CFG.h revision 2a4cd498ff9b2b126de3e370b768ab307c221ab8
1//===--- CFG.h - Classes for representing and building CFGs------*- C++ -*-===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file defines the CFG and CFGBuilder classes for representing and 11// building Control-Flow Graphs (CFGs) from ASTs. 12// 13//===----------------------------------------------------------------------===// 14 15#ifndef LLVM_CLANG_CFG_H 16#define LLVM_CLANG_CFG_H 17 18#include "llvm/ADT/PointerIntPair.h" 19#include "llvm/ADT/GraphTraits.h" 20#include "llvm/Support/Allocator.h" 21#include "llvm/Support/Casting.h" 22#include "clang/Analysis/Support/BumpVector.h" 23#include "clang/Basic/SourceLocation.h" 24#include <cassert> 25 26namespace llvm { 27 class raw_ostream; 28} 29 30namespace clang { 31 class Decl; 32 class Stmt; 33 class Expr; 34 class CFG; 35 class PrinterHelper; 36 class LangOptions; 37 class ASTContext; 38 39/// CFGElement - Represents a top-level expression in a basic block. 40class CFGElement { 41 llvm::PointerIntPair<Stmt *, 2> Data; 42public: 43 enum Type { StartScope, EndScope }; 44 explicit CFGElement() {} 45 CFGElement(Stmt *S, bool lvalue) : Data(S, lvalue ? 1 : 0) {} 46 CFGElement(Stmt *S, Type t) : Data(S, t == StartScope ? 2 : 3) {} 47 Stmt *getStmt() const { return Data.getPointer(); } 48 bool asLValue() const { return Data.getInt() == 1; } 49 bool asStartScope() const { return Data.getInt() == 2; } 50 bool asEndScope() const { return Data.getInt() == 3; } 51 bool asDtor() const { return Data.getInt() == 4; } 52 operator Stmt*() const { return getStmt(); } 53 operator bool() const { return getStmt() != 0; } 54}; 55 56/// CFGBlock - Represents a single basic block in a source-level CFG. 57/// It consists of: 58/// 59/// (1) A set of statements/expressions (which may contain subexpressions). 60/// (2) A "terminator" statement (not in the set of statements). 61/// (3) A list of successors and predecessors. 62/// 63/// Terminator: The terminator represents the type of control-flow that occurs 64/// at the end of the basic block. The terminator is a Stmt* referring to an 65/// AST node that has control-flow: if-statements, breaks, loops, etc. 66/// If the control-flow is conditional, the condition expression will appear 67/// within the set of statements in the block (usually the last statement). 68/// 69/// Predecessors: the order in the set of predecessors is arbitrary. 70/// 71/// Successors: the order in the set of successors is NOT arbitrary. We 72/// currently have the following orderings based on the terminator: 73/// 74/// Terminator Successor Ordering 75/// ----------------------------------------------------- 76/// if Then Block; Else Block 77/// ? operator LHS expression; RHS expression 78/// &&, || expression that uses result of && or ||, RHS 79/// 80class CFGBlock { 81 class StatementList { 82 typedef BumpVector<CFGElement> ImplTy; 83 ImplTy Impl; 84 public: 85 StatementList(BumpVectorContext &C) : Impl(C, 4) {} 86 87 typedef std::reverse_iterator<ImplTy::iterator> iterator; 88 typedef std::reverse_iterator<ImplTy::const_iterator> const_iterator; 89 typedef ImplTy::iterator reverse_iterator; 90 typedef ImplTy::const_iterator const_reverse_iterator; 91 92 void push_back(CFGElement e, BumpVectorContext &C) { Impl.push_back(e, C); } 93 CFGElement front() const { return Impl.back(); } 94 CFGElement back() const { return Impl.front(); } 95 96 iterator begin() { return Impl.rbegin(); } 97 iterator end() { return Impl.rend(); } 98 const_iterator begin() const { return Impl.rbegin(); } 99 const_iterator end() const { return Impl.rend(); } 100 reverse_iterator rbegin() { return Impl.begin(); } 101 reverse_iterator rend() { return Impl.end(); } 102 const_reverse_iterator rbegin() const { return Impl.begin(); } 103 const_reverse_iterator rend() const { return Impl.end(); } 104 105 CFGElement operator[](size_t i) const { 106 assert(i < Impl.size()); 107 return Impl[Impl.size() - 1 - i]; 108 } 109 110 size_t size() const { return Impl.size(); } 111 bool empty() const { return Impl.empty(); } 112 }; 113 114 /// Stmts - The set of statements in the basic block. 115 StatementList Stmts; 116 117 /// Label - An (optional) label that prefixes the executable 118 /// statements in the block. When this variable is non-NULL, it is 119 /// either an instance of LabelStmt, SwitchCase or CXXCatchStmt. 120 Stmt *Label; 121 122 /// Terminator - The terminator for a basic block that 123 /// indicates the type of control-flow that occurs between a block 124 /// and its successors. 125 Stmt *Terminator; 126 127 /// LoopTarget - Some blocks are used to represent the "loop edge" to 128 /// the start of a loop from within the loop body. This Stmt* will be 129 /// refer to the loop statement for such blocks (and be null otherwise). 130 const Stmt *LoopTarget; 131 132 /// BlockID - A numerical ID assigned to a CFGBlock during construction 133 /// of the CFG. 134 unsigned BlockID; 135 136 /// Predecessors/Successors - Keep track of the predecessor / successor 137 /// CFG blocks. 138 typedef BumpVector<CFGBlock*> AdjacentBlocks; 139 AdjacentBlocks Preds; 140 AdjacentBlocks Succs; 141 142public: 143 explicit CFGBlock(unsigned blockid, BumpVectorContext &C) 144 : Stmts(C), Label(NULL), Terminator(NULL), LoopTarget(NULL), 145 BlockID(blockid), Preds(C, 1), Succs(C, 1) {} 146 ~CFGBlock() {} 147 148 // Statement iterators 149 typedef StatementList::iterator iterator; 150 typedef StatementList::const_iterator const_iterator; 151 typedef StatementList::reverse_iterator reverse_iterator; 152 typedef StatementList::const_reverse_iterator const_reverse_iterator; 153 154 CFGElement front() const { return Stmts.front(); } 155 CFGElement back() const { return Stmts.back(); } 156 157 iterator begin() { return Stmts.begin(); } 158 iterator end() { return Stmts.end(); } 159 const_iterator begin() const { return Stmts.begin(); } 160 const_iterator end() const { return Stmts.end(); } 161 162 reverse_iterator rbegin() { return Stmts.rbegin(); } 163 reverse_iterator rend() { return Stmts.rend(); } 164 const_reverse_iterator rbegin() const { return Stmts.rbegin(); } 165 const_reverse_iterator rend() const { return Stmts.rend(); } 166 167 unsigned size() const { return Stmts.size(); } 168 bool empty() const { return Stmts.empty(); } 169 170 CFGElement operator[](size_t i) const { return Stmts[i]; } 171 172 // CFG iterators 173 typedef AdjacentBlocks::iterator pred_iterator; 174 typedef AdjacentBlocks::const_iterator const_pred_iterator; 175 typedef AdjacentBlocks::reverse_iterator pred_reverse_iterator; 176 typedef AdjacentBlocks::const_reverse_iterator const_pred_reverse_iterator; 177 178 typedef AdjacentBlocks::iterator succ_iterator; 179 typedef AdjacentBlocks::const_iterator const_succ_iterator; 180 typedef AdjacentBlocks::reverse_iterator succ_reverse_iterator; 181 typedef AdjacentBlocks::const_reverse_iterator const_succ_reverse_iterator; 182 183 pred_iterator pred_begin() { return Preds.begin(); } 184 pred_iterator pred_end() { return Preds.end(); } 185 const_pred_iterator pred_begin() const { return Preds.begin(); } 186 const_pred_iterator pred_end() const { return Preds.end(); } 187 188 pred_reverse_iterator pred_rbegin() { return Preds.rbegin(); } 189 pred_reverse_iterator pred_rend() { return Preds.rend(); } 190 const_pred_reverse_iterator pred_rbegin() const { return Preds.rbegin(); } 191 const_pred_reverse_iterator pred_rend() const { return Preds.rend(); } 192 193 succ_iterator succ_begin() { return Succs.begin(); } 194 succ_iterator succ_end() { return Succs.end(); } 195 const_succ_iterator succ_begin() const { return Succs.begin(); } 196 const_succ_iterator succ_end() const { return Succs.end(); } 197 198 succ_reverse_iterator succ_rbegin() { return Succs.rbegin(); } 199 succ_reverse_iterator succ_rend() { return Succs.rend(); } 200 const_succ_reverse_iterator succ_rbegin() const { return Succs.rbegin(); } 201 const_succ_reverse_iterator succ_rend() const { return Succs.rend(); } 202 203 unsigned succ_size() const { return Succs.size(); } 204 bool succ_empty() const { return Succs.empty(); } 205 206 unsigned pred_size() const { return Preds.size(); } 207 bool pred_empty() const { return Preds.empty(); } 208 209 210 class FilterOptions { 211 public: 212 FilterOptions() { 213 IgnoreDefaultsWithCoveredEnums = 0; 214 } 215 216 unsigned IgnoreDefaultsWithCoveredEnums : 1; 217 }; 218 219 static bool FilterEdge(const FilterOptions &F, const CFGBlock *Src, 220 const CFGBlock *Dst); 221 222 template <typename IMPL, bool IsPred> 223 class FilteredCFGBlockIterator { 224 private: 225 IMPL I, E; 226 const FilterOptions F; 227 const CFGBlock *From; 228 public: 229 explicit FilteredCFGBlockIterator(const IMPL &i, const IMPL &e, 230 const CFGBlock *from, 231 const FilterOptions &f) 232 : I(i), E(e), F(f), From(from) {} 233 234 bool hasMore() const { return I != E; } 235 236 FilteredCFGBlockIterator &operator++() { 237 do { ++I; } while (hasMore() && Filter(*I)); 238 return *this; 239 } 240 241 const CFGBlock *operator*() const { return *I; } 242 private: 243 bool Filter(const CFGBlock *To) { 244 return IsPred ? FilterEdge(F, To, From) : FilterEdge(F, From, To); 245 } 246 }; 247 248 typedef FilteredCFGBlockIterator<const_pred_iterator, true> 249 filtered_pred_iterator; 250 251 typedef FilteredCFGBlockIterator<const_succ_iterator, false> 252 filtered_succ_iterator; 253 254 filtered_pred_iterator filtered_pred_start_end(const FilterOptions &f) const { 255 return filtered_pred_iterator(pred_begin(), pred_end(), this, f); 256 } 257 258 filtered_succ_iterator filtered_succ_start_end(const FilterOptions &f) const { 259 return filtered_succ_iterator(succ_begin(), succ_end(), this, f); 260 } 261 262 // Manipulation of block contents 263 264 void setTerminator(Stmt* Statement) { Terminator = Statement; } 265 void setLabel(Stmt* Statement) { Label = Statement; } 266 void setLoopTarget(const Stmt *loopTarget) { LoopTarget = loopTarget; } 267 268 Stmt* getTerminator() { return Terminator; } 269 const Stmt* getTerminator() const { return Terminator; } 270 271 Stmt* getTerminatorCondition(); 272 273 const Stmt* getTerminatorCondition() const { 274 return const_cast<CFGBlock*>(this)->getTerminatorCondition(); 275 } 276 277 const Stmt *getLoopTarget() const { return LoopTarget; } 278 279 bool hasBinaryBranchTerminator() const; 280 281 Stmt* getLabel() { return Label; } 282 const Stmt* getLabel() const { return Label; } 283 284 unsigned getBlockID() const { return BlockID; } 285 286 void dump(const CFG *cfg, const LangOptions &LO) const; 287 void print(llvm::raw_ostream &OS, const CFG* cfg, const LangOptions &LO) const; 288 void printTerminator(llvm::raw_ostream &OS, const LangOptions &LO) const; 289 290 void addSuccessor(CFGBlock* Block, BumpVectorContext &C) { 291 if (Block) 292 Block->Preds.push_back(this, C); 293 Succs.push_back(Block, C); 294 } 295 296 void appendStmt(Stmt* Statement, BumpVectorContext &C, bool asLValue) { 297 Stmts.push_back(CFGElement(Statement, asLValue), C); 298 } 299}; 300 301 302/// CFG - Represents a source-level, intra-procedural CFG that represents the 303/// control-flow of a Stmt. The Stmt can represent an entire function body, 304/// or a single expression. A CFG will always contain one empty block that 305/// represents the Exit point of the CFG. A CFG will also contain a designated 306/// Entry block. The CFG solely represents control-flow; it consists of 307/// CFGBlocks which are simply containers of Stmt*'s in the AST the CFG 308/// was constructed from. 309class CFG { 310public: 311 //===--------------------------------------------------------------------===// 312 // CFG Construction & Manipulation. 313 //===--------------------------------------------------------------------===// 314 315 class BuildOptions { 316 public: 317 bool PruneTriviallyFalseEdges:1; 318 bool AddEHEdges:1; 319 bool AddInitializers:1; 320 bool AddImplicitDtors:1; 321 322 BuildOptions() 323 : PruneTriviallyFalseEdges(true) 324 , AddEHEdges(false) 325 , AddInitializers(false) 326 , AddImplicitDtors(false) {} 327 }; 328 329 /// buildCFG - Builds a CFG from an AST. The responsibility to free the 330 /// constructed CFG belongs to the caller. 331 static CFG* buildCFG(const Decl *D, Stmt* AST, ASTContext *C, 332 BuildOptions BO = BuildOptions()); 333 334 /// createBlock - Create a new block in the CFG. The CFG owns the block; 335 /// the caller should not directly free it. 336 CFGBlock* createBlock(); 337 338 /// setEntry - Set the entry block of the CFG. This is typically used 339 /// only during CFG construction. Most CFG clients expect that the 340 /// entry block has no predecessors and contains no statements. 341 void setEntry(CFGBlock *B) { Entry = B; } 342 343 /// setIndirectGotoBlock - Set the block used for indirect goto jumps. 344 /// This is typically used only during CFG construction. 345 void setIndirectGotoBlock(CFGBlock* B) { IndirectGotoBlock = B; } 346 347 //===--------------------------------------------------------------------===// 348 // Block Iterators 349 //===--------------------------------------------------------------------===// 350 351 typedef BumpVector<CFGBlock*> CFGBlockListTy; 352 typedef CFGBlockListTy::iterator iterator; 353 typedef CFGBlockListTy::const_iterator const_iterator; 354 typedef std::reverse_iterator<iterator> reverse_iterator; 355 typedef std::reverse_iterator<const_iterator> const_reverse_iterator; 356 357 CFGBlock& front() { return *Blocks.front(); } 358 CFGBlock& back() { return *Blocks.back(); } 359 360 iterator begin() { return Blocks.begin(); } 361 iterator end() { return Blocks.end(); } 362 const_iterator begin() const { return Blocks.begin(); } 363 const_iterator end() const { return Blocks.end(); } 364 365 reverse_iterator rbegin() { return Blocks.rbegin(); } 366 reverse_iterator rend() { return Blocks.rend(); } 367 const_reverse_iterator rbegin() const { return Blocks.rbegin(); } 368 const_reverse_iterator rend() const { return Blocks.rend(); } 369 370 CFGBlock& getEntry() { return *Entry; } 371 const CFGBlock& getEntry() const { return *Entry; } 372 CFGBlock& getExit() { return *Exit; } 373 const CFGBlock& getExit() const { return *Exit; } 374 375 CFGBlock* getIndirectGotoBlock() { return IndirectGotoBlock; } 376 const CFGBlock* getIndirectGotoBlock() const { return IndirectGotoBlock; } 377 378 //===--------------------------------------------------------------------===// 379 // Member templates useful for various batch operations over CFGs. 380 //===--------------------------------------------------------------------===// 381 382 template <typename CALLBACK> 383 void VisitBlockStmts(CALLBACK& O) const { 384 for (const_iterator I=begin(), E=end(); I != E; ++I) 385 for (CFGBlock::const_iterator BI=(*I)->begin(), BE=(*I)->end(); 386 BI != BE; ++BI) 387 O(*BI); 388 } 389 390 //===--------------------------------------------------------------------===// 391 // CFG Introspection. 392 //===--------------------------------------------------------------------===// 393 394 struct BlkExprNumTy { 395 const signed Idx; 396 explicit BlkExprNumTy(signed idx) : Idx(idx) {} 397 explicit BlkExprNumTy() : Idx(-1) {} 398 operator bool() const { return Idx >= 0; } 399 operator unsigned() const { assert(Idx >=0); return (unsigned) Idx; } 400 }; 401 402 bool isBlkExpr(const Stmt* S) { return getBlkExprNum(S); } 403 BlkExprNumTy getBlkExprNum(const Stmt* S); 404 unsigned getNumBlkExprs(); 405 406 /// getNumBlockIDs - Returns the total number of BlockIDs allocated (which 407 /// start at 0). 408 unsigned getNumBlockIDs() const { return NumBlockIDs; } 409 410 //===--------------------------------------------------------------------===// 411 // CFG Debugging: Pretty-Printing and Visualization. 412 //===--------------------------------------------------------------------===// 413 414 void viewCFG(const LangOptions &LO) const; 415 void print(llvm::raw_ostream& OS, const LangOptions &LO) const; 416 void dump(const LangOptions &LO) const; 417 418 //===--------------------------------------------------------------------===// 419 // Internal: constructors and data. 420 //===--------------------------------------------------------------------===// 421 422 CFG() : Entry(NULL), Exit(NULL), IndirectGotoBlock(NULL), NumBlockIDs(0), 423 BlkExprMap(NULL), Blocks(BlkBVC, 10) {} 424 425 ~CFG(); 426 427 llvm::BumpPtrAllocator& getAllocator() { 428 return BlkBVC.getAllocator(); 429 } 430 431 BumpVectorContext &getBumpVectorContext() { 432 return BlkBVC; 433 } 434 435private: 436 CFGBlock* Entry; 437 CFGBlock* Exit; 438 CFGBlock* IndirectGotoBlock; // Special block to contain collective dispatch 439 // for indirect gotos 440 unsigned NumBlockIDs; 441 442 // BlkExprMap - An opaque pointer to prevent inclusion of DenseMap.h. 443 // It represents a map from Expr* to integers to record the set of 444 // block-level expressions and their "statement number" in the CFG. 445 void* BlkExprMap; 446 447 BumpVectorContext BlkBVC; 448 449 CFGBlockListTy Blocks; 450 451}; 452} // end namespace clang 453 454//===----------------------------------------------------------------------===// 455// GraphTraits specializations for CFG basic block graphs (source-level CFGs) 456//===----------------------------------------------------------------------===// 457 458namespace llvm { 459 460/// Implement simplify_type for CFGElement, so that we can dyn_cast from 461/// CFGElement to a specific Stmt class. 462template <> struct simplify_type<const ::clang::CFGElement> { 463 typedef ::clang::Stmt* SimpleType; 464 static SimpleType getSimplifiedValue(const ::clang::CFGElement &Val) { 465 return Val.getStmt(); 466 } 467}; 468 469template <> struct simplify_type< ::clang::CFGElement> 470 : public simplify_type<const ::clang::CFGElement> {}; 471 472// Traits for: CFGBlock 473 474template <> struct GraphTraits< ::clang::CFGBlock* > { 475 typedef ::clang::CFGBlock NodeType; 476 typedef ::clang::CFGBlock::succ_iterator ChildIteratorType; 477 478 static NodeType* getEntryNode(::clang::CFGBlock* BB) 479 { return BB; } 480 481 static inline ChildIteratorType child_begin(NodeType* N) 482 { return N->succ_begin(); } 483 484 static inline ChildIteratorType child_end(NodeType* N) 485 { return N->succ_end(); } 486}; 487 488template <> struct GraphTraits< const ::clang::CFGBlock* > { 489 typedef const ::clang::CFGBlock NodeType; 490 typedef ::clang::CFGBlock::const_succ_iterator ChildIteratorType; 491 492 static NodeType* getEntryNode(const clang::CFGBlock* BB) 493 { return BB; } 494 495 static inline ChildIteratorType child_begin(NodeType* N) 496 { return N->succ_begin(); } 497 498 static inline ChildIteratorType child_end(NodeType* N) 499 { return N->succ_end(); } 500}; 501 502template <> struct GraphTraits<Inverse<const ::clang::CFGBlock*> > { 503 typedef const ::clang::CFGBlock NodeType; 504 typedef ::clang::CFGBlock::const_pred_iterator ChildIteratorType; 505 506 static NodeType *getEntryNode(Inverse<const ::clang::CFGBlock*> G) 507 { return G.Graph; } 508 509 static inline ChildIteratorType child_begin(NodeType* N) 510 { return N->pred_begin(); } 511 512 static inline ChildIteratorType child_end(NodeType* N) 513 { return N->pred_end(); } 514}; 515 516// Traits for: CFG 517 518template <> struct GraphTraits< ::clang::CFG* > 519 : public GraphTraits< ::clang::CFGBlock* > { 520 521 typedef ::clang::CFG::iterator nodes_iterator; 522 523 static NodeType *getEntryNode(::clang::CFG* F) { return &F->getEntry(); } 524 static nodes_iterator nodes_begin(::clang::CFG* F) { return F->begin(); } 525 static nodes_iterator nodes_end(::clang::CFG* F) { return F->end(); } 526}; 527 528template <> struct GraphTraits<const ::clang::CFG* > 529 : public GraphTraits<const ::clang::CFGBlock* > { 530 531 typedef ::clang::CFG::const_iterator nodes_iterator; 532 533 static NodeType *getEntryNode( const ::clang::CFG* F) { 534 return &F->getEntry(); 535 } 536 static nodes_iterator nodes_begin( const ::clang::CFG* F) { 537 return F->begin(); 538 } 539 static nodes_iterator nodes_end( const ::clang::CFG* F) { 540 return F->end(); 541 } 542}; 543 544template <> struct GraphTraits<Inverse<const ::clang::CFG*> > 545 : public GraphTraits<Inverse<const ::clang::CFGBlock*> > { 546 547 typedef ::clang::CFG::const_iterator nodes_iterator; 548 549 static NodeType *getEntryNode(const ::clang::CFG* F) { return &F->getExit(); } 550 static nodes_iterator nodes_begin(const ::clang::CFG* F) { return F->begin();} 551 static nodes_iterator nodes_end(const ::clang::CFG* F) { return F->end(); } 552}; 553} // end llvm namespace 554#endif 555