1 | /* |
2 | * Copyright (C) 2009-2018 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 "YarrJIT.h" |
28 | |
29 | #include <wtf/ASCIICType.h> |
30 | #include "LinkBuffer.h" |
31 | #include "Options.h" |
32 | #include "VM.h" |
33 | #include "Yarr.h" |
34 | #include "YarrCanonicalize.h" |
35 | #include "YarrDisassembler.h" |
36 | |
37 | #if ENABLE(YARR_JIT) |
38 | |
39 | namespace JSC { namespace Yarr { |
40 | |
41 | template<YarrJITCompileMode compileMode> |
42 | class YarrGenerator : public YarrJITInfo, private MacroAssembler { |
43 | |
44 | #if CPU(ARM_THUMB2) |
45 | static const RegisterID input = ARMRegisters::r0; |
46 | static const RegisterID index = ARMRegisters::r1; |
47 | static const RegisterID length = ARMRegisters::r2; |
48 | static const RegisterID output = ARMRegisters::r3; |
49 | |
50 | static const RegisterID regT0 = ARMRegisters::r4; |
51 | static const RegisterID regT1 = ARMRegisters::r5; |
52 | static const RegisterID initialStart = ARMRegisters::r8; |
53 | |
54 | static const RegisterID returnRegister = ARMRegisters::r0; |
55 | static const RegisterID returnRegister2 = ARMRegisters::r1; |
56 | |
57 | #elif CPU(ARM64) |
58 | // Argument registers |
59 | static const RegisterID input = ARM64Registers::x0; |
60 | static const RegisterID index = ARM64Registers::x1; |
61 | static const RegisterID length = ARM64Registers::x2; |
62 | static const RegisterID output = ARM64Registers::x3; |
63 | static const RegisterID freelistRegister = ARM64Registers::x4; |
64 | static const RegisterID freelistSizeRegister = ARM64Registers::x5; // Only used during initialization. |
65 | |
66 | // Scratch registers |
67 | static const RegisterID regT0 = ARM64Registers::x6; |
68 | static const RegisterID regT1 = ARM64Registers::x7; |
69 | static const RegisterID regT2 = ARM64Registers::x8; |
70 | static const RegisterID remainingMatchCount = ARM64Registers::x9; |
71 | static const RegisterID regUnicodeInputAndTrail = ARM64Registers::x10; |
72 | static const RegisterID unicodeTemp = ARM64Registers::x5; |
73 | static const RegisterID initialStart = ARM64Registers::x11; |
74 | static const RegisterID supplementaryPlanesBase = ARM64Registers::x12; |
75 | static const RegisterID leadingSurrogateTag = ARM64Registers::x13; |
76 | static const RegisterID trailingSurrogateTag = ARM64Registers::x14; |
77 | static const RegisterID endOfStringAddress = ARM64Registers::x15; |
78 | |
79 | static const RegisterID returnRegister = ARM64Registers::x0; |
80 | static const RegisterID returnRegister2 = ARM64Registers::x1; |
81 | |
82 | const TrustedImm32 surrogateTagMask = TrustedImm32(0xfffffc00); |
83 | #define JIT_UNICODE_EXPRESSIONS |
84 | #elif CPU(MIPS) |
85 | static const RegisterID input = MIPSRegisters::a0; |
86 | static const RegisterID index = MIPSRegisters::a1; |
87 | static const RegisterID length = MIPSRegisters::a2; |
88 | static const RegisterID output = MIPSRegisters::a3; |
89 | |
90 | static const RegisterID regT0 = MIPSRegisters::t4; |
91 | static const RegisterID regT1 = MIPSRegisters::t5; |
92 | static const RegisterID initialStart = MIPSRegisters::t6; |
93 | |
94 | static const RegisterID returnRegister = MIPSRegisters::v0; |
95 | static const RegisterID returnRegister2 = MIPSRegisters::v1; |
96 | |
97 | #elif CPU(X86_64) |
98 | #if !OS(WINDOWS) |
99 | // Argument registers |
100 | static const RegisterID input = X86Registers::edi; |
101 | static const RegisterID index = X86Registers::esi; |
102 | static const RegisterID length = X86Registers::edx; |
103 | static const RegisterID output = X86Registers::ecx; |
104 | static const RegisterID freelistRegister = X86Registers::r8; |
105 | static const RegisterID freelistSizeRegister = X86Registers::r9; // Only used during initialization. |
106 | #else |
107 | // If the return value doesn't fit in 64bits, its destination is pointed by rcx and the parameters are shifted. |
108 | // http://msdn.microsoft.com/en-us/library/7572ztz4.aspx |
109 | COMPILE_ASSERT(sizeof(MatchResult) > sizeof(void*), MatchResult_does_not_fit_in_64bits); |
110 | static const RegisterID input = X86Registers::edx; |
111 | static const RegisterID index = X86Registers::r8; |
112 | static const RegisterID length = X86Registers::r9; |
113 | static const RegisterID output = X86Registers::r10; |
114 | #endif |
115 | |
116 | // Scratch registers |
117 | static const RegisterID regT0 = X86Registers::eax; |
118 | #if !OS(WINDOWS) |
119 | static const RegisterID regT1 = X86Registers::r9; |
120 | static const RegisterID regT2 = X86Registers::r10; |
121 | #else |
122 | static const RegisterID regT1 = X86Registers::ecx; |
123 | static const RegisterID regT2 = X86Registers::edi; |
124 | #endif |
125 | |
126 | static const RegisterID initialStart = X86Registers::ebx; |
127 | #if !OS(WINDOWS) |
128 | static const RegisterID remainingMatchCount = X86Registers::r12; |
129 | #else |
130 | static const RegisterID remainingMatchCount = X86Registers::esi; |
131 | #endif |
132 | static const RegisterID regUnicodeInputAndTrail = X86Registers::r13; |
133 | static const RegisterID unicodeTemp = X86Registers::r14; |
134 | static const RegisterID endOfStringAddress = X86Registers::r15; |
135 | |
136 | static const RegisterID returnRegister = X86Registers::eax; |
137 | static const RegisterID returnRegister2 = X86Registers::edx; |
138 | |
139 | const TrustedImm32 supplementaryPlanesBase = TrustedImm32(0x10000); |
140 | const TrustedImm32 leadingSurrogateTag = TrustedImm32(0xd800); |
141 | const TrustedImm32 trailingSurrogateTag = TrustedImm32(0xdc00); |
142 | const TrustedImm32 surrogateTagMask = TrustedImm32(0xfffffc00); |
143 | #define JIT_UNICODE_EXPRESSIONS |
144 | #endif |
145 | |
146 | #if ENABLE(YARR_JIT_ALL_PARENS_EXPRESSIONS) |
147 | struct ParenContextSizes { |
148 | size_t m_numSubpatterns; |
149 | size_t m_frameSlots; |
150 | |
151 | ParenContextSizes(size_t numSubpatterns, size_t frameSlots) |
152 | : m_numSubpatterns(numSubpatterns) |
153 | , m_frameSlots(frameSlots) |
154 | { |
155 | } |
156 | |
157 | size_t numSubpatterns() { return m_numSubpatterns; } |
158 | |
159 | size_t frameSlots() { return m_frameSlots; } |
160 | }; |
161 | |
162 | struct ParenContext { |
163 | struct ParenContext* next; |
164 | uint32_t begin; |
165 | uint32_t matchAmount; |
166 | uintptr_t returnAddress; |
167 | struct Subpatterns { |
168 | unsigned start; |
169 | unsigned end; |
170 | } subpatterns[0]; |
171 | uintptr_t frameSlots[0]; |
172 | |
173 | static size_t sizeFor(ParenContextSizes& parenContextSizes) |
174 | { |
175 | return sizeof(ParenContext) + sizeof(Subpatterns) * parenContextSizes.numSubpatterns() + sizeof(uintptr_t) * parenContextSizes.frameSlots(); |
176 | } |
177 | |
178 | static ptrdiff_t nextOffset() |
179 | { |
180 | return offsetof(ParenContext, next); |
181 | } |
182 | |
183 | static ptrdiff_t beginOffset() |
184 | { |
185 | return offsetof(ParenContext, begin); |
186 | } |
187 | |
188 | static ptrdiff_t matchAmountOffset() |
189 | { |
190 | return offsetof(ParenContext, matchAmount); |
191 | } |
192 | |
193 | static ptrdiff_t returnAddressOffset() |
194 | { |
195 | return offsetof(ParenContext, returnAddress); |
196 | } |
197 | |
198 | static ptrdiff_t subpatternOffset(size_t subpattern) |
199 | { |
200 | return offsetof(ParenContext, subpatterns) + (subpattern - 1) * sizeof(Subpatterns); |
201 | } |
202 | |
203 | static ptrdiff_t savedFrameOffset(ParenContextSizes& parenContextSizes) |
204 | { |
205 | return offsetof(ParenContext, subpatterns) + (parenContextSizes.numSubpatterns()) * sizeof(Subpatterns); |
206 | } |
207 | }; |
208 | |
209 | void initParenContextFreeList() |
210 | { |
211 | RegisterID parenContextPointer = regT0; |
212 | RegisterID nextParenContextPointer = regT2; |
213 | |
214 | size_t parenContextSize = ParenContext::sizeFor(m_parenContextSizes); |
215 | |
216 | parenContextSize = WTF::roundUpToMultipleOf<sizeof(uintptr_t)>(parenContextSize); |
217 | |
218 | if (parenContextSize > VM::patternContextBufferSize) { |
219 | m_failureReason = JITFailureReason::ParenthesisNestedTooDeep; |
220 | return; |
221 | } |
222 | |
223 | Jump emptyFreeList = branchTestPtr(Zero, freelistRegister); |
224 | move(freelistRegister, parenContextPointer); |
225 | addPtr(TrustedImm32(parenContextSize), freelistRegister, nextParenContextPointer); |
226 | addPtr(freelistRegister, freelistSizeRegister); |
227 | subPtr(TrustedImm32(parenContextSize), freelistSizeRegister); |
228 | |
229 | Label loopTop(this); |
230 | Jump initDone = branchPtr(Above, nextParenContextPointer, freelistSizeRegister); |
231 | storePtr(nextParenContextPointer, Address(parenContextPointer, ParenContext::nextOffset())); |
232 | move(nextParenContextPointer, parenContextPointer); |
233 | addPtr(TrustedImm32(parenContextSize), parenContextPointer, nextParenContextPointer); |
234 | jump(loopTop); |
235 | |
236 | initDone.link(this); |
237 | storePtr(TrustedImmPtr(nullptr), Address(parenContextPointer, ParenContext::nextOffset())); |
238 | emptyFreeList.link(this); |
239 | } |
240 | |
241 | void allocateParenContext(RegisterID result) |
242 | { |
243 | m_abortExecution.append(branchTestPtr(Zero, freelistRegister)); |
244 | sub32(TrustedImm32(1), remainingMatchCount); |
245 | m_hitMatchLimit.append(branchTestPtr(Zero, remainingMatchCount)); |
246 | move(freelistRegister, result); |
247 | loadPtr(Address(freelistRegister, ParenContext::nextOffset()), freelistRegister); |
248 | } |
249 | |
250 | void freeParenContext(RegisterID headPtrRegister, RegisterID newHeadPtrRegister) |
251 | { |
252 | loadPtr(Address(headPtrRegister, ParenContext::nextOffset()), newHeadPtrRegister); |
253 | storePtr(freelistRegister, Address(headPtrRegister, ParenContext::nextOffset())); |
254 | move(headPtrRegister, freelistRegister); |
255 | } |
256 | |
257 | void saveParenContext(RegisterID parenContextReg, RegisterID tempReg, unsigned firstSubpattern, unsigned lastSubpattern, unsigned subpatternBaseFrameLocation) |
258 | { |
259 | store32(index, Address(parenContextReg, ParenContext::beginOffset())); |
260 | loadFromFrame(subpatternBaseFrameLocation + BackTrackInfoParentheses::matchAmountIndex(), tempReg); |
261 | store32(tempReg, Address(parenContextReg, ParenContext::matchAmountOffset())); |
262 | loadFromFrame(subpatternBaseFrameLocation + BackTrackInfoParentheses::returnAddressIndex(), tempReg); |
263 | storePtr(tempReg, Address(parenContextReg, ParenContext::returnAddressOffset())); |
264 | if (compileMode == IncludeSubpatterns) { |
265 | for (unsigned subpattern = firstSubpattern; subpattern <= lastSubpattern; subpattern++) { |
266 | loadPtr(Address(output, (subpattern << 1) * sizeof(unsigned)), tempReg); |
267 | storePtr(tempReg, Address(parenContextReg, ParenContext::subpatternOffset(subpattern))); |
268 | clearSubpatternStart(subpattern); |
269 | } |
270 | } |
271 | subpatternBaseFrameLocation += YarrStackSpaceForBackTrackInfoParentheses; |
272 | for (unsigned frameLocation = subpatternBaseFrameLocation; frameLocation < m_parenContextSizes.frameSlots(); frameLocation++) { |
273 | loadFromFrame(frameLocation, tempReg); |
274 | storePtr(tempReg, Address(parenContextReg, ParenContext::savedFrameOffset(m_parenContextSizes) + frameLocation * sizeof(uintptr_t))); |
275 | } |
276 | } |
277 | |
278 | void restoreParenContext(RegisterID parenContextReg, RegisterID tempReg, unsigned firstSubpattern, unsigned lastSubpattern, unsigned subpatternBaseFrameLocation) |
279 | { |
280 | load32(Address(parenContextReg, ParenContext::beginOffset()), index); |
281 | storeToFrame(index, subpatternBaseFrameLocation + BackTrackInfoParentheses::beginIndex()); |
282 | load32(Address(parenContextReg, ParenContext::matchAmountOffset()), tempReg); |
283 | storeToFrame(tempReg, subpatternBaseFrameLocation + BackTrackInfoParentheses::matchAmountIndex()); |
284 | loadPtr(Address(parenContextReg, ParenContext::returnAddressOffset()), tempReg); |
285 | storeToFrame(tempReg, subpatternBaseFrameLocation + BackTrackInfoParentheses::returnAddressIndex()); |
286 | if (compileMode == IncludeSubpatterns) { |
287 | for (unsigned subpattern = firstSubpattern; subpattern <= lastSubpattern; subpattern++) { |
288 | loadPtr(Address(parenContextReg, ParenContext::subpatternOffset(subpattern)), tempReg); |
289 | storePtr(tempReg, Address(output, (subpattern << 1) * sizeof(unsigned))); |
290 | } |
291 | } |
292 | subpatternBaseFrameLocation += YarrStackSpaceForBackTrackInfoParentheses; |
293 | for (unsigned frameLocation = subpatternBaseFrameLocation; frameLocation < m_parenContextSizes.frameSlots(); frameLocation++) { |
294 | loadPtr(Address(parenContextReg, ParenContext::savedFrameOffset(m_parenContextSizes) + frameLocation * sizeof(uintptr_t)), tempReg); |
295 | storeToFrame(tempReg, frameLocation); |
296 | } |
297 | } |
298 | #endif |
299 | |
300 | void optimizeAlternative(PatternAlternative* alternative) |
301 | { |
302 | if (!alternative->m_terms.size()) |
303 | return; |
304 | |
305 | for (unsigned i = 0; i < alternative->m_terms.size() - 1; ++i) { |
306 | PatternTerm& term = alternative->m_terms[i]; |
307 | PatternTerm& nextTerm = alternative->m_terms[i + 1]; |
308 | |
309 | // We can move BMP only character classes after fixed character terms. |
310 | if ((term.type == PatternTerm::TypeCharacterClass) |
311 | && (term.quantityType == QuantifierFixedCount) |
312 | && (!m_decodeSurrogatePairs || (term.characterClass->hasOneCharacterSize() && !term.m_invert)) |
313 | && (nextTerm.type == PatternTerm::TypePatternCharacter) |
314 | && (nextTerm.quantityType == QuantifierFixedCount)) { |
315 | PatternTerm termCopy = term; |
316 | alternative->m_terms[i] = nextTerm; |
317 | alternative->m_terms[i + 1] = termCopy; |
318 | } |
319 | } |
320 | } |
321 | |
322 | void matchCharacterClassRange(RegisterID character, JumpList& failures, JumpList& matchDest, const CharacterRange* ranges, unsigned count, unsigned* matchIndex, const UChar32* matches, unsigned matchCount) |
323 | { |
324 | do { |
325 | // pick which range we're going to generate |
326 | int which = count >> 1; |
327 | char lo = ranges[which].begin; |
328 | char hi = ranges[which].end; |
329 | |
330 | // check if there are any ranges or matches below lo. If not, just jl to failure - |
331 | // if there is anything else to check, check that first, if it falls through jmp to failure. |
332 | if ((*matchIndex < matchCount) && (matches[*matchIndex] < lo)) { |
333 | Jump loOrAbove = branch32(GreaterThanOrEqual, character, Imm32((unsigned short)lo)); |
334 | |
335 | // generate code for all ranges before this one |
336 | if (which) |
337 | matchCharacterClassRange(character, failures, matchDest, ranges, which, matchIndex, matches, matchCount); |
338 | |
339 | while ((*matchIndex < matchCount) && (matches[*matchIndex] < lo)) { |
340 | matchDest.append(branch32(Equal, character, Imm32((unsigned short)matches[*matchIndex]))); |
341 | ++*matchIndex; |
342 | } |
343 | failures.append(jump()); |
344 | |
345 | loOrAbove.link(this); |
346 | } else if (which) { |
347 | Jump loOrAbove = branch32(GreaterThanOrEqual, character, Imm32((unsigned short)lo)); |
348 | |
349 | matchCharacterClassRange(character, failures, matchDest, ranges, which, matchIndex, matches, matchCount); |
350 | failures.append(jump()); |
351 | |
352 | loOrAbove.link(this); |
353 | } else |
354 | failures.append(branch32(LessThan, character, Imm32((unsigned short)lo))); |
355 | |
356 | while ((*matchIndex < matchCount) && (matches[*matchIndex] <= hi)) |
357 | ++*matchIndex; |
358 | |
359 | matchDest.append(branch32(LessThanOrEqual, character, Imm32((unsigned short)hi))); |
360 | // fall through to here, the value is above hi. |
361 | |
362 | // shuffle along & loop around if there are any more matches to handle. |
363 | unsigned next = which + 1; |
364 | ranges += next; |
365 | count -= next; |
366 | } while (count); |
367 | } |
368 | |
369 | void matchCharacterClass(RegisterID character, JumpList& matchDest, const CharacterClass* charClass) |
370 | { |
371 | if (charClass->m_table && !m_decodeSurrogatePairs) { |
372 | ExtendedAddress tableEntry(character, reinterpret_cast<intptr_t>(charClass->m_table)); |
373 | matchDest.append(branchTest8(charClass->m_tableInverted ? Zero : NonZero, tableEntry)); |
374 | return; |
375 | } |
376 | |
377 | JumpList unicodeFail; |
378 | if (charClass->m_matchesUnicode.size() || charClass->m_rangesUnicode.size()) { |
379 | JumpList isAscii; |
380 | if (charClass->m_matches.size() || charClass->m_ranges.size()) |
381 | isAscii.append(branch32(LessThanOrEqual, character, TrustedImm32(0x7f))); |
382 | |
383 | if (charClass->m_matchesUnicode.size()) { |
384 | for (unsigned i = 0; i < charClass->m_matchesUnicode.size(); ++i) { |
385 | UChar32 ch = charClass->m_matchesUnicode[i]; |
386 | matchDest.append(branch32(Equal, character, Imm32(ch))); |
387 | } |
388 | } |
389 | |
390 | if (charClass->m_rangesUnicode.size()) { |
391 | for (unsigned i = 0; i < charClass->m_rangesUnicode.size(); ++i) { |
392 | UChar32 lo = charClass->m_rangesUnicode[i].begin; |
393 | UChar32 hi = charClass->m_rangesUnicode[i].end; |
394 | |
395 | Jump below = branch32(LessThan, character, Imm32(lo)); |
396 | matchDest.append(branch32(LessThanOrEqual, character, Imm32(hi))); |
397 | below.link(this); |
398 | } |
399 | } |
400 | |
401 | if (charClass->m_matches.size() || charClass->m_ranges.size()) |
402 | unicodeFail = jump(); |
403 | isAscii.link(this); |
404 | } |
405 | |
406 | if (charClass->m_ranges.size()) { |
407 | unsigned matchIndex = 0; |
408 | JumpList failures; |
409 | matchCharacterClassRange(character, failures, matchDest, charClass->m_ranges.begin(), charClass->m_ranges.size(), &matchIndex, charClass->m_matches.begin(), charClass->m_matches.size()); |
410 | while (matchIndex < charClass->m_matches.size()) |
411 | matchDest.append(branch32(Equal, character, Imm32((unsigned short)charClass->m_matches[matchIndex++]))); |
412 | |
413 | failures.link(this); |
414 | } else if (charClass->m_matches.size()) { |
415 | // optimization: gather 'a','A' etc back together, can mask & test once. |
416 | Vector<char> matchesAZaz; |
417 | |
418 | for (unsigned i = 0; i < charClass->m_matches.size(); ++i) { |
419 | char ch = charClass->m_matches[i]; |
420 | if (m_pattern.ignoreCase()) { |
421 | if (isASCIILower(ch)) { |
422 | matchesAZaz.append(ch); |
423 | continue; |
424 | } |
425 | if (isASCIIUpper(ch)) |
426 | continue; |
427 | } |
428 | matchDest.append(branch32(Equal, character, Imm32((unsigned short)ch))); |
429 | } |
430 | |
431 | if (unsigned countAZaz = matchesAZaz.size()) { |
432 | or32(TrustedImm32(32), character); |
433 | for (unsigned i = 0; i < countAZaz; ++i) |
434 | matchDest.append(branch32(Equal, character, TrustedImm32(matchesAZaz[i]))); |
435 | } |
436 | } |
437 | |
438 | if (charClass->m_matchesUnicode.size() || charClass->m_rangesUnicode.size()) |
439 | unicodeFail.link(this); |
440 | } |
441 | |
442 | #ifdef JIT_UNICODE_EXPRESSIONS |
443 | void advanceIndexAfterCharacterClassTermMatch(const PatternTerm* term, JumpList& failuresAfterIncrementingIndex, const RegisterID character) |
444 | { |
445 | ASSERT(term->type == PatternTerm::TypeCharacterClass); |
446 | |
447 | if (term->isFixedWidthCharacterClass()) |
448 | add32(TrustedImm32(term->characterClass->hasNonBMPCharacters() ? 2 : 1), index); |
449 | else { |
450 | add32(TrustedImm32(1), index); |
451 | Jump isBMPChar = branch32(LessThan, character, supplementaryPlanesBase); |
452 | failuresAfterIncrementingIndex.append(atEndOfInput()); |
453 | add32(TrustedImm32(1), index); |
454 | isBMPChar.link(this); |
455 | } |
456 | } |
457 | #endif |
458 | |
459 | // Jumps if input not available; will have (incorrectly) incremented already! |
460 | Jump jumpIfNoAvailableInput(unsigned countToCheck = 0) |
461 | { |
462 | if (countToCheck) |
463 | add32(Imm32(countToCheck), index); |
464 | return branch32(Above, index, length); |
465 | } |
466 | |
467 | Jump jumpIfAvailableInput(unsigned countToCheck) |
468 | { |
469 | add32(Imm32(countToCheck), index); |
470 | return branch32(BelowOrEqual, index, length); |
471 | } |
472 | |
473 | Jump checkNotEnoughInput(RegisterID additionalAmount) |
474 | { |
475 | add32(index, additionalAmount); |
476 | return branch32(Above, additionalAmount, length); |
477 | } |
478 | |
479 | Jump checkInput() |
480 | { |
481 | return branch32(BelowOrEqual, index, length); |
482 | } |
483 | |
484 | Jump atEndOfInput() |
485 | { |
486 | return branch32(Equal, index, length); |
487 | } |
488 | |
489 | Jump notAtEndOfInput() |
490 | { |
491 | return branch32(NotEqual, index, length); |
492 | } |
493 | |
494 | BaseIndex negativeOffsetIndexedAddress(Checked<unsigned> negativeCharacterOffset, RegisterID tempReg, RegisterID indexReg = index) |
495 | { |
496 | RegisterID base = input; |
497 | |
498 | // BaseIndex() addressing can take a int32_t offset. Given that we can have a regular |
499 | // expression that has unsigned character offsets, BaseIndex's signed offset is insufficient |
500 | // for addressing in extreme cases where we might underflow. Therefore we check to see if |
501 | // negativeCharacterOffset will underflow directly or after converting for 16 bit characters. |
502 | // If so, we do our own address calculating by adjusting the base, using the result register |
503 | // as a temp address register. |
504 | unsigned maximumNegativeOffsetForCharacterSize = m_charSize == Char8 ? 0x7fffffff : 0x3fffffff; |
505 | unsigned offsetAdjustAmount = 0x40000000; |
506 | if (negativeCharacterOffset.unsafeGet() > maximumNegativeOffsetForCharacterSize) { |
507 | base = tempReg; |
508 | move(input, base); |
509 | while (negativeCharacterOffset.unsafeGet() > maximumNegativeOffsetForCharacterSize) { |
510 | subPtr(TrustedImm32(offsetAdjustAmount), base); |
511 | if (m_charSize != Char8) |
512 | subPtr(TrustedImm32(offsetAdjustAmount), base); |
513 | negativeCharacterOffset -= offsetAdjustAmount; |
514 | } |
515 | } |
516 | |
517 | Checked<int32_t> characterOffset(-static_cast<int32_t>(negativeCharacterOffset.unsafeGet())); |
518 | |
519 | if (m_charSize == Char8) |
520 | return BaseIndex(input, indexReg, TimesOne, (characterOffset * static_cast<int32_t>(sizeof(char))).unsafeGet()); |
521 | |
522 | return BaseIndex(input, indexReg, TimesTwo, (characterOffset * static_cast<int32_t>(sizeof(UChar))).unsafeGet()); |
523 | } |
524 | |
525 | #ifdef JIT_UNICODE_EXPRESSIONS |
526 | void tryReadUnicodeCharImpl(RegisterID resultReg) |
527 | { |
528 | ASSERT(m_charSize == Char16); |
529 | |
530 | JumpList notUnicode; |
531 | |
532 | load16Unaligned(regUnicodeInputAndTrail, resultReg); |
533 | |
534 | // Is the character a leading surrogate? |
535 | and32(surrogateTagMask, resultReg, unicodeTemp); |
536 | notUnicode.append(branch32(NotEqual, unicodeTemp, leadingSurrogateTag)); |
537 | |
538 | // Is the input long enough to read a trailing surrogate? |
539 | addPtr(TrustedImm32(2), regUnicodeInputAndTrail); |
540 | notUnicode.append(branchPtr(AboveOrEqual, regUnicodeInputAndTrail, endOfStringAddress)); |
541 | |
542 | // Is the character a trailing surrogate? |
543 | load16Unaligned(Address(regUnicodeInputAndTrail), regUnicodeInputAndTrail); |
544 | and32(surrogateTagMask, regUnicodeInputAndTrail, unicodeTemp); |
545 | notUnicode.append(branch32(NotEqual, unicodeTemp, trailingSurrogateTag)); |
546 | |
547 | // Combine leading and trailing surrogates to produce a code point. |
548 | lshift32(TrustedImm32(10), resultReg); |
549 | getEffectiveAddress(BaseIndex(resultReg, regUnicodeInputAndTrail, TimesOne, -U16_SURROGATE_OFFSET), resultReg); |
550 | notUnicode.link(this); |
551 | } |
552 | |
553 | void tryReadUnicodeChar(BaseIndex address, RegisterID resultReg) |
554 | { |
555 | ASSERT(m_charSize == Char16); |
556 | |
557 | getEffectiveAddress(address, regUnicodeInputAndTrail); |
558 | |
559 | if (resultReg == regT0) |
560 | m_tryReadUnicodeCharacterCalls.append(nearCall()); |
561 | else |
562 | tryReadUnicodeCharImpl(resultReg); |
563 | } |
564 | #endif |
565 | |
566 | void readCharacter(Checked<unsigned> negativeCharacterOffset, RegisterID resultReg, RegisterID indexReg = index) |
567 | { |
568 | BaseIndex address = negativeOffsetIndexedAddress(negativeCharacterOffset, resultReg, indexReg); |
569 | |
570 | if (m_charSize == Char8) |
571 | load8(address, resultReg); |
572 | #ifdef JIT_UNICODE_EXPRESSIONS |
573 | else if (m_decodeSurrogatePairs) |
574 | tryReadUnicodeChar(address, resultReg); |
575 | #endif |
576 | else |
577 | load16Unaligned(address, resultReg); |
578 | } |
579 | |
580 | Jump jumpIfCharNotEquals(UChar32 ch, Checked<unsigned> negativeCharacterOffset, RegisterID character) |
581 | { |
582 | readCharacter(negativeCharacterOffset, character); |
583 | |
584 | // For case-insesitive compares, non-ascii characters that have different |
585 | // upper & lower case representations are converted to a character class. |
586 | ASSERT(!m_pattern.ignoreCase() || isASCIIAlpha(ch) || isCanonicallyUnique(ch, m_canonicalMode)); |
587 | if (m_pattern.ignoreCase() && isASCIIAlpha(ch)) { |
588 | or32(TrustedImm32(0x20), character); |
589 | ch |= 0x20; |
590 | } |
591 | |
592 | return branch32(NotEqual, character, Imm32(ch)); |
593 | } |
594 | |
595 | void storeToFrame(RegisterID reg, unsigned frameLocation) |
596 | { |
597 | poke(reg, frameLocation); |
598 | } |
599 | |
600 | void storeToFrame(TrustedImm32 imm, unsigned frameLocation) |
601 | { |
602 | poke(imm, frameLocation); |
603 | } |
604 | |
605 | #if CPU(ARM64) || CPU(X86_64) |
606 | void storeToFrame(TrustedImmPtr imm, unsigned frameLocation) |
607 | { |
608 | poke(imm, frameLocation); |
609 | } |
610 | #endif |
611 | |
612 | DataLabelPtr storeToFrameWithPatch(unsigned frameLocation) |
613 | { |
614 | return storePtrWithPatch(TrustedImmPtr(nullptr), Address(stackPointerRegister, frameLocation * sizeof(void*))); |
615 | } |
616 | |
617 | void loadFromFrame(unsigned frameLocation, RegisterID reg) |
618 | { |
619 | peek(reg, frameLocation); |
620 | } |
621 | |
622 | void loadFromFrameAndJump(unsigned frameLocation) |
623 | { |
624 | farJump(Address(stackPointerRegister, frameLocation * sizeof(void*)), YarrBacktrackPtrTag); |
625 | } |
626 | |
627 | unsigned alignCallFrameSizeInBytes(unsigned callFrameSize) |
628 | { |
629 | if (!callFrameSize) |
630 | return 0; |
631 | |
632 | callFrameSize *= sizeof(void*); |
633 | if (callFrameSize / sizeof(void*) != m_pattern.m_body->m_callFrameSize) |
634 | CRASH(); |
635 | callFrameSize = (callFrameSize + 0x3f) & ~0x3f; |
636 | return callFrameSize; |
637 | } |
638 | void initCallFrame() |
639 | { |
640 | unsigned callFrameSizeInBytes = alignCallFrameSizeInBytes(m_pattern.m_body->m_callFrameSize); |
641 | if (callFrameSizeInBytes) { |
642 | #if CPU(X86_64) || CPU(ARM64) |
643 | if (Options::zeroStackFrame()) { |
644 | // We need to start from the stack pointer, because we could have spilled callee saves |
645 | move(stackPointerRegister, regT0); |
646 | subPtr(Imm32(callFrameSizeInBytes), stackPointerRegister); |
647 | if (callFrameSizeInBytes <= 128) { |
648 | for (unsigned offset = 0; offset < callFrameSizeInBytes; offset += sizeof(intptr_t)) |
649 | storePtr(TrustedImm32(0), Address(regT0, -8 - offset)); |
650 | } else { |
651 | Label zeroLoop = label(); |
652 | subPtr(TrustedImm32(sizeof(intptr_t) * 2), regT0); |
653 | #if CPU(ARM64) |
654 | storePair64(ARM64Registers::zr, ARM64Registers::zr, regT0); |
655 | #else |
656 | storePtr(TrustedImm32(0), Address(regT0)); |
657 | storePtr(TrustedImm32(0), Address(regT0, sizeof(intptr_t))); |
658 | #endif |
659 | branchPtr(NotEqual, regT0, stackPointerRegister).linkTo(zeroLoop, this); |
660 | } |
661 | } else |
662 | #endif |
663 | subPtr(Imm32(callFrameSizeInBytes), stackPointerRegister); |
664 | |
665 | } |
666 | } |
667 | void removeCallFrame() |
668 | { |
669 | unsigned callFrameSizeInBytes = alignCallFrameSizeInBytes(m_pattern.m_body->m_callFrameSize); |
670 | if (callFrameSizeInBytes) |
671 | addPtr(Imm32(callFrameSizeInBytes), stackPointerRegister); |
672 | } |
673 | |
674 | void generateFailReturn() |
675 | { |
676 | move(TrustedImmPtr((void*)WTF::notFound), returnRegister); |
677 | move(TrustedImm32(0), returnRegister2); |
678 | generateReturn(); |
679 | } |
680 | |
681 | void generateJITFailReturn() |
682 | { |
683 | if (m_abortExecution.empty() && m_hitMatchLimit.empty()) |
684 | return; |
685 | |
686 | JumpList finishExiting; |
687 | if (!m_abortExecution.empty()) { |
688 | m_abortExecution.link(this); |
689 | move(TrustedImmPtr((void*)static_cast<size_t>(-2)), returnRegister); |
690 | finishExiting.append(jump()); |
691 | } |
692 | |
693 | if (!m_hitMatchLimit.empty()) { |
694 | m_hitMatchLimit.link(this); |
695 | move(TrustedImmPtr((void*)static_cast<size_t>(-1)), returnRegister); |
696 | } |
697 | |
698 | finishExiting.link(this); |
699 | removeCallFrame(); |
700 | move(TrustedImm32(0), returnRegister2); |
701 | generateReturn(); |
702 | } |
703 | |
704 | // Used to record subpatterns, should only be called if compileMode is IncludeSubpatterns. |
705 | void setSubpatternStart(RegisterID reg, unsigned subpattern) |
706 | { |
707 | ASSERT(subpattern); |
708 | // FIXME: should be able to ASSERT(compileMode == IncludeSubpatterns), but then this function is conditionally NORETURN. :-( |
709 | store32(reg, Address(output, (subpattern << 1) * sizeof(int))); |
710 | } |
711 | void setSubpatternEnd(RegisterID reg, unsigned subpattern) |
712 | { |
713 | ASSERT(subpattern); |
714 | // FIXME: should be able to ASSERT(compileMode == IncludeSubpatterns), but then this function is conditionally NORETURN. :-( |
715 | store32(reg, Address(output, ((subpattern << 1) + 1) * sizeof(int))); |
716 | } |
717 | void clearSubpatternStart(unsigned subpattern) |
718 | { |
719 | ASSERT(subpattern); |
720 | // FIXME: should be able to ASSERT(compileMode == IncludeSubpatterns), but then this function is conditionally NORETURN. :-( |
721 | store32(TrustedImm32(-1), Address(output, (subpattern << 1) * sizeof(int))); |
722 | } |
723 | |
724 | void clearMatches(unsigned subpattern, unsigned lastSubpattern) |
725 | { |
726 | for (; subpattern <= lastSubpattern; subpattern++) |
727 | clearSubpatternStart(subpattern); |
728 | } |
729 | |
730 | // We use one of three different strategies to track the start of the current match, |
731 | // while matching. |
732 | // 1) If the pattern has a fixed size, do nothing! - we calculate the value lazily |
733 | // at the end of matching. This is irrespective of compileMode, and in this case |
734 | // these methods should never be called. |
735 | // 2) If we're compiling IncludeSubpatterns, 'output' contains a pointer to an output |
736 | // vector, store the match start in the output vector. |
737 | // 3) If we're compiling MatchOnly, 'output' is unused, store the match start directly |
738 | // in this register. |
739 | void setMatchStart(RegisterID reg) |
740 | { |
741 | ASSERT(!m_pattern.m_body->m_hasFixedSize); |
742 | if (compileMode == IncludeSubpatterns) |
743 | store32(reg, output); |
744 | else |
745 | move(reg, output); |
746 | } |
747 | void getMatchStart(RegisterID reg) |
748 | { |
749 | ASSERT(!m_pattern.m_body->m_hasFixedSize); |
750 | if (compileMode == IncludeSubpatterns) |
751 | load32(output, reg); |
752 | else |
753 | move(output, reg); |
754 | } |
755 | |
756 | enum YarrOpCode : uint8_t { |
757 | // These nodes wrap body alternatives - those in the main disjunction, |
758 | // rather than subpatterns or assertions. These are chained together in |
759 | // a doubly linked list, with a 'begin' node for the first alternative, |
760 | // a 'next' node for each subsequent alternative, and an 'end' node at |
761 | // the end. In the case of repeating alternatives, the 'end' node also |
762 | // has a reference back to 'begin'. |
763 | OpBodyAlternativeBegin, |
764 | OpBodyAlternativeNext, |
765 | OpBodyAlternativeEnd, |
766 | // Similar to the body alternatives, but used for subpatterns with two |
767 | // or more alternatives. |
768 | OpNestedAlternativeBegin, |
769 | OpNestedAlternativeNext, |
770 | OpNestedAlternativeEnd, |
771 | // Used for alternatives in subpatterns where there is only a single |
772 | // alternative (backtracking is easier in these cases), or for alternatives |
773 | // which never need to be backtracked (those in parenthetical assertions, |
774 | // terminal subpatterns). |
775 | OpSimpleNestedAlternativeBegin, |
776 | OpSimpleNestedAlternativeNext, |
777 | OpSimpleNestedAlternativeEnd, |
778 | // Used to wrap 'Once' subpattern matches (quantityMaxCount == 1). |
779 | OpParenthesesSubpatternOnceBegin, |
780 | OpParenthesesSubpatternOnceEnd, |
781 | // Used to wrap 'Terminal' subpattern matches (at the end of the regexp). |
782 | OpParenthesesSubpatternTerminalBegin, |
783 | OpParenthesesSubpatternTerminalEnd, |
784 | // Used to wrap generic captured matches |
785 | OpParenthesesSubpatternBegin, |
786 | OpParenthesesSubpatternEnd, |
787 | // Used to wrap parenthetical assertions. |
788 | OpParentheticalAssertionBegin, |
789 | OpParentheticalAssertionEnd, |
790 | // Wraps all simple terms (pattern characters, character classes). |
791 | OpTerm, |
792 | // Where an expression contains only 'once through' body alternatives |
793 | // and no repeating ones, this op is used to return match failure. |
794 | OpMatchFailed |
795 | }; |
796 | |
797 | // This structure is used to hold the compiled opcode information, |
798 | // including reference back to the original PatternTerm/PatternAlternatives, |
799 | // and JIT compilation data structures. |
800 | struct YarrOp { |
801 | explicit YarrOp(PatternTerm* term) |
802 | : m_term(term) |
803 | , m_op(OpTerm) |
804 | , m_isDeadCode(false) |
805 | { |
806 | } |
807 | |
808 | explicit YarrOp(YarrOpCode op) |
809 | : m_op(op) |
810 | , m_isDeadCode(false) |
811 | { |
812 | } |
813 | |
814 | // For alternatives, this holds the PatternAlternative and doubly linked |
815 | // references to this alternative's siblings. In the case of the |
816 | // OpBodyAlternativeEnd node at the end of a section of repeating nodes, |
817 | // m_nextOp will reference the OpBodyAlternativeBegin node of the first |
818 | // repeating alternative. |
819 | PatternAlternative* m_alternative; |
820 | size_t m_previousOp; |
821 | size_t m_nextOp; |
822 | |
823 | // The operation, as a YarrOpCode, and also a reference to the PatternTerm. |
824 | PatternTerm* m_term; |
825 | YarrOpCode m_op; |
826 | |
827 | // Used to record a set of Jumps out of the generated code, typically |
828 | // used for jumps out to backtracking code, and a single reentry back |
829 | // into the code for a node (likely where a backtrack will trigger |
830 | // rematching). |
831 | Label m_reentry; |
832 | JumpList m_jumps; |
833 | |
834 | // Used for backtracking when the prior alternative did not consume any |
835 | // characters but matched. |
836 | Jump m_zeroLengthMatch; |
837 | |
838 | // This flag is used to null out the second pattern character, when |
839 | // two are fused to match a pair together. |
840 | bool m_isDeadCode; |
841 | |
842 | // Currently used in the case of some of the more complex management of |
843 | // 'm_checkedOffset', to cache the offset used in this alternative, to avoid |
844 | // recalculating it. |
845 | Checked<unsigned> m_checkAdjust; |
846 | |
847 | // Used by OpNestedAlternativeNext/End to hold the pointer to the |
848 | // value that will be pushed into the pattern's frame to return to, |
849 | // upon backtracking back into the disjunction. |
850 | DataLabelPtr m_returnAddress; |
851 | }; |
852 | |
853 | // BacktrackingState |
854 | // This class encapsulates information about the state of code generation |
855 | // whilst generating the code for backtracking, when a term fails to match. |
856 | // Upon entry to code generation of the backtracking code for a given node, |
857 | // the Backtracking state will hold references to all control flow sources |
858 | // that are outputs in need of further backtracking from the prior node |
859 | // generated (which is the subsequent operation in the regular expression, |
860 | // and in the m_ops Vector, since we generated backtracking backwards). |
861 | // These references to control flow take the form of: |
862 | // - A jump list of jumps, to be linked to code that will backtrack them |
863 | // further. |
864 | // - A set of DataLabelPtr values, to be populated with values to be |
865 | // treated effectively as return addresses backtracking into complex |
866 | // subpatterns. |
867 | // - A flag indicating that the current sequence of generated code up to |
868 | // this point requires backtracking. |
869 | class BacktrackingState { |
870 | public: |
871 | BacktrackingState() |
872 | : m_pendingFallthrough(false) |
873 | { |
874 | } |
875 | |
876 | // Add a jump or jumps, a return address, or set the flag indicating |
877 | // that the current 'fallthrough' control flow requires backtracking. |
878 | void append(const Jump& jump) |
879 | { |
880 | m_laterFailures.append(jump); |
881 | } |
882 | void append(JumpList& jumpList) |
883 | { |
884 | m_laterFailures.append(jumpList); |
885 | } |
886 | void append(const DataLabelPtr& returnAddress) |
887 | { |
888 | m_pendingReturns.append(returnAddress); |
889 | } |
890 | void fallthrough() |
891 | { |
892 | ASSERT(!m_pendingFallthrough); |
893 | m_pendingFallthrough = true; |
894 | } |
895 | |
896 | // These methods clear the backtracking state, either linking to the |
897 | // current location, a provided label, or copying the backtracking out |
898 | // to a JumpList. All actions may require code generation to take place, |
899 | // and as such are passed a pointer to the assembler. |
900 | void link(MacroAssembler* assembler) |
901 | { |
902 | if (m_pendingReturns.size()) { |
903 | Label here(assembler); |
904 | for (unsigned i = 0; i < m_pendingReturns.size(); ++i) |
905 | m_backtrackRecords.append(ReturnAddressRecord(m_pendingReturns[i], here)); |
906 | m_pendingReturns.clear(); |
907 | } |
908 | m_laterFailures.link(assembler); |
909 | m_laterFailures.clear(); |
910 | m_pendingFallthrough = false; |
911 | } |
912 | void linkTo(Label label, MacroAssembler* assembler) |
913 | { |
914 | if (m_pendingReturns.size()) { |
915 | for (unsigned i = 0; i < m_pendingReturns.size(); ++i) |
916 | m_backtrackRecords.append(ReturnAddressRecord(m_pendingReturns[i], label)); |
917 | m_pendingReturns.clear(); |
918 | } |
919 | if (m_pendingFallthrough) |
920 | assembler->jump(label); |
921 | m_laterFailures.linkTo(label, assembler); |
922 | m_laterFailures.clear(); |
923 | m_pendingFallthrough = false; |
924 | } |
925 | void takeBacktracksToJumpList(JumpList& jumpList, MacroAssembler* assembler) |
926 | { |
927 | if (m_pendingReturns.size()) { |
928 | Label here(assembler); |
929 | for (unsigned i = 0; i < m_pendingReturns.size(); ++i) |
930 | m_backtrackRecords.append(ReturnAddressRecord(m_pendingReturns[i], here)); |
931 | m_pendingReturns.clear(); |
932 | m_pendingFallthrough = true; |
933 | } |
934 | if (m_pendingFallthrough) |
935 | jumpList.append(assembler->jump()); |
936 | jumpList.append(m_laterFailures); |
937 | m_laterFailures.clear(); |
938 | m_pendingFallthrough = false; |
939 | } |
940 | |
941 | bool isEmpty() |
942 | { |
943 | return m_laterFailures.empty() && m_pendingReturns.isEmpty() && !m_pendingFallthrough; |
944 | } |
945 | |
946 | // Called at the end of code generation to link all return addresses. |
947 | void linkDataLabels(LinkBuffer& linkBuffer) |
948 | { |
949 | ASSERT(isEmpty()); |
950 | for (unsigned i = 0; i < m_backtrackRecords.size(); ++i) |
951 | linkBuffer.patch(m_backtrackRecords[i].m_dataLabel, linkBuffer.locationOf<YarrBacktrackPtrTag>(m_backtrackRecords[i].m_backtrackLocation)); |
952 | } |
953 | |
954 | private: |
955 | struct ReturnAddressRecord { |
956 | ReturnAddressRecord(DataLabelPtr dataLabel, Label backtrackLocation) |
957 | : m_dataLabel(dataLabel) |
958 | , m_backtrackLocation(backtrackLocation) |
959 | { |
960 | } |
961 | |
962 | DataLabelPtr m_dataLabel; |
963 | Label m_backtrackLocation; |
964 | }; |
965 | |
966 | JumpList m_laterFailures; |
967 | bool m_pendingFallthrough; |
968 | Vector<DataLabelPtr, 4> m_pendingReturns; |
969 | Vector<ReturnAddressRecord, 4> m_backtrackRecords; |
970 | }; |
971 | |
972 | // Generation methods: |
973 | // =================== |
974 | |
975 | // This method provides a default implementation of backtracking common |
976 | // to many terms; terms commonly jump out of the forwards matching path |
977 | // on any failed conditions, and add these jumps to the m_jumps list. If |
978 | // no special handling is required we can often just backtrack to m_jumps. |
979 | void backtrackTermDefault(size_t opIndex) |
980 | { |
981 | YarrOp& op = m_ops[opIndex]; |
982 | m_backtrackingState.append(op.m_jumps); |
983 | } |
984 | |
985 | void generateAssertionBOL(size_t opIndex) |
986 | { |
987 | YarrOp& op = m_ops[opIndex]; |
988 | PatternTerm* term = op.m_term; |
989 | |
990 | if (m_pattern.multiline()) { |
991 | const RegisterID character = regT0; |
992 | |
993 | JumpList matchDest; |
994 | if (!term->inputPosition) |
995 | matchDest.append(branch32(Equal, index, Imm32(m_checkedOffset.unsafeGet()))); |
996 | |
997 | readCharacter(m_checkedOffset - term->inputPosition + 1, character); |
998 | matchCharacterClass(character, matchDest, m_pattern.newlineCharacterClass()); |
999 | op.m_jumps.append(jump()); |
1000 | |
1001 | matchDest.link(this); |
1002 | } else { |
1003 | // Erk, really should poison out these alternatives early. :-/ |
1004 | if (term->inputPosition) |
1005 | op.m_jumps.append(jump()); |
1006 | else |
1007 | op.m_jumps.append(branch32(NotEqual, index, Imm32(m_checkedOffset.unsafeGet()))); |
1008 | } |
1009 | } |
1010 | void backtrackAssertionBOL(size_t opIndex) |
1011 | { |
1012 | backtrackTermDefault(opIndex); |
1013 | } |
1014 | |
1015 | void generateAssertionEOL(size_t opIndex) |
1016 | { |
1017 | YarrOp& op = m_ops[opIndex]; |
1018 | PatternTerm* term = op.m_term; |
1019 | |
1020 | if (m_pattern.multiline()) { |
1021 | const RegisterID character = regT0; |
1022 | |
1023 | JumpList matchDest; |
1024 | if (term->inputPosition == m_checkedOffset.unsafeGet()) |
1025 | matchDest.append(atEndOfInput()); |
1026 | |
1027 | readCharacter(m_checkedOffset - term->inputPosition, character); |
1028 | matchCharacterClass(character, matchDest, m_pattern.newlineCharacterClass()); |
1029 | op.m_jumps.append(jump()); |
1030 | |
1031 | matchDest.link(this); |
1032 | } else { |
1033 | if (term->inputPosition == m_checkedOffset.unsafeGet()) |
1034 | op.m_jumps.append(notAtEndOfInput()); |
1035 | // Erk, really should poison out these alternatives early. :-/ |
1036 | else |
1037 | op.m_jumps.append(jump()); |
1038 | } |
1039 | } |
1040 | void backtrackAssertionEOL(size_t opIndex) |
1041 | { |
1042 | backtrackTermDefault(opIndex); |
1043 | } |
1044 | |
1045 | // Also falls though on nextIsNotWordChar. |
1046 | void matchAssertionWordchar(size_t opIndex, JumpList& nextIsWordChar, JumpList& nextIsNotWordChar) |
1047 | { |
1048 | YarrOp& op = m_ops[opIndex]; |
1049 | PatternTerm* term = op.m_term; |
1050 | |
1051 | const RegisterID character = regT0; |
1052 | |
1053 | if (term->inputPosition == m_checkedOffset.unsafeGet()) |
1054 | nextIsNotWordChar.append(atEndOfInput()); |
1055 | |
1056 | readCharacter(m_checkedOffset - term->inputPosition, character); |
1057 | |
1058 | CharacterClass* wordcharCharacterClass; |
1059 | |
1060 | if (m_unicodeIgnoreCase) |
1061 | wordcharCharacterClass = m_pattern.wordUnicodeIgnoreCaseCharCharacterClass(); |
1062 | else |
1063 | wordcharCharacterClass = m_pattern.wordcharCharacterClass(); |
1064 | |
1065 | matchCharacterClass(character, nextIsWordChar, wordcharCharacterClass); |
1066 | } |
1067 | |
1068 | void generateAssertionWordBoundary(size_t opIndex) |
1069 | { |
1070 | YarrOp& op = m_ops[opIndex]; |
1071 | PatternTerm* term = op.m_term; |
1072 | |
1073 | const RegisterID character = regT0; |
1074 | |
1075 | Jump atBegin; |
1076 | JumpList matchDest; |
1077 | if (!term->inputPosition) |
1078 | atBegin = branch32(Equal, index, Imm32(m_checkedOffset.unsafeGet())); |
1079 | readCharacter(m_checkedOffset - term->inputPosition + 1, character); |
1080 | |
1081 | CharacterClass* wordcharCharacterClass; |
1082 | |
1083 | if (m_unicodeIgnoreCase) |
1084 | wordcharCharacterClass = m_pattern.wordUnicodeIgnoreCaseCharCharacterClass(); |
1085 | else |
1086 | wordcharCharacterClass = m_pattern.wordcharCharacterClass(); |
1087 | |
1088 | matchCharacterClass(character, matchDest, wordcharCharacterClass); |
1089 | if (!term->inputPosition) |
1090 | atBegin.link(this); |
1091 | |
1092 | // We fall through to here if the last character was not a wordchar. |
1093 | JumpList nonWordCharThenWordChar; |
1094 | JumpList nonWordCharThenNonWordChar; |
1095 | if (term->invert()) { |
1096 | matchAssertionWordchar(opIndex, nonWordCharThenNonWordChar, nonWordCharThenWordChar); |
1097 | nonWordCharThenWordChar.append(jump()); |
1098 | } else { |
1099 | matchAssertionWordchar(opIndex, nonWordCharThenWordChar, nonWordCharThenNonWordChar); |
1100 | nonWordCharThenNonWordChar.append(jump()); |
1101 | } |
1102 | op.m_jumps.append(nonWordCharThenNonWordChar); |
1103 | |
1104 | // We jump here if the last character was a wordchar. |
1105 | matchDest.link(this); |
1106 | JumpList wordCharThenWordChar; |
1107 | JumpList wordCharThenNonWordChar; |
1108 | if (term->invert()) { |
1109 | matchAssertionWordchar(opIndex, wordCharThenNonWordChar, wordCharThenWordChar); |
1110 | wordCharThenWordChar.append(jump()); |
1111 | } else { |
1112 | matchAssertionWordchar(opIndex, wordCharThenWordChar, wordCharThenNonWordChar); |
1113 | // This can fall-though! |
1114 | } |
1115 | |
1116 | op.m_jumps.append(wordCharThenWordChar); |
1117 | |
1118 | nonWordCharThenWordChar.link(this); |
1119 | wordCharThenNonWordChar.link(this); |
1120 | } |
1121 | void backtrackAssertionWordBoundary(size_t opIndex) |
1122 | { |
1123 | backtrackTermDefault(opIndex); |
1124 | } |
1125 | |
1126 | #if ENABLE(YARR_JIT_BACKREFERENCES) |
1127 | void matchBackreference(size_t opIndex, JumpList& characterMatchFails, RegisterID character, RegisterID patternIndex, RegisterID patternCharacter) |
1128 | { |
1129 | YarrOp& op = m_ops[opIndex]; |
1130 | PatternTerm* term = op.m_term; |
1131 | unsigned subpatternId = term->backReferenceSubpatternId; |
1132 | |
1133 | Label loop(this); |
1134 | |
1135 | readCharacter(0, patternCharacter, patternIndex); |
1136 | readCharacter(m_checkedOffset - term->inputPosition, character); |
1137 | |
1138 | if (!m_pattern.ignoreCase()) |
1139 | characterMatchFails.append(branch32(NotEqual, character, patternCharacter)); |
1140 | else { |
1141 | Jump charactersMatch = branch32(Equal, character, patternCharacter); |
1142 | ExtendedAddress characterTableEntry(character, reinterpret_cast<intptr_t>(&canonicalTableLChar)); |
1143 | load16(characterTableEntry, character); |
1144 | ExtendedAddress patternTableEntry(patternCharacter, reinterpret_cast<intptr_t>(&canonicalTableLChar)); |
1145 | load16(patternTableEntry, patternCharacter); |
1146 | characterMatchFails.append(branch32(NotEqual, character, patternCharacter)); |
1147 | charactersMatch.link(this); |
1148 | } |
1149 | |
1150 | add32(TrustedImm32(1), index); |
1151 | add32(TrustedImm32(1), patternIndex); |
1152 | |
1153 | if (m_decodeSurrogatePairs) { |
1154 | Jump isBMPChar = branch32(LessThan, character, supplementaryPlanesBase); |
1155 | add32(TrustedImm32(1), index); |
1156 | add32(TrustedImm32(1), patternIndex); |
1157 | isBMPChar.link(this); |
1158 | } |
1159 | |
1160 | branch32(NotEqual, patternIndex, Address(output, ((subpatternId << 1) + 1) * sizeof(int))).linkTo(loop, this); |
1161 | } |
1162 | |
1163 | void generateBackReference(size_t opIndex) |
1164 | { |
1165 | YarrOp& op = m_ops[opIndex]; |
1166 | PatternTerm* term = op.m_term; |
1167 | |
1168 | if (m_pattern.ignoreCase() && m_charSize != Char8) { |
1169 | m_failureReason = JITFailureReason::BackReference; |
1170 | return; |
1171 | } |
1172 | |
1173 | unsigned subpatternId = term->backReferenceSubpatternId; |
1174 | unsigned parenthesesFrameLocation = term->frameLocation; |
1175 | |
1176 | const RegisterID characterOrTemp = regT0; |
1177 | const RegisterID patternIndex = regT1; |
1178 | const RegisterID patternTemp = regT2; |
1179 | |
1180 | storeToFrame(index, parenthesesFrameLocation + BackTrackInfoBackReference::beginIndex()); |
1181 | if (term->quantityType != QuantifierFixedCount || term->quantityMaxCount != 1) |
1182 | storeToFrame(TrustedImm32(0), parenthesesFrameLocation + BackTrackInfoBackReference::matchAmountIndex()); |
1183 | |
1184 | JumpList matches; |
1185 | |
1186 | if (term->quantityType != QuantifierNonGreedy) { |
1187 | load32(Address(output, (subpatternId << 1) * sizeof(int)), patternIndex); |
1188 | load32(Address(output, ((subpatternId << 1) + 1) * sizeof(int)), patternTemp); |
1189 | |
1190 | // An empty match is successful without consuming characters |
1191 | if (term->quantityType != QuantifierFixedCount || term->quantityMaxCount != 1) { |
1192 | matches.append(branch32(Equal, TrustedImm32(-1), patternIndex)); |
1193 | matches.append(branch32(Equal, patternIndex, patternTemp)); |
1194 | } else { |
1195 | Jump zeroLengthMatch = branch32(Equal, TrustedImm32(-1), patternIndex); |
1196 | Jump tryNonZeroMatch = branch32(NotEqual, patternIndex, patternTemp); |
1197 | zeroLengthMatch.link(this); |
1198 | storeToFrame(TrustedImm32(1), parenthesesFrameLocation + BackTrackInfoBackReference::matchAmountIndex()); |
1199 | matches.append(jump()); |
1200 | tryNonZeroMatch.link(this); |
1201 | } |
1202 | } |
1203 | |
1204 | switch (term->quantityType) { |
1205 | case QuantifierFixedCount: { |
1206 | Label outerLoop(this); |
1207 | |
1208 | // PatternTemp should contain pattern end index at this point |
1209 | sub32(patternIndex, patternTemp); |
1210 | op.m_jumps.append(checkNotEnoughInput(patternTemp)); |
1211 | |
1212 | matchBackreference(opIndex, op.m_jumps, characterOrTemp, patternIndex, patternTemp); |
1213 | |
1214 | if (term->quantityMaxCount != 1) { |
1215 | loadFromFrame(parenthesesFrameLocation + BackTrackInfoBackReference::matchAmountIndex(), characterOrTemp); |
1216 | add32(TrustedImm32(1), characterOrTemp); |
1217 | storeToFrame(characterOrTemp, parenthesesFrameLocation + BackTrackInfoBackReference::matchAmountIndex()); |
1218 | matches.append(branch32(Equal, Imm32(term->quantityMaxCount.unsafeGet()), characterOrTemp)); |
1219 | load32(Address(output, (subpatternId << 1) * sizeof(int)), patternIndex); |
1220 | load32(Address(output, ((subpatternId << 1) + 1) * sizeof(int)), patternTemp); |
1221 | jump(outerLoop); |
1222 | } |
1223 | matches.link(this); |
1224 | break; |
1225 | } |
1226 | |
1227 | case QuantifierGreedy: { |
1228 | JumpList incompleteMatches; |
1229 | |
1230 | Label outerLoop(this); |
1231 | |
1232 | // PatternTemp should contain pattern end index at this point |
1233 | sub32(patternIndex, patternTemp); |
1234 | matches.append(checkNotEnoughInput(patternTemp)); |
1235 | |
1236 | matchBackreference(opIndex, incompleteMatches, characterOrTemp, patternIndex, patternTemp); |
1237 | |
1238 | loadFromFrame(parenthesesFrameLocation + BackTrackInfoBackReference::matchAmountIndex(), characterOrTemp); |
1239 | add32(TrustedImm32(1), characterOrTemp); |
1240 | storeToFrame(characterOrTemp, parenthesesFrameLocation + BackTrackInfoBackReference::matchAmountIndex()); |
1241 | if (term->quantityMaxCount != quantifyInfinite) |
1242 | matches.append(branch32(Equal, Imm32(term->quantityMaxCount.unsafeGet()), characterOrTemp)); |
1243 | load32(Address(output, (subpatternId << 1) * sizeof(int)), patternIndex); |
1244 | load32(Address(output, ((subpatternId << 1) + 1) * sizeof(int)), patternTemp); |
1245 | |
1246 | // Store current index in frame for restoring after a partial match |
1247 | storeToFrame(index, parenthesesFrameLocation + BackTrackInfoBackReference::beginIndex()); |
1248 | jump(outerLoop); |
1249 | |
1250 | incompleteMatches.link(this); |
1251 | loadFromFrame(parenthesesFrameLocation + BackTrackInfoBackReference::beginIndex(), index); |
1252 | |
1253 | matches.link(this); |
1254 | op.m_reentry = label(); |
1255 | break; |
1256 | } |
1257 | |
1258 | case QuantifierNonGreedy: { |
1259 | JumpList incompleteMatches; |
1260 | |
1261 | matches.append(jump()); |
1262 | |
1263 | op.m_reentry = label(); |
1264 | |
1265 | load32(Address(output, (subpatternId << 1) * sizeof(int)), patternIndex); |
1266 | load32(Address(output, ((subpatternId << 1) + 1) * sizeof(int)), patternTemp); |
1267 | |
1268 | // An empty match is successful without consuming characters |
1269 | Jump zeroLengthMatch = branch32(Equal, TrustedImm32(-1), patternIndex); |
1270 | Jump tryNonZeroMatch = branch32(NotEqual, patternIndex, patternTemp); |
1271 | zeroLengthMatch.link(this); |
1272 | storeToFrame(TrustedImm32(1), parenthesesFrameLocation + BackTrackInfoBackReference::matchAmountIndex()); |
1273 | matches.append(jump()); |
1274 | tryNonZeroMatch.link(this); |
1275 | |
1276 | // Check if we have input remaining to match |
1277 | sub32(patternIndex, patternTemp); |
1278 | matches.append(checkNotEnoughInput(patternTemp)); |
1279 | |
1280 | storeToFrame(index, parenthesesFrameLocation + BackTrackInfoBackReference::beginIndex()); |
1281 | |
1282 | matchBackreference(opIndex, incompleteMatches, characterOrTemp, patternIndex, patternTemp); |
1283 | |
1284 | matches.append(jump()); |
1285 | |
1286 | incompleteMatches.link(this); |
1287 | loadFromFrame(parenthesesFrameLocation + BackTrackInfoBackReference::beginIndex(), index); |
1288 | |
1289 | matches.link(this); |
1290 | break; |
1291 | } |
1292 | } |
1293 | } |
1294 | void backtrackBackReference(size_t opIndex) |
1295 | { |
1296 | YarrOp& op = m_ops[opIndex]; |
1297 | PatternTerm* term = op.m_term; |
1298 | |
1299 | unsigned subpatternId = term->backReferenceSubpatternId; |
1300 | |
1301 | m_backtrackingState.link(this); |
1302 | op.m_jumps.link(this); |
1303 | |
1304 | JumpList failures; |
1305 | |
1306 | unsigned parenthesesFrameLocation = term->frameLocation; |
1307 | switch (term->quantityType) { |
1308 | case QuantifierFixedCount: |
1309 | loadFromFrame(parenthesesFrameLocation + BackTrackInfoBackReference::beginIndex(), index); |
1310 | break; |
1311 | |
1312 | case QuantifierGreedy: { |
1313 | const RegisterID matchAmount = regT0; |
1314 | const RegisterID patternStartIndex = regT1; |
1315 | const RegisterID patternEndIndexOrLen = regT2; |
1316 | |
1317 | loadFromFrame(parenthesesFrameLocation + BackTrackInfoBackReference::matchAmountIndex(), matchAmount); |
1318 | failures.append(branchTest32(Zero, matchAmount)); |
1319 | |
1320 | load32(Address(output, (subpatternId << 1) * sizeof(int)), patternStartIndex); |
1321 | load32(Address(output, ((subpatternId << 1) + 1) * sizeof(int)), patternEndIndexOrLen); |
1322 | sub32(patternStartIndex, patternEndIndexOrLen); |
1323 | sub32(patternEndIndexOrLen, index); |
1324 | |
1325 | sub32(TrustedImm32(1), matchAmount); |
1326 | storeToFrame(matchAmount, parenthesesFrameLocation + BackTrackInfoBackReference::matchAmountIndex()); |
1327 | jump(op.m_reentry); |
1328 | break; |
1329 | } |
1330 | |
1331 | case QuantifierNonGreedy: { |
1332 | const RegisterID matchAmount = regT0; |
1333 | |
1334 | failures.append(atEndOfInput()); |
1335 | loadFromFrame(parenthesesFrameLocation + BackTrackInfoBackReference::matchAmountIndex(), matchAmount); |
1336 | if (term->quantityMaxCount != quantifyInfinite) |
1337 | failures.append(branch32(AboveOrEqual, Imm32(term->quantityMaxCount.unsafeGet()), matchAmount)); |
1338 | add32(TrustedImm32(1), matchAmount); |
1339 | storeToFrame(matchAmount, parenthesesFrameLocation + BackTrackInfoBackReference::matchAmountIndex()); |
1340 | jump(op.m_reentry); |
1341 | break; |
1342 | } |
1343 | } |
1344 | failures.link(this); |
1345 | m_backtrackingState.fallthrough(); |
1346 | } |
1347 | #endif |
1348 | |
1349 | void generatePatternCharacterOnce(size_t opIndex) |
1350 | { |
1351 | YarrOp& op = m_ops[opIndex]; |
1352 | |
1353 | if (op.m_isDeadCode) |
1354 | return; |
1355 | |
1356 | // m_ops always ends with a OpBodyAlternativeEnd or OpMatchFailed |
1357 | // node, so there must always be at least one more node. |
1358 | ASSERT(opIndex + 1 < m_ops.size()); |
1359 | YarrOp* nextOp = &m_ops[opIndex + 1]; |
1360 | |
1361 | PatternTerm* term = op.m_term; |
1362 | UChar32 ch = term->patternCharacter; |
1363 | |
1364 | if (!isLatin1(ch) && (m_charSize == Char8)) { |
1365 | // Have a 16 bit pattern character and an 8 bit string - short circuit |
1366 | op.m_jumps.append(jump()); |
1367 | return; |
1368 | } |
1369 | |
1370 | const RegisterID character = regT0; |
1371 | #if CPU(X86_64) || CPU(ARM64) |
1372 | unsigned maxCharactersAtOnce = m_charSize == Char8 ? 8 : 4; |
1373 | #else |
1374 | unsigned maxCharactersAtOnce = m_charSize == Char8 ? 4 : 2; |
1375 | #endif |
1376 | uint64_t ignoreCaseMask = 0; |
1377 | #if CPU(BIG_ENDIAN) |
1378 | uint64_t allCharacters = ch << (m_charSize == Char8 ? 24 : 16); |
1379 | #else |
1380 | uint64_t allCharacters = ch; |
1381 | #endif |
1382 | unsigned numberCharacters; |
1383 | unsigned startTermPosition = term->inputPosition; |
1384 | |
1385 | // For case-insesitive compares, non-ascii characters that have different |
1386 | // upper & lower case representations are converted to a character class. |
1387 | ASSERT(!m_pattern.ignoreCase() || isASCIIAlpha(ch) || isCanonicallyUnique(ch, m_canonicalMode)); |
1388 | |
1389 | if (m_pattern.ignoreCase() && isASCIIAlpha(ch)) { |
1390 | #if CPU(BIG_ENDIAN) |
1391 | ignoreCaseMask |= 32 << (m_charSize == Char8 ? 24 : 16); |
1392 | #else |
1393 | ignoreCaseMask |= 32; |
1394 | #endif |
1395 | } |
1396 | |
1397 | for (numberCharacters = 1; numberCharacters < maxCharactersAtOnce && nextOp->m_op == OpTerm; ++numberCharacters, nextOp = &m_ops[opIndex + numberCharacters]) { |
1398 | PatternTerm* nextTerm = nextOp->m_term; |
1399 | |
1400 | // YarrJIT handles decoded surrogate pair as one character if unicode flag is enabled. |
1401 | // Note that the numberCharacters become 1 while the width of the pattern character becomes 32bit in this case. |
1402 | if (nextTerm->type != PatternTerm::TypePatternCharacter |
1403 | || nextTerm->quantityType != QuantifierFixedCount |
1404 | || nextTerm->quantityMaxCount != 1 |
1405 | || nextTerm->inputPosition != (startTermPosition + numberCharacters) |
1406 | || (U16_LENGTH(nextTerm->patternCharacter) != 1 && m_decodeSurrogatePairs)) |
1407 | break; |
1408 | |
1409 | nextOp->m_isDeadCode = true; |
1410 | |
1411 | #if CPU(BIG_ENDIAN) |
1412 | int shiftAmount = (m_charSize == Char8 ? 24 : 16) - ((m_charSize == Char8 ? 8 : 16) * numberCharacters); |
1413 | #else |
1414 | int shiftAmount = (m_charSize == Char8 ? 8 : 16) * numberCharacters; |
1415 | #endif |
1416 | |
1417 | UChar32 currentCharacter = nextTerm->patternCharacter; |
1418 | |
1419 | if (!isLatin1(currentCharacter) && (m_charSize == Char8)) { |
1420 | // Have a 16 bit pattern character and an 8 bit string - short circuit |
1421 | op.m_jumps.append(jump()); |
1422 | return; |
1423 | } |
1424 | |
1425 | // For case-insesitive compares, non-ascii characters that have different |
1426 | // upper & lower case representations are converted to a character class. |
1427 | ASSERT(!m_pattern.ignoreCase() || isASCIIAlpha(currentCharacter) || isCanonicallyUnique(currentCharacter, m_canonicalMode)); |
1428 | |
1429 | allCharacters |= (static_cast<uint64_t>(currentCharacter) << shiftAmount); |
1430 | |
1431 | if ((m_pattern.ignoreCase()) && (isASCIIAlpha(currentCharacter))) |
1432 | ignoreCaseMask |= 32ULL << shiftAmount; |
1433 | } |
1434 | |
1435 | if (m_decodeSurrogatePairs) |
1436 | op.m_jumps.append(jumpIfNoAvailableInput()); |
1437 | |
1438 | if (m_charSize == Char8) { |
1439 | auto check1 = [&] (Checked<unsigned> offset, UChar32 characters) { |
1440 | op.m_jumps.append(jumpIfCharNotEquals(characters, offset, character)); |
1441 | }; |
1442 | |
1443 | auto check2 = [&] (Checked<unsigned> offset, uint16_t characters, uint16_t mask) { |
1444 | load16Unaligned(negativeOffsetIndexedAddress(offset, character), character); |
1445 | if (mask) |
1446 | or32(Imm32(mask), character); |
1447 | op.m_jumps.append(branch32(NotEqual, character, Imm32(characters | mask))); |
1448 | }; |
1449 | |
1450 | auto check4 = [&] (Checked<unsigned> offset, unsigned characters, unsigned mask) { |
1451 | if (mask) { |
1452 | load32WithUnalignedHalfWords(negativeOffsetIndexedAddress(offset, character), character); |
1453 | if (mask) |
1454 | or32(Imm32(mask), character); |
1455 | op.m_jumps.append(branch32(NotEqual, character, Imm32(characters | mask))); |
1456 | return; |
1457 | } |
1458 | op.m_jumps.append(branch32WithUnalignedHalfWords(NotEqual, negativeOffsetIndexedAddress(offset, character), TrustedImm32(characters))); |
1459 | }; |
1460 | |
1461 | #if CPU(X86_64) || CPU(ARM64) |
1462 | auto check8 = [&] (Checked<unsigned> offset, uint64_t characters, uint64_t mask) { |
1463 | load64(negativeOffsetIndexedAddress(offset, character), character); |
1464 | if (mask) |
1465 | or64(TrustedImm64(mask), character); |
1466 | op.m_jumps.append(branch64(NotEqual, character, TrustedImm64(characters | mask))); |
1467 | }; |
1468 | #endif |
1469 | |
1470 | switch (numberCharacters) { |
1471 | case 1: |
1472 | // Use 32bit width of allCharacters since Yarr counts surrogate pairs as one character with unicode flag. |
1473 | check1(m_checkedOffset - startTermPosition, allCharacters & 0xffffffff); |
1474 | return; |
1475 | case 2: { |
1476 | check2(m_checkedOffset - startTermPosition, allCharacters & 0xffff, ignoreCaseMask & 0xffff); |
1477 | return; |
1478 | } |
1479 | case 3: { |
1480 | check2(m_checkedOffset - startTermPosition, allCharacters & 0xffff, ignoreCaseMask & 0xffff); |
1481 | check1(m_checkedOffset - startTermPosition - 2, (allCharacters >> 16) & 0xff); |
1482 | return; |
1483 | } |
1484 | case 4: { |
1485 | check4(m_checkedOffset - startTermPosition, allCharacters & 0xffffffff, ignoreCaseMask & 0xffffffff); |
1486 | return; |
1487 | } |
1488 | #if CPU(X86_64) || CPU(ARM64) |
1489 | case 5: { |
1490 | check4(m_checkedOffset - startTermPosition, allCharacters & 0xffffffff, ignoreCaseMask & 0xffffffff); |
1491 | check1(m_checkedOffset - startTermPosition - 4, (allCharacters >> 32) & 0xff); |
1492 | return; |
1493 | } |
1494 | case 6: { |
1495 | check4(m_checkedOffset - startTermPosition, allCharacters & 0xffffffff, ignoreCaseMask & 0xffffffff); |
1496 | check2(m_checkedOffset - startTermPosition - 4, (allCharacters >> 32) & 0xffff, (ignoreCaseMask >> 32) & 0xffff); |
1497 | return; |
1498 | } |
1499 | case 7: { |
1500 | check4(m_checkedOffset - startTermPosition, allCharacters & 0xffffffff, ignoreCaseMask & 0xffffffff); |
1501 | check2(m_checkedOffset - startTermPosition - 4, (allCharacters >> 32) & 0xffff, (ignoreCaseMask >> 32) & 0xffff); |
1502 | check1(m_checkedOffset - startTermPosition - 6, (allCharacters >> 48) & 0xff); |
1503 | return; |
1504 | } |
1505 | case 8: { |
1506 | check8(m_checkedOffset - startTermPosition, allCharacters, ignoreCaseMask); |
1507 | return; |
1508 | } |
1509 | #endif |
1510 | } |
1511 | } else { |
1512 | auto check1 = [&] (Checked<unsigned> offset, UChar32 characters) { |
1513 | op.m_jumps.append(jumpIfCharNotEquals(characters, offset, character)); |
1514 | }; |
1515 | |
1516 | auto check2 = [&] (Checked<unsigned> offset, unsigned characters, unsigned mask) { |
1517 | if (mask) { |
1518 | load32WithUnalignedHalfWords(negativeOffsetIndexedAddress(offset, character), character); |
1519 | if (mask) |
1520 | or32(Imm32(mask), character); |
1521 | op.m_jumps.append(branch32(NotEqual, character, Imm32(characters | mask))); |
1522 | return; |
1523 | } |
1524 | op.m_jumps.append(branch32WithUnalignedHalfWords(NotEqual, negativeOffsetIndexedAddress(offset, character), TrustedImm32(characters))); |
1525 | }; |
1526 | |
1527 | #if CPU(X86_64) || CPU(ARM64) |
1528 | auto check4 = [&] (Checked<unsigned> offset, uint64_t characters, uint64_t mask) { |
1529 | load64(negativeOffsetIndexedAddress(offset, character), character); |
1530 | if (mask) |
1531 | or64(TrustedImm64(mask), character); |
1532 | op.m_jumps.append(branch64(NotEqual, character, TrustedImm64(characters | mask))); |
1533 | }; |
1534 | #endif |
1535 | |
1536 | switch (numberCharacters) { |
1537 | case 1: |
1538 | // Use 32bit width of allCharacters since Yarr counts surrogate pairs as one character with unicode flag. |
1539 | check1(m_checkedOffset - startTermPosition, allCharacters & 0xffffffff); |
1540 | return; |
1541 | case 2: |
1542 | check2(m_checkedOffset - startTermPosition, allCharacters & 0xffffffff, ignoreCaseMask & 0xffffffff); |
1543 | return; |
1544 | #if CPU(X86_64) || CPU(ARM64) |
1545 | case 3: |
1546 | check2(m_checkedOffset - startTermPosition, allCharacters & 0xffffffff, ignoreCaseMask & 0xffffffff); |
1547 | check1(m_checkedOffset - startTermPosition - 2, (allCharacters >> 32) & 0xffff); |
1548 | return; |
1549 | case 4: |
1550 | check4(m_checkedOffset - startTermPosition, allCharacters, ignoreCaseMask); |
1551 | return; |
1552 | #endif |
1553 | } |
1554 | } |
1555 | } |
1556 | void backtrackPatternCharacterOnce(size_t opIndex) |
1557 | { |
1558 | backtrackTermDefault(opIndex); |
1559 | } |
1560 | |
1561 | void generatePatternCharacterFixed(size_t opIndex) |
1562 | { |
1563 | YarrOp& op = m_ops[opIndex]; |
1564 | PatternTerm* term = op.m_term; |
1565 | UChar32 ch = term->patternCharacter; |
1566 | |
1567 | const RegisterID character = regT0; |
1568 | const RegisterID countRegister = regT1; |
1569 | |
1570 | if (m_decodeSurrogatePairs) |
1571 | op.m_jumps.append(jumpIfNoAvailableInput()); |
1572 | |
1573 | move(index, countRegister); |
1574 | Checked<unsigned> scaledMaxCount = term->quantityMaxCount; |
1575 | scaledMaxCount *= U_IS_BMP(ch) ? 1 : 2; |
1576 | sub32(Imm32(scaledMaxCount.unsafeGet()), countRegister); |
1577 | |
1578 | Label loop(this); |
1579 | readCharacter(m_checkedOffset - term->inputPosition - scaledMaxCount, character, countRegister); |
1580 | // For case-insesitive compares, non-ascii characters that have different |
1581 | // upper & lower case representations are converted to a character class. |
1582 | ASSERT(!m_pattern.ignoreCase() || isASCIIAlpha(ch) || isCanonicallyUnique(ch, m_canonicalMode)); |
1583 | if (m_pattern.ignoreCase() && isASCIIAlpha(ch)) { |
1584 | or32(TrustedImm32(0x20), character); |
1585 | ch |= 0x20; |
1586 | } |
1587 | |
1588 | op.m_jumps.append(branch32(NotEqual, character, Imm32(ch))); |
1589 | #ifdef JIT_UNICODE_EXPRESSIONS |
1590 | if (m_decodeSurrogatePairs && !U_IS_BMP(ch)) |
1591 | add32(TrustedImm32(2), countRegister); |
1592 | else |
1593 | #endif |
1594 | add32(TrustedImm32(1), countRegister); |
1595 | branch32(NotEqual, countRegister, index).linkTo(loop, this); |
1596 | } |
1597 | void backtrackPatternCharacterFixed(size_t opIndex) |
1598 | { |
1599 | backtrackTermDefault(opIndex); |
1600 | } |
1601 | |
1602 | void generatePatternCharacterGreedy(size_t opIndex) |
1603 | { |
1604 | YarrOp& op = m_ops[opIndex]; |
1605 | PatternTerm* term = op.m_term; |
1606 | UChar32 ch = term->patternCharacter; |
1607 | |
1608 | const RegisterID character = regT0; |
1609 | const RegisterID countRegister = regT1; |
1610 | |
1611 | move(TrustedImm32(0), countRegister); |
1612 | |
1613 | // Unless have a 16 bit pattern character and an 8 bit string - short circuit |
1614 | if (!(!isLatin1(ch) && (m_charSize == Char8))) { |
1615 | JumpList failures; |
1616 | Label loop(this); |
1617 | failures.append(atEndOfInput()); |
1618 | failures.append(jumpIfCharNotEquals(ch, m_checkedOffset - term->inputPosition, character)); |
1619 | |
1620 | add32(TrustedImm32(1), index); |
1621 | #ifdef JIT_UNICODE_EXPRESSIONS |
1622 | if (m_decodeSurrogatePairs && !U_IS_BMP(ch)) { |
1623 | Jump surrogatePairOk = notAtEndOfInput(); |
1624 | sub32(TrustedImm32(1), index); |
1625 | failures.append(jump()); |
1626 | surrogatePairOk.link(this); |
1627 | add32(TrustedImm32(1), index); |
1628 | } |
1629 | #endif |
1630 | add32(TrustedImm32(1), countRegister); |
1631 | |
1632 | if (term->quantityMaxCount == quantifyInfinite) |
1633 | jump(loop); |
1634 | else |
1635 | branch32(NotEqual, countRegister, Imm32(term->quantityMaxCount.unsafeGet())).linkTo(loop, this); |
1636 | |
1637 | failures.link(this); |
1638 | } |
1639 | op.m_reentry = label(); |
1640 | |
1641 | storeToFrame(countRegister, term->frameLocation + BackTrackInfoPatternCharacter::matchAmountIndex()); |
1642 | } |
1643 | void backtrackPatternCharacterGreedy(size_t opIndex) |
1644 | { |
1645 | YarrOp& op = m_ops[opIndex]; |
1646 | PatternTerm* term = op.m_term; |
1647 | |
1648 | const RegisterID countRegister = regT1; |
1649 | |
1650 | m_backtrackingState.link(this); |
1651 | |
1652 | loadFromFrame(term->frameLocation + BackTrackInfoPatternCharacter::matchAmountIndex(), countRegister); |
1653 | m_backtrackingState.append(branchTest32(Zero, countRegister)); |
1654 | sub32(TrustedImm32(1), countRegister); |
1655 | if (!m_decodeSurrogatePairs || U_IS_BMP(term->patternCharacter)) |
1656 | sub32(TrustedImm32(1), index); |
1657 | else |
1658 | sub32(TrustedImm32(2), index); |
1659 | jump(op.m_reentry); |
1660 | } |
1661 | |
1662 | void generatePatternCharacterNonGreedy(size_t opIndex) |
1663 | { |
1664 | YarrOp& op = m_ops[opIndex]; |
1665 | PatternTerm* term = op.m_term; |
1666 | |
1667 | const RegisterID countRegister = regT1; |
1668 | |
1669 | move(TrustedImm32(0), countRegister); |
1670 | op.m_reentry = label(); |
1671 | storeToFrame(countRegister, term->frameLocation + BackTrackInfoPatternCharacter::matchAmountIndex()); |
1672 | } |
1673 | void backtrackPatternCharacterNonGreedy(size_t opIndex) |
1674 | { |
1675 | YarrOp& op = m_ops[opIndex]; |
1676 | PatternTerm* term = op.m_term; |
1677 | UChar32 ch = term->patternCharacter; |
1678 | |
1679 | const RegisterID character = regT0; |
1680 | const RegisterID countRegister = regT1; |
1681 | |
1682 | m_backtrackingState.link(this); |
1683 | |
1684 | loadFromFrame(term->frameLocation + BackTrackInfoPatternCharacter::matchAmountIndex(), countRegister); |
1685 | |
1686 | // Unless have a 16 bit pattern character and an 8 bit string - short circuit |
1687 | if (!(!isLatin1(ch) && (m_charSize == Char8))) { |
1688 | JumpList nonGreedyFailures; |
1689 | nonGreedyFailures.append(atEndOfInput()); |
1690 | if (term->quantityMaxCount != quantifyInfinite) |
1691 | nonGreedyFailures.append(branch32(Equal, countRegister, Imm32(term->quantityMaxCount.unsafeGet()))); |
1692 | nonGreedyFailures.append(jumpIfCharNotEquals(ch, m_checkedOffset - term->inputPosition, character)); |
1693 | |
1694 | add32(TrustedImm32(1), index); |
1695 | #ifdef JIT_UNICODE_EXPRESSIONS |
1696 | if (m_decodeSurrogatePairs && !U_IS_BMP(ch)) { |
1697 | Jump surrogatePairOk = notAtEndOfInput(); |
1698 | sub32(TrustedImm32(1), index); |
1699 | nonGreedyFailures.append(jump()); |
1700 | surrogatePairOk.link(this); |
1701 | add32(TrustedImm32(1), index); |
1702 | } |
1703 | #endif |
1704 | add32(TrustedImm32(1), countRegister); |
1705 | |
1706 | jump(op.m_reentry); |
1707 | nonGreedyFailures.link(this); |
1708 | } |
1709 | |
1710 | if (m_decodeSurrogatePairs && !U_IS_BMP(ch)) { |
1711 | // subtract countRegister*2 for non-BMP characters |
1712 | lshift32(TrustedImm32(1), countRegister); |
1713 | } |
1714 | |
1715 | sub32(countRegister, index); |
1716 | m_backtrackingState.fallthrough(); |
1717 | } |
1718 | |
1719 | void generateCharacterClassOnce(size_t opIndex) |
1720 | { |
1721 | YarrOp& op = m_ops[opIndex]; |
1722 | PatternTerm* term = op.m_term; |
1723 | |
1724 | const RegisterID character = regT0; |
1725 | |
1726 | if (m_decodeSurrogatePairs) { |
1727 | op.m_jumps.append(jumpIfNoAvailableInput()); |
1728 | storeToFrame(index, term->frameLocation + BackTrackInfoCharacterClass::beginIndex()); |
1729 | } |
1730 | |
1731 | JumpList matchDest; |
1732 | readCharacter(m_checkedOffset - term->inputPosition, character); |
1733 | // If we are matching the "any character" builtin class we only need to read the |
1734 | // character and don't need to match as it will always succeed. |
1735 | if (term->invert() || !term->characterClass->m_anyCharacter) { |
1736 | matchCharacterClass(character, matchDest, term->characterClass); |
1737 | |
1738 | if (term->invert()) |
1739 | op.m_jumps.append(matchDest); |
1740 | else { |
1741 | op.m_jumps.append(jump()); |
1742 | matchDest.link(this); |
1743 | } |
1744 | } |
1745 | #ifdef JIT_UNICODE_EXPRESSIONS |
1746 | if (m_decodeSurrogatePairs && (!term->characterClass->hasOneCharacterSize() || term->invert())) { |
1747 | Jump isBMPChar = branch32(LessThan, character, supplementaryPlanesBase); |
1748 | op.m_jumps.append(atEndOfInput()); |
1749 | add32(TrustedImm32(1), index); |
1750 | isBMPChar.link(this); |
1751 | } |
1752 | #endif |
1753 | } |
1754 | void backtrackCharacterClassOnce(size_t opIndex) |
1755 | { |
1756 | #ifdef JIT_UNICODE_EXPRESSIONS |
1757 | if (m_decodeSurrogatePairs) { |
1758 | YarrOp& op = m_ops[opIndex]; |
1759 | PatternTerm* term = op.m_term; |
1760 | |
1761 | m_backtrackingState.link(this); |
1762 | loadFromFrame(term->frameLocation + BackTrackInfoCharacterClass::beginIndex(), index); |
1763 | m_backtrackingState.fallthrough(); |
1764 | } |
1765 | #endif |
1766 | backtrackTermDefault(opIndex); |
1767 | } |
1768 | |
1769 | void generateCharacterClassFixed(size_t opIndex) |
1770 | { |
1771 | YarrOp& op = m_ops[opIndex]; |
1772 | PatternTerm* term = op.m_term; |
1773 | |
1774 | const RegisterID character = regT0; |
1775 | const RegisterID countRegister = regT1; |
1776 | |
1777 | if (m_decodeSurrogatePairs) |
1778 | op.m_jumps.append(jumpIfNoAvailableInput()); |
1779 | |
1780 | move(index, countRegister); |
1781 | |
1782 | Checked<unsigned> scaledMaxCount = term->quantityMaxCount; |
1783 | |
1784 | #ifdef JIT_UNICODE_EXPRESSIONS |
1785 | if (m_decodeSurrogatePairs && term->characterClass->hasOnlyNonBMPCharacters() && !term->invert()) |
1786 | scaledMaxCount *= 2; |
1787 | #endif |
1788 | sub32(Imm32(scaledMaxCount.unsafeGet()), countRegister); |
1789 | |
1790 | Label loop(this); |
1791 | JumpList matchDest; |
1792 | readCharacter(m_checkedOffset - term->inputPosition - scaledMaxCount, character, countRegister); |
1793 | // If we are matching the "any character" builtin class we only need to read the |
1794 | // character and don't need to match as it will always succeed. |
1795 | if (term->invert() || !term->characterClass->m_anyCharacter) { |
1796 | matchCharacterClass(character, matchDest, term->characterClass); |
1797 | |
1798 | if (term->invert()) |
1799 | op.m_jumps.append(matchDest); |
1800 | else { |
1801 | op.m_jumps.append(jump()); |
1802 | matchDest.link(this); |
1803 | } |
1804 | } |
1805 | |
1806 | #ifdef JIT_UNICODE_EXPRESSIONS |
1807 | if (m_decodeSurrogatePairs) { |
1808 | if (term->isFixedWidthCharacterClass()) |
1809 | add32(TrustedImm32(term->characterClass->hasNonBMPCharacters() ? 2 : 1), countRegister); |
1810 | else { |
1811 | add32(TrustedImm32(1), countRegister); |
1812 | Jump isBMPChar = branch32(LessThan, character, supplementaryPlanesBase); |
1813 | op.m_jumps.append(atEndOfInput()); |
1814 | add32(TrustedImm32(1), countRegister); |
1815 | add32(TrustedImm32(1), index); |
1816 | isBMPChar.link(this); |
1817 | } |
1818 | } else |
1819 | #endif |
1820 | add32(TrustedImm32(1), countRegister); |
1821 | branch32(NotEqual, countRegister, index).linkTo(loop, this); |
1822 | } |
1823 | void backtrackCharacterClassFixed(size_t opIndex) |
1824 | { |
1825 | backtrackTermDefault(opIndex); |
1826 | } |
1827 | |
1828 | void generateCharacterClassGreedy(size_t opIndex) |
1829 | { |
1830 | YarrOp& op = m_ops[opIndex]; |
1831 | PatternTerm* term = op.m_term; |
1832 | |
1833 | const RegisterID character = regT0; |
1834 | const RegisterID countRegister = regT1; |
1835 | |
1836 | if (m_decodeSurrogatePairs && (!term->characterClass->hasOneCharacterSize() || term->invert())) |
1837 | storeToFrame(index, term->frameLocation + BackTrackInfoCharacterClass::beginIndex()); |
1838 | move(TrustedImm32(0), countRegister); |
1839 | |
1840 | JumpList failures; |
1841 | JumpList failuresDecrementIndex; |
1842 | Label loop(this); |
1843 | #ifdef JIT_UNICODE_EXPRESSIONS |
1844 | if (term->isFixedWidthCharacterClass() && term->characterClass->hasNonBMPCharacters()) { |
1845 | move(TrustedImm32(1), character); |
1846 | failures.append(checkNotEnoughInput(character)); |
1847 | } else |
1848 | #endif |
1849 | failures.append(atEndOfInput()); |
1850 | |
1851 | if (term->invert()) { |
1852 | readCharacter(m_checkedOffset - term->inputPosition, character); |
1853 | matchCharacterClass(character, failures, term->characterClass); |
1854 | } else { |
1855 | JumpList matchDest; |
1856 | readCharacter(m_checkedOffset - term->inputPosition, character); |
1857 | // If we are matching the "any character" builtin class for non-unicode patterns, |
1858 | // we only need to read the character and don't need to match as it will always succeed. |
1859 | if (!term->characterClass->m_anyCharacter) { |
1860 | matchCharacterClass(character, matchDest, term->characterClass); |
1861 | failures.append(jump()); |
1862 | } |
1863 | matchDest.link(this); |
1864 | } |
1865 | |
1866 | #ifdef JIT_UNICODE_EXPRESSIONS |
1867 | if (m_decodeSurrogatePairs) |
1868 | advanceIndexAfterCharacterClassTermMatch(term, failuresDecrementIndex, character); |
1869 | else |
1870 | #endif |
1871 | add32(TrustedImm32(1), index); |
1872 | add32(TrustedImm32(1), countRegister); |
1873 | |
1874 | if (term->quantityMaxCount != quantifyInfinite) { |
1875 | branch32(NotEqual, countRegister, Imm32(term->quantityMaxCount.unsafeGet())).linkTo(loop, this); |
1876 | failures.append(jump()); |
1877 | } else |
1878 | jump(loop); |
1879 | |
1880 | if (!failuresDecrementIndex.empty()) { |
1881 | failuresDecrementIndex.link(this); |
1882 | sub32(TrustedImm32(1), index); |
1883 | } |
1884 | |
1885 | failures.link(this); |
1886 | op.m_reentry = label(); |
1887 | |
1888 | storeToFrame(countRegister, term->frameLocation + BackTrackInfoCharacterClass::matchAmountIndex()); |
1889 | } |
1890 | void backtrackCharacterClassGreedy(size_t opIndex) |
1891 | { |
1892 | YarrOp& op = m_ops[opIndex]; |
1893 | PatternTerm* term = op.m_term; |
1894 | |
1895 | const RegisterID countRegister = regT1; |
1896 | |
1897 | m_backtrackingState.link(this); |
1898 | |
1899 | loadFromFrame(term->frameLocation + BackTrackInfoCharacterClass::matchAmountIndex(), countRegister); |
1900 | m_backtrackingState.append(branchTest32(Zero, countRegister)); |
1901 | sub32(TrustedImm32(1), countRegister); |
1902 | storeToFrame(countRegister, term->frameLocation + BackTrackInfoCharacterClass::matchAmountIndex()); |
1903 | |
1904 | if (!m_decodeSurrogatePairs) |
1905 | sub32(TrustedImm32(1), index); |
1906 | else if (term->isFixedWidthCharacterClass()) |
1907 | sub32(TrustedImm32(term->characterClass->hasNonBMPCharacters() ? 2 : 1), index); |
1908 | else { |
1909 | // Rematch one less |
1910 | const RegisterID character = regT0; |
1911 | |
1912 | loadFromFrame(term->frameLocation + BackTrackInfoCharacterClass::beginIndex(), index); |
1913 | |
1914 | Label rematchLoop(this); |
1915 | Jump doneRematching = branchTest32(Zero, countRegister); |
1916 | |
1917 | readCharacter(m_checkedOffset - term->inputPosition, character); |
1918 | |
1919 | sub32(TrustedImm32(1), countRegister); |
1920 | add32(TrustedImm32(1), index); |
1921 | |
1922 | #ifdef JIT_UNICODE_EXPRESSIONS |
1923 | Jump isBMPChar = branch32(LessThan, character, supplementaryPlanesBase); |
1924 | add32(TrustedImm32(1), index); |
1925 | isBMPChar.link(this); |
1926 | #endif |
1927 | |
1928 | jump(rematchLoop); |
1929 | doneRematching.link(this); |
1930 | |
1931 | loadFromFrame(term->frameLocation + BackTrackInfoCharacterClass::matchAmountIndex(), countRegister); |
1932 | } |
1933 | jump(op.m_reentry); |
1934 | } |
1935 | |
1936 | void generateCharacterClassNonGreedy(size_t opIndex) |
1937 | { |
1938 | YarrOp& op = m_ops[opIndex]; |
1939 | PatternTerm* term = op.m_term; |
1940 | |
1941 | const RegisterID countRegister = regT1; |
1942 | |
1943 | move(TrustedImm32(0), countRegister); |
1944 | op.m_reentry = label(); |
1945 | |
1946 | #ifdef JIT_UNICODE_EXPRESSIONS |
1947 | if (m_decodeSurrogatePairs) { |
1948 | if (!term->characterClass->hasOneCharacterSize() || term->invert()) |
1949 | storeToFrame(index, term->frameLocation + BackTrackInfoCharacterClass::beginIndex()); |
1950 | } |
1951 | #endif |
1952 | |
1953 | storeToFrame(countRegister, term->frameLocation + BackTrackInfoCharacterClass::matchAmountIndex()); |
1954 | } |
1955 | |
1956 | void backtrackCharacterClassNonGreedy(size_t opIndex) |
1957 | { |
1958 | YarrOp& op = m_ops[opIndex]; |
1959 | PatternTerm* term = op.m_term; |
1960 | |
1961 | const RegisterID character = regT0; |
1962 | const RegisterID countRegister = regT1; |
1963 | |
1964 | JumpList nonGreedyFailures; |
1965 | JumpList nonGreedyFailuresDecrementIndex; |
1966 | |
1967 | m_backtrackingState.link(this); |
1968 | |
1969 | #ifdef JIT_UNICODE_EXPRESSIONS |
1970 | if (m_decodeSurrogatePairs) { |
1971 | if (!term->characterClass->hasOneCharacterSize() || term->invert()) |
1972 | loadFromFrame(term->frameLocation + BackTrackInfoCharacterClass::beginIndex(), index); |
1973 | } |
1974 | #endif |
1975 | |
1976 | loadFromFrame(term->frameLocation + BackTrackInfoCharacterClass::matchAmountIndex(), countRegister); |
1977 | |
1978 | nonGreedyFailures.append(atEndOfInput()); |
1979 | nonGreedyFailures.append(branch32(Equal, countRegister, Imm32(term->quantityMaxCount.unsafeGet()))); |
1980 | |
1981 | JumpList matchDest; |
1982 | readCharacter(m_checkedOffset - term->inputPosition, character); |
1983 | // If we are matching the "any character" builtin class for non-unicode patterns, |
1984 | // we only need to read the character and don't need to match as it will always succeed. |
1985 | if (term->invert() || !term->characterClass->m_anyCharacter) { |
1986 | matchCharacterClass(character, matchDest, term->characterClass); |
1987 | |
1988 | if (term->invert()) |
1989 | nonGreedyFailures.append(matchDest); |
1990 | else { |
1991 | nonGreedyFailures.append(jump()); |
1992 | matchDest.link(this); |
1993 | } |
1994 | } |
1995 | |
1996 | #ifdef JIT_UNICODE_EXPRESSIONS |
1997 | if (m_decodeSurrogatePairs) |
1998 | advanceIndexAfterCharacterClassTermMatch(term, nonGreedyFailuresDecrementIndex, character); |
1999 | else |
2000 | #endif |
2001 | add32(TrustedImm32(1), index); |
2002 | add32(TrustedImm32(1), countRegister); |
2003 | |
2004 | jump(op.m_reentry); |
2005 | |
2006 | if (!nonGreedyFailuresDecrementIndex.empty()) { |
2007 | nonGreedyFailuresDecrementIndex.link(this); |
2008 | breakpoint(); |
2009 | } |
2010 | nonGreedyFailures.link(this); |
2011 | sub32(countRegister, index); |
2012 | m_backtrackingState.fallthrough(); |
2013 | } |
2014 | |
2015 | void generateDotStarEnclosure(size_t opIndex) |
2016 | { |
2017 | YarrOp& op = m_ops[opIndex]; |
2018 | PatternTerm* term = op.m_term; |
2019 | |
2020 | const RegisterID character = regT0; |
2021 | const RegisterID matchPos = regT1; |
2022 | JumpList foundBeginningNewLine; |
2023 | JumpList saveStartIndex; |
2024 | JumpList foundEndingNewLine; |
2025 | |
2026 | if (m_pattern.dotAll()) { |
2027 | move(TrustedImm32(0), matchPos); |
2028 | setMatchStart(matchPos); |
2029 | move(length, index); |
2030 | return; |
2031 | } |
2032 | |
2033 | ASSERT(!m_pattern.m_body->m_hasFixedSize); |
2034 | getMatchStart(matchPos); |
2035 | |
2036 | saveStartIndex.append(branch32(BelowOrEqual, matchPos, initialStart)); |
2037 | Label findBOLLoop(this); |
2038 | sub32(TrustedImm32(1), matchPos); |
2039 | if (m_charSize == Char8) |
2040 | load8(BaseIndex(input, matchPos, TimesOne, 0), character); |
2041 | else |
2042 | load16(BaseIndex(input, matchPos, TimesTwo, 0), character); |
2043 | matchCharacterClass(character, foundBeginningNewLine, m_pattern.newlineCharacterClass()); |
2044 | |
2045 | branch32(Above, matchPos, initialStart).linkTo(findBOLLoop, this); |
2046 | saveStartIndex.append(jump()); |
2047 | |
2048 | foundBeginningNewLine.link(this); |
2049 | add32(TrustedImm32(1), matchPos); // Advance past newline |
2050 | saveStartIndex.link(this); |
2051 | |
2052 | if (!m_pattern.multiline() && term->anchors.bolAnchor) |
2053 | op.m_jumps.append(branchTest32(NonZero, matchPos)); |
2054 | |
2055 | ASSERT(!m_pattern.m_body->m_hasFixedSize); |
2056 | setMatchStart(matchPos); |
2057 | |
2058 | move(index, matchPos); |
2059 | |
2060 | Label findEOLLoop(this); |
2061 | foundEndingNewLine.append(branch32(Equal, matchPos, length)); |
2062 | if (m_charSize == Char8) |
2063 | load8(BaseIndex(input, matchPos, TimesOne, 0), character); |
2064 | else |
2065 | load16(BaseIndex(input, matchPos, TimesTwo, 0), character); |
2066 | matchCharacterClass(character, foundEndingNewLine, m_pattern.newlineCharacterClass()); |
2067 | add32(TrustedImm32(1), matchPos); |
2068 | jump(findEOLLoop); |
2069 | |
2070 | foundEndingNewLine.link(this); |
2071 | |
2072 | if (!m_pattern.multiline() && term->anchors.eolAnchor) |
2073 | op.m_jumps.append(branch32(NotEqual, matchPos, length)); |
2074 | |
2075 | move(matchPos, index); |
2076 | } |
2077 | |
2078 | void backtrackDotStarEnclosure(size_t opIndex) |
2079 | { |
2080 | backtrackTermDefault(opIndex); |
2081 | } |
2082 | |
2083 | // Code generation/backtracking for simple terms |
2084 | // (pattern characters, character classes, and assertions). |
2085 | // These methods farm out work to the set of functions above. |
2086 | void generateTerm(size_t opIndex) |
2087 | { |
2088 | YarrOp& op = m_ops[opIndex]; |
2089 | PatternTerm* term = op.m_term; |
2090 | |
2091 | switch (term->type) { |
2092 | case PatternTerm::TypePatternCharacter: |
2093 | switch (term->quantityType) { |
2094 | case QuantifierFixedCount: |
2095 | if (term->quantityMaxCount == 1) |
2096 | generatePatternCharacterOnce(opIndex); |
2097 | else |
2098 | generatePatternCharacterFixed(opIndex); |
2099 | break; |
2100 | case QuantifierGreedy: |
2101 | generatePatternCharacterGreedy(opIndex); |
2102 | break; |
2103 | case QuantifierNonGreedy: |
2104 | generatePatternCharacterNonGreedy(opIndex); |
2105 | break; |
2106 | } |
2107 | break; |
2108 | |
2109 | case PatternTerm::TypeCharacterClass: |
2110 | switch (term->quantityType) { |
2111 | case QuantifierFixedCount: |
2112 | if (term->quantityMaxCount == 1) |
2113 | generateCharacterClassOnce(opIndex); |
2114 | else |
2115 | generateCharacterClassFixed(opIndex); |
2116 | break; |
2117 | case QuantifierGreedy: |
2118 | generateCharacterClassGreedy(opIndex); |
2119 | break; |
2120 | case QuantifierNonGreedy: |
2121 | generateCharacterClassNonGreedy(opIndex); |
2122 | break; |
2123 | } |
2124 | break; |
2125 | |
2126 | case PatternTerm::TypeAssertionBOL: |
2127 | generateAssertionBOL(opIndex); |
2128 | break; |
2129 | |
2130 | case PatternTerm::TypeAssertionEOL: |
2131 | generateAssertionEOL(opIndex); |
2132 | break; |
2133 | |
2134 | case PatternTerm::TypeAssertionWordBoundary: |
2135 | generateAssertionWordBoundary(opIndex); |
2136 | break; |
2137 | |
2138 | case PatternTerm::TypeForwardReference: |
2139 | m_failureReason = JITFailureReason::ForwardReference; |
2140 | break; |
2141 | |
2142 | case PatternTerm::TypeParenthesesSubpattern: |
2143 | case PatternTerm::TypeParentheticalAssertion: |
2144 | RELEASE_ASSERT_NOT_REACHED(); |
2145 | |
2146 | case PatternTerm::TypeBackReference: |
2147 | #if ENABLE(YARR_JIT_BACKREFERENCES) |
2148 | generateBackReference(opIndex); |
2149 | #else |
2150 | m_failureReason = JITFailureReason::BackReference; |
2151 | #endif |
2152 | break; |
2153 | case PatternTerm::TypeDotStarEnclosure: |
2154 | generateDotStarEnclosure(opIndex); |
2155 | break; |
2156 | } |
2157 | } |
2158 | void backtrackTerm(size_t opIndex) |
2159 | { |
2160 | YarrOp& op = m_ops[opIndex]; |
2161 | PatternTerm* term = op.m_term; |
2162 | |
2163 | switch (term->type) { |
2164 | case PatternTerm::TypePatternCharacter: |
2165 | switch (term->quantityType) { |
2166 | case QuantifierFixedCount: |
2167 | if (term->quantityMaxCount == 1) |
2168 | backtrackPatternCharacterOnce(opIndex); |
2169 | else |
2170 | backtrackPatternCharacterFixed(opIndex); |
2171 | break; |
2172 | case QuantifierGreedy: |
2173 | backtrackPatternCharacterGreedy(opIndex); |
2174 | break; |
2175 | case QuantifierNonGreedy: |
2176 | backtrackPatternCharacterNonGreedy(opIndex); |
2177 | break; |
2178 | } |
2179 | break; |
2180 | |
2181 | case PatternTerm::TypeCharacterClass: |
2182 | switch (term->quantityType) { |
2183 | case QuantifierFixedCount: |
2184 | if (term->quantityMaxCount == 1) |
2185 | backtrackCharacterClassOnce(opIndex); |
2186 | else |
2187 | backtrackCharacterClassFixed(opIndex); |
2188 | break; |
2189 | case QuantifierGreedy: |
2190 | backtrackCharacterClassGreedy(opIndex); |
2191 | break; |
2192 | case QuantifierNonGreedy: |
2193 | backtrackCharacterClassNonGreedy(opIndex); |
2194 | break; |
2195 | } |
2196 | break; |
2197 | |
2198 | case PatternTerm::TypeAssertionBOL: |
2199 | backtrackAssertionBOL(opIndex); |
2200 | break; |
2201 | |
2202 | case PatternTerm::TypeAssertionEOL: |
2203 | backtrackAssertionEOL(opIndex); |
2204 | break; |
2205 | |
2206 | case PatternTerm::TypeAssertionWordBoundary: |
2207 | backtrackAssertionWordBoundary(opIndex); |
2208 | break; |
2209 | |
2210 | case PatternTerm::TypeForwardReference: |
2211 | m_failureReason = JITFailureReason::ForwardReference; |
2212 | break; |
2213 | |
2214 | case PatternTerm::TypeParenthesesSubpattern: |
2215 | case PatternTerm::TypeParentheticalAssertion: |
2216 | RELEASE_ASSERT_NOT_REACHED(); |
2217 | |
2218 | case PatternTerm::TypeBackReference: |
2219 | #if ENABLE(YARR_JIT_BACKREFERENCES) |
2220 | backtrackBackReference(opIndex); |
2221 | #else |
2222 | m_failureReason = JITFailureReason::BackReference; |
2223 | #endif |
2224 | break; |
2225 | |
2226 | case PatternTerm::TypeDotStarEnclosure: |
2227 | backtrackDotStarEnclosure(opIndex); |
2228 | break; |
2229 | } |
2230 | } |
2231 | |
2232 | void generate() |
2233 | { |
2234 | // Forwards generate the matching code. |
2235 | ASSERT(m_ops.size()); |
2236 | size_t opIndex = 0; |
2237 | |
2238 | do { |
2239 | if (m_disassembler) |
2240 | m_disassembler->setForGenerate(opIndex, label()); |
2241 | |
2242 | YarrOp& op = m_ops[opIndex]; |
2243 | switch (op.m_op) { |
2244 | |
2245 | case OpTerm: |
2246 | generateTerm(opIndex); |
2247 | break; |
2248 | |
2249 | // OpBodyAlternativeBegin/Next/End |
2250 | // |
2251 | // These nodes wrap the set of alternatives in the body of the regular expression. |
2252 | // There may be either one or two chains of OpBodyAlternative nodes, one representing |
2253 | // the 'once through' sequence of alternatives (if any exist), and one representing |
2254 | // the repeating alternatives (again, if any exist). |
2255 | // |
2256 | // Upon normal entry to the Begin alternative, we will check that input is available. |
2257 | // Reentry to the Begin alternative will take place after the check has taken place, |
2258 | // and will assume that the input position has already been progressed as appropriate. |
2259 | // |
2260 | // Entry to subsequent Next/End alternatives occurs when the prior alternative has |
2261 | // successfully completed a match - return a success state from JIT code. |
2262 | // |
2263 | // Next alternatives allow for reentry optimized to suit backtracking from its |
2264 | // preceding alternative. It expects the input position to still be set to a position |
2265 | // appropriate to its predecessor, and it will only perform an input check if the |
2266 | // predecessor had a minimum size less than its own. |
2267 | // |
2268 | // In the case 'once through' expressions, the End node will also have a reentry |
2269 | // point to jump to when the last alternative fails. Again, this expects the input |
2270 | // position to still reflect that expected by the prior alternative. |
2271 | case OpBodyAlternativeBegin: { |
2272 | PatternAlternative* alternative = op.m_alternative; |
2273 | |
2274 | // Upon entry at the head of the set of alternatives, check if input is available |
2275 | // to run the first alternative. (This progresses the input position). |
2276 | op.m_jumps.append(jumpIfNoAvailableInput(alternative->m_minimumSize)); |
2277 | // We will reenter after the check, and assume the input position to have been |
2278 | // set as appropriate to this alternative. |
2279 | op.m_reentry = label(); |
2280 | |
2281 | m_checkedOffset += alternative->m_minimumSize; |
2282 | break; |
2283 | } |
2284 | case OpBodyAlternativeNext: |
2285 | case OpBodyAlternativeEnd: { |
2286 | PatternAlternative* priorAlternative = m_ops[op.m_previousOp].m_alternative; |
2287 | PatternAlternative* alternative = op.m_alternative; |
2288 | |
2289 | // If we get here, the prior alternative matched - return success. |
2290 | |
2291 | // Adjust the stack pointer to remove the pattern's frame. |
2292 | removeCallFrame(); |
2293 | |
2294 | // Load appropriate values into the return register and the first output |
2295 | // slot, and return. In the case of pattern with a fixed size, we will |
2296 | // not have yet set the value in the first |
2297 | ASSERT(index != returnRegister); |
2298 | if (m_pattern.m_body->m_hasFixedSize) { |
2299 | move(index, returnRegister); |
2300 | if (priorAlternative->m_minimumSize) |
2301 | sub32(Imm32(priorAlternative->m_minimumSize), returnRegister); |
2302 | if (compileMode == IncludeSubpatterns) |
2303 | store32(returnRegister, output); |
2304 | } else |
2305 | getMatchStart(returnRegister); |
2306 | if (compileMode == IncludeSubpatterns) |
2307 | store32(index, Address(output, 4)); |
2308 | move(index, returnRegister2); |
2309 | |
2310 | generateReturn(); |
2311 | |
2312 | // This is the divide between the tail of the prior alternative, above, and |
2313 | // the head of the subsequent alternative, below. |
2314 | |
2315 | if (op.m_op == OpBodyAlternativeNext) { |
2316 | // This is the reentry point for the Next alternative. We expect any code |
2317 | // that jumps here to do so with the input position matching that of the |
2318 | // PRIOR alteranative, and we will only check input availability if we |
2319 | // need to progress it forwards. |
2320 | op.m_reentry = label(); |
2321 | if (alternative->m_minimumSize > priorAlternative->m_minimumSize) { |
2322 | add32(Imm32(alternative->m_minimumSize - priorAlternative->m_minimumSize), index); |
2323 | op.m_jumps.append(jumpIfNoAvailableInput()); |
2324 | } else if (priorAlternative->m_minimumSize > alternative->m_minimumSize) |
2325 | sub32(Imm32(priorAlternative->m_minimumSize - alternative->m_minimumSize), index); |
2326 | } else if (op.m_nextOp == notFound) { |
2327 | // This is the reentry point for the End of 'once through' alternatives, |
2328 | // jumped to when the last alternative fails to match. |
2329 | op.m_reentry = label(); |
2330 | sub32(Imm32(priorAlternative->m_minimumSize), index); |
2331 | } |
2332 | |
2333 | if (op.m_op == OpBodyAlternativeNext) |
2334 | m_checkedOffset += alternative->m_minimumSize; |
2335 | m_checkedOffset -= priorAlternative->m_minimumSize; |
2336 | break; |
2337 | } |
2338 | |
2339 | // OpSimpleNestedAlternativeBegin/Next/End |
2340 | // OpNestedAlternativeBegin/Next/End |
2341 | // |
2342 | // These nodes are used to handle sets of alternatives that are nested within |
2343 | // subpatterns and parenthetical assertions. The 'simple' forms are used where |
2344 | // we do not need to be able to backtrack back into any alternative other than |
2345 | // the last, the normal forms allow backtracking into any alternative. |
2346 | // |
2347 | // Each Begin/Next node is responsible for planting an input check to ensure |
2348 | // sufficient input is available on entry. Next nodes additionally need to |
2349 | // jump to the end - Next nodes use the End node's m_jumps list to hold this |
2350 | // set of jumps. |
2351 | // |
2352 | // In the non-simple forms, successful alternative matches must store a |
2353 | // 'return address' using a DataLabelPtr, used to store the address to jump |
2354 | // to when backtracking, to get to the code for the appropriate alternative. |
2355 | case OpSimpleNestedAlternativeBegin: |
2356 | case OpNestedAlternativeBegin: { |
2357 | PatternTerm* term = op.m_term; |
2358 | PatternAlternative* alternative = op.m_alternative; |
2359 | PatternDisjunction* disjunction = term->parentheses.disjunction; |
2360 | |
2361 | // Calculate how much input we need to check for, and if non-zero check. |
2362 | op.m_checkAdjust = Checked<unsigned>(alternative->m_minimumSize); |
2363 | if ((term->quantityType == QuantifierFixedCount) && (term->type != PatternTerm::TypeParentheticalAssertion)) |
2364 | op.m_checkAdjust -= disjunction->m_minimumSize; |
2365 | if (op.m_checkAdjust) |
2366 | op.m_jumps.append(jumpIfNoAvailableInput(op.m_checkAdjust.unsafeGet())); |
2367 | |
2368 | m_checkedOffset += op.m_checkAdjust; |
2369 | break; |
2370 | } |
2371 | case OpSimpleNestedAlternativeNext: |
2372 | case OpNestedAlternativeNext: { |
2373 | PatternTerm* term = op.m_term; |
2374 | PatternAlternative* alternative = op.m_alternative; |
2375 | PatternDisjunction* disjunction = term->parentheses.disjunction; |
2376 | |
2377 | // In the non-simple case, store a 'return address' so we can backtrack correctly. |
2378 | if (op.m_op == OpNestedAlternativeNext) { |
2379 | unsigned parenthesesFrameLocation = term->frameLocation; |
2380 | op.m_returnAddress = storeToFrameWithPatch(parenthesesFrameLocation + BackTrackInfoParentheses::returnAddressIndex()); |
2381 | } |
2382 | |
2383 | if (term->quantityType != QuantifierFixedCount && !m_ops[op.m_previousOp].m_alternative->m_minimumSize) { |
2384 | // If the previous alternative matched without consuming characters then |
2385 | // backtrack to try to match while consumming some input. |
2386 | op.m_zeroLengthMatch = branch32(Equal, index, Address(stackPointerRegister, term->frameLocation * sizeof(void*))); |
2387 | } |
2388 | |
2389 | // If we reach here then the last alternative has matched - jump to the |
2390 | // End node, to skip over any further alternatives. |
2391 | // |
2392 | // FIXME: this is logically O(N^2) (though N can be expected to be very |
2393 | // small). We could avoid this either by adding an extra jump to the JIT |
2394 | // data structures, or by making backtracking code that jumps to Next |
2395 | // alternatives are responsible for checking that input is available (if |
2396 | // we didn't need to plant the input checks, then m_jumps would be free). |
2397 | YarrOp* endOp = &m_ops[op.m_nextOp]; |
2398 | while (endOp->m_nextOp != notFound) { |
2399 | ASSERT(endOp->m_op == OpSimpleNestedAlternativeNext || endOp->m_op == OpNestedAlternativeNext); |
2400 | endOp = &m_ops[endOp->m_nextOp]; |
2401 | } |
2402 | ASSERT(endOp->m_op == OpSimpleNestedAlternativeEnd || endOp->m_op == OpNestedAlternativeEnd); |
2403 | endOp->m_jumps.append(jump()); |
2404 | |
2405 | // This is the entry point for the next alternative. |
2406 | op.m_reentry = label(); |
2407 | |
2408 | // Calculate how much input we need to check for, and if non-zero check. |
2409 | op.m_checkAdjust = alternative->m_minimumSize; |
2410 | if ((term->quantityType == QuantifierFixedCount) && (term->type != PatternTerm::TypeParentheticalAssertion)) |
2411 | op.m_checkAdjust -= disjunction->m_minimumSize; |
2412 | if (op.m_checkAdjust) |
2413 | op.m_jumps.append(jumpIfNoAvailableInput(op.m_checkAdjust.unsafeGet())); |
2414 | |
2415 | YarrOp& lastOp = m_ops[op.m_previousOp]; |
2416 | m_checkedOffset -= lastOp.m_checkAdjust; |
2417 | m_checkedOffset += op.m_checkAdjust; |
2418 | break; |
2419 | } |
2420 | case OpSimpleNestedAlternativeEnd: |
2421 | case OpNestedAlternativeEnd: { |
2422 | PatternTerm* term = op.m_term; |
2423 | |
2424 | // In the non-simple case, store a 'return address' so we can backtrack correctly. |
2425 | if (op.m_op == OpNestedAlternativeEnd) { |
2426 | unsigned parenthesesFrameLocation = term->frameLocation; |
2427 | op.m_returnAddress = storeToFrameWithPatch(parenthesesFrameLocation + BackTrackInfoParentheses::returnAddressIndex()); |
2428 | } |
2429 | |
2430 | if (term->quantityType != QuantifierFixedCount && !m_ops[op.m_previousOp].m_alternative->m_minimumSize) { |
2431 | // If the previous alternative matched without consuming characters then |
2432 | // backtrack to try to match while consumming some input. |
2433 | op.m_zeroLengthMatch = branch32(Equal, index, Address(stackPointerRegister, term->frameLocation * sizeof(void*))); |
2434 | } |
2435 | |
2436 | // If this set of alternatives contains more than one alternative, |
2437 | // then the Next nodes will have planted jumps to the End, and added |
2438 | // them to this node's m_jumps list. |
2439 | op.m_jumps.link(this); |
2440 | op.m_jumps.clear(); |
2441 | |
2442 | YarrOp& lastOp = m_ops[op.m_previousOp]; |
2443 | m_checkedOffset -= lastOp.m_checkAdjust; |
2444 | break; |
2445 | } |
2446 | |
2447 | // OpParenthesesSubpatternOnceBegin/End |
2448 | // |
2449 | // These nodes support (optionally) capturing subpatterns, that have a |
2450 | // quantity count of 1 (this covers fixed once, and ?/?? quantifiers). |
2451 | case OpParenthesesSubpatternOnceBegin: { |
2452 | PatternTerm* term = op.m_term; |
2453 | unsigned parenthesesFrameLocation = term->frameLocation; |
2454 | const RegisterID indexTemporary = regT0; |
2455 | ASSERT(term->quantityMaxCount == 1); |
2456 | |
2457 | // Upon entry to a Greedy quantified set of parenthese store the index. |
2458 | // We'll use this for two purposes: |
2459 | // - To indicate which iteration we are on of mathing the remainder of |
2460 | // the expression after the parentheses - the first, including the |
2461 | // match within the parentheses, or the second having skipped over them. |
2462 | // - To check for empty matches, which must be rejected. |
2463 | // |
2464 | // At the head of a NonGreedy set of parentheses we'll immediately set the |
2465 | // value on the stack to -1 (indicating a match skipping the subpattern), |
2466 | // and plant a jump to the end. We'll also plant a label to backtrack to |
2467 | // to reenter the subpattern later, with a store to set up index on the |
2468 | // second iteration. |
2469 | // |
2470 | // FIXME: for capturing parens, could use the index in the capture array? |
2471 | if (term->quantityType == QuantifierGreedy) |
2472 | storeToFrame(index, parenthesesFrameLocation + BackTrackInfoParenthesesOnce::beginIndex()); |
2473 | else if (term->quantityType == QuantifierNonGreedy) { |
2474 | storeToFrame(TrustedImm32(-1), parenthesesFrameLocation + BackTrackInfoParenthesesOnce::beginIndex()); |
2475 | op.m_jumps.append(jump()); |
2476 | op.m_reentry = label(); |
2477 | storeToFrame(index, parenthesesFrameLocation + BackTrackInfoParenthesesOnce::beginIndex()); |
2478 | } |
2479 | |
2480 | // If the parenthese are capturing, store the starting index value to the |
2481 | // captures array, offsetting as necessary. |
2482 | // |
2483 | // FIXME: could avoid offsetting this value in JIT code, apply |
2484 | // offsets only afterwards, at the point the results array is |
2485 | // being accessed. |
2486 | if (term->capture() && compileMode == IncludeSubpatterns) { |
2487 | unsigned inputOffset = (m_checkedOffset - term->inputPosition).unsafeGet(); |
2488 | if (term->quantityType == QuantifierFixedCount) |
2489 | inputOffset += term->parentheses.disjunction->m_minimumSize; |
2490 | if (inputOffset) { |
2491 | move(index, indexTemporary); |
2492 | sub32(Imm32(inputOffset), indexTemporary); |
2493 | setSubpatternStart(indexTemporary, term->parentheses.subpatternId); |
2494 | } else |
2495 | setSubpatternStart(index, term->parentheses.subpatternId); |
2496 | } |
2497 | break; |
2498 | } |
2499 | case OpParenthesesSubpatternOnceEnd: { |
2500 | PatternTerm* term = op.m_term; |
2501 | const RegisterID indexTemporary = regT0; |
2502 | ASSERT(term->quantityMaxCount == 1); |
2503 | |
2504 | // If the nested alternative matched without consuming any characters, punt this back to the interpreter. |
2505 | // FIXME: <https://bugs.webkit.org/show_bug.cgi?id=200786> Add ability for the YARR JIT to properly |
2506 | // handle nested expressions that can match without consuming characters |
2507 | if (term->quantityType != QuantifierFixedCount && !term->parentheses.disjunction->m_minimumSize) |
2508 | m_abortExecution.append(branch32(Equal, index, Address(stackPointerRegister, term->frameLocation * sizeof(void*)))); |
2509 | |
2510 | // If the parenthese are capturing, store the ending index value to the |
2511 | // captures array, offsetting as necessary. |
2512 | // |
2513 | // FIXME: could avoid offsetting this value in JIT code, apply |
2514 | // offsets only afterwards, at the point the results array is |
2515 | // being accessed. |
2516 | if (term->capture() && compileMode == IncludeSubpatterns) { |
2517 | unsigned inputOffset = (m_checkedOffset - term->inputPosition).unsafeGet(); |
2518 | if (inputOffset) { |
2519 | move(index, indexTemporary); |
2520 | sub32(Imm32(inputOffset), indexTemporary); |
2521 | setSubpatternEnd(indexTemporary, term->parentheses.subpatternId); |
2522 | } else |
2523 | setSubpatternEnd(index, term->parentheses.subpatternId); |
2524 | } |
2525 | |
2526 | // If the parentheses are quantified Greedy then add a label to jump back |
2527 | // to if we get a failed match from after the parentheses. For NonGreedy |
2528 | // parentheses, link the jump from before the subpattern to here. |
2529 | if (term->quantityType == QuantifierGreedy) |
2530 | op.m_reentry = label(); |
2531 | else if (term->quantityType == QuantifierNonGreedy) { |
2532 | YarrOp& beginOp = m_ops[op.m_previousOp]; |
2533 | beginOp.m_jumps.link(this); |
2534 | } |
2535 | break; |
2536 | } |
2537 | |
2538 | // OpParenthesesSubpatternTerminalBegin/End |
2539 | case OpParenthesesSubpatternTerminalBegin: { |
2540 | PatternTerm* term = op.m_term; |
2541 | ASSERT(term->quantityType == QuantifierGreedy); |
2542 | ASSERT(term->quantityMaxCount == quantifyInfinite); |
2543 | ASSERT(!term->capture()); |
2544 | |
2545 | // Upon entry set a label to loop back to. |
2546 | op.m_reentry = label(); |
2547 | |
2548 | // Store the start index of the current match; we need to reject zero |
2549 | // length matches. |
2550 | storeToFrame(index, term->frameLocation + BackTrackInfoParenthesesTerminal::beginIndex()); |
2551 | break; |
2552 | } |
2553 | case OpParenthesesSubpatternTerminalEnd: { |
2554 | YarrOp& beginOp = m_ops[op.m_previousOp]; |
2555 | PatternTerm* term = op.m_term; |
2556 | |
2557 | // If the nested alternative matched without consuming any characters, punt this back to the interpreter. |
2558 | // FIXME: <https://bugs.webkit.org/show_bug.cgi?id=200786> Add ability for the YARR JIT to properly |
2559 | // handle nested expressions that can match without consuming characters |
2560 | if (term->quantityType != QuantifierFixedCount && !term->parentheses.disjunction->m_minimumSize) |
2561 | m_abortExecution.append(branch32(Equal, index, Address(stackPointerRegister, term->frameLocation * sizeof(void*)))); |
2562 | |
2563 | // We know that the match is non-zero, we can accept it and |
2564 | // loop back up to the head of the subpattern. |
2565 | jump(beginOp.m_reentry); |
2566 | |
2567 | // This is the entry point to jump to when we stop matching - we will |
2568 | // do so once the subpattern cannot match any more. |
2569 | op.m_reentry = label(); |
2570 | break; |
2571 | } |
2572 | |
2573 | // OpParenthesesSubpatternBegin/End |
2574 | // |
2575 | // These nodes support generic subpatterns. |
2576 | case OpParenthesesSubpatternBegin: { |
2577 | #if ENABLE(YARR_JIT_ALL_PARENS_EXPRESSIONS) |
2578 | PatternTerm* term = op.m_term; |
2579 | unsigned parenthesesFrameLocation = term->frameLocation; |
2580 | |
2581 | // Upon entry to a Greedy quantified set of parenthese store the index. |
2582 | // We'll use this for two purposes: |
2583 | // - To indicate which iteration we are on of mathing the remainder of |
2584 | // the expression after the parentheses - the first, including the |
2585 | // match within the parentheses, or the second having skipped over them. |
2586 | // - To check for empty matches, which must be rejected. |
2587 | // |
2588 | // At the head of a NonGreedy set of parentheses we'll immediately set 'begin' |
2589 | // in the backtrack info to -1 (indicating a match skipping the subpattern), |
2590 | // and plant a jump to the end. We'll also plant a label to backtrack to |
2591 | // to reenter the subpattern later, with a store to set 'begin' to current index |
2592 | // on the second iteration. |
2593 | // |
2594 | // FIXME: for capturing parens, could use the index in the capture array? |
2595 | if (term->quantityType == QuantifierGreedy || term->quantityType == QuantifierNonGreedy) { |
2596 | storeToFrame(TrustedImm32(0), parenthesesFrameLocation + BackTrackInfoParentheses::matchAmountIndex()); |
2597 | storeToFrame(TrustedImmPtr(nullptr), parenthesesFrameLocation + BackTrackInfoParentheses::parenContextHeadIndex()); |
2598 | |
2599 | if (term->quantityType == QuantifierNonGreedy) { |
2600 | storeToFrame(TrustedImm32(-1), parenthesesFrameLocation + BackTrackInfoParentheses::beginIndex()); |
2601 | op.m_jumps.append(jump()); |
2602 | } |
2603 | |
2604 | op.m_reentry = label(); |
2605 | RegisterID currParenContextReg = regT0; |
2606 | RegisterID newParenContextReg = regT1; |
2607 | |
2608 | loadFromFrame(parenthesesFrameLocation + BackTrackInfoParentheses::parenContextHeadIndex(), currParenContextReg); |
2609 | allocateParenContext(newParenContextReg); |
2610 | storePtr(currParenContextReg, newParenContextReg); |
2611 | storeToFrame(newParenContextReg, parenthesesFrameLocation + BackTrackInfoParentheses::parenContextHeadIndex()); |
2612 | saveParenContext(newParenContextReg, regT2, term->parentheses.subpatternId, term->parentheses.lastSubpatternId, parenthesesFrameLocation); |
2613 | storeToFrame(index, parenthesesFrameLocation + BackTrackInfoParentheses::beginIndex()); |
2614 | } |
2615 | |
2616 | // If the parenthese are capturing, store the starting index value to the |
2617 | // captures array, offsetting as necessary. |
2618 | // |
2619 | // FIXME: could avoid offsetting this value in JIT code, apply |
2620 | // offsets only afterwards, at the point the results array is |
2621 | // being accessed. |
2622 | if (term->capture() && compileMode == IncludeSubpatterns) { |
2623 | const RegisterID indexTemporary = regT0; |
2624 | unsigned inputOffset = (m_checkedOffset - term->inputPosition).unsafeGet(); |
2625 | if (term->quantityType == QuantifierFixedCount) |
2626 | inputOffset += term->parentheses.disjunction->m_minimumSize; |
2627 | if (inputOffset) { |
2628 | move(index, indexTemporary); |
2629 | sub32(Imm32(inputOffset), indexTemporary); |
2630 | setSubpatternStart(indexTemporary, term->parentheses.subpatternId); |
2631 | } else |
2632 | setSubpatternStart(index, term->parentheses.subpatternId); |
2633 | } |
2634 | #else // !YARR_JIT_ALL_PARENS_EXPRESSIONS |
2635 | RELEASE_ASSERT_NOT_REACHED(); |
2636 | #endif |
2637 | break; |
2638 | } |
2639 | case OpParenthesesSubpatternEnd: { |
2640 | #if ENABLE(YARR_JIT_ALL_PARENS_EXPRESSIONS) |
2641 | PatternTerm* term = op.m_term; |
2642 | unsigned parenthesesFrameLocation = term->frameLocation; |
2643 | |
2644 | // If the nested alternative matched without consuming any characters, punt this back to the interpreter. |
2645 | // FIXME: <https://bugs.webkit.org/show_bug.cgi?id=200786> Add ability for the YARR JIT to properly |
2646 | // handle nested expressions that can match without consuming characters |
2647 | if (term->quantityType != QuantifierFixedCount && !term->parentheses.disjunction->m_minimumSize) |
2648 | m_abortExecution.append(branch32(Equal, index, Address(stackPointerRegister, parenthesesFrameLocation * sizeof(void*)))); |
2649 | |
2650 | const RegisterID countTemporary = regT1; |
2651 | |
2652 | YarrOp& beginOp = m_ops[op.m_previousOp]; |
2653 | loadFromFrame(parenthesesFrameLocation + BackTrackInfoParentheses::matchAmountIndex(), countTemporary); |
2654 | add32(TrustedImm32(1), countTemporary); |
2655 | storeToFrame(countTemporary, parenthesesFrameLocation + BackTrackInfoParentheses::matchAmountIndex()); |
2656 | |
2657 | // If the parenthese are capturing, store the ending index value to the |
2658 | // captures array, offsetting as necessary. |
2659 | // |
2660 | // FIXME: could avoid offsetting this value in JIT code, apply |
2661 | // offsets only afterwards, at the point the results array is |
2662 | // being accessed. |
2663 | if (term->capture() && compileMode == IncludeSubpatterns) { |
2664 | const RegisterID indexTemporary = regT0; |
2665 | |
2666 | unsigned inputOffset = (m_checkedOffset - term->inputPosition).unsafeGet(); |
2667 | if (inputOffset) { |
2668 | move(index, indexTemporary); |
2669 | sub32(Imm32(inputOffset), indexTemporary); |
2670 | setSubpatternEnd(indexTemporary, term->parentheses.subpatternId); |
2671 | } else |
2672 | setSubpatternEnd(index, term->parentheses.subpatternId); |
2673 | } |
2674 | |
2675 | // If the parentheses are quantified Greedy then add a label to jump back |
2676 | // to if we get a failed match from after the parentheses. For NonGreedy |
2677 | // parentheses, link the jump from before the subpattern to here. |
2678 | if (term->quantityType == QuantifierGreedy) { |
2679 | if (term->quantityMaxCount != quantifyInfinite) |
2680 | branch32(Below, countTemporary, Imm32(term->quantityMaxCount.unsafeGet())).linkTo(beginOp.m_reentry, this); |
2681 | else |
2682 | jump(beginOp.m_reentry); |
2683 | |
2684 | op.m_reentry = label(); |
2685 | } else if (term->quantityType == QuantifierNonGreedy) { |
2686 | YarrOp& beginOp = m_ops[op.m_previousOp]; |
2687 | beginOp.m_jumps.link(this); |
2688 | op.m_reentry = label(); |
2689 | } |
2690 | #else // !YARR_JIT_ALL_PARENS_EXPRESSIONS |
2691 | RELEASE_ASSERT_NOT_REACHED(); |
2692 | #endif |
2693 | break; |
2694 | } |
2695 | |
2696 | // OpParentheticalAssertionBegin/End |
2697 | case OpParentheticalAssertionBegin: { |
2698 | PatternTerm* term = op.m_term; |
2699 | |
2700 | // Store the current index - assertions should not update index, so |
2701 | // we will need to restore it upon a successful match. |
2702 | unsigned parenthesesFrameLocation = term->frameLocation; |
2703 | storeToFrame(index, parenthesesFrameLocation + BackTrackInfoParentheticalAssertion::beginIndex()); |
2704 | |
2705 | // Check |
2706 | op.m_checkAdjust = m_checkedOffset - term->inputPosition; |
2707 | if (op.m_checkAdjust) |
2708 | sub32(Imm32(op.m_checkAdjust.unsafeGet()), index); |
2709 | |
2710 | m_checkedOffset -= op.m_checkAdjust; |
2711 | break; |
2712 | } |
2713 | case OpParentheticalAssertionEnd: { |
2714 | PatternTerm* term = op.m_term; |
2715 | |
2716 | // Restore the input index value. |
2717 | unsigned parenthesesFrameLocation = term->frameLocation; |
2718 | loadFromFrame(parenthesesFrameLocation + BackTrackInfoParentheticalAssertion::beginIndex(), index); |
2719 | |
2720 | // If inverted, a successful match of the assertion must be treated |
2721 | // as a failure, so jump to backtracking. |
2722 | if (term->invert()) { |
2723 | op.m_jumps.append(jump()); |
2724 | op.m_reentry = label(); |
2725 | } |
2726 | |
2727 | YarrOp& lastOp = m_ops[op.m_previousOp]; |
2728 | m_checkedOffset += lastOp.m_checkAdjust; |
2729 | break; |
2730 | } |
2731 | |
2732 | case OpMatchFailed: |
2733 | removeCallFrame(); |
2734 | generateFailReturn(); |
2735 | break; |
2736 | } |
2737 | |
2738 | ++opIndex; |
2739 | } while (opIndex < m_ops.size()); |
2740 | } |
2741 | |
2742 | void backtrack() |
2743 | { |
2744 | // Backwards generate the backtracking code. |
2745 | size_t opIndex = m_ops.size(); |
2746 | ASSERT(opIndex); |
2747 | |
2748 | do { |
2749 | --opIndex; |
2750 | |
2751 | if (m_disassembler) |
2752 | m_disassembler->setForBacktrack(opIndex, label()); |
2753 | |
2754 | YarrOp& op = m_ops[opIndex]; |
2755 | switch (op.m_op) { |
2756 | |
2757 | case OpTerm: |
2758 | backtrackTerm(opIndex); |
2759 | break; |
2760 | |
2761 | // OpBodyAlternativeBegin/Next/End |
2762 | // |
2763 | // For each Begin/Next node representing an alternative, we need to decide what to do |
2764 | // in two circumstances: |
2765 | // - If we backtrack back into this node, from within the alternative. |
2766 | // - If the input check at the head of the alternative fails (if this exists). |
2767 | // |
2768 | // We treat these two cases differently since in the former case we have slightly |
2769 | // more information - since we are backtracking out of a prior alternative we know |
2770 | // that at least enough input was available to run it. For example, given the regular |
2771 | // expression /a|b/, if we backtrack out of the first alternative (a failed pattern |
2772 | // character match of 'a'), then we need not perform an additional input availability |
2773 | // check before running the second alternative. |
2774 | // |
2775 | // Backtracking required differs for the last alternative, which in the case of the |
2776 | // repeating set of alternatives must loop. The code generated for the last alternative |
2777 | // will also be used to handle all input check failures from any prior alternatives - |
2778 | // these require similar functionality, in seeking the next available alternative for |
2779 | // which there is sufficient input. |
2780 | // |
2781 | // Since backtracking of all other alternatives simply requires us to link backtracks |
2782 | // to the reentry point for the subsequent alternative, we will only be generating any |
2783 | // code when backtracking the last alternative. |
2784 | case OpBodyAlternativeBegin: |
2785 | case OpBodyAlternativeNext: { |
2786 | PatternAlternative* alternative = op.m_alternative; |
2787 | |
2788 | if (op.m_op == OpBodyAlternativeNext) { |
2789 | PatternAlternative* priorAlternative = m_ops[op.m_previousOp].m_alternative; |
2790 | m_checkedOffset += priorAlternative->m_minimumSize; |
2791 | } |
2792 | m_checkedOffset -= alternative->m_minimumSize; |
2793 | |
2794 | // Is this the last alternative? If not, then if we backtrack to this point we just |
2795 | // need to jump to try to match the next alternative. |
2796 | if (m_ops[op.m_nextOp].m_op != OpBodyAlternativeEnd) { |
2797 | m_backtrackingState.linkTo(m_ops[op.m_nextOp].m_reentry, this); |
2798 | break; |
2799 | } |
2800 | YarrOp& endOp = m_ops[op.m_nextOp]; |
2801 | |
2802 | YarrOp* beginOp = &op; |
2803 | while (beginOp->m_op != OpBodyAlternativeBegin) { |
2804 | ASSERT(beginOp->m_op == OpBodyAlternativeNext); |
2805 | beginOp = &m_ops[beginOp->m_previousOp]; |
2806 | } |
2807 | |
2808 | bool onceThrough = endOp.m_nextOp == notFound; |
2809 | |
2810 | JumpList lastStickyAlternativeFailures; |
2811 | |
2812 | // First, generate code to handle cases where we backtrack out of an attempted match |
2813 | // of the last alternative. If this is a 'once through' set of alternatives then we |
2814 | // have nothing to do - link this straight through to the End. |
2815 | if (onceThrough) |
2816 | m_backtrackingState.linkTo(endOp.m_reentry, this); |
2817 | else { |
2818 | // If we don't need to move the input poistion, and the pattern has a fixed size |
2819 | // (in which case we omit the store of the start index until the pattern has matched) |
2820 | // then we can just link the backtrack out of the last alternative straight to the |
2821 | // head of the first alternative. |
2822 | if (m_pattern.m_body->m_hasFixedSize |
2823 | && (alternative->m_minimumSize > beginOp->m_alternative->m_minimumSize) |
2824 | && (alternative->m_minimumSize - beginOp->m_alternative->m_minimumSize == 1)) |
2825 | m_backtrackingState.linkTo(beginOp->m_reentry, this); |
2826 | else if (m_pattern.sticky() && m_ops[op.m_nextOp].m_op == OpBodyAlternativeEnd) { |
2827 | // It is a sticky pattern and the last alternative failed, jump to the end. |
2828 | m_backtrackingState.takeBacktracksToJumpList(lastStickyAlternativeFailures, this); |
2829 | } else { |
2830 | // We need to generate a trampoline of code to execute before looping back |
2831 | // around to the first alternative. |
2832 | m_backtrackingState.link(this); |
2833 | |
2834 | // No need to advance and retry for a sticky pattern. |
2835 | if (!m_pattern.sticky()) { |
2836 | // If the pattern size is not fixed, then store the start index for use if we match. |
2837 | if (!m_pattern.m_body->m_hasFixedSize) { |
2838 | if (alternative->m_minimumSize == 1) |
2839 | setMatchStart(index); |
2840 | else { |
2841 | move(index, regT0); |
2842 | if (alternative->m_minimumSize) |
2843 | sub32(Imm32(alternative->m_minimumSize - 1), regT0); |
2844 | else |
2845 | add32(TrustedImm32(1), regT0); |
2846 | setMatchStart(regT0); |
2847 | } |
2848 | } |
2849 | |
2850 | // Generate code to loop. Check whether the last alternative is longer than the |
2851 | // first (e.g. /a|xy/ or /a|xyz/). |
2852 | if (alternative->m_minimumSize > beginOp->m_alternative->m_minimumSize) { |
2853 | // We want to loop, and increment input position. If the delta is 1, it is |
2854 | // already correctly incremented, if more than one then decrement as appropriate. |
2855 | unsigned delta = alternative->m_minimumSize - beginOp->m_alternative->m_minimumSize; |
2856 | ASSERT(delta); |
2857 | if (delta != 1) |
2858 | sub32(Imm32(delta - 1), index); |
2859 | jump(beginOp->m_reentry); |
2860 | } else { |
2861 | // If the first alternative has minimum size 0xFFFFFFFFu, then there cannot |
2862 | // be sufficent input available to handle this, so just fall through. |
2863 | unsigned delta = beginOp->m_alternative->m_minimumSize - alternative->m_minimumSize; |
2864 | if (delta != 0xFFFFFFFFu) { |
2865 | // We need to check input because we are incrementing the input. |
2866 | add32(Imm32(delta + 1), index); |
2867 | checkInput().linkTo(beginOp->m_reentry, this); |
2868 | } |
2869 | } |
2870 | } |
2871 | } |
2872 | } |
2873 | |
2874 | // We can reach this point in the code in two ways: |
2875 | // - Fallthrough from the code above (a repeating alternative backtracked out of its |
2876 | // last alternative, and did not have sufficent input to run the first). |
2877 | // - We will loop back up to the following label when a repeating alternative loops, |
2878 | // following a failed input check. |
2879 | // |
2880 | // Either way, we have just failed the input check for the first alternative. |
2881 | Label firstInputCheckFailed(this); |
2882 | |
2883 | // Generate code to handle input check failures from alternatives except the last. |
2884 | // prevOp is the alternative we're handling a bail out from (initially Begin), and |
2885 | // nextOp is the alternative we will be attempting to reenter into. |
2886 | // |
2887 | // We will link input check failures from the forwards matching path back to the code |
2888 | // that can handle them. |
2889 | YarrOp* prevOp = beginOp; |
2890 | YarrOp* nextOp = &m_ops[beginOp->m_nextOp]; |
2891 | while (nextOp->m_op != OpBodyAlternativeEnd) { |
2892 | prevOp->m_jumps.link(this); |
2893 | |
2894 | // We only get here if an input check fails, it is only worth checking again |
2895 | // if the next alternative has a minimum size less than the last. |
2896 | if (prevOp->m_alternative->m_minimumSize > nextOp->m_alternative->m_minimumSize) { |
2897 | // FIXME: if we added an extra label to YarrOp, we could avoid needing to |
2898 | // subtract delta back out, and reduce this code. Should performance test |
2899 | // the benefit of this. |
2900 | unsigned delta = prevOp->m_alternative->m_minimumSize - nextOp->m_alternative->m_minimumSize; |
2901 | sub32(Imm32(delta), index); |
2902 | Jump fail = jumpIfNoAvailableInput(); |
2903 | add32(Imm32(delta), index); |
2904 | jump(nextOp->m_reentry); |
2905 | fail.link(this); |
2906 | } else if (prevOp->m_alternative->m_minimumSize < nextOp->m_alternative->m_minimumSize) |
2907 | add32(Imm32(nextOp->m_alternative->m_minimumSize - prevOp->m_alternative->m_minimumSize), index); |
2908 | prevOp = nextOp; |
2909 | nextOp = &m_ops[nextOp->m_nextOp]; |
2910 | } |
2911 | |
2912 | // We fall through to here if there is insufficient input to run the last alternative. |
2913 | |
2914 | // If there is insufficient input to run the last alternative, then for 'once through' |
2915 | // alternatives we are done - just jump back up into the forwards matching path at the End. |
2916 | if (onceThrough) { |
2917 | op.m_jumps.linkTo(endOp.m_reentry, this); |
2918 | jump(endOp.m_reentry); |
2919 | break; |
2920 | } |
2921 | |
2922 | // For repeating alternatives, link any input check failure from the last alternative to |
2923 | // this point. |
2924 | op.m_jumps.link(this); |
2925 | |
2926 | bool needsToUpdateMatchStart = !m_pattern.m_body->m_hasFixedSize; |
2927 | |
2928 | // Check for cases where input position is already incremented by 1 for the last |
2929 | // alternative (this is particularly useful where the minimum size of the body |
2930 | // disjunction is 0, e.g. /a*|b/). |
2931 | if (needsToUpdateMatchStart && alternative->m_minimumSize == 1) { |
2932 | // index is already incremented by 1, so just store it now! |
2933 | setMatchStart(index); |
2934 | needsToUpdateMatchStart = false; |
2935 | } |
2936 | |
2937 | if (!m_pattern.sticky()) { |
2938 | // Check whether there is sufficient input to loop. Increment the input position by |
2939 | // one, and check. Also add in the minimum disjunction size before checking - there |
2940 | // is no point in looping if we're just going to fail all the input checks around |
2941 | // the next iteration. |
2942 | ASSERT(alternative->m_minimumSize >= m_pattern.m_body->m_minimumSize); |
2943 | if (alternative->m_minimumSize == m_pattern.m_body->m_minimumSize) { |
2944 | // If the last alternative had the same minimum size as the disjunction, |
2945 | // just simply increment input pos by 1, no adjustment based on minimum size. |
2946 | add32(TrustedImm32(1), index); |
2947 | } else { |
2948 | // If the minumum for the last alternative was one greater than than that |
2949 | // for the disjunction, we're already progressed by 1, nothing to do! |
2950 | unsigned delta = (alternative->m_minimumSize - m_pattern.m_body->m_minimumSize) - 1; |
2951 | if (delta) |
2952 | sub32(Imm32(delta), index); |
2953 | } |
2954 | Jump matchFailed = jumpIfNoAvailableInput(); |
2955 | |
2956 | if (needsToUpdateMatchStart) { |
2957 | if (!m_pattern.m_body->m_minimumSize) |
2958 | setMatchStart(index); |
2959 | else { |
2960 | move(index, regT0); |
2961 | sub32(Imm32(m_pattern.m_body->m_minimumSize), regT0); |
2962 | setMatchStart(regT0); |
2963 | } |
2964 | } |
2965 | |
2966 | // Calculate how much more input the first alternative requires than the minimum |
2967 | // for the body as a whole. If no more is needed then we dont need an additional |
2968 | // input check here - jump straight back up to the start of the first alternative. |
2969 | if (beginOp->m_alternative->m_minimumSize == m_pattern.m_body->m_minimumSize) |
2970 | jump(beginOp->m_reentry); |
2971 | else { |
2972 | if (beginOp->m_alternative->m_minimumSize > m_pattern.m_body->m_minimumSize) |
2973 | add32(Imm32(beginOp->m_alternative->m_minimumSize - m_pattern.m_body->m_minimumSize), index); |
2974 | else |
2975 | sub32(Imm32(m_pattern.m_body->m_minimumSize - beginOp->m_alternative->m_minimumSize), index); |
2976 | checkInput().linkTo(beginOp->m_reentry, this); |
2977 | jump(firstInputCheckFailed); |
2978 | } |
2979 | |
2980 | // We jump to here if we iterate to the point that there is insufficient input to |
2981 | // run any matches, and need to return a failure state from JIT code. |
2982 | matchFailed.link(this); |
2983 | } |
2984 | |
2985 | lastStickyAlternativeFailures.link(this); |
2986 | removeCallFrame(); |
2987 | generateFailReturn(); |
2988 | break; |
2989 | } |
2990 | case OpBodyAlternativeEnd: { |
2991 | // We should never backtrack back into a body disjunction. |
2992 | ASSERT(m_backtrackingState.isEmpty()); |
2993 | |
2994 | PatternAlternative* priorAlternative = m_ops[op.m_previousOp].m_alternative; |
2995 | m_checkedOffset += priorAlternative->m_minimumSize; |
2996 | break; |
2997 | } |
2998 | |
2999 | // OpSimpleNestedAlternativeBegin/Next/End |
3000 | // OpNestedAlternativeBegin/Next/End |
3001 | // |
3002 | // Generate code for when we backtrack back out of an alternative into |
3003 | // a Begin or Next node, or when the entry input count check fails. If |
3004 | // there are more alternatives we need to jump to the next alternative, |
3005 | // if not we backtrack back out of the current set of parentheses. |
3006 | // |
3007 | // In the case of non-simple nested assertions we need to also link the |
3008 | // 'return address' appropriately to backtrack back out into the correct |
3009 | // alternative. |
3010 | case OpSimpleNestedAlternativeBegin: |
3011 | case OpSimpleNestedAlternativeNext: |
3012 | case OpNestedAlternativeBegin: |
3013 | case OpNestedAlternativeNext: { |
3014 | YarrOp& nextOp = m_ops[op.m_nextOp]; |
3015 | bool isBegin = op.m_previousOp == notFound; |
3016 | bool isLastAlternative = nextOp.m_nextOp == notFound; |
3017 | ASSERT(isBegin == (op.m_op == OpSimpleNestedAlternativeBegin || op.m_op == OpNestedAlternativeBegin)); |
3018 | ASSERT(isLastAlternative == (nextOp.m_op == OpSimpleNestedAlternativeEnd || nextOp.m_op == OpNestedAlternativeEnd)); |
3019 | |
3020 | // Treat an input check failure the same as a failed match. |
3021 | m_backtrackingState.append(op.m_jumps); |
3022 | |
3023 | // Set the backtracks to jump to the appropriate place. We may need |
3024 | // to link the backtracks in one of three different way depending on |
3025 | // the type of alternative we are dealing with: |
3026 | // - A single alternative, with no simplings. |
3027 | // - The last alternative of a set of two or more. |
3028 | // - An alternative other than the last of a set of two or more. |
3029 | // |
3030 | // In the case of a single alternative on its own, we don't need to |
3031 | // jump anywhere - if the alternative fails to match we can just |
3032 | // continue to backtrack out of the parentheses without jumping. |
3033 | // |
3034 | // In the case of the last alternative in a set of more than one, we |
3035 | // need to jump to return back out to the beginning. We'll do so by |
3036 | // adding a jump to the End node's m_jumps list, and linking this |
3037 | // when we come to generate the Begin node. For alternatives other |
3038 | // than the last, we need to jump to the next alternative. |
3039 | // |
3040 | // If the alternative had adjusted the input position we must link |
3041 | // backtracking to here, correct, and then jump on. If not we can |
3042 | // link the backtracks directly to their destination. |
3043 | if (op.m_checkAdjust) { |
3044 | // Handle the cases where we need to link the backtracks here. |
3045 | m_backtrackingState.link(this); |
3046 | sub32(Imm32(op.m_checkAdjust.unsafeGet()), index); |
3047 | if (!isLastAlternative) { |
3048 | // An alternative that is not the last should jump to its successor. |
3049 | jump(nextOp.m_reentry); |
3050 | } else if (!isBegin) { |
3051 | // The last of more than one alternatives must jump back to the beginning. |
3052 | nextOp.m_jumps.append(jump()); |
3053 | } else { |
3054 | // A single alternative on its own can fall through. |
3055 | m_backtrackingState.fallthrough(); |
3056 | } |
3057 | } else { |
3058 | // Handle the cases where we can link the backtracks directly to their destinations. |
3059 | if (!isLastAlternative) { |
3060 | // An alternative that is not the last should jump to its successor. |
3061 | m_backtrackingState.linkTo(nextOp.m_reentry, this); |
3062 | } else if (!isBegin) { |
3063 | // The last of more than one alternatives must jump back to the beginning. |
3064 | m_backtrackingState.takeBacktracksToJumpList(nextOp.m_jumps, this); |
3065 | } |
3066 | // In the case of a single alternative on its own do nothing - it can fall through. |
3067 | } |
3068 | |
3069 | // If there is a backtrack jump from a zero length match link it here. |
3070 | if (op.m_zeroLengthMatch.isSet()) |
3071 | m_backtrackingState.append(op.m_zeroLengthMatch); |
3072 | |
3073 | // At this point we've handled the backtracking back into this node. |
3074 | // Now link any backtracks that need to jump to here. |
3075 | |
3076 | // For non-simple alternatives, link the alternative's 'return address' |
3077 | // so that we backtrack back out into the previous alternative. |
3078 | if (op.m_op == OpNestedAlternativeNext) |
3079 | m_backtrackingState.append(op.m_returnAddress); |
3080 | |
3081 | // If there is more than one alternative, then the last alternative will |
3082 | // have planted a jump to be linked to the end. This jump was added to the |
3083 | // End node's m_jumps list. If we are back at the beginning, link it here. |
3084 | if (isBegin) { |
3085 | YarrOp* endOp = &m_ops[op.m_nextOp]; |
3086 | while (endOp->m_nextOp != notFound) { |
3087 | ASSERT(endOp->m_op == OpSimpleNestedAlternativeNext || endOp->m_op == OpNestedAlternativeNext); |
3088 | endOp = &m_ops[endOp->m_nextOp]; |
3089 | } |
3090 | ASSERT(endOp->m_op == OpSimpleNestedAlternativeEnd || endOp->m_op == OpNestedAlternativeEnd); |
3091 | m_backtrackingState.append(endOp->m_jumps); |
3092 | } |
3093 | |
3094 | if (!isBegin) { |
3095 | YarrOp& lastOp = m_ops[op.m_previousOp]; |
3096 | m_checkedOffset += lastOp.m_checkAdjust; |
3097 | } |
3098 | m_checkedOffset -= op.m_checkAdjust; |
3099 | break; |
3100 | } |
3101 | case OpSimpleNestedAlternativeEnd: |
3102 | case OpNestedAlternativeEnd: { |
3103 | PatternTerm* term = op.m_term; |
3104 | |
3105 | // If there is a backtrack jump from a zero length match link it here. |
3106 | if (op.m_zeroLengthMatch.isSet()) |
3107 | m_backtrackingState.append(op.m_zeroLengthMatch); |
3108 | |
3109 | // If we backtrack into the end of a simple subpattern do nothing; |
3110 | // just continue through into the last alternative. If we backtrack |
3111 | // into the end of a non-simple set of alterntives we need to jump |
3112 | // to the backtracking return address set up during generation. |
3113 | if (op.m_op == OpNestedAlternativeEnd) { |
3114 | m_backtrackingState.link(this); |
3115 | |
3116 | // Plant a jump to the return address. |
3117 | unsigned parenthesesFrameLocation = term->frameLocation; |
3118 | loadFromFrameAndJump(parenthesesFrameLocation + BackTrackInfoParentheses::returnAddressIndex()); |
3119 | |
3120 | // Link the DataLabelPtr associated with the end of the last |
3121 | // alternative to this point. |
3122 | m_backtrackingState.append(op.m_returnAddress); |
3123 | } |
3124 | |
3125 | YarrOp& lastOp = m_ops[op.m_previousOp]; |
3126 | m_checkedOffset += lastOp.m_checkAdjust; |
3127 | break; |
3128 | } |
3129 | |
3130 | // OpParenthesesSubpatternOnceBegin/End |
3131 | // |
3132 | // When we are backtracking back out of a capturing subpattern we need |
3133 | // to clear the start index in the matches output array, to record that |
3134 | // this subpattern has not been captured. |
3135 | // |
3136 | // When backtracking back out of a Greedy quantified subpattern we need |
3137 | // to catch this, and try running the remainder of the alternative after |
3138 | // the subpattern again, skipping the parentheses. |
3139 | // |
3140 | // Upon backtracking back into a quantified set of parentheses we need to |
3141 | // check whether we were currently skipping the subpattern. If not, we |
3142 | // can backtrack into them, if we were we need to either backtrack back |
3143 | // out of the start of the parentheses, or jump back to the forwards |
3144 | // matching start, depending of whether the match is Greedy or NonGreedy. |
3145 | case OpParenthesesSubpatternOnceBegin: { |
3146 | PatternTerm* term = op.m_term; |
3147 | ASSERT(term->quantityMaxCount == 1); |
3148 | |
3149 | // We only need to backtrack to this point if capturing or greedy. |
3150 | if ((term->capture() && compileMode == IncludeSubpatterns) || term->quantityType == QuantifierGreedy) { |
3151 | m_backtrackingState.link(this); |
3152 | |
3153 | // If capturing, clear the capture (we only need to reset start). |
3154 | if (term->capture() && compileMode == IncludeSubpatterns) |
3155 | clearSubpatternStart(term->parentheses.subpatternId); |
3156 | |
3157 | // If Greedy, jump to the end. |
3158 | if (term->quantityType == QuantifierGreedy) { |
3159 | // Clear the flag in the stackframe indicating we ran through the subpattern. |
3160 | unsigned parenthesesFrameLocation = term->frameLocation; |
3161 | storeToFrame(TrustedImm32(-1), parenthesesFrameLocation + BackTrackInfoParenthesesOnce::beginIndex()); |
3162 | // Jump to after the parentheses, skipping the subpattern. |
3163 | jump(m_ops[op.m_nextOp].m_reentry); |
3164 | // A backtrack from after the parentheses, when skipping the subpattern, |
3165 | // will jump back to here. |
3166 | op.m_jumps.link(this); |
3167 | } |
3168 | |
3169 | m_backtrackingState.fallthrough(); |
3170 | } |
3171 | break; |
3172 | } |
3173 | case OpParenthesesSubpatternOnceEnd: { |
3174 | PatternTerm* term = op.m_term; |
3175 | |
3176 | if (term->quantityType != QuantifierFixedCount) { |
3177 | m_backtrackingState.link(this); |
3178 | |
3179 | // Check whether we should backtrack back into the parentheses, or if we |
3180 | // are currently in a state where we had skipped over the subpattern |
3181 | // (in which case the flag value on the stack will be -1). |
3182 | unsigned parenthesesFrameLocation = term->frameLocation; |
3183 | Jump hadSkipped = branch32(Equal, Address(stackPointerRegister, (parenthesesFrameLocation + BackTrackInfoParenthesesOnce::beginIndex()) * sizeof(void*)), TrustedImm32(-1)); |
3184 | |
3185 | if (term->quantityType == QuantifierGreedy) { |
3186 | // For Greedy parentheses, we skip after having already tried going |
3187 | // through the subpattern, so if we get here we're done. |
3188 | YarrOp& beginOp = m_ops[op.m_previousOp]; |
3189 | beginOp.m_jumps.append(hadSkipped); |
3190 | } else { |
3191 | // For NonGreedy parentheses, we try skipping the subpattern first, |
3192 | // so if we get here we need to try running through the subpattern |
3193 | // next. Jump back to the start of the parentheses in the forwards |
3194 | // matching path. |
3195 | ASSERT(term->quantityType == QuantifierNonGreedy); |
3196 | YarrOp& beginOp = m_ops[op.m_previousOp]; |
3197 | hadSkipped.linkTo(beginOp.m_reentry, this); |
3198 | } |
3199 | |
3200 | m_backtrackingState.fallthrough(); |
3201 | } |
3202 | |
3203 | m_backtrackingState.append(op.m_jumps); |
3204 | break; |
3205 | } |
3206 | |
3207 | // OpParenthesesSubpatternTerminalBegin/End |
3208 | // |
3209 | // Terminal subpatterns will always match - there is nothing after them to |
3210 | // force a backtrack, and they have a minimum count of 0, and as such will |
3211 | // always produce an acceptable result. |
3212 | case OpParenthesesSubpatternTerminalBegin: { |
3213 | // We will backtrack to this point once the subpattern cannot match any |
3214 | // more. Since no match is accepted as a successful match (we are Greedy |
3215 | // quantified with a minimum of zero) jump back to the forwards matching |
3216 | // path at the end. |
3217 | YarrOp& endOp = m_ops[op.m_nextOp]; |
3218 | m_backtrackingState.linkTo(endOp.m_reentry, this); |
3219 | break; |
3220 | } |
3221 | case OpParenthesesSubpatternTerminalEnd: |
3222 | // We should never be backtracking to here (hence the 'terminal' in the name). |
3223 | ASSERT(m_backtrackingState.isEmpty()); |
3224 | m_backtrackingState.append(op.m_jumps); |
3225 | break; |
3226 | |
3227 | // OpParenthesesSubpatternBegin/End |
3228 | // |
3229 | // When we are backtracking back out of a capturing subpattern we need |
3230 | // to clear the start index in the matches output array, to record that |
3231 | // this subpattern has not been captured. |
3232 | // |
3233 | // When backtracking back out of a Greedy quantified subpattern we need |
3234 | // to catch this, and try running the remainder of the alternative after |
3235 | // the subpattern again, skipping the parentheses. |
3236 | // |
3237 | // Upon backtracking back into a quantified set of parentheses we need to |
3238 | // check whether we were currently skipping the subpattern. If not, we |
3239 | // can backtrack into them, if we were we need to either backtrack back |
3240 | // out of the start of the parentheses, or jump back to the forwards |
3241 | // matching start, depending of whether the match is Greedy or NonGreedy. |
3242 | case OpParenthesesSubpatternBegin: { |
3243 | #if ENABLE(YARR_JIT_ALL_PARENS_EXPRESSIONS) |
3244 | PatternTerm* term = op.m_term; |
3245 | unsigned parenthesesFrameLocation = term->frameLocation; |
3246 | |
3247 | if (term->quantityType != QuantifierFixedCount) { |
3248 | m_backtrackingState.link(this); |
3249 | |
3250 | RegisterID currParenContextReg = regT0; |
3251 | RegisterID newParenContextReg = regT1; |
3252 | |
3253 | loadFromFrame(parenthesesFrameLocation + BackTrackInfoParentheses::parenContextHeadIndex(), currParenContextReg); |
3254 | |
3255 | restoreParenContext(currParenContextReg, regT2, term->parentheses.subpatternId, term->parentheses.lastSubpatternId, parenthesesFrameLocation); |
3256 | |
3257 | freeParenContext(currParenContextReg, newParenContextReg); |
3258 | storeToFrame(newParenContextReg, parenthesesFrameLocation + BackTrackInfoParentheses::parenContextHeadIndex()); |
3259 | |
3260 | const RegisterID countTemporary = regT0; |
3261 | loadFromFrame(parenthesesFrameLocation + BackTrackInfoParentheses::matchAmountIndex(), countTemporary); |
3262 | Jump zeroLengthMatch = branchTest32(Zero, countTemporary); |
3263 | |
3264 | sub32(TrustedImm32(1), countTemporary); |
3265 | storeToFrame(countTemporary, parenthesesFrameLocation + BackTrackInfoParentheses::matchAmountIndex()); |
3266 | |
3267 | jump(m_ops[op.m_nextOp].m_reentry); |
3268 | |
3269 | zeroLengthMatch.link(this); |
3270 | |
3271 | // Clear the flag in the stackframe indicating we didn't run through the subpattern. |
3272 | storeToFrame(TrustedImm32(-1), parenthesesFrameLocation + BackTrackInfoParentheses::beginIndex()); |
3273 | |
3274 | if (term->quantityType == QuantifierGreedy) |
3275 | jump(m_ops[op.m_nextOp].m_reentry); |
3276 | |
3277 | // If Greedy, jump to the end. |
3278 | if (term->quantityType == QuantifierGreedy) { |
3279 | // A backtrack from after the parentheses, when skipping the subpattern, |
3280 | // will jump back to here. |
3281 | op.m_jumps.link(this); |
3282 | } |
3283 | |
3284 | m_backtrackingState.fallthrough(); |
3285 | } |
3286 | #else // !YARR_JIT_ALL_PARENS_EXPRESSIONS |
3287 | RELEASE_ASSERT_NOT_REACHED(); |
3288 | #endif |
3289 | break; |
3290 | } |
3291 | case OpParenthesesSubpatternEnd: { |
3292 | #if ENABLE(YARR_JIT_ALL_PARENS_EXPRESSIONS) |
3293 | PatternTerm* term = op.m_term; |
3294 | |
3295 | if (term->quantityType != QuantifierFixedCount) { |
3296 | m_backtrackingState.link(this); |
3297 | |
3298 | unsigned parenthesesFrameLocation = term->frameLocation; |
3299 | |
3300 | if (term->quantityType == QuantifierGreedy) { |
3301 | // Check whether we should backtrack back into the parentheses, or if we |
3302 | // are currently in a state where we had skipped over the subpattern |
3303 | // (in which case the flag value on the stack will be -1). |
3304 | Jump hadSkipped = branch32(Equal, Address(stackPointerRegister, (parenthesesFrameLocation + BackTrackInfoParentheses::beginIndex()) * sizeof(void*)), TrustedImm32(-1)); |
3305 | |
3306 | // For Greedy parentheses, we skip after having already tried going |
3307 | // through the subpattern, so if we get here we're done. |
3308 | YarrOp& beginOp = m_ops[op.m_previousOp]; |
3309 | beginOp.m_jumps.append(hadSkipped); |
3310 | } else { |
3311 | // For NonGreedy parentheses, we try skipping the subpattern first, |
3312 | // so if we get here we need to try running through the subpattern |
3313 | // next. Jump back to the start of the parentheses in the forwards |
3314 | // matching path. |
3315 | ASSERT(term->quantityType == QuantifierNonGreedy); |
3316 | |
3317 | const RegisterID beginTemporary = regT0; |
3318 | const RegisterID countTemporary = regT1; |
3319 | |
3320 | YarrOp& beginOp = m_ops[op.m_previousOp]; |
3321 | |
3322 | loadFromFrame(parenthesesFrameLocation + BackTrackInfoParentheses::beginIndex(), beginTemporary); |
3323 | branch32(Equal, beginTemporary, TrustedImm32(-1)).linkTo(beginOp.m_reentry, this); |
3324 | |
3325 | JumpList exceededMatchLimit; |
3326 | |
3327 | if (term->quantityMaxCount != quantifyInfinite) { |
3328 | loadFromFrame(parenthesesFrameLocation + BackTrackInfoParentheses::matchAmountIndex(), countTemporary); |
3329 | exceededMatchLimit.append(branch32(AboveOrEqual, countTemporary, Imm32(term->quantityMaxCount.unsafeGet()))); |
3330 | } |
3331 | |
3332 | branch32(Above, index, beginTemporary).linkTo(beginOp.m_reentry, this); |
3333 | |
3334 | exceededMatchLimit.link(this); |
3335 | } |
3336 | |
3337 | m_backtrackingState.fallthrough(); |
3338 | } |
3339 | |
3340 | m_backtrackingState.append(op.m_jumps); |
3341 | #else // !YARR_JIT_ALL_PARENS_EXPRESSIONS |
3342 | RELEASE_ASSERT_NOT_REACHED(); |
3343 | #endif |
3344 | break; |
3345 | } |
3346 | |
3347 | // OpParentheticalAssertionBegin/End |
3348 | case OpParentheticalAssertionBegin: { |
3349 | PatternTerm* term = op.m_term; |
3350 | YarrOp& endOp = m_ops[op.m_nextOp]; |
3351 | |
3352 | // We need to handle the backtracks upon backtracking back out |
3353 | // of a parenthetical assertion if either we need to correct |
3354 | // the input index, or the assertion was inverted. |
3355 | if (op.m_checkAdjust || term->invert()) { |
3356 | m_backtrackingState.link(this); |
3357 | |
3358 | if (op.m_checkAdjust) |
3359 | add32(Imm32(op.m_checkAdjust.unsafeGet()), index); |
3360 | |
3361 | // In an inverted assertion failure to match the subpattern |
3362 | // is treated as a successful match - jump to the end of the |
3363 | // subpattern. We already have adjusted the input position |
3364 | // back to that before the assertion, which is correct. |
3365 | if (term->invert()) |
3366 | jump(endOp.m_reentry); |
3367 | |
3368 | m_backtrackingState.fallthrough(); |
3369 | } |
3370 | |
3371 | // The End node's jump list will contain any backtracks into |
3372 | // the end of the assertion. Also, if inverted, we will have |
3373 | // added the failure caused by a successful match to this. |
3374 | m_backtrackingState.append(endOp.m_jumps); |
3375 | |
3376 | m_checkedOffset += op.m_checkAdjust; |
3377 | break; |
3378 | } |
3379 | case OpParentheticalAssertionEnd: { |
3380 | // FIXME: We should really be clearing any nested subpattern |
3381 | // matches on bailing out from after the pattern. Firefox has |
3382 | // this bug too (presumably because they use YARR!) |
3383 | |
3384 | // Never backtrack into an assertion; later failures bail to before the begin. |
3385 | m_backtrackingState.takeBacktracksToJumpList(op.m_jumps, this); |
3386 | |
3387 | YarrOp& lastOp = m_ops[op.m_previousOp]; |
3388 | m_checkedOffset -= lastOp.m_checkAdjust; |
3389 | break; |
3390 | } |
3391 | |
3392 | case OpMatchFailed: |
3393 | break; |
3394 | } |
3395 | |
3396 | } while (opIndex); |
3397 | } |
3398 | |
3399 | // Compilation methods: |
3400 | // ==================== |
3401 | |
3402 | // opCompileParenthesesSubpattern |
3403 | // Emits ops for a subpattern (set of parentheses). These consist |
3404 | // of a set of alternatives wrapped in an outer set of nodes for |
3405 | // the parentheses. |
3406 | // Supported types of parentheses are 'Once' (quantityMaxCount == 1), |
3407 | // 'Terminal' (non-capturing parentheses quantified as greedy |
3408 | // and infinite), and 0 based greedy / non-greedy quantified parentheses. |
3409 | // Alternatives will use the 'Simple' set of ops if either the |
3410 | // subpattern is terminal (in which case we will never need to |
3411 | // backtrack), or if the subpattern only contains one alternative. |
3412 | void opCompileParenthesesSubpattern(PatternTerm* term) |
3413 | { |
3414 | YarrOpCode parenthesesBeginOpCode; |
3415 | YarrOpCode parenthesesEndOpCode; |
3416 | YarrOpCode alternativeBeginOpCode = OpSimpleNestedAlternativeBegin; |
3417 | YarrOpCode alternativeNextOpCode = OpSimpleNestedAlternativeNext; |
3418 | YarrOpCode alternativeEndOpCode = OpSimpleNestedAlternativeEnd; |
3419 | |
3420 | if (UNLIKELY(!m_vm->isSafeToRecurse())) { |
3421 | m_failureReason = JITFailureReason::ParenthesisNestedTooDeep; |
3422 | return; |
3423 | } |
3424 | |
3425 | // We can currently only compile quantity 1 subpatterns that are |
3426 | // not copies. We generate a copy in the case of a range quantifier, |
3427 | // e.g. /(?:x){3,9}/, or /(?:x)+/ (These are effectively expanded to |
3428 | // /(?:x){3,3}(?:x){0,6}/ and /(?:x)(?:x)*/ repectively). The problem |
3429 | // comes where the subpattern is capturing, in which case we would |
3430 | // need to restore the capture from the first subpattern upon a |
3431 | // failure in the second. |
3432 | if (term->quantityMinCount && term->quantityMinCount != term->quantityMaxCount) { |
3433 | m_failureReason = JITFailureReason::VariableCountedParenthesisWithNonZeroMinimum; |
3434 | return; |
3435 | } |
3436 | |
3437 | if (term->quantityMaxCount == 1 && !term->parentheses.isCopy) { |
3438 | // Select the 'Once' nodes. |
3439 | parenthesesBeginOpCode = OpParenthesesSubpatternOnceBegin; |
3440 | parenthesesEndOpCode = OpParenthesesSubpatternOnceEnd; |
3441 | |
3442 | // If there is more than one alternative we cannot use the 'simple' nodes. |
3443 | if (term->parentheses.disjunction->m_alternatives.size() != 1) { |
3444 | alternativeBeginOpCode = OpNestedAlternativeBegin; |
3445 | alternativeNextOpCode = OpNestedAlternativeNext; |
3446 | alternativeEndOpCode = OpNestedAlternativeEnd; |
3447 | } |
3448 | } else if (term->parentheses.isTerminal) { |
3449 | // Select the 'Terminal' nodes. |
3450 | parenthesesBeginOpCode = OpParenthesesSubpatternTerminalBegin; |
3451 | parenthesesEndOpCode = OpParenthesesSubpatternTerminalEnd; |
3452 | } else { |
3453 | #if ENABLE(YARR_JIT_ALL_PARENS_EXPRESSIONS) |
3454 | // We only handle generic parenthesis with non-fixed counts. |
3455 | if (term->quantityType == QuantifierFixedCount) { |
3456 | // This subpattern is not supported by the JIT. |
3457 | m_failureReason = JITFailureReason::FixedCountParenthesizedSubpattern; |
3458 | return; |
3459 | } |
3460 | |
3461 | m_containsNestedSubpatterns = true; |
3462 | |
3463 | // Select the 'Generic' nodes. |
3464 | parenthesesBeginOpCode = OpParenthesesSubpatternBegin; |
3465 | parenthesesEndOpCode = OpParenthesesSubpatternEnd; |
3466 | |
3467 | // If there is more than one alternative we cannot use the 'simple' nodes. |
3468 | if (term->parentheses.disjunction->m_alternatives.size() != 1) { |
3469 | alternativeBeginOpCode = OpNestedAlternativeBegin; |
3470 | alternativeNextOpCode = OpNestedAlternativeNext; |
3471 | alternativeEndOpCode = OpNestedAlternativeEnd; |
3472 | } |
3473 | #else |
3474 | // This subpattern is not supported by the JIT. |
3475 | m_failureReason = JITFailureReason::ParenthesizedSubpattern; |
3476 | return; |
3477 | #endif |
3478 | } |
3479 | |
3480 | size_t parenBegin = m_ops.size(); |
3481 | m_ops.append(parenthesesBeginOpCode); |
3482 | |
3483 | m_ops.append(alternativeBeginOpCode); |
3484 | m_ops.last().m_previousOp = notFound; |
3485 | m_ops.last().m_term = term; |
3486 | Vector<std::unique_ptr<PatternAlternative>>& alternatives = term->parentheses.disjunction->m_alternatives; |
3487 | for (unsigned i = 0; i < alternatives.size(); ++i) { |
3488 | size_t lastOpIndex = m_ops.size() - 1; |
3489 | |
3490 | PatternAlternative* nestedAlternative = alternatives[i].get(); |
3491 | opCompileAlternative(nestedAlternative); |
3492 | |
3493 | size_t thisOpIndex = m_ops.size(); |
3494 | m_ops.append(YarrOp(alternativeNextOpCode)); |
3495 | |
3496 | YarrOp& lastOp = m_ops[lastOpIndex]; |
3497 | YarrOp& thisOp = m_ops[thisOpIndex]; |
3498 | |
3499 | lastOp.m_alternative = nestedAlternative; |
3500 | lastOp.m_nextOp = thisOpIndex; |
3501 | thisOp.m_previousOp = lastOpIndex; |
3502 | thisOp.m_term = term; |
3503 | } |
3504 | YarrOp& lastOp = m_ops.last(); |
3505 | ASSERT(lastOp.m_op == alternativeNextOpCode); |
3506 | lastOp.m_op = alternativeEndOpCode; |
3507 | lastOp.m_alternative = 0; |
3508 | lastOp.m_nextOp = notFound; |
3509 | |
3510 | size_t parenEnd = m_ops.size(); |
3511 | m_ops.append(parenthesesEndOpCode); |
3512 | |
3513 | m_ops[parenBegin].m_term = term; |
3514 | m_ops[parenBegin].m_previousOp = notFound; |
3515 | m_ops[parenBegin].m_nextOp = parenEnd; |
3516 | m_ops[parenEnd].m_term = term; |
3517 | m_ops[parenEnd].m_previousOp = parenBegin; |
3518 | m_ops[parenEnd].m_nextOp = notFound; |
3519 | } |
3520 | |
3521 | // opCompileParentheticalAssertion |
3522 | // Emits ops for a parenthetical assertion. These consist of an |
3523 | // OpSimpleNestedAlternativeBegin/Next/End set of nodes wrapping |
3524 | // the alternatives, with these wrapped by an outer pair of |
3525 | // OpParentheticalAssertionBegin/End nodes. |
3526 | // We can always use the OpSimpleNestedAlternative nodes in the |
3527 | // case of parenthetical assertions since these only ever match |
3528 | // once, and will never backtrack back into the assertion. |
3529 | void opCompileParentheticalAssertion(PatternTerm* term) |
3530 | { |
3531 | if (UNLIKELY(!m_vm->isSafeToRecurse())) { |
3532 | m_failureReason = JITFailureReason::ParenthesisNestedTooDeep; |
3533 | return; |
3534 | } |
3535 | |
3536 | size_t parenBegin = m_ops.size(); |
3537 | m_ops.append(OpParentheticalAssertionBegin); |
3538 | |
3539 | m_ops.append(OpSimpleNestedAlternativeBegin); |
3540 | m_ops.last().m_previousOp = notFound; |
3541 | m_ops.last().m_term = term; |
3542 | Vector<std::unique_ptr<PatternAlternative>>& alternatives = term->parentheses.disjunction->m_alternatives; |
3543 | for (unsigned i = 0; i < alternatives.size(); ++i) { |
3544 | size_t lastOpIndex = m_ops.size() - 1; |
3545 | |
3546 | PatternAlternative* nestedAlternative = alternatives[i].get(); |
3547 | opCompileAlternative(nestedAlternative); |
3548 | |
3549 | size_t thisOpIndex = m_ops.size(); |
3550 | m_ops.append(YarrOp(OpSimpleNestedAlternativeNext)); |
3551 | |
3552 | YarrOp& lastOp = m_ops[lastOpIndex]; |
3553 | YarrOp& thisOp = m_ops[thisOpIndex]; |
3554 | |
3555 | lastOp.m_alternative = nestedAlternative; |
3556 | lastOp.m_nextOp = thisOpIndex; |
3557 | thisOp.m_previousOp = lastOpIndex; |
3558 | thisOp.m_term = term; |
3559 | } |
3560 | YarrOp& lastOp = m_ops.last(); |
3561 | ASSERT(lastOp.m_op == OpSimpleNestedAlternativeNext); |
3562 | lastOp.m_op = OpSimpleNestedAlternativeEnd; |
3563 | lastOp.m_alternative = 0; |
3564 | lastOp.m_nextOp = notFound; |
3565 | |
3566 | size_t parenEnd = m_ops.size(); |
3567 | m_ops.append(OpParentheticalAssertionEnd); |
3568 | |
3569 | m_ops[parenBegin].m_term = term; |
3570 | m_ops[parenBegin].m_previousOp = notFound; |
3571 | m_ops[parenBegin].m_nextOp = parenEnd; |
3572 | m_ops[parenEnd].m_term = term; |
3573 | m_ops[parenEnd].m_previousOp = parenBegin; |
3574 | m_ops[parenEnd].m_nextOp = notFound; |
3575 | } |
3576 | |
3577 | // opCompileAlternative |
3578 | // Called to emit nodes for all terms in an alternative. |
3579 | void opCompileAlternative(PatternAlternative* alternative) |
3580 | { |
3581 | optimizeAlternative(alternative); |
3582 | |
3583 | for (unsigned i = 0; i < alternative->m_terms.size(); ++i) { |
3584 | PatternTerm* term = &alternative->m_terms[i]; |
3585 | |
3586 | switch (term->type) { |
3587 | case PatternTerm::TypeParenthesesSubpattern: |
3588 | opCompileParenthesesSubpattern(term); |
3589 | break; |
3590 | |
3591 | case PatternTerm::TypeParentheticalAssertion: |
3592 | opCompileParentheticalAssertion(term); |
3593 | break; |
3594 | |
3595 | default: |
3596 | m_ops.append(term); |
3597 | } |
3598 | } |
3599 | } |
3600 | |
3601 | // opCompileBody |
3602 | // This method compiles the body disjunction of the regular expression. |
3603 | // The body consists of two sets of alternatives - zero or more 'once |
3604 | // through' (BOL anchored) alternatives, followed by zero or more |
3605 | // repeated alternatives. |
3606 | // For each of these two sets of alteratives, if not empty they will be |
3607 | // wrapped in a set of OpBodyAlternativeBegin/Next/End nodes (with the |
3608 | // 'begin' node referencing the first alternative, and 'next' nodes |
3609 | // referencing any further alternatives. The begin/next/end nodes are |
3610 | // linked together in a doubly linked list. In the case of repeating |
3611 | // alternatives, the end node is also linked back to the beginning. |
3612 | // If no repeating alternatives exist, then a OpMatchFailed node exists |
3613 | // to return the failing result. |
3614 | void opCompileBody(PatternDisjunction* disjunction) |
3615 | { |
3616 | if (UNLIKELY(!m_vm->isSafeToRecurse())) { |
3617 | m_failureReason = JITFailureReason::ParenthesisNestedTooDeep; |
3618 | return; |
3619 | } |
3620 | |
3621 | Vector<std::unique_ptr<PatternAlternative>>& alternatives = disjunction->m_alternatives; |
3622 | size_t currentAlternativeIndex = 0; |
3623 | |
3624 | // Emit the 'once through' alternatives. |
3625 | if (alternatives.size() && alternatives[0]->onceThrough()) { |
3626 | m_ops.append(YarrOp(OpBodyAlternativeBegin)); |
3627 | m_ops.last().m_previousOp = notFound; |
3628 | |
3629 | do { |
3630 | size_t lastOpIndex = m_ops.size() - 1; |
3631 | PatternAlternative* alternative = alternatives[currentAlternativeIndex].get(); |
3632 | opCompileAlternative(alternative); |
3633 | |
3634 | size_t thisOpIndex = m_ops.size(); |
3635 | m_ops.append(YarrOp(OpBodyAlternativeNext)); |
3636 | |
3637 | YarrOp& lastOp = m_ops[lastOpIndex]; |
3638 | YarrOp& thisOp = m_ops[thisOpIndex]; |
3639 | |
3640 | lastOp.m_alternative = alternative; |
3641 | lastOp.m_nextOp = thisOpIndex; |
3642 | thisOp.m_previousOp = lastOpIndex; |
3643 | |
3644 | ++currentAlternativeIndex; |
3645 | } while (currentAlternativeIndex < alternatives.size() && alternatives[currentAlternativeIndex]->onceThrough()); |
3646 | |
3647 | YarrOp& lastOp = m_ops.last(); |
3648 | |
3649 | ASSERT(lastOp.m_op == OpBodyAlternativeNext); |
3650 | lastOp.m_op = OpBodyAlternativeEnd; |
3651 | lastOp.m_alternative = 0; |
3652 | lastOp.m_nextOp = notFound; |
3653 | } |
3654 | |
3655 | if (currentAlternativeIndex == alternatives.size()) { |
3656 | m_ops.append(YarrOp(OpMatchFailed)); |
3657 | return; |
3658 | } |
3659 | |
3660 | // Emit the repeated alternatives. |
3661 | size_t repeatLoop = m_ops.size(); |
3662 | m_ops.append(YarrOp(OpBodyAlternativeBegin)); |
3663 | m_ops.last().m_previousOp = notFound; |
3664 | do { |
3665 | size_t lastOpIndex = m_ops.size() - 1; |
3666 | PatternAlternative* alternative = alternatives[currentAlternativeIndex].get(); |
3667 | ASSERT(!alternative->onceThrough()); |
3668 | opCompileAlternative(alternative); |
3669 | |
3670 | size_t thisOpIndex = m_ops.size(); |
3671 | m_ops.append(YarrOp(OpBodyAlternativeNext)); |
3672 | |
3673 | YarrOp& lastOp = m_ops[lastOpIndex]; |
3674 | YarrOp& thisOp = m_ops[thisOpIndex]; |
3675 | |
3676 | lastOp.m_alternative = alternative; |
3677 | lastOp.m_nextOp = thisOpIndex; |
3678 | thisOp.m_previousOp = lastOpIndex; |
3679 | |
3680 | ++currentAlternativeIndex; |
3681 | } while (currentAlternativeIndex < alternatives.size()); |
3682 | YarrOp& lastOp = m_ops.last(); |
3683 | ASSERT(lastOp.m_op == OpBodyAlternativeNext); |
3684 | lastOp.m_op = OpBodyAlternativeEnd; |
3685 | lastOp.m_alternative = 0; |
3686 | lastOp.m_nextOp = repeatLoop; |
3687 | } |
3688 | |
3689 | void generateTryReadUnicodeCharacterHelper() |
3690 | { |
3691 | #ifdef JIT_UNICODE_EXPRESSIONS |
3692 | if (m_tryReadUnicodeCharacterCalls.isEmpty()) |
3693 | return; |
3694 | |
3695 | ASSERT(m_decodeSurrogatePairs); |
3696 | |
3697 | m_tryReadUnicodeCharacterEntry = label(); |
3698 | |
3699 | tagReturnAddress(); |
3700 | |
3701 | tryReadUnicodeCharImpl(regT0); |
3702 | |
3703 | ret(); |
3704 | #endif |
3705 | } |
3706 | |
3707 | void generateEnter() |
3708 | { |
3709 | #if CPU(X86_64) |
3710 | push(X86Registers::ebp); |
3711 | move(stackPointerRegister, X86Registers::ebp); |
3712 | |
3713 | if (m_pattern.m_saveInitialStartValue) |
3714 | push(X86Registers::ebx); |
3715 | |
3716 | #if OS(WINDOWS) |
3717 | push(X86Registers::edi); |
3718 | #endif |
3719 | #if ENABLE(YARR_JIT_ALL_PARENS_EXPRESSIONS) |
3720 | if (m_containsNestedSubpatterns) { |
3721 | #if OS(WINDOWS) |
3722 | push(X86Registers::esi); |
3723 | #endif |
3724 | push(X86Registers::r12); |
3725 | } |
3726 | #endif |
3727 | |
3728 | if (m_decodeSurrogatePairs) { |
3729 | push(X86Registers::r13); |
3730 | push(X86Registers::r14); |
3731 | push(X86Registers::r15); |
3732 | } |
3733 | // The ABI doesn't guarantee the upper bits are zero on unsigned arguments, so clear them ourselves. |
3734 | zeroExtend32ToPtr(index, index); |
3735 | zeroExtend32ToPtr(length, length); |
3736 | #if OS(WINDOWS) |
3737 | if (compileMode == IncludeSubpatterns) |
3738 | loadPtr(Address(X86Registers::ebp, 6 * sizeof(void*)), output); |
3739 | // rcx is the pointer to the allocated space for result in x64 Windows. |
3740 | push(X86Registers::ecx); |
3741 | #endif |
3742 | #elif CPU(ARM64) |
3743 | tagReturnAddress(); |
3744 | if (m_decodeSurrogatePairs) { |
3745 | pushPair(framePointerRegister, linkRegister); |
3746 | move(TrustedImm32(0x10000), supplementaryPlanesBase); |
3747 | move(TrustedImm32(0xd800), leadingSurrogateTag); |
3748 | move(TrustedImm32(0xdc00), trailingSurrogateTag); |
3749 | } |
3750 | |
3751 | // The ABI doesn't guarantee the upper bits are zero on unsigned arguments, so clear them ourselves. |
3752 | zeroExtend32ToPtr(index, index); |
3753 | zeroExtend32ToPtr(length, length); |
3754 | #elif CPU(ARM_THUMB2) |
3755 | push(ARMRegisters::r4); |
3756 | push(ARMRegisters::r5); |
3757 | push(ARMRegisters::r6); |
3758 | push(ARMRegisters::r8); |
3759 | #elif CPU(MIPS) |
3760 | // Do nothing. |
3761 | #endif |
3762 | |
3763 | store8(TrustedImm32(1), &m_vm->isExecutingInRegExpJIT); |
3764 | } |
3765 | |
3766 | void generateReturn() |
3767 | { |
3768 | store8(TrustedImm32(0), &m_vm->isExecutingInRegExpJIT); |
3769 | |
3770 | #if CPU(X86_64) |
3771 | #if OS(WINDOWS) |
3772 | // Store the return value in the allocated space pointed by rcx. |
3773 | pop(X86Registers::ecx); |
3774 | store64(returnRegister, Address(X86Registers::ecx)); |
3775 | store64(returnRegister2, Address(X86Registers::ecx, sizeof(void*))); |
3776 | move(X86Registers::ecx, returnRegister); |
3777 | #endif |
3778 | if (m_decodeSurrogatePairs) { |
3779 | pop(X86Registers::r15); |
3780 | pop(X86Registers::r14); |
3781 | pop(X86Registers::r13); |
3782 | } |
3783 | |
3784 | #if ENABLE(YARR_JIT_ALL_PARENS_EXPRESSIONS) |
3785 | if (m_containsNestedSubpatterns) { |
3786 | pop(X86Registers::r12); |
3787 | #if OS(WINDOWS) |
3788 | pop(X86Registers::esi); |
3789 | #endif |
3790 | } |
3791 | #endif |
3792 | #if OS(WINDOWS) |
3793 | pop(X86Registers::edi); |
3794 | #endif |
3795 | |
3796 | if (m_pattern.m_saveInitialStartValue) |
3797 | pop(X86Registers::ebx); |
3798 | pop(X86Registers::ebp); |
3799 | #elif CPU(ARM64) |
3800 | if (m_decodeSurrogatePairs) |
3801 | popPair(framePointerRegister, linkRegister); |
3802 | #elif CPU(ARM_THUMB2) |
3803 | pop(ARMRegisters::r8); |
3804 | pop(ARMRegisters::r6); |
3805 | pop(ARMRegisters::r5); |
3806 | pop(ARMRegisters::r4); |
3807 | #elif CPU(MIPS) |
3808 | // Do nothing |
3809 | #endif |
3810 | ret(); |
3811 | } |
3812 | |
3813 | public: |
3814 | YarrGenerator(VM* vm, YarrPattern& pattern, String& patternString, YarrCodeBlock& codeBlock, YarrCharSize charSize) |
3815 | : m_vm(vm) |
3816 | , m_pattern(pattern) |
3817 | , m_patternString(patternString) |
3818 | , m_codeBlock(codeBlock) |
3819 | , m_charSize(charSize) |
3820 | , m_decodeSurrogatePairs(m_charSize == Char16 && m_pattern.unicode()) |
3821 | , m_unicodeIgnoreCase(m_pattern.unicode() && m_pattern.ignoreCase()) |
3822 | , m_fixedSizedAlternative(false) |
3823 | , m_canonicalMode(m_pattern.unicode() ? CanonicalMode::Unicode : CanonicalMode::UCS2) |
3824 | #if ENABLE(YARR_JIT_ALL_PARENS_EXPRESSIONS) |
3825 | , m_containsNestedSubpatterns(false) |
3826 | , m_parenContextSizes(compileMode == IncludeSubpatterns ? m_pattern.m_numSubpatterns : 0, m_pattern.m_body->m_callFrameSize) |
3827 | #endif |
3828 | { |
3829 | } |
3830 | |
3831 | void compile() |
3832 | { |
3833 | YarrCodeBlock& codeBlock = m_codeBlock; |
3834 | |
3835 | #ifndef JIT_UNICODE_EXPRESSIONS |
3836 | if (m_decodeSurrogatePairs) { |
3837 | codeBlock.setFallBackWithFailureReason(JITFailureReason::DecodeSurrogatePair); |
3838 | return; |
3839 | } |
3840 | #endif |
3841 | |
3842 | if (m_pattern.m_containsBackreferences |
3843 | #if ENABLE(YARR_JIT_BACKREFERENCES) |
3844 | && (compileMode == MatchOnly || (m_pattern.ignoreCase() && m_charSize != Char8)) |
3845 | #endif |
3846 | ) { |
3847 | codeBlock.setFallBackWithFailureReason(JITFailureReason::BackReference); |
3848 | return; |
3849 | } |
3850 | |
3851 | // We need to compile before generating code since we set flags based on compilation that |
3852 | // are used during generation. |
3853 | opCompileBody(m_pattern.m_body); |
3854 | |
3855 | if (m_failureReason) { |
3856 | codeBlock.setFallBackWithFailureReason(*m_failureReason); |
3857 | return; |
3858 | } |
3859 | |
3860 | if (UNLIKELY(Options::dumpDisassembly() || Options::dumpRegExpDisassembly())) |
3861 | m_disassembler = makeUnique<YarrDisassembler>(this); |
3862 | |
3863 | if (m_disassembler) |
3864 | m_disassembler->setStartOfCode(label()); |
3865 | |
3866 | #if ENABLE(YARR_JIT_ALL_PARENS_EXPRESSIONS) |
3867 | if (m_containsNestedSubpatterns) |
3868 | codeBlock.setUsesPatternContextBuffer(); |
3869 | #endif |
3870 | |
3871 | generateEnter(); |
3872 | |
3873 | Jump hasInput = checkInput(); |
3874 | generateFailReturn(); |
3875 | hasInput.link(this); |
3876 | |
3877 | #ifdef JIT_UNICODE_EXPRESSIONS |
3878 | if (m_decodeSurrogatePairs) |
3879 | getEffectiveAddress(BaseIndex(input, length, TimesTwo), endOfStringAddress); |
3880 | #endif |
3881 | |
3882 | #if ENABLE(YARR_JIT_ALL_PARENS_EXPRESSIONS) |
3883 | if (m_containsNestedSubpatterns) |
3884 | move(TrustedImm32(matchLimit), remainingMatchCount); |
3885 | #endif |
3886 | |
3887 | if (compileMode == IncludeSubpatterns) { |
3888 | for (unsigned i = 0; i < m_pattern.m_numSubpatterns + 1; ++i) |
3889 | store32(TrustedImm32(-1), Address(output, (i << 1) * sizeof(int))); |
3890 | } |
3891 | |
3892 | if (!m_pattern.m_body->m_hasFixedSize) |
3893 | setMatchStart(index); |
3894 | |
3895 | initCallFrame(); |
3896 | |
3897 | #if ENABLE(YARR_JIT_ALL_PARENS_EXPRESSIONS) |
3898 | if (m_containsNestedSubpatterns) { |
3899 | initParenContextFreeList(); |
3900 | if (m_failureReason) { |
3901 | codeBlock.setFallBackWithFailureReason(*m_failureReason); |
3902 | return; |
3903 | } |
3904 | } |
3905 | #endif |
3906 | |
3907 | if (m_pattern.m_saveInitialStartValue) |
3908 | move(index, initialStart); |
3909 | |
3910 | generate(); |
3911 | if (m_disassembler) |
3912 | m_disassembler->setEndOfGenerate(label()); |
3913 | backtrack(); |
3914 | if (m_disassembler) |
3915 | m_disassembler->setEndOfBacktrack(label()); |
3916 | |
3917 | generateTryReadUnicodeCharacterHelper(); |
3918 | |
3919 | generateJITFailReturn(); |
3920 | |
3921 | if (m_disassembler) |
3922 | m_disassembler->setEndOfCode(label()); |
3923 | |
3924 | LinkBuffer linkBuffer(*this, REGEXP_CODE_ID, JITCompilationCanFail); |
3925 | if (linkBuffer.didFailToAllocate()) { |
3926 | codeBlock.setFallBackWithFailureReason(JITFailureReason::ExecutableMemoryAllocationFailure); |
3927 | return; |
3928 | } |
3929 | |
3930 | if (!m_tryReadUnicodeCharacterCalls.isEmpty()) { |
3931 | CodeLocationLabel<NoPtrTag> tryReadUnicodeCharacterHelper = linkBuffer.locationOf<NoPtrTag>(m_tryReadUnicodeCharacterEntry); |
3932 | |
3933 | for (auto call : m_tryReadUnicodeCharacterCalls) |
3934 | linkBuffer.link(call, tryReadUnicodeCharacterHelper); |
3935 | } |
3936 | |
3937 | m_backtrackingState.linkDataLabels(linkBuffer); |
3938 | |
3939 | if (m_disassembler) |
3940 | m_disassembler->dump(linkBuffer); |
3941 | |
3942 | if (compileMode == MatchOnly) { |
3943 | if (m_charSize == Char8) |
3944 | codeBlock.set8BitCodeMatchOnly(FINALIZE_REGEXP_CODE(linkBuffer, YarrMatchOnly8BitPtrTag, "Match-only 8-bit regular expression" )); |
3945 | else |
3946 | codeBlock.set16BitCodeMatchOnly(FINALIZE_REGEXP_CODE(linkBuffer, YarrMatchOnly16BitPtrTag, "Match-only 16-bit regular expression" )); |
3947 | } else { |
3948 | if (m_charSize == Char8) |
3949 | codeBlock.set8BitCode(FINALIZE_REGEXP_CODE(linkBuffer, Yarr8BitPtrTag, "8-bit regular expression" )); |
3950 | else |
3951 | codeBlock.set16BitCode(FINALIZE_REGEXP_CODE(linkBuffer, Yarr16BitPtrTag, "16-bit regular expression" )); |
3952 | } |
3953 | if (m_failureReason) |
3954 | codeBlock.setFallBackWithFailureReason(*m_failureReason); |
3955 | } |
3956 | |
3957 | const char* variant() override |
3958 | { |
3959 | if (compileMode == MatchOnly) { |
3960 | if (m_charSize == Char8) |
3961 | return "Match-only 8-bit regular expression" ; |
3962 | |
3963 | return "Match-only 16-bit regular expression" ; |
3964 | } |
3965 | |
3966 | if (m_charSize == Char8) |
3967 | return "8-bit regular expression" ; |
3968 | |
3969 | return "16-bit regular expression" ; |
3970 | } |
3971 | |
3972 | unsigned opCount() override |
3973 | { |
3974 | return m_ops.size(); |
3975 | } |
3976 | |
3977 | void dumpPatternString(PrintStream& out) override |
3978 | { |
3979 | m_pattern.dumpPatternString(out, m_patternString); |
3980 | } |
3981 | |
3982 | int dumpFor(PrintStream& out, unsigned opIndex) override |
3983 | { |
3984 | if (opIndex >= opCount()) |
3985 | return 0; |
3986 | |
3987 | out.printf("%4d:" , opIndex); |
3988 | |
3989 | YarrOp& op = m_ops[opIndex]; |
3990 | PatternTerm* term = op.m_term; |
3991 | switch (op.m_op) { |
3992 | case OpTerm: { |
3993 | out.print("OpTerm " ); |
3994 | switch (term->type) { |
3995 | case PatternTerm::TypeAssertionBOL: |
3996 | out.print("Assert BOL" ); |
3997 | break; |
3998 | |
3999 | case PatternTerm::TypeAssertionEOL: |
4000 | out.print("Assert EOL" ); |
4001 | break; |
4002 | |
4003 | case PatternTerm::TypeBackReference: |
4004 | out.printf("BackReference pattern #%u" , term->backReferenceSubpatternId); |
4005 | term->dumpQuantifier(out); |
4006 | break; |
4007 | |
4008 | case PatternTerm::TypePatternCharacter: |
4009 | out.print("TypePatternCharacter " ); |
4010 | dumpUChar32(out, term->patternCharacter); |
4011 | if (m_pattern.ignoreCase()) |
4012 | out.print(" ignore case" ); |
4013 | |
4014 | term->dumpQuantifier(out); |
4015 | break; |
4016 | |
4017 | case PatternTerm::TypeCharacterClass: |
4018 | out.print("TypePatternCharacterClass " ); |
4019 | if (term->invert()) |
4020 | out.print("not " ); |
4021 | dumpCharacterClass(out, &m_pattern, term->characterClass); |
4022 | term->dumpQuantifier(out); |
4023 | break; |
4024 | |
4025 | case PatternTerm::TypeAssertionWordBoundary: |
4026 | out.printf("%sword boundary" , term->invert() ? "non-" : "" ); |
4027 | break; |
4028 | |
4029 | case PatternTerm::TypeDotStarEnclosure: |
4030 | out.print(".* enclosure" ); |
4031 | break; |
4032 | |
4033 | case PatternTerm::TypeForwardReference: |
4034 | out.print("TypeForwardReference <not handled>" ); |
4035 | break; |
4036 | |
4037 | case PatternTerm::TypeParenthesesSubpattern: |
4038 | case PatternTerm::TypeParentheticalAssertion: |
4039 | RELEASE_ASSERT_NOT_REACHED(); |
4040 | break; |
4041 | } |
4042 | |
4043 | if (op.m_isDeadCode) |
4044 | out.print(" already handled" ); |
4045 | out.print("\n" ); |
4046 | return(0); |
4047 | } |
4048 | |
4049 | case OpBodyAlternativeBegin: |
4050 | out.printf("OpBodyAlternativeBegin minimum size %u\n" , op.m_alternative->m_minimumSize); |
4051 | return(0); |
4052 | |
4053 | case OpBodyAlternativeNext: |
4054 | out.printf("OpBodyAlternativeNext minimum size %u\n" , op.m_alternative->m_minimumSize); |
4055 | return(0); |
4056 | |
4057 | case OpBodyAlternativeEnd: |
4058 | out.print("OpBodyAlternativeEnd\n" ); |
4059 | return(0); |
4060 | |
4061 | case OpSimpleNestedAlternativeBegin: |
4062 | out.printf("OpSimpleNestedAlternativeBegin minimum size %u\n" , op.m_alternative->m_minimumSize); |
4063 | return(1); |
4064 | |
4065 | case OpNestedAlternativeBegin: |
4066 | out.printf("OpNestedAlternativeBegin minimum size %u\n" , op.m_alternative->m_minimumSize); |
4067 | return(1); |
4068 | |
4069 | case OpSimpleNestedAlternativeNext: |
4070 | out.printf("OpSimpleNestedAlternativeNext minimum size %u\n" , op.m_alternative->m_minimumSize); |
4071 | return(0); |
4072 | |
4073 | case OpNestedAlternativeNext: |
4074 | out.printf("OpNestedAlternativeNext minimum size %u\n" , op.m_alternative->m_minimumSize); |
4075 | return(0); |
4076 | |
4077 | case OpSimpleNestedAlternativeEnd: |
4078 | out.print("OpSimpleNestedAlternativeEnd" ); |
4079 | term->dumpQuantifier(out); |
4080 | out.print("\n" ); |
4081 | return(-1); |
4082 | |
4083 | case OpNestedAlternativeEnd: |
4084 | out.print("OpNestedAlternativeEnd" ); |
4085 | term->dumpQuantifier(out); |
4086 | out.print("\n" ); |
4087 | return(-1); |
4088 | |
4089 | case OpParenthesesSubpatternOnceBegin: |
4090 | out.print("OpParenthesesSubpatternOnceBegin " ); |
4091 | if (term->capture()) |
4092 | out.printf("capturing pattern #%u" , term->parentheses.subpatternId); |
4093 | else |
4094 | out.print("non-capturing" ); |
4095 | term->dumpQuantifier(out); |
4096 | out.print("\n" ); |
4097 | return(0); |
4098 | |
4099 | case OpParenthesesSubpatternOnceEnd: |
4100 | out.print("OpParenthesesSubpatternOnceEnd " ); |
4101 | if (term->capture()) |
4102 | out.printf("capturing pattern #%u" , term->parentheses.subpatternId); |
4103 | else |
4104 | out.print("non-capturing" ); |
4105 | term->dumpQuantifier(out); |
4106 | out.print("\n" ); |
4107 | return(0); |
4108 | |
4109 | case OpParenthesesSubpatternTerminalBegin: |
4110 | out.print("OpParenthesesSubpatternTerminalBegin " ); |
4111 | if (term->capture()) |
4112 | out.printf("capturing pattern #%u\n" , term->parentheses.subpatternId); |
4113 | else |
4114 | out.print("non-capturing\n" ); |
4115 | return(0); |
4116 | |
4117 | case OpParenthesesSubpatternTerminalEnd: |
4118 | out.print("OpParenthesesSubpatternTerminalEnd " ); |
4119 | if (term->capture()) |
4120 | out.printf("capturing pattern #%u\n" , term->parentheses.subpatternId); |
4121 | else |
4122 | out.print("non-capturing\n" ); |
4123 | return(0); |
4124 | |
4125 | case OpParenthesesSubpatternBegin: |
4126 | out.print("OpParenthesesSubpatternBegin " ); |
4127 | if (term->capture()) |
4128 | out.printf("capturing pattern #%u" , term->parentheses.subpatternId); |
4129 | else |
4130 | out.print("non-capturing" ); |
4131 | term->dumpQuantifier(out); |
4132 | out.print("\n" ); |
4133 | return(0); |
4134 | |
4135 | case OpParenthesesSubpatternEnd: |
4136 | out.print("OpParenthesesSubpatternEnd " ); |
4137 | if (term->capture()) |
4138 | out.printf("capturing pattern #%u" , term->parentheses.subpatternId); |
4139 | else |
4140 | out.print("non-capturing" ); |
4141 | term->dumpQuantifier(out); |
4142 | out.print("\n" ); |
4143 | return(0); |
4144 | |
4145 | case OpParentheticalAssertionBegin: |
4146 | out.printf("OpParentheticalAssertionBegin%s\n" , term->invert() ? " inverted" : "" ); |
4147 | return(0); |
4148 | |
4149 | case OpParentheticalAssertionEnd: |
4150 | out.printf("OpParentheticalAssertionEnd%s\n" , term->invert() ? " inverted" : "" ); |
4151 | return(0); |
4152 | |
4153 | case OpMatchFailed: |
4154 | out.print("OpMatchFailed\n" ); |
4155 | return(0); |
4156 | } |
4157 | |
4158 | return(0); |
4159 | } |
4160 | |
4161 | private: |
4162 | VM* m_vm; |
4163 | |
4164 | YarrPattern& m_pattern; |
4165 | String& m_patternString; |
4166 | |
4167 | YarrCodeBlock& m_codeBlock; |
4168 | YarrCharSize m_charSize; |
4169 | |
4170 | // Used to detect regular expression constructs that are not currently |
4171 | // supported in the JIT; fall back to the interpreter when this is detected. |
4172 | Optional<JITFailureReason> m_failureReason; |
4173 | |
4174 | bool m_decodeSurrogatePairs; |
4175 | bool m_unicodeIgnoreCase; |
4176 | bool m_fixedSizedAlternative; |
4177 | CanonicalMode m_canonicalMode; |
4178 | #if ENABLE(YARR_JIT_ALL_PARENS_EXPRESSIONS) |
4179 | bool m_containsNestedSubpatterns; |
4180 | ParenContextSizes m_parenContextSizes; |
4181 | #endif |
4182 | JumpList m_abortExecution; |
4183 | JumpList m_hitMatchLimit; |
4184 | Vector<Call> m_tryReadUnicodeCharacterCalls; |
4185 | Label m_tryReadUnicodeCharacterEntry; |
4186 | |
4187 | // The regular expression expressed as a linear sequence of operations. |
4188 | Vector<YarrOp, 128> m_ops; |
4189 | |
4190 | // This records the current input offset being applied due to the current |
4191 | // set of alternatives we are nested within. E.g. when matching the |
4192 | // character 'b' within the regular expression /abc/, we will know that |
4193 | // the minimum size for the alternative is 3, checked upon entry to the |
4194 | // alternative, and that 'b' is at offset 1 from the start, and as such |
4195 | // when matching 'b' we need to apply an offset of -2 to the load. |
4196 | // |
4197 | // FIXME: This should go away. Rather than tracking this value throughout |
4198 | // code generation, we should gather this information up front & store it |
4199 | // on the YarrOp structure. |
4200 | Checked<unsigned> m_checkedOffset; |
4201 | |
4202 | // This class records state whilst generating the backtracking path of code. |
4203 | BacktrackingState m_backtrackingState; |
4204 | |
4205 | std::unique_ptr<YarrDisassembler> m_disassembler; |
4206 | }; |
4207 | |
4208 | static void dumpCompileFailure(JITFailureReason failure) |
4209 | { |
4210 | switch (failure) { |
4211 | case JITFailureReason::DecodeSurrogatePair: |
4212 | dataLog("Can't JIT a pattern decoding surrogate pairs\n" ); |
4213 | break; |
4214 | case JITFailureReason::BackReference: |
4215 | dataLog("Can't JIT some patterns containing back references\n" ); |
4216 | break; |
4217 | case JITFailureReason::ForwardReference: |
4218 | dataLog("Can't JIT a pattern containing forward references\n" ); |
4219 | break; |
4220 | case JITFailureReason::VariableCountedParenthesisWithNonZeroMinimum: |
4221 | dataLog("Can't JIT a pattern containing a variable counted parenthesis with a non-zero minimum\n" ); |
4222 | break; |
4223 | case JITFailureReason::ParenthesizedSubpattern: |
4224 | dataLog("Can't JIT a pattern containing parenthesized subpatterns\n" ); |
4225 | break; |
4226 | case JITFailureReason::FixedCountParenthesizedSubpattern: |
4227 | dataLog("Can't JIT a pattern containing fixed count parenthesized subpatterns\n" ); |
4228 | break; |
4229 | case JITFailureReason::ParenthesisNestedTooDeep: |
4230 | dataLog("Can't JIT pattern due to parentheses nested too deeply\n" ); |
4231 | break; |
4232 | case JITFailureReason::ExecutableMemoryAllocationFailure: |
4233 | dataLog("Can't JIT because of failure of allocation of executable memory\n" ); |
4234 | break; |
4235 | } |
4236 | } |
4237 | |
4238 | void jitCompile(YarrPattern& pattern, String& patternString, YarrCharSize charSize, VM* vm, YarrCodeBlock& codeBlock, YarrJITCompileMode mode) |
4239 | { |
4240 | if (mode == MatchOnly) |
4241 | YarrGenerator<MatchOnly>(vm, pattern, patternString, codeBlock, charSize).compile(); |
4242 | else |
4243 | YarrGenerator<IncludeSubpatterns>(vm, pattern, patternString, codeBlock, charSize).compile(); |
4244 | |
4245 | if (auto failureReason = codeBlock.failureReason()) { |
4246 | if (Options::dumpCompiledRegExpPatterns()) { |
4247 | pattern.dumpPatternString(WTF::dataFile(), patternString); |
4248 | dataLog(" : " ); |
4249 | dumpCompileFailure(*failureReason); |
4250 | } |
4251 | } |
4252 | } |
4253 | |
4254 | }} |
4255 | |
4256 | #endif |
4257 | |