1 //===- CallGraph.cpp - Build a Module's call graph ------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by the LLVM research group and is distributed under
6 // the University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This interface is used to build and manipulate a call graph, which is a very
11 // useful tool for interprocedural optimization.
13 // Every function in a module is represented as a node in the call graph. The
14 // callgraph node keeps track of which functions the are called by the function
15 // corresponding to the node.
17 // A call graph will contain nodes where the function that they correspond to is
18 // null. This 'external' node is used to represent control flow that is not
19 // represented (or analyzable) in the module. As such, the external node will
20 // have edges to functions with the following properties:
21 // 1. All functions in the module without internal linkage, since they could
22 // be called by functions outside of the our analysis capability.
23 // 2. All functions whose address is used for something more than a direct
24 // call, for example being stored into a memory location. Since they may
25 // be called by an unknown caller later, they must be tracked as such.
27 // Similarly, functions have a call edge to the external node iff:
28 // 1. The function is external, reflecting the fact that they could call
29 // anything without internal linkage or that has its address taken.
30 // 2. The function contains an indirect function call.
32 // As an extension in the future, there may be multiple nodes with a null
33 // function. These will be used when we can prove (through pointer analysis)
34 // that an indirect call site can call only a specific set of functions.
36 // Because of these properties, the CallGraph captures a conservative superset
37 // of all of the caller-callee relationships, which is useful for
40 // The CallGraph class also attempts to figure out what the root of the
41 // CallGraph is, which is currently does by looking for a function named 'main'.
42 // If no function named 'main' is found, the external node is used as the entry
43 // node, reflecting the fact that any function without internal linkage could
44 // be called into (which is common for libraries).
46 //===----------------------------------------------------------------------===//
48 #include "llvm/Analysis/CallGraph.h"
49 #include "llvm/Constants.h" // Remove when ConstantPointerRefs are gone
50 #include "llvm/Module.h"
51 #include "llvm/iOther.h"
52 #include "llvm/iTerminators.h"
53 #include "llvm/Support/CallSite.h"
54 #include "Support/STLExtras.h"
56 static RegisterAnalysis<CallGraph> X("callgraph", "Call Graph Construction");
58 static const char * const KnownExternalFunctions[] = {
59 // Low-level system calls
68 // Low-level stdc library functions
73 // Standard IO functions
136 // gettext functions used by libstdc++
149 /// ExternalFunctionDoesntCallIntoProgram - This hack is used to indicate to the
150 /// call graph that the specified external function is _KNOWN_ to not call back
151 /// into the program. This is important, because otherwise functions which call
152 /// "printf" for example, end up in a great big SCC that goes from the function
155 static bool ExternalFunctionDoesntCallIntoProgram(const std::string &Name) {
156 static std::vector<std::string> Funcs;
158 // First time this is called?
160 // Add a whole bunch of functions which are often used...
161 Funcs.insert(Funcs.end(), KnownExternalFunctions,
162 KnownExternalFunctions+
163 sizeof(KnownExternalFunctions)/sizeof(KnownExternalFunctions[0]));
164 // Sort the list for efficient access
165 std::sort(Funcs.begin(), Funcs.end());
168 if (Name.size() > 7 && !memcmp("__llvm_", Name.c_str(), 7))
171 // Binary search for the function name...
172 std::vector<std::string>::iterator I =
173 std::lower_bound(Funcs.begin(), Funcs.end(), Name);
176 return I != Funcs.end() && *I == Name;
181 // getNodeFor - Return the node for the specified function or create one if it
182 // does not already exist.
184 CallGraphNode *CallGraph::getNodeFor(Function *F) {
185 CallGraphNode *&CGN = FunctionMap[F];
188 assert((!F || F->getParent() == Mod) && "Function not in current module!");
189 return CGN = new CallGraphNode(F);
192 static bool isOnlyADirectCall(Function *F, CallSite CS) {
193 if (!CS.getInstruction()) return false;
194 for (CallSite::arg_iterator I = CS.arg_begin(), E = CS.arg_end(); I != E; ++I)
195 if (*I == F) return false;
199 // addToCallGraph - Add a function to the call graph, and link the node to all
200 // of the functions that it calls.
202 void CallGraph::addToCallGraph(Function *F) {
203 CallGraphNode *Node = getNodeFor(F);
205 // If this function has external linkage, anything could call it...
206 if (!F->hasInternalLinkage()) {
207 ExternalNode->addCalledFunction(Node);
209 // Found the entry point?
210 if (F->getName() == "main") {
212 Root = ExternalNode; // Found multiple external mains? Don't pick one.
214 Root = Node; // Found a main, keep track of it!
218 // If this function is not defined in this translation unit, it could call
220 if (F->isExternal() && !F->getIntrinsicID() &&
221 !ExternalFunctionDoesntCallIntoProgram(F->getName()))
222 Node->addCalledFunction(ExternalNode);
224 // Loop over all of the users of the function... looking for callers...
226 bool isUsedExternally = false;
227 for (Value::use_iterator I = F->use_begin(), E = F->use_end(); I != E; ++I) {
228 if (Instruction *Inst = dyn_cast<Instruction>(*I)) {
229 if (isOnlyADirectCall(F, CallSite::get(Inst)))
230 getNodeFor(Inst->getParent()->getParent())->addCalledFunction(Node);
232 isUsedExternally = true;
233 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(*I)) {
234 // THIS IS A DISGUSTING HACK. Brought to you by the power of
235 // ConstantPointerRefs!
236 for (Value::use_iterator I = CPR->use_begin(), E = CPR->use_end();
238 if (Instruction *Inst = dyn_cast<Instruction>(*I)) {
239 if (isOnlyADirectCall(F, CallSite::get(Inst)))
240 getNodeFor(Inst->getParent()->getParent())->addCalledFunction(Node);
242 isUsedExternally = true;
244 isUsedExternally = true;
246 } else { // Can't classify the user!
247 isUsedExternally = true;
250 if (isUsedExternally)
251 ExternalNode->addCalledFunction(Node);
253 // Look for an indirect function call...
254 for (Function::iterator BB = F->begin(), BBE = F->end(); BB != BBE; ++BB)
255 for (BasicBlock::iterator II = BB->begin(), IE = BB->end(); II != IE; ++II){
256 CallSite CS = CallSite::get(II);
257 if (CS.getInstruction() && !CS.getCalledFunction())
258 Node->addCalledFunction(ExternalNode);
262 bool CallGraph::run(Module &M) {
266 ExternalNode = getNodeFor(0);
269 // Add every function to the call graph...
270 for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
273 // If we didn't find a main function, use the external call graph node
274 if (Root == 0) Root = ExternalNode;
279 void CallGraph::destroy() {
280 for (FunctionMapTy::iterator I = FunctionMap.begin(), E = FunctionMap.end();
286 static void WriteToOutput(const CallGraphNode *CGN, std::ostream &o) {
287 if (CGN->getFunction())
288 o << "Call graph node for function: '"
289 << CGN->getFunction()->getName() <<"'\n";
291 o << "Call graph node <<null function: 0x" << CGN << ">>:\n";
293 for (unsigned i = 0; i < CGN->size(); ++i)
294 if ((*CGN)[i]->getFunction())
295 o << " Calls function '" << (*CGN)[i]->getFunction()->getName() << "'\n";
297 o << " Calls external node\n";
301 void CallGraph::print(std::ostream &o, const Module *M) const {
302 o << "CallGraph Root is: ";
303 if (getRoot()->getFunction())
304 o << getRoot()->getFunction()->getName() << "\n";
306 o << "<<null function: 0x" << getRoot() << ">>\n";
308 for (CallGraph::const_iterator I = begin(), E = end(); I != E; ++I)
309 WriteToOutput(I->second, o);
313 //===----------------------------------------------------------------------===//
314 // Implementations of public modification methods
317 // Functions to keep a call graph up to date with a function that has been
320 void CallGraph::addFunctionToModule(Function *Meth) {
321 assert(0 && "not implemented");
325 // removeFunctionFromModule - Unlink the function from this module, returning
326 // it. Because this removes the function from the module, the call graph node
327 // is destroyed. This is only valid if the function does not call any other
328 // functions (ie, there are no edges in it's CGN). The easiest way to do this
329 // is to dropAllReferences before calling this.
331 Function *CallGraph::removeFunctionFromModule(CallGraphNode *CGN) {
332 assert(CGN->CalledFunctions.empty() && "Cannot remove function from call "
333 "graph if it references other functions!");
334 Function *F = CGN->getFunction(); // Get the function for the call graph node
335 delete CGN; // Delete the call graph node for this func
336 FunctionMap.erase(F); // Remove the call graph node from the map
338 Mod->getFunctionList().remove(F);
342 void CallGraph::stub() {}