1 | /* |
2 | * Copyright (C) 2009 Apple Inc. All rights reserved. |
3 | * |
4 | * Redistribution and use in source and binary forms, with or without |
5 | * modification, are permitted provided that the following conditions |
6 | * are met: |
7 | * 1. Redistributions of source code must retain the above copyright |
8 | * notice, this list of conditions and the following disclaimer. |
9 | * 2. Redistributions in binary form must reproduce the above copyright |
10 | * notice, this list of conditions and the following disclaimer in the |
11 | * documentation and/or other materials provided with the distribution. |
12 | * |
13 | * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY |
14 | * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
15 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR |
16 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR |
17 | * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, |
18 | * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
19 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR |
20 | * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY |
21 | * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
22 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
23 | * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
24 | */ |
25 | |
26 | #include "config.h" |
27 | #include "RegexJIT.h" |
28 | |
29 | #include "ASCIICType.h" |
30 | #include "JSGlobalData.h" |
31 | #include "LinkBuffer.h" |
32 | #include "MacroAssembler.h" |
33 | #include "RegexCompiler.h" |
34 | |
35 | #include "pcre.h" // temporary, remove when fallback is removed. |
36 | |
37 | #if ENABLE(YARR_JIT) |
38 | |
39 | using namespace WTF; |
40 | |
41 | namespace JSC { namespace Yarr { |
42 | |
43 | |
44 | class RegexGenerator : private MacroAssembler { |
45 | friend void jitCompileRegex(JSGlobalData* globalData, RegexCodeBlock& jitObject, const UString& pattern, unsigned& numSubpatterns, const char*& error, bool ignoreCase, bool multiline); |
46 | |
47 | #if CPU(ARM) |
48 | static const RegisterID input = ARMRegisters::r0; |
49 | static const RegisterID index = ARMRegisters::r1; |
50 | static const RegisterID length = ARMRegisters::r2; |
51 | static const RegisterID output = ARMRegisters::r4; |
52 | |
53 | static const RegisterID regT0 = ARMRegisters::r5; |
54 | static const RegisterID regT1 = ARMRegisters::r6; |
55 | |
56 | static const RegisterID returnRegister = ARMRegisters::r0; |
57 | #elif CPU(X86) |
58 | static const RegisterID input = X86Registers::eax; |
59 | static const RegisterID index = X86Registers::edx; |
60 | static const RegisterID length = X86Registers::ecx; |
61 | static const RegisterID output = X86Registers::edi; |
62 | |
63 | static const RegisterID regT0 = X86Registers::ebx; |
64 | static const RegisterID regT1 = X86Registers::esi; |
65 | |
66 | static const RegisterID returnRegister = X86Registers::eax; |
67 | #elif CPU(X86_64) |
68 | static const RegisterID input = X86Registers::edi; |
69 | static const RegisterID index = X86Registers::esi; |
70 | static const RegisterID length = X86Registers::edx; |
71 | static const RegisterID output = X86Registers::ecx; |
72 | |
73 | static const RegisterID regT0 = X86Registers::eax; |
74 | static const RegisterID regT1 = X86Registers::ebx; |
75 | |
76 | static const RegisterID returnRegister = X86Registers::eax; |
77 | #endif |
78 | |
79 | void optimizeAlternative(PatternAlternative* alternative) |
80 | { |
81 | if (!alternative->m_terms.size()) |
82 | return; |
83 | |
84 | for (unsigned i = 0; i < alternative->m_terms.size() - 1; ++i) { |
85 | PatternTerm& term = alternative->m_terms[i]; |
86 | PatternTerm& nextTerm = alternative->m_terms[i + 1]; |
87 | |
88 | if ((term.type == PatternTerm::TypeCharacterClass) |
89 | && (term.quantityType == QuantifierFixedCount) |
90 | && (nextTerm.type == PatternTerm::TypePatternCharacter) |
91 | && (nextTerm.quantityType == QuantifierFixedCount)) { |
92 | PatternTerm termCopy = term; |
93 | alternative->m_terms[i] = nextTerm; |
94 | alternative->m_terms[i + 1] = termCopy; |
95 | } |
96 | } |
97 | } |
98 | |
99 | void matchCharacterClassRange(RegisterID character, JumpList& failures, JumpList& matchDest, const CharacterRange* ranges, unsigned count, unsigned* matchIndex, const UChar* matches, unsigned matchCount) |
100 | { |
101 | do { |
102 | // pick which range we're going to generate |
103 | int which = count >> 1; |
104 | char lo = ranges[which].begin; |
105 | char hi = ranges[which].end; |
106 | |
107 | // check if there are any ranges or matches below lo. If not, just jl to failure - |
108 | // if there is anything else to check, check that first, if it falls through jmp to failure. |
109 | if ((*matchIndex < matchCount) && (matches[*matchIndex] < lo)) { |
110 | Jump loOrAbove = branch32(cond: GreaterThanOrEqual, left: character, right: Imm32((unsigned short)lo)); |
111 | |
112 | // generate code for all ranges before this one |
113 | if (which) |
114 | matchCharacterClassRange(character, failures, matchDest, ranges, count: which, matchIndex, matches, matchCount); |
115 | |
116 | while ((*matchIndex < matchCount) && (matches[*matchIndex] < lo)) { |
117 | matchDest.append(jump: branch32(cond: Equal, left: character, right: Imm32((unsigned short)matches[*matchIndex]))); |
118 | ++*matchIndex; |
119 | } |
120 | failures.append(jump: jump()); |
121 | |
122 | loOrAbove.link(masm: this); |
123 | } else if (which) { |
124 | Jump loOrAbove = branch32(cond: GreaterThanOrEqual, left: character, right: Imm32((unsigned short)lo)); |
125 | |
126 | matchCharacterClassRange(character, failures, matchDest, ranges, count: which, matchIndex, matches, matchCount); |
127 | failures.append(jump: jump()); |
128 | |
129 | loOrAbove.link(masm: this); |
130 | } else |
131 | failures.append(jump: branch32(cond: LessThan, left: character, right: Imm32((unsigned short)lo))); |
132 | |
133 | while ((*matchIndex < matchCount) && (matches[*matchIndex] <= hi)) |
134 | ++*matchIndex; |
135 | |
136 | matchDest.append(jump: branch32(cond: LessThanOrEqual, left: character, right: Imm32((unsigned short)hi))); |
137 | // fall through to here, the value is above hi. |
138 | |
139 | // shuffle along & loop around if there are any more matches to handle. |
140 | unsigned next = which + 1; |
141 | ranges += next; |
142 | count -= next; |
143 | } while (count); |
144 | } |
145 | |
146 | void matchCharacterClass(RegisterID character, JumpList& matchDest, const CharacterClass* charClass) |
147 | { |
148 | Jump unicodeFail; |
149 | if (charClass->m_matchesUnicode.size() || charClass->m_rangesUnicode.size()) { |
150 | Jump isAscii = branch32(cond: LessThanOrEqual, left: character, right: Imm32(0x7f)); |
151 | |
152 | if (charClass->m_matchesUnicode.size()) { |
153 | for (unsigned i = 0; i < charClass->m_matchesUnicode.size(); ++i) { |
154 | UChar ch = charClass->m_matchesUnicode[i]; |
155 | matchDest.append(jump: branch32(cond: Equal, left: character, right: Imm32(ch))); |
156 | } |
157 | } |
158 | |
159 | if (charClass->m_rangesUnicode.size()) { |
160 | for (unsigned i = 0; i < charClass->m_rangesUnicode.size(); ++i) { |
161 | UChar lo = charClass->m_rangesUnicode[i].begin; |
162 | UChar hi = charClass->m_rangesUnicode[i].end; |
163 | |
164 | Jump below = branch32(cond: LessThan, left: character, right: Imm32(lo)); |
165 | matchDest.append(jump: branch32(cond: LessThanOrEqual, left: character, right: Imm32(hi))); |
166 | below.link(masm: this); |
167 | } |
168 | } |
169 | |
170 | unicodeFail = jump(); |
171 | isAscii.link(masm: this); |
172 | } |
173 | |
174 | if (charClass->m_ranges.size()) { |
175 | unsigned matchIndex = 0; |
176 | JumpList failures; |
177 | matchCharacterClassRange(character, failures, matchDest, ranges: charClass->m_ranges.begin(), count: charClass->m_ranges.size(), matchIndex: &matchIndex, matches: charClass->m_matches.begin(), matchCount: charClass->m_matches.size()); |
178 | while (matchIndex < charClass->m_matches.size()) |
179 | matchDest.append(jump: branch32(cond: Equal, left: character, right: Imm32((unsigned short)charClass->m_matches[matchIndex++]))); |
180 | |
181 | failures.link(masm: this); |
182 | } else if (charClass->m_matches.size()) { |
183 | // optimization: gather 'a','A' etc back together, can mask & test once. |
184 | Vector<char> matchesAZaz; |
185 | |
186 | for (unsigned i = 0; i < charClass->m_matches.size(); ++i) { |
187 | char ch = charClass->m_matches[i]; |
188 | if (m_pattern.m_ignoreCase) { |
189 | if (isASCIILower(c: ch)) { |
190 | matchesAZaz.append(val: ch); |
191 | continue; |
192 | } |
193 | if (isASCIIUpper(c: ch)) |
194 | continue; |
195 | } |
196 | matchDest.append(jump: branch32(cond: Equal, left: character, right: Imm32((unsigned short)ch))); |
197 | } |
198 | |
199 | if (unsigned countAZaz = matchesAZaz.size()) { |
200 | or32(imm: Imm32(32), dest: character); |
201 | for (unsigned i = 0; i < countAZaz; ++i) |
202 | matchDest.append(jump: branch32(cond: Equal, left: character, right: Imm32(matchesAZaz[i]))); |
203 | } |
204 | } |
205 | |
206 | if (charClass->m_matchesUnicode.size() || charClass->m_rangesUnicode.size()) |
207 | unicodeFail.link(masm: this); |
208 | } |
209 | |
210 | // Jumps if input not available; will have (incorrectly) incremented already! |
211 | Jump jumpIfNoAvailableInput(unsigned countToCheck) |
212 | { |
213 | add32(imm: Imm32(countToCheck), dest: index); |
214 | return branch32(cond: Above, left: index, right: length); |
215 | } |
216 | |
217 | Jump jumpIfAvailableInput(unsigned countToCheck) |
218 | { |
219 | add32(imm: Imm32(countToCheck), dest: index); |
220 | return branch32(cond: BelowOrEqual, left: index, right: length); |
221 | } |
222 | |
223 | Jump checkInput() |
224 | { |
225 | return branch32(cond: BelowOrEqual, left: index, right: length); |
226 | } |
227 | |
228 | Jump atEndOfInput() |
229 | { |
230 | return branch32(cond: Equal, left: index, right: length); |
231 | } |
232 | |
233 | Jump notAtEndOfInput() |
234 | { |
235 | return branch32(cond: NotEqual, left: index, right: length); |
236 | } |
237 | |
238 | Jump jumpIfCharEquals(UChar ch, int inputPosition) |
239 | { |
240 | return branch16(cond: Equal, left: BaseIndex(input, index, TimesTwo, inputPosition * sizeof(UChar)), right: Imm32(ch)); |
241 | } |
242 | |
243 | Jump jumpIfCharNotEquals(UChar ch, int inputPosition) |
244 | { |
245 | return branch16(cond: NotEqual, left: BaseIndex(input, index, TimesTwo, inputPosition * sizeof(UChar)), right: Imm32(ch)); |
246 | } |
247 | |
248 | void readCharacter(int inputPosition, RegisterID reg) |
249 | { |
250 | load16(address: BaseIndex(input, index, TimesTwo, inputPosition * sizeof(UChar)), dest: reg); |
251 | } |
252 | |
253 | void storeToFrame(RegisterID reg, unsigned frameLocation) |
254 | { |
255 | poke(src: reg, index: frameLocation); |
256 | } |
257 | |
258 | void storeToFrame(Imm32 imm, unsigned frameLocation) |
259 | { |
260 | poke(value: imm, index: frameLocation); |
261 | } |
262 | |
263 | DataLabelPtr storeToFrameWithPatch(unsigned frameLocation) |
264 | { |
265 | return storePtrWithPatch(initialValue: ImmPtr(0), address: Address(stackPointerRegister, frameLocation * sizeof(void*))); |
266 | } |
267 | |
268 | void loadFromFrame(unsigned frameLocation, RegisterID reg) |
269 | { |
270 | peek(dest: reg, index: frameLocation); |
271 | } |
272 | |
273 | void loadFromFrameAndJump(unsigned frameLocation) |
274 | { |
275 | jump(address: Address(stackPointerRegister, frameLocation * sizeof(void*))); |
276 | } |
277 | |
278 | struct AlternativeBacktrackRecord { |
279 | DataLabelPtr dataLabel; |
280 | Label backtrackLocation; |
281 | |
282 | AlternativeBacktrackRecord(DataLabelPtr dataLabel, Label backtrackLocation) |
283 | : dataLabel(dataLabel) |
284 | , backtrackLocation(backtrackLocation) |
285 | { |
286 | } |
287 | }; |
288 | |
289 | struct TermGenerationState { |
290 | TermGenerationState(PatternDisjunction* disjunction, unsigned checkedTotal) |
291 | : disjunction(disjunction) |
292 | , checkedTotal(checkedTotal) |
293 | { |
294 | } |
295 | |
296 | void resetAlternative() |
297 | { |
298 | isBackTrackGenerated = false; |
299 | alt = 0; |
300 | } |
301 | bool alternativeValid() |
302 | { |
303 | return alt < disjunction->m_alternatives.size(); |
304 | } |
305 | void nextAlternative() |
306 | { |
307 | ++alt; |
308 | } |
309 | PatternAlternative* alternative() |
310 | { |
311 | return disjunction->m_alternatives[alt]; |
312 | } |
313 | |
314 | void resetTerm() |
315 | { |
316 | ASSERT(alternativeValid()); |
317 | t = 0; |
318 | } |
319 | bool termValid() |
320 | { |
321 | ASSERT(alternativeValid()); |
322 | return t < alternative()->m_terms.size(); |
323 | } |
324 | void nextTerm() |
325 | { |
326 | ASSERT(alternativeValid()); |
327 | ++t; |
328 | } |
329 | PatternTerm& term() |
330 | { |
331 | ASSERT(alternativeValid()); |
332 | return alternative()->m_terms[t]; |
333 | } |
334 | |
335 | PatternTerm& lookaheadTerm() |
336 | { |
337 | ASSERT(alternativeValid()); |
338 | ASSERT((t + 1) < alternative()->m_terms.size()); |
339 | return alternative()->m_terms[t + 1]; |
340 | } |
341 | bool isSinglePatternCharacterLookaheadTerm() |
342 | { |
343 | ASSERT(alternativeValid()); |
344 | return ((t + 1) < alternative()->m_terms.size()) |
345 | && (lookaheadTerm().type == PatternTerm::TypePatternCharacter) |
346 | && (lookaheadTerm().quantityType == QuantifierFixedCount) |
347 | && (lookaheadTerm().quantityCount == 1); |
348 | } |
349 | |
350 | int inputOffset() |
351 | { |
352 | return term().inputPosition - checkedTotal; |
353 | } |
354 | |
355 | void jumpToBacktrack(Jump jump, MacroAssembler* masm) |
356 | { |
357 | if (isBackTrackGenerated) |
358 | jump.linkTo(label: backtrackLabel, masm); |
359 | else |
360 | backTrackJumps.append(jump); |
361 | } |
362 | void jumpToBacktrack(JumpList& jumps, MacroAssembler* masm) |
363 | { |
364 | if (isBackTrackGenerated) |
365 | jumps.linkTo(label: backtrackLabel, masm); |
366 | else |
367 | backTrackJumps.append(other&: jumps); |
368 | } |
369 | bool plantJumpToBacktrackIfExists(MacroAssembler* masm) |
370 | { |
371 | if (isBackTrackGenerated) { |
372 | masm->jump(target: backtrackLabel); |
373 | return true; |
374 | } |
375 | return false; |
376 | } |
377 | void addBacktrackJump(Jump jump) |
378 | { |
379 | backTrackJumps.append(jump); |
380 | } |
381 | void setBacktrackGenerated(Label label) |
382 | { |
383 | isBackTrackGenerated = true; |
384 | backtrackLabel = label; |
385 | } |
386 | void linkAlternativeBacktracks(MacroAssembler* masm) |
387 | { |
388 | isBackTrackGenerated = false; |
389 | backTrackJumps.link(masm); |
390 | } |
391 | void linkAlternativeBacktracksTo(Label label, MacroAssembler* masm) |
392 | { |
393 | isBackTrackGenerated = false; |
394 | backTrackJumps.linkTo(label, masm); |
395 | } |
396 | void propagateBacktrackingFrom(TermGenerationState& nestedParenthesesState, MacroAssembler* masm) |
397 | { |
398 | jumpToBacktrack(jumps&: nestedParenthesesState.backTrackJumps, masm); |
399 | if (nestedParenthesesState.isBackTrackGenerated) |
400 | setBacktrackGenerated(nestedParenthesesState.backtrackLabel); |
401 | } |
402 | |
403 | PatternDisjunction* disjunction; |
404 | int checkedTotal; |
405 | private: |
406 | unsigned alt; |
407 | unsigned t; |
408 | JumpList backTrackJumps; |
409 | Label backtrackLabel; |
410 | bool isBackTrackGenerated; |
411 | }; |
412 | |
413 | void generateAssertionBOL(TermGenerationState& state) |
414 | { |
415 | PatternTerm& term = state.term(); |
416 | |
417 | if (m_pattern.m_multiline) { |
418 | const RegisterID character = regT0; |
419 | |
420 | JumpList matchDest; |
421 | if (!term.inputPosition) |
422 | matchDest.append(jump: branch32(cond: Equal, left: index, right: Imm32(state.checkedTotal))); |
423 | |
424 | readCharacter(inputPosition: state.inputOffset() - 1, reg: character); |
425 | matchCharacterClass(character, matchDest, charClass: m_pattern.newlineCharacterClass()); |
426 | state.jumpToBacktrack(jump: jump(), masm: this); |
427 | |
428 | matchDest.link(masm: this); |
429 | } else { |
430 | // Erk, really should poison out these alternatives early. :-/ |
431 | if (term.inputPosition) |
432 | state.jumpToBacktrack(jump: jump(), masm: this); |
433 | else |
434 | state.jumpToBacktrack(jump: branch32(cond: NotEqual, left: index, right: Imm32(state.checkedTotal)), masm: this); |
435 | } |
436 | } |
437 | |
438 | void generateAssertionEOL(TermGenerationState& state) |
439 | { |
440 | PatternTerm& term = state.term(); |
441 | |
442 | if (m_pattern.m_multiline) { |
443 | const RegisterID character = regT0; |
444 | |
445 | JumpList matchDest; |
446 | if (term.inputPosition == state.checkedTotal) |
447 | matchDest.append(jump: atEndOfInput()); |
448 | |
449 | readCharacter(inputPosition: state.inputOffset(), reg: character); |
450 | matchCharacterClass(character, matchDest, charClass: m_pattern.newlineCharacterClass()); |
451 | state.jumpToBacktrack(jump: jump(), masm: this); |
452 | |
453 | matchDest.link(masm: this); |
454 | } else { |
455 | if (term.inputPosition == state.checkedTotal) |
456 | state.jumpToBacktrack(jump: notAtEndOfInput(), masm: this); |
457 | // Erk, really should poison out these alternatives early. :-/ |
458 | else |
459 | state.jumpToBacktrack(jump: jump(), masm: this); |
460 | } |
461 | } |
462 | |
463 | // Also falls though on nextIsNotWordChar. |
464 | void matchAssertionWordchar(TermGenerationState& state, JumpList& nextIsWordChar, JumpList& nextIsNotWordChar) |
465 | { |
466 | const RegisterID character = regT0; |
467 | PatternTerm& term = state.term(); |
468 | |
469 | if (term.inputPosition == state.checkedTotal) |
470 | nextIsNotWordChar.append(jump: atEndOfInput()); |
471 | |
472 | readCharacter(inputPosition: state.inputOffset(), reg: character); |
473 | matchCharacterClass(character, matchDest&: nextIsWordChar, charClass: m_pattern.wordcharCharacterClass()); |
474 | } |
475 | |
476 | void generateAssertionWordBoundary(TermGenerationState& state) |
477 | { |
478 | const RegisterID character = regT0; |
479 | PatternTerm& term = state.term(); |
480 | |
481 | Jump atBegin; |
482 | JumpList matchDest; |
483 | if (!term.inputPosition) |
484 | atBegin = branch32(cond: Equal, left: index, right: Imm32(state.checkedTotal)); |
485 | readCharacter(inputPosition: state.inputOffset() - 1, reg: character); |
486 | matchCharacterClass(character, matchDest, charClass: m_pattern.wordcharCharacterClass()); |
487 | if (!term.inputPosition) |
488 | atBegin.link(masm: this); |
489 | |
490 | // We fall through to here if the last character was not a wordchar. |
491 | JumpList nonWordCharThenWordChar; |
492 | JumpList nonWordCharThenNonWordChar; |
493 | if (term.invertOrCapture) { |
494 | matchAssertionWordchar(state, nextIsWordChar&: nonWordCharThenNonWordChar, nextIsNotWordChar&: nonWordCharThenWordChar); |
495 | nonWordCharThenWordChar.append(jump: jump()); |
496 | } else { |
497 | matchAssertionWordchar(state, nextIsWordChar&: nonWordCharThenWordChar, nextIsNotWordChar&: nonWordCharThenNonWordChar); |
498 | nonWordCharThenNonWordChar.append(jump: jump()); |
499 | } |
500 | state.jumpToBacktrack(jumps&: nonWordCharThenNonWordChar, masm: this); |
501 | |
502 | // We jump here if the last character was a wordchar. |
503 | matchDest.link(masm: this); |
504 | JumpList wordCharThenWordChar; |
505 | JumpList wordCharThenNonWordChar; |
506 | if (term.invertOrCapture) { |
507 | matchAssertionWordchar(state, nextIsWordChar&: wordCharThenNonWordChar, nextIsNotWordChar&: wordCharThenWordChar); |
508 | wordCharThenWordChar.append(jump: jump()); |
509 | } else { |
510 | matchAssertionWordchar(state, nextIsWordChar&: wordCharThenWordChar, nextIsNotWordChar&: wordCharThenNonWordChar); |
511 | // This can fall-though! |
512 | } |
513 | |
514 | state.jumpToBacktrack(jumps&: wordCharThenWordChar, masm: this); |
515 | |
516 | nonWordCharThenWordChar.link(masm: this); |
517 | wordCharThenNonWordChar.link(masm: this); |
518 | } |
519 | |
520 | void generatePatternCharacterSingle(TermGenerationState& state) |
521 | { |
522 | const RegisterID character = regT0; |
523 | UChar ch = state.term().patternCharacter; |
524 | |
525 | if (m_pattern.m_ignoreCase && isASCIIAlpha(c: ch)) { |
526 | readCharacter(inputPosition: state.inputOffset(), reg: character); |
527 | or32(imm: Imm32(32), dest: character); |
528 | state.jumpToBacktrack(jump: branch32(cond: NotEqual, left: character, right: Imm32(Unicode::toLower(ch))), masm: this); |
529 | } else { |
530 | ASSERT(!m_pattern.m_ignoreCase || (Unicode::toLower(ch) == Unicode::toUpper(ch))); |
531 | state.jumpToBacktrack(jump: jumpIfCharNotEquals(ch, inputPosition: state.inputOffset()), masm: this); |
532 | } |
533 | } |
534 | |
535 | void generatePatternCharacterPair(TermGenerationState& state) |
536 | { |
537 | const RegisterID character = regT0; |
538 | UChar ch1 = state.term().patternCharacter; |
539 | UChar ch2 = state.lookaheadTerm().patternCharacter; |
540 | |
541 | int mask = 0; |
542 | int chPair = ch1 | (ch2 << 16); |
543 | |
544 | if (m_pattern.m_ignoreCase) { |
545 | if (isASCIIAlpha(c: ch1)) |
546 | mask |= 32; |
547 | if (isASCIIAlpha(c: ch2)) |
548 | mask |= 32 << 16; |
549 | } |
550 | |
551 | if (mask) { |
552 | load32WithUnalignedHalfWords(address: BaseIndex(input, index, TimesTwo, state.inputOffset() * sizeof(UChar)), dest: character); |
553 | or32(imm: Imm32(mask), dest: character); |
554 | state.jumpToBacktrack(jump: branch32(cond: NotEqual, left: character, right: Imm32(chPair | mask)), masm: this); |
555 | } else |
556 | state.jumpToBacktrack(jump: branch32WithUnalignedHalfWords(cond: NotEqual, left: BaseIndex(input, index, TimesTwo, state.inputOffset() * sizeof(UChar)), right: Imm32(chPair)), masm: this); |
557 | } |
558 | |
559 | void generatePatternCharacterFixed(TermGenerationState& state) |
560 | { |
561 | const RegisterID character = regT0; |
562 | const RegisterID countRegister = regT1; |
563 | PatternTerm& term = state.term(); |
564 | UChar ch = term.patternCharacter; |
565 | |
566 | move(src: index, dest: countRegister); |
567 | sub32(imm: Imm32(term.quantityCount), dest: countRegister); |
568 | |
569 | Label loop(this); |
570 | if (m_pattern.m_ignoreCase && isASCIIAlpha(c: ch)) { |
571 | load16(address: BaseIndex(input, countRegister, TimesTwo, (state.inputOffset() + term.quantityCount) * sizeof(UChar)), dest: character); |
572 | or32(imm: Imm32(32), dest: character); |
573 | state.jumpToBacktrack(jump: branch32(cond: NotEqual, left: character, right: Imm32(Unicode::toLower(ch))), masm: this); |
574 | } else { |
575 | ASSERT(!m_pattern.m_ignoreCase || (Unicode::toLower(ch) == Unicode::toUpper(ch))); |
576 | state.jumpToBacktrack(jump: branch16(cond: NotEqual, left: BaseIndex(input, countRegister, TimesTwo, (state.inputOffset() + term.quantityCount) * sizeof(UChar)), right: Imm32(ch)), masm: this); |
577 | } |
578 | add32(imm: Imm32(1), dest: countRegister); |
579 | branch32(cond: NotEqual, left: countRegister, right: index).linkTo(label: loop, masm: this); |
580 | } |
581 | |
582 | void generatePatternCharacterGreedy(TermGenerationState& state) |
583 | { |
584 | const RegisterID character = regT0; |
585 | const RegisterID countRegister = regT1; |
586 | PatternTerm& term = state.term(); |
587 | UChar ch = term.patternCharacter; |
588 | |
589 | move(imm: Imm32(0), dest: countRegister); |
590 | |
591 | JumpList failures; |
592 | Label loop(this); |
593 | failures.append(jump: atEndOfInput()); |
594 | if (m_pattern.m_ignoreCase && isASCIIAlpha(c: ch)) { |
595 | readCharacter(inputPosition: state.inputOffset(), reg: character); |
596 | or32(imm: Imm32(32), dest: character); |
597 | failures.append(jump: branch32(cond: NotEqual, left: character, right: Imm32(Unicode::toLower(ch)))); |
598 | } else { |
599 | ASSERT(!m_pattern.m_ignoreCase || (Unicode::toLower(ch) == Unicode::toUpper(ch))); |
600 | failures.append(jump: jumpIfCharNotEquals(ch, inputPosition: state.inputOffset())); |
601 | } |
602 | add32(imm: Imm32(1), dest: countRegister); |
603 | add32(imm: Imm32(1), dest: index); |
604 | branch32(cond: NotEqual, left: countRegister, right: Imm32(term.quantityCount)).linkTo(label: loop, masm: this); |
605 | failures.append(jump: jump()); |
606 | |
607 | Label backtrackBegin(this); |
608 | loadFromFrame(frameLocation: term.frameLocation, reg: countRegister); |
609 | state.jumpToBacktrack(jump: branchTest32(cond: Zero, reg: countRegister), masm: this); |
610 | sub32(imm: Imm32(1), dest: countRegister); |
611 | sub32(imm: Imm32(1), dest: index); |
612 | |
613 | failures.link(masm: this); |
614 | |
615 | storeToFrame(reg: countRegister, frameLocation: term.frameLocation); |
616 | |
617 | state.setBacktrackGenerated(backtrackBegin); |
618 | } |
619 | |
620 | void generatePatternCharacterNonGreedy(TermGenerationState& state) |
621 | { |
622 | const RegisterID character = regT0; |
623 | const RegisterID countRegister = regT1; |
624 | PatternTerm& term = state.term(); |
625 | UChar ch = term.patternCharacter; |
626 | |
627 | move(imm: Imm32(0), dest: countRegister); |
628 | |
629 | Jump firstTimeDoNothing = jump(); |
630 | |
631 | Label hardFail(this); |
632 | sub32(src: countRegister, dest: index); |
633 | state.jumpToBacktrack(jump: jump(), masm: this); |
634 | |
635 | Label backtrackBegin(this); |
636 | loadFromFrame(frameLocation: term.frameLocation, reg: countRegister); |
637 | |
638 | atEndOfInput().linkTo(label: hardFail, masm: this); |
639 | branch32(cond: Equal, op1: countRegister, imm: Imm32(term.quantityCount), target: hardFail); |
640 | if (m_pattern.m_ignoreCase && isASCIIAlpha(c: ch)) { |
641 | readCharacter(inputPosition: state.inputOffset(), reg: character); |
642 | or32(imm: Imm32(32), dest: character); |
643 | branch32(cond: NotEqual, left: character, right: Imm32(Unicode::toLower(ch))).linkTo(label: hardFail, masm: this); |
644 | } else { |
645 | ASSERT(!m_pattern.m_ignoreCase || (Unicode::toLower(ch) == Unicode::toUpper(ch))); |
646 | jumpIfCharNotEquals(ch, inputPosition: state.inputOffset()).linkTo(label: hardFail, masm: this); |
647 | } |
648 | |
649 | add32(imm: Imm32(1), dest: countRegister); |
650 | add32(imm: Imm32(1), dest: index); |
651 | |
652 | firstTimeDoNothing.link(masm: this); |
653 | storeToFrame(reg: countRegister, frameLocation: term.frameLocation); |
654 | |
655 | state.setBacktrackGenerated(backtrackBegin); |
656 | } |
657 | |
658 | void generateCharacterClassSingle(TermGenerationState& state) |
659 | { |
660 | const RegisterID character = regT0; |
661 | PatternTerm& term = state.term(); |
662 | |
663 | JumpList matchDest; |
664 | readCharacter(inputPosition: state.inputOffset(), reg: character); |
665 | matchCharacterClass(character, matchDest, charClass: term.characterClass); |
666 | |
667 | if (term.invertOrCapture) |
668 | state.jumpToBacktrack(jumps&: matchDest, masm: this); |
669 | else { |
670 | state.jumpToBacktrack(jump: jump(), masm: this); |
671 | matchDest.link(masm: this); |
672 | } |
673 | } |
674 | |
675 | void generateCharacterClassFixed(TermGenerationState& state) |
676 | { |
677 | const RegisterID character = regT0; |
678 | const RegisterID countRegister = regT1; |
679 | PatternTerm& term = state.term(); |
680 | |
681 | move(src: index, dest: countRegister); |
682 | sub32(imm: Imm32(term.quantityCount), dest: countRegister); |
683 | |
684 | Label loop(this); |
685 | JumpList matchDest; |
686 | load16(address: BaseIndex(input, countRegister, TimesTwo, (state.inputOffset() + term.quantityCount) * sizeof(UChar)), dest: character); |
687 | matchCharacterClass(character, matchDest, charClass: term.characterClass); |
688 | |
689 | if (term.invertOrCapture) |
690 | state.jumpToBacktrack(jumps&: matchDest, masm: this); |
691 | else { |
692 | state.jumpToBacktrack(jump: jump(), masm: this); |
693 | matchDest.link(masm: this); |
694 | } |
695 | |
696 | add32(imm: Imm32(1), dest: countRegister); |
697 | branch32(cond: NotEqual, left: countRegister, right: index).linkTo(label: loop, masm: this); |
698 | } |
699 | |
700 | void generateCharacterClassGreedy(TermGenerationState& state) |
701 | { |
702 | const RegisterID character = regT0; |
703 | const RegisterID countRegister = regT1; |
704 | PatternTerm& term = state.term(); |
705 | |
706 | move(imm: Imm32(0), dest: countRegister); |
707 | |
708 | JumpList failures; |
709 | Label loop(this); |
710 | failures.append(jump: atEndOfInput()); |
711 | |
712 | if (term.invertOrCapture) { |
713 | readCharacter(inputPosition: state.inputOffset(), reg: character); |
714 | matchCharacterClass(character, matchDest&: failures, charClass: term.characterClass); |
715 | } else { |
716 | JumpList matchDest; |
717 | readCharacter(inputPosition: state.inputOffset(), reg: character); |
718 | matchCharacterClass(character, matchDest, charClass: term.characterClass); |
719 | failures.append(jump: jump()); |
720 | matchDest.link(masm: this); |
721 | } |
722 | |
723 | add32(imm: Imm32(1), dest: countRegister); |
724 | add32(imm: Imm32(1), dest: index); |
725 | branch32(cond: NotEqual, left: countRegister, right: Imm32(term.quantityCount)).linkTo(label: loop, masm: this); |
726 | failures.append(jump: jump()); |
727 | |
728 | Label backtrackBegin(this); |
729 | loadFromFrame(frameLocation: term.frameLocation, reg: countRegister); |
730 | state.jumpToBacktrack(jump: branchTest32(cond: Zero, reg: countRegister), masm: this); |
731 | sub32(imm: Imm32(1), dest: countRegister); |
732 | sub32(imm: Imm32(1), dest: index); |
733 | |
734 | failures.link(masm: this); |
735 | |
736 | storeToFrame(reg: countRegister, frameLocation: term.frameLocation); |
737 | |
738 | state.setBacktrackGenerated(backtrackBegin); |
739 | } |
740 | |
741 | void generateCharacterClassNonGreedy(TermGenerationState& state) |
742 | { |
743 | const RegisterID character = regT0; |
744 | const RegisterID countRegister = regT1; |
745 | PatternTerm& term = state.term(); |
746 | |
747 | move(imm: Imm32(0), dest: countRegister); |
748 | |
749 | Jump firstTimeDoNothing = jump(); |
750 | |
751 | Label hardFail(this); |
752 | sub32(src: countRegister, dest: index); |
753 | state.jumpToBacktrack(jump: jump(), masm: this); |
754 | |
755 | Label backtrackBegin(this); |
756 | loadFromFrame(frameLocation: term.frameLocation, reg: countRegister); |
757 | |
758 | atEndOfInput().linkTo(label: hardFail, masm: this); |
759 | branch32(cond: Equal, op1: countRegister, imm: Imm32(term.quantityCount), target: hardFail); |
760 | |
761 | JumpList matchDest; |
762 | readCharacter(inputPosition: state.inputOffset(), reg: character); |
763 | matchCharacterClass(character, matchDest, charClass: term.characterClass); |
764 | |
765 | if (term.invertOrCapture) |
766 | matchDest.linkTo(label: hardFail, masm: this); |
767 | else { |
768 | jump(target: hardFail); |
769 | matchDest.link(masm: this); |
770 | } |
771 | |
772 | add32(imm: Imm32(1), dest: countRegister); |
773 | add32(imm: Imm32(1), dest: index); |
774 | |
775 | firstTimeDoNothing.link(masm: this); |
776 | storeToFrame(reg: countRegister, frameLocation: term.frameLocation); |
777 | |
778 | state.setBacktrackGenerated(backtrackBegin); |
779 | } |
780 | |
781 | void generateParenthesesDisjunction(PatternTerm& parenthesesTerm, TermGenerationState& state, unsigned alternativeFrameLocation) |
782 | { |
783 | ASSERT((parenthesesTerm.type == PatternTerm::TypeParenthesesSubpattern) || (parenthesesTerm.type == PatternTerm::TypeParentheticalAssertion)); |
784 | ASSERT(parenthesesTerm.quantityCount == 1); |
785 | |
786 | PatternDisjunction* disjunction = parenthesesTerm.parentheses.disjunction; |
787 | unsigned preCheckedCount = ((parenthesesTerm.quantityType == QuantifierFixedCount) && (parenthesesTerm.type != PatternTerm::TypeParentheticalAssertion)) ? disjunction->m_minimumSize : 0; |
788 | |
789 | if (disjunction->m_alternatives.size() == 1) { |
790 | state.resetAlternative(); |
791 | ASSERT(state.alternativeValid()); |
792 | PatternAlternative* alternative = state.alternative(); |
793 | optimizeAlternative(alternative); |
794 | |
795 | int countToCheck = alternative->m_minimumSize - preCheckedCount; |
796 | if (countToCheck) { |
797 | ASSERT((parenthesesTerm.type == PatternTerm::TypeParentheticalAssertion) || (parenthesesTerm.quantityType != QuantifierFixedCount)); |
798 | |
799 | // FIXME: This is quite horrible. The call to 'plantJumpToBacktrackIfExists' |
800 | // will be forced to always trampoline into here, just to decrement the index. |
801 | // Ick. |
802 | Jump skip = jump(); |
803 | |
804 | Label backtrackBegin(this); |
805 | sub32(imm: Imm32(countToCheck), dest: index); |
806 | state.addBacktrackJump(jump: jump()); |
807 | |
808 | skip.link(masm: this); |
809 | |
810 | state.setBacktrackGenerated(backtrackBegin); |
811 | |
812 | state.jumpToBacktrack(jump: jumpIfNoAvailableInput(countToCheck), masm: this); |
813 | state.checkedTotal += countToCheck; |
814 | } |
815 | |
816 | for (state.resetTerm(); state.termValid(); state.nextTerm()) |
817 | generateTerm(state); |
818 | |
819 | state.checkedTotal -= countToCheck; |
820 | } else { |
821 | JumpList successes; |
822 | |
823 | for (state.resetAlternative(); state.alternativeValid(); state.nextAlternative()) { |
824 | |
825 | PatternAlternative* alternative = state.alternative(); |
826 | optimizeAlternative(alternative); |
827 | |
828 | ASSERT(alternative->m_minimumSize >= preCheckedCount); |
829 | int countToCheck = alternative->m_minimumSize - preCheckedCount; |
830 | if (countToCheck) { |
831 | state.addBacktrackJump(jump: jumpIfNoAvailableInput(countToCheck)); |
832 | state.checkedTotal += countToCheck; |
833 | } |
834 | |
835 | for (state.resetTerm(); state.termValid(); state.nextTerm()) |
836 | generateTerm(state); |
837 | |
838 | // Matched an alternative. |
839 | DataLabelPtr dataLabel = storeToFrameWithPatch(frameLocation: alternativeFrameLocation); |
840 | successes.append(jump: jump()); |
841 | |
842 | // Alternative did not match. |
843 | Label backtrackLocation(this); |
844 | |
845 | // Can we backtrack the alternative? - if so, do so. If not, just fall through to the next one. |
846 | state.plantJumpToBacktrackIfExists(masm: this); |
847 | |
848 | state.linkAlternativeBacktracks(masm: this); |
849 | |
850 | if (countToCheck) { |
851 | sub32(imm: Imm32(countToCheck), dest: index); |
852 | state.checkedTotal -= countToCheck; |
853 | } |
854 | |
855 | m_backtrackRecords.append(val: AlternativeBacktrackRecord(dataLabel, backtrackLocation)); |
856 | } |
857 | // We fall through to here when the last alternative fails. |
858 | // Add a backtrack out of here for the parenthese handling code to link up. |
859 | state.addBacktrackJump(jump: jump()); |
860 | |
861 | // Generate a trampoline for the parens code to backtrack to, to retry the |
862 | // next alternative. |
863 | state.setBacktrackGenerated(label()); |
864 | loadFromFrameAndJump(frameLocation: alternativeFrameLocation); |
865 | |
866 | // FIXME: both of the above hooks are a little inefficient, in that you |
867 | // may end up trampolining here, just to trampoline back out to the |
868 | // parentheses code, or vice versa. We can probably eliminate a jump |
869 | // by restructuring, but coding this way for now for simplicity during |
870 | // development. |
871 | |
872 | successes.link(masm: this); |
873 | } |
874 | } |
875 | |
876 | void generateParenthesesSingle(TermGenerationState& state) |
877 | { |
878 | const RegisterID indexTemporary = regT0; |
879 | PatternTerm& term = state.term(); |
880 | PatternDisjunction* disjunction = term.parentheses.disjunction; |
881 | ASSERT(term.quantityCount == 1); |
882 | |
883 | unsigned preCheckedCount = ((term.quantityCount == 1) && (term.quantityType == QuantifierFixedCount)) ? disjunction->m_minimumSize : 0; |
884 | |
885 | unsigned parenthesesFrameLocation = term.frameLocation; |
886 | unsigned alternativeFrameLocation = parenthesesFrameLocation; |
887 | if (term.quantityType != QuantifierFixedCount) |
888 | alternativeFrameLocation += RegexStackSpaceForBackTrackInfoParenthesesOnce; |
889 | |
890 | // optimized case - no capture & no quantifier can be handled in a light-weight manner. |
891 | if (!term.invertOrCapture && (term.quantityType == QuantifierFixedCount)) { |
892 | TermGenerationState parenthesesState(disjunction, state.checkedTotal); |
893 | generateParenthesesDisjunction(parenthesesTerm&: state.term(), state&: parenthesesState, alternativeFrameLocation); |
894 | // this expects that any backtracks back out of the parentheses will be in the |
895 | // parenthesesState's backTrackJumps vector, and that if they need backtracking |
896 | // they will have set an entry point on the parenthesesState's backtrackLabel. |
897 | state.propagateBacktrackingFrom(nestedParenthesesState&: parenthesesState, masm: this); |
898 | } else { |
899 | Jump nonGreedySkipParentheses; |
900 | Label nonGreedyTryParentheses; |
901 | if (term.quantityType == QuantifierGreedy) |
902 | storeToFrame(imm: Imm32(1), frameLocation: parenthesesFrameLocation); |
903 | else if (term.quantityType == QuantifierNonGreedy) { |
904 | storeToFrame(imm: Imm32(0), frameLocation: parenthesesFrameLocation); |
905 | nonGreedySkipParentheses = jump(); |
906 | nonGreedyTryParentheses = label(); |
907 | storeToFrame(imm: Imm32(1), frameLocation: parenthesesFrameLocation); |
908 | } |
909 | |
910 | // store the match start index |
911 | if (term.invertOrCapture) { |
912 | int inputOffset = state.inputOffset() - preCheckedCount; |
913 | if (inputOffset) { |
914 | move(src: index, dest: indexTemporary); |
915 | add32(imm: Imm32(inputOffset), dest: indexTemporary); |
916 | store32(src: indexTemporary, address: Address(output, (term.parentheses.subpatternId << 1) * sizeof(int))); |
917 | } else |
918 | store32(src: index, address: Address(output, (term.parentheses.subpatternId << 1) * sizeof(int))); |
919 | } |
920 | |
921 | // generate the body of the parentheses |
922 | TermGenerationState parenthesesState(disjunction, state.checkedTotal); |
923 | generateParenthesesDisjunction(parenthesesTerm&: state.term(), state&: parenthesesState, alternativeFrameLocation); |
924 | |
925 | // store the match end index |
926 | if (term.invertOrCapture) { |
927 | int inputOffset = state.inputOffset(); |
928 | if (inputOffset) { |
929 | move(src: index, dest: indexTemporary); |
930 | add32(imm: Imm32(state.inputOffset()), dest: indexTemporary); |
931 | store32(src: indexTemporary, address: Address(output, ((term.parentheses.subpatternId << 1) + 1) * sizeof(int))); |
932 | } else |
933 | store32(src: index, address: Address(output, ((term.parentheses.subpatternId << 1) + 1) * sizeof(int))); |
934 | } |
935 | Jump success = jump(); |
936 | |
937 | // A failure AFTER the parens jumps here |
938 | Label backtrackFromAfterParens(this); |
939 | |
940 | if (term.quantityType == QuantifierGreedy) { |
941 | // If this is zero we have now tested with both with and without the parens. |
942 | loadFromFrame(frameLocation: parenthesesFrameLocation, reg: indexTemporary); |
943 | state.jumpToBacktrack(jump: branchTest32(cond: Zero, reg: indexTemporary), masm: this); |
944 | } else if (term.quantityType == QuantifierNonGreedy) { |
945 | // If this is zero we have now tested with both with and without the parens. |
946 | loadFromFrame(frameLocation: parenthesesFrameLocation, reg: indexTemporary); |
947 | branchTest32(cond: Zero, reg: indexTemporary).linkTo(label: nonGreedyTryParentheses, masm: this); |
948 | } |
949 | |
950 | parenthesesState.plantJumpToBacktrackIfExists(masm: this); |
951 | // A failure WITHIN the parens jumps here |
952 | parenthesesState.linkAlternativeBacktracks(masm: this); |
953 | if (term.invertOrCapture) { |
954 | store32(imm: Imm32(-1), address: Address(output, (term.parentheses.subpatternId << 1) * sizeof(int))); |
955 | store32(imm: Imm32(-1), address: Address(output, ((term.parentheses.subpatternId << 1) + 1) * sizeof(int))); |
956 | } |
957 | |
958 | if (term.quantityType == QuantifierGreedy) |
959 | storeToFrame(imm: Imm32(0), frameLocation: parenthesesFrameLocation); |
960 | else |
961 | state.jumpToBacktrack(jump: jump(), masm: this); |
962 | |
963 | state.setBacktrackGenerated(backtrackFromAfterParens); |
964 | if (term.quantityType == QuantifierNonGreedy) |
965 | nonGreedySkipParentheses.link(masm: this); |
966 | success.link(masm: this); |
967 | } |
968 | } |
969 | |
970 | void generateParentheticalAssertion(TermGenerationState& state) |
971 | { |
972 | PatternTerm& term = state.term(); |
973 | PatternDisjunction* disjunction = term.parentheses.disjunction; |
974 | ASSERT(term.quantityCount == 1); |
975 | ASSERT(term.quantityType == QuantifierFixedCount); |
976 | |
977 | unsigned parenthesesFrameLocation = term.frameLocation; |
978 | unsigned alternativeFrameLocation = parenthesesFrameLocation + RegexStackSpaceForBackTrackInfoParentheticalAssertion; |
979 | |
980 | int countCheckedAfterAssertion = state.checkedTotal - term.inputPosition; |
981 | |
982 | if (term.invertOrCapture) { |
983 | // Inverted case |
984 | storeToFrame(reg: index, frameLocation: parenthesesFrameLocation); |
985 | |
986 | state.checkedTotal -= countCheckedAfterAssertion; |
987 | if (countCheckedAfterAssertion) |
988 | sub32(imm: Imm32(countCheckedAfterAssertion), dest: index); |
989 | |
990 | TermGenerationState parenthesesState(disjunction, state.checkedTotal); |
991 | generateParenthesesDisjunction(parenthesesTerm&: state.term(), state&: parenthesesState, alternativeFrameLocation); |
992 | // Success! - which means - Fail! |
993 | loadFromFrame(frameLocation: parenthesesFrameLocation, reg: index); |
994 | state.jumpToBacktrack(jump: jump(), masm: this); |
995 | |
996 | // And fail means success. |
997 | parenthesesState.linkAlternativeBacktracks(masm: this); |
998 | loadFromFrame(frameLocation: parenthesesFrameLocation, reg: index); |
999 | |
1000 | state.checkedTotal += countCheckedAfterAssertion; |
1001 | } else { |
1002 | // Normal case |
1003 | storeToFrame(reg: index, frameLocation: parenthesesFrameLocation); |
1004 | |
1005 | state.checkedTotal -= countCheckedAfterAssertion; |
1006 | if (countCheckedAfterAssertion) |
1007 | sub32(imm: Imm32(countCheckedAfterAssertion), dest: index); |
1008 | |
1009 | TermGenerationState parenthesesState(disjunction, state.checkedTotal); |
1010 | generateParenthesesDisjunction(parenthesesTerm&: state.term(), state&: parenthesesState, alternativeFrameLocation); |
1011 | // Success! - which means - Success! |
1012 | loadFromFrame(frameLocation: parenthesesFrameLocation, reg: index); |
1013 | Jump success = jump(); |
1014 | |
1015 | parenthesesState.linkAlternativeBacktracks(masm: this); |
1016 | loadFromFrame(frameLocation: parenthesesFrameLocation, reg: index); |
1017 | state.jumpToBacktrack(jump: jump(), masm: this); |
1018 | |
1019 | success.link(masm: this); |
1020 | |
1021 | state.checkedTotal += countCheckedAfterAssertion; |
1022 | } |
1023 | } |
1024 | |
1025 | void generateTerm(TermGenerationState& state) |
1026 | { |
1027 | PatternTerm& term = state.term(); |
1028 | |
1029 | switch (term.type) { |
1030 | case PatternTerm::TypeAssertionBOL: |
1031 | generateAssertionBOL(state); |
1032 | break; |
1033 | |
1034 | case PatternTerm::TypeAssertionEOL: |
1035 | generateAssertionEOL(state); |
1036 | break; |
1037 | |
1038 | case PatternTerm::TypeAssertionWordBoundary: |
1039 | generateAssertionWordBoundary(state); |
1040 | break; |
1041 | |
1042 | case PatternTerm::TypePatternCharacter: |
1043 | switch (term.quantityType) { |
1044 | case QuantifierFixedCount: |
1045 | if (term.quantityCount == 1) { |
1046 | if (state.isSinglePatternCharacterLookaheadTerm() && (state.lookaheadTerm().inputPosition == (term.inputPosition + 1))) { |
1047 | generatePatternCharacterPair(state); |
1048 | state.nextTerm(); |
1049 | } else |
1050 | generatePatternCharacterSingle(state); |
1051 | } else |
1052 | generatePatternCharacterFixed(state); |
1053 | break; |
1054 | case QuantifierGreedy: |
1055 | generatePatternCharacterGreedy(state); |
1056 | break; |
1057 | case QuantifierNonGreedy: |
1058 | generatePatternCharacterNonGreedy(state); |
1059 | break; |
1060 | } |
1061 | break; |
1062 | |
1063 | case PatternTerm::TypeCharacterClass: |
1064 | switch (term.quantityType) { |
1065 | case QuantifierFixedCount: |
1066 | if (term.quantityCount == 1) |
1067 | generateCharacterClassSingle(state); |
1068 | else |
1069 | generateCharacterClassFixed(state); |
1070 | break; |
1071 | case QuantifierGreedy: |
1072 | generateCharacterClassGreedy(state); |
1073 | break; |
1074 | case QuantifierNonGreedy: |
1075 | generateCharacterClassNonGreedy(state); |
1076 | break; |
1077 | } |
1078 | break; |
1079 | |
1080 | case PatternTerm::TypeBackReference: |
1081 | m_generationFailed = true; |
1082 | break; |
1083 | |
1084 | case PatternTerm::TypeForwardReference: |
1085 | break; |
1086 | |
1087 | case PatternTerm::TypeParenthesesSubpattern: |
1088 | if ((term.quantityCount == 1) && !term.parentheses.isCopy) |
1089 | generateParenthesesSingle(state); |
1090 | else |
1091 | m_generationFailed = true; |
1092 | break; |
1093 | |
1094 | case PatternTerm::TypeParentheticalAssertion: |
1095 | generateParentheticalAssertion(state); |
1096 | break; |
1097 | } |
1098 | } |
1099 | |
1100 | void generateDisjunction(PatternDisjunction* disjunction) |
1101 | { |
1102 | TermGenerationState state(disjunction, 0); |
1103 | state.resetAlternative(); |
1104 | |
1105 | // Plant a check to see if there is sufficient input available to run the first alternative. |
1106 | // Jumping back to the label 'firstAlternative' will get to this check, jumping to |
1107 | // 'firstAlternativeInputChecked' will jump directly to matching the alternative having |
1108 | // skipped this check. |
1109 | |
1110 | Label firstAlternative(this); |
1111 | |
1112 | // check availability for the next alternative |
1113 | int countCheckedForCurrentAlternative = 0; |
1114 | int countToCheckForFirstAlternative = 0; |
1115 | bool hasShorterAlternatives = false; |
1116 | JumpList notEnoughInputForPreviousAlternative; |
1117 | |
1118 | if (state.alternativeValid()) { |
1119 | PatternAlternative* alternative = state.alternative(); |
1120 | countToCheckForFirstAlternative = alternative->m_minimumSize; |
1121 | state.checkedTotal += countToCheckForFirstAlternative; |
1122 | if (countToCheckForFirstAlternative) |
1123 | notEnoughInputForPreviousAlternative.append(jump: jumpIfNoAvailableInput(countToCheck: countToCheckForFirstAlternative)); |
1124 | countCheckedForCurrentAlternative = countToCheckForFirstAlternative; |
1125 | } |
1126 | |
1127 | Label firstAlternativeInputChecked(this); |
1128 | |
1129 | while (state.alternativeValid()) { |
1130 | // Track whether any alternatives are shorter than the first one. |
1131 | hasShorterAlternatives = hasShorterAlternatives || (countCheckedForCurrentAlternative < countToCheckForFirstAlternative); |
1132 | |
1133 | PatternAlternative* alternative = state.alternative(); |
1134 | optimizeAlternative(alternative); |
1135 | |
1136 | for (state.resetTerm(); state.termValid(); state.nextTerm()) |
1137 | generateTerm(state); |
1138 | |
1139 | // If we get here, the alternative matched. |
1140 | if (m_pattern.m_body->m_callFrameSize) |
1141 | addPtr(imm: Imm32(m_pattern.m_body->m_callFrameSize * sizeof(void*)), srcDest: stackPointerRegister); |
1142 | |
1143 | ASSERT(index != returnRegister); |
1144 | if (m_pattern.m_body->m_hasFixedSize) { |
1145 | move(src: index, dest: returnRegister); |
1146 | if (alternative->m_minimumSize) |
1147 | sub32(imm: Imm32(alternative->m_minimumSize), dest: returnRegister); |
1148 | } else |
1149 | pop(dest: returnRegister); |
1150 | store32(src: index, address: Address(output, 4)); |
1151 | store32(src: returnRegister, address: output); |
1152 | |
1153 | generateReturn(); |
1154 | |
1155 | state.nextAlternative(); |
1156 | |
1157 | // if there are any more alternatives, plant the check for input before looping. |
1158 | if (state.alternativeValid()) { |
1159 | PatternAlternative* nextAlternative = state.alternative(); |
1160 | int countToCheckForNextAlternative = nextAlternative->m_minimumSize; |
1161 | |
1162 | if (countCheckedForCurrentAlternative > countToCheckForNextAlternative) { // CASE 1: current alternative was longer than the next one. |
1163 | // If we get here, there the last input checked failed. |
1164 | notEnoughInputForPreviousAlternative.link(masm: this); |
1165 | |
1166 | // Check if sufficent input available to run the next alternative |
1167 | notEnoughInputForPreviousAlternative.append(jump: jumpIfNoAvailableInput(countToCheck: countToCheckForNextAlternative - countCheckedForCurrentAlternative)); |
1168 | // We are now in the correct state to enter the next alternative; this add is only required |
1169 | // to mirror and revert operation of the sub32, just below. |
1170 | add32(imm: Imm32(countCheckedForCurrentAlternative - countToCheckForNextAlternative), dest: index); |
1171 | |
1172 | // If we get here, there the last input checked passed. |
1173 | state.linkAlternativeBacktracks(masm: this); |
1174 | // No need to check if we can run the next alternative, since it is shorter - |
1175 | // just update index. |
1176 | sub32(imm: Imm32(countCheckedForCurrentAlternative - countToCheckForNextAlternative), dest: index); |
1177 | } else if (countCheckedForCurrentAlternative < countToCheckForNextAlternative) { // CASE 2: next alternative is longer than the current one. |
1178 | // If we get here, there the last input checked failed. |
1179 | // If there is insufficient input to run the current alternative, and the next alternative is longer, |
1180 | // then there is definitely not enough input to run it - don't even check. Just adjust index, as if |
1181 | // we had checked. |
1182 | notEnoughInputForPreviousAlternative.link(masm: this); |
1183 | add32(imm: Imm32(countToCheckForNextAlternative - countCheckedForCurrentAlternative), dest: index); |
1184 | notEnoughInputForPreviousAlternative.append(jump: jump()); |
1185 | |
1186 | // The next alternative is longer than the current one; check the difference. |
1187 | state.linkAlternativeBacktracks(masm: this); |
1188 | notEnoughInputForPreviousAlternative.append(jump: jumpIfNoAvailableInput(countToCheck: countToCheckForNextAlternative - countCheckedForCurrentAlternative)); |
1189 | } else { // CASE 3: Both alternatives are the same length. |
1190 | ASSERT(countCheckedForCurrentAlternative == countToCheckForNextAlternative); |
1191 | |
1192 | // If the next alterative is the same length as this one, then no need to check the input - |
1193 | // if there was sufficent input to run the current alternative then there is sufficient |
1194 | // input to run the next one; if not, there isn't. |
1195 | state.linkAlternativeBacktracks(masm: this); |
1196 | } |
1197 | |
1198 | state.checkedTotal -= countCheckedForCurrentAlternative; |
1199 | countCheckedForCurrentAlternative = countToCheckForNextAlternative; |
1200 | state.checkedTotal += countCheckedForCurrentAlternative; |
1201 | } |
1202 | } |
1203 | |
1204 | // If we get here, all Alternatives failed... |
1205 | |
1206 | state.checkedTotal -= countCheckedForCurrentAlternative; |
1207 | |
1208 | // How much more input need there be to be able to retry from the first alternative? |
1209 | // examples: |
1210 | // /yarr_jit/ or /wrec|pcre/ |
1211 | // In these examples we need check for one more input before looping. |
1212 | // /yarr_jit|pcre/ |
1213 | // In this case we need check for 5 more input to loop (+4 to allow for the first alterative |
1214 | // being four longer than the last alternative checked, and another +1 to effectively move |
1215 | // the start position along by one). |
1216 | // /yarr|rules/ or /wrec|notsomuch/ |
1217 | // In these examples, provided that there was sufficient input to have just been matching for |
1218 | // the second alternative we can loop without checking for available input (since the second |
1219 | // alternative is longer than the first). In the latter example we need to decrement index |
1220 | // (by 4) so the start position is only progressed by 1 from the last iteration. |
1221 | int incrementForNextIter = (countToCheckForFirstAlternative - countCheckedForCurrentAlternative) + 1; |
1222 | |
1223 | // First, deal with the cases where there was sufficient input to try the last alternative. |
1224 | if (incrementForNextIter > 0) // We need to check for more input anyway, fall through to the checking below. |
1225 | state.linkAlternativeBacktracks(masm: this); |
1226 | else if (m_pattern.m_body->m_hasFixedSize && !incrementForNextIter) // No need to update anything, link these backtracks straight to the to pof the loop! |
1227 | state.linkAlternativeBacktracksTo(label: firstAlternativeInputChecked, masm: this); |
1228 | else { // no need to check the input, but we do have some bookkeeping to do first. |
1229 | state.linkAlternativeBacktracks(masm: this); |
1230 | |
1231 | // Where necessary update our preserved start position. |
1232 | if (!m_pattern.m_body->m_hasFixedSize) { |
1233 | move(src: index, dest: regT0); |
1234 | sub32(imm: Imm32(countCheckedForCurrentAlternative - 1), dest: regT0); |
1235 | poke(src: regT0, index: m_pattern.m_body->m_callFrameSize); |
1236 | } |
1237 | |
1238 | // Update index if necessary, and loop (without checking). |
1239 | if (incrementForNextIter) |
1240 | add32(imm: Imm32(incrementForNextIter), dest: index); |
1241 | jump().linkTo(label: firstAlternativeInputChecked, masm: this); |
1242 | } |
1243 | |
1244 | notEnoughInputForPreviousAlternative.link(masm: this); |
1245 | // Update our idea of the start position, if we're tracking this. |
1246 | if (!m_pattern.m_body->m_hasFixedSize) { |
1247 | if (countCheckedForCurrentAlternative - 1) { |
1248 | move(src: index, dest: regT0); |
1249 | sub32(imm: Imm32(countCheckedForCurrentAlternative - 1), dest: regT0); |
1250 | poke(src: regT0, index: m_pattern.m_body->m_callFrameSize); |
1251 | } else |
1252 | poke(src: index, index: m_pattern.m_body->m_callFrameSize); |
1253 | } |
1254 | // Check if there is sufficent input to run the first alternative again. |
1255 | jumpIfAvailableInput(countToCheck: incrementForNextIter).linkTo(label: firstAlternativeInputChecked, masm: this); |
1256 | // No - insufficent input to run the first alteranative, are there any other alternatives we |
1257 | // might need to check? If so, the last check will have left the index incremented by |
1258 | // (countToCheckForFirstAlternative + 1), so we need test whether countToCheckForFirstAlternative |
1259 | // LESS input is available, to have the effect of just progressing the start position by 1 |
1260 | // from the last iteration. If this check passes we can just jump up to the check associated |
1261 | // with the first alternative in the loop. This is a bit sad, since we'll end up trying the |
1262 | // first alternative again, and this check will fail (otherwise the check planted just above |
1263 | // here would have passed). This is a bit sad, however it saves trying to do something more |
1264 | // complex here in compilation, and in the common case we should end up coallescing the checks. |
1265 | // |
1266 | // FIXME: a nice improvement here may be to stop trying to match sooner, based on the least |
1267 | // of the minimum-alternative-lengths. E.g. if I have two alternatives of length 200 and 150, |
1268 | // and a string of length 100, we'll end up looping index from 0 to 100, checking whether there |
1269 | // is sufficient input to run either alternative (constantly failing). If there had been only |
1270 | // one alternative, or if the shorter alternative had come first, we would have terminated |
1271 | // immediately. :-/ |
1272 | if (hasShorterAlternatives) |
1273 | jumpIfAvailableInput(countToCheck: -countToCheckForFirstAlternative).linkTo(label: firstAlternative, masm: this); |
1274 | // index will now be a bit garbled (depending on whether 'hasShorterAlternatives' is true, |
1275 | // it has either been incremented by 1 or by (countToCheckForFirstAlternative + 1) ... |
1276 | // but since we're about to return a failure this doesn't really matter!) |
1277 | |
1278 | unsigned frameSize = m_pattern.m_body->m_callFrameSize; |
1279 | if (!m_pattern.m_body->m_hasFixedSize) |
1280 | ++frameSize; |
1281 | if (frameSize) |
1282 | addPtr(imm: Imm32(frameSize * sizeof(void*)), srcDest: stackPointerRegister); |
1283 | |
1284 | move(imm: Imm32(-1), dest: returnRegister); |
1285 | |
1286 | generateReturn(); |
1287 | } |
1288 | |
1289 | void generateEnter() |
1290 | { |
1291 | #if CPU(X86_64) |
1292 | push(src: X86Registers::ebp); |
1293 | move(src: stackPointerRegister, dest: X86Registers::ebp); |
1294 | push(src: X86Registers::ebx); |
1295 | #elif CPU(X86) |
1296 | push(X86Registers::ebp); |
1297 | move(stackPointerRegister, X86Registers::ebp); |
1298 | // TODO: do we need spill registers to fill the output pointer if there are no sub captures? |
1299 | push(X86Registers::ebx); |
1300 | push(X86Registers::edi); |
1301 | push(X86Registers::esi); |
1302 | // load output into edi (2 = saved ebp + return address). |
1303 | #if COMPILER(MSVC) |
1304 | loadPtr(Address(X86Registers::ebp, 2 * sizeof(void*)), input); |
1305 | loadPtr(Address(X86Registers::ebp, 3 * sizeof(void*)), index); |
1306 | loadPtr(Address(X86Registers::ebp, 4 * sizeof(void*)), length); |
1307 | loadPtr(Address(X86Registers::ebp, 5 * sizeof(void*)), output); |
1308 | #else |
1309 | loadPtr(Address(X86Registers::ebp, 2 * sizeof(void*)), output); |
1310 | #endif |
1311 | #elif CPU(ARM) |
1312 | push(ARMRegisters::r4); |
1313 | push(ARMRegisters::r5); |
1314 | push(ARMRegisters::r6); |
1315 | #if CPU(ARM_TRADITIONAL) |
1316 | push(ARMRegisters::r8); // scratch register |
1317 | #endif |
1318 | move(ARMRegisters::r3, output); |
1319 | #endif |
1320 | } |
1321 | |
1322 | void generateReturn() |
1323 | { |
1324 | #if CPU(X86_64) |
1325 | pop(dest: X86Registers::ebx); |
1326 | pop(dest: X86Registers::ebp); |
1327 | #elif CPU(X86) |
1328 | pop(X86Registers::esi); |
1329 | pop(X86Registers::edi); |
1330 | pop(X86Registers::ebx); |
1331 | pop(X86Registers::ebp); |
1332 | #elif CPU(ARM) |
1333 | #if CPU(ARM_TRADITIONAL) |
1334 | pop(ARMRegisters::r8); // scratch register |
1335 | #endif |
1336 | pop(ARMRegisters::r6); |
1337 | pop(ARMRegisters::r5); |
1338 | pop(ARMRegisters::r4); |
1339 | #endif |
1340 | ret(); |
1341 | } |
1342 | |
1343 | public: |
1344 | RegexGenerator(RegexPattern& pattern) |
1345 | : m_pattern(pattern) |
1346 | , m_generationFailed(false) |
1347 | { |
1348 | } |
1349 | |
1350 | void generate() |
1351 | { |
1352 | generateEnter(); |
1353 | |
1354 | // TODO: do I really want this on the stack? |
1355 | if (!m_pattern.m_body->m_hasFixedSize) |
1356 | push(src: index); |
1357 | |
1358 | if (m_pattern.m_body->m_callFrameSize) |
1359 | subPtr(imm: Imm32(m_pattern.m_body->m_callFrameSize * sizeof(void*)), dest: stackPointerRegister); |
1360 | |
1361 | generateDisjunction(disjunction: m_pattern.m_body); |
1362 | } |
1363 | |
1364 | void compile(JSGlobalData* globalData, RegexCodeBlock& jitObject) |
1365 | { |
1366 | generate(); |
1367 | |
1368 | LinkBuffer patchBuffer(this, globalData->executableAllocator.poolForSize(n: size())); |
1369 | |
1370 | for (unsigned i = 0; i < m_backtrackRecords.size(); ++i) |
1371 | patchBuffer.patch(label: m_backtrackRecords[i].dataLabel, value: patchBuffer.locationOf(label: m_backtrackRecords[i].backtrackLocation)); |
1372 | |
1373 | jitObject.set(patchBuffer.finalizeCode()); |
1374 | } |
1375 | |
1376 | bool generationFailed() |
1377 | { |
1378 | return m_generationFailed; |
1379 | } |
1380 | |
1381 | private: |
1382 | RegexPattern& m_pattern; |
1383 | Vector<AlternativeBacktrackRecord> m_backtrackRecords; |
1384 | bool m_generationFailed; |
1385 | }; |
1386 | |
1387 | void jitCompileRegex(JSGlobalData* globalData, RegexCodeBlock& jitObject, const UString& patternString, unsigned& numSubpatterns, const char*& error, bool ignoreCase, bool multiline) |
1388 | { |
1389 | RegexPattern pattern(ignoreCase, multiline); |
1390 | |
1391 | if ((error = compileRegex(patternString, pattern))) |
1392 | return; |
1393 | |
1394 | numSubpatterns = pattern.m_numSubpatterns; |
1395 | |
1396 | RegexGenerator generator(pattern); |
1397 | generator.compile(globalData, jitObject); |
1398 | |
1399 | if (generator.generationFailed()) { |
1400 | JSRegExpIgnoreCaseOption ignoreCaseOption = ignoreCase ? JSRegExpIgnoreCase : JSRegExpDoNotIgnoreCase; |
1401 | JSRegExpMultilineOption multilineOption = multiline ? JSRegExpMultiline : JSRegExpSingleLine; |
1402 | jitObject.setFallback(jsRegExpCompile(pattern: reinterpret_cast<const UChar*>(patternString.data()), patternLength: patternString.size(), ignoreCaseOption, multilineOption, numSubpatterns: &numSubpatterns, errorMessage: &error)); |
1403 | } |
1404 | } |
1405 | |
1406 | }} |
1407 | |
1408 | #endif |
1409 | |
1410 | |
1411 | |
1412 | |
1413 | |
1414 | |