1// Copyright 2012 the V8 project authors. All rights reserved.
2// Use of this source code is governed by a BSD-style license that can be
3// found in the LICENSE file.
4
5#if V8_TARGET_ARCH_ARM
6
7#include "src/regexp/arm/regexp-macro-assembler-arm.h"
8
9#include "src/code-stubs.h"
10#include "src/log.h"
11#include "src/macro-assembler.h"
12#include "src/regexp/regexp-macro-assembler.h"
13#include "src/regexp/regexp-stack.h"
14#include "src/unicode.h"
15
16namespace v8 {
17namespace internal {
18
19#ifndef V8_INTERPRETED_REGEXP
20/*
21 * This assembler uses the following register assignment convention
22 * - r4 : Temporarily stores the index of capture start after a matching pass
23 *        for a global regexp.
24 * - r5 : Pointer to current code object (Code*) including heap object tag.
25 * - r6 : 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 * - r7 : Currently loaded character. Must be loaded using
28 *        LoadCurrentCharacter before using any of the dispatch methods.
29 * - r8 : Points to tip of backtrack stack
30 * - r9 : Unused, might be used by C code and expected unchanged.
31 * - r10 : End of input (points to byte after last character in input).
32 * - r11 : Frame pointer. Used to access arguments, local variables and
33 *         RegExp registers.
34 * - r12 : IP register, used by assembler. Very volatile.
35 * - r13/sp : Points to tip of C stack.
36 *
37 * The remaining registers are free for computations.
38 * Each call to a public method should retain this convention.
39 *
40 * The stack will have the following structure:
41 *  - fp[56]  Isolate* isolate   (address of the current isolate)
42 *  - fp[52]  direct_call        (if 1, direct call from JavaScript code,
43 *                                if 0, call through the runtime system).
44 *  - fp[48]  stack_area_base    (high end of the memory area to use as
45 *                                backtracking stack).
46 *  - fp[44]  capture array size (may fit multiple sets of matches)
47 *  - fp[40]  int* capture_array (int[num_saved_registers_], for output).
48 *  - fp[36]  secondary link/return address used by native call.
49 *  --- sp when called ---
50 *  - fp[32]  return address     (lr).
51 *  - fp[28]  old frame pointer  (r11).
52 *  - fp[0..24]  backup of registers r4..r10.
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 *            string start - 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 arm/simulator-arm.h.
88 * When calling as a non-direct call (i.e., from C++ code), the return address
89 * area is overwritten with the LR 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
96RegExpMacroAssemblerARM::RegExpMacroAssemblerARM(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  DCHECK_EQ(0, registers_to_save % 2);
111  __ jmp(&entry_label_);   // We'll write the entry code later.
112  __ bind(&start_label_);  // And then continue from here.
113}
114
115
116RegExpMacroAssemblerARM::~RegExpMacroAssemblerARM() {
117  delete masm_;
118  // Unuse labels in case we throw away the assembler without calling GetCode.
119  entry_label_.Unuse();
120  start_label_.Unuse();
121  success_label_.Unuse();
122  backtrack_label_.Unuse();
123  exit_label_.Unuse();
124  check_preempt_label_.Unuse();
125  stack_overflow_label_.Unuse();
126}
127
128
129int RegExpMacroAssemblerARM::stack_limit_slack()  {
130  return RegExpStack::kStackLimitSlack;
131}
132
133
134void RegExpMacroAssemblerARM::AdvanceCurrentPosition(int by) {
135  if (by != 0) {
136    __ add(current_input_offset(),
137           current_input_offset(), Operand(by * char_size()));
138  }
139}
140
141
142void RegExpMacroAssemblerARM::AdvanceRegister(int reg, int by) {
143  DCHECK(reg >= 0);
144  DCHECK(reg < num_registers_);
145  if (by != 0) {
146    __ ldr(r0, register_location(reg));
147    __ add(r0, r0, Operand(by));
148    __ str(r0, register_location(reg));
149  }
150}
151
152
153void RegExpMacroAssemblerARM::Backtrack() {
154  CheckPreemption();
155  // Pop Code* offset from backtrack stack, add Code* and jump to location.
156  Pop(r0);
157  __ add(pc, r0, Operand(code_pointer()));
158}
159
160
161void RegExpMacroAssemblerARM::Bind(Label* label) {
162  __ bind(label);
163}
164
165
166void RegExpMacroAssemblerARM::CheckCharacter(uint32_t c, Label* on_equal) {
167  __ cmp(current_character(), Operand(c));
168  BranchOrBacktrack(eq, on_equal);
169}
170
171
172void RegExpMacroAssemblerARM::CheckCharacterGT(uc16 limit, Label* on_greater) {
173  __ cmp(current_character(), Operand(limit));
174  BranchOrBacktrack(gt, on_greater);
175}
176
177
178void RegExpMacroAssemblerARM::CheckAtStart(Label* on_at_start) {
179  __ ldr(r1, MemOperand(frame_pointer(), kStringStartMinusOne));
180  __ add(r0, current_input_offset(), Operand(-char_size()));
181  __ cmp(r0, r1);
182  BranchOrBacktrack(eq, on_at_start);
183}
184
185
186void RegExpMacroAssemblerARM::CheckNotAtStart(int cp_offset,
187                                              Label* on_not_at_start) {
188  __ ldr(r1, MemOperand(frame_pointer(), kStringStartMinusOne));
189  __ add(r0, current_input_offset(),
190         Operand(-char_size() + cp_offset * char_size()));
191  __ cmp(r0, r1);
192  BranchOrBacktrack(ne, on_not_at_start);
193}
194
195
196void RegExpMacroAssemblerARM::CheckCharacterLT(uc16 limit, Label* on_less) {
197  __ cmp(current_character(), Operand(limit));
198  BranchOrBacktrack(lt, on_less);
199}
200
201
202void RegExpMacroAssemblerARM::CheckGreedyLoop(Label* on_equal) {
203  __ ldr(r0, MemOperand(backtrack_stackpointer(), 0));
204  __ cmp(current_input_offset(), r0);
205  __ add(backtrack_stackpointer(),
206         backtrack_stackpointer(), Operand(kPointerSize), LeaveCC, eq);
207  BranchOrBacktrack(eq, on_equal);
208}
209
210
211void RegExpMacroAssemblerARM::CheckNotBackReferenceIgnoreCase(
212    int start_reg, bool read_backward, bool unicode, Label* on_no_match) {
213  Label fallthrough;
214  __ ldr(r0, register_location(start_reg));  // Index of start of capture
215  __ ldr(r1, register_location(start_reg + 1));  // Index of end of capture
216  __ sub(r1, r1, r0, SetCC);  // Length of capture.
217
218  // At this point, the capture registers are either both set or both cleared.
219  // If the capture length is zero, then the capture is either empty or cleared.
220  // Fall through in both cases.
221  __ b(eq, &fallthrough);
222
223  // Check that there are enough characters left in the input.
224  if (read_backward) {
225    __ ldr(r3, MemOperand(frame_pointer(), kStringStartMinusOne));
226    __ add(r3, r3, r1);
227    __ cmp(current_input_offset(), r3);
228    BranchOrBacktrack(le, on_no_match);
229  } else {
230    __ cmn(r1, Operand(current_input_offset()));
231    BranchOrBacktrack(gt, on_no_match);
232  }
233
234  if (mode_ == LATIN1) {
235    Label success;
236    Label fail;
237    Label loop_check;
238
239    // r0 - offset of start of capture
240    // r1 - length of capture
241    __ add(r0, r0, end_of_input_address());
242    __ add(r2, end_of_input_address(), current_input_offset());
243    if (read_backward) {
244      __ sub(r2, r2, r1);  // Offset by length when matching backwards.
245    }
246    __ add(r1, r0, r1);
247
248    // r0 - Address of start of capture.
249    // r1 - Address of end of capture
250    // r2 - Address of current input position.
251
252    Label loop;
253    __ bind(&loop);
254    __ ldrb(r3, MemOperand(r0, char_size(), PostIndex));
255    __ ldrb(r4, MemOperand(r2, char_size(), PostIndex));
256    __ cmp(r4, r3);
257    __ b(eq, &loop_check);
258
259    // Mismatch, try case-insensitive match (converting letters to lower-case).
260    __ orr(r3, r3, Operand(0x20));  // Convert capture character to lower-case.
261    __ orr(r4, r4, Operand(0x20));  // Also convert input character.
262    __ cmp(r4, r3);
263    __ b(ne, &fail);
264    __ sub(r3, r3, Operand('a'));
265    __ cmp(r3, Operand('z' - 'a'));  // Is r3 a lowercase letter?
266    __ b(ls, &loop_check);  // In range 'a'-'z'.
267    // Latin-1: Check for values in range [224,254] but not 247.
268    __ sub(r3, r3, Operand(224 - 'a'));
269    __ cmp(r3, Operand(254 - 224));
270    __ b(hi, &fail);  // Weren't Latin-1 letters.
271    __ cmp(r3, Operand(247 - 224));  // Check for 247.
272    __ b(eq, &fail);
273
274    __ bind(&loop_check);
275    __ cmp(r0, r1);
276    __ b(lt, &loop);
277    __ jmp(&success);
278
279    __ bind(&fail);
280    BranchOrBacktrack(al, on_no_match);
281
282    __ bind(&success);
283    // Compute new value of character position after the matched part.
284    __ sub(current_input_offset(), r2, end_of_input_address());
285    if (read_backward) {
286      __ ldr(r0, register_location(start_reg));  // Index of start of capture
287      __ ldr(r1, register_location(start_reg + 1));  // Index of end of capture
288      __ add(current_input_offset(), current_input_offset(), r0);
289      __ sub(current_input_offset(), current_input_offset(), r1);
290    }
291  } else {
292    DCHECK(mode_ == UC16);
293    int argument_count = 4;
294    __ PrepareCallCFunction(argument_count, r2);
295
296    // r0 - offset of start of capture
297    // r1 - length of capture
298
299    // Put arguments into arguments registers.
300    // Parameters are
301    //   r0: Address byte_offset1 - Address captured substring's start.
302    //   r1: Address byte_offset2 - Address of current character position.
303    //   r2: size_t byte_length - length of capture in bytes(!)
304    //   r3: Isolate* isolate or 0 if unicode flag.
305
306    // Address of start of capture.
307    __ add(r0, r0, Operand(end_of_input_address()));
308    // Length of capture.
309    __ mov(r2, Operand(r1));
310    // Save length in callee-save register for use on return.
311    __ mov(r4, Operand(r1));
312    // Address of current input position.
313    __ add(r1, current_input_offset(), end_of_input_address());
314    if (read_backward) {
315      __ sub(r1, r1, r4);
316    }
317    // Isolate.
318#ifdef V8_I18N_SUPPORT
319    if (unicode) {
320      __ mov(r3, Operand(0));
321    } else  // NOLINT
322#endif      // V8_I18N_SUPPORT
323    {
324      __ mov(r3, Operand(ExternalReference::isolate_address(isolate())));
325    }
326
327    {
328      AllowExternalCallThatCantCauseGC scope(masm_);
329      ExternalReference function =
330          ExternalReference::re_case_insensitive_compare_uc16(isolate());
331      __ CallCFunction(function, argument_count);
332    }
333
334    // Check if function returned non-zero for success or zero for failure.
335    __ cmp(r0, Operand::Zero());
336    BranchOrBacktrack(eq, on_no_match);
337
338    // On success, advance position by length of capture.
339    if (read_backward) {
340      __ sub(current_input_offset(), current_input_offset(), r4);
341    } else {
342      __ add(current_input_offset(), current_input_offset(), r4);
343    }
344  }
345
346  __ bind(&fallthrough);
347}
348
349
350void RegExpMacroAssemblerARM::CheckNotBackReference(int start_reg,
351                                                    bool read_backward,
352                                                    Label* on_no_match) {
353  Label fallthrough;
354  Label success;
355
356  // Find length of back-referenced capture.
357  __ ldr(r0, register_location(start_reg));
358  __ ldr(r1, register_location(start_reg + 1));
359  __ sub(r1, r1, r0, SetCC);  // Length to check.
360
361  // At this point, the capture registers are either both set or both cleared.
362  // If the capture length is zero, then the capture is either empty or cleared.
363  // Fall through in both cases.
364  __ b(eq, &fallthrough);
365
366  // Check that there are enough characters left in the input.
367  if (read_backward) {
368    __ ldr(r3, MemOperand(frame_pointer(), kStringStartMinusOne));
369    __ add(r3, r3, r1);
370    __ cmp(current_input_offset(), r3);
371    BranchOrBacktrack(lt, on_no_match);
372  } else {
373    __ cmn(r1, Operand(current_input_offset()));
374    BranchOrBacktrack(gt, on_no_match);
375  }
376
377  // r0 - offset of start of capture
378  // r1 - length of capture
379  __ add(r0, r0, end_of_input_address());
380  __ add(r2, end_of_input_address(), current_input_offset());
381  if (read_backward) {
382    __ sub(r2, r2, r1);  // Offset by length when matching backwards.
383  }
384  __ add(r1, r0, r1);
385
386  Label loop;
387  __ bind(&loop);
388  if (mode_ == LATIN1) {
389    __ ldrb(r3, MemOperand(r0, char_size(), PostIndex));
390    __ ldrb(r4, MemOperand(r2, char_size(), PostIndex));
391  } else {
392    DCHECK(mode_ == UC16);
393    __ ldrh(r3, MemOperand(r0, char_size(), PostIndex));
394    __ ldrh(r4, MemOperand(r2, char_size(), PostIndex));
395  }
396  __ cmp(r3, r4);
397  BranchOrBacktrack(ne, on_no_match);
398  __ cmp(r0, r1);
399  __ b(lt, &loop);
400
401  // Move current character position to position after match.
402  __ sub(current_input_offset(), r2, end_of_input_address());
403  if (read_backward) {
404    __ ldr(r0, register_location(start_reg));      // Index of start of capture
405    __ ldr(r1, register_location(start_reg + 1));  // Index of end of capture
406    __ add(current_input_offset(), current_input_offset(), r0);
407    __ sub(current_input_offset(), current_input_offset(), r1);
408  }
409
410  __ bind(&fallthrough);
411}
412
413
414void RegExpMacroAssemblerARM::CheckNotCharacter(unsigned c,
415                                                Label* on_not_equal) {
416  __ cmp(current_character(), Operand(c));
417  BranchOrBacktrack(ne, on_not_equal);
418}
419
420
421void RegExpMacroAssemblerARM::CheckCharacterAfterAnd(uint32_t c,
422                                                     uint32_t mask,
423                                                     Label* on_equal) {
424  if (c == 0) {
425    __ tst(current_character(), Operand(mask));
426  } else {
427    __ and_(r0, current_character(), Operand(mask));
428    __ cmp(r0, Operand(c));
429  }
430  BranchOrBacktrack(eq, on_equal);
431}
432
433
434void RegExpMacroAssemblerARM::CheckNotCharacterAfterAnd(unsigned c,
435                                                        unsigned mask,
436                                                        Label* on_not_equal) {
437  if (c == 0) {
438    __ tst(current_character(), Operand(mask));
439  } else {
440    __ and_(r0, current_character(), Operand(mask));
441    __ cmp(r0, Operand(c));
442  }
443  BranchOrBacktrack(ne, on_not_equal);
444}
445
446
447void RegExpMacroAssemblerARM::CheckNotCharacterAfterMinusAnd(
448    uc16 c,
449    uc16 minus,
450    uc16 mask,
451    Label* on_not_equal) {
452  DCHECK(minus < String::kMaxUtf16CodeUnit);
453  __ sub(r0, current_character(), Operand(minus));
454  __ and_(r0, r0, Operand(mask));
455  __ cmp(r0, Operand(c));
456  BranchOrBacktrack(ne, on_not_equal);
457}
458
459
460void RegExpMacroAssemblerARM::CheckCharacterInRange(
461    uc16 from,
462    uc16 to,
463    Label* on_in_range) {
464  __ sub(r0, current_character(), Operand(from));
465  __ cmp(r0, Operand(to - from));
466  BranchOrBacktrack(ls, on_in_range);  // Unsigned lower-or-same condition.
467}
468
469
470void RegExpMacroAssemblerARM::CheckCharacterNotInRange(
471    uc16 from,
472    uc16 to,
473    Label* on_not_in_range) {
474  __ sub(r0, current_character(), Operand(from));
475  __ cmp(r0, Operand(to - from));
476  BranchOrBacktrack(hi, on_not_in_range);  // Unsigned higher condition.
477}
478
479
480void RegExpMacroAssemblerARM::CheckBitInTable(
481    Handle<ByteArray> table,
482    Label* on_bit_set) {
483  __ mov(r0, Operand(table));
484  if (mode_ != LATIN1 || kTableMask != String::kMaxOneByteCharCode) {
485    __ and_(r1, current_character(), Operand(kTableSize - 1));
486    __ add(r1, r1, Operand(ByteArray::kHeaderSize - kHeapObjectTag));
487  } else {
488    __ add(r1,
489           current_character(),
490           Operand(ByteArray::kHeaderSize - kHeapObjectTag));
491  }
492  __ ldrb(r0, MemOperand(r0, r1));
493  __ cmp(r0, Operand::Zero());
494  BranchOrBacktrack(ne, on_bit_set);
495}
496
497
498bool RegExpMacroAssemblerARM::CheckSpecialCharacterClass(uc16 type,
499                                                         Label* on_no_match) {
500  // Range checks (c in min..max) are generally implemented by an unsigned
501  // (c - min) <= (max - min) check
502  switch (type) {
503  case 's':
504    // Match space-characters
505    if (mode_ == LATIN1) {
506      // One byte space characters are '\t'..'\r', ' ' and \u00a0.
507      Label success;
508      __ cmp(current_character(), Operand(' '));
509      __ b(eq, &success);
510      // Check range 0x09..0x0d
511      __ sub(r0, current_character(), Operand('\t'));
512      __ cmp(r0, Operand('\r' - '\t'));
513      __ b(ls, &success);
514      // \u00a0 (NBSP).
515      __ cmp(r0, Operand(0x00a0 - '\t'));
516      BranchOrBacktrack(ne, on_no_match);
517      __ bind(&success);
518      return true;
519    }
520    return false;
521  case 'S':
522    // The emitted code for generic character classes is good enough.
523    return false;
524  case 'd':
525    // Match ASCII digits ('0'..'9')
526    __ sub(r0, current_character(), Operand('0'));
527    __ cmp(r0, Operand('9' - '0'));
528    BranchOrBacktrack(hi, on_no_match);
529    return true;
530  case 'D':
531    // Match non ASCII-digits
532    __ sub(r0, current_character(), Operand('0'));
533    __ cmp(r0, Operand('9' - '0'));
534    BranchOrBacktrack(ls, on_no_match);
535    return true;
536  case '.': {
537    // Match non-newlines (not 0x0a('\n'), 0x0d('\r'), 0x2028 and 0x2029)
538    __ eor(r0, current_character(), Operand(0x01));
539    // See if current character is '\n'^1 or '\r'^1, i.e., 0x0b or 0x0c
540    __ sub(r0, r0, Operand(0x0b));
541    __ cmp(r0, Operand(0x0c - 0x0b));
542    BranchOrBacktrack(ls, on_no_match);
543    if (mode_ == UC16) {
544      // Compare original value to 0x2028 and 0x2029, using the already
545      // computed (current_char ^ 0x01 - 0x0b). I.e., check for
546      // 0x201d (0x2028 - 0x0b) or 0x201e.
547      __ sub(r0, r0, Operand(0x2028 - 0x0b));
548      __ cmp(r0, Operand(1));
549      BranchOrBacktrack(ls, on_no_match);
550    }
551    return true;
552  }
553  case 'n': {
554    // Match newlines (0x0a('\n'), 0x0d('\r'), 0x2028 and 0x2029)
555    __ eor(r0, current_character(), Operand(0x01));
556    // See if current character is '\n'^1 or '\r'^1, i.e., 0x0b or 0x0c
557    __ sub(r0, r0, Operand(0x0b));
558    __ cmp(r0, Operand(0x0c - 0x0b));
559    if (mode_ == LATIN1) {
560      BranchOrBacktrack(hi, on_no_match);
561    } else {
562      Label done;
563      __ b(ls, &done);
564      // Compare original value to 0x2028 and 0x2029, using the already
565      // computed (current_char ^ 0x01 - 0x0b). I.e., check for
566      // 0x201d (0x2028 - 0x0b) or 0x201e.
567      __ sub(r0, r0, Operand(0x2028 - 0x0b));
568      __ cmp(r0, Operand(1));
569      BranchOrBacktrack(hi, on_no_match);
570      __ bind(&done);
571    }
572    return true;
573  }
574  case 'w': {
575    if (mode_ != LATIN1) {
576      // Table is 256 entries, so all Latin1 characters can be tested.
577      __ cmp(current_character(), Operand('z'));
578      BranchOrBacktrack(hi, on_no_match);
579    }
580    ExternalReference map = ExternalReference::re_word_character_map();
581    __ mov(r0, Operand(map));
582    __ ldrb(r0, MemOperand(r0, current_character()));
583    __ cmp(r0, Operand::Zero());
584    BranchOrBacktrack(eq, on_no_match);
585    return true;
586  }
587  case 'W': {
588    Label done;
589    if (mode_ != LATIN1) {
590      // Table is 256 entries, so all Latin1 characters can be tested.
591      __ cmp(current_character(), Operand('z'));
592      __ b(hi, &done);
593    }
594    ExternalReference map = ExternalReference::re_word_character_map();
595    __ mov(r0, Operand(map));
596    __ ldrb(r0, MemOperand(r0, current_character()));
597    __ cmp(r0, Operand::Zero());
598    BranchOrBacktrack(ne, on_no_match);
599    if (mode_ != LATIN1) {
600      __ bind(&done);
601    }
602    return true;
603  }
604  case '*':
605    // Match any character.
606    return true;
607  // No custom implementation (yet): s(UC16), S(UC16).
608  default:
609    return false;
610  }
611}
612
613
614void RegExpMacroAssemblerARM::Fail() {
615  __ mov(r0, Operand(FAILURE));
616  __ jmp(&exit_label_);
617}
618
619
620Handle<HeapObject> RegExpMacroAssemblerARM::GetCode(Handle<String> source) {
621  Label return_r0;
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, no
629  // 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 = r4.bit() | r5.bit() | r6.bit() |
639      r7.bit() | r8.bit() | r9.bit() | r10.bit() | fp.bit();
640  RegList argument_registers = r0.bit() | r1.bit() | r2.bit() | r3.bit();
641  __ stm(db_w, sp, argument_registers | registers_to_retain | lr.bit());
642  // Set frame pointer in space for it if this is not a direct call
643  // from generated code.
644  __ add(frame_pointer(), sp, Operand(4 * kPointerSize));
645  __ mov(r0, Operand::Zero());
646  __ push(r0);  // Make room for success counter and initialize it to 0.
647  __ push(r0);  // Make room for "string start - 1" constant.
648  // Check if we have space on the stack for registers.
649  Label stack_limit_hit;
650  Label stack_ok;
651
652  ExternalReference stack_limit =
653      ExternalReference::address_of_stack_limit(isolate());
654  __ mov(r0, Operand(stack_limit));
655  __ ldr(r0, MemOperand(r0));
656  __ sub(r0, sp, r0, SetCC);
657  // Handle it if the stack pointer is already below the stack limit.
658  __ b(ls, &stack_limit_hit);
659  // Check if there is room for the variable number of registers above
660  // the stack limit.
661  __ cmp(r0, Operand(num_registers_ * kPointerSize));
662  __ b(hs, &stack_ok);
663  // Exit with OutOfMemory exception. There is not enough space on the stack
664  // for our working registers.
665  __ mov(r0, Operand(EXCEPTION));
666  __ jmp(&return_r0);
667
668  __ bind(&stack_limit_hit);
669  CallCheckStackGuardState(r0);
670  __ cmp(r0, Operand::Zero());
671  // If returned value is non-zero, we exit with the returned value as result.
672  __ b(ne, &return_r0);
673
674  __ bind(&stack_ok);
675
676  // Allocate space on stack for registers.
677  __ sub(sp, sp, Operand(num_registers_ * kPointerSize));
678  // Load string end.
679  __ ldr(end_of_input_address(), MemOperand(frame_pointer(), kInputEnd));
680  // Load input start.
681  __ ldr(r0, MemOperand(frame_pointer(), kInputStart));
682  // Find negative length (offset of start relative to end).
683  __ sub(current_input_offset(), r0, end_of_input_address());
684  // Set r0 to address of char before start of the input string
685  // (effectively string position -1).
686  __ ldr(r1, MemOperand(frame_pointer(), kStartIndex));
687  __ sub(r0, current_input_offset(), Operand(char_size()));
688  __ sub(r0, r0, Operand(r1, LSL, (mode_ == UC16) ? 1 : 0));
689  // Store this value in a local variable, for use when clearing
690  // position registers.
691  __ str(r0, MemOperand(frame_pointer(), kStringStartMinusOne));
692
693  // Initialize code pointer register
694  __ mov(code_pointer(), Operand(masm_->CodeObject()));
695
696  Label load_char_start_regexp, start_regexp;
697  // Load newline if index is at start, previous character otherwise.
698  __ cmp(r1, Operand::Zero());
699  __ b(ne, &load_char_start_regexp);
700  __ mov(current_character(), Operand('\n'), LeaveCC, eq);
701  __ jmp(&start_regexp);
702
703  // Global regexp restarts matching here.
704  __ bind(&load_char_start_regexp);
705  // Load previous char as initial value of current character register.
706  LoadCurrentCharacterUnchecked(-1, 1);
707  __ bind(&start_regexp);
708
709  // Initialize on-stack registers.
710  if (num_saved_registers_ > 0) {  // Always is, if generated from a regexp.
711    // Fill saved registers with initial value = start offset - 1
712    if (num_saved_registers_ > 8) {
713      // Address of register 0.
714      __ add(r1, frame_pointer(), Operand(kRegisterZero));
715      __ mov(r2, Operand(num_saved_registers_));
716      Label init_loop;
717      __ bind(&init_loop);
718      __ str(r0, MemOperand(r1, kPointerSize, NegPostIndex));
719      __ sub(r2, r2, Operand(1), SetCC);
720      __ b(ne, &init_loop);
721    } else {
722      for (int i = 0; i < num_saved_registers_; i++) {
723        __ str(r0, register_location(i));
724      }
725    }
726  }
727
728  // Initialize backtrack stack pointer.
729  __ ldr(backtrack_stackpointer(), MemOperand(frame_pointer(), kStackHighEnd));
730
731  __ jmp(&start_label_);
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      __ ldr(r1, MemOperand(frame_pointer(), kInputStart));
740      __ ldr(r0, MemOperand(frame_pointer(), kRegisterOutput));
741      __ ldr(r2, MemOperand(frame_pointer(), kStartIndex));
742      __ sub(r1, end_of_input_address(), r1);
743      // r1 is length of input in bytes.
744      if (mode_ == UC16) {
745        __ mov(r1, Operand(r1, LSR, 1));
746      }
747      // r1 is length of input in characters.
748      __ add(r1, r1, Operand(r2));
749      // r1 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        __ ldr(r2, register_location(i));
757        __ ldr(r3, register_location(i + 1));
758        if (i == 0 && global_with_zero_length_check()) {
759          // Keep capture start in r4 for the zero-length check later.
760          __ mov(r4, r2);
761        }
762        if (mode_ == UC16) {
763          __ add(r2, r1, Operand(r2, ASR, 1));
764          __ add(r3, r1, Operand(r3, ASR, 1));
765        } else {
766          __ add(r2, r1, Operand(r2));
767          __ add(r3, r1, Operand(r3));
768        }
769        __ str(r2, MemOperand(r0, kPointerSize, PostIndex));
770        __ str(r3, MemOperand(r0, kPointerSize, PostIndex));
771      }
772    }
773
774    if (global()) {
775      // Restart matching if the regular expression is flagged as global.
776      __ ldr(r0, MemOperand(frame_pointer(), kSuccessfulCaptures));
777      __ ldr(r1, MemOperand(frame_pointer(), kNumOutputRegisters));
778      __ ldr(r2, MemOperand(frame_pointer(), kRegisterOutput));
779      // Increment success counter.
780      __ add(r0, r0, Operand(1));
781      __ str(r0, MemOperand(frame_pointer(), kSuccessfulCaptures));
782      // Capture results have been stored, so the number of remaining global
783      // output registers is reduced by the number of stored captures.
784      __ sub(r1, r1, Operand(num_saved_registers_));
785      // Check whether we have enough room for another set of capture results.
786      __ cmp(r1, Operand(num_saved_registers_));
787      __ b(lt, &return_r0);
788
789      __ str(r1, MemOperand(frame_pointer(), kNumOutputRegisters));
790      // Advance the location for output.
791      __ add(r2, r2, Operand(num_saved_registers_ * kPointerSize));
792      __ str(r2, MemOperand(frame_pointer(), kRegisterOutput));
793
794      // Prepare r0 to initialize registers with its value in the next run.
795      __ ldr(r0, MemOperand(frame_pointer(), kStringStartMinusOne));
796
797      if (global_with_zero_length_check()) {
798        // Special case for zero-length matches.
799        // r4: capture start index
800        __ cmp(current_input_offset(), r4);
801        // Not a zero-length match, restart.
802        __ b(ne, &load_char_start_regexp);
803        // Offset from the end is zero if we already reached the end.
804        __ cmp(current_input_offset(), Operand::Zero());
805        __ b(eq, &exit_label_);
806        // Advance current position after a zero-length match.
807        Label advance;
808        __ bind(&advance);
809        __ add(current_input_offset(),
810               current_input_offset(),
811               Operand((mode_ == UC16) ? 2 : 1));
812        if (global_unicode()) CheckNotInSurrogatePair(0, &advance);
813      }
814
815      __ b(&load_char_start_regexp);
816    } else {
817      __ mov(r0, Operand(SUCCESS));
818    }
819  }
820
821  // Exit and return r0
822  __ bind(&exit_label_);
823  if (global()) {
824    __ ldr(r0, MemOperand(frame_pointer(), kSuccessfulCaptures));
825  }
826
827  __ bind(&return_r0);
828  // Skip sp past regexp registers and local variables..
829  __ mov(sp, frame_pointer());
830  // Restore registers r4..r11 and return (restoring lr to pc).
831  __ ldm(ia_w, sp, registers_to_retain | pc.bit());
832
833  // Backtrack code (branch target for conditional backtracks).
834  if (backtrack_label_.is_linked()) {
835    __ bind(&backtrack_label_);
836    Backtrack();
837  }
838
839  Label exit_with_exception;
840
841  // Preempt-code
842  if (check_preempt_label_.is_linked()) {
843    SafeCallTarget(&check_preempt_label_);
844
845    CallCheckStackGuardState(r0);
846    __ cmp(r0, Operand::Zero());
847    // If returning non-zero, we should end execution with the given
848    // result as return value.
849    __ b(ne, &return_r0);
850
851    // String might have moved: Reload end of string from frame.
852    __ ldr(end_of_input_address(), MemOperand(frame_pointer(), kInputEnd));
853    SafeReturn();
854  }
855
856  // Backtrack stack overflow code.
857  if (stack_overflow_label_.is_linked()) {
858    SafeCallTarget(&stack_overflow_label_);
859    // Reached if the backtrack-stack limit has been hit.
860    Label grow_failed;
861
862    // Call GrowStack(backtrack_stackpointer(), &stack_base)
863    static const int num_arguments = 3;
864    __ PrepareCallCFunction(num_arguments, r0);
865    __ mov(r0, backtrack_stackpointer());
866    __ add(r1, frame_pointer(), Operand(kStackHighEnd));
867    __ mov(r2, Operand(ExternalReference::isolate_address(isolate())));
868    ExternalReference grow_stack =
869        ExternalReference::re_grow_stack(isolate());
870    __ CallCFunction(grow_stack, num_arguments);
871    // If return NULL, we have failed to grow the stack, and
872    // must exit with a stack-overflow exception.
873    __ cmp(r0, Operand::Zero());
874    __ b(eq, &exit_with_exception);
875    // Otherwise use return value as new stack pointer.
876    __ mov(backtrack_stackpointer(), r0);
877    // Restore saved registers and continue.
878    SafeReturn();
879  }
880
881  if (exit_with_exception.is_linked()) {
882    // If any of the code above needed to exit with an exception.
883    __ bind(&exit_with_exception);
884    // Exit with Result EXCEPTION(-1) to signal thrown exception.
885    __ mov(r0, Operand(EXCEPTION));
886    __ jmp(&return_r0);
887  }
888
889  CodeDesc code_desc;
890  masm_->GetCode(&code_desc);
891  Handle<Code> code = isolate()->factory()->NewCode(
892      code_desc, Code::ComputeFlags(Code::REGEXP), masm_->CodeObject());
893  PROFILE(masm_->isolate(),
894          RegExpCodeCreateEvent(AbstractCode::cast(*code), *source));
895  return Handle<HeapObject>::cast(code);
896}
897
898
899void RegExpMacroAssemblerARM::GoTo(Label* to) {
900  BranchOrBacktrack(al, to);
901}
902
903
904void RegExpMacroAssemblerARM::IfRegisterGE(int reg,
905                                           int comparand,
906                                           Label* if_ge) {
907  __ ldr(r0, register_location(reg));
908  __ cmp(r0, Operand(comparand));
909  BranchOrBacktrack(ge, if_ge);
910}
911
912
913void RegExpMacroAssemblerARM::IfRegisterLT(int reg,
914                                           int comparand,
915                                           Label* if_lt) {
916  __ ldr(r0, register_location(reg));
917  __ cmp(r0, Operand(comparand));
918  BranchOrBacktrack(lt, if_lt);
919}
920
921
922void RegExpMacroAssemblerARM::IfRegisterEqPos(int reg,
923                                              Label* if_eq) {
924  __ ldr(r0, register_location(reg));
925  __ cmp(r0, Operand(current_input_offset()));
926  BranchOrBacktrack(eq, if_eq);
927}
928
929
930RegExpMacroAssembler::IrregexpImplementation
931    RegExpMacroAssemblerARM::Implementation() {
932  return kARMImplementation;
933}
934
935
936void RegExpMacroAssemblerARM::LoadCurrentCharacter(int cp_offset,
937                                                   Label* on_end_of_input,
938                                                   bool check_bounds,
939                                                   int characters) {
940  DCHECK(cp_offset < (1<<30));  // Be sane! (And ensure negation works)
941  if (check_bounds) {
942    if (cp_offset >= 0) {
943      CheckPosition(cp_offset + characters - 1, on_end_of_input);
944    } else {
945      CheckPosition(cp_offset, on_end_of_input);
946    }
947  }
948  LoadCurrentCharacterUnchecked(cp_offset, characters);
949}
950
951
952void RegExpMacroAssemblerARM::PopCurrentPosition() {
953  Pop(current_input_offset());
954}
955
956
957void RegExpMacroAssemblerARM::PopRegister(int register_index) {
958  Pop(r0);
959  __ str(r0, register_location(register_index));
960}
961
962
963void RegExpMacroAssemblerARM::PushBacktrack(Label* label) {
964  __ mov_label_offset(r0, label);
965  Push(r0);
966  CheckStackLimit();
967}
968
969
970void RegExpMacroAssemblerARM::PushCurrentPosition() {
971  Push(current_input_offset());
972}
973
974
975void RegExpMacroAssemblerARM::PushRegister(int register_index,
976                                           StackCheckFlag check_stack_limit) {
977  __ ldr(r0, register_location(register_index));
978  Push(r0);
979  if (check_stack_limit) CheckStackLimit();
980}
981
982
983void RegExpMacroAssemblerARM::ReadCurrentPositionFromRegister(int reg) {
984  __ ldr(current_input_offset(), register_location(reg));
985}
986
987
988void RegExpMacroAssemblerARM::ReadStackPointerFromRegister(int reg) {
989  __ ldr(backtrack_stackpointer(), register_location(reg));
990  __ ldr(r0, MemOperand(frame_pointer(), kStackHighEnd));
991  __ add(backtrack_stackpointer(), backtrack_stackpointer(), Operand(r0));
992}
993
994
995void RegExpMacroAssemblerARM::SetCurrentPositionFromEnd(int by) {
996  Label after_position;
997  __ cmp(current_input_offset(), Operand(-by * char_size()));
998  __ b(ge, &after_position);
999  __ mov(current_input_offset(), Operand(-by * char_size()));
1000  // On RegExp code entry (where this operation is used), the character before
1001  // the current position is expected to be already loaded.
1002  // We have advanced the position, so it's safe to read backwards.
1003  LoadCurrentCharacterUnchecked(-1, 1);
1004  __ bind(&after_position);
1005}
1006
1007
1008void RegExpMacroAssemblerARM::SetRegister(int register_index, int to) {
1009  DCHECK(register_index >= num_saved_registers_);  // Reserved for positions!
1010  __ mov(r0, Operand(to));
1011  __ str(r0, register_location(register_index));
1012}
1013
1014
1015bool RegExpMacroAssemblerARM::Succeed() {
1016  __ jmp(&success_label_);
1017  return global();
1018}
1019
1020
1021void RegExpMacroAssemblerARM::WriteCurrentPositionToRegister(int reg,
1022                                                             int cp_offset) {
1023  if (cp_offset == 0) {
1024    __ str(current_input_offset(), register_location(reg));
1025  } else {
1026    __ add(r0, current_input_offset(), Operand(cp_offset * char_size()));
1027    __ str(r0, register_location(reg));
1028  }
1029}
1030
1031
1032void RegExpMacroAssemblerARM::ClearRegisters(int reg_from, int reg_to) {
1033  DCHECK(reg_from <= reg_to);
1034  __ ldr(r0, MemOperand(frame_pointer(), kStringStartMinusOne));
1035  for (int reg = reg_from; reg <= reg_to; reg++) {
1036    __ str(r0, register_location(reg));
1037  }
1038}
1039
1040
1041void RegExpMacroAssemblerARM::WriteStackPointerToRegister(int reg) {
1042  __ ldr(r1, MemOperand(frame_pointer(), kStackHighEnd));
1043  __ sub(r0, backtrack_stackpointer(), r1);
1044  __ str(r0, register_location(reg));
1045}
1046
1047
1048// Private methods:
1049
1050void RegExpMacroAssemblerARM::CallCheckStackGuardState(Register scratch) {
1051  __ PrepareCallCFunction(3, scratch);
1052
1053  // RegExp code frame pointer.
1054  __ mov(r2, frame_pointer());
1055  // Code* of self.
1056  __ mov(r1, Operand(masm_->CodeObject()));
1057
1058  // We need to make room for the return address on the stack.
1059  int stack_alignment = base::OS::ActivationFrameAlignment();
1060  DCHECK(IsAligned(stack_alignment, kPointerSize));
1061  __ sub(sp, sp, Operand(stack_alignment));
1062
1063  // r0 will point to the return address, placed by DirectCEntry.
1064  __ mov(r0, sp);
1065
1066  ExternalReference stack_guard_check =
1067      ExternalReference::re_check_stack_guard_state(isolate());
1068  __ mov(ip, Operand(stack_guard_check));
1069  DirectCEntryStub stub(isolate());
1070  stub.GenerateCall(masm_, ip);
1071
1072  // Drop the return address from the stack.
1073  __ add(sp, sp, Operand(stack_alignment));
1074
1075  DCHECK(stack_alignment != 0);
1076  __ ldr(sp, MemOperand(sp, 0));
1077
1078  __ mov(code_pointer(), Operand(masm_->CodeObject()));
1079}
1080
1081
1082// Helper function for reading a value out of a stack frame.
1083template <typename T>
1084static T& frame_entry(Address re_frame, int frame_offset) {
1085  return reinterpret_cast<T&>(Memory::int32_at(re_frame + frame_offset));
1086}
1087
1088
1089template <typename T>
1090static T* frame_entry_address(Address re_frame, int frame_offset) {
1091  return reinterpret_cast<T*>(re_frame + frame_offset);
1092}
1093
1094
1095int RegExpMacroAssemblerARM::CheckStackGuardState(Address* return_address,
1096                                                  Code* re_code,
1097                                                  Address re_frame) {
1098  return NativeRegExpMacroAssembler::CheckStackGuardState(
1099      frame_entry<Isolate*>(re_frame, kIsolate),
1100      frame_entry<int>(re_frame, kStartIndex),
1101      frame_entry<int>(re_frame, kDirectCall) == 1, return_address, re_code,
1102      frame_entry_address<String*>(re_frame, kInputString),
1103      frame_entry_address<const byte*>(re_frame, kInputStart),
1104      frame_entry_address<const byte*>(re_frame, kInputEnd));
1105}
1106
1107
1108MemOperand RegExpMacroAssemblerARM::register_location(int register_index) {
1109  DCHECK(register_index < (1<<30));
1110  if (num_registers_ <= register_index) {
1111    num_registers_ = register_index + 1;
1112  }
1113  return MemOperand(frame_pointer(),
1114                    kRegisterZero - register_index * kPointerSize);
1115}
1116
1117
1118void RegExpMacroAssemblerARM::CheckPosition(int cp_offset,
1119                                            Label* on_outside_input) {
1120  if (cp_offset >= 0) {
1121    __ cmp(current_input_offset(), Operand(-cp_offset * char_size()));
1122    BranchOrBacktrack(ge, on_outside_input);
1123  } else {
1124    __ ldr(r1, MemOperand(frame_pointer(), kStringStartMinusOne));
1125    __ add(r0, current_input_offset(), Operand(cp_offset * char_size()));
1126    __ cmp(r0, r1);
1127    BranchOrBacktrack(le, on_outside_input);
1128  }
1129}
1130
1131
1132void RegExpMacroAssemblerARM::BranchOrBacktrack(Condition condition,
1133                                                Label* to) {
1134  if (condition == al) {  // Unconditional.
1135    if (to == NULL) {
1136      Backtrack();
1137      return;
1138    }
1139    __ jmp(to);
1140    return;
1141  }
1142  if (to == NULL) {
1143    __ b(condition, &backtrack_label_);
1144    return;
1145  }
1146  __ b(condition, to);
1147}
1148
1149
1150void RegExpMacroAssemblerARM::SafeCall(Label* to, Condition cond) {
1151  __ bl(to, cond);
1152}
1153
1154
1155void RegExpMacroAssemblerARM::SafeReturn() {
1156  __ pop(lr);
1157  __ add(pc, lr, Operand(masm_->CodeObject()));
1158}
1159
1160
1161void RegExpMacroAssemblerARM::SafeCallTarget(Label* name) {
1162  __ bind(name);
1163  __ sub(lr, lr, Operand(masm_->CodeObject()));
1164  __ push(lr);
1165}
1166
1167
1168void RegExpMacroAssemblerARM::Push(Register source) {
1169  DCHECK(!source.is(backtrack_stackpointer()));
1170  __ str(source,
1171         MemOperand(backtrack_stackpointer(), kPointerSize, NegPreIndex));
1172}
1173
1174
1175void RegExpMacroAssemblerARM::Pop(Register target) {
1176  DCHECK(!target.is(backtrack_stackpointer()));
1177  __ ldr(target,
1178         MemOperand(backtrack_stackpointer(), kPointerSize, PostIndex));
1179}
1180
1181
1182void RegExpMacroAssemblerARM::CheckPreemption() {
1183  // Check for preemption.
1184  ExternalReference stack_limit =
1185      ExternalReference::address_of_stack_limit(isolate());
1186  __ mov(r0, Operand(stack_limit));
1187  __ ldr(r0, MemOperand(r0));
1188  __ cmp(sp, r0);
1189  SafeCall(&check_preempt_label_, ls);
1190}
1191
1192
1193void RegExpMacroAssemblerARM::CheckStackLimit() {
1194  ExternalReference stack_limit =
1195      ExternalReference::address_of_regexp_stack_limit(isolate());
1196  __ mov(r0, Operand(stack_limit));
1197  __ ldr(r0, MemOperand(r0));
1198  __ cmp(backtrack_stackpointer(), Operand(r0));
1199  SafeCall(&stack_overflow_label_, ls);
1200}
1201
1202
1203void RegExpMacroAssemblerARM::LoadCurrentCharacterUnchecked(int cp_offset,
1204                                                            int characters) {
1205  Register offset = current_input_offset();
1206  if (cp_offset != 0) {
1207    // r4 is not being used to store the capture start index at this point.
1208    __ add(r4, current_input_offset(), Operand(cp_offset * char_size()));
1209    offset = r4;
1210  }
1211  // The ldr, str, ldrh, strh instructions can do unaligned accesses, if the CPU
1212  // and the operating system running on the target allow it.
1213  // If unaligned load/stores are not supported then this function must only
1214  // be used to load a single character at a time.
1215  if (!CanReadUnaligned()) {
1216    DCHECK(characters == 1);
1217  }
1218
1219  if (mode_ == LATIN1) {
1220    if (characters == 4) {
1221      __ ldr(current_character(), MemOperand(end_of_input_address(), offset));
1222    } else if (characters == 2) {
1223      __ ldrh(current_character(), MemOperand(end_of_input_address(), offset));
1224    } else {
1225      DCHECK(characters == 1);
1226      __ ldrb(current_character(), MemOperand(end_of_input_address(), offset));
1227    }
1228  } else {
1229    DCHECK(mode_ == UC16);
1230    if (characters == 2) {
1231      __ ldr(current_character(), MemOperand(end_of_input_address(), offset));
1232    } else {
1233      DCHECK(characters == 1);
1234      __ ldrh(current_character(), MemOperand(end_of_input_address(), offset));
1235    }
1236  }
1237}
1238
1239
1240#undef __
1241
1242#endif  // V8_INTERPRETED_REGEXP
1243
1244}  // namespace internal
1245}  // namespace v8
1246
1247#endif  // V8_TARGET_ARCH_ARM
1248