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 "X86ModRMFilters.h" 21 #include "llvm/Support/raw_ostream.h" 22 #include <map> 23 #include <vector> 24 25 namespace llvm { 26 27 namespace X86Disassembler { 28 29 /// DisassemblerTables - Encapsulates all the decode tables being generated by 30 /// the table emitter. Contains functions to populate the tables as well as 31 /// to emit them as hierarchical C structures suitable for consumption by the 32 /// runtime. 33 class DisassemblerTables { 34 private: 35 /// The decoder tables. There is one for each opcode type: 36 /// [0] one-byte opcodes 37 /// [1] two-byte opcodes of the form 0f __ 38 /// [2] three-byte opcodes of the form 0f 38 __ 39 /// [3] three-byte opcodes of the form 0f 3a __ 40 /// [4] XOP8 map opcode 41 /// [5] XOP9 map opcode 42 /// [6] XOPA map opcode 43 /// [7] 3dnow map opcode 44 std::unique_ptr<ContextDecision> Tables[8]; 45 46 // Table of ModRM encodings. 47 typedef std::map<std::vector<unsigned>, unsigned> ModRMMapTy; 48 mutable ModRMMapTy ModRMTable; 49 50 /// The instruction information table 51 std::vector<InstructionSpecifier> InstructionSpecifiers; 52 53 /// True if there are primary decode conflicts in the instruction set 54 bool HasConflicts; 55 56 /// emitModRMDecision - Emits a table of entries corresponding to a single 57 /// ModR/M decision. Compacts the ModR/M decision if possible. ModR/M 58 /// decisions are printed as: 59 /// 60 /// { /* struct ModRMDecision */ 61 /// TYPE, 62 /// modRMTablennnn 63 /// } 64 /// 65 /// where nnnn is a unique ID for the corresponding table of IDs. 66 /// TYPE indicates whether the table has one entry that is the same 67 /// regardless of ModR/M byte, two entries - one for bytes 0x00-0xbf and one 68 /// for bytes 0xc0-0xff -, or 256 entries, one for each possible byte. 69 /// nnnn is the number of a table for looking up these values. The tables 70 /// are written separately so that tables consisting entirely of zeros will 71 /// not be duplicated. (These all have the name modRMEmptyTable.) A table 72 /// is printed as: 73 /// 74 /// InstrUID modRMTablennnn[k] = { 75 /// nnnn, /* MNEMONIC */ 76 /// ... 77 /// nnnn /* MNEMONIC */ 78 /// }; 79 /// 80 /// @param o1 - The output stream to print the ID table to. 81 /// @param o2 - The output stream to print the decision structure to. 82 /// @param i1 - The indentation level to use with stream o1. 83 /// @param i2 - The indentation level to use with stream o2. 84 /// @param ModRMTableNum - next table number for adding to ModRMTable. 85 /// @param decision - The ModR/M decision to emit. This decision has 256 86 /// entries - emitModRMDecision decides how to compact it. 87 void emitModRMDecision(raw_ostream &o1, raw_ostream &o2, 88 unsigned &i1, unsigned &i2, unsigned &ModRMTableNum, 89 ModRMDecision &decision) const; 90 91 /// emitOpcodeDecision - Emits an OpcodeDecision and all its subsidiary ModR/M 92 /// decisions. An OpcodeDecision is printed as: 93 /// 94 /// { /* struct OpcodeDecision */ 95 /// /* 0x00 */ 96 /// { /* struct ModRMDecision */ 97 /// ... 98 /// } 99 /// ... 100 /// } 101 /// 102 /// where the ModRMDecision structure is printed as described in the 103 /// documentation for emitModRMDecision(). emitOpcodeDecision() passes on a 104 /// stream and indent level for the UID tables generated by 105 /// emitModRMDecision(), but does not use them itself. 106 /// 107 /// @param o1 - The output stream to print the ID tables generated by 108 /// emitModRMDecision() to. 109 /// @param o2 - The output stream for the decision structure itself. 110 /// @param i1 - The indent level to use with stream o1. 111 /// @param i2 - The indent level to use with stream o2. 112 /// @param ModRMTableNum - next table number for adding to ModRMTable. 113 /// @param decision - The OpcodeDecision to emit along with its subsidiary 114 /// structures. 115 void emitOpcodeDecision(raw_ostream &o1, raw_ostream &o2, 116 unsigned &i1, unsigned &i2, unsigned &ModRMTableNum, 117 OpcodeDecision &decision) const; 118 119 /// emitContextDecision - Emits a ContextDecision and all its subsidiary 120 /// Opcode and ModRMDecisions. A ContextDecision is printed as: 121 /// 122 /// struct ContextDecision NAME = { 123 /// { /* OpcodeDecisions */ 124 /// /* IC */ 125 /// { /* struct OpcodeDecision */ 126 /// ... 127 /// }, 128 /// ... 129 /// } 130 /// } 131 /// 132 /// NAME is the name of the ContextDecision (typically one of the four names 133 /// ONEBYTE_SYM, TWOBYTE_SYM, THREEBYTE38_SYM, THREEBYTE3A_SYM from 134 /// X86DisassemblerDecoderCommon.h). 135 /// IC is one of the contexts in InstructionContext. There is an opcode 136 /// decision for each possible context. 137 /// The OpcodeDecision structures are printed as described in the 138 /// documentation for emitOpcodeDecision. 139 /// 140 /// @param o1 - The output stream to print the ID tables generated by 141 /// emitModRMDecision() to. 142 /// @param o2 - The output stream to print the decision structure to. 143 /// @param i1 - The indent level to use with stream o1. 144 /// @param i2 - The indent level to use with stream o2. 145 /// @param ModRMTableNum - next table number for adding to ModRMTable. 146 /// @param decision - The ContextDecision to emit along with its subsidiary 147 /// structures. 148 /// @param name - The name for the ContextDecision. 149 void emitContextDecision(raw_ostream &o1, raw_ostream &o2, 150 unsigned &i1, unsigned &i2, unsigned &ModRMTableNum, 151 ContextDecision &decision, const char* name) const; 152 153 /// emitInstructionInfo - Prints the instruction specifier table, which has 154 /// one entry for each instruction, and contains name and operand 155 /// information. This table is printed as: 156 /// 157 /// struct InstructionSpecifier CONTEXTS_SYM[k] = { 158 /// { 159 /// /* nnnn */ 160 /// "MNEMONIC", 161 /// 0xnn, 162 /// { 163 /// { 164 /// ENCODING, 165 /// TYPE 166 /// }, 167 /// ... 168 /// } 169 /// }, 170 /// }; 171 /// 172 /// k is the total number of instructions. 173 /// nnnn is the ID of the current instruction (0-based). This table 174 /// includes entries for non-instructions like PHINODE. 175 /// 0xnn is the lowest possible opcode for the current instruction, used for 176 /// AddRegFrm instructions to compute the operand's value. 177 /// ENCODING and TYPE describe the encoding and type for a single operand. 178 /// 179 /// @param o - The output stream to which the instruction table should be 180 /// written. 181 /// @param i - The indent level for use with the stream. 182 void emitInstructionInfo(raw_ostream &o, unsigned &i) const; 183 184 /// emitContextTable - Prints the table that is used to translate from an 185 /// instruction attribute mask to an instruction context. This table is 186 /// printed as: 187 /// 188 /// InstructionContext CONTEXTS_STR[256] = { 189 /// IC, /* 0x00 */ 190 /// ... 191 /// }; 192 /// 193 /// IC is the context corresponding to the mask 0x00, and there are 256 194 /// possible masks. 195 /// 196 /// @param o - The output stream to which the context table should be written. 197 /// @param i - The indent level for use with the stream. 198 void emitContextTable(raw_ostream &o, uint32_t &i) const; 199 200 /// emitContextDecisions - Prints all four ContextDecision structures using 201 /// emitContextDecision(). 202 /// 203 /// @param o1 - The output stream to print the ID tables generated by 204 /// emitModRMDecision() to. 205 /// @param o2 - The output stream to print the decision structures to. 206 /// @param i1 - The indent level to use with stream o1. 207 /// @param i2 - The indent level to use with stream o2. 208 /// @param ModRMTableNum - next table number for adding to ModRMTable. 209 void emitContextDecisions(raw_ostream &o1, raw_ostream &o2, 210 unsigned &i1, unsigned &i2, 211 unsigned &ModRMTableNum) const; 212 213 /// setTableFields - Uses a ModRMFilter to set the appropriate entries in a 214 /// ModRMDecision to refer to a particular instruction ID. 215 /// 216 /// @param decision - The ModRMDecision to populate. 217 /// @param filter - The filter to use in deciding which entries to populate. 218 /// @param uid - The unique ID to set matching entries to. 219 /// @param opcode - The opcode of the instruction, for error reporting. 220 void setTableFields(ModRMDecision &decision, 221 const ModRMFilter &filter, 222 InstrUID uid, 223 uint8_t opcode); 224 public: 225 /// Constructor - Allocates space for the class decisions and clears them. 226 DisassemblerTables(); 227 228 ~DisassemblerTables(); 229 230 /// emit - Emits the instruction table, context table, and class decisions. 231 /// 232 /// @param o - The output stream to print the tables to. 233 void emit(raw_ostream &o) const; 234 235 /// setTableFields - Uses the opcode type, instruction context, opcode, and a 236 /// ModRMFilter as criteria to set a particular set of entries in the 237 /// decode tables to point to a specific uid. 238 /// 239 /// @param type - The opcode type (ONEBYTE, TWOBYTE, etc.) 240 /// @param insnContext - The context to use (IC, IC_64BIT, etc.) 241 /// @param opcode - The last byte of the opcode (not counting any escape 242 /// or extended opcodes). 243 /// @param filter - The ModRMFilter that decides which ModR/M byte values 244 /// correspond to the desired instruction. 245 /// @param uid - The unique ID of the instruction. 246 /// @param is32bit - Instructon is only 32-bit 247 /// @param noPrefix - Instruction record has no prefix. 248 /// @param ignoresVEX_L - Instruction ignores VEX.L 249 /// @param ignoresVEX_W - Instruction ignores VEX.W 250 /// @param AddrSize - Instructions address size 16/32/64. 0 is unspecified 251 void setTableFields(OpcodeType type, 252 InstructionContext insnContext, 253 uint8_t opcode, 254 const ModRMFilter &filter, 255 InstrUID uid, 256 bool is32bit, 257 bool noPrefix, 258 bool ignoresVEX_L, 259 bool ignoresVEX_W, 260 unsigned AddrSize); 261 262 /// specForUID - Returns the instruction specifier for a given unique 263 /// instruction ID. Used when resolving collisions. 264 /// 265 /// @param uid - The unique ID of the instruction. 266 /// @return - A reference to the instruction specifier. 267 InstructionSpecifier& specForUID(InstrUID uid) { 268 if (uid >= InstructionSpecifiers.size()) 269 InstructionSpecifiers.resize(uid + 1); 270 271 return InstructionSpecifiers[uid]; 272 } 273 274 // hasConflicts - Reports whether there were primary decode conflicts 275 // from any instructions added to the tables. 276 // @return - true if there were; false otherwise. 277 278 bool hasConflicts() { 279 return HasConflicts; 280 } 281 }; 282 283 } // namespace X86Disassembler 284 285 } // namespace llvm 286 287 #endif 288