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_X64_REGEXP_MACRO_ASSEMBLER_X64_H_ 6 #define V8_X64_REGEXP_MACRO_ASSEMBLER_X64_H_ 7 8 #include "src/macro-assembler.h" 9 #include "src/x64/assembler-x64-inl.h" 10 #include "src/x64/assembler-x64.h" 11 #include "src/x64/macro-assembler-x64.h" 12 13 namespace v8 { 14 namespace internal { 15 16 #ifndef V8_INTERPRETED_REGEXP 17 18 class RegExpMacroAssemblerX64: public NativeRegExpMacroAssembler { 19 public: 20 RegExpMacroAssemblerX64(Mode mode, int registers_to_save, Zone* zone); 21 virtual ~RegExpMacroAssemblerX64(); 22 virtual int stack_limit_slack(); 23 virtual void AdvanceCurrentPosition(int by); 24 virtual void AdvanceRegister(int reg, int by); 25 virtual void Backtrack(); 26 virtual void Bind(Label* label); 27 virtual void CheckAtStart(Label* on_at_start); 28 virtual void CheckCharacter(uint32_t c, Label* on_equal); 29 virtual void CheckCharacterAfterAnd(uint32_t c, 30 uint32_t mask, 31 Label* on_equal); 32 virtual void CheckCharacterGT(uc16 limit, Label* on_greater); 33 virtual void CheckCharacterLT(uc16 limit, Label* on_less); 34 // A "greedy loop" is a loop that is both greedy and with a simple 35 // body. It has a particularly simple implementation. 36 virtual void CheckGreedyLoop(Label* on_tos_equals_current_position); 37 virtual void CheckNotAtStart(Label* on_not_at_start); 38 virtual void CheckNotBackReference(int start_reg, Label* on_no_match); 39 virtual void CheckNotBackReferenceIgnoreCase(int start_reg, 40 Label* on_no_match); 41 virtual void CheckNotCharacter(uint32_t c, Label* on_not_equal); 42 virtual void CheckNotCharacterAfterAnd(uint32_t c, 43 uint32_t mask, 44 Label* on_not_equal); 45 virtual void CheckNotCharacterAfterMinusAnd(uc16 c, 46 uc16 minus, 47 uc16 mask, 48 Label* on_not_equal); 49 virtual void CheckCharacterInRange(uc16 from, 50 uc16 to, 51 Label* on_in_range); 52 virtual void CheckCharacterNotInRange(uc16 from, 53 uc16 to, 54 Label* on_not_in_range); 55 virtual void CheckBitInTable(Handle<ByteArray> table, Label* on_bit_set); 56 57 // Checks whether the given offset from the current position is before 58 // the end of the string. 59 virtual void CheckPosition(int cp_offset, Label* on_outside_input); 60 virtual bool CheckSpecialCharacterClass(uc16 type, 61 Label* on_no_match); 62 virtual void Fail(); 63 virtual Handle<HeapObject> GetCode(Handle<String> source); 64 virtual void GoTo(Label* label); 65 virtual void IfRegisterGE(int reg, int comparand, Label* if_ge); 66 virtual void IfRegisterLT(int reg, int comparand, Label* if_lt); 67 virtual void IfRegisterEqPos(int reg, Label* if_eq); 68 virtual IrregexpImplementation Implementation(); 69 virtual void LoadCurrentCharacter(int cp_offset, 70 Label* on_end_of_input, 71 bool check_bounds = true, 72 int characters = 1); 73 virtual void PopCurrentPosition(); 74 virtual void PopRegister(int register_index); 75 virtual void PushBacktrack(Label* label); 76 virtual void PushCurrentPosition(); 77 virtual void PushRegister(int register_index, 78 StackCheckFlag check_stack_limit); 79 virtual void ReadCurrentPositionFromRegister(int reg); 80 virtual void ReadStackPointerFromRegister(int reg); 81 virtual void SetCurrentPositionFromEnd(int by); 82 virtual void SetRegister(int register_index, int to); 83 virtual bool Succeed(); 84 virtual void WriteCurrentPositionToRegister(int reg, int cp_offset); 85 virtual void ClearRegisters(int reg_from, int reg_to); 86 virtual void WriteStackPointerToRegister(int reg); 87 88 static Result Match(Handle<Code> regexp, 89 Handle<String> subject, 90 int* offsets_vector, 91 int offsets_vector_length, 92 int previous_index, 93 Isolate* isolate); 94 95 static Result Execute(Code* code, 96 String* input, 97 int start_offset, 98 const byte* input_start, 99 const byte* input_end, 100 int* output, 101 bool at_start); 102 103 // Called from RegExp if the stack-guard is triggered. 104 // If the code object is relocated, the return address is fixed before 105 // returning. 106 static int CheckStackGuardState(Address* return_address, 107 Code* re_code, 108 Address re_frame); 109 110 private: 111 // Offsets from rbp of function parameters and stored registers. 112 static const int kFramePointer = 0; 113 // Above the frame pointer - function parameters and return address. 114 static const int kReturn_eip = kFramePointer + kRegisterSize; 115 static const int kFrameAlign = kReturn_eip + kRegisterSize; 116 117 #ifdef _WIN64 118 // Parameters (first four passed as registers, but with room on stack). 119 // In Microsoft 64-bit Calling Convention, there is room on the callers 120 // stack (before the return address) to spill parameter registers. We 121 // use this space to store the register passed parameters. 122 static const int kInputString = kFrameAlign; 123 // StartIndex is passed as 32 bit int. 124 static const int kStartIndex = kInputString + kRegisterSize; 125 static const int kInputStart = kStartIndex + kRegisterSize; 126 static const int kInputEnd = kInputStart + kRegisterSize; 127 static const int kRegisterOutput = kInputEnd + kRegisterSize; 128 // For the case of global regular expression, we have room to store at least 129 // one set of capture results. For the case of non-global regexp, we ignore 130 // this value. NumOutputRegisters is passed as 32-bit value. The upper 131 // 32 bit of this 64-bit stack slot may contain garbage. 132 static const int kNumOutputRegisters = kRegisterOutput + kRegisterSize; 133 static const int kStackHighEnd = kNumOutputRegisters + kRegisterSize; 134 // DirectCall is passed as 32 bit int (values 0 or 1). 135 static const int kDirectCall = kStackHighEnd + kRegisterSize; 136 static const int kIsolate = kDirectCall + kRegisterSize; 137 #else 138 // In AMD64 ABI Calling Convention, the first six integer parameters 139 // are passed as registers, and caller must allocate space on the stack 140 // if it wants them stored. We push the parameters after the frame pointer. 141 static const int kInputString = kFramePointer - kRegisterSize; 142 static const int kStartIndex = kInputString - kRegisterSize; 143 static const int kInputStart = kStartIndex - kRegisterSize; 144 static const int kInputEnd = kInputStart - kRegisterSize; 145 static const int kRegisterOutput = kInputEnd - kRegisterSize; 146 147 // For the case of global regular expression, we have room to store at least 148 // one set of capture results. For the case of non-global regexp, we ignore 149 // this value. 150 static const int kNumOutputRegisters = kRegisterOutput - kRegisterSize; 151 static const int kStackHighEnd = kFrameAlign; 152 static const int kDirectCall = kStackHighEnd + kRegisterSize; 153 static const int kIsolate = kDirectCall + kRegisterSize; 154 #endif 155 156 #ifdef _WIN64 157 // Microsoft calling convention has three callee-saved registers 158 // (that we are using). We push these after the frame pointer. 159 static const int kBackup_rsi = kFramePointer - kRegisterSize; 160 static const int kBackup_rdi = kBackup_rsi - kRegisterSize; 161 static const int kBackup_rbx = kBackup_rdi - kRegisterSize; 162 static const int kLastCalleeSaveRegister = kBackup_rbx; 163 #else 164 // AMD64 Calling Convention has only one callee-save register that 165 // we use. We push this after the frame pointer (and after the 166 // parameters). 167 static const int kBackup_rbx = kNumOutputRegisters - kRegisterSize; 168 static const int kLastCalleeSaveRegister = kBackup_rbx; 169 #endif 170 171 static const int kSuccessfulCaptures = kLastCalleeSaveRegister - kPointerSize; 172 // When adding local variables remember to push space for them in 173 // the frame in GetCode. 174 static const int kInputStartMinusOne = kSuccessfulCaptures - kPointerSize; 175 176 // First register address. Following registers are below it on the stack. 177 static const int kRegisterZero = kInputStartMinusOne - kPointerSize; 178 179 // Initial size of code buffer. 180 static const size_t kRegExpCodeSize = 1024; 181 182 // Load a number of characters at the given offset from the 183 // current position, into the current-character register. 184 void LoadCurrentCharacterUnchecked(int cp_offset, int character_count); 185 186 // Check whether preemption has been requested. 187 void CheckPreemption(); 188 189 // Check whether we are exceeding the stack limit on the backtrack stack. 190 void CheckStackLimit(); 191 192 // Generate a call to CheckStackGuardState. 193 void CallCheckStackGuardState(); 194 195 // The rbp-relative location of a regexp register. 196 Operand register_location(int register_index); 197 198 // The register containing the current character after LoadCurrentCharacter. current_character()199 inline Register current_character() { return rdx; } 200 201 // The register containing the backtrack stack top. Provides a meaningful 202 // name to the register. backtrack_stackpointer()203 inline Register backtrack_stackpointer() { return rcx; } 204 205 // The registers containing a self pointer to this code's Code object. code_object_pointer()206 inline Register code_object_pointer() { return r8; } 207 208 // Byte size of chars in the string to match (decided by the Mode argument) char_size()209 inline int char_size() { return static_cast<int>(mode_); } 210 211 // Equivalent to a conditional branch to the label, unless the label 212 // is NULL, in which case it is a conditional Backtrack. 213 void BranchOrBacktrack(Condition condition, Label* to); 214 MarkPositionForCodeRelativeFixup()215 void MarkPositionForCodeRelativeFixup() { 216 code_relative_fixup_positions_.Add(masm_.pc_offset(), zone()); 217 } 218 219 void FixupCodeRelativePositions(); 220 221 // Call and return internally in the generated code in a way that 222 // is GC-safe (i.e., doesn't leave absolute code addresses on the stack) 223 inline void SafeCall(Label* to); 224 inline void SafeCallTarget(Label* label); 225 inline void SafeReturn(); 226 227 // Pushes the value of a register on the backtrack stack. Decrements the 228 // stack pointer (rcx) by a word size and stores the register's value there. 229 inline void Push(Register source); 230 231 // Pushes a value on the backtrack stack. Decrements the stack pointer (rcx) 232 // by a word size and stores the value there. 233 inline void Push(Immediate value); 234 235 // Pushes the Code object relative offset of a label on the backtrack stack 236 // (i.e., a backtrack target). Decrements the stack pointer (rcx) 237 // by a word size and stores the value there. 238 inline void Push(Label* label); 239 240 // Pops a value from the backtrack stack. Reads the word at the stack pointer 241 // (rcx) and increments it by a word size. 242 inline void Pop(Register target); 243 244 // Drops the top value from the backtrack stack without reading it. 245 // Increments the stack pointer (rcx) by a word size. 246 inline void Drop(); 247 248 inline void ReadPositionFromRegister(Register dst, int reg); 249 isolate()250 Isolate* isolate() const { return masm_.isolate(); } 251 252 MacroAssembler masm_; 253 MacroAssembler::NoRootArrayScope no_root_array_scope_; 254 255 ZoneList<int> code_relative_fixup_positions_; 256 257 // Which mode to generate code for (LATIN1 or UC16). 258 Mode mode_; 259 260 // One greater than maximal register index actually used. 261 int num_registers_; 262 263 // Number of registers to output at the end (the saved registers 264 // are always 0..num_saved_registers_-1) 265 int num_saved_registers_; 266 267 // Labels used internally. 268 Label entry_label_; 269 Label start_label_; 270 Label success_label_; 271 Label backtrack_label_; 272 Label exit_label_; 273 Label check_preempt_label_; 274 Label stack_overflow_label_; 275 }; 276 277 #endif // V8_INTERPRETED_REGEXP 278 279 }} // namespace v8::internal 280 281 #endif // V8_X64_REGEXP_MACRO_ASSEMBLER_X64_H_ 282