OLD | NEW |
(Empty) | |
| 1 // Copyright (c) 2014, the Dart project authors. Please see the AUTHORS file |
| 2 // for details. All rights reserved. Use of this source code is governed by a |
| 3 // BSD-style license that can be found in the LICENSE file. |
| 4 |
| 5 #include "vm/regexp.h" |
| 6 |
| 7 #include "vm/dart_entry.h" |
| 8 #include "vm/regexp_assembler.h" |
| 9 #include "vm/regexp_ast.h" |
| 10 #include "vm/unibrow-inl.h" |
| 11 #include "vm/unicode.h" |
| 12 #include "vm/symbols.h" |
| 13 |
| 14 #define I (isolate()) |
| 15 #define CI (compiler->isolate()) |
| 16 |
| 17 namespace dart { |
| 18 |
| 19 DECLARE_FLAG(bool, trace_irregexp); |
| 20 |
| 21 // Default to generating optimized regexp code. |
| 22 static const bool kRegexpOptimization = true; |
| 23 |
| 24 // More makes code generation slower, less makes V8 benchmark score lower. |
| 25 static const intptr_t kMaxLookaheadForBoyerMoore = 8; |
| 26 |
| 27 ContainedInLattice AddRange(ContainedInLattice containment, |
| 28 const intptr_t* ranges, |
| 29 intptr_t ranges_length, |
| 30 Interval new_range) { |
| 31 ASSERT((ranges_length & 1) == 1); |
| 32 ASSERT(ranges[ranges_length - 1] == Utf16::kMaxCodeUnit + 1); |
| 33 if (containment == kLatticeUnknown) return containment; |
| 34 bool inside = false; |
| 35 intptr_t last = 0; |
| 36 for (intptr_t i = 0; i < ranges_length; |
| 37 inside = !inside, last = ranges[i], i++) { |
| 38 // Consider the range from last to ranges[i]. |
| 39 // We haven't got to the new range yet. |
| 40 if (ranges[i] <= new_range.from()) continue; |
| 41 // New range is wholly inside last-ranges[i]. Note that new_range.to() is |
| 42 // inclusive, but the values in ranges are not. |
| 43 if (last <= new_range.from() && new_range.to() < ranges[i]) { |
| 44 return Combine(containment, inside ? kLatticeIn : kLatticeOut); |
| 45 } |
| 46 return kLatticeUnknown; |
| 47 } |
| 48 return containment; |
| 49 } |
| 50 |
| 51 // ------------------------------------------------------------------- |
| 52 // Implementation of the Irregexp regular expression engine. |
| 53 // |
| 54 // The Irregexp regular expression engine is intended to be a complete |
| 55 // implementation of ECMAScript regular expressions. It generates |
| 56 // IR code that is subsequently compiled to native code. |
| 57 |
| 58 // The Irregexp regexp engine is structured in three steps. |
| 59 // 1) The parser generates an abstract syntax tree. See regexp_ast.cc. |
| 60 // 2) From the AST a node network is created. The nodes are all |
| 61 // subclasses of RegExpNode. The nodes represent states when |
| 62 // executing a regular expression. Several optimizations are |
| 63 // performed on the node network. |
| 64 // 3) From the nodes we generate IR instructions that can actually |
| 65 // execute the regular expression (perform the search). The |
| 66 // code generation step is described in more detail below. |
| 67 |
| 68 // Code generation. |
| 69 // |
| 70 // The nodes are divided into four main categories. |
| 71 // * Choice nodes |
| 72 // These represent places where the regular expression can |
| 73 // match in more than one way. For example on entry to an |
| 74 // alternation (foo|bar) or a repetition (*, +, ? or {}). |
| 75 // * Action nodes |
| 76 // These represent places where some action should be |
| 77 // performed. Examples include recording the current position |
| 78 // in the input string to a register (in order to implement |
| 79 // captures) or other actions on register for example in order |
| 80 // to implement the counters needed for {} repetitions. |
| 81 // * Matching nodes |
| 82 // These attempt to match some element part of the input string. |
| 83 // Examples of elements include character classes, plain strings |
| 84 // or back references. |
| 85 // * End nodes |
| 86 // These are used to implement the actions required on finding |
| 87 // a successful match or failing to find a match. |
| 88 // |
| 89 // The code generated maintains some state as it runs. This consists of the |
| 90 // following elements: |
| 91 // |
| 92 // * The capture registers. Used for string captures. |
| 93 // * Other registers. Used for counters etc. |
| 94 // * The current position. |
| 95 // * The stack of backtracking information. Used when a matching node |
| 96 // fails to find a match and needs to try an alternative. |
| 97 // |
| 98 // Conceptual regular expression execution model: |
| 99 // |
| 100 // There is a simple conceptual model of regular expression execution |
| 101 // which will be presented first. The actual code generated is a more |
| 102 // efficient simulation of the simple conceptual model: |
| 103 // |
| 104 // * Choice nodes are implemented as follows: |
| 105 // For each choice except the last { |
| 106 // push current position |
| 107 // push backtrack code location |
| 108 // <generate code to test for choice> |
| 109 // backtrack code location: |
| 110 // pop current position |
| 111 // } |
| 112 // <generate code to test for last choice> |
| 113 // |
| 114 // * Actions nodes are generated as follows |
| 115 // <push affected registers on backtrack stack> |
| 116 // <generate code to perform action> |
| 117 // push backtrack code location |
| 118 // <generate code to test for following nodes> |
| 119 // backtrack code location: |
| 120 // <pop affected registers to restore their state> |
| 121 // <pop backtrack location from stack and go to it> |
| 122 // |
| 123 // * Matching nodes are generated as follows: |
| 124 // if input string matches at current position |
| 125 // update current position |
| 126 // <generate code to test for following nodes> |
| 127 // else |
| 128 // <pop backtrack location from stack and go to it> |
| 129 // |
| 130 // Thus it can be seen that the current position is saved and restored |
| 131 // by the choice nodes, whereas the registers are saved and restored by |
| 132 // by the action nodes that manipulate them. |
| 133 // |
| 134 // The other interesting aspect of this model is that nodes are generated |
| 135 // at the point where they are needed by a recursive call to Emit(). If |
| 136 // the node has already been code generated then the Emit() call will |
| 137 // generate a jump to the previously generated code instead. In order to |
| 138 // limit recursion it is possible for the Emit() function to put the node |
| 139 // on a work list for later generation and instead generate a jump. The |
| 140 // destination of the jump is resolved later when the code is generated. |
| 141 // |
| 142 // Actual regular expression code generation. |
| 143 // |
| 144 // Code generation is actually more complicated than the above. In order |
| 145 // to improve the efficiency of the generated code some optimizations are |
| 146 // performed |
| 147 // |
| 148 // * Choice nodes have 1-character lookahead. |
| 149 // A choice node looks at the following character and eliminates some of |
| 150 // the choices immediately based on that character. This is not yet |
| 151 // implemented. |
| 152 // * Simple greedy loops store reduced backtracking information. |
| 153 // A quantifier like /.*foo/m will greedily match the whole input. It will |
| 154 // then need to backtrack to a point where it can match "foo". The naive |
| 155 // implementation of this would push each character position onto the |
| 156 // backtracking stack, then pop them off one by one. This would use space |
| 157 // proportional to the length of the input string. However since the "." |
| 158 // can only match in one way and always has a constant length (in this case |
| 159 // of 1) it suffices to store the current position on the top of the stack |
| 160 // once. Matching now becomes merely incrementing the current position and |
| 161 // backtracking becomes decrementing the current position and checking the |
| 162 // result against the stored current position. This is faster and saves |
| 163 // space. |
| 164 // * The current state is virtualized. |
| 165 // This is used to defer expensive operations until it is clear that they |
| 166 // are needed and to generate code for a node more than once, allowing |
| 167 // specialized an efficient versions of the code to be created. This is |
| 168 // explained in the section below. |
| 169 // |
| 170 // Execution state virtualization. |
| 171 // |
| 172 // Instead of emitting code, nodes that manipulate the state can record their |
| 173 // manipulation in an object called the Trace. The Trace object can record a |
| 174 // current position offset, an optional backtrack code location on the top of |
| 175 // the virtualized backtrack stack and some register changes. When a node is |
| 176 // to be emitted it can flush the Trace or update it. Flushing the Trace |
| 177 // will emit code to bring the actual state into line with the virtual state. |
| 178 // Avoiding flushing the state can postpone some work (e.g. updates of capture |
| 179 // registers). Postponing work can save time when executing the regular |
| 180 // expression since it may be found that the work never has to be done as a |
| 181 // failure to match can occur. In addition it is much faster to jump to a |
| 182 // known backtrack code location than it is to pop an unknown backtrack |
| 183 // location from the stack and jump there. |
| 184 // |
| 185 // The virtual state found in the Trace affects code generation. For example |
| 186 // the virtual state contains the difference between the actual current |
| 187 // position and the virtual current position, and matching code needs to use |
| 188 // this offset to attempt a match in the correct location of the input |
| 189 // string. Therefore code generated for a non-trivial trace is specialized |
| 190 // to that trace. The code generator therefore has the ability to generate |
| 191 // code for each node several times. In order to limit the size of the |
| 192 // generated code there is an arbitrary limit on how many specialized sets of |
| 193 // code may be generated for a given node. If the limit is reached, the |
| 194 // trace is flushed and a generic version of the code for a node is emitted. |
| 195 // This is subsequently used for that node. The code emitted for non-generic |
| 196 // trace is not recorded in the node and so it cannot currently be reused in |
| 197 // the event that code generation is requested for an identical trace. |
| 198 |
| 199 |
| 200 void RegExpTree::AppendToText(RegExpText* text) { |
| 201 UNREACHABLE(); |
| 202 } |
| 203 |
| 204 |
| 205 void RegExpAtom::AppendToText(RegExpText* text) { |
| 206 text->AddElement(TextElement::Atom(this)); |
| 207 } |
| 208 |
| 209 |
| 210 void RegExpCharacterClass::AppendToText(RegExpText* text) { |
| 211 text->AddElement(TextElement::CharClass(this)); |
| 212 } |
| 213 |
| 214 |
| 215 void RegExpText::AppendToText(RegExpText* text) { |
| 216 for (intptr_t i = 0; i < elements()->length(); i++) |
| 217 text->AddElement((*elements())[i]); |
| 218 } |
| 219 |
| 220 |
| 221 TextElement TextElement::Atom(RegExpAtom* atom) { |
| 222 return TextElement(ATOM, atom); |
| 223 } |
| 224 |
| 225 |
| 226 TextElement TextElement::CharClass(RegExpCharacterClass* char_class) { |
| 227 return TextElement(CHAR_CLASS, char_class); |
| 228 } |
| 229 |
| 230 |
| 231 intptr_t TextElement::length() const { |
| 232 switch (text_type()) { |
| 233 case ATOM: |
| 234 return atom()->length(); |
| 235 |
| 236 case CHAR_CLASS: |
| 237 return 1; |
| 238 } |
| 239 UNREACHABLE(); |
| 240 return 0; |
| 241 } |
| 242 |
| 243 |
| 244 class FrequencyCollator : public ValueObject { |
| 245 public: |
| 246 FrequencyCollator() : total_samples_(0) { |
| 247 for (intptr_t i = 0; i < RegExpMacroAssembler::kTableSize; i++) { |
| 248 frequencies_[i] = CharacterFrequency(i); |
| 249 } |
| 250 } |
| 251 |
| 252 void CountCharacter(intptr_t character) { |
| 253 intptr_t index = (character & RegExpMacroAssembler::kTableMask); |
| 254 frequencies_[index].Increment(); |
| 255 total_samples_++; |
| 256 } |
| 257 |
| 258 // Does not measure in percent, but rather per-128 (the table size from the |
| 259 // regexp macro assembler). |
| 260 intptr_t Frequency(intptr_t in_character) { |
| 261 ASSERT((in_character & RegExpMacroAssembler::kTableMask) == in_character); |
| 262 if (total_samples_ < 1) return 1; // Division by zero. |
| 263 intptr_t freq_in_per128 = |
| 264 (frequencies_[in_character].counter() * 128) / total_samples_; |
| 265 return freq_in_per128; |
| 266 } |
| 267 |
| 268 private: |
| 269 class CharacterFrequency { |
| 270 public: |
| 271 CharacterFrequency() : counter_(0), character_(-1) { } |
| 272 explicit CharacterFrequency(intptr_t character) |
| 273 : counter_(0), character_(character) { } |
| 274 |
| 275 void Increment() { counter_++; } |
| 276 intptr_t counter() { return counter_; } |
| 277 intptr_t character() { return character_; } |
| 278 |
| 279 private: |
| 280 intptr_t counter_; |
| 281 intptr_t character_; |
| 282 |
| 283 DISALLOW_ALLOCATION(); |
| 284 }; |
| 285 |
| 286 |
| 287 private: |
| 288 CharacterFrequency frequencies_[RegExpMacroAssembler::kTableSize]; |
| 289 intptr_t total_samples_; |
| 290 }; |
| 291 |
| 292 |
| 293 class RegExpCompiler : public ValueObject { |
| 294 public: |
| 295 RegExpCompiler(intptr_t capture_count, |
| 296 bool ignore_case, |
| 297 intptr_t specialization_cid); |
| 298 |
| 299 intptr_t AllocateRegister() { |
| 300 return next_register_++; |
| 301 } |
| 302 |
| 303 RegExpEngine::CompilationResult Assemble(IRRegExpMacroAssembler* assembler, |
| 304 RegExpNode* start, |
| 305 intptr_t capture_count, |
| 306 const String& pattern); |
| 307 |
| 308 inline void AddWork(RegExpNode* node) { work_list_->Add(node); } |
| 309 |
| 310 static const intptr_t kImplementationOffset = 0; |
| 311 static const intptr_t kNumberOfRegistersOffset = 0; |
| 312 static const intptr_t kCodeOffset = 1; |
| 313 |
| 314 IRRegExpMacroAssembler* macro_assembler() { return macro_assembler_; } |
| 315 EndNode* accept() { return accept_; } |
| 316 |
| 317 static const intptr_t kMaxRecursion = 100; |
| 318 inline intptr_t recursion_depth() { return recursion_depth_; } |
| 319 inline void IncrementRecursionDepth() { recursion_depth_++; } |
| 320 inline void DecrementRecursionDepth() { recursion_depth_--; } |
| 321 |
| 322 void SetRegExpTooBig() { reg_exp_too_big_ = true; } |
| 323 |
| 324 inline bool ignore_case() { return ignore_case_; } |
| 325 inline bool one_byte() const { |
| 326 return (specialization_cid_ == kOneByteStringCid || |
| 327 specialization_cid_ == kExternalOneByteStringCid); |
| 328 } |
| 329 inline intptr_t specialization_cid() { return specialization_cid_; } |
| 330 FrequencyCollator* frequency_collator() { return &frequency_collator_; } |
| 331 |
| 332 intptr_t current_expansion_factor() { return current_expansion_factor_; } |
| 333 void set_current_expansion_factor(intptr_t value) { |
| 334 current_expansion_factor_ = value; |
| 335 } |
| 336 |
| 337 Isolate* isolate() const { return isolate_; } |
| 338 |
| 339 static const intptr_t kNoRegister = -1; |
| 340 |
| 341 private: |
| 342 EndNode* accept_; |
| 343 intptr_t next_register_; |
| 344 ZoneGrowableArray<RegExpNode*>* work_list_; |
| 345 intptr_t recursion_depth_; |
| 346 IRRegExpMacroAssembler* macro_assembler_; |
| 347 bool ignore_case_; |
| 348 intptr_t specialization_cid_; |
| 349 bool reg_exp_too_big_; |
| 350 intptr_t current_expansion_factor_; |
| 351 FrequencyCollator frequency_collator_; |
| 352 Isolate* isolate_; |
| 353 }; |
| 354 |
| 355 |
| 356 class RecursionCheck : public ValueObject { |
| 357 public: |
| 358 explicit RecursionCheck(RegExpCompiler* compiler) : compiler_(compiler) { |
| 359 compiler->IncrementRecursionDepth(); |
| 360 } |
| 361 ~RecursionCheck() { compiler_->DecrementRecursionDepth(); } |
| 362 private: |
| 363 RegExpCompiler* compiler_; |
| 364 }; |
| 365 |
| 366 |
| 367 static RegExpEngine::CompilationResult IrregexpRegExpTooBig() { |
| 368 return RegExpEngine::CompilationResult("RegExp too big"); |
| 369 } |
| 370 |
| 371 |
| 372 // Attempts to compile the regexp using an Irregexp code generator. Returns |
| 373 // a fixed array or a null handle depending on whether it succeeded. |
| 374 RegExpCompiler::RegExpCompiler(intptr_t capture_count, bool ignore_case, |
| 375 intptr_t specialization_cid) |
| 376 : next_register_(2 * (capture_count + 1)), |
| 377 work_list_(NULL), |
| 378 recursion_depth_(0), |
| 379 ignore_case_(ignore_case), |
| 380 specialization_cid_(specialization_cid), |
| 381 reg_exp_too_big_(false), |
| 382 current_expansion_factor_(1), |
| 383 isolate_(Isolate::Current()) { |
| 384 accept_ = new(I) EndNode(EndNode::ACCEPT, I); |
| 385 } |
| 386 |
| 387 |
| 388 RegExpEngine::CompilationResult RegExpCompiler::Assemble( |
| 389 IRRegExpMacroAssembler* macro_assembler, |
| 390 RegExpNode* start, |
| 391 intptr_t capture_count, |
| 392 const String& pattern) { |
| 393 static const bool use_slow_safe_regexp_compiler = false; |
| 394 |
| 395 macro_assembler->set_slow_safe(use_slow_safe_regexp_compiler); |
| 396 macro_assembler_ = macro_assembler; |
| 397 |
| 398 ZoneGrowableArray<RegExpNode*> work_list(0); |
| 399 work_list_ = &work_list; |
| 400 BlockLabel fail; |
| 401 macro_assembler_->PushBacktrack(&fail); |
| 402 Trace new_trace; |
| 403 start->Emit(this, &new_trace); |
| 404 macro_assembler_->BindBlock(&fail); |
| 405 macro_assembler_->Fail(); |
| 406 while (!work_list.is_empty()) { |
| 407 work_list.RemoveLast()->Emit(this, &new_trace); |
| 408 } |
| 409 if (reg_exp_too_big_) return IrregexpRegExpTooBig(); |
| 410 |
| 411 macro_assembler->FinalizeIndirectGotos(); |
| 412 |
| 413 return RegExpEngine::CompilationResult(macro_assembler, |
| 414 macro_assembler->graph_entry(), |
| 415 macro_assembler->num_blocks(), |
| 416 macro_assembler->num_stack_locals()); |
| 417 } |
| 418 |
| 419 |
| 420 bool Trace::DeferredAction::Mentions(intptr_t that) { |
| 421 if (action_type() == ActionNode::CLEAR_CAPTURES) { |
| 422 Interval range = static_cast<DeferredClearCaptures*>(this)->range(); |
| 423 return range.Contains(that); |
| 424 } else { |
| 425 return reg() == that; |
| 426 } |
| 427 } |
| 428 |
| 429 |
| 430 bool Trace::mentions_reg(intptr_t reg) { |
| 431 for (DeferredAction* action = actions_; |
| 432 action != NULL; |
| 433 action = action->next()) { |
| 434 if (action->Mentions(reg)) |
| 435 return true; |
| 436 } |
| 437 return false; |
| 438 } |
| 439 |
| 440 |
| 441 bool Trace::GetStoredPosition(intptr_t reg, intptr_t* cp_offset) { |
| 442 ASSERT(*cp_offset == 0); |
| 443 for (DeferredAction* action = actions_; |
| 444 action != NULL; |
| 445 action = action->next()) { |
| 446 if (action->Mentions(reg)) { |
| 447 if (action->action_type() == ActionNode::STORE_POSITION) { |
| 448 *cp_offset = static_cast<DeferredCapture*>(action)->cp_offset(); |
| 449 return true; |
| 450 } else { |
| 451 return false; |
| 452 } |
| 453 } |
| 454 } |
| 455 return false; |
| 456 } |
| 457 |
| 458 |
| 459 // This is called as we come into a loop choice node and some other tricky |
| 460 // nodes. It normalizes the state of the code generator to ensure we can |
| 461 // generate generic code. |
| 462 intptr_t Trace::FindAffectedRegisters(OutSet* affected_registers, |
| 463 Isolate* isolate) { |
| 464 intptr_t max_register = RegExpCompiler::kNoRegister; |
| 465 for (DeferredAction* action = actions_; |
| 466 action != NULL; |
| 467 action = action->next()) { |
| 468 if (action->action_type() == ActionNode::CLEAR_CAPTURES) { |
| 469 Interval range = static_cast<DeferredClearCaptures*>(action)->range(); |
| 470 for (intptr_t i = range.from(); i <= range.to(); i++) |
| 471 affected_registers->Set(i, isolate); |
| 472 if (range.to() > max_register) max_register = range.to(); |
| 473 } else { |
| 474 affected_registers->Set(action->reg(), isolate); |
| 475 if (action->reg() > max_register) max_register = action->reg(); |
| 476 } |
| 477 } |
| 478 return max_register; |
| 479 } |
| 480 |
| 481 |
| 482 void Trace::RestoreAffectedRegisters(RegExpMacroAssembler* assembler, |
| 483 intptr_t max_register, |
| 484 const OutSet& registers_to_pop, |
| 485 const OutSet& registers_to_clear) { |
| 486 for (intptr_t reg = max_register; reg >= 0; reg--) { |
| 487 if (registers_to_pop.Get(reg)) { |
| 488 assembler->PopRegister(reg); |
| 489 } else if (registers_to_clear.Get(reg)) { |
| 490 intptr_t clear_to = reg; |
| 491 while (reg > 0 && registers_to_clear.Get(reg - 1)) { |
| 492 reg--; |
| 493 } |
| 494 assembler->ClearRegisters(reg, clear_to); |
| 495 } |
| 496 } |
| 497 } |
| 498 |
| 499 |
| 500 void Trace::PerformDeferredActions(RegExpMacroAssembler* assembler, |
| 501 intptr_t max_register, |
| 502 const OutSet& affected_registers, |
| 503 OutSet* registers_to_pop, |
| 504 OutSet* registers_to_clear, |
| 505 Isolate* isolate) { |
| 506 for (intptr_t reg = 0; reg <= max_register; reg++) { |
| 507 if (!affected_registers.Get(reg)) { |
| 508 continue; |
| 509 } |
| 510 |
| 511 // The chronologically first deferred action in the trace |
| 512 // is used to infer the action needed to restore a register |
| 513 // to its previous state (or not, if it's safe to ignore it). |
| 514 enum DeferredActionUndoType { IGNORE, RESTORE, CLEAR }; |
| 515 DeferredActionUndoType undo_action = IGNORE; |
| 516 |
| 517 intptr_t value = 0; |
| 518 bool absolute = false; |
| 519 bool clear = false; |
| 520 intptr_t store_position = -1; |
| 521 // This is a little tricky because we are scanning the actions in reverse |
| 522 // historical order (newest first). |
| 523 for (DeferredAction* action = actions_; |
| 524 action != NULL; |
| 525 action = action->next()) { |
| 526 if (action->Mentions(reg)) { |
| 527 switch (action->action_type()) { |
| 528 case ActionNode::SET_REGISTER: { |
| 529 Trace::DeferredSetRegister* psr = |
| 530 static_cast<Trace::DeferredSetRegister*>(action); |
| 531 if (!absolute) { |
| 532 value += psr->value(); |
| 533 absolute = true; |
| 534 } |
| 535 // SET_REGISTER is currently only used for newly introduced loop |
| 536 // counters. They can have a significant previous value if they |
| 537 // occour in a loop. TODO(lrn): Propagate this information, so |
| 538 // we can set undo_action to IGNORE if we know there is no value to |
| 539 // restore. |
| 540 undo_action = RESTORE; |
| 541 ASSERT(store_position == -1); |
| 542 ASSERT(!clear); |
| 543 break; |
| 544 } |
| 545 case ActionNode::INCREMENT_REGISTER: |
| 546 if (!absolute) { |
| 547 value++; |
| 548 } |
| 549 ASSERT(store_position == -1); |
| 550 ASSERT(!clear); |
| 551 undo_action = RESTORE; |
| 552 break; |
| 553 case ActionNode::STORE_POSITION: { |
| 554 Trace::DeferredCapture* pc = |
| 555 static_cast<Trace::DeferredCapture*>(action); |
| 556 if (!clear && store_position == -1) { |
| 557 store_position = pc->cp_offset(); |
| 558 } |
| 559 |
| 560 // For captures we know that stores and clears alternate. |
| 561 // Other register, are never cleared, and if the occur |
| 562 // inside a loop, they might be assigned more than once. |
| 563 if (reg <= 1) { |
| 564 // Registers zero and one, aka "capture zero", is |
| 565 // always set correctly if we succeed. There is no |
| 566 // need to undo a setting on backtrack, because we |
| 567 // will set it again or fail. |
| 568 undo_action = IGNORE; |
| 569 } else { |
| 570 undo_action = pc->is_capture() ? CLEAR : RESTORE; |
| 571 } |
| 572 ASSERT(!absolute); |
| 573 ASSERT(value == 0); |
| 574 break; |
| 575 } |
| 576 case ActionNode::CLEAR_CAPTURES: { |
| 577 // Since we're scanning in reverse order, if we've already |
| 578 // set the position we have to ignore historically earlier |
| 579 // clearing operations. |
| 580 if (store_position == -1) { |
| 581 clear = true; |
| 582 } |
| 583 undo_action = RESTORE; |
| 584 ASSERT(!absolute); |
| 585 ASSERT(value == 0); |
| 586 break; |
| 587 } |
| 588 default: |
| 589 UNREACHABLE(); |
| 590 break; |
| 591 } |
| 592 } |
| 593 } |
| 594 // Prepare for the undo-action (e.g., push if it's going to be popped). |
| 595 if (undo_action == RESTORE) { |
| 596 assembler->PushRegister(reg); |
| 597 registers_to_pop->Set(reg, isolate); |
| 598 } else if (undo_action == CLEAR) { |
| 599 registers_to_clear->Set(reg, isolate); |
| 600 } |
| 601 // Perform the chronologically last action (or accumulated increment) |
| 602 // for the register. |
| 603 if (store_position != -1) { |
| 604 assembler->WriteCurrentPositionToRegister(reg, store_position); |
| 605 } else if (clear) { |
| 606 assembler->ClearRegisters(reg, reg); |
| 607 } else if (absolute) { |
| 608 assembler->SetRegister(reg, value); |
| 609 } else if (value != 0) { |
| 610 assembler->AdvanceRegister(reg, value); |
| 611 } |
| 612 } |
| 613 } |
| 614 |
| 615 |
| 616 // This is called as we come into a loop choice node and some other tricky |
| 617 // nodes. It normalizes the state of the code generator to ensure we can |
| 618 // generate generic code. |
| 619 void Trace::Flush(RegExpCompiler* compiler, RegExpNode* successor) { |
| 620 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 621 |
| 622 ASSERT(!is_trivial()); |
| 623 |
| 624 if (actions_ == NULL && backtrack() == NULL) { |
| 625 // Here we just have some deferred cp advances to fix and we are back to |
| 626 // a normal situation. We may also have to forget some information gained |
| 627 // through a quick check that was already performed. |
| 628 if (cp_offset_ != 0) assembler->AdvanceCurrentPosition(cp_offset_); |
| 629 // Create a new trivial state and generate the node with that. |
| 630 Trace new_state; |
| 631 successor->Emit(compiler, &new_state); |
| 632 return; |
| 633 } |
| 634 |
| 635 // Generate deferred actions here along with code to undo them again. |
| 636 OutSet affected_registers; |
| 637 |
| 638 if (backtrack() != NULL) { |
| 639 // Here we have a concrete backtrack location. These are set up by choice |
| 640 // nodes and so they indicate that we have a deferred save of the current |
| 641 // position which we may need to emit here. |
| 642 assembler->PushCurrentPosition(); |
| 643 } |
| 644 |
| 645 intptr_t max_register = FindAffectedRegisters(&affected_registers, CI); |
| 646 OutSet registers_to_pop; |
| 647 OutSet registers_to_clear; |
| 648 PerformDeferredActions(assembler, |
| 649 max_register, |
| 650 affected_registers, |
| 651 ®isters_to_pop, |
| 652 ®isters_to_clear, |
| 653 CI); |
| 654 if (cp_offset_ != 0) { |
| 655 assembler->AdvanceCurrentPosition(cp_offset_); |
| 656 } |
| 657 |
| 658 // Create a new trivial state and generate the node with that. |
| 659 BlockLabel undo; |
| 660 assembler->PushBacktrack(&undo); |
| 661 Trace new_state; |
| 662 successor->Emit(compiler, &new_state); |
| 663 |
| 664 // On backtrack we need to restore state. |
| 665 assembler->BindBlock(&undo); |
| 666 RestoreAffectedRegisters(assembler, |
| 667 max_register, |
| 668 registers_to_pop, |
| 669 registers_to_clear); |
| 670 if (backtrack() == NULL) { |
| 671 assembler->Backtrack(); |
| 672 } else { |
| 673 assembler->PopCurrentPosition(); |
| 674 assembler->GoTo(backtrack()); |
| 675 } |
| 676 } |
| 677 |
| 678 |
| 679 void NegativeSubmatchSuccess::Emit(RegExpCompiler* compiler, Trace* trace) { |
| 680 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 681 |
| 682 // Omit flushing the trace. We discard the entire stack frame anyway. |
| 683 |
| 684 if (!label()->IsBound()) { |
| 685 // We are completely independent of the trace, since we ignore it, |
| 686 // so this code can be used as the generic version. |
| 687 assembler->BindBlock(label()); |
| 688 } |
| 689 |
| 690 // Throw away everything on the backtrack stack since the start |
| 691 // of the negative submatch and restore the character position. |
| 692 assembler->ReadCurrentPositionFromRegister(current_position_register_); |
| 693 assembler->ReadStackPointerFromRegister(stack_pointer_register_); |
| 694 if (clear_capture_count_ > 0) { |
| 695 // Clear any captures that might have been performed during the success |
| 696 // of the body of the negative look-ahead. |
| 697 int clear_capture_end = clear_capture_start_ + clear_capture_count_ - 1; |
| 698 assembler->ClearRegisters(clear_capture_start_, clear_capture_end); |
| 699 } |
| 700 // Now that we have unwound the stack we find at the top of the stack the |
| 701 // backtrack that the BeginSubmatch node got. |
| 702 assembler->Backtrack(); |
| 703 } |
| 704 |
| 705 |
| 706 void EndNode::Emit(RegExpCompiler* compiler, Trace* trace) { |
| 707 if (!trace->is_trivial()) { |
| 708 trace->Flush(compiler, this); |
| 709 return; |
| 710 } |
| 711 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 712 if (!label()->IsBound()) { |
| 713 assembler->BindBlock(label()); |
| 714 } |
| 715 switch (action_) { |
| 716 case ACCEPT: |
| 717 assembler->Succeed(); |
| 718 return; |
| 719 case BACKTRACK: |
| 720 assembler->GoTo(trace->backtrack()); |
| 721 return; |
| 722 case NEGATIVE_SUBMATCH_SUCCESS: |
| 723 // This case is handled in a different virtual method. |
| 724 UNREACHABLE(); |
| 725 } |
| 726 UNIMPLEMENTED(); |
| 727 } |
| 728 |
| 729 |
| 730 void GuardedAlternative::AddGuard(Guard* guard, Isolate* isolate) { |
| 731 if (guards_ == NULL) |
| 732 guards_ = new(isolate) ZoneGrowableArray<Guard*>(1); |
| 733 guards_->Add(guard); |
| 734 } |
| 735 |
| 736 |
| 737 ActionNode* ActionNode::SetRegister(intptr_t reg, |
| 738 intptr_t val, |
| 739 RegExpNode* on_success) { |
| 740 ActionNode* result = |
| 741 new(on_success->isolate()) ActionNode(SET_REGISTER, on_success); |
| 742 result->data_.u_store_register.reg = reg; |
| 743 result->data_.u_store_register.value = val; |
| 744 return result; |
| 745 } |
| 746 |
| 747 |
| 748 ActionNode* ActionNode::IncrementRegister(intptr_t reg, |
| 749 RegExpNode* on_success) { |
| 750 ActionNode* result = |
| 751 new(on_success->isolate()) ActionNode(INCREMENT_REGISTER, on_success); |
| 752 result->data_.u_increment_register.reg = reg; |
| 753 return result; |
| 754 } |
| 755 |
| 756 |
| 757 ActionNode* ActionNode::StorePosition(intptr_t reg, |
| 758 bool is_capture, |
| 759 RegExpNode* on_success) { |
| 760 ActionNode* result = |
| 761 new(on_success->isolate()) ActionNode(STORE_POSITION, on_success); |
| 762 result->data_.u_position_register.reg = reg; |
| 763 result->data_.u_position_register.is_capture = is_capture; |
| 764 return result; |
| 765 } |
| 766 |
| 767 |
| 768 ActionNode* ActionNode::ClearCaptures(Interval range, |
| 769 RegExpNode* on_success) { |
| 770 ActionNode* result = |
| 771 new(on_success->isolate()) ActionNode(CLEAR_CAPTURES, on_success); |
| 772 result->data_.u_clear_captures.range_from = range.from(); |
| 773 result->data_.u_clear_captures.range_to = range.to(); |
| 774 return result; |
| 775 } |
| 776 |
| 777 |
| 778 ActionNode* ActionNode::BeginSubmatch(intptr_t stack_reg, |
| 779 intptr_t position_reg, |
| 780 RegExpNode* on_success) { |
| 781 ActionNode* result = |
| 782 new(on_success->isolate()) ActionNode(BEGIN_SUBMATCH, on_success); |
| 783 result->data_.u_submatch.stack_pointer_register = stack_reg; |
| 784 result->data_.u_submatch.current_position_register = position_reg; |
| 785 return result; |
| 786 } |
| 787 |
| 788 |
| 789 ActionNode* ActionNode::PositiveSubmatchSuccess(intptr_t stack_reg, |
| 790 intptr_t position_reg, |
| 791 intptr_t clear_register_count, |
| 792 intptr_t clear_register_from, |
| 793 RegExpNode* on_success) { |
| 794 ActionNode* result = |
| 795 new(on_success->isolate()) ActionNode(POSITIVE_SUBMATCH_SUCCESS, |
| 796 on_success); |
| 797 result->data_.u_submatch.stack_pointer_register = stack_reg; |
| 798 result->data_.u_submatch.current_position_register = position_reg; |
| 799 result->data_.u_submatch.clear_register_count = clear_register_count; |
| 800 result->data_.u_submatch.clear_register_from = clear_register_from; |
| 801 return result; |
| 802 } |
| 803 |
| 804 |
| 805 ActionNode* ActionNode::EmptyMatchCheck(intptr_t start_register, |
| 806 intptr_t repetition_register, |
| 807 intptr_t repetition_limit, |
| 808 RegExpNode* on_success) { |
| 809 ActionNode* result = |
| 810 new(on_success->isolate()) ActionNode(EMPTY_MATCH_CHECK, on_success); |
| 811 result->data_.u_empty_match_check.start_register = start_register; |
| 812 result->data_.u_empty_match_check.repetition_register = repetition_register; |
| 813 result->data_.u_empty_match_check.repetition_limit = repetition_limit; |
| 814 return result; |
| 815 } |
| 816 |
| 817 |
| 818 #define DEFINE_ACCEPT(Type) \ |
| 819 void Type##Node::Accept(NodeVisitor* visitor) { \ |
| 820 visitor->Visit##Type(this); \ |
| 821 } |
| 822 FOR_EACH_NODE_TYPE(DEFINE_ACCEPT) |
| 823 #undef DEFINE_ACCEPT |
| 824 |
| 825 |
| 826 void LoopChoiceNode::Accept(NodeVisitor* visitor) { |
| 827 visitor->VisitLoopChoice(this); |
| 828 } |
| 829 |
| 830 |
| 831 // ------------------------------------------------------------------- |
| 832 // Emit code. |
| 833 |
| 834 |
| 835 void ChoiceNode::GenerateGuard(RegExpMacroAssembler* macro_assembler, |
| 836 Guard* guard, |
| 837 Trace* trace) { |
| 838 switch (guard->op()) { |
| 839 case Guard::LT: |
| 840 ASSERT(!trace->mentions_reg(guard->reg())); |
| 841 macro_assembler->IfRegisterGE(guard->reg(), |
| 842 guard->value(), |
| 843 trace->backtrack()); |
| 844 break; |
| 845 case Guard::GEQ: |
| 846 ASSERT(!trace->mentions_reg(guard->reg())); |
| 847 macro_assembler->IfRegisterLT(guard->reg(), |
| 848 guard->value(), |
| 849 trace->backtrack()); |
| 850 break; |
| 851 } |
| 852 } |
| 853 |
| 854 |
| 855 // Returns the number of characters in the equivalence class, omitting those |
| 856 // that cannot occur in the source string because it is ASCII. |
| 857 static intptr_t GetCaseIndependentLetters(uint16_t character, |
| 858 bool one_byte_subject, |
| 859 int32_t* letters) { |
| 860 unibrow::Mapping<unibrow::Ecma262UnCanonicalize> jsregexp_uncanonicalize; |
| 861 intptr_t length = jsregexp_uncanonicalize.get(character, '\0', letters); |
| 862 // Unibrow returns 0 or 1 for characters where case independence is |
| 863 // trivial. |
| 864 if (length == 0) { |
| 865 letters[0] = character; |
| 866 length = 1; |
| 867 } |
| 868 if (!one_byte_subject || character <= Symbols::kMaxOneCharCodeSymbol) { |
| 869 return length; |
| 870 } |
| 871 |
| 872 // The standard requires that non-ASCII characters cannot have ASCII |
| 873 // character codes in their equivalence class. |
| 874 // TODO(dcarney): issue 3550 this is not actually true for Latin1 anymore, |
| 875 // is it? For example, \u00C5 is equivalent to \u212B. |
| 876 return 0; |
| 877 } |
| 878 |
| 879 |
| 880 static inline bool EmitSimpleCharacter(Isolate* isolate, |
| 881 RegExpCompiler* compiler, |
| 882 uint16_t c, |
| 883 BlockLabel* on_failure, |
| 884 intptr_t cp_offset, |
| 885 bool check, |
| 886 bool preloaded) { |
| 887 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 888 bool bound_checked = false; |
| 889 if (!preloaded) { |
| 890 assembler->LoadCurrentCharacter( |
| 891 cp_offset, |
| 892 on_failure, |
| 893 check); |
| 894 bound_checked = true; |
| 895 } |
| 896 assembler->CheckNotCharacter(c, on_failure); |
| 897 return bound_checked; |
| 898 } |
| 899 |
| 900 |
| 901 // Only emits non-letters (things that don't have case). Only used for case |
| 902 // independent matches. |
| 903 static inline bool EmitAtomNonLetter(Isolate* isolate, |
| 904 RegExpCompiler* compiler, |
| 905 uint16_t c, |
| 906 BlockLabel* on_failure, |
| 907 intptr_t cp_offset, |
| 908 bool check, |
| 909 bool preloaded) { |
| 910 RegExpMacroAssembler* macro_assembler = compiler->macro_assembler(); |
| 911 bool one_byte = compiler->one_byte(); |
| 912 int32_t chars[unibrow::Ecma262UnCanonicalize::kMaxWidth]; |
| 913 intptr_t length = GetCaseIndependentLetters(c, one_byte, chars); |
| 914 if (length < 1) { |
| 915 // This can't match. Must be an one-byte subject and a non-one-byte |
| 916 // character. We do not need to do anything since the one-byte pass |
| 917 // already handled this. |
| 918 return false; // Bounds not checked. |
| 919 } |
| 920 bool checked = false; |
| 921 // We handle the length > 1 case in a later pass. |
| 922 if (length == 1) { |
| 923 if (one_byte && c > Symbols::kMaxOneCharCodeSymbol) { |
| 924 // Can't match - see above. |
| 925 return false; // Bounds not checked. |
| 926 } |
| 927 if (!preloaded) { |
| 928 macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check); |
| 929 checked = check; |
| 930 } |
| 931 macro_assembler->CheckNotCharacter(c, on_failure); |
| 932 } |
| 933 return checked; |
| 934 } |
| 935 |
| 936 |
| 937 static bool ShortCutEmitCharacterPair(RegExpMacroAssembler* macro_assembler, |
| 938 bool one_byte, |
| 939 uint16_t c1, |
| 940 uint16_t c2, |
| 941 BlockLabel* on_failure) { |
| 942 uint16_t char_mask; |
| 943 if (one_byte) { |
| 944 char_mask = Symbols::kMaxOneCharCodeSymbol; |
| 945 } else { |
| 946 char_mask = Utf16::kMaxCodeUnit; |
| 947 } |
| 948 uint16_t exor = c1 ^ c2; |
| 949 // Check whether exor has only one bit set. |
| 950 if (((exor - 1) & exor) == 0) { |
| 951 // If c1 and c2 differ only by one bit. |
| 952 // Ecma262UnCanonicalize always gives the highest number last. |
| 953 ASSERT(c2 > c1); |
| 954 uint16_t mask = char_mask ^ exor; |
| 955 macro_assembler->CheckNotCharacterAfterAnd(c1, mask, on_failure); |
| 956 return true; |
| 957 } |
| 958 ASSERT(c2 > c1); |
| 959 uint16_t diff = c2 - c1; |
| 960 if (((diff - 1) & diff) == 0 && c1 >= diff) { |
| 961 // If the characters differ by 2^n but don't differ by one bit then |
| 962 // subtract the difference from the found character, then do the or |
| 963 // trick. We avoid the theoretical case where negative numbers are |
| 964 // involved in order to simplify code generation. |
| 965 uint16_t mask = char_mask ^ diff; |
| 966 macro_assembler->CheckNotCharacterAfterMinusAnd(c1 - diff, |
| 967 diff, |
| 968 mask, |
| 969 on_failure); |
| 970 return true; |
| 971 } |
| 972 return false; |
| 973 } |
| 974 |
| 975 |
| 976 typedef bool EmitCharacterFunction(Isolate* isolate, |
| 977 RegExpCompiler* compiler, |
| 978 uint16_t c, |
| 979 BlockLabel* on_failure, |
| 980 intptr_t cp_offset, |
| 981 bool check, |
| 982 bool preloaded); |
| 983 |
| 984 // Only emits letters (things that have case). Only used for case independent |
| 985 // matches. |
| 986 static inline bool EmitAtomLetter(Isolate* isolate, |
| 987 RegExpCompiler* compiler, |
| 988 uint16_t c, |
| 989 BlockLabel* on_failure, |
| 990 intptr_t cp_offset, |
| 991 bool check, |
| 992 bool preloaded) { |
| 993 RegExpMacroAssembler* macro_assembler = compiler->macro_assembler(); |
| 994 bool one_byte = compiler->one_byte(); |
| 995 int32_t chars[unibrow::Ecma262UnCanonicalize::kMaxWidth]; |
| 996 intptr_t length = GetCaseIndependentLetters(c, one_byte, chars); |
| 997 if (length <= 1) return false; |
| 998 // We may not need to check against the end of the input string |
| 999 // if this character lies before a character that matched. |
| 1000 if (!preloaded) { |
| 1001 macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check); |
| 1002 } |
| 1003 BlockLabel ok; |
| 1004 ASSERT(unibrow::Ecma262UnCanonicalize::kMaxWidth == 4); |
| 1005 switch (length) { |
| 1006 case 2: { |
| 1007 if (ShortCutEmitCharacterPair(macro_assembler, |
| 1008 one_byte, |
| 1009 chars[0], |
| 1010 chars[1], |
| 1011 on_failure)) { |
| 1012 } else { |
| 1013 macro_assembler->CheckCharacter(chars[0], &ok); |
| 1014 macro_assembler->CheckNotCharacter(chars[1], on_failure); |
| 1015 macro_assembler->BindBlock(&ok); |
| 1016 } |
| 1017 break; |
| 1018 } |
| 1019 case 4: |
| 1020 macro_assembler->CheckCharacter(chars[3], &ok); |
| 1021 // Fall through! |
| 1022 case 3: |
| 1023 macro_assembler->CheckCharacter(chars[0], &ok); |
| 1024 macro_assembler->CheckCharacter(chars[1], &ok); |
| 1025 macro_assembler->CheckNotCharacter(chars[2], on_failure); |
| 1026 macro_assembler->BindBlock(&ok); |
| 1027 break; |
| 1028 default: |
| 1029 UNREACHABLE(); |
| 1030 break; |
| 1031 } |
| 1032 return true; |
| 1033 } |
| 1034 |
| 1035 |
| 1036 static void EmitBoundaryTest(RegExpMacroAssembler* masm, |
| 1037 intptr_t border, |
| 1038 BlockLabel* fall_through, |
| 1039 BlockLabel* above_or_equal, |
| 1040 BlockLabel* below) { |
| 1041 if (below != fall_through) { |
| 1042 masm->CheckCharacterLT(border, below); |
| 1043 if (above_or_equal != fall_through) masm->GoTo(above_or_equal); |
| 1044 } else { |
| 1045 masm->CheckCharacterGT(border - 1, above_or_equal); |
| 1046 } |
| 1047 } |
| 1048 |
| 1049 |
| 1050 static void EmitDoubleBoundaryTest(RegExpMacroAssembler* masm, |
| 1051 intptr_t first, |
| 1052 intptr_t last, |
| 1053 BlockLabel* fall_through, |
| 1054 BlockLabel* in_range, |
| 1055 BlockLabel* out_of_range) { |
| 1056 if (in_range == fall_through) { |
| 1057 if (first == last) { |
| 1058 masm->CheckNotCharacter(first, out_of_range); |
| 1059 } else { |
| 1060 masm->CheckCharacterNotInRange(first, last, out_of_range); |
| 1061 } |
| 1062 } else { |
| 1063 if (first == last) { |
| 1064 masm->CheckCharacter(first, in_range); |
| 1065 } else { |
| 1066 masm->CheckCharacterInRange(first, last, in_range); |
| 1067 } |
| 1068 if (out_of_range != fall_through) masm->GoTo(out_of_range); |
| 1069 } |
| 1070 } |
| 1071 |
| 1072 |
| 1073 // even_label is for ranges[i] to ranges[i + 1] where i - start_index is even. |
| 1074 // odd_label is for ranges[i] to ranges[i + 1] where i - start_index is odd. |
| 1075 static void EmitUseLookupTable( |
| 1076 RegExpMacroAssembler* masm, |
| 1077 ZoneGrowableArray<int>* ranges, |
| 1078 intptr_t start_index, |
| 1079 intptr_t end_index, |
| 1080 intptr_t min_char, |
| 1081 BlockLabel* fall_through, |
| 1082 BlockLabel* even_label, |
| 1083 BlockLabel* odd_label) { |
| 1084 static const intptr_t kSize = RegExpMacroAssembler::kTableSize; |
| 1085 static const intptr_t kMask = RegExpMacroAssembler::kTableMask; |
| 1086 |
| 1087 intptr_t base = (min_char & ~kMask); |
| 1088 |
| 1089 // Assert that everything is on one kTableSize page. |
| 1090 for (intptr_t i = start_index; i <= end_index; i++) { |
| 1091 ASSERT((ranges->At(i) & ~kMask) == base); |
| 1092 } |
| 1093 ASSERT(start_index == 0 || (ranges->At(start_index - 1) & ~kMask) <= base); |
| 1094 |
| 1095 char templ[kSize]; |
| 1096 BlockLabel* on_bit_set; |
| 1097 BlockLabel* on_bit_clear; |
| 1098 intptr_t bit; |
| 1099 if (even_label == fall_through) { |
| 1100 on_bit_set = odd_label; |
| 1101 on_bit_clear = even_label; |
| 1102 bit = 1; |
| 1103 } else { |
| 1104 on_bit_set = even_label; |
| 1105 on_bit_clear = odd_label; |
| 1106 bit = 0; |
| 1107 } |
| 1108 for (intptr_t i = 0; i < (ranges->At(start_index) & kMask) && i < kSize; |
| 1109 i++) { |
| 1110 templ[i] = bit; |
| 1111 } |
| 1112 intptr_t j = 0; |
| 1113 bit ^= 1; |
| 1114 for (intptr_t i = start_index; i < end_index; i++) { |
| 1115 for (j = (ranges->At(i) & kMask); j < (ranges->At(i + 1) & kMask); j++) { |
| 1116 templ[j] = bit; |
| 1117 } |
| 1118 bit ^= 1; |
| 1119 } |
| 1120 for (intptr_t i = j; i < kSize; i++) { |
| 1121 templ[i] = bit; |
| 1122 } |
| 1123 // TODO(erikcorry): Cache these. |
| 1124 const TypedData& ba = TypedData::ZoneHandle( |
| 1125 masm->isolate(), |
| 1126 TypedData::New(kTypedDataUint8ArrayCid, kSize, Heap::kOld)); |
| 1127 for (intptr_t i = 0; i < kSize; i++) { |
| 1128 ba.SetUint8(i, templ[i]); |
| 1129 } |
| 1130 masm->CheckBitInTable(ba, on_bit_set); |
| 1131 if (on_bit_clear != fall_through) masm->GoTo(on_bit_clear); |
| 1132 } |
| 1133 |
| 1134 |
| 1135 static void CutOutRange(RegExpMacroAssembler* masm, |
| 1136 ZoneGrowableArray<int>* ranges, |
| 1137 intptr_t start_index, |
| 1138 intptr_t end_index, |
| 1139 intptr_t cut_index, |
| 1140 BlockLabel* even_label, |
| 1141 BlockLabel* odd_label) { |
| 1142 bool odd = (((cut_index - start_index) & 1) == 1); |
| 1143 BlockLabel* in_range_label = odd ? odd_label : even_label; |
| 1144 BlockLabel dummy; |
| 1145 EmitDoubleBoundaryTest(masm, |
| 1146 ranges->At(cut_index), |
| 1147 ranges->At(cut_index + 1) - 1, |
| 1148 &dummy, |
| 1149 in_range_label, |
| 1150 &dummy); |
| 1151 ASSERT(!dummy.IsLinked()); |
| 1152 // Cut out the single range by rewriting the array. This creates a new |
| 1153 // range that is a merger of the two ranges on either side of the one we |
| 1154 // are cutting out. The oddity of the labels is preserved. |
| 1155 for (intptr_t j = cut_index; j > start_index; j--) { |
| 1156 (*ranges)[j] = ranges->At(j - 1); |
| 1157 } |
| 1158 for (intptr_t j = cut_index + 1; j < end_index; j++) { |
| 1159 (*ranges)[j] = ranges->At(j + 1); |
| 1160 } |
| 1161 } |
| 1162 |
| 1163 |
| 1164 // Unicode case. Split the search space into kSize spaces that are handled |
| 1165 // with recursion. |
| 1166 static void SplitSearchSpace(ZoneGrowableArray<int>* ranges, |
| 1167 intptr_t start_index, |
| 1168 intptr_t end_index, |
| 1169 intptr_t* new_start_index, |
| 1170 intptr_t* new_end_index, |
| 1171 intptr_t* border) { |
| 1172 static const intptr_t kSize = RegExpMacroAssembler::kTableSize; |
| 1173 static const intptr_t kMask = RegExpMacroAssembler::kTableMask; |
| 1174 |
| 1175 intptr_t first = ranges->At(start_index); |
| 1176 intptr_t last = ranges->At(end_index) - 1; |
| 1177 |
| 1178 *new_start_index = start_index; |
| 1179 *border = (ranges->At(start_index) & ~kMask) + kSize; |
| 1180 while (*new_start_index < end_index) { |
| 1181 if (ranges->At(*new_start_index) > *border) break; |
| 1182 (*new_start_index)++; |
| 1183 } |
| 1184 // new_start_index is the index of the first edge that is beyond the |
| 1185 // current kSize space. |
| 1186 |
| 1187 // For very large search spaces we do a binary chop search of the non-Latin1 |
| 1188 // space instead of just going to the end of the current kSize space. The |
| 1189 // heuristics are complicated a little by the fact that any 128-character |
| 1190 // encoding space can be quickly tested with a table lookup, so we don't |
| 1191 // wish to do binary chop search at a smaller granularity than that. A |
| 1192 // 128-character space can take up a lot of space in the ranges array if, |
| 1193 // for example, we only want to match every second character (eg. the lower |
| 1194 // case characters on some Unicode pages). |
| 1195 intptr_t binary_chop_index = (end_index + start_index) / 2; |
| 1196 // The first test ensures that we get to the code that handles the Latin1 |
| 1197 // range with a single not-taken branch, speeding up this important |
| 1198 // character range (even non-Latin1 charset-based text has spaces and |
| 1199 // punctuation). |
| 1200 if (*border - 1 > Symbols::kMaxOneCharCodeSymbol && // Latin1 case. |
| 1201 end_index - start_index > (*new_start_index - start_index) * 2 && |
| 1202 last - first > kSize * 2 && |
| 1203 binary_chop_index > *new_start_index && |
| 1204 ranges->At(binary_chop_index) >= first + 2 * kSize) { |
| 1205 intptr_t scan_forward_for_section_border = binary_chop_index;; |
| 1206 intptr_t new_border = (ranges->At(binary_chop_index) | kMask) + 1; |
| 1207 |
| 1208 while (scan_forward_for_section_border < end_index) { |
| 1209 if (ranges->At(scan_forward_for_section_border) > new_border) { |
| 1210 *new_start_index = scan_forward_for_section_border; |
| 1211 *border = new_border; |
| 1212 break; |
| 1213 } |
| 1214 scan_forward_for_section_border++; |
| 1215 } |
| 1216 } |
| 1217 |
| 1218 ASSERT(*new_start_index > start_index); |
| 1219 *new_end_index = *new_start_index - 1; |
| 1220 if (ranges->At(*new_end_index) == *border) { |
| 1221 (*new_end_index)--; |
| 1222 } |
| 1223 if (*border >= ranges->At(end_index)) { |
| 1224 *border = ranges->At(end_index); |
| 1225 *new_start_index = end_index; // Won't be used. |
| 1226 *new_end_index = end_index - 1; |
| 1227 } |
| 1228 } |
| 1229 |
| 1230 |
| 1231 // Gets a series of segment boundaries representing a character class. If the |
| 1232 // character is in the range between an even and an odd boundary (counting from |
| 1233 // start_index) then go to even_label, otherwise go to odd_label. We already |
| 1234 // know that the character is in the range of min_char to max_char inclusive. |
| 1235 // Either label can be NULL indicating backtracking. Either label can also be |
| 1236 // equal to the fall_through label. |
| 1237 static void GenerateBranches(RegExpMacroAssembler* masm, |
| 1238 ZoneGrowableArray<int>* ranges, |
| 1239 intptr_t start_index, |
| 1240 intptr_t end_index, |
| 1241 uint16_t min_char, |
| 1242 uint16_t max_char, |
| 1243 BlockLabel* fall_through, |
| 1244 BlockLabel* even_label, |
| 1245 BlockLabel* odd_label) { |
| 1246 intptr_t first = ranges->At(start_index); |
| 1247 intptr_t last = ranges->At(end_index) - 1; |
| 1248 |
| 1249 ASSERT(min_char < first); |
| 1250 |
| 1251 // Just need to test if the character is before or on-or-after |
| 1252 // a particular character. |
| 1253 if (start_index == end_index) { |
| 1254 EmitBoundaryTest(masm, first, fall_through, even_label, odd_label); |
| 1255 return; |
| 1256 } |
| 1257 |
| 1258 // Another almost trivial case: There is one interval in the middle that is |
| 1259 // different from the end intervals. |
| 1260 if (start_index + 1 == end_index) { |
| 1261 EmitDoubleBoundaryTest( |
| 1262 masm, first, last, fall_through, even_label, odd_label); |
| 1263 return; |
| 1264 } |
| 1265 |
| 1266 // It's not worth using table lookup if there are very few intervals in the |
| 1267 // character class. |
| 1268 if (end_index - start_index <= 6) { |
| 1269 // It is faster to test for individual characters, so we look for those |
| 1270 // first, then try arbitrary ranges in the second round. |
| 1271 static intptr_t kNoCutIndex = -1; |
| 1272 intptr_t cut = kNoCutIndex; |
| 1273 for (intptr_t i = start_index; i < end_index; i++) { |
| 1274 if (ranges->At(i) == ranges->At(i + 1) - 1) { |
| 1275 cut = i; |
| 1276 break; |
| 1277 } |
| 1278 } |
| 1279 if (cut == kNoCutIndex) cut = start_index; |
| 1280 CutOutRange( |
| 1281 masm, ranges, start_index, end_index, cut, even_label, odd_label); |
| 1282 ASSERT(end_index - start_index >= 2); |
| 1283 GenerateBranches(masm, |
| 1284 ranges, |
| 1285 start_index + 1, |
| 1286 end_index - 1, |
| 1287 min_char, |
| 1288 max_char, |
| 1289 fall_through, |
| 1290 even_label, |
| 1291 odd_label); |
| 1292 return; |
| 1293 } |
| 1294 |
| 1295 // If there are a lot of intervals in the regexp, then we will use tables to |
| 1296 // determine whether the character is inside or outside the character class. |
| 1297 static const intptr_t kBits = RegExpMacroAssembler::kTableSizeBits; |
| 1298 |
| 1299 if ((max_char >> kBits) == (min_char >> kBits)) { |
| 1300 EmitUseLookupTable(masm, |
| 1301 ranges, |
| 1302 start_index, |
| 1303 end_index, |
| 1304 min_char, |
| 1305 fall_through, |
| 1306 even_label, |
| 1307 odd_label); |
| 1308 return; |
| 1309 } |
| 1310 |
| 1311 if ((min_char >> kBits) != (first >> kBits)) { |
| 1312 masm->CheckCharacterLT(first, odd_label); |
| 1313 GenerateBranches(masm, |
| 1314 ranges, |
| 1315 start_index + 1, |
| 1316 end_index, |
| 1317 first, |
| 1318 max_char, |
| 1319 fall_through, |
| 1320 odd_label, |
| 1321 even_label); |
| 1322 return; |
| 1323 } |
| 1324 |
| 1325 intptr_t new_start_index = 0; |
| 1326 intptr_t new_end_index = 0; |
| 1327 intptr_t border = 0; |
| 1328 |
| 1329 SplitSearchSpace(ranges, |
| 1330 start_index, |
| 1331 end_index, |
| 1332 &new_start_index, |
| 1333 &new_end_index, |
| 1334 &border); |
| 1335 |
| 1336 BlockLabel handle_rest; |
| 1337 BlockLabel* above = &handle_rest; |
| 1338 if (border == last + 1) { |
| 1339 // We didn't find any section that started after the limit, so everything |
| 1340 // above the border is one of the terminal labels. |
| 1341 above = (end_index & 1) != (start_index & 1) ? odd_label : even_label; |
| 1342 ASSERT(new_end_index == end_index - 1); |
| 1343 } |
| 1344 |
| 1345 ASSERT(start_index <= new_end_index); |
| 1346 ASSERT(new_start_index <= end_index); |
| 1347 ASSERT(start_index < new_start_index); |
| 1348 ASSERT(new_end_index < end_index); |
| 1349 ASSERT(new_end_index + 1 == new_start_index || |
| 1350 (new_end_index + 2 == new_start_index && |
| 1351 border == ranges->At(new_end_index + 1))); |
| 1352 ASSERT(min_char < border - 1); |
| 1353 ASSERT(border < max_char); |
| 1354 ASSERT(ranges->At(new_end_index) < border); |
| 1355 ASSERT(border < ranges->At(new_start_index) || |
| 1356 (border == ranges->At(new_start_index) && |
| 1357 new_start_index == end_index && |
| 1358 new_end_index == end_index - 1 && |
| 1359 border == last + 1)); |
| 1360 ASSERT(new_start_index == 0 || border >= ranges->At(new_start_index - 1)); |
| 1361 |
| 1362 masm->CheckCharacterGT(border - 1, above); |
| 1363 BlockLabel dummy; |
| 1364 GenerateBranches(masm, |
| 1365 ranges, |
| 1366 start_index, |
| 1367 new_end_index, |
| 1368 min_char, |
| 1369 border - 1, |
| 1370 &dummy, |
| 1371 even_label, |
| 1372 odd_label); |
| 1373 |
| 1374 if (handle_rest.IsLinked()) { |
| 1375 masm->BindBlock(&handle_rest); |
| 1376 bool flip = (new_start_index & 1) != (start_index & 1); |
| 1377 GenerateBranches(masm, |
| 1378 ranges, |
| 1379 new_start_index, |
| 1380 end_index, |
| 1381 border, |
| 1382 max_char, |
| 1383 &dummy, |
| 1384 flip ? odd_label : even_label, |
| 1385 flip ? even_label : odd_label); |
| 1386 } |
| 1387 } |
| 1388 |
| 1389 |
| 1390 static void EmitCharClass(RegExpMacroAssembler* macro_assembler, |
| 1391 RegExpCharacterClass* cc, |
| 1392 bool one_byte, |
| 1393 BlockLabel* on_failure, |
| 1394 intptr_t cp_offset, |
| 1395 bool check_offset, |
| 1396 bool preloaded, |
| 1397 Isolate* isolate) { |
| 1398 ZoneGrowableArray<CharacterRange>* ranges = cc->ranges(); |
| 1399 if (!CharacterRange::IsCanonical(ranges)) { |
| 1400 CharacterRange::Canonicalize(ranges); |
| 1401 } |
| 1402 |
| 1403 intptr_t max_char; |
| 1404 if (one_byte) { |
| 1405 max_char = Symbols::kMaxOneCharCodeSymbol; |
| 1406 } else { |
| 1407 max_char = Utf16::kMaxCodeUnit; |
| 1408 } |
| 1409 |
| 1410 intptr_t range_count = ranges->length(); |
| 1411 |
| 1412 intptr_t last_valid_range = range_count - 1; |
| 1413 while (last_valid_range >= 0) { |
| 1414 CharacterRange& range = (*ranges)[last_valid_range]; |
| 1415 if (range.from() <= max_char) { |
| 1416 break; |
| 1417 } |
| 1418 last_valid_range--; |
| 1419 } |
| 1420 |
| 1421 if (last_valid_range < 0) { |
| 1422 if (!cc->is_negated()) { |
| 1423 macro_assembler->GoTo(on_failure); |
| 1424 } |
| 1425 if (check_offset) { |
| 1426 macro_assembler->CheckPosition(cp_offset, on_failure); |
| 1427 } |
| 1428 return; |
| 1429 } |
| 1430 |
| 1431 if (last_valid_range == 0 && |
| 1432 ranges->At(0).IsEverything(max_char)) { |
| 1433 if (cc->is_negated()) { |
| 1434 macro_assembler->GoTo(on_failure); |
| 1435 } else { |
| 1436 // This is a common case hit by non-anchored expressions. |
| 1437 if (check_offset) { |
| 1438 macro_assembler->CheckPosition(cp_offset, on_failure); |
| 1439 } |
| 1440 } |
| 1441 return; |
| 1442 } |
| 1443 if (last_valid_range == 0 && |
| 1444 !cc->is_negated() && |
| 1445 ranges->At(0).IsEverything(max_char)) { |
| 1446 // This is a common case hit by non-anchored expressions. |
| 1447 if (check_offset) { |
| 1448 macro_assembler->CheckPosition(cp_offset, on_failure); |
| 1449 } |
| 1450 return; |
| 1451 } |
| 1452 |
| 1453 if (!preloaded) { |
| 1454 macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check_offset); |
| 1455 } |
| 1456 |
| 1457 if (cc->is_standard() && |
| 1458 macro_assembler->CheckSpecialCharacterClass(cc->standard_type(), |
| 1459 on_failure)) { |
| 1460 return; |
| 1461 } |
| 1462 |
| 1463 |
| 1464 // A new list with ascending entries. Each entry is a code unit |
| 1465 // where there is a boundary between code units that are part of |
| 1466 // the class and code units that are not. Normally we insert an |
| 1467 // entry at zero which goes to the failure label, but if there |
| 1468 // was already one there we fall through for success on that entry. |
| 1469 // Subsequent entries have alternating meaning (success/failure). |
| 1470 ZoneGrowableArray<int>* range_boundaries = |
| 1471 new(isolate) ZoneGrowableArray<int>(last_valid_range); |
| 1472 |
| 1473 bool zeroth_entry_is_failure = !cc->is_negated(); |
| 1474 |
| 1475 for (intptr_t i = 0; i <= last_valid_range; i++) { |
| 1476 CharacterRange& range = (*ranges)[i]; |
| 1477 if (range.from() == 0) { |
| 1478 ASSERT(i == 0); |
| 1479 zeroth_entry_is_failure = !zeroth_entry_is_failure; |
| 1480 } else { |
| 1481 range_boundaries->Add(range.from()); |
| 1482 } |
| 1483 range_boundaries->Add(range.to() + 1); |
| 1484 } |
| 1485 intptr_t end_index = range_boundaries->length() - 1; |
| 1486 if (range_boundaries->At(end_index) > max_char) { |
| 1487 end_index--; |
| 1488 } |
| 1489 |
| 1490 BlockLabel fall_through; |
| 1491 GenerateBranches(macro_assembler, |
| 1492 range_boundaries, |
| 1493 0, // start_index. |
| 1494 end_index, |
| 1495 0, // min_char. |
| 1496 max_char, |
| 1497 &fall_through, |
| 1498 zeroth_entry_is_failure ? &fall_through : on_failure, |
| 1499 zeroth_entry_is_failure ? on_failure : &fall_through); |
| 1500 macro_assembler->BindBlock(&fall_through); |
| 1501 } |
| 1502 |
| 1503 |
| 1504 RegExpNode::~RegExpNode() { |
| 1505 } |
| 1506 |
| 1507 |
| 1508 RegExpNode::LimitResult RegExpNode::LimitVersions(RegExpCompiler* compiler, |
| 1509 Trace* trace) { |
| 1510 // If we are generating a greedy loop then don't stop and don't reuse code. |
| 1511 if (trace->stop_node() != NULL) { |
| 1512 return CONTINUE; |
| 1513 } |
| 1514 |
| 1515 RegExpMacroAssembler* macro_assembler = compiler->macro_assembler(); |
| 1516 if (trace->is_trivial()) { |
| 1517 if (label_.IsBound()) { |
| 1518 // We are being asked to generate a generic version, but that's already |
| 1519 // been done so just go to it. |
| 1520 macro_assembler->GoTo(&label_); |
| 1521 return DONE; |
| 1522 } |
| 1523 if (compiler->recursion_depth() >= RegExpCompiler::kMaxRecursion) { |
| 1524 // To avoid too deep recursion we push the node to the work queue and just |
| 1525 // generate a goto here. |
| 1526 compiler->AddWork(this); |
| 1527 macro_assembler->GoTo(&label_); |
| 1528 return DONE; |
| 1529 } |
| 1530 // Generate generic version of the node and bind the label for later use. |
| 1531 macro_assembler->BindBlock(&label_); |
| 1532 return CONTINUE; |
| 1533 } |
| 1534 |
| 1535 // We are being asked to make a non-generic version. Keep track of how many |
| 1536 // non-generic versions we generate so as not to overdo it. |
| 1537 trace_count_++; |
| 1538 if (kRegexpOptimization && |
| 1539 trace_count_ < kMaxCopiesCodeGenerated && |
| 1540 compiler->recursion_depth() <= RegExpCompiler::kMaxRecursion) { |
| 1541 return CONTINUE; |
| 1542 } |
| 1543 |
| 1544 // If we get here code has been generated for this node too many times or |
| 1545 // recursion is too deep. Time to switch to a generic version. The code for |
| 1546 // generic versions above can handle deep recursion properly. |
| 1547 trace->Flush(compiler, this); |
| 1548 return DONE; |
| 1549 } |
| 1550 |
| 1551 |
| 1552 intptr_t ActionNode::EatsAtLeast(intptr_t still_to_find, |
| 1553 intptr_t budget, |
| 1554 bool not_at_start) { |
| 1555 if (budget <= 0) return 0; |
| 1556 if (action_type_ == POSITIVE_SUBMATCH_SUCCESS) return 0; // Rewinds input! |
| 1557 return on_success()->EatsAtLeast(still_to_find, |
| 1558 budget - 1, |
| 1559 not_at_start); |
| 1560 } |
| 1561 |
| 1562 |
| 1563 void ActionNode::FillInBMInfo(intptr_t offset, |
| 1564 intptr_t budget, |
| 1565 BoyerMooreLookahead* bm, |
| 1566 bool not_at_start) { |
| 1567 if (action_type_ == BEGIN_SUBMATCH) { |
| 1568 bm->SetRest(offset); |
| 1569 } else if (action_type_ != POSITIVE_SUBMATCH_SUCCESS) { |
| 1570 on_success()->FillInBMInfo(offset, budget - 1, bm, not_at_start); |
| 1571 } |
| 1572 SaveBMInfo(bm, not_at_start, offset); |
| 1573 } |
| 1574 |
| 1575 |
| 1576 intptr_t AssertionNode::EatsAtLeast(intptr_t still_to_find, |
| 1577 intptr_t budget, |
| 1578 bool not_at_start) { |
| 1579 if (budget <= 0) return 0; |
| 1580 // If we know we are not at the start and we are asked "how many characters |
| 1581 // will you match if you succeed?" then we can answer anything since false |
| 1582 // implies false. So lets just return the max answer (still_to_find) since |
| 1583 // that won't prevent us from preloading a lot of characters for the other |
| 1584 // branches in the node graph. |
| 1585 if (assertion_type() == AT_START && not_at_start) return still_to_find; |
| 1586 return on_success()->EatsAtLeast(still_to_find, |
| 1587 budget - 1, |
| 1588 not_at_start); |
| 1589 } |
| 1590 |
| 1591 |
| 1592 void AssertionNode::FillInBMInfo(intptr_t offset, |
| 1593 intptr_t budget, |
| 1594 BoyerMooreLookahead* bm, |
| 1595 bool not_at_start) { |
| 1596 // Match the behaviour of EatsAtLeast on this node. |
| 1597 if (assertion_type() == AT_START && not_at_start) return; |
| 1598 on_success()->FillInBMInfo(offset, budget - 1, bm, not_at_start); |
| 1599 SaveBMInfo(bm, not_at_start, offset); |
| 1600 } |
| 1601 |
| 1602 |
| 1603 intptr_t BackReferenceNode::EatsAtLeast(intptr_t still_to_find, |
| 1604 intptr_t budget, |
| 1605 bool not_at_start) { |
| 1606 if (budget <= 0) return 0; |
| 1607 return on_success()->EatsAtLeast(still_to_find, |
| 1608 budget - 1, |
| 1609 not_at_start); |
| 1610 } |
| 1611 |
| 1612 |
| 1613 intptr_t TextNode::EatsAtLeast(intptr_t still_to_find, |
| 1614 intptr_t budget, |
| 1615 bool not_at_start) { |
| 1616 intptr_t answer = Length(); |
| 1617 if (answer >= still_to_find) return answer; |
| 1618 if (budget <= 0) return answer; |
| 1619 // We are not at start after this node so we set the last argument to 'true'. |
| 1620 return answer + on_success()->EatsAtLeast(still_to_find - answer, |
| 1621 budget - 1, |
| 1622 true); |
| 1623 } |
| 1624 |
| 1625 |
| 1626 intptr_t NegativeLookaheadChoiceNode::EatsAtLeast(intptr_t still_to_find, |
| 1627 intptr_t budget, |
| 1628 bool not_at_start) { |
| 1629 if (budget <= 0) return 0; |
| 1630 // Alternative 0 is the negative lookahead, alternative 1 is what comes |
| 1631 // afterwards. |
| 1632 RegExpNode* node = (*alternatives_)[1].node(); |
| 1633 return node->EatsAtLeast(still_to_find, budget - 1, not_at_start); |
| 1634 } |
| 1635 |
| 1636 |
| 1637 void NegativeLookaheadChoiceNode::GetQuickCheckDetails( |
| 1638 QuickCheckDetails* details, |
| 1639 RegExpCompiler* compiler, |
| 1640 intptr_t filled_in, |
| 1641 bool not_at_start) { |
| 1642 // Alternative 0 is the negative lookahead, alternative 1 is what comes |
| 1643 // afterwards. |
| 1644 RegExpNode* node = (*alternatives_)[1].node(); |
| 1645 return node->GetQuickCheckDetails(details, compiler, filled_in, not_at_start); |
| 1646 } |
| 1647 |
| 1648 |
| 1649 intptr_t ChoiceNode::EatsAtLeastHelper(intptr_t still_to_find, |
| 1650 intptr_t budget, |
| 1651 RegExpNode* ignore_this_node, |
| 1652 bool not_at_start) { |
| 1653 if (budget <= 0) return 0; |
| 1654 intptr_t min = 100; |
| 1655 intptr_t choice_count = alternatives_->length(); |
| 1656 budget = (budget - 1) / choice_count; |
| 1657 for (intptr_t i = 0; i < choice_count; i++) { |
| 1658 RegExpNode* node = (*alternatives_)[i].node(); |
| 1659 if (node == ignore_this_node) continue; |
| 1660 intptr_t node_eats_at_least = |
| 1661 node->EatsAtLeast(still_to_find, budget, not_at_start); |
| 1662 if (node_eats_at_least < min) min = node_eats_at_least; |
| 1663 if (min == 0) return 0; |
| 1664 } |
| 1665 return min; |
| 1666 } |
| 1667 |
| 1668 |
| 1669 intptr_t LoopChoiceNode::EatsAtLeast(intptr_t still_to_find, |
| 1670 intptr_t budget, |
| 1671 bool not_at_start) { |
| 1672 return EatsAtLeastHelper(still_to_find, |
| 1673 budget - 1, |
| 1674 loop_node_, |
| 1675 not_at_start); |
| 1676 } |
| 1677 |
| 1678 |
| 1679 intptr_t ChoiceNode::EatsAtLeast(intptr_t still_to_find, |
| 1680 intptr_t budget, |
| 1681 bool not_at_start) { |
| 1682 return EatsAtLeastHelper(still_to_find, |
| 1683 budget, |
| 1684 NULL, |
| 1685 not_at_start); |
| 1686 } |
| 1687 |
| 1688 |
| 1689 // Takes the left-most 1-bit and smears it out, setting all bits to its right. |
| 1690 static inline uint32_t SmearBitsRight(uint32_t v) { |
| 1691 v |= v >> 1; |
| 1692 v |= v >> 2; |
| 1693 v |= v >> 4; |
| 1694 v |= v >> 8; |
| 1695 v |= v >> 16; |
| 1696 return v; |
| 1697 } |
| 1698 |
| 1699 |
| 1700 bool QuickCheckDetails::Rationalize(bool asc) { |
| 1701 bool found_useful_op = false; |
| 1702 uint32_t char_mask; |
| 1703 if (asc) { |
| 1704 char_mask = Symbols::kMaxOneCharCodeSymbol; |
| 1705 } else { |
| 1706 char_mask = Utf16::kMaxCodeUnit; |
| 1707 } |
| 1708 mask_ = 0; |
| 1709 value_ = 0; |
| 1710 intptr_t char_shift = 0; |
| 1711 for (intptr_t i = 0; i < characters_; i++) { |
| 1712 Position* pos = &positions_[i]; |
| 1713 if ((pos->mask & Symbols::kMaxOneCharCodeSymbol) != 0) { |
| 1714 found_useful_op = true; |
| 1715 } |
| 1716 mask_ |= (pos->mask & char_mask) << char_shift; |
| 1717 value_ |= (pos->value & char_mask) << char_shift; |
| 1718 char_shift += asc ? 8 : 16; |
| 1719 } |
| 1720 return found_useful_op; |
| 1721 } |
| 1722 |
| 1723 |
| 1724 bool RegExpNode::EmitQuickCheck(RegExpCompiler* compiler, |
| 1725 Trace* bounds_check_trace, |
| 1726 Trace* trace, |
| 1727 bool preload_has_checked_bounds, |
| 1728 BlockLabel* on_possible_success, |
| 1729 QuickCheckDetails* details, |
| 1730 bool fall_through_on_failure) { |
| 1731 if (details->characters() == 0) return false; |
| 1732 GetQuickCheckDetails( |
| 1733 details, compiler, 0, trace->at_start() == Trace::FALSE_VALUE); |
| 1734 if (details->cannot_match()) return false; |
| 1735 if (!details->Rationalize(compiler->one_byte())) return false; |
| 1736 ASSERT(details->characters() == 1 || |
| 1737 compiler->macro_assembler()->CanReadUnaligned()); |
| 1738 uint32_t mask = details->mask(); |
| 1739 uint32_t value = details->value(); |
| 1740 |
| 1741 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 1742 |
| 1743 if (trace->characters_preloaded() != details->characters()) { |
| 1744 ASSERT(trace->cp_offset() == bounds_check_trace->cp_offset()); |
| 1745 // We are attempting to preload the minimum number of characters |
| 1746 // any choice would eat, so if the bounds check fails, then none of the |
| 1747 // choices can succeed, so we can just immediately backtrack, rather |
| 1748 // than go to the next choice. |
| 1749 assembler->LoadCurrentCharacter(trace->cp_offset(), |
| 1750 bounds_check_trace->backtrack(), |
| 1751 !preload_has_checked_bounds, |
| 1752 details->characters()); |
| 1753 } |
| 1754 |
| 1755 |
| 1756 bool need_mask = true; |
| 1757 |
| 1758 if (details->characters() == 1) { |
| 1759 // If number of characters preloaded is 1 then we used a byte or 16 bit |
| 1760 // load so the value is already masked down. |
| 1761 uint32_t char_mask; |
| 1762 if (compiler->one_byte()) { |
| 1763 char_mask = Symbols::kMaxOneCharCodeSymbol; |
| 1764 } else { |
| 1765 char_mask = Utf16::kMaxCodeUnit; |
| 1766 } |
| 1767 if ((mask & char_mask) == char_mask) need_mask = false; |
| 1768 mask &= char_mask; |
| 1769 } else { |
| 1770 // For 2-character preloads in one-byte mode or 1-character preloads in |
| 1771 // two-byte mode we also use a 16 bit load with zero extend. |
| 1772 if (details->characters() == 2 && compiler->one_byte()) { |
| 1773 if ((mask & 0xffff) == 0xffff) need_mask = false; |
| 1774 } else if (details->characters() == 1 && !compiler->one_byte()) { |
| 1775 if ((mask & 0xffff) == 0xffff) need_mask = false; |
| 1776 } else { |
| 1777 if (mask == 0xffffffff) need_mask = false; |
| 1778 } |
| 1779 } |
| 1780 |
| 1781 if (fall_through_on_failure) { |
| 1782 if (need_mask) { |
| 1783 assembler->CheckCharacterAfterAnd(value, mask, on_possible_success); |
| 1784 } else { |
| 1785 assembler->CheckCharacter(value, on_possible_success); |
| 1786 } |
| 1787 } else { |
| 1788 if (need_mask) { |
| 1789 assembler->CheckNotCharacterAfterAnd(value, mask, trace->backtrack()); |
| 1790 } else { |
| 1791 assembler->CheckNotCharacter(value, trace->backtrack()); |
| 1792 } |
| 1793 } |
| 1794 return true; |
| 1795 } |
| 1796 |
| 1797 |
| 1798 // Here is the meat of GetQuickCheckDetails (see also the comment on the |
| 1799 // super-class in the .h file). |
| 1800 // |
| 1801 // We iterate along the text object, building up for each character a |
| 1802 // mask and value that can be used to test for a quick failure to match. |
| 1803 // The masks and values for the positions will be combined into a single |
| 1804 // machine word for the current character width in order to be used in |
| 1805 // generating a quick check. |
| 1806 void TextNode::GetQuickCheckDetails(QuickCheckDetails* details, |
| 1807 RegExpCompiler* compiler, |
| 1808 intptr_t characters_filled_in, |
| 1809 bool not_at_start) { |
| 1810 ASSERT(characters_filled_in < details->characters()); |
| 1811 intptr_t characters = details->characters(); |
| 1812 intptr_t char_mask; |
| 1813 if (compiler->one_byte()) { |
| 1814 char_mask = Symbols::kMaxOneCharCodeSymbol; |
| 1815 } else { |
| 1816 char_mask = Utf16::kMaxCodeUnit; |
| 1817 } |
| 1818 for (intptr_t k = 0; k < elms_->length(); k++) { |
| 1819 TextElement elm = elms_->At(k); |
| 1820 if (elm.text_type() == TextElement::ATOM) { |
| 1821 ZoneGrowableArray<uint16_t>* quarks = elm.atom()->data(); |
| 1822 for (intptr_t i = 0; i < characters && i < quarks->length(); i++) { |
| 1823 QuickCheckDetails::Position* pos = |
| 1824 details->positions(characters_filled_in); |
| 1825 uint16_t c = quarks->At(i); |
| 1826 if (c > char_mask) { |
| 1827 // If we expect a non-Latin1 character from an one-byte string, |
| 1828 // there is no way we can match. Not even case independent |
| 1829 // matching can turn an Latin1 character into non-Latin1 or |
| 1830 // vice versa. |
| 1831 // TODO(dcarney): issue 3550. Verify that this works as expected. |
| 1832 // For example, \u0178 is uppercase of \u00ff (y-umlaut). |
| 1833 details->set_cannot_match(); |
| 1834 pos->determines_perfectly = false; |
| 1835 return; |
| 1836 } |
| 1837 if (compiler->ignore_case()) { |
| 1838 int32_t chars[unibrow::Ecma262UnCanonicalize::kMaxWidth]; |
| 1839 intptr_t length = |
| 1840 GetCaseIndependentLetters(c, compiler->one_byte(), chars); |
| 1841 ASSERT(length != 0); // Can only happen if c > char_mask (see above). |
| 1842 if (length == 1) { |
| 1843 // This letter has no case equivalents, so it's nice and simple |
| 1844 // and the mask-compare will determine definitely whether we have |
| 1845 // a match at this character position. |
| 1846 pos->mask = char_mask; |
| 1847 pos->value = c; |
| 1848 pos->determines_perfectly = true; |
| 1849 } else { |
| 1850 uint32_t common_bits = char_mask; |
| 1851 uint32_t bits = chars[0]; |
| 1852 for (intptr_t j = 1; j < length; j++) { |
| 1853 uint32_t differing_bits = ((chars[j] & common_bits) ^ bits); |
| 1854 common_bits ^= differing_bits; |
| 1855 bits &= common_bits; |
| 1856 } |
| 1857 // If length is 2 and common bits has only one zero in it then |
| 1858 // our mask and compare instruction will determine definitely |
| 1859 // whether we have a match at this character position. Otherwise |
| 1860 // it can only be an approximate check. |
| 1861 uint32_t one_zero = (common_bits | ~char_mask); |
| 1862 if (length == 2 && ((~one_zero) & ((~one_zero) - 1)) == 0) { |
| 1863 pos->determines_perfectly = true; |
| 1864 } |
| 1865 pos->mask = common_bits; |
| 1866 pos->value = bits; |
| 1867 } |
| 1868 } else { |
| 1869 // Don't ignore case. Nice simple case where the mask-compare will |
| 1870 // determine definitely whether we have a match at this character |
| 1871 // position. |
| 1872 pos->mask = char_mask; |
| 1873 pos->value = c; |
| 1874 pos->determines_perfectly = true; |
| 1875 } |
| 1876 characters_filled_in++; |
| 1877 ASSERT(characters_filled_in <= details->characters()); |
| 1878 if (characters_filled_in == details->characters()) { |
| 1879 return; |
| 1880 } |
| 1881 } |
| 1882 } else { |
| 1883 QuickCheckDetails::Position* pos = |
| 1884 details->positions(characters_filled_in); |
| 1885 RegExpCharacterClass* tree = elm.char_class(); |
| 1886 ZoneGrowableArray<CharacterRange>* ranges = tree->ranges(); |
| 1887 if (tree->is_negated()) { |
| 1888 // A quick check uses multi-character mask and compare. There is no |
| 1889 // useful way to incorporate a negative char class into this scheme |
| 1890 // so we just conservatively create a mask and value that will always |
| 1891 // succeed. |
| 1892 pos->mask = 0; |
| 1893 pos->value = 0; |
| 1894 } else { |
| 1895 intptr_t first_range = 0; |
| 1896 while (ranges->At(first_range).from() > char_mask) { |
| 1897 first_range++; |
| 1898 if (first_range == ranges->length()) { |
| 1899 details->set_cannot_match(); |
| 1900 pos->determines_perfectly = false; |
| 1901 return; |
| 1902 } |
| 1903 } |
| 1904 CharacterRange range = ranges->At(first_range); |
| 1905 uint16_t from = range.from(); |
| 1906 uint16_t to = range.to(); |
| 1907 if (to > char_mask) { |
| 1908 to = char_mask; |
| 1909 } |
| 1910 uint32_t differing_bits = (from ^ to); |
| 1911 // A mask and compare is only perfect if the differing bits form a |
| 1912 // number like 00011111 with one single block of trailing 1s. |
| 1913 if ((differing_bits & (differing_bits + 1)) == 0 && |
| 1914 from + differing_bits == to) { |
| 1915 pos->determines_perfectly = true; |
| 1916 } |
| 1917 uint32_t common_bits = ~SmearBitsRight(differing_bits); |
| 1918 uint32_t bits = (from & common_bits); |
| 1919 for (intptr_t i = first_range + 1; i < ranges->length(); i++) { |
| 1920 CharacterRange range = ranges->At(i); |
| 1921 uint16_t from = range.from(); |
| 1922 uint16_t to = range.to(); |
| 1923 if (from > char_mask) continue; |
| 1924 if (to > char_mask) to = char_mask; |
| 1925 // Here we are combining more ranges into the mask and compare |
| 1926 // value. With each new range the mask becomes more sparse and |
| 1927 // so the chances of a false positive rise. A character class |
| 1928 // with multiple ranges is assumed never to be equivalent to a |
| 1929 // mask and compare operation. |
| 1930 pos->determines_perfectly = false; |
| 1931 uint32_t new_common_bits = (from ^ to); |
| 1932 new_common_bits = ~SmearBitsRight(new_common_bits); |
| 1933 common_bits &= new_common_bits; |
| 1934 bits &= new_common_bits; |
| 1935 uint32_t differing_bits = (from & common_bits) ^ bits; |
| 1936 common_bits ^= differing_bits; |
| 1937 bits &= common_bits; |
| 1938 } |
| 1939 pos->mask = common_bits; |
| 1940 pos->value = bits; |
| 1941 } |
| 1942 characters_filled_in++; |
| 1943 ASSERT(characters_filled_in <= details->characters()); |
| 1944 if (characters_filled_in == details->characters()) { |
| 1945 return; |
| 1946 } |
| 1947 } |
| 1948 } |
| 1949 ASSERT(characters_filled_in != details->characters()); |
| 1950 if (!details->cannot_match()) { |
| 1951 on_success()-> GetQuickCheckDetails(details, |
| 1952 compiler, |
| 1953 characters_filled_in, |
| 1954 true); |
| 1955 } |
| 1956 } |
| 1957 |
| 1958 |
| 1959 void QuickCheckDetails::Clear() { |
| 1960 for (int i = 0; i < characters_; i++) { |
| 1961 positions_[i].mask = 0; |
| 1962 positions_[i].value = 0; |
| 1963 positions_[i].determines_perfectly = false; |
| 1964 } |
| 1965 characters_ = 0; |
| 1966 } |
| 1967 |
| 1968 |
| 1969 void QuickCheckDetails::Advance(intptr_t by, bool one_byte) { |
| 1970 ASSERT(by >= 0); |
| 1971 if (by >= characters_) { |
| 1972 Clear(); |
| 1973 return; |
| 1974 } |
| 1975 for (intptr_t i = 0; i < characters_ - by; i++) { |
| 1976 positions_[i] = positions_[by + i]; |
| 1977 } |
| 1978 for (intptr_t i = characters_ - by; i < characters_; i++) { |
| 1979 positions_[i].mask = 0; |
| 1980 positions_[i].value = 0; |
| 1981 positions_[i].determines_perfectly = false; |
| 1982 } |
| 1983 characters_ -= by; |
| 1984 // We could change mask_ and value_ here but we would never advance unless |
| 1985 // they had already been used in a check and they won't be used again because |
| 1986 // it would gain us nothing. So there's no point. |
| 1987 } |
| 1988 |
| 1989 |
| 1990 void QuickCheckDetails::Merge(QuickCheckDetails* other, intptr_t from_index) { |
| 1991 ASSERT(characters_ == other->characters_); |
| 1992 if (other->cannot_match_) { |
| 1993 return; |
| 1994 } |
| 1995 if (cannot_match_) { |
| 1996 *this = *other; |
| 1997 return; |
| 1998 } |
| 1999 for (intptr_t i = from_index; i < characters_; i++) { |
| 2000 QuickCheckDetails::Position* pos = positions(i); |
| 2001 QuickCheckDetails::Position* other_pos = other->positions(i); |
| 2002 if (pos->mask != other_pos->mask || |
| 2003 pos->value != other_pos->value || |
| 2004 !other_pos->determines_perfectly) { |
| 2005 // Our mask-compare operation will be approximate unless we have the |
| 2006 // exact same operation on both sides of the alternation. |
| 2007 pos->determines_perfectly = false; |
| 2008 } |
| 2009 pos->mask &= other_pos->mask; |
| 2010 pos->value &= pos->mask; |
| 2011 other_pos->value &= pos->mask; |
| 2012 uint16_t differing_bits = (pos->value ^ other_pos->value); |
| 2013 pos->mask &= ~differing_bits; |
| 2014 pos->value &= pos->mask; |
| 2015 } |
| 2016 } |
| 2017 |
| 2018 |
| 2019 class VisitMarker : public ValueObject { |
| 2020 public: |
| 2021 explicit VisitMarker(NodeInfo* info) : info_(info) { |
| 2022 ASSERT(!info->visited); |
| 2023 info->visited = true; |
| 2024 } |
| 2025 ~VisitMarker() { |
| 2026 info_->visited = false; |
| 2027 } |
| 2028 private: |
| 2029 NodeInfo* info_; |
| 2030 }; |
| 2031 |
| 2032 |
| 2033 RegExpNode* SeqRegExpNode::FilterOneByte(intptr_t depth, bool ignore_case) { |
| 2034 if (info()->replacement_calculated) return replacement(); |
| 2035 if (depth < 0) return this; |
| 2036 ASSERT(!info()->visited); |
| 2037 VisitMarker marker(info()); |
| 2038 return FilterSuccessor(depth - 1, ignore_case); |
| 2039 } |
| 2040 |
| 2041 |
| 2042 RegExpNode* SeqRegExpNode::FilterSuccessor(intptr_t depth, bool ignore_case) { |
| 2043 RegExpNode* next = on_success_->FilterOneByte(depth - 1, ignore_case); |
| 2044 if (next == NULL) return set_replacement(NULL); |
| 2045 on_success_ = next; |
| 2046 return set_replacement(this); |
| 2047 } |
| 2048 |
| 2049 |
| 2050 // We need to check for the following characters: 0x39c 0x3bc 0x178. |
| 2051 static inline bool RangeContainsLatin1Equivalents(CharacterRange range) { |
| 2052 // TODO(dcarney): this could be a lot more efficient. |
| 2053 return range.Contains(0x39c) || |
| 2054 range.Contains(0x3bc) || range.Contains(0x178); |
| 2055 } |
| 2056 |
| 2057 |
| 2058 static bool RangesContainLatin1Equivalents( |
| 2059 ZoneGrowableArray<CharacterRange>* ranges) { |
| 2060 for (intptr_t i = 0; i < ranges->length(); i++) { |
| 2061 // TODO(dcarney): this could be a lot more efficient. |
| 2062 if (RangeContainsLatin1Equivalents(ranges->At(i))) return true; |
| 2063 } |
| 2064 return false; |
| 2065 } |
| 2066 |
| 2067 |
| 2068 static uint16_t ConvertNonLatin1ToLatin1(uint16_t c) { |
| 2069 ASSERT(c > Symbols::kMaxOneCharCodeSymbol); |
| 2070 switch (c) { |
| 2071 // This are equivalent characters in unicode. |
| 2072 case 0x39c: |
| 2073 case 0x3bc: |
| 2074 return 0xb5; |
| 2075 // This is an uppercase of a Latin-1 character |
| 2076 // outside of Latin-1. |
| 2077 case 0x178: |
| 2078 return 0xff; |
| 2079 } |
| 2080 return 0; |
| 2081 } |
| 2082 |
| 2083 |
| 2084 RegExpNode* TextNode::FilterOneByte(intptr_t depth, bool ignore_case) { |
| 2085 if (info()->replacement_calculated) return replacement(); |
| 2086 if (depth < 0) return this; |
| 2087 ASSERT(!info()->visited); |
| 2088 VisitMarker marker(info()); |
| 2089 intptr_t element_count = elms_->length(); |
| 2090 for (intptr_t i = 0; i < element_count; i++) { |
| 2091 TextElement elm = elms_->At(i); |
| 2092 if (elm.text_type() == TextElement::ATOM) { |
| 2093 ZoneGrowableArray<uint16_t>* quarks = elm.atom()->data(); |
| 2094 for (intptr_t j = 0; j < quarks->length(); j++) { |
| 2095 uint16_t c = quarks->At(j); |
| 2096 if (c <= Symbols::kMaxOneCharCodeSymbol) continue; |
| 2097 if (!ignore_case) return set_replacement(NULL); |
| 2098 // Here, we need to check for characters whose upper and lower cases |
| 2099 // are outside the Latin-1 range. |
| 2100 uint16_t converted = ConvertNonLatin1ToLatin1(c); |
| 2101 // Character is outside Latin-1 completely |
| 2102 if (converted == 0) return set_replacement(NULL); |
| 2103 // Convert quark to Latin-1 in place. |
| 2104 (*quarks)[0] = converted; |
| 2105 } |
| 2106 } else { |
| 2107 ASSERT(elm.text_type() == TextElement::CHAR_CLASS); |
| 2108 RegExpCharacterClass* cc = elm.char_class(); |
| 2109 ZoneGrowableArray<CharacterRange>* ranges = cc->ranges(); |
| 2110 if (!CharacterRange::IsCanonical(ranges)) { |
| 2111 CharacterRange::Canonicalize(ranges); |
| 2112 } |
| 2113 // Now they are in order so we only need to look at the first. |
| 2114 intptr_t range_count = ranges->length(); |
| 2115 if (cc->is_negated()) { |
| 2116 if (range_count != 0 && |
| 2117 ranges->At(0).from() == 0 && |
| 2118 ranges->At(0).to() >= Symbols::kMaxOneCharCodeSymbol) { |
| 2119 // This will be handled in a later filter. |
| 2120 if (ignore_case && RangesContainLatin1Equivalents(ranges)) continue; |
| 2121 return set_replacement(NULL); |
| 2122 } |
| 2123 } else { |
| 2124 if (range_count == 0 || |
| 2125 ranges->At(0).from() > Symbols::kMaxOneCharCodeSymbol) { |
| 2126 // This will be handled in a later filter. |
| 2127 if (ignore_case && RangesContainLatin1Equivalents(ranges)) continue; |
| 2128 return set_replacement(NULL); |
| 2129 } |
| 2130 } |
| 2131 } |
| 2132 } |
| 2133 return FilterSuccessor(depth - 1, ignore_case); |
| 2134 } |
| 2135 |
| 2136 |
| 2137 RegExpNode* LoopChoiceNode::FilterOneByte(intptr_t depth, bool ignore_case) { |
| 2138 if (info()->replacement_calculated) return replacement(); |
| 2139 if (depth < 0) return this; |
| 2140 if (info()->visited) return this; |
| 2141 { |
| 2142 VisitMarker marker(info()); |
| 2143 |
| 2144 RegExpNode* continue_replacement = |
| 2145 continue_node_->FilterOneByte(depth - 1, ignore_case); |
| 2146 // If we can't continue after the loop then there is no sense in doing the |
| 2147 // loop. |
| 2148 if (continue_replacement == NULL) return set_replacement(NULL); |
| 2149 } |
| 2150 |
| 2151 return ChoiceNode::FilterOneByte(depth - 1, ignore_case); |
| 2152 } |
| 2153 |
| 2154 |
| 2155 RegExpNode* ChoiceNode::FilterOneByte(intptr_t depth, bool ignore_case) { |
| 2156 if (info()->replacement_calculated) return replacement(); |
| 2157 if (depth < 0) return this; |
| 2158 if (info()->visited) return this; |
| 2159 VisitMarker marker(info()); |
| 2160 intptr_t choice_count = alternatives_->length(); |
| 2161 |
| 2162 for (intptr_t i = 0; i < choice_count; i++) { |
| 2163 GuardedAlternative alternative = alternatives_->At(i); |
| 2164 if (alternative.guards() != NULL && alternative.guards()->length() != 0) { |
| 2165 set_replacement(this); |
| 2166 return this; |
| 2167 } |
| 2168 } |
| 2169 |
| 2170 intptr_t surviving = 0; |
| 2171 RegExpNode* survivor = NULL; |
| 2172 for (intptr_t i = 0; i < choice_count; i++) { |
| 2173 GuardedAlternative alternative = alternatives_->At(i); |
| 2174 RegExpNode* replacement = |
| 2175 alternative.node()->FilterOneByte(depth - 1, ignore_case); |
| 2176 ASSERT(replacement != this); // No missing EMPTY_MATCH_CHECK. |
| 2177 if (replacement != NULL) { |
| 2178 (*alternatives_)[i].set_node(replacement); |
| 2179 surviving++; |
| 2180 survivor = replacement; |
| 2181 } |
| 2182 } |
| 2183 if (surviving < 2) return set_replacement(survivor); |
| 2184 |
| 2185 set_replacement(this); |
| 2186 if (surviving == choice_count) { |
| 2187 return this; |
| 2188 } |
| 2189 // Only some of the nodes survived the filtering. We need to rebuild the |
| 2190 // alternatives list. |
| 2191 ZoneGrowableArray<GuardedAlternative>* new_alternatives = |
| 2192 new(I) ZoneGrowableArray<GuardedAlternative>(surviving); |
| 2193 for (intptr_t i = 0; i < choice_count; i++) { |
| 2194 RegExpNode* replacement = |
| 2195 (*alternatives_)[i].node()->FilterOneByte(depth - 1, ignore_case); |
| 2196 if (replacement != NULL) { |
| 2197 (*alternatives_)[i].set_node(replacement); |
| 2198 new_alternatives->Add((*alternatives_)[i]); |
| 2199 } |
| 2200 } |
| 2201 alternatives_ = new_alternatives; |
| 2202 return this; |
| 2203 } |
| 2204 |
| 2205 |
| 2206 RegExpNode* NegativeLookaheadChoiceNode::FilterOneByte(intptr_t depth, |
| 2207 bool ignore_case) { |
| 2208 if (info()->replacement_calculated) return replacement(); |
| 2209 if (depth < 0) return this; |
| 2210 if (info()->visited) return this; |
| 2211 VisitMarker marker(info()); |
| 2212 // Alternative 0 is the negative lookahead, alternative 1 is what comes |
| 2213 // afterwards. |
| 2214 RegExpNode* node = (*alternatives_)[1].node(); |
| 2215 RegExpNode* replacement = node->FilterOneByte(depth - 1, ignore_case); |
| 2216 if (replacement == NULL) return set_replacement(NULL); |
| 2217 (*alternatives_)[1].set_node(replacement); |
| 2218 |
| 2219 RegExpNode* neg_node = (*alternatives_)[0].node(); |
| 2220 RegExpNode* neg_replacement = neg_node->FilterOneByte(depth - 1, ignore_case); |
| 2221 // If the negative lookahead is always going to fail then |
| 2222 // we don't need to check it. |
| 2223 if (neg_replacement == NULL) return set_replacement(replacement); |
| 2224 (*alternatives_)[0].set_node(neg_replacement); |
| 2225 return set_replacement(this); |
| 2226 } |
| 2227 |
| 2228 |
| 2229 void LoopChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details, |
| 2230 RegExpCompiler* compiler, |
| 2231 intptr_t characters_filled_in, |
| 2232 bool not_at_start) { |
| 2233 if (body_can_be_zero_length_ || info()->visited) return; |
| 2234 VisitMarker marker(info()); |
| 2235 return ChoiceNode::GetQuickCheckDetails(details, |
| 2236 compiler, |
| 2237 characters_filled_in, |
| 2238 not_at_start); |
| 2239 } |
| 2240 |
| 2241 |
| 2242 void LoopChoiceNode::FillInBMInfo(intptr_t offset, |
| 2243 intptr_t budget, |
| 2244 BoyerMooreLookahead* bm, |
| 2245 bool not_at_start) { |
| 2246 if (body_can_be_zero_length_ || budget <= 0) { |
| 2247 bm->SetRest(offset); |
| 2248 SaveBMInfo(bm, not_at_start, offset); |
| 2249 return; |
| 2250 } |
| 2251 ChoiceNode::FillInBMInfo(offset, budget - 1, bm, not_at_start); |
| 2252 SaveBMInfo(bm, not_at_start, offset); |
| 2253 } |
| 2254 |
| 2255 |
| 2256 void ChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details, |
| 2257 RegExpCompiler* compiler, |
| 2258 intptr_t characters_filled_in, |
| 2259 bool not_at_start) { |
| 2260 not_at_start = (not_at_start || not_at_start_); |
| 2261 intptr_t choice_count = alternatives_->length(); |
| 2262 ASSERT(choice_count > 0); |
| 2263 (*alternatives_)[0].node()->GetQuickCheckDetails(details, |
| 2264 compiler, |
| 2265 characters_filled_in, |
| 2266 not_at_start); |
| 2267 for (intptr_t i = 1; i < choice_count; i++) { |
| 2268 QuickCheckDetails new_details(details->characters()); |
| 2269 RegExpNode* node = (*alternatives_)[i].node(); |
| 2270 node->GetQuickCheckDetails(&new_details, compiler, |
| 2271 characters_filled_in, |
| 2272 not_at_start); |
| 2273 // Here we merge the quick match details of the two branches. |
| 2274 details->Merge(&new_details, characters_filled_in); |
| 2275 } |
| 2276 } |
| 2277 |
| 2278 |
| 2279 // Check for [0-9A-Z_a-z]. |
| 2280 static void EmitWordCheck(RegExpMacroAssembler* assembler, |
| 2281 BlockLabel* word, |
| 2282 BlockLabel* non_word, |
| 2283 bool fall_through_on_word) { |
| 2284 if (assembler->CheckSpecialCharacterClass( |
| 2285 fall_through_on_word ? 'w' : 'W', |
| 2286 fall_through_on_word ? non_word : word)) { |
| 2287 // Optimized implementation available. |
| 2288 return; |
| 2289 } |
| 2290 assembler->CheckCharacterGT('z', non_word); |
| 2291 assembler->CheckCharacterLT('0', non_word); |
| 2292 assembler->CheckCharacterGT('a' - 1, word); |
| 2293 assembler->CheckCharacterLT('9' + 1, word); |
| 2294 assembler->CheckCharacterLT('A', non_word); |
| 2295 assembler->CheckCharacterLT('Z' + 1, word); |
| 2296 if (fall_through_on_word) { |
| 2297 assembler->CheckNotCharacter('_', non_word); |
| 2298 } else { |
| 2299 assembler->CheckCharacter('_', word); |
| 2300 } |
| 2301 } |
| 2302 |
| 2303 |
| 2304 // Emit the code to check for a ^ in multiline mode (1-character lookbehind |
| 2305 // that matches newline or the start of input). |
| 2306 static void EmitHat(RegExpCompiler* compiler, |
| 2307 RegExpNode* on_success, |
| 2308 Trace* trace) { |
| 2309 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 2310 // We will be loading the previous character into the current character |
| 2311 // register. |
| 2312 Trace new_trace(*trace); |
| 2313 new_trace.InvalidateCurrentCharacter(); |
| 2314 |
| 2315 BlockLabel ok; |
| 2316 if (new_trace.cp_offset() == 0) { |
| 2317 // The start of input counts as a newline in this context, so skip to |
| 2318 // ok if we are at the start. |
| 2319 assembler->CheckAtStart(&ok); |
| 2320 } |
| 2321 // We already checked that we are not at the start of input so it must be |
| 2322 // OK to load the previous character. |
| 2323 assembler->LoadCurrentCharacter(new_trace.cp_offset() -1, |
| 2324 new_trace.backtrack(), |
| 2325 false); |
| 2326 if (!assembler->CheckSpecialCharacterClass('n', |
| 2327 new_trace.backtrack())) { |
| 2328 // Newline means \n, \r, 0x2028 or 0x2029. |
| 2329 if (!compiler->one_byte()) { |
| 2330 assembler->CheckCharacterAfterAnd(0x2028, 0xfffe, &ok); |
| 2331 } |
| 2332 assembler->CheckCharacter('\n', &ok); |
| 2333 assembler->CheckNotCharacter('\r', new_trace.backtrack()); |
| 2334 } |
| 2335 assembler->BindBlock(&ok); |
| 2336 on_success->Emit(compiler, &new_trace); |
| 2337 } |
| 2338 |
| 2339 |
| 2340 // Emit the code to handle \b and \B (word-boundary or non-word-boundary). |
| 2341 void AssertionNode::EmitBoundaryCheck(RegExpCompiler* compiler, Trace* trace) { |
| 2342 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 2343 Trace::TriBool next_is_word_character = Trace::UNKNOWN; |
| 2344 bool not_at_start = (trace->at_start() == Trace::FALSE_VALUE); |
| 2345 BoyerMooreLookahead* lookahead = bm_info(not_at_start); |
| 2346 if (lookahead == NULL) { |
| 2347 intptr_t eats_at_least = |
| 2348 Utils::Minimum(kMaxLookaheadForBoyerMoore, |
| 2349 EatsAtLeast(kMaxLookaheadForBoyerMoore, |
| 2350 kRecursionBudget, |
| 2351 not_at_start)); |
| 2352 if (eats_at_least >= 1) { |
| 2353 BoyerMooreLookahead* bm = |
| 2354 new(I) BoyerMooreLookahead(eats_at_least, compiler, I); |
| 2355 FillInBMInfo(0, kRecursionBudget, bm, not_at_start); |
| 2356 if (bm->at(0)->is_non_word()) |
| 2357 next_is_word_character = Trace::FALSE_VALUE; |
| 2358 if (bm->at(0)->is_word()) next_is_word_character = Trace::TRUE_VALUE; |
| 2359 } |
| 2360 } else { |
| 2361 if (lookahead->at(0)->is_non_word()) |
| 2362 next_is_word_character = Trace::FALSE_VALUE; |
| 2363 if (lookahead->at(0)->is_word()) |
| 2364 next_is_word_character = Trace::TRUE_VALUE; |
| 2365 } |
| 2366 bool at_boundary = (assertion_type_ == AssertionNode::AT_BOUNDARY); |
| 2367 if (next_is_word_character == Trace::UNKNOWN) { |
| 2368 BlockLabel before_non_word; |
| 2369 BlockLabel before_word; |
| 2370 if (trace->characters_preloaded() != 1) { |
| 2371 assembler->LoadCurrentCharacter(trace->cp_offset(), &before_non_word); |
| 2372 } |
| 2373 // Fall through on non-word. |
| 2374 EmitWordCheck(assembler, &before_word, &before_non_word, false); |
| 2375 // Next character is not a word character. |
| 2376 assembler->BindBlock(&before_non_word); |
| 2377 BlockLabel ok; |
| 2378 // Backtrack on \B (non-boundary check) if previous is a word, |
| 2379 // since we know next *is not* a word and this would be a boundary. |
| 2380 BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord); |
| 2381 |
| 2382 if (!assembler->IsClosed()) { |
| 2383 assembler->GoTo(&ok); |
| 2384 } |
| 2385 |
| 2386 assembler->BindBlock(&before_word); |
| 2387 BacktrackIfPrevious(compiler, trace, at_boundary ? kIsWord : kIsNonWord); |
| 2388 assembler->BindBlock(&ok); |
| 2389 } else if (next_is_word_character == Trace::TRUE_VALUE) { |
| 2390 BacktrackIfPrevious(compiler, trace, at_boundary ? kIsWord : kIsNonWord); |
| 2391 } else { |
| 2392 ASSERT(next_is_word_character == Trace::FALSE_VALUE); |
| 2393 BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord); |
| 2394 } |
| 2395 } |
| 2396 |
| 2397 |
| 2398 void AssertionNode::BacktrackIfPrevious( |
| 2399 RegExpCompiler* compiler, |
| 2400 Trace* trace, |
| 2401 AssertionNode::IfPrevious backtrack_if_previous) { |
| 2402 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 2403 Trace new_trace(*trace); |
| 2404 new_trace.InvalidateCurrentCharacter(); |
| 2405 |
| 2406 BlockLabel fall_through, dummy; |
| 2407 |
| 2408 BlockLabel* non_word = backtrack_if_previous == kIsNonWord ? |
| 2409 new_trace.backtrack() : |
| 2410 &fall_through; |
| 2411 BlockLabel* word = backtrack_if_previous == kIsNonWord ? |
| 2412 &fall_through : |
| 2413 new_trace.backtrack(); |
| 2414 |
| 2415 if (new_trace.cp_offset() == 0) { |
| 2416 // The start of input counts as a non-word character, so the question is |
| 2417 // decided if we are at the start. |
| 2418 assembler->CheckAtStart(non_word); |
| 2419 } |
| 2420 // We already checked that we are not at the start of input so it must be |
| 2421 // OK to load the previous character. |
| 2422 assembler->LoadCurrentCharacter(new_trace.cp_offset() - 1, &dummy, false); |
| 2423 EmitWordCheck(assembler, word, non_word, backtrack_if_previous == kIsNonWord); |
| 2424 |
| 2425 assembler->BindBlock(&fall_through); |
| 2426 on_success()->Emit(compiler, &new_trace); |
| 2427 } |
| 2428 |
| 2429 |
| 2430 void AssertionNode::GetQuickCheckDetails(QuickCheckDetails* details, |
| 2431 RegExpCompiler* compiler, |
| 2432 intptr_t filled_in, |
| 2433 bool not_at_start) { |
| 2434 if (assertion_type_ == AT_START && not_at_start) { |
| 2435 details->set_cannot_match(); |
| 2436 return; |
| 2437 } |
| 2438 return on_success()->GetQuickCheckDetails(details, |
| 2439 compiler, |
| 2440 filled_in, |
| 2441 not_at_start); |
| 2442 } |
| 2443 |
| 2444 |
| 2445 void AssertionNode::Emit(RegExpCompiler* compiler, Trace* trace) { |
| 2446 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 2447 switch (assertion_type_) { |
| 2448 case AT_END: { |
| 2449 BlockLabel ok; |
| 2450 assembler->CheckPosition(trace->cp_offset(), &ok); |
| 2451 assembler->GoTo(trace->backtrack()); |
| 2452 assembler->BindBlock(&ok); |
| 2453 break; |
| 2454 } |
| 2455 case AT_START: { |
| 2456 if (trace->at_start() == Trace::FALSE_VALUE) { |
| 2457 assembler->GoTo(trace->backtrack()); |
| 2458 return; |
| 2459 } |
| 2460 if (trace->at_start() == Trace::UNKNOWN) { |
| 2461 assembler->CheckNotAtStart(trace->backtrack()); |
| 2462 Trace at_start_trace = *trace; |
| 2463 at_start_trace.set_at_start(true); |
| 2464 on_success()->Emit(compiler, &at_start_trace); |
| 2465 return; |
| 2466 } |
| 2467 } |
| 2468 break; |
| 2469 case AFTER_NEWLINE: |
| 2470 EmitHat(compiler, on_success(), trace); |
| 2471 return; |
| 2472 case AT_BOUNDARY: |
| 2473 case AT_NON_BOUNDARY: { |
| 2474 EmitBoundaryCheck(compiler, trace); |
| 2475 return; |
| 2476 } |
| 2477 } |
| 2478 on_success()->Emit(compiler, trace); |
| 2479 } |
| 2480 |
| 2481 |
| 2482 static bool DeterminedAlready(QuickCheckDetails* quick_check, intptr_t offset) { |
| 2483 if (quick_check == NULL) return false; |
| 2484 if (offset >= quick_check->characters()) return false; |
| 2485 return quick_check->positions(offset)->determines_perfectly; |
| 2486 } |
| 2487 |
| 2488 |
| 2489 static void UpdateBoundsCheck(intptr_t index, intptr_t* checked_up_to) { |
| 2490 if (index > *checked_up_to) { |
| 2491 *checked_up_to = index; |
| 2492 } |
| 2493 } |
| 2494 |
| 2495 |
| 2496 // We call this repeatedly to generate code for each pass over the text node. |
| 2497 // The passes are in increasing order of difficulty because we hope one |
| 2498 // of the first passes will fail in which case we are saved the work of the |
| 2499 // later passes. for example for the case independent regexp /%[asdfghjkl]a/ |
| 2500 // we will check the '%' in the first pass, the case independent 'a' in the |
| 2501 // second pass and the character class in the last pass. |
| 2502 // |
| 2503 // The passes are done from right to left, so for example to test for /bar/ |
| 2504 // we will first test for an 'r' with offset 2, then an 'a' with offset 1 |
| 2505 // and then a 'b' with offset 0. This means we can avoid the end-of-input |
| 2506 // bounds check most of the time. In the example we only need to check for |
| 2507 // end-of-input when loading the putative 'r'. |
| 2508 // |
| 2509 // A slight complication involves the fact that the first character may already |
| 2510 // be fetched into a register by the previous node. In this case we want to |
| 2511 // do the test for that character first. We do this in separate passes. The |
| 2512 // 'preloaded' argument indicates that we are doing such a 'pass'. If such a |
| 2513 // pass has been performed then subsequent passes will have true in |
| 2514 // first_element_checked to indicate that that character does not need to be |
| 2515 // checked again. |
| 2516 // |
| 2517 // In addition to all this we are passed a Trace, which can |
| 2518 // contain an AlternativeGeneration object. In this AlternativeGeneration |
| 2519 // object we can see details of any quick check that was already passed in |
| 2520 // order to get to the code we are now generating. The quick check can involve |
| 2521 // loading characters, which means we do not need to recheck the bounds |
| 2522 // up to the limit the quick check already checked. In addition the quick |
| 2523 // check can have involved a mask and compare operation which may simplify |
| 2524 // or obviate the need for further checks at some character positions. |
| 2525 void TextNode::TextEmitPass(RegExpCompiler* compiler, |
| 2526 TextEmitPassType pass, |
| 2527 bool preloaded, |
| 2528 Trace* trace, |
| 2529 bool first_element_checked, |
| 2530 intptr_t* checked_up_to) { |
| 2531 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 2532 bool one_byte = compiler->one_byte(); |
| 2533 BlockLabel* backtrack = trace->backtrack(); |
| 2534 QuickCheckDetails* quick_check = trace->quick_check_performed(); |
| 2535 intptr_t element_count = elms_->length(); |
| 2536 for (intptr_t i = preloaded ? 0 : element_count - 1; i >= 0; i--) { |
| 2537 TextElement elm = elms_->At(i); |
| 2538 intptr_t cp_offset = trace->cp_offset() + elm.cp_offset(); |
| 2539 if (elm.text_type() == TextElement::ATOM) { |
| 2540 ZoneGrowableArray<uint16_t>* quarks = elm.atom()->data(); |
| 2541 for (intptr_t j = preloaded ? 0 : quarks->length() - 1; j >= 0; j--) { |
| 2542 if (first_element_checked && i == 0 && j == 0) continue; |
| 2543 if (DeterminedAlready(quick_check, elm.cp_offset() + j)) continue; |
| 2544 EmitCharacterFunction* emit_function = NULL; |
| 2545 switch (pass) { |
| 2546 case NON_LATIN1_MATCH: |
| 2547 ASSERT(one_byte); |
| 2548 if (quarks->At(j) > Symbols::kMaxOneCharCodeSymbol) { |
| 2549 assembler->GoTo(backtrack); |
| 2550 return; |
| 2551 } |
| 2552 break; |
| 2553 case NON_LETTER_CHARACTER_MATCH: |
| 2554 emit_function = &EmitAtomNonLetter; |
| 2555 break; |
| 2556 case SIMPLE_CHARACTER_MATCH: |
| 2557 emit_function = &EmitSimpleCharacter; |
| 2558 break; |
| 2559 case CASE_CHARACTER_MATCH: |
| 2560 emit_function = &EmitAtomLetter; |
| 2561 break; |
| 2562 default: |
| 2563 break; |
| 2564 } |
| 2565 if (emit_function != NULL) { |
| 2566 bool bound_checked = emit_function(I, |
| 2567 compiler, |
| 2568 quarks->At(j), |
| 2569 backtrack, |
| 2570 cp_offset + j, |
| 2571 *checked_up_to < cp_offset + j, |
| 2572 preloaded); |
| 2573 if (bound_checked) UpdateBoundsCheck(cp_offset + j, checked_up_to); |
| 2574 } |
| 2575 } |
| 2576 } else { |
| 2577 ASSERT(elm.text_type() == TextElement::CHAR_CLASS); |
| 2578 if (pass == CHARACTER_CLASS_MATCH) { |
| 2579 if (first_element_checked && i == 0) continue; |
| 2580 if (DeterminedAlready(quick_check, elm.cp_offset())) continue; |
| 2581 RegExpCharacterClass* cc = elm.char_class(); |
| 2582 EmitCharClass(assembler, |
| 2583 cc, |
| 2584 one_byte, |
| 2585 backtrack, |
| 2586 cp_offset, |
| 2587 *checked_up_to < cp_offset, |
| 2588 preloaded, |
| 2589 I); |
| 2590 UpdateBoundsCheck(cp_offset, checked_up_to); |
| 2591 } |
| 2592 } |
| 2593 } |
| 2594 } |
| 2595 |
| 2596 |
| 2597 intptr_t TextNode::Length() { |
| 2598 TextElement elm = elms_->Last(); |
| 2599 ASSERT(elm.cp_offset() >= 0); |
| 2600 return elm.cp_offset() + elm.length(); |
| 2601 } |
| 2602 |
| 2603 |
| 2604 bool TextNode::SkipPass(intptr_t intptr_t_pass, bool ignore_case) { |
| 2605 TextEmitPassType pass = static_cast<TextEmitPassType>(intptr_t_pass); |
| 2606 if (ignore_case) { |
| 2607 return pass == SIMPLE_CHARACTER_MATCH; |
| 2608 } else { |
| 2609 return pass == NON_LETTER_CHARACTER_MATCH || pass == CASE_CHARACTER_MATCH; |
| 2610 } |
| 2611 } |
| 2612 |
| 2613 |
| 2614 // This generates the code to match a text node. A text node can contain |
| 2615 // straight character sequences (possibly to be matched in a case-independent |
| 2616 // way) and character classes. For efficiency we do not do this in a single |
| 2617 // pass from left to right. Instead we pass over the text node several times, |
| 2618 // emitting code for some character positions every time. See the comment on |
| 2619 // TextEmitPass for details. |
| 2620 void TextNode::Emit(RegExpCompiler* compiler, Trace* trace) { |
| 2621 LimitResult limit_result = LimitVersions(compiler, trace); |
| 2622 if (limit_result == DONE) return; |
| 2623 ASSERT(limit_result == CONTINUE); |
| 2624 |
| 2625 if (trace->cp_offset() + Length() > RegExpMacroAssembler::kMaxCPOffset) { |
| 2626 compiler->SetRegExpTooBig(); |
| 2627 return; |
| 2628 } |
| 2629 |
| 2630 if (compiler->one_byte()) { |
| 2631 intptr_t dummy = 0; |
| 2632 TextEmitPass(compiler, NON_LATIN1_MATCH, false, trace, false, &dummy); |
| 2633 } |
| 2634 |
| 2635 bool first_elt_done = false; |
| 2636 intptr_t bound_checked_to = trace->cp_offset() - 1; |
| 2637 bound_checked_to += trace->bound_checked_up_to(); |
| 2638 |
| 2639 // If a character is preloaded into the current character register then |
| 2640 // check that now. |
| 2641 if (trace->characters_preloaded() == 1) { |
| 2642 for (intptr_t pass = kFirstRealPass; pass <= kLastPass; pass++) { |
| 2643 if (!SkipPass(pass, compiler->ignore_case())) { |
| 2644 TextEmitPass(compiler, |
| 2645 static_cast<TextEmitPassType>(pass), |
| 2646 true, |
| 2647 trace, |
| 2648 false, |
| 2649 &bound_checked_to); |
| 2650 } |
| 2651 } |
| 2652 first_elt_done = true; |
| 2653 } |
| 2654 |
| 2655 for (intptr_t pass = kFirstRealPass; pass <= kLastPass; pass++) { |
| 2656 if (!SkipPass(pass, compiler->ignore_case())) { |
| 2657 TextEmitPass(compiler, |
| 2658 static_cast<TextEmitPassType>(pass), |
| 2659 false, |
| 2660 trace, |
| 2661 first_elt_done, |
| 2662 &bound_checked_to); |
| 2663 } |
| 2664 } |
| 2665 |
| 2666 Trace successor_trace(*trace); |
| 2667 successor_trace.set_at_start(false); |
| 2668 successor_trace.AdvanceCurrentPositionInTrace(Length(), compiler); |
| 2669 RecursionCheck rc(compiler); |
| 2670 on_success()->Emit(compiler, &successor_trace); |
| 2671 } |
| 2672 |
| 2673 |
| 2674 void Trace::InvalidateCurrentCharacter() { |
| 2675 characters_preloaded_ = 0; |
| 2676 } |
| 2677 |
| 2678 |
| 2679 void Trace::AdvanceCurrentPositionInTrace(intptr_t by, |
| 2680 RegExpCompiler* compiler) { |
| 2681 ASSERT(by > 0); |
| 2682 // We don't have an instruction for shifting the current character register |
| 2683 // down or for using a shifted value for anything so lets just forget that |
| 2684 // we preloaded any characters into it. |
| 2685 characters_preloaded_ = 0; |
| 2686 // Adjust the offsets of the quick check performed information. This |
| 2687 // information is used to find out what we already determined about the |
| 2688 // characters by means of mask and compare. |
| 2689 quick_check_performed_.Advance(by, compiler->one_byte()); |
| 2690 cp_offset_ += by; |
| 2691 if (cp_offset_ > RegExpMacroAssembler::kMaxCPOffset) { |
| 2692 compiler->SetRegExpTooBig(); |
| 2693 cp_offset_ = 0; |
| 2694 } |
| 2695 bound_checked_up_to_ = Utils::Maximum(static_cast<intptr_t>(0), |
| 2696 bound_checked_up_to_ - by); |
| 2697 } |
| 2698 |
| 2699 |
| 2700 void TextNode::MakeCaseIndependent(bool is_one_byte) { |
| 2701 intptr_t element_count = elms_->length(); |
| 2702 for (intptr_t i = 0; i < element_count; i++) { |
| 2703 TextElement elm = elms_->At(i); |
| 2704 if (elm.text_type() == TextElement::CHAR_CLASS) { |
| 2705 RegExpCharacterClass* cc = elm.char_class(); |
| 2706 // None of the standard character classes is different in the case |
| 2707 // independent case and it slows us down if we don't know that. |
| 2708 if (cc->is_standard()) continue; |
| 2709 ZoneGrowableArray<CharacterRange>* ranges = cc->ranges(); |
| 2710 intptr_t range_count = ranges->length(); |
| 2711 for (intptr_t j = 0; j < range_count; j++) { |
| 2712 (*ranges)[j].AddCaseEquivalents(ranges, is_one_byte, I); |
| 2713 } |
| 2714 } |
| 2715 } |
| 2716 } |
| 2717 |
| 2718 |
| 2719 intptr_t TextNode::GreedyLoopTextLength() { |
| 2720 TextElement elm = elms_->At(elms_->length() - 1); |
| 2721 return elm.cp_offset() + elm.length(); |
| 2722 } |
| 2723 |
| 2724 |
| 2725 RegExpNode* TextNode::GetSuccessorOfOmnivorousTextNode( |
| 2726 RegExpCompiler* compiler) { |
| 2727 if (elms_->length() != 1) return NULL; |
| 2728 TextElement elm = elms_->At(0); |
| 2729 if (elm.text_type() != TextElement::CHAR_CLASS) return NULL; |
| 2730 RegExpCharacterClass* node = elm.char_class(); |
| 2731 ZoneGrowableArray<CharacterRange>* ranges = node->ranges(); |
| 2732 if (!CharacterRange::IsCanonical(ranges)) { |
| 2733 CharacterRange::Canonicalize(ranges); |
| 2734 } |
| 2735 if (node->is_negated()) { |
| 2736 return ranges->length() == 0 ? on_success() : NULL; |
| 2737 } |
| 2738 if (ranges->length() != 1) return NULL; |
| 2739 uint32_t max_char; |
| 2740 if (compiler->one_byte()) { |
| 2741 max_char = Symbols::kMaxOneCharCodeSymbol; |
| 2742 } else { |
| 2743 max_char = Utf16::kMaxCodeUnit; |
| 2744 } |
| 2745 return ranges->At(0).IsEverything(max_char) ? on_success() : NULL; |
| 2746 } |
| 2747 |
| 2748 |
| 2749 // Finds the fixed match length of a sequence of nodes that goes from |
| 2750 // this alternative and back to this choice node. If there are variable |
| 2751 // length nodes or other complications in the way then return a sentinel |
| 2752 // value indicating that a greedy loop cannot be constructed. |
| 2753 intptr_t ChoiceNode::GreedyLoopTextLengthForAlternative( |
| 2754 GuardedAlternative* alternative) { |
| 2755 intptr_t length = 0; |
| 2756 RegExpNode* node = alternative->node(); |
| 2757 // Later we will generate code for all these text nodes using recursion |
| 2758 // so we have to limit the max number. |
| 2759 intptr_t recursion_depth = 0; |
| 2760 while (node != this) { |
| 2761 if (recursion_depth++ > RegExpCompiler::kMaxRecursion) { |
| 2762 return kNodeIsTooComplexForGreedyLoops; |
| 2763 } |
| 2764 intptr_t node_length = node->GreedyLoopTextLength(); |
| 2765 if (node_length == kNodeIsTooComplexForGreedyLoops) { |
| 2766 return kNodeIsTooComplexForGreedyLoops; |
| 2767 } |
| 2768 length += node_length; |
| 2769 SeqRegExpNode* seq_node = static_cast<SeqRegExpNode*>(node); |
| 2770 node = seq_node->on_success(); |
| 2771 } |
| 2772 return length; |
| 2773 } |
| 2774 |
| 2775 |
| 2776 void LoopChoiceNode::AddLoopAlternative(GuardedAlternative alt) { |
| 2777 ASSERT(loop_node_ == NULL); |
| 2778 AddAlternative(alt); |
| 2779 loop_node_ = alt.node(); |
| 2780 } |
| 2781 |
| 2782 |
| 2783 void LoopChoiceNode::AddContinueAlternative(GuardedAlternative alt) { |
| 2784 ASSERT(continue_node_ == NULL); |
| 2785 AddAlternative(alt); |
| 2786 continue_node_ = alt.node(); |
| 2787 } |
| 2788 |
| 2789 |
| 2790 void LoopChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) { |
| 2791 RegExpMacroAssembler* macro_assembler = compiler->macro_assembler(); |
| 2792 if (trace->stop_node() == this) { |
| 2793 // Back edge of greedy optimized loop node graph. |
| 2794 intptr_t text_length = |
| 2795 GreedyLoopTextLengthForAlternative(&((*alternatives_)[0])); |
| 2796 ASSERT(text_length != kNodeIsTooComplexForGreedyLoops); |
| 2797 // Update the counter-based backtracking info on the stack. This is an |
| 2798 // optimization for greedy loops (see below). |
| 2799 ASSERT(trace->cp_offset() == text_length); |
| 2800 macro_assembler->AdvanceCurrentPosition(text_length); |
| 2801 macro_assembler->GoTo(trace->loop_label()); |
| 2802 return; |
| 2803 } |
| 2804 ASSERT(trace->stop_node() == NULL); |
| 2805 if (!trace->is_trivial()) { |
| 2806 trace->Flush(compiler, this); |
| 2807 return; |
| 2808 } |
| 2809 ChoiceNode::Emit(compiler, trace); |
| 2810 } |
| 2811 |
| 2812 |
| 2813 intptr_t ChoiceNode::CalculatePreloadCharacters(RegExpCompiler* compiler, |
| 2814 intptr_t eats_at_least) { |
| 2815 intptr_t preload_characters = Utils::Minimum(static_cast<intptr_t>(4), |
| 2816 eats_at_least); |
| 2817 if (compiler->macro_assembler()->CanReadUnaligned()) { |
| 2818 bool one_byte = compiler->one_byte(); |
| 2819 if (one_byte) { |
| 2820 if (preload_characters > 4) preload_characters = 4; |
| 2821 // We can't preload 3 characters because there is no machine instruction |
| 2822 // to do that. We can't just load 4 because we could be reading |
| 2823 // beyond the end of the string, which could cause a memory fault. |
| 2824 if (preload_characters == 3) preload_characters = 2; |
| 2825 } else { |
| 2826 if (preload_characters > 2) preload_characters = 2; |
| 2827 } |
| 2828 } else { |
| 2829 if (preload_characters > 1) preload_characters = 1; |
| 2830 } |
| 2831 return preload_characters; |
| 2832 } |
| 2833 |
| 2834 |
| 2835 // This structure is used when generating the alternatives in a choice node. It |
| 2836 // records the way the alternative is being code generated. |
| 2837 struct AlternativeGeneration { |
| 2838 AlternativeGeneration() |
| 2839 : possible_success(), |
| 2840 expects_preload(false), |
| 2841 after(), |
| 2842 quick_check_details() { } |
| 2843 BlockLabel possible_success; |
| 2844 bool expects_preload; |
| 2845 BlockLabel after; |
| 2846 QuickCheckDetails quick_check_details; |
| 2847 }; |
| 2848 |
| 2849 |
| 2850 // Creates a list of AlternativeGenerations. If the list has a reasonable |
| 2851 // size then it is on the stack, otherwise the excess is on the heap. |
| 2852 class AlternativeGenerationList { |
| 2853 public: |
| 2854 explicit AlternativeGenerationList(intptr_t count) |
| 2855 : alt_gens_(count) { |
| 2856 for (intptr_t i = 0; i < count && i < kAFew; i++) { |
| 2857 alt_gens_.Add(a_few_alt_gens_ + i); |
| 2858 } |
| 2859 for (intptr_t i = kAFew; i < count; i++) { |
| 2860 alt_gens_.Add(new AlternativeGeneration()); |
| 2861 } |
| 2862 } |
| 2863 ~AlternativeGenerationList() { |
| 2864 for (intptr_t i = kAFew; i < alt_gens_.length(); i++) { |
| 2865 delete alt_gens_[i]; |
| 2866 alt_gens_[i] = NULL; |
| 2867 } |
| 2868 } |
| 2869 |
| 2870 AlternativeGeneration* at(intptr_t i) { |
| 2871 return alt_gens_[i]; |
| 2872 } |
| 2873 |
| 2874 private: |
| 2875 static const intptr_t kAFew = 10; |
| 2876 GrowableArray<AlternativeGeneration*> alt_gens_; |
| 2877 AlternativeGeneration a_few_alt_gens_[kAFew]; |
| 2878 |
| 2879 DISALLOW_ALLOCATION(); |
| 2880 }; |
| 2881 |
| 2882 |
| 2883 // The '2' variant is inclusive from and exclusive to. |
| 2884 // This covers \s as defined in ECMA-262 5.1, 15.10.2.12, |
| 2885 // which include WhiteSpace (7.2) or LineTerminator (7.3) values. |
| 2886 static const intptr_t kSpaceRanges[] = { '\t', '\r' + 1, ' ', ' ' + 1, |
| 2887 0x00A0, 0x00A1, 0x1680, 0x1681, 0x180E, 0x180F, 0x2000, 0x200B, |
| 2888 0x2028, 0x202A, 0x202F, 0x2030, 0x205F, 0x2060, 0x3000, 0x3001, |
| 2889 0xFEFF, 0xFF00, 0x10000 }; |
| 2890 static const intptr_t kSpaceRangeCount = ARRAY_SIZE(kSpaceRanges); |
| 2891 static const intptr_t kWordRanges[] = { |
| 2892 '0', '9' + 1, 'A', 'Z' + 1, '_', '_' + 1, 'a', 'z' + 1, 0x10000 }; |
| 2893 static const intptr_t kWordRangeCount = ARRAY_SIZE(kWordRanges); |
| 2894 static const intptr_t kDigitRanges[] = { '0', '9' + 1, 0x10000 }; |
| 2895 static const intptr_t kDigitRangeCount = ARRAY_SIZE(kDigitRanges); |
| 2896 static const intptr_t kSurrogateRanges[] = { 0xd800, 0xe000, 0x10000 }; |
| 2897 static const intptr_t kSurrogateRangeCount = ARRAY_SIZE(kSurrogateRanges); |
| 2898 static const intptr_t kLineTerminatorRanges[] = { |
| 2899 0x000A, 0x000B, 0x000D, 0x000E, 0x2028, 0x202A, 0x10000 }; |
| 2900 static const intptr_t kLineTerminatorRangeCount = |
| 2901 ARRAY_SIZE(kLineTerminatorRanges); |
| 2902 |
| 2903 |
| 2904 void BoyerMoorePositionInfo::Set(intptr_t character) { |
| 2905 SetInterval(Interval(character, character)); |
| 2906 } |
| 2907 |
| 2908 |
| 2909 void BoyerMoorePositionInfo::SetInterval(const Interval& interval) { |
| 2910 s_ = AddRange(s_, kSpaceRanges, kSpaceRangeCount, interval); |
| 2911 w_ = AddRange(w_, kWordRanges, kWordRangeCount, interval); |
| 2912 d_ = AddRange(d_, kDigitRanges, kDigitRangeCount, interval); |
| 2913 surrogate_ = |
| 2914 AddRange(surrogate_, kSurrogateRanges, kSurrogateRangeCount, interval); |
| 2915 if (interval.to() - interval.from() >= kMapSize - 1) { |
| 2916 if (map_count_ != kMapSize) { |
| 2917 map_count_ = kMapSize; |
| 2918 for (intptr_t i = 0; i < kMapSize; i++) (*map_)[i] = true; |
| 2919 } |
| 2920 return; |
| 2921 } |
| 2922 for (intptr_t i = interval.from(); i <= interval.to(); i++) { |
| 2923 intptr_t mod_character = (i & kMask); |
| 2924 if (!map_->At(mod_character)) { |
| 2925 map_count_++; |
| 2926 (*map_)[mod_character] = true; |
| 2927 } |
| 2928 if (map_count_ == kMapSize) return; |
| 2929 } |
| 2930 } |
| 2931 |
| 2932 |
| 2933 void BoyerMoorePositionInfo::SetAll() { |
| 2934 s_ = w_ = d_ = kLatticeUnknown; |
| 2935 if (map_count_ != kMapSize) { |
| 2936 map_count_ = kMapSize; |
| 2937 for (intptr_t i = 0; i < kMapSize; i++) (*map_)[i] = true; |
| 2938 } |
| 2939 } |
| 2940 |
| 2941 |
| 2942 BoyerMooreLookahead::BoyerMooreLookahead( |
| 2943 intptr_t length, RegExpCompiler* compiler, Isolate* isolate) |
| 2944 : length_(length), |
| 2945 compiler_(compiler) { |
| 2946 if (compiler->one_byte()) { |
| 2947 max_char_ = Symbols::kMaxOneCharCodeSymbol; |
| 2948 } else { |
| 2949 max_char_ = Utf16::kMaxCodeUnit; |
| 2950 } |
| 2951 bitmaps_ = new(isolate) ZoneGrowableArray<BoyerMoorePositionInfo*>(length); |
| 2952 for (intptr_t i = 0; i < length; i++) { |
| 2953 bitmaps_->Add(new(isolate) BoyerMoorePositionInfo(isolate)); |
| 2954 } |
| 2955 } |
| 2956 |
| 2957 |
| 2958 // Find the longest range of lookahead that has the fewest number of different |
| 2959 // characters that can occur at a given position. Since we are optimizing two |
| 2960 // different parameters at once this is a tradeoff. |
| 2961 bool BoyerMooreLookahead::FindWorthwhileInterval(intptr_t* from, intptr_t* to) { |
| 2962 intptr_t biggest_points = 0; |
| 2963 // If more than 32 characters out of 128 can occur it is unlikely that we can |
| 2964 // be lucky enough to step forwards much of the time. |
| 2965 const intptr_t kMaxMax = 32; |
| 2966 for (intptr_t max_number_of_chars = 4; |
| 2967 max_number_of_chars < kMaxMax; |
| 2968 max_number_of_chars *= 2) { |
| 2969 biggest_points = |
| 2970 FindBestInterval(max_number_of_chars, biggest_points, from, to); |
| 2971 } |
| 2972 if (biggest_points == 0) return false; |
| 2973 return true; |
| 2974 } |
| 2975 |
| 2976 |
| 2977 // Find the highest-points range between 0 and length_ where the character |
| 2978 // information is not too vague. 'Too vague' means that there are more than |
| 2979 // max_number_of_chars that can occur at this position. Calculates the number |
| 2980 // of points as the product of width-of-the-range and |
| 2981 // probability-of-finding-one-of-the-characters, where the probability is |
| 2982 // calculated using the frequency distribution of the sample subject string. |
| 2983 intptr_t BoyerMooreLookahead::FindBestInterval( |
| 2984 intptr_t max_number_of_chars, |
| 2985 intptr_t old_biggest_points, |
| 2986 intptr_t* from, |
| 2987 intptr_t* to) { |
| 2988 intptr_t biggest_points = old_biggest_points; |
| 2989 static const intptr_t kSize = RegExpMacroAssembler::kTableSize; |
| 2990 for (intptr_t i = 0; i < length_; ) { |
| 2991 while (i < length_ && Count(i) > max_number_of_chars) i++; |
| 2992 if (i == length_) break; |
| 2993 intptr_t remembered_from = i; |
| 2994 bool union_map[kSize]; |
| 2995 for (intptr_t j = 0; j < kSize; j++) union_map[j] = false; |
| 2996 while (i < length_ && Count(i) <= max_number_of_chars) { |
| 2997 BoyerMoorePositionInfo* map = bitmaps_->At(i); |
| 2998 for (intptr_t j = 0; j < kSize; j++) union_map[j] |= map->at(j); |
| 2999 i++; |
| 3000 } |
| 3001 intptr_t frequency = 0; |
| 3002 for (intptr_t j = 0; j < kSize; j++) { |
| 3003 if (union_map[j]) { |
| 3004 // Add 1 to the frequency to give a small per-character boost for |
| 3005 // the cases where our sampling is not good enough and many |
| 3006 // characters have a frequency of zero. This means the frequency |
| 3007 // can theoretically be up to 2*kSize though we treat it mostly as |
| 3008 // a fraction of kSize. |
| 3009 frequency += compiler_->frequency_collator()->Frequency(j) + 1; |
| 3010 } |
| 3011 } |
| 3012 // We use the probability of skipping times the distance we are skipping to |
| 3013 // judge the effectiveness of this. Actually we have a cut-off: By |
| 3014 // dividing by 2 we switch off the skipping if the probability of skipping |
| 3015 // is less than 50%. This is because the multibyte mask-and-compare |
| 3016 // skipping in quickcheck is more likely to do well on this case. |
| 3017 bool in_quickcheck_range = ((i - remembered_from < 4) || |
| 3018 (compiler_->one_byte() ? remembered_from <= 4 : remembered_from <= 2)); |
| 3019 // Called 'probability' but it is only a rough estimate and can actually |
| 3020 // be outside the 0-kSize range. |
| 3021 intptr_t probability = |
| 3022 (in_quickcheck_range ? kSize / 2 : kSize) - frequency; |
| 3023 intptr_t points = (i - remembered_from) * probability; |
| 3024 if (points > biggest_points) { |
| 3025 *from = remembered_from; |
| 3026 *to = i - 1; |
| 3027 biggest_points = points; |
| 3028 } |
| 3029 } |
| 3030 return biggest_points; |
| 3031 } |
| 3032 |
| 3033 |
| 3034 // Take all the characters that will not prevent a successful match if they |
| 3035 // occur in the subject string in the range between min_lookahead and |
| 3036 // max_lookahead (inclusive) measured from the current position. If the |
| 3037 // character at max_lookahead offset is not one of these characters, then we |
| 3038 // can safely skip forwards by the number of characters in the range. |
| 3039 intptr_t BoyerMooreLookahead::GetSkipTable( |
| 3040 intptr_t min_lookahead, |
| 3041 intptr_t max_lookahead, |
| 3042 const TypedData& boolean_skip_table) { |
| 3043 const intptr_t kSize = RegExpMacroAssembler::kTableSize; |
| 3044 |
| 3045 const intptr_t kSkipArrayEntry = 0; |
| 3046 const intptr_t kDontSkipArrayEntry = 1; |
| 3047 |
| 3048 for (intptr_t i = 0; i < kSize; i++) { |
| 3049 boolean_skip_table.SetUint8(i, kSkipArrayEntry); |
| 3050 } |
| 3051 intptr_t skip = max_lookahead + 1 - min_lookahead; |
| 3052 |
| 3053 for (intptr_t i = max_lookahead; i >= min_lookahead; i--) { |
| 3054 BoyerMoorePositionInfo* map = bitmaps_->At(i); |
| 3055 for (intptr_t j = 0; j < kSize; j++) { |
| 3056 if (map->at(j)) { |
| 3057 boolean_skip_table.SetUint8(j, kDontSkipArrayEntry); |
| 3058 } |
| 3059 } |
| 3060 } |
| 3061 |
| 3062 return skip; |
| 3063 } |
| 3064 |
| 3065 |
| 3066 // See comment above on the implementation of GetSkipTable. |
| 3067 void BoyerMooreLookahead::EmitSkipInstructions(RegExpMacroAssembler* masm) { |
| 3068 const intptr_t kSize = RegExpMacroAssembler::kTableSize; |
| 3069 |
| 3070 intptr_t min_lookahead = 0; |
| 3071 intptr_t max_lookahead = 0; |
| 3072 |
| 3073 if (!FindWorthwhileInterval(&min_lookahead, &max_lookahead)) return; |
| 3074 |
| 3075 bool found_single_character = false; |
| 3076 intptr_t single_character = 0; |
| 3077 for (intptr_t i = max_lookahead; i >= min_lookahead; i--) { |
| 3078 BoyerMoorePositionInfo* map = bitmaps_->At(i); |
| 3079 if (map->map_count() > 1 || |
| 3080 (found_single_character && map->map_count() != 0)) { |
| 3081 found_single_character = false; |
| 3082 break; |
| 3083 } |
| 3084 for (intptr_t j = 0; j < kSize; j++) { |
| 3085 if (map->at(j)) { |
| 3086 found_single_character = true; |
| 3087 single_character = j; |
| 3088 break; |
| 3089 } |
| 3090 } |
| 3091 } |
| 3092 |
| 3093 intptr_t lookahead_width = max_lookahead + 1 - min_lookahead; |
| 3094 |
| 3095 if (found_single_character && lookahead_width == 1 && max_lookahead < 3) { |
| 3096 // The mask-compare can probably handle this better. |
| 3097 return; |
| 3098 } |
| 3099 |
| 3100 if (found_single_character) { |
| 3101 BlockLabel cont, again; |
| 3102 masm->BindBlock(&again); |
| 3103 masm->LoadCurrentCharacter(max_lookahead, &cont, true); |
| 3104 if (max_char_ > kSize) { |
| 3105 masm->CheckCharacterAfterAnd(single_character, |
| 3106 RegExpMacroAssembler::kTableMask, |
| 3107 &cont); |
| 3108 } else { |
| 3109 masm->CheckCharacter(single_character, &cont); |
| 3110 } |
| 3111 masm->AdvanceCurrentPosition(lookahead_width); |
| 3112 masm->GoTo(&again); |
| 3113 masm->BindBlock(&cont); |
| 3114 return; |
| 3115 } |
| 3116 |
| 3117 const TypedData& boolean_skip_table = TypedData::ZoneHandle( |
| 3118 compiler_->isolate(), |
| 3119 TypedData::New(kTypedDataUint8ArrayCid, kSize, Heap::kOld)); |
| 3120 intptr_t skip_distance = GetSkipTable( |
| 3121 min_lookahead, max_lookahead, boolean_skip_table); |
| 3122 ASSERT(skip_distance != 0); |
| 3123 |
| 3124 BlockLabel cont, again; |
| 3125 |
| 3126 masm->BindBlock(&again); |
| 3127 masm->LoadCurrentCharacter(max_lookahead, &cont, true); |
| 3128 masm->CheckBitInTable(boolean_skip_table, &cont); |
| 3129 masm->AdvanceCurrentPosition(skip_distance); |
| 3130 masm->GoTo(&again); |
| 3131 masm->BindBlock(&cont); |
| 3132 |
| 3133 return; |
| 3134 } |
| 3135 |
| 3136 |
| 3137 /* Code generation for choice nodes. |
| 3138 * |
| 3139 * We generate quick checks that do a mask and compare to eliminate a |
| 3140 * choice. If the quick check succeeds then it jumps to the continuation to |
| 3141 * do slow checks and check subsequent nodes. If it fails (the common case) |
| 3142 * it falls through to the next choice. |
| 3143 * |
| 3144 * Here is the desired flow graph. Nodes directly below each other imply |
| 3145 * fallthrough. Alternatives 1 and 2 have quick checks. Alternative |
| 3146 * 3 doesn't have a quick check so we have to call the slow check. |
| 3147 * Nodes are marked Qn for quick checks and Sn for slow checks. The entire |
| 3148 * regexp continuation is generated directly after the Sn node, up to the |
| 3149 * next GoTo if we decide to reuse some already generated code. Some |
| 3150 * nodes expect preload_characters to be preloaded into the current |
| 3151 * character register. R nodes do this preloading. Vertices are marked |
| 3152 * F for failures and S for success (possible success in the case of quick |
| 3153 * nodes). L, V, < and > are used as arrow heads. |
| 3154 * |
| 3155 * ----------> R |
| 3156 * | |
| 3157 * V |
| 3158 * Q1 -----> S1 |
| 3159 * | S / |
| 3160 * F| / |
| 3161 * | F/ |
| 3162 * | / |
| 3163 * | R |
| 3164 * | / |
| 3165 * V L |
| 3166 * Q2 -----> S2 |
| 3167 * | S / |
| 3168 * F| / |
| 3169 * | F/ |
| 3170 * | / |
| 3171 * | R |
| 3172 * | / |
| 3173 * V L |
| 3174 * S3 |
| 3175 * | |
| 3176 * F| |
| 3177 * | |
| 3178 * R |
| 3179 * | |
| 3180 * backtrack V |
| 3181 * <----------Q4 |
| 3182 * \ F | |
| 3183 * \ |S |
| 3184 * \ F V |
| 3185 * \-----S4 |
| 3186 * |
| 3187 * For greedy loops we push the current position, then generate the code that |
| 3188 * eats the input specially in EmitGreedyLoop. The other choice (the |
| 3189 * continuation) is generated by the normal code in EmitChoices, and steps back |
| 3190 * in the input to the starting position when it fails to match. The loop code |
| 3191 * looks like this (U is the unwind code that steps back in the greedy loop). |
| 3192 * |
| 3193 * _____ |
| 3194 * / \ |
| 3195 * V | |
| 3196 * ----------> S1 | |
| 3197 * /| | |
| 3198 * / |S | |
| 3199 * F/ \_____/ |
| 3200 * / |
| 3201 * |<----- |
| 3202 * | \ |
| 3203 * V |S |
| 3204 * Q2 ---> U----->backtrack |
| 3205 * | F / |
| 3206 * S| / |
| 3207 * V F / |
| 3208 * S2--/ |
| 3209 */ |
| 3210 |
| 3211 GreedyLoopState::GreedyLoopState(bool not_at_start) { |
| 3212 counter_backtrack_trace_.set_backtrack(&label_); |
| 3213 if (not_at_start) counter_backtrack_trace_.set_at_start(false); |
| 3214 } |
| 3215 |
| 3216 |
| 3217 void ChoiceNode::AssertGuardsMentionRegisters(Trace* trace) { |
| 3218 #ifdef DEBUG |
| 3219 intptr_t choice_count = alternatives_->length(); |
| 3220 for (intptr_t i = 0; i < choice_count - 1; i++) { |
| 3221 GuardedAlternative alternative = alternatives_->At(i); |
| 3222 ZoneGrowableArray<Guard*>* guards = alternative.guards(); |
| 3223 intptr_t guard_count = (guards == NULL) ? 0 : guards->length(); |
| 3224 for (intptr_t j = 0; j < guard_count; j++) { |
| 3225 ASSERT(!trace->mentions_reg(guards->At(j)->reg())); |
| 3226 } |
| 3227 } |
| 3228 #endif |
| 3229 } |
| 3230 |
| 3231 |
| 3232 void ChoiceNode::SetUpPreLoad(RegExpCompiler* compiler, |
| 3233 Trace* current_trace, |
| 3234 PreloadState* state) { |
| 3235 if (state->eats_at_least_ == PreloadState::kEatsAtLeastNotYetInitialized) { |
| 3236 // Save some time by looking at most one machine word ahead. |
| 3237 state->eats_at_least_ = |
| 3238 EatsAtLeast(compiler->one_byte() ? 4 : 2, kRecursionBudget, |
| 3239 current_trace->at_start() == Trace::FALSE_VALUE); |
| 3240 } |
| 3241 state->preload_characters_ = |
| 3242 CalculatePreloadCharacters(compiler, state->eats_at_least_); |
| 3243 |
| 3244 state->preload_is_current_ = |
| 3245 (current_trace->characters_preloaded() == state->preload_characters_); |
| 3246 state->preload_has_checked_bounds_ = state->preload_is_current_; |
| 3247 } |
| 3248 |
| 3249 |
| 3250 void ChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) { |
| 3251 intptr_t choice_count = alternatives_->length(); |
| 3252 |
| 3253 AssertGuardsMentionRegisters(trace); |
| 3254 |
| 3255 LimitResult limit_result = LimitVersions(compiler, trace); |
| 3256 if (limit_result == DONE) return; |
| 3257 ASSERT(limit_result == CONTINUE); |
| 3258 |
| 3259 // For loop nodes we already flushed (see LoopChoiceNode::Emit), but for |
| 3260 // other choice nodes we only flush if we are out of code size budget. |
| 3261 if (trace->flush_budget() == 0 && trace->actions() != NULL) { |
| 3262 trace->Flush(compiler, this); |
| 3263 return; |
| 3264 } |
| 3265 |
| 3266 RecursionCheck rc(compiler); |
| 3267 |
| 3268 PreloadState preload; |
| 3269 preload.init(); |
| 3270 GreedyLoopState greedy_loop_state(not_at_start()); |
| 3271 |
| 3272 intptr_t text_length = |
| 3273 GreedyLoopTextLengthForAlternative(&((*alternatives_)[0])); |
| 3274 AlternativeGenerationList alt_gens(choice_count); |
| 3275 |
| 3276 if (choice_count > 1 && text_length != kNodeIsTooComplexForGreedyLoops) { |
| 3277 trace = EmitGreedyLoop(compiler, |
| 3278 trace, |
| 3279 &alt_gens, |
| 3280 &preload, |
| 3281 &greedy_loop_state, |
| 3282 text_length); |
| 3283 } else { |
| 3284 // TODO(erikcorry): Delete this. We don't need this label, but it makes us |
| 3285 // match the traces produced pre-cleanup. |
| 3286 BlockLabel second_choice; |
| 3287 compiler->macro_assembler()->BindBlock(&second_choice); |
| 3288 |
| 3289 preload.eats_at_least_ = EmitOptimizedUnanchoredSearch(compiler, trace); |
| 3290 |
| 3291 EmitChoices(compiler, |
| 3292 &alt_gens, |
| 3293 0, |
| 3294 trace, |
| 3295 &preload); |
| 3296 } |
| 3297 |
| 3298 // At this point we need to generate slow checks for the alternatives where |
| 3299 // the quick check was inlined. We can recognize these because the associated |
| 3300 // label was bound. |
| 3301 intptr_t new_flush_budget = trace->flush_budget() / choice_count; |
| 3302 for (intptr_t i = 0; i < choice_count; i++) { |
| 3303 AlternativeGeneration* alt_gen = alt_gens.at(i); |
| 3304 Trace new_trace(*trace); |
| 3305 // If there are actions to be flushed we have to limit how many times |
| 3306 // they are flushed. Take the budget of the parent trace and distribute |
| 3307 // it fairly amongst the children. |
| 3308 if (new_trace.actions() != NULL) { |
| 3309 new_trace.set_flush_budget(new_flush_budget); |
| 3310 } |
| 3311 bool next_expects_preload = |
| 3312 i == choice_count - 1 ? false : alt_gens.at(i + 1)->expects_preload; |
| 3313 EmitOutOfLineContinuation(compiler, |
| 3314 &new_trace, |
| 3315 alternatives_->At(i), |
| 3316 alt_gen, |
| 3317 preload.preload_characters_, |
| 3318 next_expects_preload); |
| 3319 } |
| 3320 } |
| 3321 |
| 3322 Trace* ChoiceNode::EmitGreedyLoop(RegExpCompiler* compiler, |
| 3323 Trace* trace, |
| 3324 AlternativeGenerationList* alt_gens, |
| 3325 PreloadState* preload, |
| 3326 GreedyLoopState* greedy_loop_state, |
| 3327 intptr_t text_length) { |
| 3328 RegExpMacroAssembler* macro_assembler = compiler->macro_assembler(); |
| 3329 // Here we have special handling for greedy loops containing only text nodes |
| 3330 // and other simple nodes. These are handled by pushing the current |
| 3331 // position on the stack and then incrementing the current position each |
| 3332 // time around the switch. On backtrack we decrement the current position |
| 3333 // and check it against the pushed value. This avoids pushing backtrack |
| 3334 // information for each iteration of the loop, which could take up a lot of |
| 3335 // space. |
| 3336 ASSERT(trace->stop_node() == NULL); |
| 3337 macro_assembler->PushCurrentPosition(); |
| 3338 BlockLabel greedy_match_failed; |
| 3339 Trace greedy_match_trace; |
| 3340 if (not_at_start()) greedy_match_trace.set_at_start(false); |
| 3341 greedy_match_trace.set_backtrack(&greedy_match_failed); |
| 3342 BlockLabel loop_label; |
| 3343 macro_assembler->BindBlock(&loop_label); |
| 3344 greedy_match_trace.set_stop_node(this); |
| 3345 greedy_match_trace.set_loop_label(&loop_label); |
| 3346 (*alternatives_)[0].node()->Emit(compiler, &greedy_match_trace); |
| 3347 macro_assembler->BindBlock(&greedy_match_failed); |
| 3348 |
| 3349 BlockLabel second_choice; // For use in greedy matches. |
| 3350 macro_assembler->BindBlock(&second_choice); |
| 3351 |
| 3352 Trace* new_trace = greedy_loop_state->counter_backtrack_trace(); |
| 3353 |
| 3354 EmitChoices(compiler, |
| 3355 alt_gens, |
| 3356 1, |
| 3357 new_trace, |
| 3358 preload); |
| 3359 |
| 3360 macro_assembler->BindBlock(greedy_loop_state->label()); |
| 3361 // If we have unwound to the bottom then backtrack. |
| 3362 macro_assembler->CheckGreedyLoop(trace->backtrack()); |
| 3363 // Otherwise try the second priority at an earlier position. |
| 3364 macro_assembler->AdvanceCurrentPosition(-text_length); |
| 3365 macro_assembler->GoTo(&second_choice); |
| 3366 return new_trace; |
| 3367 } |
| 3368 |
| 3369 |
| 3370 intptr_t ChoiceNode::EmitOptimizedUnanchoredSearch(RegExpCompiler* compiler, |
| 3371 Trace* trace) { |
| 3372 intptr_t eats_at_least = PreloadState::kEatsAtLeastNotYetInitialized; |
| 3373 if (alternatives_->length() != 2) return eats_at_least; |
| 3374 |
| 3375 GuardedAlternative alt1 = alternatives_->At(1); |
| 3376 if (alt1.guards() != NULL && alt1.guards()->length() != 0) { |
| 3377 return eats_at_least; |
| 3378 } |
| 3379 RegExpNode* eats_anything_node = alt1.node(); |
| 3380 if (eats_anything_node->GetSuccessorOfOmnivorousTextNode(compiler) != this) { |
| 3381 return eats_at_least; |
| 3382 } |
| 3383 |
| 3384 // Really we should be creating a new trace when we execute this function, |
| 3385 // but there is no need, because the code it generates cannot backtrack, and |
| 3386 // we always arrive here with a trivial trace (since it's the entry to a |
| 3387 // loop. That also implies that there are no preloaded characters, which is |
| 3388 // good, because it means we won't be violating any assumptions by |
| 3389 // overwriting those characters with new load instructions. |
| 3390 ASSERT(trace->is_trivial()); |
| 3391 |
| 3392 RegExpMacroAssembler* macro_assembler = compiler->macro_assembler(); |
| 3393 // At this point we know that we are at a non-greedy loop that will eat |
| 3394 // any character one at a time. Any non-anchored regexp has such a |
| 3395 // loop prepended to it in order to find where it starts. We look for |
| 3396 // a pattern of the form ...abc... where we can look 6 characters ahead |
| 3397 // and step forwards 3 if the character is not one of abc. Abc need |
| 3398 // not be atoms, they can be any reasonably limited character class or |
| 3399 // small alternation. |
| 3400 BoyerMooreLookahead* bm = bm_info(false); |
| 3401 if (bm == NULL) { |
| 3402 eats_at_least = Utils::Minimum(kMaxLookaheadForBoyerMoore, |
| 3403 EatsAtLeast(kMaxLookaheadForBoyerMoore, |
| 3404 kRecursionBudget, |
| 3405 false)); |
| 3406 if (eats_at_least >= 1) { |
| 3407 bm = new(I) BoyerMooreLookahead(eats_at_least, compiler, I); |
| 3408 GuardedAlternative alt0 = alternatives_->At(0); |
| 3409 alt0.node()->FillInBMInfo(0, kRecursionBudget, bm, false); |
| 3410 } |
| 3411 } |
| 3412 if (bm != NULL) { |
| 3413 bm->EmitSkipInstructions(macro_assembler); |
| 3414 } |
| 3415 return eats_at_least; |
| 3416 } |
| 3417 |
| 3418 |
| 3419 void ChoiceNode::EmitChoices(RegExpCompiler* compiler, |
| 3420 AlternativeGenerationList* alt_gens, |
| 3421 intptr_t first_choice, |
| 3422 Trace* trace, |
| 3423 PreloadState* preload) { |
| 3424 RegExpMacroAssembler* macro_assembler = compiler->macro_assembler(); |
| 3425 SetUpPreLoad(compiler, trace, preload); |
| 3426 |
| 3427 // For now we just call all choices one after the other. The idea ultimately |
| 3428 // is to use the Dispatch table to try only the relevant ones. |
| 3429 intptr_t choice_count = alternatives_->length(); |
| 3430 |
| 3431 intptr_t new_flush_budget = trace->flush_budget() / choice_count; |
| 3432 |
| 3433 for (intptr_t i = first_choice; i < choice_count; i++) { |
| 3434 bool is_last = i == choice_count - 1; |
| 3435 bool fall_through_on_failure = !is_last; |
| 3436 GuardedAlternative alternative = alternatives_->At(i); |
| 3437 AlternativeGeneration* alt_gen = alt_gens->at(i); |
| 3438 alt_gen->quick_check_details.set_characters(preload->preload_characters_); |
| 3439 ZoneGrowableArray<Guard*>* guards = alternative.guards(); |
| 3440 intptr_t guard_count = (guards == NULL) ? 0 : guards->length(); |
| 3441 Trace new_trace(*trace); |
| 3442 new_trace.set_characters_preloaded(preload->preload_is_current_ ? |
| 3443 preload->preload_characters_ : |
| 3444 0); |
| 3445 if (preload->preload_has_checked_bounds_) { |
| 3446 new_trace.set_bound_checked_up_to(preload->preload_characters_); |
| 3447 } |
| 3448 new_trace.quick_check_performed()->Clear(); |
| 3449 if (not_at_start_) new_trace.set_at_start(Trace::FALSE_VALUE); |
| 3450 if (!is_last) { |
| 3451 new_trace.set_backtrack(&alt_gen->after); |
| 3452 } |
| 3453 alt_gen->expects_preload = preload->preload_is_current_; |
| 3454 bool generate_full_check_inline = false; |
| 3455 if (kRegexpOptimization && |
| 3456 try_to_emit_quick_check_for_alternative(i == 0) && |
| 3457 alternative.node()->EmitQuickCheck(compiler, |
| 3458 trace, |
| 3459 &new_trace, |
| 3460 preload->preload_has_checked_bounds_, |
| 3461 &alt_gen->possible_success, |
| 3462 &alt_gen->quick_check_details, |
| 3463 fall_through_on_failure)) { |
| 3464 // Quick check was generated for this choice. |
| 3465 preload->preload_is_current_ = true; |
| 3466 preload->preload_has_checked_bounds_ = true; |
| 3467 // If we generated the quick check to fall through on possible success, |
| 3468 // we now need to generate the full check inline. |
| 3469 if (!fall_through_on_failure) { |
| 3470 macro_assembler->BindBlock(&alt_gen->possible_success); |
| 3471 new_trace.set_quick_check_performed(&alt_gen->quick_check_details); |
| 3472 new_trace.set_characters_preloaded(preload->preload_characters_); |
| 3473 new_trace.set_bound_checked_up_to(preload->preload_characters_); |
| 3474 generate_full_check_inline = true; |
| 3475 } |
| 3476 } else if (alt_gen->quick_check_details.cannot_match()) { |
| 3477 if (!fall_through_on_failure) { |
| 3478 macro_assembler->GoTo(trace->backtrack()); |
| 3479 } |
| 3480 continue; |
| 3481 } else { |
| 3482 // No quick check was generated. Put the full code here. |
| 3483 // If this is not the first choice then there could be slow checks from |
| 3484 // previous cases that go here when they fail. There's no reason to |
| 3485 // insist that they preload characters since the slow check we are about |
| 3486 // to generate probably can't use it. |
| 3487 if (i != first_choice) { |
| 3488 alt_gen->expects_preload = false; |
| 3489 new_trace.InvalidateCurrentCharacter(); |
| 3490 } |
| 3491 generate_full_check_inline = true; |
| 3492 } |
| 3493 if (generate_full_check_inline) { |
| 3494 if (new_trace.actions() != NULL) { |
| 3495 new_trace.set_flush_budget(new_flush_budget); |
| 3496 } |
| 3497 for (intptr_t j = 0; j < guard_count; j++) { |
| 3498 GenerateGuard(macro_assembler, guards->At(j), &new_trace); |
| 3499 } |
| 3500 alternative.node()->Emit(compiler, &new_trace); |
| 3501 preload->preload_is_current_ = false; |
| 3502 } |
| 3503 macro_assembler->BindBlock(&alt_gen->after); |
| 3504 } |
| 3505 } |
| 3506 |
| 3507 |
| 3508 void ChoiceNode::EmitOutOfLineContinuation(RegExpCompiler* compiler, |
| 3509 Trace* trace, |
| 3510 GuardedAlternative alternative, |
| 3511 AlternativeGeneration* alt_gen, |
| 3512 intptr_t preload_characters, |
| 3513 bool next_expects_preload) { |
| 3514 if (!alt_gen->possible_success.IsLinked()) return; |
| 3515 |
| 3516 RegExpMacroAssembler* macro_assembler = compiler->macro_assembler(); |
| 3517 macro_assembler->BindBlock(&alt_gen->possible_success); |
| 3518 Trace out_of_line_trace(*trace); |
| 3519 out_of_line_trace.set_characters_preloaded(preload_characters); |
| 3520 out_of_line_trace.set_quick_check_performed(&alt_gen->quick_check_details); |
| 3521 if (not_at_start_) out_of_line_trace.set_at_start(Trace::FALSE_VALUE); |
| 3522 ZoneGrowableArray<Guard*>* guards = alternative.guards(); |
| 3523 intptr_t guard_count = (guards == NULL) ? 0 : guards->length(); |
| 3524 if (next_expects_preload) { |
| 3525 BlockLabel reload_current_char; |
| 3526 out_of_line_trace.set_backtrack(&reload_current_char); |
| 3527 for (intptr_t j = 0; j < guard_count; j++) { |
| 3528 GenerateGuard(macro_assembler, guards->At(j), &out_of_line_trace); |
| 3529 } |
| 3530 alternative.node()->Emit(compiler, &out_of_line_trace); |
| 3531 macro_assembler->BindBlock(&reload_current_char); |
| 3532 // Reload the current character, since the next quick check expects that. |
| 3533 // We don't need to check bounds here because we only get into this |
| 3534 // code through a quick check which already did the checked load. |
| 3535 macro_assembler->LoadCurrentCharacter(trace->cp_offset(), |
| 3536 NULL, |
| 3537 false, |
| 3538 preload_characters); |
| 3539 macro_assembler->GoTo(&(alt_gen->after)); |
| 3540 } else { |
| 3541 out_of_line_trace.set_backtrack(&(alt_gen->after)); |
| 3542 for (intptr_t j = 0; j < guard_count; j++) { |
| 3543 GenerateGuard(macro_assembler, guards->At(j), &out_of_line_trace); |
| 3544 } |
| 3545 alternative.node()->Emit(compiler, &out_of_line_trace); |
| 3546 } |
| 3547 } |
| 3548 |
| 3549 |
| 3550 void ActionNode::Emit(RegExpCompiler* compiler, Trace* trace) { |
| 3551 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 3552 LimitResult limit_result = LimitVersions(compiler, trace); |
| 3553 if (limit_result == DONE) return; |
| 3554 ASSERT(limit_result == CONTINUE); |
| 3555 |
| 3556 RecursionCheck rc(compiler); |
| 3557 |
| 3558 switch (action_type_) { |
| 3559 case STORE_POSITION: { |
| 3560 Trace::DeferredCapture |
| 3561 new_capture(data_.u_position_register.reg, |
| 3562 data_.u_position_register.is_capture, |
| 3563 trace); |
| 3564 Trace new_trace = *trace; |
| 3565 new_trace.add_action(&new_capture); |
| 3566 on_success()->Emit(compiler, &new_trace); |
| 3567 break; |
| 3568 } |
| 3569 case INCREMENT_REGISTER: { |
| 3570 Trace::DeferredIncrementRegister |
| 3571 new_increment(data_.u_increment_register.reg); |
| 3572 Trace new_trace = *trace; |
| 3573 new_trace.add_action(&new_increment); |
| 3574 on_success()->Emit(compiler, &new_trace); |
| 3575 break; |
| 3576 } |
| 3577 case SET_REGISTER: { |
| 3578 Trace::DeferredSetRegister |
| 3579 new_set(data_.u_store_register.reg, data_.u_store_register.value); |
| 3580 Trace new_trace = *trace; |
| 3581 new_trace.add_action(&new_set); |
| 3582 on_success()->Emit(compiler, &new_trace); |
| 3583 break; |
| 3584 } |
| 3585 case CLEAR_CAPTURES: { |
| 3586 Trace::DeferredClearCaptures |
| 3587 new_capture(Interval(data_.u_clear_captures.range_from, |
| 3588 data_.u_clear_captures.range_to)); |
| 3589 Trace new_trace = *trace; |
| 3590 new_trace.add_action(&new_capture); |
| 3591 on_success()->Emit(compiler, &new_trace); |
| 3592 break; |
| 3593 } |
| 3594 case BEGIN_SUBMATCH: |
| 3595 if (!trace->is_trivial()) { |
| 3596 trace->Flush(compiler, this); |
| 3597 } else { |
| 3598 assembler->WriteCurrentPositionToRegister( |
| 3599 data_.u_submatch.current_position_register, 0); |
| 3600 assembler->WriteStackPointerToRegister( |
| 3601 data_.u_submatch.stack_pointer_register); |
| 3602 on_success()->Emit(compiler, trace); |
| 3603 } |
| 3604 break; |
| 3605 case EMPTY_MATCH_CHECK: { |
| 3606 intptr_t start_pos_reg = data_.u_empty_match_check.start_register; |
| 3607 intptr_t stored_pos = 0; |
| 3608 intptr_t rep_reg = data_.u_empty_match_check.repetition_register; |
| 3609 bool has_minimum = (rep_reg != RegExpCompiler::kNoRegister); |
| 3610 bool know_dist = trace->GetStoredPosition(start_pos_reg, &stored_pos); |
| 3611 if (know_dist && !has_minimum && stored_pos == trace->cp_offset()) { |
| 3612 // If we know we haven't advanced and there is no minimum we |
| 3613 // can just backtrack immediately. |
| 3614 assembler->GoTo(trace->backtrack()); |
| 3615 } else if (know_dist && stored_pos < trace->cp_offset()) { |
| 3616 // If we know we've advanced we can generate the continuation |
| 3617 // immediately. |
| 3618 on_success()->Emit(compiler, trace); |
| 3619 } else if (!trace->is_trivial()) { |
| 3620 trace->Flush(compiler, this); |
| 3621 } else { |
| 3622 BlockLabel skip_empty_check; |
| 3623 // If we have a minimum number of repetitions we check the current |
| 3624 // number first and skip the empty check if it's not enough. |
| 3625 if (has_minimum) { |
| 3626 intptr_t limit = data_.u_empty_match_check.repetition_limit; |
| 3627 assembler->IfRegisterLT(rep_reg, limit, &skip_empty_check); |
| 3628 } |
| 3629 // If the match is empty we bail out, otherwise we fall through |
| 3630 // to the on-success continuation. |
| 3631 assembler->IfRegisterEqPos(data_.u_empty_match_check.start_register, |
| 3632 trace->backtrack()); |
| 3633 assembler->BindBlock(&skip_empty_check); |
| 3634 on_success()->Emit(compiler, trace); |
| 3635 } |
| 3636 break; |
| 3637 } |
| 3638 case POSITIVE_SUBMATCH_SUCCESS: { |
| 3639 if (!trace->is_trivial()) { |
| 3640 trace->Flush(compiler, this); |
| 3641 return; |
| 3642 } |
| 3643 assembler->ReadCurrentPositionFromRegister( |
| 3644 data_.u_submatch.current_position_register); |
| 3645 assembler->ReadStackPointerFromRegister( |
| 3646 data_.u_submatch.stack_pointer_register); |
| 3647 intptr_t clear_register_count = data_.u_submatch.clear_register_count; |
| 3648 if (clear_register_count == 0) { |
| 3649 on_success()->Emit(compiler, trace); |
| 3650 return; |
| 3651 } |
| 3652 intptr_t clear_registers_from = data_.u_submatch.clear_register_from; |
| 3653 BlockLabel clear_registers_backtrack; |
| 3654 Trace new_trace = *trace; |
| 3655 new_trace.set_backtrack(&clear_registers_backtrack); |
| 3656 on_success()->Emit(compiler, &new_trace); |
| 3657 |
| 3658 assembler->BindBlock(&clear_registers_backtrack); |
| 3659 intptr_t clear_registers_to = |
| 3660 clear_registers_from + clear_register_count - 1; |
| 3661 assembler->ClearRegisters(clear_registers_from, clear_registers_to); |
| 3662 |
| 3663 ASSERT(trace->backtrack() == NULL); |
| 3664 assembler->Backtrack(); |
| 3665 return; |
| 3666 } |
| 3667 default: |
| 3668 UNREACHABLE(); |
| 3669 } |
| 3670 } |
| 3671 |
| 3672 |
| 3673 void BackReferenceNode::Emit(RegExpCompiler* compiler, Trace* trace) { |
| 3674 RegExpMacroAssembler* assembler = compiler->macro_assembler(); |
| 3675 if (!trace->is_trivial()) { |
| 3676 trace->Flush(compiler, this); |
| 3677 return; |
| 3678 } |
| 3679 |
| 3680 LimitResult limit_result = LimitVersions(compiler, trace); |
| 3681 if (limit_result == DONE) return; |
| 3682 ASSERT(limit_result == CONTINUE); |
| 3683 |
| 3684 RecursionCheck rc(compiler); |
| 3685 |
| 3686 ASSERT(start_reg_ + 1 == end_reg_); |
| 3687 if (compiler->ignore_case()) { |
| 3688 assembler->CheckNotBackReferenceIgnoreCase(start_reg_, |
| 3689 trace->backtrack()); |
| 3690 } else { |
| 3691 assembler->CheckNotBackReference(start_reg_, trace->backtrack()); |
| 3692 } |
| 3693 on_success()->Emit(compiler, trace); |
| 3694 } |
| 3695 |
| 3696 |
| 3697 // ------------------------------------------------------------------- |
| 3698 // Dot/dotty output |
| 3699 |
| 3700 |
| 3701 #ifdef DEBUG |
| 3702 |
| 3703 |
| 3704 class DotPrinter: public NodeVisitor { |
| 3705 public: |
| 3706 explicit DotPrinter(bool ignore_case) |
| 3707 : ignore_case_(ignore_case) {} |
| 3708 void PrintNode(const char* label, RegExpNode* node); |
| 3709 void Visit(RegExpNode* node); |
| 3710 void PrintAttributes(RegExpNode* from); |
| 3711 void PrintOnFailure(RegExpNode* from, RegExpNode* to); |
| 3712 #define DECLARE_VISIT(Type) \ |
| 3713 virtual void Visit##Type(Type##Node* that); |
| 3714 FOR_EACH_NODE_TYPE(DECLARE_VISIT) |
| 3715 #undef DECLARE_VISIT |
| 3716 private: |
| 3717 bool ignore_case_; |
| 3718 }; |
| 3719 |
| 3720 |
| 3721 void DotPrinter::PrintNode(const char* label, RegExpNode* node) { |
| 3722 OS::Print("digraph G {\n graph [label=\""); |
| 3723 for (intptr_t i = 0; label[i]; i++) { |
| 3724 switch (label[i]) { |
| 3725 case '\\': |
| 3726 OS::Print("\\\\"); |
| 3727 break; |
| 3728 case '"': |
| 3729 OS::Print("\""); |
| 3730 break; |
| 3731 default: |
| 3732 OS::Print("%c", label[i]); |
| 3733 break; |
| 3734 } |
| 3735 } |
| 3736 OS::Print("\"];\n"); |
| 3737 Visit(node); |
| 3738 OS::Print("}\n"); |
| 3739 } |
| 3740 |
| 3741 |
| 3742 void DotPrinter::Visit(RegExpNode* node) { |
| 3743 if (node->info()->visited) return; |
| 3744 node->info()->visited = true; |
| 3745 node->Accept(this); |
| 3746 } |
| 3747 |
| 3748 |
| 3749 void DotPrinter::PrintOnFailure(RegExpNode* from, RegExpNode* on_failure) { |
| 3750 OS::Print(" n%p -> n%p [style=dotted];\n", from, on_failure); |
| 3751 Visit(on_failure); |
| 3752 } |
| 3753 |
| 3754 |
| 3755 class AttributePrinter : public ValueObject { |
| 3756 public: |
| 3757 AttributePrinter() : first_(true) {} |
| 3758 void PrintSeparator() { |
| 3759 if (first_) { |
| 3760 first_ = false; |
| 3761 } else { |
| 3762 OS::Print("|"); |
| 3763 } |
| 3764 } |
| 3765 void PrintBit(const char* name, bool value) { |
| 3766 if (!value) return; |
| 3767 PrintSeparator(); |
| 3768 OS::Print("{%s}", name); |
| 3769 } |
| 3770 void PrintPositive(const char* name, intptr_t value) { |
| 3771 if (value < 0) return; |
| 3772 PrintSeparator(); |
| 3773 OS::Print("{%s|%" Pd "}", name, value); |
| 3774 } |
| 3775 |
| 3776 private: |
| 3777 bool first_; |
| 3778 }; |
| 3779 |
| 3780 |
| 3781 void DotPrinter::PrintAttributes(RegExpNode* that) { |
| 3782 OS::Print(" a%p [shape=Mrecord, color=grey, fontcolor=grey, " |
| 3783 "margin=0.1, fontsize=10, label=\"{", that); |
| 3784 AttributePrinter printer; |
| 3785 NodeInfo* info = that->info(); |
| 3786 printer.PrintBit("NI", info->follows_newline_interest); |
| 3787 printer.PrintBit("WI", info->follows_word_interest); |
| 3788 printer.PrintBit("SI", info->follows_start_interest); |
| 3789 BlockLabel* label = that->label(); |
| 3790 if (label->IsBound()) |
| 3791 printer.PrintPositive("@", label->Position()); |
| 3792 OS::Print("}\"];\n" |
| 3793 " a%p -> n%p [style=dashed, color=grey, arrowhead=none];\n", |
| 3794 that, that); |
| 3795 } |
| 3796 |
| 3797 |
| 3798 void DotPrinter::VisitChoice(ChoiceNode* that) { |
| 3799 OS::Print(" n%p [shape=Mrecord, label=\"?\"];\n", that); |
| 3800 for (intptr_t i = 0; i < that->alternatives()->length(); i++) { |
| 3801 GuardedAlternative alt = that->alternatives()->At(i); |
| 3802 OS::Print(" n%p -> n%p", that, alt.node()); |
| 3803 } |
| 3804 for (intptr_t i = 0; i < that->alternatives()->length(); i++) { |
| 3805 GuardedAlternative alt = that->alternatives()->At(i); |
| 3806 alt.node()->Accept(this); |
| 3807 } |
| 3808 } |
| 3809 |
| 3810 |
| 3811 void DotPrinter::VisitText(TextNode* that) { |
| 3812 OS::Print(" n%p [label=\"", that); |
| 3813 for (intptr_t i = 0; i < that->elements()->length(); i++) { |
| 3814 if (i > 0) OS::Print(" "); |
| 3815 TextElement elm = that->elements()->At(i); |
| 3816 switch (elm.text_type()) { |
| 3817 case TextElement::ATOM: { |
| 3818 ZoneGrowableArray<uint16_t>* data = elm.atom()->data(); |
| 3819 for (intptr_t i = 0; i < data->length(); i++) { |
| 3820 OS::Print("%c", static_cast<char>(data->At(i))); |
| 3821 } |
| 3822 break; |
| 3823 } |
| 3824 case TextElement::CHAR_CLASS: { |
| 3825 RegExpCharacterClass* node = elm.char_class(); |
| 3826 OS::Print("["); |
| 3827 if (node->is_negated()) OS::Print("^"); |
| 3828 for (intptr_t j = 0; j < node->ranges()->length(); j++) { |
| 3829 CharacterRange range = node->ranges()->At(j); |
| 3830 PrintUtf16(range.from()); |
| 3831 OS::Print("-"); |
| 3832 PrintUtf16(range.to()); |
| 3833 } |
| 3834 OS::Print("]"); |
| 3835 break; |
| 3836 } |
| 3837 default: |
| 3838 UNREACHABLE(); |
| 3839 } |
| 3840 } |
| 3841 OS::Print("\", shape=box, peripheries=2];\n"); |
| 3842 PrintAttributes(that); |
| 3843 OS::Print(" n%p -> n%p;\n", that, that->on_success()); |
| 3844 Visit(that->on_success()); |
| 3845 } |
| 3846 |
| 3847 |
| 3848 void DotPrinter::VisitBackReference(BackReferenceNode* that) { |
| 3849 OS::Print(" n%p [label=\"$%" Pd "..$%" Pd "\", shape=doubleoctagon];\n", |
| 3850 that, that->start_register(), that->end_register()); |
| 3851 PrintAttributes(that); |
| 3852 OS::Print(" n%p -> n%p;\n", that, that->on_success()); |
| 3853 Visit(that->on_success()); |
| 3854 } |
| 3855 |
| 3856 |
| 3857 void DotPrinter::VisitEnd(EndNode* that) { |
| 3858 OS::Print(" n%p [style=bold, shape=point];\n", that); |
| 3859 PrintAttributes(that); |
| 3860 } |
| 3861 |
| 3862 |
| 3863 void DotPrinter::VisitAssertion(AssertionNode* that) { |
| 3864 OS::Print(" n%p [", that); |
| 3865 switch (that->assertion_type()) { |
| 3866 case AssertionNode::AT_END: |
| 3867 OS::Print("label=\"$\", shape=septagon"); |
| 3868 break; |
| 3869 case AssertionNode::AT_START: |
| 3870 OS::Print("label=\"^\", shape=septagon"); |
| 3871 break; |
| 3872 case AssertionNode::AT_BOUNDARY: |
| 3873 OS::Print("label=\"\\b\", shape=septagon"); |
| 3874 break; |
| 3875 case AssertionNode::AT_NON_BOUNDARY: |
| 3876 OS::Print("label=\"\\B\", shape=septagon"); |
| 3877 break; |
| 3878 case AssertionNode::AFTER_NEWLINE: |
| 3879 OS::Print("label=\"(?<=\\n)\", shape=septagon"); |
| 3880 break; |
| 3881 } |
| 3882 OS::Print("];\n"); |
| 3883 PrintAttributes(that); |
| 3884 RegExpNode* successor = that->on_success(); |
| 3885 OS::Print(" n%p -> n%p;\n", that, successor); |
| 3886 Visit(successor); |
| 3887 } |
| 3888 |
| 3889 |
| 3890 void DotPrinter::VisitAction(ActionNode* that) { |
| 3891 OS::Print(" n%p [", that); |
| 3892 switch (that->action_type_) { |
| 3893 case ActionNode::SET_REGISTER: |
| 3894 OS::Print("label=\"$%" Pd ":=%" Pd "\", shape=octagon", |
| 3895 that->data_.u_store_register.reg, |
| 3896 that->data_.u_store_register.value); |
| 3897 break; |
| 3898 case ActionNode::INCREMENT_REGISTER: |
| 3899 OS::Print("label=\"$%" Pd "++\", shape=octagon", |
| 3900 that->data_.u_increment_register.reg); |
| 3901 break; |
| 3902 case ActionNode::STORE_POSITION: |
| 3903 OS::Print("label=\"$%" Pd ":=$pos\", shape=octagon", |
| 3904 that->data_.u_position_register.reg); |
| 3905 break; |
| 3906 case ActionNode::BEGIN_SUBMATCH: |
| 3907 OS::Print("label=\"$%" Pd ":=$pos,begin\", shape=septagon", |
| 3908 that->data_.u_submatch.current_position_register); |
| 3909 break; |
| 3910 case ActionNode::POSITIVE_SUBMATCH_SUCCESS: |
| 3911 OS::Print("label=\"escape\", shape=septagon"); |
| 3912 break; |
| 3913 case ActionNode::EMPTY_MATCH_CHECK: |
| 3914 OS::Print("label=\"$%" Pd "=$pos?,$%" Pd "<%" Pd "?\", shape=septagon", |
| 3915 that->data_.u_empty_match_check.start_register, |
| 3916 that->data_.u_empty_match_check.repetition_register, |
| 3917 that->data_.u_empty_match_check.repetition_limit); |
| 3918 break; |
| 3919 case ActionNode::CLEAR_CAPTURES: { |
| 3920 OS::Print("label=\"clear $%" Pd " to $%" Pd "\", shape=septagon", |
| 3921 that->data_.u_clear_captures.range_from, |
| 3922 that->data_.u_clear_captures.range_to); |
| 3923 break; |
| 3924 } |
| 3925 } |
| 3926 OS::Print("];\n"); |
| 3927 PrintAttributes(that); |
| 3928 RegExpNode* successor = that->on_success(); |
| 3929 OS::Print(" n%p -> n%p;\n", that, successor); |
| 3930 Visit(successor); |
| 3931 } |
| 3932 |
| 3933 |
| 3934 void RegExpEngine::DotPrint(const char* label, |
| 3935 RegExpNode* node, |
| 3936 bool ignore_case) { |
| 3937 DotPrinter printer(ignore_case); |
| 3938 printer.PrintNode(label, node); |
| 3939 } |
| 3940 |
| 3941 |
| 3942 #endif // DEBUG |
| 3943 |
| 3944 |
| 3945 // ------------------------------------------------------------------- |
| 3946 // Tree to graph conversion |
| 3947 |
| 3948 RegExpNode* RegExpAtom::ToNode(RegExpCompiler* compiler, |
| 3949 RegExpNode* on_success) { |
| 3950 ZoneGrowableArray<TextElement>* elms = |
| 3951 new(CI) ZoneGrowableArray<TextElement>(1); |
| 3952 elms->Add(TextElement::Atom(this)); |
| 3953 return new(CI) TextNode(elms, on_success); |
| 3954 } |
| 3955 |
| 3956 |
| 3957 RegExpNode* RegExpText::ToNode(RegExpCompiler* compiler, |
| 3958 RegExpNode* on_success) { |
| 3959 ZoneGrowableArray<TextElement>* elms = |
| 3960 new(CI) ZoneGrowableArray<TextElement>(1); |
| 3961 for (intptr_t i = 0; i < elements()->length(); i++) { |
| 3962 elms->Add(elements()->At(i)); |
| 3963 } |
| 3964 return new(CI) TextNode(elms, on_success); |
| 3965 } |
| 3966 |
| 3967 |
| 3968 static bool CompareInverseRanges(ZoneGrowableArray<CharacterRange>* ranges, |
| 3969 const intptr_t* special_class, |
| 3970 intptr_t length) { |
| 3971 length--; // Remove final 0x10000. |
| 3972 ASSERT(special_class[length] == 0x10000); |
| 3973 ASSERT(ranges->length() != 0); |
| 3974 ASSERT(length != 0); |
| 3975 ASSERT(special_class[0] != 0); |
| 3976 if (ranges->length() != (length >> 1) + 1) { |
| 3977 return false; |
| 3978 } |
| 3979 CharacterRange range = ranges->At(0); |
| 3980 if (range.from() != 0) { |
| 3981 return false; |
| 3982 } |
| 3983 for (intptr_t i = 0; i < length; i += 2) { |
| 3984 if (special_class[i] != (range.to() + 1)) { |
| 3985 return false; |
| 3986 } |
| 3987 range = ranges->At((i >> 1) + 1); |
| 3988 if (special_class[i+1] != range.from()) { |
| 3989 return false; |
| 3990 } |
| 3991 } |
| 3992 if (range.to() != 0xffff) { |
| 3993 return false; |
| 3994 } |
| 3995 return true; |
| 3996 } |
| 3997 |
| 3998 |
| 3999 static bool CompareRanges(ZoneGrowableArray<CharacterRange>* ranges, |
| 4000 const intptr_t* special_class, |
| 4001 intptr_t length) { |
| 4002 length--; // Remove final 0x10000. |
| 4003 ASSERT(special_class[length] == 0x10000); |
| 4004 if (ranges->length() * 2 != length) { |
| 4005 return false; |
| 4006 } |
| 4007 for (intptr_t i = 0; i < length; i += 2) { |
| 4008 CharacterRange range = ranges->At(i >> 1); |
| 4009 if (range.from() != special_class[i] || |
| 4010 range.to() != special_class[i + 1] - 1) { |
| 4011 return false; |
| 4012 } |
| 4013 } |
| 4014 return true; |
| 4015 } |
| 4016 |
| 4017 |
| 4018 bool RegExpCharacterClass::is_standard() { |
| 4019 // TODO(lrn): Remove need for this function, by not throwing away information |
| 4020 // along the way. |
| 4021 if (is_negated_) { |
| 4022 return false; |
| 4023 } |
| 4024 if (set_.is_standard()) { |
| 4025 return true; |
| 4026 } |
| 4027 if (CompareRanges(set_.ranges(), kSpaceRanges, kSpaceRangeCount)) { |
| 4028 set_.set_standard_set_type('s'); |
| 4029 return true; |
| 4030 } |
| 4031 if (CompareInverseRanges(set_.ranges(), kSpaceRanges, kSpaceRangeCount)) { |
| 4032 set_.set_standard_set_type('S'); |
| 4033 return true; |
| 4034 } |
| 4035 if (CompareInverseRanges(set_.ranges(), |
| 4036 kLineTerminatorRanges, |
| 4037 kLineTerminatorRangeCount)) { |
| 4038 set_.set_standard_set_type('.'); |
| 4039 return true; |
| 4040 } |
| 4041 if (CompareRanges(set_.ranges(), |
| 4042 kLineTerminatorRanges, |
| 4043 kLineTerminatorRangeCount)) { |
| 4044 set_.set_standard_set_type('n'); |
| 4045 return true; |
| 4046 } |
| 4047 if (CompareRanges(set_.ranges(), kWordRanges, kWordRangeCount)) { |
| 4048 set_.set_standard_set_type('w'); |
| 4049 return true; |
| 4050 } |
| 4051 if (CompareInverseRanges(set_.ranges(), kWordRanges, kWordRangeCount)) { |
| 4052 set_.set_standard_set_type('W'); |
| 4053 return true; |
| 4054 } |
| 4055 return false; |
| 4056 } |
| 4057 |
| 4058 |
| 4059 RegExpNode* RegExpCharacterClass::ToNode(RegExpCompiler* compiler, |
| 4060 RegExpNode* on_success) { |
| 4061 return new(CI) TextNode(this, on_success); |
| 4062 } |
| 4063 |
| 4064 |
| 4065 RegExpNode* RegExpDisjunction::ToNode(RegExpCompiler* compiler, |
| 4066 RegExpNode* on_success) { |
| 4067 ZoneGrowableArray<RegExpTree*>* alternatives = this->alternatives(); |
| 4068 intptr_t length = alternatives->length(); |
| 4069 ChoiceNode* result = |
| 4070 new(CI) ChoiceNode(length, CI); |
| 4071 for (intptr_t i = 0; i < length; i++) { |
| 4072 GuardedAlternative alternative(alternatives->At(i)->ToNode(compiler, |
| 4073 on_success)); |
| 4074 result->AddAlternative(alternative); |
| 4075 } |
| 4076 return result; |
| 4077 } |
| 4078 |
| 4079 |
| 4080 RegExpNode* RegExpQuantifier::ToNode(RegExpCompiler* compiler, |
| 4081 RegExpNode* on_success) { |
| 4082 return ToNode(min(), |
| 4083 max(), |
| 4084 is_greedy(), |
| 4085 body(), |
| 4086 compiler, |
| 4087 on_success); |
| 4088 } |
| 4089 |
| 4090 |
| 4091 // Scoped object to keep track of how much we unroll quantifier loops in the |
| 4092 // regexp graph generator. |
| 4093 class RegExpExpansionLimiter : public ValueObject { |
| 4094 public: |
| 4095 static const intptr_t kMaxExpansionFactor = 6; |
| 4096 RegExpExpansionLimiter(RegExpCompiler* compiler, intptr_t factor) |
| 4097 : compiler_(compiler), |
| 4098 saved_expansion_factor_(compiler->current_expansion_factor()), |
| 4099 ok_to_expand_(saved_expansion_factor_ <= kMaxExpansionFactor) { |
| 4100 ASSERT(factor > 0); |
| 4101 if (ok_to_expand_) { |
| 4102 if (factor > kMaxExpansionFactor) { |
| 4103 // Avoid integer overflow of the current expansion factor. |
| 4104 ok_to_expand_ = false; |
| 4105 compiler->set_current_expansion_factor(kMaxExpansionFactor + 1); |
| 4106 } else { |
| 4107 intptr_t new_factor = saved_expansion_factor_ * factor; |
| 4108 ok_to_expand_ = (new_factor <= kMaxExpansionFactor); |
| 4109 compiler->set_current_expansion_factor(new_factor); |
| 4110 } |
| 4111 } |
| 4112 } |
| 4113 |
| 4114 ~RegExpExpansionLimiter() { |
| 4115 compiler_->set_current_expansion_factor(saved_expansion_factor_); |
| 4116 } |
| 4117 |
| 4118 bool ok_to_expand() { return ok_to_expand_; } |
| 4119 |
| 4120 private: |
| 4121 RegExpCompiler* compiler_; |
| 4122 intptr_t saved_expansion_factor_; |
| 4123 bool ok_to_expand_; |
| 4124 |
| 4125 DISALLOW_IMPLICIT_CONSTRUCTORS(RegExpExpansionLimiter); |
| 4126 }; |
| 4127 |
| 4128 |
| 4129 RegExpNode* RegExpQuantifier::ToNode(intptr_t min, |
| 4130 intptr_t max, |
| 4131 bool is_greedy, |
| 4132 RegExpTree* body, |
| 4133 RegExpCompiler* compiler, |
| 4134 RegExpNode* on_success, |
| 4135 bool not_at_start) { |
| 4136 // x{f, t} becomes this: |
| 4137 // |
| 4138 // (r++)<-. |
| 4139 // | ` |
| 4140 // | (x) |
| 4141 // v ^ |
| 4142 // (r=0)-->(?)---/ [if r < t] |
| 4143 // | |
| 4144 // [if r >= f] \----> ... |
| 4145 // |
| 4146 |
| 4147 // 15.10.2.5 RepeatMatcher algorithm. |
| 4148 // The parser has already eliminated the case where max is 0. In the case |
| 4149 // where max_match is zero the parser has removed the quantifier if min was |
| 4150 // > 0 and removed the atom if min was 0. See AddQuantifierToAtom. |
| 4151 |
| 4152 // If we know that we cannot match zero length then things are a little |
| 4153 // simpler since we don't need to make the special zero length match check |
| 4154 // from step 2.1. If the min and max are small we can unroll a little in |
| 4155 // this case. |
| 4156 // Unroll (foo)+ and (foo){3,} |
| 4157 static const intptr_t kMaxUnrolledMinMatches = 3; |
| 4158 // Unroll (foo)? and (foo){x,3} |
| 4159 static const intptr_t kMaxUnrolledMaxMatches = 3; |
| 4160 if (max == 0) return on_success; // This can happen due to recursion. |
| 4161 bool body_can_be_empty = (body->min_match() == 0); |
| 4162 intptr_t body_start_reg = RegExpCompiler::kNoRegister; |
| 4163 Interval capture_registers = body->CaptureRegisters(); |
| 4164 bool needs_capture_clearing = !capture_registers.is_empty(); |
| 4165 Isolate* isolate = compiler->isolate(); |
| 4166 |
| 4167 if (body_can_be_empty) { |
| 4168 body_start_reg = compiler->AllocateRegister(); |
| 4169 } else if (kRegexpOptimization && !needs_capture_clearing) { |
| 4170 // Only unroll if there are no captures and the body can't be |
| 4171 // empty. |
| 4172 { |
| 4173 RegExpExpansionLimiter limiter( |
| 4174 compiler, min + ((max != min) ? 1 : 0)); |
| 4175 if (min > 0 && min <= kMaxUnrolledMinMatches && limiter.ok_to_expand()) { |
| 4176 intptr_t new_max = (max == kInfinity) ? max : max - min; |
| 4177 // Recurse once to get the loop or optional matches after the fixed |
| 4178 // ones. |
| 4179 RegExpNode* answer = ToNode( |
| 4180 0, new_max, is_greedy, body, compiler, on_success, true); |
| 4181 // Unroll the forced matches from 0 to min. This can cause chains of |
| 4182 // TextNodes (which the parser does not generate). These should be |
| 4183 // combined if it turns out they hinder good code generation. |
| 4184 for (intptr_t i = 0; i < min; i++) { |
| 4185 answer = body->ToNode(compiler, answer); |
| 4186 } |
| 4187 return answer; |
| 4188 } |
| 4189 } |
| 4190 if (max <= kMaxUnrolledMaxMatches && min == 0) { |
| 4191 ASSERT(max > 0); // Due to the 'if' above. |
| 4192 RegExpExpansionLimiter limiter(compiler, max); |
| 4193 if (limiter.ok_to_expand()) { |
| 4194 // Unroll the optional matches up to max. |
| 4195 RegExpNode* answer = on_success; |
| 4196 for (intptr_t i = 0; i < max; i++) { |
| 4197 ChoiceNode* alternation = new(isolate) ChoiceNode(2, isolate); |
| 4198 if (is_greedy) { |
| 4199 alternation->AddAlternative( |
| 4200 GuardedAlternative(body->ToNode(compiler, answer))); |
| 4201 alternation->AddAlternative(GuardedAlternative(on_success)); |
| 4202 } else { |
| 4203 alternation->AddAlternative(GuardedAlternative(on_success)); |
| 4204 alternation->AddAlternative( |
| 4205 GuardedAlternative(body->ToNode(compiler, answer))); |
| 4206 } |
| 4207 answer = alternation; |
| 4208 if (not_at_start) alternation->set_not_at_start(); |
| 4209 } |
| 4210 return answer; |
| 4211 } |
| 4212 } |
| 4213 } |
| 4214 bool has_min = min > 0; |
| 4215 bool has_max = max < RegExpTree::kInfinity; |
| 4216 bool needs_counter = has_min || has_max; |
| 4217 intptr_t reg_ctr = needs_counter |
| 4218 ? compiler->AllocateRegister() |
| 4219 : RegExpCompiler::kNoRegister; |
| 4220 LoopChoiceNode* center = new(isolate) LoopChoiceNode(body->min_match() == 0, |
| 4221 isolate); |
| 4222 if (not_at_start) center->set_not_at_start(); |
| 4223 RegExpNode* loop_return = needs_counter |
| 4224 ? static_cast<RegExpNode*>(ActionNode::IncrementRegister(reg_ctr, center)) |
| 4225 : static_cast<RegExpNode*>(center); |
| 4226 if (body_can_be_empty) { |
| 4227 // If the body can be empty we need to check if it was and then |
| 4228 // backtrack. |
| 4229 loop_return = ActionNode::EmptyMatchCheck(body_start_reg, |
| 4230 reg_ctr, |
| 4231 min, |
| 4232 loop_return); |
| 4233 } |
| 4234 RegExpNode* body_node = body->ToNode(compiler, loop_return); |
| 4235 if (body_can_be_empty) { |
| 4236 // If the body can be empty we need to store the start position |
| 4237 // so we can bail out if it was empty. |
| 4238 body_node = ActionNode::StorePosition(body_start_reg, false, body_node); |
| 4239 } |
| 4240 if (needs_capture_clearing) { |
| 4241 // Before entering the body of this loop we need to clear captures. |
| 4242 body_node = ActionNode::ClearCaptures(capture_registers, body_node); |
| 4243 } |
| 4244 GuardedAlternative body_alt(body_node); |
| 4245 if (has_max) { |
| 4246 Guard* body_guard = |
| 4247 new(isolate) Guard(reg_ctr, Guard::LT, max); |
| 4248 body_alt.AddGuard(body_guard, isolate); |
| 4249 } |
| 4250 GuardedAlternative rest_alt(on_success); |
| 4251 if (has_min) { |
| 4252 Guard* rest_guard = new(isolate) Guard(reg_ctr, Guard::GEQ, min); |
| 4253 rest_alt.AddGuard(rest_guard, isolate); |
| 4254 } |
| 4255 if (is_greedy) { |
| 4256 center->AddLoopAlternative(body_alt); |
| 4257 center->AddContinueAlternative(rest_alt); |
| 4258 } else { |
| 4259 center->AddContinueAlternative(rest_alt); |
| 4260 center->AddLoopAlternative(body_alt); |
| 4261 } |
| 4262 if (needs_counter) { |
| 4263 return ActionNode::SetRegister(reg_ctr, 0, center); |
| 4264 } else { |
| 4265 return center; |
| 4266 } |
| 4267 } |
| 4268 |
| 4269 |
| 4270 RegExpNode* RegExpAssertion::ToNode(RegExpCompiler* compiler, |
| 4271 RegExpNode* on_success) { |
| 4272 switch (assertion_type()) { |
| 4273 case START_OF_LINE: |
| 4274 return AssertionNode::AfterNewline(on_success); |
| 4275 case START_OF_INPUT: |
| 4276 return AssertionNode::AtStart(on_success); |
| 4277 case BOUNDARY: |
| 4278 return AssertionNode::AtBoundary(on_success); |
| 4279 case NON_BOUNDARY: |
| 4280 return AssertionNode::AtNonBoundary(on_success); |
| 4281 case END_OF_INPUT: |
| 4282 return AssertionNode::AtEnd(on_success); |
| 4283 case END_OF_LINE: { |
| 4284 // Compile $ in multiline regexps as an alternation with a positive |
| 4285 // lookahead in one side and an end-of-input on the other side. |
| 4286 // We need two registers for the lookahead. |
| 4287 intptr_t stack_pointer_register = compiler->AllocateRegister(); |
| 4288 intptr_t position_register = compiler->AllocateRegister(); |
| 4289 // The ChoiceNode to distinguish between a newline and end-of-input. |
| 4290 ChoiceNode* result = new ChoiceNode(2, on_success->isolate()); |
| 4291 // Create a newline atom. |
| 4292 ZoneGrowableArray<CharacterRange>* newline_ranges = |
| 4293 new ZoneGrowableArray<CharacterRange>(3); |
| 4294 CharacterRange::AddClassEscape('n', newline_ranges); |
| 4295 RegExpCharacterClass* newline_atom = new RegExpCharacterClass('n'); |
| 4296 TextNode* newline_matcher = new TextNode( |
| 4297 newline_atom, |
| 4298 ActionNode::PositiveSubmatchSuccess(stack_pointer_register, |
| 4299 position_register, |
| 4300 0, // No captures inside. |
| 4301 -1, // Ignored if no captures. |
| 4302 on_success)); |
| 4303 // Create an end-of-input matcher. |
| 4304 RegExpNode* end_of_line = ActionNode::BeginSubmatch( |
| 4305 stack_pointer_register, |
| 4306 position_register, |
| 4307 newline_matcher); |
| 4308 // Add the two alternatives to the ChoiceNode. |
| 4309 GuardedAlternative eol_alternative(end_of_line); |
| 4310 result->AddAlternative(eol_alternative); |
| 4311 GuardedAlternative end_alternative(AssertionNode::AtEnd(on_success)); |
| 4312 result->AddAlternative(end_alternative); |
| 4313 return result; |
| 4314 } |
| 4315 default: |
| 4316 UNREACHABLE(); |
| 4317 } |
| 4318 return on_success; |
| 4319 } |
| 4320 |
| 4321 |
| 4322 RegExpNode* RegExpBackReference::ToNode(RegExpCompiler* compiler, |
| 4323 RegExpNode* on_success) { |
| 4324 return new(CI) |
| 4325 BackReferenceNode(RegExpCapture::StartRegister(index()), |
| 4326 RegExpCapture::EndRegister(index()), |
| 4327 on_success); |
| 4328 } |
| 4329 |
| 4330 |
| 4331 RegExpNode* RegExpEmpty::ToNode(RegExpCompiler* compiler, |
| 4332 RegExpNode* on_success) { |
| 4333 return on_success; |
| 4334 } |
| 4335 |
| 4336 |
| 4337 RegExpNode* RegExpLookahead::ToNode(RegExpCompiler* compiler, |
| 4338 RegExpNode* on_success) { |
| 4339 intptr_t stack_pointer_register = compiler->AllocateRegister(); |
| 4340 intptr_t position_register = compiler->AllocateRegister(); |
| 4341 |
| 4342 const intptr_t registers_per_capture = 2; |
| 4343 const intptr_t register_of_first_capture = 2; |
| 4344 intptr_t register_count = capture_count_ * registers_per_capture; |
| 4345 intptr_t register_start = |
| 4346 register_of_first_capture + capture_from_ * registers_per_capture; |
| 4347 |
| 4348 RegExpNode* success; |
| 4349 if (is_positive()) { |
| 4350 RegExpNode* node = ActionNode::BeginSubmatch( |
| 4351 stack_pointer_register, |
| 4352 position_register, |
| 4353 body()->ToNode( |
| 4354 compiler, |
| 4355 ActionNode::PositiveSubmatchSuccess(stack_pointer_register, |
| 4356 position_register, |
| 4357 register_count, |
| 4358 register_start, |
| 4359 on_success))); |
| 4360 return node; |
| 4361 } else { |
| 4362 // We use a ChoiceNode for a negative lookahead because it has most of |
| 4363 // the characteristics we need. It has the body of the lookahead as its |
| 4364 // first alternative and the expression after the lookahead of the second |
| 4365 // alternative. If the first alternative succeeds then the |
| 4366 // NegativeSubmatchSuccess will unwind the stack including everything the |
| 4367 // choice node set up and backtrack. If the first alternative fails then |
| 4368 // the second alternative is tried, which is exactly the desired result |
| 4369 // for a negative lookahead. The NegativeLookaheadChoiceNode is a special |
| 4370 // ChoiceNode that knows to ignore the first exit when calculating quick |
| 4371 // checks. |
| 4372 |
| 4373 GuardedAlternative body_alt( |
| 4374 body()->ToNode( |
| 4375 compiler, |
| 4376 success = new(CI) NegativeSubmatchSuccess(stack_pointer_register, |
| 4377 position_register, |
| 4378 register_count, |
| 4379 register_start, |
| 4380 CI))); |
| 4381 ChoiceNode* choice_node = |
| 4382 new(CI) NegativeLookaheadChoiceNode(body_alt, |
| 4383 GuardedAlternative(on_success), |
| 4384 CI); |
| 4385 return ActionNode::BeginSubmatch(stack_pointer_register, |
| 4386 position_register, |
| 4387 choice_node); |
| 4388 } |
| 4389 } |
| 4390 |
| 4391 |
| 4392 RegExpNode* RegExpCapture::ToNode(RegExpCompiler* compiler, |
| 4393 RegExpNode* on_success) { |
| 4394 return ToNode(body(), index(), compiler, on_success); |
| 4395 } |
| 4396 |
| 4397 |
| 4398 RegExpNode* RegExpCapture::ToNode(RegExpTree* body, |
| 4399 intptr_t index, |
| 4400 RegExpCompiler* compiler, |
| 4401 RegExpNode* on_success) { |
| 4402 intptr_t start_reg = RegExpCapture::StartRegister(index); |
| 4403 intptr_t end_reg = RegExpCapture::EndRegister(index); |
| 4404 RegExpNode* store_end = ActionNode::StorePosition(end_reg, true, on_success); |
| 4405 RegExpNode* body_node = body->ToNode(compiler, store_end); |
| 4406 return ActionNode::StorePosition(start_reg, true, body_node); |
| 4407 } |
| 4408 |
| 4409 |
| 4410 RegExpNode* RegExpAlternative::ToNode(RegExpCompiler* compiler, |
| 4411 RegExpNode* on_success) { |
| 4412 ZoneGrowableArray<RegExpTree*>* children = nodes(); |
| 4413 RegExpNode* current = on_success; |
| 4414 for (intptr_t i = children->length() - 1; i >= 0; i--) { |
| 4415 current = children->At(i)->ToNode(compiler, current); |
| 4416 } |
| 4417 return current; |
| 4418 } |
| 4419 |
| 4420 |
| 4421 static void AddClass(const intptr_t* elmv, |
| 4422 intptr_t elmc, |
| 4423 ZoneGrowableArray<CharacterRange>* ranges) { |
| 4424 elmc--; |
| 4425 ASSERT(elmv[elmc] == 0x10000); |
| 4426 for (intptr_t i = 0; i < elmc; i += 2) { |
| 4427 ASSERT(elmv[i] < elmv[i + 1]); |
| 4428 ranges->Add(CharacterRange(elmv[i], elmv[i + 1] - 1)); |
| 4429 } |
| 4430 } |
| 4431 |
| 4432 |
| 4433 static void AddClassNegated(const intptr_t *elmv, |
| 4434 intptr_t elmc, |
| 4435 ZoneGrowableArray<CharacterRange>* ranges) { |
| 4436 elmc--; |
| 4437 ASSERT(elmv[elmc] == 0x10000); |
| 4438 ASSERT(elmv[0] != 0x0000); |
| 4439 ASSERT(elmv[elmc-1] != Utf16::kMaxCodeUnit); |
| 4440 uint16_t last = 0x0000; |
| 4441 for (intptr_t i = 0; i < elmc; i += 2) { |
| 4442 ASSERT(last <= elmv[i] - 1); |
| 4443 ASSERT(elmv[i] < elmv[i + 1]); |
| 4444 ranges->Add(CharacterRange(last, elmv[i] - 1)); |
| 4445 last = elmv[i + 1]; |
| 4446 } |
| 4447 ranges->Add(CharacterRange(last, Utf16::kMaxCodeUnit)); |
| 4448 } |
| 4449 |
| 4450 |
| 4451 void CharacterRange::AddClassEscape(uint16_t type, |
| 4452 ZoneGrowableArray<CharacterRange>* ranges) { |
| 4453 switch (type) { |
| 4454 case 's': |
| 4455 AddClass(kSpaceRanges, kSpaceRangeCount, ranges); |
| 4456 break; |
| 4457 case 'S': |
| 4458 AddClassNegated(kSpaceRanges, kSpaceRangeCount, ranges); |
| 4459 break; |
| 4460 case 'w': |
| 4461 AddClass(kWordRanges, kWordRangeCount, ranges); |
| 4462 break; |
| 4463 case 'W': |
| 4464 AddClassNegated(kWordRanges, kWordRangeCount, ranges); |
| 4465 break; |
| 4466 case 'd': |
| 4467 AddClass(kDigitRanges, kDigitRangeCount, ranges); |
| 4468 break; |
| 4469 case 'D': |
| 4470 AddClassNegated(kDigitRanges, kDigitRangeCount, ranges); |
| 4471 break; |
| 4472 case '.': |
| 4473 AddClassNegated(kLineTerminatorRanges, |
| 4474 kLineTerminatorRangeCount, |
| 4475 ranges); |
| 4476 break; |
| 4477 // This is not a character range as defined by the spec but a |
| 4478 // convenient shorthand for a character class that matches any |
| 4479 // character. |
| 4480 case '*': |
| 4481 ranges->Add(CharacterRange::Everything()); |
| 4482 break; |
| 4483 // This is the set of characters matched by the $ and ^ symbols |
| 4484 // in multiline mode. |
| 4485 case 'n': |
| 4486 AddClass(kLineTerminatorRanges, |
| 4487 kLineTerminatorRangeCount, |
| 4488 ranges); |
| 4489 break; |
| 4490 default: |
| 4491 UNREACHABLE(); |
| 4492 } |
| 4493 } |
| 4494 |
| 4495 |
| 4496 void CharacterRange::AddCaseEquivalents( |
| 4497 ZoneGrowableArray<CharacterRange>* ranges, |
| 4498 bool is_one_byte, |
| 4499 Isolate* isolate) { |
| 4500 uint16_t bottom = from(); |
| 4501 uint16_t top = to(); |
| 4502 if (is_one_byte && !RangeContainsLatin1Equivalents(*this)) { |
| 4503 if (bottom > Symbols::kMaxOneCharCodeSymbol) return; |
| 4504 if (top > Symbols::kMaxOneCharCodeSymbol) { |
| 4505 top = Symbols::kMaxOneCharCodeSymbol; |
| 4506 } |
| 4507 } |
| 4508 |
| 4509 unibrow::Mapping<unibrow::Ecma262UnCanonicalize> jsregexp_uncanonicalize; |
| 4510 unibrow::Mapping<unibrow::CanonicalizationRange> jsregexp_canonrange; |
| 4511 int32_t chars[unibrow::Ecma262UnCanonicalize::kMaxWidth]; |
| 4512 if (top == bottom) { |
| 4513 // If this is a singleton we just expand the one character. |
| 4514 intptr_t length = jsregexp_uncanonicalize.get(bottom, '\0', chars); // NOLIN
T |
| 4515 for (intptr_t i = 0; i < length; i++) { |
| 4516 uint32_t chr = chars[i]; |
| 4517 if (chr != bottom) { |
| 4518 ranges->Add(CharacterRange::Singleton(chars[i])); |
| 4519 } |
| 4520 } |
| 4521 } else { |
| 4522 // If this is a range we expand the characters block by block, |
| 4523 // expanding contiguous subranges (blocks) one at a time. |
| 4524 // The approach is as follows. For a given start character we |
| 4525 // look up the remainder of the block that contains it (represented |
| 4526 // by the end point), for instance we find 'z' if the character |
| 4527 // is 'c'. A block is characterized by the property |
| 4528 // that all characters uncanonicalize in the same way, except that |
| 4529 // each entry in the result is incremented by the distance from the first |
| 4530 // element. So a-z is a block because 'a' uncanonicalizes to ['a', 'A'] and |
| 4531 // the k'th letter uncanonicalizes to ['a' + k, 'A' + k]. |
| 4532 // Once we've found the end point we look up its uncanonicalization |
| 4533 // and produce a range for each element. For instance for [c-f] |
| 4534 // we look up ['z', 'Z'] and produce [c-f] and [C-F]. We then only |
| 4535 // add a range if it is not already contained in the input, so [c-f] |
| 4536 // will be skipped but [C-F] will be added. If this range is not |
| 4537 // completely contained in a block we do this for all the blocks |
| 4538 // covered by the range (handling characters that is not in a block |
| 4539 // as a "singleton block"). |
| 4540 int32_t range[unibrow::Ecma262UnCanonicalize::kMaxWidth]; |
| 4541 intptr_t pos = bottom; |
| 4542 while (pos <= top) { |
| 4543 intptr_t length = jsregexp_canonrange.get(pos, '\0', range); |
| 4544 uint16_t block_end; |
| 4545 if (length == 0) { |
| 4546 block_end = pos; |
| 4547 } else { |
| 4548 ASSERT(length == 1); |
| 4549 block_end = range[0]; |
| 4550 } |
| 4551 intptr_t end = (block_end > top) ? top : block_end; |
| 4552 length = jsregexp_uncanonicalize.get(block_end, '\0', range); // NOLINT |
| 4553 for (intptr_t i = 0; i < length; i++) { |
| 4554 uint32_t c = range[i]; |
| 4555 uint16_t range_from = c - (block_end - pos); |
| 4556 uint16_t range_to = c - (block_end - end); |
| 4557 if (!(bottom <= range_from && range_to <= top)) { |
| 4558 ranges->Add(CharacterRange(range_from, range_to)); |
| 4559 } |
| 4560 } |
| 4561 pos = end + 1; |
| 4562 } |
| 4563 } |
| 4564 } |
| 4565 |
| 4566 |
| 4567 bool CharacterRange::IsCanonical(ZoneGrowableArray<CharacterRange>* ranges) { |
| 4568 ASSERT(ranges != NULL); |
| 4569 intptr_t n = ranges->length(); |
| 4570 if (n <= 1) return true; |
| 4571 intptr_t max = ranges->At(0).to(); |
| 4572 for (intptr_t i = 1; i < n; i++) { |
| 4573 CharacterRange next_range = ranges->At(i); |
| 4574 if (next_range.from() <= max + 1) return false; |
| 4575 max = next_range.to(); |
| 4576 } |
| 4577 return true; |
| 4578 } |
| 4579 |
| 4580 |
| 4581 ZoneGrowableArray<CharacterRange>* CharacterSet::ranges() { |
| 4582 if (ranges_ == NULL) { |
| 4583 ranges_ = new ZoneGrowableArray<CharacterRange>(2); |
| 4584 CharacterRange::AddClassEscape(standard_set_type_, ranges_); |
| 4585 } |
| 4586 return ranges_; |
| 4587 } |
| 4588 |
| 4589 |
| 4590 // Move a number of elements in a zone array to another position |
| 4591 // in the same array. Handles overlapping source and target areas. |
| 4592 static void MoveRanges(ZoneGrowableArray<CharacterRange>* list, |
| 4593 intptr_t from, |
| 4594 intptr_t to, |
| 4595 intptr_t count) { |
| 4596 // Ranges are potentially overlapping. |
| 4597 if (from < to) { |
| 4598 for (intptr_t i = count - 1; i >= 0; i--) { |
| 4599 (*list)[to + i] = list->At(from + i); |
| 4600 } |
| 4601 } else { |
| 4602 for (intptr_t i = 0; i < count; i++) { |
| 4603 (*list)[to + i] = list->At(from + i); |
| 4604 } |
| 4605 } |
| 4606 } |
| 4607 |
| 4608 |
| 4609 static intptr_t InsertRangeInCanonicalList( |
| 4610 ZoneGrowableArray<CharacterRange>* list, |
| 4611 intptr_t count, |
| 4612 CharacterRange insert) { |
| 4613 // Inserts a range into list[0..count[, which must be sorted |
| 4614 // by from value and non-overlapping and non-adjacent, using at most |
| 4615 // list[0..count] for the result. Returns the number of resulting |
| 4616 // canonicalized ranges. Inserting a range may collapse existing ranges into |
| 4617 // fewer ranges, so the return value can be anything in the range 1..count+1. |
| 4618 uint16_t from = insert.from(); |
| 4619 uint16_t to = insert.to(); |
| 4620 intptr_t start_pos = 0; |
| 4621 intptr_t end_pos = count; |
| 4622 for (intptr_t i = count - 1; i >= 0; i--) { |
| 4623 CharacterRange current = list->At(i); |
| 4624 if (current.from() > to + 1) { |
| 4625 end_pos = i; |
| 4626 } else if (current.to() + 1 < from) { |
| 4627 start_pos = i + 1; |
| 4628 break; |
| 4629 } |
| 4630 } |
| 4631 |
| 4632 // Inserted range overlaps, or is adjacent to, ranges at positions |
| 4633 // [start_pos..end_pos[. Ranges before start_pos or at or after end_pos are |
| 4634 // not affected by the insertion. |
| 4635 // If start_pos == end_pos, the range must be inserted before start_pos. |
| 4636 // if start_pos < end_pos, the entire range from start_pos to end_pos |
| 4637 // must be merged with the insert range. |
| 4638 |
| 4639 if (start_pos == end_pos) { |
| 4640 // Insert between existing ranges at position start_pos. |
| 4641 if (start_pos < count) { |
| 4642 MoveRanges(list, start_pos, start_pos + 1, count - start_pos); |
| 4643 } |
| 4644 (*list)[start_pos] = insert; |
| 4645 return count + 1; |
| 4646 } |
| 4647 if (start_pos + 1 == end_pos) { |
| 4648 // Replace single existing range at position start_pos. |
| 4649 CharacterRange to_replace = list->At(start_pos); |
| 4650 intptr_t new_from = Utils::Minimum(to_replace.from(), from); |
| 4651 intptr_t new_to = Utils::Maximum(to_replace.to(), to); |
| 4652 (*list)[start_pos] = CharacterRange(new_from, new_to); |
| 4653 return count; |
| 4654 } |
| 4655 // Replace a number of existing ranges from start_pos to end_pos - 1. |
| 4656 // Move the remaining ranges down. |
| 4657 |
| 4658 intptr_t new_from = Utils::Minimum(list->At(start_pos).from(), from); |
| 4659 intptr_t new_to = Utils::Maximum(list->At(end_pos - 1).to(), to); |
| 4660 if (end_pos < count) { |
| 4661 MoveRanges(list, end_pos, start_pos + 1, count - end_pos); |
| 4662 } |
| 4663 (*list)[start_pos] = CharacterRange(new_from, new_to); |
| 4664 return count - (end_pos - start_pos) + 1; |
| 4665 } |
| 4666 |
| 4667 |
| 4668 void CharacterSet::Canonicalize() { |
| 4669 // Special/default classes are always considered canonical. The result |
| 4670 // of calling ranges() will be sorted. |
| 4671 if (ranges_ == NULL) return; |
| 4672 CharacterRange::Canonicalize(ranges_); |
| 4673 } |
| 4674 |
| 4675 |
| 4676 void CharacterRange::Canonicalize( |
| 4677 ZoneGrowableArray<CharacterRange>* character_ranges) { |
| 4678 if (character_ranges->length() <= 1) return; |
| 4679 // Check whether ranges are already canonical (increasing, non-overlapping, |
| 4680 // non-adjacent). |
| 4681 intptr_t n = character_ranges->length(); |
| 4682 intptr_t max = character_ranges->At(0).to(); |
| 4683 intptr_t i = 1; |
| 4684 while (i < n) { |
| 4685 CharacterRange current = character_ranges->At(i); |
| 4686 if (current.from() <= max + 1) { |
| 4687 break; |
| 4688 } |
| 4689 max = current.to(); |
| 4690 i++; |
| 4691 } |
| 4692 // Canonical until the i'th range. If that's all of them, we are done. |
| 4693 if (i == n) return; |
| 4694 |
| 4695 // The ranges at index i and forward are not canonicalized. Make them so by |
| 4696 // doing the equivalent of insertion sort (inserting each into the previous |
| 4697 // list, in order). |
| 4698 // Notice that inserting a range can reduce the number of ranges in the |
| 4699 // result due to combining of adjacent and overlapping ranges. |
| 4700 intptr_t read = i; // Range to insert. |
| 4701 intptr_t num_canonical = i; // Length of canonicalized part of list. |
| 4702 do { |
| 4703 num_canonical = InsertRangeInCanonicalList(character_ranges, |
| 4704 num_canonical, |
| 4705 character_ranges->At(read)); |
| 4706 read++; |
| 4707 } while (read < n); |
| 4708 character_ranges->TruncateTo(num_canonical); |
| 4709 |
| 4710 ASSERT(CharacterRange::IsCanonical(character_ranges)); |
| 4711 } |
| 4712 |
| 4713 |
| 4714 void CharacterRange::Negate(ZoneGrowableArray<CharacterRange>* ranges, |
| 4715 ZoneGrowableArray<CharacterRange>* negated_ranges) { |
| 4716 ASSERT(CharacterRange::IsCanonical(ranges)); |
| 4717 ASSERT(negated_ranges->length() == 0); |
| 4718 intptr_t range_count = ranges->length(); |
| 4719 uint16_t from = 0; |
| 4720 intptr_t i = 0; |
| 4721 if (range_count > 0 && ranges->At(0).from() == 0) { |
| 4722 from = ranges->At(0).to(); |
| 4723 i = 1; |
| 4724 } |
| 4725 while (i < range_count) { |
| 4726 CharacterRange range = ranges->At(i); |
| 4727 negated_ranges->Add(CharacterRange(from + 1, range.from() - 1)); |
| 4728 from = range.to(); |
| 4729 i++; |
| 4730 } |
| 4731 if (from < Utf16::kMaxCodeUnit) { |
| 4732 negated_ranges->Add(CharacterRange(from + 1, Utf16::kMaxCodeUnit)); |
| 4733 } |
| 4734 } |
| 4735 |
| 4736 |
| 4737 // ------------------------------------------------------------------- |
| 4738 // Splay tree |
| 4739 |
| 4740 |
| 4741 // Workaround for the fact that ZoneGrowableArray does not have contains(). |
| 4742 static bool ArrayContains(ZoneGrowableArray<unsigned>* array, |
| 4743 unsigned value) { |
| 4744 for (intptr_t i = 0; i < array->length(); i++) { |
| 4745 if (array->At(i) == value) { |
| 4746 return true; |
| 4747 } |
| 4748 } |
| 4749 return false; |
| 4750 } |
| 4751 |
| 4752 |
| 4753 void OutSet::Set(unsigned value, Isolate* isolate) { |
| 4754 if (value < kFirstLimit) { |
| 4755 first_ |= (1 << value); |
| 4756 } else { |
| 4757 if (remaining_ == NULL) |
| 4758 remaining_ = new(isolate) ZoneGrowableArray<unsigned>(1); |
| 4759 |
| 4760 bool remaining_contains_value = ArrayContains(remaining_, value); |
| 4761 if (remaining_->is_empty() || !remaining_contains_value) { |
| 4762 remaining_->Add(value); |
| 4763 } |
| 4764 } |
| 4765 } |
| 4766 |
| 4767 |
| 4768 bool OutSet::Get(unsigned value) const { |
| 4769 if (value < kFirstLimit) { |
| 4770 return (first_ & (1 << value)) != 0; |
| 4771 } else if (remaining_ == NULL) { |
| 4772 return false; |
| 4773 } else { |
| 4774 return ArrayContains(remaining_, value); |
| 4775 } |
| 4776 } |
| 4777 |
| 4778 |
| 4779 // ------------------------------------------------------------------- |
| 4780 // Analysis |
| 4781 |
| 4782 |
| 4783 void Analysis::EnsureAnalyzed(RegExpNode* that) { |
| 4784 if (that->info()->been_analyzed || that->info()->being_analyzed) |
| 4785 return; |
| 4786 that->info()->being_analyzed = true; |
| 4787 that->Accept(this); |
| 4788 that->info()->being_analyzed = false; |
| 4789 that->info()->been_analyzed = true; |
| 4790 } |
| 4791 |
| 4792 |
| 4793 void Analysis::VisitEnd(EndNode* that) { |
| 4794 // nothing to do |
| 4795 } |
| 4796 |
| 4797 |
| 4798 void TextNode::CalculateOffsets() { |
| 4799 intptr_t element_count = elements()->length(); |
| 4800 // Set up the offsets of the elements relative to the start. This is a fixed |
| 4801 // quantity since a TextNode can only contain fixed-width things. |
| 4802 intptr_t cp_offset = 0; |
| 4803 for (intptr_t i = 0; i < element_count; i++) { |
| 4804 TextElement& elm = (*elements())[i]; |
| 4805 elm.set_cp_offset(cp_offset); |
| 4806 cp_offset += elm.length(); |
| 4807 } |
| 4808 } |
| 4809 |
| 4810 |
| 4811 void Analysis::VisitText(TextNode* that) { |
| 4812 if (ignore_case_) { |
| 4813 that->MakeCaseIndependent(is_one_byte_); |
| 4814 } |
| 4815 EnsureAnalyzed(that->on_success()); |
| 4816 if (!has_failed()) { |
| 4817 that->CalculateOffsets(); |
| 4818 } |
| 4819 } |
| 4820 |
| 4821 |
| 4822 void Analysis::VisitAction(ActionNode* that) { |
| 4823 RegExpNode* target = that->on_success(); |
| 4824 EnsureAnalyzed(target); |
| 4825 if (!has_failed()) { |
| 4826 // If the next node is interested in what it follows then this node |
| 4827 // has to be interested too so it can pass the information on. |
| 4828 that->info()->AddFromFollowing(target->info()); |
| 4829 } |
| 4830 } |
| 4831 |
| 4832 |
| 4833 void Analysis::VisitChoice(ChoiceNode* that) { |
| 4834 NodeInfo* info = that->info(); |
| 4835 for (intptr_t i = 0; i < that->alternatives()->length(); i++) { |
| 4836 RegExpNode* node = (*that->alternatives())[i].node(); |
| 4837 EnsureAnalyzed(node); |
| 4838 if (has_failed()) return; |
| 4839 // Anything the following nodes need to know has to be known by |
| 4840 // this node also, so it can pass it on. |
| 4841 info->AddFromFollowing(node->info()); |
| 4842 } |
| 4843 } |
| 4844 |
| 4845 |
| 4846 void Analysis::VisitLoopChoice(LoopChoiceNode* that) { |
| 4847 NodeInfo* info = that->info(); |
| 4848 for (intptr_t i = 0; i < that->alternatives()->length(); i++) { |
| 4849 RegExpNode* node = (*that->alternatives())[i].node(); |
| 4850 if (node != that->loop_node()) { |
| 4851 EnsureAnalyzed(node); |
| 4852 if (has_failed()) return; |
| 4853 info->AddFromFollowing(node->info()); |
| 4854 } |
| 4855 } |
| 4856 // Check the loop last since it may need the value of this node |
| 4857 // to get a correct result. |
| 4858 EnsureAnalyzed(that->loop_node()); |
| 4859 if (!has_failed()) { |
| 4860 info->AddFromFollowing(that->loop_node()->info()); |
| 4861 } |
| 4862 } |
| 4863 |
| 4864 |
| 4865 void Analysis::VisitBackReference(BackReferenceNode* that) { |
| 4866 EnsureAnalyzed(that->on_success()); |
| 4867 } |
| 4868 |
| 4869 |
| 4870 void Analysis::VisitAssertion(AssertionNode* that) { |
| 4871 EnsureAnalyzed(that->on_success()); |
| 4872 } |
| 4873 |
| 4874 |
| 4875 void BackReferenceNode::FillInBMInfo(intptr_t offset, |
| 4876 intptr_t budget, |
| 4877 BoyerMooreLookahead* bm, |
| 4878 bool not_at_start) { |
| 4879 // Working out the set of characters that a backreference can match is too |
| 4880 // hard, so we just say that any character can match. |
| 4881 bm->SetRest(offset); |
| 4882 SaveBMInfo(bm, not_at_start, offset); |
| 4883 } |
| 4884 |
| 4885 |
| 4886 COMPILE_ASSERT(BoyerMoorePositionInfo::kMapSize == |
| 4887 RegExpMacroAssembler::kTableSize); |
| 4888 |
| 4889 |
| 4890 void ChoiceNode::FillInBMInfo(intptr_t offset, |
| 4891 intptr_t budget, |
| 4892 BoyerMooreLookahead* bm, |
| 4893 bool not_at_start) { |
| 4894 ZoneGrowableArray<GuardedAlternative>* alts = alternatives(); |
| 4895 budget = (budget - 1) / alts->length(); |
| 4896 for (intptr_t i = 0; i < alts->length(); i++) { |
| 4897 GuardedAlternative& alt = (*alts)[i]; |
| 4898 if (alt.guards() != NULL && alt.guards()->length() != 0) { |
| 4899 bm->SetRest(offset); // Give up trying to fill in info. |
| 4900 SaveBMInfo(bm, not_at_start, offset); |
| 4901 return; |
| 4902 } |
| 4903 alt.node()->FillInBMInfo(offset, budget, bm, not_at_start); |
| 4904 } |
| 4905 SaveBMInfo(bm, not_at_start, offset); |
| 4906 } |
| 4907 |
| 4908 |
| 4909 void TextNode::FillInBMInfo(intptr_t initial_offset, |
| 4910 intptr_t budget, |
| 4911 BoyerMooreLookahead* bm, |
| 4912 bool not_at_start) { |
| 4913 if (initial_offset >= bm->length()) return; |
| 4914 intptr_t offset = initial_offset; |
| 4915 intptr_t max_char = bm->max_char(); |
| 4916 for (intptr_t i = 0; i < elements()->length(); i++) { |
| 4917 if (offset >= bm->length()) { |
| 4918 if (initial_offset == 0) set_bm_info(not_at_start, bm); |
| 4919 return; |
| 4920 } |
| 4921 TextElement text = elements()->At(i); |
| 4922 if (text.text_type() == TextElement::ATOM) { |
| 4923 RegExpAtom* atom = text.atom(); |
| 4924 for (intptr_t j = 0; j < atom->length(); j++, offset++) { |
| 4925 if (offset >= bm->length()) { |
| 4926 if (initial_offset == 0) set_bm_info(not_at_start, bm); |
| 4927 return; |
| 4928 } |
| 4929 uint16_t character = atom->data()->At(j); |
| 4930 if (bm->compiler()->ignore_case()) { |
| 4931 int32_t chars[unibrow::Ecma262UnCanonicalize::kMaxWidth]; |
| 4932 intptr_t length = GetCaseIndependentLetters( |
| 4933 character, |
| 4934 bm->max_char() == Symbols::kMaxOneCharCodeSymbol, |
| 4935 chars); |
| 4936 for (intptr_t j = 0; j < length; j++) { |
| 4937 bm->Set(offset, chars[j]); |
| 4938 } |
| 4939 } else { |
| 4940 if (character <= max_char) bm->Set(offset, character); |
| 4941 } |
| 4942 } |
| 4943 } else { |
| 4944 ASSERT(text.text_type() == TextElement::CHAR_CLASS); |
| 4945 RegExpCharacterClass* char_class = text.char_class(); |
| 4946 ZoneGrowableArray<CharacterRange>* ranges = char_class->ranges(); |
| 4947 if (char_class->is_negated()) { |
| 4948 bm->SetAll(offset); |
| 4949 } else { |
| 4950 for (intptr_t k = 0; k < ranges->length(); k++) { |
| 4951 CharacterRange& range = (*ranges)[k]; |
| 4952 if (range.from() > max_char) continue; |
| 4953 intptr_t to = Utils::Minimum(max_char, |
| 4954 static_cast<intptr_t>(range.to())); |
| 4955 bm->SetInterval(offset, Interval(range.from(), to)); |
| 4956 } |
| 4957 } |
| 4958 offset++; |
| 4959 } |
| 4960 } |
| 4961 if (offset >= bm->length()) { |
| 4962 if (initial_offset == 0) set_bm_info(not_at_start, bm); |
| 4963 return; |
| 4964 } |
| 4965 on_success()->FillInBMInfo(offset, |
| 4966 budget - 1, |
| 4967 bm, |
| 4968 true); // Not at start after a text node. |
| 4969 if (initial_offset == 0) set_bm_info(not_at_start, bm); |
| 4970 } |
| 4971 |
| 4972 |
| 4973 RegExpEngine::CompilationResult RegExpEngine::Compile( |
| 4974 RegExpCompileData* data, |
| 4975 const ParsedFunction* parsed_function, |
| 4976 const ZoneGrowableArray<const ICData*>& ic_data_array) { |
| 4977 Isolate* isolate = Isolate::Current(); |
| 4978 |
| 4979 const Function& function = parsed_function->function(); |
| 4980 const intptr_t specialization_cid = function.regexp_cid(); |
| 4981 const bool is_one_byte = (specialization_cid == kOneByteStringCid || |
| 4982 specialization_cid == kExternalOneByteStringCid); |
| 4983 JSRegExp& regexp = JSRegExp::Handle(isolate, function.regexp()); |
| 4984 const String& pattern = String::Handle(isolate, regexp.pattern()); |
| 4985 |
| 4986 ASSERT(!regexp.IsNull()); |
| 4987 ASSERT(!pattern.IsNull()); |
| 4988 |
| 4989 const bool ignore_case = regexp.is_ignore_case(); |
| 4990 const bool is_global = regexp.is_global(); |
| 4991 |
| 4992 RegExpCompiler compiler(data->capture_count, ignore_case, specialization_cid); |
| 4993 |
| 4994 // TODO(zerny): Frequency sampling is currently disabled because of several |
| 4995 // issues. We do not want to store subject strings in the regexp object since |
| 4996 // they might be long and we should not prevent their garbage collection. |
| 4997 // Passing them to this function explicitly does not help, since we must |
| 4998 // generate exactly the same IR for both the unoptimizing and optimizing |
| 4999 // pipelines (otherwise it gets confused when i.e. deopt id's differ). |
| 5000 // An option would be to store sampling results in the regexp object, but |
| 5001 // I'm not sure the performance gains are relevant enough. |
| 5002 |
| 5003 // Wrap the body of the regexp in capture #0. |
| 5004 RegExpNode* captured_body = RegExpCapture::ToNode(data->tree, |
| 5005 0, |
| 5006 &compiler, |
| 5007 compiler.accept()); |
| 5008 |
| 5009 RegExpNode* node = captured_body; |
| 5010 bool is_end_anchored = data->tree->IsAnchoredAtEnd(); |
| 5011 bool is_start_anchored = data->tree->IsAnchoredAtStart(); |
| 5012 intptr_t max_length = data->tree->max_match(); |
| 5013 if (!is_start_anchored) { |
| 5014 // Add a .*? at the beginning, outside the body capture, unless |
| 5015 // this expression is anchored at the beginning. |
| 5016 RegExpNode* loop_node = |
| 5017 RegExpQuantifier::ToNode(0, |
| 5018 RegExpTree::kInfinity, |
| 5019 false, |
| 5020 new(isolate) RegExpCharacterClass('*'), |
| 5021 &compiler, |
| 5022 captured_body, |
| 5023 data->contains_anchor); |
| 5024 |
| 5025 if (data->contains_anchor) { |
| 5026 // Unroll loop once, to take care of the case that might start |
| 5027 // at the start of input. |
| 5028 ChoiceNode* first_step_node = new(isolate) ChoiceNode(2, isolate); |
| 5029 first_step_node->AddAlternative(GuardedAlternative(captured_body)); |
| 5030 first_step_node->AddAlternative(GuardedAlternative( |
| 5031 new(isolate) TextNode( |
| 5032 new(isolate) RegExpCharacterClass('*'), loop_node))); |
| 5033 node = first_step_node; |
| 5034 } else { |
| 5035 node = loop_node; |
| 5036 } |
| 5037 } |
| 5038 if (is_one_byte) { |
| 5039 node = node->FilterOneByte(RegExpCompiler::kMaxRecursion, ignore_case); |
| 5040 // Do it again to propagate the new nodes to places where they were not |
| 5041 // put because they had not been calculated yet. |
| 5042 if (node != NULL) { |
| 5043 node = node->FilterOneByte(RegExpCompiler::kMaxRecursion, ignore_case); |
| 5044 } |
| 5045 } |
| 5046 |
| 5047 if (node == NULL) node = new(isolate) EndNode(EndNode::BACKTRACK, isolate); |
| 5048 data->node = node; |
| 5049 Analysis analysis(ignore_case, is_one_byte); |
| 5050 analysis.EnsureAnalyzed(node); |
| 5051 if (analysis.has_failed()) { |
| 5052 const char* error_message = analysis.error_message(); |
| 5053 return CompilationResult(error_message); |
| 5054 } |
| 5055 |
| 5056 // Native regexp implementation. |
| 5057 |
| 5058 IRRegExpMacroAssembler* macro_assembler = |
| 5059 new(isolate) IRRegExpMacroAssembler(specialization_cid, |
| 5060 data->capture_count, |
| 5061 parsed_function, |
| 5062 ic_data_array, |
| 5063 isolate); |
| 5064 |
| 5065 // Inserted here, instead of in Assembler, because it depends on information |
| 5066 // in the AST that isn't replicated in the Node structure. |
| 5067 static const intptr_t kMaxBacksearchLimit = 1024; |
| 5068 if (is_end_anchored && |
| 5069 !is_start_anchored && |
| 5070 max_length < kMaxBacksearchLimit) { |
| 5071 macro_assembler->SetCurrentPositionFromEnd(max_length); |
| 5072 } |
| 5073 |
| 5074 if (is_global) { |
| 5075 macro_assembler->set_global_mode( |
| 5076 (data->tree->min_match() > 0) |
| 5077 ? RegExpMacroAssembler::GLOBAL_NO_ZERO_LENGTH_CHECK |
| 5078 : RegExpMacroAssembler::GLOBAL); |
| 5079 } |
| 5080 |
| 5081 RegExpEngine::CompilationResult result = |
| 5082 compiler.Assemble(macro_assembler, |
| 5083 node, |
| 5084 data->capture_count, |
| 5085 pattern); |
| 5086 |
| 5087 if (FLAG_trace_irregexp) { |
| 5088 macro_assembler->PrintBlocks(); |
| 5089 } |
| 5090 |
| 5091 return result; |
| 5092 } |
| 5093 |
| 5094 |
| 5095 static void CreateSpecializedFunction(Isolate* isolate, |
| 5096 const JSRegExp& regexp, |
| 5097 intptr_t specialization_cid, |
| 5098 const Object& owner) { |
| 5099 const intptr_t kParamCount = RegExpMacroAssembler::kParamCount; |
| 5100 |
| 5101 Function& fn = Function::Handle(isolate, |
| 5102 Function::New(String::Handle(isolate, Symbols::New("RegExp")), |
| 5103 RawFunction::kIrregexpFunction, |
| 5104 true, // Static. |
| 5105 false, // Not const. |
| 5106 false, // Not abstract. |
| 5107 false, // Not external. |
| 5108 false, // Not native. |
| 5109 owner, |
| 5110 0)); // Requires a non-negative token position. |
| 5111 |
| 5112 // TODO(zerny): Share these arrays between all irregexp functions. |
| 5113 fn.set_num_fixed_parameters(kParamCount); |
| 5114 fn.set_parameter_types(Array::Handle(isolate, Array::New(kParamCount, |
| 5115 Heap::kOld))); |
| 5116 fn.set_parameter_names(Array::Handle(isolate, Array::New(kParamCount, |
| 5117 Heap::kOld))); |
| 5118 fn.SetParameterTypeAt(0, Type::Handle(isolate, Type::DynamicType())); |
| 5119 fn.SetParameterNameAt(0, Symbols::string_param_()); |
| 5120 fn.SetParameterTypeAt(1, Type::Handle(isolate, Type::DynamicType())); |
| 5121 fn.SetParameterNameAt(1, Symbols::start_index_param_()); |
| 5122 fn.set_result_type(Type::Handle(isolate, Type::ArrayType())); |
| 5123 |
| 5124 // Cache the result. |
| 5125 regexp.set_function(specialization_cid, fn); |
| 5126 |
| 5127 fn.set_regexp(regexp); |
| 5128 fn.set_regexp_cid(specialization_cid); |
| 5129 |
| 5130 // The function is compiled lazily during the first call. |
| 5131 } |
| 5132 |
| 5133 |
| 5134 RawJSRegExp* RegExpEngine::CreateJSRegExp(Isolate* isolate, |
| 5135 const String& pattern, |
| 5136 bool multi_line, |
| 5137 bool ignore_case) { |
| 5138 const JSRegExp& regexp = JSRegExp::Handle(JSRegExp::New(0)); |
| 5139 |
| 5140 regexp.set_pattern(pattern); |
| 5141 |
| 5142 if (multi_line) { |
| 5143 regexp.set_is_multi_line(); |
| 5144 } |
| 5145 if (ignore_case) { |
| 5146 regexp.set_is_ignore_case(); |
| 5147 } |
| 5148 |
| 5149 // TODO(zerny): We might want to use normal string searching algorithms |
| 5150 // for simple patterns. |
| 5151 regexp.set_is_complex(); |
| 5152 regexp.set_is_global(); // All dart regexps are global. |
| 5153 |
| 5154 const Library& lib = Library::Handle(isolate, Library::CoreLibrary()); |
| 5155 const Class& owner = Class::Handle(isolate, |
| 5156 lib.LookupClass(String::Handle(isolate, Symbols::New("RegExp")))); |
| 5157 |
| 5158 CreateSpecializedFunction(isolate, regexp, kOneByteStringCid, owner); |
| 5159 CreateSpecializedFunction(isolate, regexp, kTwoByteStringCid, owner); |
| 5160 CreateSpecializedFunction(isolate, regexp, kExternalOneByteStringCid, owner); |
| 5161 CreateSpecializedFunction(isolate, regexp, kExternalTwoByteStringCid, owner); |
| 5162 |
| 5163 return regexp.raw(); |
| 5164 } |
| 5165 |
| 5166 |
| 5167 } // namespace dart |
OLD | NEW |