1//===- X86RecognizableInstr.h - Disassembler instruction spec ----*- C++ -*-===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file is part of the X86 Disassembler Emitter. 11// It contains the interface of a single recognizable instruction. 12// Documentation for the disassembler emitter in general can be found in 13// X86DisasemblerEmitter.h. 14// 15//===----------------------------------------------------------------------===// 16 17#ifndef X86RECOGNIZABLEINSTR_H 18#define X86RECOGNIZABLEINSTR_H 19 20#include "X86DisassemblerTables.h" 21 22#include "CodeGenTarget.h" 23 24#include "llvm/TableGen/Record.h" 25#include "llvm/Support/DataTypes.h" 26#include "llvm/ADT/SmallVector.h" 27 28namespace llvm { 29 30namespace X86Disassembler { 31 32/// RecognizableInstr - Encapsulates all information required to decode a single 33/// instruction, as extracted from the LLVM instruction tables. Has methods 34/// to interpret the information available in the LLVM tables, and to emit the 35/// instruction into DisassemblerTables. 36class RecognizableInstr { 37private: 38 /// The opcode of the instruction, as used in an MCInst 39 InstrUID UID; 40 /// The record from the .td files corresponding to this instruction 41 const Record* Rec; 42 /// The prefix field from the record 43 uint8_t Prefix; 44 /// The opcode field from the record; this is the opcode used in the Intel 45 /// encoding and therefore distinct from the UID 46 uint8_t Opcode; 47 /// The form field from the record 48 uint8_t Form; 49 /// The segment override field from the record 50 uint8_t SegOvr; 51 /// The hasOpSizePrefix field from the record 52 bool HasOpSizePrefix; 53 /// The hasAdSizePrefix field from the record 54 bool HasAdSizePrefix; 55 /// The hasREX_WPrefix field from the record 56 bool HasREX_WPrefix; 57 /// The hasVEXPrefix field from the record 58 bool HasVEXPrefix; 59 /// The hasVEX_4VPrefix field from the record 60 bool HasVEX_4VPrefix; 61 /// The hasVEX_4VOp3Prefix field from the record 62 bool HasVEX_4VOp3Prefix; 63 /// The hasVEX_WPrefix field from the record 64 bool HasVEX_WPrefix; 65 /// Inferred from the operands; indicates whether the L bit in the VEX prefix is set 66 bool HasVEX_LPrefix; 67 /// The hasMemOp4Prefix field from the record 68 bool HasMemOp4Prefix; 69 /// The ignoreVEX_L field from the record 70 bool IgnoresVEX_L; 71 /// The hasLockPrefix field from the record 72 bool HasLockPrefix; 73 /// The isCodeGenOnly filed from the record 74 bool IsCodeGenOnly; 75 // Whether the instruction has the predicate "In64BitMode" 76 bool Is64Bit; 77 // Whether the instruction has the predicate "In32BitMode" 78 bool Is32Bit; 79 80 /// The instruction name as listed in the tables 81 std::string Name; 82 /// The AT&T AsmString for the instruction 83 std::string AsmString; 84 85 /// Indicates whether the instruction is SSE 86 bool IsSSE; 87 /// Indicates whether the instruction has FR operands - MOVs with FR operands 88 /// are typically ignored 89 bool HasFROperands; 90 /// Indicates whether the instruction should be emitted into the decode 91 /// tables; regardless, it will be emitted into the instruction info table 92 bool ShouldBeEmitted; 93 94 /// The operands of the instruction, as listed in the CodeGenInstruction. 95 /// They are not one-to-one with operands listed in the MCInst; for example, 96 /// memory operands expand to 5 operands in the MCInst 97 const std::vector<CGIOperandList::OperandInfo>* Operands; 98 99 /// The description of the instruction that is emitted into the instruction 100 /// info table 101 InstructionSpecifier* Spec; 102 103 /// insnContext - Returns the primary context in which the instruction is 104 /// valid. 105 /// 106 /// @return - The context in which the instruction is valid. 107 InstructionContext insnContext() const; 108 109 enum filter_ret { 110 FILTER_STRONG, // instruction has no place in the instruction tables 111 FILTER_WEAK, // instruction may conflict, and should be eliminated if 112 // it does 113 FILTER_NORMAL // instruction should have high priority and generate an 114 // error if it conflcits with any other FILTER_NORMAL 115 // instruction 116 }; 117 118 /// filter - Determines whether the instruction should be decodable. Some 119 /// instructions are pure intrinsics and use unencodable operands; many 120 /// synthetic instructions are duplicates of other instructions; other 121 /// instructions only differ in the logical way in which they are used, and 122 /// have the same decoding. Because these would cause decode conflicts, 123 /// they must be filtered out. 124 /// 125 /// @return - The degree of filtering to be applied (see filter_ret). 126 filter_ret filter() const; 127 128 /// hasFROperands - Returns true if any operand is a FR operand. 129 bool hasFROperands() const; 130 131 /// has256BitOperands - Returns true if any operand is a 256-bit SSE operand. 132 bool has256BitOperands() const; 133 134 /// typeFromString - Translates an operand type from the string provided in 135 /// the LLVM tables to an OperandType for use in the operand specifier. 136 /// 137 /// @param s - The string, as extracted by calling Rec->getName() 138 /// on a CodeGenInstruction::OperandInfo. 139 /// @param isSSE - Indicates whether the instruction is an SSE 140 /// instruction. For SSE instructions, immediates are 141 /// fixed-size rather than being affected by the 142 /// mandatory OpSize prefix. 143 /// @param hasREX_WPrefix - Indicates whether the instruction has a REX.W 144 /// prefix. If it does, 32-bit register operands stay 145 /// 32-bit regardless of the operand size. 146 /// @param hasOpSizePrefix- Indicates whether the instruction has an OpSize 147 /// prefix. If it does not, then 16-bit register 148 /// operands stay 16-bit. 149 /// @return - The operand's type. 150 static OperandType typeFromString(const std::string& s, 151 bool isSSE, 152 bool hasREX_WPrefix, 153 bool hasOpSizePrefix); 154 155 /// immediateEncodingFromString - Translates an immediate encoding from the 156 /// string provided in the LLVM tables to an OperandEncoding for use in 157 /// the operand specifier. 158 /// 159 /// @param s - See typeFromString(). 160 /// @param hasOpSizePrefix - Indicates whether the instruction has an OpSize 161 /// prefix. If it does not, then 16-bit immediate 162 /// operands stay 16-bit. 163 /// @return - The operand's encoding. 164 static OperandEncoding immediateEncodingFromString(const std::string &s, 165 bool hasOpSizePrefix); 166 167 /// rmRegisterEncodingFromString - Like immediateEncodingFromString, but 168 /// handles operands that are in the REG field of the ModR/M byte. 169 static OperandEncoding rmRegisterEncodingFromString(const std::string &s, 170 bool hasOpSizePrefix); 171 172 /// rmRegisterEncodingFromString - Like immediateEncodingFromString, but 173 /// handles operands that are in the REG field of the ModR/M byte. 174 static OperandEncoding roRegisterEncodingFromString(const std::string &s, 175 bool hasOpSizePrefix); 176 static OperandEncoding memoryEncodingFromString(const std::string &s, 177 bool hasOpSizePrefix); 178 static OperandEncoding relocationEncodingFromString(const std::string &s, 179 bool hasOpSizePrefix); 180 static OperandEncoding opcodeModifierEncodingFromString(const std::string &s, 181 bool hasOpSizePrefix); 182 static OperandEncoding vvvvRegisterEncodingFromString(const std::string &s, 183 bool HasOpSizePrefix); 184 185 /// handleOperand - Converts a single operand from the LLVM table format to 186 /// the emitted table format, handling any duplicate operands it encounters 187 /// and then one non-duplicate. 188 /// 189 /// @param optional - Determines whether to assert that the 190 /// operand exists. 191 /// @param operandIndex - The index into the generated operand table. 192 /// Incremented by this function one or more 193 /// times to reflect possible duplicate 194 /// operands). 195 /// @param physicalOperandIndex - The index of the current operand into the 196 /// set of non-duplicate ('physical') operands. 197 /// Incremented by this function once. 198 /// @param numPhysicalOperands - The number of non-duplicate operands in the 199 /// instructions. 200 /// @param operandMapping - The operand mapping, which has an entry for 201 /// each operand that indicates whether it is a 202 /// duplicate, and of what. 203 void handleOperand(bool optional, 204 unsigned &operandIndex, 205 unsigned &physicalOperandIndex, 206 unsigned &numPhysicalOperands, 207 unsigned *operandMapping, 208 OperandEncoding (*encodingFromString) 209 (const std::string&, 210 bool hasOpSizePrefix)); 211 212 /// shouldBeEmitted - Returns the shouldBeEmitted field. Although filter() 213 /// filters out many instructions, at various points in decoding we 214 /// determine that the instruction should not actually be decodable. In 215 /// particular, MMX MOV instructions aren't emitted, but they're only 216 /// identified during operand parsing. 217 /// 218 /// @return - true if at this point we believe the instruction should be 219 /// emitted; false if not. This will return false if filter() returns false 220 /// once emitInstructionSpecifier() has been called. 221 bool shouldBeEmitted() const { 222 return ShouldBeEmitted; 223 } 224 225 /// emitInstructionSpecifier - Loads the instruction specifier for the current 226 /// instruction into a DisassemblerTables. 227 /// 228 /// @arg tables - The DisassemblerTables to populate with the specifier for 229 /// the current instruction. 230 void emitInstructionSpecifier(DisassemblerTables &tables); 231 232 /// emitDecodePath - Populates the proper fields in the decode tables 233 /// corresponding to the decode paths for this instruction. 234 /// 235 /// @arg tables - The DisassemblerTables to populate with the decode 236 /// decode information for the current instruction. 237 void emitDecodePath(DisassemblerTables &tables) const; 238 239 /// Constructor - Initializes a RecognizableInstr with the appropriate fields 240 /// from a CodeGenInstruction. 241 /// 242 /// @arg tables - The DisassemblerTables that the specifier will be added to. 243 /// @arg insn - The CodeGenInstruction to extract information from. 244 /// @arg uid - The unique ID of the current instruction. 245 RecognizableInstr(DisassemblerTables &tables, 246 const CodeGenInstruction &insn, 247 InstrUID uid); 248public: 249 /// processInstr - Accepts a CodeGenInstruction and loads decode information 250 /// for it into a DisassemblerTables if appropriate. 251 /// 252 /// @arg tables - The DiassemblerTables to be populated with decode 253 /// information. 254 /// @arg insn - The CodeGenInstruction to be used as a source for this 255 /// information. 256 /// @uid - The unique ID of the instruction. 257 static void processInstr(DisassemblerTables &tables, 258 const CodeGenInstruction &insn, 259 InstrUID uid); 260}; 261 262} // namespace X86Disassembler 263 264} // namespace llvm 265 266#endif 267