X86RecognizableInstr.h revision 4da632e6e09b96db4b3f9202cde4e6ca732001c1
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#include "Record.h" 24 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 hasREX_WPrefix field from the record 54 bool HasREX_WPrefix; 55 /// The hasVEXPrefix field from the record 56 bool HasVEXPrefix; 57 /// The hasVEX_4VPrefix field from the record 58 bool HasVEX_4VPrefix; 59 /// The hasVEX_WPrefix field from the record 60 bool HasVEX_WPrefix; 61 /// Inferred from the operands; indicates whether the L bit in the VEX prefix is set 62 bool HasVEX_LPrefix; 63 /// The hasLockPrefix field from the record 64 bool HasLockPrefix; 65 /// The isCodeGenOnly filed from the record 66 bool IsCodeGenOnly; 67 // Whether the instruction has the predicate "In64BitMode" 68 bool Is64Bit; 69 // Whether the instruction has the predicate "In32BitMode" 70 bool Is32Bit; 71 72 /// The instruction name as listed in the tables 73 std::string Name; 74 /// The AT&T AsmString for the instruction 75 std::string AsmString; 76 77 /// Indicates whether the instruction is SSE 78 bool IsSSE; 79 /// Indicates whether the instruction has FR operands - MOVs with FR operands 80 /// are typically ignored 81 bool HasFROperands; 82 /// Indicates whether the instruction should be emitted into the decode 83 /// tables; regardless, it will be emitted into the instruction info table 84 bool ShouldBeEmitted; 85 86 /// The operands of the instruction, as listed in the CodeGenInstruction. 87 /// They are not one-to-one with operands listed in the MCInst; for example, 88 /// memory operands expand to 5 operands in the MCInst 89 const std::vector<CGIOperandList::OperandInfo>* Operands; 90 91 /// The description of the instruction that is emitted into the instruction 92 /// info table 93 InstructionSpecifier* Spec; 94 95 /// insnContext - Returns the primary context in which the instruction is 96 /// valid. 97 /// 98 /// @return - The context in which the instruction is valid. 99 InstructionContext insnContext() const; 100 101 enum filter_ret { 102 FILTER_STRONG, // instruction has no place in the instruction tables 103 FILTER_WEAK, // instruction may conflict, and should be eliminated if 104 // it does 105 FILTER_NORMAL // instruction should have high priority and generate an 106 // error if it conflcits with any other FILTER_NORMAL 107 // instruction 108 }; 109 110 /// filter - Determines whether the instruction should be decodable. Some 111 /// instructions are pure intrinsics and use unencodable operands; many 112 /// synthetic instructions are duplicates of other instructions; other 113 /// instructions only differ in the logical way in which they are used, and 114 /// have the same decoding. Because these would cause decode conflicts, 115 /// they must be filtered out. 116 /// 117 /// @return - The degree of filtering to be applied (see filter_ret). 118 filter_ret filter() const; 119 120 /// hasFROperands - Returns true if any operand is a FR operand. 121 bool hasFROperands() const; 122 123 /// has256BitOperands - Returns true if any operand is a 256-bit SSE operand. 124 bool has256BitOperands() const; 125 126 /// typeFromString - Translates an operand type from the string provided in 127 /// the LLVM tables to an OperandType for use in the operand specifier. 128 /// 129 /// @param s - The string, as extracted by calling Rec->getName() 130 /// on a CodeGenInstruction::OperandInfo. 131 /// @param isSSE - Indicates whether the instruction is an SSE 132 /// instruction. For SSE instructions, immediates are 133 /// fixed-size rather than being affected by the 134 /// mandatory OpSize prefix. 135 /// @param hasREX_WPrefix - Indicates whether the instruction has a REX.W 136 /// prefix. If it does, 32-bit register operands stay 137 /// 32-bit regardless of the operand size. 138 /// @param hasOpSizePrefix- Indicates whether the instruction has an OpSize 139 /// prefix. If it does not, then 16-bit register 140 /// operands stay 16-bit. 141 /// @return - The operand's type. 142 static OperandType typeFromString(const std::string& s, 143 bool isSSE, 144 bool hasREX_WPrefix, 145 bool hasOpSizePrefix); 146 147 /// immediateEncodingFromString - Translates an immediate encoding from the 148 /// string provided in the LLVM tables to an OperandEncoding for use in 149 /// the operand specifier. 150 /// 151 /// @param s - See typeFromString(). 152 /// @param hasOpSizePrefix - Indicates whether the instruction has an OpSize 153 /// prefix. If it does not, then 16-bit immediate 154 /// operands stay 16-bit. 155 /// @return - The operand's encoding. 156 static OperandEncoding immediateEncodingFromString(const std::string &s, 157 bool hasOpSizePrefix); 158 159 /// rmRegisterEncodingFromString - Like immediateEncodingFromString, but 160 /// handles operands that are in the REG field of the ModR/M byte. 161 static OperandEncoding rmRegisterEncodingFromString(const std::string &s, 162 bool hasOpSizePrefix); 163 164 /// rmRegisterEncodingFromString - Like immediateEncodingFromString, but 165 /// handles operands that are in the REG field of the ModR/M byte. 166 static OperandEncoding roRegisterEncodingFromString(const std::string &s, 167 bool hasOpSizePrefix); 168 static OperandEncoding memoryEncodingFromString(const std::string &s, 169 bool hasOpSizePrefix); 170 static OperandEncoding relocationEncodingFromString(const std::string &s, 171 bool hasOpSizePrefix); 172 static OperandEncoding opcodeModifierEncodingFromString(const std::string &s, 173 bool hasOpSizePrefix); 174 static OperandEncoding vvvvRegisterEncodingFromString(const std::string &s, 175 bool HasOpSizePrefix); 176 177 /// handleOperand - Converts a single operand from the LLVM table format to 178 /// the emitted table format, handling any duplicate operands it encounters 179 /// and then one non-duplicate. 180 /// 181 /// @param optional - Determines whether to assert that the 182 /// operand exists. 183 /// @param operandIndex - The index into the generated operand table. 184 /// Incremented by this function one or more 185 /// times to reflect possible duplicate 186 /// operands). 187 /// @param physicalOperandIndex - The index of the current operand into the 188 /// set of non-duplicate ('physical') operands. 189 /// Incremented by this function once. 190 /// @param numPhysicalOperands - The number of non-duplicate operands in the 191 /// instructions. 192 /// @param operandMapping - The operand mapping, which has an entry for 193 /// each operand that indicates whether it is a 194 /// duplicate, and of what. 195 void handleOperand(bool optional, 196 unsigned &operandIndex, 197 unsigned &physicalOperandIndex, 198 unsigned &numPhysicalOperands, 199 unsigned *operandMapping, 200 OperandEncoding (*encodingFromString) 201 (const std::string&, 202 bool hasOpSizePrefix)); 203 204 /// shouldBeEmitted - Returns the shouldBeEmitted field. Although filter() 205 /// filters out many instructions, at various points in decoding we 206 /// determine that the instruction should not actually be decodable. In 207 /// particular, MMX MOV instructions aren't emitted, but they're only 208 /// identified during operand parsing. 209 /// 210 /// @return - true if at this point we believe the instruction should be 211 /// emitted; false if not. This will return false if filter() returns false 212 /// once emitInstructionSpecifier() has been called. 213 bool shouldBeEmitted() const { 214 return ShouldBeEmitted; 215 } 216 217 /// emitInstructionSpecifier - Loads the instruction specifier for the current 218 /// instruction into a DisassemblerTables. 219 /// 220 /// @arg tables - The DisassemblerTables to populate with the specifier for 221 /// the current instruction. 222 void emitInstructionSpecifier(DisassemblerTables &tables); 223 224 /// emitDecodePath - Populates the proper fields in the decode tables 225 /// corresponding to the decode paths for this instruction. 226 /// 227 /// @arg tables - The DisassemblerTables to populate with the decode 228 /// decode information for the current instruction. 229 void emitDecodePath(DisassemblerTables &tables) const; 230 231 /// Constructor - Initializes a RecognizableInstr with the appropriate fields 232 /// from a CodeGenInstruction. 233 /// 234 /// @arg tables - The DisassemblerTables that the specifier will be added to. 235 /// @arg insn - The CodeGenInstruction to extract information from. 236 /// @arg uid - The unique ID of the current instruction. 237 RecognizableInstr(DisassemblerTables &tables, 238 const CodeGenInstruction &insn, 239 InstrUID uid); 240public: 241 /// processInstr - Accepts a CodeGenInstruction and loads decode information 242 /// for it into a DisassemblerTables if appropriate. 243 /// 244 /// @arg tables - The DiassemblerTables to be populated with decode 245 /// information. 246 /// @arg insn - The CodeGenInstruction to be used as a source for this 247 /// information. 248 /// @uid - The unique ID of the instruction. 249 static void processInstr(DisassemblerTables &tables, 250 const CodeGenInstruction &insn, 251 InstrUID uid); 252}; 253 254} // namespace X86Disassembler 255 256} // namespace llvm 257 258#endif 259