts them before previous instructions
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
dbgs() << "InsertInstrumentationCall (\"" << BB->getName ()
<< "\", \"" << F->getName() << "\", " << BBNumber << ")\n";
//On Exit we emit the FunRet block
TerminatorInst* TI=BB->getTerminator();
if(TI->getNumSuccessors()==0) {
InsertRetInstrumentationCall(TI,InstrFn);
}
InsertInstrumentationCall (BB, InstrFn, BBNumber);
if(TraceMemory) {
InsertMemoryTracingCall (BB,InstrFn);
}
++BBNumber;
}
//on Entry we emit the FunCall block.
BasicBlock* EntryBlock=&F->getEntryBlock();
InsertInstrumentationCall (EntryBlock, InstrFn, BBTraceStream::FunCallID);
}
// Add the initialization call to main.
InsertProfilingInitCall(Main, "llvm_start_basic_block_tracing");
return true;
}
/// matchEdges - Link every profile counter with an edge.
unsigned ProfileMetadataLoaderPass::matchEdges(Module &M, ProfileData &PB,
ArrayRef<unsigned> Counters) {
if (Counters.size() == 0) return 0;
unsigned ReadCount = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
DEBUG(dbgs() << "Loading edges in '" << F->getName() << "'\n");
readEdge(ReadCount++, PB, PB.getEdge(0, &F->getEntryBlock()), Counters);
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
TerminatorInst *TI = BB->getTerminator();
for (unsigned s = 0, e = TI->getNumSuccessors(); s != e; ++s) {
readEdge(ReadCount++, PB, PB.getEdge(BB,TI->getSuccessor(s)),
Counters);
}
}
}
return ReadCount;
}
bool MLStatic::runOnModule(Module &M) {
for(Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
errs()<<"---Running for Function:"<<F->getName()<<"\n";
DT = &getAnalysis<DominatorTree>(*F);
BasicBlock *BB = &(F->getEntryBlock());
tracePath(BB);
}
DEBUG(dbgs()<<"PATH collection Complete\n");
for(Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
for(int i=0;i<pathCollecn2[F].size();++i){
for(int j=0;j<pathCollecn2[F][i].size();++j){
DEBUG(dbgs()<<pathCollecn2[F][i][j]->getName()<<" ");
}
DEBUG(dbgs()<<"\n");
}
}
expandPathAll(M);
return false;
}
bool PrintFunctionNames::runOnModule(Module &M) {
errs() << "Pass PrintFunctionNames\n";
// iterating through functions
for(Module::iterator f = M.begin(), fe = M.end(); f != fe; f++) {
if (!f->isDeclaration()) {
// finding and printing file information
BasicBlock &block = f->getEntryBlock();
Instruction &inst = block.front();
if (MDNode *node = inst.getMetadata("dbg")) {
DILocation loc(node);
errs() << "File name: " << loc.getFilename() << "\t";
}
}
// finding and printing function informatonion
errs() << "Function name: " << f->getName() << "\n";
}
return false;
}
bool OptimalEdgeProfiler::runOnModule(Module &M) {
Function *Main = M.getFunction("main");
if (Main == 0) {
errs() << "WARNING: cannot insert edge profiling into a module"
<< " with no main function!\n";
return false; // No main, no instrumentation!
}
// NumEdges counts all the edges that may be instrumented. Later on its
// decided which edges to actually instrument, to achieve optimal profiling.
// For the entry block a virtual edge (0,entry) is reserved, for each block
// with no successors an edge (BB,0) is reserved. These edges are necessary
// to calculate a truly optimal maximum spanning tree and thus an optimal
// instrumentation.
unsigned NumEdges = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
// Reserve space for (0,entry) edge.
++NumEdges;
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
// Keep track of which blocks need to be instrumented. We don't want to
// instrument blocks that are added as the result of breaking critical
// edges!
if (BB->getTerminator()->getNumSuccessors() == 0) {
// Reserve space for (BB,0) edge.
++NumEdges;
} else {
NumEdges += BB->getTerminator()->getNumSuccessors();
}
}
}
// In the profiling output a counter for each edge is reserved, but only few
// are used. This is done to be able to read back in the profile without
// calulating the maximum spanning tree again, instead each edge counter that
// is not used is initialised with -1 to signal that this edge counter has to
// be calculated from other edge counters on reading the profile info back
// in.
const Type *Int32 = Type::getInt32Ty(M.getContext());
const ArrayType *ATy = ArrayType::get(Int32, NumEdges);
GlobalVariable *Counters =
new GlobalVariable(M, ATy, false, GlobalValue::InternalLinkage,
Constant::getNullValue(ATy), "OptEdgeProfCounters");
NumEdgesInserted = 0;
std::vector<Constant*> Initializer(NumEdges);
Constant* Zero = ConstantInt::get(Int32, 0);
Constant* Uncounted = ConstantInt::get(Int32, ProfileInfoLoader::Uncounted);
// Instrument all of the edges not in MST...
unsigned i = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
DEBUG(dbgs()<<"Working on "<<F->getNameStr()<<"\n");
// Calculate a Maximum Spanning Tree with the edge weights determined by
// ProfileEstimator. ProfileEstimator also assign weights to the virtual
// edges (0,entry) and (BB,0) (for blocks with no successors) and this
// edges also participate in the maximum spanning tree calculation.
// The third parameter of MaximumSpanningTree() has the effect that not the
// actual MST is returned but the edges _not_ in the MST.
ProfileInfo::EdgeWeights ECs =
getAnalysis<ProfileInfo>(*F).getEdgeWeights(F);
std::vector<ProfileInfo::EdgeWeight> EdgeVector(ECs.begin(), ECs.end());
MaximumSpanningTree<BasicBlock> MST (EdgeVector);
std::stable_sort(MST.begin(),MST.end());
// Check if (0,entry) not in the MST. If not, instrument edge
// (IncrementCounterInBlock()) and set the counter initially to zero, if
// the edge is in the MST the counter is initialised to -1.
BasicBlock *entry = &(F->getEntryBlock());
ProfileInfo::Edge edge = ProfileInfo::getEdge(0,entry);
if (!std::binary_search(MST.begin(), MST.end(), edge)) {
printEdgeCounter(edge,entry,i);
IncrementCounterInBlock(entry, i, Counters); ++NumEdgesInserted;
Initializer[i++] = (Zero);
} else{
Initializer[i++] = (Uncounted);
}
// InsertedBlocks contains all blocks that were inserted for splitting an
// edge, this blocks do not have to be instrumented.
DenseSet<BasicBlock*> InsertedBlocks;
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
// Check if block was not inserted and thus does not have to be
// instrumented.
if (InsertedBlocks.count(BB)) continue;
// Okay, we have to add a counter of each outgoing edge not in MST. If
// the outgoing edge is not critical don't split it, just insert the
// counter in the source or destination of the edge. Also, if the block
// has no successors, the virtual edge (BB,0) is processed.
TerminatorInst *TI = BB->getTerminator();
if (TI->getNumSuccessors() == 0) {
ProfileInfo::Edge edge = ProfileInfo::getEdge(BB,0);
if (!std::binary_search(MST.begin(), MST.end(), edge)) {
printEdgeCounter(edge,BB,i);
IncrementCounterInBlock(BB, i, Counters); ++NumEdgesInserted;
Initializer[i++] = (Zero);
} else{
Initializer[i++] = (Uncounted);
}
}
for (unsigned s = 0, e = TI->getNumSuccessors(); s != e; ++s) {
BasicBlock *Succ = TI->getSuccessor(s);
ProfileInfo::Edge edge = ProfileInfo::getEdge(BB,Succ);
if (!std::binary_search(MST.begin(), MST.end(), edge)) {
// If the edge is critical, split it.
bool wasInserted = SplitCriticalEdge(TI, s, this);
Succ = TI->getSuccessor(s);
if (wasInserted)
InsertedBlocks.insert(Succ);
// Okay, we are guaranteed that the edge is no longer critical. If
// we only have a single successor, insert the counter in this block,
// otherwise insert it in the successor block.
if (TI->getNumSuccessors() == 1) {
// Insert counter at the start of the block
printEdgeCounter(edge,BB,i);
IncrementCounterInBlock(BB, i, Counters); ++NumEdgesInserted;
} else {
// Insert counter at the start of the block
printEdgeCounter(edge,Succ,i);
IncrementCounterInBlock(Succ, i, Counters); ++NumEdgesInserted;
}
Initializer[i++] = (Zero);
} else {
Initializer[i++] = (Uncounted);
}
}
}
}
// Check if the number of edges counted at first was the number of edges we
// considered for instrumentation.
assert(i==NumEdges && "the number of edges in counting array is wrong");
// Assing the now completely defined initialiser to the array.
Constant *init = ConstantArray::get(ATy, Initializer);
Counters->setInitializer(init);
// Add the initialization call to main.
InsertProfilingInitCall(Main, "llvm_start_opt_edge_profiling", Counters);
return true;
}
bool EdgeProfiler::runOnModule(Module &M) {
Function *Main = M.getFunction("main");
if (Main == 0) {
errs() << "WARNING: cannot insert edge profiling into a module"
<< " with no main function!\n";
return false; // No main, no instrumentation!
}
std::set<BasicBlock*> BlocksToInstrument;
unsigned NumEdges = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
// Reserve space for (0,entry) edge.
++NumEdges;
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
// Keep track of which blocks need to be instrumented. We don't want to
// instrument blocks that are added as the result of breaking critical
// edges!
BlocksToInstrument.insert(BB);
NumEdges += BB->getTerminator()->getNumSuccessors();
}
}
Type *ATy = ArrayType::get(Type::getInt32Ty(M.getContext()), NumEdges);
GlobalVariable *Counters =
new GlobalVariable(M, ATy, false, GlobalValue::InternalLinkage,
Constant::getNullValue(ATy), "EdgeProfCounters");
NumEdgesInserted = NumEdges;
// Instrument all of the edges...
unsigned i = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
// Create counter for (0,entry) edge.
IncrementCounterInBlock(&F->getEntryBlock(), i++, Counters);
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
if (BlocksToInstrument.count(BB)) { // Don't instrument inserted blocks
// Okay, we have to add a counter of each outgoing edge. If the
// outgoing edge is not critical don't split it, just insert the counter
// in the source or destination of the edge.
TerminatorInst *TI = BB->getTerminator();
for (unsigned s = 0, e = TI->getNumSuccessors(); s != e; ++s) {
// If the edge is critical, split it.
SplitCriticalEdge(TI, s, this);
// Okay, we are guaranteed that the edge is no longer critical. If we
// only have a single successor, insert the counter in this block,
// otherwise insert it in the successor block.
if (TI->getNumSuccessors() == 1) {
// Insert counter at the start of the block
IncrementCounterInBlock(BB, i++, Counters, false);
} else {
// Insert counter at the start of the block
IncrementCounterInBlock(TI->getSuccessor(s), i++, Counters);
}
}
}
}
// Add the initialization call to main.
InsertProfilingInitCall(Main, "llvm_start_edge_profiling", Counters);
return true;
}
bool LoaderPass::runOnModule(Module &M) {
ProfileInfoLoader PIL("profile-loader", Filename);
EdgeInformation.clear();
std::vector<uint64_t> Counters64 = PIL.getRawEdgeCounts();
if (Counters64.size() > 0) {
ReadCount = 0;
std::vector<uint64_t>& Counters = Counters64;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
DEBUG(dbgs() << "Working on " << F->getName() << "\n");
readEdge(getEdge(0,&F->getEntryBlock()), Counters);
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
TerminatorInst *TI = BB->getTerminator();
for (unsigned s = 0, e = TI->getNumSuccessors(); s != e; ++s) {
readEdge(getEdge(BB,TI->getSuccessor(s)), Counters);
}
}
}
if (ReadCount != Counters.size()) {
errs() << "WARNING: profile information is inconsistent with "
<< "the current program!\n";
}
NumEdgesRead = ReadCount;
}
#if 0
std::vector<unsigned> Counters = PIL.getRawOptimalEdgeCounts();
if (Counters.size() > 0) {
ReadCount = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
DEBUG(dbgs() << "Working on " << F->getName() << "\n");
readEdge(getEdge(0,&F->getEntryBlock()), Counters);
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
TerminatorInst *TI = BB->getTerminator();
if (TI->getNumSuccessors() == 0) {
readEdge(getEdge(BB,0), Counters);
}
for (unsigned s = 0, e = TI->getNumSuccessors(); s != e; ++s) {
readEdge(getEdge(BB,TI->getSuccessor(s)), Counters);
}
}
while (SpanningTree.size() > 0) {
unsigned size = SpanningTree.size();
BBisUnvisited.clear();
for (std::set<Edge>::iterator ei = SpanningTree.begin(),
ee = SpanningTree.end(); ei != ee; ++ei) {
BBisUnvisited.insert(ei->first);
BBisUnvisited.insert(ei->second);
}
while (BBisUnvisited.size() > 0) {
recurseBasicBlock(*BBisUnvisited.begin());
}
if (SpanningTree.size() == size) {
DEBUG(dbgs()<<"{");
for (std::set<Edge>::iterator ei = SpanningTree.begin(),
ee = SpanningTree.end(); ei != ee; ++ei) {
DEBUG(dbgs()<< *ei <<",");
}
assert(0 && "No edge calculated!");
}
}
}
if (ReadCount != Counters.size()) {
errs() << "WARNING: profile information is inconsistent with "
<< "the current program!\n";
}
NumEdgesRead = ReadCount;
}
#endif
BlockInformation.clear();
Counters64 = PIL.getRawBlockCounts();
if (Counters64.size() > 0) {
std::vector<uint64_t>& Counters = Counters64;
ReadCount = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
if (ReadCount < Counters.size())
// Here the data realm changes from the unsigned of the file to the
// double of the ProfileInfo. This conversion is save because we know
// that everything thats representable in unsinged is also
// representable in double.
BlockInformation[F][BB] = (double)Counters[ReadCount++];
}
if (ReadCount != Counters.size()) {
errs() << "WARNING: profile information is inconsistent with "
<< "the current program!\n";
}
}
FunctionInformation.clear();
std::vector<unsigned> Counters = PIL.getRawFunctionCounts();
if (Counters.size() > 0) {
ReadCount = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
if (ReadCount < Counters.size())
// Here the data realm changes from the unsigned of the file to the
// double of the ProfileInfo. This conversion is save because we know
// that everything thats representable in unsinged is also
// representable in double.
FunctionInformation[F] = (double)Counters[ReadCount++];
}
if (ReadCount != Counters.size()) {
errs() << "WARNING: profile information is inconsistent with "
<< "the current program!\n";
}
}
ValueInformation.clear();
Counters = PIL.getRawValueCounts();
if(Counters.size() > 0) {
ReadCount = 0;
for(Module::iterator F = M.begin(),E = M.end(); F!= E; ++F) {
if (F->isDeclaration()) continue;
for(inst_iterator I = inst_begin(F),IE = inst_end(F); I!=IE; ++I){
CallInst* Call = dyn_cast<CallInst>(&*I);
if(!Call) continue;
if(Call->getCalledValue()->getName() != "llvm_profiling_trap_value") continue;
unsigned index = getTrapedIndex(Call);
ValueCounts Ins;
Ins.Nums = Counters[index];
const std::vector<int>& content = PIL.getRawValueContent(index);
Ins.flags = (ProfilingFlags)content.front();
Ins.Contents.resize(content.size()-1);
copy(content.begin()+1,content.end(),Ins.Contents.begin());
//should NOT insert two values into one cell.
ValueInformation[Call] = Ins;
}
}
}
SLGInformation.clear();
Counters = PIL.getRawSLGCounts();
if(Counters.size() > 0) {
unsigned MaxStore = std::accumulate(Counters.begin(), Counters.end(), 0, sig_max);
std::vector<const Instruction*> Cache(MaxStore+1);
ReadCount = 0;
unsigned load_idx = 0, store_idx = 1;
for(Module::iterator F = M.begin(), E = M.end(); F!=E; ++F){
for(inst_iterator I = inst_begin(F), IE = inst_end(F); I!=IE; ++I){
if(!isa<StoreInst>(&*I) && !isa<LoadInst>(&*I)) continue;
if(lle::access_global_variable(&*I)){
unsigned index = 0;
Instruction* SLI = &*I;
if(isa<StoreInst>(&*I)){
index = store_idx++;
}else if(LoadInst* LI = dyn_cast<LoadInst>(&*I)){
SLGInformation[LI] = std::make_pair(load_idx, (Instruction*)NULL);
index = Counters[load_idx++];
if(index == 0 || index == ~0U/*unsigned -1*/){
continue;
}
}
if(index >= Cache.size()) continue;
if(Cache[index]){
const Instruction* SLJ = Cache[index];
if(isa<StoreInst>(SLJ) && isa<LoadInst>(SLI)) SLGInformation[SLI].second = SLJ;
else if(isa<StoreInst>(SLI) && isa<LoadInst>(SLJ)) SLGInformation[SLJ].second = SLI;
else
assert(0 && "It shouldn't happen");
}else
Cache[index] = &*I;
}
}
}
}
MPInformation.clear();
Counters = PIL.getRawMPICounts();
if(Counters.size() > 0) {
ReadCount = 0;
for(auto F = M.begin(), E = M.end(); F!=E; ++F){
for(auto I = inst_begin(F), IE = inst_end(F); I!=IE; ++I){
CallInst* CI = dyn_cast<CallInst>(&*I);
if(CI == NULL) continue;
if(lle::get_mpi_count_idx(CI)){
MPInformation[CI] = std::make_pair(ReadCount, Counters[ReadCount]);
++ReadCount;
}
}
}
}
MPIFullInformation.clear();
Counters = PIL.getRawMPIFullCounts();
if(Counters.size() > 0) {
ReadCount = 0;
for(auto F = M.begin(), E = M.end(); F!=E; ++F){
for(auto I = inst_begin(F), IE = inst_end(F); I!=IE; ++I){
CallInst* CI = dyn_cast<CallInst>(&*I);
if(CI == NULL) continue;
if(lle::get_mpi_count_idx(CI)){
MPIFullInformation[CI] = std::make_pair(ReadCount, Counters[ReadCount]);
++ReadCount;
}
}
}
}
return false;
}
// the verifier iterates through each path to gather the total
// number of edge frequencies
bool PathProfileVerifier::runOnModule (Module &M) {
PathProfileInfo& pathProfileInfo = getAnalysis<PathProfileInfo>();
// setup a data structure to map path edges which index an
// array of edge counters
NestedBlockToIndexMap arrayMap;
unsigned i = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
arrayMap[0][F->begin()][0] = i++;
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
TerminatorInst *TI = BB->getTerminator();
unsigned duplicate = 0;
BasicBlock* prev = 0;
for (unsigned s = 0, e = TI->getNumSuccessors(); s != e;
prev = TI->getSuccessor(s), ++s) {
if (prev == TI->getSuccessor(s))
duplicate++;
else duplicate = 0;
arrayMap[BB][TI->getSuccessor(s)][duplicate] = i++;
}
}
}
std::vector<unsigned> edgeArray(i);
// iterate through each path and increment the edge counters as needed
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
pathProfileInfo.setCurrentFunction(F);
DEBUG(dbgs() << "function '" << F->getName() << "' ran "
<< pathProfileInfo.pathsRun()
<< "/" << pathProfileInfo.getPotentialPathCount()
<< " potential paths\n");
for( ProfilePathIterator nextPath = pathProfileInfo.pathBegin(),
endPath = pathProfileInfo.pathEnd();
nextPath != endPath; nextPath++ ) {
ProfilePath* currentPath = nextPath->second;
ProfilePathEdgeVector* pev = currentPath->getPathEdges();
DEBUG(dbgs () << "path #" << currentPath->getNumber() << ": "
<< currentPath->getCount() << "\n");
// setup the entry edge (normally path profiling doesn't care about this)
if (currentPath->getFirstBlockInPath() == &F->getEntryBlock())
edgeArray[arrayMap[0][currentPath->getFirstBlockInPath()][0]]
+= currentPath->getCount();
for( ProfilePathEdgeIterator nextEdge = pev->begin(),
endEdge = pev->end(); nextEdge != endEdge; nextEdge++ ) {
if (nextEdge != pev->begin())
DEBUG(dbgs() << " :: ");
BasicBlock* source = nextEdge->getSource();
BasicBlock* target = nextEdge->getTarget();
unsigned duplicateNumber = nextEdge->getDuplicateNumber();
DEBUG(dbgs() << source->getName() << " --{" << duplicateNumber
<< "}--> " << target->getName());
// Ensure all the referenced edges exist
// TODO: make this a separate function
if( !arrayMap.count(source) ) {
errs() << " error [" << F->getName() << "()]: source '"
<< source->getName()
<< "' does not exist in the array map.\n";
} else if( !arrayMap[source].count(target) ) {
errs() << " error [" << F->getName() << "()]: target '"
<< target->getName()
<< "' does not exist in the array map.\n";
} else if( !arrayMap[source][target].count(duplicateNumber) ) {
errs() << " error [" << F->getName() << "()]: edge "
<< source->getName() << " -> " << target->getName()
<< " duplicate number " << duplicateNumber
<< " does not exist in the array map.\n";
} else {
edgeArray[arrayMap[source][target][duplicateNumber]]
+= currentPath->getCount();
}
}
DEBUG(errs() << "\n");
delete pev;
}
}
std::string errorInfo;
std::string filename = EdgeProfileFilename;
// Open a handle to the file
FILE* edgeFile = fopen(filename.c_str(),"wb");
if (!edgeFile) {
errs() << "error: unable to open file '" << filename << "' for output.\n";
return false;
}
errs() << "Generating edge profile '" << filename << "' ...\n";
// write argument info
unsigned type = ArgumentInfo;
unsigned num = pathProfileInfo.argList.size();
int zeros = 0;
fwrite(&type,sizeof(unsigned),1,edgeFile);
fwrite(&num,sizeof(unsigned),1,edgeFile);
fwrite(pathProfileInfo.argList.c_str(),1,num,edgeFile);
if (num&3)
fwrite(&zeros, 1, 4-(num&3), edgeFile);
type = EdgeInfo;
num = edgeArray.size();
fwrite(&type,sizeof(unsigned),1,edgeFile);
fwrite(&num,sizeof(unsigned),1,edgeFile);
// write each edge to the file
for( std::vector<unsigned>::iterator s = edgeArray.begin(),
e = edgeArray.end(); s != e; s++)
fwrite(&*s, sizeof (unsigned), 1, edgeFile);
fclose (edgeFile);
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->getValueType(), 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();
}
// Copy GVMap over to a standard value map.
ValueToValueMapTy VM;
for (auto I = GVMap.begin(), E = GVMap.end(); I != E; ++I)
VM[I->first] = I->second;
// Walk through the metadata section and update the debug information
// associated with the global variables in the default address space.
for (NamedMDNode &I : M.named_metadata()) {
remapNamedMDNode(VM, &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 LoaderPass::runOnModule(Module &M) {
ProfileInfoLoader PIL("profile-loader", Filename, M);
EdgeInformation.clear();
std::vector<unsigned> Counters = PIL.getRawEdgeCounts();
if (Counters.size() > 0) {
ReadCount = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
DEBUG(dbgs()<<"Working on "<<F->getNameStr()<<"\n");
readEdge(getEdge(0,&F->getEntryBlock()), Counters);
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
TerminatorInst *TI = BB->getTerminator();
for (unsigned s = 0, e = TI->getNumSuccessors(); s != e; ++s) {
readEdge(getEdge(BB,TI->getSuccessor(s)), Counters);
}
}
}
if (ReadCount != Counters.size()) {
errs() << "WARNING: profile information is inconsistent with "
<< "the current program!\n";
}
NumEdgesRead = ReadCount;
}
Counters = PIL.getRawOptimalEdgeCounts();
if (Counters.size() > 0) {
ReadCount = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
DEBUG(dbgs()<<"Working on "<<F->getNameStr()<<"\n");
readEdge(getEdge(0,&F->getEntryBlock()), Counters);
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
TerminatorInst *TI = BB->getTerminator();
if (TI->getNumSuccessors() == 0) {
readEdge(getEdge(BB,0), Counters);
}
for (unsigned s = 0, e = TI->getNumSuccessors(); s != e; ++s) {
readEdge(getEdge(BB,TI->getSuccessor(s)), Counters);
}
}
while (SpanningTree.size() > 0) {
unsigned size = SpanningTree.size();
BBisUnvisited.clear();
for (std::set<Edge>::iterator ei = SpanningTree.begin(),
ee = SpanningTree.end(); ei != ee; ++ei) {
BBisUnvisited.insert(ei->first);
BBisUnvisited.insert(ei->second);
}
while (BBisUnvisited.size() > 0) {
recurseBasicBlock(*BBisUnvisited.begin());
}
if (SpanningTree.size() == size) {
DEBUG(dbgs()<<"{");
for (std::set<Edge>::iterator ei = SpanningTree.begin(),
ee = SpanningTree.end(); ei != ee; ++ei) {
DEBUG(dbgs()<< *ei <<",");
}
assert(0 && "No edge calculated!");
}
}
}
if (ReadCount != Counters.size()) {
errs() << "WARNING: profile information is inconsistent with "
<< "the current program!\n";
}
NumEdgesRead = ReadCount;
}
BlockInformation.clear();
Counters = PIL.getRawBlockCounts();
if (Counters.size() > 0) {
ReadCount = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
if (ReadCount < Counters.size())
// Here the data realm changes from the unsigned of the file to the
// double of the ProfileInfo. This conversion is save because we know
// that everything thats representable in unsinged is also
// representable in double.
BlockInformation[F][BB] = (double)Counters[ReadCount++];
}
if (ReadCount != Counters.size()) {
errs() << "WARNING: profile information is inconsistent with "
<< "the current program!\n";
}
}
FunctionInformation.clear();
Counters = PIL.getRawFunctionCounts();
if (Counters.size() > 0) {
ReadCount = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
if (ReadCount < Counters.size())
// Here the data realm changes from the unsigned of the file to the
// double of the ProfileInfo. This conversion is save because we know
// that everything thats representable in unsinged is also
// representable in double.
FunctionInformation[F] = (double)Counters[ReadCount++];
}
if (ReadCount != Counters.size()) {
errs() << "WARNING: profile information is inconsistent with "
<< "the current program!\n";
}
}
return false;
}
//
// Method: runOnModule()
//
// Description:
// Entry point for this LLVM pass.
// Clone functions that take LoadInsts as arguments
//
// 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 LoadArgs::runOnModule(Module& M) {
std::map<std::pair<Function*, const Type * > , Function* > fnCache;
bool changed;
do {
changed = false;
for (Module::iterator Func = M.begin(); Func != M.end(); ++Func) {
for (Function::iterator B = Func->begin(), FE = Func->end(); B != FE; ++B) {
for (BasicBlock::iterator I = B->begin(), BE = B->end(); I != BE;) {
CallInst *CI = dyn_cast<CallInst>(I++);
if(!CI)
continue;
if(CI->hasByValArgument())
continue;
// if the CallInst calls a function, that is externally defined,
// or might be changed, ignore this call site.
Function *F = CI->getCalledFunction();
if (!F || (F->isDeclaration() || F->mayBeOverridden()))
continue;
if(F->hasStructRetAttr())
continue;
if(F->isVarArg())
continue;
// find the argument we must replace
Function::arg_iterator ai = F->arg_begin(), ae = F->arg_end();
unsigned argNum = 0;
for(; argNum < CI->getNumArgOperands();argNum++, ++ai) {
// do not care about dead arguments
if(ai->use_empty())
continue;
if(F->getAttributes().getParamAttributes(argNum).hasAttrSomewhere(Attribute::SExt) ||
F->getAttributes().getParamAttributes(argNum).hasAttrSomewhere(Attribute::ZExt))
continue;
if (isa<LoadInst>(CI->getArgOperand(argNum)))
break;
}
// if no argument was a GEP operator to be changed
if(ai == ae)
continue;
LoadInst *LI = dyn_cast<LoadInst>(CI->getArgOperand(argNum));
Instruction * InsertPt = &(Func->getEntryBlock().front());
AllocaInst *NewVal = new AllocaInst(LI->getType(), "",InsertPt);
StoreInst *Copy = new StoreInst(LI, NewVal);
Copy->insertAfter(LI);
/*if(LI->getParent() != CI->getParent())
continue;
// Also check that there is no store after the load.
// TODO: Check if the load/store do not alias.
BasicBlock::iterator bii = LI->getParent()->begin();
Instruction *BII = bii;
while(BII != LI) {
++bii;
BII = bii;
}
while(BII != CI) {
if(isa<StoreInst>(BII))
break;
++bii;
BII = bii;
}
if(isa<StoreInst>(bii)){
continue;
}*/
// Construct the new Type
// Appends the struct Type at the beginning
std::vector<Type*>TP;
for(unsigned c = 0; c < CI->getNumArgOperands();c++) {
if(c == argNum)
TP.push_back(LI->getPointerOperand()->getType());
TP.push_back(CI->getArgOperand(c)->getType());
}
//return type is same as that of original instruction
FunctionType *NewFTy = FunctionType::get(CI->getType(), TP, false);
numSimplified++;
//if(numSimplified > 1000)
//return true;
Function *NewF;
std::map<std::pair<Function*, const Type* > , Function* >::iterator Test;
Test = fnCache.find(std::make_pair(F, NewFTy));
if(Test != fnCache.end()) {
NewF = Test->second;
} else {
NewF = Function::Create(NewFTy,
GlobalValue::InternalLinkage,
F->getName().str() + ".TEST",
&M);
fnCache[std::make_pair(F, NewFTy)] = NewF;
Function::arg_iterator NI = NewF->arg_begin();
ValueToValueMapTy ValueMap;
unsigned count = 0;
for (Function::arg_iterator II = F->arg_begin(); NI != NewF->arg_end(); ++count, ++NI) {
if(count == argNum) {
NI->setName("LDarg");
continue;
}
ValueMap[II] = NI;
NI->setName(II->getName());
NI->addAttr(F->getAttributes().getParamAttributes(II->getArgNo() + 1));
++II;
}
// Perform the cloning.
SmallVector<ReturnInst*,100> Returns;
CloneFunctionInto(NewF, F, ValueMap, false, Returns);
std::vector<Value*> fargs;
for(Function::arg_iterator ai = NewF->arg_begin(),
ae= NewF->arg_end(); ai != ae; ++ai) {
fargs.push_back(ai);
}
NewF->setAttributes(NewF->getAttributes().addAttributes(
F->getContext(), 0, F->getAttributes().getRetAttributes()));
NewF->setAttributes(NewF->getAttributes().addAttributes(
F->getContext(), ~0, F->getAttributes().getFnAttributes()));
//Get the point to insert the GEP instr.
Instruction *InsertPoint;
for (BasicBlock::iterator insrt = NewF->front().begin(); isa<AllocaInst>(InsertPoint = insrt); ++insrt) {;}
LoadInst *LI_new = new LoadInst(fargs.at(argNum), "", InsertPoint);
fargs.at(argNum+1)->replaceAllUsesWith(LI_new);
}
//this does not seem to be a good idea
AttributeSet NewCallPAL=AttributeSet();
// Get the initial attributes of the call
AttributeSet CallPAL = CI->getAttributes();
AttributeSet RAttrs = CallPAL.getRetAttributes();
AttributeSet FnAttrs = CallPAL.getFnAttributes();
if (!RAttrs.isEmpty())
NewCallPAL=NewCallPAL.addAttributes(F->getContext(),0, RAttrs);
SmallVector<Value*, 8> Args;
for(unsigned j =0;j<CI->getNumArgOperands();j++) {
if(j == argNum) {
Args.push_back(NewVal);
}
Args.push_back(CI->getArgOperand(j));
// position in the NewCallPAL
AttributeSet Attrs = CallPAL.getParamAttributes(j+1);
if (!Attrs.isEmpty())
NewCallPAL=NewCallPAL.addAttributes(F->getContext(),Args.size(), Attrs);
}
// Create the new attributes vec.
if (!FnAttrs.isEmpty())
NewCallPAL=NewCallPAL.addAttributes(F->getContext(),~0, FnAttrs);
CallInst *CallI = CallInst::Create(NewF,Args,"", CI);
CallI->setCallingConv(CI->getCallingConv());
CallI->setAttributes(NewCallPAL);
CI->replaceAllUsesWith(CallI);
CI->eraseFromParent();
changed = true;
}
}
}
} while(changed);
return true;
}
bool LoaderPass::runOnModule(Module &M) {
ProfileInfoLoader PIL("profile-loader", Filename, M);
EdgeInformation.clear();
std::vector<unsigned> ECs = PIL.getRawEdgeCounts();
if (ECs.size() > 0) {
unsigned ei = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
if (ei < ECs.size())
EdgeInformation[F][ProfileInfo::getEdge(0, &F->getEntryBlock())] +=
ECs[ei++];
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
// Okay, we have to add a counter of each outgoing edge. If the
// outgoing edge is not critical don't split it, just insert the counter
// in the source or destination of the edge.
TerminatorInst *TI = BB->getTerminator();
for (unsigned s = 0, e = TI->getNumSuccessors(); s != e; ++s) {
if (ei < ECs.size())
EdgeInformation[F][ProfileInfo::getEdge(BB, TI->getSuccessor(s))] +=
ECs[ei++];
}
}
}
if (ei != ECs.size()) {
cerr << "WARNING: profile information is inconsistent with "
<< "the current program!\n";
}
}
BlockInformation.clear();
std::vector<unsigned> BCs = PIL.getRawBlockCounts();
if (BCs.size() > 0) {
unsigned bi = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
if (bi < BCs.size())
BlockInformation[F][BB] = BCs[bi++];
}
if (bi != BCs.size()) {
cerr << "WARNING: profile information is inconsistent with "
<< "the current program!\n";
}
}
FunctionInformation.clear();
std::vector<unsigned> FCs = PIL.getRawFunctionCounts();
if (FCs.size() > 0) {
unsigned fi = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
if (fi < FCs.size())
FunctionInformation[F] = FCs[fi++];
}
if (fi != FCs.size()) {
cerr << "WARNING: profile information is inconsistent with "
<< "the current program!\n";
}
}
return false;
}