1 //===- CallGraph.h - Build a Module's call graph ----------------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
11 /// This file provides interfaces used to build and manipulate a call graph,
12 /// which is a very useful tool for interprocedural optimization.
14 /// Every function in a module is represented as a node in the call graph. The
15 /// callgraph node keeps track of which functions the are called by the
16 /// function corresponding to the node.
18 /// A call graph may contain nodes where the function that they correspond to
19 /// is null. These 'external' nodes are used to represent control flow that is
20 /// not represented (or analyzable) in the module. In particular, this
21 /// analysis builds one external node such that:
22 /// 1. All functions in the module without internal linkage will have edges
23 /// from this external node, indicating that they could be called by
24 /// functions outside of the module.
25 /// 2. All functions whose address is used for something more than a direct
26 /// call, for example being stored into a memory location will also have
27 /// an edge from this external node. Since they may be called by an
28 /// unknown caller later, they must be tracked as such.
30 /// There is a second external node added for calls that leave this module.
31 /// Functions have a call edge to the external node iff:
32 /// 1. The function is external, reflecting the fact that they could call
33 /// anything without internal linkage or that has its address taken.
34 /// 2. The function contains an indirect function call.
36 /// As an extension in the future, there may be multiple nodes with a null
37 /// function. These will be used when we can prove (through pointer analysis)
38 /// that an indirect call site can call only a specific set of functions.
40 /// Because of these properties, the CallGraph captures a conservative superset
41 /// of all of the caller-callee relationships, which is useful for
44 /// The CallGraph class also attempts to figure out what the root of the
45 /// CallGraph is, which it currently does by looking for a function named
46 /// 'main'. If no function named 'main' is found, the external node is used as
47 /// the entry node, reflecting the fact that any function without internal
48 /// linkage could be called into (which is common for libraries).
50 //===----------------------------------------------------------------------===//
52 #ifndef LLVM_ANALYSIS_CALLGRAPH_H
53 #define LLVM_ANALYSIS_CALLGRAPH_H
55 #include "llvm/ADT/GraphTraits.h"
56 #include "llvm/ADT/OwningPtr.h"
57 #include "llvm/ADT/STLExtras.h"
58 #include "llvm/IR/Function.h"
59 #include "llvm/Pass.h"
60 #include "llvm/Support/CallSite.h"
61 #include "llvm/Support/IncludeFile.h"
62 #include "llvm/Support/ValueHandle.h"
71 /// \brief The basic data container for the call graph of a \c Module of IR.
73 /// This class exposes both the interface to the call graph for a module of IR.
75 /// The core call graph itself can also be updated to reflect changes to the IR.
79 typedef std::map<const Function *, CallGraphNode *> FunctionMapTy;
81 /// \brief A map from \c Function* to \c CallGraphNode*.
82 FunctionMapTy FunctionMap;
84 /// \brief Root is root of the call graph, or the external node if a 'main'
85 /// function couldn't be found.
88 /// \brief This node has edges to all external functions and those internal
89 /// functions that have their address taken.
90 CallGraphNode *ExternalCallingNode;
92 /// \brief This node has edges to it from all functions making indirect calls
93 /// or calling an external function.
94 CallGraphNode *CallsExternalNode;
96 /// \brief Replace the function represented by this node by another.
98 /// This does not rescan the body of the function, so it is suitable when
99 /// splicing the body of one function to another while also updating all
100 /// callers from the old function to the new.
101 void spliceFunction(const Function *From, const Function *To);
103 /// \brief Add a function to the call graph, and link the node to all of the
104 /// functions that it calls.
105 void addToCallGraph(Function *F);
108 CallGraph(Module &M);
111 void print(raw_ostream &OS) const;
114 typedef FunctionMapTy::iterator iterator;
115 typedef FunctionMapTy::const_iterator const_iterator;
117 /// \brief Returns the module the call graph corresponds to.
118 Module &getModule() const { return M; }
120 inline iterator begin() { return FunctionMap.begin(); }
121 inline iterator end() { return FunctionMap.end(); }
122 inline const_iterator begin() const { return FunctionMap.begin(); }
123 inline const_iterator end() const { return FunctionMap.end(); }
125 /// \brief Returns the call graph node for the provided function.
126 inline const CallGraphNode *operator[](const Function *F) const {
127 const_iterator I = FunctionMap.find(F);
128 assert(I != FunctionMap.end() && "Function not in callgraph!");
132 /// \brief Returns the call graph node for the provided function.
133 inline CallGraphNode *operator[](const Function *F) {
134 const_iterator I = FunctionMap.find(F);
135 assert(I != FunctionMap.end() && "Function not in callgraph!");
139 /// \brief Returns the \c CallGraphNode which is used to represent
140 /// undetermined calls into the callgraph.
141 CallGraphNode *getExternalCallingNode() const { return ExternalCallingNode; }
143 CallGraphNode *getCallsExternalNode() const { return CallsExternalNode; }
145 //===---------------------------------------------------------------------
146 // Functions to keep a call graph up to date with a function that has been
150 /// \brief Unlink the function from this module, returning it.
152 /// Because this removes the function from the module, the call graph node is
153 /// destroyed. This is only valid if the function does not call any other
154 /// functions (ie, there are no edges in it's CGN). The easiest way to do
155 /// this is to dropAllReferences before calling this.
156 Function *removeFunctionFromModule(CallGraphNode *CGN);
158 /// \brief Similar to operator[], but this will insert a new CallGraphNode for
159 /// \c F if one does not already exist.
160 CallGraphNode *getOrInsertFunction(const Function *F);
163 /// \brief A node in the call graph for a module.
165 /// Typically represents a function in the call graph. There are also special
166 /// "null" nodes used to represent theoretical entries in the call graph.
167 class CallGraphNode {
169 /// \brief A pair of the calling instruction (a call or invoke)
170 /// and the call graph node being called.
171 typedef std::pair<WeakVH, CallGraphNode *> CallRecord;
174 typedef std::vector<CallRecord> CalledFunctionsVector;
176 /// \brief Creates a node for the specified function.
177 inline CallGraphNode(Function *F) : F(F), NumReferences(0) {}
180 assert(NumReferences == 0 && "Node deleted while references remain");
183 typedef std::vector<CallRecord>::iterator iterator;
184 typedef std::vector<CallRecord>::const_iterator const_iterator;
186 /// \brief Returns the function that this call graph node represents.
187 Function *getFunction() const { return F; }
189 inline iterator begin() { return CalledFunctions.begin(); }
190 inline iterator end() { return CalledFunctions.end(); }
191 inline const_iterator begin() const { return CalledFunctions.begin(); }
192 inline const_iterator end() const { return CalledFunctions.end(); }
193 inline bool empty() const { return CalledFunctions.empty(); }
194 inline unsigned size() const { return (unsigned)CalledFunctions.size(); }
196 /// \brief Returns the number of other CallGraphNodes in this CallGraph that
197 /// reference this node in their callee list.
198 unsigned getNumReferences() const { return NumReferences; }
200 /// \brief Returns the i'th called function.
201 CallGraphNode *operator[](unsigned i) const {
202 assert(i < CalledFunctions.size() && "Invalid index");
203 return CalledFunctions[i].second;
206 /// \brief Print out this call graph node.
208 void print(raw_ostream &OS) const;
210 //===---------------------------------------------------------------------
211 // Methods to keep a call graph up to date with a function that has been
215 /// \brief Removes all edges from this CallGraphNode to any functions it
217 void removeAllCalledFunctions() {
218 while (!CalledFunctions.empty()) {
219 CalledFunctions.back().second->DropRef();
220 CalledFunctions.pop_back();
224 /// \brief Moves all the callee information from N to this node.
225 void stealCalledFunctionsFrom(CallGraphNode *N) {
226 assert(CalledFunctions.empty() &&
227 "Cannot steal callsite information if I already have some");
228 std::swap(CalledFunctions, N->CalledFunctions);
231 /// \brief Adds a function to the list of functions called by this one.
232 void addCalledFunction(CallSite CS, CallGraphNode *M) {
233 assert(!CS.getInstruction() || !CS.getCalledFunction() ||
234 !CS.getCalledFunction()->isIntrinsic());
235 CalledFunctions.push_back(std::make_pair(CS.getInstruction(), M));
239 void removeCallEdge(iterator I) {
240 I->second->DropRef();
241 *I = CalledFunctions.back();
242 CalledFunctions.pop_back();
245 /// \brief Removes the edge in the node for the specified call site.
247 /// Note that this method takes linear time, so it should be used sparingly.
248 void removeCallEdgeFor(CallSite CS);
250 /// \brief Removes all call edges from this node to the specified callee
253 /// This takes more time to execute than removeCallEdgeTo, so it should not
254 /// be used unless necessary.
255 void removeAnyCallEdgeTo(CallGraphNode *Callee);
257 /// \brief Removes one edge associated with a null callsite from this node to
258 /// the specified callee function.
259 void removeOneAbstractEdgeTo(CallGraphNode *Callee);
261 /// \brief Replaces the edge in the node for the specified call site with a
264 /// Note that this method takes linear time, so it should be used sparingly.
265 void replaceCallEdge(CallSite CS, CallSite NewCS, CallGraphNode *NewNode);
268 friend class CallGraph;
270 AssertingVH<Function> F;
272 std::vector<CallRecord> CalledFunctions;
274 /// \brief The number of times that this CallGraphNode occurs in the
275 /// CalledFunctions array of this or other CallGraphNodes.
276 unsigned NumReferences;
278 CallGraphNode(const CallGraphNode &) LLVM_DELETED_FUNCTION;
279 void operator=(const CallGraphNode &) LLVM_DELETED_FUNCTION;
281 void DropRef() { --NumReferences; }
282 void AddRef() { ++NumReferences; }
284 /// \brief A special function that should only be used by the CallGraph class.
285 void allReferencesDropped() { NumReferences = 0; }
288 /// \brief An analysis pass to compute the \c CallGraph for a \c Module.
290 /// This class implements the concept of an analysis pass used by the \c
291 /// ModuleAnalysisManager to run an analysis over a module and cache the
293 class CallGraphAnalysis {
295 /// \brief A formulaic typedef to inform clients of the result type.
296 typedef CallGraph Result;
298 static void *ID() { return (void *)&PassID; }
300 /// \brief Compute the \c CallGraph for the module \c M.
302 /// The real work here is done in the \c CallGraph constructor.
303 CallGraph run(Module *M) { return CallGraph(*M); }
309 /// \brief The \c ModulePass which wraps up a \c CallGraph and the logic to
312 /// This class exposes both the interface to the call graph container and the
313 /// module pass which runs over a module of IR and produces the call graph. The
314 /// call graph interface is entirelly a wrapper around a \c CallGraph object
315 /// which is stored internally for each module.
316 class CallGraphWrapperPass : public ModulePass {
317 OwningPtr<CallGraph> G;
320 static char ID; // Class identification, replacement for typeinfo
322 CallGraphWrapperPass();
323 virtual ~CallGraphWrapperPass();
325 /// \brief The internal \c CallGraph around which the rest of this interface
327 const CallGraph &getCallGraph() const { return *G; }
328 CallGraph &getCallGraph() { return *G; }
330 typedef CallGraph::iterator iterator;
331 typedef CallGraph::const_iterator const_iterator;
333 /// \brief Returns the module the call graph corresponds to.
334 Module &getModule() const { return G->getModule(); }
336 inline iterator begin() { return G->begin(); }
337 inline iterator end() { return G->end(); }
338 inline const_iterator begin() const { return G->begin(); }
339 inline const_iterator end() const { return G->end(); }
341 /// \brief Returns the call graph node for the provided function.
342 inline const CallGraphNode *operator[](const Function *F) const {
346 /// \brief Returns the call graph node for the provided function.
347 inline CallGraphNode *operator[](const Function *F) { return (*G)[F]; }
349 /// \brief Returns the \c CallGraphNode which is used to represent
350 /// undetermined calls into the callgraph.
351 CallGraphNode *getExternalCallingNode() const {
352 return G->getExternalCallingNode();
355 CallGraphNode *getCallsExternalNode() const {
356 return G->getCallsExternalNode();
359 //===---------------------------------------------------------------------
360 // Functions to keep a call graph up to date with a function that has been
364 /// \brief Unlink the function from this module, returning it.
366 /// Because this removes the function from the module, the call graph node is
367 /// destroyed. This is only valid if the function does not call any other
368 /// functions (ie, there are no edges in it's CGN). The easiest way to do
369 /// this is to dropAllReferences before calling this.
370 Function *removeFunctionFromModule(CallGraphNode *CGN) {
371 return G->removeFunctionFromModule(CGN);
374 /// \brief Similar to operator[], but this will insert a new CallGraphNode for
375 /// \c F if one does not already exist.
376 CallGraphNode *getOrInsertFunction(const Function *F) {
377 return G->getOrInsertFunction(F);
380 //===---------------------------------------------------------------------
381 // Implementation of the ModulePass interface needed here.
384 virtual void getAnalysisUsage(AnalysisUsage &AU) const;
385 virtual bool runOnModule(Module &M);
386 virtual void releaseMemory();
388 void print(raw_ostream &o, const Module *) const;
392 //===----------------------------------------------------------------------===//
393 // GraphTraits specializations for call graphs so that they can be treated as
394 // graphs by the generic graph algorithms.
397 // Provide graph traits for tranversing call graphs using standard graph
399 template <> struct GraphTraits<CallGraphNode *> {
400 typedef CallGraphNode NodeType;
402 typedef CallGraphNode::CallRecord CGNPairTy;
403 typedef std::pointer_to_unary_function<CGNPairTy, CallGraphNode *>
406 static NodeType *getEntryNode(CallGraphNode *CGN) { return CGN; }
408 typedef mapped_iterator<NodeType::iterator, CGNDerefFun> ChildIteratorType;
410 static inline ChildIteratorType child_begin(NodeType *N) {
411 return map_iterator(N->begin(), CGNDerefFun(CGNDeref));
413 static inline ChildIteratorType child_end(NodeType *N) {
414 return map_iterator(N->end(), CGNDerefFun(CGNDeref));
417 static CallGraphNode *CGNDeref(CGNPairTy P) { return P.second; }
420 template <> struct GraphTraits<const CallGraphNode *> {
421 typedef const CallGraphNode NodeType;
422 typedef NodeType::const_iterator ChildIteratorType;
424 static NodeType *getEntryNode(const CallGraphNode *CGN) { return CGN; }
425 static inline ChildIteratorType child_begin(NodeType *N) {
428 static inline ChildIteratorType child_end(NodeType *N) { return N->end(); }
432 struct GraphTraits<CallGraph *> : public GraphTraits<CallGraphNode *> {
433 static NodeType *getEntryNode(CallGraph *CGN) {
434 return CGN->getExternalCallingNode(); // Start at the external node!
436 typedef std::pair<const Function *, CallGraphNode *> PairTy;
437 typedef std::pointer_to_unary_function<PairTy, CallGraphNode &> DerefFun;
439 // nodes_iterator/begin/end - Allow iteration over all nodes in the graph
440 typedef mapped_iterator<CallGraph::iterator, DerefFun> nodes_iterator;
441 static nodes_iterator nodes_begin(CallGraph *CG) {
442 return map_iterator(CG->begin(), DerefFun(CGdereference));
444 static nodes_iterator nodes_end(CallGraph *CG) {
445 return map_iterator(CG->end(), DerefFun(CGdereference));
448 static CallGraphNode &CGdereference(PairTy P) { return *P.second; }
452 struct GraphTraits<const CallGraph *> : public GraphTraits<
453 const CallGraphNode *> {
454 static NodeType *getEntryNode(const CallGraph *CGN) {
455 return CGN->getExternalCallingNode();
457 // nodes_iterator/begin/end - Allow iteration over all nodes in the graph
458 typedef CallGraph::const_iterator nodes_iterator;
459 static nodes_iterator nodes_begin(const CallGraph *CG) { return CG->begin(); }
460 static nodes_iterator nodes_end(const CallGraph *CG) { return CG->end(); }
463 } // End llvm namespace
465 // Make sure that any clients of this file link in CallGraph.cpp
466 FORCE_DEFINING_FILE_TO_BE_LINKED(CallGraph)