regexp-macro-assembler.h 10.7 KB
Newer Older
1
// Copyright 2012 the V8 project authors. All rights reserved.
2 3
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
4 5 6 7

#ifndef V8_REGEXP_MACRO_ASSEMBLER_H_
#define V8_REGEXP_MACRO_ASSEMBLER_H_

8
#include "src/ast.h"
9

10 11
namespace v8 {
namespace internal {
12 13 14 15 16 17 18 19 20

struct DisjunctDecisionRow {
  RegExpCharacterClass cc;
  Label* on_match;
};


class RegExpMacroAssembler {
 public:
21 22 23 24
  // The implementation must be able to handle at least:
  static const int kMaxRegister = (1 << 16) - 1;
  static const int kMaxCPOffset = (1 << 15) - 1;
  static const int kMinCPOffset = -(1 << 15);
25 26 27 28 29

  static const int kTableSizeBits = 7;
  static const int kTableSize = 1 << kTableSizeBits;
  static const int kTableMask = kTableSize - 1;

30 31 32
  enum IrregexpImplementation {
    kIA32Implementation,
    kARMImplementation,
33
    kARM64Implementation,
34
    kMIPSImplementation,
35
    kPPCImplementation,
36
    kX64Implementation,
danno@chromium.org's avatar
danno@chromium.org committed
37
    kX87Implementation,
38 39 40 41 42 43 44
    kBytecodeImplementation
  };

  enum StackCheckFlag {
    kNoStackLimitCheck = false,
    kCheckStackLimit = true
  };
45

46
  RegExpMacroAssembler(Isolate* isolate, Zone* zone);
47
  virtual ~RegExpMacroAssembler();
48 49 50
  // This function is called when code generation is aborted, so that
  // the assembler could clean up internal data structures.
  virtual void AbortedCodeGeneration() {}
51 52 53 54
  // The maximal number of pushes between stack checks. Users must supply
  // kCheckStackLimit flag to push operations (instead of kNoStackLimitCheck)
  // at least once for every stack_limit() pushes that are executed.
  virtual int stack_limit_slack() = 0;
55
  virtual bool CanReadUnaligned() = 0;
56 57
  virtual void AdvanceCurrentPosition(int by) = 0;  // Signed cp change.
  virtual void AdvanceRegister(int reg, int by) = 0;  // r[reg] += by.
58 59
  // Continues execution from the position pushed on the top of the backtrack
  // stack by an earlier PushBacktrack(Label*).
60 61
  virtual void Backtrack() = 0;
  virtual void Bind(Label* label) = 0;
62
  virtual void CheckAtStart(Label* on_at_start) = 0;
63 64
  // Dispatch after looking the current character up in a 2-bits-per-entry
  // map.  The destinations vector has up to 4 labels.
65
  virtual void CheckCharacter(unsigned c, Label* on_equal) = 0;
66 67
  // Bitwise and the current character with the given constant and then
  // check for a match with c.
68 69
  virtual void CheckCharacterAfterAnd(unsigned c,
                                      unsigned and_with,
70
                                      Label* on_equal) = 0;
71 72
  virtual void CheckCharacterGT(uc16 limit, Label* on_greater) = 0;
  virtual void CheckCharacterLT(uc16 limit, Label* on_less) = 0;
erik.corry@gmail.com's avatar
erik.corry@gmail.com committed
73
  virtual void CheckGreedyLoop(Label* on_tos_equals_current_position) = 0;
74
  virtual void CheckNotAtStart(Label* on_not_at_start) = 0;
75
  virtual void CheckNotBackReference(int start_reg, Label* on_no_match) = 0;
76 77
  virtual void CheckNotBackReferenceIgnoreCase(int start_reg,
                                               Label* on_no_match) = 0;
78 79 80 81
  // Check the current character for a match with a literal character.  If we
  // fail to match then goto the on_failure label.  End of input always
  // matches.  If the label is NULL then we should pop a backtrack address off
  // the stack and go to that.
82 83 84
  virtual void CheckNotCharacter(unsigned c, Label* on_not_equal) = 0;
  virtual void CheckNotCharacterAfterAnd(unsigned c,
                                         unsigned and_with,
85
                                         Label* on_not_equal) = 0;
86
  // Subtract a constant from the current character, then and with the given
87
  // constant and then check for a match with c.
88 89 90 91
  virtual void CheckNotCharacterAfterMinusAnd(uc16 c,
                                              uc16 minus,
                                              uc16 and_with,
                                              Label* on_not_equal) = 0;
92 93 94 95 96 97 98 99 100 101 102
  virtual void CheckCharacterInRange(uc16 from,
                                     uc16 to,  // Both inclusive.
                                     Label* on_in_range) = 0;
  virtual void CheckCharacterNotInRange(uc16 from,
                                        uc16 to,  // Both inclusive.
                                        Label* on_not_in_range) = 0;

  // The current character (modulus the kTableSize) is looked up in the byte
  // array, and if the found byte is non-zero, we jump to the on_bit_set label.
  virtual void CheckBitInTable(Handle<ByteArray> table, Label* on_bit_set) = 0;

103 104 105 106 107
  // Checks whether the given offset from the current position is before
  // the end of the string.  May overwrite the current character.
  virtual void CheckPosition(int cp_offset, Label* on_outside_input) {
    LoadCurrentCharacter(cp_offset, on_outside_input, true);
  }
108 109 110 111 112 113 114 115
  // Check whether a standard/default character class matches the current
  // character. Returns false if the type of special character class does
  // not have custom support.
  // May clobber the current loaded character.
  virtual bool CheckSpecialCharacterClass(uc16 type,
                                          Label* on_no_match) {
    return false;
  }
116
  virtual void Fail() = 0;
117
  virtual Handle<HeapObject> GetCode(Handle<String> source) = 0;
118 119 120 121 122 123 124
  virtual void GoTo(Label* label) = 0;
  // Check whether a register is >= a given constant and go to a label if it
  // is.  Backtracks instead if the label is NULL.
  virtual void IfRegisterGE(int reg, int comparand, Label* if_ge) = 0;
  // Check whether a register is < a given constant and go to a label if it is.
  // Backtracks instead if the label is NULL.
  virtual void IfRegisterLT(int reg, int comparand, Label* if_lt) = 0;
125 126 127
  // Check whether a register is == to the current position and go to a
  // label if it is.
  virtual void IfRegisterEqPos(int reg, Label* if_eq) = 0;
128
  virtual IrregexpImplementation Implementation() = 0;
129 130 131 132
  virtual void LoadCurrentCharacter(int cp_offset,
                                    Label* on_end_of_input,
                                    bool check_bounds = true,
                                    int characters = 1) = 0;
133 134
  virtual void PopCurrentPosition() = 0;
  virtual void PopRegister(int register_index) = 0;
135 136
  // Pushes the label on the backtrack stack, so that a following Backtrack
  // will go to this label. Always checks the backtrack stack limit.
137 138
  virtual void PushBacktrack(Label* label) = 0;
  virtual void PushCurrentPosition() = 0;
139 140
  virtual void PushRegister(int register_index,
                            StackCheckFlag check_stack_limit) = 0;
141 142
  virtual void ReadCurrentPositionFromRegister(int reg) = 0;
  virtual void ReadStackPointerFromRegister(int reg) = 0;
143
  virtual void SetCurrentPositionFromEnd(int by) = 0;
144
  virtual void SetRegister(int register_index, int to) = 0;
145 146
  // Return whether the matching (with a global regexp) will be restarted.
  virtual bool Succeed() = 0;
erik.corry@gmail.com's avatar
erik.corry@gmail.com committed
147
  virtual void WriteCurrentPositionToRegister(int reg, int cp_offset) = 0;
148
  virtual void ClearRegisters(int reg_from, int reg_to) = 0;
149
  virtual void WriteStackPointerToRegister(int reg) = 0;
150 151 152 153 154

  // Controls the generation of large inlined constants in the code.
  void set_slow_safe(bool ssc) { slow_safe_compiler_ = ssc; }
  bool slow_safe() { return slow_safe_compiler_; }

155
  enum GlobalMode { NOT_GLOBAL, GLOBAL, GLOBAL_NO_ZERO_LENGTH_CHECK };
156 157
  // Set whether the regular expression has the global flag.  Exiting due to
  // a failure in a global regexp may still mean success overall.
158 159 160 161 162
  inline void set_global_mode(GlobalMode mode) { global_mode_ = mode; }
  inline bool global() { return global_mode_ != NOT_GLOBAL; }
  inline bool global_with_zero_length_check() {
    return global_mode_ == GLOBAL;
  }
163

164
  Isolate* isolate() const { return isolate_; }
165 166
  Zone* zone() const { return zone_; }

167 168
 private:
  bool slow_safe_compiler_;
169
  bool global_mode_;
170
  Isolate* isolate_;
171
  Zone* zone_;
172 173 174
};


175
#ifndef V8_INTERPRETED_REGEXP  // Avoid compiling unused code.
176 177

class NativeRegExpMacroAssembler: public RegExpMacroAssembler {
178
 public:
179
  // Type of input string to generate code for.
180
  enum Mode { LATIN1 = 1, UC16 = 2 };
181

182 183 184 185 186 187 188 189 190 191
  // Result of calling generated native RegExp code.
  // RETRY: Something significant changed during execution, and the matching
  //        should be retried from scratch.
  // EXCEPTION: Something failed during execution. If no exception has been
  //        thrown, it's an internal out-of-memory, and the caller should
  //        throw the exception.
  // FAILURE: Matching failed.
  // SUCCESS: Matching succeeded, and the output array has been filled with
  //        capture positions.
  enum Result { RETRY = -2, EXCEPTION = -1, FAILURE = 0, SUCCESS = 1 };
192

193
  NativeRegExpMacroAssembler(Isolate* isolate, Zone* zone);
194
  virtual ~NativeRegExpMacroAssembler();
lrn@chromium.org's avatar
lrn@chromium.org committed
195
  virtual bool CanReadUnaligned();
196 197 198 199 200

  static Result Match(Handle<Code> regexp,
                      Handle<String> subject,
                      int* offsets_vector,
                      int offsets_vector_length,
201 202
                      int previous_index,
                      Isolate* isolate);
203

204 205 206 207
  // Compares two-byte strings case insensitively.
  // Called from generated RegExp code.
  static int CaseInsensitiveCompareUC16(Address byte_offset1,
                                        Address byte_offset2,
208 209
                                        size_t byte_length,
                                        Isolate* isolate);
210

lrn@chromium.org's avatar
lrn@chromium.org committed
211 212 213 214 215
  // Called from RegExp if the backtrack stack limit is hit.
  // Tries to expand the stack. Returns the new stack-pointer if
  // successful, and updates the stack_top address, or returns 0 if unable
  // to grow the stack.
  // This function must not trigger a garbage collection.
216 217
  static Address GrowStack(Address stack_pointer, Address* stack_top,
                           Isolate* isolate);
lrn@chromium.org's avatar
lrn@chromium.org committed
218

219 220
  static const byte* StringCharacterPosition(String* subject, int start_index);

221 222 223 224 225 226
  static int CheckStackGuardState(Isolate* isolate, int start_index,
                                  bool is_direct_call, Address* return_address,
                                  Code* re_code, String** subject,
                                  const byte** input_start,
                                  const byte** input_end);

227
  // Byte map of one byte characters with a 0xff if the character is a word
228 229
  // character (digit, letter or underscore) and 0x00 otherwise.
  // Used by generated RegExp code.
230
  static const byte word_character_map[256];
231 232

  static Address word_character_map_address() {
233
    return const_cast<Address>(&word_character_map[0]);
234 235
  }

236 237 238 239 240
  static Result Execute(Code* code,
                        String* input,
                        int start_offset,
                        const byte* input_start,
                        const byte* input_end,
241
                        int* output,
242
                        int output_size,
243
                        Isolate* isolate);
244
};
lrn@chromium.org's avatar
lrn@chromium.org committed
245

246
#endif  // V8_INTERPRETED_REGEXP
lrn@chromium.org's avatar
lrn@chromium.org committed
247

248 249 250
} }  // namespace v8::internal

#endif  // V8_REGEXP_MACRO_ASSEMBLER_H_