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_MIPS |
| 6 | |
| 7 | #include "src/regexp/mips/regexp-macro-assembler-mips.h" |
| 8 | |
| 9 | #include "src/code-stubs.h" |
| 10 | #include "src/log.h" |
| 11 | #include "src/macro-assembler.h" |
| 12 | #include "src/regexp/regexp-macro-assembler.h" |
| 13 | #include "src/regexp/regexp-stack.h" |
| 14 | #include "src/unicode.h" |
| 15 | |
| 16 | namespace v8 { |
| 17 | namespace internal { |
| 18 | |
| 19 | #ifndef V8_INTERPRETED_REGEXP |
| 20 | /* |
| 21 | * This assembler uses the following register assignment convention |
| 22 | * - t7 : Temporarily stores the index of capture start after a matching pass |
| 23 | * for a global regexp. |
| 24 | * - t1 : Pointer to current code object (Code*) including heap object tag. |
| 25 | * - t2 : Current position in input, as negative offset from end of string. |
| 26 | * Please notice that this is the byte offset, not the character offset! |
| 27 | * - t3 : Currently loaded character. Must be loaded using |
| 28 | * LoadCurrentCharacter before using any of the dispatch methods. |
| 29 | * - t4 : Points to tip of backtrack stack |
| 30 | * - t5 : Unused. |
| 31 | * - t6 : End of input (points to byte after last character in input). |
| 32 | * - fp : Frame pointer. Used to access arguments, local variables and |
| 33 | * RegExp registers. |
| 34 | * - sp : Points to tip of C stack. |
| 35 | * |
| 36 | * The remaining registers are free for computations. |
| 37 | * Each call to a public method should retain this convention. |
| 38 | * |
| 39 | * The stack will have the following structure: |
| 40 | * |
| 41 | * - fp[64] Isolate* isolate (address of the current isolate) |
| 42 | * - fp[60] direct_call (if 1, direct call from JavaScript code, |
| 43 | * if 0, call through the runtime system). |
| 44 | * - fp[56] stack_area_base (High end of the memory area to use as |
| 45 | * backtracking stack). |
| 46 | * - fp[52] capture array size (may fit multiple sets of matches) |
| 47 | * - fp[48] int* capture_array (int[num_saved_registers_], for output). |
| 48 | * - fp[44] secondary link/return address used by native call. |
| 49 | * --- sp when called --- |
| 50 | * - fp[40] return address (lr). |
| 51 | * - fp[36] old frame pointer (r11). |
| 52 | * - fp[0..32] backup of registers s0..s7. |
| 53 | * --- frame pointer ---- |
| 54 | * - fp[-4] end of input (address of end of string). |
| 55 | * - fp[-8] start of input (address of first character in string). |
| 56 | * - fp[-12] start index (character index of start). |
| 57 | * - fp[-16] void* input_string (location of a handle containing the string). |
| 58 | * - fp[-20] success counter (only for global regexps to count matches). |
| 59 | * - fp[-24] Offset of location before start of input (effectively character |
| 60 | * position -1). Used to initialize capture registers to a |
| 61 | * non-position. |
| 62 | * - fp[-28] At start (if 1, we are starting at the start of the |
| 63 | * string, otherwise 0) |
| 64 | * - fp[-32] register 0 (Only positions must be stored in the first |
| 65 | * - register 1 num_saved_registers_ registers) |
| 66 | * - ... |
| 67 | * - register num_registers-1 |
| 68 | * --- sp --- |
| 69 | * |
| 70 | * The first num_saved_registers_ registers are initialized to point to |
| 71 | * "character -1" in the string (i.e., char_size() bytes before the first |
| 72 | * character of the string). The remaining registers start out as garbage. |
| 73 | * |
| 74 | * The data up to the return address must be placed there by the calling |
| 75 | * code and the remaining arguments are passed in registers, e.g. by calling the |
| 76 | * code entry as cast to a function with the signature: |
| 77 | * int (*match)(String* input_string, |
| 78 | * int start_index, |
| 79 | * Address start, |
| 80 | * Address end, |
| 81 | * Address secondary_return_address, // Only used by native call. |
| 82 | * int* capture_output_array, |
| 83 | * byte* stack_area_base, |
| 84 | * bool direct_call = false) |
| 85 | * The call is performed by NativeRegExpMacroAssembler::Execute() |
| 86 | * (in regexp-macro-assembler.cc) via the CALL_GENERATED_REGEXP_CODE macro |
| 87 | * in mips/simulator-mips.h. |
| 88 | * When calling as a non-direct call (i.e., from C++ code), the return address |
| 89 | * area is overwritten with the ra register by the RegExp code. When doing a |
| 90 | * direct call from generated code, the return address is placed there by |
| 91 | * the calling code, as in a normal exit frame. |
| 92 | */ |
| 93 | |
| 94 | #define __ ACCESS_MASM(masm_) |
| 95 | |
| 96 | RegExpMacroAssemblerMIPS::RegExpMacroAssemblerMIPS(Isolate* isolate, Zone* zone, |
| 97 | Mode mode, |
| 98 | int registers_to_save) |
| 99 | : NativeRegExpMacroAssembler(isolate, zone), |
| 100 | masm_(new MacroAssembler(isolate, NULL, kRegExpCodeSize, |
| 101 | CodeObjectRequired::kYes)), |
| 102 | mode_(mode), |
| 103 | num_registers_(registers_to_save), |
| 104 | num_saved_registers_(registers_to_save), |
| 105 | entry_label_(), |
| 106 | start_label_(), |
| 107 | success_label_(), |
| 108 | backtrack_label_(), |
| 109 | exit_label_(), |
| 110 | internal_failure_label_() { |
| 111 | DCHECK_EQ(0, registers_to_save % 2); |
| 112 | __ jmp(&entry_label_); // We'll write the entry code later. |
| 113 | // If the code gets too big or corrupted, an internal exception will be |
| 114 | // raised, and we will exit right away. |
| 115 | __ bind(&internal_failure_label_); |
| 116 | __ li(v0, Operand(FAILURE)); |
| 117 | __ Ret(); |
| 118 | __ bind(&start_label_); // And then continue from here. |
| 119 | } |
| 120 | |
| 121 | |
| 122 | RegExpMacroAssemblerMIPS::~RegExpMacroAssemblerMIPS() { |
| 123 | delete masm_; |
| 124 | // Unuse labels in case we throw away the assembler without calling GetCode. |
| 125 | entry_label_.Unuse(); |
| 126 | start_label_.Unuse(); |
| 127 | success_label_.Unuse(); |
| 128 | backtrack_label_.Unuse(); |
| 129 | exit_label_.Unuse(); |
| 130 | check_preempt_label_.Unuse(); |
| 131 | stack_overflow_label_.Unuse(); |
| 132 | internal_failure_label_.Unuse(); |
| 133 | } |
| 134 | |
| 135 | |
| 136 | int RegExpMacroAssemblerMIPS::stack_limit_slack() { |
| 137 | return RegExpStack::kStackLimitSlack; |
| 138 | } |
| 139 | |
| 140 | |
| 141 | void RegExpMacroAssemblerMIPS::AdvanceCurrentPosition(int by) { |
| 142 | if (by != 0) { |
| 143 | __ Addu(current_input_offset(), |
| 144 | current_input_offset(), Operand(by * char_size())); |
| 145 | } |
| 146 | } |
| 147 | |
| 148 | |
| 149 | void RegExpMacroAssemblerMIPS::AdvanceRegister(int reg, int by) { |
| 150 | DCHECK(reg >= 0); |
| 151 | DCHECK(reg < num_registers_); |
| 152 | if (by != 0) { |
| 153 | __ lw(a0, register_location(reg)); |
| 154 | __ Addu(a0, a0, Operand(by)); |
| 155 | __ sw(a0, register_location(reg)); |
| 156 | } |
| 157 | } |
| 158 | |
| 159 | |
| 160 | void RegExpMacroAssemblerMIPS::Backtrack() { |
| 161 | CheckPreemption(); |
| 162 | // Pop Code* offset from backtrack stack, add Code* and jump to location. |
| 163 | Pop(a0); |
| 164 | __ Addu(a0, a0, code_pointer()); |
| 165 | __ Jump(a0); |
| 166 | } |
| 167 | |
| 168 | |
| 169 | void RegExpMacroAssemblerMIPS::Bind(Label* label) { |
| 170 | __ bind(label); |
| 171 | } |
| 172 | |
| 173 | |
| 174 | void RegExpMacroAssemblerMIPS::CheckCharacter(uint32_t c, Label* on_equal) { |
| 175 | BranchOrBacktrack(on_equal, eq, current_character(), Operand(c)); |
| 176 | } |
| 177 | |
| 178 | |
| 179 | void RegExpMacroAssemblerMIPS::CheckCharacterGT(uc16 limit, Label* on_greater) { |
| 180 | BranchOrBacktrack(on_greater, gt, current_character(), Operand(limit)); |
| 181 | } |
| 182 | |
| 183 | |
| 184 | void RegExpMacroAssemblerMIPS::CheckAtStart(Label* on_at_start) { |
| 185 | __ lw(a1, MemOperand(frame_pointer(), kStringStartMinusOne)); |
| 186 | __ Addu(a0, current_input_offset(), Operand(-char_size())); |
| 187 | BranchOrBacktrack(on_at_start, eq, a0, Operand(a1)); |
| 188 | } |
| 189 | |
| 190 | |
| 191 | void RegExpMacroAssemblerMIPS::CheckNotAtStart(int cp_offset, |
| 192 | Label* on_not_at_start) { |
| 193 | __ lw(a1, MemOperand(frame_pointer(), kStringStartMinusOne)); |
| 194 | __ Addu(a0, current_input_offset(), |
| 195 | Operand(-char_size() + cp_offset * char_size())); |
| 196 | BranchOrBacktrack(on_not_at_start, ne, a0, Operand(a1)); |
| 197 | } |
| 198 | |
| 199 | |
| 200 | void RegExpMacroAssemblerMIPS::CheckCharacterLT(uc16 limit, Label* on_less) { |
| 201 | BranchOrBacktrack(on_less, lt, current_character(), Operand(limit)); |
| 202 | } |
| 203 | |
| 204 | |
| 205 | void RegExpMacroAssemblerMIPS::CheckGreedyLoop(Label* on_equal) { |
| 206 | Label backtrack_non_equal; |
| 207 | __ lw(a0, MemOperand(backtrack_stackpointer(), 0)); |
| 208 | __ Branch(&backtrack_non_equal, ne, current_input_offset(), Operand(a0)); |
| 209 | __ Addu(backtrack_stackpointer(), |
| 210 | backtrack_stackpointer(), |
| 211 | Operand(kPointerSize)); |
| 212 | __ bind(&backtrack_non_equal); |
| 213 | BranchOrBacktrack(on_equal, eq, current_input_offset(), Operand(a0)); |
| 214 | } |
| 215 | |
| 216 | |
| 217 | void RegExpMacroAssemblerMIPS::CheckNotBackReferenceIgnoreCase( |
Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 218 | int start_reg, bool read_backward, bool unicode, Label* on_no_match) { |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 219 | Label fallthrough; |
| 220 | __ lw(a0, register_location(start_reg)); // Index of start of capture. |
| 221 | __ lw(a1, register_location(start_reg + 1)); // Index of end of capture. |
| 222 | __ Subu(a1, a1, a0); // Length of capture. |
| 223 | |
| 224 | // At this point, the capture registers are either both set or both cleared. |
| 225 | // If the capture length is zero, then the capture is either empty or cleared. |
| 226 | // Fall through in both cases. |
| 227 | __ Branch(&fallthrough, eq, a1, Operand(zero_reg)); |
| 228 | |
| 229 | if (read_backward) { |
| 230 | __ lw(t0, MemOperand(frame_pointer(), kStringStartMinusOne)); |
| 231 | __ Addu(t0, t0, a1); |
| 232 | BranchOrBacktrack(on_no_match, le, current_input_offset(), Operand(t0)); |
| 233 | } else { |
| 234 | __ Addu(t5, a1, current_input_offset()); |
| 235 | // Check that there are enough characters left in the input. |
| 236 | BranchOrBacktrack(on_no_match, gt, t5, Operand(zero_reg)); |
| 237 | } |
| 238 | |
| 239 | if (mode_ == LATIN1) { |
| 240 | Label success; |
| 241 | Label fail; |
| 242 | Label loop_check; |
| 243 | |
| 244 | // a0 - offset of start of capture. |
| 245 | // a1 - length of capture. |
| 246 | __ Addu(a0, a0, Operand(end_of_input_address())); |
| 247 | __ Addu(a2, end_of_input_address(), Operand(current_input_offset())); |
| 248 | if (read_backward) { |
| 249 | __ Subu(a2, a2, Operand(a1)); |
| 250 | } |
| 251 | __ Addu(a1, a0, Operand(a1)); |
| 252 | |
| 253 | // a0 - Address of start of capture. |
| 254 | // a1 - Address of end of capture. |
| 255 | // a2 - Address of current input position. |
| 256 | |
| 257 | Label loop; |
| 258 | __ bind(&loop); |
| 259 | __ lbu(a3, MemOperand(a0, 0)); |
| 260 | __ addiu(a0, a0, char_size()); |
| 261 | __ lbu(t0, MemOperand(a2, 0)); |
| 262 | __ addiu(a2, a2, char_size()); |
| 263 | |
| 264 | __ Branch(&loop_check, eq, t0, Operand(a3)); |
| 265 | |
| 266 | // Mismatch, try case-insensitive match (converting letters to lower-case). |
| 267 | __ Or(a3, a3, Operand(0x20)); // Convert capture character to lower-case. |
| 268 | __ Or(t0, t0, Operand(0x20)); // Also convert input character. |
| 269 | __ Branch(&fail, ne, t0, Operand(a3)); |
| 270 | __ Subu(a3, a3, Operand('a')); |
| 271 | __ Branch(&loop_check, ls, a3, Operand('z' - 'a')); |
| 272 | // Latin-1: Check for values in range [224,254] but not 247. |
| 273 | __ Subu(a3, a3, Operand(224 - 'a')); |
| 274 | // Weren't Latin-1 letters. |
| 275 | __ Branch(&fail, hi, a3, Operand(254 - 224)); |
| 276 | // Check for 247. |
| 277 | __ Branch(&fail, eq, a3, Operand(247 - 224)); |
| 278 | |
| 279 | __ bind(&loop_check); |
| 280 | __ Branch(&loop, lt, a0, Operand(a1)); |
| 281 | __ jmp(&success); |
| 282 | |
| 283 | __ bind(&fail); |
| 284 | GoTo(on_no_match); |
| 285 | |
| 286 | __ bind(&success); |
| 287 | // Compute new value of character position after the matched part. |
| 288 | __ Subu(current_input_offset(), a2, end_of_input_address()); |
| 289 | if (read_backward) { |
| 290 | __ lw(t0, register_location(start_reg)); // Index of start of capture. |
| 291 | __ lw(t5, register_location(start_reg + 1)); // Index of end of capture. |
| 292 | __ Addu(current_input_offset(), current_input_offset(), Operand(t0)); |
| 293 | __ Subu(current_input_offset(), current_input_offset(), Operand(t5)); |
| 294 | } |
| 295 | } else { |
| 296 | DCHECK(mode_ == UC16); |
| 297 | // Put regexp engine registers on stack. |
| 298 | RegList regexp_registers_to_retain = current_input_offset().bit() | |
| 299 | current_character().bit() | backtrack_stackpointer().bit(); |
| 300 | __ MultiPush(regexp_registers_to_retain); |
| 301 | |
| 302 | int argument_count = 4; |
| 303 | __ PrepareCallCFunction(argument_count, a2); |
| 304 | |
| 305 | // a0 - offset of start of capture. |
| 306 | // a1 - length of capture. |
| 307 | |
| 308 | // Put arguments into arguments registers. |
| 309 | // Parameters are |
| 310 | // a0: Address byte_offset1 - Address captured substring's start. |
| 311 | // a1: Address byte_offset2 - Address of current character position. |
| 312 | // a2: size_t byte_length - length of capture in bytes(!). |
Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 313 | // a3: Isolate* isolate or 0 if unicode flag. |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 314 | |
| 315 | // Address of start of capture. |
| 316 | __ Addu(a0, a0, Operand(end_of_input_address())); |
| 317 | // Length of capture. |
| 318 | __ mov(a2, a1); |
| 319 | // Save length in callee-save register for use on return. |
| 320 | __ mov(s3, a1); |
| 321 | // Address of current input position. |
| 322 | __ Addu(a1, current_input_offset(), Operand(end_of_input_address())); |
| 323 | if (read_backward) { |
| 324 | __ Subu(a1, a1, Operand(s3)); |
| 325 | } |
| 326 | // Isolate. |
Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 327 | #ifdef V8_I18N_SUPPORT |
| 328 | if (unicode) { |
| 329 | __ mov(a3, zero_reg); |
| 330 | } else // NOLINT |
| 331 | #endif // V8_I18N_SUPPORT |
| 332 | { |
| 333 | __ li(a3, Operand(ExternalReference::isolate_address(masm_->isolate()))); |
| 334 | } |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 335 | |
| 336 | { |
| 337 | AllowExternalCallThatCantCauseGC scope(masm_); |
| 338 | ExternalReference function = |
| 339 | ExternalReference::re_case_insensitive_compare_uc16(masm_->isolate()); |
| 340 | __ CallCFunction(function, argument_count); |
| 341 | } |
| 342 | |
| 343 | // Restore regexp engine registers. |
| 344 | __ MultiPop(regexp_registers_to_retain); |
| 345 | __ li(code_pointer(), Operand(masm_->CodeObject()), CONSTANT_SIZE); |
| 346 | __ lw(end_of_input_address(), MemOperand(frame_pointer(), kInputEnd)); |
| 347 | |
| 348 | // Check if function returned non-zero for success or zero for failure. |
| 349 | BranchOrBacktrack(on_no_match, eq, v0, Operand(zero_reg)); |
| 350 | // On success, advance position by length of capture. |
| 351 | if (read_backward) { |
| 352 | __ Subu(current_input_offset(), current_input_offset(), Operand(s3)); |
| 353 | } else { |
| 354 | __ Addu(current_input_offset(), current_input_offset(), Operand(s3)); |
| 355 | } |
| 356 | } |
| 357 | |
| 358 | __ bind(&fallthrough); |
| 359 | } |
| 360 | |
| 361 | |
| 362 | void RegExpMacroAssemblerMIPS::CheckNotBackReference(int start_reg, |
| 363 | bool read_backward, |
| 364 | Label* on_no_match) { |
| 365 | Label fallthrough; |
| 366 | Label success; |
| 367 | |
| 368 | // Find length of back-referenced capture. |
| 369 | __ lw(a0, register_location(start_reg)); |
| 370 | __ lw(a1, register_location(start_reg + 1)); |
| 371 | __ Subu(a1, a1, a0); // Length to check. |
| 372 | |
| 373 | // At this point, the capture registers are either both set or both cleared. |
| 374 | // If the capture length is zero, then the capture is either empty or cleared. |
| 375 | // Fall through in both cases. |
| 376 | __ Branch(&fallthrough, le, a1, Operand(zero_reg)); |
| 377 | |
| 378 | if (read_backward) { |
| 379 | __ lw(t0, MemOperand(frame_pointer(), kStringStartMinusOne)); |
| 380 | __ Addu(t0, t0, a1); |
| 381 | BranchOrBacktrack(on_no_match, le, current_input_offset(), Operand(t0)); |
| 382 | } else { |
| 383 | __ Addu(t5, a1, current_input_offset()); |
| 384 | // Check that there are enough characters left in the input. |
| 385 | BranchOrBacktrack(on_no_match, gt, t5, Operand(zero_reg)); |
| 386 | } |
| 387 | |
| 388 | // a0 - offset of start of capture. |
| 389 | // a1 - length of capture. |
| 390 | __ Addu(a0, a0, Operand(end_of_input_address())); |
| 391 | __ Addu(a2, end_of_input_address(), Operand(current_input_offset())); |
| 392 | if (read_backward) { |
| 393 | __ Subu(a2, a2, Operand(a1)); |
| 394 | } |
| 395 | __ Addu(a1, a0, Operand(a1)); |
| 396 | |
| 397 | // a0 - Address of start of capture. |
| 398 | // a1 - Address of end of capture. |
| 399 | // a2 - Address of current input position. |
| 400 | |
| 401 | |
| 402 | Label loop; |
| 403 | __ bind(&loop); |
| 404 | if (mode_ == LATIN1) { |
| 405 | __ lbu(a3, MemOperand(a0, 0)); |
| 406 | __ addiu(a0, a0, char_size()); |
| 407 | __ lbu(t0, MemOperand(a2, 0)); |
| 408 | __ addiu(a2, a2, char_size()); |
| 409 | } else { |
| 410 | DCHECK(mode_ == UC16); |
| 411 | __ lhu(a3, MemOperand(a0, 0)); |
| 412 | __ addiu(a0, a0, char_size()); |
| 413 | __ lhu(t0, MemOperand(a2, 0)); |
| 414 | __ addiu(a2, a2, char_size()); |
| 415 | } |
| 416 | BranchOrBacktrack(on_no_match, ne, a3, Operand(t0)); |
| 417 | __ Branch(&loop, lt, a0, Operand(a1)); |
| 418 | |
| 419 | // Move current character position to position after match. |
| 420 | __ Subu(current_input_offset(), a2, end_of_input_address()); |
| 421 | if (read_backward) { |
| 422 | __ lw(t0, register_location(start_reg)); // Index of start of capture. |
| 423 | __ lw(t5, register_location(start_reg + 1)); // Index of end of capture. |
| 424 | __ Addu(current_input_offset(), current_input_offset(), Operand(t0)); |
| 425 | __ Subu(current_input_offset(), current_input_offset(), Operand(t5)); |
| 426 | } |
| 427 | __ bind(&fallthrough); |
| 428 | } |
| 429 | |
| 430 | |
| 431 | void RegExpMacroAssemblerMIPS::CheckNotCharacter(uint32_t c, |
| 432 | Label* on_not_equal) { |
| 433 | BranchOrBacktrack(on_not_equal, ne, current_character(), Operand(c)); |
| 434 | } |
| 435 | |
| 436 | |
| 437 | void RegExpMacroAssemblerMIPS::CheckCharacterAfterAnd(uint32_t c, |
| 438 | uint32_t mask, |
| 439 | Label* on_equal) { |
| 440 | __ And(a0, current_character(), Operand(mask)); |
| 441 | Operand rhs = (c == 0) ? Operand(zero_reg) : Operand(c); |
| 442 | BranchOrBacktrack(on_equal, eq, a0, rhs); |
| 443 | } |
| 444 | |
| 445 | |
| 446 | void RegExpMacroAssemblerMIPS::CheckNotCharacterAfterAnd(uint32_t c, |
| 447 | uint32_t mask, |
| 448 | Label* on_not_equal) { |
| 449 | __ And(a0, current_character(), Operand(mask)); |
| 450 | Operand rhs = (c == 0) ? Operand(zero_reg) : Operand(c); |
| 451 | BranchOrBacktrack(on_not_equal, ne, a0, rhs); |
| 452 | } |
| 453 | |
| 454 | |
| 455 | void RegExpMacroAssemblerMIPS::CheckNotCharacterAfterMinusAnd( |
| 456 | uc16 c, |
| 457 | uc16 minus, |
| 458 | uc16 mask, |
| 459 | Label* on_not_equal) { |
| 460 | DCHECK(minus < String::kMaxUtf16CodeUnit); |
| 461 | __ Subu(a0, current_character(), Operand(minus)); |
| 462 | __ And(a0, a0, Operand(mask)); |
| 463 | BranchOrBacktrack(on_not_equal, ne, a0, Operand(c)); |
| 464 | } |
| 465 | |
| 466 | |
| 467 | void RegExpMacroAssemblerMIPS::CheckCharacterInRange( |
| 468 | uc16 from, |
| 469 | uc16 to, |
| 470 | Label* on_in_range) { |
| 471 | __ Subu(a0, current_character(), Operand(from)); |
| 472 | // Unsigned lower-or-same condition. |
| 473 | BranchOrBacktrack(on_in_range, ls, a0, Operand(to - from)); |
| 474 | } |
| 475 | |
| 476 | |
| 477 | void RegExpMacroAssemblerMIPS::CheckCharacterNotInRange( |
| 478 | uc16 from, |
| 479 | uc16 to, |
| 480 | Label* on_not_in_range) { |
| 481 | __ Subu(a0, current_character(), Operand(from)); |
| 482 | // Unsigned higher condition. |
| 483 | BranchOrBacktrack(on_not_in_range, hi, a0, Operand(to - from)); |
| 484 | } |
| 485 | |
| 486 | |
| 487 | void RegExpMacroAssemblerMIPS::CheckBitInTable( |
| 488 | Handle<ByteArray> table, |
| 489 | Label* on_bit_set) { |
| 490 | __ li(a0, Operand(table)); |
| 491 | if (mode_ != LATIN1 || kTableMask != String::kMaxOneByteCharCode) { |
| 492 | __ And(a1, current_character(), Operand(kTableSize - 1)); |
| 493 | __ Addu(a0, a0, a1); |
| 494 | } else { |
| 495 | __ Addu(a0, a0, current_character()); |
| 496 | } |
| 497 | |
| 498 | __ lbu(a0, FieldMemOperand(a0, ByteArray::kHeaderSize)); |
| 499 | BranchOrBacktrack(on_bit_set, ne, a0, Operand(zero_reg)); |
| 500 | } |
| 501 | |
| 502 | |
| 503 | bool RegExpMacroAssemblerMIPS::CheckSpecialCharacterClass(uc16 type, |
| 504 | Label* on_no_match) { |
| 505 | // Range checks (c in min..max) are generally implemented by an unsigned |
| 506 | // (c - min) <= (max - min) check. |
| 507 | switch (type) { |
| 508 | case 's': |
| 509 | // Match space-characters. |
| 510 | if (mode_ == LATIN1) { |
| 511 | // One byte space characters are '\t'..'\r', ' ' and \u00a0. |
| 512 | Label success; |
| 513 | __ Branch(&success, eq, current_character(), Operand(' ')); |
| 514 | // Check range 0x09..0x0d. |
| 515 | __ Subu(a0, current_character(), Operand('\t')); |
| 516 | __ Branch(&success, ls, a0, Operand('\r' - '\t')); |
| 517 | // \u00a0 (NBSP). |
| 518 | BranchOrBacktrack(on_no_match, ne, a0, Operand(0x00a0 - '\t')); |
| 519 | __ bind(&success); |
| 520 | return true; |
| 521 | } |
| 522 | return false; |
| 523 | case 'S': |
| 524 | // The emitted code for generic character classes is good enough. |
| 525 | return false; |
| 526 | case 'd': |
| 527 | // Match Latin1 digits ('0'..'9'). |
| 528 | __ Subu(a0, current_character(), Operand('0')); |
| 529 | BranchOrBacktrack(on_no_match, hi, a0, Operand('9' - '0')); |
| 530 | return true; |
| 531 | case 'D': |
| 532 | // Match non Latin1-digits. |
| 533 | __ Subu(a0, current_character(), Operand('0')); |
| 534 | BranchOrBacktrack(on_no_match, ls, a0, Operand('9' - '0')); |
| 535 | return true; |
| 536 | case '.': { |
| 537 | // Match non-newlines (not 0x0a('\n'), 0x0d('\r'), 0x2028 and 0x2029). |
| 538 | __ Xor(a0, current_character(), Operand(0x01)); |
| 539 | // See if current character is '\n'^1 or '\r'^1, i.e., 0x0b or 0x0c. |
| 540 | __ Subu(a0, a0, Operand(0x0b)); |
| 541 | BranchOrBacktrack(on_no_match, ls, a0, Operand(0x0c - 0x0b)); |
| 542 | if (mode_ == UC16) { |
| 543 | // Compare original value to 0x2028 and 0x2029, using the already |
| 544 | // computed (current_char ^ 0x01 - 0x0b). I.e., check for |
| 545 | // 0x201d (0x2028 - 0x0b) or 0x201e. |
| 546 | __ Subu(a0, a0, Operand(0x2028 - 0x0b)); |
| 547 | BranchOrBacktrack(on_no_match, ls, a0, Operand(1)); |
| 548 | } |
| 549 | return true; |
| 550 | } |
| 551 | case 'n': { |
| 552 | // Match newlines (0x0a('\n'), 0x0d('\r'), 0x2028 and 0x2029). |
| 553 | __ Xor(a0, current_character(), Operand(0x01)); |
| 554 | // See if current character is '\n'^1 or '\r'^1, i.e., 0x0b or 0x0c. |
| 555 | __ Subu(a0, a0, Operand(0x0b)); |
| 556 | if (mode_ == LATIN1) { |
| 557 | BranchOrBacktrack(on_no_match, hi, a0, Operand(0x0c - 0x0b)); |
| 558 | } else { |
| 559 | Label done; |
| 560 | BranchOrBacktrack(&done, ls, a0, Operand(0x0c - 0x0b)); |
| 561 | // Compare original value to 0x2028 and 0x2029, using the already |
| 562 | // computed (current_char ^ 0x01 - 0x0b). I.e., check for |
| 563 | // 0x201d (0x2028 - 0x0b) or 0x201e. |
| 564 | __ Subu(a0, a0, Operand(0x2028 - 0x0b)); |
| 565 | BranchOrBacktrack(on_no_match, hi, a0, Operand(1)); |
| 566 | __ bind(&done); |
| 567 | } |
| 568 | return true; |
| 569 | } |
| 570 | case 'w': { |
| 571 | if (mode_ != LATIN1) { |
| 572 | // Table is 256 entries, so all Latin1 characters can be tested. |
| 573 | BranchOrBacktrack(on_no_match, hi, current_character(), Operand('z')); |
| 574 | } |
| 575 | ExternalReference map = ExternalReference::re_word_character_map(); |
| 576 | __ li(a0, Operand(map)); |
| 577 | __ Addu(a0, a0, current_character()); |
| 578 | __ lbu(a0, MemOperand(a0, 0)); |
| 579 | BranchOrBacktrack(on_no_match, eq, a0, Operand(zero_reg)); |
| 580 | return true; |
| 581 | } |
| 582 | case 'W': { |
| 583 | Label done; |
| 584 | if (mode_ != LATIN1) { |
| 585 | // Table is 256 entries, so all Latin1 characters can be tested. |
| 586 | __ Branch(&done, hi, current_character(), Operand('z')); |
| 587 | } |
| 588 | ExternalReference map = ExternalReference::re_word_character_map(); |
| 589 | __ li(a0, Operand(map)); |
| 590 | __ Addu(a0, a0, current_character()); |
| 591 | __ lbu(a0, MemOperand(a0, 0)); |
| 592 | BranchOrBacktrack(on_no_match, ne, a0, Operand(zero_reg)); |
| 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 RegExpMacroAssemblerMIPS::Fail() { |
| 609 | __ li(v0, Operand(FAILURE)); |
| 610 | __ jmp(&exit_label_); |
| 611 | } |
| 612 | |
| 613 | |
| 614 | Handle<HeapObject> RegExpMacroAssemblerMIPS::GetCode(Handle<String> source) { |
| 615 | Label return_v0; |
| 616 | if (masm_->has_exception()) { |
| 617 | // If the code gets corrupted due to long regular expressions and lack of |
| 618 | // space on trampolines, an internal exception flag is set. If this case |
| 619 | // is detected, we will jump into exit sequence right away. |
| 620 | __ bind_to(&entry_label_, internal_failure_label_.pos()); |
| 621 | } else { |
| 622 | // Finalize code - write the entry point code now we know how many |
| 623 | // registers we need. |
| 624 | |
| 625 | // Entry code: |
| 626 | __ bind(&entry_label_); |
| 627 | |
| 628 | // Tell the system that we have a stack frame. Because the type is MANUAL, |
| 629 | // no is generated. |
| 630 | FrameScope scope(masm_, StackFrame::MANUAL); |
| 631 | |
| 632 | // Actually emit code to start a new stack frame. |
| 633 | // Push arguments |
| 634 | // Save callee-save registers. |
| 635 | // Start new stack frame. |
| 636 | // Store link register in existing stack-cell. |
| 637 | // Order here should correspond to order of offset constants in header file. |
| 638 | RegList registers_to_retain = s0.bit() | s1.bit() | s2.bit() | |
| 639 | s3.bit() | s4.bit() | s5.bit() | s6.bit() | s7.bit() | fp.bit(); |
| 640 | RegList argument_registers = a0.bit() | a1.bit() | a2.bit() | a3.bit(); |
| 641 | __ MultiPush(argument_registers | registers_to_retain | ra.bit()); |
| 642 | // Set frame pointer in space for it if this is not a direct call |
| 643 | // from generated code. |
| 644 | __ Addu(frame_pointer(), sp, Operand(4 * kPointerSize)); |
| 645 | __ mov(a0, zero_reg); |
| 646 | __ push(a0); // Make room for success counter and initialize it to 0. |
| 647 | __ push(a0); // Make room for "string start - 1" constant. |
| 648 | |
| 649 | // Check if we have space on the stack for registers. |
| 650 | Label stack_limit_hit; |
| 651 | Label stack_ok; |
| 652 | |
| 653 | ExternalReference stack_limit = |
| 654 | ExternalReference::address_of_stack_limit(masm_->isolate()); |
| 655 | __ li(a0, Operand(stack_limit)); |
| 656 | __ lw(a0, MemOperand(a0)); |
| 657 | __ Subu(a0, sp, a0); |
| 658 | // Handle it if the stack pointer is already below the stack limit. |
| 659 | __ Branch(&stack_limit_hit, le, a0, Operand(zero_reg)); |
| 660 | // Check if there is room for the variable number of registers above |
| 661 | // the stack limit. |
| 662 | __ Branch(&stack_ok, hs, a0, Operand(num_registers_ * kPointerSize)); |
| 663 | // Exit with OutOfMemory exception. There is not enough space on the stack |
| 664 | // for our working registers. |
| 665 | __ li(v0, Operand(EXCEPTION)); |
| 666 | __ jmp(&return_v0); |
| 667 | |
| 668 | __ bind(&stack_limit_hit); |
| 669 | CallCheckStackGuardState(a0); |
| 670 | // If returned value is non-zero, we exit with the returned value as result. |
| 671 | __ Branch(&return_v0, ne, v0, Operand(zero_reg)); |
| 672 | |
| 673 | __ bind(&stack_ok); |
| 674 | // Allocate space on stack for registers. |
| 675 | __ Subu(sp, sp, Operand(num_registers_ * kPointerSize)); |
| 676 | // Load string end. |
| 677 | __ lw(end_of_input_address(), MemOperand(frame_pointer(), kInputEnd)); |
| 678 | // Load input start. |
| 679 | __ lw(a0, MemOperand(frame_pointer(), kInputStart)); |
| 680 | // Find negative length (offset of start relative to end). |
| 681 | __ Subu(current_input_offset(), a0, end_of_input_address()); |
| 682 | // Set a0 to address of char before start of the input string |
| 683 | // (effectively string position -1). |
| 684 | __ lw(a1, MemOperand(frame_pointer(), kStartIndex)); |
| 685 | __ Subu(a0, current_input_offset(), Operand(char_size())); |
| 686 | __ sll(t5, a1, (mode_ == UC16) ? 1 : 0); |
| 687 | __ Subu(a0, a0, t5); |
| 688 | // Store this value in a local variable, for use when clearing |
| 689 | // position registers. |
| 690 | __ sw(a0, MemOperand(frame_pointer(), kStringStartMinusOne)); |
| 691 | |
| 692 | // Initialize code pointer register |
| 693 | __ li(code_pointer(), Operand(masm_->CodeObject()), CONSTANT_SIZE); |
| 694 | |
| 695 | Label load_char_start_regexp, start_regexp; |
| 696 | // Load newline if index is at start, previous character otherwise. |
| 697 | __ Branch(&load_char_start_regexp, ne, a1, Operand(zero_reg)); |
| 698 | __ li(current_character(), Operand('\n')); |
| 699 | __ jmp(&start_regexp); |
| 700 | |
| 701 | // Global regexp restarts matching here. |
| 702 | __ bind(&load_char_start_regexp); |
| 703 | // Load previous char as initial value of current character register. |
| 704 | LoadCurrentCharacterUnchecked(-1, 1); |
| 705 | __ bind(&start_regexp); |
| 706 | |
| 707 | // Initialize on-stack registers. |
| 708 | if (num_saved_registers_ > 0) { // Always is, if generated from a regexp. |
| 709 | // Fill saved registers with initial value = start offset - 1. |
| 710 | if (num_saved_registers_ > 8) { |
| 711 | // Address of register 0. |
| 712 | __ Addu(a1, frame_pointer(), Operand(kRegisterZero)); |
| 713 | __ li(a2, Operand(num_saved_registers_)); |
| 714 | Label init_loop; |
| 715 | __ bind(&init_loop); |
| 716 | __ sw(a0, MemOperand(a1)); |
| 717 | __ Addu(a1, a1, Operand(-kPointerSize)); |
| 718 | __ Subu(a2, a2, Operand(1)); |
| 719 | __ Branch(&init_loop, ne, a2, Operand(zero_reg)); |
| 720 | } else { |
| 721 | for (int i = 0; i < num_saved_registers_; i++) { |
| 722 | __ sw(a0, register_location(i)); |
| 723 | } |
| 724 | } |
| 725 | } |
| 726 | |
| 727 | // Initialize backtrack stack pointer. |
| 728 | __ lw(backtrack_stackpointer(), MemOperand(frame_pointer(), kStackHighEnd)); |
| 729 | |
| 730 | __ jmp(&start_label_); |
| 731 | |
| 732 | |
| 733 | // Exit code: |
| 734 | if (success_label_.is_linked()) { |
| 735 | // Save captures when successful. |
| 736 | __ bind(&success_label_); |
| 737 | if (num_saved_registers_ > 0) { |
| 738 | // Copy captures to output. |
| 739 | __ lw(a1, MemOperand(frame_pointer(), kInputStart)); |
| 740 | __ lw(a0, MemOperand(frame_pointer(), kRegisterOutput)); |
| 741 | __ lw(a2, MemOperand(frame_pointer(), kStartIndex)); |
| 742 | __ Subu(a1, end_of_input_address(), a1); |
| 743 | // a1 is length of input in bytes. |
| 744 | if (mode_ == UC16) { |
| 745 | __ srl(a1, a1, 1); |
| 746 | } |
| 747 | // a1 is length of input in characters. |
| 748 | __ Addu(a1, a1, Operand(a2)); |
| 749 | // a1 is length of string in characters. |
| 750 | |
| 751 | DCHECK_EQ(0, num_saved_registers_ % 2); |
| 752 | // Always an even number of capture registers. This allows us to |
| 753 | // unroll the loop once to add an operation between a load of a register |
| 754 | // and the following use of that register. |
| 755 | for (int i = 0; i < num_saved_registers_; i += 2) { |
| 756 | __ lw(a2, register_location(i)); |
| 757 | __ lw(a3, register_location(i + 1)); |
| 758 | if (i == 0 && global_with_zero_length_check()) { |
| 759 | // Keep capture start in a4 for the zero-length check later. |
| 760 | __ mov(t7, a2); |
| 761 | } |
| 762 | if (mode_ == UC16) { |
| 763 | __ sra(a2, a2, 1); |
| 764 | __ Addu(a2, a2, a1); |
| 765 | __ sra(a3, a3, 1); |
| 766 | __ Addu(a3, a3, a1); |
| 767 | } else { |
| 768 | __ Addu(a2, a1, Operand(a2)); |
| 769 | __ Addu(a3, a1, Operand(a3)); |
| 770 | } |
| 771 | __ sw(a2, MemOperand(a0)); |
| 772 | __ Addu(a0, a0, kPointerSize); |
| 773 | __ sw(a3, MemOperand(a0)); |
| 774 | __ Addu(a0, a0, kPointerSize); |
| 775 | } |
| 776 | } |
| 777 | |
| 778 | if (global()) { |
| 779 | // Restart matching if the regular expression is flagged as global. |
| 780 | __ lw(a0, MemOperand(frame_pointer(), kSuccessfulCaptures)); |
| 781 | __ lw(a1, MemOperand(frame_pointer(), kNumOutputRegisters)); |
| 782 | __ lw(a2, MemOperand(frame_pointer(), kRegisterOutput)); |
| 783 | // Increment success counter. |
| 784 | __ Addu(a0, a0, 1); |
| 785 | __ sw(a0, MemOperand(frame_pointer(), kSuccessfulCaptures)); |
| 786 | // Capture results have been stored, so the number of remaining global |
| 787 | // output registers is reduced by the number of stored captures. |
| 788 | __ Subu(a1, a1, num_saved_registers_); |
| 789 | // Check whether we have enough room for another set of capture results. |
| 790 | __ mov(v0, a0); |
| 791 | __ Branch(&return_v0, lt, a1, Operand(num_saved_registers_)); |
| 792 | |
| 793 | __ sw(a1, MemOperand(frame_pointer(), kNumOutputRegisters)); |
| 794 | // Advance the location for output. |
| 795 | __ Addu(a2, a2, num_saved_registers_ * kPointerSize); |
| 796 | __ sw(a2, MemOperand(frame_pointer(), kRegisterOutput)); |
| 797 | |
| 798 | // Prepare a0 to initialize registers with its value in the next run. |
| 799 | __ lw(a0, MemOperand(frame_pointer(), kStringStartMinusOne)); |
| 800 | |
| 801 | if (global_with_zero_length_check()) { |
| 802 | // Special case for zero-length matches. |
| 803 | // t7: capture start index |
| 804 | // Not a zero-length match, restart. |
| 805 | __ Branch( |
| 806 | &load_char_start_regexp, ne, current_input_offset(), Operand(t7)); |
| 807 | // Offset from the end is zero if we already reached the end. |
| 808 | __ Branch(&exit_label_, eq, current_input_offset(), |
| 809 | Operand(zero_reg)); |
| 810 | // Advance current position after a zero-length match. |
Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 811 | Label advance; |
| 812 | __ bind(&advance); |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 813 | __ Addu(current_input_offset(), |
| 814 | current_input_offset(), |
| 815 | Operand((mode_ == UC16) ? 2 : 1)); |
Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 816 | if (global_unicode()) CheckNotInSurrogatePair(0, &advance); |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 817 | } |
| 818 | |
| 819 | __ Branch(&load_char_start_regexp); |
| 820 | } else { |
| 821 | __ li(v0, Operand(SUCCESS)); |
| 822 | } |
| 823 | } |
| 824 | // Exit and return v0. |
| 825 | __ bind(&exit_label_); |
| 826 | if (global()) { |
| 827 | __ lw(v0, MemOperand(frame_pointer(), kSuccessfulCaptures)); |
| 828 | } |
| 829 | |
| 830 | __ bind(&return_v0); |
| 831 | // Skip sp past regexp registers and local variables.. |
| 832 | __ mov(sp, frame_pointer()); |
| 833 | // Restore registers s0..s7 and return (restoring ra to pc). |
| 834 | __ MultiPop(registers_to_retain | ra.bit()); |
| 835 | __ Ret(); |
| 836 | |
| 837 | // Backtrack code (branch target for conditional backtracks). |
| 838 | if (backtrack_label_.is_linked()) { |
| 839 | __ bind(&backtrack_label_); |
| 840 | Backtrack(); |
| 841 | } |
| 842 | |
| 843 | Label exit_with_exception; |
| 844 | |
| 845 | // Preempt-code. |
| 846 | if (check_preempt_label_.is_linked()) { |
| 847 | SafeCallTarget(&check_preempt_label_); |
| 848 | // Put regexp engine registers on stack. |
| 849 | RegList regexp_registers_to_retain = current_input_offset().bit() | |
| 850 | current_character().bit() | backtrack_stackpointer().bit(); |
| 851 | __ MultiPush(regexp_registers_to_retain); |
| 852 | CallCheckStackGuardState(a0); |
| 853 | __ MultiPop(regexp_registers_to_retain); |
| 854 | // If returning non-zero, we should end execution with the given |
| 855 | // result as return value. |
| 856 | __ Branch(&return_v0, ne, v0, Operand(zero_reg)); |
| 857 | |
| 858 | // String might have moved: Reload end of string from frame. |
| 859 | __ lw(end_of_input_address(), MemOperand(frame_pointer(), kInputEnd)); |
| 860 | __ li(code_pointer(), Operand(masm_->CodeObject()), CONSTANT_SIZE); |
| 861 | SafeReturn(); |
| 862 | } |
| 863 | |
| 864 | // Backtrack stack overflow code. |
| 865 | if (stack_overflow_label_.is_linked()) { |
| 866 | SafeCallTarget(&stack_overflow_label_); |
| 867 | // Reached if the backtrack-stack limit has been hit. |
| 868 | // Put regexp engine registers on stack first. |
| 869 | RegList regexp_registers = current_input_offset().bit() | |
| 870 | current_character().bit(); |
| 871 | __ MultiPush(regexp_registers); |
| 872 | Label grow_failed; |
| 873 | // Call GrowStack(backtrack_stackpointer(), &stack_base) |
| 874 | static const int num_arguments = 3; |
| 875 | __ PrepareCallCFunction(num_arguments, a0); |
| 876 | __ mov(a0, backtrack_stackpointer()); |
| 877 | __ Addu(a1, frame_pointer(), Operand(kStackHighEnd)); |
| 878 | __ li(a2, Operand(ExternalReference::isolate_address(masm_->isolate()))); |
| 879 | ExternalReference grow_stack = |
| 880 | ExternalReference::re_grow_stack(masm_->isolate()); |
| 881 | __ CallCFunction(grow_stack, num_arguments); |
| 882 | // Restore regexp registers. |
| 883 | __ MultiPop(regexp_registers); |
| 884 | // If return NULL, we have failed to grow the stack, and |
| 885 | // must exit with a stack-overflow exception. |
| 886 | __ Branch(&exit_with_exception, eq, v0, Operand(zero_reg)); |
| 887 | // Otherwise use return value as new stack pointer. |
| 888 | __ mov(backtrack_stackpointer(), v0); |
| 889 | // Restore saved registers and continue. |
| 890 | __ li(code_pointer(), Operand(masm_->CodeObject()), CONSTANT_SIZE); |
| 891 | __ lw(end_of_input_address(), MemOperand(frame_pointer(), kInputEnd)); |
| 892 | SafeReturn(); |
| 893 | } |
| 894 | |
| 895 | if (exit_with_exception.is_linked()) { |
| 896 | // If any of the code above needed to exit with an exception. |
| 897 | __ bind(&exit_with_exception); |
| 898 | // Exit with Result EXCEPTION(-1) to signal thrown exception. |
| 899 | __ li(v0, Operand(EXCEPTION)); |
| 900 | __ jmp(&return_v0); |
| 901 | } |
| 902 | } |
| 903 | |
| 904 | CodeDesc code_desc; |
| 905 | masm_->GetCode(&code_desc); |
| 906 | Handle<Code> code = isolate()->factory()->NewCode( |
| 907 | code_desc, Code::ComputeFlags(Code::REGEXP), masm_->CodeObject()); |
Ben Murdoch | da12d29 | 2016-06-02 14:46:10 +0100 | [diff] [blame] | 908 | LOG(masm_->isolate(), |
| 909 | RegExpCodeCreateEvent(AbstractCode::cast(*code), *source)); |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 910 | return Handle<HeapObject>::cast(code); |
| 911 | } |
| 912 | |
| 913 | |
| 914 | void RegExpMacroAssemblerMIPS::GoTo(Label* to) { |
| 915 | if (to == NULL) { |
| 916 | Backtrack(); |
| 917 | return; |
| 918 | } |
| 919 | __ jmp(to); |
| 920 | return; |
| 921 | } |
| 922 | |
| 923 | |
| 924 | void RegExpMacroAssemblerMIPS::IfRegisterGE(int reg, |
| 925 | int comparand, |
| 926 | Label* if_ge) { |
| 927 | __ lw(a0, register_location(reg)); |
| 928 | BranchOrBacktrack(if_ge, ge, a0, Operand(comparand)); |
| 929 | } |
| 930 | |
| 931 | |
| 932 | void RegExpMacroAssemblerMIPS::IfRegisterLT(int reg, |
| 933 | int comparand, |
| 934 | Label* if_lt) { |
| 935 | __ lw(a0, register_location(reg)); |
| 936 | BranchOrBacktrack(if_lt, lt, a0, Operand(comparand)); |
| 937 | } |
| 938 | |
| 939 | |
| 940 | void RegExpMacroAssemblerMIPS::IfRegisterEqPos(int reg, |
| 941 | Label* if_eq) { |
| 942 | __ lw(a0, register_location(reg)); |
| 943 | BranchOrBacktrack(if_eq, eq, a0, Operand(current_input_offset())); |
| 944 | } |
| 945 | |
| 946 | |
| 947 | RegExpMacroAssembler::IrregexpImplementation |
| 948 | RegExpMacroAssemblerMIPS::Implementation() { |
| 949 | return kMIPSImplementation; |
| 950 | } |
| 951 | |
| 952 | |
| 953 | void RegExpMacroAssemblerMIPS::LoadCurrentCharacter(int cp_offset, |
| 954 | Label* on_end_of_input, |
| 955 | bool check_bounds, |
| 956 | int characters) { |
| 957 | DCHECK(cp_offset < (1<<30)); // Be sane! (And ensure negation works). |
| 958 | if (check_bounds) { |
| 959 | if (cp_offset >= 0) { |
| 960 | CheckPosition(cp_offset + characters - 1, on_end_of_input); |
| 961 | } else { |
| 962 | CheckPosition(cp_offset, on_end_of_input); |
| 963 | } |
| 964 | } |
| 965 | LoadCurrentCharacterUnchecked(cp_offset, characters); |
| 966 | } |
| 967 | |
| 968 | |
| 969 | void RegExpMacroAssemblerMIPS::PopCurrentPosition() { |
| 970 | Pop(current_input_offset()); |
| 971 | } |
| 972 | |
| 973 | |
| 974 | void RegExpMacroAssemblerMIPS::PopRegister(int register_index) { |
| 975 | Pop(a0); |
| 976 | __ sw(a0, register_location(register_index)); |
| 977 | } |
| 978 | |
| 979 | |
| 980 | void RegExpMacroAssemblerMIPS::PushBacktrack(Label* label) { |
| 981 | if (label->is_bound()) { |
| 982 | int target = label->pos(); |
| 983 | __ li(a0, Operand(target + Code::kHeaderSize - kHeapObjectTag)); |
| 984 | } else { |
| 985 | Assembler::BlockTrampolinePoolScope block_trampoline_pool(masm_); |
| 986 | Label after_constant; |
| 987 | __ Branch(&after_constant); |
| 988 | int offset = masm_->pc_offset(); |
| 989 | int cp_offset = offset + Code::kHeaderSize - kHeapObjectTag; |
| 990 | __ emit(0); |
| 991 | masm_->label_at_put(label, offset); |
| 992 | __ bind(&after_constant); |
| 993 | if (is_int16(cp_offset)) { |
| 994 | __ lw(a0, MemOperand(code_pointer(), cp_offset)); |
| 995 | } else { |
| 996 | __ Addu(a0, code_pointer(), cp_offset); |
| 997 | __ lw(a0, MemOperand(a0, 0)); |
| 998 | } |
| 999 | } |
| 1000 | Push(a0); |
| 1001 | CheckStackLimit(); |
| 1002 | } |
| 1003 | |
| 1004 | |
| 1005 | void RegExpMacroAssemblerMIPS::PushCurrentPosition() { |
| 1006 | Push(current_input_offset()); |
| 1007 | } |
| 1008 | |
| 1009 | |
| 1010 | void RegExpMacroAssemblerMIPS::PushRegister(int register_index, |
| 1011 | StackCheckFlag check_stack_limit) { |
| 1012 | __ lw(a0, register_location(register_index)); |
| 1013 | Push(a0); |
| 1014 | if (check_stack_limit) CheckStackLimit(); |
| 1015 | } |
| 1016 | |
| 1017 | |
| 1018 | void RegExpMacroAssemblerMIPS::ReadCurrentPositionFromRegister(int reg) { |
| 1019 | __ lw(current_input_offset(), register_location(reg)); |
| 1020 | } |
| 1021 | |
| 1022 | |
| 1023 | void RegExpMacroAssemblerMIPS::ReadStackPointerFromRegister(int reg) { |
| 1024 | __ lw(backtrack_stackpointer(), register_location(reg)); |
| 1025 | __ lw(a0, MemOperand(frame_pointer(), kStackHighEnd)); |
| 1026 | __ Addu(backtrack_stackpointer(), backtrack_stackpointer(), Operand(a0)); |
| 1027 | } |
| 1028 | |
| 1029 | |
| 1030 | void RegExpMacroAssemblerMIPS::SetCurrentPositionFromEnd(int by) { |
| 1031 | Label after_position; |
| 1032 | __ Branch(&after_position, |
| 1033 | ge, |
| 1034 | current_input_offset(), |
| 1035 | Operand(-by * char_size())); |
| 1036 | __ li(current_input_offset(), -by * char_size()); |
| 1037 | // On RegExp code entry (where this operation is used), the character before |
| 1038 | // the current position is expected to be already loaded. |
| 1039 | // We have advanced the position, so it's safe to read backwards. |
| 1040 | LoadCurrentCharacterUnchecked(-1, 1); |
| 1041 | __ bind(&after_position); |
| 1042 | } |
| 1043 | |
| 1044 | |
| 1045 | void RegExpMacroAssemblerMIPS::SetRegister(int register_index, int to) { |
| 1046 | DCHECK(register_index >= num_saved_registers_); // Reserved for positions! |
| 1047 | __ li(a0, Operand(to)); |
| 1048 | __ sw(a0, register_location(register_index)); |
| 1049 | } |
| 1050 | |
| 1051 | |
| 1052 | bool RegExpMacroAssemblerMIPS::Succeed() { |
| 1053 | __ jmp(&success_label_); |
| 1054 | return global(); |
| 1055 | } |
| 1056 | |
| 1057 | |
| 1058 | void RegExpMacroAssemblerMIPS::WriteCurrentPositionToRegister(int reg, |
| 1059 | int cp_offset) { |
| 1060 | if (cp_offset == 0) { |
| 1061 | __ sw(current_input_offset(), register_location(reg)); |
| 1062 | } else { |
| 1063 | __ Addu(a0, current_input_offset(), Operand(cp_offset * char_size())); |
| 1064 | __ sw(a0, register_location(reg)); |
| 1065 | } |
| 1066 | } |
| 1067 | |
| 1068 | |
| 1069 | void RegExpMacroAssemblerMIPS::ClearRegisters(int reg_from, int reg_to) { |
| 1070 | DCHECK(reg_from <= reg_to); |
| 1071 | __ lw(a0, MemOperand(frame_pointer(), kStringStartMinusOne)); |
| 1072 | for (int reg = reg_from; reg <= reg_to; reg++) { |
| 1073 | __ sw(a0, register_location(reg)); |
| 1074 | } |
| 1075 | } |
| 1076 | |
| 1077 | |
| 1078 | void RegExpMacroAssemblerMIPS::WriteStackPointerToRegister(int reg) { |
| 1079 | __ lw(a1, MemOperand(frame_pointer(), kStackHighEnd)); |
| 1080 | __ Subu(a0, backtrack_stackpointer(), a1); |
| 1081 | __ sw(a0, register_location(reg)); |
| 1082 | } |
| 1083 | |
| 1084 | |
| 1085 | bool RegExpMacroAssemblerMIPS::CanReadUnaligned() { |
| 1086 | return false; |
| 1087 | } |
| 1088 | |
| 1089 | |
| 1090 | // Private methods: |
| 1091 | |
| 1092 | void RegExpMacroAssemblerMIPS::CallCheckStackGuardState(Register scratch) { |
| 1093 | int stack_alignment = base::OS::ActivationFrameAlignment(); |
| 1094 | |
| 1095 | // Align the stack pointer and save the original sp value on the stack. |
| 1096 | __ mov(scratch, sp); |
| 1097 | __ Subu(sp, sp, Operand(kPointerSize)); |
| 1098 | DCHECK(base::bits::IsPowerOfTwo32(stack_alignment)); |
| 1099 | __ And(sp, sp, Operand(-stack_alignment)); |
| 1100 | __ sw(scratch, MemOperand(sp)); |
| 1101 | |
| 1102 | __ mov(a2, frame_pointer()); |
| 1103 | // Code* of self. |
| 1104 | __ li(a1, Operand(masm_->CodeObject()), CONSTANT_SIZE); |
| 1105 | |
| 1106 | // We need to make room for the return address on the stack. |
| 1107 | DCHECK(IsAligned(stack_alignment, kPointerSize)); |
| 1108 | __ Subu(sp, sp, Operand(stack_alignment)); |
| 1109 | |
| 1110 | // Stack pointer now points to cell where return address is to be written. |
| 1111 | // Arguments are in registers, meaning we teat the return address as |
| 1112 | // argument 5. Since DirectCEntryStub will handleallocating space for the C |
| 1113 | // argument slots, we don't need to care about that here. This is how the |
| 1114 | // stack will look (sp meaning the value of sp at this moment): |
| 1115 | // [sp + 3] - empty slot if needed for alignment. |
| 1116 | // [sp + 2] - saved sp. |
| 1117 | // [sp + 1] - second word reserved for return value. |
| 1118 | // [sp + 0] - first word reserved for return value. |
| 1119 | |
| 1120 | // a0 will point to the return address, placed by DirectCEntry. |
| 1121 | __ mov(a0, sp); |
| 1122 | |
| 1123 | ExternalReference stack_guard_check = |
| 1124 | ExternalReference::re_check_stack_guard_state(masm_->isolate()); |
| 1125 | __ li(t9, Operand(stack_guard_check)); |
| 1126 | DirectCEntryStub stub(isolate()); |
| 1127 | stub.GenerateCall(masm_, t9); |
| 1128 | |
| 1129 | // DirectCEntryStub allocated space for the C argument slots so we have to |
| 1130 | // drop them with the return address from the stack with loading saved sp. |
| 1131 | // At this point stack must look: |
| 1132 | // [sp + 7] - empty slot if needed for alignment. |
| 1133 | // [sp + 6] - saved sp. |
| 1134 | // [sp + 5] - second word reserved for return value. |
| 1135 | // [sp + 4] - first word reserved for return value. |
| 1136 | // [sp + 3] - C argument slot. |
| 1137 | // [sp + 2] - C argument slot. |
| 1138 | // [sp + 1] - C argument slot. |
| 1139 | // [sp + 0] - C argument slot. |
| 1140 | __ lw(sp, MemOperand(sp, stack_alignment + kCArgsSlotsSize)); |
| 1141 | |
| 1142 | __ li(code_pointer(), Operand(masm_->CodeObject())); |
| 1143 | } |
| 1144 | |
| 1145 | |
| 1146 | // Helper function for reading a value out of a stack frame. |
| 1147 | template <typename T> |
| 1148 | static T& frame_entry(Address re_frame, int frame_offset) { |
| 1149 | return reinterpret_cast<T&>(Memory::int32_at(re_frame + frame_offset)); |
| 1150 | } |
| 1151 | |
| 1152 | |
| 1153 | template <typename T> |
| 1154 | static T* frame_entry_address(Address re_frame, int frame_offset) { |
| 1155 | return reinterpret_cast<T*>(re_frame + frame_offset); |
| 1156 | } |
| 1157 | |
| 1158 | |
| 1159 | int RegExpMacroAssemblerMIPS::CheckStackGuardState(Address* return_address, |
| 1160 | Code* re_code, |
| 1161 | Address re_frame) { |
| 1162 | return NativeRegExpMacroAssembler::CheckStackGuardState( |
| 1163 | frame_entry<Isolate*>(re_frame, kIsolate), |
| 1164 | frame_entry<int>(re_frame, kStartIndex), |
| 1165 | frame_entry<int>(re_frame, kDirectCall) == 1, return_address, re_code, |
| 1166 | frame_entry_address<String*>(re_frame, kInputString), |
| 1167 | frame_entry_address<const byte*>(re_frame, kInputStart), |
| 1168 | frame_entry_address<const byte*>(re_frame, kInputEnd)); |
| 1169 | } |
| 1170 | |
| 1171 | |
| 1172 | MemOperand RegExpMacroAssemblerMIPS::register_location(int register_index) { |
| 1173 | DCHECK(register_index < (1<<30)); |
| 1174 | if (num_registers_ <= register_index) { |
| 1175 | num_registers_ = register_index + 1; |
| 1176 | } |
| 1177 | return MemOperand(frame_pointer(), |
| 1178 | kRegisterZero - register_index * kPointerSize); |
| 1179 | } |
| 1180 | |
| 1181 | |
| 1182 | void RegExpMacroAssemblerMIPS::CheckPosition(int cp_offset, |
| 1183 | Label* on_outside_input) { |
| 1184 | if (cp_offset >= 0) { |
| 1185 | BranchOrBacktrack(on_outside_input, ge, current_input_offset(), |
| 1186 | Operand(-cp_offset * char_size())); |
| 1187 | } else { |
| 1188 | __ lw(a1, MemOperand(frame_pointer(), kStringStartMinusOne)); |
| 1189 | __ Addu(a0, current_input_offset(), Operand(cp_offset * char_size())); |
| 1190 | BranchOrBacktrack(on_outside_input, le, a0, Operand(a1)); |
| 1191 | } |
| 1192 | } |
| 1193 | |
| 1194 | |
| 1195 | void RegExpMacroAssemblerMIPS::BranchOrBacktrack(Label* to, |
| 1196 | Condition condition, |
| 1197 | Register rs, |
| 1198 | const Operand& rt) { |
| 1199 | if (condition == al) { // Unconditional. |
| 1200 | if (to == NULL) { |
| 1201 | Backtrack(); |
| 1202 | return; |
| 1203 | } |
| 1204 | __ jmp(to); |
| 1205 | return; |
| 1206 | } |
| 1207 | if (to == NULL) { |
| 1208 | __ Branch(&backtrack_label_, condition, rs, rt); |
| 1209 | return; |
| 1210 | } |
| 1211 | __ Branch(to, condition, rs, rt); |
| 1212 | } |
| 1213 | |
| 1214 | |
| 1215 | void RegExpMacroAssemblerMIPS::SafeCall(Label* to, |
| 1216 | Condition cond, |
| 1217 | Register rs, |
| 1218 | const Operand& rt) { |
| 1219 | __ BranchAndLink(to, cond, rs, rt); |
| 1220 | } |
| 1221 | |
| 1222 | |
| 1223 | void RegExpMacroAssemblerMIPS::SafeReturn() { |
| 1224 | __ pop(ra); |
| 1225 | __ Addu(t5, ra, Operand(masm_->CodeObject())); |
| 1226 | __ Jump(t5); |
| 1227 | } |
| 1228 | |
| 1229 | |
| 1230 | void RegExpMacroAssemblerMIPS::SafeCallTarget(Label* name) { |
| 1231 | __ bind(name); |
| 1232 | __ Subu(ra, ra, Operand(masm_->CodeObject())); |
| 1233 | __ push(ra); |
| 1234 | } |
| 1235 | |
| 1236 | |
| 1237 | void RegExpMacroAssemblerMIPS::Push(Register source) { |
| 1238 | DCHECK(!source.is(backtrack_stackpointer())); |
| 1239 | __ Addu(backtrack_stackpointer(), |
| 1240 | backtrack_stackpointer(), |
| 1241 | Operand(-kPointerSize)); |
| 1242 | __ sw(source, MemOperand(backtrack_stackpointer())); |
| 1243 | } |
| 1244 | |
| 1245 | |
| 1246 | void RegExpMacroAssemblerMIPS::Pop(Register target) { |
| 1247 | DCHECK(!target.is(backtrack_stackpointer())); |
| 1248 | __ lw(target, MemOperand(backtrack_stackpointer())); |
| 1249 | __ Addu(backtrack_stackpointer(), backtrack_stackpointer(), kPointerSize); |
| 1250 | } |
| 1251 | |
| 1252 | |
| 1253 | void RegExpMacroAssemblerMIPS::CheckPreemption() { |
| 1254 | // Check for preemption. |
| 1255 | ExternalReference stack_limit = |
| 1256 | ExternalReference::address_of_stack_limit(masm_->isolate()); |
| 1257 | __ li(a0, Operand(stack_limit)); |
| 1258 | __ lw(a0, MemOperand(a0)); |
| 1259 | SafeCall(&check_preempt_label_, ls, sp, Operand(a0)); |
| 1260 | } |
| 1261 | |
| 1262 | |
| 1263 | void RegExpMacroAssemblerMIPS::CheckStackLimit() { |
| 1264 | ExternalReference stack_limit = |
| 1265 | ExternalReference::address_of_regexp_stack_limit(masm_->isolate()); |
| 1266 | |
| 1267 | __ li(a0, Operand(stack_limit)); |
| 1268 | __ lw(a0, MemOperand(a0)); |
| 1269 | SafeCall(&stack_overflow_label_, ls, backtrack_stackpointer(), Operand(a0)); |
| 1270 | } |
| 1271 | |
| 1272 | |
| 1273 | void RegExpMacroAssemblerMIPS::LoadCurrentCharacterUnchecked(int cp_offset, |
| 1274 | int characters) { |
| 1275 | Register offset = current_input_offset(); |
| 1276 | if (cp_offset != 0) { |
| 1277 | // t7 is not being used to store the capture start index at this point. |
| 1278 | __ Addu(t7, current_input_offset(), Operand(cp_offset * char_size())); |
| 1279 | offset = t7; |
| 1280 | } |
| 1281 | // We assume that we cannot do unaligned loads on MIPS, so this function |
| 1282 | // must only be used to load a single character at a time. |
| 1283 | DCHECK(characters == 1); |
| 1284 | __ Addu(t5, end_of_input_address(), Operand(offset)); |
| 1285 | if (mode_ == LATIN1) { |
| 1286 | __ lbu(current_character(), MemOperand(t5, 0)); |
| 1287 | } else { |
| 1288 | DCHECK(mode_ == UC16); |
| 1289 | __ lhu(current_character(), MemOperand(t5, 0)); |
| 1290 | } |
| 1291 | } |
| 1292 | |
| 1293 | |
| 1294 | #undef __ |
| 1295 | |
| 1296 | #endif // V8_INTERPRETED_REGEXP |
| 1297 | |
| 1298 | } // namespace internal |
| 1299 | } // namespace v8 |
| 1300 | |
| 1301 | #endif // V8_TARGET_ARCH_MIPS |