OLD | NEW |
(Empty) | |
| 1 // Copyright 2008-2009 the V8 project authors. All rights reserved. |
| 2 // Redistribution and use in source and binary forms, with or without |
| 3 // modification, are permitted provided that the following conditions are |
| 4 // met: |
| 5 // |
| 6 // * Redistributions of source code must retain the above copyright |
| 7 // notice, this list of conditions and the following disclaimer. |
| 8 // * Redistributions in binary form must reproduce the above |
| 9 // copyright notice, this list of conditions and the following |
| 10 // disclaimer in the documentation and/or other materials provided |
| 11 // with the distribution. |
| 12 // * Neither the name of Google Inc. nor the names of its |
| 13 // contributors may be used to endorse or promote products derived |
| 14 // from this software without specific prior written permission. |
| 15 // |
| 16 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 17 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 18 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
| 19 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT |
| 20 // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
| 21 // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
| 22 // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
| 23 // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
| 24 // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| 25 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| 26 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 27 |
| 28 #include "v8.h" |
| 29 #include "unicode.h" |
| 30 #include "log.h" |
| 31 #include "ast.h" |
| 32 #include "regexp-stack.h" |
| 33 #include "macro-assembler.h" |
| 34 #include "regexp-macro-assembler.h" |
| 35 #include "ia32/macro-assembler-ia32.h" |
| 36 #include "ia32/regexp-macro-assembler-ia32.h" |
| 37 |
| 38 namespace v8 { namespace internal { |
| 39 |
| 40 /* |
| 41 * This assembler uses the following register assignment convention |
| 42 * - edx : current character. Must be loaded using LoadCurrentCharacter |
| 43 * before using any of the dispatch methods. |
| 44 * - edi : current position in input, as negative offset from end of string. |
| 45 * Please notice that this is the byte offset, not the character offset! |
| 46 * - esi : end of input (points to byte after last character in input). |
| 47 * - ebp : frame pointer. Used to access arguments, local variables and |
| 48 * RegExp registers. |
| 49 * - esp : points to tip of C stack. |
| 50 * - ecx : points to tip of backtrack stack |
| 51 * |
| 52 * The registers eax, ebx and ecx are free to use for computations. |
| 53 * |
| 54 * Each call to a public method should retain this convention. |
| 55 * The stack will have the following structure: |
| 56 * - stack_area_top (High end of the memory area to use as |
| 57 * backtracking stack) |
| 58 * - at_start (if 1, start at start of string, if 0, don't) |
| 59 * - int* capture_array (int[num_saved_registers_], for output). |
| 60 * - end of input (Address of end of string) |
| 61 * - start of input (Address of first character in string) |
| 62 * - void* input_string (location of a handle containing the string) |
| 63 * --- frame alignment (if applicable) --- |
| 64 * - return address |
| 65 * ebp-> - old ebp |
| 66 * - backup of caller esi |
| 67 * - backup of caller edi |
| 68 * - backup of caller ebx |
| 69 * - Offset of location before start of input (effectively character |
| 70 * position -1). Used to initialize capture registers to a non-position. |
| 71 * - register 0 ebp[-4] (Only positions must be stored in the first |
| 72 * - register 1 ebp[-8] num_saved_registers_ registers) |
| 73 * - ... |
| 74 * |
| 75 * The first num_saved_registers_ registers are initialized to point to |
| 76 * "character -1" in the string (i.e., char_size() bytes before the first |
| 77 * character of the string). The remaining registers starts out as garbage. |
| 78 * |
| 79 * The data up to the return address must be placed there by the calling |
| 80 * code, e.g., by calling the code entry as cast to: |
| 81 * int (*match)(String* input_string, |
| 82 * Address start, |
| 83 * Address end, |
| 84 * int* capture_output_array, |
| 85 * bool at_start, |
| 86 * byte* stack_area_top) |
| 87 */ |
| 88 |
| 89 #define __ ACCESS_MASM(masm_) |
| 90 |
| 91 RegExpMacroAssemblerIA32::RegExpMacroAssemblerIA32( |
| 92 Mode mode, |
| 93 int registers_to_save) |
| 94 : masm_(new MacroAssembler(NULL, kRegExpCodeSize)), |
| 95 constants_(kRegExpConstantsSize), |
| 96 mode_(mode), |
| 97 num_registers_(registers_to_save), |
| 98 num_saved_registers_(registers_to_save), |
| 99 entry_label_(), |
| 100 start_label_(), |
| 101 success_label_(), |
| 102 backtrack_label_(), |
| 103 exit_label_() { |
| 104 __ jmp(&entry_label_); // We'll write the entry code later. |
| 105 __ bind(&start_label_); // And then continue from here. |
| 106 } |
| 107 |
| 108 |
| 109 RegExpMacroAssemblerIA32::~RegExpMacroAssemblerIA32() { |
| 110 delete masm_; |
| 111 // Unuse labels in case we throw away the assembler without calling GetCode. |
| 112 entry_label_.Unuse(); |
| 113 start_label_.Unuse(); |
| 114 success_label_.Unuse(); |
| 115 backtrack_label_.Unuse(); |
| 116 exit_label_.Unuse(); |
| 117 check_preempt_label_.Unuse(); |
| 118 stack_overflow_label_.Unuse(); |
| 119 } |
| 120 |
| 121 |
| 122 int RegExpMacroAssemblerIA32::stack_limit_slack() { |
| 123 return RegExpStack::kStackLimitSlack; |
| 124 } |
| 125 |
| 126 |
| 127 void RegExpMacroAssemblerIA32::AdvanceCurrentPosition(int by) { |
| 128 if (by != 0) { |
| 129 Label inside_string; |
| 130 __ add(Operand(edi), Immediate(by * char_size())); |
| 131 } |
| 132 } |
| 133 |
| 134 |
| 135 void RegExpMacroAssemblerIA32::AdvanceRegister(int reg, int by) { |
| 136 ASSERT(reg >= 0); |
| 137 ASSERT(reg < num_registers_); |
| 138 if (by != 0) { |
| 139 __ add(register_location(reg), Immediate(by)); |
| 140 } |
| 141 } |
| 142 |
| 143 |
| 144 void RegExpMacroAssemblerIA32::Backtrack() { |
| 145 CheckPreemption(); |
| 146 // Pop Code* offset from backtrack stack, add Code* and jump to location. |
| 147 Pop(ebx); |
| 148 __ add(Operand(ebx), Immediate(masm_->CodeObject())); |
| 149 __ jmp(Operand(ebx)); |
| 150 } |
| 151 |
| 152 |
| 153 void RegExpMacroAssemblerIA32::Bind(Label* label) { |
| 154 __ bind(label); |
| 155 } |
| 156 |
| 157 |
| 158 void RegExpMacroAssemblerIA32::CheckBitmap(uc16 start, |
| 159 Label* bitmap, |
| 160 Label* on_zero) { |
| 161 UNIMPLEMENTED(); |
| 162 } |
| 163 |
| 164 |
| 165 void RegExpMacroAssemblerIA32::CheckCharacter(uint32_t c, Label* on_equal) { |
| 166 __ cmp(current_character(), c); |
| 167 BranchOrBacktrack(equal, on_equal); |
| 168 } |
| 169 |
| 170 |
| 171 void RegExpMacroAssemblerIA32::CheckCharacterGT(uc16 limit, Label* on_greater) { |
| 172 __ cmp(current_character(), limit); |
| 173 BranchOrBacktrack(greater, on_greater); |
| 174 } |
| 175 |
| 176 |
| 177 void RegExpMacroAssemblerIA32::CheckAtStart(Label* on_at_start) { |
| 178 Label not_at_start; |
| 179 // Did we start the match at the start of the string at all? |
| 180 __ cmp(Operand(ebp, kAtStart), Immediate(0)); |
| 181 BranchOrBacktrack(equal, ¬_at_start); |
| 182 // If we did, are we still at the start of the input? |
| 183 __ lea(eax, Operand(esi, edi, times_1, 0)); |
| 184 __ cmp(eax, Operand(ebp, kInputStart)); |
| 185 BranchOrBacktrack(equal, on_at_start); |
| 186 __ bind(¬_at_start); |
| 187 } |
| 188 |
| 189 |
| 190 void RegExpMacroAssemblerIA32::CheckNotAtStart(Label* on_not_at_start) { |
| 191 // Did we start the match at the start of the string at all? |
| 192 __ cmp(Operand(ebp, kAtStart), Immediate(0)); |
| 193 BranchOrBacktrack(equal, on_not_at_start); |
| 194 // If we did, are we still at the start of the input? |
| 195 __ lea(eax, Operand(esi, edi, times_1, 0)); |
| 196 __ cmp(eax, Operand(ebp, kInputStart)); |
| 197 BranchOrBacktrack(not_equal, on_not_at_start); |
| 198 } |
| 199 |
| 200 |
| 201 void RegExpMacroAssemblerIA32::CheckCharacterLT(uc16 limit, Label* on_less) { |
| 202 __ cmp(current_character(), limit); |
| 203 BranchOrBacktrack(less, on_less); |
| 204 } |
| 205 |
| 206 |
| 207 void RegExpMacroAssemblerIA32::CheckCharacters(Vector<const uc16> str, |
| 208 int cp_offset, |
| 209 Label* on_failure, |
| 210 bool check_end_of_string) { |
| 211 int byte_length = str.length() * char_size(); |
| 212 int byte_offset = cp_offset * char_size(); |
| 213 if (check_end_of_string) { |
| 214 // Check that there are at least str.length() characters left in the input. |
| 215 __ cmp(Operand(edi), Immediate(-(byte_offset + byte_length))); |
| 216 BranchOrBacktrack(greater, on_failure); |
| 217 } |
| 218 |
| 219 Label backtrack; |
| 220 if (on_failure == NULL) { |
| 221 // Avoid inlining the Backtrack macro for each test. |
| 222 Label skip_backtrack; |
| 223 __ jmp(&skip_backtrack); |
| 224 __ bind(&backtrack); |
| 225 Backtrack(); |
| 226 __ bind(&skip_backtrack); |
| 227 on_failure = &backtrack; |
| 228 } |
| 229 |
| 230 for (int i = 0; i < str.length(); i++) { |
| 231 if (mode_ == ASCII) { |
| 232 __ cmpb(Operand(esi, edi, times_1, byte_offset + i), |
| 233 static_cast<int8_t>(str[i])); |
| 234 } else { |
| 235 ASSERT(mode_ == UC16); |
| 236 __ cmpw(Operand(esi, edi, times_1, byte_offset + i * sizeof(uc16)), |
| 237 Immediate(str[i])); |
| 238 } |
| 239 BranchOrBacktrack(not_equal, on_failure); |
| 240 } |
| 241 } |
| 242 |
| 243 |
| 244 void RegExpMacroAssemblerIA32::CheckGreedyLoop(Label* on_equal) { |
| 245 Label fallthrough; |
| 246 __ cmp(edi, Operand(backtrack_stackpointer(), 0)); |
| 247 __ j(not_equal, &fallthrough); |
| 248 __ add(Operand(backtrack_stackpointer()), Immediate(kPointerSize)); // Pop. |
| 249 BranchOrBacktrack(no_condition, on_equal); |
| 250 __ bind(&fallthrough); |
| 251 } |
| 252 |
| 253 |
| 254 void RegExpMacroAssemblerIA32::CheckNotBackReferenceIgnoreCase( |
| 255 int start_reg, |
| 256 Label* on_no_match) { |
| 257 Label fallthrough; |
| 258 __ mov(edx, register_location(start_reg)); // Index of start of capture |
| 259 __ mov(ebx, register_location(start_reg + 1)); // Index of end of capture |
| 260 __ sub(ebx, Operand(edx)); // Length of capture. |
| 261 |
| 262 // The length of a capture should not be negative. This can only happen |
| 263 // if the end of the capture is unrecorded, or at a point earlier than |
| 264 // the start of the capture. |
| 265 BranchOrBacktrack(less, on_no_match, not_taken); |
| 266 |
| 267 // If length is zero, either the capture is empty or it is completely |
| 268 // uncaptured. In either case succeed immediately. |
| 269 __ j(equal, &fallthrough); |
| 270 |
| 271 if (mode_ == ASCII) { |
| 272 Label success; |
| 273 Label fail; |
| 274 Label loop_increment; |
| 275 // Save register contents to make the registers available below. |
| 276 __ push(edi); |
| 277 __ push(backtrack_stackpointer()); |
| 278 // After this, the eax, ecx, and edi registers are available. |
| 279 |
| 280 __ add(edx, Operand(esi)); // Start of capture |
| 281 __ add(edi, Operand(esi)); // Start of text to match against capture. |
| 282 __ add(ebx, Operand(edi)); // End of text to match against capture. |
| 283 |
| 284 Label loop; |
| 285 __ bind(&loop); |
| 286 __ movzx_b(eax, Operand(edi, 0)); |
| 287 __ cmpb_al(Operand(edx, 0)); |
| 288 __ j(equal, &loop_increment); |
| 289 |
| 290 // Mismatch, try case-insensitive match (converting letters to lower-case). |
| 291 __ or_(eax, 0x20); // Convert match character to lower-case. |
| 292 __ lea(ecx, Operand(eax, -'a')); |
| 293 __ cmp(ecx, static_cast<int32_t>('z' - 'a')); // Is eax a lowercase letter? |
| 294 __ j(above, &fail); |
| 295 // Also convert capture character. |
| 296 __ movzx_b(ecx, Operand(edx, 0)); |
| 297 __ or_(ecx, 0x20); |
| 298 |
| 299 __ cmp(eax, Operand(ecx)); |
| 300 __ j(not_equal, &fail); |
| 301 |
| 302 __ bind(&loop_increment); |
| 303 // Increment pointers into match and capture strings. |
| 304 __ add(Operand(edx), Immediate(1)); |
| 305 __ add(Operand(edi), Immediate(1)); |
| 306 // Compare to end of match, and loop if not done. |
| 307 __ cmp(edi, Operand(ebx)); |
| 308 __ j(below, &loop, taken); |
| 309 __ jmp(&success); |
| 310 |
| 311 __ bind(&fail); |
| 312 // Restore original values before failing. |
| 313 __ pop(backtrack_stackpointer()); |
| 314 __ pop(edi); |
| 315 BranchOrBacktrack(no_condition, on_no_match); |
| 316 |
| 317 __ bind(&success); |
| 318 // Restore original value before continuing. |
| 319 __ pop(backtrack_stackpointer()); |
| 320 // Drop original value of character position. |
| 321 __ add(Operand(esp), Immediate(kPointerSize)); |
| 322 // Compute new value of character position after the matched part. |
| 323 __ sub(edi, Operand(esi)); |
| 324 } else { |
| 325 ASSERT(mode_ == UC16); |
| 326 // Save registers before calling C function. |
| 327 __ push(esi); |
| 328 __ push(edi); |
| 329 __ push(backtrack_stackpointer()); |
| 330 __ push(ebx); |
| 331 |
| 332 const int argument_count = 3; |
| 333 FrameAlign(argument_count, ecx); |
| 334 // Put arguments into allocated stack area, last argument highest on stack. |
| 335 // Parameters are |
| 336 // Address byte_offset1 - Address captured substring's start. |
| 337 // Address byte_offset2 - Address of current character position. |
| 338 // size_t byte_length - length of capture in bytes(!) |
| 339 |
| 340 // Set byte_length. |
| 341 __ mov(Operand(esp, 2 * kPointerSize), ebx); |
| 342 // Set byte_offset2. |
| 343 // Found by adding negative string-end offset of current position (edi) |
| 344 // to end of string. |
| 345 __ add(edi, Operand(esi)); |
| 346 __ mov(Operand(esp, 1 * kPointerSize), edi); |
| 347 // Set byte_offset1. |
| 348 // Start of capture, where edx already holds string-end negative offset. |
| 349 __ add(edx, Operand(esi)); |
| 350 __ mov(Operand(esp, 0 * kPointerSize), edx); |
| 351 |
| 352 Address function_address = FUNCTION_ADDR(&CaseInsensitiveCompareUC16); |
| 353 CallCFunction(function_address, argument_count); |
| 354 // Pop original values before reacting on result value. |
| 355 __ pop(ebx); |
| 356 __ pop(backtrack_stackpointer()); |
| 357 __ pop(edi); |
| 358 __ pop(esi); |
| 359 |
| 360 // Check if function returned non-zero for success or zero for failure. |
| 361 __ or_(eax, Operand(eax)); |
| 362 BranchOrBacktrack(zero, on_no_match); |
| 363 // On success, increment position by length of capture. |
| 364 __ add(edi, Operand(ebx)); |
| 365 } |
| 366 __ bind(&fallthrough); |
| 367 } |
| 368 |
| 369 |
| 370 void RegExpMacroAssemblerIA32::CheckNotBackReference( |
| 371 int start_reg, |
| 372 Label* on_no_match) { |
| 373 Label fallthrough; |
| 374 Label success; |
| 375 Label fail; |
| 376 |
| 377 // Find length of back-referenced capture. |
| 378 __ mov(edx, register_location(start_reg)); |
| 379 __ mov(eax, register_location(start_reg + 1)); |
| 380 __ sub(eax, Operand(edx)); // Length to check. |
| 381 // Fail on partial or illegal capture (start of capture after end of capture). |
| 382 BranchOrBacktrack(less, on_no_match); |
| 383 // Succeed on empty capture (including no capture) |
| 384 __ j(equal, &fallthrough); |
| 385 |
| 386 // Check that there are sufficient characters left in the input. |
| 387 __ mov(ebx, edi); |
| 388 __ add(ebx, Operand(eax)); |
| 389 BranchOrBacktrack(greater, on_no_match); |
| 390 |
| 391 // Save register to make it available below. |
| 392 __ push(backtrack_stackpointer()); |
| 393 |
| 394 // Compute pointers to match string and capture string |
| 395 __ lea(ebx, Operand(esi, edi, times_1, 0)); // Start of match. |
| 396 __ add(edx, Operand(esi)); // Start of capture. |
| 397 __ lea(ecx, Operand(eax, ebx, times_1, 0)); // End of match |
| 398 |
| 399 Label loop; |
| 400 __ bind(&loop); |
| 401 if (mode_ == ASCII) { |
| 402 __ movzx_b(eax, Operand(edx, 0)); |
| 403 __ cmpb_al(Operand(ebx, 0)); |
| 404 } else { |
| 405 ASSERT(mode_ == UC16); |
| 406 __ movzx_w(eax, Operand(edx, 0)); |
| 407 __ cmpw_ax(Operand(ebx, 0)); |
| 408 } |
| 409 __ j(not_equal, &fail); |
| 410 // Increment pointers into capture and match string. |
| 411 __ add(Operand(edx), Immediate(char_size())); |
| 412 __ add(Operand(ebx), Immediate(char_size())); |
| 413 // Check if we have reached end of match area. |
| 414 __ cmp(ebx, Operand(ecx)); |
| 415 __ j(below, &loop); |
| 416 __ jmp(&success); |
| 417 |
| 418 __ bind(&fail); |
| 419 // Restore backtrack stackpointer. |
| 420 __ pop(backtrack_stackpointer()); |
| 421 BranchOrBacktrack(no_condition, on_no_match); |
| 422 |
| 423 __ bind(&success); |
| 424 // Move current character position to position after match. |
| 425 __ mov(edi, ecx); |
| 426 __ sub(Operand(edi), esi); |
| 427 // Restore backtrack stackpointer. |
| 428 __ pop(backtrack_stackpointer()); |
| 429 |
| 430 __ bind(&fallthrough); |
| 431 } |
| 432 |
| 433 |
| 434 void RegExpMacroAssemblerIA32::CheckNotRegistersEqual(int reg1, |
| 435 int reg2, |
| 436 Label* on_not_equal) { |
| 437 __ mov(eax, register_location(reg1)); |
| 438 __ cmp(eax, register_location(reg2)); |
| 439 BranchOrBacktrack(not_equal, on_not_equal); |
| 440 } |
| 441 |
| 442 |
| 443 void RegExpMacroAssemblerIA32::CheckNotCharacter(uint32_t c, |
| 444 Label* on_not_equal) { |
| 445 __ cmp(current_character(), c); |
| 446 BranchOrBacktrack(not_equal, on_not_equal); |
| 447 } |
| 448 |
| 449 |
| 450 void RegExpMacroAssemblerIA32::CheckCharacterAfterAnd(uint32_t c, |
| 451 uint32_t mask, |
| 452 Label* on_equal) { |
| 453 __ mov(eax, current_character()); |
| 454 __ and_(eax, mask); |
| 455 __ cmp(eax, c); |
| 456 BranchOrBacktrack(equal, on_equal); |
| 457 } |
| 458 |
| 459 |
| 460 void RegExpMacroAssemblerIA32::CheckNotCharacterAfterAnd(uint32_t c, |
| 461 uint32_t mask, |
| 462 Label* on_not_equal) { |
| 463 __ mov(eax, current_character()); |
| 464 __ and_(eax, mask); |
| 465 __ cmp(eax, c); |
| 466 BranchOrBacktrack(not_equal, on_not_equal); |
| 467 } |
| 468 |
| 469 |
| 470 void RegExpMacroAssemblerIA32::CheckNotCharacterAfterMinusAnd( |
| 471 uc16 c, |
| 472 uc16 minus, |
| 473 uc16 mask, |
| 474 Label* on_not_equal) { |
| 475 ASSERT(minus < String::kMaxUC16CharCode); |
| 476 __ lea(eax, Operand(current_character(), -minus)); |
| 477 __ and_(eax, mask); |
| 478 __ cmp(eax, c); |
| 479 BranchOrBacktrack(not_equal, on_not_equal); |
| 480 } |
| 481 |
| 482 |
| 483 bool RegExpMacroAssemblerIA32::CheckSpecialCharacterClass(uc16 type, |
| 484 int cp_offset, |
| 485 bool check_offset, |
| 486 Label* on_no_match) { |
| 487 // Range checks (c in min..max) are generally implemented by an unsigned |
| 488 // (c - min) <= (max - min) check |
| 489 switch (type) { |
| 490 case 's': |
| 491 // Match space-characters |
| 492 if (mode_ == ASCII) { |
| 493 // ASCII space characters are '\t'..'\r' and ' '. |
| 494 if (check_offset) { |
| 495 LoadCurrentCharacter(cp_offset, on_no_match); |
| 496 } else { |
| 497 LoadCurrentCharacterUnchecked(cp_offset, 1); |
| 498 } |
| 499 Label success; |
| 500 __ cmp(current_character(), ' '); |
| 501 __ j(equal, &success); |
| 502 // Check range 0x09..0x0d |
| 503 __ sub(Operand(current_character()), Immediate('\t')); |
| 504 __ cmp(current_character(), '\r' - '\t'); |
| 505 BranchOrBacktrack(above, on_no_match); |
| 506 __ bind(&success); |
| 507 return true; |
| 508 } |
| 509 return false; |
| 510 case 'S': |
| 511 // Match non-space characters. |
| 512 if (check_offset) { |
| 513 LoadCurrentCharacter(cp_offset, on_no_match, 1); |
| 514 } else { |
| 515 LoadCurrentCharacterUnchecked(cp_offset, 1); |
| 516 } |
| 517 if (mode_ == ASCII) { |
| 518 // ASCII space characters are '\t'..'\r' and ' '. |
| 519 __ cmp(current_character(), ' '); |
| 520 BranchOrBacktrack(equal, on_no_match); |
| 521 __ sub(Operand(current_character()), Immediate('\t')); |
| 522 __ cmp(current_character(), '\r' - '\t'); |
| 523 BranchOrBacktrack(below_equal, on_no_match); |
| 524 return true; |
| 525 } |
| 526 return false; |
| 527 case 'd': |
| 528 // Match ASCII digits ('0'..'9') |
| 529 if (check_offset) { |
| 530 LoadCurrentCharacter(cp_offset, on_no_match, 1); |
| 531 } else { |
| 532 LoadCurrentCharacterUnchecked(cp_offset, 1); |
| 533 } |
| 534 __ sub(Operand(current_character()), Immediate('0')); |
| 535 __ cmp(current_character(), '9' - '0'); |
| 536 BranchOrBacktrack(above, on_no_match); |
| 537 return true; |
| 538 case 'D': |
| 539 // Match non ASCII-digits |
| 540 if (check_offset) { |
| 541 LoadCurrentCharacter(cp_offset, on_no_match, 1); |
| 542 } else { |
| 543 LoadCurrentCharacterUnchecked(cp_offset, 1); |
| 544 } |
| 545 __ sub(Operand(current_character()), Immediate('0')); |
| 546 __ cmp(current_character(), '9' - '0'); |
| 547 BranchOrBacktrack(below_equal, on_no_match); |
| 548 return true; |
| 549 case '.': { |
| 550 // Match non-newlines (not 0x0a('\n'), 0x0d('\r'), 0x2028 and 0x2029) |
| 551 if (check_offset) { |
| 552 LoadCurrentCharacter(cp_offset, on_no_match, 1); |
| 553 } else { |
| 554 LoadCurrentCharacterUnchecked(cp_offset, 1); |
| 555 } |
| 556 __ xor_(Operand(current_character()), Immediate(0x01)); |
| 557 // See if current character is '\n'^1 or '\r'^1, i.e., 0x0b or 0x0c |
| 558 __ sub(Operand(current_character()), Immediate(0x0b)); |
| 559 __ cmp(current_character(), 0x0c - 0x0b); |
| 560 BranchOrBacktrack(below_equal, on_no_match); |
| 561 if (mode_ == UC16) { |
| 562 // Compare original value to 0x2028 and 0x2029, using the already |
| 563 // computed (current_char ^ 0x01 - 0x0b). I.e., check for |
| 564 // 0x201d (0x2028 - 0x0b) or 0x201e. |
| 565 __ sub(Operand(current_character()), Immediate(0x2028 - 0x0b)); |
| 566 __ cmp(current_character(), 1); |
| 567 BranchOrBacktrack(below_equal, on_no_match); |
| 568 } |
| 569 return true; |
| 570 } |
| 571 case '*': |
| 572 // Match any character. |
| 573 if (check_offset) { |
| 574 CheckPosition(cp_offset, on_no_match); |
| 575 } |
| 576 return true; |
| 577 // No custom implementation (yet): w, W, s(UC16), S(UC16). |
| 578 default: |
| 579 return false; |
| 580 } |
| 581 } |
| 582 |
| 583 void RegExpMacroAssemblerIA32::DispatchHalfNibbleMap( |
| 584 uc16 start, |
| 585 Label* half_nibble_map, |
| 586 const Vector<Label*>& destinations) { |
| 587 UNIMPLEMENTED(); |
| 588 } |
| 589 |
| 590 |
| 591 void RegExpMacroAssemblerIA32::DispatchByteMap( |
| 592 uc16 start, |
| 593 Label* byte_map, |
| 594 const Vector<Label*>& destinations) { |
| 595 UNIMPLEMENTED(); |
| 596 } |
| 597 |
| 598 |
| 599 void RegExpMacroAssemblerIA32::DispatchHighByteMap( |
| 600 byte start, |
| 601 Label* byte_map, |
| 602 const Vector<Label*>& destinations) { |
| 603 UNIMPLEMENTED(); |
| 604 } |
| 605 |
| 606 |
| 607 void RegExpMacroAssemblerIA32::EmitOrLink(Label* label) { |
| 608 UNIMPLEMENTED(); // Has no use. |
| 609 } |
| 610 |
| 611 |
| 612 void RegExpMacroAssemblerIA32::Fail() { |
| 613 ASSERT(FAILURE == 0); // Return value for failure is zero. |
| 614 __ xor_(eax, Operand(eax)); // zero eax. |
| 615 __ jmp(&exit_label_); |
| 616 } |
| 617 |
| 618 |
| 619 Handle<Object> RegExpMacroAssemblerIA32::GetCode(Handle<String> source) { |
| 620 // Finalize code - write the entry point code now we know how many |
| 621 // registers we need. |
| 622 |
| 623 // Entry code: |
| 624 __ bind(&entry_label_); |
| 625 // Start new stack frame. |
| 626 __ push(ebp); |
| 627 __ mov(ebp, esp); |
| 628 // Save callee-save registers. Order here should correspond to order of |
| 629 // kBackup_ebx etc. |
| 630 __ push(esi); |
| 631 __ push(edi); |
| 632 __ push(ebx); // Callee-save on MacOS. |
| 633 __ push(Immediate(0)); // Make room for "input start - 1" constant. |
| 634 |
| 635 // Check if we have space on the stack for registers. |
| 636 Label retry_stack_check; |
| 637 Label stack_limit_hit; |
| 638 Label stack_ok; |
| 639 |
| 640 __ bind(&retry_stack_check); |
| 641 ExternalReference stack_guard_limit = |
| 642 ExternalReference::address_of_stack_guard_limit(); |
| 643 __ mov(ecx, esp); |
| 644 __ sub(ecx, Operand::StaticVariable(stack_guard_limit)); |
| 645 // Handle it if the stack pointer is already below the stack limit. |
| 646 __ j(below_equal, &stack_limit_hit, not_taken); |
| 647 // Check if there is room for the variable number of registers above |
| 648 // the stack limit. |
| 649 __ cmp(ecx, num_registers_ * kPointerSize); |
| 650 __ j(above_equal, &stack_ok, taken); |
| 651 // Exit with OutOfMemory exception. There is not enough space on the stack |
| 652 // for our working registers. |
| 653 __ mov(eax, EXCEPTION); |
| 654 __ jmp(&exit_label_); |
| 655 |
| 656 __ bind(&stack_limit_hit); |
| 657 CallCheckStackGuardState(ebx); |
| 658 __ or_(eax, Operand(eax)); |
| 659 // If returned value is non-zero, we exit with the returned value as result. |
| 660 // Otherwise it was a preemption and we just check the limit again. |
| 661 __ j(equal, &retry_stack_check); |
| 662 // Return value was non-zero. Exit with exception or retry. |
| 663 __ jmp(&exit_label_); |
| 664 |
| 665 __ bind(&stack_ok); |
| 666 |
| 667 // Allocate space on stack for registers. |
| 668 __ sub(Operand(esp), Immediate(num_registers_ * kPointerSize)); |
| 669 // Load string length. |
| 670 __ mov(esi, Operand(ebp, kInputEnd)); |
| 671 // Load input position. |
| 672 __ mov(edi, Operand(ebp, kInputStart)); |
| 673 // Set up edi to be negative offset from string end. |
| 674 __ sub(edi, Operand(esi)); |
| 675 if (num_saved_registers_ > 0) { |
| 676 // Fill saved registers with initial value = start offset - 1 |
| 677 // Fill in stack push order, to avoid accessing across an unwritten |
| 678 // page (a problem on Windows). |
| 679 __ mov(ecx, kRegisterZero); |
| 680 // Set eax to address of char before start of input |
| 681 // (effectively string position -1). |
| 682 __ lea(eax, Operand(edi, -char_size())); |
| 683 // Store this value in a local variable, for use when clearing |
| 684 // position registers. |
| 685 __ mov(Operand(ebp, kInputStartMinusOne), eax); |
| 686 Label init_loop; |
| 687 __ bind(&init_loop); |
| 688 __ mov(Operand(ebp, ecx, times_1, +0), eax); |
| 689 __ sub(Operand(ecx), Immediate(kPointerSize)); |
| 690 __ cmp(ecx, kRegisterZero - num_saved_registers_ * kPointerSize); |
| 691 __ j(greater, &init_loop); |
| 692 } |
| 693 // Ensure that we have written to each stack page, in order. Skipping a page |
| 694 // on Windows can cause segmentation faults. Assuming page size is 4k. |
| 695 const int kPageSize = 4096; |
| 696 const int kRegistersPerPage = kPageSize / kPointerSize; |
| 697 for (int i = num_saved_registers_ + kRegistersPerPage - 1; |
| 698 i < num_registers_; |
| 699 i += kRegistersPerPage) { |
| 700 __ mov(register_location(i), eax); // One write every page. |
| 701 } |
| 702 |
| 703 |
| 704 // Initialize backtrack stack pointer. |
| 705 __ mov(backtrack_stackpointer(), Operand(ebp, kStackHighEnd)); |
| 706 // Load previous char as initial value of current-character. |
| 707 Label at_start; |
| 708 __ cmp(Operand(ebp, kAtStart), Immediate(0)); |
| 709 __ j(not_equal, &at_start); |
| 710 LoadCurrentCharacterUnchecked(-1, 1); // Load previous char. |
| 711 __ jmp(&start_label_); |
| 712 __ bind(&at_start); |
| 713 __ mov(current_character(), '\n'); |
| 714 __ jmp(&start_label_); |
| 715 |
| 716 |
| 717 // Exit code: |
| 718 if (success_label_.is_linked()) { |
| 719 // Save captures when successful. |
| 720 __ bind(&success_label_); |
| 721 if (num_saved_registers_ > 0) { |
| 722 // copy captures to output |
| 723 __ mov(ebx, Operand(ebp, kRegisterOutput)); |
| 724 __ mov(ecx, Operand(ebp, kInputEnd)); |
| 725 __ sub(ecx, Operand(ebp, kInputStart)); |
| 726 for (int i = 0; i < num_saved_registers_; i++) { |
| 727 __ mov(eax, register_location(i)); |
| 728 __ add(eax, Operand(ecx)); // Convert to index from start, not end. |
| 729 if (mode_ == UC16) { |
| 730 __ sar(eax, 1); // Convert byte index to character index. |
| 731 } |
| 732 __ mov(Operand(ebx, i * kPointerSize), eax); |
| 733 } |
| 734 } |
| 735 __ mov(eax, Immediate(SUCCESS)); |
| 736 } |
| 737 // Exit and return eax |
| 738 __ bind(&exit_label_); |
| 739 // Skip esp past regexp registers. |
| 740 __ lea(esp, Operand(ebp, kBackup_ebx)); |
| 741 // Restore callee-save registers. |
| 742 __ pop(ebx); |
| 743 __ pop(edi); |
| 744 __ pop(esi); |
| 745 // Exit function frame, restore previous one. |
| 746 __ pop(ebp); |
| 747 __ ret(0); |
| 748 |
| 749 // Backtrack code (branch target for conditional backtracks). |
| 750 if (backtrack_label_.is_linked()) { |
| 751 __ bind(&backtrack_label_); |
| 752 Backtrack(); |
| 753 } |
| 754 |
| 755 Label exit_with_exception; |
| 756 |
| 757 // Preempt-code |
| 758 if (check_preempt_label_.is_linked()) { |
| 759 __ bind(&check_preempt_label_); |
| 760 |
| 761 __ push(backtrack_stackpointer()); |
| 762 __ push(edi); |
| 763 |
| 764 Label retry; |
| 765 |
| 766 __ bind(&retry); |
| 767 CallCheckStackGuardState(ebx); |
| 768 __ or_(eax, Operand(eax)); |
| 769 // If returning non-zero, we should end execution with the given |
| 770 // result as return value. |
| 771 __ j(not_zero, &exit_label_); |
| 772 // Check if we are still preempted. |
| 773 ExternalReference stack_guard_limit = |
| 774 ExternalReference::address_of_stack_guard_limit(); |
| 775 __ cmp(esp, Operand::StaticVariable(stack_guard_limit)); |
| 776 __ j(below_equal, &retry); |
| 777 |
| 778 __ pop(edi); |
| 779 __ pop(backtrack_stackpointer()); |
| 780 // String might have moved: Reload esi from frame. |
| 781 __ mov(esi, Operand(ebp, kInputEnd)); |
| 782 SafeReturn(); |
| 783 } |
| 784 |
| 785 // Backtrack stack overflow code. |
| 786 if (stack_overflow_label_.is_linked()) { |
| 787 __ bind(&stack_overflow_label_); |
| 788 // Reached if the backtrack-stack limit has been hit. |
| 789 |
| 790 Label grow_failed; |
| 791 // Save registers before calling C function |
| 792 __ push(esi); |
| 793 __ push(edi); |
| 794 |
| 795 // Call GrowStack(backtrack_stackpointer()) |
| 796 int num_arguments = 2; |
| 797 FrameAlign(num_arguments, ebx); |
| 798 __ lea(eax, Operand(ebp, kStackHighEnd)); |
| 799 __ mov(Operand(esp, 1 * kPointerSize), eax); |
| 800 __ mov(Operand(esp, 0 * kPointerSize), backtrack_stackpointer()); |
| 801 CallCFunction(FUNCTION_ADDR(&GrowStack), num_arguments); |
| 802 // If return NULL, we have failed to grow the stack, and |
| 803 // must exit with a stack-overflow exception. |
| 804 __ or_(eax, Operand(eax)); |
| 805 __ j(equal, &exit_with_exception); |
| 806 // Otherwise use return value as new stack pointer. |
| 807 __ mov(backtrack_stackpointer(), eax); |
| 808 // Restore saved registers and continue. |
| 809 __ pop(edi); |
| 810 __ pop(esi); |
| 811 SafeReturn(); |
| 812 } |
| 813 |
| 814 if (exit_with_exception.is_linked()) { |
| 815 // If any of the code above needed to exit with an exception. |
| 816 __ bind(&exit_with_exception); |
| 817 // Exit with Result EXCEPTION(-1) to signal thrown exception. |
| 818 __ mov(eax, EXCEPTION); |
| 819 __ jmp(&exit_label_); |
| 820 } |
| 821 |
| 822 CodeDesc code_desc; |
| 823 masm_->GetCode(&code_desc); |
| 824 Handle<Code> code = Factory::NewCode(code_desc, |
| 825 NULL, |
| 826 Code::ComputeFlags(Code::REGEXP), |
| 827 masm_->CodeObject()); |
| 828 LOG(RegExpCodeCreateEvent(*code, *source)); |
| 829 return Handle<Object>::cast(code); |
| 830 } |
| 831 |
| 832 |
| 833 void RegExpMacroAssemblerIA32::GoTo(Label* to) { |
| 834 BranchOrBacktrack(no_condition, to); |
| 835 } |
| 836 |
| 837 |
| 838 void RegExpMacroAssemblerIA32::IfRegisterGE(int reg, |
| 839 int comparand, |
| 840 Label* if_ge) { |
| 841 __ cmp(register_location(reg), Immediate(comparand)); |
| 842 BranchOrBacktrack(greater_equal, if_ge); |
| 843 } |
| 844 |
| 845 |
| 846 void RegExpMacroAssemblerIA32::IfRegisterLT(int reg, |
| 847 int comparand, |
| 848 Label* if_lt) { |
| 849 __ cmp(register_location(reg), Immediate(comparand)); |
| 850 BranchOrBacktrack(less, if_lt); |
| 851 } |
| 852 |
| 853 |
| 854 void RegExpMacroAssemblerIA32::IfRegisterEqPos(int reg, |
| 855 Label* if_eq) { |
| 856 __ cmp(edi, register_location(reg)); |
| 857 BranchOrBacktrack(equal, if_eq); |
| 858 } |
| 859 |
| 860 |
| 861 RegExpMacroAssembler::IrregexpImplementation |
| 862 RegExpMacroAssemblerIA32::Implementation() { |
| 863 return kIA32Implementation; |
| 864 } |
| 865 |
| 866 |
| 867 void RegExpMacroAssemblerIA32::LoadCurrentCharacter(int cp_offset, |
| 868 Label* on_end_of_input, |
| 869 bool check_bounds, |
| 870 int characters) { |
| 871 ASSERT(cp_offset >= -1); // ^ and \b can look behind one character. |
| 872 ASSERT(cp_offset < (1<<30)); // Be sane! (And ensure negation works) |
| 873 CheckPosition(cp_offset + characters - 1, on_end_of_input); |
| 874 LoadCurrentCharacterUnchecked(cp_offset, characters); |
| 875 } |
| 876 |
| 877 |
| 878 void RegExpMacroAssemblerIA32::PopCurrentPosition() { |
| 879 Pop(edi); |
| 880 } |
| 881 |
| 882 |
| 883 void RegExpMacroAssemblerIA32::PopRegister(int register_index) { |
| 884 Pop(eax); |
| 885 __ mov(register_location(register_index), eax); |
| 886 } |
| 887 |
| 888 |
| 889 void RegExpMacroAssemblerIA32::PushBacktrack(Label* label) { |
| 890 Push(Immediate::CodeRelativeOffset(label)); |
| 891 CheckStackLimit(); |
| 892 } |
| 893 |
| 894 |
| 895 void RegExpMacroAssemblerIA32::PushCurrentPosition() { |
| 896 Push(edi); |
| 897 } |
| 898 |
| 899 |
| 900 void RegExpMacroAssemblerIA32::PushRegister(int register_index, |
| 901 StackCheckFlag check_stack_limit) { |
| 902 __ mov(eax, register_location(register_index)); |
| 903 Push(eax); |
| 904 if (check_stack_limit) CheckStackLimit(); |
| 905 } |
| 906 |
| 907 |
| 908 void RegExpMacroAssemblerIA32::ReadCurrentPositionFromRegister(int reg) { |
| 909 __ mov(edi, register_location(reg)); |
| 910 } |
| 911 |
| 912 |
| 913 void RegExpMacroAssemblerIA32::ReadStackPointerFromRegister(int reg) { |
| 914 __ mov(backtrack_stackpointer(), register_location(reg)); |
| 915 __ add(backtrack_stackpointer(), Operand(ebp, kStackHighEnd)); |
| 916 } |
| 917 |
| 918 |
| 919 void RegExpMacroAssemblerIA32::SetRegister(int register_index, int to) { |
| 920 ASSERT(register_index >= num_saved_registers_); // Reserved for positions! |
| 921 __ mov(register_location(register_index), Immediate(to)); |
| 922 } |
| 923 |
| 924 |
| 925 void RegExpMacroAssemblerIA32::Succeed() { |
| 926 __ jmp(&success_label_); |
| 927 } |
| 928 |
| 929 |
| 930 void RegExpMacroAssemblerIA32::WriteCurrentPositionToRegister(int reg, |
| 931 int cp_offset) { |
| 932 if (cp_offset == 0) { |
| 933 __ mov(register_location(reg), edi); |
| 934 } else { |
| 935 __ lea(eax, Operand(edi, cp_offset * char_size())); |
| 936 __ mov(register_location(reg), eax); |
| 937 } |
| 938 } |
| 939 |
| 940 |
| 941 void RegExpMacroAssemblerIA32::ClearRegisters(int reg_from, int reg_to) { |
| 942 ASSERT(reg_from <= reg_to); |
| 943 __ mov(eax, Operand(ebp, kInputStartMinusOne)); |
| 944 for (int reg = reg_from; reg <= reg_to; reg++) { |
| 945 __ mov(register_location(reg), eax); |
| 946 } |
| 947 } |
| 948 |
| 949 |
| 950 void RegExpMacroAssemblerIA32::WriteStackPointerToRegister(int reg) { |
| 951 __ mov(eax, backtrack_stackpointer()); |
| 952 __ sub(eax, Operand(ebp, kStackHighEnd)); |
| 953 __ mov(register_location(reg), eax); |
| 954 } |
| 955 |
| 956 |
| 957 RegExpMacroAssemblerIA32::Result RegExpMacroAssemblerIA32::Match( |
| 958 Handle<Code> regexp_code, |
| 959 Handle<String> subject, |
| 960 int* offsets_vector, |
| 961 int offsets_vector_length, |
| 962 int previous_index) { |
| 963 |
| 964 ASSERT(subject->IsFlat()); |
| 965 ASSERT(previous_index >= 0); |
| 966 ASSERT(previous_index <= subject->length()); |
| 967 |
| 968 // No allocations before calling the regexp, but we can't use |
| 969 // AssertNoAllocation, since regexps might be preempted, and another thread |
| 970 // might do allocation anyway. |
| 971 |
| 972 String* subject_ptr = *subject; |
| 973 // Character offsets into string. |
| 974 int start_offset = previous_index; |
| 975 int end_offset = subject_ptr->length(); |
| 976 |
| 977 bool is_ascii = StringShape(*subject).IsAsciiRepresentation(); |
| 978 |
| 979 if (StringShape(subject_ptr).IsCons()) { |
| 980 subject_ptr = ConsString::cast(subject_ptr)->first(); |
| 981 } else if (StringShape(subject_ptr).IsSliced()) { |
| 982 SlicedString* slice = SlicedString::cast(subject_ptr); |
| 983 start_offset += slice->start(); |
| 984 end_offset += slice->start(); |
| 985 subject_ptr = slice->buffer(); |
| 986 } |
| 987 // Ensure that an underlying string has the same ascii-ness. |
| 988 ASSERT(StringShape(subject_ptr).IsAsciiRepresentation() == is_ascii); |
| 989 ASSERT(subject_ptr->IsExternalString() || subject_ptr->IsSeqString()); |
| 990 // String is now either Sequential or External |
| 991 int char_size_shift = is_ascii ? 0 : 1; |
| 992 int char_length = end_offset - start_offset; |
| 993 |
| 994 const byte* input_start = |
| 995 StringCharacterPosition(subject_ptr, start_offset); |
| 996 int byte_length = char_length << char_size_shift; |
| 997 const byte* input_end = input_start + byte_length; |
| 998 RegExpMacroAssemblerIA32::Result res = Execute(*regexp_code, |
| 999 subject_ptr, |
| 1000 start_offset, |
| 1001 input_start, |
| 1002 input_end, |
| 1003 offsets_vector, |
| 1004 previous_index == 0); |
| 1005 |
| 1006 if (res == SUCCESS) { |
| 1007 // Capture values are relative to start_offset only. |
| 1008 // Convert them to be relative to start of string. |
| 1009 for (int i = 0; i < offsets_vector_length; i++) { |
| 1010 if (offsets_vector[i] >= 0) { |
| 1011 offsets_vector[i] += previous_index; |
| 1012 } |
| 1013 } |
| 1014 } |
| 1015 |
| 1016 return res; |
| 1017 } |
| 1018 |
| 1019 // Private methods: |
| 1020 |
| 1021 static unibrow::Mapping<unibrow::Ecma262Canonicalize> canonicalize; |
| 1022 |
| 1023 RegExpMacroAssemblerIA32::Result RegExpMacroAssemblerIA32::Execute( |
| 1024 Code* code, |
| 1025 String* input, |
| 1026 int start_offset, |
| 1027 const byte* input_start, |
| 1028 const byte* input_end, |
| 1029 int* output, |
| 1030 bool at_start) { |
| 1031 typedef int (*matcher)(String*, int, const byte*, |
| 1032 const byte*, int*, int, Address); |
| 1033 matcher matcher_func = FUNCTION_CAST<matcher>(code->entry()); |
| 1034 |
| 1035 int at_start_val = at_start ? 1 : 0; |
| 1036 |
| 1037 // Ensure that the minimum stack has been allocated. |
| 1038 RegExpStack stack; |
| 1039 Address stack_top = RegExpStack::stack_top(); |
| 1040 |
| 1041 int result = matcher_func(input, |
| 1042 start_offset, |
| 1043 input_start, |
| 1044 input_end, |
| 1045 output, |
| 1046 at_start_val, |
| 1047 stack_top); |
| 1048 ASSERT(result <= SUCCESS); |
| 1049 ASSERT(result >= RETRY); |
| 1050 |
| 1051 if (result == EXCEPTION && !Top::has_pending_exception()) { |
| 1052 // We detected a stack overflow (on the backtrack stack) in RegExp code, |
| 1053 // but haven't created the exception yet. |
| 1054 Top::StackOverflow(); |
| 1055 } |
| 1056 return static_cast<Result>(result); |
| 1057 } |
| 1058 |
| 1059 |
| 1060 int RegExpMacroAssemblerIA32::CaseInsensitiveCompareUC16(Address byte_offset1, |
| 1061 Address byte_offset2, |
| 1062 size_t byte_length) { |
| 1063 // This function is not allowed to cause a garbage collection. |
| 1064 // A GC might move the calling generated code and invalidate the |
| 1065 // return address on the stack. |
| 1066 ASSERT(byte_length % 2 == 0); |
| 1067 uc16* substring1 = reinterpret_cast<uc16*>(byte_offset1); |
| 1068 uc16* substring2 = reinterpret_cast<uc16*>(byte_offset2); |
| 1069 size_t length = byte_length >> 1; |
| 1070 |
| 1071 for (size_t i = 0; i < length; i++) { |
| 1072 unibrow::uchar c1 = substring1[i]; |
| 1073 unibrow::uchar c2 = substring2[i]; |
| 1074 if (c1 != c2) { |
| 1075 canonicalize.get(c1, '\0', &c1); |
| 1076 if (c1 != c2) { |
| 1077 canonicalize.get(c2, '\0', &c2); |
| 1078 if (c1 != c2) { |
| 1079 return 0; |
| 1080 } |
| 1081 } |
| 1082 } |
| 1083 } |
| 1084 return 1; |
| 1085 } |
| 1086 |
| 1087 |
| 1088 void RegExpMacroAssemblerIA32::CallCheckStackGuardState(Register scratch) { |
| 1089 int num_arguments = 3; |
| 1090 FrameAlign(num_arguments, scratch); |
| 1091 // RegExp code frame pointer. |
| 1092 __ mov(Operand(esp, 2 * kPointerSize), ebp); |
| 1093 // Code* of self. |
| 1094 __ mov(Operand(esp, 1 * kPointerSize), Immediate(masm_->CodeObject())); |
| 1095 // Next address on the stack (will be address of return address). |
| 1096 __ lea(eax, Operand(esp, -kPointerSize)); |
| 1097 __ mov(Operand(esp, 0 * kPointerSize), eax); |
| 1098 CallCFunction(FUNCTION_ADDR(&CheckStackGuardState), num_arguments); |
| 1099 } |
| 1100 |
| 1101 |
| 1102 // Helper function for reading a value out of a stack frame. |
| 1103 template <typename T> |
| 1104 static T& frame_entry(Address re_frame, int frame_offset) { |
| 1105 return reinterpret_cast<T&>(Memory::int32_at(re_frame + frame_offset)); |
| 1106 } |
| 1107 |
| 1108 |
| 1109 const byte* RegExpMacroAssemblerIA32::StringCharacterPosition(String* subject, |
| 1110 int start_index) { |
| 1111 // Not just flat, but ultra flat. |
| 1112 ASSERT(subject->IsExternalString() || subject->IsSeqString()); |
| 1113 ASSERT(start_index >= 0); |
| 1114 ASSERT(start_index <= subject->length()); |
| 1115 if (StringShape(subject).IsAsciiRepresentation()) { |
| 1116 const byte* address; |
| 1117 if (StringShape(subject).IsExternal()) { |
| 1118 const char* data = ExternalAsciiString::cast(subject)->resource()->data(); |
| 1119 address = reinterpret_cast<const byte*>(data); |
| 1120 } else { |
| 1121 ASSERT(subject->IsSeqAsciiString()); |
| 1122 char* data = SeqAsciiString::cast(subject)->GetChars(); |
| 1123 address = reinterpret_cast<const byte*>(data); |
| 1124 } |
| 1125 return address + start_index; |
| 1126 } |
| 1127 const uc16* data; |
| 1128 if (StringShape(subject).IsExternal()) { |
| 1129 data = ExternalTwoByteString::cast(subject)->resource()->data(); |
| 1130 } else { |
| 1131 ASSERT(subject->IsSeqTwoByteString()); |
| 1132 data = SeqTwoByteString::cast(subject)->GetChars(); |
| 1133 } |
| 1134 return reinterpret_cast<const byte*>(data + start_index); |
| 1135 } |
| 1136 |
| 1137 |
| 1138 int RegExpMacroAssemblerIA32::CheckStackGuardState(Address* return_address, |
| 1139 Code* re_code, |
| 1140 Address re_frame) { |
| 1141 if (StackGuard::IsStackOverflow()) { |
| 1142 Top::StackOverflow(); |
| 1143 return EXCEPTION; |
| 1144 } |
| 1145 |
| 1146 // If not real stack overflow the stack guard was used to interrupt |
| 1147 // execution for another purpose. |
| 1148 |
| 1149 // Prepare for possible GC. |
| 1150 HandleScope handles; |
| 1151 Handle<Code> code_handle(re_code); |
| 1152 |
| 1153 Handle<String> subject(frame_entry<String*>(re_frame, kInputString)); |
| 1154 // Current string. |
| 1155 bool is_ascii = StringShape(*subject).IsAsciiRepresentation(); |
| 1156 |
| 1157 ASSERT(re_code->instruction_start() <= *return_address); |
| 1158 ASSERT(*return_address <= |
| 1159 re_code->instruction_start() + re_code->instruction_size()); |
| 1160 |
| 1161 Object* result = Execution::HandleStackGuardInterrupt(); |
| 1162 |
| 1163 if (*code_handle != re_code) { // Return address no longer valid |
| 1164 int delta = *code_handle - re_code; |
| 1165 // Overwrite the return address on the stack. |
| 1166 *return_address += delta; |
| 1167 } |
| 1168 |
| 1169 if (result->IsException()) { |
| 1170 return EXCEPTION; |
| 1171 } |
| 1172 |
| 1173 // String might have changed. |
| 1174 if (StringShape(*subject).IsAsciiRepresentation() != is_ascii) { |
| 1175 // If we changed between an ASCII and an UC16 string, the specialized |
| 1176 // code cannot be used, and we need to restart regexp matching from |
| 1177 // scratch (including, potentially, compiling a new version of the code). |
| 1178 return RETRY; |
| 1179 } |
| 1180 |
| 1181 // Otherwise, the content of the string might have moved. It must still |
| 1182 // be a sequential or external string with the same content. |
| 1183 // Update the start and end pointers in the stack frame to the current |
| 1184 // location (whether it has actually moved or not). |
| 1185 ASSERT(StringShape(*subject).IsSequential() || |
| 1186 StringShape(*subject).IsExternal()); |
| 1187 |
| 1188 // The original start address of the characters to match. |
| 1189 const byte* start_address = frame_entry<const byte*>(re_frame, kInputStart); |
| 1190 |
| 1191 // Find the current start address of the same character at the current string |
| 1192 // position. |
| 1193 int start_index = frame_entry<int>(re_frame, kStartIndex); |
| 1194 const byte* new_address = StringCharacterPosition(*subject, start_index); |
| 1195 |
| 1196 if (start_address != new_address) { |
| 1197 // If there is a difference, update start and end addresses in the |
| 1198 // RegExp stack frame to match the new value. |
| 1199 const byte* end_address = frame_entry<const byte* >(re_frame, kInputEnd); |
| 1200 int byte_length = end_address - start_address; |
| 1201 frame_entry<const byte*>(re_frame, kInputStart) = new_address; |
| 1202 frame_entry<const byte*>(re_frame, kInputEnd) = new_address + byte_length; |
| 1203 } |
| 1204 |
| 1205 return 0; |
| 1206 } |
| 1207 |
| 1208 |
| 1209 Address RegExpMacroAssemblerIA32::GrowStack(Address stack_pointer, |
| 1210 Address* stack_top) { |
| 1211 size_t size = RegExpStack::stack_capacity(); |
| 1212 Address old_stack_top = RegExpStack::stack_top(); |
| 1213 ASSERT(old_stack_top == *stack_top); |
| 1214 ASSERT(stack_pointer <= old_stack_top); |
| 1215 ASSERT(static_cast<size_t>(old_stack_top - stack_pointer) <= size); |
| 1216 Address new_stack_top = RegExpStack::EnsureCapacity(size * 2); |
| 1217 if (new_stack_top == NULL) { |
| 1218 return NULL; |
| 1219 } |
| 1220 *stack_top = new_stack_top; |
| 1221 return new_stack_top - (old_stack_top - stack_pointer); |
| 1222 } |
| 1223 |
| 1224 |
| 1225 Operand RegExpMacroAssemblerIA32::register_location(int register_index) { |
| 1226 ASSERT(register_index < (1<<30)); |
| 1227 if (num_registers_ <= register_index) { |
| 1228 num_registers_ = register_index + 1; |
| 1229 } |
| 1230 return Operand(ebp, kRegisterZero - register_index * kPointerSize); |
| 1231 } |
| 1232 |
| 1233 |
| 1234 void RegExpMacroAssemblerIA32::CheckPosition(int cp_offset, |
| 1235 Label* on_outside_input) { |
| 1236 __ cmp(edi, -cp_offset * char_size()); |
| 1237 BranchOrBacktrack(greater_equal, on_outside_input); |
| 1238 } |
| 1239 |
| 1240 |
| 1241 void RegExpMacroAssemblerIA32::BranchOrBacktrack(Condition condition, |
| 1242 Label* to, |
| 1243 Hint hint) { |
| 1244 if (condition < 0) { // No condition |
| 1245 if (to == NULL) { |
| 1246 Backtrack(); |
| 1247 return; |
| 1248 } |
| 1249 __ jmp(to); |
| 1250 return; |
| 1251 } |
| 1252 if (to == NULL) { |
| 1253 __ j(condition, &backtrack_label_, hint); |
| 1254 return; |
| 1255 } |
| 1256 __ j(condition, to, hint); |
| 1257 } |
| 1258 |
| 1259 |
| 1260 void RegExpMacroAssemblerIA32::SafeCall(Label* to) { |
| 1261 Label return_to; |
| 1262 __ push(Immediate::CodeRelativeOffset(&return_to)); |
| 1263 __ jmp(to); |
| 1264 __ bind(&return_to); |
| 1265 } |
| 1266 |
| 1267 |
| 1268 void RegExpMacroAssemblerIA32::SafeReturn() { |
| 1269 __ pop(ebx); |
| 1270 __ add(Operand(ebx), Immediate(masm_->CodeObject())); |
| 1271 __ jmp(Operand(ebx)); |
| 1272 } |
| 1273 |
| 1274 |
| 1275 void RegExpMacroAssemblerIA32::Push(Register source) { |
| 1276 ASSERT(!source.is(backtrack_stackpointer())); |
| 1277 // Notice: This updates flags, unlike normal Push. |
| 1278 __ sub(Operand(backtrack_stackpointer()), Immediate(kPointerSize)); |
| 1279 __ mov(Operand(backtrack_stackpointer(), 0), source); |
| 1280 } |
| 1281 |
| 1282 |
| 1283 void RegExpMacroAssemblerIA32::Push(Immediate value) { |
| 1284 // Notice: This updates flags, unlike normal Push. |
| 1285 __ sub(Operand(backtrack_stackpointer()), Immediate(kPointerSize)); |
| 1286 __ mov(Operand(backtrack_stackpointer(), 0), value); |
| 1287 } |
| 1288 |
| 1289 |
| 1290 void RegExpMacroAssemblerIA32::Pop(Register target) { |
| 1291 ASSERT(!target.is(backtrack_stackpointer())); |
| 1292 __ mov(target, Operand(backtrack_stackpointer(), 0)); |
| 1293 // Notice: This updates flags, unlike normal Pop. |
| 1294 __ add(Operand(backtrack_stackpointer()), Immediate(kPointerSize)); |
| 1295 } |
| 1296 |
| 1297 |
| 1298 void RegExpMacroAssemblerIA32::CheckPreemption() { |
| 1299 // Check for preemption. |
| 1300 Label no_preempt; |
| 1301 ExternalReference stack_guard_limit = |
| 1302 ExternalReference::address_of_stack_guard_limit(); |
| 1303 __ cmp(esp, Operand::StaticVariable(stack_guard_limit)); |
| 1304 __ j(above, &no_preempt, taken); |
| 1305 |
| 1306 SafeCall(&check_preempt_label_); |
| 1307 |
| 1308 __ bind(&no_preempt); |
| 1309 } |
| 1310 |
| 1311 |
| 1312 void RegExpMacroAssemblerIA32::CheckStackLimit() { |
| 1313 if (FLAG_check_stack) { |
| 1314 Label no_stack_overflow; |
| 1315 ExternalReference stack_limit = |
| 1316 ExternalReference::address_of_regexp_stack_limit(); |
| 1317 __ cmp(backtrack_stackpointer(), Operand::StaticVariable(stack_limit)); |
| 1318 __ j(above, &no_stack_overflow); |
| 1319 |
| 1320 SafeCall(&stack_overflow_label_); |
| 1321 |
| 1322 __ bind(&no_stack_overflow); |
| 1323 } |
| 1324 } |
| 1325 |
| 1326 |
| 1327 void RegExpMacroAssemblerIA32::FrameAlign(int num_arguments, Register scratch) { |
| 1328 // TODO(lrn): Since we no longer use the system stack arbitrarily (but we do |
| 1329 // use it, e.g., for SafeCall), we know the number of elements on the stack |
| 1330 // since the last frame alignment. We might be able to do this simpler then. |
| 1331 int frameAlignment = OS::ActivationFrameAlignment(); |
| 1332 if (frameAlignment != 0) { |
| 1333 // Make stack end at alignment and make room for num_arguments words |
| 1334 // and the original value of esp. |
| 1335 __ mov(scratch, esp); |
| 1336 __ sub(Operand(esp), Immediate((num_arguments + 1) * kPointerSize)); |
| 1337 ASSERT(IsPowerOf2(frameAlignment)); |
| 1338 __ and_(esp, -frameAlignment); |
| 1339 __ mov(Operand(esp, num_arguments * kPointerSize), scratch); |
| 1340 } else { |
| 1341 __ sub(Operand(esp), Immediate(num_arguments * kPointerSize)); |
| 1342 } |
| 1343 } |
| 1344 |
| 1345 |
| 1346 void RegExpMacroAssemblerIA32::CallCFunction(Address function_address, |
| 1347 int num_arguments) { |
| 1348 __ mov(Operand(eax), Immediate(reinterpret_cast<int32_t>(function_address))); |
| 1349 __ call(Operand(eax)); |
| 1350 if (OS::ActivationFrameAlignment() != 0) { |
| 1351 __ mov(esp, Operand(esp, num_arguments * kPointerSize)); |
| 1352 } else { |
| 1353 __ add(Operand(esp), Immediate(num_arguments * sizeof(int32_t))); |
| 1354 } |
| 1355 } |
| 1356 |
| 1357 |
| 1358 void RegExpMacroAssemblerIA32::LoadCurrentCharacterUnchecked(int cp_offset, |
| 1359 int characters) { |
| 1360 if (mode_ == ASCII) { |
| 1361 if (characters == 4) { |
| 1362 __ mov(current_character(), Operand(esi, edi, times_1, cp_offset)); |
| 1363 } else if (characters == 2) { |
| 1364 __ movzx_w(current_character(), Operand(esi, edi, times_1, cp_offset)); |
| 1365 } else { |
| 1366 ASSERT(characters == 1); |
| 1367 __ movzx_b(current_character(), Operand(esi, edi, times_1, cp_offset)); |
| 1368 } |
| 1369 } else { |
| 1370 ASSERT(mode_ == UC16); |
| 1371 if (characters == 2) { |
| 1372 __ mov(current_character(), |
| 1373 Operand(esi, edi, times_1, cp_offset * sizeof(uc16))); |
| 1374 } else { |
| 1375 ASSERT(characters == 1); |
| 1376 __ movzx_w(current_character(), |
| 1377 Operand(esi, edi, times_1, cp_offset * sizeof(uc16))); |
| 1378 } |
| 1379 } |
| 1380 } |
| 1381 |
| 1382 |
| 1383 void RegExpMacroAssemblerIA32::LoadConstantBufferAddress(Register reg, |
| 1384 ArraySlice* buffer) { |
| 1385 __ mov(reg, buffer->array()); |
| 1386 __ add(Operand(reg), Immediate(buffer->base_offset())); |
| 1387 } |
| 1388 |
| 1389 #undef __ |
| 1390 }} // namespace v8::internal |
OLD | NEW |