gn build: Merge r372706
[llvm-complete.git] / utils / TableGen / X86RecognizableInstr.h
blobb15bef4e1931d208319310547c3db3aa0d0dc054
1 //===- X86RecognizableInstr.h - Disassembler instruction spec ----*- C++ -*-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file is part of the X86 Disassembler Emitter.
10 // It contains the interface of a single recognizable instruction.
11 // Documentation for the disassembler emitter in general can be found in
12 // X86DisassemblerEmitter.h.
14 //===----------------------------------------------------------------------===//
16 #ifndef LLVM_UTILS_TABLEGEN_X86RECOGNIZABLEINSTR_H
17 #define LLVM_UTILS_TABLEGEN_X86RECOGNIZABLEINSTR_H
19 #include "CodeGenTarget.h"
20 #include "X86DisassemblerTables.h"
21 #include "llvm/Support/DataTypes.h"
22 #include "llvm/TableGen/Record.h"
24 namespace llvm {
26 #define X86_INSTR_MRM_MAPPING \
27 MAP(C0, 64) \
28 MAP(C1, 65) \
29 MAP(C2, 66) \
30 MAP(C3, 67) \
31 MAP(C4, 68) \
32 MAP(C5, 69) \
33 MAP(C6, 70) \
34 MAP(C7, 71) \
35 MAP(C8, 72) \
36 MAP(C9, 73) \
37 MAP(CA, 74) \
38 MAP(CB, 75) \
39 MAP(CC, 76) \
40 MAP(CD, 77) \
41 MAP(CE, 78) \
42 MAP(CF, 79) \
43 MAP(D0, 80) \
44 MAP(D1, 81) \
45 MAP(D2, 82) \
46 MAP(D3, 83) \
47 MAP(D4, 84) \
48 MAP(D5, 85) \
49 MAP(D6, 86) \
50 MAP(D7, 87) \
51 MAP(D8, 88) \
52 MAP(D9, 89) \
53 MAP(DA, 90) \
54 MAP(DB, 91) \
55 MAP(DC, 92) \
56 MAP(DD, 93) \
57 MAP(DE, 94) \
58 MAP(DF, 95) \
59 MAP(E0, 96) \
60 MAP(E1, 97) \
61 MAP(E2, 98) \
62 MAP(E3, 99) \
63 MAP(E4, 100) \
64 MAP(E5, 101) \
65 MAP(E6, 102) \
66 MAP(E7, 103) \
67 MAP(E8, 104) \
68 MAP(E9, 105) \
69 MAP(EA, 106) \
70 MAP(EB, 107) \
71 MAP(EC, 108) \
72 MAP(ED, 109) \
73 MAP(EE, 110) \
74 MAP(EF, 111) \
75 MAP(F0, 112) \
76 MAP(F1, 113) \
77 MAP(F2, 114) \
78 MAP(F3, 115) \
79 MAP(F4, 116) \
80 MAP(F5, 117) \
81 MAP(F6, 118) \
82 MAP(F7, 119) \
83 MAP(F8, 120) \
84 MAP(F9, 121) \
85 MAP(FA, 122) \
86 MAP(FB, 123) \
87 MAP(FC, 124) \
88 MAP(FD, 125) \
89 MAP(FE, 126) \
90 MAP(FF, 127)
92 // A clone of X86 since we can't depend on something that is generated.
93 namespace X86Local {
94 enum {
95 Pseudo = 0,
96 RawFrm = 1,
97 AddRegFrm = 2,
98 RawFrmMemOffs = 3,
99 RawFrmSrc = 4,
100 RawFrmDst = 5,
101 RawFrmDstSrc = 6,
102 RawFrmImm8 = 7,
103 RawFrmImm16 = 8,
104 AddCCFrm = 9,
105 MRMDestMem = 32,
106 MRMSrcMem = 33,
107 MRMSrcMem4VOp3 = 34,
108 MRMSrcMemOp4 = 35,
109 MRMSrcMemCC = 36,
110 MRMXmCC = 38, MRMXm = 39,
111 MRM0m = 40, MRM1m = 41, MRM2m = 42, MRM3m = 43,
112 MRM4m = 44, MRM5m = 45, MRM6m = 46, MRM7m = 47,
113 MRMDestReg = 48,
114 MRMSrcReg = 49,
115 MRMSrcReg4VOp3 = 50,
116 MRMSrcRegOp4 = 51,
117 MRMSrcRegCC = 52,
118 MRMXrCC = 54, MRMXr = 55,
119 MRM0r = 56, MRM1r = 57, MRM2r = 58, MRM3r = 59,
120 MRM4r = 60, MRM5r = 61, MRM6r = 62, MRM7r = 63,
121 #define MAP(from, to) MRM_##from = to,
122 X86_INSTR_MRM_MAPPING
123 #undef MAP
126 enum {
127 OB = 0, TB = 1, T8 = 2, TA = 3, XOP8 = 4, XOP9 = 5, XOPA = 6, ThreeDNow = 7
130 enum {
131 PD = 1, XS = 2, XD = 3, PS = 4
134 enum {
135 VEX = 1, XOP = 2, EVEX = 3
138 enum {
139 OpSize16 = 1, OpSize32 = 2
142 enum {
143 AdSize16 = 1, AdSize32 = 2, AdSize64 = 3
147 namespace X86Disassembler {
149 /// RecognizableInstr - Encapsulates all information required to decode a single
150 /// instruction, as extracted from the LLVM instruction tables. Has methods
151 /// to interpret the information available in the LLVM tables, and to emit the
152 /// instruction into DisassemblerTables.
153 class RecognizableInstr {
154 private:
155 /// The opcode of the instruction, as used in an MCInst
156 InstrUID UID;
157 /// The record from the .td files corresponding to this instruction
158 const Record* Rec;
159 /// The OpPrefix field from the record
160 uint8_t OpPrefix;
161 /// The OpMap field from the record
162 uint8_t OpMap;
163 /// The opcode field from the record; this is the opcode used in the Intel
164 /// encoding and therefore distinct from the UID
165 uint8_t Opcode;
166 /// The form field from the record
167 uint8_t Form;
168 // The encoding field from the record
169 uint8_t Encoding;
170 /// The OpSize field from the record
171 uint8_t OpSize;
172 /// The AdSize field from the record
173 uint8_t AdSize;
174 /// The hasREX_WPrefix field from the record
175 bool HasREX_WPrefix;
176 /// The hasVEX_4V field from the record
177 bool HasVEX_4V;
178 /// The HasVEX_WPrefix field from the record
179 bool HasVEX_W;
180 /// The IgnoresVEX_W field from the record
181 bool IgnoresVEX_W;
182 /// Inferred from the operands; indicates whether the L bit in the VEX prefix is set
183 bool HasVEX_LPrefix;
184 /// The ignoreVEX_L field from the record
185 bool IgnoresVEX_L;
186 /// The hasEVEX_L2Prefix field from the record
187 bool HasEVEX_L2Prefix;
188 /// The hasEVEX_K field from the record
189 bool HasEVEX_K;
190 /// The hasEVEX_KZ field from the record
191 bool HasEVEX_KZ;
192 /// The hasEVEX_B field from the record
193 bool HasEVEX_B;
194 /// Indicates that the instruction uses the L and L' fields for RC.
195 bool EncodeRC;
196 /// The isCodeGenOnly field from the record
197 bool IsCodeGenOnly;
198 /// The ForceDisassemble field from the record
199 bool ForceDisassemble;
200 // The CD8_Scale field from the record
201 uint8_t CD8_Scale;
202 // Whether the instruction has the predicate "In64BitMode"
203 bool Is64Bit;
204 // Whether the instruction has the predicate "In32BitMode"
205 bool Is32Bit;
207 /// The instruction name as listed in the tables
208 std::string Name;
210 /// Indicates whether the instruction should be emitted into the decode
211 /// tables; regardless, it will be emitted into the instruction info table
212 bool ShouldBeEmitted;
214 /// The operands of the instruction, as listed in the CodeGenInstruction.
215 /// They are not one-to-one with operands listed in the MCInst; for example,
216 /// memory operands expand to 5 operands in the MCInst
217 const std::vector<CGIOperandList::OperandInfo>* Operands;
219 /// The description of the instruction that is emitted into the instruction
220 /// info table
221 InstructionSpecifier* Spec;
223 /// insnContext - Returns the primary context in which the instruction is
224 /// valid.
226 /// @return - The context in which the instruction is valid.
227 InstructionContext insnContext() const;
229 /// typeFromString - Translates an operand type from the string provided in
230 /// the LLVM tables to an OperandType for use in the operand specifier.
232 /// @param s - The string, as extracted by calling Rec->getName()
233 /// on a CodeGenInstruction::OperandInfo.
234 /// @param hasREX_WPrefix - Indicates whether the instruction has a REX.W
235 /// prefix. If it does, 32-bit register operands stay
236 /// 32-bit regardless of the operand size.
237 /// @param OpSize Indicates the operand size of the instruction.
238 /// If register size does not match OpSize, then
239 /// register sizes keep their size.
240 /// @return - The operand's type.
241 static OperandType typeFromString(const std::string& s,
242 bool hasREX_WPrefix, uint8_t OpSize);
244 /// immediateEncodingFromString - Translates an immediate encoding from the
245 /// string provided in the LLVM tables to an OperandEncoding for use in
246 /// the operand specifier.
248 /// @param s - See typeFromString().
249 /// @param OpSize - Indicates whether this is an OpSize16 instruction.
250 /// If it is not, then 16-bit immediate operands stay 16-bit.
251 /// @return - The operand's encoding.
252 static OperandEncoding immediateEncodingFromString(const std::string &s,
253 uint8_t OpSize);
255 /// rmRegisterEncodingFromString - Like immediateEncodingFromString, but
256 /// handles operands that are in the REG field of the ModR/M byte.
257 static OperandEncoding rmRegisterEncodingFromString(const std::string &s,
258 uint8_t OpSize);
260 /// rmRegisterEncodingFromString - Like immediateEncodingFromString, but
261 /// handles operands that are in the REG field of the ModR/M byte.
262 static OperandEncoding roRegisterEncodingFromString(const std::string &s,
263 uint8_t OpSize);
264 static OperandEncoding memoryEncodingFromString(const std::string &s,
265 uint8_t OpSize);
266 static OperandEncoding relocationEncodingFromString(const std::string &s,
267 uint8_t OpSize);
268 static OperandEncoding opcodeModifierEncodingFromString(const std::string &s,
269 uint8_t OpSize);
270 static OperandEncoding vvvvRegisterEncodingFromString(const std::string &s,
271 uint8_t OpSize);
272 static OperandEncoding writemaskRegisterEncodingFromString(const std::string &s,
273 uint8_t OpSize);
275 /// Adjust the encoding type for an operand based on the instruction.
276 void adjustOperandEncoding(OperandEncoding &encoding);
278 /// handleOperand - Converts a single operand from the LLVM table format to
279 /// the emitted table format, handling any duplicate operands it encounters
280 /// and then one non-duplicate.
282 /// @param optional - Determines whether to assert that the
283 /// operand exists.
284 /// @param operandIndex - The index into the generated operand table.
285 /// Incremented by this function one or more
286 /// times to reflect possible duplicate
287 /// operands).
288 /// @param physicalOperandIndex - The index of the current operand into the
289 /// set of non-duplicate ('physical') operands.
290 /// Incremented by this function once.
291 /// @param numPhysicalOperands - The number of non-duplicate operands in the
292 /// instructions.
293 /// @param operandMapping - The operand mapping, which has an entry for
294 /// each operand that indicates whether it is a
295 /// duplicate, and of what.
296 void handleOperand(bool optional,
297 unsigned &operandIndex,
298 unsigned &physicalOperandIndex,
299 unsigned numPhysicalOperands,
300 const unsigned *operandMapping,
301 OperandEncoding (*encodingFromString)
302 (const std::string&,
303 uint8_t OpSize));
305 /// shouldBeEmitted - Returns the shouldBeEmitted field. Although filter()
306 /// filters out many instructions, at various points in decoding we
307 /// determine that the instruction should not actually be decodable. In
308 /// particular, MMX MOV instructions aren't emitted, but they're only
309 /// identified during operand parsing.
311 /// @return - true if at this point we believe the instruction should be
312 /// emitted; false if not. This will return false if filter() returns false
313 /// once emitInstructionSpecifier() has been called.
314 bool shouldBeEmitted() const {
315 return ShouldBeEmitted;
318 /// emitInstructionSpecifier - Loads the instruction specifier for the current
319 /// instruction into a DisassemblerTables.
321 void emitInstructionSpecifier();
323 /// emitDecodePath - Populates the proper fields in the decode tables
324 /// corresponding to the decode paths for this instruction.
326 /// \param tables The DisassemblerTables to populate with the decode
327 /// decode information for the current instruction.
328 void emitDecodePath(DisassemblerTables &tables) const;
330 /// Constructor - Initializes a RecognizableInstr with the appropriate fields
331 /// from a CodeGenInstruction.
333 /// \param tables The DisassemblerTables that the specifier will be added to.
334 /// \param insn The CodeGenInstruction to extract information from.
335 /// \param uid The unique ID of the current instruction.
336 RecognizableInstr(DisassemblerTables &tables,
337 const CodeGenInstruction &insn,
338 InstrUID uid);
339 public:
340 /// processInstr - Accepts a CodeGenInstruction and loads decode information
341 /// for it into a DisassemblerTables if appropriate.
343 /// \param tables The DiassemblerTables to be populated with decode
344 /// information.
345 /// \param insn The CodeGenInstruction to be used as a source for this
346 /// information.
347 /// \param uid The unique ID of the instruction.
348 static void processInstr(DisassemblerTables &tables,
349 const CodeGenInstruction &insn,
350 InstrUID uid);
353 } // namespace X86Disassembler
355 } // namespace llvm
357 #endif