1 //===- X86InstrInfo.td - Describe the X86 Instruction Set -------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by the LLVM research group and is distributed under
6 // the University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86 instruction set, defining the instructions, and
11 // properties of the instructions which are needed for code generation, machine
12 // code emission, and analysis.
14 //===----------------------------------------------------------------------===//
16 // *mem - Operand definitions for the funky X86 addressing mode operands.
18 class X86MemOperand<ValueType Ty, string printMethod> : Operand<Ty> {
19 let PrintMethod = printMethod;
20 let NumMIOperands = 4;
21 let MIOperandInfo = (ops R32, i8imm, R32, i32imm);
24 def i8mem : X86MemOperand<i32, "printi8mem">;
25 def i16mem : X86MemOperand<i32, "printi16mem">;
26 def i32mem : X86MemOperand<i32, "printi32mem">;
27 def i64mem : X86MemOperand<i32, "printi64mem">;
28 def f32mem : X86MemOperand<i32, "printf32mem">;
29 def f64mem : X86MemOperand<i32, "printf64mem">;
30 def f80mem : X86MemOperand<i32, "printf80mem">;
32 def SSECC : Operand<i8> {
33 let PrintMethod = "printSSECC";
36 // A couple of more descriptive operand definitions.
37 // 16-bits but only 8 bits are significant.
38 def i16i8imm : Operand<i16>;
39 // 32-bits but only 8 bits are significant.
40 def i32i8imm : Operand<i32>;
42 // PCRelative calls need special operand formatting.
43 let PrintMethod = "printCallOperand" in
44 def calltarget : Operand<i32>;
46 // Branch targets have OtherVT type.
47 def brtarget : Operand<OtherVT>;
49 // Define X86 specific addressing mode.
50 def addr : ComplexPattern<i32, 4, "SelectAddr", []>;
51 def leaaddr : ComplexPattern<i32, 4, "SelectLEAAddr", [add]>;
53 // Format specifies the encoding used by the instruction. This is part of the
54 // ad-hoc solution used to emit machine instruction encodings by our machine
56 class Format<bits<5> val> {
60 def Pseudo : Format<0>; def RawFrm : Format<1>;
61 def AddRegFrm : Format<2>; def MRMDestReg : Format<3>;
62 def MRMDestMem : Format<4>; def MRMSrcReg : Format<5>;
63 def MRMSrcMem : Format<6>;
64 def MRM0r : Format<16>; def MRM1r : Format<17>; def MRM2r : Format<18>;
65 def MRM3r : Format<19>; def MRM4r : Format<20>; def MRM5r : Format<21>;
66 def MRM6r : Format<22>; def MRM7r : Format<23>;
67 def MRM0m : Format<24>; def MRM1m : Format<25>; def MRM2m : Format<26>;
68 def MRM3m : Format<27>; def MRM4m : Format<28>; def MRM5m : Format<29>;
69 def MRM6m : Format<30>; def MRM7m : Format<31>;
71 // ImmType - This specifies the immediate type used by an instruction. This is
72 // part of the ad-hoc solution used to emit machine instruction encodings by our
73 // machine code emitter.
74 class ImmType<bits<2> val> {
77 def NoImm : ImmType<0>;
78 def Imm8 : ImmType<1>;
79 def Imm16 : ImmType<2>;
80 def Imm32 : ImmType<3>;
82 // FPFormat - This specifies what form this FP instruction has. This is used by
83 // the Floating-Point stackifier pass.
84 class FPFormat<bits<3> val> {
87 def NotFP : FPFormat<0>;
88 def ZeroArgFP : FPFormat<1>;
89 def OneArgFP : FPFormat<2>;
90 def OneArgFPRW : FPFormat<3>;
91 def TwoArgFP : FPFormat<4>;
92 def CompareFP : FPFormat<5>;
93 def CondMovFP : FPFormat<6>;
94 def SpecialFP : FPFormat<7>;
97 class X86Inst<bits<8> opcod, Format f, ImmType i, dag ops, string AsmStr>
99 let Namespace = "X86";
101 bits<8> Opcode = opcod;
103 bits<5> FormBits = Form.Value;
105 bits<2> ImmTypeBits = ImmT.Value;
107 dag OperandList = ops;
108 string AsmString = AsmStr;
111 // Attributes specific to X86 instructions...
113 bit hasOpSizePrefix = 0; // Does this inst have a 0x66 prefix?
115 bits<4> Prefix = 0; // Which prefix byte does this inst have?
116 FPFormat FPForm; // What flavor of FP instruction is this?
117 bits<3> FPFormBits = 0;
120 class Imp<list<Register> uses, list<Register> defs> {
121 list<Register> Uses = uses;
122 list<Register> Defs = defs;
126 // Prefix byte classes which are used to indicate to the ad-hoc machine code
127 // emitter that various prefix bytes are required.
128 class OpSize { bit hasOpSizePrefix = 1; }
129 class TB { bits<4> Prefix = 1; }
130 class REP { bits<4> Prefix = 2; }
131 class D8 { bits<4> Prefix = 3; }
132 class D9 { bits<4> Prefix = 4; }
133 class DA { bits<4> Prefix = 5; }
134 class DB { bits<4> Prefix = 6; }
135 class DC { bits<4> Prefix = 7; }
136 class DD { bits<4> Prefix = 8; }
137 class DE { bits<4> Prefix = 9; }
138 class DF { bits<4> Prefix = 10; }
139 class XD { bits<4> Prefix = 11; }
140 class XS { bits<4> Prefix = 12; }
143 //===----------------------------------------------------------------------===//
144 // Pattern fragments...
146 def i16SExt8 : PatLeaf<(i16 imm), [{
147 // i16SExt8 predicate - True if the 16-bit immediate fits in a 8-bit
148 // sign extended field.
149 return (int)N->getValue() == (signed char)N->getValue();
152 def i32SExt8 : PatLeaf<(i32 imm), [{
153 // i32SExt8 predicate - True if the 32-bit immediate fits in a 8-bit
154 // sign extended field.
155 return (int)N->getValue() == (signed char)N->getValue();
158 def i16ZExt8 : PatLeaf<(i16 imm), [{
159 // i16ZExt8 predicate - True if the 16-bit immediate fits in a 8-bit zero
161 return (unsigned)N->getValue() == (unsigned char)N->getValue();
164 //===----------------------------------------------------------------------===//
165 // Instruction templates...
167 class I<bits<8> o, Format f, dag ops, string asm, list<dag> pattern>
168 : X86Inst<o, f, NoImm, ops, asm> {
169 let Pattern = pattern;
171 class Ii8 <bits<8> o, Format f, dag ops, string asm, list<dag> pattern>
172 : X86Inst<o, f, Imm8 , ops, asm> {
173 let Pattern = pattern;
175 class Ii16<bits<8> o, Format f, dag ops, string asm, list<dag> pattern>
176 : X86Inst<o, f, Imm16, ops, asm> {
177 let Pattern = pattern;
179 class Ii32<bits<8> o, Format f, dag ops, string asm, list<dag> pattern>
180 : X86Inst<o, f, Imm32, ops, asm> {
181 let Pattern = pattern;
184 //===----------------------------------------------------------------------===//
185 // Instruction list...
188 def PHI : I<0, Pseudo, (ops variable_ops), "PHINODE", []>; // PHI node.
189 def NOOP : I<0x90, RawFrm, (ops), "nop", []>; // nop
191 def ADJCALLSTACKDOWN : I<0, Pseudo, (ops i32imm:$amt), "#ADJCALLSTACKDOWN", []>;
192 def ADJCALLSTACKUP : I<0, Pseudo, (ops i32imm:$amt1, i32imm:$amt2),
193 "#ADJCALLSTACKUP", []>;
194 def IMPLICIT_USE : I<0, Pseudo, (ops variable_ops), "#IMPLICIT_USE", []>;
195 def IMPLICIT_DEF : I<0, Pseudo, (ops variable_ops), "#IMPLICIT_DEF", []>;
196 let isTerminator = 1 in
197 let Defs = [FP0, FP1, FP2, FP3, FP4, FP5, FP6] in
198 def FP_REG_KILL : I<0, Pseudo, (ops), "#FP_REG_KILL", []>;
200 //===----------------------------------------------------------------------===//
201 // Control Flow Instructions...
204 // Return instructions.
205 let isTerminator = 1, isReturn = 1, isBarrier = 1 in
206 def RET : I<0xC3, RawFrm, (ops), "ret", [(ret)]>;
207 let isTerminator = 1, isReturn = 1, isBarrier = 1 in
208 def RETI : Ii16<0xC2, RawFrm, (ops i16imm:$amt), "ret $amt", []>;
210 // All branches are RawFrm, Void, Branch, and Terminators
211 let isBranch = 1, isTerminator = 1 in
212 class IBr<bits<8> opcode, dag ops, string asm, list<dag> pattern> :
213 I<opcode, RawFrm, ops, asm, pattern>;
216 def JMP : IBr<0xE9, (ops brtarget:$dst), "jmp $dst", [(br bb:$dst)]>;
217 def JB : IBr<0x82, (ops brtarget:$dst), "jb $dst",
219 def JAE : IBr<0x83, (ops brtarget:$dst), "jae $dst", []>, TB;
220 def JE : IBr<0x84, (ops brtarget:$dst), "je $dst", []>, TB;
221 def JNE : IBr<0x85, (ops brtarget:$dst), "jne $dst", []>, TB;
222 def JBE : IBr<0x86, (ops brtarget:$dst), "jbe $dst", []>, TB;
223 def JA : IBr<0x87, (ops brtarget:$dst), "ja $dst", []>, TB;
224 def JS : IBr<0x88, (ops brtarget:$dst), "js $dst", []>, TB;
225 def JNS : IBr<0x89, (ops brtarget:$dst), "jns $dst", []>, TB;
226 def JP : IBr<0x8A, (ops brtarget:$dst), "jp $dst", []>, TB;
227 def JNP : IBr<0x8B, (ops brtarget:$dst), "jnp $dst", []>, TB;
228 def JL : IBr<0x8C, (ops brtarget:$dst), "jl $dst", []>, TB;
229 def JGE : IBr<0x8D, (ops brtarget:$dst), "jge $dst", []>, TB;
230 def JLE : IBr<0x8E, (ops brtarget:$dst), "jle $dst", []>, TB;
231 def JG : IBr<0x8F, (ops brtarget:$dst), "jg $dst", []>, TB;
233 //===----------------------------------------------------------------------===//
234 // Call Instructions...
237 // All calls clobber the non-callee saved registers...
238 let Defs = [EAX, ECX, EDX, FP0, FP1, FP2, FP3, FP4, FP5, FP6, ST0,
239 XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, XMM6, XMM7] in {
240 def CALLpcrel32 : I<0xE8, RawFrm, (ops calltarget:$dst), "call $dst", []>;
241 def CALL32r : I<0xFF, MRM2r, (ops R32:$dst), "call {*}$dst", []>;
242 def CALL32m : I<0xFF, MRM2m, (ops i32mem:$dst), "call {*}$dst", []>;
246 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1 in
247 def TAILJMPd : IBr<0xE9, (ops calltarget:$dst), "jmp $dst # TAIL CALL", []>;
248 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1 in
249 def TAILJMPr : I<0xFF, MRM4r, (ops R32:$dst), "jmp {*}$dst # TAIL CALL", []>;
250 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1 in
251 def TAILJMPm : I<0xFF, MRM4m, (ops i32mem:$dst),
252 "jmp {*}$dst # TAIL CALL", []>;
254 // ADJSTACKPTRri - This is a standard ADD32ri instruction, identical in every
255 // way, except that it is marked as being a terminator. This causes the epilog
256 // inserter to insert reloads of callee saved registers BEFORE this. We need
257 // this until we have a more accurate way of tracking where the stack pointer is
258 // within a function.
259 let isTerminator = 1, isTwoAddress = 1 in
260 def ADJSTACKPTRri : Ii32<0x81, MRM0r, (ops R32:$dst, R32:$src1, i32imm:$src2),
261 "add{l} {$src2, $dst|$dst, $src2}", []>;
263 //===----------------------------------------------------------------------===//
264 // Miscellaneous Instructions...
266 def LEAVE : I<0xC9, RawFrm,
267 (ops), "leave", []>, Imp<[EBP,ESP],[EBP,ESP]>;
268 def POP32r : I<0x58, AddRegFrm,
269 (ops R32:$reg), "pop{l} $reg", []>, Imp<[ESP],[ESP]>;
271 let isTwoAddress = 1 in // R32 = bswap R32
272 def BSWAP32r : I<0xC8, AddRegFrm,
273 (ops R32:$dst, R32:$src), "bswap{l} $dst", []>, TB;
275 def XCHG8rr : I<0x86, MRMDestReg, // xchg R8, R8
276 (ops R8:$src1, R8:$src2),
277 "xchg{b} {$src2|$src1}, {$src1|$src2}", []>;
278 def XCHG16rr : I<0x87, MRMDestReg, // xchg R16, R16
279 (ops R16:$src1, R16:$src2),
280 "xchg{w} {$src2|$src1}, {$src1|$src2}", []>, OpSize;
281 def XCHG32rr : I<0x87, MRMDestReg, // xchg R32, R32
282 (ops R32:$src1, R32:$src2),
283 "xchg{l} {$src2|$src1}, {$src1|$src2}", []>;
285 def XCHG8mr : I<0x86, MRMDestMem,
286 (ops i8mem:$src1, R8:$src2),
287 "xchg{b} {$src2|$src1}, {$src1|$src2}", []>;
288 def XCHG16mr : I<0x87, MRMDestMem,
289 (ops i16mem:$src1, R16:$src2),
290 "xchg{w} {$src2|$src1}, {$src1|$src2}", []>, OpSize;
291 def XCHG32mr : I<0x87, MRMDestMem,
292 (ops i32mem:$src1, R32:$src2),
293 "xchg{l} {$src2|$src1}, {$src1|$src2}", []>;
294 def XCHG8rm : I<0x86, MRMSrcMem,
295 (ops R8:$src1, i8mem:$src2),
296 "xchg{b} {$src2|$src1}, {$src1|$src2}", []>;
297 def XCHG16rm : I<0x87, MRMSrcMem,
298 (ops R16:$src1, i16mem:$src2),
299 "xchg{w} {$src2|$src1}, {$src1|$src2}", []>, OpSize;
300 def XCHG32rm : I<0x87, MRMSrcMem,
301 (ops R32:$src1, i32mem:$src2),
302 "xchg{l} {$src2|$src1}, {$src1|$src2}", []>;
304 def LEA16r : I<0x8D, MRMSrcMem,
305 (ops R16:$dst, i32mem:$src),
306 "lea{w} {$src|$dst}, {$dst|$src}", []>, OpSize;
307 def LEA32r : I<0x8D, MRMSrcMem,
308 (ops R32:$dst, i32mem:$src),
309 "lea{l} {$src|$dst}, {$dst|$src}",
310 [(set R32:$dst, leaaddr:$src)]>;
313 def REP_MOVSB : I<0xA4, RawFrm, (ops), "{rep;movsb|rep movsb}", []>,
314 Imp<[ECX,EDI,ESI], [ECX,EDI,ESI]>, REP;
315 def REP_MOVSW : I<0xA5, RawFrm, (ops), "{rep;movsw|rep movsw}", []>,
316 Imp<[ECX,EDI,ESI], [ECX,EDI,ESI]>, REP, OpSize;
317 def REP_MOVSD : I<0xA5, RawFrm, (ops), "{rep;movsd|rep movsd}", []>,
318 Imp<[ECX,EDI,ESI], [ECX,EDI,ESI]>, REP;
320 def REP_STOSB : I<0xAA, RawFrm, (ops), "{rep;stosb|rep stosb}", []>,
321 Imp<[AL,ECX,EDI], [ECX,EDI]>, REP;
322 def REP_STOSW : I<0xAB, RawFrm, (ops), "{rep;stosw|rep stosw}", []>,
323 Imp<[AX,ECX,EDI], [ECX,EDI]>, REP, OpSize;
324 def REP_STOSD : I<0xAB, RawFrm, (ops), "{rep;stosl|rep stosd}", []>,
325 Imp<[EAX,ECX,EDI], [ECX,EDI]>, REP;
328 //===----------------------------------------------------------------------===//
329 // Input/Output Instructions...
331 def IN8rr : I<0xEC, RawFrm, (ops),
332 "in{b} {%dx, %al|%AL, %DX}", []>, Imp<[DX], [AL]>;
333 def IN16rr : I<0xED, RawFrm, (ops),
334 "in{w} {%dx, %ax|%AX, %DX}", []>, Imp<[DX], [AX]>, OpSize;
335 def IN32rr : I<0xED, RawFrm, (ops),
336 "in{l} {%dx, %eax|%EAX, %DX}", []>, Imp<[DX],[EAX]>;
338 def IN8ri : Ii8<0xE4, RawFrm, (ops i8imm:$port),
339 "in{b} {$port, %al|%AL, $port}", []>, Imp<[], [AL]>;
340 def IN16ri : Ii8<0xE5, RawFrm, (ops i8imm:$port),
341 "in{w} {$port, %ax|%AX, $port}", []>, Imp<[], [AX]>, OpSize;
342 def IN32ri : Ii8<0xE5, RawFrm, (ops i8imm:$port),
343 "in{l} {$port, %eax|%EAX, $port}", []>, Imp<[],[EAX]>;
345 def OUT8rr : I<0xEE, RawFrm, (ops),
346 "out{b} {%al, %dx|%DX, %AL}",
347 [(writeport AL, DX)]>, Imp<[DX, AL], []>;
348 def OUT16rr : I<0xEF, RawFrm, (ops),
349 "out{w} {%ax, %dx|%DX, %AX}",
350 [(writeport AX, DX)]>, Imp<[DX, AX], []>, OpSize;
351 def OUT32rr : I<0xEF, RawFrm, (ops),
352 "out{l} {%eax, %dx|%DX, %EAX}",
353 [(writeport EAX, DX)]>, Imp<[DX, EAX], []>;
355 def OUT8ir : Ii8<0xE6, RawFrm, (ops i16i8imm:$port),
356 "out{b} {%al, $port|$port, %AL}",
357 [(writeport AL, (i16 i16ZExt8:$port))]>,
359 def OUT16ir : Ii8<0xE7, RawFrm, (ops i16i8imm:$port),
360 "out{w} {%ax, $port|$port, %AX}",
361 [(writeport AX, (i16 i16ZExt8:$port))]>,
362 Imp<[AX], []>, OpSize;
363 def OUT32ir : Ii8<0xE7, RawFrm, (ops i16i8imm:$port),
364 "out{l} {%eax, $port|$port, %EAX}",
365 [(writeport EAX, (i16 i16ZExt8:$port))]>,
368 //===----------------------------------------------------------------------===//
369 // Move Instructions...
371 def MOV8rr : I<0x88, MRMDestReg, (ops R8 :$dst, R8 :$src),
372 "mov{b} {$src, $dst|$dst, $src}", []>;
373 def MOV16rr : I<0x89, MRMDestReg, (ops R16:$dst, R16:$src),
374 "mov{w} {$src, $dst|$dst, $src}", []>, OpSize;
375 def MOV32rr : I<0x89, MRMDestReg, (ops R32:$dst, R32:$src),
376 "mov{l} {$src, $dst|$dst, $src}", []>;
377 def MOV8ri : Ii8 <0xB0, AddRegFrm, (ops R8 :$dst, i8imm :$src),
378 "mov{b} {$src, $dst|$dst, $src}",
379 [(set R8:$dst, imm:$src)]>;
380 def MOV16ri : Ii16<0xB8, AddRegFrm, (ops R16:$dst, i16imm:$src),
381 "mov{w} {$src, $dst|$dst, $src}",
382 [(set R16:$dst, imm:$src)]>, OpSize;
383 def MOV32ri : Ii32<0xB8, AddRegFrm, (ops R32:$dst, i32imm:$src),
384 "mov{l} {$src, $dst|$dst, $src}",
385 [(set R32:$dst, imm:$src)]>;
386 def MOV8mi : Ii8 <0xC6, MRM0m, (ops i8mem :$dst, i8imm :$src),
387 "mov{b} {$src, $dst|$dst, $src}",
388 [(store (i8 imm:$src), addr:$dst)]>;
389 def MOV16mi : Ii16<0xC7, MRM0m, (ops i16mem:$dst, i16imm:$src),
390 "mov{w} {$src, $dst|$dst, $src}",
391 [(store (i16 imm:$src), addr:$dst)]>, OpSize;
392 def MOV32mi : Ii32<0xC7, MRM0m, (ops i32mem:$dst, i32imm:$src),
393 "mov{l} {$src, $dst|$dst, $src}",
394 [(store (i32 imm:$src), addr:$dst)]>;
396 def MOV8rm : I<0x8A, MRMSrcMem, (ops R8 :$dst, i8mem :$src),
397 "mov{b} {$src, $dst|$dst, $src}",
398 [(set R8:$dst, (load addr:$src))]>;
399 def MOV16rm : I<0x8B, MRMSrcMem, (ops R16:$dst, i16mem:$src),
400 "mov{w} {$src, $dst|$dst, $src}",
401 [(set R16:$dst, (load addr:$src))]>, OpSize;
402 def MOV32rm : I<0x8B, MRMSrcMem, (ops R32:$dst, i32mem:$src),
403 "mov{l} {$src, $dst|$dst, $src}",
404 [(set R32:$dst, (load addr:$src))]>;
406 def MOV8mr : I<0x88, MRMDestMem, (ops i8mem :$dst, R8 :$src),
407 "mov{b} {$src, $dst|$dst, $src}",
408 [(store R8:$src, addr:$dst)]>;
409 def MOV16mr : I<0x89, MRMDestMem, (ops i16mem:$dst, R16:$src),
410 "mov{w} {$src, $dst|$dst, $src}",
411 [(store R16:$src, addr:$dst)]>, OpSize;
412 def MOV32mr : I<0x89, MRMDestMem, (ops i32mem:$dst, R32:$src),
413 "mov{l} {$src, $dst|$dst, $src}",
414 [(store R32:$src, addr:$dst)]>;
416 //===----------------------------------------------------------------------===//
417 // Fixed-Register Multiplication and Division Instructions...
420 // Extra precision multiplication
421 def MUL8r : I<0xF6, MRM4r, (ops R8:$src), "mul{b} $src", []>,
422 Imp<[AL],[AX]>; // AL,AH = AL*R8
423 def MUL16r : I<0xF7, MRM4r, (ops R16:$src), "mul{w} $src", []>,
424 Imp<[AX],[AX,DX]>, OpSize; // AX,DX = AX*R16
425 def MUL32r : I<0xF7, MRM4r, (ops R32:$src), "mul{l} $src", []>,
426 Imp<[EAX],[EAX,EDX]>; // EAX,EDX = EAX*R32
427 def MUL8m : I<0xF6, MRM4m, (ops i8mem :$src),
428 "mul{b} $src", []>, Imp<[AL],[AX]>; // AL,AH = AL*[mem8]
429 def MUL16m : I<0xF7, MRM4m, (ops i16mem:$src),
430 "mul{w} $src", []>, Imp<[AX],[AX,DX]>,
431 OpSize; // AX,DX = AX*[mem16]
432 def MUL32m : I<0xF7, MRM4m, (ops i32mem:$src),
433 "mul{l} $src", []>, Imp<[EAX],[EAX,EDX]>;// EAX,EDX = EAX*[mem32]
435 def IMUL8r : I<0xF6, MRM5r, (ops R8:$src), "imul{b} $src", []>,
436 Imp<[AL],[AX]>; // AL,AH = AL*R8
437 def IMUL16r : I<0xF7, MRM5r, (ops R16:$src), "imul{w} $src", []>,
438 Imp<[AX],[AX,DX]>, OpSize; // AX,DX = AX*R16
439 def IMUL32r : I<0xF7, MRM5r, (ops R32:$src), "imul{l} $src", []>,
440 Imp<[EAX],[EAX,EDX]>; // EAX,EDX = EAX*R32
441 def IMUL8m : I<0xF6, MRM5m, (ops i8mem :$src),
442 "imul{b} $src", []>, Imp<[AL],[AX]>; // AL,AH = AL*[mem8]
443 def IMUL16m : I<0xF7, MRM5m, (ops i16mem:$src),
444 "imul{w} $src", []>, Imp<[AX],[AX,DX]>,
445 OpSize; // AX,DX = AX*[mem16]
446 def IMUL32m : I<0xF7, MRM5m, (ops i32mem:$src),
448 Imp<[EAX],[EAX,EDX]>; // EAX,EDX = EAX*[mem32]
450 // unsigned division/remainder
451 def DIV8r : I<0xF6, MRM6r, (ops R8:$src), // AX/r8 = AL,AH
452 "div{b} $src", []>, Imp<[AX],[AX]>;
453 def DIV16r : I<0xF7, MRM6r, (ops R16:$src), // DX:AX/r16 = AX,DX
454 "div{w} $src", []>, Imp<[AX,DX],[AX,DX]>, OpSize;
455 def DIV32r : I<0xF7, MRM6r, (ops R32:$src), // EDX:EAX/r32 = EAX,EDX
456 "div{l} $src", []>, Imp<[EAX,EDX],[EAX,EDX]>;
457 def DIV8m : I<0xF6, MRM6m, (ops i8mem:$src), // AX/[mem8] = AL,AH
458 "div{b} $src", []>, Imp<[AX],[AX]>;
459 def DIV16m : I<0xF7, MRM6m, (ops i16mem:$src), // DX:AX/[mem16] = AX,DX
460 "div{w} $src", []>, Imp<[AX,DX],[AX,DX]>, OpSize;
461 def DIV32m : I<0xF7, MRM6m, (ops i32mem:$src), // EDX:EAX/[mem32] = EAX,EDX
462 "div{l} $src", []>, Imp<[EAX,EDX],[EAX,EDX]>;
464 // Signed division/remainder.
465 def IDIV8r : I<0xF6, MRM7r, (ops R8:$src), // AX/r8 = AL,AH
466 "idiv{b} $src", []>, Imp<[AX],[AX]>;
467 def IDIV16r: I<0xF7, MRM7r, (ops R16:$src), // DX:AX/r16 = AX,DX
468 "idiv{w} $src", []>, Imp<[AX,DX],[AX,DX]>, OpSize;
469 def IDIV32r: I<0xF7, MRM7r, (ops R32:$src), // EDX:EAX/r32 = EAX,EDX
470 "idiv{l} $src", []>, Imp<[EAX,EDX],[EAX,EDX]>;
471 def IDIV8m : I<0xF6, MRM7m, (ops i8mem:$src), // AX/[mem8] = AL,AH
472 "idiv{b} $src", []>, Imp<[AX],[AX]>;
473 def IDIV16m: I<0xF7, MRM7m, (ops i16mem:$src), // DX:AX/[mem16] = AX,DX
474 "idiv{w} $src", []>, Imp<[AX,DX],[AX,DX]>, OpSize;
475 def IDIV32m: I<0xF7, MRM7m, (ops i32mem:$src), // EDX:EAX/[mem32] = EAX,EDX
476 "idiv{l} $src", []>, Imp<[EAX,EDX],[EAX,EDX]>;
478 // Sign-extenders for division.
479 def CBW : I<0x98, RawFrm, (ops),
480 "{cbtw|cbw}", []>, Imp<[AL],[AH]>; // AX = signext(AL)
481 def CWD : I<0x99, RawFrm, (ops),
482 "{cwtd|cwd}", []>, Imp<[AX],[DX]>; // DX:AX = signext(AX)
483 def CDQ : I<0x99, RawFrm, (ops),
484 "{cltd|cdq}", []>, Imp<[EAX],[EDX]>; // EDX:EAX = signext(EAX)
487 //===----------------------------------------------------------------------===//
488 // Two address Instructions...
490 let isTwoAddress = 1 in {
493 def CMOVB16rr : I<0x42, MRMSrcReg, // if <u, R16 = R16
494 (ops R16:$dst, R16:$src1, R16:$src2),
495 "cmovb {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
496 def CMOVB16rm : I<0x42, MRMSrcMem, // if <u, R16 = [mem16]
497 (ops R16:$dst, R16:$src1, i16mem:$src2),
498 "cmovb {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
499 def CMOVB32rr : I<0x42, MRMSrcReg, // if <u, R32 = R32
500 (ops R32:$dst, R32:$src1, R32:$src2),
501 "cmovb {$src2, $dst|$dst, $src2}", []>, TB;
502 def CMOVB32rm : I<0x42, MRMSrcMem, // if <u, R32 = [mem32]
503 (ops R32:$dst, R32:$src1, i32mem:$src2),
504 "cmovb {$src2, $dst|$dst, $src2}", []>, TB;
506 def CMOVAE16rr: I<0x43, MRMSrcReg, // if >=u, R16 = R16
507 (ops R16:$dst, R16:$src1, R16:$src2),
508 "cmovae {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
509 def CMOVAE16rm: I<0x43, MRMSrcMem, // if >=u, R16 = [mem16]
510 (ops R16:$dst, R16:$src1, i16mem:$src2),
511 "cmovae {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
512 def CMOVAE32rr: I<0x43, MRMSrcReg, // if >=u, R32 = R32
513 (ops R32:$dst, R32:$src1, R32:$src2),
514 "cmovae {$src2, $dst|$dst, $src2}", []>, TB;
515 def CMOVAE32rm: I<0x43, MRMSrcMem, // if >=u, R32 = [mem32]
516 (ops R32:$dst, R32:$src1, i32mem:$src2),
517 "cmovae {$src2, $dst|$dst, $src2}", []>, TB;
519 def CMOVE16rr : I<0x44, MRMSrcReg, // if ==, R16 = R16
520 (ops R16:$dst, R16:$src1, R16:$src2),
521 "cmove {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
522 def CMOVE16rm : I<0x44, MRMSrcMem, // if ==, R16 = [mem16]
523 (ops R16:$dst, R16:$src1, i16mem:$src2),
524 "cmove {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
525 def CMOVE32rr : I<0x44, MRMSrcReg, // if ==, R32 = R32
526 (ops R32:$dst, R32:$src1, R32:$src2),
527 "cmove {$src2, $dst|$dst, $src2}", []>, TB;
528 def CMOVE32rm : I<0x44, MRMSrcMem, // if ==, R32 = [mem32]
529 (ops R32:$dst, R32:$src1, i32mem:$src2),
530 "cmove {$src2, $dst|$dst, $src2}", []>, TB;
532 def CMOVNE16rr: I<0x45, MRMSrcReg, // if !=, R16 = R16
533 (ops R16:$dst, R16:$src1, R16:$src2),
534 "cmovne {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
535 def CMOVNE16rm: I<0x45, MRMSrcMem, // if !=, R16 = [mem16]
536 (ops R16:$dst, R16:$src1, i16mem:$src2),
537 "cmovne {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
538 def CMOVNE32rr: I<0x45, MRMSrcReg, // if !=, R32 = R32
539 (ops R32:$dst, R32:$src1, R32:$src2),
540 "cmovne {$src2, $dst|$dst, $src2}", []>, TB;
541 def CMOVNE32rm: I<0x45, MRMSrcMem, // if !=, R32 = [mem32]
542 (ops R32:$dst, R32:$src1, i32mem:$src2),
543 "cmovne {$src2, $dst|$dst, $src2}", []>, TB;
545 def CMOVBE16rr: I<0x46, MRMSrcReg, // if <=u, R16 = R16
546 (ops R16:$dst, R16:$src1, R16:$src2),
547 "cmovbe {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
548 def CMOVBE16rm: I<0x46, MRMSrcMem, // if <=u, R16 = [mem16]
549 (ops R16:$dst, R16:$src1, i16mem:$src2),
550 "cmovbe {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
551 def CMOVBE32rr: I<0x46, MRMSrcReg, // if <=u, R32 = R32
552 (ops R32:$dst, R32:$src1, R32:$src2),
553 "cmovbe {$src2, $dst|$dst, $src2}", []>, TB;
554 def CMOVBE32rm: I<0x46, MRMSrcMem, // if <=u, R32 = [mem32]
555 (ops R32:$dst, R32:$src1, i32mem:$src2),
556 "cmovbe {$src2, $dst|$dst, $src2}", []>, TB;
558 def CMOVA16rr : I<0x47, MRMSrcReg, // if >u, R16 = R16
559 (ops R16:$dst, R16:$src1, R16:$src2),
560 "cmova {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
561 def CMOVA16rm : I<0x47, MRMSrcMem, // if >u, R16 = [mem16]
562 (ops R16:$dst, R16:$src1, i16mem:$src2),
563 "cmova {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
564 def CMOVA32rr : I<0x47, MRMSrcReg, // if >u, R32 = R32
565 (ops R32:$dst, R32:$src1, R32:$src2),
566 "cmova {$src2, $dst|$dst, $src2}", []>, TB;
567 def CMOVA32rm : I<0x47, MRMSrcMem, // if >u, R32 = [mem32]
568 (ops R32:$dst, R32:$src1, i32mem:$src2),
569 "cmova {$src2, $dst|$dst, $src2}", []>, TB;
571 def CMOVS16rr : I<0x48, MRMSrcReg, // if signed, R16 = R16
572 (ops R16:$dst, R16:$src1, R16:$src2),
573 "cmovs {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
574 def CMOVS16rm : I<0x48, MRMSrcMem, // if signed, R16 = [mem16]
575 (ops R16:$dst, R16:$src1, i16mem:$src2),
576 "cmovs {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
577 def CMOVS32rr : I<0x48, MRMSrcReg, // if signed, R32 = R32
578 (ops R32:$dst, R32:$src1, R32:$src2),
579 "cmovs {$src2, $dst|$dst, $src2}", []>, TB;
580 def CMOVS32rm : I<0x48, MRMSrcMem, // if signed, R32 = [mem32]
581 (ops R32:$dst, R32:$src1, i32mem:$src2),
582 "cmovs {$src2, $dst|$dst, $src2}", []>, TB;
584 def CMOVNS16rr: I<0x49, MRMSrcReg, // if !signed, R16 = R16
585 (ops R16:$dst, R16:$src1, R16:$src2),
586 "cmovns {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
587 def CMOVNS16rm: I<0x49, MRMSrcMem, // if !signed, R16 = [mem16]
588 (ops R16:$dst, R16:$src1, i16mem:$src2),
589 "cmovns {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
590 def CMOVNS32rr: I<0x49, MRMSrcReg, // if !signed, R32 = R32
591 (ops R32:$dst, R32:$src1, R32:$src2),
592 "cmovns {$src2, $dst|$dst, $src2}", []>, TB;
593 def CMOVNS32rm: I<0x49, MRMSrcMem, // if !signed, R32 = [mem32]
594 (ops R32:$dst, R32:$src1, i32mem:$src2),
595 "cmovns {$src2, $dst|$dst, $src2}", []>, TB;
597 def CMOVP16rr : I<0x4A, MRMSrcReg, // if parity, R16 = R16
598 (ops R16:$dst, R16:$src1, R16:$src2),
599 "cmovp {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
600 def CMOVP16rm : I<0x4A, MRMSrcMem, // if parity, R16 = [mem16]
601 (ops R16:$dst, R16:$src1, i16mem:$src2),
602 "cmovp {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
603 def CMOVP32rr : I<0x4A, MRMSrcReg, // if parity, R32 = R32
604 (ops R32:$dst, R32:$src1, R32:$src2),
605 "cmovp {$src2, $dst|$dst, $src2}", []>, TB;
606 def CMOVP32rm : I<0x4A, MRMSrcMem, // if parity, R32 = [mem32]
607 (ops R32:$dst, R32:$src1, i32mem:$src2),
608 "cmovp {$src2, $dst|$dst, $src2}", []>, TB;
611 def CMOVNP16rr : I<0x4B, MRMSrcReg, // if !parity, R16 = R16
612 (ops R16:$dst, R16:$src1, R16:$src2),
613 "cmovnp {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
614 def CMOVNP16rm : I<0x4B, MRMSrcMem, // if !parity, R16 = [mem16]
615 (ops R16:$dst, R16:$src1, i16mem:$src2),
616 "cmovnp {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
617 def CMOVNP32rr : I<0x4B, MRMSrcReg, // if !parity, R32 = R32
618 (ops R32:$dst, R32:$src1, R32:$src2),
619 "cmovnp {$src2, $dst|$dst, $src2}", []>, TB;
620 def CMOVNP32rm : I<0x4B, MRMSrcMem, // if !parity, R32 = [mem32]
621 (ops R32:$dst, R32:$src1, i32mem:$src2),
622 "cmovnp {$src2, $dst|$dst, $src2}", []>, TB;
625 def CMOVL16rr : I<0x4C, MRMSrcReg, // if <s, R16 = R16
626 (ops R16:$dst, R16:$src1, R16:$src2),
627 "cmovl {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
628 def CMOVL16rm : I<0x4C, MRMSrcMem, // if <s, R16 = [mem16]
629 (ops R16:$dst, R16:$src1, i16mem:$src2),
630 "cmovl {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
631 def CMOVL32rr : I<0x4C, MRMSrcReg, // if <s, R32 = R32
632 (ops R32:$dst, R32:$src1, R32:$src2),
633 "cmovl {$src2, $dst|$dst, $src2}", []>, TB;
634 def CMOVL32rm : I<0x4C, MRMSrcMem, // if <s, R32 = [mem32]
635 (ops R32:$dst, R32:$src1, i32mem:$src2),
636 "cmovl {$src2, $dst|$dst, $src2}", []>, TB;
638 def CMOVGE16rr: I<0x4D, MRMSrcReg, // if >=s, R16 = R16
639 (ops R16:$dst, R16:$src1, R16:$src2),
640 "cmovge {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
641 def CMOVGE16rm: I<0x4D, MRMSrcMem, // if >=s, R16 = [mem16]
642 (ops R16:$dst, R16:$src1, i16mem:$src2),
643 "cmovge {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
644 def CMOVGE32rr: I<0x4D, MRMSrcReg, // if >=s, R32 = R32
645 (ops R32:$dst, R32:$src1, R32:$src2),
646 "cmovge {$src2, $dst|$dst, $src2}", []>, TB;
647 def CMOVGE32rm: I<0x4D, MRMSrcMem, // if >=s, R32 = [mem32]
648 (ops R32:$dst, R32:$src1, i32mem:$src2),
649 "cmovge {$src2, $dst|$dst, $src2}", []>, TB;
651 def CMOVLE16rr: I<0x4E, MRMSrcReg, // if <=s, R16 = R16
652 (ops R16:$dst, R16:$src1, R16:$src2),
653 "cmovle {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
654 def CMOVLE16rm: I<0x4E, MRMSrcMem, // if <=s, R16 = [mem16]
655 (ops R16:$dst, R16:$src1, i16mem:$src2),
656 "cmovle {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
657 def CMOVLE32rr: I<0x4E, MRMSrcReg, // if <=s, R32 = R32
658 (ops R32:$dst, R32:$src1, R32:$src2),
659 "cmovle {$src2, $dst|$dst, $src2}", []>, TB;
660 def CMOVLE32rm: I<0x4E, MRMSrcMem, // if <=s, R32 = [mem32]
661 (ops R32:$dst, R32:$src1, i32mem:$src2),
662 "cmovle {$src2, $dst|$dst, $src2}", []>, TB;
664 def CMOVG16rr : I<0x4F, MRMSrcReg, // if >s, R16 = R16
665 (ops R16:$dst, R16:$src1, R16:$src2),
666 "cmovg {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
667 def CMOVG16rm : I<0x4F, MRMSrcMem, // if >s, R16 = [mem16]
668 (ops R16:$dst, R16:$src1, i16mem:$src2),
669 "cmovg {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
670 def CMOVG32rr : I<0x4F, MRMSrcReg, // if >s, R32 = R32
671 (ops R32:$dst, R32:$src1, R32:$src2),
672 "cmovg {$src2, $dst|$dst, $src2}", []>, TB;
673 def CMOVG32rm : I<0x4F, MRMSrcMem, // if >s, R32 = [mem32]
674 (ops R32:$dst, R32:$src1, i32mem:$src2),
675 "cmovg {$src2, $dst|$dst, $src2}", []>, TB;
677 // unary instructions
678 def NEG8r : I<0xF6, MRM3r, (ops R8 :$dst, R8 :$src), "neg{b} $dst",
679 [(set R8:$dst, (ineg R8:$src))]>;
680 def NEG16r : I<0xF7, MRM3r, (ops R16:$dst, R16:$src), "neg{w} $dst",
681 [(set R16:$dst, (ineg R16:$src))]>, OpSize;
682 def NEG32r : I<0xF7, MRM3r, (ops R32:$dst, R32:$src), "neg{l} $dst",
683 [(set R32:$dst, (ineg R32:$src))]>;
684 let isTwoAddress = 0 in {
685 def NEG8m : I<0xF6, MRM3m, (ops i8mem :$dst), "neg{b} $dst", []>;
686 def NEG16m : I<0xF7, MRM3m, (ops i16mem:$dst), "neg{w} $dst", []>, OpSize;
687 def NEG32m : I<0xF7, MRM3m, (ops i32mem:$dst), "neg{l} $dst", []>;
690 def NOT8r : I<0xF6, MRM2r, (ops R8 :$dst, R8 :$src), "not{b} $dst",
691 [(set R8:$dst, (not R8:$src))]>;
692 def NOT16r : I<0xF7, MRM2r, (ops R16:$dst, R16:$src), "not{w} $dst",
693 [(set R16:$dst, (not R16:$src))]>, OpSize;
694 def NOT32r : I<0xF7, MRM2r, (ops R32:$dst, R32:$src), "not{l} $dst",
695 [(set R32:$dst, (not R32:$src))]>;
696 let isTwoAddress = 0 in {
697 def NOT8m : I<0xF6, MRM2m, (ops i8mem :$dst), "not{b} $dst", []>;
698 def NOT16m : I<0xF7, MRM2m, (ops i16mem:$dst), "not{w} $dst", []>, OpSize;
699 def NOT32m : I<0xF7, MRM2m, (ops i32mem:$dst), "not{l} $dst", []>;
702 // TODO: inc/dec is slow for P4, but fast for Pentium-M.
703 def INC8r : I<0xFE, MRM0r, (ops R8 :$dst, R8 :$src), "inc{b} $dst",
704 [(set R8:$dst, (add R8:$src, 1))]>;
705 let isConvertibleToThreeAddress = 1 in { // Can transform into LEA.
706 def INC16r : I<0xFF, MRM0r, (ops R16:$dst, R16:$src), "inc{w} $dst",
707 [(set R16:$dst, (add R16:$src, 1))]>, OpSize;
708 def INC32r : I<0xFF, MRM0r, (ops R32:$dst, R32:$src), "inc{l} $dst",
709 [(set R32:$dst, (add R32:$src, 1))]>;
711 let isTwoAddress = 0 in {
712 def INC8m : I<0xFE, MRM0m, (ops i8mem :$dst), "inc{b} $dst", []>;
713 def INC16m : I<0xFF, MRM0m, (ops i16mem:$dst), "inc{w} $dst", []>, OpSize;
714 def INC32m : I<0xFF, MRM0m, (ops i32mem:$dst), "inc{l} $dst", []>;
717 def DEC8r : I<0xFE, MRM1r, (ops R8 :$dst, R8 :$src), "dec{b} $dst",
718 [(set R8:$dst, (add R8:$src, -1))]>;
719 let isConvertibleToThreeAddress = 1 in { // Can transform into LEA.
720 def DEC16r : I<0xFF, MRM1r, (ops R16:$dst, R16:$src), "dec{w} $dst",
721 [(set R16:$dst, (add R16:$src, -1))]>, OpSize;
722 def DEC32r : I<0xFF, MRM1r, (ops R32:$dst, R32:$src), "dec{l} $dst",
723 [(set R32:$dst, (add R32:$src, -1))]>;
726 let isTwoAddress = 0 in {
727 def DEC8m : I<0xFE, MRM1m, (ops i8mem :$dst), "dec{b} $dst", []>;
728 def DEC16m : I<0xFF, MRM1m, (ops i16mem:$dst), "dec{w} $dst", []>, OpSize;
729 def DEC32m : I<0xFF, MRM1m, (ops i32mem:$dst), "dec{l} $dst", []>;
732 // Logical operators...
733 let isCommutable = 1 in { // X = AND Y, Z --> X = AND Z, Y
734 def AND8rr : I<0x20, MRMDestReg,
735 (ops R8 :$dst, R8 :$src1, R8 :$src2),
736 "and{b} {$src2, $dst|$dst, $src2}",
737 [(set R8:$dst, (and R8:$src1, R8:$src2))]>;
738 def AND16rr : I<0x21, MRMDestReg,
739 (ops R16:$dst, R16:$src1, R16:$src2),
740 "and{w} {$src2, $dst|$dst, $src2}",
741 [(set R16:$dst, (and R16:$src1, R16:$src2))]>, OpSize;
742 def AND32rr : I<0x21, MRMDestReg,
743 (ops R32:$dst, R32:$src1, R32:$src2),
744 "and{l} {$src2, $dst|$dst, $src2}",
745 [(set R32:$dst, (and R32:$src1, R32:$src2))]>;
748 def AND8rm : I<0x22, MRMSrcMem,
749 (ops R8 :$dst, R8 :$src1, i8mem :$src2),
750 "and{b} {$src2, $dst|$dst, $src2}",[]>;
751 def AND16rm : I<0x23, MRMSrcMem,
752 (ops R16:$dst, R16:$src1, i16mem:$src2),
753 "and{w} {$src2, $dst|$dst, $src2}", []>, OpSize;
754 def AND32rm : I<0x23, MRMSrcMem,
755 (ops R32:$dst, R32:$src1, i32mem:$src2),
756 "and{l} {$src2, $dst|$dst, $src2}", []>;
758 def AND8ri : Ii8<0x80, MRM4r,
759 (ops R8 :$dst, R8 :$src1, i8imm :$src2),
760 "and{b} {$src2, $dst|$dst, $src2}",
761 [(set R8:$dst, (and R8:$src1, imm:$src2))]>;
762 def AND16ri : Ii16<0x81, MRM4r,
763 (ops R16:$dst, R16:$src1, i16imm:$src2),
764 "and{w} {$src2, $dst|$dst, $src2}",
765 [(set R16:$dst, (and R16:$src1, imm:$src2))]>, OpSize;
766 def AND32ri : Ii32<0x81, MRM4r,
767 (ops R32:$dst, R32:$src1, i32imm:$src2),
768 "and{l} {$src2, $dst|$dst, $src2}",
769 [(set R32:$dst, (and R32:$src1, imm:$src2))]>;
770 def AND16ri8 : Ii8<0x83, MRM4r,
771 (ops R16:$dst, R16:$src1, i16i8imm:$src2),
772 "and{w} {$src2, $dst|$dst, $src2}",
773 [(set R16:$dst, (and R16:$src1, i16SExt8:$src2))]>, OpSize;
774 def AND32ri8 : Ii8<0x83, MRM4r,
775 (ops R32:$dst, R32:$src1, i32i8imm:$src2),
776 "and{l} {$src2, $dst|$dst, $src2}",
777 [(set R32:$dst, (and R32:$src1, i32SExt8:$src2))]>;
779 let isTwoAddress = 0 in {
780 def AND8mr : I<0x20, MRMDestMem,
781 (ops i8mem :$dst, R8 :$src),
782 "and{b} {$src, $dst|$dst, $src}", []>;
783 def AND16mr : I<0x21, MRMDestMem,
784 (ops i16mem:$dst, R16:$src),
785 "and{w} {$src, $dst|$dst, $src}", []>, OpSize;
786 def AND32mr : I<0x21, MRMDestMem,
787 (ops i32mem:$dst, R32:$src),
788 "and{l} {$src, $dst|$dst, $src}", []>;
789 def AND8mi : Ii8<0x80, MRM4m,
790 (ops i8mem :$dst, i8imm :$src),
791 "and{b} {$src, $dst|$dst, $src}", []>;
792 def AND16mi : Ii16<0x81, MRM4m,
793 (ops i16mem:$dst, i16imm:$src),
794 "and{w} {$src, $dst|$dst, $src}", []>, OpSize;
795 def AND32mi : Ii32<0x81, MRM4m,
796 (ops i32mem:$dst, i32imm:$src),
797 "and{l} {$src, $dst|$dst, $src}", []>;
798 def AND16mi8 : Ii8<0x83, MRM4m,
799 (ops i16mem:$dst, i8imm :$src),
800 "and{w} {$src, $dst|$dst, $src}", []>, OpSize;
801 def AND32mi8 : Ii8<0x83, MRM4m,
802 (ops i32mem:$dst, i8imm :$src),
803 "and{l} {$src, $dst|$dst, $src}", []>;
807 let isCommutable = 1 in { // X = OR Y, Z --> X = OR Z, Y
808 def OR8rr : I<0x08, MRMDestReg, (ops R8 :$dst, R8 :$src1, R8 :$src2),
809 "or{b} {$src2, $dst|$dst, $src2}",
810 [(set R8:$dst, (or R8:$src1, R8:$src2))]>;
811 def OR16rr : I<0x09, MRMDestReg, (ops R16:$dst, R16:$src1, R16:$src2),
812 "or{w} {$src2, $dst|$dst, $src2}",
813 [(set R16:$dst, (or R16:$src1, R16:$src2))]>, OpSize;
814 def OR32rr : I<0x09, MRMDestReg, (ops R32:$dst, R32:$src1, R32:$src2),
815 "or{l} {$src2, $dst|$dst, $src2}",
816 [(set R32:$dst, (or R32:$src1, R32:$src2))]>;
818 def OR8rm : I<0x0A, MRMSrcMem , (ops R8 :$dst, R8 :$src1, i8mem :$src2),
819 "or{b} {$src2, $dst|$dst, $src2}", []>;
820 def OR16rm : I<0x0B, MRMSrcMem , (ops R16:$dst, R16:$src1, i16mem:$src2),
821 "or{w} {$src2, $dst|$dst, $src2}", []>, OpSize;
822 def OR32rm : I<0x0B, MRMSrcMem , (ops R32:$dst, R32:$src1, i32mem:$src2),
823 "or{l} {$src2, $dst|$dst, $src2}", []>;
825 def OR8ri : Ii8 <0x80, MRM1r, (ops R8 :$dst, R8 :$src1, i8imm:$src2),
826 "or{b} {$src2, $dst|$dst, $src2}",
827 [(set R8:$dst, (or R8:$src1, imm:$src2))]>;
828 def OR16ri : Ii16<0x81, MRM1r, (ops R16:$dst, R16:$src1, i16imm:$src2),
829 "or{w} {$src2, $dst|$dst, $src2}",
830 [(set R16:$dst, (or R16:$src1, imm:$src2))]>, OpSize;
831 def OR32ri : Ii32<0x81, MRM1r, (ops R32:$dst, R32:$src1, i32imm:$src2),
832 "or{l} {$src2, $dst|$dst, $src2}",
833 [(set R32:$dst, (or R32:$src1, imm:$src2))]>;
835 def OR16ri8 : Ii8<0x83, MRM1r, (ops R16:$dst, R16:$src1, i16i8imm:$src2),
836 "or{w} {$src2, $dst|$dst, $src2}",
837 [(set R16:$dst, (or R16:$src1, i16SExt8:$src2))]>, OpSize;
838 def OR32ri8 : Ii8<0x83, MRM1r, (ops R32:$dst, R32:$src1, i32i8imm:$src2),
839 "or{l} {$src2, $dst|$dst, $src2}",
840 [(set R32:$dst, (or R32:$src1, i32SExt8:$src2))]>;
841 let isTwoAddress = 0 in {
842 def OR8mr : I<0x08, MRMDestMem, (ops i8mem:$dst, R8:$src),
843 "or{b} {$src, $dst|$dst, $src}", []>;
844 def OR16mr : I<0x09, MRMDestMem, (ops i16mem:$dst, R16:$src),
845 "or{w} {$src, $dst|$dst, $src}", []>, OpSize;
846 def OR32mr : I<0x09, MRMDestMem, (ops i32mem:$dst, R32:$src),
847 "or{l} {$src, $dst|$dst, $src}", []>;
848 def OR8mi : Ii8<0x80, MRM1m, (ops i8mem :$dst, i8imm:$src),
849 "or{b} {$src, $dst|$dst, $src}", []>;
850 def OR16mi : Ii16<0x81, MRM1m, (ops i16mem:$dst, i16imm:$src),
851 "or{w} {$src, $dst|$dst, $src}", []>, OpSize;
852 def OR32mi : Ii32<0x81, MRM1m, (ops i32mem:$dst, i32imm:$src),
853 "or{l} {$src, $dst|$dst, $src}", []>;
854 def OR16mi8 : Ii8<0x83, MRM1m, (ops i16mem:$dst, i8imm:$src),
855 "or{w} {$src, $dst|$dst, $src}", []>, OpSize;
856 def OR32mi8 : Ii8<0x83, MRM1m, (ops i32mem:$dst, i8imm:$src),
857 "or{l} {$src, $dst|$dst, $src}", []>;
861 let isCommutable = 1 in { // X = XOR Y, Z --> X = XOR Z, Y
862 def XOR8rr : I<0x30, MRMDestReg,
863 (ops R8 :$dst, R8 :$src1, R8 :$src2),
864 "xor{b} {$src2, $dst|$dst, $src2}",
865 [(set R8:$dst, (xor R8:$src1, R8:$src2))]>;
866 def XOR16rr : I<0x31, MRMDestReg,
867 (ops R16:$dst, R16:$src1, R16:$src2),
868 "xor{w} {$src2, $dst|$dst, $src2}",
869 [(set R16:$dst, (xor R16:$src1, R16:$src2))]>, OpSize;
870 def XOR32rr : I<0x31, MRMDestReg,
871 (ops R32:$dst, R32:$src1, R32:$src2),
872 "xor{l} {$src2, $dst|$dst, $src2}",
873 [(set R32:$dst, (xor R32:$src1, R32:$src2))]>;
876 def XOR8rm : I<0x32, MRMSrcMem ,
877 (ops R8 :$dst, R8:$src1, i8mem :$src2),
878 "xor{b} {$src2, $dst|$dst, $src2}", []>;
879 def XOR16rm : I<0x33, MRMSrcMem ,
880 (ops R16:$dst, R8:$src1, i16mem:$src2),
881 "xor{w} {$src2, $dst|$dst, $src2}", []>, OpSize;
882 def XOR32rm : I<0x33, MRMSrcMem ,
883 (ops R32:$dst, R8:$src1, i32mem:$src2),
884 "xor{l} {$src2, $dst|$dst, $src2}", []>;
886 def XOR8ri : Ii8<0x80, MRM6r,
887 (ops R8:$dst, R8:$src1, i8imm:$src2),
888 "xor{b} {$src2, $dst|$dst, $src2}",
889 [(set R8:$dst, (xor R8:$src1, imm:$src2))]>;
890 def XOR16ri : Ii16<0x81, MRM6r,
891 (ops R16:$dst, R16:$src1, i16imm:$src2),
892 "xor{w} {$src2, $dst|$dst, $src2}",
893 [(set R16:$dst, (xor R16:$src1, imm:$src2))]>, OpSize;
894 def XOR32ri : Ii32<0x81, MRM6r,
895 (ops R32:$dst, R32:$src1, i32imm:$src2),
896 "xor{l} {$src2, $dst|$dst, $src2}",
897 [(set R32:$dst, (xor R32:$src1, imm:$src2))]>;
898 def XOR16ri8 : Ii8<0x83, MRM6r,
899 (ops R16:$dst, R16:$src1, i16i8imm:$src2),
900 "xor{w} {$src2, $dst|$dst, $src2}",
901 [(set R16:$dst, (xor R16:$src1, i16SExt8:$src2))]>, OpSize;
902 def XOR32ri8 : Ii8<0x83, MRM6r,
903 (ops R32:$dst, R32:$src1, i32i8imm:$src2),
904 "xor{l} {$src2, $dst|$dst, $src2}",
905 [(set R32:$dst, (xor R32:$src1, i32SExt8:$src2))]>;
906 let isTwoAddress = 0 in {
907 def XOR8mr : I<0x30, MRMDestMem,
908 (ops i8mem :$dst, R8 :$src),
909 "xor{b} {$src, $dst|$dst, $src}", []>;
910 def XOR16mr : I<0x31, MRMDestMem,
911 (ops i16mem:$dst, R16:$src),
912 "xor{w} {$src, $dst|$dst, $src}", []>, OpSize;
913 def XOR32mr : I<0x31, MRMDestMem,
914 (ops i32mem:$dst, R32:$src),
915 "xor{l} {$src, $dst|$dst, $src}", []>;
916 def XOR8mi : Ii8<0x80, MRM6m,
917 (ops i8mem :$dst, i8imm :$src),
918 "xor{b} {$src, $dst|$dst, $src}", []>;
919 def XOR16mi : Ii16<0x81, MRM6m,
920 (ops i16mem:$dst, i16imm:$src),
921 "xor{w} {$src, $dst|$dst, $src}", []>, OpSize;
922 def XOR32mi : Ii32<0x81, MRM6m,
923 (ops i32mem:$dst, i32imm:$src),
924 "xor{l} {$src, $dst|$dst, $src}", []>;
925 def XOR16mi8 : Ii8<0x83, MRM6m,
926 (ops i16mem:$dst, i8imm :$src),
927 "xor{w} {$src, $dst|$dst, $src}", []>, OpSize;
928 def XOR32mi8 : Ii8<0x83, MRM6m,
929 (ops i32mem:$dst, i8imm :$src),
930 "xor{l} {$src, $dst|$dst, $src}", []>;
933 // Shift instructions
934 // FIXME: provide shorter instructions when imm8 == 1
935 def SHL8rCL : I<0xD2, MRM4r, (ops R8 :$dst, R8 :$src),
936 "shl{b} {%cl, $dst|$dst, %CL}",
937 [(set R8:$dst, (shl R8:$src, CL))]>, Imp<[CL],[]>;
938 def SHL16rCL : I<0xD3, MRM4r, (ops R16:$dst, R16:$src),
939 "shl{w} {%cl, $dst|$dst, %CL}",
940 [(set R16:$dst, (shl R16:$src, CL))]>, Imp<[CL],[]>, OpSize;
941 def SHL32rCL : I<0xD3, MRM4r, (ops R32:$dst, R32:$src),
942 "shl{l} {%cl, $dst|$dst, %CL}",
943 [(set R32:$dst, (shl R32:$src, CL))]>, Imp<[CL],[]>;
945 def SHL8ri : Ii8<0xC0, MRM4r, (ops R8 :$dst, R8 :$src1, i8imm:$src2),
946 "shl{b} {$src2, $dst|$dst, $src2}",
947 [(set R8:$dst, (shl R8:$src1, (i8 imm:$src2)))]>;
948 let isConvertibleToThreeAddress = 1 in { // Can transform into LEA.
949 def SHL16ri : Ii8<0xC1, MRM4r, (ops R16:$dst, R16:$src1, i8imm:$src2),
950 "shl{w} {$src2, $dst|$dst, $src2}",
951 [(set R16:$dst, (shl R16:$src1, (i8 imm:$src2)))]>, OpSize;
952 def SHL32ri : Ii8<0xC1, MRM4r, (ops R32:$dst, R32:$src1, i8imm:$src2),
953 "shl{l} {$src2, $dst|$dst, $src2}",
954 [(set R32:$dst, (shl R32:$src1, (i8 imm:$src2)))]>;
957 let isTwoAddress = 0 in {
958 def SHL8mCL : I<0xD2, MRM4m, (ops i8mem :$dst),
959 "shl{b} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>;
960 def SHL16mCL : I<0xD3, MRM4m, (ops i16mem:$dst),
961 "shl{w} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>, OpSize;
962 def SHL32mCL : I<0xD3, MRM4m, (ops i32mem:$dst),
963 "shl{l} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>;
964 def SHL8mi : Ii8<0xC0, MRM4m, (ops i8mem :$dst, i8imm:$src),
965 "shl{b} {$src, $dst|$dst, $src}", []>;
966 def SHL16mi : Ii8<0xC1, MRM4m, (ops i16mem:$dst, i8imm:$src),
967 "shl{w} {$src, $dst|$dst, $src}", []>, OpSize;
968 def SHL32mi : Ii8<0xC1, MRM4m, (ops i32mem:$dst, i8imm:$src),
969 "shl{l} {$src, $dst|$dst, $src}", []>;
972 def SHR8rCL : I<0xD2, MRM5r, (ops R8 :$dst, R8 :$src),
973 "shr{b} {%cl, $dst|$dst, %CL}",
974 [(set R8:$dst, (srl R8:$src, CL))]>, Imp<[CL],[]>;
975 def SHR16rCL : I<0xD3, MRM5r, (ops R16:$dst, R16:$src),
976 "shr{w} {%cl, $dst|$dst, %CL}",
977 [(set R16:$dst, (srl R16:$src, CL))]>, Imp<[CL],[]>, OpSize;
978 def SHR32rCL : I<0xD3, MRM5r, (ops R32:$dst, R32:$src),
979 "shr{l} {%cl, $dst|$dst, %CL}",
980 [(set R32:$dst, (srl R32:$src, CL))]>, Imp<[CL],[]>;
982 def SHR8ri : Ii8<0xC0, MRM5r, (ops R8:$dst, R8:$src1, i8imm:$src2),
983 "shr{b} {$src2, $dst|$dst, $src2}",
984 [(set R8:$dst, (srl R8:$src1, (i8 imm:$src2)))]>;
985 def SHR16ri : Ii8<0xC1, MRM5r, (ops R16:$dst, R16:$src1, i8imm:$src2),
986 "shr{w} {$src2, $dst|$dst, $src2}",
987 [(set R16:$dst, (srl R16:$src1, (i8 imm:$src2)))]>, OpSize;
988 def SHR32ri : Ii8<0xC1, MRM5r, (ops R32:$dst, R32:$src1, i8imm:$src2),
989 "shr{l} {$src2, $dst|$dst, $src2}",
990 [(set R32:$dst, (srl R32:$src1, (i8 imm:$src2)))]>;
992 let isTwoAddress = 0 in {
993 def SHR8mCL : I<0xD2, MRM5m, (ops i8mem :$dst),
994 "shr{b} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>;
995 def SHR16mCL : I<0xD3, MRM5m, (ops i16mem:$dst),
996 "shr{w} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>, OpSize;
997 def SHR32mCL : I<0xD3, MRM5m, (ops i32mem:$dst),
998 "shr{l} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>;
999 def SHR8mi : Ii8<0xC0, MRM5m, (ops i8mem :$dst, i8imm:$src),
1000 "shr{b} {$src, $dst|$dst, $src}", []>;
1001 def SHR16mi : Ii8<0xC1, MRM5m, (ops i16mem:$dst, i8imm:$src),
1002 "shr{w} {$src, $dst|$dst, $src}", []>, OpSize;
1003 def SHR32mi : Ii8<0xC1, MRM5m, (ops i32mem:$dst, i8imm:$src),
1004 "shr{l} {$src, $dst|$dst, $src}", []>;
1007 def SAR8rCL : I<0xD2, MRM7r, (ops R8 :$dst, R8 :$src),
1008 "sar{b} {%cl, $dst|$dst, %CL}",
1009 [(set R8:$dst, (sra R8:$src, CL))]>, Imp<[CL],[]>;
1010 def SAR16rCL : I<0xD3, MRM7r, (ops R16:$dst, R16:$src),
1011 "sar{w} {%cl, $dst|$dst, %CL}",
1012 [(set R16:$dst, (sra R16:$src, CL))]>, Imp<[CL],[]>, OpSize;
1013 def SAR32rCL : I<0xD3, MRM7r, (ops R32:$dst, R32:$src),
1014 "sar{l} {%cl, $dst|$dst, %CL}",
1015 [(set R32:$dst, (sra R32:$src, CL))]>, Imp<[CL],[]>;
1017 def SAR8ri : Ii8<0xC0, MRM7r, (ops R8 :$dst, R8 :$src1, i8imm:$src2),
1018 "sar{b} {$src2, $dst|$dst, $src2}",
1019 [(set R8:$dst, (sra R8:$src1, (i8 imm:$src2)))]>;
1020 def SAR16ri : Ii8<0xC1, MRM7r, (ops R16:$dst, R16:$src1, i8imm:$src2),
1021 "sar{w} {$src2, $dst|$dst, $src2}",
1022 [(set R16:$dst, (sra R16:$src1, (i8 imm:$src2)))]>,
1024 def SAR32ri : Ii8<0xC1, MRM7r, (ops R32:$dst, R32:$src1, i8imm:$src2),
1025 "sar{l} {$src2, $dst|$dst, $src2}",
1026 [(set R32:$dst, (sra R32:$src1, (i8 imm:$src2)))]>;
1027 let isTwoAddress = 0 in {
1028 def SAR8mCL : I<0xD2, MRM7m, (ops i8mem :$dst),
1029 "sar{b} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>;
1030 def SAR16mCL : I<0xD3, MRM7m, (ops i16mem:$dst),
1031 "sar{w} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>, OpSize;
1032 def SAR32mCL : I<0xD3, MRM7m, (ops i32mem:$dst),
1033 "sar{l} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>;
1034 def SAR8mi : Ii8<0xC0, MRM7m, (ops i8mem :$dst, i8imm:$src),
1035 "sar{b} {$src, $dst|$dst, $src}", []>;
1036 def SAR16mi : Ii8<0xC1, MRM7m, (ops i16mem:$dst, i8imm:$src),
1037 "sar{w} {$src, $dst|$dst, $src}", []>, OpSize;
1038 def SAR32mi : Ii8<0xC1, MRM7m, (ops i32mem:$dst, i8imm:$src),
1039 "sar{l} {$src, $dst|$dst, $src}", []>;
1042 // Rotate instructions
1043 // FIXME: provide shorter instructions when imm8 == 1
1044 def ROL8rCL : I<0xD2, MRM0r, (ops R8 :$dst, R8 :$src),
1045 "rol{b} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>;
1046 def ROL16rCL : I<0xD3, MRM0r, (ops R16:$dst, R16:$src),
1047 "rol{w} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>, OpSize;
1048 def ROL32rCL : I<0xD3, MRM0r, (ops R32:$dst, R32:$src),
1049 "rol{l} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>;
1051 def ROL8ri : Ii8<0xC0, MRM0r, (ops R8 :$dst, R8 :$src1, i8imm:$src2),
1052 "rol{b} {$src2, $dst|$dst, $src2}", []>;
1053 def ROL16ri : Ii8<0xC1, MRM0r, (ops R16:$dst, R16:$src1, i8imm:$src2),
1054 "rol{w} {$src2, $dst|$dst, $src2}", []>, OpSize;
1055 def ROL32ri : Ii8<0xC1, MRM0r, (ops R32:$dst, R32:$src1, i8imm:$src2),
1056 "rol{l} {$src2, $dst|$dst, $src2}", []>;
1058 let isTwoAddress = 0 in {
1059 def ROL8mCL : I<0xD2, MRM0m, (ops i8mem :$dst),
1060 "rol{b} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>;
1061 def ROL16mCL : I<0xD3, MRM0m, (ops i16mem:$dst),
1062 "rol{w} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>, OpSize;
1063 def ROL32mCL : I<0xD3, MRM0m, (ops i32mem:$dst),
1064 "rol{l} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>;
1065 def ROL8mi : Ii8<0xC0, MRM0m, (ops i8mem :$dst, i8imm:$src),
1066 "rol{b} {$src, $dst|$dst, $src}", []>;
1067 def ROL16mi : Ii8<0xC1, MRM0m, (ops i16mem:$dst, i8imm:$src),
1068 "rol{w} {$src, $dst|$dst, $src}", []>, OpSize;
1069 def ROL32mi : Ii8<0xC1, MRM0m, (ops i32mem:$dst, i8imm:$src),
1070 "rol{l} {$src, $dst|$dst, $src}", []>;
1073 def ROR8rCL : I<0xD2, MRM1r, (ops R8 :$dst, R8 :$src),
1074 "ror{b} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>;
1075 def ROR16rCL : I<0xD3, MRM1r, (ops R16:$dst, R16:$src),
1076 "ror{w} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>, OpSize;
1077 def ROR32rCL : I<0xD3, MRM1r, (ops R32:$dst, R32:$src),
1078 "ror{l} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>;
1080 def ROR8ri : Ii8<0xC0, MRM1r, (ops R8 :$dst, R8 :$src1, i8imm:$src2),
1081 "ror{b} {$src2, $dst|$dst, $src2}", []>;
1082 def ROR16ri : Ii8<0xC1, MRM1r, (ops R16:$dst, R16:$src1, i8imm:$src2),
1083 "ror{w} {$src2, $dst|$dst, $src2}", []>, OpSize;
1084 def ROR32ri : Ii8<0xC1, MRM1r, (ops R32:$dst, R32:$src1, i8imm:$src2),
1085 "ror{l} {$src2, $dst|$dst, $src2}", []>;
1086 let isTwoAddress = 0 in {
1087 def ROR8mCL : I<0xD2, MRM1m, (ops i8mem :$dst),
1088 "ror{b} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>;
1089 def ROR16mCL : I<0xD3, MRM1m, (ops i16mem:$dst),
1090 "ror{w} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>, OpSize;
1091 def ROR32mCL : I<0xD3, MRM1m, (ops i32mem:$dst),
1092 "ror{l} {%cl, $dst|$dst, %CL}", []>, Imp<[CL],[]>;
1093 def ROR8mi : Ii8<0xC0, MRM1m, (ops i8mem :$dst, i8imm:$src),
1094 "ror{b} {$src, $dst|$dst, $src}", []>;
1095 def ROR16mi : Ii8<0xC1, MRM1m, (ops i16mem:$dst, i8imm:$src),
1096 "ror{w} {$src, $dst|$dst, $src}", []>, OpSize;
1097 def ROR32mi : Ii8<0xC1, MRM1m, (ops i32mem:$dst, i8imm:$src),
1098 "ror{l} {$src, $dst|$dst, $src}", []>;
1103 // Double shift instructions (generalizations of rotate)
1105 def SHLD32rrCL : I<0xA5, MRMDestReg, (ops R32:$dst, R32:$src1, R32:$src2),
1106 "shld{l} {%cl, $src2, $dst|$dst, $src2, %CL}", []>,
1108 def SHRD32rrCL : I<0xAD, MRMDestReg, (ops R32:$dst, R32:$src1, R32:$src2),
1109 "shrd{l} {%cl, $src2, $dst|$dst, $src2, %CL}", []>,
1111 def SHLD16rrCL : I<0xA5, MRMDestReg, (ops R16:$dst, R16:$src1, R16:$src2),
1112 "shld{w} {%cl, $src2, $dst|$dst, $src2, %CL}", []>,
1113 Imp<[CL],[]>, TB, OpSize;
1114 def SHRD16rrCL : I<0xAD, MRMDestReg, (ops R16:$dst, R16:$src1, R16:$src2),
1115 "shrd{w} {%cl, $src2, $dst|$dst, $src2, %CL}", []>,
1116 Imp<[CL],[]>, TB, OpSize;
1118 let isCommutable = 1 in { // These instructions commute to each other.
1119 def SHLD32rri8 : Ii8<0xA4, MRMDestReg,
1120 (ops R32:$dst, R32:$src1, R32:$src2, i8imm:$src3),
1121 "shld{l} {$src3, $src2, $dst|$dst, $src2, $src3}", []>, TB;
1122 def SHRD32rri8 : Ii8<0xAC, MRMDestReg,
1123 (ops R32:$dst, R32:$src1, R32:$src2, i8imm:$src3),
1124 "shrd{l} {$src3, $src2, $dst|$dst, $src2, $src3}", []>, TB;
1125 def SHLD16rri8 : Ii8<0xA4, MRMDestReg,
1126 (ops R16:$dst, R16:$src1, R16:$src2, i8imm:$src3),
1127 "shld{w} {$src3, $src2, $dst|$dst, $src2, $src3}", []>,
1129 def SHRD16rri8 : Ii8<0xAC, MRMDestReg,
1130 (ops R16:$dst, R16:$src1, R16:$src2, i8imm:$src3),
1131 "shrd{w} {$src3, $src2, $dst|$dst, $src2, $src3}", []>,
1135 let isTwoAddress = 0 in {
1136 def SHLD32mrCL : I<0xA5, MRMDestMem, (ops i32mem:$dst, R32:$src2),
1137 "shld{l} {%cl, $src2, $dst|$dst, $src2, %CL}", []>,
1139 def SHRD32mrCL : I<0xAD, MRMDestMem, (ops i32mem:$dst, R32:$src2),
1140 "shrd{l} {%cl, $src2, $dst|$dst, $src2, %CL}", []>,
1142 def SHLD32mri8 : Ii8<0xA4, MRMDestMem,
1143 (ops i32mem:$dst, R32:$src2, i8imm:$src3),
1144 "shld{l} {$src3, $src2, $dst|$dst, $src2, $src3}", []>,
1146 def SHRD32mri8 : Ii8<0xAC, MRMDestMem,
1147 (ops i32mem:$dst, R32:$src2, i8imm:$src3),
1148 "shrd{l} {$src3, $src2, $dst|$dst, $src2, $src3}", []>,
1151 def SHLD16mrCL : I<0xA5, MRMDestMem, (ops i16mem:$dst, R16:$src2),
1152 "shld{w} {%cl, $src2, $dst|$dst, $src2, %CL}", []>,
1153 Imp<[CL],[]>, TB, OpSize;
1154 def SHRD16mrCL : I<0xAD, MRMDestMem, (ops i16mem:$dst, R16:$src2),
1155 "shrd{w} {%cl, $src2, $dst|$dst, $src2, %CL}", []>,
1156 Imp<[CL],[]>, TB, OpSize;
1157 def SHLD16mri8 : Ii8<0xA4, MRMDestMem,
1158 (ops i16mem:$dst, R16:$src2, i8imm:$src3),
1159 "shld{w} {$src3, $src2, $dst|$dst, $src2, $src3}", []>,
1161 def SHRD16mri8 : Ii8<0xAC, MRMDestMem,
1162 (ops i16mem:$dst, R16:$src2, i8imm:$src3),
1163 "shrd{w} {$src3, $src2, $dst|$dst, $src2, $src3}", []>,
1169 let isCommutable = 1 in { // X = ADD Y, Z --> X = ADD Z, Y
1170 def ADD8rr : I<0x00, MRMDestReg, (ops R8 :$dst, R8 :$src1, R8 :$src2),
1171 "add{b} {$src2, $dst|$dst, $src2}",
1172 [(set R8:$dst, (add R8:$src1, R8:$src2))]>;
1173 let isConvertibleToThreeAddress = 1 in { // Can transform into LEA.
1174 def ADD16rr : I<0x01, MRMDestReg, (ops R16:$dst, R16:$src1, R16:$src2),
1175 "add{w} {$src2, $dst|$dst, $src2}",
1176 [(set R16:$dst, (add R16:$src1, R16:$src2))]>, OpSize;
1177 def ADD32rr : I<0x01, MRMDestReg, (ops R32:$dst, R32:$src1, R32:$src2),
1178 "add{l} {$src2, $dst|$dst, $src2}",
1179 [(set R32:$dst, (add R32:$src1, R32:$src2))]>;
1180 } // end isConvertibleToThreeAddress
1181 } // end isCommutable
1182 def ADD8rm : I<0x02, MRMSrcMem, (ops R8 :$dst, R8 :$src1, i8mem :$src2),
1183 "add{b} {$src2, $dst|$dst, $src2}",
1184 [(set R8:$dst, (add R8:$src1, (load addr:$src2)))]>;
1185 def ADD16rm : I<0x03, MRMSrcMem, (ops R16:$dst, R16:$src1, i16mem:$src2),
1186 "add{w} {$src2, $dst|$dst, $src2}",
1187 [(set R16:$dst, (add R16:$src1, (load addr:$src2)))]>, OpSize;
1188 def ADD32rm : I<0x03, MRMSrcMem, (ops R32:$dst, R32:$src1, i32mem:$src2),
1189 "add{l} {$src2, $dst|$dst, $src2}",
1190 [(set R32:$dst, (add R32:$src1, (load addr:$src2)))]>;
1192 def ADD8ri : Ii8<0x80, MRM0r, (ops R8:$dst, R8:$src1, i8imm:$src2),
1193 "add{b} {$src2, $dst|$dst, $src2}",
1194 [(set R8:$dst, (add R8:$src1, imm:$src2))]>;
1196 let isConvertibleToThreeAddress = 1 in { // Can transform into LEA.
1197 def ADD16ri : Ii16<0x81, MRM0r, (ops R16:$dst, R16:$src1, i16imm:$src2),
1198 "add{w} {$src2, $dst|$dst, $src2}",
1199 [(set R16:$dst, (add R16:$src1, imm:$src2))]>, OpSize;
1200 def ADD32ri : Ii32<0x81, MRM0r, (ops R32:$dst, R32:$src1, i32imm:$src2),
1201 "add{l} {$src2, $dst|$dst, $src2}",
1202 [(set R32:$dst, (add R32:$src1, imm:$src2))]>;
1205 // FIXME: move ADD16ri8 above ADD16ri to optimize for space.
1206 def ADD16ri8 : Ii8<0x83, MRM0r, (ops R16:$dst, R16:$src1, i16i8imm:$src2),
1207 "add{w} {$src2, $dst|$dst, $src2}",
1208 [(set R16:$dst, (add R16:$src1, i16SExt8:$src2))]>, OpSize;
1209 def ADD32ri8 : Ii8<0x83, MRM0r, (ops R32:$dst, R32:$src1, i32i8imm:$src2),
1210 "add{l} {$src2, $dst|$dst, $src2}",
1211 [(set R32:$dst, (add R32:$src1, i32SExt8:$src2))]>;
1213 let isTwoAddress = 0 in {
1214 def ADD8mr : I<0x00, MRMDestMem, (ops i8mem :$dst, R8 :$src2),
1215 "add{b} {$src2, $dst|$dst, $src2}",
1216 [(store (add (load addr:$dst), R8:$src2), addr:$dst)]>;
1217 def ADD16mr : I<0x01, MRMDestMem, (ops i16mem:$dst, R16:$src2),
1218 "add{w} {$src2, $dst|$dst, $src2}",
1219 [(store (add (load addr:$dst), R16:$src2), addr:$dst)]>, OpSize;
1220 def ADD32mr : I<0x01, MRMDestMem, (ops i32mem:$dst, R32:$src2),
1221 "add{l} {$src2, $dst|$dst, $src2}",
1222 [(store (add (load addr:$dst), R32:$src2), addr:$dst)]>;
1223 def ADD8mi : Ii8<0x80, MRM0m, (ops i8mem :$dst, i8imm :$src2),
1224 "add{b} {$src2, $dst|$dst, $src2}",
1225 [(store (add (load addr:$dst), (i8 imm:$src2)), addr:$dst)]>;
1226 def ADD16mi : Ii16<0x81, MRM0m, (ops i16mem:$dst, i16imm:$src2),
1227 "add{w} {$src2, $dst|$dst, $src2}",
1228 [(store (add (load addr:$dst), (i16 imm:$src2)), addr:$dst)]>, OpSize;
1229 def ADD32mi : Ii32<0x81, MRM0m, (ops i32mem:$dst, i32imm:$src2),
1230 "add{l} {$src2, $dst|$dst, $src2}",
1231 [(store (add (load addr:$dst), (i32 imm:$src2)), addr:$dst)]>;
1232 def ADD16mi8 : Ii8<0x83, MRM0m, (ops i16mem:$dst, i16i8imm :$src2),
1233 "add{w} {$src2, $dst|$dst, $src2}",
1234 [(store (add (load addr:$dst), i16SExt8:$src2), addr:$dst)]>, OpSize;
1235 def ADD32mi8 : Ii8<0x83, MRM0m, (ops i32mem:$dst, i32i8imm :$src2),
1236 "add{l} {$src2, $dst|$dst, $src2}",
1237 [(store (add (load addr:$dst), i32SExt8:$src2), addr:$dst)]>;
1240 let isCommutable = 1 in { // X = ADC Y, Z --> X = ADC Z, Y
1241 def ADC32rr : I<0x11, MRMDestReg, (ops R32:$dst, R32:$src1, R32:$src2),
1242 "adc{l} {$src2, $dst|$dst, $src2}", []>;
1244 def ADC32rm : I<0x13, MRMSrcMem , (ops R32:$dst, R32:$src1, i32mem:$src2),
1245 "adc{l} {$src2, $dst|$dst, $src2}", []>;
1246 def ADC32ri : Ii32<0x81, MRM2r, (ops R32:$dst, R32:$src1, i32imm:$src2),
1247 "adc{l} {$src2, $dst|$dst, $src2}", []>;
1248 def ADC32ri8 : Ii8<0x83, MRM2r, (ops R32:$dst, R32:$src1, i8imm:$src2),
1249 "adc{l} {$src2, $dst|$dst, $src2}", []>;
1251 let isTwoAddress = 0 in {
1252 def ADC32mr : I<0x11, MRMDestMem, (ops i32mem:$dst, R32:$src2),
1253 "adc{l} {$src2, $dst|$dst, $src2}", []>;
1254 def ADC32mi : Ii32<0x81, MRM2m, (ops i32mem:$dst, i32imm:$src2),
1255 "adc{l} {$src2, $dst|$dst, $src2}", []>;
1256 def ADC32mi8 : Ii8<0x83, MRM2m, (ops i32mem:$dst, i8imm :$src2),
1257 "adc{l} {$src2, $dst|$dst, $src2}", []>;
1260 def SUB8rr : I<0x28, MRMDestReg, (ops R8 :$dst, R8 :$src1, R8 :$src2),
1261 "sub{b} {$src2, $dst|$dst, $src2}",
1262 [(set R8:$dst, (sub R8:$src1, R8:$src2))]>;
1263 def SUB16rr : I<0x29, MRMDestReg, (ops R16:$dst, R16:$src1, R16:$src2),
1264 "sub{w} {$src2, $dst|$dst, $src2}",
1265 [(set R16:$dst, (sub R16:$src1, R16:$src2))]>, OpSize;
1266 def SUB32rr : I<0x29, MRMDestReg, (ops R32:$dst, R32:$src1, R32:$src2),
1267 "sub{l} {$src2, $dst|$dst, $src2}",
1268 [(set R32:$dst, (sub R32:$src1, R32:$src2))]>;
1269 def SUB8rm : I<0x2A, MRMSrcMem, (ops R8 :$dst, R8 :$src1, i8mem :$src2),
1270 "sub{b} {$src2, $dst|$dst, $src2}",
1271 [(set R8:$dst, (sub R8:$src1, (load addr:$src2)))]>;
1272 def SUB16rm : I<0x2B, MRMSrcMem, (ops R16:$dst, R16:$src1, i16mem:$src2),
1273 "sub{w} {$src2, $dst|$dst, $src2}",
1274 [(set R16:$dst, (sub R16:$src1, (load addr:$src2)))]>, OpSize;
1275 def SUB32rm : I<0x2B, MRMSrcMem, (ops R32:$dst, R32:$src1, i32mem:$src2),
1276 "sub{l} {$src2, $dst|$dst, $src2}",
1277 [(set R32:$dst, (sub R32:$src1, (load addr:$src2)))]>;
1279 def SUB8ri : Ii8 <0x80, MRM5r, (ops R8:$dst, R8:$src1, i8imm:$src2),
1280 "sub{b} {$src2, $dst|$dst, $src2}",
1281 [(set R8:$dst, (sub R8:$src1, imm:$src2))]>;
1282 def SUB16ri : Ii16<0x81, MRM5r, (ops R16:$dst, R16:$src1, i16imm:$src2),
1283 "sub{w} {$src2, $dst|$dst, $src2}",
1284 [(set R16:$dst, (sub R16:$src1, imm:$src2))]>, OpSize;
1285 def SUB32ri : Ii32<0x81, MRM5r, (ops R32:$dst, R32:$src1, i32imm:$src2),
1286 "sub{l} {$src2, $dst|$dst, $src2}",
1287 [(set R32:$dst, (sub R32:$src1, imm:$src2))]>;
1288 def SUB16ri8 : Ii8<0x83, MRM5r, (ops R16:$dst, R16:$src1, i16i8imm:$src2),
1289 "sub{w} {$src2, $dst|$dst, $src2}",
1290 [(set R16:$dst, (sub R16:$src1, i16SExt8:$src2))]>, OpSize;
1291 def SUB32ri8 : Ii8<0x83, MRM5r, (ops R32:$dst, R32:$src1, i32i8imm:$src2),
1292 "sub{l} {$src2, $dst|$dst, $src2}",
1293 [(set R32:$dst, (sub R32:$src1, i32SExt8:$src2))]>;
1294 let isTwoAddress = 0 in {
1295 def SUB8mr : I<0x28, MRMDestMem, (ops i8mem :$dst, R8 :$src2),
1296 "sub{b} {$src2, $dst|$dst, $src2}",
1297 [(store (sub (load addr:$dst), R8:$src2), addr:$dst)]>;
1298 def SUB16mr : I<0x29, MRMDestMem, (ops i16mem:$dst, R16:$src2),
1299 "sub{w} {$src2, $dst|$dst, $src2}",
1300 [(store (sub (load addr:$dst), R16:$src2), addr:$dst)]>, OpSize;
1301 def SUB32mr : I<0x29, MRMDestMem, (ops i32mem:$dst, R32:$src2),
1302 "sub{l} {$src2, $dst|$dst, $src2}",
1303 [(store (sub (load addr:$dst), R32:$src2), addr:$dst)]>;
1304 def SUB8mi : Ii8<0x80, MRM5m, (ops i8mem :$dst, i8imm:$src2),
1305 "sub{b} {$src2, $dst|$dst, $src2}",
1306 [(store (sub (load addr:$dst), (i8 imm:$src2)), addr:$dst)]>;
1307 def SUB16mi : Ii16<0x81, MRM5m, (ops i16mem:$dst, i16imm:$src2),
1308 "sub{w} {$src2, $dst|$dst, $src2}",
1309 [(store (sub (load addr:$dst), (i16 imm:$src2)), addr:$dst)]>, OpSize;
1310 def SUB32mi : Ii32<0x81, MRM5m, (ops i32mem:$dst, i32imm:$src2),
1311 "sub{l} {$src2, $dst|$dst, $src2}",
1312 [(store (sub (load addr:$dst), (i32 imm:$src2)), addr:$dst)]>;
1313 def SUB16mi8 : Ii8<0x83, MRM5m, (ops i16mem:$dst, i16i8imm :$src2),
1314 "sub{w} {$src2, $dst|$dst, $src2}",
1315 [(store (sub (load addr:$dst), i16SExt8:$src2), addr:$dst)]>, OpSize;
1316 def SUB32mi8 : Ii8<0x83, MRM5m, (ops i32mem:$dst, i32i8imm :$src2),
1317 "sub{l} {$src2, $dst|$dst, $src2}",
1318 [(store (sub (load addr:$dst), i32SExt8:$src2), addr:$dst)]>;
1321 def SBB32rr : I<0x19, MRMDestReg, (ops R32:$dst, R32:$src1, R32:$src2),
1322 "sbb{l} {$src2, $dst|$dst, $src2}", []>;
1324 let isTwoAddress = 0 in {
1325 def SBB32mr : I<0x19, MRMDestMem, (ops i32mem:$dst, R32:$src2),
1326 "sbb{l} {$src2, $dst|$dst, $src2}", []>;
1327 def SBB8mi : Ii32<0x80, MRM3m, (ops i8mem:$dst, i8imm:$src2),
1328 "sbb{b} {$src2, $dst|$dst, $src2}", []>;
1329 def SBB16mi : Ii32<0x81, MRM3m, (ops i16mem:$dst, i16imm:$src2),
1330 "sbb{w} {$src2, $dst|$dst, $src2}", []>, OpSize;
1331 def SBB32mi : Ii32<0x81, MRM3m, (ops i32mem:$dst, i32imm:$src2),
1332 "sbb{l} {$src2, $dst|$dst, $src2}", []>;
1333 def SBB16mi8 : Ii8<0x83, MRM3m, (ops i16mem:$dst, i8imm :$src2),
1334 "sbb{w} {$src2, $dst|$dst, $src2}", []>, OpSize;
1335 def SBB32mi8 : Ii8<0x83, MRM3m, (ops i32mem:$dst, i8imm :$src2),
1336 "sbb{l} {$src2, $dst|$dst, $src2}", []>;
1338 def SBB8ri : Ii8<0x80, MRM3r, (ops R8:$dst, R8:$src1, i8imm:$src2),
1339 "sbb{b} {$src2, $dst|$dst, $src2}", []>;
1340 def SBB16ri : Ii16<0x81, MRM3r, (ops R16:$dst, R16:$src1, i16imm:$src2),
1341 "sbb{w} {$src2, $dst|$dst, $src2}", []>, OpSize;
1343 def SBB32rm : I<0x1B, MRMSrcMem, (ops R32:$dst, R32:$src1, i32mem:$src2),
1344 "sbb{l} {$src2, $dst|$dst, $src2}", []>;
1345 def SBB32ri : Ii32<0x81, MRM3r, (ops R32:$dst, R32:$src1, i32imm:$src2),
1346 "sbb{l} {$src2, $dst|$dst, $src2}", []>;
1348 def SBB16ri8 : Ii8<0x83, MRM3r, (ops R16:$dst, R16:$src1, i8imm:$src2),
1349 "sbb{w} {$src2, $dst|$dst, $src2}", []>, OpSize;
1350 def SBB32ri8 : Ii8<0x83, MRM3r, (ops R32:$dst, R32:$src1, i8imm:$src2),
1351 "sbb{l} {$src2, $dst|$dst, $src2}", []>;
1353 let isCommutable = 1 in { // X = IMUL Y, Z --> X = IMUL Z, Y
1354 def IMUL16rr : I<0xAF, MRMSrcReg, (ops R16:$dst, R16:$src1, R16:$src2),
1355 "imul{w} {$src2, $dst|$dst, $src2}",
1356 [(set R16:$dst, (mul R16:$src1, R16:$src2))]>, TB, OpSize;
1357 def IMUL32rr : I<0xAF, MRMSrcReg, (ops R32:$dst, R32:$src1, R32:$src2),
1358 "imul{l} {$src2, $dst|$dst, $src2}",
1359 [(set R32:$dst, (mul R32:$src1, R32:$src2))]>, TB;
1361 def IMUL16rm : I<0xAF, MRMSrcMem, (ops R16:$dst, R16:$src1, i16mem:$src2),
1362 "imul{w} {$src2, $dst|$dst, $src2}",
1363 [(set R16:$dst, (mul R16:$src1, (load addr:$src2)))]>, TB, OpSize;
1364 def IMUL32rm : I<0xAF, MRMSrcMem, (ops R32:$dst, R32:$src1, i32mem:$src2),
1365 "imul{l} {$src2, $dst|$dst, $src2}",
1366 [(set R32:$dst, (mul R32:$src1, (load addr:$src2)))]>, TB;
1368 } // end Two Address instructions
1370 // Suprisingly enough, these are not two address instructions!
1371 def IMUL16rri : Ii16<0x69, MRMSrcReg, // R16 = R16*I16
1372 (ops R16:$dst, R16:$src1, i16imm:$src2),
1373 "imul{w} {$src2, $src1, $dst|$dst, $src1, $src2}",
1374 [(set R16:$dst, (mul R16:$src1, imm:$src2))]>, OpSize;
1375 def IMUL32rri : Ii32<0x69, MRMSrcReg, // R32 = R32*I32
1376 (ops R32:$dst, R32:$src1, i32imm:$src2),
1377 "imul{l} {$src2, $src1, $dst|$dst, $src1, $src2}",
1378 [(set R32:$dst, (mul R32:$src1, imm:$src2))]>;
1379 def IMUL16rri8 : Ii8<0x6B, MRMSrcReg, // R16 = R16*I8
1380 (ops R16:$dst, R16:$src1, i16i8imm:$src2),
1381 "imul{w} {$src2, $src1, $dst|$dst, $src1, $src2}",
1382 [(set R16:$dst, (mul R16:$src1, i16SExt8:$src2))]>, OpSize;
1383 def IMUL32rri8 : Ii8<0x6B, MRMSrcReg, // R32 = R32*I8
1384 (ops R32:$dst, R32:$src1, i32i8imm:$src2),
1385 "imul{l} {$src2, $src1, $dst|$dst, $src1, $src2}",
1386 [(set R32:$dst, (mul R32:$src1, i32SExt8:$src2))]>;
1388 def IMUL16rmi : Ii16<0x69, MRMSrcMem, // R16 = [mem16]*I16
1389 (ops R16:$dst, i16mem:$src1, i16imm:$src2),
1390 "imul{w} {$src2, $src1, $dst|$dst, $src1, $src2}",
1391 [(set R16:$dst, (mul (load addr:$src1), imm:$src2))]>,
1393 def IMUL32rmi : Ii32<0x69, MRMSrcMem, // R32 = [mem32]*I32
1394 (ops R32:$dst, i32mem:$src1, i32imm:$src2),
1395 "imul{l} {$src2, $src1, $dst|$dst, $src1, $src2}",
1396 [(set R32:$dst, (mul (load addr:$src1), imm:$src2))]>;
1397 def IMUL16rmi8 : Ii8<0x6B, MRMSrcMem, // R16 = [mem16]*I8
1398 (ops R16:$dst, i16mem:$src1, i16i8imm :$src2),
1399 "imul{w} {$src2, $src1, $dst|$dst, $src1, $src2}",
1400 [(set R16:$dst, (mul (load addr:$src1), i16SExt8:$src2))]>, OpSize;
1401 def IMUL32rmi8 : Ii8<0x6B, MRMSrcMem, // R32 = [mem32]*I8
1402 (ops R32:$dst, i32mem:$src1, i32i8imm: $src2),
1403 "imul{l} {$src2, $src1, $dst|$dst, $src1, $src2}",
1404 [(set R32:$dst, (mul (load addr:$src1), i32SExt8:$src2))]>;
1406 //===----------------------------------------------------------------------===//
1407 // Test instructions are just like AND, except they don't generate a result.
1409 let isCommutable = 1 in { // TEST X, Y --> TEST Y, X
1410 def TEST8rr : I<0x84, MRMDestReg, (ops R8:$src1, R8:$src2),
1411 "test{b} {$src2, $src1|$src1, $src2}", []>;
1412 def TEST16rr : I<0x85, MRMDestReg, (ops R16:$src1, R16:$src2),
1413 "test{w} {$src2, $src1|$src1, $src2}", []>, OpSize;
1414 def TEST32rr : I<0x85, MRMDestReg, (ops R32:$src1, R32:$src2),
1415 "test{l} {$src2, $src1|$src1, $src2}", []>;
1417 def TEST8mr : I<0x84, MRMDestMem, (ops i8mem :$src1, R8 :$src2),
1418 "test{b} {$src2, $src1|$src1, $src2}", []>;
1419 def TEST16mr : I<0x85, MRMDestMem, (ops i16mem:$src1, R16:$src2),
1420 "test{w} {$src2, $src1|$src1, $src2}", []>, OpSize;
1421 def TEST32mr : I<0x85, MRMDestMem, (ops i32mem:$src1, R32:$src2),
1422 "test{l} {$src2, $src1|$src1, $src2}", []>;
1423 def TEST8rm : I<0x84, MRMSrcMem, (ops R8 :$src1, i8mem :$src2),
1424 "test{b} {$src2, $src1|$src1, $src2}", []>;
1425 def TEST16rm : I<0x85, MRMSrcMem, (ops R16:$src1, i16mem:$src2),
1426 "test{w} {$src2, $src1|$src1, $src2}", []>, OpSize;
1427 def TEST32rm : I<0x85, MRMSrcMem, (ops R32:$src1, i32mem:$src2),
1428 "test{l} {$src2, $src1|$src1, $src2}", []>;
1430 def TEST8ri : Ii8 <0xF6, MRM0r, // flags = R8 & imm8
1431 (ops R8:$src1, i8imm:$src2),
1432 "test{b} {$src2, $src1|$src1, $src2}", []>;
1433 def TEST16ri : Ii16<0xF7, MRM0r, // flags = R16 & imm16
1434 (ops R16:$src1, i16imm:$src2),
1435 "test{w} {$src2, $src1|$src1, $src2}", []>, OpSize;
1436 def TEST32ri : Ii32<0xF7, MRM0r, // flags = R32 & imm32
1437 (ops R32:$src1, i32imm:$src2),
1438 "test{l} {$src2, $src1|$src1, $src2}", []>;
1439 def TEST8mi : Ii8 <0xF6, MRM0m, // flags = [mem8] & imm8
1440 (ops i32mem:$src1, i8imm:$src2),
1441 "test{b} {$src2, $src1|$src1, $src2}", []>;
1442 def TEST16mi : Ii16<0xF7, MRM0m, // flags = [mem16] & imm16
1443 (ops i16mem:$src1, i16imm:$src2),
1444 "test{w} {$src2, $src1|$src1, $src2}", []>, OpSize;
1445 def TEST32mi : Ii32<0xF7, MRM0m, // flags = [mem32] & imm32
1446 (ops i32mem:$src1, i32imm:$src2),
1447 "test{l} {$src2, $src1|$src1, $src2}", []>;
1451 // Condition code ops, incl. set if equal/not equal/...
1452 def SAHF : I<0x9E, RawFrm, (ops), "sahf", []>, Imp<[AH],[]>; // flags = AH
1453 def LAHF : I<0x9F, RawFrm, (ops), "lahf", []>, Imp<[],[AH]>; // AH = flags
1455 def SETBr : I<0x92, MRM0r,
1456 (ops R8 :$dst), "setb $dst", []>, TB; // R8 = < unsign
1457 def SETBm : I<0x92, MRM0m,
1458 (ops i8mem:$dst), "setb $dst", []>, TB; // [mem8] = < unsign
1459 def SETAEr : I<0x93, MRM0r,
1460 (ops R8 :$dst), "setae $dst", []>, TB; // R8 = >= unsign
1461 def SETAEm : I<0x93, MRM0m,
1462 (ops i8mem:$dst), "setae $dst", []>, TB; // [mem8] = >= unsign
1463 def SETEr : I<0x94, MRM0r,
1464 (ops R8 :$dst), "sete $dst", []>, TB; // R8 = ==
1465 def SETEm : I<0x94, MRM0m,
1466 (ops i8mem:$dst), "sete $dst", []>, TB; // [mem8] = ==
1467 def SETNEr : I<0x95, MRM0r,
1468 (ops R8 :$dst), "setne $dst", []>, TB; // R8 = !=
1469 def SETNEm : I<0x95, MRM0m,
1470 (ops i8mem:$dst), "setne $dst", []>, TB; // [mem8] = !=
1471 def SETBEr : I<0x96, MRM0r,
1472 (ops R8 :$dst), "setbe $dst", []>, TB; // R8 = <= unsign
1473 def SETBEm : I<0x96, MRM0m,
1474 (ops i8mem:$dst), "setbe $dst", []>, TB; // [mem8] = <= unsign
1475 def SETAr : I<0x97, MRM0r,
1476 (ops R8 :$dst), "seta $dst", []>, TB; // R8 = > signed
1477 def SETAm : I<0x97, MRM0m,
1478 (ops i8mem:$dst), "seta $dst", []>, TB; // [mem8] = > signed
1479 def SETSr : I<0x98, MRM0r,
1480 (ops R8 :$dst), "sets $dst", []>, TB; // R8 = <sign bit>
1481 def SETSm : I<0x98, MRM0m,
1482 (ops i8mem:$dst), "sets $dst", []>, TB; // [mem8] = <sign bit>
1483 def SETNSr : I<0x99, MRM0r,
1484 (ops R8 :$dst), "setns $dst", []>, TB; // R8 = !<sign bit>
1485 def SETNSm : I<0x99, MRM0m,
1486 (ops i8mem:$dst), "setns $dst", []>, TB; // [mem8] = !<sign bit>
1487 def SETPr : I<0x9A, MRM0r,
1488 (ops R8 :$dst), "setp $dst", []>, TB; // R8 = parity
1489 def SETPm : I<0x9A, MRM0m,
1490 (ops i8mem:$dst), "setp $dst", []>, TB; // [mem8] = parity
1491 def SETNPr : I<0x9B, MRM0r,
1492 (ops R8 :$dst), "setnp $dst", []>, TB; // R8 = not parity
1493 def SETNPm : I<0x9B, MRM0m,
1494 (ops i8mem:$dst), "setnp $dst", []>, TB; // [mem8] = not parity
1495 def SETLr : I<0x9C, MRM0r,
1496 (ops R8 :$dst), "setl $dst", []>, TB; // R8 = < signed
1497 def SETLm : I<0x9C, MRM0m,
1498 (ops i8mem:$dst), "setl $dst", []>, TB; // [mem8] = < signed
1499 def SETGEr : I<0x9D, MRM0r,
1500 (ops R8 :$dst), "setge $dst", []>, TB; // R8 = >= signed
1501 def SETGEm : I<0x9D, MRM0m,
1502 (ops i8mem:$dst), "setge $dst", []>, TB; // [mem8] = >= signed
1503 def SETLEr : I<0x9E, MRM0r,
1504 (ops R8 :$dst), "setle $dst", []>, TB; // R8 = <= signed
1505 def SETLEm : I<0x9E, MRM0m,
1506 (ops i8mem:$dst), "setle $dst", []>, TB; // [mem8] = <= signed
1507 def SETGr : I<0x9F, MRM0r,
1508 (ops R8 :$dst), "setg $dst", []>, TB; // R8 = < signed
1509 def SETGm : I<0x9F, MRM0m,
1510 (ops i8mem:$dst), "setg $dst", []>, TB; // [mem8] = < signed
1512 // Integer comparisons
1513 def CMP8rr : I<0x38, MRMDestReg,
1514 (ops R8 :$src1, R8 :$src2),
1515 "cmp{b} {$src2, $src1|$src1, $src2}", []>;
1516 def CMP16rr : I<0x39, MRMDestReg,
1517 (ops R16:$src1, R16:$src2),
1518 "cmp{w} {$src2, $src1|$src1, $src2}", []>, OpSize;
1519 def CMP32rr : I<0x39, MRMDestReg,
1520 (ops R32:$src1, R32:$src2),
1521 "cmp{l} {$src2, $src1|$src1, $src2}", []>;
1522 def CMP8mr : I<0x38, MRMDestMem,
1523 (ops i8mem :$src1, R8 :$src2),
1524 "cmp{b} {$src2, $src1|$src1, $src2}", []>;
1525 def CMP16mr : I<0x39, MRMDestMem,
1526 (ops i16mem:$src1, R16:$src2),
1527 "cmp{w} {$src2, $src1|$src1, $src2}", []>, OpSize;
1528 def CMP32mr : I<0x39, MRMDestMem,
1529 (ops i32mem:$src1, R32:$src2),
1530 "cmp{l} {$src2, $src1|$src1, $src2}", []>;
1531 def CMP8rm : I<0x3A, MRMSrcMem,
1532 (ops R8 :$src1, i8mem :$src2),
1533 "cmp{b} {$src2, $src1|$src1, $src2}", []>;
1534 def CMP16rm : I<0x3B, MRMSrcMem,
1535 (ops R16:$src1, i16mem:$src2),
1536 "cmp{w} {$src2, $src1|$src1, $src2}", []>, OpSize;
1537 def CMP32rm : I<0x3B, MRMSrcMem,
1538 (ops R32:$src1, i32mem:$src2),
1539 "cmp{l} {$src2, $src1|$src1, $src2}", []>;
1540 def CMP8ri : Ii8<0x80, MRM7r,
1541 (ops R16:$src1, i8imm:$src2),
1542 "cmp{b} {$src2, $src1|$src1, $src2}", []>;
1543 def CMP16ri : Ii16<0x81, MRM7r,
1544 (ops R16:$src1, i16imm:$src2),
1545 "cmp{w} {$src2, $src1|$src1, $src2}", []>, OpSize;
1546 def CMP32ri : Ii32<0x81, MRM7r,
1547 (ops R32:$src1, i32imm:$src2),
1548 "cmp{l} {$src2, $src1|$src1, $src2}", []>;
1549 def CMP8mi : Ii8 <0x80, MRM7m,
1550 (ops i8mem :$src1, i8imm :$src2),
1551 "cmp{b} {$src2, $src1|$src1, $src2}", []>;
1552 def CMP16mi : Ii16<0x81, MRM7m,
1553 (ops i16mem:$src1, i16imm:$src2),
1554 "cmp{w} {$src2, $src1|$src1, $src2}", []>, OpSize;
1555 def CMP32mi : Ii32<0x81, MRM7m,
1556 (ops i32mem:$src1, i32imm:$src2),
1557 "cmp{l} {$src2, $src1|$src1, $src2}", []>;
1559 // Sign/Zero extenders
1560 def MOVSX16rr8 : I<0xBE, MRMSrcReg, (ops R16:$dst, R8 :$src),
1561 "movs{bw|x} {$src, $dst|$dst, $src}",
1562 [(set R16:$dst, (sext R8:$src))]>, TB, OpSize;
1563 def MOVSX16rm8 : I<0xBE, MRMSrcMem, (ops R16:$dst, i8mem :$src),
1564 "movs{bw|x} {$src, $dst|$dst, $src}", []>, TB, OpSize;
1565 def MOVSX32rr8 : I<0xBE, MRMSrcReg, (ops R32:$dst, R8 :$src),
1566 "movs{bl|x} {$src, $dst|$dst, $src}",
1567 [(set R32:$dst, (sext R8:$src))]>, TB;
1568 def MOVSX32rm8 : I<0xBE, MRMSrcMem, (ops R32:$dst, i8mem :$src),
1569 "movs{bl|x} {$src, $dst|$dst, $src}", []>, TB;
1570 def MOVSX32rr16: I<0xBF, MRMSrcReg, (ops R32:$dst, R16:$src),
1571 "movs{wl|x} {$src, $dst|$dst, $src}",
1572 [(set R32:$dst, (sext R16:$src))]>, TB;
1573 def MOVSX32rm16: I<0xBF, MRMSrcMem, (ops R32:$dst, i16mem:$src),
1574 "movs{wl|x} {$src, $dst|$dst, $src}", []>, TB;
1576 def MOVZX16rr8 : I<0xB6, MRMSrcReg, (ops R16:$dst, R8 :$src),
1577 "movz{bw|x} {$src, $dst|$dst, $src}",
1578 [(set R16:$dst, (zext R8:$src))]>, TB, OpSize;
1579 def MOVZX16rm8 : I<0xB6, MRMSrcMem, (ops R16:$dst, i8mem :$src),
1580 "movz{bw|x} {$src, $dst|$dst, $src}", []>, TB, OpSize;
1581 def MOVZX32rr8 : I<0xB6, MRMSrcReg, (ops R32:$dst, R8 :$src),
1582 "movz{bl|x} {$src, $dst|$dst, $src}",
1583 [(set R32:$dst, (zext R8:$src))]>, TB;
1584 def MOVZX32rm8 : I<0xB6, MRMSrcMem, (ops R32:$dst, i8mem :$src),
1585 "movz{bl|x} {$src, $dst|$dst, $src}", []>, TB;
1586 def MOVZX32rr16: I<0xB7, MRMSrcReg, (ops R32:$dst, R16:$src),
1587 "movz{wl|x} {$src, $dst|$dst, $src}",
1588 [(set R32:$dst, (zext R16:$src))]>, TB;
1589 def MOVZX32rm16: I<0xB7, MRMSrcMem, (ops R32:$dst, i16mem:$src),
1590 "movz{wl|x} {$src, $dst|$dst, $src}", []>, TB;
1592 //===----------------------------------------------------------------------===//
1593 // XMM Floating point support (requires SSE2)
1594 //===----------------------------------------------------------------------===//
1596 def MOVSSrr : I<0x10, MRMSrcReg, (ops V4F4:$dst, V4F4:$src),
1597 "movss {$src, $dst|$dst, $src}", []>, XS;
1598 def MOVSSrm : I<0x10, MRMSrcMem, (ops V4F4:$dst, f32mem:$src),
1599 "movss {$src, $dst|$dst, $src}", []>, XS;
1600 def MOVSSmr : I<0x11, MRMDestMem, (ops f32mem:$dst, V4F4:$src),
1601 "movss {$src, $dst|$dst, $src}", []>, XS;
1602 def MOVSDrr : I<0x10, MRMSrcReg, (ops V2F8:$dst, V2F8:$src),
1603 "movsd {$src, $dst|$dst, $src}", []>, XD;
1604 def MOVSDrm : I<0x10, MRMSrcMem, (ops V2F8:$dst, f64mem:$src),
1605 "movsd {$src, $dst|$dst, $src}", []>, XD;
1606 def MOVSDmr : I<0x11, MRMDestMem, (ops f64mem:$dst, V2F8:$src),
1607 "movsd {$src, $dst|$dst, $src}", []>, XD;
1609 def CVTTSD2SIrr: I<0x2C, MRMSrcReg, (ops R32:$dst, V2F8:$src),
1610 "cvttsd2si {$src, $dst|$dst, $src}",
1611 [(set R32:$dst, (fp_to_sint V2F8:$src))]>, XD;
1612 def CVTTSD2SIrm: I<0x2C, MRMSrcMem, (ops R32:$dst, f64mem:$src),
1613 "cvttsd2si {$src, $dst|$dst, $src}", []>, XD;
1614 def CVTTSS2SIrr: I<0x2C, MRMSrcReg, (ops R32:$dst, V4F4:$src),
1615 "cvttss2si {$src, $dst|$dst, $src}",
1616 [(set R32:$dst, (fp_to_sint V4F4:$src))]>, XS;
1617 def CVTTSS2SIrm: I<0x2C, MRMSrcMem, (ops R32:$dst, f32mem:$src),
1618 "cvttss2si {$src, $dst|$dst, $src}", []>, XS;
1619 def CVTSD2SSrr: I<0x5A, MRMSrcReg, (ops V4F4:$dst, V2F8:$src),
1620 "cvtsd2ss {$src, $dst|$dst, $src}",
1621 [(set V4F4:$dst, (fround V2F8:$src))]>, XS;
1622 def CVTSD2SSrm: I<0x5A, MRMSrcMem, (ops V4F4:$dst, f64mem:$src),
1623 "cvtsd2ss {$src, $dst|$dst, $src}", []>, XS;
1624 def CVTSS2SDrr: I<0x5A, MRMSrcReg, (ops V2F8:$dst, V4F4:$src),
1625 "cvtss2sd {$src, $dst|$dst, $src}",
1626 [(set V2F8:$dst, (fextend V4F4:$src))]>, XD;
1627 def CVTSS2SDrm: I<0x5A, MRMSrcMem, (ops V2F8:$dst, f32mem:$src),
1628 "cvtss2sd {$src, $dst|$dst, $src}", []>, XD;
1629 def CVTSI2SSrr: I<0x2A, MRMSrcReg, (ops V4F4:$dst, R32:$src),
1630 "cvtsi2ss {$src, $dst|$dst, $src}",
1631 [(set V4F4:$dst, (sint_to_fp R32:$src))]>, XS;
1632 def CVTSI2SSrm: I<0x2A, MRMSrcMem, (ops V4F4:$dst, i32mem:$src),
1633 "cvtsi2ss {$src, $dst|$dst, $src}", []>, XS;
1634 def CVTSI2SDrr: I<0x2A, MRMSrcReg, (ops V2F8:$dst, R32:$src),
1635 "cvtsi2sd {$src, $dst|$dst, $src}",
1636 [(set V2F8:$dst, (sint_to_fp R32:$src))]>, XD;
1637 def CVTSI2SDrm: I<0x2A, MRMSrcMem, (ops V2F8:$dst, i32mem:$src),
1638 "cvtsi2sd {$src, $dst|$dst, $src}", []>, XD;
1640 def SQRTSSrm : I<0x51, MRMSrcMem, (ops V4F4:$dst, f32mem:$src),
1641 "sqrtss {$src, $dst|$dst, $src}", []>, XS;
1642 def SQRTSSrr : I<0x51, MRMSrcReg, (ops V4F4:$dst, V4F4:$src),
1643 "sqrtss {$src, $dst|$dst, $src}",
1644 [(set V4F4:$dst, (fsqrt V4F4:$src))]>, XS;
1645 def SQRTSDrm : I<0x51, MRMSrcMem, (ops V2F8:$dst, f64mem:$src),
1646 "sqrtsd {$src, $dst|$dst, $src}", []>, XD;
1647 def SQRTSDrr : I<0x51, MRMSrcReg, (ops V2F8:$dst, V2F8:$src),
1648 "sqrtsd {$src, $dst|$dst, $src}",
1649 [(set V2F8:$dst, (fsqrt V2F8:$src))]>, XD;
1651 def UCOMISDrr: I<0x2E, MRMSrcReg, (ops V2F8:$dst, V2F8:$src),
1652 "ucomisd {$src, $dst|$dst, $src}", []>, TB, OpSize;
1653 def UCOMISDrm: I<0x2E, MRMSrcMem, (ops V2F8:$dst, f64mem:$src),
1654 "ucomisd {$src, $dst|$dst, $src}", []>, TB, OpSize;
1655 def UCOMISSrr: I<0x2E, MRMSrcReg, (ops V4F4:$dst, V4F4:$src),
1656 "ucomiss {$src, $dst|$dst, $src}", []>, TB;
1657 def UCOMISSrm: I<0x2E, MRMSrcMem, (ops V4F4:$dst, f32mem:$src),
1658 "ucomiss {$src, $dst|$dst, $src}", []>, TB;
1660 // Pseudo-instructions that map fld0 to xorps/xorpd for sse.
1661 // FIXME: remove when we can teach regalloc that xor reg, reg is ok.
1662 def FLD0SS : I<0x57, MRMSrcReg, (ops V4F4:$dst),
1663 "xorps $dst, $dst", []>, TB;
1664 def FLD0SD : I<0x57, MRMSrcReg, (ops V2F8:$dst),
1665 "xorpd $dst, $dst", []>, TB, OpSize;
1667 let isTwoAddress = 1 in {
1668 let isCommutable = 1 in {
1669 def ADDSSrr : I<0x58, MRMSrcReg, (ops V4F4:$dst, V4F4:$src1, V4F4:$src2),
1670 "addss {$src2, $dst|$dst, $src2}",
1671 [(set V4F4:$dst, (fadd V4F4:$src1, V4F4:$src2))]>, XS;
1672 def ADDSDrr : I<0x58, MRMSrcReg, (ops V2F8:$dst, V2F8:$src1, V2F8:$src2),
1673 "addsd {$src2, $dst|$dst, $src2}",
1674 [(set V2F8:$dst, (fadd V2F8:$src1, V2F8:$src2))]>, XD;
1675 def ANDPSrr : I<0x54, MRMSrcReg, (ops V4F4:$dst, V4F4:$src1, V4F4:$src2),
1676 "andps {$src2, $dst|$dst, $src2}", []>, TB;
1677 def ANDPDrr : I<0x54, MRMSrcReg, (ops V2F8:$dst, V2F8:$src1, V2F8:$src2),
1678 "andpd {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
1679 def MULSSrr : I<0x59, MRMSrcReg, (ops V4F4:$dst, V4F4:$src1, V4F4:$src2),
1680 "mulss {$src2, $dst|$dst, $src2}",
1681 [(set V4F4:$dst, (fmul V4F4:$src1, V4F4:$src2))]>, XS;
1682 def MULSDrr : I<0x59, MRMSrcReg, (ops V2F8:$dst, V2F8:$src1, V2F8:$src2),
1683 "mulsd {$src2, $dst|$dst, $src2}",
1684 [(set V2F8:$dst, (fmul V2F8:$src1, V2F8:$src2))]>, XD;
1685 def ORPSrr : I<0x56, MRMSrcReg, (ops V4F4:$dst, V4F4:$src1, V4F4:$src2),
1686 "orps {$src2, $dst|$dst, $src2}", []>, TB;
1687 def ORPDrr : I<0x56, MRMSrcReg, (ops V2F8:$dst, V2F8:$src1, V2F8:$src2),
1688 "orpd {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
1689 def XORPSrr : I<0x57, MRMSrcReg, (ops V4F4:$dst, V4F4:$src1, V4F4:$src2),
1690 "xorps {$src2, $dst|$dst, $src2}", []>, TB;
1691 def XORPDrr : I<0x57, MRMSrcReg, (ops V2F8:$dst, V2F8:$src1, V2F8:$src2),
1692 "xorpd {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
1694 def ANDNPSrr : I<0x55, MRMSrcReg, (ops V4F4:$dst, V4F4:$src1, V4F4:$src2),
1695 "andnps {$src2, $dst|$dst, $src2}", []>, TB;
1696 def ANDNPDrr : I<0x55, MRMSrcReg, (ops V2F8:$dst, V2F8:$src1, V2F8:$src2),
1697 "andnpd {$src2, $dst|$dst, $src2}", []>, TB, OpSize;
1698 def ADDSSrm : I<0x58, MRMSrcMem, (ops V4F4:$dst, V4F4:$src1, f32mem:$src2),
1699 "addss {$src2, $dst|$dst, $src2}", []>, XS;
1700 def ADDSDrm : I<0x58, MRMSrcMem, (ops V2F8:$dst, V2F8:$src1, f64mem:$src2),
1701 "addsd {$src2, $dst|$dst, $src2}", []>, XD;
1702 def MULSSrm : I<0x59, MRMSrcMem, (ops V4F4:$dst, V4F4:$src1, f32mem:$src2),
1703 "mulss {$src2, $dst|$dst, $src2}", []>, XS;
1704 def MULSDrm : I<0x59, MRMSrcMem, (ops V2F8:$dst, V2F8:$src1, f64mem:$src2),
1705 "mulsd {$src2, $dst|$dst, $src2}", []>, XD;
1707 def DIVSSrm : I<0x5E, MRMSrcMem, (ops V4F4:$dst, V4F4:$src1, f32mem:$src2),
1708 "divss {$src2, $dst|$dst, $src2}", []>, XS;
1709 def DIVSSrr : I<0x5E, MRMSrcReg, (ops V4F4:$dst, V4F4:$src1, V4F4:$src2),
1710 "divss {$src2, $dst|$dst, $src2}",
1711 [(set V4F4:$dst, (fdiv V4F4:$src1, V4F4:$src2))]>, XS;
1712 def DIVSDrm : I<0x5E, MRMSrcMem, (ops V2F8:$dst, V2F8:$src1, f64mem:$src2),
1713 "divsd {$src2, $dst|$dst, $src2}", []>, XD;
1714 def DIVSDrr : I<0x5E, MRMSrcReg, (ops V2F8:$dst, V2F8:$src1, V2F8:$src2),
1715 "divsd {$src2, $dst|$dst, $src2}",
1716 [(set V2F8:$dst, (fdiv V2F8:$src1, V2F8:$src2))]>, XD;
1718 def SUBSSrm : I<0x5C, MRMSrcMem, (ops V4F4:$dst, V4F4:$src1, f32mem:$src2),
1719 "subss {$src2, $dst|$dst, $src2}", []>, XS;
1720 def SUBSSrr : I<0x5C, MRMSrcReg, (ops V4F4:$dst, V4F4:$src1, V4F4:$src2),
1721 "subss {$src2, $dst|$dst, $src2}",
1722 [(set V4F4:$dst, (fsub V4F4:$src1, V4F4:$src2))]>, XS;
1723 def SUBSDrm : I<0x5C, MRMSrcMem, (ops V2F8:$dst, V2F8:$src1, f64mem:$src2),
1724 "subsd {$src2, $dst|$dst, $src2}", []>, XD;
1725 def SUBSDrr : I<0x5C, MRMSrcReg, (ops V2F8:$dst, V2F8:$src1, V2F8:$src2),
1726 "subsd {$src2, $dst|$dst, $src2}",
1727 [(set V2F8:$dst, (fsub V2F8:$src1, V2F8:$src2))]>, XD;
1729 def CMPSSrr : I<0xC2, MRMSrcReg,
1730 (ops V4F4:$dst, V4F4:$src1, V4F4:$src, SSECC:$cc),
1731 "cmp${cc}ss {$src, $dst|$dst, $src}", []>, XS;
1732 def CMPSSrm : I<0xC2, MRMSrcMem,
1733 (ops V4F4:$dst, V4F4:$src1, f32mem:$src, SSECC:$cc),
1734 "cmp${cc}ss {$src, $dst|$dst, $src}", []>, XS;
1735 def CMPSDrr : I<0xC2, MRMSrcReg,
1736 (ops V2F8:$dst, V2F8:$src1, V2F8:$src, SSECC:$cc),
1737 "cmp${cc}sd {$src, $dst|$dst, $src}", []>, XD;
1738 def CMPSDrm : I<0xC2, MRMSrcMem,
1739 (ops V2F8:$dst, V2F8:$src1, f64mem:$src, SSECC:$cc),
1740 "cmp${cc}sd {$src, $dst|$dst, $src}", []>, XD;
1743 //===----------------------------------------------------------------------===//
1744 // Miscellaneous Instructions
1745 //===----------------------------------------------------------------------===//
1747 def RDTSC : I<0x31, RawFrm, (ops), "rdtsc", []>, TB, Imp<[],[EAX,EDX]>;
1750 //===----------------------------------------------------------------------===//
1751 // Stack-based Floating point support
1752 //===----------------------------------------------------------------------===//
1754 // FIXME: These need to indicate mod/ref sets for FP regs... & FP 'TOP'
1756 // Floating point instruction template
1757 class FPI<bits<8> o, Format F, FPFormat fp, dag ops, string asm>
1758 : X86Inst<o, F, NoImm, ops, asm> {
1759 let FPForm = fp; let FPFormBits = FPForm.Value;
1762 // Pseudo instructions for floating point. We use these pseudo instructions
1763 // because they can be expanded by the fp spackifier into one of many different
1764 // forms of instructions for doing these operations. Until the stackifier runs,
1765 // we prefer to be abstract.
1766 def FpMOV : FPI<0, Pseudo, SpecialFP,
1767 (ops RFP:$dst, RFP:$src), "">; // f1 = fmov f2
1768 def FpADD : FPI<0, Pseudo, TwoArgFP ,
1769 (ops RFP:$dst, RFP:$src1, RFP:$src2), "">; // f1 = fadd f2, f3
1770 def FpSUB : FPI<0, Pseudo, TwoArgFP ,
1771 (ops RFP:$dst, RFP:$src1, RFP:$src2), "">; // f1 = fsub f2, f3
1772 def FpMUL : FPI<0, Pseudo, TwoArgFP ,
1773 (ops RFP:$dst, RFP:$src1, RFP:$src2), "">; // f1 = fmul f2, f3
1774 def FpDIV : FPI<0, Pseudo, TwoArgFP ,
1775 (ops RFP:$dst, RFP:$src1, RFP:$src2), "">; // f1 = fdiv f2, f3
1777 def FpGETRESULT : FPI<0, Pseudo, SpecialFP, (ops RFP:$dst), "">,
1778 Imp<[ST0], []>; // FPR = ST(0)
1780 def FpSETRESULT : FPI<0, Pseudo, SpecialFP, (ops RFP:$src), "">,
1781 Imp<[], [ST0]>; // ST(0) = FPR
1783 // FADD reg, mem: Before stackification, these are represented by:
1784 // R1 = FADD* R2, [mem]
1785 def FADD32m : FPI<0xD8, MRM0m, OneArgFPRW, // ST(0) = ST(0) + [mem32real]
1786 (ops f32mem:$src, variable_ops),
1788 def FADD64m : FPI<0xDC, MRM0m, OneArgFPRW, // ST(0) = ST(0) + [mem64real]
1789 (ops f64mem:$src, variable_ops),
1791 //def FIADD16m : FPI<0xDE, MRM0m, OneArgFPRW>; // ST(0) = ST(0) + [mem16int]
1792 //def FIADD32m : FPI<0xDA, MRM0m, OneArgFPRW>; // ST(0) = ST(0) + [mem32int]
1794 // FMUL reg, mem: Before stackification, these are represented by:
1795 // R1 = FMUL* R2, [mem]
1796 def FMUL32m : FPI<0xD8, MRM1m, OneArgFPRW, // ST(0) = ST(0) * [mem32real]
1797 (ops f32mem:$src, variable_ops),
1799 def FMUL64m : FPI<0xDC, MRM1m, OneArgFPRW, // ST(0) = ST(0) * [mem64real]
1800 (ops f64mem:$src, variable_ops),
1802 // ST(0) = ST(0) * [mem16int]
1803 //def FIMUL16m : FPI16m<"fimul", 0xDE, MRM1m, OneArgFPRW>;
1804 // ST(0) = ST(0) * [mem32int]
1805 //def FIMUL32m : FPI32m<"fimul", 0xDA, MRM1m, OneArgFPRW>;
1807 // FSUB reg, mem: Before stackification, these are represented by:
1808 // R1 = FSUB* R2, [mem]
1809 def FSUB32m : FPI<0xD8, MRM4m, OneArgFPRW, // ST(0) = ST(0) - [mem32real]
1810 (ops f32mem:$src, variable_ops),
1812 def FSUB64m : FPI<0xDC, MRM4m, OneArgFPRW, // ST(0) = ST(0) - [mem64real]
1813 (ops f64mem:$src, variable_ops),
1815 // ST(0) = ST(0) - [mem16int]
1816 //def FISUB16m : FPI16m<"fisub", 0xDE, MRM4m, OneArgFPRW>;
1817 // ST(0) = ST(0) - [mem32int]
1818 //def FISUB32m : FPI32m<"fisub", 0xDA, MRM4m, OneArgFPRW>;
1820 // FSUBR reg, mem: Before stackification, these are represented by:
1821 // R1 = FSUBR* R2, [mem]
1823 // Note that the order of operands does not reflect the operation being
1825 def FSUBR32m : FPI<0xD8, MRM5m, OneArgFPRW, // ST(0) = [mem32real] - ST(0)
1826 (ops f32mem:$src, variable_ops),
1828 def FSUBR64m : FPI<0xDC, MRM5m, OneArgFPRW, // ST(0) = [mem64real] - ST(0)
1829 (ops f64mem:$src, variable_ops),
1831 // ST(0) = [mem16int] - ST(0)
1832 //def FISUBR16m : FPI16m<"fisubr", 0xDE, MRM5m, OneArgFPRW>;
1833 // ST(0) = [mem32int] - ST(0)
1834 //def FISUBR32m : FPI32m<"fisubr", 0xDA, MRM5m, OneArgFPRW>;
1836 // FDIV reg, mem: Before stackification, these are represented by:
1837 // R1 = FDIV* R2, [mem]
1838 def FDIV32m : FPI<0xD8, MRM6m, OneArgFPRW, // ST(0) = ST(0) / [mem32real]
1839 (ops f32mem:$src, variable_ops),
1841 def FDIV64m : FPI<0xDC, MRM6m, OneArgFPRW, // ST(0) = ST(0) / [mem64real]
1842 (ops f64mem:$src, variable_ops),
1844 // ST(0) = ST(0) / [mem16int]
1845 //def FIDIV16m : FPI16m<"fidiv", 0xDE, MRM6m, OneArgFPRW>;
1846 // ST(0) = ST(0) / [mem32int]
1847 //def FIDIV32m : FPI32m<"fidiv", 0xDA, MRM6m, OneArgFPRW>;
1849 // FDIVR reg, mem: Before stackification, these are represented by:
1850 // R1 = FDIVR* R2, [mem]
1851 // Note that the order of operands does not reflect the operation being
1853 def FDIVR32m : FPI<0xD8, MRM7m, OneArgFPRW, // ST(0) = [mem32real] / ST(0)
1854 (ops f32mem:$src, variable_ops),
1856 def FDIVR64m : FPI<0xDC, MRM7m, OneArgFPRW, // ST(0) = [mem64real] / ST(0)
1857 (ops f64mem:$src, variable_ops),
1859 // ST(0) = [mem16int] / ST(0)
1860 //def FIDIVR16m : FPI16m<"fidivr", 0xDE, MRM7m, OneArgFPRW>;
1861 // ST(0) = [mem32int] / ST(0)
1862 //def FIDIVR32m : FPI32m<"fidivr", 0xDA, MRM7m, OneArgFPRW>;
1865 // Floating point cmovs...
1866 let isTwoAddress = 1, Uses = [ST0], Defs = [ST0] in {
1867 def FCMOVB : FPI<0xC0, AddRegFrm, CondMovFP,
1868 (ops RST:$op, variable_ops),
1869 "fcmovb {$op, %ST(0)|%ST(0), $op}">, DA;
1870 def FCMOVBE : FPI<0xD0, AddRegFrm, CondMovFP,
1871 (ops RST:$op, variable_ops),
1872 "fcmovbe {$op, %ST(0)|%ST(0), $op}">, DA;
1873 def FCMOVE : FPI<0xC8, AddRegFrm, CondMovFP,
1874 (ops RST:$op, variable_ops),
1875 "fcmove {$op, %ST(0)|%ST(0), $op}">, DA;
1876 def FCMOVP : FPI<0xD8, AddRegFrm, CondMovFP,
1877 (ops RST:$op, variable_ops),
1878 "fcmovu {$op, %ST(0)|%ST(0), $op}">, DA;
1879 def FCMOVAE : FPI<0xC0, AddRegFrm, CondMovFP,
1880 (ops RST:$op, variable_ops),
1881 "fcmovae {$op, %ST(0)|%ST(0), $op}">, DB;
1882 def FCMOVA : FPI<0xD0, AddRegFrm, CondMovFP,
1883 (ops RST:$op, variable_ops),
1884 "fcmova {$op, %ST(0)|%ST(0), $op}">, DB;
1885 def FCMOVNE : FPI<0xC8, AddRegFrm, CondMovFP,
1886 (ops RST:$op, variable_ops),
1887 "fcmovne {$op, %ST(0)|%ST(0), $op}">, DB;
1888 def FCMOVNP : FPI<0xD8, AddRegFrm, CondMovFP,
1889 (ops RST:$op, variable_ops),
1890 "fcmovnu {$op, %ST(0)|%ST(0), $op}">, DB;
1893 // Floating point loads & stores...
1894 // FIXME: these are all marked variable_ops because they have an implicit
1895 // destination. Instructions like FILD* that are generated by the instruction
1896 // selector (not the fp stackifier) need more accurate operand accounting.
1897 def FLDrr : FPI<0xC0, AddRegFrm, NotFP,
1898 (ops RST:$src, variable_ops),
1900 def FLD32m : FPI<0xD9, MRM0m, ZeroArgFP,
1901 (ops f32mem:$src, variable_ops),
1903 def FLD64m : FPI<0xDD, MRM0m, ZeroArgFP,
1904 (ops f64mem:$src, variable_ops),
1906 def FLD80m : FPI<0xDB, MRM5m, ZeroArgFP,
1907 (ops f80mem:$src, variable_ops),
1909 def FILD16m : FPI<0xDF, MRM0m, ZeroArgFP,
1910 (ops i16mem:$src, variable_ops),
1912 def FILD32m : FPI<0xDB, MRM0m, ZeroArgFP,
1913 (ops i32mem:$src, variable_ops),
1915 def FILD64m : FPI<0xDF, MRM5m, ZeroArgFP,
1916 (ops i64mem:$src, variable_ops),
1919 def FSTrr : FPI<0xD0, AddRegFrm, NotFP,
1920 (ops RST:$op, variable_ops),
1922 def FSTPrr : FPI<0xD8, AddRegFrm, NotFP,
1923 (ops RST:$op, variable_ops),
1925 def FST32m : FPI<0xD9, MRM2m, OneArgFP,
1926 (ops f32mem:$op, variable_ops),
1928 def FST64m : FPI<0xDD, MRM2m, OneArgFP,
1929 (ops f64mem:$op, variable_ops),
1931 def FSTP32m : FPI<0xD9, MRM3m, OneArgFP,
1932 (ops f32mem:$op, variable_ops),
1934 def FSTP64m : FPI<0xDD, MRM3m, OneArgFP,
1935 (ops f64mem:$op, variable_ops),
1937 def FSTP80m : FPI<0xDB, MRM7m, OneArgFP,
1938 (ops f80mem:$op, variable_ops),
1941 def FIST16m : FPI<0xDF, MRM2m , OneArgFP,
1942 (ops i16mem:$op, variable_ops),
1944 def FIST32m : FPI<0xDB, MRM2m , OneArgFP,
1945 (ops i32mem:$op, variable_ops),
1947 def FISTP16m : FPI<0xDF, MRM3m , NotFP ,
1948 (ops i16mem:$op, variable_ops),
1950 def FISTP32m : FPI<0xDB, MRM3m , NotFP ,
1951 (ops i32mem:$op, variable_ops),
1953 def FISTP64m : FPI<0xDF, MRM7m , OneArgFP,
1954 (ops i64mem:$op, variable_ops),
1957 def FXCH : FPI<0xC8, AddRegFrm, NotFP,
1958 (ops RST:$op), "fxch $op">, D9; // fxch ST(i), ST(0)
1960 // Floating point constant loads...
1961 def FLD0 : FPI<0xEE, RawFrm, ZeroArgFP, (ops variable_ops), "fldz">, D9;
1962 def FLD1 : FPI<0xE8, RawFrm, ZeroArgFP, (ops variable_ops), "fld1">, D9;
1965 // Unary operations...
1966 def FCHS : FPI<0xE0, RawFrm, OneArgFPRW, // f1 = fchs f2
1969 def FABS : FPI<0xE1, RawFrm, OneArgFPRW, // f1 = fabs f2
1972 def FSQRT : FPI<0xFA, RawFrm, OneArgFPRW, // fsqrt ST(0)
1975 def FSIN : FPI<0xFE, RawFrm, OneArgFPRW, // fsin ST(0)
1978 def FCOS : FPI<0xFF, RawFrm, OneArgFPRW, // fcos ST(0)
1981 def FTST : FPI<0xE4, RawFrm, OneArgFP , // ftst ST(0)
1985 // Binary arithmetic operations...
1986 class FPST0rInst<bits<8> o, dag ops, string asm>
1987 : I<o, AddRegFrm, ops, asm, []>, D8 {
1988 list<Register> Uses = [ST0];
1989 list<Register> Defs = [ST0];
1991 class FPrST0Inst<bits<8> o, dag ops, string asm>
1992 : I<o, AddRegFrm, ops, asm, []>, DC {
1993 list<Register> Uses = [ST0];
1995 class FPrST0PInst<bits<8> o, dag ops, string asm>
1996 : I<o, AddRegFrm, ops, asm, []>, DE {
1997 list<Register> Uses = [ST0];
2000 def FADDST0r : FPST0rInst <0xC0, (ops RST:$op),
2002 def FADDrST0 : FPrST0Inst <0xC0, (ops RST:$op),
2003 "fadd {%ST(0), $op|$op, %ST(0)}">;
2004 def FADDPrST0 : FPrST0PInst<0xC0, (ops RST:$op),
2007 // NOTE: GAS and apparently all other AT&T style assemblers have a broken notion
2008 // of some of the 'reverse' forms of the fsub and fdiv instructions. As such,
2009 // we have to put some 'r's in and take them out of weird places.
2010 def FSUBRST0r : FPST0rInst <0xE8, (ops RST:$op),
2012 def FSUBrST0 : FPrST0Inst <0xE8, (ops RST:$op),
2013 "fsub{r} {%ST(0), $op|$op, %ST(0)}">;
2014 def FSUBPrST0 : FPrST0PInst<0xE8, (ops RST:$op),
2017 def FSUBST0r : FPST0rInst <0xE0, (ops RST:$op),
2019 def FSUBRrST0 : FPrST0Inst <0xE0, (ops RST:$op),
2020 "fsub{|r} {%ST(0), $op|$op, %ST(0)}">;
2021 def FSUBRPrST0 : FPrST0PInst<0xE0, (ops RST:$op),
2024 def FMULST0r : FPST0rInst <0xC8, (ops RST:$op),
2026 def FMULrST0 : FPrST0Inst <0xC8, (ops RST:$op),
2027 "fmul {%ST(0), $op|$op, %ST(0)}">;
2028 def FMULPrST0 : FPrST0PInst<0xC8, (ops RST:$op),
2031 def FDIVRST0r : FPST0rInst <0xF8, (ops RST:$op),
2033 def FDIVrST0 : FPrST0Inst <0xF8, (ops RST:$op),
2034 "fdiv{r} {%ST(0), $op|$op, %ST(0)}">;
2035 def FDIVPrST0 : FPrST0PInst<0xF8, (ops RST:$op),
2038 def FDIVST0r : FPST0rInst <0xF0, (ops RST:$op), // ST(0) = ST(0) / ST(i)
2040 def FDIVRrST0 : FPrST0Inst <0xF0, (ops RST:$op), // ST(i) = ST(0) / ST(i)
2041 "fdiv{|r} {%ST(0), $op|$op, %ST(0)}">;
2042 def FDIVRPrST0 : FPrST0PInst<0xF0, (ops RST:$op), // ST(i) = ST(0) / ST(i), pop
2045 // Floating point compares
2046 def FUCOMr : FPI<0xE0, AddRegFrm, CompareFP, // FPSW = cmp ST(0) with ST(i)
2047 (ops RST:$reg, variable_ops),
2048 "fucom $reg">, DD, Imp<[ST0],[]>;
2049 def FUCOMPr : I<0xE8, AddRegFrm, // FPSW = cmp ST(0) with ST(i), pop
2050 (ops RST:$reg, variable_ops),
2051 "fucomp $reg", []>, DD, Imp<[ST0],[]>;
2052 def FUCOMPPr : I<0xE9, RawFrm, // cmp ST(0) with ST(1), pop, pop
2054 "fucompp", []>, DA, Imp<[ST0],[]>;
2056 def FUCOMIr : FPI<0xE8, AddRegFrm, CompareFP, // CC = cmp ST(0) with ST(i)
2057 (ops RST:$reg, variable_ops),
2058 "fucomi {$reg, %ST(0)|%ST(0), $reg}">, DB, Imp<[ST0],[]>;
2059 def FUCOMIPr : I<0xE8, AddRegFrm, // CC = cmp ST(0) with ST(i), pop
2060 (ops RST:$reg, variable_ops),
2061 "fucomip {$reg, %ST(0)|%ST(0), $reg}", []>, DF, Imp<[ST0],[]>;
2064 // Floating point flag ops
2065 def FNSTSW8r : I<0xE0, RawFrm, // AX = fp flags
2066 (ops), "fnstsw", []>, DF, Imp<[],[AX]>;
2068 def FNSTCW16m : I<0xD9, MRM7m, // [mem16] = X87 control world
2069 (ops i16mem:$dst), "fnstcw $dst", []>;
2070 def FLDCW16m : I<0xD9, MRM5m, // X87 control world = [mem16]
2071 (ops i16mem:$dst), "fldcw $dst", []>;