Line data Source code
1 : // Copyright 2012 the V8 project authors. All rights reserved.
2 : // Use of this source code is governed by a BSD-style license that can be
3 : // found in the LICENSE file.
4 :
5 : #if V8_TARGET_ARCH_X64
6 :
7 : #include "src/regexp/x64/regexp-macro-assembler-x64.h"
8 :
9 : #include "src/heap/factory.h"
10 : #include "src/log.h"
11 : #include "src/macro-assembler.h"
12 : #include "src/objects-inl.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 : /*
23 : * This assembler uses the following register assignment convention
24 : * - rdx : Currently loaded character(s) as Latin1 or UC16. Must be loaded
25 : * using LoadCurrentCharacter before using any of the dispatch methods.
26 : * Temporarily stores the index of capture start after a matching pass
27 : * for a global regexp.
28 : * - rdi : Current position in input, as negative offset from end of string.
29 : * Please notice that this is the byte offset, not the character
30 : * offset! Is always a 32-bit signed (negative) offset, but must be
31 : * maintained sign-extended to 64 bits, since it is used as index.
32 : * - rsi : End of input (points to byte after last character in input),
33 : * so that rsi+rdi points to the current character.
34 : * - rbp : Frame pointer. Used to access arguments, local variables and
35 : * RegExp registers.
36 : * - rsp : Points to tip of C stack.
37 : * - rcx : Points to tip of backtrack stack. The backtrack stack contains
38 : * only 32-bit values. Most are offsets from some base (e.g., character
39 : * positions from end of string or code location from Code pointer).
40 : * - r8 : Code object pointer. Used to convert between absolute and
41 : * code-object-relative addresses.
42 : *
43 : * The registers rax, rbx, r9 and r11 are free to use for computations.
44 : * If changed to use r12+, they should be saved as callee-save registers.
45 : * The macro assembler special register r13 (kRootRegister) isn't special
46 : * during execution of RegExp code (it doesn't hold the value assumed when
47 : * creating JS code), so Root related macro operations can be used.
48 : *
49 : * Each call to a C++ method should retain these registers.
50 : *
51 : * The stack will have the following content, in some order, indexable from the
52 : * frame pointer (see, e.g., kStackHighEnd):
53 : * - Isolate* isolate (address of the current isolate)
54 : * - direct_call (if 1, direct call from JavaScript code, if 0 call
55 : * through the runtime system)
56 : * - stack_area_base (high end of the memory area to use as
57 : * backtracking stack)
58 : * - capture array size (may fit multiple sets of matches)
59 : * - int* capture_array (int[num_saved_registers_], for output).
60 : * - end of input (address of end of string)
61 : * - start of input (address of first character in string)
62 : * - start index (character index of start)
63 : * - String input_string (input string)
64 : * - return address
65 : * - backup of callee save registers (rbx, possibly rsi and rdi).
66 : * - success counter (only useful for global regexp to count matches)
67 : * - Offset of location before start of input (effectively character
68 : * string start - 1). Used to initialize capture registers to a
69 : * non-position.
70 : * - At start of string (if 1, we are starting at the start of the
71 : * string, otherwise 0)
72 : * - register 0 rbp[-n] (Only positions must be stored in the first
73 : * - register 1 rbp[-n-8] num_saved_registers_ registers)
74 : * - ...
75 : *
76 : * The first num_saved_registers_ registers are initialized to point to
77 : * "character -1" in the string (i.e., char_size() bytes before the first
78 : * character of the string). The remaining registers starts out uninitialized.
79 : *
80 : * The first seven values must be provided by the calling code by
81 : * calling the code's entry address cast to a function pointer with the
82 : * following signature:
83 : * int (*match)(String input_string,
84 : * int start_index,
85 : * Address start,
86 : * Address end,
87 : * int* capture_output_array,
88 : * int num_capture_registers,
89 : * byte* stack_area_base,
90 : * bool direct_call = false,
91 : * Isolate* isolate);
92 : */
93 :
94 : #define __ ACCESS_MASM((&masm_))
95 :
96 : const int RegExpMacroAssemblerX64::kRegExpCodeSize;
97 :
98 85333 : RegExpMacroAssemblerX64::RegExpMacroAssemblerX64(Isolate* isolate, Zone* zone,
99 : Mode mode,
100 : int registers_to_save)
101 : : NativeRegExpMacroAssembler(isolate, zone),
102 : masm_(isolate, CodeObjectRequired::kYes,
103 : NewAssemblerBuffer(kRegExpCodeSize)),
104 : no_root_array_scope_(&masm_),
105 : code_relative_fixup_positions_(zone),
106 : mode_(mode),
107 : num_registers_(registers_to_save),
108 : num_saved_registers_(registers_to_save),
109 : entry_label_(),
110 : start_label_(),
111 : success_label_(),
112 : backtrack_label_(),
113 341332 : exit_label_() {
114 : DCHECK_EQ(0, registers_to_save % 2);
115 85333 : __ jmp(&entry_label_); // We'll write the entry code when we know more.
116 85333 : __ bind(&start_label_); // And then continue from here.
117 85333 : }
118 :
119 170666 : RegExpMacroAssemblerX64::~RegExpMacroAssemblerX64() {
120 : // Unuse labels in case we throw away the assembler without calling GetCode.
121 : entry_label_.Unuse();
122 : start_label_.Unuse();
123 : success_label_.Unuse();
124 : backtrack_label_.Unuse();
125 : exit_label_.Unuse();
126 : check_preempt_label_.Unuse();
127 : stack_overflow_label_.Unuse();
128 85333 : }
129 :
130 :
131 510256 : int RegExpMacroAssemblerX64::stack_limit_slack() {
132 510256 : return RegExpStack::kStackLimitSlack;
133 : }
134 :
135 :
136 1016806 : void RegExpMacroAssemblerX64::AdvanceCurrentPosition(int by) {
137 508403 : if (by != 0) {
138 1016806 : __ addq(rdi, Immediate(by * char_size()));
139 : }
140 508403 : }
141 :
142 :
143 3754 : void RegExpMacroAssemblerX64::AdvanceRegister(int reg, int by) {
144 : DCHECK_LE(0, reg);
145 : DCHECK_GT(num_registers_, reg);
146 3754 : if (by != 0) {
147 3754 : __ addp(register_location(reg), Immediate(by));
148 : }
149 3754 : }
150 :
151 :
152 214255 : void RegExpMacroAssemblerX64::Backtrack() {
153 214255 : CheckPreemption();
154 : // Pop Code offset from backtrack stack, add Code and jump to location.
155 214255 : Pop(rbx);
156 214255 : __ addp(rbx, code_object_pointer());
157 214255 : __ jmp(rbx);
158 214255 : }
159 :
160 :
161 2464023 : void RegExpMacroAssemblerX64::Bind(Label* label) {
162 2464023 : __ bind(label);
163 2464023 : }
164 :
165 :
166 164810 : void RegExpMacroAssemblerX64::CheckCharacter(uint32_t c, Label* on_equal) {
167 329620 : __ cmpl(current_character(), Immediate(c));
168 164810 : BranchOrBacktrack(equal, on_equal);
169 164810 : }
170 :
171 :
172 19830 : void RegExpMacroAssemblerX64::CheckCharacterGT(uc16 limit, Label* on_greater) {
173 39660 : __ cmpl(current_character(), Immediate(limit));
174 19830 : BranchOrBacktrack(greater, on_greater);
175 19830 : }
176 :
177 :
178 288 : void RegExpMacroAssemblerX64::CheckAtStart(Label* on_at_start) {
179 576 : __ leap(rax, Operand(rdi, -char_size()));
180 288 : __ cmpp(rax, Operand(rbp, kStringStartMinusOne));
181 288 : BranchOrBacktrack(equal, on_at_start);
182 288 : }
183 :
184 :
185 3068 : void RegExpMacroAssemblerX64::CheckNotAtStart(int cp_offset,
186 3068 : Label* on_not_at_start) {
187 6136 : __ leap(rax, Operand(rdi, -char_size() + cp_offset * char_size()));
188 3068 : __ cmpp(rax, Operand(rbp, kStringStartMinusOne));
189 3068 : BranchOrBacktrack(not_equal, on_not_at_start);
190 3068 : }
191 :
192 :
193 15958 : void RegExpMacroAssemblerX64::CheckCharacterLT(uc16 limit, Label* on_less) {
194 31916 : __ cmpl(current_character(), Immediate(limit));
195 15958 : BranchOrBacktrack(less, on_less);
196 15958 : }
197 :
198 :
199 11700 : void RegExpMacroAssemblerX64::CheckGreedyLoop(Label* on_equal) {
200 11700 : Label fallthrough;
201 11700 : __ cmpl(rdi, Operand(backtrack_stackpointer(), 0));
202 11700 : __ j(not_equal, &fallthrough);
203 : Drop();
204 11700 : BranchOrBacktrack(no_condition, on_equal);
205 11700 : __ bind(&fallthrough);
206 11700 : }
207 :
208 :
209 1701 : void RegExpMacroAssemblerX64::CheckNotBackReferenceIgnoreCase(
210 : int start_reg, bool read_backward, bool unicode, Label* on_no_match) {
211 1701 : Label fallthrough;
212 1701 : ReadPositionFromRegister(rdx, start_reg); // Offset of start of capture
213 1701 : ReadPositionFromRegister(rbx, start_reg + 1); // Offset of end of capture
214 1701 : __ subp(rbx, rdx); // Length of capture.
215 :
216 : // -----------------------
217 : // rdx = Start offset of capture.
218 : // rbx = Length of capture
219 :
220 : // At this point, the capture registers are either both set or both cleared.
221 : // If the capture length is zero, then the capture is either empty or cleared.
222 : // Fall through in both cases.
223 1701 : __ j(equal, &fallthrough);
224 :
225 : // -----------------------
226 : // rdx - Start of capture
227 : // rbx - length of capture
228 : // Check that there are sufficient characters left in the input.
229 1701 : if (read_backward) {
230 130 : __ movl(rax, Operand(rbp, kStringStartMinusOne));
231 65 : __ addl(rax, rbx);
232 65 : __ cmpl(rdi, rax);
233 65 : BranchOrBacktrack(less_equal, on_no_match);
234 : } else {
235 : __ movl(rax, rdi);
236 1636 : __ addl(rax, rbx);
237 1636 : BranchOrBacktrack(greater, on_no_match);
238 : }
239 :
240 1701 : if (mode_ == LATIN1) {
241 1456 : Label loop_increment;
242 1456 : if (on_no_match == nullptr) {
243 1436 : on_no_match = &backtrack_label_;
244 : }
245 :
246 2912 : __ leap(r9, Operand(rsi, rdx, times_1, 0));
247 2912 : __ leap(r11, Operand(rsi, rdi, times_1, 0));
248 1456 : if (read_backward) {
249 60 : __ subp(r11, rbx); // Offset by length when matching backwards.
250 : }
251 1456 : __ addp(rbx, r9); // End of capture
252 : // ---------------------
253 : // r11 - current input character address
254 : // r9 - current capture character address
255 : // rbx - end of capture
256 :
257 1456 : Label loop;
258 1456 : __ bind(&loop);
259 2912 : __ movzxbl(rdx, Operand(r9, 0));
260 2912 : __ movzxbl(rax, Operand(r11, 0));
261 : // al - input character
262 : // dl - capture character
263 : __ cmpb(rax, rdx);
264 1456 : __ j(equal, &loop_increment);
265 :
266 : // Mismatch, try case-insensitive match (converting letters to lower-case).
267 : // I.e., if or-ing with 0x20 makes values equal and in range 'a'-'z', it's
268 : // a match.
269 1456 : __ orp(rax, Immediate(0x20)); // Convert match character to lower-case.
270 1456 : __ orp(rdx, Immediate(0x20)); // Convert capture character to lower-case.
271 : __ cmpb(rax, rdx);
272 1456 : __ j(not_equal, on_no_match); // Definitely not equal.
273 : __ subb(rax, Immediate('a'));
274 : __ cmpb(rax, Immediate('z' - 'a'));
275 1456 : __ j(below_equal, &loop_increment); // In range 'a'-'z'.
276 : // Latin-1: Check for values in range [224,254] but not 247.
277 : __ subb(rax, Immediate(224 - 'a'));
278 : __ cmpb(rax, Immediate(254 - 224));
279 1456 : __ j(above, on_no_match); // Weren't Latin-1 letters.
280 : __ cmpb(rax, Immediate(247 - 224)); // Check for 247.
281 1456 : __ j(equal, on_no_match);
282 1456 : __ bind(&loop_increment);
283 : // Increment pointers into match and capture strings.
284 1456 : __ addp(r11, Immediate(1));
285 1456 : __ addp(r9, Immediate(1));
286 : // Compare to end of capture, and loop if not done.
287 1456 : __ cmpp(r9, rbx);
288 1456 : __ j(below, &loop);
289 :
290 : // Compute new value of character position after the matched part.
291 : __ movp(rdi, r11);
292 1456 : __ subq(rdi, rsi);
293 1456 : if (read_backward) {
294 : // Subtract match length if we matched backward.
295 60 : __ addq(rdi, register_location(start_reg));
296 60 : __ subq(rdi, register_location(start_reg + 1));
297 : }
298 : } else {
299 : DCHECK(mode_ == UC16);
300 : // Save important/volatile registers before calling C function.
301 : #ifndef _WIN64
302 : // Caller save on Linux and callee save in Windows.
303 245 : __ pushq(rsi);
304 245 : __ pushq(rdi);
305 : #endif
306 245 : __ pushq(backtrack_stackpointer());
307 :
308 : static const int num_arguments = 4;
309 245 : __ PrepareCallCFunction(num_arguments);
310 :
311 : // Put arguments into parameter registers. Parameters are
312 : // Address byte_offset1 - Address captured substring's start.
313 : // Address byte_offset2 - Address of current character position.
314 : // size_t byte_length - length of capture in bytes(!)
315 : // Isolate* isolate or 0 if unicode flag.
316 : #ifdef _WIN64
317 : DCHECK(rcx == arg_reg_1);
318 : DCHECK(rdx == arg_reg_2);
319 : // Compute and set byte_offset1 (start of capture).
320 : __ leap(rcx, Operand(rsi, rdx, times_1, 0));
321 : // Set byte_offset2.
322 : __ leap(rdx, Operand(rsi, rdi, times_1, 0));
323 : if (read_backward) {
324 : __ subq(rdx, rbx);
325 : }
326 : #else // AMD64 calling convention
327 : DCHECK(rdi == arg_reg_1);
328 : DCHECK(rsi == arg_reg_2);
329 : // Compute byte_offset2 (current position = rsi+rdi).
330 490 : __ leap(rax, Operand(rsi, rdi, times_1, 0));
331 : // Compute and set byte_offset1 (start of capture).
332 490 : __ leap(rdi, Operand(rsi, rdx, times_1, 0));
333 : // Set byte_offset2.
334 : __ movp(rsi, rax);
335 245 : if (read_backward) {
336 5 : __ subq(rsi, rbx);
337 : }
338 : #endif // _WIN64
339 :
340 : // Set byte_length.
341 : __ movp(arg_reg_3, rbx);
342 : // Isolate.
343 : #ifdef V8_INTL_SUPPORT
344 245 : if (unicode) {
345 : __ movp(arg_reg_4, Immediate(0));
346 : } else // NOLINT
347 : #endif // V8_INTL_SUPPORT
348 : {
349 230 : __ LoadAddress(arg_reg_4, ExternalReference::isolate_address(isolate()));
350 : }
351 :
352 : { // NOLINT: Can't find a way to open this scope without confusing the
353 : // linter.
354 245 : AllowExternalCallThatCantCauseGC scope(&masm_);
355 : ExternalReference compare =
356 245 : ExternalReference::re_case_insensitive_compare_uc16(isolate());
357 245 : __ CallCFunction(compare, num_arguments);
358 : }
359 :
360 : // Restore original values before reacting on result value.
361 245 : __ Move(code_object_pointer(), masm_.CodeObject());
362 245 : __ popq(backtrack_stackpointer());
363 : #ifndef _WIN64
364 245 : __ popq(rdi);
365 245 : __ popq(rsi);
366 : #endif
367 :
368 : // Check if function returned non-zero for success or zero for failure.
369 : __ testp(rax, rax);
370 245 : BranchOrBacktrack(zero, on_no_match);
371 : // On success, advance position by length of capture.
372 : // Requires that rbx is callee save (true for both Win64 and AMD64 ABIs).
373 245 : if (read_backward) {
374 5 : __ subq(rdi, rbx);
375 : } else {
376 240 : __ addq(rdi, rbx);
377 : }
378 : }
379 1701 : __ bind(&fallthrough);
380 1701 : }
381 :
382 :
383 634 : void RegExpMacroAssemblerX64::CheckNotBackReference(int start_reg,
384 : bool read_backward,
385 1268 : Label* on_no_match) {
386 634 : Label fallthrough;
387 :
388 : // Find length of back-referenced capture.
389 634 : ReadPositionFromRegister(rdx, start_reg); // Offset of start of capture
390 634 : ReadPositionFromRegister(rax, start_reg + 1); // Offset of end of capture
391 634 : __ subp(rax, rdx); // Length to check.
392 :
393 : // At this point, the capture registers are either both set or both cleared.
394 : // If the capture length is zero, then the capture is either empty or cleared.
395 : // Fall through in both cases.
396 634 : __ j(equal, &fallthrough);
397 :
398 : // -----------------------
399 : // rdx - Start of capture
400 : // rax - length of capture
401 : // Check that there are sufficient characters left in the input.
402 634 : if (read_backward) {
403 240 : __ movl(rbx, Operand(rbp, kStringStartMinusOne));
404 120 : __ addl(rbx, rax);
405 120 : __ cmpl(rdi, rbx);
406 120 : BranchOrBacktrack(less_equal, on_no_match);
407 : } else {
408 : __ movl(rbx, rdi);
409 514 : __ addl(rbx, rax);
410 514 : BranchOrBacktrack(greater, on_no_match);
411 : }
412 :
413 : // Compute pointers to match string and capture string
414 1268 : __ leap(rbx, Operand(rsi, rdi, times_1, 0)); // Start of match.
415 634 : if (read_backward) {
416 120 : __ subq(rbx, rax); // Offset by length when matching backwards.
417 : }
418 634 : __ addp(rdx, rsi); // Start of capture.
419 1268 : __ leap(r9, Operand(rdx, rax, times_1, 0)); // End of capture
420 :
421 : // -----------------------
422 : // rbx - current capture character address.
423 : // rbx - current input character address .
424 : // r9 - end of input to match (capture length after rbx).
425 :
426 634 : Label loop;
427 634 : __ bind(&loop);
428 634 : if (mode_ == LATIN1) {
429 1118 : __ movzxbl(rax, Operand(rdx, 0));
430 1118 : __ cmpb(rax, Operand(rbx, 0));
431 : } else {
432 : DCHECK(mode_ == UC16);
433 150 : __ movzxwl(rax, Operand(rdx, 0));
434 150 : __ cmpw(rax, Operand(rbx, 0));
435 : }
436 634 : BranchOrBacktrack(not_equal, on_no_match);
437 : // Increment pointers into capture and match string.
438 634 : __ addp(rbx, Immediate(char_size()));
439 634 : __ addp(rdx, Immediate(char_size()));
440 : // Check if we have reached end of match area.
441 634 : __ cmpp(rdx, r9);
442 634 : __ j(below, &loop);
443 :
444 : // Success.
445 : // Set current character position to position after match.
446 : __ movp(rdi, rbx);
447 634 : __ subq(rdi, rsi);
448 634 : if (read_backward) {
449 : // Subtract match length if we matched backward.
450 120 : __ addq(rdi, register_location(start_reg));
451 120 : __ subq(rdi, register_location(start_reg + 1));
452 : }
453 :
454 634 : __ bind(&fallthrough);
455 634 : }
456 :
457 :
458 595480 : void RegExpMacroAssemblerX64::CheckNotCharacter(uint32_t c,
459 : Label* on_not_equal) {
460 1190960 : __ cmpl(current_character(), Immediate(c));
461 595480 : BranchOrBacktrack(not_equal, on_not_equal);
462 595480 : }
463 :
464 :
465 51746 : void RegExpMacroAssemblerX64::CheckCharacterAfterAnd(uint32_t c,
466 : uint32_t mask,
467 : Label* on_equal) {
468 51746 : if (c == 0) {
469 756 : __ testl(current_character(), Immediate(mask));
470 : } else {
471 50990 : __ movl(rax, Immediate(mask));
472 50990 : __ andp(rax, current_character());
473 101980 : __ cmpl(rax, Immediate(c));
474 : }
475 51746 : BranchOrBacktrack(equal, on_equal);
476 51746 : }
477 :
478 :
479 8394 : void RegExpMacroAssemblerX64::CheckNotCharacterAfterAnd(uint32_t c,
480 : uint32_t mask,
481 : Label* on_not_equal) {
482 8394 : if (c == 0) {
483 583 : __ testl(current_character(), Immediate(mask));
484 : } else {
485 7811 : __ movl(rax, Immediate(mask));
486 7811 : __ andp(rax, current_character());
487 15622 : __ cmpl(rax, Immediate(c));
488 : }
489 8394 : BranchOrBacktrack(not_equal, on_not_equal);
490 8394 : }
491 :
492 :
493 85 : void RegExpMacroAssemblerX64::CheckNotCharacterAfterMinusAnd(
494 : uc16 c,
495 : uc16 minus,
496 : uc16 mask,
497 : Label* on_not_equal) {
498 : DCHECK_GT(String::kMaxUtf16CodeUnit, minus);
499 170 : __ leap(rax, Operand(current_character(), -minus));
500 170 : __ andp(rax, Immediate(mask));
501 170 : __ cmpl(rax, Immediate(c));
502 85 : BranchOrBacktrack(not_equal, on_not_equal);
503 85 : }
504 :
505 :
506 22694 : void RegExpMacroAssemblerX64::CheckCharacterInRange(
507 : uc16 from,
508 : uc16 to,
509 : Label* on_in_range) {
510 45388 : __ leal(rax, Operand(current_character(), -from));
511 45388 : __ cmpl(rax, Immediate(to - from));
512 22694 : BranchOrBacktrack(below_equal, on_in_range);
513 22694 : }
514 :
515 :
516 92935 : void RegExpMacroAssemblerX64::CheckCharacterNotInRange(
517 : uc16 from,
518 : uc16 to,
519 : Label* on_not_in_range) {
520 185870 : __ leal(rax, Operand(current_character(), -from));
521 185870 : __ cmpl(rax, Immediate(to - from));
522 92935 : BranchOrBacktrack(above, on_not_in_range);
523 92935 : }
524 :
525 :
526 10302 : void RegExpMacroAssemblerX64::CheckBitInTable(
527 : Handle<ByteArray> table,
528 : Label* on_bit_set) {
529 10302 : __ Move(rax, table);
530 : Register index = current_character();
531 : if (mode_ != LATIN1 || kTableMask != String::kMaxOneByteCharCode) {
532 10302 : __ movp(rbx, current_character());
533 10302 : __ andp(rbx, Immediate(kTableMask));
534 : index = rbx;
535 : }
536 : __ cmpb(FieldOperand(rax, index, times_1, ByteArray::kHeaderSize),
537 : Immediate(0));
538 10302 : BranchOrBacktrack(not_equal, on_bit_set);
539 10302 : }
540 :
541 :
542 11699 : bool RegExpMacroAssemblerX64::CheckSpecialCharacterClass(uc16 type,
543 : Label* on_no_match) {
544 : // Range checks (c in min..max) are generally implemented by an unsigned
545 : // (c - min) <= (max - min) check, using the sequence:
546 : // leap(rax, Operand(current_character(), -min)) or sub(rax, Immediate(min))
547 : // cmp(rax, Immediate(max - min))
548 11699 : switch (type) {
549 : case 's':
550 : // Match space-characters
551 697 : if (mode_ == LATIN1) {
552 : // One byte space characters are '\t'..'\r', ' ' and \u00a0.
553 582 : Label success;
554 582 : __ cmpl(current_character(), Immediate(' '));
555 582 : __ j(equal, &success, Label::kNear);
556 : // Check range 0x09..0x0D
557 1164 : __ leap(rax, Operand(current_character(), -'\t'));
558 582 : __ cmpl(rax, Immediate('\r' - '\t'));
559 582 : __ j(below_equal, &success, Label::kNear);
560 : // \u00a0 (NBSP).
561 582 : __ cmpl(rax, Immediate(0x00A0 - '\t'));
562 582 : BranchOrBacktrack(not_equal, on_no_match);
563 582 : __ bind(&success);
564 : return true;
565 : }
566 : return false;
567 : case 'S':
568 : // The emitted code for generic character classes is good enough.
569 : return false;
570 : case 'd':
571 : // Match ASCII digits ('0'..'9')
572 0 : __ leap(rax, Operand(current_character(), -'0'));
573 0 : __ cmpl(rax, Immediate('9' - '0'));
574 0 : BranchOrBacktrack(above, on_no_match);
575 0 : return true;
576 : case 'D':
577 : // Match non ASCII-digits
578 0 : __ leap(rax, Operand(current_character(), -'0'));
579 0 : __ cmpl(rax, Immediate('9' - '0'));
580 0 : BranchOrBacktrack(below_equal, on_no_match);
581 0 : return true;
582 : case '.': {
583 : // Match non-newlines (not 0x0A('\n'), 0x0D('\r'), 0x2028 and 0x2029)
584 8069 : __ movl(rax, current_character());
585 8069 : __ xorp(rax, Immediate(0x01));
586 : // See if current character is '\n'^1 or '\r'^1, i.e., 0x0B or 0x0C
587 8069 : __ subl(rax, Immediate(0x0B));
588 8069 : __ cmpl(rax, Immediate(0x0C - 0x0B));
589 8069 : BranchOrBacktrack(below_equal, on_no_match);
590 8069 : if (mode_ == UC16) {
591 : // Compare original value to 0x2028 and 0x2029, using the already
592 : // computed (current_char ^ 0x01 - 0x0B). I.e., check for
593 : // 0x201D (0x2028 - 0x0B) or 0x201E.
594 432 : __ subl(rax, Immediate(0x2028 - 0x0B));
595 432 : __ cmpl(rax, Immediate(0x2029 - 0x2028));
596 432 : BranchOrBacktrack(below_equal, on_no_match);
597 : }
598 : return true;
599 : }
600 : case 'n': {
601 : // Match newlines (0x0A('\n'), 0x0D('\r'), 0x2028 and 0x2029)
602 233 : __ movl(rax, current_character());
603 233 : __ xorp(rax, Immediate(0x01));
604 : // See if current character is '\n'^1 or '\r'^1, i.e., 0x0B or 0x0C
605 233 : __ subl(rax, Immediate(0x0B));
606 233 : __ cmpl(rax, Immediate(0x0C - 0x0B));
607 233 : if (mode_ == LATIN1) {
608 208 : BranchOrBacktrack(above, on_no_match);
609 : } else {
610 25 : Label done;
611 25 : BranchOrBacktrack(below_equal, &done);
612 : // Compare original value to 0x2028 and 0x2029, using the already
613 : // computed (current_char ^ 0x01 - 0x0B). I.e., check for
614 : // 0x201D (0x2028 - 0x0B) or 0x201E.
615 25 : __ subl(rax, Immediate(0x2028 - 0x0B));
616 25 : __ cmpl(rax, Immediate(0x2029 - 0x2028));
617 25 : BranchOrBacktrack(above, on_no_match);
618 25 : __ bind(&done);
619 : }
620 : return true;
621 : }
622 : case 'w': {
623 1903 : if (mode_ != LATIN1) {
624 : // Table is 256 entries, so all Latin1 characters can be tested.
625 61 : __ cmpl(current_character(), Immediate('z'));
626 61 : BranchOrBacktrack(above, on_no_match);
627 : }
628 1903 : __ Move(rbx, ExternalReference::re_word_character_map(isolate()));
629 : DCHECK_EQ(0, word_character_map[0]); // Character '\0' is not a word char.
630 : __ testb(Operand(rbx, current_character(), times_1, 0),
631 1903 : current_character());
632 1903 : BranchOrBacktrack(zero, on_no_match);
633 1903 : return true;
634 : }
635 : case 'W': {
636 521 : Label done;
637 521 : if (mode_ != LATIN1) {
638 : // Table is 256 entries, so all Latin1 characters can be tested.
639 104 : __ cmpl(current_character(), Immediate('z'));
640 104 : __ j(above, &done);
641 : }
642 521 : __ Move(rbx, ExternalReference::re_word_character_map(isolate()));
643 : DCHECK_EQ(0, word_character_map[0]); // Character '\0' is not a word char.
644 : __ testb(Operand(rbx, current_character(), times_1, 0),
645 521 : current_character());
646 521 : BranchOrBacktrack(not_zero, on_no_match);
647 521 : if (mode_ != LATIN1) {
648 104 : __ bind(&done);
649 : }
650 : return true;
651 : }
652 :
653 : case '*':
654 : // Match any character.
655 0 : return true;
656 : // No custom implementation (yet): s(UC16), S(UC16).
657 : default:
658 : return false;
659 : }
660 : }
661 :
662 :
663 85338 : void RegExpMacroAssemblerX64::Fail() {
664 : STATIC_ASSERT(FAILURE == 0); // Return value for failure is zero.
665 85338 : if (!global()) {
666 81515 : __ Set(rax, FAILURE);
667 : }
668 85338 : __ jmp(&exit_label_);
669 85338 : }
670 :
671 :
672 85333 : Handle<HeapObject> RegExpMacroAssemblerX64::GetCode(Handle<String> source) {
673 85333 : Label return_rax;
674 : // Finalize code - write the entry point code now we know how many
675 : // registers we need.
676 : // Entry code:
677 85333 : __ bind(&entry_label_);
678 :
679 : // Tell the system that we have a stack frame. Because the type is MANUAL, no
680 : // is generated.
681 170666 : FrameScope scope(&masm_, StackFrame::MANUAL);
682 :
683 : // Actually emit code to start a new stack frame.
684 85333 : __ pushq(rbp);
685 : __ movp(rbp, rsp);
686 : // Save parameters and callee-save registers. Order here should correspond
687 : // to order of kBackup_ebx etc.
688 : #ifdef _WIN64
689 : // MSVC passes arguments in rcx, rdx, r8, r9, with backing stack slots.
690 : // Store register parameters in pre-allocated stack slots,
691 : __ movq(Operand(rbp, kInputString), rcx);
692 : __ movq(Operand(rbp, kStartIndex), rdx); // Passed as int32 in edx.
693 : __ movq(Operand(rbp, kInputStart), r8);
694 : __ movq(Operand(rbp, kInputEnd), r9);
695 : // Callee-save on Win64.
696 : __ pushq(rsi);
697 : __ pushq(rdi);
698 : __ pushq(rbx);
699 : #else
700 : // GCC passes arguments in rdi, rsi, rdx, rcx, r8, r9 (and then on stack).
701 : // Push register parameters on stack for reference.
702 : DCHECK_EQ(kInputString, -1 * kRegisterSize);
703 : DCHECK_EQ(kStartIndex, -2 * kRegisterSize);
704 : DCHECK_EQ(kInputStart, -3 * kRegisterSize);
705 : DCHECK_EQ(kInputEnd, -4 * kRegisterSize);
706 : DCHECK_EQ(kRegisterOutput, -5 * kRegisterSize);
707 : DCHECK_EQ(kNumOutputRegisters, -6 * kRegisterSize);
708 85333 : __ pushq(rdi);
709 85333 : __ pushq(rsi);
710 85333 : __ pushq(rdx);
711 85333 : __ pushq(rcx);
712 85333 : __ pushq(r8);
713 85333 : __ pushq(r9);
714 :
715 85333 : __ pushq(rbx); // Callee-save
716 : #endif
717 :
718 85333 : __ Push(Immediate(0)); // Number of successful matches in a global regexp.
719 85333 : __ Push(Immediate(0)); // Make room for "string start - 1" constant.
720 :
721 : // Check if we have space on the stack for registers.
722 85333 : Label stack_limit_hit;
723 85333 : Label stack_ok;
724 :
725 : ExternalReference stack_limit =
726 85333 : ExternalReference::address_of_stack_limit(isolate());
727 : __ movp(rcx, rsp);
728 85333 : __ Move(kScratchRegister, stack_limit);
729 85333 : __ subp(rcx, Operand(kScratchRegister, 0));
730 : // Handle it if the stack pointer is already below the stack limit.
731 85333 : __ j(below_equal, &stack_limit_hit);
732 : // Check if there is room for the variable number of registers above
733 : // the stack limit.
734 170666 : __ cmpp(rcx, Immediate(num_registers_ * kSystemPointerSize));
735 85333 : __ j(above_equal, &stack_ok);
736 : // Exit with OutOfMemory exception. There is not enough space on the stack
737 : // for our working registers.
738 85333 : __ Set(rax, EXCEPTION);
739 85333 : __ jmp(&return_rax);
740 :
741 85333 : __ bind(&stack_limit_hit);
742 85333 : __ Move(code_object_pointer(), masm_.CodeObject());
743 85333 : CallCheckStackGuardState(); // Preserves no registers beside rbp and rsp.
744 : __ testp(rax, rax);
745 : // If returned value is non-zero, we exit with the returned value as result.
746 85333 : __ j(not_zero, &return_rax);
747 :
748 85333 : __ bind(&stack_ok);
749 :
750 : // Allocate space on stack for registers.
751 170666 : __ subp(rsp, Immediate(num_registers_ * kSystemPointerSize));
752 : // Load string length.
753 170666 : __ movp(rsi, Operand(rbp, kInputEnd));
754 : // Load input position.
755 170666 : __ movp(rdi, Operand(rbp, kInputStart));
756 : // Set up rdi to be negative offset from string end.
757 85333 : __ subq(rdi, rsi);
758 : // Set rax to address of char before start of the string
759 : // (effectively string position -1).
760 170666 : __ movp(rbx, Operand(rbp, kStartIndex));
761 : __ negq(rbx);
762 85333 : if (mode_ == UC16) {
763 142070 : __ leap(rax, Operand(rdi, rbx, times_2, -char_size()));
764 : } else {
765 28596 : __ leap(rax, Operand(rdi, rbx, times_1, -char_size()));
766 : }
767 : // Store this value in a local variable, for use when clearing
768 : // position registers.
769 170666 : __ movp(Operand(rbp, kStringStartMinusOne), rax);
770 :
771 : #if V8_OS_WIN
772 : // Ensure that we have written to each stack page, in order. Skipping a page
773 : // on Windows can cause segmentation faults. Assuming page size is 4k.
774 : const int kPageSize = 4096;
775 : const int kRegistersPerPage = kPageSize / kSystemPointerSize;
776 : for (int i = num_saved_registers_ + kRegistersPerPage - 1;
777 : i < num_registers_;
778 : i += kRegistersPerPage) {
779 : __ movp(register_location(i), rax); // One write every page.
780 : }
781 : #endif // V8_OS_WIN
782 :
783 : // Initialize code object pointer.
784 85333 : __ Move(code_object_pointer(), masm_.CodeObject());
785 :
786 85333 : Label load_char_start_regexp, start_regexp;
787 : // Load newline if index is at start, previous character otherwise.
788 85333 : __ cmpl(Operand(rbp, kStartIndex), Immediate(0));
789 85333 : __ j(not_equal, &load_char_start_regexp, Label::kNear);
790 85333 : __ Set(current_character(), '\n');
791 85333 : __ jmp(&start_regexp, Label::kNear);
792 :
793 : // Global regexp restarts matching here.
794 85333 : __ bind(&load_char_start_regexp);
795 : // Load previous char as initial value of current character register.
796 85333 : LoadCurrentCharacterUnchecked(-1, 1);
797 85333 : __ bind(&start_regexp);
798 :
799 : // Initialize on-stack registers.
800 85333 : if (num_saved_registers_ > 0) {
801 : // Fill saved registers with initial value = start offset - 1
802 : // Fill in stack push order, to avoid accessing across an unwritten
803 : // page (a problem on Windows).
804 85318 : if (num_saved_registers_ > 8) {
805 480 : __ Set(rcx, kRegisterZero);
806 480 : Label init_loop;
807 480 : __ bind(&init_loop);
808 960 : __ movp(Operand(rbp, rcx, times_1, 0), rax);
809 480 : __ subq(rcx, Immediate(kSystemPointerSize));
810 : __ cmpq(rcx, Immediate(kRegisterZero -
811 960 : num_saved_registers_ * kSystemPointerSize));
812 480 : __ j(greater, &init_loop);
813 : } else { // Unroll the loop.
814 186818 : for (int i = 0; i < num_saved_registers_; i++) {
815 186818 : __ movp(register_location(i), rax);
816 : }
817 : }
818 : }
819 :
820 : // Initialize backtrack stack pointer.
821 170666 : __ movp(backtrack_stackpointer(), Operand(rbp, kStackHighEnd));
822 :
823 85333 : __ jmp(&start_label_);
824 :
825 : // Exit code:
826 85333 : if (success_label_.is_linked()) {
827 : // Save captures when successful.
828 85022 : __ bind(&success_label_);
829 85022 : if (num_saved_registers_ > 0) {
830 : // copy captures to output
831 170024 : __ movp(rdx, Operand(rbp, kStartIndex));
832 170024 : __ movp(rbx, Operand(rbp, kRegisterOutput));
833 170024 : __ movp(rcx, Operand(rbp, kInputEnd));
834 85012 : __ subp(rcx, Operand(rbp, kInputStart));
835 85012 : if (mode_ == UC16) {
836 142070 : __ leap(rcx, Operand(rcx, rdx, times_2, 0));
837 : } else {
838 13977 : __ addp(rcx, rdx);
839 : }
840 221572 : for (int i = 0; i < num_saved_registers_; i++) {
841 221572 : __ movp(rax, register_location(i));
842 480899 : if (i == 0 && global_with_zero_length_check()) {
843 : // Keep capture start in rdx for the zero-length check later.
844 : __ movp(rdx, rax);
845 : }
846 221572 : __ addp(rax, rcx); // Convert to index from start, not end.
847 221572 : if (mode_ == UC16) {
848 : __ sarp(rax, Immediate(1)); // Convert byte index to character index.
849 : }
850 443144 : __ movl(Operand(rbx, i * kIntSize), rax);
851 : }
852 : }
853 :
854 85022 : if (global()) {
855 : // Restart matching if the regular expression is flagged as global.
856 : // Increment success counter.
857 7644 : __ incp(Operand(rbp, kSuccessfulCaptures));
858 : // Capture results have been stored, so the number of remaining global
859 : // output registers is reduced by the number of stored captures.
860 3822 : __ movsxlq(rcx, Operand(rbp, kNumOutputRegisters));
861 7644 : __ subp(rcx, Immediate(num_saved_registers_));
862 : // Check whether we have enough room for another set of capture results.
863 7644 : __ cmpp(rcx, Immediate(num_saved_registers_));
864 3822 : __ j(less, &exit_label_);
865 :
866 7644 : __ movp(Operand(rbp, kNumOutputRegisters), rcx);
867 : // Advance the location for output.
868 : __ addp(Operand(rbp, kRegisterOutput),
869 7644 : Immediate(num_saved_registers_ * kIntSize));
870 :
871 : // Prepare rax to initialize registers with its value in the next run.
872 7644 : __ movp(rax, Operand(rbp, kStringStartMinusOne));
873 :
874 3822 : if (global_with_zero_length_check()) {
875 : // Special case for zero-length matches.
876 : // rdx: capture start index
877 138 : __ cmpp(rdi, rdx);
878 : // Not a zero-length match, restart.
879 138 : __ j(not_equal, &load_char_start_regexp);
880 : // rdi (offset from the end) is zero if we already reached the end.
881 : __ testp(rdi, rdi);
882 138 : __ j(zero, &exit_label_, Label::kNear);
883 : // Advance current position after a zero-length match.
884 138 : Label advance;
885 138 : __ bind(&advance);
886 138 : if (mode_ == UC16) {
887 23 : __ addq(rdi, Immediate(2));
888 : } else {
889 : __ incq(rdi);
890 : }
891 138 : if (global_unicode()) CheckNotInSurrogatePair(0, &advance);
892 : }
893 :
894 3822 : __ jmp(&load_char_start_regexp);
895 : } else {
896 : __ movp(rax, Immediate(SUCCESS));
897 : }
898 : }
899 :
900 85333 : __ bind(&exit_label_);
901 85333 : if (global()) {
902 : // Return the number of successful captures.
903 7646 : __ movp(rax, Operand(rbp, kSuccessfulCaptures));
904 : }
905 :
906 85333 : __ bind(&return_rax);
907 : #ifdef _WIN64
908 : // Restore callee save registers.
909 : __ leap(rsp, Operand(rbp, kLastCalleeSaveRegister));
910 : __ popq(rbx);
911 : __ popq(rdi);
912 : __ popq(rsi);
913 : // Stack now at rbp.
914 : #else
915 : // Restore callee save register.
916 170666 : __ movp(rbx, Operand(rbp, kBackup_rbx));
917 : // Skip rsp to rbp.
918 : __ movp(rsp, rbp);
919 : #endif
920 : // Exit function frame, restore previous one.
921 85333 : __ popq(rbp);
922 85333 : __ ret(0);
923 :
924 : // Backtrack code (branch target for conditional backtracks).
925 85333 : if (backtrack_label_.is_linked()) {
926 84992 : __ bind(&backtrack_label_);
927 84992 : Backtrack();
928 : }
929 :
930 85333 : Label exit_with_exception;
931 :
932 : // Preempt-code
933 85333 : if (check_preempt_label_.is_linked()) {
934 85298 : SafeCallTarget(&check_preempt_label_);
935 :
936 85298 : __ pushq(backtrack_stackpointer());
937 85298 : __ pushq(rdi);
938 :
939 85298 : CallCheckStackGuardState();
940 : __ testp(rax, rax);
941 : // If returning non-zero, we should end execution with the given
942 : // result as return value.
943 85298 : __ j(not_zero, &return_rax);
944 :
945 : // Restore registers.
946 85298 : __ Move(code_object_pointer(), masm_.CodeObject());
947 85298 : __ popq(rdi);
948 85298 : __ popq(backtrack_stackpointer());
949 : // String might have moved: Reload esi from frame.
950 170596 : __ movp(rsi, Operand(rbp, kInputEnd));
951 85298 : SafeReturn();
952 : }
953 :
954 : // Backtrack stack overflow code.
955 85333 : if (stack_overflow_label_.is_linked()) {
956 85303 : SafeCallTarget(&stack_overflow_label_);
957 : // Reached if the backtrack-stack limit has been hit.
958 :
959 : Label grow_failed;
960 : // Save registers before calling C function
961 : #ifndef _WIN64
962 : // Callee-save in Microsoft 64-bit ABI, but not in AMD64 ABI.
963 85303 : __ pushq(rsi);
964 85303 : __ pushq(rdi);
965 : #endif
966 :
967 : // Call GrowStack(backtrack_stackpointer())
968 : static const int num_arguments = 3;
969 85303 : __ PrepareCallCFunction(num_arguments);
970 : #ifdef _WIN64
971 : // Microsoft passes parameters in rcx, rdx, r8.
972 : // First argument, backtrack stackpointer, is already in rcx.
973 : __ leap(rdx, Operand(rbp, kStackHighEnd)); // Second argument
974 : __ LoadAddress(r8, ExternalReference::isolate_address(isolate()));
975 : #else
976 : // AMD64 ABI passes parameters in rdi, rsi, rdx.
977 : __ movp(rdi, backtrack_stackpointer()); // First argument.
978 170606 : __ leap(rsi, Operand(rbp, kStackHighEnd)); // Second argument.
979 85303 : __ LoadAddress(rdx, ExternalReference::isolate_address(isolate()));
980 : #endif
981 : ExternalReference grow_stack =
982 85303 : ExternalReference::re_grow_stack(isolate());
983 85303 : __ CallCFunction(grow_stack, num_arguments);
984 : // If return nullptr, we have failed to grow the stack, and
985 : // must exit with a stack-overflow exception.
986 : __ testp(rax, rax);
987 85303 : __ j(equal, &exit_with_exception);
988 : // Otherwise use return value as new stack pointer.
989 : __ movp(backtrack_stackpointer(), rax);
990 : // Restore saved registers and continue.
991 85303 : __ Move(code_object_pointer(), masm_.CodeObject());
992 : #ifndef _WIN64
993 85303 : __ popq(rdi);
994 85303 : __ popq(rsi);
995 : #endif
996 85303 : SafeReturn();
997 : }
998 :
999 85333 : if (exit_with_exception.is_linked()) {
1000 : // If any of the code above needed to exit with an exception.
1001 85303 : __ bind(&exit_with_exception);
1002 : // Exit with Result EXCEPTION(-1) to signal thrown exception.
1003 85303 : __ Set(rax, EXCEPTION);
1004 85303 : __ jmp(&return_rax);
1005 : }
1006 :
1007 85333 : FixupCodeRelativePositions();
1008 :
1009 85333 : CodeDesc code_desc;
1010 : Isolate* isolate = this->isolate();
1011 85333 : masm_.GetCode(isolate, &code_desc);
1012 : Handle<Code> code =
1013 170666 : isolate->factory()->NewCode(code_desc, Code::REGEXP, masm_.CodeObject());
1014 85333 : PROFILE(isolate, RegExpCodeCreateEvent(AbstractCode::cast(*code), *source));
1015 170666 : return Handle<HeapObject>::cast(code);
1016 : }
1017 :
1018 :
1019 1207172 : void RegExpMacroAssemblerX64::GoTo(Label* to) {
1020 1207172 : BranchOrBacktrack(no_condition, to);
1021 1207172 : }
1022 :
1023 :
1024 2609 : void RegExpMacroAssemblerX64::IfRegisterGE(int reg,
1025 : int comparand,
1026 : Label* if_ge) {
1027 2609 : __ cmpp(register_location(reg), Immediate(comparand));
1028 2609 : BranchOrBacktrack(greater_equal, if_ge);
1029 2609 : }
1030 :
1031 :
1032 1544 : void RegExpMacroAssemblerX64::IfRegisterLT(int reg,
1033 : int comparand,
1034 : Label* if_lt) {
1035 1544 : __ cmpp(register_location(reg), Immediate(comparand));
1036 1544 : BranchOrBacktrack(less, if_lt);
1037 1544 : }
1038 :
1039 :
1040 281 : void RegExpMacroAssemblerX64::IfRegisterEqPos(int reg,
1041 : Label* if_eq) {
1042 281 : __ cmpp(rdi, register_location(reg));
1043 281 : BranchOrBacktrack(equal, if_eq);
1044 281 : }
1045 :
1046 :
1047 : RegExpMacroAssembler::IrregexpImplementation
1048 0 : RegExpMacroAssemblerX64::Implementation() {
1049 0 : return kX64Implementation;
1050 : }
1051 :
1052 :
1053 900376 : void RegExpMacroAssemblerX64::LoadCurrentCharacter(int cp_offset,
1054 : Label* on_end_of_input,
1055 : bool check_bounds,
1056 : int characters) {
1057 : DCHECK(cp_offset < (1<<30)); // Be sane! (And ensure negation works)
1058 900376 : if (check_bounds) {
1059 423190 : if (cp_offset >= 0) {
1060 409692 : CheckPosition(cp_offset + characters - 1, on_end_of_input);
1061 : } else {
1062 13498 : CheckPosition(cp_offset, on_end_of_input);
1063 : }
1064 : }
1065 900376 : LoadCurrentCharacterUnchecked(cp_offset, characters);
1066 900376 : }
1067 :
1068 :
1069 399328 : void RegExpMacroAssemblerX64::PopCurrentPosition() {
1070 399328 : Pop(rdi);
1071 399328 : }
1072 :
1073 :
1074 52754 : void RegExpMacroAssemblerX64::PopRegister(int register_index) {
1075 52754 : Pop(rax);
1076 52754 : __ movp(register_location(register_index), rax);
1077 52754 : }
1078 :
1079 :
1080 595574 : void RegExpMacroAssemblerX64::PushBacktrack(Label* label) {
1081 595574 : Push(label);
1082 595574 : CheckStackLimit();
1083 595574 : }
1084 :
1085 :
1086 411018 : void RegExpMacroAssemblerX64::PushCurrentPosition() {
1087 411018 : Push(rdi);
1088 411018 : }
1089 :
1090 :
1091 52769 : void RegExpMacroAssemblerX64::PushRegister(int register_index,
1092 : StackCheckFlag check_stack_limit) {
1093 52769 : __ movp(rax, register_location(register_index));
1094 52769 : Push(rax);
1095 52769 : if (check_stack_limit) CheckStackLimit();
1096 52769 : }
1097 :
1098 : STATIC_ASSERT(kSystemPointerSize == kInt64Size ||
1099 : kSystemPointerSize == kInt32Size);
1100 :
1101 4478 : void RegExpMacroAssemblerX64::ReadCurrentPositionFromRegister(int reg) {
1102 : if (kSystemPointerSize == kInt64Size) {
1103 4478 : __ movq(rdi, register_location(reg));
1104 : } else {
1105 : // Need sign extension for x32 as rdi might be used as an index register.
1106 : __ movsxlq(rdi, register_location(reg));
1107 : }
1108 4478 : }
1109 :
1110 :
1111 4670 : void RegExpMacroAssemblerX64::ReadPositionFromRegister(Register dst, int reg) {
1112 : if (kSystemPointerSize == kInt64Size) {
1113 4670 : __ movq(dst, register_location(reg));
1114 : } else {
1115 : // Need sign extension for x32 as dst might be used as an index register.
1116 : __ movsxlq(dst, register_location(reg));
1117 : }
1118 4670 : }
1119 :
1120 :
1121 4473 : void RegExpMacroAssemblerX64::ReadStackPointerFromRegister(int reg) {
1122 4473 : __ movp(backtrack_stackpointer(), register_location(reg));
1123 4473 : __ addp(backtrack_stackpointer(), Operand(rbp, kStackHighEnd));
1124 4473 : }
1125 :
1126 :
1127 630 : void RegExpMacroAssemblerX64::SetCurrentPositionFromEnd(int by) {
1128 210 : Label after_position;
1129 630 : __ cmpp(rdi, Immediate(-by * char_size()));
1130 210 : __ j(greater_equal, &after_position, Label::kNear);
1131 210 : __ movq(rdi, Immediate(-by * char_size()));
1132 : // On RegExp code entry (where this operation is used), the character before
1133 : // the current position is expected to be already loaded.
1134 : // We have advanced the position, so it's safe to read backwards.
1135 210 : LoadCurrentCharacterUnchecked(-1, 1);
1136 210 : __ bind(&after_position);
1137 210 : }
1138 :
1139 :
1140 3475 : void RegExpMacroAssemblerX64::SetRegister(int register_index, int to) {
1141 : DCHECK(register_index >= num_saved_registers_); // Reserved for positions!
1142 3475 : __ movp(register_location(register_index), Immediate(to));
1143 3475 : }
1144 :
1145 :
1146 91177 : bool RegExpMacroAssemblerX64::Succeed() {
1147 91177 : __ jmp(&success_label_);
1148 182354 : return global();
1149 : }
1150 :
1151 :
1152 412974 : void RegExpMacroAssemblerX64::WriteCurrentPositionToRegister(int reg,
1153 169360 : int cp_offset) {
1154 412974 : if (cp_offset == 0) {
1155 243614 : __ movp(register_location(reg), rdi);
1156 : } else {
1157 338720 : __ leap(rax, Operand(rdi, cp_offset * char_size()));
1158 169360 : __ movp(register_location(reg), rax);
1159 : }
1160 412974 : }
1161 :
1162 :
1163 116918 : void RegExpMacroAssemblerX64::ClearRegisters(int reg_from, int reg_to) {
1164 : DCHECK(reg_from <= reg_to);
1165 233836 : __ movp(rax, Operand(rbp, kStringStartMinusOne));
1166 340348 : for (int reg = reg_from; reg <= reg_to; reg++) {
1167 223430 : __ movp(register_location(reg), rax);
1168 : }
1169 116918 : }
1170 :
1171 :
1172 4437 : void RegExpMacroAssemblerX64::WriteStackPointerToRegister(int reg) {
1173 4437 : __ movp(rax, backtrack_stackpointer());
1174 4437 : __ subp(rax, Operand(rbp, kStackHighEnd));
1175 4437 : __ movp(register_location(reg), rax);
1176 4437 : }
1177 :
1178 :
1179 : // Private methods:
1180 :
1181 170631 : void RegExpMacroAssemblerX64::CallCheckStackGuardState() {
1182 : // This function call preserves no register values. Caller should
1183 : // store anything volatile in a C call or overwritten by this function.
1184 : static const int num_arguments = 3;
1185 170631 : __ PrepareCallCFunction(num_arguments);
1186 : #ifdef _WIN64
1187 : // Second argument: Code of self. (Do this before overwriting r8).
1188 : __ movp(rdx, code_object_pointer());
1189 : // Third argument: RegExp code frame pointer.
1190 : __ movp(r8, rbp);
1191 : // First argument: Next address on the stack (will be address of
1192 : // return address).
1193 : __ leap(rcx, Operand(rsp, -kSystemPointerSize));
1194 : #else
1195 : // Third argument: RegExp code frame pointer.
1196 170631 : __ movp(rdx, rbp);
1197 : // Second argument: Code of self.
1198 : __ movp(rsi, code_object_pointer());
1199 : // First argument: Next address on the stack (will be address of
1200 : // return address).
1201 341262 : __ leap(rdi, Operand(rsp, -kRegisterSize));
1202 : #endif
1203 : ExternalReference stack_check =
1204 170631 : ExternalReference::re_check_stack_guard_state(isolate());
1205 170631 : __ CallCFunction(stack_check, num_arguments);
1206 170631 : }
1207 :
1208 :
1209 : // Helper function for reading a value out of a stack frame.
1210 : template <typename T>
1211 : static T& frame_entry(Address re_frame, int frame_offset) {
1212 1200 : return reinterpret_cast<T&>(Memory<int32_t>(re_frame + frame_offset));
1213 : }
1214 :
1215 :
1216 : template <typename T>
1217 : static T* frame_entry_address(Address re_frame, int frame_offset) {
1218 400 : return reinterpret_cast<T*>(re_frame + frame_offset);
1219 : }
1220 :
1221 400 : int RegExpMacroAssemblerX64::CheckStackGuardState(Address* return_address,
1222 : Address raw_code,
1223 : Address re_frame) {
1224 400 : Code re_code = Code::cast(Object(raw_code));
1225 : return NativeRegExpMacroAssembler::CheckStackGuardState(
1226 : frame_entry<Isolate*>(re_frame, kIsolate),
1227 : frame_entry<int>(re_frame, kStartIndex),
1228 : frame_entry<int>(re_frame, kDirectCall) == 1, return_address, re_code,
1229 : frame_entry_address<Address>(re_frame, kInputString),
1230 : frame_entry_address<const byte*>(re_frame, kInputStart),
1231 1200 : frame_entry_address<const byte*>(re_frame, kInputEnd));
1232 : }
1233 :
1234 :
1235 1180398 : Operand RegExpMacroAssemblerX64::register_location(int register_index) {
1236 : DCHECK(register_index < (1<<30));
1237 1180398 : if (num_registers_ <= register_index) {
1238 5109 : num_registers_ = register_index + 1;
1239 : }
1240 1180398 : return Operand(rbp, kRegisterZero - register_index * kSystemPointerSize);
1241 : }
1242 :
1243 :
1244 479422 : void RegExpMacroAssemblerX64::CheckPosition(int cp_offset,
1245 479422 : Label* on_outside_input) {
1246 479422 : if (cp_offset >= 0) {
1247 1397757 : __ cmpl(rdi, Immediate(-cp_offset * char_size()));
1248 465919 : BranchOrBacktrack(greater_equal, on_outside_input);
1249 : } else {
1250 27006 : __ leap(rax, Operand(rdi, cp_offset * char_size()));
1251 13503 : __ cmpp(rax, Operand(rbp, kStringStartMinusOne));
1252 13503 : BranchOrBacktrack(less_equal, on_outside_input);
1253 : }
1254 479422 : }
1255 :
1256 :
1257 2703358 : void RegExpMacroAssemblerX64::BranchOrBacktrack(Condition condition,
1258 : Label* to) {
1259 2703358 : if (condition < 0) { // No condition
1260 1218872 : if (to == nullptr) {
1261 14984 : Backtrack();
1262 14984 : return;
1263 : }
1264 1203888 : __ jmp(to);
1265 1203888 : return;
1266 : }
1267 1484486 : if (to == nullptr) {
1268 479654 : __ j(condition, &backtrack_label_);
1269 479654 : return;
1270 : }
1271 1004832 : __ j(condition, to);
1272 : }
1273 :
1274 :
1275 : void RegExpMacroAssemblerX64::SafeCall(Label* to) {
1276 812139 : __ call(to);
1277 : }
1278 :
1279 :
1280 170601 : void RegExpMacroAssemblerX64::SafeCallTarget(Label* label) {
1281 170601 : __ bind(label);
1282 170601 : __ subp(Operand(rsp, 0), code_object_pointer());
1283 170601 : }
1284 :
1285 :
1286 170601 : void RegExpMacroAssemblerX64::SafeReturn() {
1287 170601 : __ addp(Operand(rsp, 0), code_object_pointer());
1288 170601 : __ ret(0);
1289 170601 : }
1290 :
1291 :
1292 463787 : void RegExpMacroAssemblerX64::Push(Register source) {
1293 : DCHECK(source != backtrack_stackpointer());
1294 : // Notice: This updates flags, unlike normal Push.
1295 463787 : __ subp(backtrack_stackpointer(), Immediate(kIntSize));
1296 927574 : __ movl(Operand(backtrack_stackpointer(), 0), source);
1297 463787 : }
1298 :
1299 :
1300 : void RegExpMacroAssemblerX64::Push(Immediate value) {
1301 : // Notice: This updates flags, unlike normal Push.
1302 : __ subp(backtrack_stackpointer(), Immediate(kIntSize));
1303 : __ movl(Operand(backtrack_stackpointer(), 0), value);
1304 : }
1305 :
1306 :
1307 85333 : void RegExpMacroAssemblerX64::FixupCodeRelativePositions() {
1308 1361814 : for (int position : code_relative_fixup_positions_) {
1309 : // The position succeeds a relative label offset from position.
1310 : // Patch the relative offset to be relative to the Code object pointer
1311 : // instead.
1312 595574 : int patch_position = position - kIntSize;
1313 595574 : int offset = masm_.long_at(patch_position);
1314 : masm_.long_at_put(patch_position,
1315 : offset
1316 595574 : + position
1317 : + Code::kHeaderSize
1318 595574 : - kHeapObjectTag);
1319 : }
1320 85333 : code_relative_fixup_positions_.Rewind(0);
1321 85333 : }
1322 :
1323 :
1324 595574 : void RegExpMacroAssemblerX64::Push(Label* backtrack_target) {
1325 595574 : __ subp(backtrack_stackpointer(), Immediate(kIntSize));
1326 595574 : __ movl(Operand(backtrack_stackpointer(), 0), backtrack_target);
1327 : MarkPositionForCodeRelativeFixup();
1328 595574 : }
1329 :
1330 :
1331 666337 : void RegExpMacroAssemblerX64::Pop(Register target) {
1332 : DCHECK(target != backtrack_stackpointer());
1333 666337 : __ movsxlq(target, Operand(backtrack_stackpointer(), 0));
1334 : // Notice: This updates flags, unlike normal Pop.
1335 666337 : __ addp(backtrack_stackpointer(), Immediate(kIntSize));
1336 666337 : }
1337 :
1338 :
1339 : void RegExpMacroAssemblerX64::Drop() {
1340 11700 : __ addp(backtrack_stackpointer(), Immediate(kIntSize));
1341 : }
1342 :
1343 :
1344 214255 : void RegExpMacroAssemblerX64::CheckPreemption() {
1345 : // Check for preemption.
1346 214255 : Label no_preempt;
1347 : ExternalReference stack_limit =
1348 214255 : ExternalReference::address_of_stack_limit(isolate());
1349 214255 : __ load_rax(stack_limit);
1350 214255 : __ cmpp(rsp, rax);
1351 214255 : __ j(above, &no_preempt);
1352 :
1353 214255 : SafeCall(&check_preempt_label_);
1354 :
1355 214255 : __ bind(&no_preempt);
1356 214255 : }
1357 :
1358 :
1359 597884 : void RegExpMacroAssemblerX64::CheckStackLimit() {
1360 597884 : Label no_stack_overflow;
1361 : ExternalReference stack_limit =
1362 597884 : ExternalReference::address_of_regexp_stack_limit(isolate());
1363 597884 : __ load_rax(stack_limit);
1364 597884 : __ cmpp(backtrack_stackpointer(), rax);
1365 597884 : __ j(above, &no_stack_overflow);
1366 :
1367 597884 : SafeCall(&stack_overflow_label_);
1368 :
1369 597884 : __ bind(&no_stack_overflow);
1370 597884 : }
1371 :
1372 :
1373 985919 : void RegExpMacroAssemblerX64::LoadCurrentCharacterUnchecked(int cp_offset,
1374 : int characters) {
1375 985919 : if (mode_ == LATIN1) {
1376 709441 : if (characters == 4) {
1377 7192 : __ movl(current_character(), Operand(rsi, rdi, times_1, cp_offset));
1378 705845 : } else if (characters == 2) {
1379 317522 : __ movzxwl(current_character(), Operand(rsi, rdi, times_1, cp_offset));
1380 : } else {
1381 : DCHECK_EQ(1, characters);
1382 1094168 : __ movzxbl(current_character(), Operand(rsi, rdi, times_1, cp_offset));
1383 : }
1384 : } else {
1385 : DCHECK(mode_ == UC16);
1386 276478 : if (characters == 2) {
1387 : __ movl(current_character(),
1388 29622 : Operand(rsi, rdi, times_1, cp_offset * sizeof(uc16)));
1389 : } else {
1390 : DCHECK_EQ(1, characters);
1391 : __ movzxwl(current_character(),
1392 523334 : Operand(rsi, rdi, times_1, cp_offset * sizeof(uc16)));
1393 : }
1394 : }
1395 985919 : }
1396 :
1397 : #undef __
1398 :
1399 : #endif // V8_INTERPRETED_REGEXP
1400 :
1401 : } // namespace internal
1402 183867 : } // namespace v8
1403 :
1404 : #endif // V8_TARGET_ARCH_X64
|