void TDDataStructures::ComputePostOrder(const Function &F,
DenseSet<DSGraph*> &Visited,
std::vector<DSGraph*> &PostOrder) {
if (F.isDeclaration()) return;
DSGraph* G = getOrCreateGraph(&F);
if (!Visited.insert(G).second) return;
// Recursively traverse all of the callee graphs.
svset<const Function*> Callees;
// Go through all of the callsites in this graph and find all callees
// Here we're trying to capture all possible callees so that we can ensure
// each function has all possible callers inlined into it.
for (DSGraph::fc_iterator CI = G->fc_begin(), E = G->fc_end();
CI != E; ++CI) {
// Direct calls are easy, no reason to look at DSCallGraph
// or anything to do with SCC's
if (CI->isDirectCall()) {
ComputePostOrder(*CI->getCalleeFunc(), Visited, PostOrder);
}
else {
// Otherwise, ask the DSCallGraph for the full set of possible
// callees for this callsite.
// This includes all members of the SCC's of those callees,
// and well as others in F's SCC, since we must assume
// any indirect call might be intra-SCC.
callgraph.addFullFunctionSet(CI->getCallSite(), Callees);
}
}
for (svset<const Function*>::iterator I = Callees.begin(),
E = Callees.end(); I != E; ++I)
ComputePostOrder(**I, Visited, PostOrder);
PostOrder.push_back(G);
}
/// InlineCallersIntoGraph - Inline all of the callers of the specified DS graph
/// into it, then recompute completeness of nodes in the resultant graph.
void TDDataStructures::InlineCallersIntoGraph(DSGraph* DSG) {
// Inline caller graphs into this graph. First step, get the list of call
// sites that call into this graph.
std::vector<CallerCallEdge> EdgesFromCaller;
std::map<DSGraph*, std::vector<CallerCallEdge> >::iterator
CEI = CallerEdges.find(DSG);
if (CEI != CallerEdges.end()) {
std::swap(CEI->second, EdgesFromCaller);
CallerEdges.erase(CEI);
}
// Sort the caller sites to provide a by-caller-graph ordering.
std::sort(EdgesFromCaller.begin(), EdgesFromCaller.end());
// Merge information from the globals graph into this graph. FIXME: This is
// stupid. Instead of us cloning information from the GG into this graph,
// then having RemoveDeadNodes clone it back, we should do all of this as a
// post-pass over all of the graphs. We need to take cloning out of
// removeDeadNodes and gut removeDeadNodes at the same time first though. :(
cloneGlobalsInto(DSG, DSGraph::DontCloneCallNodes |
DSGraph::DontCloneAuxCallNodes);
DEBUG(errs() << "[TD] Inlining callers into '"
<< DSG->getFunctionNames() << "'\n");
DSG->maskIncompleteMarkers();
// Iteratively inline caller graphs into this graph.
while (!EdgesFromCaller.empty()) {
DSGraph* CallerGraph = EdgesFromCaller.back().CallerGraph;
// Iterate through all of the call sites of this graph, cloning and merging
// any nodes required by the call.
ReachabilityCloner RC(DSG, CallerGraph,
DSGraph::DontCloneCallNodes |
DSGraph::DontCloneAuxCallNodes);
// Inline all call sites from this caller graph.
do {
const DSCallSite &CS = *EdgesFromCaller.back().CS;
const Function &CF = *EdgesFromCaller.back().CalledFunction;
DEBUG(errs() << " [TD] Inlining graph into Fn '"
<< CF.getName().str() << "' from ");
if (CallerGraph->getReturnNodes().empty()) {
DEBUG(errs() << "SYNTHESIZED INDIRECT GRAPH");
} else {
DEBUG(errs() << "Fn '" << CS.getCallSite().getInstruction()->
getParent()->getParent()->getName().str() << "'");
}
DEBUG(errs() << ": " << CF.getFunctionType()->getNumParams()
<< " args\n");
// Get the formal argument and return nodes for the called function and
// merge them with the cloned subgraph.
DSCallSite T1 = DSG->getCallSiteForArguments(CF);
RC.mergeCallSite(T1, CS);
++NumTDInlines;
EdgesFromCaller.pop_back();
} while (!EdgesFromCaller.empty() &&
EdgesFromCaller.back().CallerGraph == CallerGraph);
}
// Next, now that this graph is finalized, we need to recompute the
// incompleteness markers for this graph and remove unreachable nodes.
// If any of the functions is externally callable, treat everything in its
// SCC as externally callable.
bool isExternallyCallable = false;
for (DSGraph::retnodes_iterator I = DSG->retnodes_begin(),
E = DSG->retnodes_end(); I != E; ++I)
if (ExternallyCallable.count(I->first)) {
isExternallyCallable = true;
break;
}
// Recompute the Incomplete markers. Depends on whether args are complete
unsigned IncFlags = DSGraph::IgnoreFormalArgs;
IncFlags |= DSGraph::IgnoreGlobals | DSGraph::MarkVAStart;
DSG->markIncompleteNodes(IncFlags);
// If this graph contains functions that are externally callable, now is the time to mark
// their arguments and return values as external. At this point TD is inlining all caller information,
// and that means External callers too.
unsigned ExtFlags
= isExternallyCallable ? DSGraph::MarkFormalsExternal : DSGraph::DontMarkFormalsExternal;
DSG->computeExternalFlags(ExtFlags);
DSG->computeIntPtrFlags();
cloneIntoGlobals(DSG, DSGraph::DontCloneCallNodes |
DSGraph::DontCloneAuxCallNodes);
//
// Delete dead nodes. Treat globals that are unreachable as dead also.
//
// FIXME:
// Do not delete unreachable globals as the comment describes. For its
// alignment checks on the results of load instructions, SAFECode must be
// able to find the DSNode of both the result of the load as well as the
// pointer dereferenced by the load. If we remove unreachable globals, then
// if the dereferenced pointer is a global, its DSNode will not reachable
// from the local graph's scalar map, and chaos ensues.
//
// So, for now, just remove dead nodes but leave the globals alone.
//
DSG->removeDeadNodes(0);
// We are done with computing the current TD Graph! Finally, before we can
// finish processing this function, we figure out which functions it calls and
// records these call graph edges, so that we have them when we process the
// callee graphs.
if (DSG->fc_begin() == DSG->fc_end()) return;
// Loop over all the call sites and all the callees at each call site, and add
// edges to the CallerEdges structure for each callee.
for (DSGraph::fc_iterator CI = DSG->fc_begin(), E = DSG->fc_end();
CI != E; ++CI) {
// Handle direct calls efficiently.
if (CI->isDirectCall()) {
if (!CI->getCalleeFunc()->isDeclaration() &&
!DSG->getReturnNodes().count(CI->getCalleeFunc()))
CallerEdges[getOrCreateGraph(CI->getCalleeFunc())]
.push_back(CallerCallEdge(DSG, &*CI, CI->getCalleeFunc()));
continue;
}
svset<const Function*> AllCallees;
std::vector<const Function*> Callees;
// Get the list of callees
callgraph.addFullFunctionSet(CI->getCallSite(), AllCallees);
// Filter all non-declarations, and calls within this DSGraph
for (svset<const Function*>::iterator I = AllCallees.begin(),
E = AllCallees.end(); I != E; ++I) {
const Function *F = *I;
if (!F->isDeclaration() && getDSGraph(**I) != DSG)
Callees.push_back(F);
}
AllCallees.clear();
// If there is exactly one callee from this call site, remember the edge in
// CallerEdges.
if (Callees.size() == 1) {
const Function * Callee = Callees[0];
CallerEdges[getOrCreateGraph(Callee)]
.push_back(CallerCallEdge(DSG, &*CI, Callee));
}
if (Callees.size() <= 1) continue;
// Otherwise, there are multiple callees from this call site, so it must be
// an indirect call. Chances are that there will be other call sites with
// this set of targets. If so, we don't want to do M*N inlining operations,
// so we build up a new, private, graph that represents the calls of all
// calls to this set of functions.
std::map<std::vector<const Function*>, DSGraph*>::iterator IndCallRecI =
IndCallMap.lower_bound(Callees);
// If we already have this graph, recycle it.
if (IndCallRecI != IndCallMap.end() && IndCallRecI->first == Callees) {
DEBUG(errs() << " [TD] *** Reuse of indcall graph for " << Callees.size()
<< " callees!\n");
DSGraph * IndCallGraph = IndCallRecI->second;
assert(IndCallGraph->getFunctionCalls().size() == 1);
// Merge the call into the CS already in the IndCallGraph
ReachabilityCloner RC(IndCallGraph, DSG, 0);
RC.mergeCallSite(IndCallGraph->getFunctionCalls().front(), *CI);
} else {
// Otherwise, create a new DSGraph to represent this.
DSGraph* IndCallGraph = new DSGraph(DSG->getGlobalECs(),
DSG->getDataLayout(), *TypeSS);
// Clone over the call into the new DSGraph
ReachabilityCloner RC(IndCallGraph, DSG, 0);
DSCallSite ClonedCS = RC.cloneCallSite(*CI);
// Add the cloned CS to the graph, as if it were an original call.
IndCallGraph->getFunctionCalls().push_back(ClonedCS);
// Save this graph for use later, should we need it.
IndCallRecI = IndCallMap.insert(IndCallRecI,
std::make_pair(Callees, IndCallGraph));
// Additionally, make sure that each of the callees inlines this graph
// exactly once.
DSCallSite *NCS = &IndCallGraph->getFunctionCalls().front();
for (unsigned i = 0, e = Callees.size(); i != e; ++i) {
DSGraph* CalleeGraph = getDSGraph(*Callees[i]);
if (CalleeGraph != DSG)
CallerEdges[CalleeGraph].push_back(CallerCallEdge(IndCallGraph, NCS,
Callees[i]));
}
}
}
}
/// InlineCallersIntoGraph - Inline all of the callers of the specified DS graph
/// into it, then recompute completeness of nodes in the resultant graph.
void TDDataStructures::InlineCallersIntoGraph(DSGraph* DSG) {
// Inline caller graphs into this graph. First step, get the list of call
// sites that call into this graph.
std::vector<CallerCallEdge> EdgesFromCaller;
std::map<DSGraph*, std::vector<CallerCallEdge> >::iterator
CEI = CallerEdges.find(DSG);
if (CEI != CallerEdges.end()) {
std::swap(CEI->second, EdgesFromCaller);
CallerEdges.erase(CEI);
}
// Sort the caller sites to provide a by-caller-graph ordering.
std::sort(EdgesFromCaller.begin(), EdgesFromCaller.end());
// Merge information from the globals graph into this graph. FIXME: This is
// stupid. Instead of us cloning information from the GG into this graph,
// then having RemoveDeadNodes clone it back, we should do all of this as a
// post-pass over all of the graphs. We need to take cloning out of
// removeDeadNodes and gut removeDeadNodes at the same time first though. :(
{
DSGraph* GG = DSG->getGlobalsGraph();
ReachabilityCloner RC(DSG, GG,
DSGraph::DontCloneCallNodes |
DSGraph::DontCloneAuxCallNodes);
for (DSScalarMap::global_iterator
GI = DSG->getScalarMap().global_begin(),
E = DSG->getScalarMap().global_end(); GI != E; ++GI)
RC.getClonedNH(GG->getNodeForValue(*GI));
}
DEBUG(errs() << "[TD] Inlining callers into '"
<< DSG->getFunctionNames() << "'\n");
// Iteratively inline caller graphs into this graph.
while (!EdgesFromCaller.empty()) {
DSGraph* CallerGraph = EdgesFromCaller.back().CallerGraph;
// Iterate through all of the call sites of this graph, cloning and merging
// any nodes required by the call.
ReachabilityCloner RC(DSG, CallerGraph,
DSGraph::DontCloneCallNodes |
DSGraph::DontCloneAuxCallNodes);
// Inline all call sites from this caller graph.
do {
const DSCallSite &CS = *EdgesFromCaller.back().CS;
const Function &CF = *EdgesFromCaller.back().CalledFunction;
DEBUG(errs() << " [TD] Inlining graph into Fn '"
<< CF.getNameStr() << "' from ");
if (CallerGraph->getReturnNodes().empty()) {
DEBUG(errs() << "SYNTHESIZED INDIRECT GRAPH");
} else {
DEBUG(errs() << "Fn '" << CS.getCallSite().getInstruction()->
getParent()->getParent()->getNameStr() << "'");
}
DEBUG(errs() << ": " << CF.getFunctionType()->getNumParams()
<< " args\n");
// Get the formal argument and return nodes for the called function and
// merge them with the cloned subgraph.
DSCallSite T1 = DSG->getCallSiteForArguments(CF);
RC.mergeCallSite(T1, CS);
++NumTDInlines;
EdgesFromCaller.pop_back();
} while (!EdgesFromCaller.empty() &&
EdgesFromCaller.back().CallerGraph == CallerGraph);
}
{
DSGraph* GG = DSG->getGlobalsGraph();
ReachabilityCloner RC(GG, DSG,
DSGraph::DontCloneCallNodes |
DSGraph::DontCloneAuxCallNodes);
for (DSScalarMap::global_iterator
GI = DSG->getScalarMap().global_begin(),
E = DSG->getScalarMap().global_end(); GI != E; ++GI)
RC.getClonedNH(DSG->getNodeForValue(*GI));
}
// Next, now that this graph is finalized, we need to recompute the
// incompleteness markers for this graph and remove unreachable nodes.
DSG->maskIncompleteMarkers();
// If any of the functions has incomplete incoming arguments, don't mark any
// of them as complete.
bool HasIncompleteArgs = false;
for (DSGraph::retnodes_iterator I = DSG->retnodes_begin(),
E = DSG->retnodes_end(); I != E; ++I)
if (ArgsRemainIncomplete.count(I->first)) {
HasIncompleteArgs = true;
break;
}
// Recompute the Incomplete markers. Depends on whether args are complete
unsigned Flags
= HasIncompleteArgs ? DSGraph::MarkFormalArgs : DSGraph::IgnoreFormalArgs;
Flags |= DSGraph::IgnoreGlobals | DSGraph::MarkVAStart;
DSG->markIncompleteNodes(Flags);
// Delete dead nodes. Treat globals that are unreachable as dead also.
DSG->removeDeadNodes(DSGraph::RemoveUnreachableGlobals);
// We are done with computing the current TD Graph! Finally, before we can
// finish processing this function, we figure out which functions it calls and
// records these call graph edges, so that we have them when we process the
// callee graphs.
if (DSG->fc_begin() == DSG->fc_end()) return;
// Loop over all the call sites and all the callees at each call site, and add
// edges to the CallerEdges structure for each callee.
for (DSGraph::fc_iterator CI = DSG->fc_begin(), E = DSG->fc_end();
CI != E; ++CI) {
// Handle direct calls efficiently.
if (CI->isDirectCall()) {
if (!CI->getCalleeFunc()->isDeclaration() &&
!DSG->getReturnNodes().count(CI->getCalleeFunc()))
CallerEdges[getOrFetchDSGraph(CI->getCalleeFunc())]
.push_back(CallerCallEdge(DSG, &*CI, CI->getCalleeFunc()));
continue;
}
Instruction *CallI = CI->getCallSite().getInstruction();
// For each function in the invoked function list at this call site...
calleeTy::iterator IPI =
callee.begin(CallI), IPE = callee.end(CallI);
// Skip over all calls to this graph (SCC calls).
while (IPI != IPE && getDSGraph(*IPI) == DSG)
++IPI;
// All SCC calls?
if (IPI == IPE) continue;
const Function *FirstCallee = *IPI;
++IPI;
// Skip over more SCC calls.
while (IPI != IPE && getDSGraph(*IPI) == DSG)
++IPI;
// If there is exactly one callee from this call site, remember the edge in
// CallerEdges.
if (IPI == IPE) {
if (!FirstCallee->isDeclaration())
CallerEdges[getOrFetchDSGraph(FirstCallee)]
.push_back(CallerCallEdge(DSG, &*CI, FirstCallee));
continue;
}
// Otherwise, there are multiple callees from this call site, so it must be
// an indirect call. Chances are that there will be other call sites with
// this set of targets. If so, we don't want to do M*N inlining operations,
// so we build up a new, private, graph that represents the calls of all
// calls to this set of functions.
std::vector<const Function*> Callees;
for (calleeTy::iterator I = callee.begin(CallI), E = callee.end(CallI);
I != E; ++I)
if (!(*I)->isDeclaration())
Callees.push_back(*I);
std::sort(Callees.begin(), Callees.end());
std::map<std::vector<const Function*>, DSGraph*>::iterator IndCallRecI =
IndCallMap.lower_bound(Callees);
DSGraph *IndCallGraph;
// If we already have this graph, recycle it.
if (IndCallRecI != IndCallMap.end() && IndCallRecI->first == Callees) {
DEBUG(errs() << " [TD] *** Reuse of indcall graph for " << Callees.size()
<< " callees!\n");
IndCallGraph = IndCallRecI->second;
} else {
// Otherwise, create a new DSGraph to represent this.
IndCallGraph = new DSGraph(DSG->getGlobalECs(), DSG->getTargetData(), GlobalsGraph);
// Make a nullary dummy call site, which will eventually get some content
// merged into it. The actual callee function doesn't matter here, so we
// just pass it something to keep the ctor happy.
std::vector<DSNodeHandle> ArgDummyVec;
DSCallSite DummyCS(CI->getCallSite(), DSNodeHandle(), Callees[0]/*dummy*/,
ArgDummyVec);
IndCallGraph->getFunctionCalls().push_back(DummyCS);
IndCallRecI = IndCallMap.insert(IndCallRecI,
std::make_pair(Callees, IndCallGraph));
// Additionally, make sure that each of the callees inlines this graph
// exactly once.
DSCallSite *NCS = &IndCallGraph->getFunctionCalls().front();
for (unsigned i = 0, e = Callees.size(); i != e; ++i) {
DSGraph* CalleeGraph = getDSGraph(Callees[i]);
if (CalleeGraph != DSG)
CallerEdges[CalleeGraph].push_back(CallerCallEdge(IndCallGraph, NCS,
Callees[i]));
}
}
// Now that we know which graph to use for this, merge the caller
// information into the graph, based on information from the call site.
ReachabilityCloner RC(IndCallGraph, DSG, 0);
RC.mergeCallSite(IndCallGraph->getFunctionCalls().front(), *CI);
}
}