1// Copyright 2013 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_ARM64_REGEXP_MACRO_ASSEMBLER_ARM64_H_
6#define V8_REGEXP_ARM64_REGEXP_MACRO_ASSEMBLER_ARM64_H_
7
8#include "src/base/strings.h"
9#include "src/codegen/arm64/assembler-arm64.h"
10#include "src/codegen/macro-assembler.h"
11#include "src/regexp/regexp-macro-assembler.h"
12
13namespace v8 {
14namespace internal {
15
16class V8_EXPORT_PRIVATE RegExpMacroAssemblerARM64
17    : public NativeRegExpMacroAssembler {
18 public:
19  RegExpMacroAssemblerARM64(Isolate* isolate, Zone* zone, Mode mode,
20                            int registers_to_save);
21  ~RegExpMacroAssemblerARM64() override;
22  void AbortedCodeGeneration() override;
23  int stack_limit_slack() override;
24  void AdvanceCurrentPosition(int by) override;
25  void AdvanceRegister(int reg, int by) override;
26  void Backtrack() override;
27  void Bind(Label* label) override;
28  void CheckAtStart(int cp_offset, Label* on_at_start) override;
29  void CheckCharacter(unsigned c, Label* on_equal) override;
30  void CheckCharacterAfterAnd(unsigned c, unsigned mask,
31                              Label* on_equal) override;
32  void CheckCharacterGT(base::uc16 limit, Label* on_greater) override;
33  void CheckCharacterLT(base::uc16 limit, Label* on_less) override;
34  void CheckCharacters(base::Vector<const base::uc16> str, int cp_offset,
35                       Label* on_failure, bool check_end_of_string);
36  // A "greedy loop" is a loop that is both greedy and with a simple
37  // body. It has a particularly simple implementation.
38  void CheckGreedyLoop(Label* on_tos_equals_current_position) override;
39  void CheckNotAtStart(int cp_offset, Label* on_not_at_start) override;
40  void CheckNotBackReference(int start_reg, bool read_backward,
41                             Label* on_no_match) override;
42  void CheckNotBackReferenceIgnoreCase(int start_reg, bool read_backward,
43                                       bool unicode,
44                                       Label* on_no_match) override;
45  void CheckNotCharacter(unsigned c, Label* on_not_equal) override;
46  void CheckNotCharacterAfterAnd(unsigned c, unsigned mask,
47                                 Label* on_not_equal) override;
48  void CheckNotCharacterAfterMinusAnd(base::uc16 c, base::uc16 minus,
49                                      base::uc16 mask,
50                                      Label* on_not_equal) override;
51  void CheckCharacterInRange(base::uc16 from, base::uc16 to,
52                             Label* on_in_range) override;
53  void CheckCharacterNotInRange(base::uc16 from, base::uc16 to,
54                                Label* on_not_in_range) override;
55  bool CheckCharacterInRangeArray(const ZoneList<CharacterRange>* ranges,
56                                  Label* on_in_range) override;
57  bool CheckCharacterNotInRangeArray(const ZoneList<CharacterRange>* ranges,
58                                     Label* on_not_in_range) override;
59  void CheckBitInTable(Handle<ByteArray> table, Label* on_bit_set) override;
60
61  // Checks whether the given offset from the current position is before
62  // the end of the string.
63  void CheckPosition(int cp_offset, Label* on_outside_input) override;
64  bool CheckSpecialCharacterClass(StandardCharacterSet type,
65                                  Label* on_no_match) override;
66  void BindJumpTarget(Label* label = nullptr) override;
67  void Fail() override;
68  Handle<HeapObject> GetCode(Handle<String> source) override;
69  void GoTo(Label* label) override;
70  void IfRegisterGE(int reg, int comparand, Label* if_ge) override;
71  void IfRegisterLT(int reg, int comparand, Label* if_lt) override;
72  void IfRegisterEqPos(int reg, Label* if_eq) override;
73  IrregexpImplementation Implementation() override;
74  void LoadCurrentCharacterUnchecked(int cp_offset,
75                                     int character_count) override;
76  void PopCurrentPosition() override;
77  void PopRegister(int register_index) override;
78  void PushBacktrack(Label* label) override;
79  void PushCurrentPosition() override;
80  void PushRegister(int register_index,
81                    StackCheckFlag check_stack_limit) override;
82  void ReadCurrentPositionFromRegister(int reg) override;
83  void ReadStackPointerFromRegister(int reg) override;
84  void SetCurrentPositionFromEnd(int by) override;
85  void SetRegister(int register_index, int to) override;
86  bool Succeed() override;
87  void WriteCurrentPositionToRegister(int reg, int cp_offset) override;
88  void ClearRegisters(int reg_from, int reg_to) override;
89  void WriteStackPointerToRegister(int reg) override;
90
91  // Called from RegExp if the stack-guard is triggered.
92  // If the code object is relocated, the return address is fixed before
93  // returning.
94  // {raw_code} is an Address because this is called via ExternalReference.
95  static int CheckStackGuardState(Address* return_address, Address raw_code,
96                                  Address re_frame, int start_offset,
97                                  const byte** input_start,
98                                  const byte** input_end);
99
100 private:
101  // Above the frame pointer - Stored registers and stack passed parameters.
102  static const int kFramePointer = 0;
103  static const int kReturnAddress = kFramePointer + kSystemPointerSize;
104  // Callee-saved registers (x19-x28).
105  static const int kNumCalleeSavedRegisters = 10;
106  static const int kCalleeSavedRegisters = kReturnAddress + kSystemPointerSize;
107
108  // Below the frame pointer.
109  // Register parameters stored by setup code.
110  static const int kIsolate = -kSystemPointerSize;
111  static const int kDirectCall = kIsolate - kSystemPointerSize;
112  static const int kOutputSize = kDirectCall - kSystemPointerSize;
113  static const int kInput = kOutputSize - kSystemPointerSize;
114  // When adding local variables remember to push space for them in
115  // the frame in GetCode.
116  static const int kSuccessCounter = kInput - kSystemPointerSize;
117  static const int kBacktrackCount = kSuccessCounter - kSystemPointerSize;
118  // Stores the initial value of the regexp stack pointer in a
119  // position-independent representation (in case the regexp stack grows and
120  // thus moves).
121  static const int kRegExpStackBasePointer =
122      kBacktrackCount - kSystemPointerSize;
123  // A padding slot to preserve alignment.
124  static const int kStackLocalPadding =
125      kRegExpStackBasePointer - kSystemPointerSize;
126  static constexpr int kNumberOfStackLocals = 4;
127
128  // First position register address on the stack. Following positions are
129  // below it. A position is a 32 bit value.
130  static const int kFirstRegisterOnStack = kStackLocalPadding - kWRegSize;
131  // A capture is a 64 bit value holding two position.
132  static const int kFirstCaptureOnStack = kStackLocalPadding - kXRegSize;
133
134  // Initial size of code buffer.
135  static const int kRegExpCodeSize = 1024;
136
137  // Registers x0 to x7 are used to store the first captures, they need to be
138  // retained over calls to C++ code.
139  void PushCachedRegisters();
140  void PopCachedRegisters();
141
142  // When initializing registers to a non-position value we can unroll
143  // the loop. Set the limit of registers to unroll.
144  static const int kNumRegistersToUnroll = 16;
145
146  // We are using x0 to x7 as a register cache. Each hardware register must
147  // contain one capture, that is two 32 bit registers. We can cache at most
148  // 16 registers.
149  static const int kNumCachedRegisters = 16;
150
151  // Check whether preemption has been requested.
152  void CheckPreemption();
153
154  // Check whether we are exceeding the stack limit on the backtrack stack.
155  void CheckStackLimit();
156
157  void CallCheckStackGuardState(Register scratch);
158  void CallIsCharacterInRangeArray(const ZoneList<CharacterRange>* ranges);
159
160  // Location of a 32 bit position register.
161  MemOperand register_location(int register_index);
162
163  // Location of a 64 bit capture, combining two position registers.
164  MemOperand capture_location(int register_index, Register scratch);
165
166  // Register holding the current input position as negative offset from
167  // the end of the string.
168  static constexpr Register current_input_offset() { return w21; }
169
170  // The register containing the current character after LoadCurrentCharacter.
171  static constexpr Register current_character() { return w22; }
172
173  // Register holding address of the end of the input string.
174  static constexpr Register input_end() { return x25; }
175
176  // Register holding address of the start of the input string.
177  static constexpr Register input_start() { return x26; }
178
179  // Register holding the offset from the start of the string where we should
180  // start matching.
181  static constexpr Register start_offset() { return w27; }
182
183  // Pointer to the output array's first element.
184  static constexpr Register output_array() { return x28; }
185
186  // Register holding the frame address. Local variables, parameters and
187  // regexp registers are addressed relative to this.
188  static constexpr Register frame_pointer() { return fp; }
189
190  // The register containing the backtrack stack top. Provides a meaningful
191  // name to the register.
192  static constexpr Register backtrack_stackpointer() { return x23; }
193
194  // Register holding pointer to the current code object.
195  static constexpr Register code_pointer() { return x20; }
196
197  // Register holding the value used for clearing capture registers.
198  static constexpr Register string_start_minus_one() { return w24; }
199  // The top 32 bit of this register is used to store this value
200  // twice. This is used for clearing more than one register at a time.
201  static constexpr Register twice_non_position_value() { return x24; }
202
203  // Byte size of chars in the string to match (decided by the Mode argument)
204  int char_size() const { return static_cast<int>(mode_); }
205
206  // Equivalent to a conditional branch to the label, unless the label
207  // is nullptr, in which case it is a conditional Backtrack.
208  void BranchOrBacktrack(Condition condition, Label* to);
209
210  // Compares reg against immmediate before calling BranchOrBacktrack.
211  // It makes use of the Cbz and Cbnz instructions.
212  void CompareAndBranchOrBacktrack(Register reg,
213                                   int immediate,
214                                   Condition condition,
215                                   Label* to);
216
217  inline void CallIf(Label* to, Condition condition);
218
219  // Save and restore the link register on the stack in a way that
220  // is GC-safe.
221  inline void SaveLinkRegister();
222  inline void RestoreLinkRegister();
223
224  // Pushes the value of a register on the backtrack stack. Decrements the
225  // stack pointer by a word size and stores the register's value there.
226  inline void Push(Register source);
227
228  // Pops a value from the backtrack stack. Reads the word at the stack pointer
229  // and increments it by a word size.
230  inline void Pop(Register target);
231
232  // This state indicates where the register actually is.
233  enum RegisterState {
234    STACKED,     // Resides in memory.
235    CACHED_LSW,  // Least Significant Word of a 64 bit hardware register.
236    CACHED_MSW   // Most Significant Word of a 64 bit hardware register.
237  };
238
239  RegisterState GetRegisterState(int register_index) {
240    DCHECK_LE(0, register_index);
241    if (register_index >= kNumCachedRegisters) {
242      return STACKED;
243    } else {
244      if ((register_index % 2) == 0) {
245        return CACHED_LSW;
246      } else {
247        return CACHED_MSW;
248      }
249    }
250  }
251
252  // Store helper that takes the state of the register into account.
253  inline void StoreRegister(int register_index, Register source);
254
255  // Returns a hardware W register that holds the value of the capture
256  // register.
257  //
258  // This function will try to use an existing cache register (w0-w7) for the
259  // result. Otherwise, it will load the value into maybe_result.
260  //
261  // If the returned register is anything other than maybe_result, calling code
262  // must not write to it.
263  inline Register GetRegister(int register_index, Register maybe_result);
264
265  // Returns the harware register (x0-x7) holding the value of the capture
266  // register.
267  // This assumes that the state of the register is not STACKED.
268  inline Register GetCachedRegister(int register_index);
269
270  void LoadRegExpStackPointerFromMemory(Register dst);
271  void StoreRegExpStackPointerToMemory(Register src, Register scratch);
272  void PushRegExpBasePointer(Register stack_pointer, Register scratch);
273  void PopRegExpBasePointer(Register stack_pointer_out, Register scratch);
274
275  Isolate* isolate() const { return masm_->isolate(); }
276
277  const std::unique_ptr<MacroAssembler> masm_;
278  const NoRootArrayScope no_root_array_scope_;
279
280  // Which mode to generate code for (LATIN1 or UC16).
281  const Mode mode_;
282
283  // One greater than maximal register index actually used.
284  int num_registers_;
285
286  // Number of registers to output at the end (the saved registers
287  // are always 0..num_saved_registers_-1)
288  const int num_saved_registers_;
289
290  // Labels used internally.
291  Label entry_label_;
292  Label start_label_;
293  Label success_label_;
294  Label backtrack_label_;
295  Label exit_label_;
296  Label check_preempt_label_;
297  Label stack_overflow_label_;
298  Label fallback_label_;
299};
300
301}  // namespace internal
302}  // namespace v8
303
304#endif  // V8_REGEXP_ARM64_REGEXP_MACRO_ASSEMBLER_ARM64_H_
305