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#ifndef V8_REGEXP_REGEXP_MACRO_ASSEMBLER_H_
6#define V8_REGEXP_REGEXP_MACRO_ASSEMBLER_H_
7
8#include "src/label.h"
9#include "src/regexp/regexp-ast.h"
10
11namespace v8 {
12namespace internal {
13
14static const uc32 kLeadSurrogateStart = 0xd800;
15static const uc32 kLeadSurrogateEnd = 0xdbff;
16static const uc32 kTrailSurrogateStart = 0xdc00;
17static const uc32 kTrailSurrogateEnd = 0xdfff;
18static const uc32 kNonBmpStart = 0x10000;
19static const uc32 kNonBmpEnd = 0x10ffff;
20
21struct DisjunctDecisionRow {
22 RegExpCharacterClass cc;
23 Label* on_match;
24};
25
26
27class RegExpMacroAssembler {
28 public:
29 // The implementation must be able to handle at least:
30 static const int kMaxRegister = (1 << 16) - 1;
31 static const int kMaxCPOffset = (1 << 15) - 1;
32 static const int kMinCPOffset = -(1 << 15);
33
34 static const int kTableSizeBits = 7;
35 static const int kTableSize = 1 << kTableSizeBits;
36 static const int kTableMask = kTableSize - 1;
37
38 enum IrregexpImplementation {
39 kIA32Implementation,
40 kARMImplementation,
41 kARM64Implementation,
42 kMIPSImplementation,
43 kS390Implementation,
44 kPPCImplementation,
45 kX64Implementation,
46 kX87Implementation,
47 kBytecodeImplementation
48 };
49
50 enum StackCheckFlag {
51 kNoStackLimitCheck = false,
52 kCheckStackLimit = true
53 };
54
55 RegExpMacroAssembler(Isolate* isolate, Zone* zone);
56 virtual ~RegExpMacroAssembler();
57 // This function is called when code generation is aborted, so that
58 // the assembler could clean up internal data structures.
59 virtual void AbortedCodeGeneration() {}
60 // The maximal number of pushes between stack checks. Users must supply
61 // kCheckStackLimit flag to push operations (instead of kNoStackLimitCheck)
62 // at least once for every stack_limit() pushes that are executed.
63 virtual int stack_limit_slack() = 0;
64 virtual bool CanReadUnaligned() = 0;
65 virtual void AdvanceCurrentPosition(int by) = 0; // Signed cp change.
66 virtual void AdvanceRegister(int reg, int by) = 0; // r[reg] += by.
67 // Continues execution from the position pushed on the top of the backtrack
68 // stack by an earlier PushBacktrack(Label*).
69 virtual void Backtrack() = 0;
70 virtual void Bind(Label* label) = 0;
71 virtual void CheckAtStart(Label* on_at_start) = 0;
72 // Dispatch after looking the current character up in a 2-bits-per-entry
73 // map. The destinations vector has up to 4 labels.
74 virtual void CheckCharacter(unsigned c, Label* on_equal) = 0;
75 // Bitwise and the current character with the given constant and then
76 // check for a match with c.
77 virtual void CheckCharacterAfterAnd(unsigned c,
78 unsigned and_with,
79 Label* on_equal) = 0;
80 virtual void CheckCharacterGT(uc16 limit, Label* on_greater) = 0;
81 virtual void CheckCharacterLT(uc16 limit, Label* on_less) = 0;
82 virtual void CheckGreedyLoop(Label* on_tos_equals_current_position) = 0;
83 virtual void CheckNotAtStart(int cp_offset, Label* on_not_at_start) = 0;
84 virtual void CheckNotBackReference(int start_reg, bool read_backward,
85 Label* on_no_match) = 0;
86 virtual void CheckNotBackReferenceIgnoreCase(int start_reg,
87 bool read_backward, bool unicode,
88 Label* on_no_match) = 0;
89 // Check the current character for a match with a literal character. If we
90 // fail to match then goto the on_failure label. End of input always
91 // matches. If the label is nullptr then we should pop a backtrack address
92 // off the stack and go to that.
93 virtual void CheckNotCharacter(unsigned c, Label* on_not_equal) = 0;
94 virtual void CheckNotCharacterAfterAnd(unsigned c,
95 unsigned and_with,
96 Label* on_not_equal) = 0;
97 // Subtract a constant from the current character, then and with the given
98 // constant and then check for a match with c.
99 virtual void CheckNotCharacterAfterMinusAnd(uc16 c,
100 uc16 minus,
101 uc16 and_with,
102 Label* on_not_equal) = 0;
103 virtual void CheckCharacterInRange(uc16 from,
104 uc16 to, // Both inclusive.
105 Label* on_in_range) = 0;
106 virtual void CheckCharacterNotInRange(uc16 from,
107 uc16 to, // Both inclusive.
108 Label* on_not_in_range) = 0;
109
110 // The current character (modulus the kTableSize) is looked up in the byte
111 // array, and if the found byte is non-zero, we jump to the on_bit_set label.
112 virtual void CheckBitInTable(Handle<ByteArray> table, Label* on_bit_set) = 0;
113
114 // Checks whether the given offset from the current position is before
115 // the end of the string. May overwrite the current character.
116 virtual void CheckPosition(int cp_offset, Label* on_outside_input);
117 // Check whether a standard/default character class matches the current
118 // character. Returns false if the type of special character class does
119 // not have custom support.
120 // May clobber the current loaded character.
121 virtual bool CheckSpecialCharacterClass(uc16 type, Label* on_no_match);
122 virtual void Fail() = 0;
123 virtual Handle<HeapObject> GetCode(Handle<String> source) = 0;
124 virtual void GoTo(Label* label) = 0;
125 // Check whether a register is >= a given constant and go to a label if it
126 // is. Backtracks instead if the label is nullptr.
127 virtual void IfRegisterGE(int reg, int comparand, Label* if_ge) = 0;
128 // Check whether a register is < a given constant and go to a label if it is.
129 // Backtracks instead if the label is nullptr.
130 virtual void IfRegisterLT(int reg, int comparand, Label* if_lt) = 0;
131 // Check whether a register is == to the current position and go to a
132 // label if it is.
133 virtual void IfRegisterEqPos(int reg, Label* if_eq) = 0;
134 virtual IrregexpImplementation Implementation() = 0;
135 virtual void LoadCurrentCharacter(int cp_offset,
136 Label* on_end_of_input,
137 bool check_bounds = true,
138 int characters = 1) = 0;
139 virtual void PopCurrentPosition() = 0;
140 virtual void PopRegister(int register_index) = 0;
141 // Pushes the label on the backtrack stack, so that a following Backtrack
142 // will go to this label. Always checks the backtrack stack limit.
143 virtual void PushBacktrack(Label* label) = 0;
144 virtual void PushCurrentPosition() = 0;
145 virtual void PushRegister(int register_index,
146 StackCheckFlag check_stack_limit) = 0;
147 virtual void ReadCurrentPositionFromRegister(int reg) = 0;
148 virtual void ReadStackPointerFromRegister(int reg) = 0;
149 virtual void SetCurrentPositionFromEnd(int by) = 0;
150 virtual void SetRegister(int register_index, int to) = 0;
151 // Return whether the matching (with a global regexp) will be restarted.
152 virtual bool Succeed() = 0;
153 virtual void WriteCurrentPositionToRegister(int reg, int cp_offset) = 0;
154 virtual void ClearRegisters(int reg_from, int reg_to) = 0;
155 virtual void WriteStackPointerToRegister(int reg) = 0;
156
157 // Compares two-byte strings case insensitively.
158 // Called from generated RegExp code.
159 static int CaseInsensitiveCompareUC16(Address byte_offset1,
160 Address byte_offset2,
161 size_t byte_length, Isolate* isolate);
162
163 // Check that we are not in the middle of a surrogate pair.
164 void CheckNotInSurrogatePair(int cp_offset, Label* on_failure);
165
166 // Controls the generation of large inlined constants in the code.
167 void set_slow_safe(bool ssc) { slow_safe_compiler_ = ssc; }
168 bool slow_safe() { return slow_safe_compiler_; }
169
170 enum GlobalMode {
171 NOT_GLOBAL,
172 GLOBAL_NO_ZERO_LENGTH_CHECK,
173 GLOBAL,
174 GLOBAL_UNICODE
175 };
176 // Set whether the regular expression has the global flag. Exiting due to
177 // a failure in a global regexp may still mean success overall.
178 inline void set_global_mode(GlobalMode mode) { global_mode_ = mode; }
179 inline bool global() { return global_mode_ != NOT_GLOBAL; }
180 inline bool global_with_zero_length_check() {
181 return global_mode_ == GLOBAL || global_mode_ == GLOBAL_UNICODE;
182 }
183 inline bool global_unicode() { return global_mode_ == GLOBAL_UNICODE; }
184
185 Isolate* isolate() const { return isolate_; }
186 Zone* zone() const { return zone_; }
187
188 private:
189 bool slow_safe_compiler_;
190 GlobalMode global_mode_;
191 Isolate* isolate_;
192 Zone* zone_;
193};
194
195class NativeRegExpMacroAssembler: public RegExpMacroAssembler {
196 public:
197 // Type of input string to generate code for.
198 enum Mode { LATIN1 = 1, UC16 = 2 };
199
200 // Result of calling generated native RegExp code.
201 // RETRY: Something significant changed during execution, and the matching
202 // should be retried from scratch.
203 // EXCEPTION: Something failed during execution. If no exception has been
204 // thrown, it's an internal out-of-memory, and the caller should
205 // throw the exception.
206 // FAILURE: Matching failed.
207 // SUCCESS: Matching succeeded, and the output array has been filled with
208 // capture positions.
209 enum Result { RETRY = -2, EXCEPTION = -1, FAILURE = 0, SUCCESS = 1 };
210
211 NativeRegExpMacroAssembler(Isolate* isolate, Zone* zone);
212 ~NativeRegExpMacroAssembler() override;
213 bool CanReadUnaligned() override;
214
215 // Returns a {Result} sentinel, or the number of successful matches.
216 static int Match(Handle<Code> regexp, Handle<String> subject,
217 int* offsets_vector, int offsets_vector_length,
218 int previous_index, Isolate* isolate);
219
220 // Called from RegExp if the backtrack stack limit is hit.
221 // Tries to expand the stack. Returns the new stack-pointer if
222 // successful, and updates the stack_top address, or returns 0 if unable
223 // to grow the stack.
224 // This function must not trigger a garbage collection.
225 static Address GrowStack(Address stack_pointer, Address* stack_top,
226 Isolate* isolate);
227
228 static const byte* StringCharacterPosition(
229 String subject, int start_index, const DisallowHeapAllocation& no_gc);
230
231 static int CheckStackGuardState(Isolate* isolate, int start_index,
232 bool is_direct_call, Address* return_address,
233 Code re_code, Address* subject,
234 const byte** input_start,
235 const byte** input_end);
236
237 // Byte map of one byte characters with a 0xff if the character is a word
238 // character (digit, letter or underscore) and 0x00 otherwise.
239 // Used by generated RegExp code.
240 static const byte word_character_map[256];
241
242 static Address word_character_map_address() {
243 return reinterpret_cast<Address>(&word_character_map[0]);
244 }
245
246 // Returns a {Result} sentinel, or the number of successful matches.
247 V8_EXPORT_PRIVATE static int Execute(Code code, String input,
248 int start_offset,
249 const byte* input_start,
250 const byte* input_end, int* output,
251 int output_size, Isolate* isolate);
252};
253
254} // namespace internal
255} // namespace v8
256
257#endif // V8_REGEXP_REGEXP_MACRO_ASSEMBLER_H_
258