When removing a function from the function set and adding it to deferred, we
[llvm.git] / utils / TableGen / X86RecognizableInstr.h
blobc043b909b42fc5c6a6741f16c8c07d2a37c6900a
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.
15 //===----------------------------------------------------------------------===//
17 #ifndef X86RECOGNIZABLEINSTR_H
18 #define X86RECOGNIZABLEINSTR_H
20 #include "X86DisassemblerTables.h"
22 #include "CodeGenTarget.h"
23 #include "Record.h"
25 #include "llvm/Support/DataTypes.h"
26 #include "llvm/ADT/SmallVector.h"
28 namespace llvm {
30 namespace X86Disassembler {
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.
36 class RecognizableInstr {
37 private:
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 hasVEX_4VPrefix field from the record
56 bool HasVEX_4VPrefix;
57 /// The hasLockPrefix field from the record
58 bool HasLockPrefix;
59 /// The isCodeGenOnly filed from the record
60 bool IsCodeGenOnly;
62 /// The instruction name as listed in the tables
63 std::string Name;
64 /// The AT&T AsmString for the instruction
65 std::string AsmString;
67 /// Indicates whether the instruction is SSE
68 bool IsSSE;
69 /// Indicates whether the instruction has FR operands - MOVs with FR operands
70 /// are typically ignored
71 bool HasFROperands;
72 /// Indicates whether the instruction should be emitted into the decode
73 /// tables; regardless, it will be emitted into the instruction info table
74 bool ShouldBeEmitted;
76 /// The operands of the instruction, as listed in the CodeGenInstruction.
77 /// They are not one-to-one with operands listed in the MCInst; for example,
78 /// memory operands expand to 5 operands in the MCInst
79 const std::vector<CGIOperandList::OperandInfo>* Operands;
81 /// The description of the instruction that is emitted into the instruction
82 /// info table
83 InstructionSpecifier* Spec;
85 /// insnContext - Returns the primary context in which the instruction is
86 /// valid.
87 ///
88 /// @return - The context in which the instruction is valid.
89 InstructionContext insnContext() const;
91 enum filter_ret {
92 FILTER_STRONG, // instruction has no place in the instruction tables
93 FILTER_WEAK, // instruction may conflict, and should be eliminated if
94 // it does
95 FILTER_NORMAL // instruction should have high priority and generate an
96 // error if it conflcits with any other FILTER_NORMAL
97 // instruction
100 /// filter - Determines whether the instruction should be decodable. Some
101 /// instructions are pure intrinsics and use unencodable operands; many
102 /// synthetic instructions are duplicates of other instructions; other
103 /// instructions only differ in the logical way in which they are used, and
104 /// have the same decoding. Because these would cause decode conflicts,
105 /// they must be filtered out.
107 /// @return - The degree of filtering to be applied (see filter_ret).
108 filter_ret filter() const;
110 /// typeFromString - Translates an operand type from the string provided in
111 /// the LLVM tables to an OperandType for use in the operand specifier.
113 /// @param s - The string, as extracted by calling Rec->getName()
114 /// on a CodeGenInstruction::OperandInfo.
115 /// @param isSSE - Indicates whether the instruction is an SSE
116 /// instruction. For SSE instructions, immediates are
117 /// fixed-size rather than being affected by the
118 /// mandatory OpSize prefix.
119 /// @param hasREX_WPrefix - Indicates whether the instruction has a REX.W
120 /// prefix. If it does, 32-bit register operands stay
121 /// 32-bit regardless of the operand size.
122 /// @param hasOpSizePrefix- Indicates whether the instruction has an OpSize
123 /// prefix. If it does not, then 16-bit register
124 /// operands stay 16-bit.
125 /// @return - The operand's type.
126 static OperandType typeFromString(const std::string& s,
127 bool isSSE,
128 bool hasREX_WPrefix,
129 bool hasOpSizePrefix);
131 /// immediateEncodingFromString - Translates an immediate encoding from the
132 /// string provided in the LLVM tables to an OperandEncoding for use in
133 /// the operand specifier.
135 /// @param s - See typeFromString().
136 /// @param hasOpSizePrefix - Indicates whether the instruction has an OpSize
137 /// prefix. If it does not, then 16-bit immediate
138 /// operands stay 16-bit.
139 /// @return - The operand's encoding.
140 static OperandEncoding immediateEncodingFromString(const std::string &s,
141 bool hasOpSizePrefix);
143 /// rmRegisterEncodingFromString - Like immediateEncodingFromString, but
144 /// handles operands that are in the REG field of the ModR/M byte.
145 static OperandEncoding rmRegisterEncodingFromString(const std::string &s,
146 bool hasOpSizePrefix);
148 /// rmRegisterEncodingFromString - Like immediateEncodingFromString, but
149 /// handles operands that are in the REG field of the ModR/M byte.
150 static OperandEncoding roRegisterEncodingFromString(const std::string &s,
151 bool hasOpSizePrefix);
152 static OperandEncoding memoryEncodingFromString(const std::string &s,
153 bool hasOpSizePrefix);
154 static OperandEncoding relocationEncodingFromString(const std::string &s,
155 bool hasOpSizePrefix);
156 static OperandEncoding opcodeModifierEncodingFromString(const std::string &s,
157 bool hasOpSizePrefix);
159 /// handleOperand - Converts a single operand from the LLVM table format to
160 /// the emitted table format, handling any duplicate operands it encounters
161 /// and then one non-duplicate.
163 /// @param optional - Determines whether to assert that the
164 /// operand exists.
165 /// @param operandIndex - The index into the generated operand table.
166 /// Incremented by this function one or more
167 /// times to reflect possible duplicate
168 /// operands).
169 /// @param physicalOperandIndex - The index of the current operand into the
170 /// set of non-duplicate ('physical') operands.
171 /// Incremented by this function once.
172 /// @param numPhysicalOperands - The number of non-duplicate operands in the
173 /// instructions.
174 /// @param operandMapping - The operand mapping, which has an entry for
175 /// each operand that indicates whether it is a
176 /// duplicate, and of what.
177 void handleOperand(bool optional,
178 unsigned &operandIndex,
179 unsigned &physicalOperandIndex,
180 unsigned &numPhysicalOperands,
181 unsigned *operandMapping,
182 OperandEncoding (*encodingFromString)
183 (const std::string&,
184 bool hasOpSizePrefix));
186 /// shouldBeEmitted - Returns the shouldBeEmitted field. Although filter()
187 /// filters out many instructions, at various points in decoding we
188 /// determine that the instruction should not actually be decodable. In
189 /// particular, MMX MOV instructions aren't emitted, but they're only
190 /// identified during operand parsing.
192 /// @return - true if at this point we believe the instruction should be
193 /// emitted; false if not. This will return false if filter() returns false
194 /// once emitInstructionSpecifier() has been called.
195 bool shouldBeEmitted() const {
196 return ShouldBeEmitted;
199 /// emitInstructionSpecifier - Loads the instruction specifier for the current
200 /// instruction into a DisassemblerTables.
202 /// @arg tables - The DisassemblerTables to populate with the specifier for
203 /// the current instruction.
204 void emitInstructionSpecifier(DisassemblerTables &tables);
206 /// emitDecodePath - Populates the proper fields in the decode tables
207 /// corresponding to the decode paths for this instruction.
209 /// @arg tables - The DisassemblerTables to populate with the decode
210 /// decode information for the current instruction.
211 void emitDecodePath(DisassemblerTables &tables) const;
213 /// Constructor - Initializes a RecognizableInstr with the appropriate fields
214 /// from a CodeGenInstruction.
216 /// @arg tables - The DisassemblerTables that the specifier will be added to.
217 /// @arg insn - The CodeGenInstruction to extract information from.
218 /// @arg uid - The unique ID of the current instruction.
219 RecognizableInstr(DisassemblerTables &tables,
220 const CodeGenInstruction &insn,
221 InstrUID uid);
222 public:
223 /// processInstr - Accepts a CodeGenInstruction and loads decode information
224 /// for it into a DisassemblerTables if appropriate.
226 /// @arg tables - The DiassemblerTables to be populated with decode
227 /// information.
228 /// @arg insn - The CodeGenInstruction to be used as a source for this
229 /// information.
230 /// @uid - The unique ID of the instruction.
231 static void processInstr(DisassemblerTables &tables,
232 const CodeGenInstruction &insn,
233 InstrUID uid);
236 } // namespace X86Disassembler
238 } // namespace llvm
240 #endif