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_X64_REGEXP_MACRO_ASSEMBLER_X64_H_ 6 #define V8_REGEXP_X64_REGEXP_MACRO_ASSEMBLER_X64_H_ 7 8 #include "src/macro-assembler.h" 9 #include "src/regexp/regexp-macro-assembler.h" 10 #include "src/x64/assembler-x64.h" 11 12 namespace v8 { 13 namespace internal { 14 15 #ifndef V8_INTERPRETED_REGEXP 16 17 class RegExpMacroAssemblerX64: public NativeRegExpMacroAssembler { 18 public: 19 RegExpMacroAssemblerX64(Isolate* isolate, Zone* zone, Mode mode, 20 int registers_to_save); 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(int cp_offset, Label* on_not_at_start); 38 virtual void CheckNotBackReference(int start_reg, bool read_backward, 39 Label* on_no_match); 40 virtual void CheckNotBackReferenceIgnoreCase(int start_reg, 41 bool read_backward, bool unicode, 42 Label* on_no_match); 43 virtual void CheckNotCharacter(uint32_t c, Label* on_not_equal); 44 virtual void CheckNotCharacterAfterAnd(uint32_t c, 45 uint32_t mask, 46 Label* on_not_equal); 47 virtual void CheckNotCharacterAfterMinusAnd(uc16 c, 48 uc16 minus, 49 uc16 mask, 50 Label* on_not_equal); 51 virtual void CheckCharacterInRange(uc16 from, 52 uc16 to, 53 Label* on_in_range); 54 virtual void CheckCharacterNotInRange(uc16 from, 55 uc16 to, 56 Label* on_not_in_range); 57 virtual void CheckBitInTable(Handle<ByteArray> table, Label* on_bit_set); 58 59 // Checks whether the given offset from the current position is before 60 // the end of the string. 61 virtual void CheckPosition(int cp_offset, Label* on_outside_input); 62 virtual bool CheckSpecialCharacterClass(uc16 type, 63 Label* on_no_match); 64 virtual void Fail(); 65 virtual Handle<HeapObject> GetCode(Handle<String> source); 66 virtual void GoTo(Label* label); 67 virtual void IfRegisterGE(int reg, int comparand, Label* if_ge); 68 virtual void IfRegisterLT(int reg, int comparand, Label* if_lt); 69 virtual void IfRegisterEqPos(int reg, Label* if_eq); 70 virtual IrregexpImplementation Implementation(); 71 virtual void LoadCurrentCharacter(int cp_offset, 72 Label* on_end_of_input, 73 bool check_bounds = true, 74 int characters = 1); 75 virtual void PopCurrentPosition(); 76 virtual void PopRegister(int register_index); 77 virtual void PushBacktrack(Label* label); 78 virtual void PushCurrentPosition(); 79 virtual void PushRegister(int register_index, 80 StackCheckFlag check_stack_limit); 81 virtual void ReadCurrentPositionFromRegister(int reg); 82 virtual void ReadStackPointerFromRegister(int reg); 83 virtual void SetCurrentPositionFromEnd(int by); 84 virtual void SetRegister(int register_index, int to); 85 virtual bool Succeed(); 86 virtual void WriteCurrentPositionToRegister(int reg, int cp_offset); 87 virtual void ClearRegisters(int reg_from, int reg_to); 88 virtual void WriteStackPointerToRegister(int reg); 89 90 static Result Match(Handle<Code> regexp, 91 Handle<String> subject, 92 int* offsets_vector, 93 int offsets_vector_length, 94 int previous_index, 95 Isolate* isolate); 96 97 static Result Execute(Code* code, 98 String* input, 99 int start_offset, 100 const byte* input_start, 101 const byte* input_end, 102 int* output, 103 bool at_start); 104 105 // Called from RegExp if the stack-guard is triggered. 106 // If the code object is relocated, the return address is fixed before 107 // returning. 108 static int CheckStackGuardState(Address* return_address, 109 Code* re_code, 110 Address re_frame); 111 112 private: 113 // Offsets from rbp of function parameters and stored registers. 114 static const int kFramePointer = 0; 115 // Above the frame pointer - function parameters and return address. 116 static const int kReturn_eip = kFramePointer + kRegisterSize; 117 static const int kFrameAlign = kReturn_eip + kRegisterSize; 118 119 #ifdef _WIN64 120 // Parameters (first four passed as registers, but with room on stack). 121 // In Microsoft 64-bit Calling Convention, there is room on the callers 122 // stack (before the return address) to spill parameter registers. We 123 // use this space to store the register passed parameters. 124 static const int kInputString = kFrameAlign; 125 // StartIndex is passed as 32 bit int. 126 static const int kStartIndex = kInputString + kRegisterSize; 127 static const int kInputStart = kStartIndex + kRegisterSize; 128 static const int kInputEnd = kInputStart + kRegisterSize; 129 static const int kRegisterOutput = kInputEnd + kRegisterSize; 130 // For the case of global regular expression, we have room to store at least 131 // one set of capture results. For the case of non-global regexp, we ignore 132 // this value. NumOutputRegisters is passed as 32-bit value. The upper 133 // 32 bit of this 64-bit stack slot may contain garbage. 134 static const int kNumOutputRegisters = kRegisterOutput + kRegisterSize; 135 static const int kStackHighEnd = kNumOutputRegisters + kRegisterSize; 136 // DirectCall is passed as 32 bit int (values 0 or 1). 137 static const int kDirectCall = kStackHighEnd + kRegisterSize; 138 static const int kIsolate = kDirectCall + kRegisterSize; 139 #else 140 // In AMD64 ABI Calling Convention, the first six integer parameters 141 // are passed as registers, and caller must allocate space on the stack 142 // if it wants them stored. We push the parameters after the frame pointer. 143 static const int kInputString = kFramePointer - kRegisterSize; 144 static const int kStartIndex = kInputString - kRegisterSize; 145 static const int kInputStart = kStartIndex - kRegisterSize; 146 static const int kInputEnd = kInputStart - kRegisterSize; 147 static const int kRegisterOutput = kInputEnd - kRegisterSize; 148 149 // For the case of global regular expression, we have room to store at least 150 // one set of capture results. For the case of non-global regexp, we ignore 151 // this value. 152 static const int kNumOutputRegisters = kRegisterOutput - kRegisterSize; 153 static const int kStackHighEnd = kFrameAlign; 154 static const int kDirectCall = kStackHighEnd + kRegisterSize; 155 static const int kIsolate = kDirectCall + kRegisterSize; 156 #endif 157 158 #ifdef _WIN64 159 // Microsoft calling convention has three callee-saved registers 160 // (that we are using). We push these after the frame pointer. 161 static const int kBackup_rsi = kFramePointer - kRegisterSize; 162 static const int kBackup_rdi = kBackup_rsi - kRegisterSize; 163 static const int kBackup_rbx = kBackup_rdi - kRegisterSize; 164 static const int kLastCalleeSaveRegister = kBackup_rbx; 165 #else 166 // AMD64 Calling Convention has only one callee-save register that 167 // we use. We push this after the frame pointer (and after the 168 // parameters). 169 static const int kBackup_rbx = kNumOutputRegisters - kRegisterSize; 170 static const int kLastCalleeSaveRegister = kBackup_rbx; 171 #endif 172 173 static const int kSuccessfulCaptures = kLastCalleeSaveRegister - kPointerSize; 174 // When adding local variables remember to push space for them in 175 // the frame in GetCode. 176 static const int kStringStartMinusOne = kSuccessfulCaptures - kPointerSize; 177 178 // First register address. Following registers are below it on the stack. 179 static const int kRegisterZero = kStringStartMinusOne - kPointerSize; 180 181 // Initial size of code buffer. 182 static const size_t kRegExpCodeSize = 1024; 183 184 // Load a number of characters at the given offset from the 185 // current position, into the current-character register. 186 void LoadCurrentCharacterUnchecked(int cp_offset, int character_count); 187 188 // Check whether preemption has been requested. 189 void CheckPreemption(); 190 191 // Check whether we are exceeding the stack limit on the backtrack stack. 192 void CheckStackLimit(); 193 194 // Generate a call to CheckStackGuardState. 195 void CallCheckStackGuardState(); 196 197 // The rbp-relative location of a regexp register. 198 Operand register_location(int register_index); 199 200 // The register containing the current character after LoadCurrentCharacter. current_character()201 inline Register current_character() { return rdx; } 202 203 // The register containing the backtrack stack top. Provides a meaningful 204 // name to the register. backtrack_stackpointer()205 inline Register backtrack_stackpointer() { return rcx; } 206 207 // The registers containing a self pointer to this code's Code object. code_object_pointer()208 inline Register code_object_pointer() { return r8; } 209 210 // Byte size of chars in the string to match (decided by the Mode argument) char_size()211 inline int char_size() { return static_cast<int>(mode_); } 212 213 // Equivalent to a conditional branch to the label, unless the label 214 // is NULL, in which case it is a conditional Backtrack. 215 void BranchOrBacktrack(Condition condition, Label* to); 216 MarkPositionForCodeRelativeFixup()217 void MarkPositionForCodeRelativeFixup() { 218 code_relative_fixup_positions_.Add(masm_.pc_offset(), zone()); 219 } 220 221 void FixupCodeRelativePositions(); 222 223 // Call and return internally in the generated code in a way that 224 // is GC-safe (i.e., doesn't leave absolute code addresses on the stack) 225 inline void SafeCall(Label* to); 226 inline void SafeCallTarget(Label* label); 227 inline void SafeReturn(); 228 229 // Pushes the value of a register on the backtrack stack. Decrements the 230 // stack pointer (rcx) by a word size and stores the register's value there. 231 inline void Push(Register source); 232 233 // Pushes a value on the backtrack stack. Decrements the stack pointer (rcx) 234 // by a word size and stores the value there. 235 inline void Push(Immediate value); 236 237 // Pushes the Code object relative offset of a label on the backtrack stack 238 // (i.e., a backtrack target). Decrements the stack pointer (rcx) 239 // by a word size and stores the value there. 240 inline void Push(Label* label); 241 242 // Pops a value from the backtrack stack. Reads the word at the stack pointer 243 // (rcx) and increments it by a word size. 244 inline void Pop(Register target); 245 246 // Drops the top value from the backtrack stack without reading it. 247 // Increments the stack pointer (rcx) by a word size. 248 inline void Drop(); 249 250 inline void ReadPositionFromRegister(Register dst, int reg); 251 isolate()252 Isolate* isolate() const { return masm_.isolate(); } 253 254 MacroAssembler masm_; 255 MacroAssembler::NoRootArrayScope no_root_array_scope_; 256 257 ZoneList<int> code_relative_fixup_positions_; 258 259 // Which mode to generate code for (LATIN1 or UC16). 260 Mode mode_; 261 262 // One greater than maximal register index actually used. 263 int num_registers_; 264 265 // Number of registers to output at the end (the saved registers 266 // are always 0..num_saved_registers_-1) 267 int num_saved_registers_; 268 269 // Labels used internally. 270 Label entry_label_; 271 Label start_label_; 272 Label success_label_; 273 Label backtrack_label_; 274 Label exit_label_; 275 Label check_preempt_label_; 276 Label stack_overflow_label_; 277 }; 278 279 #endif // V8_INTERPRETED_REGEXP 280 281 } // namespace internal 282 } // namespace v8 283 284 #endif // V8_REGEXP_X64_REGEXP_MACRO_ASSEMBLER_X64_H_ 285