Steve Block | a7e24c1 | 2009-10-30 11:49:00 +0000 | [diff] [blame] | 1 | // Copyright 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 "serialize.h" |
| 30 | #include "unicode.h" |
| 31 | #include "log.h" |
| 32 | #include "ast.h" |
| 33 | #include "regexp-stack.h" |
| 34 | #include "macro-assembler.h" |
| 35 | #include "regexp-macro-assembler.h" |
| 36 | #include "x64/macro-assembler-x64.h" |
| 37 | #include "x64/regexp-macro-assembler-x64.h" |
| 38 | |
| 39 | namespace v8 { |
| 40 | namespace internal { |
| 41 | |
| 42 | #ifdef V8_NATIVE_REGEXP |
| 43 | |
| 44 | /* |
| 45 | * This assembler uses the following register assignment convention |
| 46 | * - rdx : currently loaded character(s) as ASCII or UC16. Must be loaded using |
| 47 | * LoadCurrentCharacter before using any of the dispatch methods. |
| 48 | * - rdi : current position in input, as negative offset from end of string. |
| 49 | * Please notice that this is the byte offset, not the character |
| 50 | * offset! Is always a 32-bit signed (negative) offset, but must be |
| 51 | * maintained sign-extended to 64 bits, since it is used as index. |
| 52 | * - rsi : end of input (points to byte after last character in input), |
| 53 | * so that rsi+rdi points to the current character. |
| 54 | * - rbp : frame pointer. Used to access arguments, local variables and |
| 55 | * RegExp registers. |
| 56 | * - rsp : points to tip of C stack. |
| 57 | * - rcx : points to tip of backtrack stack. The backtrack stack contains |
| 58 | * only 32-bit values. Most are offsets from some base (e.g., character |
| 59 | * positions from end of string or code location from Code* pointer). |
| 60 | * - r8 : code object pointer. Used to convert between absolute and |
| 61 | * code-object-relative addresses. |
| 62 | * |
| 63 | * The registers rax, rbx, rcx, r9 and r11 are free to use for computations. |
| 64 | * If changed to use r12+, they should be saved as callee-save registers. |
| 65 | * |
| 66 | * Each call to a C++ method should retain these registers. |
| 67 | * |
| 68 | * The stack will have the following content, in some order, indexable from the |
| 69 | * frame pointer (see, e.g., kStackHighEnd): |
| 70 | * - stack_area_base (High end of the memory area to use as |
| 71 | * backtracking stack) |
| 72 | * - at_start (if 1, start at start of string, if 0, don't) |
| 73 | * - int* capture_array (int[num_saved_registers_], for output). |
| 74 | * - end of input (Address of end of string) |
| 75 | * - start of input (Address of first character in string) |
| 76 | * - String** input_string (location of a handle containing the string) |
| 77 | * - return address |
| 78 | * - backup of callee save registers (rbx, possibly rsi and rdi). |
| 79 | * - Offset of location before start of input (effectively character |
| 80 | * position -1). Used to initialize capture registers to a non-position. |
| 81 | * - register 0 rbp[-n] (Only positions must be stored in the first |
| 82 | * - register 1 rbp[-n-8] num_saved_registers_ registers) |
| 83 | * - ... |
| 84 | * |
| 85 | * The first num_saved_registers_ registers are initialized to point to |
| 86 | * "character -1" in the string (i.e., char_size() bytes before the first |
| 87 | * character of the string). The remaining registers starts out uninitialized. |
| 88 | * |
| 89 | * The first seven values must be provided by the calling code by |
| 90 | * calling the code's entry address cast to a function pointer with the |
| 91 | * following signature: |
| 92 | * int (*match)(String* input_string, |
| 93 | * Address start, |
| 94 | * Address end, |
| 95 | * int* capture_output_array, |
| 96 | * bool at_start, |
| 97 | * byte* stack_area_base) |
| 98 | */ |
| 99 | |
| 100 | #define __ ACCESS_MASM(masm_) |
| 101 | |
| 102 | RegExpMacroAssemblerX64::RegExpMacroAssemblerX64( |
| 103 | Mode mode, |
| 104 | int registers_to_save) |
| 105 | : masm_(new MacroAssembler(NULL, kRegExpCodeSize)), |
| 106 | code_relative_fixup_positions_(4), |
| 107 | mode_(mode), |
| 108 | num_registers_(registers_to_save), |
| 109 | num_saved_registers_(registers_to_save), |
| 110 | entry_label_(), |
| 111 | start_label_(), |
| 112 | success_label_(), |
| 113 | backtrack_label_(), |
| 114 | exit_label_() { |
| 115 | ASSERT_EQ(0, registers_to_save % 2); |
| 116 | __ jmp(&entry_label_); // We'll write the entry code when we know more. |
| 117 | __ bind(&start_label_); // And then continue from here. |
| 118 | } |
| 119 | |
| 120 | |
| 121 | RegExpMacroAssemblerX64::~RegExpMacroAssemblerX64() { |
| 122 | delete masm_; |
| 123 | // Unuse labels in case we throw away the assembler without calling GetCode. |
| 124 | entry_label_.Unuse(); |
| 125 | start_label_.Unuse(); |
| 126 | success_label_.Unuse(); |
| 127 | backtrack_label_.Unuse(); |
| 128 | exit_label_.Unuse(); |
| 129 | check_preempt_label_.Unuse(); |
| 130 | stack_overflow_label_.Unuse(); |
| 131 | } |
| 132 | |
| 133 | |
| 134 | int RegExpMacroAssemblerX64::stack_limit_slack() { |
| 135 | return RegExpStack::kStackLimitSlack; |
| 136 | } |
| 137 | |
| 138 | |
| 139 | void RegExpMacroAssemblerX64::AdvanceCurrentPosition(int by) { |
| 140 | if (by != 0) { |
| 141 | Label inside_string; |
| 142 | __ addq(rdi, Immediate(by * char_size())); |
| 143 | } |
| 144 | } |
| 145 | |
| 146 | |
| 147 | void RegExpMacroAssemblerX64::AdvanceRegister(int reg, int by) { |
| 148 | ASSERT(reg >= 0); |
| 149 | ASSERT(reg < num_registers_); |
| 150 | if (by != 0) { |
| 151 | __ addq(register_location(reg), Immediate(by)); |
| 152 | } |
| 153 | } |
| 154 | |
| 155 | |
| 156 | void RegExpMacroAssemblerX64::Backtrack() { |
| 157 | CheckPreemption(); |
| 158 | // Pop Code* offset from backtrack stack, add Code* and jump to location. |
| 159 | Pop(rbx); |
| 160 | __ addq(rbx, code_object_pointer()); |
| 161 | __ jmp(rbx); |
| 162 | } |
| 163 | |
| 164 | |
| 165 | void RegExpMacroAssemblerX64::Bind(Label* label) { |
| 166 | __ bind(label); |
| 167 | } |
| 168 | |
| 169 | |
| 170 | void RegExpMacroAssemblerX64::CheckCharacter(uint32_t c, Label* on_equal) { |
| 171 | __ cmpl(current_character(), Immediate(c)); |
| 172 | BranchOrBacktrack(equal, on_equal); |
| 173 | } |
| 174 | |
| 175 | |
| 176 | void RegExpMacroAssemblerX64::CheckCharacterGT(uc16 limit, Label* on_greater) { |
| 177 | __ cmpl(current_character(), Immediate(limit)); |
| 178 | BranchOrBacktrack(greater, on_greater); |
| 179 | } |
| 180 | |
| 181 | |
| 182 | void RegExpMacroAssemblerX64::CheckAtStart(Label* on_at_start) { |
| 183 | Label not_at_start; |
| 184 | // Did we start the match at the start of the string at all? |
| 185 | __ cmpb(Operand(rbp, kAtStart), Immediate(0)); |
| 186 | BranchOrBacktrack(equal, ¬_at_start); |
| 187 | // If we did, are we still at the start of the input? |
| 188 | __ lea(rax, Operand(rsi, rdi, times_1, 0)); |
| 189 | __ cmpq(rax, Operand(rbp, kInputStart)); |
| 190 | BranchOrBacktrack(equal, on_at_start); |
| 191 | __ bind(¬_at_start); |
| 192 | } |
| 193 | |
| 194 | |
| 195 | void RegExpMacroAssemblerX64::CheckNotAtStart(Label* on_not_at_start) { |
| 196 | // Did we start the match at the start of the string at all? |
| 197 | __ cmpb(Operand(rbp, kAtStart), Immediate(0)); |
| 198 | BranchOrBacktrack(equal, on_not_at_start); |
| 199 | // If we did, are we still at the start of the input? |
| 200 | __ lea(rax, Operand(rsi, rdi, times_1, 0)); |
| 201 | __ cmpq(rax, Operand(rbp, kInputStart)); |
| 202 | BranchOrBacktrack(not_equal, on_not_at_start); |
| 203 | } |
| 204 | |
| 205 | |
| 206 | void RegExpMacroAssemblerX64::CheckCharacterLT(uc16 limit, Label* on_less) { |
| 207 | __ cmpl(current_character(), Immediate(limit)); |
| 208 | BranchOrBacktrack(less, on_less); |
| 209 | } |
| 210 | |
| 211 | |
| 212 | void RegExpMacroAssemblerX64::CheckCharacters(Vector<const uc16> str, |
| 213 | int cp_offset, |
| 214 | Label* on_failure, |
| 215 | bool check_end_of_string) { |
| 216 | int byte_length = str.length() * char_size(); |
| 217 | int byte_offset = cp_offset * char_size(); |
| 218 | if (check_end_of_string) { |
| 219 | // Check that there are at least str.length() characters left in the input. |
| 220 | __ cmpl(rdi, Immediate(-(byte_offset + byte_length))); |
| 221 | BranchOrBacktrack(greater, on_failure); |
| 222 | } |
| 223 | |
| 224 | if (on_failure == NULL) { |
| 225 | // Instead of inlining a backtrack, (re)use the global backtrack target. |
| 226 | on_failure = &backtrack_label_; |
| 227 | } |
| 228 | |
| 229 | // TODO(lrn): Test multiple characters at a time by loading 4 or 8 bytes |
| 230 | // at a time. |
| 231 | for (int i = 0; i < str.length(); i++) { |
| 232 | if (mode_ == ASCII) { |
| 233 | __ cmpb(Operand(rsi, rdi, times_1, byte_offset + i), |
| 234 | Immediate(static_cast<int8_t>(str[i]))); |
| 235 | } else { |
| 236 | ASSERT(mode_ == UC16); |
| 237 | __ cmpw(Operand(rsi, rdi, times_1, byte_offset + i * sizeof(uc16)), |
| 238 | Immediate(str[i])); |
| 239 | } |
| 240 | BranchOrBacktrack(not_equal, on_failure); |
| 241 | } |
| 242 | } |
| 243 | |
| 244 | |
| 245 | void RegExpMacroAssemblerX64::CheckGreedyLoop(Label* on_equal) { |
| 246 | Label fallthrough; |
| 247 | __ cmpl(rdi, Operand(backtrack_stackpointer(), 0)); |
| 248 | __ j(not_equal, &fallthrough); |
| 249 | Drop(); |
| 250 | BranchOrBacktrack(no_condition, on_equal); |
| 251 | __ bind(&fallthrough); |
| 252 | } |
| 253 | |
| 254 | |
| 255 | void RegExpMacroAssemblerX64::CheckNotBackReferenceIgnoreCase( |
| 256 | int start_reg, |
| 257 | Label* on_no_match) { |
| 258 | Label fallthrough; |
| 259 | __ movq(rdx, register_location(start_reg)); // Offset of start of capture |
| 260 | __ movq(rbx, register_location(start_reg + 1)); // Offset of end of capture |
| 261 | __ subq(rbx, rdx); // Length of capture. |
| 262 | |
| 263 | // ----------------------- |
| 264 | // rdx = Start offset of capture. |
| 265 | // rbx = Length of capture |
| 266 | |
| 267 | // If length is negative, this code will fail (it's a symptom of a partial or |
| 268 | // illegal capture where start of capture after end of capture). |
| 269 | // This must not happen (no back-reference can reference a capture that wasn't |
| 270 | // closed before in the reg-exp, and we must not generate code that can cause |
| 271 | // this condition). |
| 272 | |
| 273 | // If length is zero, either the capture is empty or it is nonparticipating. |
| 274 | // In either case succeed immediately. |
| 275 | __ j(equal, &fallthrough); |
| 276 | |
| 277 | if (mode_ == ASCII) { |
| 278 | Label loop_increment; |
| 279 | if (on_no_match == NULL) { |
| 280 | on_no_match = &backtrack_label_; |
| 281 | } |
| 282 | |
| 283 | __ lea(r9, Operand(rsi, rdx, times_1, 0)); |
| 284 | __ lea(r11, Operand(rsi, rdi, times_1, 0)); |
| 285 | __ addq(rbx, r9); // End of capture |
| 286 | // --------------------- |
| 287 | // r11 - current input character address |
| 288 | // r9 - current capture character address |
| 289 | // rbx - end of capture |
| 290 | |
| 291 | Label loop; |
| 292 | __ bind(&loop); |
| 293 | __ movzxbl(rdx, Operand(r9, 0)); |
| 294 | __ movzxbl(rax, Operand(r11, 0)); |
| 295 | // al - input character |
| 296 | // dl - capture character |
| 297 | __ cmpb(rax, rdx); |
| 298 | __ j(equal, &loop_increment); |
| 299 | |
| 300 | // Mismatch, try case-insensitive match (converting letters to lower-case). |
| 301 | // I.e., if or-ing with 0x20 makes values equal and in range 'a'-'z', it's |
| 302 | // a match. |
| 303 | __ or_(rax, Immediate(0x20)); // Convert match character to lower-case. |
| 304 | __ or_(rdx, Immediate(0x20)); // Convert capture character to lower-case. |
| 305 | __ cmpb(rax, rdx); |
| 306 | __ j(not_equal, on_no_match); // Definitely not equal. |
| 307 | __ subb(rax, Immediate('a')); |
| 308 | __ cmpb(rax, Immediate('z' - 'a')); |
| 309 | __ j(above, on_no_match); // Weren't letters anyway. |
| 310 | |
| 311 | __ bind(&loop_increment); |
| 312 | // Increment pointers into match and capture strings. |
| 313 | __ addq(r11, Immediate(1)); |
| 314 | __ addq(r9, Immediate(1)); |
| 315 | // Compare to end of capture, and loop if not done. |
| 316 | __ cmpq(r9, rbx); |
| 317 | __ j(below, &loop); |
| 318 | |
| 319 | // Compute new value of character position after the matched part. |
| 320 | __ movq(rdi, r11); |
| 321 | __ subq(rdi, rsi); |
| 322 | } else { |
| 323 | ASSERT(mode_ == UC16); |
| 324 | // Save important/volatile registers before calling C function. |
| 325 | #ifndef _WIN64 |
| 326 | // Callee save on Win64 |
| 327 | __ push(rsi); |
| 328 | __ push(rdi); |
| 329 | #endif |
| 330 | __ push(backtrack_stackpointer()); |
| 331 | |
| 332 | int num_arguments = 3; |
| 333 | FrameAlign(num_arguments); |
| 334 | |
| 335 | // Put arguments into parameter registers. 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 | #ifdef _WIN64 |
| 340 | // Compute and set byte_offset1 (start of capture). |
| 341 | __ lea(rcx, Operand(rsi, rdx, times_1, 0)); |
| 342 | // Set byte_offset2. |
| 343 | __ lea(rdx, Operand(rsi, rdi, times_1, 0)); |
| 344 | // Set byte_length. |
| 345 | __ movq(r8, rbx); |
| 346 | #else // AMD64 calling convention |
| 347 | // Compute byte_offset2 (current position = rsi+rdi). |
| 348 | __ lea(rax, Operand(rsi, rdi, times_1, 0)); |
| 349 | // Compute and set byte_offset1 (start of capture). |
| 350 | __ lea(rdi, Operand(rsi, rdx, times_1, 0)); |
| 351 | // Set byte_offset2. |
| 352 | __ movq(rsi, rax); |
| 353 | // Set byte_length. |
| 354 | __ movq(rdx, rbx); |
| 355 | #endif |
| 356 | ExternalReference compare = |
| 357 | ExternalReference::re_case_insensitive_compare_uc16(); |
| 358 | CallCFunction(compare, num_arguments); |
| 359 | |
| 360 | // Restore original values before reacting on result value. |
| 361 | __ Move(code_object_pointer(), masm_->CodeObject()); |
| 362 | __ pop(backtrack_stackpointer()); |
| 363 | #ifndef _WIN64 |
| 364 | __ pop(rdi); |
| 365 | __ pop(rsi); |
| 366 | #endif |
| 367 | |
| 368 | // Check if function returned non-zero for success or zero for failure. |
| 369 | __ testq(rax, rax); |
| 370 | BranchOrBacktrack(zero, on_no_match); |
| 371 | // On success, increment position by length of capture. |
| 372 | // Requires that rbx is callee save (true for both Win64 and AMD64 ABIs). |
| 373 | __ addq(rdi, rbx); |
| 374 | } |
| 375 | __ bind(&fallthrough); |
| 376 | } |
| 377 | |
| 378 | |
| 379 | void RegExpMacroAssemblerX64::CheckNotBackReference( |
| 380 | int start_reg, |
| 381 | Label* on_no_match) { |
| 382 | Label fallthrough; |
| 383 | |
| 384 | // Find length of back-referenced capture. |
| 385 | __ movq(rdx, register_location(start_reg)); |
| 386 | __ movq(rax, register_location(start_reg + 1)); |
| 387 | __ subq(rax, rdx); // Length to check. |
| 388 | |
| 389 | // Fail on partial or illegal capture (start of capture after end of capture). |
| 390 | // This must not happen (no back-reference can reference a capture that wasn't |
| 391 | // closed before in the reg-exp). |
| 392 | __ Check(greater_equal, "Invalid capture referenced"); |
| 393 | |
| 394 | // Succeed on empty capture (including non-participating capture) |
| 395 | __ j(equal, &fallthrough); |
| 396 | |
| 397 | // ----------------------- |
| 398 | // rdx - Start of capture |
| 399 | // rax - length of capture |
| 400 | |
| 401 | // Check that there are sufficient characters left in the input. |
| 402 | __ movl(rbx, rdi); |
| 403 | __ addl(rbx, rax); |
| 404 | BranchOrBacktrack(greater, on_no_match); |
| 405 | |
| 406 | // Compute pointers to match string and capture string |
| 407 | __ lea(rbx, Operand(rsi, rdi, times_1, 0)); // Start of match. |
| 408 | __ addq(rdx, rsi); // Start of capture. |
| 409 | __ lea(r9, Operand(rdx, rax, times_1, 0)); // End of capture |
| 410 | |
| 411 | // ----------------------- |
| 412 | // rbx - current capture character address. |
| 413 | // rbx - current input character address . |
| 414 | // r9 - end of input to match (capture length after rbx). |
| 415 | |
| 416 | Label loop; |
| 417 | __ bind(&loop); |
| 418 | if (mode_ == ASCII) { |
| 419 | __ movzxbl(rax, Operand(rdx, 0)); |
| 420 | __ cmpb(rax, Operand(rbx, 0)); |
| 421 | } else { |
| 422 | ASSERT(mode_ == UC16); |
| 423 | __ movzxwl(rax, Operand(rdx, 0)); |
| 424 | __ cmpw(rax, Operand(rbx, 0)); |
| 425 | } |
| 426 | BranchOrBacktrack(not_equal, on_no_match); |
| 427 | // Increment pointers into capture and match string. |
| 428 | __ addq(rbx, Immediate(char_size())); |
| 429 | __ addq(rdx, Immediate(char_size())); |
| 430 | // Check if we have reached end of match area. |
| 431 | __ cmpq(rdx, r9); |
| 432 | __ j(below, &loop); |
| 433 | |
| 434 | // Success. |
| 435 | // Set current character position to position after match. |
| 436 | __ movq(rdi, rbx); |
| 437 | __ subq(rdi, rsi); |
| 438 | |
| 439 | __ bind(&fallthrough); |
| 440 | } |
| 441 | |
| 442 | |
| 443 | void RegExpMacroAssemblerX64::CheckNotRegistersEqual(int reg1, |
| 444 | int reg2, |
| 445 | Label* on_not_equal) { |
| 446 | __ movq(rax, register_location(reg1)); |
| 447 | __ cmpq(rax, register_location(reg2)); |
| 448 | BranchOrBacktrack(not_equal, on_not_equal); |
| 449 | } |
| 450 | |
| 451 | |
| 452 | void RegExpMacroAssemblerX64::CheckNotCharacter(uint32_t c, |
| 453 | Label* on_not_equal) { |
| 454 | __ cmpl(current_character(), Immediate(c)); |
| 455 | BranchOrBacktrack(not_equal, on_not_equal); |
| 456 | } |
| 457 | |
| 458 | |
| 459 | void RegExpMacroAssemblerX64::CheckCharacterAfterAnd(uint32_t c, |
| 460 | uint32_t mask, |
| 461 | Label* on_equal) { |
| 462 | __ movl(rax, current_character()); |
| 463 | __ and_(rax, Immediate(mask)); |
| 464 | __ cmpl(rax, Immediate(c)); |
| 465 | BranchOrBacktrack(equal, on_equal); |
| 466 | } |
| 467 | |
| 468 | |
| 469 | void RegExpMacroAssemblerX64::CheckNotCharacterAfterAnd(uint32_t c, |
| 470 | uint32_t mask, |
| 471 | Label* on_not_equal) { |
| 472 | __ movl(rax, current_character()); |
| 473 | __ and_(rax, Immediate(mask)); |
| 474 | __ cmpl(rax, Immediate(c)); |
| 475 | BranchOrBacktrack(not_equal, on_not_equal); |
| 476 | } |
| 477 | |
| 478 | |
| 479 | void RegExpMacroAssemblerX64::CheckNotCharacterAfterMinusAnd( |
| 480 | uc16 c, |
| 481 | uc16 minus, |
| 482 | uc16 mask, |
| 483 | Label* on_not_equal) { |
| 484 | ASSERT(minus < String::kMaxUC16CharCode); |
| 485 | __ lea(rax, Operand(current_character(), -minus)); |
| 486 | __ and_(rax, Immediate(mask)); |
| 487 | __ cmpl(rax, Immediate(c)); |
| 488 | BranchOrBacktrack(not_equal, on_not_equal); |
| 489 | } |
| 490 | |
| 491 | |
| 492 | bool RegExpMacroAssemblerX64::CheckSpecialCharacterClass(uc16 type, |
| 493 | int cp_offset, |
| 494 | bool check_offset, |
| 495 | Label* on_no_match) { |
| 496 | // Range checks (c in min..max) are generally implemented by an unsigned |
| 497 | // (c - min) <= (max - min) check |
| 498 | switch (type) { |
| 499 | case 's': |
| 500 | // Match space-characters |
| 501 | if (mode_ == ASCII) { |
| 502 | // ASCII space characters are '\t'..'\r' and ' '. |
| 503 | if (check_offset) { |
| 504 | LoadCurrentCharacter(cp_offset, on_no_match); |
| 505 | } else { |
| 506 | LoadCurrentCharacterUnchecked(cp_offset, 1); |
| 507 | } |
| 508 | Label success; |
| 509 | __ cmpl(current_character(), Immediate(' ')); |
| 510 | __ j(equal, &success); |
| 511 | // Check range 0x09..0x0d |
| 512 | __ subl(current_character(), Immediate('\t')); |
| 513 | __ cmpl(current_character(), Immediate('\r' - '\t')); |
| 514 | BranchOrBacktrack(above, on_no_match); |
| 515 | __ bind(&success); |
| 516 | return true; |
| 517 | } |
| 518 | return false; |
| 519 | case 'S': |
| 520 | // Match non-space characters. |
| 521 | if (check_offset) { |
| 522 | LoadCurrentCharacter(cp_offset, on_no_match, 1); |
| 523 | } else { |
| 524 | LoadCurrentCharacterUnchecked(cp_offset, 1); |
| 525 | } |
| 526 | if (mode_ == ASCII) { |
| 527 | // ASCII space characters are '\t'..'\r' and ' '. |
| 528 | __ cmpl(current_character(), Immediate(' ')); |
| 529 | BranchOrBacktrack(equal, on_no_match); |
| 530 | __ subl(current_character(), Immediate('\t')); |
| 531 | __ cmpl(current_character(), Immediate('\r' - '\t')); |
| 532 | BranchOrBacktrack(below_equal, on_no_match); |
| 533 | return true; |
| 534 | } |
| 535 | return false; |
| 536 | case 'd': |
| 537 | // Match ASCII digits ('0'..'9') |
| 538 | if (check_offset) { |
| 539 | LoadCurrentCharacter(cp_offset, on_no_match, 1); |
| 540 | } else { |
| 541 | LoadCurrentCharacterUnchecked(cp_offset, 1); |
| 542 | } |
| 543 | __ subl(current_character(), Immediate('0')); |
| 544 | __ cmpl(current_character(), Immediate('9' - '0')); |
| 545 | BranchOrBacktrack(above, on_no_match); |
| 546 | return true; |
| 547 | case 'D': |
| 548 | // Match non ASCII-digits |
| 549 | if (check_offset) { |
| 550 | LoadCurrentCharacter(cp_offset, on_no_match, 1); |
| 551 | } else { |
| 552 | LoadCurrentCharacterUnchecked(cp_offset, 1); |
| 553 | } |
| 554 | __ subl(current_character(), Immediate('0')); |
| 555 | __ cmpl(current_character(), Immediate('9' - '0')); |
| 556 | BranchOrBacktrack(below_equal, on_no_match); |
| 557 | return true; |
| 558 | case '.': { |
| 559 | // Match non-newlines (not 0x0a('\n'), 0x0d('\r'), 0x2028 and 0x2029) |
| 560 | if (check_offset) { |
| 561 | LoadCurrentCharacter(cp_offset, on_no_match, 1); |
| 562 | } else { |
| 563 | LoadCurrentCharacterUnchecked(cp_offset, 1); |
| 564 | } |
| 565 | __ xor_(current_character(), Immediate(0x01)); |
| 566 | // See if current character is '\n'^1 or '\r'^1, i.e., 0x0b or 0x0c |
| 567 | __ subl(current_character(), Immediate(0x0b)); |
| 568 | __ cmpl(current_character(), Immediate(0x0c - 0x0b)); |
| 569 | BranchOrBacktrack(below_equal, on_no_match); |
| 570 | if (mode_ == UC16) { |
| 571 | // Compare original value to 0x2028 and 0x2029, using the already |
| 572 | // computed (current_char ^ 0x01 - 0x0b). I.e., check for |
| 573 | // 0x201d (0x2028 - 0x0b) or 0x201e. |
| 574 | __ subl(current_character(), Immediate(0x2028 - 0x0b)); |
| 575 | __ cmpl(current_character(), Immediate(1)); |
| 576 | BranchOrBacktrack(below_equal, on_no_match); |
| 577 | } |
| 578 | return true; |
| 579 | } |
| 580 | case '*': |
| 581 | // Match any character. |
| 582 | if (check_offset) { |
| 583 | CheckPosition(cp_offset, on_no_match); |
| 584 | } |
| 585 | return true; |
| 586 | // No custom implementation (yet): w, W, s(UC16), S(UC16). |
| 587 | default: |
| 588 | return false; |
| 589 | } |
| 590 | } |
| 591 | |
| 592 | |
| 593 | void RegExpMacroAssemblerX64::Fail() { |
| 594 | ASSERT(FAILURE == 0); // Return value for failure is zero. |
| 595 | __ xor_(rax, rax); // zero rax. |
| 596 | __ jmp(&exit_label_); |
| 597 | } |
| 598 | |
| 599 | |
| 600 | Handle<Object> RegExpMacroAssemblerX64::GetCode(Handle<String> source) { |
| 601 | // Finalize code - write the entry point code now we know how many |
| 602 | // registers we need. |
| 603 | |
| 604 | // Entry code: |
| 605 | __ bind(&entry_label_); |
| 606 | // Start new stack frame. |
| 607 | __ push(rbp); |
| 608 | __ movq(rbp, rsp); |
| 609 | // Save parameters and callee-save registers. Order here should correspond |
| 610 | // to order of kBackup_ebx etc. |
| 611 | #ifdef _WIN64 |
| 612 | // MSVC passes arguments in rcx, rdx, r8, r9, with backing stack slots. |
| 613 | // Store register parameters in pre-allocated stack slots, |
| 614 | __ movq(Operand(rbp, kInputString), rcx); |
| 615 | __ movq(Operand(rbp, kStartIndex), rdx); // Passed as int32 in edx. |
| 616 | __ movq(Operand(rbp, kInputStart), r8); |
| 617 | __ movq(Operand(rbp, kInputEnd), r9); |
| 618 | // Callee-save on Win64. |
| 619 | __ push(rsi); |
| 620 | __ push(rdi); |
| 621 | __ push(rbx); |
| 622 | #else |
| 623 | // GCC passes arguments in rdi, rsi, rdx, rcx, r8, r9 (and then on stack). |
| 624 | // Push register parameters on stack for reference. |
| 625 | ASSERT_EQ(kInputString, -1 * kPointerSize); |
| 626 | ASSERT_EQ(kStartIndex, -2 * kPointerSize); |
| 627 | ASSERT_EQ(kInputStart, -3 * kPointerSize); |
| 628 | ASSERT_EQ(kInputEnd, -4 * kPointerSize); |
| 629 | ASSERT_EQ(kRegisterOutput, -5 * kPointerSize); |
| 630 | ASSERT_EQ(kAtStart, -6 * kPointerSize); |
| 631 | __ push(rdi); |
| 632 | __ push(rsi); |
| 633 | __ push(rdx); |
| 634 | __ push(rcx); |
| 635 | __ push(r8); |
| 636 | __ push(r9); |
| 637 | |
| 638 | __ push(rbx); // Callee-save |
| 639 | #endif |
| 640 | __ push(Immediate(0)); // Make room for "input start - 1" constant. |
| 641 | |
| 642 | // Check if we have space on the stack for registers. |
| 643 | Label stack_limit_hit; |
| 644 | Label stack_ok; |
| 645 | |
Steve Block | d0582a6 | 2009-12-15 09:54:21 +0000 | [diff] [blame^] | 646 | ExternalReference stack_limit = |
| 647 | ExternalReference::address_of_stack_limit(); |
Steve Block | a7e24c1 | 2009-10-30 11:49:00 +0000 | [diff] [blame] | 648 | __ movq(rcx, rsp); |
Steve Block | d0582a6 | 2009-12-15 09:54:21 +0000 | [diff] [blame^] | 649 | __ movq(kScratchRegister, stack_limit); |
Steve Block | a7e24c1 | 2009-10-30 11:49:00 +0000 | [diff] [blame] | 650 | __ subq(rcx, Operand(kScratchRegister, 0)); |
| 651 | // Handle it if the stack pointer is already below the stack limit. |
| 652 | __ j(below_equal, &stack_limit_hit); |
| 653 | // Check if there is room for the variable number of registers above |
| 654 | // the stack limit. |
| 655 | __ cmpq(rcx, Immediate(num_registers_ * kPointerSize)); |
| 656 | __ j(above_equal, &stack_ok); |
| 657 | // Exit with OutOfMemory exception. There is not enough space on the stack |
| 658 | // for our working registers. |
| 659 | __ movq(rax, Immediate(EXCEPTION)); |
| 660 | __ jmp(&exit_label_); |
| 661 | |
| 662 | __ bind(&stack_limit_hit); |
| 663 | __ Move(code_object_pointer(), masm_->CodeObject()); |
| 664 | CallCheckStackGuardState(); // Preserves no registers beside rbp and rsp. |
| 665 | __ testq(rax, rax); |
| 666 | // If returned value is non-zero, we exit with the returned value as result. |
| 667 | __ j(not_zero, &exit_label_); |
| 668 | |
| 669 | __ bind(&stack_ok); |
| 670 | |
| 671 | // Allocate space on stack for registers. |
| 672 | __ subq(rsp, Immediate(num_registers_ * kPointerSize)); |
| 673 | // Load string length. |
| 674 | __ movq(rsi, Operand(rbp, kInputEnd)); |
| 675 | // Load input position. |
| 676 | __ movq(rdi, Operand(rbp, kInputStart)); |
| 677 | // Set up rdi to be negative offset from string end. |
| 678 | __ subq(rdi, rsi); |
| 679 | // Set rax to address of char before start of input |
| 680 | // (effectively string position -1). |
| 681 | __ lea(rax, Operand(rdi, -char_size())); |
| 682 | // Store this value in a local variable, for use when clearing |
| 683 | // position registers. |
| 684 | __ movq(Operand(rbp, kInputStartMinusOne), rax); |
| 685 | if (num_saved_registers_ > 0) { |
| 686 | // Fill saved registers with initial value = start offset - 1 |
| 687 | // Fill in stack push order, to avoid accessing across an unwritten |
| 688 | // page (a problem on Windows). |
| 689 | __ movq(rcx, Immediate(kRegisterZero)); |
| 690 | Label init_loop; |
| 691 | __ bind(&init_loop); |
| 692 | __ movq(Operand(rbp, rcx, times_1, 0), rax); |
| 693 | __ subq(rcx, Immediate(kPointerSize)); |
| 694 | __ cmpq(rcx, |
| 695 | Immediate(kRegisterZero - num_saved_registers_ * kPointerSize)); |
| 696 | __ j(greater, &init_loop); |
| 697 | } |
| 698 | // Ensure that we have written to each stack page, in order. Skipping a page |
| 699 | // on Windows can cause segmentation faults. Assuming page size is 4k. |
| 700 | const int kPageSize = 4096; |
| 701 | const int kRegistersPerPage = kPageSize / kPointerSize; |
| 702 | for (int i = num_saved_registers_ + kRegistersPerPage - 1; |
| 703 | i < num_registers_; |
| 704 | i += kRegistersPerPage) { |
| 705 | __ movq(register_location(i), rax); // One write every page. |
| 706 | } |
| 707 | |
| 708 | // Initialize backtrack stack pointer. |
| 709 | __ movq(backtrack_stackpointer(), Operand(rbp, kStackHighEnd)); |
| 710 | // Initialize code object pointer. |
| 711 | __ Move(code_object_pointer(), masm_->CodeObject()); |
| 712 | // Load previous char as initial value of current-character. |
| 713 | Label at_start; |
| 714 | __ cmpb(Operand(rbp, kAtStart), Immediate(0)); |
| 715 | __ j(not_equal, &at_start); |
| 716 | LoadCurrentCharacterUnchecked(-1, 1); // Load previous char. |
| 717 | __ jmp(&start_label_); |
| 718 | __ bind(&at_start); |
| 719 | __ movq(current_character(), Immediate('\n')); |
| 720 | __ jmp(&start_label_); |
| 721 | |
| 722 | |
| 723 | // Exit code: |
| 724 | if (success_label_.is_linked()) { |
| 725 | // Save captures when successful. |
| 726 | __ bind(&success_label_); |
| 727 | if (num_saved_registers_ > 0) { |
| 728 | // copy captures to output |
| 729 | __ movq(rbx, Operand(rbp, kRegisterOutput)); |
| 730 | __ movq(rcx, Operand(rbp, kInputEnd)); |
| 731 | __ subq(rcx, Operand(rbp, kInputStart)); |
| 732 | for (int i = 0; i < num_saved_registers_; i++) { |
| 733 | __ movq(rax, register_location(i)); |
| 734 | __ addq(rax, rcx); // Convert to index from start, not end. |
| 735 | if (mode_ == UC16) { |
| 736 | __ sar(rax, Immediate(1)); // Convert byte index to character index. |
| 737 | } |
| 738 | __ movl(Operand(rbx, i * kIntSize), rax); |
| 739 | } |
| 740 | } |
| 741 | __ movq(rax, Immediate(SUCCESS)); |
| 742 | } |
| 743 | |
| 744 | // Exit and return rax |
| 745 | __ bind(&exit_label_); |
| 746 | |
| 747 | #ifdef _WIN64 |
| 748 | // Restore callee save registers. |
| 749 | __ lea(rsp, Operand(rbp, kLastCalleeSaveRegister)); |
| 750 | __ pop(rbx); |
| 751 | __ pop(rdi); |
| 752 | __ pop(rsi); |
| 753 | // Stack now at rbp. |
| 754 | #else |
| 755 | // Restore callee save register. |
| 756 | __ movq(rbx, Operand(rbp, kBackup_rbx)); |
| 757 | // Skip rsp to rbp. |
| 758 | __ movq(rsp, rbp); |
| 759 | #endif |
| 760 | // Exit function frame, restore previous one. |
| 761 | __ pop(rbp); |
| 762 | __ ret(0); |
| 763 | |
| 764 | // Backtrack code (branch target for conditional backtracks). |
| 765 | if (backtrack_label_.is_linked()) { |
| 766 | __ bind(&backtrack_label_); |
| 767 | Backtrack(); |
| 768 | } |
| 769 | |
| 770 | Label exit_with_exception; |
| 771 | |
| 772 | // Preempt-code |
| 773 | if (check_preempt_label_.is_linked()) { |
| 774 | SafeCallTarget(&check_preempt_label_); |
| 775 | |
| 776 | __ push(backtrack_stackpointer()); |
| 777 | __ push(rdi); |
| 778 | |
| 779 | CallCheckStackGuardState(); |
| 780 | __ testq(rax, rax); |
| 781 | // If returning non-zero, we should end execution with the given |
| 782 | // result as return value. |
| 783 | __ j(not_zero, &exit_label_); |
| 784 | |
| 785 | // Restore registers. |
| 786 | __ Move(code_object_pointer(), masm_->CodeObject()); |
| 787 | __ pop(rdi); |
| 788 | __ pop(backtrack_stackpointer()); |
| 789 | // String might have moved: Reload esi from frame. |
| 790 | __ movq(rsi, Operand(rbp, kInputEnd)); |
| 791 | SafeReturn(); |
| 792 | } |
| 793 | |
| 794 | // Backtrack stack overflow code. |
| 795 | if (stack_overflow_label_.is_linked()) { |
| 796 | SafeCallTarget(&stack_overflow_label_); |
| 797 | // Reached if the backtrack-stack limit has been hit. |
| 798 | |
| 799 | Label grow_failed; |
| 800 | // Save registers before calling C function |
| 801 | #ifndef _WIN64 |
| 802 | // Callee-save in Microsoft 64-bit ABI, but not in AMD64 ABI. |
| 803 | __ push(rsi); |
| 804 | __ push(rdi); |
| 805 | #endif |
| 806 | |
| 807 | // Call GrowStack(backtrack_stackpointer()) |
| 808 | int num_arguments = 2; |
| 809 | FrameAlign(num_arguments); |
| 810 | #ifdef _WIN64 |
| 811 | // Microsoft passes parameters in rcx, rdx. |
| 812 | // First argument, backtrack stackpointer, is already in rcx. |
| 813 | __ lea(rdx, Operand(rbp, kStackHighEnd)); // Second argument |
| 814 | #else |
| 815 | // AMD64 ABI passes parameters in rdi, rsi. |
| 816 | __ movq(rdi, backtrack_stackpointer()); // First argument. |
| 817 | __ lea(rsi, Operand(rbp, kStackHighEnd)); // Second argument. |
| 818 | #endif |
| 819 | ExternalReference grow_stack = ExternalReference::re_grow_stack(); |
| 820 | CallCFunction(grow_stack, num_arguments); |
| 821 | // If return NULL, we have failed to grow the stack, and |
| 822 | // must exit with a stack-overflow exception. |
| 823 | __ testq(rax, rax); |
| 824 | __ j(equal, &exit_with_exception); |
| 825 | // Otherwise use return value as new stack pointer. |
| 826 | __ movq(backtrack_stackpointer(), rax); |
| 827 | // Restore saved registers and continue. |
| 828 | __ Move(code_object_pointer(), masm_->CodeObject()); |
| 829 | #ifndef _WIN64 |
| 830 | __ pop(rdi); |
| 831 | __ pop(rsi); |
| 832 | #endif |
| 833 | SafeReturn(); |
| 834 | } |
| 835 | |
| 836 | if (exit_with_exception.is_linked()) { |
| 837 | // If any of the code above needed to exit with an exception. |
| 838 | __ bind(&exit_with_exception); |
| 839 | // Exit with Result EXCEPTION(-1) to signal thrown exception. |
| 840 | __ movq(rax, Immediate(EXCEPTION)); |
| 841 | __ jmp(&exit_label_); |
| 842 | } |
| 843 | |
| 844 | FixupCodeRelativePositions(); |
| 845 | |
| 846 | CodeDesc code_desc; |
| 847 | masm_->GetCode(&code_desc); |
| 848 | Handle<Code> code = Factory::NewCode(code_desc, |
| 849 | NULL, |
| 850 | Code::ComputeFlags(Code::REGEXP), |
| 851 | masm_->CodeObject()); |
| 852 | LOG(RegExpCodeCreateEvent(*code, *source)); |
| 853 | return Handle<Object>::cast(code); |
| 854 | } |
| 855 | |
| 856 | |
| 857 | void RegExpMacroAssemblerX64::GoTo(Label* to) { |
| 858 | BranchOrBacktrack(no_condition, to); |
| 859 | } |
| 860 | |
| 861 | |
| 862 | void RegExpMacroAssemblerX64::IfRegisterGE(int reg, |
| 863 | int comparand, |
| 864 | Label* if_ge) { |
| 865 | __ cmpq(register_location(reg), Immediate(comparand)); |
| 866 | BranchOrBacktrack(greater_equal, if_ge); |
| 867 | } |
| 868 | |
| 869 | |
| 870 | void RegExpMacroAssemblerX64::IfRegisterLT(int reg, |
| 871 | int comparand, |
| 872 | Label* if_lt) { |
| 873 | __ cmpq(register_location(reg), Immediate(comparand)); |
| 874 | BranchOrBacktrack(less, if_lt); |
| 875 | } |
| 876 | |
| 877 | |
| 878 | void RegExpMacroAssemblerX64::IfRegisterEqPos(int reg, |
| 879 | Label* if_eq) { |
| 880 | __ cmpq(rdi, register_location(reg)); |
| 881 | BranchOrBacktrack(equal, if_eq); |
| 882 | } |
| 883 | |
| 884 | |
| 885 | RegExpMacroAssembler::IrregexpImplementation |
| 886 | RegExpMacroAssemblerX64::Implementation() { |
| 887 | return kX64Implementation; |
| 888 | } |
| 889 | |
| 890 | |
| 891 | void RegExpMacroAssemblerX64::LoadCurrentCharacter(int cp_offset, |
| 892 | Label* on_end_of_input, |
| 893 | bool check_bounds, |
| 894 | int characters) { |
| 895 | ASSERT(cp_offset >= -1); // ^ and \b can look behind one character. |
| 896 | ASSERT(cp_offset < (1<<30)); // Be sane! (And ensure negation works) |
| 897 | if (check_bounds) { |
| 898 | CheckPosition(cp_offset + characters - 1, on_end_of_input); |
| 899 | } |
| 900 | LoadCurrentCharacterUnchecked(cp_offset, characters); |
| 901 | } |
| 902 | |
| 903 | |
| 904 | void RegExpMacroAssemblerX64::PopCurrentPosition() { |
| 905 | Pop(rdi); |
| 906 | } |
| 907 | |
| 908 | |
| 909 | void RegExpMacroAssemblerX64::PopRegister(int register_index) { |
| 910 | Pop(rax); |
| 911 | __ movq(register_location(register_index), rax); |
| 912 | } |
| 913 | |
| 914 | |
| 915 | void RegExpMacroAssemblerX64::PushBacktrack(Label* label) { |
| 916 | Push(label); |
| 917 | CheckStackLimit(); |
| 918 | } |
| 919 | |
| 920 | |
| 921 | void RegExpMacroAssemblerX64::PushCurrentPosition() { |
| 922 | Push(rdi); |
| 923 | } |
| 924 | |
| 925 | |
| 926 | void RegExpMacroAssemblerX64::PushRegister(int register_index, |
| 927 | StackCheckFlag check_stack_limit) { |
| 928 | __ movq(rax, register_location(register_index)); |
| 929 | Push(rax); |
| 930 | if (check_stack_limit) CheckStackLimit(); |
| 931 | } |
| 932 | |
| 933 | |
| 934 | void RegExpMacroAssemblerX64::ReadCurrentPositionFromRegister(int reg) { |
| 935 | __ movq(rdi, register_location(reg)); |
| 936 | } |
| 937 | |
| 938 | |
| 939 | void RegExpMacroAssemblerX64::ReadStackPointerFromRegister(int reg) { |
| 940 | __ movq(backtrack_stackpointer(), register_location(reg)); |
| 941 | __ addq(backtrack_stackpointer(), Operand(rbp, kStackHighEnd)); |
| 942 | } |
| 943 | |
| 944 | |
| 945 | void RegExpMacroAssemblerX64::SetRegister(int register_index, int to) { |
| 946 | ASSERT(register_index >= num_saved_registers_); // Reserved for positions! |
| 947 | __ movq(register_location(register_index), Immediate(to)); |
| 948 | } |
| 949 | |
| 950 | |
| 951 | void RegExpMacroAssemblerX64::Succeed() { |
| 952 | __ jmp(&success_label_); |
| 953 | } |
| 954 | |
| 955 | |
| 956 | void RegExpMacroAssemblerX64::WriteCurrentPositionToRegister(int reg, |
| 957 | int cp_offset) { |
| 958 | if (cp_offset == 0) { |
| 959 | __ movq(register_location(reg), rdi); |
| 960 | } else { |
| 961 | __ lea(rax, Operand(rdi, cp_offset * char_size())); |
| 962 | __ movq(register_location(reg), rax); |
| 963 | } |
| 964 | } |
| 965 | |
| 966 | |
| 967 | void RegExpMacroAssemblerX64::ClearRegisters(int reg_from, int reg_to) { |
| 968 | ASSERT(reg_from <= reg_to); |
| 969 | __ movq(rax, Operand(rbp, kInputStartMinusOne)); |
| 970 | for (int reg = reg_from; reg <= reg_to; reg++) { |
| 971 | __ movq(register_location(reg), rax); |
| 972 | } |
| 973 | } |
| 974 | |
| 975 | |
| 976 | void RegExpMacroAssemblerX64::WriteStackPointerToRegister(int reg) { |
| 977 | __ movq(rax, backtrack_stackpointer()); |
| 978 | __ subq(rax, Operand(rbp, kStackHighEnd)); |
| 979 | __ movq(register_location(reg), rax); |
| 980 | } |
| 981 | |
| 982 | |
| 983 | // Private methods: |
| 984 | |
| 985 | void RegExpMacroAssemblerX64::CallCheckStackGuardState() { |
| 986 | // This function call preserves no register values. Caller should |
| 987 | // store anything volatile in a C call or overwritten by this function. |
| 988 | int num_arguments = 3; |
| 989 | FrameAlign(num_arguments); |
| 990 | #ifdef _WIN64 |
| 991 | // Second argument: Code* of self. (Do this before overwriting r8). |
| 992 | __ movq(rdx, code_object_pointer()); |
| 993 | // Third argument: RegExp code frame pointer. |
| 994 | __ movq(r8, rbp); |
| 995 | // First argument: Next address on the stack (will be address of |
| 996 | // return address). |
| 997 | __ lea(rcx, Operand(rsp, -kPointerSize)); |
| 998 | #else |
| 999 | // Third argument: RegExp code frame pointer. |
| 1000 | __ movq(rdx, rbp); |
| 1001 | // Second argument: Code* of self. |
| 1002 | __ movq(rsi, code_object_pointer()); |
| 1003 | // First argument: Next address on the stack (will be address of |
| 1004 | // return address). |
| 1005 | __ lea(rdi, Operand(rsp, -kPointerSize)); |
| 1006 | #endif |
| 1007 | ExternalReference stack_check = |
| 1008 | ExternalReference::re_check_stack_guard_state(); |
| 1009 | CallCFunction(stack_check, num_arguments); |
| 1010 | } |
| 1011 | |
| 1012 | |
| 1013 | // Helper function for reading a value out of a stack frame. |
| 1014 | template <typename T> |
| 1015 | static T& frame_entry(Address re_frame, int frame_offset) { |
| 1016 | return reinterpret_cast<T&>(Memory::int32_at(re_frame + frame_offset)); |
| 1017 | } |
| 1018 | |
| 1019 | |
| 1020 | int RegExpMacroAssemblerX64::CheckStackGuardState(Address* return_address, |
| 1021 | Code* re_code, |
| 1022 | Address re_frame) { |
| 1023 | if (StackGuard::IsStackOverflow()) { |
| 1024 | Top::StackOverflow(); |
| 1025 | return EXCEPTION; |
| 1026 | } |
| 1027 | |
| 1028 | // If not real stack overflow the stack guard was used to interrupt |
| 1029 | // execution for another purpose. |
| 1030 | |
| 1031 | // Prepare for possible GC. |
| 1032 | HandleScope handles; |
| 1033 | Handle<Code> code_handle(re_code); |
| 1034 | |
| 1035 | Handle<String> subject(frame_entry<String*>(re_frame, kInputString)); |
| 1036 | // Current string. |
| 1037 | bool is_ascii = subject->IsAsciiRepresentation(); |
| 1038 | |
| 1039 | ASSERT(re_code->instruction_start() <= *return_address); |
| 1040 | ASSERT(*return_address <= |
| 1041 | re_code->instruction_start() + re_code->instruction_size()); |
| 1042 | |
| 1043 | Object* result = Execution::HandleStackGuardInterrupt(); |
| 1044 | |
| 1045 | if (*code_handle != re_code) { // Return address no longer valid |
| 1046 | intptr_t delta = *code_handle - re_code; |
| 1047 | // Overwrite the return address on the stack. |
| 1048 | *return_address += delta; |
| 1049 | } |
| 1050 | |
| 1051 | if (result->IsException()) { |
| 1052 | return EXCEPTION; |
| 1053 | } |
| 1054 | |
| 1055 | // String might have changed. |
| 1056 | if (subject->IsAsciiRepresentation() != is_ascii) { |
| 1057 | // If we changed between an ASCII and an UC16 string, the specialized |
| 1058 | // code cannot be used, and we need to restart regexp matching from |
| 1059 | // scratch (including, potentially, compiling a new version of the code). |
| 1060 | return RETRY; |
| 1061 | } |
| 1062 | |
| 1063 | // Otherwise, the content of the string might have moved. It must still |
| 1064 | // be a sequential or external string with the same content. |
| 1065 | // Update the start and end pointers in the stack frame to the current |
| 1066 | // location (whether it has actually moved or not). |
| 1067 | ASSERT(StringShape(*subject).IsSequential() || |
| 1068 | StringShape(*subject).IsExternal()); |
| 1069 | |
| 1070 | // The original start address of the characters to match. |
| 1071 | const byte* start_address = frame_entry<const byte*>(re_frame, kInputStart); |
| 1072 | |
| 1073 | // Find the current start address of the same character at the current string |
| 1074 | // position. |
| 1075 | int start_index = frame_entry<int>(re_frame, kStartIndex); |
| 1076 | const byte* new_address = StringCharacterPosition(*subject, start_index); |
| 1077 | |
| 1078 | if (start_address != new_address) { |
| 1079 | // If there is a difference, update the object pointer and start and end |
| 1080 | // addresses in the RegExp stack frame to match the new value. |
| 1081 | const byte* end_address = frame_entry<const byte* >(re_frame, kInputEnd); |
Steve Block | d0582a6 | 2009-12-15 09:54:21 +0000 | [diff] [blame^] | 1082 | int byte_length = static_cast<int>(end_address - start_address); |
Steve Block | a7e24c1 | 2009-10-30 11:49:00 +0000 | [diff] [blame] | 1083 | frame_entry<const String*>(re_frame, kInputString) = *subject; |
| 1084 | frame_entry<const byte*>(re_frame, kInputStart) = new_address; |
| 1085 | frame_entry<const byte*>(re_frame, kInputEnd) = new_address + byte_length; |
| 1086 | } |
| 1087 | |
| 1088 | return 0; |
| 1089 | } |
| 1090 | |
| 1091 | |
| 1092 | Operand RegExpMacroAssemblerX64::register_location(int register_index) { |
| 1093 | ASSERT(register_index < (1<<30)); |
| 1094 | if (num_registers_ <= register_index) { |
| 1095 | num_registers_ = register_index + 1; |
| 1096 | } |
| 1097 | return Operand(rbp, kRegisterZero - register_index * kPointerSize); |
| 1098 | } |
| 1099 | |
| 1100 | |
| 1101 | void RegExpMacroAssemblerX64::CheckPosition(int cp_offset, |
| 1102 | Label* on_outside_input) { |
| 1103 | __ cmpl(rdi, Immediate(-cp_offset * char_size())); |
| 1104 | BranchOrBacktrack(greater_equal, on_outside_input); |
| 1105 | } |
| 1106 | |
| 1107 | |
| 1108 | void RegExpMacroAssemblerX64::BranchOrBacktrack(Condition condition, |
| 1109 | Label* to) { |
| 1110 | if (condition < 0) { // No condition |
| 1111 | if (to == NULL) { |
| 1112 | Backtrack(); |
| 1113 | return; |
| 1114 | } |
| 1115 | __ jmp(to); |
| 1116 | return; |
| 1117 | } |
| 1118 | if (to == NULL) { |
| 1119 | __ j(condition, &backtrack_label_); |
| 1120 | return; |
| 1121 | } |
| 1122 | __ j(condition, to); |
| 1123 | } |
| 1124 | |
| 1125 | |
| 1126 | void RegExpMacroAssemblerX64::SafeCall(Label* to) { |
| 1127 | __ call(to); |
| 1128 | } |
| 1129 | |
| 1130 | |
| 1131 | void RegExpMacroAssemblerX64::SafeCallTarget(Label* label) { |
| 1132 | __ bind(label); |
| 1133 | __ subq(Operand(rsp, 0), code_object_pointer()); |
| 1134 | } |
| 1135 | |
| 1136 | |
| 1137 | void RegExpMacroAssemblerX64::SafeReturn() { |
| 1138 | __ addq(Operand(rsp, 0), code_object_pointer()); |
| 1139 | __ ret(0); |
| 1140 | } |
| 1141 | |
| 1142 | |
| 1143 | void RegExpMacroAssemblerX64::Push(Register source) { |
| 1144 | ASSERT(!source.is(backtrack_stackpointer())); |
| 1145 | // Notice: This updates flags, unlike normal Push. |
| 1146 | __ subq(backtrack_stackpointer(), Immediate(kIntSize)); |
| 1147 | __ movl(Operand(backtrack_stackpointer(), 0), source); |
| 1148 | } |
| 1149 | |
| 1150 | |
| 1151 | void RegExpMacroAssemblerX64::Push(Immediate value) { |
| 1152 | // Notice: This updates flags, unlike normal Push. |
| 1153 | __ subq(backtrack_stackpointer(), Immediate(kIntSize)); |
| 1154 | __ movl(Operand(backtrack_stackpointer(), 0), value); |
| 1155 | } |
| 1156 | |
| 1157 | |
| 1158 | void RegExpMacroAssemblerX64::FixupCodeRelativePositions() { |
| 1159 | for (int i = 0, n = code_relative_fixup_positions_.length(); i < n; i++) { |
| 1160 | int position = code_relative_fixup_positions_[i]; |
| 1161 | // The position succeeds a relative label offset from position. |
| 1162 | // Patch the relative offset to be relative to the Code object pointer |
| 1163 | // instead. |
| 1164 | int patch_position = position - kIntSize; |
| 1165 | int offset = masm_->long_at(patch_position); |
| 1166 | masm_->long_at_put(patch_position, |
| 1167 | offset |
| 1168 | + position |
| 1169 | + Code::kHeaderSize |
| 1170 | - kHeapObjectTag); |
| 1171 | } |
| 1172 | code_relative_fixup_positions_.Clear(); |
| 1173 | } |
| 1174 | |
| 1175 | |
| 1176 | void RegExpMacroAssemblerX64::Push(Label* backtrack_target) { |
| 1177 | __ subq(backtrack_stackpointer(), Immediate(kIntSize)); |
| 1178 | __ movl(Operand(backtrack_stackpointer(), 0), backtrack_target); |
| 1179 | MarkPositionForCodeRelativeFixup(); |
| 1180 | } |
| 1181 | |
| 1182 | |
| 1183 | void RegExpMacroAssemblerX64::Pop(Register target) { |
| 1184 | ASSERT(!target.is(backtrack_stackpointer())); |
| 1185 | __ movsxlq(target, Operand(backtrack_stackpointer(), 0)); |
| 1186 | // Notice: This updates flags, unlike normal Pop. |
| 1187 | __ addq(backtrack_stackpointer(), Immediate(kIntSize)); |
| 1188 | } |
| 1189 | |
| 1190 | |
| 1191 | void RegExpMacroAssemblerX64::Drop() { |
| 1192 | __ addq(backtrack_stackpointer(), Immediate(kIntSize)); |
| 1193 | } |
| 1194 | |
| 1195 | |
| 1196 | void RegExpMacroAssemblerX64::CheckPreemption() { |
| 1197 | // Check for preemption. |
| 1198 | Label no_preempt; |
Steve Block | d0582a6 | 2009-12-15 09:54:21 +0000 | [diff] [blame^] | 1199 | ExternalReference stack_limit = |
| 1200 | ExternalReference::address_of_stack_limit(); |
| 1201 | __ load_rax(stack_limit); |
Steve Block | a7e24c1 | 2009-10-30 11:49:00 +0000 | [diff] [blame] | 1202 | __ cmpq(rsp, rax); |
| 1203 | __ j(above, &no_preempt); |
| 1204 | |
| 1205 | SafeCall(&check_preempt_label_); |
| 1206 | |
| 1207 | __ bind(&no_preempt); |
| 1208 | } |
| 1209 | |
| 1210 | |
| 1211 | void RegExpMacroAssemblerX64::CheckStackLimit() { |
Steve Block | d0582a6 | 2009-12-15 09:54:21 +0000 | [diff] [blame^] | 1212 | Label no_stack_overflow; |
| 1213 | ExternalReference stack_limit = |
| 1214 | ExternalReference::address_of_regexp_stack_limit(); |
| 1215 | __ load_rax(stack_limit); |
| 1216 | __ cmpq(backtrack_stackpointer(), rax); |
| 1217 | __ j(above, &no_stack_overflow); |
Steve Block | a7e24c1 | 2009-10-30 11:49:00 +0000 | [diff] [blame] | 1218 | |
Steve Block | d0582a6 | 2009-12-15 09:54:21 +0000 | [diff] [blame^] | 1219 | SafeCall(&stack_overflow_label_); |
Steve Block | a7e24c1 | 2009-10-30 11:49:00 +0000 | [diff] [blame] | 1220 | |
Steve Block | d0582a6 | 2009-12-15 09:54:21 +0000 | [diff] [blame^] | 1221 | __ bind(&no_stack_overflow); |
Steve Block | a7e24c1 | 2009-10-30 11:49:00 +0000 | [diff] [blame] | 1222 | } |
| 1223 | |
| 1224 | |
| 1225 | void RegExpMacroAssemblerX64::FrameAlign(int num_arguments) { |
| 1226 | // TODO(lrn): Since we no longer use the system stack arbitrarily (but we do |
| 1227 | // use it, e.g., for SafeCall), we know the number of elements on the stack |
| 1228 | // since the last frame alignment. We might be able to do this simpler then. |
| 1229 | int frameAlignment = OS::ActivationFrameAlignment(); |
| 1230 | ASSERT(frameAlignment != 0); |
| 1231 | // Make stack end at alignment and make room for num_arguments pointers |
| 1232 | // (on Win64 only) and the original value of rsp. |
| 1233 | __ movq(kScratchRegister, rsp); |
| 1234 | ASSERT(IsPowerOf2(frameAlignment)); |
| 1235 | #ifdef _WIN64 |
| 1236 | // Allocate space for parameters and old rsp. |
| 1237 | __ subq(rsp, Immediate((num_arguments + 1) * kPointerSize)); |
| 1238 | __ and_(rsp, Immediate(-frameAlignment)); |
| 1239 | __ movq(Operand(rsp, num_arguments * kPointerSize), kScratchRegister); |
| 1240 | #else |
| 1241 | // Allocate space for old rsp. |
| 1242 | __ subq(rsp, Immediate(kPointerSize)); |
| 1243 | __ and_(rsp, Immediate(-frameAlignment)); |
| 1244 | __ movq(Operand(rsp, 0), kScratchRegister); |
| 1245 | #endif |
| 1246 | } |
| 1247 | |
| 1248 | |
| 1249 | void RegExpMacroAssemblerX64::CallCFunction(ExternalReference function, |
| 1250 | int num_arguments) { |
| 1251 | __ movq(rax, function); |
| 1252 | __ call(rax); |
| 1253 | ASSERT(OS::ActivationFrameAlignment() != 0); |
| 1254 | #ifdef _WIN64 |
| 1255 | __ movq(rsp, Operand(rsp, num_arguments * kPointerSize)); |
| 1256 | #else |
| 1257 | // All arguments passed in registers. |
| 1258 | ASSERT(num_arguments <= 6); |
| 1259 | __ pop(rsp); |
| 1260 | #endif |
| 1261 | } |
| 1262 | |
| 1263 | |
| 1264 | void RegExpMacroAssemblerX64::LoadCurrentCharacterUnchecked(int cp_offset, |
| 1265 | int characters) { |
| 1266 | if (mode_ == ASCII) { |
| 1267 | if (characters == 4) { |
| 1268 | __ movl(current_character(), Operand(rsi, rdi, times_1, cp_offset)); |
| 1269 | } else if (characters == 2) { |
| 1270 | __ movzxwl(current_character(), Operand(rsi, rdi, times_1, cp_offset)); |
| 1271 | } else { |
| 1272 | ASSERT(characters == 1); |
| 1273 | __ movzxbl(current_character(), Operand(rsi, rdi, times_1, cp_offset)); |
| 1274 | } |
| 1275 | } else { |
| 1276 | ASSERT(mode_ == UC16); |
| 1277 | if (characters == 2) { |
| 1278 | __ movl(current_character(), |
| 1279 | Operand(rsi, rdi, times_1, cp_offset * sizeof(uc16))); |
| 1280 | } else { |
| 1281 | ASSERT(characters == 1); |
| 1282 | __ movzxwl(current_character(), |
| 1283 | Operand(rsi, rdi, times_1, cp_offset * sizeof(uc16))); |
| 1284 | } |
| 1285 | } |
| 1286 | } |
| 1287 | |
Steve Block | a7e24c1 | 2009-10-30 11:49:00 +0000 | [diff] [blame] | 1288 | #undef __ |
| 1289 | |
| 1290 | #endif // V8_NATIVE_REGEXP |
| 1291 | |
| 1292 | }} // namespace v8::internal |