Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 1 | // Copyright 2012 the V8 project authors. All rights reserved. |
| 2 | // Use of this source code is governed by a BSD-style license that can be |
| 3 | // found in the LICENSE file. |
| 4 | |
| 5 | #if V8_TARGET_ARCH_X64 |
| 6 | |
| 7 | #include "src/regexp/x64/regexp-macro-assembler-x64.h" |
| 8 | |
| 9 | #include "src/log.h" |
| 10 | #include "src/macro-assembler.h" |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 11 | #include "src/regexp/regexp-macro-assembler.h" |
| 12 | #include "src/regexp/regexp-stack.h" |
| 13 | #include "src/unicode.h" |
| 14 | |
| 15 | namespace v8 { |
| 16 | namespace internal { |
| 17 | |
| 18 | #ifndef V8_INTERPRETED_REGEXP |
| 19 | |
| 20 | /* |
| 21 | * This assembler uses the following register assignment convention |
| 22 | * - rdx : Currently loaded character(s) as Latin1 or UC16. Must be loaded |
| 23 | * using LoadCurrentCharacter before using any of the dispatch methods. |
| 24 | * Temporarily stores the index of capture start after a matching pass |
| 25 | * for a global regexp. |
| 26 | * - rdi : Current position in input, as negative offset from end of string. |
| 27 | * Please notice that this is the byte offset, not the character |
| 28 | * offset! Is always a 32-bit signed (negative) offset, but must be |
| 29 | * maintained sign-extended to 64 bits, since it is used as index. |
| 30 | * - rsi : End of input (points to byte after last character in input), |
| 31 | * so that rsi+rdi points to the current character. |
| 32 | * - rbp : Frame pointer. Used to access arguments, local variables and |
| 33 | * RegExp registers. |
| 34 | * - rsp : Points to tip of C stack. |
| 35 | * - rcx : Points to tip of backtrack stack. The backtrack stack contains |
| 36 | * only 32-bit values. Most are offsets from some base (e.g., character |
| 37 | * positions from end of string or code location from Code* pointer). |
| 38 | * - r8 : Code object pointer. Used to convert between absolute and |
| 39 | * code-object-relative addresses. |
| 40 | * |
| 41 | * The registers rax, rbx, r9 and r11 are free to use for computations. |
| 42 | * If changed to use r12+, they should be saved as callee-save registers. |
| 43 | * The macro assembler special register r13 (kRootRegister) isn't special |
| 44 | * during execution of RegExp code (it doesn't hold the value assumed when |
| 45 | * creating JS code), so Root related macro operations can be used. |
| 46 | * |
| 47 | * Each call to a C++ method should retain these registers. |
| 48 | * |
| 49 | * The stack will have the following content, in some order, indexable from the |
| 50 | * frame pointer (see, e.g., kStackHighEnd): |
| 51 | * - Isolate* isolate (address of the current isolate) |
| 52 | * - direct_call (if 1, direct call from JavaScript code, if 0 call |
| 53 | * through the runtime system) |
| 54 | * - stack_area_base (high end of the memory area to use as |
| 55 | * backtracking stack) |
| 56 | * - capture array size (may fit multiple sets of matches) |
| 57 | * - int* capture_array (int[num_saved_registers_], for output). |
| 58 | * - end of input (address of end of string) |
| 59 | * - start of input (address of first character in string) |
| 60 | * - start index (character index of start) |
| 61 | * - String* input_string (input string) |
| 62 | * - return address |
| 63 | * - backup of callee save registers (rbx, possibly rsi and rdi). |
| 64 | * - success counter (only useful for global regexp to count matches) |
| 65 | * - Offset of location before start of input (effectively character |
| 66 | * string start - 1). Used to initialize capture registers to a |
| 67 | * non-position. |
| 68 | * - At start of string (if 1, we are starting at the start of the |
| 69 | * string, otherwise 0) |
| 70 | * - register 0 rbp[-n] (Only positions must be stored in the first |
| 71 | * - register 1 rbp[-n-8] num_saved_registers_ registers) |
| 72 | * - ... |
| 73 | * |
| 74 | * The first num_saved_registers_ registers are initialized to point to |
| 75 | * "character -1" in the string (i.e., char_size() bytes before the first |
| 76 | * character of the string). The remaining registers starts out uninitialized. |
| 77 | * |
| 78 | * The first seven values must be provided by the calling code by |
| 79 | * calling the code's entry address cast to a function pointer with the |
| 80 | * following signature: |
| 81 | * int (*match)(String* input_string, |
| 82 | * int start_index, |
| 83 | * Address start, |
| 84 | * Address end, |
| 85 | * int* capture_output_array, |
| 86 | * bool at_start, |
| 87 | * byte* stack_area_base, |
| 88 | * bool direct_call) |
| 89 | */ |
| 90 | |
| 91 | #define __ ACCESS_MASM((&masm_)) |
| 92 | |
| 93 | RegExpMacroAssemblerX64::RegExpMacroAssemblerX64(Isolate* isolate, Zone* zone, |
| 94 | Mode mode, |
| 95 | int registers_to_save) |
| 96 | : NativeRegExpMacroAssembler(isolate, zone), |
| 97 | masm_(isolate, NULL, kRegExpCodeSize, CodeObjectRequired::kYes), |
| 98 | no_root_array_scope_(&masm_), |
| 99 | code_relative_fixup_positions_(4, zone), |
| 100 | mode_(mode), |
| 101 | num_registers_(registers_to_save), |
| 102 | num_saved_registers_(registers_to_save), |
| 103 | entry_label_(), |
| 104 | start_label_(), |
| 105 | success_label_(), |
| 106 | backtrack_label_(), |
| 107 | exit_label_() { |
| 108 | DCHECK_EQ(0, registers_to_save % 2); |
| 109 | __ jmp(&entry_label_); // We'll write the entry code when we know more. |
| 110 | __ bind(&start_label_); // And then continue from here. |
| 111 | } |
| 112 | |
| 113 | |
| 114 | RegExpMacroAssemblerX64::~RegExpMacroAssemblerX64() { |
| 115 | // Unuse labels in case we throw away the assembler without calling GetCode. |
| 116 | entry_label_.Unuse(); |
| 117 | start_label_.Unuse(); |
| 118 | success_label_.Unuse(); |
| 119 | backtrack_label_.Unuse(); |
| 120 | exit_label_.Unuse(); |
| 121 | check_preempt_label_.Unuse(); |
| 122 | stack_overflow_label_.Unuse(); |
| 123 | } |
| 124 | |
| 125 | |
| 126 | int RegExpMacroAssemblerX64::stack_limit_slack() { |
| 127 | return RegExpStack::kStackLimitSlack; |
| 128 | } |
| 129 | |
| 130 | |
| 131 | void RegExpMacroAssemblerX64::AdvanceCurrentPosition(int by) { |
| 132 | if (by != 0) { |
| 133 | __ addq(rdi, Immediate(by * char_size())); |
| 134 | } |
| 135 | } |
| 136 | |
| 137 | |
| 138 | void RegExpMacroAssemblerX64::AdvanceRegister(int reg, int by) { |
| 139 | DCHECK(reg >= 0); |
| 140 | DCHECK(reg < num_registers_); |
| 141 | if (by != 0) { |
| 142 | __ addp(register_location(reg), Immediate(by)); |
| 143 | } |
| 144 | } |
| 145 | |
| 146 | |
| 147 | void RegExpMacroAssemblerX64::Backtrack() { |
| 148 | CheckPreemption(); |
| 149 | // Pop Code* offset from backtrack stack, add Code* and jump to location. |
| 150 | Pop(rbx); |
| 151 | __ addp(rbx, code_object_pointer()); |
| 152 | __ jmp(rbx); |
| 153 | } |
| 154 | |
| 155 | |
| 156 | void RegExpMacroAssemblerX64::Bind(Label* label) { |
| 157 | __ bind(label); |
| 158 | } |
| 159 | |
| 160 | |
| 161 | void RegExpMacroAssemblerX64::CheckCharacter(uint32_t c, Label* on_equal) { |
| 162 | __ cmpl(current_character(), Immediate(c)); |
| 163 | BranchOrBacktrack(equal, on_equal); |
| 164 | } |
| 165 | |
| 166 | |
| 167 | void RegExpMacroAssemblerX64::CheckCharacterGT(uc16 limit, Label* on_greater) { |
| 168 | __ cmpl(current_character(), Immediate(limit)); |
| 169 | BranchOrBacktrack(greater, on_greater); |
| 170 | } |
| 171 | |
| 172 | |
| 173 | void RegExpMacroAssemblerX64::CheckAtStart(Label* on_at_start) { |
| 174 | __ leap(rax, Operand(rdi, -char_size())); |
| 175 | __ cmpp(rax, Operand(rbp, kStringStartMinusOne)); |
| 176 | BranchOrBacktrack(equal, on_at_start); |
| 177 | } |
| 178 | |
| 179 | |
| 180 | void RegExpMacroAssemblerX64::CheckNotAtStart(int cp_offset, |
| 181 | Label* on_not_at_start) { |
| 182 | __ leap(rax, Operand(rdi, -char_size() + cp_offset * char_size())); |
| 183 | __ cmpp(rax, Operand(rbp, kStringStartMinusOne)); |
| 184 | BranchOrBacktrack(not_equal, on_not_at_start); |
| 185 | } |
| 186 | |
| 187 | |
| 188 | void RegExpMacroAssemblerX64::CheckCharacterLT(uc16 limit, Label* on_less) { |
| 189 | __ cmpl(current_character(), Immediate(limit)); |
| 190 | BranchOrBacktrack(less, on_less); |
| 191 | } |
| 192 | |
| 193 | |
| 194 | void RegExpMacroAssemblerX64::CheckGreedyLoop(Label* on_equal) { |
| 195 | Label fallthrough; |
| 196 | __ cmpl(rdi, Operand(backtrack_stackpointer(), 0)); |
| 197 | __ j(not_equal, &fallthrough); |
| 198 | Drop(); |
| 199 | BranchOrBacktrack(no_condition, on_equal); |
| 200 | __ bind(&fallthrough); |
| 201 | } |
| 202 | |
| 203 | |
| 204 | void RegExpMacroAssemblerX64::CheckNotBackReferenceIgnoreCase( |
Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 205 | int start_reg, bool read_backward, bool unicode, Label* on_no_match) { |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 206 | Label fallthrough; |
| 207 | ReadPositionFromRegister(rdx, start_reg); // Offset of start of capture |
| 208 | ReadPositionFromRegister(rbx, start_reg + 1); // Offset of end of capture |
| 209 | __ subp(rbx, rdx); // Length of capture. |
| 210 | |
| 211 | // ----------------------- |
| 212 | // rdx = Start offset of capture. |
| 213 | // rbx = Length of capture |
| 214 | |
| 215 | // At this point, the capture registers are either both set or both cleared. |
| 216 | // If the capture length is zero, then the capture is either empty or cleared. |
| 217 | // Fall through in both cases. |
| 218 | __ j(equal, &fallthrough); |
| 219 | |
| 220 | // ----------------------- |
| 221 | // rdx - Start of capture |
| 222 | // rbx - length of capture |
| 223 | // Check that there are sufficient characters left in the input. |
| 224 | if (read_backward) { |
| 225 | __ movl(rax, Operand(rbp, kStringStartMinusOne)); |
| 226 | __ addl(rax, rbx); |
| 227 | __ cmpl(rdi, rax); |
| 228 | BranchOrBacktrack(less_equal, on_no_match); |
| 229 | } else { |
| 230 | __ movl(rax, rdi); |
| 231 | __ addl(rax, rbx); |
| 232 | BranchOrBacktrack(greater, on_no_match); |
| 233 | } |
| 234 | |
| 235 | if (mode_ == LATIN1) { |
| 236 | Label loop_increment; |
| 237 | if (on_no_match == NULL) { |
| 238 | on_no_match = &backtrack_label_; |
| 239 | } |
| 240 | |
| 241 | __ leap(r9, Operand(rsi, rdx, times_1, 0)); |
| 242 | __ leap(r11, Operand(rsi, rdi, times_1, 0)); |
| 243 | if (read_backward) { |
| 244 | __ subp(r11, rbx); // Offset by length when matching backwards. |
| 245 | } |
| 246 | __ addp(rbx, r9); // End of capture |
| 247 | // --------------------- |
| 248 | // r11 - current input character address |
| 249 | // r9 - current capture character address |
| 250 | // rbx - end of capture |
| 251 | |
| 252 | Label loop; |
| 253 | __ bind(&loop); |
| 254 | __ movzxbl(rdx, Operand(r9, 0)); |
| 255 | __ movzxbl(rax, Operand(r11, 0)); |
| 256 | // al - input character |
| 257 | // dl - capture character |
| 258 | __ cmpb(rax, rdx); |
| 259 | __ j(equal, &loop_increment); |
| 260 | |
| 261 | // Mismatch, try case-insensitive match (converting letters to lower-case). |
| 262 | // I.e., if or-ing with 0x20 makes values equal and in range 'a'-'z', it's |
| 263 | // a match. |
| 264 | __ orp(rax, Immediate(0x20)); // Convert match character to lower-case. |
| 265 | __ orp(rdx, Immediate(0x20)); // Convert capture character to lower-case. |
| 266 | __ cmpb(rax, rdx); |
| 267 | __ j(not_equal, on_no_match); // Definitely not equal. |
| 268 | __ subb(rax, Immediate('a')); |
| 269 | __ cmpb(rax, Immediate('z' - 'a')); |
| 270 | __ j(below_equal, &loop_increment); // In range 'a'-'z'. |
| 271 | // Latin-1: Check for values in range [224,254] but not 247. |
| 272 | __ subb(rax, Immediate(224 - 'a')); |
| 273 | __ cmpb(rax, Immediate(254 - 224)); |
| 274 | __ j(above, on_no_match); // Weren't Latin-1 letters. |
| 275 | __ cmpb(rax, Immediate(247 - 224)); // Check for 247. |
| 276 | __ j(equal, on_no_match); |
| 277 | __ bind(&loop_increment); |
| 278 | // Increment pointers into match and capture strings. |
| 279 | __ addp(r11, Immediate(1)); |
| 280 | __ addp(r9, Immediate(1)); |
| 281 | // Compare to end of capture, and loop if not done. |
| 282 | __ cmpp(r9, rbx); |
| 283 | __ j(below, &loop); |
| 284 | |
| 285 | // Compute new value of character position after the matched part. |
| 286 | __ movp(rdi, r11); |
| 287 | __ subq(rdi, rsi); |
| 288 | if (read_backward) { |
| 289 | // Subtract match length if we matched backward. |
| 290 | __ addq(rdi, register_location(start_reg)); |
| 291 | __ subq(rdi, register_location(start_reg + 1)); |
| 292 | } |
| 293 | } else { |
| 294 | DCHECK(mode_ == UC16); |
| 295 | // Save important/volatile registers before calling C function. |
| 296 | #ifndef _WIN64 |
| 297 | // Caller save on Linux and callee save in Windows. |
| 298 | __ pushq(rsi); |
| 299 | __ pushq(rdi); |
| 300 | #endif |
| 301 | __ pushq(backtrack_stackpointer()); |
| 302 | |
| 303 | static const int num_arguments = 4; |
| 304 | __ PrepareCallCFunction(num_arguments); |
| 305 | |
| 306 | // Put arguments into parameter registers. Parameters are |
| 307 | // Address byte_offset1 - Address captured substring's start. |
| 308 | // Address byte_offset2 - Address of current character position. |
| 309 | // size_t byte_length - length of capture in bytes(!) |
Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 310 | // Isolate* isolate or 0 if unicode flag. |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 311 | #ifdef _WIN64 |
Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 312 | DCHECK(rcx.is(arg_reg_1)); |
| 313 | DCHECK(rdx.is(arg_reg_2)); |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 314 | // Compute and set byte_offset1 (start of capture). |
| 315 | __ leap(rcx, Operand(rsi, rdx, times_1, 0)); |
| 316 | // Set byte_offset2. |
| 317 | __ leap(rdx, Operand(rsi, rdi, times_1, 0)); |
| 318 | if (read_backward) { |
| 319 | __ subq(rdx, rbx); |
| 320 | } |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 321 | #else // AMD64 calling convention |
Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 322 | DCHECK(rdi.is(arg_reg_1)); |
| 323 | DCHECK(rsi.is(arg_reg_2)); |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 324 | // Compute byte_offset2 (current position = rsi+rdi). |
| 325 | __ leap(rax, Operand(rsi, rdi, times_1, 0)); |
| 326 | // Compute and set byte_offset1 (start of capture). |
| 327 | __ leap(rdi, Operand(rsi, rdx, times_1, 0)); |
| 328 | // Set byte_offset2. |
| 329 | __ movp(rsi, rax); |
| 330 | if (read_backward) { |
| 331 | __ subq(rsi, rbx); |
| 332 | } |
Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 333 | #endif // _WIN64 |
| 334 | |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 335 | // Set byte_length. |
Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 336 | __ movp(arg_reg_3, rbx); |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 337 | // Isolate. |
Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 338 | #ifdef V8_I18N_SUPPORT |
| 339 | if (unicode) { |
| 340 | __ movp(arg_reg_4, Immediate(0)); |
| 341 | } else // NOLINT |
| 342 | #endif // V8_I18N_SUPPORT |
| 343 | { |
| 344 | __ LoadAddress(arg_reg_4, ExternalReference::isolate_address(isolate())); |
| 345 | } |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 346 | |
| 347 | { // NOLINT: Can't find a way to open this scope without confusing the |
| 348 | // linter. |
| 349 | AllowExternalCallThatCantCauseGC scope(&masm_); |
| 350 | ExternalReference compare = |
| 351 | ExternalReference::re_case_insensitive_compare_uc16(isolate()); |
| 352 | __ CallCFunction(compare, num_arguments); |
| 353 | } |
| 354 | |
| 355 | // Restore original values before reacting on result value. |
| 356 | __ Move(code_object_pointer(), masm_.CodeObject()); |
| 357 | __ popq(backtrack_stackpointer()); |
| 358 | #ifndef _WIN64 |
| 359 | __ popq(rdi); |
| 360 | __ popq(rsi); |
| 361 | #endif |
| 362 | |
| 363 | // Check if function returned non-zero for success or zero for failure. |
| 364 | __ testp(rax, rax); |
| 365 | BranchOrBacktrack(zero, on_no_match); |
| 366 | // On success, advance position by length of capture. |
| 367 | // Requires that rbx is callee save (true for both Win64 and AMD64 ABIs). |
| 368 | if (read_backward) { |
| 369 | __ subq(rdi, rbx); |
| 370 | } else { |
| 371 | __ addq(rdi, rbx); |
| 372 | } |
| 373 | } |
| 374 | __ bind(&fallthrough); |
| 375 | } |
| 376 | |
| 377 | |
| 378 | void RegExpMacroAssemblerX64::CheckNotBackReference(int start_reg, |
| 379 | bool read_backward, |
| 380 | Label* on_no_match) { |
| 381 | Label fallthrough; |
| 382 | |
| 383 | // Find length of back-referenced capture. |
| 384 | ReadPositionFromRegister(rdx, start_reg); // Offset of start of capture |
| 385 | ReadPositionFromRegister(rax, start_reg + 1); // Offset of end of capture |
| 386 | __ subp(rax, rdx); // Length to check. |
| 387 | |
| 388 | // At this point, the capture registers are either both set or both cleared. |
| 389 | // If the capture length is zero, then the capture is either empty or cleared. |
| 390 | // Fall through in both cases. |
| 391 | __ j(equal, &fallthrough); |
| 392 | |
| 393 | // ----------------------- |
| 394 | // rdx - Start of capture |
| 395 | // rax - length of capture |
| 396 | // Check that there are sufficient characters left in the input. |
| 397 | if (read_backward) { |
| 398 | __ movl(rbx, Operand(rbp, kStringStartMinusOne)); |
| 399 | __ addl(rbx, rax); |
| 400 | __ cmpl(rdi, rbx); |
| 401 | BranchOrBacktrack(less_equal, on_no_match); |
| 402 | } else { |
| 403 | __ movl(rbx, rdi); |
| 404 | __ addl(rbx, rax); |
| 405 | BranchOrBacktrack(greater, on_no_match); |
| 406 | } |
| 407 | |
| 408 | // Compute pointers to match string and capture string |
| 409 | __ leap(rbx, Operand(rsi, rdi, times_1, 0)); // Start of match. |
| 410 | if (read_backward) { |
| 411 | __ subq(rbx, rax); // Offset by length when matching backwards. |
| 412 | } |
| 413 | __ addp(rdx, rsi); // Start of capture. |
| 414 | __ leap(r9, Operand(rdx, rax, times_1, 0)); // End of capture |
| 415 | |
| 416 | // ----------------------- |
| 417 | // rbx - current capture character address. |
| 418 | // rbx - current input character address . |
| 419 | // r9 - end of input to match (capture length after rbx). |
| 420 | |
| 421 | Label loop; |
| 422 | __ bind(&loop); |
| 423 | if (mode_ == LATIN1) { |
| 424 | __ movzxbl(rax, Operand(rdx, 0)); |
| 425 | __ cmpb(rax, Operand(rbx, 0)); |
| 426 | } else { |
| 427 | DCHECK(mode_ == UC16); |
| 428 | __ movzxwl(rax, Operand(rdx, 0)); |
| 429 | __ cmpw(rax, Operand(rbx, 0)); |
| 430 | } |
| 431 | BranchOrBacktrack(not_equal, on_no_match); |
| 432 | // Increment pointers into capture and match string. |
| 433 | __ addp(rbx, Immediate(char_size())); |
| 434 | __ addp(rdx, Immediate(char_size())); |
| 435 | // Check if we have reached end of match area. |
| 436 | __ cmpp(rdx, r9); |
| 437 | __ j(below, &loop); |
| 438 | |
| 439 | // Success. |
| 440 | // Set current character position to position after match. |
| 441 | __ movp(rdi, rbx); |
| 442 | __ subq(rdi, rsi); |
| 443 | if (read_backward) { |
| 444 | // Subtract match length if we matched backward. |
| 445 | __ addq(rdi, register_location(start_reg)); |
| 446 | __ subq(rdi, register_location(start_reg + 1)); |
| 447 | } |
| 448 | |
| 449 | __ bind(&fallthrough); |
| 450 | } |
| 451 | |
| 452 | |
| 453 | void RegExpMacroAssemblerX64::CheckNotCharacter(uint32_t c, |
| 454 | Label* on_not_equal) { |
| 455 | __ cmpl(current_character(), Immediate(c)); |
| 456 | BranchOrBacktrack(not_equal, on_not_equal); |
| 457 | } |
| 458 | |
| 459 | |
| 460 | void RegExpMacroAssemblerX64::CheckCharacterAfterAnd(uint32_t c, |
| 461 | uint32_t mask, |
| 462 | Label* on_equal) { |
| 463 | if (c == 0) { |
| 464 | __ testl(current_character(), Immediate(mask)); |
| 465 | } else { |
| 466 | __ movl(rax, Immediate(mask)); |
| 467 | __ andp(rax, current_character()); |
| 468 | __ cmpl(rax, Immediate(c)); |
| 469 | } |
| 470 | BranchOrBacktrack(equal, on_equal); |
| 471 | } |
| 472 | |
| 473 | |
| 474 | void RegExpMacroAssemblerX64::CheckNotCharacterAfterAnd(uint32_t c, |
| 475 | uint32_t mask, |
| 476 | Label* on_not_equal) { |
| 477 | if (c == 0) { |
| 478 | __ testl(current_character(), Immediate(mask)); |
| 479 | } else { |
| 480 | __ movl(rax, Immediate(mask)); |
| 481 | __ andp(rax, current_character()); |
| 482 | __ cmpl(rax, Immediate(c)); |
| 483 | } |
| 484 | BranchOrBacktrack(not_equal, on_not_equal); |
| 485 | } |
| 486 | |
| 487 | |
| 488 | void RegExpMacroAssemblerX64::CheckNotCharacterAfterMinusAnd( |
| 489 | uc16 c, |
| 490 | uc16 minus, |
| 491 | uc16 mask, |
| 492 | Label* on_not_equal) { |
| 493 | DCHECK(minus < String::kMaxUtf16CodeUnit); |
| 494 | __ leap(rax, Operand(current_character(), -minus)); |
| 495 | __ andp(rax, Immediate(mask)); |
| 496 | __ cmpl(rax, Immediate(c)); |
| 497 | BranchOrBacktrack(not_equal, on_not_equal); |
| 498 | } |
| 499 | |
| 500 | |
| 501 | void RegExpMacroAssemblerX64::CheckCharacterInRange( |
| 502 | uc16 from, |
| 503 | uc16 to, |
| 504 | Label* on_in_range) { |
| 505 | __ leal(rax, Operand(current_character(), -from)); |
| 506 | __ cmpl(rax, Immediate(to - from)); |
| 507 | BranchOrBacktrack(below_equal, on_in_range); |
| 508 | } |
| 509 | |
| 510 | |
| 511 | void RegExpMacroAssemblerX64::CheckCharacterNotInRange( |
| 512 | uc16 from, |
| 513 | uc16 to, |
| 514 | Label* on_not_in_range) { |
| 515 | __ leal(rax, Operand(current_character(), -from)); |
| 516 | __ cmpl(rax, Immediate(to - from)); |
| 517 | BranchOrBacktrack(above, on_not_in_range); |
| 518 | } |
| 519 | |
| 520 | |
| 521 | void RegExpMacroAssemblerX64::CheckBitInTable( |
| 522 | Handle<ByteArray> table, |
| 523 | Label* on_bit_set) { |
| 524 | __ Move(rax, table); |
| 525 | Register index = current_character(); |
| 526 | if (mode_ != LATIN1 || kTableMask != String::kMaxOneByteCharCode) { |
| 527 | __ movp(rbx, current_character()); |
| 528 | __ andp(rbx, Immediate(kTableMask)); |
| 529 | index = rbx; |
| 530 | } |
| 531 | __ cmpb(FieldOperand(rax, index, times_1, ByteArray::kHeaderSize), |
| 532 | Immediate(0)); |
| 533 | BranchOrBacktrack(not_equal, on_bit_set); |
| 534 | } |
| 535 | |
| 536 | |
| 537 | bool RegExpMacroAssemblerX64::CheckSpecialCharacterClass(uc16 type, |
| 538 | Label* on_no_match) { |
| 539 | // Range checks (c in min..max) are generally implemented by an unsigned |
| 540 | // (c - min) <= (max - min) check, using the sequence: |
| 541 | // leap(rax, Operand(current_character(), -min)) or sub(rax, Immediate(min)) |
| 542 | // cmp(rax, Immediate(max - min)) |
| 543 | switch (type) { |
| 544 | case 's': |
| 545 | // Match space-characters |
| 546 | if (mode_ == LATIN1) { |
| 547 | // One byte space characters are '\t'..'\r', ' ' and \u00a0. |
| 548 | Label success; |
| 549 | __ cmpl(current_character(), Immediate(' ')); |
| 550 | __ j(equal, &success, Label::kNear); |
| 551 | // Check range 0x09..0x0d |
| 552 | __ leap(rax, Operand(current_character(), -'\t')); |
| 553 | __ cmpl(rax, Immediate('\r' - '\t')); |
| 554 | __ j(below_equal, &success, Label::kNear); |
| 555 | // \u00a0 (NBSP). |
| 556 | __ cmpl(rax, Immediate(0x00a0 - '\t')); |
| 557 | BranchOrBacktrack(not_equal, on_no_match); |
| 558 | __ bind(&success); |
| 559 | return true; |
| 560 | } |
| 561 | return false; |
| 562 | case 'S': |
| 563 | // The emitted code for generic character classes is good enough. |
| 564 | return false; |
| 565 | case 'd': |
| 566 | // Match ASCII digits ('0'..'9') |
| 567 | __ leap(rax, Operand(current_character(), -'0')); |
| 568 | __ cmpl(rax, Immediate('9' - '0')); |
| 569 | BranchOrBacktrack(above, on_no_match); |
| 570 | return true; |
| 571 | case 'D': |
| 572 | // Match non ASCII-digits |
| 573 | __ leap(rax, Operand(current_character(), -'0')); |
| 574 | __ cmpl(rax, Immediate('9' - '0')); |
| 575 | BranchOrBacktrack(below_equal, on_no_match); |
| 576 | return true; |
| 577 | case '.': { |
| 578 | // Match non-newlines (not 0x0a('\n'), 0x0d('\r'), 0x2028 and 0x2029) |
| 579 | __ movl(rax, current_character()); |
| 580 | __ xorp(rax, Immediate(0x01)); |
| 581 | // See if current character is '\n'^1 or '\r'^1, i.e., 0x0b or 0x0c |
| 582 | __ subl(rax, Immediate(0x0b)); |
| 583 | __ cmpl(rax, Immediate(0x0c - 0x0b)); |
| 584 | BranchOrBacktrack(below_equal, on_no_match); |
| 585 | if (mode_ == UC16) { |
| 586 | // Compare original value to 0x2028 and 0x2029, using the already |
| 587 | // computed (current_char ^ 0x01 - 0x0b). I.e., check for |
| 588 | // 0x201d (0x2028 - 0x0b) or 0x201e. |
| 589 | __ subl(rax, Immediate(0x2028 - 0x0b)); |
| 590 | __ cmpl(rax, Immediate(0x2029 - 0x2028)); |
| 591 | BranchOrBacktrack(below_equal, on_no_match); |
| 592 | } |
| 593 | return true; |
| 594 | } |
| 595 | case 'n': { |
| 596 | // Match newlines (0x0a('\n'), 0x0d('\r'), 0x2028 and 0x2029) |
| 597 | __ movl(rax, current_character()); |
| 598 | __ xorp(rax, Immediate(0x01)); |
| 599 | // See if current character is '\n'^1 or '\r'^1, i.e., 0x0b or 0x0c |
| 600 | __ subl(rax, Immediate(0x0b)); |
| 601 | __ cmpl(rax, Immediate(0x0c - 0x0b)); |
| 602 | if (mode_ == LATIN1) { |
| 603 | BranchOrBacktrack(above, on_no_match); |
| 604 | } else { |
| 605 | Label done; |
| 606 | BranchOrBacktrack(below_equal, &done); |
| 607 | // Compare original value to 0x2028 and 0x2029, using the already |
| 608 | // computed (current_char ^ 0x01 - 0x0b). I.e., check for |
| 609 | // 0x201d (0x2028 - 0x0b) or 0x201e. |
| 610 | __ subl(rax, Immediate(0x2028 - 0x0b)); |
| 611 | __ cmpl(rax, Immediate(0x2029 - 0x2028)); |
| 612 | BranchOrBacktrack(above, on_no_match); |
| 613 | __ bind(&done); |
| 614 | } |
| 615 | return true; |
| 616 | } |
| 617 | case 'w': { |
| 618 | if (mode_ != LATIN1) { |
| 619 | // Table is 256 entries, so all Latin1 characters can be tested. |
| 620 | __ cmpl(current_character(), Immediate('z')); |
| 621 | BranchOrBacktrack(above, on_no_match); |
| 622 | } |
| 623 | __ Move(rbx, ExternalReference::re_word_character_map()); |
| 624 | DCHECK_EQ(0, word_character_map[0]); // Character '\0' is not a word char. |
| 625 | __ testb(Operand(rbx, current_character(), times_1, 0), |
| 626 | current_character()); |
| 627 | BranchOrBacktrack(zero, on_no_match); |
| 628 | return true; |
| 629 | } |
| 630 | case 'W': { |
| 631 | Label done; |
| 632 | if (mode_ != LATIN1) { |
| 633 | // Table is 256 entries, so all Latin1 characters can be tested. |
| 634 | __ cmpl(current_character(), Immediate('z')); |
| 635 | __ j(above, &done); |
| 636 | } |
| 637 | __ Move(rbx, ExternalReference::re_word_character_map()); |
| 638 | DCHECK_EQ(0, word_character_map[0]); // Character '\0' is not a word char. |
| 639 | __ testb(Operand(rbx, current_character(), times_1, 0), |
| 640 | current_character()); |
| 641 | BranchOrBacktrack(not_zero, on_no_match); |
| 642 | if (mode_ != LATIN1) { |
| 643 | __ bind(&done); |
| 644 | } |
| 645 | return true; |
| 646 | } |
| 647 | |
| 648 | case '*': |
| 649 | // Match any character. |
| 650 | return true; |
| 651 | // No custom implementation (yet): s(UC16), S(UC16). |
| 652 | default: |
| 653 | return false; |
| 654 | } |
| 655 | } |
| 656 | |
| 657 | |
| 658 | void RegExpMacroAssemblerX64::Fail() { |
| 659 | STATIC_ASSERT(FAILURE == 0); // Return value for failure is zero. |
| 660 | if (!global()) { |
| 661 | __ Set(rax, FAILURE); |
| 662 | } |
| 663 | __ jmp(&exit_label_); |
| 664 | } |
| 665 | |
| 666 | |
| 667 | Handle<HeapObject> RegExpMacroAssemblerX64::GetCode(Handle<String> source) { |
| 668 | Label return_rax; |
| 669 | // Finalize code - write the entry point code now we know how many |
| 670 | // registers we need. |
| 671 | // Entry code: |
| 672 | __ bind(&entry_label_); |
| 673 | |
| 674 | // Tell the system that we have a stack frame. Because the type is MANUAL, no |
| 675 | // is generated. |
| 676 | FrameScope scope(&masm_, StackFrame::MANUAL); |
| 677 | |
| 678 | // Actually emit code to start a new stack frame. |
| 679 | __ pushq(rbp); |
| 680 | __ movp(rbp, rsp); |
| 681 | // Save parameters and callee-save registers. Order here should correspond |
| 682 | // to order of kBackup_ebx etc. |
| 683 | #ifdef _WIN64 |
| 684 | // MSVC passes arguments in rcx, rdx, r8, r9, with backing stack slots. |
| 685 | // Store register parameters in pre-allocated stack slots, |
| 686 | __ movq(Operand(rbp, kInputString), rcx); |
| 687 | __ movq(Operand(rbp, kStartIndex), rdx); // Passed as int32 in edx. |
| 688 | __ movq(Operand(rbp, kInputStart), r8); |
| 689 | __ movq(Operand(rbp, kInputEnd), r9); |
| 690 | // Callee-save on Win64. |
| 691 | __ pushq(rsi); |
| 692 | __ pushq(rdi); |
| 693 | __ pushq(rbx); |
| 694 | #else |
| 695 | // GCC passes arguments in rdi, rsi, rdx, rcx, r8, r9 (and then on stack). |
| 696 | // Push register parameters on stack for reference. |
| 697 | DCHECK_EQ(kInputString, -1 * kRegisterSize); |
| 698 | DCHECK_EQ(kStartIndex, -2 * kRegisterSize); |
| 699 | DCHECK_EQ(kInputStart, -3 * kRegisterSize); |
| 700 | DCHECK_EQ(kInputEnd, -4 * kRegisterSize); |
| 701 | DCHECK_EQ(kRegisterOutput, -5 * kRegisterSize); |
| 702 | DCHECK_EQ(kNumOutputRegisters, -6 * kRegisterSize); |
| 703 | __ pushq(rdi); |
| 704 | __ pushq(rsi); |
| 705 | __ pushq(rdx); |
| 706 | __ pushq(rcx); |
| 707 | __ pushq(r8); |
| 708 | __ pushq(r9); |
| 709 | |
| 710 | __ pushq(rbx); // Callee-save |
| 711 | #endif |
| 712 | |
| 713 | __ Push(Immediate(0)); // Number of successful matches in a global regexp. |
| 714 | __ Push(Immediate(0)); // Make room for "string start - 1" constant. |
| 715 | |
| 716 | // Check if we have space on the stack for registers. |
| 717 | Label stack_limit_hit; |
| 718 | Label stack_ok; |
| 719 | |
| 720 | ExternalReference stack_limit = |
| 721 | ExternalReference::address_of_stack_limit(isolate()); |
| 722 | __ movp(rcx, rsp); |
| 723 | __ Move(kScratchRegister, stack_limit); |
| 724 | __ subp(rcx, Operand(kScratchRegister, 0)); |
| 725 | // Handle it if the stack pointer is already below the stack limit. |
| 726 | __ j(below_equal, &stack_limit_hit); |
| 727 | // Check if there is room for the variable number of registers above |
| 728 | // the stack limit. |
| 729 | __ cmpp(rcx, Immediate(num_registers_ * kPointerSize)); |
| 730 | __ j(above_equal, &stack_ok); |
| 731 | // Exit with OutOfMemory exception. There is not enough space on the stack |
| 732 | // for our working registers. |
| 733 | __ Set(rax, EXCEPTION); |
| 734 | __ jmp(&return_rax); |
| 735 | |
| 736 | __ bind(&stack_limit_hit); |
| 737 | __ Move(code_object_pointer(), masm_.CodeObject()); |
| 738 | CallCheckStackGuardState(); // Preserves no registers beside rbp and rsp. |
| 739 | __ testp(rax, rax); |
| 740 | // If returned value is non-zero, we exit with the returned value as result. |
| 741 | __ j(not_zero, &return_rax); |
| 742 | |
| 743 | __ bind(&stack_ok); |
| 744 | |
| 745 | // Allocate space on stack for registers. |
| 746 | __ subp(rsp, Immediate(num_registers_ * kPointerSize)); |
| 747 | // Load string length. |
| 748 | __ movp(rsi, Operand(rbp, kInputEnd)); |
| 749 | // Load input position. |
| 750 | __ movp(rdi, Operand(rbp, kInputStart)); |
| 751 | // Set up rdi to be negative offset from string end. |
| 752 | __ subq(rdi, rsi); |
| 753 | // Set rax to address of char before start of the string |
| 754 | // (effectively string position -1). |
| 755 | __ movp(rbx, Operand(rbp, kStartIndex)); |
| 756 | __ negq(rbx); |
| 757 | if (mode_ == UC16) { |
| 758 | __ leap(rax, Operand(rdi, rbx, times_2, -char_size())); |
| 759 | } else { |
| 760 | __ leap(rax, Operand(rdi, rbx, times_1, -char_size())); |
| 761 | } |
| 762 | // Store this value in a local variable, for use when clearing |
| 763 | // position registers. |
| 764 | __ movp(Operand(rbp, kStringStartMinusOne), rax); |
| 765 | |
| 766 | #if V8_OS_WIN |
| 767 | // Ensure that we have written to each stack page, in order. Skipping a page |
| 768 | // on Windows can cause segmentation faults. Assuming page size is 4k. |
| 769 | const int kPageSize = 4096; |
| 770 | const int kRegistersPerPage = kPageSize / kPointerSize; |
| 771 | for (int i = num_saved_registers_ + kRegistersPerPage - 1; |
| 772 | i < num_registers_; |
| 773 | i += kRegistersPerPage) { |
| 774 | __ movp(register_location(i), rax); // One write every page. |
| 775 | } |
| 776 | #endif // V8_OS_WIN |
| 777 | |
| 778 | // Initialize code object pointer. |
| 779 | __ Move(code_object_pointer(), masm_.CodeObject()); |
| 780 | |
| 781 | Label load_char_start_regexp, start_regexp; |
| 782 | // Load newline if index is at start, previous character otherwise. |
| 783 | __ cmpl(Operand(rbp, kStartIndex), Immediate(0)); |
| 784 | __ j(not_equal, &load_char_start_regexp, Label::kNear); |
| 785 | __ Set(current_character(), '\n'); |
| 786 | __ jmp(&start_regexp, Label::kNear); |
| 787 | |
| 788 | // Global regexp restarts matching here. |
| 789 | __ bind(&load_char_start_regexp); |
| 790 | // Load previous char as initial value of current character register. |
| 791 | LoadCurrentCharacterUnchecked(-1, 1); |
| 792 | __ bind(&start_regexp); |
| 793 | |
| 794 | // Initialize on-stack registers. |
| 795 | if (num_saved_registers_ > 0) { |
| 796 | // Fill saved registers with initial value = start offset - 1 |
| 797 | // Fill in stack push order, to avoid accessing across an unwritten |
| 798 | // page (a problem on Windows). |
| 799 | if (num_saved_registers_ > 8) { |
| 800 | __ Set(rcx, kRegisterZero); |
| 801 | Label init_loop; |
| 802 | __ bind(&init_loop); |
| 803 | __ movp(Operand(rbp, rcx, times_1, 0), rax); |
| 804 | __ subq(rcx, Immediate(kPointerSize)); |
| 805 | __ cmpq(rcx, |
| 806 | Immediate(kRegisterZero - num_saved_registers_ * kPointerSize)); |
| 807 | __ j(greater, &init_loop); |
| 808 | } else { // Unroll the loop. |
| 809 | for (int i = 0; i < num_saved_registers_; i++) { |
| 810 | __ movp(register_location(i), rax); |
| 811 | } |
| 812 | } |
| 813 | } |
| 814 | |
| 815 | // Initialize backtrack stack pointer. |
| 816 | __ movp(backtrack_stackpointer(), Operand(rbp, kStackHighEnd)); |
| 817 | |
| 818 | __ jmp(&start_label_); |
| 819 | |
| 820 | // Exit code: |
| 821 | if (success_label_.is_linked()) { |
| 822 | // Save captures when successful. |
| 823 | __ bind(&success_label_); |
| 824 | if (num_saved_registers_ > 0) { |
| 825 | // copy captures to output |
| 826 | __ movp(rdx, Operand(rbp, kStartIndex)); |
| 827 | __ movp(rbx, Operand(rbp, kRegisterOutput)); |
| 828 | __ movp(rcx, Operand(rbp, kInputEnd)); |
| 829 | __ subp(rcx, Operand(rbp, kInputStart)); |
| 830 | if (mode_ == UC16) { |
| 831 | __ leap(rcx, Operand(rcx, rdx, times_2, 0)); |
| 832 | } else { |
| 833 | __ addp(rcx, rdx); |
| 834 | } |
| 835 | for (int i = 0; i < num_saved_registers_; i++) { |
| 836 | __ movp(rax, register_location(i)); |
| 837 | if (i == 0 && global_with_zero_length_check()) { |
| 838 | // Keep capture start in rdx for the zero-length check later. |
| 839 | __ movp(rdx, rax); |
| 840 | } |
| 841 | __ addp(rax, rcx); // Convert to index from start, not end. |
| 842 | if (mode_ == UC16) { |
| 843 | __ sarp(rax, Immediate(1)); // Convert byte index to character index. |
| 844 | } |
| 845 | __ movl(Operand(rbx, i * kIntSize), rax); |
| 846 | } |
| 847 | } |
| 848 | |
| 849 | if (global()) { |
| 850 | // Restart matching if the regular expression is flagged as global. |
| 851 | // Increment success counter. |
| 852 | __ incp(Operand(rbp, kSuccessfulCaptures)); |
| 853 | // Capture results have been stored, so the number of remaining global |
| 854 | // output registers is reduced by the number of stored captures. |
| 855 | __ movsxlq(rcx, Operand(rbp, kNumOutputRegisters)); |
| 856 | __ subp(rcx, Immediate(num_saved_registers_)); |
| 857 | // Check whether we have enough room for another set of capture results. |
| 858 | __ cmpp(rcx, Immediate(num_saved_registers_)); |
| 859 | __ j(less, &exit_label_); |
| 860 | |
| 861 | __ movp(Operand(rbp, kNumOutputRegisters), rcx); |
| 862 | // Advance the location for output. |
| 863 | __ addp(Operand(rbp, kRegisterOutput), |
| 864 | Immediate(num_saved_registers_ * kIntSize)); |
| 865 | |
| 866 | // Prepare rax to initialize registers with its value in the next run. |
| 867 | __ movp(rax, Operand(rbp, kStringStartMinusOne)); |
| 868 | |
| 869 | if (global_with_zero_length_check()) { |
| 870 | // Special case for zero-length matches. |
| 871 | // rdx: capture start index |
| 872 | __ cmpp(rdi, rdx); |
| 873 | // Not a zero-length match, restart. |
| 874 | __ j(not_equal, &load_char_start_regexp); |
| 875 | // rdi (offset from the end) is zero if we already reached the end. |
| 876 | __ testp(rdi, rdi); |
| 877 | __ j(zero, &exit_label_, Label::kNear); |
| 878 | // Advance current position after a zero-length match. |
Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 879 | Label advance; |
| 880 | __ bind(&advance); |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 881 | if (mode_ == UC16) { |
| 882 | __ addq(rdi, Immediate(2)); |
| 883 | } else { |
| 884 | __ incq(rdi); |
| 885 | } |
Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 886 | if (global_unicode()) CheckNotInSurrogatePair(0, &advance); |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 887 | } |
| 888 | |
| 889 | __ jmp(&load_char_start_regexp); |
| 890 | } else { |
| 891 | __ movp(rax, Immediate(SUCCESS)); |
| 892 | } |
| 893 | } |
| 894 | |
| 895 | __ bind(&exit_label_); |
| 896 | if (global()) { |
| 897 | // Return the number of successful captures. |
| 898 | __ movp(rax, Operand(rbp, kSuccessfulCaptures)); |
| 899 | } |
| 900 | |
| 901 | __ bind(&return_rax); |
| 902 | #ifdef _WIN64 |
| 903 | // Restore callee save registers. |
| 904 | __ leap(rsp, Operand(rbp, kLastCalleeSaveRegister)); |
| 905 | __ popq(rbx); |
| 906 | __ popq(rdi); |
| 907 | __ popq(rsi); |
| 908 | // Stack now at rbp. |
| 909 | #else |
| 910 | // Restore callee save register. |
| 911 | __ movp(rbx, Operand(rbp, kBackup_rbx)); |
| 912 | // Skip rsp to rbp. |
| 913 | __ movp(rsp, rbp); |
| 914 | #endif |
| 915 | // Exit function frame, restore previous one. |
| 916 | __ popq(rbp); |
| 917 | __ ret(0); |
| 918 | |
| 919 | // Backtrack code (branch target for conditional backtracks). |
| 920 | if (backtrack_label_.is_linked()) { |
| 921 | __ bind(&backtrack_label_); |
| 922 | Backtrack(); |
| 923 | } |
| 924 | |
| 925 | Label exit_with_exception; |
| 926 | |
| 927 | // Preempt-code |
| 928 | if (check_preempt_label_.is_linked()) { |
| 929 | SafeCallTarget(&check_preempt_label_); |
| 930 | |
| 931 | __ pushq(backtrack_stackpointer()); |
| 932 | __ pushq(rdi); |
| 933 | |
| 934 | CallCheckStackGuardState(); |
| 935 | __ testp(rax, rax); |
| 936 | // If returning non-zero, we should end execution with the given |
| 937 | // result as return value. |
| 938 | __ j(not_zero, &return_rax); |
| 939 | |
| 940 | // Restore registers. |
| 941 | __ Move(code_object_pointer(), masm_.CodeObject()); |
| 942 | __ popq(rdi); |
| 943 | __ popq(backtrack_stackpointer()); |
| 944 | // String might have moved: Reload esi from frame. |
| 945 | __ movp(rsi, Operand(rbp, kInputEnd)); |
| 946 | SafeReturn(); |
| 947 | } |
| 948 | |
| 949 | // Backtrack stack overflow code. |
| 950 | if (stack_overflow_label_.is_linked()) { |
| 951 | SafeCallTarget(&stack_overflow_label_); |
| 952 | // Reached if the backtrack-stack limit has been hit. |
| 953 | |
| 954 | Label grow_failed; |
| 955 | // Save registers before calling C function |
| 956 | #ifndef _WIN64 |
| 957 | // Callee-save in Microsoft 64-bit ABI, but not in AMD64 ABI. |
| 958 | __ pushq(rsi); |
| 959 | __ pushq(rdi); |
| 960 | #endif |
| 961 | |
| 962 | // Call GrowStack(backtrack_stackpointer()) |
| 963 | static const int num_arguments = 3; |
| 964 | __ PrepareCallCFunction(num_arguments); |
| 965 | #ifdef _WIN64 |
| 966 | // Microsoft passes parameters in rcx, rdx, r8. |
| 967 | // First argument, backtrack stackpointer, is already in rcx. |
| 968 | __ leap(rdx, Operand(rbp, kStackHighEnd)); // Second argument |
| 969 | __ LoadAddress(r8, ExternalReference::isolate_address(isolate())); |
| 970 | #else |
| 971 | // AMD64 ABI passes parameters in rdi, rsi, rdx. |
| 972 | __ movp(rdi, backtrack_stackpointer()); // First argument. |
| 973 | __ leap(rsi, Operand(rbp, kStackHighEnd)); // Second argument. |
| 974 | __ LoadAddress(rdx, ExternalReference::isolate_address(isolate())); |
| 975 | #endif |
| 976 | ExternalReference grow_stack = |
| 977 | ExternalReference::re_grow_stack(isolate()); |
| 978 | __ CallCFunction(grow_stack, num_arguments); |
| 979 | // If return NULL, we have failed to grow the stack, and |
| 980 | // must exit with a stack-overflow exception. |
| 981 | __ testp(rax, rax); |
| 982 | __ j(equal, &exit_with_exception); |
| 983 | // Otherwise use return value as new stack pointer. |
| 984 | __ movp(backtrack_stackpointer(), rax); |
| 985 | // Restore saved registers and continue. |
| 986 | __ Move(code_object_pointer(), masm_.CodeObject()); |
| 987 | #ifndef _WIN64 |
| 988 | __ popq(rdi); |
| 989 | __ popq(rsi); |
| 990 | #endif |
| 991 | SafeReturn(); |
| 992 | } |
| 993 | |
| 994 | if (exit_with_exception.is_linked()) { |
| 995 | // If any of the code above needed to exit with an exception. |
| 996 | __ bind(&exit_with_exception); |
| 997 | // Exit with Result EXCEPTION(-1) to signal thrown exception. |
| 998 | __ Set(rax, EXCEPTION); |
| 999 | __ jmp(&return_rax); |
| 1000 | } |
| 1001 | |
| 1002 | FixupCodeRelativePositions(); |
| 1003 | |
| 1004 | CodeDesc code_desc; |
| 1005 | masm_.GetCode(&code_desc); |
| 1006 | Isolate* isolate = this->isolate(); |
| 1007 | Handle<Code> code = isolate->factory()->NewCode( |
| 1008 | code_desc, Code::ComputeFlags(Code::REGEXP), |
| 1009 | masm_.CodeObject()); |
Ben Murdoch | da12d29 | 2016-06-02 14:46:10 +0100 | [diff] [blame] | 1010 | PROFILE(isolate, RegExpCodeCreateEvent(AbstractCode::cast(*code), *source)); |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 1011 | return Handle<HeapObject>::cast(code); |
| 1012 | } |
| 1013 | |
| 1014 | |
| 1015 | void RegExpMacroAssemblerX64::GoTo(Label* to) { |
| 1016 | BranchOrBacktrack(no_condition, to); |
| 1017 | } |
| 1018 | |
| 1019 | |
| 1020 | void RegExpMacroAssemblerX64::IfRegisterGE(int reg, |
| 1021 | int comparand, |
| 1022 | Label* if_ge) { |
| 1023 | __ cmpp(register_location(reg), Immediate(comparand)); |
| 1024 | BranchOrBacktrack(greater_equal, if_ge); |
| 1025 | } |
| 1026 | |
| 1027 | |
| 1028 | void RegExpMacroAssemblerX64::IfRegisterLT(int reg, |
| 1029 | int comparand, |
| 1030 | Label* if_lt) { |
| 1031 | __ cmpp(register_location(reg), Immediate(comparand)); |
| 1032 | BranchOrBacktrack(less, if_lt); |
| 1033 | } |
| 1034 | |
| 1035 | |
| 1036 | void RegExpMacroAssemblerX64::IfRegisterEqPos(int reg, |
| 1037 | Label* if_eq) { |
| 1038 | __ cmpp(rdi, register_location(reg)); |
| 1039 | BranchOrBacktrack(equal, if_eq); |
| 1040 | } |
| 1041 | |
| 1042 | |
| 1043 | RegExpMacroAssembler::IrregexpImplementation |
| 1044 | RegExpMacroAssemblerX64::Implementation() { |
| 1045 | return kX64Implementation; |
| 1046 | } |
| 1047 | |
| 1048 | |
| 1049 | void RegExpMacroAssemblerX64::LoadCurrentCharacter(int cp_offset, |
| 1050 | Label* on_end_of_input, |
| 1051 | bool check_bounds, |
| 1052 | int characters) { |
| 1053 | DCHECK(cp_offset < (1<<30)); // Be sane! (And ensure negation works) |
| 1054 | if (check_bounds) { |
| 1055 | if (cp_offset >= 0) { |
| 1056 | CheckPosition(cp_offset + characters - 1, on_end_of_input); |
| 1057 | } else { |
| 1058 | CheckPosition(cp_offset, on_end_of_input); |
| 1059 | } |
| 1060 | } |
| 1061 | LoadCurrentCharacterUnchecked(cp_offset, characters); |
| 1062 | } |
| 1063 | |
| 1064 | |
| 1065 | void RegExpMacroAssemblerX64::PopCurrentPosition() { |
| 1066 | Pop(rdi); |
| 1067 | } |
| 1068 | |
| 1069 | |
| 1070 | void RegExpMacroAssemblerX64::PopRegister(int register_index) { |
| 1071 | Pop(rax); |
| 1072 | __ movp(register_location(register_index), rax); |
| 1073 | } |
| 1074 | |
| 1075 | |
| 1076 | void RegExpMacroAssemblerX64::PushBacktrack(Label* label) { |
| 1077 | Push(label); |
| 1078 | CheckStackLimit(); |
| 1079 | } |
| 1080 | |
| 1081 | |
| 1082 | void RegExpMacroAssemblerX64::PushCurrentPosition() { |
| 1083 | Push(rdi); |
| 1084 | } |
| 1085 | |
| 1086 | |
| 1087 | void RegExpMacroAssemblerX64::PushRegister(int register_index, |
| 1088 | StackCheckFlag check_stack_limit) { |
| 1089 | __ movp(rax, register_location(register_index)); |
| 1090 | Push(rax); |
| 1091 | if (check_stack_limit) CheckStackLimit(); |
| 1092 | } |
| 1093 | |
| 1094 | |
| 1095 | STATIC_ASSERT(kPointerSize == kInt64Size || kPointerSize == kInt32Size); |
| 1096 | |
| 1097 | |
| 1098 | void RegExpMacroAssemblerX64::ReadCurrentPositionFromRegister(int reg) { |
| 1099 | if (kPointerSize == kInt64Size) { |
| 1100 | __ movq(rdi, register_location(reg)); |
| 1101 | } else { |
| 1102 | // Need sign extension for x32 as rdi might be used as an index register. |
| 1103 | __ movsxlq(rdi, register_location(reg)); |
| 1104 | } |
| 1105 | } |
| 1106 | |
| 1107 | |
| 1108 | void RegExpMacroAssemblerX64::ReadPositionFromRegister(Register dst, int reg) { |
| 1109 | if (kPointerSize == kInt64Size) { |
| 1110 | __ movq(dst, register_location(reg)); |
| 1111 | } else { |
| 1112 | // Need sign extension for x32 as dst might be used as an index register. |
| 1113 | __ movsxlq(dst, register_location(reg)); |
| 1114 | } |
| 1115 | } |
| 1116 | |
| 1117 | |
| 1118 | void RegExpMacroAssemblerX64::ReadStackPointerFromRegister(int reg) { |
| 1119 | __ movp(backtrack_stackpointer(), register_location(reg)); |
| 1120 | __ addp(backtrack_stackpointer(), Operand(rbp, kStackHighEnd)); |
| 1121 | } |
| 1122 | |
| 1123 | |
| 1124 | void RegExpMacroAssemblerX64::SetCurrentPositionFromEnd(int by) { |
| 1125 | Label after_position; |
| 1126 | __ cmpp(rdi, Immediate(-by * char_size())); |
| 1127 | __ j(greater_equal, &after_position, Label::kNear); |
| 1128 | __ movq(rdi, Immediate(-by * char_size())); |
| 1129 | // On RegExp code entry (where this operation is used), the character before |
| 1130 | // the current position is expected to be already loaded. |
| 1131 | // We have advanced the position, so it's safe to read backwards. |
| 1132 | LoadCurrentCharacterUnchecked(-1, 1); |
| 1133 | __ bind(&after_position); |
| 1134 | } |
| 1135 | |
| 1136 | |
| 1137 | void RegExpMacroAssemblerX64::SetRegister(int register_index, int to) { |
| 1138 | DCHECK(register_index >= num_saved_registers_); // Reserved for positions! |
| 1139 | __ movp(register_location(register_index), Immediate(to)); |
| 1140 | } |
| 1141 | |
| 1142 | |
| 1143 | bool RegExpMacroAssemblerX64::Succeed() { |
| 1144 | __ jmp(&success_label_); |
| 1145 | return global(); |
| 1146 | } |
| 1147 | |
| 1148 | |
| 1149 | void RegExpMacroAssemblerX64::WriteCurrentPositionToRegister(int reg, |
| 1150 | int cp_offset) { |
| 1151 | if (cp_offset == 0) { |
| 1152 | __ movp(register_location(reg), rdi); |
| 1153 | } else { |
| 1154 | __ leap(rax, Operand(rdi, cp_offset * char_size())); |
| 1155 | __ movp(register_location(reg), rax); |
| 1156 | } |
| 1157 | } |
| 1158 | |
| 1159 | |
| 1160 | void RegExpMacroAssemblerX64::ClearRegisters(int reg_from, int reg_to) { |
| 1161 | DCHECK(reg_from <= reg_to); |
| 1162 | __ movp(rax, Operand(rbp, kStringStartMinusOne)); |
| 1163 | for (int reg = reg_from; reg <= reg_to; reg++) { |
| 1164 | __ movp(register_location(reg), rax); |
| 1165 | } |
| 1166 | } |
| 1167 | |
| 1168 | |
| 1169 | void RegExpMacroAssemblerX64::WriteStackPointerToRegister(int reg) { |
| 1170 | __ movp(rax, backtrack_stackpointer()); |
| 1171 | __ subp(rax, Operand(rbp, kStackHighEnd)); |
| 1172 | __ movp(register_location(reg), rax); |
| 1173 | } |
| 1174 | |
| 1175 | |
| 1176 | // Private methods: |
| 1177 | |
| 1178 | void RegExpMacroAssemblerX64::CallCheckStackGuardState() { |
| 1179 | // This function call preserves no register values. Caller should |
| 1180 | // store anything volatile in a C call or overwritten by this function. |
| 1181 | static const int num_arguments = 3; |
| 1182 | __ PrepareCallCFunction(num_arguments); |
| 1183 | #ifdef _WIN64 |
| 1184 | // Second argument: Code* of self. (Do this before overwriting r8). |
| 1185 | __ movp(rdx, code_object_pointer()); |
| 1186 | // Third argument: RegExp code frame pointer. |
| 1187 | __ movp(r8, rbp); |
| 1188 | // First argument: Next address on the stack (will be address of |
| 1189 | // return address). |
| 1190 | __ leap(rcx, Operand(rsp, -kPointerSize)); |
| 1191 | #else |
| 1192 | // Third argument: RegExp code frame pointer. |
| 1193 | __ movp(rdx, rbp); |
| 1194 | // Second argument: Code* of self. |
| 1195 | __ movp(rsi, code_object_pointer()); |
| 1196 | // First argument: Next address on the stack (will be address of |
| 1197 | // return address). |
| 1198 | __ leap(rdi, Operand(rsp, -kRegisterSize)); |
| 1199 | #endif |
| 1200 | ExternalReference stack_check = |
| 1201 | ExternalReference::re_check_stack_guard_state(isolate()); |
| 1202 | __ CallCFunction(stack_check, num_arguments); |
| 1203 | } |
| 1204 | |
| 1205 | |
| 1206 | // Helper function for reading a value out of a stack frame. |
| 1207 | template <typename T> |
| 1208 | static T& frame_entry(Address re_frame, int frame_offset) { |
| 1209 | return reinterpret_cast<T&>(Memory::int32_at(re_frame + frame_offset)); |
| 1210 | } |
| 1211 | |
| 1212 | |
| 1213 | template <typename T> |
| 1214 | static T* frame_entry_address(Address re_frame, int frame_offset) { |
| 1215 | return reinterpret_cast<T*>(re_frame + frame_offset); |
| 1216 | } |
| 1217 | |
| 1218 | |
| 1219 | int RegExpMacroAssemblerX64::CheckStackGuardState(Address* return_address, |
| 1220 | Code* re_code, |
| 1221 | Address re_frame) { |
| 1222 | return NativeRegExpMacroAssembler::CheckStackGuardState( |
| 1223 | frame_entry<Isolate*>(re_frame, kIsolate), |
| 1224 | frame_entry<int>(re_frame, kStartIndex), |
| 1225 | frame_entry<int>(re_frame, kDirectCall) == 1, return_address, re_code, |
| 1226 | frame_entry_address<String*>(re_frame, kInputString), |
| 1227 | frame_entry_address<const byte*>(re_frame, kInputStart), |
| 1228 | frame_entry_address<const byte*>(re_frame, kInputEnd)); |
| 1229 | } |
| 1230 | |
| 1231 | |
| 1232 | Operand RegExpMacroAssemblerX64::register_location(int register_index) { |
| 1233 | DCHECK(register_index < (1<<30)); |
| 1234 | if (num_registers_ <= register_index) { |
| 1235 | num_registers_ = register_index + 1; |
| 1236 | } |
| 1237 | return Operand(rbp, kRegisterZero - register_index * kPointerSize); |
| 1238 | } |
| 1239 | |
| 1240 | |
| 1241 | void RegExpMacroAssemblerX64::CheckPosition(int cp_offset, |
| 1242 | Label* on_outside_input) { |
| 1243 | if (cp_offset >= 0) { |
| 1244 | __ cmpl(rdi, Immediate(-cp_offset * char_size())); |
| 1245 | BranchOrBacktrack(greater_equal, on_outside_input); |
| 1246 | } else { |
| 1247 | __ leap(rax, Operand(rdi, cp_offset * char_size())); |
| 1248 | __ cmpp(rax, Operand(rbp, kStringStartMinusOne)); |
| 1249 | BranchOrBacktrack(less_equal, on_outside_input); |
| 1250 | } |
| 1251 | } |
| 1252 | |
| 1253 | |
| 1254 | void RegExpMacroAssemblerX64::BranchOrBacktrack(Condition condition, |
| 1255 | Label* to) { |
| 1256 | if (condition < 0) { // No condition |
| 1257 | if (to == NULL) { |
| 1258 | Backtrack(); |
| 1259 | return; |
| 1260 | } |
| 1261 | __ jmp(to); |
| 1262 | return; |
| 1263 | } |
| 1264 | if (to == NULL) { |
| 1265 | __ j(condition, &backtrack_label_); |
| 1266 | return; |
| 1267 | } |
| 1268 | __ j(condition, to); |
| 1269 | } |
| 1270 | |
| 1271 | |
| 1272 | void RegExpMacroAssemblerX64::SafeCall(Label* to) { |
| 1273 | __ call(to); |
| 1274 | } |
| 1275 | |
| 1276 | |
| 1277 | void RegExpMacroAssemblerX64::SafeCallTarget(Label* label) { |
| 1278 | __ bind(label); |
| 1279 | __ subp(Operand(rsp, 0), code_object_pointer()); |
| 1280 | } |
| 1281 | |
| 1282 | |
| 1283 | void RegExpMacroAssemblerX64::SafeReturn() { |
| 1284 | __ addp(Operand(rsp, 0), code_object_pointer()); |
| 1285 | __ ret(0); |
| 1286 | } |
| 1287 | |
| 1288 | |
| 1289 | void RegExpMacroAssemblerX64::Push(Register source) { |
| 1290 | DCHECK(!source.is(backtrack_stackpointer())); |
| 1291 | // Notice: This updates flags, unlike normal Push. |
| 1292 | __ subp(backtrack_stackpointer(), Immediate(kIntSize)); |
| 1293 | __ movl(Operand(backtrack_stackpointer(), 0), source); |
| 1294 | } |
| 1295 | |
| 1296 | |
| 1297 | void RegExpMacroAssemblerX64::Push(Immediate value) { |
| 1298 | // Notice: This updates flags, unlike normal Push. |
| 1299 | __ subp(backtrack_stackpointer(), Immediate(kIntSize)); |
| 1300 | __ movl(Operand(backtrack_stackpointer(), 0), value); |
| 1301 | } |
| 1302 | |
| 1303 | |
| 1304 | void RegExpMacroAssemblerX64::FixupCodeRelativePositions() { |
| 1305 | for (int i = 0, n = code_relative_fixup_positions_.length(); i < n; i++) { |
| 1306 | int position = code_relative_fixup_positions_[i]; |
| 1307 | // The position succeeds a relative label offset from position. |
| 1308 | // Patch the relative offset to be relative to the Code object pointer |
| 1309 | // instead. |
| 1310 | int patch_position = position - kIntSize; |
| 1311 | int offset = masm_.long_at(patch_position); |
| 1312 | masm_.long_at_put(patch_position, |
| 1313 | offset |
| 1314 | + position |
| 1315 | + Code::kHeaderSize |
| 1316 | - kHeapObjectTag); |
| 1317 | } |
| 1318 | code_relative_fixup_positions_.Clear(); |
| 1319 | } |
| 1320 | |
| 1321 | |
| 1322 | void RegExpMacroAssemblerX64::Push(Label* backtrack_target) { |
| 1323 | __ subp(backtrack_stackpointer(), Immediate(kIntSize)); |
| 1324 | __ movl(Operand(backtrack_stackpointer(), 0), backtrack_target); |
| 1325 | MarkPositionForCodeRelativeFixup(); |
| 1326 | } |
| 1327 | |
| 1328 | |
| 1329 | void RegExpMacroAssemblerX64::Pop(Register target) { |
| 1330 | DCHECK(!target.is(backtrack_stackpointer())); |
| 1331 | __ movsxlq(target, Operand(backtrack_stackpointer(), 0)); |
| 1332 | // Notice: This updates flags, unlike normal Pop. |
| 1333 | __ addp(backtrack_stackpointer(), Immediate(kIntSize)); |
| 1334 | } |
| 1335 | |
| 1336 | |
| 1337 | void RegExpMacroAssemblerX64::Drop() { |
| 1338 | __ addp(backtrack_stackpointer(), Immediate(kIntSize)); |
| 1339 | } |
| 1340 | |
| 1341 | |
| 1342 | void RegExpMacroAssemblerX64::CheckPreemption() { |
| 1343 | // Check for preemption. |
| 1344 | Label no_preempt; |
| 1345 | ExternalReference stack_limit = |
| 1346 | ExternalReference::address_of_stack_limit(isolate()); |
| 1347 | __ load_rax(stack_limit); |
| 1348 | __ cmpp(rsp, rax); |
| 1349 | __ j(above, &no_preempt); |
| 1350 | |
| 1351 | SafeCall(&check_preempt_label_); |
| 1352 | |
| 1353 | __ bind(&no_preempt); |
| 1354 | } |
| 1355 | |
| 1356 | |
| 1357 | void RegExpMacroAssemblerX64::CheckStackLimit() { |
| 1358 | Label no_stack_overflow; |
| 1359 | ExternalReference stack_limit = |
| 1360 | ExternalReference::address_of_regexp_stack_limit(isolate()); |
| 1361 | __ load_rax(stack_limit); |
| 1362 | __ cmpp(backtrack_stackpointer(), rax); |
| 1363 | __ j(above, &no_stack_overflow); |
| 1364 | |
| 1365 | SafeCall(&stack_overflow_label_); |
| 1366 | |
| 1367 | __ bind(&no_stack_overflow); |
| 1368 | } |
| 1369 | |
| 1370 | |
| 1371 | void RegExpMacroAssemblerX64::LoadCurrentCharacterUnchecked(int cp_offset, |
| 1372 | int characters) { |
| 1373 | if (mode_ == LATIN1) { |
| 1374 | if (characters == 4) { |
| 1375 | __ movl(current_character(), Operand(rsi, rdi, times_1, cp_offset)); |
| 1376 | } else if (characters == 2) { |
| 1377 | __ movzxwl(current_character(), Operand(rsi, rdi, times_1, cp_offset)); |
| 1378 | } else { |
| 1379 | DCHECK(characters == 1); |
| 1380 | __ movzxbl(current_character(), Operand(rsi, rdi, times_1, cp_offset)); |
| 1381 | } |
| 1382 | } else { |
| 1383 | DCHECK(mode_ == UC16); |
| 1384 | if (characters == 2) { |
| 1385 | __ movl(current_character(), |
| 1386 | Operand(rsi, rdi, times_1, cp_offset * sizeof(uc16))); |
| 1387 | } else { |
| 1388 | DCHECK(characters == 1); |
| 1389 | __ movzxwl(current_character(), |
| 1390 | Operand(rsi, rdi, times_1, cp_offset * sizeof(uc16))); |
| 1391 | } |
| 1392 | } |
| 1393 | } |
| 1394 | |
| 1395 | #undef __ |
| 1396 | |
| 1397 | #endif // V8_INTERPRETED_REGEXP |
| 1398 | |
| 1399 | } // namespace internal |
| 1400 | } // namespace v8 |
| 1401 | |
| 1402 | #endif // V8_TARGET_ARCH_X64 |