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_ARM |
| 6 | |
| 7 | #include "src/regexp/arm/regexp-macro-assembler-arm.h" |
| 8 | |
| 9 | #include "src/code-stubs.h" |
| 10 | #include "src/log.h" |
| 11 | #include "src/macro-assembler.h" |
| 12 | #include "src/profiler/cpu-profiler.h" |
| 13 | #include "src/regexp/regexp-macro-assembler.h" |
| 14 | #include "src/regexp/regexp-stack.h" |
| 15 | #include "src/unicode.h" |
| 16 | |
| 17 | namespace v8 { |
| 18 | namespace internal { |
| 19 | |
| 20 | #ifndef V8_INTERPRETED_REGEXP |
| 21 | /* |
| 22 | * This assembler uses the following register assignment convention |
| 23 | * - r4 : Temporarily stores the index of capture start after a matching pass |
| 24 | * for a global regexp. |
| 25 | * - r5 : Pointer to current code object (Code*) including heap object tag. |
| 26 | * - r6 : Current position in input, as negative offset from end of string. |
| 27 | * Please notice that this is the byte offset, not the character offset! |
| 28 | * - r7 : Currently loaded character. Must be loaded using |
| 29 | * LoadCurrentCharacter before using any of the dispatch methods. |
| 30 | * - r8 : Points to tip of backtrack stack |
| 31 | * - r9 : Unused, might be used by C code and expected unchanged. |
| 32 | * - r10 : End of input (points to byte after last character in input). |
| 33 | * - r11 : Frame pointer. Used to access arguments, local variables and |
| 34 | * RegExp registers. |
| 35 | * - r12 : IP register, used by assembler. Very volatile. |
| 36 | * - r13/sp : Points to tip of C stack. |
| 37 | * |
| 38 | * The remaining registers are free for computations. |
| 39 | * Each call to a public method should retain this convention. |
| 40 | * |
| 41 | * The stack will have the following structure: |
| 42 | * - fp[56] Isolate* isolate (address of the current isolate) |
| 43 | * - fp[52] direct_call (if 1, direct call from JavaScript code, |
| 44 | * if 0, call through the runtime system). |
| 45 | * - fp[48] stack_area_base (high end of the memory area to use as |
| 46 | * backtracking stack). |
| 47 | * - fp[44] capture array size (may fit multiple sets of matches) |
| 48 | * - fp[40] int* capture_array (int[num_saved_registers_], for output). |
| 49 | * - fp[36] secondary link/return address used by native call. |
| 50 | * --- sp when called --- |
| 51 | * - fp[32] return address (lr). |
| 52 | * - fp[28] old frame pointer (r11). |
| 53 | * - fp[0..24] backup of registers r4..r10. |
| 54 | * --- frame pointer ---- |
| 55 | * - fp[-4] end of input (address of end of string). |
| 56 | * - fp[-8] start of input (address of first character in string). |
| 57 | * - fp[-12] start index (character index of start). |
| 58 | * - fp[-16] void* input_string (location of a handle containing the string). |
| 59 | * - fp[-20] success counter (only for global regexps to count matches). |
| 60 | * - fp[-24] Offset of location before start of input (effectively character |
| 61 | * string start - 1). Used to initialize capture registers to a |
| 62 | * non-position. |
| 63 | * - fp[-28] At start (if 1, we are starting at the start of the |
| 64 | * string, otherwise 0) |
| 65 | * - fp[-32] register 0 (Only positions must be stored in the first |
| 66 | * - register 1 num_saved_registers_ registers) |
| 67 | * - ... |
| 68 | * - register num_registers-1 |
| 69 | * --- sp --- |
| 70 | * |
| 71 | * The first num_saved_registers_ registers are initialized to point to |
| 72 | * "character -1" in the string (i.e., char_size() bytes before the first |
| 73 | * character of the string). The remaining registers start out as garbage. |
| 74 | * |
| 75 | * The data up to the return address must be placed there by the calling |
| 76 | * code and the remaining arguments are passed in registers, e.g. by calling the |
| 77 | * code entry as cast to a function with the signature: |
| 78 | * int (*match)(String* input_string, |
| 79 | * int start_index, |
| 80 | * Address start, |
| 81 | * Address end, |
| 82 | * Address secondary_return_address, // Only used by native call. |
| 83 | * int* capture_output_array, |
| 84 | * byte* stack_area_base, |
| 85 | * bool direct_call = false) |
| 86 | * The call is performed by NativeRegExpMacroAssembler::Execute() |
| 87 | * (in regexp-macro-assembler.cc) via the CALL_GENERATED_REGEXP_CODE macro |
| 88 | * in arm/simulator-arm.h. |
| 89 | * When calling as a non-direct call (i.e., from C++ code), the return address |
| 90 | * area is overwritten with the LR register by the RegExp code. When doing a |
| 91 | * direct call from generated code, the return address is placed there by |
| 92 | * the calling code, as in a normal exit frame. |
| 93 | */ |
| 94 | |
| 95 | #define __ ACCESS_MASM(masm_) |
| 96 | |
| 97 | RegExpMacroAssemblerARM::RegExpMacroAssemblerARM(Isolate* isolate, Zone* zone, |
| 98 | Mode mode, |
| 99 | int registers_to_save) |
| 100 | : NativeRegExpMacroAssembler(isolate, zone), |
| 101 | masm_(new MacroAssembler(isolate, NULL, kRegExpCodeSize, |
| 102 | CodeObjectRequired::kYes)), |
| 103 | mode_(mode), |
| 104 | num_registers_(registers_to_save), |
| 105 | num_saved_registers_(registers_to_save), |
| 106 | entry_label_(), |
| 107 | start_label_(), |
| 108 | success_label_(), |
| 109 | backtrack_label_(), |
| 110 | exit_label_() { |
| 111 | DCHECK_EQ(0, registers_to_save % 2); |
| 112 | __ jmp(&entry_label_); // We'll write the entry code later. |
| 113 | __ bind(&start_label_); // And then continue from here. |
| 114 | } |
| 115 | |
| 116 | |
| 117 | RegExpMacroAssemblerARM::~RegExpMacroAssemblerARM() { |
| 118 | delete masm_; |
| 119 | // Unuse labels in case we throw away the assembler without calling GetCode. |
| 120 | entry_label_.Unuse(); |
| 121 | start_label_.Unuse(); |
| 122 | success_label_.Unuse(); |
| 123 | backtrack_label_.Unuse(); |
| 124 | exit_label_.Unuse(); |
| 125 | check_preempt_label_.Unuse(); |
| 126 | stack_overflow_label_.Unuse(); |
| 127 | } |
| 128 | |
| 129 | |
| 130 | int RegExpMacroAssemblerARM::stack_limit_slack() { |
| 131 | return RegExpStack::kStackLimitSlack; |
| 132 | } |
| 133 | |
| 134 | |
| 135 | void RegExpMacroAssemblerARM::AdvanceCurrentPosition(int by) { |
| 136 | if (by != 0) { |
| 137 | __ add(current_input_offset(), |
| 138 | current_input_offset(), Operand(by * char_size())); |
| 139 | } |
| 140 | } |
| 141 | |
| 142 | |
| 143 | void RegExpMacroAssemblerARM::AdvanceRegister(int reg, int by) { |
| 144 | DCHECK(reg >= 0); |
| 145 | DCHECK(reg < num_registers_); |
| 146 | if (by != 0) { |
| 147 | __ ldr(r0, register_location(reg)); |
| 148 | __ add(r0, r0, Operand(by)); |
| 149 | __ str(r0, register_location(reg)); |
| 150 | } |
| 151 | } |
| 152 | |
| 153 | |
| 154 | void RegExpMacroAssemblerARM::Backtrack() { |
| 155 | CheckPreemption(); |
| 156 | // Pop Code* offset from backtrack stack, add Code* and jump to location. |
| 157 | Pop(r0); |
| 158 | __ add(pc, r0, Operand(code_pointer())); |
| 159 | } |
| 160 | |
| 161 | |
| 162 | void RegExpMacroAssemblerARM::Bind(Label* label) { |
| 163 | __ bind(label); |
| 164 | } |
| 165 | |
| 166 | |
| 167 | void RegExpMacroAssemblerARM::CheckCharacter(uint32_t c, Label* on_equal) { |
| 168 | __ cmp(current_character(), Operand(c)); |
| 169 | BranchOrBacktrack(eq, on_equal); |
| 170 | } |
| 171 | |
| 172 | |
| 173 | void RegExpMacroAssemblerARM::CheckCharacterGT(uc16 limit, Label* on_greater) { |
| 174 | __ cmp(current_character(), Operand(limit)); |
| 175 | BranchOrBacktrack(gt, on_greater); |
| 176 | } |
| 177 | |
| 178 | |
| 179 | void RegExpMacroAssemblerARM::CheckAtStart(Label* on_at_start) { |
| 180 | __ ldr(r1, MemOperand(frame_pointer(), kStringStartMinusOne)); |
| 181 | __ add(r0, current_input_offset(), Operand(-char_size())); |
| 182 | __ cmp(r0, r1); |
| 183 | BranchOrBacktrack(eq, on_at_start); |
| 184 | } |
| 185 | |
| 186 | |
| 187 | void RegExpMacroAssemblerARM::CheckNotAtStart(int cp_offset, |
| 188 | Label* on_not_at_start) { |
| 189 | __ ldr(r1, MemOperand(frame_pointer(), kStringStartMinusOne)); |
| 190 | __ add(r0, current_input_offset(), |
| 191 | Operand(-char_size() + cp_offset * char_size())); |
| 192 | __ cmp(r0, r1); |
| 193 | BranchOrBacktrack(ne, on_not_at_start); |
| 194 | } |
| 195 | |
| 196 | |
| 197 | void RegExpMacroAssemblerARM::CheckCharacterLT(uc16 limit, Label* on_less) { |
| 198 | __ cmp(current_character(), Operand(limit)); |
| 199 | BranchOrBacktrack(lt, on_less); |
| 200 | } |
| 201 | |
| 202 | |
| 203 | void RegExpMacroAssemblerARM::CheckGreedyLoop(Label* on_equal) { |
| 204 | __ ldr(r0, MemOperand(backtrack_stackpointer(), 0)); |
| 205 | __ cmp(current_input_offset(), r0); |
| 206 | __ add(backtrack_stackpointer(), |
| 207 | backtrack_stackpointer(), Operand(kPointerSize), LeaveCC, eq); |
| 208 | BranchOrBacktrack(eq, on_equal); |
| 209 | } |
| 210 | |
| 211 | |
| 212 | void RegExpMacroAssemblerARM::CheckNotBackReferenceIgnoreCase( |
| 213 | int start_reg, bool read_backward, Label* on_no_match) { |
| 214 | Label fallthrough; |
| 215 | __ ldr(r0, register_location(start_reg)); // Index of start of capture |
| 216 | __ ldr(r1, register_location(start_reg + 1)); // Index of end of capture |
| 217 | __ sub(r1, r1, r0, SetCC); // Length of capture. |
| 218 | |
| 219 | // At this point, the capture registers are either both set or both cleared. |
| 220 | // If the capture length is zero, then the capture is either empty or cleared. |
| 221 | // Fall through in both cases. |
| 222 | __ b(eq, &fallthrough); |
| 223 | |
| 224 | // Check that there are enough characters left in the input. |
| 225 | if (read_backward) { |
| 226 | __ ldr(r3, MemOperand(frame_pointer(), kStringStartMinusOne)); |
| 227 | __ add(r3, r3, r1); |
| 228 | __ cmp(current_input_offset(), r3); |
| 229 | BranchOrBacktrack(le, on_no_match); |
| 230 | } else { |
| 231 | __ cmn(r1, Operand(current_input_offset())); |
| 232 | BranchOrBacktrack(gt, on_no_match); |
| 233 | } |
| 234 | |
| 235 | if (mode_ == LATIN1) { |
| 236 | Label success; |
| 237 | Label fail; |
| 238 | Label loop_check; |
| 239 | |
| 240 | // r0 - offset of start of capture |
| 241 | // r1 - length of capture |
| 242 | __ add(r0, r0, end_of_input_address()); |
| 243 | __ add(r2, end_of_input_address(), current_input_offset()); |
| 244 | if (read_backward) { |
| 245 | __ sub(r2, r2, r1); // Offset by length when matching backwards. |
| 246 | } |
| 247 | __ add(r1, r0, r1); |
| 248 | |
| 249 | // r0 - Address of start of capture. |
| 250 | // r1 - Address of end of capture |
| 251 | // r2 - Address of current input position. |
| 252 | |
| 253 | Label loop; |
| 254 | __ bind(&loop); |
| 255 | __ ldrb(r3, MemOperand(r0, char_size(), PostIndex)); |
| 256 | __ ldrb(r4, MemOperand(r2, char_size(), PostIndex)); |
| 257 | __ cmp(r4, r3); |
| 258 | __ b(eq, &loop_check); |
| 259 | |
| 260 | // Mismatch, try case-insensitive match (converting letters to lower-case). |
| 261 | __ orr(r3, r3, Operand(0x20)); // Convert capture character to lower-case. |
| 262 | __ orr(r4, r4, Operand(0x20)); // Also convert input character. |
| 263 | __ cmp(r4, r3); |
| 264 | __ b(ne, &fail); |
| 265 | __ sub(r3, r3, Operand('a')); |
| 266 | __ cmp(r3, Operand('z' - 'a')); // Is r3 a lowercase letter? |
| 267 | __ b(ls, &loop_check); // In range 'a'-'z'. |
| 268 | // Latin-1: Check for values in range [224,254] but not 247. |
| 269 | __ sub(r3, r3, Operand(224 - 'a')); |
| 270 | __ cmp(r3, Operand(254 - 224)); |
| 271 | __ b(hi, &fail); // Weren't Latin-1 letters. |
| 272 | __ cmp(r3, Operand(247 - 224)); // Check for 247. |
| 273 | __ b(eq, &fail); |
| 274 | |
| 275 | __ bind(&loop_check); |
| 276 | __ cmp(r0, r1); |
| 277 | __ b(lt, &loop); |
| 278 | __ jmp(&success); |
| 279 | |
| 280 | __ bind(&fail); |
| 281 | BranchOrBacktrack(al, on_no_match); |
| 282 | |
| 283 | __ bind(&success); |
| 284 | // Compute new value of character position after the matched part. |
| 285 | __ sub(current_input_offset(), r2, end_of_input_address()); |
| 286 | if (read_backward) { |
| 287 | __ ldr(r0, register_location(start_reg)); // Index of start of capture |
| 288 | __ ldr(r1, register_location(start_reg + 1)); // Index of end of capture |
| 289 | __ add(current_input_offset(), current_input_offset(), r0); |
| 290 | __ sub(current_input_offset(), current_input_offset(), r1); |
| 291 | } |
| 292 | } else { |
| 293 | DCHECK(mode_ == UC16); |
| 294 | int argument_count = 4; |
| 295 | __ PrepareCallCFunction(argument_count, r2); |
| 296 | |
| 297 | // r0 - offset of start of capture |
| 298 | // r1 - length of capture |
| 299 | |
| 300 | // Put arguments into arguments registers. |
| 301 | // Parameters are |
| 302 | // r0: Address byte_offset1 - Address captured substring's start. |
| 303 | // r1: Address byte_offset2 - Address of current character position. |
| 304 | // r2: size_t byte_length - length of capture in bytes(!) |
| 305 | // r3: Isolate* isolate |
| 306 | |
| 307 | // Address of start of capture. |
| 308 | __ add(r0, r0, Operand(end_of_input_address())); |
| 309 | // Length of capture. |
| 310 | __ mov(r2, Operand(r1)); |
| 311 | // Save length in callee-save register for use on return. |
| 312 | __ mov(r4, Operand(r1)); |
| 313 | // Address of current input position. |
| 314 | __ add(r1, current_input_offset(), end_of_input_address()); |
| 315 | if (read_backward) { |
| 316 | __ sub(r1, r1, r4); |
| 317 | } |
| 318 | // Isolate. |
| 319 | __ mov(r3, Operand(ExternalReference::isolate_address(isolate()))); |
| 320 | |
| 321 | { |
| 322 | AllowExternalCallThatCantCauseGC scope(masm_); |
| 323 | ExternalReference function = |
| 324 | ExternalReference::re_case_insensitive_compare_uc16(isolate()); |
| 325 | __ CallCFunction(function, argument_count); |
| 326 | } |
| 327 | |
| 328 | // Check if function returned non-zero for success or zero for failure. |
| 329 | __ cmp(r0, Operand::Zero()); |
| 330 | BranchOrBacktrack(eq, on_no_match); |
| 331 | |
| 332 | // On success, advance position by length of capture. |
| 333 | if (read_backward) { |
| 334 | __ sub(current_input_offset(), current_input_offset(), r4); |
| 335 | } else { |
| 336 | __ add(current_input_offset(), current_input_offset(), r4); |
| 337 | } |
| 338 | } |
| 339 | |
| 340 | __ bind(&fallthrough); |
| 341 | } |
| 342 | |
| 343 | |
| 344 | void RegExpMacroAssemblerARM::CheckNotBackReference(int start_reg, |
| 345 | bool read_backward, |
| 346 | Label* on_no_match) { |
| 347 | Label fallthrough; |
| 348 | Label success; |
| 349 | |
| 350 | // Find length of back-referenced capture. |
| 351 | __ ldr(r0, register_location(start_reg)); |
| 352 | __ ldr(r1, register_location(start_reg + 1)); |
| 353 | __ sub(r1, r1, r0, SetCC); // Length to check. |
| 354 | |
| 355 | // At this point, the capture registers are either both set or both cleared. |
| 356 | // If the capture length is zero, then the capture is either empty or cleared. |
| 357 | // Fall through in both cases. |
| 358 | __ b(eq, &fallthrough); |
| 359 | |
| 360 | // Check that there are enough characters left in the input. |
| 361 | if (read_backward) { |
| 362 | __ ldr(r3, MemOperand(frame_pointer(), kStringStartMinusOne)); |
| 363 | __ add(r3, r3, r1); |
| 364 | __ cmp(current_input_offset(), r3); |
| 365 | BranchOrBacktrack(lt, on_no_match); |
| 366 | } else { |
| 367 | __ cmn(r1, Operand(current_input_offset())); |
| 368 | BranchOrBacktrack(gt, on_no_match); |
| 369 | } |
| 370 | |
| 371 | // r0 - offset of start of capture |
| 372 | // r1 - length of capture |
| 373 | __ add(r0, r0, end_of_input_address()); |
| 374 | __ add(r2, end_of_input_address(), current_input_offset()); |
| 375 | if (read_backward) { |
| 376 | __ sub(r2, r2, r1); // Offset by length when matching backwards. |
| 377 | } |
| 378 | __ add(r1, r0, r1); |
| 379 | |
| 380 | Label loop; |
| 381 | __ bind(&loop); |
| 382 | if (mode_ == LATIN1) { |
| 383 | __ ldrb(r3, MemOperand(r0, char_size(), PostIndex)); |
| 384 | __ ldrb(r4, MemOperand(r2, char_size(), PostIndex)); |
| 385 | } else { |
| 386 | DCHECK(mode_ == UC16); |
| 387 | __ ldrh(r3, MemOperand(r0, char_size(), PostIndex)); |
| 388 | __ ldrh(r4, MemOperand(r2, char_size(), PostIndex)); |
| 389 | } |
| 390 | __ cmp(r3, r4); |
| 391 | BranchOrBacktrack(ne, on_no_match); |
| 392 | __ cmp(r0, r1); |
| 393 | __ b(lt, &loop); |
| 394 | |
| 395 | // Move current character position to position after match. |
| 396 | __ sub(current_input_offset(), r2, end_of_input_address()); |
| 397 | if (read_backward) { |
| 398 | __ ldr(r0, register_location(start_reg)); // Index of start of capture |
| 399 | __ ldr(r1, register_location(start_reg + 1)); // Index of end of capture |
| 400 | __ add(current_input_offset(), current_input_offset(), r0); |
| 401 | __ sub(current_input_offset(), current_input_offset(), r1); |
| 402 | } |
| 403 | |
| 404 | __ bind(&fallthrough); |
| 405 | } |
| 406 | |
| 407 | |
| 408 | void RegExpMacroAssemblerARM::CheckNotCharacter(unsigned c, |
| 409 | Label* on_not_equal) { |
| 410 | __ cmp(current_character(), Operand(c)); |
| 411 | BranchOrBacktrack(ne, on_not_equal); |
| 412 | } |
| 413 | |
| 414 | |
| 415 | void RegExpMacroAssemblerARM::CheckCharacterAfterAnd(uint32_t c, |
| 416 | uint32_t mask, |
| 417 | Label* on_equal) { |
| 418 | if (c == 0) { |
| 419 | __ tst(current_character(), Operand(mask)); |
| 420 | } else { |
| 421 | __ and_(r0, current_character(), Operand(mask)); |
| 422 | __ cmp(r0, Operand(c)); |
| 423 | } |
| 424 | BranchOrBacktrack(eq, on_equal); |
| 425 | } |
| 426 | |
| 427 | |
| 428 | void RegExpMacroAssemblerARM::CheckNotCharacterAfterAnd(unsigned c, |
| 429 | unsigned mask, |
| 430 | Label* on_not_equal) { |
| 431 | if (c == 0) { |
| 432 | __ tst(current_character(), Operand(mask)); |
| 433 | } else { |
| 434 | __ and_(r0, current_character(), Operand(mask)); |
| 435 | __ cmp(r0, Operand(c)); |
| 436 | } |
| 437 | BranchOrBacktrack(ne, on_not_equal); |
| 438 | } |
| 439 | |
| 440 | |
| 441 | void RegExpMacroAssemblerARM::CheckNotCharacterAfterMinusAnd( |
| 442 | uc16 c, |
| 443 | uc16 minus, |
| 444 | uc16 mask, |
| 445 | Label* on_not_equal) { |
| 446 | DCHECK(minus < String::kMaxUtf16CodeUnit); |
| 447 | __ sub(r0, current_character(), Operand(minus)); |
| 448 | __ and_(r0, r0, Operand(mask)); |
| 449 | __ cmp(r0, Operand(c)); |
| 450 | BranchOrBacktrack(ne, on_not_equal); |
| 451 | } |
| 452 | |
| 453 | |
| 454 | void RegExpMacroAssemblerARM::CheckCharacterInRange( |
| 455 | uc16 from, |
| 456 | uc16 to, |
| 457 | Label* on_in_range) { |
| 458 | __ sub(r0, current_character(), Operand(from)); |
| 459 | __ cmp(r0, Operand(to - from)); |
| 460 | BranchOrBacktrack(ls, on_in_range); // Unsigned lower-or-same condition. |
| 461 | } |
| 462 | |
| 463 | |
| 464 | void RegExpMacroAssemblerARM::CheckCharacterNotInRange( |
| 465 | uc16 from, |
| 466 | uc16 to, |
| 467 | Label* on_not_in_range) { |
| 468 | __ sub(r0, current_character(), Operand(from)); |
| 469 | __ cmp(r0, Operand(to - from)); |
| 470 | BranchOrBacktrack(hi, on_not_in_range); // Unsigned higher condition. |
| 471 | } |
| 472 | |
| 473 | |
| 474 | void RegExpMacroAssemblerARM::CheckBitInTable( |
| 475 | Handle<ByteArray> table, |
| 476 | Label* on_bit_set) { |
| 477 | __ mov(r0, Operand(table)); |
| 478 | if (mode_ != LATIN1 || kTableMask != String::kMaxOneByteCharCode) { |
| 479 | __ and_(r1, current_character(), Operand(kTableSize - 1)); |
| 480 | __ add(r1, r1, Operand(ByteArray::kHeaderSize - kHeapObjectTag)); |
| 481 | } else { |
| 482 | __ add(r1, |
| 483 | current_character(), |
| 484 | Operand(ByteArray::kHeaderSize - kHeapObjectTag)); |
| 485 | } |
| 486 | __ ldrb(r0, MemOperand(r0, r1)); |
| 487 | __ cmp(r0, Operand::Zero()); |
| 488 | BranchOrBacktrack(ne, on_bit_set); |
| 489 | } |
| 490 | |
| 491 | |
| 492 | bool RegExpMacroAssemblerARM::CheckSpecialCharacterClass(uc16 type, |
| 493 | Label* on_no_match) { |
| 494 | // Range checks (c in min..max) are generally implemented by an unsigned |
| 495 | // (c - min) <= (max - min) check |
| 496 | switch (type) { |
| 497 | case 's': |
| 498 | // Match space-characters |
| 499 | if (mode_ == LATIN1) { |
| 500 | // One byte space characters are '\t'..'\r', ' ' and \u00a0. |
| 501 | Label success; |
| 502 | __ cmp(current_character(), Operand(' ')); |
| 503 | __ b(eq, &success); |
| 504 | // Check range 0x09..0x0d |
| 505 | __ sub(r0, current_character(), Operand('\t')); |
| 506 | __ cmp(r0, Operand('\r' - '\t')); |
| 507 | __ b(ls, &success); |
| 508 | // \u00a0 (NBSP). |
| 509 | __ cmp(r0, Operand(0x00a0 - '\t')); |
| 510 | BranchOrBacktrack(ne, on_no_match); |
| 511 | __ bind(&success); |
| 512 | return true; |
| 513 | } |
| 514 | return false; |
| 515 | case 'S': |
| 516 | // The emitted code for generic character classes is good enough. |
| 517 | return false; |
| 518 | case 'd': |
| 519 | // Match ASCII digits ('0'..'9') |
| 520 | __ sub(r0, current_character(), Operand('0')); |
| 521 | __ cmp(r0, Operand('9' - '0')); |
| 522 | BranchOrBacktrack(hi, on_no_match); |
| 523 | return true; |
| 524 | case 'D': |
| 525 | // Match non ASCII-digits |
| 526 | __ sub(r0, current_character(), Operand('0')); |
| 527 | __ cmp(r0, Operand('9' - '0')); |
| 528 | BranchOrBacktrack(ls, on_no_match); |
| 529 | return true; |
| 530 | case '.': { |
| 531 | // Match non-newlines (not 0x0a('\n'), 0x0d('\r'), 0x2028 and 0x2029) |
| 532 | __ eor(r0, current_character(), Operand(0x01)); |
| 533 | // See if current character is '\n'^1 or '\r'^1, i.e., 0x0b or 0x0c |
| 534 | __ sub(r0, r0, Operand(0x0b)); |
| 535 | __ cmp(r0, Operand(0x0c - 0x0b)); |
| 536 | BranchOrBacktrack(ls, on_no_match); |
| 537 | if (mode_ == UC16) { |
| 538 | // Compare original value to 0x2028 and 0x2029, using the already |
| 539 | // computed (current_char ^ 0x01 - 0x0b). I.e., check for |
| 540 | // 0x201d (0x2028 - 0x0b) or 0x201e. |
| 541 | __ sub(r0, r0, Operand(0x2028 - 0x0b)); |
| 542 | __ cmp(r0, Operand(1)); |
| 543 | BranchOrBacktrack(ls, on_no_match); |
| 544 | } |
| 545 | return true; |
| 546 | } |
| 547 | case 'n': { |
| 548 | // Match newlines (0x0a('\n'), 0x0d('\r'), 0x2028 and 0x2029) |
| 549 | __ eor(r0, current_character(), Operand(0x01)); |
| 550 | // See if current character is '\n'^1 or '\r'^1, i.e., 0x0b or 0x0c |
| 551 | __ sub(r0, r0, Operand(0x0b)); |
| 552 | __ cmp(r0, Operand(0x0c - 0x0b)); |
| 553 | if (mode_ == LATIN1) { |
| 554 | BranchOrBacktrack(hi, on_no_match); |
| 555 | } else { |
| 556 | Label done; |
| 557 | __ b(ls, &done); |
| 558 | // Compare original value to 0x2028 and 0x2029, using the already |
| 559 | // computed (current_char ^ 0x01 - 0x0b). I.e., check for |
| 560 | // 0x201d (0x2028 - 0x0b) or 0x201e. |
| 561 | __ sub(r0, r0, Operand(0x2028 - 0x0b)); |
| 562 | __ cmp(r0, Operand(1)); |
| 563 | BranchOrBacktrack(hi, on_no_match); |
| 564 | __ bind(&done); |
| 565 | } |
| 566 | return true; |
| 567 | } |
| 568 | case 'w': { |
| 569 | if (mode_ != LATIN1) { |
| 570 | // Table is 256 entries, so all Latin1 characters can be tested. |
| 571 | __ cmp(current_character(), Operand('z')); |
| 572 | BranchOrBacktrack(hi, on_no_match); |
| 573 | } |
| 574 | ExternalReference map = ExternalReference::re_word_character_map(); |
| 575 | __ mov(r0, Operand(map)); |
| 576 | __ ldrb(r0, MemOperand(r0, current_character())); |
| 577 | __ cmp(r0, Operand::Zero()); |
| 578 | BranchOrBacktrack(eq, on_no_match); |
| 579 | return true; |
| 580 | } |
| 581 | case 'W': { |
| 582 | Label done; |
| 583 | if (mode_ != LATIN1) { |
| 584 | // Table is 256 entries, so all Latin1 characters can be tested. |
| 585 | __ cmp(current_character(), Operand('z')); |
| 586 | __ b(hi, &done); |
| 587 | } |
| 588 | ExternalReference map = ExternalReference::re_word_character_map(); |
| 589 | __ mov(r0, Operand(map)); |
| 590 | __ ldrb(r0, MemOperand(r0, current_character())); |
| 591 | __ cmp(r0, Operand::Zero()); |
| 592 | BranchOrBacktrack(ne, on_no_match); |
| 593 | if (mode_ != LATIN1) { |
| 594 | __ bind(&done); |
| 595 | } |
| 596 | return true; |
| 597 | } |
| 598 | case '*': |
| 599 | // Match any character. |
| 600 | return true; |
| 601 | // No custom implementation (yet): s(UC16), S(UC16). |
| 602 | default: |
| 603 | return false; |
| 604 | } |
| 605 | } |
| 606 | |
| 607 | |
| 608 | void RegExpMacroAssemblerARM::Fail() { |
| 609 | __ mov(r0, Operand(FAILURE)); |
| 610 | __ jmp(&exit_label_); |
| 611 | } |
| 612 | |
| 613 | |
| 614 | Handle<HeapObject> RegExpMacroAssemblerARM::GetCode(Handle<String> source) { |
| 615 | Label return_r0; |
| 616 | // Finalize code - write the entry point code now we know how many |
| 617 | // registers we need. |
| 618 | |
| 619 | // Entry code: |
| 620 | __ bind(&entry_label_); |
| 621 | |
| 622 | // Tell the system that we have a stack frame. Because the type is MANUAL, no |
| 623 | // is generated. |
| 624 | FrameScope scope(masm_, StackFrame::MANUAL); |
| 625 | |
| 626 | // Actually emit code to start a new stack frame. |
| 627 | // Push arguments |
| 628 | // Save callee-save registers. |
| 629 | // Start new stack frame. |
| 630 | // Store link register in existing stack-cell. |
| 631 | // Order here should correspond to order of offset constants in header file. |
| 632 | RegList registers_to_retain = r4.bit() | r5.bit() | r6.bit() | |
| 633 | r7.bit() | r8.bit() | r9.bit() | r10.bit() | fp.bit(); |
| 634 | RegList argument_registers = r0.bit() | r1.bit() | r2.bit() | r3.bit(); |
| 635 | __ stm(db_w, sp, argument_registers | registers_to_retain | lr.bit()); |
| 636 | // Set frame pointer in space for it if this is not a direct call |
| 637 | // from generated code. |
| 638 | __ add(frame_pointer(), sp, Operand(4 * kPointerSize)); |
| 639 | __ mov(r0, Operand::Zero()); |
| 640 | __ push(r0); // Make room for success counter and initialize it to 0. |
| 641 | __ push(r0); // Make room for "string start - 1" constant. |
| 642 | // Check if we have space on the stack for registers. |
| 643 | Label stack_limit_hit; |
| 644 | Label stack_ok; |
| 645 | |
| 646 | ExternalReference stack_limit = |
| 647 | ExternalReference::address_of_stack_limit(isolate()); |
| 648 | __ mov(r0, Operand(stack_limit)); |
| 649 | __ ldr(r0, MemOperand(r0)); |
| 650 | __ sub(r0, sp, r0, SetCC); |
| 651 | // Handle it if the stack pointer is already below the stack limit. |
| 652 | __ b(ls, &stack_limit_hit); |
| 653 | // Check if there is room for the variable number of registers above |
| 654 | // the stack limit. |
| 655 | __ cmp(r0, Operand(num_registers_ * kPointerSize)); |
| 656 | __ b(hs, &stack_ok); |
| 657 | // Exit with OutOfMemory exception. There is not enough space on the stack |
| 658 | // for our working registers. |
| 659 | __ mov(r0, Operand(EXCEPTION)); |
| 660 | __ jmp(&return_r0); |
| 661 | |
| 662 | __ bind(&stack_limit_hit); |
| 663 | CallCheckStackGuardState(r0); |
| 664 | __ cmp(r0, Operand::Zero()); |
| 665 | // If returned value is non-zero, we exit with the returned value as result. |
| 666 | __ b(ne, &return_r0); |
| 667 | |
| 668 | __ bind(&stack_ok); |
| 669 | |
| 670 | // Allocate space on stack for registers. |
| 671 | __ sub(sp, sp, Operand(num_registers_ * kPointerSize)); |
| 672 | // Load string end. |
| 673 | __ ldr(end_of_input_address(), MemOperand(frame_pointer(), kInputEnd)); |
| 674 | // Load input start. |
| 675 | __ ldr(r0, MemOperand(frame_pointer(), kInputStart)); |
| 676 | // Find negative length (offset of start relative to end). |
| 677 | __ sub(current_input_offset(), r0, end_of_input_address()); |
| 678 | // Set r0 to address of char before start of the input string |
| 679 | // (effectively string position -1). |
| 680 | __ ldr(r1, MemOperand(frame_pointer(), kStartIndex)); |
| 681 | __ sub(r0, current_input_offset(), Operand(char_size())); |
| 682 | __ sub(r0, r0, Operand(r1, LSL, (mode_ == UC16) ? 1 : 0)); |
| 683 | // Store this value in a local variable, for use when clearing |
| 684 | // position registers. |
| 685 | __ str(r0, MemOperand(frame_pointer(), kStringStartMinusOne)); |
| 686 | |
| 687 | // Initialize code pointer register |
| 688 | __ mov(code_pointer(), Operand(masm_->CodeObject())); |
| 689 | |
| 690 | Label load_char_start_regexp, start_regexp; |
| 691 | // Load newline if index is at start, previous character otherwise. |
| 692 | __ cmp(r1, Operand::Zero()); |
| 693 | __ b(ne, &load_char_start_regexp); |
| 694 | __ mov(current_character(), Operand('\n'), LeaveCC, eq); |
| 695 | __ jmp(&start_regexp); |
| 696 | |
| 697 | // Global regexp restarts matching here. |
| 698 | __ bind(&load_char_start_regexp); |
| 699 | // Load previous char as initial value of current character register. |
| 700 | LoadCurrentCharacterUnchecked(-1, 1); |
| 701 | __ bind(&start_regexp); |
| 702 | |
| 703 | // Initialize on-stack registers. |
| 704 | if (num_saved_registers_ > 0) { // Always is, if generated from a regexp. |
| 705 | // Fill saved registers with initial value = start offset - 1 |
| 706 | if (num_saved_registers_ > 8) { |
| 707 | // Address of register 0. |
| 708 | __ add(r1, frame_pointer(), Operand(kRegisterZero)); |
| 709 | __ mov(r2, Operand(num_saved_registers_)); |
| 710 | Label init_loop; |
| 711 | __ bind(&init_loop); |
| 712 | __ str(r0, MemOperand(r1, kPointerSize, NegPostIndex)); |
| 713 | __ sub(r2, r2, Operand(1), SetCC); |
| 714 | __ b(ne, &init_loop); |
| 715 | } else { |
| 716 | for (int i = 0; i < num_saved_registers_; i++) { |
| 717 | __ str(r0, register_location(i)); |
| 718 | } |
| 719 | } |
| 720 | } |
| 721 | |
| 722 | // Initialize backtrack stack pointer. |
| 723 | __ ldr(backtrack_stackpointer(), MemOperand(frame_pointer(), kStackHighEnd)); |
| 724 | |
| 725 | __ jmp(&start_label_); |
| 726 | |
| 727 | // Exit code: |
| 728 | if (success_label_.is_linked()) { |
| 729 | // Save captures when successful. |
| 730 | __ bind(&success_label_); |
| 731 | if (num_saved_registers_ > 0) { |
| 732 | // copy captures to output |
| 733 | __ ldr(r1, MemOperand(frame_pointer(), kInputStart)); |
| 734 | __ ldr(r0, MemOperand(frame_pointer(), kRegisterOutput)); |
| 735 | __ ldr(r2, MemOperand(frame_pointer(), kStartIndex)); |
| 736 | __ sub(r1, end_of_input_address(), r1); |
| 737 | // r1 is length of input in bytes. |
| 738 | if (mode_ == UC16) { |
| 739 | __ mov(r1, Operand(r1, LSR, 1)); |
| 740 | } |
| 741 | // r1 is length of input in characters. |
| 742 | __ add(r1, r1, Operand(r2)); |
| 743 | // r1 is length of string in characters. |
| 744 | |
| 745 | DCHECK_EQ(0, num_saved_registers_ % 2); |
| 746 | // Always an even number of capture registers. This allows us to |
| 747 | // unroll the loop once to add an operation between a load of a register |
| 748 | // and the following use of that register. |
| 749 | for (int i = 0; i < num_saved_registers_; i += 2) { |
| 750 | __ ldr(r2, register_location(i)); |
| 751 | __ ldr(r3, register_location(i + 1)); |
| 752 | if (i == 0 && global_with_zero_length_check()) { |
| 753 | // Keep capture start in r4 for the zero-length check later. |
| 754 | __ mov(r4, r2); |
| 755 | } |
| 756 | if (mode_ == UC16) { |
| 757 | __ add(r2, r1, Operand(r2, ASR, 1)); |
| 758 | __ add(r3, r1, Operand(r3, ASR, 1)); |
| 759 | } else { |
| 760 | __ add(r2, r1, Operand(r2)); |
| 761 | __ add(r3, r1, Operand(r3)); |
| 762 | } |
| 763 | __ str(r2, MemOperand(r0, kPointerSize, PostIndex)); |
| 764 | __ str(r3, MemOperand(r0, kPointerSize, PostIndex)); |
| 765 | } |
| 766 | } |
| 767 | |
| 768 | if (global()) { |
| 769 | // Restart matching if the regular expression is flagged as global. |
| 770 | __ ldr(r0, MemOperand(frame_pointer(), kSuccessfulCaptures)); |
| 771 | __ ldr(r1, MemOperand(frame_pointer(), kNumOutputRegisters)); |
| 772 | __ ldr(r2, MemOperand(frame_pointer(), kRegisterOutput)); |
| 773 | // Increment success counter. |
| 774 | __ add(r0, r0, Operand(1)); |
| 775 | __ str(r0, MemOperand(frame_pointer(), kSuccessfulCaptures)); |
| 776 | // Capture results have been stored, so the number of remaining global |
| 777 | // output registers is reduced by the number of stored captures. |
| 778 | __ sub(r1, r1, Operand(num_saved_registers_)); |
| 779 | // Check whether we have enough room for another set of capture results. |
| 780 | __ cmp(r1, Operand(num_saved_registers_)); |
| 781 | __ b(lt, &return_r0); |
| 782 | |
| 783 | __ str(r1, MemOperand(frame_pointer(), kNumOutputRegisters)); |
| 784 | // Advance the location for output. |
| 785 | __ add(r2, r2, Operand(num_saved_registers_ * kPointerSize)); |
| 786 | __ str(r2, MemOperand(frame_pointer(), kRegisterOutput)); |
| 787 | |
| 788 | // Prepare r0 to initialize registers with its value in the next run. |
| 789 | __ ldr(r0, MemOperand(frame_pointer(), kStringStartMinusOne)); |
| 790 | |
| 791 | if (global_with_zero_length_check()) { |
| 792 | // Special case for zero-length matches. |
| 793 | // r4: capture start index |
| 794 | __ cmp(current_input_offset(), r4); |
| 795 | // Not a zero-length match, restart. |
| 796 | __ b(ne, &load_char_start_regexp); |
| 797 | // Offset from the end is zero if we already reached the end. |
| 798 | __ cmp(current_input_offset(), Operand::Zero()); |
| 799 | __ b(eq, &exit_label_); |
| 800 | // Advance current position after a zero-length match. |
| 801 | __ add(current_input_offset(), |
| 802 | current_input_offset(), |
| 803 | Operand((mode_ == UC16) ? 2 : 1)); |
| 804 | } |
| 805 | |
| 806 | __ b(&load_char_start_regexp); |
| 807 | } else { |
| 808 | __ mov(r0, Operand(SUCCESS)); |
| 809 | } |
| 810 | } |
| 811 | |
| 812 | // Exit and return r0 |
| 813 | __ bind(&exit_label_); |
| 814 | if (global()) { |
| 815 | __ ldr(r0, MemOperand(frame_pointer(), kSuccessfulCaptures)); |
| 816 | } |
| 817 | |
| 818 | __ bind(&return_r0); |
| 819 | // Skip sp past regexp registers and local variables.. |
| 820 | __ mov(sp, frame_pointer()); |
| 821 | // Restore registers r4..r11 and return (restoring lr to pc). |
| 822 | __ ldm(ia_w, sp, registers_to_retain | pc.bit()); |
| 823 | |
| 824 | // Backtrack code (branch target for conditional backtracks). |
| 825 | if (backtrack_label_.is_linked()) { |
| 826 | __ bind(&backtrack_label_); |
| 827 | Backtrack(); |
| 828 | } |
| 829 | |
| 830 | Label exit_with_exception; |
| 831 | |
| 832 | // Preempt-code |
| 833 | if (check_preempt_label_.is_linked()) { |
| 834 | SafeCallTarget(&check_preempt_label_); |
| 835 | |
| 836 | CallCheckStackGuardState(r0); |
| 837 | __ cmp(r0, Operand::Zero()); |
| 838 | // If returning non-zero, we should end execution with the given |
| 839 | // result as return value. |
| 840 | __ b(ne, &return_r0); |
| 841 | |
| 842 | // String might have moved: Reload end of string from frame. |
| 843 | __ ldr(end_of_input_address(), MemOperand(frame_pointer(), kInputEnd)); |
| 844 | SafeReturn(); |
| 845 | } |
| 846 | |
| 847 | // Backtrack stack overflow code. |
| 848 | if (stack_overflow_label_.is_linked()) { |
| 849 | SafeCallTarget(&stack_overflow_label_); |
| 850 | // Reached if the backtrack-stack limit has been hit. |
| 851 | Label grow_failed; |
| 852 | |
| 853 | // Call GrowStack(backtrack_stackpointer(), &stack_base) |
| 854 | static const int num_arguments = 3; |
| 855 | __ PrepareCallCFunction(num_arguments, r0); |
| 856 | __ mov(r0, backtrack_stackpointer()); |
| 857 | __ add(r1, frame_pointer(), Operand(kStackHighEnd)); |
| 858 | __ mov(r2, Operand(ExternalReference::isolate_address(isolate()))); |
| 859 | ExternalReference grow_stack = |
| 860 | ExternalReference::re_grow_stack(isolate()); |
| 861 | __ CallCFunction(grow_stack, num_arguments); |
| 862 | // If return NULL, we have failed to grow the stack, and |
| 863 | // must exit with a stack-overflow exception. |
| 864 | __ cmp(r0, Operand::Zero()); |
| 865 | __ b(eq, &exit_with_exception); |
| 866 | // Otherwise use return value as new stack pointer. |
| 867 | __ mov(backtrack_stackpointer(), r0); |
| 868 | // Restore saved registers and continue. |
| 869 | SafeReturn(); |
| 870 | } |
| 871 | |
| 872 | if (exit_with_exception.is_linked()) { |
| 873 | // If any of the code above needed to exit with an exception. |
| 874 | __ bind(&exit_with_exception); |
| 875 | // Exit with Result EXCEPTION(-1) to signal thrown exception. |
| 876 | __ mov(r0, Operand(EXCEPTION)); |
| 877 | __ jmp(&return_r0); |
| 878 | } |
| 879 | |
| 880 | CodeDesc code_desc; |
| 881 | masm_->GetCode(&code_desc); |
| 882 | Handle<Code> code = isolate()->factory()->NewCode( |
| 883 | code_desc, Code::ComputeFlags(Code::REGEXP), masm_->CodeObject()); |
| 884 | PROFILE(masm_->isolate(), RegExpCodeCreateEvent(*code, *source)); |
| 885 | return Handle<HeapObject>::cast(code); |
| 886 | } |
| 887 | |
| 888 | |
| 889 | void RegExpMacroAssemblerARM::GoTo(Label* to) { |
| 890 | BranchOrBacktrack(al, to); |
| 891 | } |
| 892 | |
| 893 | |
| 894 | void RegExpMacroAssemblerARM::IfRegisterGE(int reg, |
| 895 | int comparand, |
| 896 | Label* if_ge) { |
| 897 | __ ldr(r0, register_location(reg)); |
| 898 | __ cmp(r0, Operand(comparand)); |
| 899 | BranchOrBacktrack(ge, if_ge); |
| 900 | } |
| 901 | |
| 902 | |
| 903 | void RegExpMacroAssemblerARM::IfRegisterLT(int reg, |
| 904 | int comparand, |
| 905 | Label* if_lt) { |
| 906 | __ ldr(r0, register_location(reg)); |
| 907 | __ cmp(r0, Operand(comparand)); |
| 908 | BranchOrBacktrack(lt, if_lt); |
| 909 | } |
| 910 | |
| 911 | |
| 912 | void RegExpMacroAssemblerARM::IfRegisterEqPos(int reg, |
| 913 | Label* if_eq) { |
| 914 | __ ldr(r0, register_location(reg)); |
| 915 | __ cmp(r0, Operand(current_input_offset())); |
| 916 | BranchOrBacktrack(eq, if_eq); |
| 917 | } |
| 918 | |
| 919 | |
| 920 | RegExpMacroAssembler::IrregexpImplementation |
| 921 | RegExpMacroAssemblerARM::Implementation() { |
| 922 | return kARMImplementation; |
| 923 | } |
| 924 | |
| 925 | |
| 926 | void RegExpMacroAssemblerARM::LoadCurrentCharacter(int cp_offset, |
| 927 | Label* on_end_of_input, |
| 928 | bool check_bounds, |
| 929 | int characters) { |
| 930 | DCHECK(cp_offset < (1<<30)); // Be sane! (And ensure negation works) |
| 931 | if (check_bounds) { |
| 932 | if (cp_offset >= 0) { |
| 933 | CheckPosition(cp_offset + characters - 1, on_end_of_input); |
| 934 | } else { |
| 935 | CheckPosition(cp_offset, on_end_of_input); |
| 936 | } |
| 937 | } |
| 938 | LoadCurrentCharacterUnchecked(cp_offset, characters); |
| 939 | } |
| 940 | |
| 941 | |
| 942 | void RegExpMacroAssemblerARM::PopCurrentPosition() { |
| 943 | Pop(current_input_offset()); |
| 944 | } |
| 945 | |
| 946 | |
| 947 | void RegExpMacroAssemblerARM::PopRegister(int register_index) { |
| 948 | Pop(r0); |
| 949 | __ str(r0, register_location(register_index)); |
| 950 | } |
| 951 | |
| 952 | |
| 953 | void RegExpMacroAssemblerARM::PushBacktrack(Label* label) { |
| 954 | __ mov_label_offset(r0, label); |
| 955 | Push(r0); |
| 956 | CheckStackLimit(); |
| 957 | } |
| 958 | |
| 959 | |
| 960 | void RegExpMacroAssemblerARM::PushCurrentPosition() { |
| 961 | Push(current_input_offset()); |
| 962 | } |
| 963 | |
| 964 | |
| 965 | void RegExpMacroAssemblerARM::PushRegister(int register_index, |
| 966 | StackCheckFlag check_stack_limit) { |
| 967 | __ ldr(r0, register_location(register_index)); |
| 968 | Push(r0); |
| 969 | if (check_stack_limit) CheckStackLimit(); |
| 970 | } |
| 971 | |
| 972 | |
| 973 | void RegExpMacroAssemblerARM::ReadCurrentPositionFromRegister(int reg) { |
| 974 | __ ldr(current_input_offset(), register_location(reg)); |
| 975 | } |
| 976 | |
| 977 | |
| 978 | void RegExpMacroAssemblerARM::ReadStackPointerFromRegister(int reg) { |
| 979 | __ ldr(backtrack_stackpointer(), register_location(reg)); |
| 980 | __ ldr(r0, MemOperand(frame_pointer(), kStackHighEnd)); |
| 981 | __ add(backtrack_stackpointer(), backtrack_stackpointer(), Operand(r0)); |
| 982 | } |
| 983 | |
| 984 | |
| 985 | void RegExpMacroAssemblerARM::SetCurrentPositionFromEnd(int by) { |
| 986 | Label after_position; |
| 987 | __ cmp(current_input_offset(), Operand(-by * char_size())); |
| 988 | __ b(ge, &after_position); |
| 989 | __ mov(current_input_offset(), Operand(-by * char_size())); |
| 990 | // On RegExp code entry (where this operation is used), the character before |
| 991 | // the current position is expected to be already loaded. |
| 992 | // We have advanced the position, so it's safe to read backwards. |
| 993 | LoadCurrentCharacterUnchecked(-1, 1); |
| 994 | __ bind(&after_position); |
| 995 | } |
| 996 | |
| 997 | |
| 998 | void RegExpMacroAssemblerARM::SetRegister(int register_index, int to) { |
| 999 | DCHECK(register_index >= num_saved_registers_); // Reserved for positions! |
| 1000 | __ mov(r0, Operand(to)); |
| 1001 | __ str(r0, register_location(register_index)); |
| 1002 | } |
| 1003 | |
| 1004 | |
| 1005 | bool RegExpMacroAssemblerARM::Succeed() { |
| 1006 | __ jmp(&success_label_); |
| 1007 | return global(); |
| 1008 | } |
| 1009 | |
| 1010 | |
| 1011 | void RegExpMacroAssemblerARM::WriteCurrentPositionToRegister(int reg, |
| 1012 | int cp_offset) { |
| 1013 | if (cp_offset == 0) { |
| 1014 | __ str(current_input_offset(), register_location(reg)); |
| 1015 | } else { |
| 1016 | __ add(r0, current_input_offset(), Operand(cp_offset * char_size())); |
| 1017 | __ str(r0, register_location(reg)); |
| 1018 | } |
| 1019 | } |
| 1020 | |
| 1021 | |
| 1022 | void RegExpMacroAssemblerARM::ClearRegisters(int reg_from, int reg_to) { |
| 1023 | DCHECK(reg_from <= reg_to); |
| 1024 | __ ldr(r0, MemOperand(frame_pointer(), kStringStartMinusOne)); |
| 1025 | for (int reg = reg_from; reg <= reg_to; reg++) { |
| 1026 | __ str(r0, register_location(reg)); |
| 1027 | } |
| 1028 | } |
| 1029 | |
| 1030 | |
| 1031 | void RegExpMacroAssemblerARM::WriteStackPointerToRegister(int reg) { |
| 1032 | __ ldr(r1, MemOperand(frame_pointer(), kStackHighEnd)); |
| 1033 | __ sub(r0, backtrack_stackpointer(), r1); |
| 1034 | __ str(r0, register_location(reg)); |
| 1035 | } |
| 1036 | |
| 1037 | |
| 1038 | // Private methods: |
| 1039 | |
| 1040 | void RegExpMacroAssemblerARM::CallCheckStackGuardState(Register scratch) { |
| 1041 | __ PrepareCallCFunction(3, scratch); |
| 1042 | |
| 1043 | // RegExp code frame pointer. |
| 1044 | __ mov(r2, frame_pointer()); |
| 1045 | // Code* of self. |
| 1046 | __ mov(r1, Operand(masm_->CodeObject())); |
| 1047 | |
| 1048 | // We need to make room for the return address on the stack. |
| 1049 | int stack_alignment = base::OS::ActivationFrameAlignment(); |
| 1050 | DCHECK(IsAligned(stack_alignment, kPointerSize)); |
| 1051 | __ sub(sp, sp, Operand(stack_alignment)); |
| 1052 | |
| 1053 | // r0 will point to the return address, placed by DirectCEntry. |
| 1054 | __ mov(r0, sp); |
| 1055 | |
| 1056 | ExternalReference stack_guard_check = |
| 1057 | ExternalReference::re_check_stack_guard_state(isolate()); |
| 1058 | __ mov(ip, Operand(stack_guard_check)); |
| 1059 | DirectCEntryStub stub(isolate()); |
| 1060 | stub.GenerateCall(masm_, ip); |
| 1061 | |
| 1062 | // Drop the return address from the stack. |
| 1063 | __ add(sp, sp, Operand(stack_alignment)); |
| 1064 | |
| 1065 | DCHECK(stack_alignment != 0); |
| 1066 | __ ldr(sp, MemOperand(sp, 0)); |
| 1067 | |
| 1068 | __ mov(code_pointer(), Operand(masm_->CodeObject())); |
| 1069 | } |
| 1070 | |
| 1071 | |
| 1072 | // Helper function for reading a value out of a stack frame. |
| 1073 | template <typename T> |
| 1074 | static T& frame_entry(Address re_frame, int frame_offset) { |
| 1075 | return reinterpret_cast<T&>(Memory::int32_at(re_frame + frame_offset)); |
| 1076 | } |
| 1077 | |
| 1078 | |
| 1079 | template <typename T> |
| 1080 | static T* frame_entry_address(Address re_frame, int frame_offset) { |
| 1081 | return reinterpret_cast<T*>(re_frame + frame_offset); |
| 1082 | } |
| 1083 | |
| 1084 | |
| 1085 | int RegExpMacroAssemblerARM::CheckStackGuardState(Address* return_address, |
| 1086 | Code* re_code, |
| 1087 | Address re_frame) { |
| 1088 | return NativeRegExpMacroAssembler::CheckStackGuardState( |
| 1089 | frame_entry<Isolate*>(re_frame, kIsolate), |
| 1090 | frame_entry<int>(re_frame, kStartIndex), |
| 1091 | frame_entry<int>(re_frame, kDirectCall) == 1, return_address, re_code, |
| 1092 | frame_entry_address<String*>(re_frame, kInputString), |
| 1093 | frame_entry_address<const byte*>(re_frame, kInputStart), |
| 1094 | frame_entry_address<const byte*>(re_frame, kInputEnd)); |
| 1095 | } |
| 1096 | |
| 1097 | |
| 1098 | MemOperand RegExpMacroAssemblerARM::register_location(int register_index) { |
| 1099 | DCHECK(register_index < (1<<30)); |
| 1100 | if (num_registers_ <= register_index) { |
| 1101 | num_registers_ = register_index + 1; |
| 1102 | } |
| 1103 | return MemOperand(frame_pointer(), |
| 1104 | kRegisterZero - register_index * kPointerSize); |
| 1105 | } |
| 1106 | |
| 1107 | |
| 1108 | void RegExpMacroAssemblerARM::CheckPosition(int cp_offset, |
| 1109 | Label* on_outside_input) { |
| 1110 | if (cp_offset >= 0) { |
| 1111 | __ cmp(current_input_offset(), Operand(-cp_offset * char_size())); |
| 1112 | BranchOrBacktrack(ge, on_outside_input); |
| 1113 | } else { |
| 1114 | __ ldr(r1, MemOperand(frame_pointer(), kStringStartMinusOne)); |
| 1115 | __ add(r0, current_input_offset(), Operand(cp_offset * char_size())); |
| 1116 | __ cmp(r0, r1); |
| 1117 | BranchOrBacktrack(le, on_outside_input); |
| 1118 | } |
| 1119 | } |
| 1120 | |
| 1121 | |
| 1122 | void RegExpMacroAssemblerARM::BranchOrBacktrack(Condition condition, |
| 1123 | Label* to) { |
| 1124 | if (condition == al) { // Unconditional. |
| 1125 | if (to == NULL) { |
| 1126 | Backtrack(); |
| 1127 | return; |
| 1128 | } |
| 1129 | __ jmp(to); |
| 1130 | return; |
| 1131 | } |
| 1132 | if (to == NULL) { |
| 1133 | __ b(condition, &backtrack_label_); |
| 1134 | return; |
| 1135 | } |
| 1136 | __ b(condition, to); |
| 1137 | } |
| 1138 | |
| 1139 | |
| 1140 | void RegExpMacroAssemblerARM::SafeCall(Label* to, Condition cond) { |
| 1141 | __ bl(to, cond); |
| 1142 | } |
| 1143 | |
| 1144 | |
| 1145 | void RegExpMacroAssemblerARM::SafeReturn() { |
| 1146 | __ pop(lr); |
| 1147 | __ add(pc, lr, Operand(masm_->CodeObject())); |
| 1148 | } |
| 1149 | |
| 1150 | |
| 1151 | void RegExpMacroAssemblerARM::SafeCallTarget(Label* name) { |
| 1152 | __ bind(name); |
| 1153 | __ sub(lr, lr, Operand(masm_->CodeObject())); |
| 1154 | __ push(lr); |
| 1155 | } |
| 1156 | |
| 1157 | |
| 1158 | void RegExpMacroAssemblerARM::Push(Register source) { |
| 1159 | DCHECK(!source.is(backtrack_stackpointer())); |
| 1160 | __ str(source, |
| 1161 | MemOperand(backtrack_stackpointer(), kPointerSize, NegPreIndex)); |
| 1162 | } |
| 1163 | |
| 1164 | |
| 1165 | void RegExpMacroAssemblerARM::Pop(Register target) { |
| 1166 | DCHECK(!target.is(backtrack_stackpointer())); |
| 1167 | __ ldr(target, |
| 1168 | MemOperand(backtrack_stackpointer(), kPointerSize, PostIndex)); |
| 1169 | } |
| 1170 | |
| 1171 | |
| 1172 | void RegExpMacroAssemblerARM::CheckPreemption() { |
| 1173 | // Check for preemption. |
| 1174 | ExternalReference stack_limit = |
| 1175 | ExternalReference::address_of_stack_limit(isolate()); |
| 1176 | __ mov(r0, Operand(stack_limit)); |
| 1177 | __ ldr(r0, MemOperand(r0)); |
| 1178 | __ cmp(sp, r0); |
| 1179 | SafeCall(&check_preempt_label_, ls); |
| 1180 | } |
| 1181 | |
| 1182 | |
| 1183 | void RegExpMacroAssemblerARM::CheckStackLimit() { |
| 1184 | ExternalReference stack_limit = |
| 1185 | ExternalReference::address_of_regexp_stack_limit(isolate()); |
| 1186 | __ mov(r0, Operand(stack_limit)); |
| 1187 | __ ldr(r0, MemOperand(r0)); |
| 1188 | __ cmp(backtrack_stackpointer(), Operand(r0)); |
| 1189 | SafeCall(&stack_overflow_label_, ls); |
| 1190 | } |
| 1191 | |
| 1192 | |
| 1193 | bool RegExpMacroAssemblerARM::CanReadUnaligned() { |
| 1194 | return CpuFeatures::IsSupported(UNALIGNED_ACCESSES) && !slow_safe(); |
| 1195 | } |
| 1196 | |
| 1197 | |
| 1198 | void RegExpMacroAssemblerARM::LoadCurrentCharacterUnchecked(int cp_offset, |
| 1199 | int characters) { |
| 1200 | Register offset = current_input_offset(); |
| 1201 | if (cp_offset != 0) { |
| 1202 | // r4 is not being used to store the capture start index at this point. |
| 1203 | __ add(r4, current_input_offset(), Operand(cp_offset * char_size())); |
| 1204 | offset = r4; |
| 1205 | } |
| 1206 | // The ldr, str, ldrh, strh instructions can do unaligned accesses, if the CPU |
| 1207 | // and the operating system running on the target allow it. |
| 1208 | // If unaligned load/stores are not supported then this function must only |
| 1209 | // be used to load a single character at a time. |
| 1210 | if (!CanReadUnaligned()) { |
| 1211 | DCHECK(characters == 1); |
| 1212 | } |
| 1213 | |
| 1214 | if (mode_ == LATIN1) { |
| 1215 | if (characters == 4) { |
| 1216 | __ ldr(current_character(), MemOperand(end_of_input_address(), offset)); |
| 1217 | } else if (characters == 2) { |
| 1218 | __ ldrh(current_character(), MemOperand(end_of_input_address(), offset)); |
| 1219 | } else { |
| 1220 | DCHECK(characters == 1); |
| 1221 | __ ldrb(current_character(), MemOperand(end_of_input_address(), offset)); |
| 1222 | } |
| 1223 | } else { |
| 1224 | DCHECK(mode_ == UC16); |
| 1225 | if (characters == 2) { |
| 1226 | __ ldr(current_character(), MemOperand(end_of_input_address(), offset)); |
| 1227 | } else { |
| 1228 | DCHECK(characters == 1); |
| 1229 | __ ldrh(current_character(), MemOperand(end_of_input_address(), offset)); |
| 1230 | } |
| 1231 | } |
| 1232 | } |
| 1233 | |
| 1234 | |
| 1235 | #undef __ |
| 1236 | |
| 1237 | #endif // V8_INTERPRETED_REGEXP |
| 1238 | |
| 1239 | } // namespace internal |
| 1240 | } // namespace v8 |
| 1241 | |
| 1242 | #endif // V8_TARGET_ARCH_ARM |