1 //===- X86DisassemblerTables.h - Disassembler tables ------------*- 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 the disassembler tables. 11 // Documentation for the disassembler emitter in general can be found in 12 // X86DisassemblerEmitter.h. 13 // 14 //===----------------------------------------------------------------------===// 15 16 #ifndef LLVM_UTILS_TABLEGEN_X86DISASSEMBLERTABLES_H 17 #define LLVM_UTILS_TABLEGEN_X86DISASSEMBLERTABLES_H 18 19 #include "X86DisassemblerShared.h" 20 #include "llvm/Support/X86DisassemblerDecoderCommon.h" 21 #include <map> 22 #include <memory> 23 #include <vector> 24 25 namespace llvm { 26 class raw_ostream; 27 28 namespace X86Disassembler { 29 30 class ModRMFilter; 31 32 /// DisassemblerTables - Encapsulates all the decode tables being generated by 33 /// the table emitter. Contains functions to populate the tables as well as 34 /// to emit them as hierarchical C structures suitable for consumption by the 35 /// runtime. 36 class DisassemblerTables { 37 private: 38 /// The decoder tables. There is one for each opcode type: 39 /// [0] one-byte opcodes 40 /// [1] two-byte opcodes of the form 0f __ 41 /// [2] three-byte opcodes of the form 0f 38 __ 42 /// [3] three-byte opcodes of the form 0f 3a __ 43 /// [4] XOP8 map opcode 44 /// [5] XOP9 map opcode 45 /// [6] XOPA map opcode 46 /// [7] 3dnow map opcode 47 /// [8] fixed length MAP5 opcode 48 /// [9] fixed length MAP6 opcode 49 std::unique_ptr<ContextDecision> Tables[10]; 50 51 // Table of ModRM encodings. 52 typedef std::map<std::vector<unsigned>, unsigned> ModRMMapTy; 53 mutable ModRMMapTy ModRMTable; 54 55 /// The instruction information table 56 std::vector<InstructionSpecifier> InstructionSpecifiers; 57 58 /// True if there are primary decode conflicts in the instruction set 59 bool HasConflicts; 60 61 /// emitModRMDecision - Emits a table of entries corresponding to a single 62 /// ModR/M decision. Compacts the ModR/M decision if possible. ModR/M 63 /// decisions are printed as: 64 /// 65 /// { /* struct ModRMDecision */ 66 /// TYPE, 67 /// modRMTablennnn 68 /// } 69 /// 70 /// where nnnn is a unique ID for the corresponding table of IDs. 71 /// TYPE indicates whether the table has one entry that is the same 72 /// regardless of ModR/M byte, two entries - one for bytes 0x00-0xbf and one 73 /// for bytes 0xc0-0xff -, or 256 entries, one for each possible byte. 74 /// nnnn is the number of a table for looking up these values. The tables 75 /// are written separately so that tables consisting entirely of zeros will 76 /// not be duplicated. (These all have the name modRMEmptyTable.) A table 77 /// is printed as: 78 /// 79 /// InstrUID modRMTablennnn[k] = { 80 /// nnnn, /* MNEMONIC */ 81 /// ... 82 /// nnnn /* MNEMONIC */ 83 /// }; 84 /// 85 /// @param o1 - The output stream to print the ID table to. 86 /// @param o2 - The output stream to print the decision structure to. 87 /// @param i1 - The indentation level to use with stream o1. 88 /// @param i2 - The indentation level to use with stream o2. 89 /// @param ModRMTableNum - next table number for adding to ModRMTable. 90 /// @param decision - The ModR/M decision to emit. This decision has 256 91 /// entries - emitModRMDecision decides how to compact it. 92 void emitModRMDecision(raw_ostream &o1, raw_ostream &o2, 93 unsigned &i1, unsigned &i2, unsigned &ModRMTableNum, 94 ModRMDecision &decision) const; 95 96 /// emitOpcodeDecision - Emits an OpcodeDecision and all its subsidiary ModR/M 97 /// decisions. An OpcodeDecision is printed as: 98 /// 99 /// { /* struct OpcodeDecision */ 100 /// /* 0x00 */ 101 /// { /* struct ModRMDecision */ 102 /// ... 103 /// } 104 /// ... 105 /// } 106 /// 107 /// where the ModRMDecision structure is printed as described in the 108 /// documentation for emitModRMDecision(). emitOpcodeDecision() passes on a 109 /// stream and indent level for the UID tables generated by 110 /// emitModRMDecision(), but does not use them itself. 111 /// 112 /// @param o1 - The output stream to print the ID tables generated by 113 /// emitModRMDecision() to. 114 /// @param o2 - The output stream for the decision structure itself. 115 /// @param i1 - The indent level to use with stream o1. 116 /// @param i2 - The indent level to use with stream o2. 117 /// @param ModRMTableNum - next table number for adding to ModRMTable. 118 /// @param decision - The OpcodeDecision to emit along with its subsidiary 119 /// structures. 120 void emitOpcodeDecision(raw_ostream &o1, raw_ostream &o2, 121 unsigned &i1, unsigned &i2, unsigned &ModRMTableNum, 122 OpcodeDecision &decision) const; 123 124 /// emitContextDecision - Emits a ContextDecision and all its subsidiary 125 /// Opcode and ModRMDecisions. A ContextDecision is printed as: 126 /// 127 /// struct ContextDecision NAME = { 128 /// { /* OpcodeDecisions */ 129 /// /* IC */ 130 /// { /* struct OpcodeDecision */ 131 /// ... 132 /// }, 133 /// ... 134 /// } 135 /// } 136 /// 137 /// NAME is the name of the ContextDecision (typically one of the four names 138 /// ONEBYTE_SYM, TWOBYTE_SYM, THREEBYTE38_SYM, THREEBYTE3A_SYM from 139 /// X86DisassemblerDecoderCommon.h). 140 /// IC is one of the contexts in InstructionContext. There is an opcode 141 /// decision for each possible context. 142 /// The OpcodeDecision structures are printed as described in the 143 /// documentation for emitOpcodeDecision. 144 /// 145 /// @param o1 - The output stream to print the ID tables generated by 146 /// emitModRMDecision() to. 147 /// @param o2 - The output stream to print the decision structure to. 148 /// @param i1 - The indent level to use with stream o1. 149 /// @param i2 - The indent level to use with stream o2. 150 /// @param ModRMTableNum - next table number for adding to ModRMTable. 151 /// @param decision - The ContextDecision to emit along with its subsidiary 152 /// structures. 153 /// @param name - The name for the ContextDecision. 154 void emitContextDecision(raw_ostream &o1, raw_ostream &o2, 155 unsigned &i1, unsigned &i2, unsigned &ModRMTableNum, 156 ContextDecision &decision, const char* name) const; 157 158 /// emitInstructionInfo - Prints the instruction specifier table, which has 159 /// one entry for each instruction, and contains name and operand 160 /// information. This table is printed as: 161 /// 162 /// struct InstructionSpecifier CONTEXTS_SYM[k] = { 163 /// { 164 /// /* nnnn */ 165 /// "MNEMONIC", 166 /// 0xnn, 167 /// { 168 /// { 169 /// ENCODING, 170 /// TYPE 171 /// }, 172 /// ... 173 /// } 174 /// }, 175 /// }; 176 /// 177 /// k is the total number of instructions. 178 /// nnnn is the ID of the current instruction (0-based). This table 179 /// includes entries for non-instructions like PHINODE. 180 /// 0xnn is the lowest possible opcode for the current instruction, used for 181 /// AddRegFrm instructions to compute the operand's value. 182 /// ENCODING and TYPE describe the encoding and type for a single operand. 183 /// 184 /// @param o - The output stream to which the instruction table should be 185 /// written. 186 /// @param i - The indent level for use with the stream. 187 void emitInstructionInfo(raw_ostream &o, unsigned &i) const; 188 189 /// emitContextTable - Prints the table that is used to translate from an 190 /// instruction attribute mask to an instruction context. This table is 191 /// printed as: 192 /// 193 /// InstructionContext CONTEXTS_STR[256] = { 194 /// IC, /* 0x00 */ 195 /// ... 196 /// }; 197 /// 198 /// IC is the context corresponding to the mask 0x00, and there are 256 199 /// possible masks. 200 /// 201 /// @param o - The output stream to which the context table should be written. 202 /// @param i - The indent level for use with the stream. 203 void emitContextTable(raw_ostream &o, uint32_t &i) const; 204 205 /// emitContextDecisions - Prints all four ContextDecision structures using 206 /// emitContextDecision(). 207 /// 208 /// @param o1 - The output stream to print the ID tables generated by 209 /// emitModRMDecision() to. 210 /// @param o2 - The output stream to print the decision structures to. 211 /// @param i1 - The indent level to use with stream o1. 212 /// @param i2 - The indent level to use with stream o2. 213 /// @param ModRMTableNum - next table number for adding to ModRMTable. 214 void emitContextDecisions(raw_ostream &o1, raw_ostream &o2, 215 unsigned &i1, unsigned &i2, 216 unsigned &ModRMTableNum) const; 217 218 /// setTableFields - Uses a ModRMFilter to set the appropriate entries in a 219 /// ModRMDecision to refer to a particular instruction ID. 220 /// 221 /// @param decision - The ModRMDecision to populate. 222 /// @param filter - The filter to use in deciding which entries to populate. 223 /// @param uid - The unique ID to set matching entries to. 224 /// @param opcode - The opcode of the instruction, for error reporting. 225 void setTableFields(ModRMDecision &decision, 226 const ModRMFilter &filter, 227 InstrUID uid, 228 uint8_t opcode); 229 public: 230 /// Constructor - Allocates space for the class decisions and clears them. 231 DisassemblerTables(); 232 233 ~DisassemblerTables(); 234 235 /// emit - Emits the instruction table, context table, and class decisions. 236 /// 237 /// @param o - The output stream to print the tables to. 238 void emit(raw_ostream &o) const; 239 240 /// setTableFields - Uses the opcode type, instruction context, opcode, and a 241 /// ModRMFilter as criteria to set a particular set of entries in the 242 /// decode tables to point to a specific uid. 243 /// 244 /// @param type - The opcode type (ONEBYTE, TWOBYTE, etc.) 245 /// @param insnContext - The context to use (IC, IC_64BIT, etc.) 246 /// @param opcode - The last byte of the opcode (not counting any escape 247 /// or extended opcodes). 248 /// @param filter - The ModRMFilter that decides which ModR/M byte values 249 /// correspond to the desired instruction. 250 /// @param uid - The unique ID of the instruction. 251 /// @param is32bit - Instructon is only 32-bit 252 /// @param noPrefix - Instruction record has no prefix. 253 /// @param ignoresVEX_L - Instruction ignores VEX.L 254 /// @param ignoresVEX_W - Instruction ignores VEX.W 255 /// @param AddrSize - Instructions address size 16/32/64. 0 is unspecified 256 void setTableFields(OpcodeType type, 257 InstructionContext insnContext, 258 uint8_t opcode, 259 const ModRMFilter &filter, 260 InstrUID uid, 261 bool is32bit, 262 bool noPrefix, 263 bool ignoresVEX_L, 264 bool ignoresVEX_W, 265 unsigned AddrSize); 266 267 /// specForUID - Returns the instruction specifier for a given unique 268 /// instruction ID. Used when resolving collisions. 269 /// 270 /// @param uid - The unique ID of the instruction. 271 /// @return - A reference to the instruction specifier. 272 InstructionSpecifier& specForUID(InstrUID uid) { 273 if (uid >= InstructionSpecifiers.size()) 274 InstructionSpecifiers.resize(uid + 1); 275 276 return InstructionSpecifiers[uid]; 277 } 278 279 // hasConflicts - Reports whether there were primary decode conflicts 280 // from any instructions added to the tables. 281 // @return - true if there were; false otherwise. 282 283 bool hasConflicts() { 284 return HasConflicts; 285 } 286 }; 287 288 } // namespace X86Disassembler 289 290 } // namespace llvm 291 292 #endif 293