1 //====- X86Instr64bit.td - Describe X86-64 Instructions ----*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86-64 instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
16 //===----------------------------------------------------------------------===//
17 // Operand Definitions.
20 // 64-bits but only 32 bits are significant.
21 def i64i32imm : Operand<i64>;
23 // 64-bits but only 32 bits are significant, and those bits are treated as being
25 def i64i32imm_pcrel : Operand<i64> {
26 let PrintMethod = "print_pcrel_imm";
30 // 64-bits but only 8 bits are significant.
31 def i64i8imm : Operand<i64> {
32 let ParserMatchClass = ImmSExt8AsmOperand;
35 def lea64mem : Operand<i64> {
36 let PrintMethod = "printlea64mem";
37 let MIOperandInfo = (ops GR64, i8imm, GR64_NOSP, i32imm);
38 let ParserMatchClass = X86MemAsmOperand;
41 def lea64_32mem : Operand<i32> {
42 let PrintMethod = "printlea64_32mem";
43 let AsmOperandLowerMethod = "lower_lea64_32mem";
44 let MIOperandInfo = (ops GR32, i8imm, GR32_NOSP, i32imm);
45 let ParserMatchClass = X86MemAsmOperand;
48 //===----------------------------------------------------------------------===//
49 // Complex Pattern Definitions.
51 def lea64addr : ComplexPattern<i64, 4, "SelectLEAAddr",
52 [add, sub, mul, X86mul_imm, shl, or, frameindex,
55 def tls64addr : ComplexPattern<i64, 4, "SelectTLSADDRAddr",
56 [tglobaltlsaddr], []>;
58 //===----------------------------------------------------------------------===//
62 def i64immSExt8 : PatLeaf<(i64 imm), [{
63 // i64immSExt8 predicate - True if the 64-bit immediate fits in a 8-bit
64 // sign extended field.
65 return (int64_t)N->getZExtValue() == (int8_t)N->getZExtValue();
68 def i64immSExt32 : PatLeaf<(i64 imm), [{
69 // i64immSExt32 predicate - True if the 64-bit immediate fits in a 32-bit
70 // sign extended field.
71 return (int64_t)N->getZExtValue() == (int32_t)N->getZExtValue();
74 def i64immZExt32 : PatLeaf<(i64 imm), [{
75 // i64immZExt32 predicate - True if the 64-bit immediate fits in a 32-bit
76 // unsignedsign extended field.
77 return (uint64_t)N->getZExtValue() == (uint32_t)N->getZExtValue();
80 def sextloadi64i8 : PatFrag<(ops node:$ptr), (i64 (sextloadi8 node:$ptr))>;
81 def sextloadi64i16 : PatFrag<(ops node:$ptr), (i64 (sextloadi16 node:$ptr))>;
82 def sextloadi64i32 : PatFrag<(ops node:$ptr), (i64 (sextloadi32 node:$ptr))>;
84 def zextloadi64i1 : PatFrag<(ops node:$ptr), (i64 (zextloadi1 node:$ptr))>;
85 def zextloadi64i8 : PatFrag<(ops node:$ptr), (i64 (zextloadi8 node:$ptr))>;
86 def zextloadi64i16 : PatFrag<(ops node:$ptr), (i64 (zextloadi16 node:$ptr))>;
87 def zextloadi64i32 : PatFrag<(ops node:$ptr), (i64 (zextloadi32 node:$ptr))>;
89 def extloadi64i1 : PatFrag<(ops node:$ptr), (i64 (extloadi1 node:$ptr))>;
90 def extloadi64i8 : PatFrag<(ops node:$ptr), (i64 (extloadi8 node:$ptr))>;
91 def extloadi64i16 : PatFrag<(ops node:$ptr), (i64 (extloadi16 node:$ptr))>;
92 def extloadi64i32 : PatFrag<(ops node:$ptr), (i64 (extloadi32 node:$ptr))>;
94 //===----------------------------------------------------------------------===//
95 // Instruction list...
98 // ADJCALLSTACKDOWN/UP implicitly use/def RSP because they may be expanded into
99 // a stack adjustment and the codegen must know that they may modify the stack
100 // pointer before prolog-epilog rewriting occurs.
101 // Pessimistically assume ADJCALLSTACKDOWN / ADJCALLSTACKUP will become
102 // sub / add which can clobber EFLAGS.
103 let Defs = [RSP, EFLAGS], Uses = [RSP] in {
104 def ADJCALLSTACKDOWN64 : I<0, Pseudo, (outs), (ins i32imm:$amt),
106 [(X86callseq_start timm:$amt)]>,
107 Requires<[In64BitMode]>;
108 def ADJCALLSTACKUP64 : I<0, Pseudo, (outs), (ins i32imm:$amt1, i32imm:$amt2),
110 [(X86callseq_end timm:$amt1, timm:$amt2)]>,
111 Requires<[In64BitMode]>;
114 //===----------------------------------------------------------------------===//
115 // Call Instructions...
118 // All calls clobber the non-callee saved registers. RSP is marked as
119 // a use to prevent stack-pointer assignments that appear immediately
120 // before calls from potentially appearing dead. Uses for argument
121 // registers are added manually.
122 let Defs = [RAX, RCX, RDX, RSI, RDI, R8, R9, R10, R11,
123 FP0, FP1, FP2, FP3, FP4, FP5, FP6, ST0, ST1,
124 MM0, MM1, MM2, MM3, MM4, MM5, MM6, MM7,
125 XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, XMM6, XMM7,
126 XMM8, XMM9, XMM10, XMM11, XMM12, XMM13, XMM14, XMM15, EFLAGS],
129 // NOTE: this pattern doesn't match "X86call imm", because we do not know
130 // that the offset between an arbitrary immediate and the call will fit in
131 // the 32-bit pcrel field that we have.
132 def CALL64pcrel32 : Ii32<0xE8, RawFrm,
133 (outs), (ins i64i32imm_pcrel:$dst, variable_ops),
135 Requires<[In64BitMode, NotWin64]>;
136 def CALL64r : I<0xFF, MRM2r, (outs), (ins GR64:$dst, variable_ops),
137 "call\t{*}$dst", [(X86call GR64:$dst)]>,
138 Requires<[NotWin64]>;
139 def CALL64m : I<0xFF, MRM2m, (outs), (ins i64mem:$dst, variable_ops),
140 "call\t{*}$dst", [(X86call (loadi64 addr:$dst))]>,
141 Requires<[NotWin64]>;
143 def FARCALL64 : RI<0xFF, MRM3m, (outs), (ins opaque80mem:$dst),
144 "lcall{q}\t{*}$dst", []>;
147 // FIXME: We need to teach codegen about single list of call-clobbered registers.
149 // All calls clobber the non-callee saved registers. RSP is marked as
150 // a use to prevent stack-pointer assignments that appear immediately
151 // before calls from potentially appearing dead. Uses for argument
152 // registers are added manually.
153 let Defs = [RAX, RCX, RDX, R8, R9, R10, R11,
154 FP0, FP1, FP2, FP3, FP4, FP5, FP6, ST0, ST1,
155 MM0, MM1, MM2, MM3, MM4, MM5, MM6, MM7,
156 XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, EFLAGS],
158 def WINCALL64pcrel32 : I<0xE8, RawFrm,
159 (outs), (ins i64i32imm_pcrel:$dst, variable_ops),
162 def WINCALL64r : I<0xFF, MRM2r, (outs), (ins GR64:$dst, variable_ops),
164 [(X86call GR64:$dst)]>, Requires<[IsWin64]>;
165 def WINCALL64m : I<0xFF, MRM2m, (outs), (ins i64mem:$dst, variable_ops),
167 [(X86call (loadi64 addr:$dst))]>, Requires<[IsWin64]>;
171 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1 in
172 def TCRETURNdi64 : I<0, Pseudo, (outs), (ins i64imm:$dst, i32imm:$offset,
174 "#TC_RETURN $dst $offset",
177 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1 in
178 def TCRETURNri64 : I<0, Pseudo, (outs), (ins GR64:$dst, i32imm:$offset,
180 "#TC_RETURN $dst $offset",
184 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1 in
185 def TAILJMPr64 : I<0xFF, MRM4r, (outs), (ins GR64:$dst),
186 "jmp{q}\t{*}$dst # TAILCALL",
190 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
191 def JMP64r : I<0xFF, MRM4r, (outs), (ins GR64:$dst), "jmp{q}\t{*}$dst",
192 [(brind GR64:$dst)]>;
193 def JMP64m : I<0xFF, MRM4m, (outs), (ins i64mem:$dst), "jmp{q}\t{*}$dst",
194 [(brind (loadi64 addr:$dst))]>;
195 def FARJMP64 : RI<0xFF, MRM5m, (outs), (ins opaque80mem:$dst),
196 "ljmp{q}\t{*}$dst", []>;
199 //===----------------------------------------------------------------------===//
200 // EH Pseudo Instructions
202 let isTerminator = 1, isReturn = 1, isBarrier = 1,
204 def EH_RETURN64 : I<0xC3, RawFrm, (outs), (ins GR64:$addr),
205 "ret\t#eh_return, addr: $addr",
206 [(X86ehret GR64:$addr)]>;
210 //===----------------------------------------------------------------------===//
211 // Miscellaneous Instructions...
213 let Defs = [RBP,RSP], Uses = [RBP,RSP], mayLoad = 1, neverHasSideEffects = 1 in
214 def LEAVE64 : I<0xC9, RawFrm,
215 (outs), (ins), "leave", []>;
216 let Defs = [RSP], Uses = [RSP], neverHasSideEffects=1 in {
218 def POP64r : I<0x58, AddRegFrm,
219 (outs GR64:$reg), (ins), "pop{q}\t$reg", []>;
220 def POP64rmr: I<0x8F, MRM0r, (outs GR64:$reg), (ins), "pop{q}\t$reg", []>;
221 def POP64rmm: I<0x8F, MRM0m, (outs i64mem:$dst), (ins), "pop{q}\t$dst", []>;
223 let mayStore = 1 in {
224 def PUSH64r : I<0x50, AddRegFrm,
225 (outs), (ins GR64:$reg), "push{q}\t$reg", []>;
226 def PUSH64rmr: I<0xFF, MRM6r, (outs), (ins GR64:$reg), "push{q}\t$reg", []>;
227 def PUSH64rmm: I<0xFF, MRM6m, (outs), (ins i64mem:$src), "push{q}\t$src", []>;
231 let Defs = [RSP], Uses = [RSP], neverHasSideEffects = 1, mayStore = 1 in {
232 def PUSH64i8 : Ii8<0x6a, RawFrm, (outs), (ins i8imm:$imm),
233 "push{q}\t$imm", []>;
234 def PUSH64i16 : Ii16<0x68, RawFrm, (outs), (ins i16imm:$imm),
235 "push{q}\t$imm", []>;
236 def PUSH64i32 : Ii32<0x68, RawFrm, (outs), (ins i32imm:$imm),
237 "push{q}\t$imm", []>;
240 let Defs = [RSP, EFLAGS], Uses = [RSP], mayLoad = 1 in
241 def POPFQ : I<0x9D, RawFrm, (outs), (ins), "popf", []>, REX_W;
242 let Defs = [RSP], Uses = [RSP, EFLAGS], mayStore = 1 in
243 def PUSHFQ : I<0x9C, RawFrm, (outs), (ins), "pushf", []>;
245 def LEA64_32r : I<0x8D, MRMSrcMem,
246 (outs GR32:$dst), (ins lea64_32mem:$src),
247 "lea{l}\t{$src|$dst}, {$dst|$src}",
248 [(set GR32:$dst, lea32addr:$src)]>, Requires<[In64BitMode]>;
250 let isReMaterializable = 1 in
251 def LEA64r : RI<0x8D, MRMSrcMem, (outs GR64:$dst), (ins lea64mem:$src),
252 "lea{q}\t{$src|$dst}, {$dst|$src}",
253 [(set GR64:$dst, lea64addr:$src)]>;
255 let isTwoAddress = 1 in
256 def BSWAP64r : RI<0xC8, AddRegFrm, (outs GR64:$dst), (ins GR64:$src),
258 [(set GR64:$dst, (bswap GR64:$src))]>, TB;
260 // Bit scan instructions.
261 let Defs = [EFLAGS] in {
262 def BSF64rr : RI<0xBC, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
263 "bsf{q}\t{$src, $dst|$dst, $src}",
264 [(set GR64:$dst, (X86bsf GR64:$src)), (implicit EFLAGS)]>, TB;
265 def BSF64rm : RI<0xBC, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
266 "bsf{q}\t{$src, $dst|$dst, $src}",
267 [(set GR64:$dst, (X86bsf (loadi64 addr:$src))),
268 (implicit EFLAGS)]>, TB;
270 def BSR64rr : RI<0xBD, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
271 "bsr{q}\t{$src, $dst|$dst, $src}",
272 [(set GR64:$dst, (X86bsr GR64:$src)), (implicit EFLAGS)]>, TB;
273 def BSR64rm : RI<0xBD, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
274 "bsr{q}\t{$src, $dst|$dst, $src}",
275 [(set GR64:$dst, (X86bsr (loadi64 addr:$src))),
276 (implicit EFLAGS)]>, TB;
280 let Defs = [RCX,RDI,RSI], Uses = [RCX,RDI,RSI] in
281 def REP_MOVSQ : RI<0xA5, RawFrm, (outs), (ins), "{rep;movsq|rep movsq}",
282 [(X86rep_movs i64)]>, REP;
283 let Defs = [RCX,RDI], Uses = [RAX,RCX,RDI] in
284 def REP_STOSQ : RI<0xAB, RawFrm, (outs), (ins), "{rep;stosq|rep stosq}",
285 [(X86rep_stos i64)]>, REP;
287 // Fast system-call instructions
288 def SYSEXIT64 : RI<0x35, RawFrm,
289 (outs), (ins), "sysexit", []>, TB;
291 //===----------------------------------------------------------------------===//
292 // Move Instructions...
295 let neverHasSideEffects = 1 in
296 def MOV64rr : RI<0x89, MRMDestReg, (outs GR64:$dst), (ins GR64:$src),
297 "mov{q}\t{$src, $dst|$dst, $src}", []>;
299 let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
300 def MOV64ri : RIi64<0xB8, AddRegFrm, (outs GR64:$dst), (ins i64imm:$src),
301 "movabs{q}\t{$src, $dst|$dst, $src}",
302 [(set GR64:$dst, imm:$src)]>;
303 def MOV64ri32 : RIi32<0xC7, MRM0r, (outs GR64:$dst), (ins i64i32imm:$src),
304 "mov{q}\t{$src, $dst|$dst, $src}",
305 [(set GR64:$dst, i64immSExt32:$src)]>;
308 let canFoldAsLoad = 1 in
309 def MOV64rm : RI<0x8B, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
310 "mov{q}\t{$src, $dst|$dst, $src}",
311 [(set GR64:$dst, (load addr:$src))]>;
313 def MOV64mr : RI<0x89, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
314 "mov{q}\t{$src, $dst|$dst, $src}",
315 [(store GR64:$src, addr:$dst)]>;
316 def MOV64mi32 : RIi32<0xC7, MRM0m, (outs), (ins i64mem:$dst, i64i32imm:$src),
317 "mov{q}\t{$src, $dst|$dst, $src}",
318 [(store i64immSExt32:$src, addr:$dst)]>;
320 // Sign/Zero extenders
322 // MOVSX64rr8 always has a REX prefix and it has an 8-bit register
323 // operand, which makes it a rare instruction with an 8-bit register
324 // operand that can never access an h register. If support for h registers
325 // were generalized, this would require a special register class.
326 def MOVSX64rr8 : RI<0xBE, MRMSrcReg, (outs GR64:$dst), (ins GR8 :$src),
327 "movs{bq|x}\t{$src, $dst|$dst, $src}",
328 [(set GR64:$dst, (sext GR8:$src))]>, TB;
329 def MOVSX64rm8 : RI<0xBE, MRMSrcMem, (outs GR64:$dst), (ins i8mem :$src),
330 "movs{bq|x}\t{$src, $dst|$dst, $src}",
331 [(set GR64:$dst, (sextloadi64i8 addr:$src))]>, TB;
332 def MOVSX64rr16: RI<0xBF, MRMSrcReg, (outs GR64:$dst), (ins GR16:$src),
333 "movs{wq|x}\t{$src, $dst|$dst, $src}",
334 [(set GR64:$dst, (sext GR16:$src))]>, TB;
335 def MOVSX64rm16: RI<0xBF, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src),
336 "movs{wq|x}\t{$src, $dst|$dst, $src}",
337 [(set GR64:$dst, (sextloadi64i16 addr:$src))]>, TB;
338 def MOVSX64rr32: RI<0x63, MRMSrcReg, (outs GR64:$dst), (ins GR32:$src),
339 "movs{lq|xd}\t{$src, $dst|$dst, $src}",
340 [(set GR64:$dst, (sext GR32:$src))]>;
341 def MOVSX64rm32: RI<0x63, MRMSrcMem, (outs GR64:$dst), (ins i32mem:$src),
342 "movs{lq|xd}\t{$src, $dst|$dst, $src}",
343 [(set GR64:$dst, (sextloadi64i32 addr:$src))]>;
345 // Use movzbl instead of movzbq when the destination is a register; it's
346 // equivalent due to implicit zero-extending, and it has a smaller encoding.
347 def MOVZX64rr8 : I<0xB6, MRMSrcReg, (outs GR64:$dst), (ins GR8 :$src),
348 "movz{bl|x}\t{$src, ${dst:subreg32}|${dst:subreg32}, $src}",
349 [(set GR64:$dst, (zext GR8:$src))]>, TB;
350 def MOVZX64rm8 : I<0xB6, MRMSrcMem, (outs GR64:$dst), (ins i8mem :$src),
351 "movz{bl|x}\t{$src, ${dst:subreg32}|${dst:subreg32}, $src}",
352 [(set GR64:$dst, (zextloadi64i8 addr:$src))]>, TB;
353 // Use movzwl instead of movzwq when the destination is a register; it's
354 // equivalent due to implicit zero-extending, and it has a smaller encoding.
355 def MOVZX64rr16: I<0xB7, MRMSrcReg, (outs GR64:$dst), (ins GR16:$src),
356 "movz{wl|x}\t{$src, ${dst:subreg32}|${dst:subreg32}, $src}",
357 [(set GR64:$dst, (zext GR16:$src))]>, TB;
358 def MOVZX64rm16: I<0xB7, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src),
359 "movz{wl|x}\t{$src, ${dst:subreg32}|${dst:subreg32}, $src}",
360 [(set GR64:$dst, (zextloadi64i16 addr:$src))]>, TB;
362 // There's no movzlq instruction, but movl can be used for this purpose, using
363 // implicit zero-extension. The preferred way to do 32-bit-to-64-bit zero
364 // extension on x86-64 is to use a SUBREG_TO_REG to utilize implicit
365 // zero-extension, however this isn't possible when the 32-bit value is
366 // defined by a truncate or is copied from something where the high bits aren't
367 // necessarily all zero. In such cases, we fall back to these explicit zext
369 def MOVZX64rr32 : I<0x89, MRMDestReg, (outs GR64:$dst), (ins GR32:$src),
370 "mov{l}\t{$src, ${dst:subreg32}|${dst:subreg32}, $src}",
371 [(set GR64:$dst, (zext GR32:$src))]>;
372 def MOVZX64rm32 : I<0x8B, MRMSrcMem, (outs GR64:$dst), (ins i32mem:$src),
373 "mov{l}\t{$src, ${dst:subreg32}|${dst:subreg32}, $src}",
374 [(set GR64:$dst, (zextloadi64i32 addr:$src))]>;
376 // Any instruction that defines a 32-bit result leaves the high half of the
377 // register. Truncate can be lowered to EXTRACT_SUBREG, and CopyFromReg may
378 // be copying from a truncate, but any other 32-bit operation will zero-extend
380 def def32 : PatLeaf<(i32 GR32:$src), [{
381 return N->getOpcode() != ISD::TRUNCATE &&
382 N->getOpcode() != TargetInstrInfo::EXTRACT_SUBREG &&
383 N->getOpcode() != ISD::CopyFromReg;
386 // In the case of a 32-bit def that is known to implicitly zero-extend,
387 // we can use a SUBREG_TO_REG.
388 def : Pat<(i64 (zext def32:$src)),
389 (SUBREG_TO_REG (i64 0), GR32:$src, x86_subreg_32bit)>;
391 let neverHasSideEffects = 1 in {
392 let Defs = [RAX], Uses = [EAX] in
393 def CDQE : RI<0x98, RawFrm, (outs), (ins),
394 "{cltq|cdqe}", []>; // RAX = signext(EAX)
396 let Defs = [RAX,RDX], Uses = [RAX] in
397 def CQO : RI<0x99, RawFrm, (outs), (ins),
398 "{cqto|cqo}", []>; // RDX:RAX = signext(RAX)
401 //===----------------------------------------------------------------------===//
402 // Arithmetic Instructions...
405 let Defs = [EFLAGS] in {
407 def ADD64i32 : RI<0x05, RawFrm, (outs), (ins i32imm:$src),
408 "add{q}\t{$src, %rax|%rax, $src}", []>;
410 let isTwoAddress = 1 in {
411 let isConvertibleToThreeAddress = 1 in {
412 let isCommutable = 1 in
413 // Register-Register Addition
414 def ADD64rr : RI<0x01, MRMDestReg, (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
415 "add{q}\t{$src2, $dst|$dst, $src2}",
416 [(set GR64:$dst, (add GR64:$src1, GR64:$src2)),
419 // Register-Integer Addition
420 def ADD64ri8 : RIi8<0x83, MRM0r, (outs GR64:$dst), (ins GR64:$src1, i64i8imm:$src2),
421 "add{q}\t{$src2, $dst|$dst, $src2}",
422 [(set GR64:$dst, (add GR64:$src1, i64immSExt8:$src2)),
424 def ADD64ri32 : RIi32<0x81, MRM0r, (outs GR64:$dst), (ins GR64:$src1, i64i32imm:$src2),
425 "add{q}\t{$src2, $dst|$dst, $src2}",
426 [(set GR64:$dst, (add GR64:$src1, i64immSExt32:$src2)),
428 } // isConvertibleToThreeAddress
430 // Register-Memory Addition
431 def ADD64rm : RI<0x03, MRMSrcMem, (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
432 "add{q}\t{$src2, $dst|$dst, $src2}",
433 [(set GR64:$dst, (add GR64:$src1, (load addr:$src2))),
437 // Memory-Register Addition
438 def ADD64mr : RI<0x01, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src2),
439 "add{q}\t{$src2, $dst|$dst, $src2}",
440 [(store (add (load addr:$dst), GR64:$src2), addr:$dst),
442 def ADD64mi8 : RIi8<0x83, MRM0m, (outs), (ins i64mem:$dst, i64i8imm :$src2),
443 "add{q}\t{$src2, $dst|$dst, $src2}",
444 [(store (add (load addr:$dst), i64immSExt8:$src2), addr:$dst),
446 def ADD64mi32 : RIi32<0x81, MRM0m, (outs), (ins i64mem:$dst, i64i32imm :$src2),
447 "add{q}\t{$src2, $dst|$dst, $src2}",
448 [(store (add (load addr:$dst), i64immSExt32:$src2), addr:$dst),
451 let Uses = [EFLAGS] in {
452 let isTwoAddress = 1 in {
453 let isCommutable = 1 in
454 def ADC64rr : RI<0x11, MRMDestReg, (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
455 "adc{q}\t{$src2, $dst|$dst, $src2}",
456 [(set GR64:$dst, (adde GR64:$src1, GR64:$src2))]>;
458 def ADC64rm : RI<0x13, MRMSrcMem , (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
459 "adc{q}\t{$src2, $dst|$dst, $src2}",
460 [(set GR64:$dst, (adde GR64:$src1, (load addr:$src2)))]>;
462 def ADC64ri8 : RIi8<0x83, MRM2r, (outs GR64:$dst), (ins GR64:$src1, i64i8imm:$src2),
463 "adc{q}\t{$src2, $dst|$dst, $src2}",
464 [(set GR64:$dst, (adde GR64:$src1, i64immSExt8:$src2))]>;
465 def ADC64ri32 : RIi32<0x81, MRM2r, (outs GR64:$dst), (ins GR64:$src1, i64i32imm:$src2),
466 "adc{q}\t{$src2, $dst|$dst, $src2}",
467 [(set GR64:$dst, (adde GR64:$src1, i64immSExt32:$src2))]>;
470 def ADC64mr : RI<0x11, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src2),
471 "adc{q}\t{$src2, $dst|$dst, $src2}",
472 [(store (adde (load addr:$dst), GR64:$src2), addr:$dst)]>;
473 def ADC64mi8 : RIi8<0x83, MRM2m, (outs), (ins i64mem:$dst, i64i8imm :$src2),
474 "adc{q}\t{$src2, $dst|$dst, $src2}",
475 [(store (adde (load addr:$dst), i64immSExt8:$src2), addr:$dst)]>;
476 def ADC64mi32 : RIi32<0x81, MRM2m, (outs), (ins i64mem:$dst, i64i32imm:$src2),
477 "adc{q}\t{$src2, $dst|$dst, $src2}",
478 [(store (adde (load addr:$dst), i64immSExt8:$src2), addr:$dst)]>;
481 let isTwoAddress = 1 in {
482 // Register-Register Subtraction
483 def SUB64rr : RI<0x29, MRMDestReg, (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
484 "sub{q}\t{$src2, $dst|$dst, $src2}",
485 [(set GR64:$dst, (sub GR64:$src1, GR64:$src2)),
488 // Register-Memory Subtraction
489 def SUB64rm : RI<0x2B, MRMSrcMem, (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
490 "sub{q}\t{$src2, $dst|$dst, $src2}",
491 [(set GR64:$dst, (sub GR64:$src1, (load addr:$src2))),
494 // Register-Integer Subtraction
495 def SUB64ri8 : RIi8<0x83, MRM5r, (outs GR64:$dst),
496 (ins GR64:$src1, i64i8imm:$src2),
497 "sub{q}\t{$src2, $dst|$dst, $src2}",
498 [(set GR64:$dst, (sub GR64:$src1, i64immSExt8:$src2)),
500 def SUB64ri32 : RIi32<0x81, MRM5r, (outs GR64:$dst),
501 (ins GR64:$src1, i64i32imm:$src2),
502 "sub{q}\t{$src2, $dst|$dst, $src2}",
503 [(set GR64:$dst, (sub GR64:$src1, i64immSExt32:$src2)),
507 // Memory-Register Subtraction
508 def SUB64mr : RI<0x29, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src2),
509 "sub{q}\t{$src2, $dst|$dst, $src2}",
510 [(store (sub (load addr:$dst), GR64:$src2), addr:$dst),
513 // Memory-Integer Subtraction
514 def SUB64mi8 : RIi8<0x83, MRM5m, (outs), (ins i64mem:$dst, i64i8imm :$src2),
515 "sub{q}\t{$src2, $dst|$dst, $src2}",
516 [(store (sub (load addr:$dst), i64immSExt8:$src2),
519 def SUB64mi32 : RIi32<0x81, MRM5m, (outs), (ins i64mem:$dst, i64i32imm:$src2),
520 "sub{q}\t{$src2, $dst|$dst, $src2}",
521 [(store (sub (load addr:$dst), i64immSExt32:$src2),
525 let Uses = [EFLAGS] in {
526 let isTwoAddress = 1 in {
527 def SBB64rr : RI<0x19, MRMDestReg, (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
528 "sbb{q}\t{$src2, $dst|$dst, $src2}",
529 [(set GR64:$dst, (sube GR64:$src1, GR64:$src2))]>;
531 def SBB64rm : RI<0x1B, MRMSrcMem, (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
532 "sbb{q}\t{$src2, $dst|$dst, $src2}",
533 [(set GR64:$dst, (sube GR64:$src1, (load addr:$src2)))]>;
535 def SBB64ri8 : RIi8<0x83, MRM3r, (outs GR64:$dst), (ins GR64:$src1, i64i8imm:$src2),
536 "sbb{q}\t{$src2, $dst|$dst, $src2}",
537 [(set GR64:$dst, (sube GR64:$src1, i64immSExt8:$src2))]>;
538 def SBB64ri32 : RIi32<0x81, MRM3r, (outs GR64:$dst), (ins GR64:$src1, i64i32imm:$src2),
539 "sbb{q}\t{$src2, $dst|$dst, $src2}",
540 [(set GR64:$dst, (sube GR64:$src1, i64immSExt32:$src2))]>;
543 def SBB64mr : RI<0x19, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src2),
544 "sbb{q}\t{$src2, $dst|$dst, $src2}",
545 [(store (sube (load addr:$dst), GR64:$src2), addr:$dst)]>;
546 def SBB64mi8 : RIi8<0x83, MRM3m, (outs), (ins i64mem:$dst, i64i8imm :$src2),
547 "sbb{q}\t{$src2, $dst|$dst, $src2}",
548 [(store (sube (load addr:$dst), i64immSExt8:$src2), addr:$dst)]>;
549 def SBB64mi32 : RIi32<0x81, MRM3m, (outs), (ins i64mem:$dst, i64i32imm:$src2),
550 "sbb{q}\t{$src2, $dst|$dst, $src2}",
551 [(store (sube (load addr:$dst), i64immSExt32:$src2), addr:$dst)]>;
555 // Unsigned multiplication
556 let Defs = [RAX,RDX,EFLAGS], Uses = [RAX], neverHasSideEffects = 1 in {
557 def MUL64r : RI<0xF7, MRM4r, (outs), (ins GR64:$src),
558 "mul{q}\t$src", []>; // RAX,RDX = RAX*GR64
560 def MUL64m : RI<0xF7, MRM4m, (outs), (ins i64mem:$src),
561 "mul{q}\t$src", []>; // RAX,RDX = RAX*[mem64]
563 // Signed multiplication
564 def IMUL64r : RI<0xF7, MRM5r, (outs), (ins GR64:$src),
565 "imul{q}\t$src", []>; // RAX,RDX = RAX*GR64
567 def IMUL64m : RI<0xF7, MRM5m, (outs), (ins i64mem:$src),
568 "imul{q}\t$src", []>; // RAX,RDX = RAX*[mem64]
571 let Defs = [EFLAGS] in {
572 let isTwoAddress = 1 in {
573 let isCommutable = 1 in
574 // Register-Register Signed Integer Multiplication
575 def IMUL64rr : RI<0xAF, MRMSrcReg, (outs GR64:$dst),
576 (ins GR64:$src1, GR64:$src2),
577 "imul{q}\t{$src2, $dst|$dst, $src2}",
578 [(set GR64:$dst, (mul GR64:$src1, GR64:$src2)),
579 (implicit EFLAGS)]>, TB;
581 // Register-Memory Signed Integer Multiplication
582 def IMUL64rm : RI<0xAF, MRMSrcMem, (outs GR64:$dst),
583 (ins GR64:$src1, i64mem:$src2),
584 "imul{q}\t{$src2, $dst|$dst, $src2}",
585 [(set GR64:$dst, (mul GR64:$src1, (load addr:$src2))),
586 (implicit EFLAGS)]>, TB;
589 // Suprisingly enough, these are not two address instructions!
591 // Register-Integer Signed Integer Multiplication
592 def IMUL64rri8 : RIi8<0x6B, MRMSrcReg, // GR64 = GR64*I8
593 (outs GR64:$dst), (ins GR64:$src1, i64i8imm:$src2),
594 "imul{q}\t{$src2, $src1, $dst|$dst, $src1, $src2}",
595 [(set GR64:$dst, (mul GR64:$src1, i64immSExt8:$src2)),
597 def IMUL64rri32 : RIi32<0x69, MRMSrcReg, // GR64 = GR64*I32
598 (outs GR64:$dst), (ins GR64:$src1, i64i32imm:$src2),
599 "imul{q}\t{$src2, $src1, $dst|$dst, $src1, $src2}",
600 [(set GR64:$dst, (mul GR64:$src1, i64immSExt32:$src2)),
603 // Memory-Integer Signed Integer Multiplication
604 def IMUL64rmi8 : RIi8<0x6B, MRMSrcMem, // GR64 = [mem64]*I8
605 (outs GR64:$dst), (ins i64mem:$src1, i64i8imm: $src2),
606 "imul{q}\t{$src2, $src1, $dst|$dst, $src1, $src2}",
607 [(set GR64:$dst, (mul (load addr:$src1),
610 def IMUL64rmi32 : RIi32<0x69, MRMSrcMem, // GR64 = [mem64]*I32
611 (outs GR64:$dst), (ins i64mem:$src1, i64i32imm:$src2),
612 "imul{q}\t{$src2, $src1, $dst|$dst, $src1, $src2}",
613 [(set GR64:$dst, (mul (load addr:$src1),
614 i64immSExt32:$src2)),
618 // Unsigned division / remainder
619 let Defs = [RAX,RDX,EFLAGS], Uses = [RAX,RDX] in {
620 def DIV64r : RI<0xF7, MRM6r, (outs), (ins GR64:$src), // RDX:RAX/r64 = RAX,RDX
622 // Signed division / remainder
623 def IDIV64r: RI<0xF7, MRM7r, (outs), (ins GR64:$src), // RDX:RAX/r64 = RAX,RDX
624 "idiv{q}\t$src", []>;
626 def DIV64m : RI<0xF7, MRM6m, (outs), (ins i64mem:$src), // RDX:RAX/[mem64] = RAX,RDX
628 def IDIV64m: RI<0xF7, MRM7m, (outs), (ins i64mem:$src), // RDX:RAX/[mem64] = RAX,RDX
629 "idiv{q}\t$src", []>;
633 // Unary instructions
634 let Defs = [EFLAGS], CodeSize = 2 in {
635 let isTwoAddress = 1 in
636 def NEG64r : RI<0xF7, MRM3r, (outs GR64:$dst), (ins GR64:$src), "neg{q}\t$dst",
637 [(set GR64:$dst, (ineg GR64:$src)),
639 def NEG64m : RI<0xF7, MRM3m, (outs), (ins i64mem:$dst), "neg{q}\t$dst",
640 [(store (ineg (loadi64 addr:$dst)), addr:$dst),
643 let isTwoAddress = 1, isConvertibleToThreeAddress = 1 in
644 def INC64r : RI<0xFF, MRM0r, (outs GR64:$dst), (ins GR64:$src), "inc{q}\t$dst",
645 [(set GR64:$dst, (add GR64:$src, 1)),
647 def INC64m : RI<0xFF, MRM0m, (outs), (ins i64mem:$dst), "inc{q}\t$dst",
648 [(store (add (loadi64 addr:$dst), 1), addr:$dst),
651 let isTwoAddress = 1, isConvertibleToThreeAddress = 1 in
652 def DEC64r : RI<0xFF, MRM1r, (outs GR64:$dst), (ins GR64:$src), "dec{q}\t$dst",
653 [(set GR64:$dst, (add GR64:$src, -1)),
655 def DEC64m : RI<0xFF, MRM1m, (outs), (ins i64mem:$dst), "dec{q}\t$dst",
656 [(store (add (loadi64 addr:$dst), -1), addr:$dst),
659 // In 64-bit mode, single byte INC and DEC cannot be encoded.
660 let isTwoAddress = 1, isConvertibleToThreeAddress = 1 in {
661 // Can transform into LEA.
662 def INC64_16r : I<0xFF, MRM0r, (outs GR16:$dst), (ins GR16:$src), "inc{w}\t$dst",
663 [(set GR16:$dst, (add GR16:$src, 1)),
665 OpSize, Requires<[In64BitMode]>;
666 def INC64_32r : I<0xFF, MRM0r, (outs GR32:$dst), (ins GR32:$src), "inc{l}\t$dst",
667 [(set GR32:$dst, (add GR32:$src, 1)),
669 Requires<[In64BitMode]>;
670 def DEC64_16r : I<0xFF, MRM1r, (outs GR16:$dst), (ins GR16:$src), "dec{w}\t$dst",
671 [(set GR16:$dst, (add GR16:$src, -1)),
673 OpSize, Requires<[In64BitMode]>;
674 def DEC64_32r : I<0xFF, MRM1r, (outs GR32:$dst), (ins GR32:$src), "dec{l}\t$dst",
675 [(set GR32:$dst, (add GR32:$src, -1)),
677 Requires<[In64BitMode]>;
678 } // isConvertibleToThreeAddress
680 // These are duplicates of their 32-bit counterparts. Only needed so X86 knows
681 // how to unfold them.
682 let isTwoAddress = 0, CodeSize = 2 in {
683 def INC64_16m : I<0xFF, MRM0m, (outs), (ins i16mem:$dst), "inc{w}\t$dst",
684 [(store (add (loadi16 addr:$dst), 1), addr:$dst),
686 OpSize, Requires<[In64BitMode]>;
687 def INC64_32m : I<0xFF, MRM0m, (outs), (ins i32mem:$dst), "inc{l}\t$dst",
688 [(store (add (loadi32 addr:$dst), 1), addr:$dst),
690 Requires<[In64BitMode]>;
691 def DEC64_16m : I<0xFF, MRM1m, (outs), (ins i16mem:$dst), "dec{w}\t$dst",
692 [(store (add (loadi16 addr:$dst), -1), addr:$dst),
694 OpSize, Requires<[In64BitMode]>;
695 def DEC64_32m : I<0xFF, MRM1m, (outs), (ins i32mem:$dst), "dec{l}\t$dst",
696 [(store (add (loadi32 addr:$dst), -1), addr:$dst),
698 Requires<[In64BitMode]>;
700 } // Defs = [EFLAGS], CodeSize
703 let Defs = [EFLAGS] in {
704 // Shift instructions
705 let isTwoAddress = 1 in {
707 def SHL64rCL : RI<0xD3, MRM4r, (outs GR64:$dst), (ins GR64:$src),
708 "shl{q}\t{%cl, $dst|$dst, %CL}",
709 [(set GR64:$dst, (shl GR64:$src, CL))]>;
710 let isConvertibleToThreeAddress = 1 in // Can transform into LEA.
711 def SHL64ri : RIi8<0xC1, MRM4r, (outs GR64:$dst), (ins GR64:$src1, i8imm:$src2),
712 "shl{q}\t{$src2, $dst|$dst, $src2}",
713 [(set GR64:$dst, (shl GR64:$src1, (i8 imm:$src2)))]>;
714 // NOTE: We don't use shifts of a register by one, because 'add reg,reg' is
719 def SHL64mCL : RI<0xD3, MRM4m, (outs), (ins i64mem:$dst),
720 "shl{q}\t{%cl, $dst|$dst, %CL}",
721 [(store (shl (loadi64 addr:$dst), CL), addr:$dst)]>;
722 def SHL64mi : RIi8<0xC1, MRM4m, (outs), (ins i64mem:$dst, i8imm:$src),
723 "shl{q}\t{$src, $dst|$dst, $src}",
724 [(store (shl (loadi64 addr:$dst), (i8 imm:$src)), addr:$dst)]>;
725 def SHL64m1 : RI<0xD1, MRM4m, (outs), (ins i64mem:$dst),
727 [(store (shl (loadi64 addr:$dst), (i8 1)), addr:$dst)]>;
729 let isTwoAddress = 1 in {
731 def SHR64rCL : RI<0xD3, MRM5r, (outs GR64:$dst), (ins GR64:$src),
732 "shr{q}\t{%cl, $dst|$dst, %CL}",
733 [(set GR64:$dst, (srl GR64:$src, CL))]>;
734 def SHR64ri : RIi8<0xC1, MRM5r, (outs GR64:$dst), (ins GR64:$src1, i8imm:$src2),
735 "shr{q}\t{$src2, $dst|$dst, $src2}",
736 [(set GR64:$dst, (srl GR64:$src1, (i8 imm:$src2)))]>;
737 def SHR64r1 : RI<0xD1, MRM5r, (outs GR64:$dst), (ins GR64:$src1),
739 [(set GR64:$dst, (srl GR64:$src1, (i8 1)))]>;
743 def SHR64mCL : RI<0xD3, MRM5m, (outs), (ins i64mem:$dst),
744 "shr{q}\t{%cl, $dst|$dst, %CL}",
745 [(store (srl (loadi64 addr:$dst), CL), addr:$dst)]>;
746 def SHR64mi : RIi8<0xC1, MRM5m, (outs), (ins i64mem:$dst, i8imm:$src),
747 "shr{q}\t{$src, $dst|$dst, $src}",
748 [(store (srl (loadi64 addr:$dst), (i8 imm:$src)), addr:$dst)]>;
749 def SHR64m1 : RI<0xD1, MRM5m, (outs), (ins i64mem:$dst),
751 [(store (srl (loadi64 addr:$dst), (i8 1)), addr:$dst)]>;
753 let isTwoAddress = 1 in {
755 def SAR64rCL : RI<0xD3, MRM7r, (outs GR64:$dst), (ins GR64:$src),
756 "sar{q}\t{%cl, $dst|$dst, %CL}",
757 [(set GR64:$dst, (sra GR64:$src, CL))]>;
758 def SAR64ri : RIi8<0xC1, MRM7r, (outs GR64:$dst), (ins GR64:$src1, i8imm:$src2),
759 "sar{q}\t{$src2, $dst|$dst, $src2}",
760 [(set GR64:$dst, (sra GR64:$src1, (i8 imm:$src2)))]>;
761 def SAR64r1 : RI<0xD1, MRM7r, (outs GR64:$dst), (ins GR64:$src1),
763 [(set GR64:$dst, (sra GR64:$src1, (i8 1)))]>;
767 def SAR64mCL : RI<0xD3, MRM7m, (outs), (ins i64mem:$dst),
768 "sar{q}\t{%cl, $dst|$dst, %CL}",
769 [(store (sra (loadi64 addr:$dst), CL), addr:$dst)]>;
770 def SAR64mi : RIi8<0xC1, MRM7m, (outs), (ins i64mem:$dst, i8imm:$src),
771 "sar{q}\t{$src, $dst|$dst, $src}",
772 [(store (sra (loadi64 addr:$dst), (i8 imm:$src)), addr:$dst)]>;
773 def SAR64m1 : RI<0xD1, MRM7m, (outs), (ins i64mem:$dst),
775 [(store (sra (loadi64 addr:$dst), (i8 1)), addr:$dst)]>;
777 // Rotate instructions
778 let isTwoAddress = 1 in {
780 def ROL64rCL : RI<0xD3, MRM0r, (outs GR64:$dst), (ins GR64:$src),
781 "rol{q}\t{%cl, $dst|$dst, %CL}",
782 [(set GR64:$dst, (rotl GR64:$src, CL))]>;
783 def ROL64ri : RIi8<0xC1, MRM0r, (outs GR64:$dst), (ins GR64:$src1, i8imm:$src2),
784 "rol{q}\t{$src2, $dst|$dst, $src2}",
785 [(set GR64:$dst, (rotl GR64:$src1, (i8 imm:$src2)))]>;
786 def ROL64r1 : RI<0xD1, MRM0r, (outs GR64:$dst), (ins GR64:$src1),
788 [(set GR64:$dst, (rotl GR64:$src1, (i8 1)))]>;
792 def ROL64mCL : I<0xD3, MRM0m, (outs), (ins i64mem:$dst),
793 "rol{q}\t{%cl, $dst|$dst, %CL}",
794 [(store (rotl (loadi64 addr:$dst), CL), addr:$dst)]>;
795 def ROL64mi : RIi8<0xC1, MRM0m, (outs), (ins i64mem:$dst, i8imm:$src),
796 "rol{q}\t{$src, $dst|$dst, $src}",
797 [(store (rotl (loadi64 addr:$dst), (i8 imm:$src)), addr:$dst)]>;
798 def ROL64m1 : RI<0xD1, MRM0m, (outs), (ins i64mem:$dst),
800 [(store (rotl (loadi64 addr:$dst), (i8 1)), addr:$dst)]>;
802 let isTwoAddress = 1 in {
804 def ROR64rCL : RI<0xD3, MRM1r, (outs GR64:$dst), (ins GR64:$src),
805 "ror{q}\t{%cl, $dst|$dst, %CL}",
806 [(set GR64:$dst, (rotr GR64:$src, CL))]>;
807 def ROR64ri : RIi8<0xC1, MRM1r, (outs GR64:$dst), (ins GR64:$src1, i8imm:$src2),
808 "ror{q}\t{$src2, $dst|$dst, $src2}",
809 [(set GR64:$dst, (rotr GR64:$src1, (i8 imm:$src2)))]>;
810 def ROR64r1 : RI<0xD1, MRM1r, (outs GR64:$dst), (ins GR64:$src1),
812 [(set GR64:$dst, (rotr GR64:$src1, (i8 1)))]>;
816 def ROR64mCL : RI<0xD3, MRM1m, (outs), (ins i64mem:$dst),
817 "ror{q}\t{%cl, $dst|$dst, %CL}",
818 [(store (rotr (loadi64 addr:$dst), CL), addr:$dst)]>;
819 def ROR64mi : RIi8<0xC1, MRM1m, (outs), (ins i64mem:$dst, i8imm:$src),
820 "ror{q}\t{$src, $dst|$dst, $src}",
821 [(store (rotr (loadi64 addr:$dst), (i8 imm:$src)), addr:$dst)]>;
822 def ROR64m1 : RI<0xD1, MRM1m, (outs), (ins i64mem:$dst),
824 [(store (rotr (loadi64 addr:$dst), (i8 1)), addr:$dst)]>;
826 // Double shift instructions (generalizations of rotate)
827 let isTwoAddress = 1 in {
829 def SHLD64rrCL : RI<0xA5, MRMDestReg, (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
830 "shld{q}\t{%cl, $src2, $dst|$dst, $src2, %CL}",
831 [(set GR64:$dst, (X86shld GR64:$src1, GR64:$src2, CL))]>, TB;
832 def SHRD64rrCL : RI<0xAD, MRMDestReg, (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
833 "shrd{q}\t{%cl, $src2, $dst|$dst, $src2, %CL}",
834 [(set GR64:$dst, (X86shrd GR64:$src1, GR64:$src2, CL))]>, TB;
837 let isCommutable = 1 in { // FIXME: Update X86InstrInfo::commuteInstruction
838 def SHLD64rri8 : RIi8<0xA4, MRMDestReg,
839 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2, i8imm:$src3),
840 "shld{q}\t{$src3, $src2, $dst|$dst, $src2, $src3}",
841 [(set GR64:$dst, (X86shld GR64:$src1, GR64:$src2,
844 def SHRD64rri8 : RIi8<0xAC, MRMDestReg,
845 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2, i8imm:$src3),
846 "shrd{q}\t{$src3, $src2, $dst|$dst, $src2, $src3}",
847 [(set GR64:$dst, (X86shrd GR64:$src1, GR64:$src2,
854 def SHLD64mrCL : RI<0xA5, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src2),
855 "shld{q}\t{%cl, $src2, $dst|$dst, $src2, %CL}",
856 [(store (X86shld (loadi64 addr:$dst), GR64:$src2, CL),
858 def SHRD64mrCL : RI<0xAD, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src2),
859 "shrd{q}\t{%cl, $src2, $dst|$dst, $src2, %CL}",
860 [(store (X86shrd (loadi64 addr:$dst), GR64:$src2, CL),
863 def SHLD64mri8 : RIi8<0xA4, MRMDestMem,
864 (outs), (ins i64mem:$dst, GR64:$src2, i8imm:$src3),
865 "shld{q}\t{$src3, $src2, $dst|$dst, $src2, $src3}",
866 [(store (X86shld (loadi64 addr:$dst), GR64:$src2,
867 (i8 imm:$src3)), addr:$dst)]>,
869 def SHRD64mri8 : RIi8<0xAC, MRMDestMem,
870 (outs), (ins i64mem:$dst, GR64:$src2, i8imm:$src3),
871 "shrd{q}\t{$src3, $src2, $dst|$dst, $src2, $src3}",
872 [(store (X86shrd (loadi64 addr:$dst), GR64:$src2,
873 (i8 imm:$src3)), addr:$dst)]>,
877 //===----------------------------------------------------------------------===//
878 // Logical Instructions...
881 let isTwoAddress = 1 , AddedComplexity = 15 in
882 def NOT64r : RI<0xF7, MRM2r, (outs GR64:$dst), (ins GR64:$src), "not{q}\t$dst",
883 [(set GR64:$dst, (not GR64:$src))]>;
884 def NOT64m : RI<0xF7, MRM2m, (outs), (ins i64mem:$dst), "not{q}\t$dst",
885 [(store (not (loadi64 addr:$dst)), addr:$dst)]>;
887 let Defs = [EFLAGS] in {
888 def AND64i32 : RI<0x25, RawFrm, (outs), (ins i32imm:$src),
889 "and{q}\t{$src, %rax|%rax, $src}", []>;
891 let isTwoAddress = 1 in {
892 let isCommutable = 1 in
893 def AND64rr : RI<0x21, MRMDestReg,
894 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
895 "and{q}\t{$src2, $dst|$dst, $src2}",
896 [(set GR64:$dst, (and GR64:$src1, GR64:$src2)),
898 def AND64rm : RI<0x23, MRMSrcMem,
899 (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
900 "and{q}\t{$src2, $dst|$dst, $src2}",
901 [(set GR64:$dst, (and GR64:$src1, (load addr:$src2))),
903 def AND64ri8 : RIi8<0x83, MRM4r,
904 (outs GR64:$dst), (ins GR64:$src1, i64i8imm:$src2),
905 "and{q}\t{$src2, $dst|$dst, $src2}",
906 [(set GR64:$dst, (and GR64:$src1, i64immSExt8:$src2)),
908 def AND64ri32 : RIi32<0x81, MRM4r,
909 (outs GR64:$dst), (ins GR64:$src1, i64i32imm:$src2),
910 "and{q}\t{$src2, $dst|$dst, $src2}",
911 [(set GR64:$dst, (and GR64:$src1, i64immSExt32:$src2)),
915 def AND64mr : RI<0x21, MRMDestMem,
916 (outs), (ins i64mem:$dst, GR64:$src),
917 "and{q}\t{$src, $dst|$dst, $src}",
918 [(store (and (load addr:$dst), GR64:$src), addr:$dst),
920 def AND64mi8 : RIi8<0x83, MRM4m,
921 (outs), (ins i64mem:$dst, i64i8imm :$src),
922 "and{q}\t{$src, $dst|$dst, $src}",
923 [(store (and (load addr:$dst), i64immSExt8:$src), addr:$dst),
925 def AND64mi32 : RIi32<0x81, MRM4m,
926 (outs), (ins i64mem:$dst, i64i32imm:$src),
927 "and{q}\t{$src, $dst|$dst, $src}",
928 [(store (and (loadi64 addr:$dst), i64immSExt32:$src), addr:$dst),
931 let isTwoAddress = 1 in {
932 let isCommutable = 1 in
933 def OR64rr : RI<0x09, MRMDestReg, (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
934 "or{q}\t{$src2, $dst|$dst, $src2}",
935 [(set GR64:$dst, (or GR64:$src1, GR64:$src2)),
937 def OR64rm : RI<0x0B, MRMSrcMem , (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
938 "or{q}\t{$src2, $dst|$dst, $src2}",
939 [(set GR64:$dst, (or GR64:$src1, (load addr:$src2))),
941 def OR64ri8 : RIi8<0x83, MRM1r, (outs GR64:$dst), (ins GR64:$src1, i64i8imm:$src2),
942 "or{q}\t{$src2, $dst|$dst, $src2}",
943 [(set GR64:$dst, (or GR64:$src1, i64immSExt8:$src2)),
945 def OR64ri32 : RIi32<0x81, MRM1r, (outs GR64:$dst), (ins GR64:$src1, i64i32imm:$src2),
946 "or{q}\t{$src2, $dst|$dst, $src2}",
947 [(set GR64:$dst, (or GR64:$src1, i64immSExt32:$src2)),
951 def OR64mr : RI<0x09, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
952 "or{q}\t{$src, $dst|$dst, $src}",
953 [(store (or (load addr:$dst), GR64:$src), addr:$dst),
955 def OR64mi8 : RIi8<0x83, MRM1m, (outs), (ins i64mem:$dst, i64i8imm:$src),
956 "or{q}\t{$src, $dst|$dst, $src}",
957 [(store (or (load addr:$dst), i64immSExt8:$src), addr:$dst),
959 def OR64mi32 : RIi32<0x81, MRM1m, (outs), (ins i64mem:$dst, i64i32imm:$src),
960 "or{q}\t{$src, $dst|$dst, $src}",
961 [(store (or (loadi64 addr:$dst), i64immSExt32:$src), addr:$dst),
964 let isTwoAddress = 1 in {
965 let isCommutable = 1 in
966 def XOR64rr : RI<0x31, MRMDestReg, (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
967 "xor{q}\t{$src2, $dst|$dst, $src2}",
968 [(set GR64:$dst, (xor GR64:$src1, GR64:$src2)),
970 def XOR64rm : RI<0x33, MRMSrcMem, (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
971 "xor{q}\t{$src2, $dst|$dst, $src2}",
972 [(set GR64:$dst, (xor GR64:$src1, (load addr:$src2))),
974 def XOR64ri8 : RIi8<0x83, MRM6r, (outs GR64:$dst), (ins GR64:$src1, i64i8imm:$src2),
975 "xor{q}\t{$src2, $dst|$dst, $src2}",
976 [(set GR64:$dst, (xor GR64:$src1, i64immSExt8:$src2)),
978 def XOR64ri32 : RIi32<0x81, MRM6r,
979 (outs GR64:$dst), (ins GR64:$src1, i64i32imm:$src2),
980 "xor{q}\t{$src2, $dst|$dst, $src2}",
981 [(set GR64:$dst, (xor GR64:$src1, i64immSExt32:$src2)),
985 def XOR64mr : RI<0x31, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
986 "xor{q}\t{$src, $dst|$dst, $src}",
987 [(store (xor (load addr:$dst), GR64:$src), addr:$dst),
989 def XOR64mi8 : RIi8<0x83, MRM6m, (outs), (ins i64mem:$dst, i64i8imm :$src),
990 "xor{q}\t{$src, $dst|$dst, $src}",
991 [(store (xor (load addr:$dst), i64immSExt8:$src), addr:$dst),
993 def XOR64mi32 : RIi32<0x81, MRM6m, (outs), (ins i64mem:$dst, i64i32imm:$src),
994 "xor{q}\t{$src, $dst|$dst, $src}",
995 [(store (xor (loadi64 addr:$dst), i64immSExt32:$src), addr:$dst),
999 //===----------------------------------------------------------------------===//
1000 // Comparison Instructions...
1003 // Integer comparison
1004 let Defs = [EFLAGS] in {
1005 def TEST64i32 : RI<0xa9, RawFrm, (outs), (ins i32imm:$src),
1006 "test{q}\t{$src, %rax|%rax, $src}", []>;
1007 let isCommutable = 1 in
1008 def TEST64rr : RI<0x85, MRMDestReg, (outs), (ins GR64:$src1, GR64:$src2),
1009 "test{q}\t{$src2, $src1|$src1, $src2}",
1010 [(X86cmp (and GR64:$src1, GR64:$src2), 0),
1011 (implicit EFLAGS)]>;
1012 def TEST64rm : RI<0x85, MRMSrcMem, (outs), (ins GR64:$src1, i64mem:$src2),
1013 "test{q}\t{$src2, $src1|$src1, $src2}",
1014 [(X86cmp (and GR64:$src1, (loadi64 addr:$src2)), 0),
1015 (implicit EFLAGS)]>;
1016 def TEST64ri32 : RIi32<0xF7, MRM0r, (outs),
1017 (ins GR64:$src1, i64i32imm:$src2),
1018 "test{q}\t{$src2, $src1|$src1, $src2}",
1019 [(X86cmp (and GR64:$src1, i64immSExt32:$src2), 0),
1020 (implicit EFLAGS)]>;
1021 def TEST64mi32 : RIi32<0xF7, MRM0m, (outs),
1022 (ins i64mem:$src1, i64i32imm:$src2),
1023 "test{q}\t{$src2, $src1|$src1, $src2}",
1024 [(X86cmp (and (loadi64 addr:$src1), i64immSExt32:$src2), 0),
1025 (implicit EFLAGS)]>;
1028 def CMP64i32 : RI<0x3D, RawFrm, (outs), (ins i32imm:$src),
1029 "cmp{q}\t{$src, %rax|%rax, $src}", []>;
1030 def CMP64rr : RI<0x39, MRMDestReg, (outs), (ins GR64:$src1, GR64:$src2),
1031 "cmp{q}\t{$src2, $src1|$src1, $src2}",
1032 [(X86cmp GR64:$src1, GR64:$src2),
1033 (implicit EFLAGS)]>;
1034 def CMP64mr : RI<0x39, MRMDestMem, (outs), (ins i64mem:$src1, GR64:$src2),
1035 "cmp{q}\t{$src2, $src1|$src1, $src2}",
1036 [(X86cmp (loadi64 addr:$src1), GR64:$src2),
1037 (implicit EFLAGS)]>;
1038 def CMP64rm : RI<0x3B, MRMSrcMem, (outs), (ins GR64:$src1, i64mem:$src2),
1039 "cmp{q}\t{$src2, $src1|$src1, $src2}",
1040 [(X86cmp GR64:$src1, (loadi64 addr:$src2)),
1041 (implicit EFLAGS)]>;
1042 def CMP64ri8 : RIi8<0x83, MRM7r, (outs), (ins GR64:$src1, i64i8imm:$src2),
1043 "cmp{q}\t{$src2, $src1|$src1, $src2}",
1044 [(X86cmp GR64:$src1, i64immSExt8:$src2),
1045 (implicit EFLAGS)]>;
1046 def CMP64ri32 : RIi32<0x81, MRM7r, (outs), (ins GR64:$src1, i64i32imm:$src2),
1047 "cmp{q}\t{$src2, $src1|$src1, $src2}",
1048 [(X86cmp GR64:$src1, i64immSExt32:$src2),
1049 (implicit EFLAGS)]>;
1050 def CMP64mi8 : RIi8<0x83, MRM7m, (outs), (ins i64mem:$src1, i64i8imm:$src2),
1051 "cmp{q}\t{$src2, $src1|$src1, $src2}",
1052 [(X86cmp (loadi64 addr:$src1), i64immSExt8:$src2),
1053 (implicit EFLAGS)]>;
1054 def CMP64mi32 : RIi32<0x81, MRM7m, (outs),
1055 (ins i64mem:$src1, i64i32imm:$src2),
1056 "cmp{q}\t{$src2, $src1|$src1, $src2}",
1057 [(X86cmp (loadi64 addr:$src1), i64immSExt32:$src2),
1058 (implicit EFLAGS)]>;
1059 } // Defs = [EFLAGS]
1062 // TODO: BTC, BTR, and BTS
1063 let Defs = [EFLAGS] in {
1064 def BT64rr : RI<0xA3, MRMDestReg, (outs), (ins GR64:$src1, GR64:$src2),
1065 "bt{q}\t{$src2, $src1|$src1, $src2}",
1066 [(X86bt GR64:$src1, GR64:$src2),
1067 (implicit EFLAGS)]>, TB;
1069 // Unlike with the register+register form, the memory+register form of the
1070 // bt instruction does not ignore the high bits of the index. From ISel's
1071 // perspective, this is pretty bizarre. Disable these instructions for now.
1072 //def BT64mr : RI<0xA3, MRMDestMem, (outs), (ins i64mem:$src1, GR64:$src2),
1073 // "bt{q}\t{$src2, $src1|$src1, $src2}",
1074 // [(X86bt (loadi64 addr:$src1), GR64:$src2),
1075 // (implicit EFLAGS)]>, TB;
1077 def BT64ri8 : Ii8<0xBA, MRM4r, (outs), (ins GR64:$src1, i64i8imm:$src2),
1078 "bt{q}\t{$src2, $src1|$src1, $src2}",
1079 [(X86bt GR64:$src1, i64immSExt8:$src2),
1080 (implicit EFLAGS)]>, TB;
1081 // Note that these instructions don't need FastBTMem because that
1082 // only applies when the other operand is in a register. When it's
1083 // an immediate, bt is still fast.
1084 def BT64mi8 : Ii8<0xBA, MRM4m, (outs), (ins i64mem:$src1, i64i8imm:$src2),
1085 "bt{q}\t{$src2, $src1|$src1, $src2}",
1086 [(X86bt (loadi64 addr:$src1), i64immSExt8:$src2),
1087 (implicit EFLAGS)]>, TB;
1088 } // Defs = [EFLAGS]
1090 // Conditional moves
1091 let Uses = [EFLAGS], isTwoAddress = 1 in {
1092 let isCommutable = 1 in {
1093 def CMOVB64rr : RI<0x42, MRMSrcReg, // if <u, GR64 = GR64
1094 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
1095 "cmovb\t{$src2, $dst|$dst, $src2}",
1096 [(set GR64:$dst, (X86cmov GR64:$src1, GR64:$src2,
1097 X86_COND_B, EFLAGS))]>, TB;
1098 def CMOVAE64rr: RI<0x43, MRMSrcReg, // if >=u, GR64 = GR64
1099 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
1100 "cmovae\t{$src2, $dst|$dst, $src2}",
1101 [(set GR64:$dst, (X86cmov GR64:$src1, GR64:$src2,
1102 X86_COND_AE, EFLAGS))]>, TB;
1103 def CMOVE64rr : RI<0x44, MRMSrcReg, // if ==, GR64 = GR64
1104 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
1105 "cmove\t{$src2, $dst|$dst, $src2}",
1106 [(set GR64:$dst, (X86cmov GR64:$src1, GR64:$src2,
1107 X86_COND_E, EFLAGS))]>, TB;
1108 def CMOVNE64rr: RI<0x45, MRMSrcReg, // if !=, GR64 = GR64
1109 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
1110 "cmovne\t{$src2, $dst|$dst, $src2}",
1111 [(set GR64:$dst, (X86cmov GR64:$src1, GR64:$src2,
1112 X86_COND_NE, EFLAGS))]>, TB;
1113 def CMOVBE64rr: RI<0x46, MRMSrcReg, // if <=u, GR64 = GR64
1114 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
1115 "cmovbe\t{$src2, $dst|$dst, $src2}",
1116 [(set GR64:$dst, (X86cmov GR64:$src1, GR64:$src2,
1117 X86_COND_BE, EFLAGS))]>, TB;
1118 def CMOVA64rr : RI<0x47, MRMSrcReg, // if >u, GR64 = GR64
1119 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
1120 "cmova\t{$src2, $dst|$dst, $src2}",
1121 [(set GR64:$dst, (X86cmov GR64:$src1, GR64:$src2,
1122 X86_COND_A, EFLAGS))]>, TB;
1123 def CMOVL64rr : RI<0x4C, MRMSrcReg, // if <s, GR64 = GR64
1124 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
1125 "cmovl\t{$src2, $dst|$dst, $src2}",
1126 [(set GR64:$dst, (X86cmov GR64:$src1, GR64:$src2,
1127 X86_COND_L, EFLAGS))]>, TB;
1128 def CMOVGE64rr: RI<0x4D, MRMSrcReg, // if >=s, GR64 = GR64
1129 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
1130 "cmovge\t{$src2, $dst|$dst, $src2}",
1131 [(set GR64:$dst, (X86cmov GR64:$src1, GR64:$src2,
1132 X86_COND_GE, EFLAGS))]>, TB;
1133 def CMOVLE64rr: RI<0x4E, MRMSrcReg, // if <=s, GR64 = GR64
1134 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
1135 "cmovle\t{$src2, $dst|$dst, $src2}",
1136 [(set GR64:$dst, (X86cmov GR64:$src1, GR64:$src2,
1137 X86_COND_LE, EFLAGS))]>, TB;
1138 def CMOVG64rr : RI<0x4F, MRMSrcReg, // if >s, GR64 = GR64
1139 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
1140 "cmovg\t{$src2, $dst|$dst, $src2}",
1141 [(set GR64:$dst, (X86cmov GR64:$src1, GR64:$src2,
1142 X86_COND_G, EFLAGS))]>, TB;
1143 def CMOVS64rr : RI<0x48, MRMSrcReg, // if signed, GR64 = GR64
1144 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
1145 "cmovs\t{$src2, $dst|$dst, $src2}",
1146 [(set GR64:$dst, (X86cmov GR64:$src1, GR64:$src2,
1147 X86_COND_S, EFLAGS))]>, TB;
1148 def CMOVNS64rr: RI<0x49, MRMSrcReg, // if !signed, GR64 = GR64
1149 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
1150 "cmovns\t{$src2, $dst|$dst, $src2}",
1151 [(set GR64:$dst, (X86cmov GR64:$src1, GR64:$src2,
1152 X86_COND_NS, EFLAGS))]>, TB;
1153 def CMOVP64rr : RI<0x4A, MRMSrcReg, // if parity, GR64 = GR64
1154 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
1155 "cmovp\t{$src2, $dst|$dst, $src2}",
1156 [(set GR64:$dst, (X86cmov GR64:$src1, GR64:$src2,
1157 X86_COND_P, EFLAGS))]>, TB;
1158 def CMOVNP64rr : RI<0x4B, MRMSrcReg, // if !parity, GR64 = GR64
1159 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
1160 "cmovnp\t{$src2, $dst|$dst, $src2}",
1161 [(set GR64:$dst, (X86cmov GR64:$src1, GR64:$src2,
1162 X86_COND_NP, EFLAGS))]>, TB;
1163 def CMOVO64rr : RI<0x40, MRMSrcReg, // if overflow, GR64 = GR64
1164 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
1165 "cmovo\t{$src2, $dst|$dst, $src2}",
1166 [(set GR64:$dst, (X86cmov GR64:$src1, GR64:$src2,
1167 X86_COND_O, EFLAGS))]>, TB;
1168 def CMOVNO64rr : RI<0x41, MRMSrcReg, // if !overflow, GR64 = GR64
1169 (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
1170 "cmovno\t{$src2, $dst|$dst, $src2}",
1171 [(set GR64:$dst, (X86cmov GR64:$src1, GR64:$src2,
1172 X86_COND_NO, EFLAGS))]>, TB;
1173 } // isCommutable = 1
1175 def CMOVB64rm : RI<0x42, MRMSrcMem, // if <u, GR64 = [mem64]
1176 (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
1177 "cmovb\t{$src2, $dst|$dst, $src2}",
1178 [(set GR64:$dst, (X86cmov GR64:$src1, (loadi64 addr:$src2),
1179 X86_COND_B, EFLAGS))]>, TB;
1180 def CMOVAE64rm: RI<0x43, MRMSrcMem, // if >=u, GR64 = [mem64]
1181 (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
1182 "cmovae\t{$src2, $dst|$dst, $src2}",
1183 [(set GR64:$dst, (X86cmov GR64:$src1, (loadi64 addr:$src2),
1184 X86_COND_AE, EFLAGS))]>, TB;
1185 def CMOVE64rm : RI<0x44, MRMSrcMem, // if ==, GR64 = [mem64]
1186 (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
1187 "cmove\t{$src2, $dst|$dst, $src2}",
1188 [(set GR64:$dst, (X86cmov GR64:$src1, (loadi64 addr:$src2),
1189 X86_COND_E, EFLAGS))]>, TB;
1190 def CMOVNE64rm: RI<0x45, MRMSrcMem, // if !=, GR64 = [mem64]
1191 (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
1192 "cmovne\t{$src2, $dst|$dst, $src2}",
1193 [(set GR64:$dst, (X86cmov GR64:$src1, (loadi64 addr:$src2),
1194 X86_COND_NE, EFLAGS))]>, TB;
1195 def CMOVBE64rm: RI<0x46, MRMSrcMem, // if <=u, GR64 = [mem64]
1196 (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
1197 "cmovbe\t{$src2, $dst|$dst, $src2}",
1198 [(set GR64:$dst, (X86cmov GR64:$src1, (loadi64 addr:$src2),
1199 X86_COND_BE, EFLAGS))]>, TB;
1200 def CMOVA64rm : RI<0x47, MRMSrcMem, // if >u, GR64 = [mem64]
1201 (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
1202 "cmova\t{$src2, $dst|$dst, $src2}",
1203 [(set GR64:$dst, (X86cmov GR64:$src1, (loadi64 addr:$src2),
1204 X86_COND_A, EFLAGS))]>, TB;
1205 def CMOVL64rm : RI<0x4C, MRMSrcMem, // if <s, GR64 = [mem64]
1206 (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
1207 "cmovl\t{$src2, $dst|$dst, $src2}",
1208 [(set GR64:$dst, (X86cmov GR64:$src1, (loadi64 addr:$src2),
1209 X86_COND_L, EFLAGS))]>, TB;
1210 def CMOVGE64rm: RI<0x4D, MRMSrcMem, // if >=s, GR64 = [mem64]
1211 (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
1212 "cmovge\t{$src2, $dst|$dst, $src2}",
1213 [(set GR64:$dst, (X86cmov GR64:$src1, (loadi64 addr:$src2),
1214 X86_COND_GE, EFLAGS))]>, TB;
1215 def CMOVLE64rm: RI<0x4E, MRMSrcMem, // if <=s, GR64 = [mem64]
1216 (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
1217 "cmovle\t{$src2, $dst|$dst, $src2}",
1218 [(set GR64:$dst, (X86cmov GR64:$src1, (loadi64 addr:$src2),
1219 X86_COND_LE, EFLAGS))]>, TB;
1220 def CMOVG64rm : RI<0x4F, MRMSrcMem, // if >s, GR64 = [mem64]
1221 (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
1222 "cmovg\t{$src2, $dst|$dst, $src2}",
1223 [(set GR64:$dst, (X86cmov GR64:$src1, (loadi64 addr:$src2),
1224 X86_COND_G, EFLAGS))]>, TB;
1225 def CMOVS64rm : RI<0x48, MRMSrcMem, // if signed, GR64 = [mem64]
1226 (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
1227 "cmovs\t{$src2, $dst|$dst, $src2}",
1228 [(set GR64:$dst, (X86cmov GR64:$src1, (loadi64 addr:$src2),
1229 X86_COND_S, EFLAGS))]>, TB;
1230 def CMOVNS64rm: RI<0x49, MRMSrcMem, // if !signed, GR64 = [mem64]
1231 (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
1232 "cmovns\t{$src2, $dst|$dst, $src2}",
1233 [(set GR64:$dst, (X86cmov GR64:$src1, (loadi64 addr:$src2),
1234 X86_COND_NS, EFLAGS))]>, TB;
1235 def CMOVP64rm : RI<0x4A, MRMSrcMem, // if parity, GR64 = [mem64]
1236 (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
1237 "cmovp\t{$src2, $dst|$dst, $src2}",
1238 [(set GR64:$dst, (X86cmov GR64:$src1, (loadi64 addr:$src2),
1239 X86_COND_P, EFLAGS))]>, TB;
1240 def CMOVNP64rm : RI<0x4B, MRMSrcMem, // if !parity, GR64 = [mem64]
1241 (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
1242 "cmovnp\t{$src2, $dst|$dst, $src2}",
1243 [(set GR64:$dst, (X86cmov GR64:$src1, (loadi64 addr:$src2),
1244 X86_COND_NP, EFLAGS))]>, TB;
1245 def CMOVO64rm : RI<0x40, MRMSrcMem, // if overflow, GR64 = [mem64]
1246 (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
1247 "cmovo\t{$src2, $dst|$dst, $src2}",
1248 [(set GR64:$dst, (X86cmov GR64:$src1, (loadi64 addr:$src2),
1249 X86_COND_O, EFLAGS))]>, TB;
1250 def CMOVNO64rm : RI<0x41, MRMSrcMem, // if !overflow, GR64 = [mem64]
1251 (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
1252 "cmovno\t{$src2, $dst|$dst, $src2}",
1253 [(set GR64:$dst, (X86cmov GR64:$src1, (loadi64 addr:$src2),
1254 X86_COND_NO, EFLAGS))]>, TB;
1257 //===----------------------------------------------------------------------===//
1258 // Conversion Instructions...
1261 // f64 -> signed i64
1262 def Int_CVTSD2SI64rr: RSDI<0x2D, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1263 "cvtsd2si{q}\t{$src, $dst|$dst, $src}",
1265 (int_x86_sse2_cvtsd2si64 VR128:$src))]>;
1266 def Int_CVTSD2SI64rm: RSDI<0x2D, MRMSrcMem, (outs GR64:$dst), (ins f128mem:$src),
1267 "cvtsd2si{q}\t{$src, $dst|$dst, $src}",
1268 [(set GR64:$dst, (int_x86_sse2_cvtsd2si64
1269 (load addr:$src)))]>;
1270 def CVTTSD2SI64rr: RSDI<0x2C, MRMSrcReg, (outs GR64:$dst), (ins FR64:$src),
1271 "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1272 [(set GR64:$dst, (fp_to_sint FR64:$src))]>;
1273 def CVTTSD2SI64rm: RSDI<0x2C, MRMSrcMem, (outs GR64:$dst), (ins f64mem:$src),
1274 "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1275 [(set GR64:$dst, (fp_to_sint (loadf64 addr:$src)))]>;
1276 def Int_CVTTSD2SI64rr: RSDI<0x2C, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1277 "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1279 (int_x86_sse2_cvttsd2si64 VR128:$src))]>;
1280 def Int_CVTTSD2SI64rm: RSDI<0x2C, MRMSrcMem, (outs GR64:$dst), (ins f128mem:$src),
1281 "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1283 (int_x86_sse2_cvttsd2si64
1284 (load addr:$src)))]>;
1286 // Signed i64 -> f64
1287 def CVTSI2SD64rr: RSDI<0x2A, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
1288 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
1289 [(set FR64:$dst, (sint_to_fp GR64:$src))]>;
1290 def CVTSI2SD64rm: RSDI<0x2A, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
1291 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
1292 [(set FR64:$dst, (sint_to_fp (loadi64 addr:$src)))]>;
1294 let isTwoAddress = 1 in {
1295 def Int_CVTSI2SD64rr: RSDI<0x2A, MRMSrcReg,
1296 (outs VR128:$dst), (ins VR128:$src1, GR64:$src2),
1297 "cvtsi2sd{q}\t{$src2, $dst|$dst, $src2}",
1299 (int_x86_sse2_cvtsi642sd VR128:$src1,
1301 def Int_CVTSI2SD64rm: RSDI<0x2A, MRMSrcMem,
1302 (outs VR128:$dst), (ins VR128:$src1, i64mem:$src2),
1303 "cvtsi2sd{q}\t{$src2, $dst|$dst, $src2}",
1305 (int_x86_sse2_cvtsi642sd VR128:$src1,
1306 (loadi64 addr:$src2)))]>;
1309 // Signed i64 -> f32
1310 def CVTSI2SS64rr: RSSI<0x2A, MRMSrcReg, (outs FR32:$dst), (ins GR64:$src),
1311 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
1312 [(set FR32:$dst, (sint_to_fp GR64:$src))]>;
1313 def CVTSI2SS64rm: RSSI<0x2A, MRMSrcMem, (outs FR32:$dst), (ins i64mem:$src),
1314 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
1315 [(set FR32:$dst, (sint_to_fp (loadi64 addr:$src)))]>;
1317 let isTwoAddress = 1 in {
1318 def Int_CVTSI2SS64rr : RSSI<0x2A, MRMSrcReg,
1319 (outs VR128:$dst), (ins VR128:$src1, GR64:$src2),
1320 "cvtsi2ss{q}\t{$src2, $dst|$dst, $src2}",
1322 (int_x86_sse_cvtsi642ss VR128:$src1,
1324 def Int_CVTSI2SS64rm : RSSI<0x2A, MRMSrcMem,
1325 (outs VR128:$dst), (ins VR128:$src1, i64mem:$src2),
1326 "cvtsi2ss{q}\t{$src2, $dst|$dst, $src2}",
1328 (int_x86_sse_cvtsi642ss VR128:$src1,
1329 (loadi64 addr:$src2)))]>;
1332 // f32 -> signed i64
1333 def Int_CVTSS2SI64rr: RSSI<0x2D, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1334 "cvtss2si{q}\t{$src, $dst|$dst, $src}",
1336 (int_x86_sse_cvtss2si64 VR128:$src))]>;
1337 def Int_CVTSS2SI64rm: RSSI<0x2D, MRMSrcMem, (outs GR64:$dst), (ins f32mem:$src),
1338 "cvtss2si{q}\t{$src, $dst|$dst, $src}",
1339 [(set GR64:$dst, (int_x86_sse_cvtss2si64
1340 (load addr:$src)))]>;
1341 def CVTTSS2SI64rr: RSSI<0x2C, MRMSrcReg, (outs GR64:$dst), (ins FR32:$src),
1342 "cvttss2si{q}\t{$src, $dst|$dst, $src}",
1343 [(set GR64:$dst, (fp_to_sint FR32:$src))]>;
1344 def CVTTSS2SI64rm: RSSI<0x2C, MRMSrcMem, (outs GR64:$dst), (ins f32mem:$src),
1345 "cvttss2si{q}\t{$src, $dst|$dst, $src}",
1346 [(set GR64:$dst, (fp_to_sint (loadf32 addr:$src)))]>;
1347 def Int_CVTTSS2SI64rr: RSSI<0x2C, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1348 "cvttss2si{q}\t{$src, $dst|$dst, $src}",
1350 (int_x86_sse_cvttss2si64 VR128:$src))]>;
1351 def Int_CVTTSS2SI64rm: RSSI<0x2C, MRMSrcMem, (outs GR64:$dst), (ins f32mem:$src),
1352 "cvttss2si{q}\t{$src, $dst|$dst, $src}",
1354 (int_x86_sse_cvttss2si64 (load addr:$src)))]>;
1356 //===----------------------------------------------------------------------===//
1357 // Alias Instructions
1358 //===----------------------------------------------------------------------===//
1360 // Alias instructions that map movr0 to xor. Use xorl instead of xorq; it's
1361 // equivalent due to implicit zero-extending, and it sometimes has a smaller
1363 // FIXME: AddedComplexity gives this a higher priority than MOV64ri32. Remove
1364 // when we have a better way to specify isel priority.
1365 let AddedComplexity = 1 in
1367 (SUBREG_TO_REG (i64 0), (MOV32r0), x86_subreg_32bit)>;
1370 // Materialize i64 constant where top 32-bits are zero.
1371 let AddedComplexity = 1, isReMaterializable = 1, isAsCheapAsAMove = 1 in
1372 def MOV64ri64i32 : Ii32<0xB8, AddRegFrm, (outs GR64:$dst), (ins i64i32imm:$src),
1373 "mov{l}\t{$src, ${dst:subreg32}|${dst:subreg32}, $src}",
1374 [(set GR64:$dst, i64immZExt32:$src)]>;
1376 //===----------------------------------------------------------------------===//
1377 // Thread Local Storage Instructions
1378 //===----------------------------------------------------------------------===//
1380 // All calls clobber the non-callee saved registers. RSP is marked as
1381 // a use to prevent stack-pointer assignments that appear immediately
1382 // before calls from potentially appearing dead.
1383 let Defs = [RAX, RCX, RDX, RSI, RDI, R8, R9, R10, R11,
1384 FP0, FP1, FP2, FP3, FP4, FP5, FP6, ST0, ST1,
1385 MM0, MM1, MM2, MM3, MM4, MM5, MM6, MM7,
1386 XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, XMM6, XMM7,
1387 XMM8, XMM9, XMM10, XMM11, XMM12, XMM13, XMM14, XMM15, EFLAGS],
1389 def TLS_addr64 : I<0, Pseudo, (outs), (ins lea64mem:$sym),
1391 "leaq\t$sym(%rip), %rdi; "
1394 "call\t__tls_get_addr@PLT",
1395 [(X86tlsaddr tls64addr:$sym)]>,
1396 Requires<[In64BitMode]>;
1398 let AddedComplexity = 5, isCodeGenOnly = 1 in
1399 def MOV64GSrm : RI<0x8B, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
1400 "movq\t%gs:$src, $dst",
1401 [(set GR64:$dst, (gsload addr:$src))]>, SegGS;
1403 let AddedComplexity = 5, isCodeGenOnly = 1 in
1404 def MOV64FSrm : RI<0x8B, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
1405 "movq\t%fs:$src, $dst",
1406 [(set GR64:$dst, (fsload addr:$src))]>, SegFS;
1408 //===----------------------------------------------------------------------===//
1409 // Atomic Instructions
1410 //===----------------------------------------------------------------------===//
1412 let Defs = [RAX, EFLAGS], Uses = [RAX] in {
1413 def LCMPXCHG64 : RI<0xB1, MRMDestMem, (outs), (ins i64mem:$ptr, GR64:$swap),
1415 "cmpxchgq\t$swap,$ptr",
1416 [(X86cas addr:$ptr, GR64:$swap, 8)]>, TB, LOCK;
1419 let Constraints = "$val = $dst" in {
1420 let Defs = [EFLAGS] in
1421 def LXADD64 : RI<0xC1, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$ptr,GR64:$val),
1424 [(set GR64:$dst, (atomic_load_add_64 addr:$ptr, GR64:$val))]>,
1427 def XCHG64rm : RI<0x87, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$ptr,GR64:$val),
1429 [(set GR64:$dst, (atomic_swap_64 addr:$ptr, GR64:$val))]>;
1432 // Optimized codegen when the non-memory output is not used.
1433 // FIXME: Use normal add / sub instructions and add lock prefix dynamically.
1434 def LOCK_ADD64mr : RI<0x03, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src2),
1436 "add{q}\t{$src2, $dst|$dst, $src2}", []>, LOCK;
1437 def LOCK_ADD64mi8 : RIi8<0x83, MRM0m, (outs),
1438 (ins i64mem:$dst, i64i8imm :$src2),
1440 "add{q}\t{$src2, $dst|$dst, $src2}", []>, LOCK;
1441 def LOCK_ADD64mi32 : RIi32<0x81, MRM0m, (outs),
1442 (ins i64mem:$dst, i64i32imm :$src2),
1444 "add{q}\t{$src2, $dst|$dst, $src2}", []>, LOCK;
1445 def LOCK_SUB64mr : RI<0x29, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src2),
1447 "sub{q}\t{$src2, $dst|$dst, $src2}", []>, LOCK;
1448 def LOCK_SUB64mi8 : RIi8<0x83, MRM5m, (outs),
1449 (ins i64mem:$dst, i64i8imm :$src2),
1451 "sub{q}\t{$src2, $dst|$dst, $src2}", []>, LOCK;
1452 def LOCK_SUB64mi32 : RIi32<0x81, MRM5m, (outs),
1453 (ins i64mem:$dst, i64i32imm:$src2),
1455 "sub{q}\t{$src2, $dst|$dst, $src2}", []>, LOCK;
1456 def LOCK_INC64m : RI<0xFF, MRM0m, (outs), (ins i64mem:$dst),
1458 "inc{q}\t$dst", []>, LOCK;
1459 def LOCK_DEC64m : RI<0xFF, MRM1m, (outs), (ins i64mem:$dst),
1461 "dec{q}\t$dst", []>, LOCK;
1463 // Atomic exchange, and, or, xor
1464 let Constraints = "$val = $dst", Defs = [EFLAGS],
1465 usesCustomDAGSchedInserter = 1 in {
1466 def ATOMAND64 : I<0, Pseudo, (outs GR64:$dst),(ins i64mem:$ptr, GR64:$val),
1467 "#ATOMAND64 PSEUDO!",
1468 [(set GR64:$dst, (atomic_load_and_64 addr:$ptr, GR64:$val))]>;
1469 def ATOMOR64 : I<0, Pseudo, (outs GR64:$dst),(ins i64mem:$ptr, GR64:$val),
1470 "#ATOMOR64 PSEUDO!",
1471 [(set GR64:$dst, (atomic_load_or_64 addr:$ptr, GR64:$val))]>;
1472 def ATOMXOR64 : I<0, Pseudo,(outs GR64:$dst),(ins i64mem:$ptr, GR64:$val),
1473 "#ATOMXOR64 PSEUDO!",
1474 [(set GR64:$dst, (atomic_load_xor_64 addr:$ptr, GR64:$val))]>;
1475 def ATOMNAND64 : I<0, Pseudo,(outs GR64:$dst),(ins i64mem:$ptr, GR64:$val),
1476 "#ATOMNAND64 PSEUDO!",
1477 [(set GR64:$dst, (atomic_load_nand_64 addr:$ptr, GR64:$val))]>;
1478 def ATOMMIN64: I<0, Pseudo, (outs GR64:$dst), (ins i64mem:$ptr, GR64:$val),
1479 "#ATOMMIN64 PSEUDO!",
1480 [(set GR64:$dst, (atomic_load_min_64 addr:$ptr, GR64:$val))]>;
1481 def ATOMMAX64: I<0, Pseudo, (outs GR64:$dst),(ins i64mem:$ptr, GR64:$val),
1482 "#ATOMMAX64 PSEUDO!",
1483 [(set GR64:$dst, (atomic_load_max_64 addr:$ptr, GR64:$val))]>;
1484 def ATOMUMIN64: I<0, Pseudo, (outs GR64:$dst),(ins i64mem:$ptr, GR64:$val),
1485 "#ATOMUMIN64 PSEUDO!",
1486 [(set GR64:$dst, (atomic_load_umin_64 addr:$ptr, GR64:$val))]>;
1487 def ATOMUMAX64: I<0, Pseudo, (outs GR64:$dst),(ins i64mem:$ptr, GR64:$val),
1488 "#ATOMUMAX64 PSEUDO!",
1489 [(set GR64:$dst, (atomic_load_umax_64 addr:$ptr, GR64:$val))]>;
1492 //===----------------------------------------------------------------------===//
1493 // Non-Instruction Patterns
1494 //===----------------------------------------------------------------------===//
1496 // ConstantPool GlobalAddress, ExternalSymbol, and JumpTable when not in small
1497 // code model mode, should use 'movabs'. FIXME: This is really a hack, the
1498 // 'movabs' predicate should handle this sort of thing.
1499 def : Pat<(i64 (X86Wrapper tconstpool :$dst)),
1500 (MOV64ri tconstpool :$dst)>, Requires<[FarData]>;
1501 def : Pat<(i64 (X86Wrapper tjumptable :$dst)),
1502 (MOV64ri tjumptable :$dst)>, Requires<[FarData]>;
1503 def : Pat<(i64 (X86Wrapper tglobaladdr :$dst)),
1504 (MOV64ri tglobaladdr :$dst)>, Requires<[FarData]>;
1505 def : Pat<(i64 (X86Wrapper texternalsym:$dst)),
1506 (MOV64ri texternalsym:$dst)>, Requires<[FarData]>;
1508 // In static codegen with small code model, we can get the address of a label
1509 // into a register with 'movl'. FIXME: This is a hack, the 'imm' predicate of
1510 // the MOV64ri64i32 should accept these.
1511 def : Pat<(i64 (X86Wrapper tconstpool :$dst)),
1512 (MOV64ri64i32 tconstpool :$dst)>, Requires<[SmallCode]>;
1513 def : Pat<(i64 (X86Wrapper tjumptable :$dst)),
1514 (MOV64ri64i32 tjumptable :$dst)>, Requires<[SmallCode]>;
1515 def : Pat<(i64 (X86Wrapper tglobaladdr :$dst)),
1516 (MOV64ri64i32 tglobaladdr :$dst)>, Requires<[SmallCode]>;
1517 def : Pat<(i64 (X86Wrapper texternalsym:$dst)),
1518 (MOV64ri64i32 texternalsym:$dst)>, Requires<[SmallCode]>;
1520 // In kernel code model, we can get the address of a label
1521 // into a register with 'movq'. FIXME: This is a hack, the 'imm' predicate of
1522 // the MOV64ri32 should accept these.
1523 def : Pat<(i64 (X86Wrapper tconstpool :$dst)),
1524 (MOV64ri32 tconstpool :$dst)>, Requires<[KernelCode]>;
1525 def : Pat<(i64 (X86Wrapper tjumptable :$dst)),
1526 (MOV64ri32 tjumptable :$dst)>, Requires<[KernelCode]>;
1527 def : Pat<(i64 (X86Wrapper tglobaladdr :$dst)),
1528 (MOV64ri32 tglobaladdr :$dst)>, Requires<[KernelCode]>;
1529 def : Pat<(i64 (X86Wrapper texternalsym:$dst)),
1530 (MOV64ri32 texternalsym:$dst)>, Requires<[KernelCode]>;
1532 // If we have small model and -static mode, it is safe to store global addresses
1533 // directly as immediates. FIXME: This is really a hack, the 'imm' predicate
1534 // for MOV64mi32 should handle this sort of thing.
1535 def : Pat<(store (i64 (X86Wrapper tconstpool:$src)), addr:$dst),
1536 (MOV64mi32 addr:$dst, tconstpool:$src)>,
1537 Requires<[NearData, IsStatic]>;
1538 def : Pat<(store (i64 (X86Wrapper tjumptable:$src)), addr:$dst),
1539 (MOV64mi32 addr:$dst, tjumptable:$src)>,
1540 Requires<[NearData, IsStatic]>;
1541 def : Pat<(store (i64 (X86Wrapper tglobaladdr:$src)), addr:$dst),
1542 (MOV64mi32 addr:$dst, tglobaladdr:$src)>,
1543 Requires<[NearData, IsStatic]>;
1544 def : Pat<(store (i64 (X86Wrapper texternalsym:$src)), addr:$dst),
1545 (MOV64mi32 addr:$dst, texternalsym:$src)>,
1546 Requires<[NearData, IsStatic]>;
1549 // Direct PC relative function call for small code model. 32-bit displacement
1550 // sign extended to 64-bit.
1551 def : Pat<(X86call (i64 tglobaladdr:$dst)),
1552 (CALL64pcrel32 tglobaladdr:$dst)>, Requires<[NotWin64]>;
1553 def : Pat<(X86call (i64 texternalsym:$dst)),
1554 (CALL64pcrel32 texternalsym:$dst)>, Requires<[NotWin64]>;
1556 def : Pat<(X86call (i64 tglobaladdr:$dst)),
1557 (WINCALL64pcrel32 tglobaladdr:$dst)>, Requires<[IsWin64]>;
1558 def : Pat<(X86call (i64 texternalsym:$dst)),
1559 (WINCALL64pcrel32 texternalsym:$dst)>, Requires<[IsWin64]>;
1562 def : Pat<(X86tcret GR64:$dst, imm:$off),
1563 (TCRETURNri64 GR64:$dst, imm:$off)>;
1565 def : Pat<(X86tcret (i64 tglobaladdr:$dst), imm:$off),
1566 (TCRETURNdi64 texternalsym:$dst, imm:$off)>;
1568 def : Pat<(X86tcret (i64 texternalsym:$dst), imm:$off),
1569 (TCRETURNdi64 texternalsym:$dst, imm:$off)>;
1573 // TEST R,R is smaller than CMP R,0
1574 def : Pat<(parallel (X86cmp GR64:$src1, 0), (implicit EFLAGS)),
1575 (TEST64rr GR64:$src1, GR64:$src1)>;
1577 // Conditional moves with folded loads with operands swapped and conditions
1579 def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, X86_COND_B, EFLAGS),
1580 (CMOVAE64rm GR64:$src2, addr:$src1)>;
1581 def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, X86_COND_AE, EFLAGS),
1582 (CMOVB64rm GR64:$src2, addr:$src1)>;
1583 def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, X86_COND_E, EFLAGS),
1584 (CMOVNE64rm GR64:$src2, addr:$src1)>;
1585 def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, X86_COND_NE, EFLAGS),
1586 (CMOVE64rm GR64:$src2, addr:$src1)>;
1587 def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, X86_COND_BE, EFLAGS),
1588 (CMOVA64rm GR64:$src2, addr:$src1)>;
1589 def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, X86_COND_A, EFLAGS),
1590 (CMOVBE64rm GR64:$src2, addr:$src1)>;
1591 def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, X86_COND_L, EFLAGS),
1592 (CMOVGE64rm GR64:$src2, addr:$src1)>;
1593 def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, X86_COND_GE, EFLAGS),
1594 (CMOVL64rm GR64:$src2, addr:$src1)>;
1595 def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, X86_COND_LE, EFLAGS),
1596 (CMOVG64rm GR64:$src2, addr:$src1)>;
1597 def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, X86_COND_G, EFLAGS),
1598 (CMOVLE64rm GR64:$src2, addr:$src1)>;
1599 def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, X86_COND_P, EFLAGS),
1600 (CMOVNP64rm GR64:$src2, addr:$src1)>;
1601 def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, X86_COND_NP, EFLAGS),
1602 (CMOVP64rm GR64:$src2, addr:$src1)>;
1603 def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, X86_COND_S, EFLAGS),
1604 (CMOVNS64rm GR64:$src2, addr:$src1)>;
1605 def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, X86_COND_NS, EFLAGS),
1606 (CMOVS64rm GR64:$src2, addr:$src1)>;
1607 def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, X86_COND_O, EFLAGS),
1608 (CMOVNO64rm GR64:$src2, addr:$src1)>;
1609 def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, X86_COND_NO, EFLAGS),
1610 (CMOVO64rm GR64:$src2, addr:$src1)>;
1612 // zextload bool -> zextload byte
1613 def : Pat<(zextloadi64i1 addr:$src), (MOVZX64rm8 addr:$src)>;
1616 // When extloading from 16-bit and smaller memory locations into 64-bit registers,
1617 // use zero-extending loads so that the entire 64-bit register is defined, avoiding
1618 // partial-register updates.
1619 def : Pat<(extloadi64i1 addr:$src), (MOVZX64rm8 addr:$src)>;
1620 def : Pat<(extloadi64i8 addr:$src), (MOVZX64rm8 addr:$src)>;
1621 def : Pat<(extloadi64i16 addr:$src), (MOVZX64rm16 addr:$src)>;
1622 // For other extloads, use subregs, since the high contents of the register are
1623 // defined after an extload.
1624 def : Pat<(extloadi64i32 addr:$src),
1625 (SUBREG_TO_REG (i64 0), (MOV32rm addr:$src),
1628 // anyext. Define these to do an explicit zero-extend to
1629 // avoid partial-register updates.
1630 def : Pat<(i64 (anyext GR8 :$src)), (MOVZX64rr8 GR8 :$src)>;
1631 def : Pat<(i64 (anyext GR16:$src)), (MOVZX64rr16 GR16 :$src)>;
1632 def : Pat<(i64 (anyext GR32:$src)),
1633 (SUBREG_TO_REG (i64 0), GR32:$src, x86_subreg_32bit)>;
1635 //===----------------------------------------------------------------------===//
1637 //===----------------------------------------------------------------------===//
1639 // Odd encoding trick: -128 fits into an 8-bit immediate field while
1640 // +128 doesn't, so in this special case use a sub instead of an add.
1641 def : Pat<(add GR64:$src1, 128),
1642 (SUB64ri8 GR64:$src1, -128)>;
1643 def : Pat<(store (add (loadi64 addr:$dst), 128), addr:$dst),
1644 (SUB64mi8 addr:$dst, -128)>;
1646 // The same trick applies for 32-bit immediate fields in 64-bit
1648 def : Pat<(add GR64:$src1, 0x0000000080000000),
1649 (SUB64ri32 GR64:$src1, 0xffffffff80000000)>;
1650 def : Pat<(store (add (loadi64 addr:$dst), 0x00000000800000000), addr:$dst),
1651 (SUB64mi32 addr:$dst, 0xffffffff80000000)>;
1653 // r & (2^32-1) ==> movz
1654 def : Pat<(and GR64:$src, 0x00000000FFFFFFFF),
1655 (MOVZX64rr32 (EXTRACT_SUBREG GR64:$src, x86_subreg_32bit))>;
1656 // r & (2^16-1) ==> movz
1657 def : Pat<(and GR64:$src, 0xffff),
1658 (MOVZX64rr16 (i16 (EXTRACT_SUBREG GR64:$src, x86_subreg_16bit)))>;
1659 // r & (2^8-1) ==> movz
1660 def : Pat<(and GR64:$src, 0xff),
1661 (MOVZX64rr8 (i8 (EXTRACT_SUBREG GR64:$src, x86_subreg_8bit)))>;
1662 // r & (2^8-1) ==> movz
1663 def : Pat<(and GR32:$src1, 0xff),
1664 (MOVZX32rr8 (EXTRACT_SUBREG GR32:$src1, x86_subreg_8bit))>,
1665 Requires<[In64BitMode]>;
1666 // r & (2^8-1) ==> movz
1667 def : Pat<(and GR16:$src1, 0xff),
1668 (MOVZX16rr8 (i8 (EXTRACT_SUBREG GR16:$src1, x86_subreg_8bit)))>,
1669 Requires<[In64BitMode]>;
1671 // sext_inreg patterns
1672 def : Pat<(sext_inreg GR64:$src, i32),
1673 (MOVSX64rr32 (EXTRACT_SUBREG GR64:$src, x86_subreg_32bit))>;
1674 def : Pat<(sext_inreg GR64:$src, i16),
1675 (MOVSX64rr16 (EXTRACT_SUBREG GR64:$src, x86_subreg_16bit))>;
1676 def : Pat<(sext_inreg GR64:$src, i8),
1677 (MOVSX64rr8 (EXTRACT_SUBREG GR64:$src, x86_subreg_8bit))>;
1678 def : Pat<(sext_inreg GR32:$src, i8),
1679 (MOVSX32rr8 (EXTRACT_SUBREG GR32:$src, x86_subreg_8bit))>,
1680 Requires<[In64BitMode]>;
1681 def : Pat<(sext_inreg GR16:$src, i8),
1682 (MOVSX16rr8 (i8 (EXTRACT_SUBREG GR16:$src, x86_subreg_8bit)))>,
1683 Requires<[In64BitMode]>;
1686 def : Pat<(i32 (trunc GR64:$src)),
1687 (EXTRACT_SUBREG GR64:$src, x86_subreg_32bit)>;
1688 def : Pat<(i16 (trunc GR64:$src)),
1689 (EXTRACT_SUBREG GR64:$src, x86_subreg_16bit)>;
1690 def : Pat<(i8 (trunc GR64:$src)),
1691 (EXTRACT_SUBREG GR64:$src, x86_subreg_8bit)>;
1692 def : Pat<(i8 (trunc GR32:$src)),
1693 (EXTRACT_SUBREG GR32:$src, x86_subreg_8bit)>,
1694 Requires<[In64BitMode]>;
1695 def : Pat<(i8 (trunc GR16:$src)),
1696 (EXTRACT_SUBREG GR16:$src, x86_subreg_8bit)>,
1697 Requires<[In64BitMode]>;
1699 // h-register tricks.
1700 // For now, be conservative on x86-64 and use an h-register extract only if the
1701 // value is immediately zero-extended or stored, which are somewhat common
1702 // cases. This uses a bunch of code to prevent a register requiring a REX prefix
1703 // from being allocated in the same instruction as the h register, as there's
1704 // currently no way to describe this requirement to the register allocator.
1706 // h-register extract and zero-extend.
1707 def : Pat<(and (srl_su GR64:$src, (i8 8)), (i64 255)),
1711 (EXTRACT_SUBREG (COPY_TO_REGCLASS GR64:$src, GR64_ABCD),
1712 x86_subreg_8bit_hi)),
1714 def : Pat<(and (srl_su GR32:$src, (i8 8)), (i32 255)),
1716 (EXTRACT_SUBREG (COPY_TO_REGCLASS GR32:$src, GR32_ABCD),
1717 x86_subreg_8bit_hi))>,
1718 Requires<[In64BitMode]>;
1719 def : Pat<(srl_su GR16:$src, (i8 8)),
1722 (EXTRACT_SUBREG (COPY_TO_REGCLASS GR16:$src, GR16_ABCD),
1723 x86_subreg_8bit_hi)),
1725 Requires<[In64BitMode]>;
1726 def : Pat<(i32 (zext (srl_su GR16:$src, (i8 8)))),
1728 (EXTRACT_SUBREG (COPY_TO_REGCLASS GR16:$src, GR16_ABCD),
1729 x86_subreg_8bit_hi))>,
1730 Requires<[In64BitMode]>;
1731 def : Pat<(i32 (anyext (srl_su GR16:$src, (i8 8)))),
1733 (EXTRACT_SUBREG (COPY_TO_REGCLASS GR16:$src, GR16_ABCD),
1734 x86_subreg_8bit_hi))>,
1735 Requires<[In64BitMode]>;
1736 def : Pat<(i64 (zext (srl_su GR16:$src, (i8 8)))),
1740 (EXTRACT_SUBREG (COPY_TO_REGCLASS GR16:$src, GR16_ABCD),
1741 x86_subreg_8bit_hi)),
1743 def : Pat<(i64 (anyext (srl_su GR16:$src, (i8 8)))),
1747 (EXTRACT_SUBREG (COPY_TO_REGCLASS GR16:$src, GR16_ABCD),
1748 x86_subreg_8bit_hi)),
1751 // h-register extract and store.
1752 def : Pat<(store (i8 (trunc_su (srl_su GR64:$src, (i8 8)))), addr:$dst),
1755 (EXTRACT_SUBREG (COPY_TO_REGCLASS GR64:$src, GR64_ABCD),
1756 x86_subreg_8bit_hi))>;
1757 def : Pat<(store (i8 (trunc_su (srl_su GR32:$src, (i8 8)))), addr:$dst),
1760 (EXTRACT_SUBREG (COPY_TO_REGCLASS GR32:$src, GR32_ABCD),
1761 x86_subreg_8bit_hi))>,
1762 Requires<[In64BitMode]>;
1763 def : Pat<(store (i8 (trunc_su (srl_su GR16:$src, (i8 8)))), addr:$dst),
1766 (EXTRACT_SUBREG (COPY_TO_REGCLASS GR16:$src, GR16_ABCD),
1767 x86_subreg_8bit_hi))>,
1768 Requires<[In64BitMode]>;
1770 // (shl x, 1) ==> (add x, x)
1771 def : Pat<(shl GR64:$src1, (i8 1)), (ADD64rr GR64:$src1, GR64:$src1)>;
1773 // (shl x (and y, 63)) ==> (shl x, y)
1774 def : Pat<(shl GR64:$src1, (and CL:$amt, 63)),
1775 (SHL64rCL GR64:$src1)>;
1776 def : Pat<(store (shl (loadi64 addr:$dst), (and CL:$amt, 63)), addr:$dst),
1777 (SHL64mCL addr:$dst)>;
1779 def : Pat<(srl GR64:$src1, (and CL:$amt, 63)),
1780 (SHR64rCL GR64:$src1)>;
1781 def : Pat<(store (srl (loadi64 addr:$dst), (and CL:$amt, 63)), addr:$dst),
1782 (SHR64mCL addr:$dst)>;
1784 def : Pat<(sra GR64:$src1, (and CL:$amt, 63)),
1785 (SAR64rCL GR64:$src1)>;
1786 def : Pat<(store (sra (loadi64 addr:$dst), (and CL:$amt, 63)), addr:$dst),
1787 (SAR64mCL addr:$dst)>;
1789 // (or (x >> c) | (y << (64 - c))) ==> (shrd64 x, y, c)
1790 def : Pat<(or (srl GR64:$src1, CL:$amt),
1791 (shl GR64:$src2, (sub 64, CL:$amt))),
1792 (SHRD64rrCL GR64:$src1, GR64:$src2)>;
1794 def : Pat<(store (or (srl (loadi64 addr:$dst), CL:$amt),
1795 (shl GR64:$src2, (sub 64, CL:$amt))), addr:$dst),
1796 (SHRD64mrCL addr:$dst, GR64:$src2)>;
1798 def : Pat<(or (srl GR64:$src1, (i8 (trunc RCX:$amt))),
1799 (shl GR64:$src2, (i8 (trunc (sub 64, RCX:$amt))))),
1800 (SHRD64rrCL GR64:$src1, GR64:$src2)>;
1802 def : Pat<(store (or (srl (loadi64 addr:$dst), (i8 (trunc RCX:$amt))),
1803 (shl GR64:$src2, (i8 (trunc (sub 64, RCX:$amt))))),
1805 (SHRD64mrCL addr:$dst, GR64:$src2)>;
1807 def : Pat<(shrd GR64:$src1, (i8 imm:$amt1), GR64:$src2, (i8 imm:$amt2)),
1808 (SHRD64rri8 GR64:$src1, GR64:$src2, (i8 imm:$amt1))>;
1810 def : Pat<(store (shrd (loadi64 addr:$dst), (i8 imm:$amt1),
1811 GR64:$src2, (i8 imm:$amt2)), addr:$dst),
1812 (SHRD64mri8 addr:$dst, GR64:$src2, (i8 imm:$amt1))>;
1814 // (or (x << c) | (y >> (64 - c))) ==> (shld64 x, y, c)
1815 def : Pat<(or (shl GR64:$src1, CL:$amt),
1816 (srl GR64:$src2, (sub 64, CL:$amt))),
1817 (SHLD64rrCL GR64:$src1, GR64:$src2)>;
1819 def : Pat<(store (or (shl (loadi64 addr:$dst), CL:$amt),
1820 (srl GR64:$src2, (sub 64, CL:$amt))), addr:$dst),
1821 (SHLD64mrCL addr:$dst, GR64:$src2)>;
1823 def : Pat<(or (shl GR64:$src1, (i8 (trunc RCX:$amt))),
1824 (srl GR64:$src2, (i8 (trunc (sub 64, RCX:$amt))))),
1825 (SHLD64rrCL GR64:$src1, GR64:$src2)>;
1827 def : Pat<(store (or (shl (loadi64 addr:$dst), (i8 (trunc RCX:$amt))),
1828 (srl GR64:$src2, (i8 (trunc (sub 64, RCX:$amt))))),
1830 (SHLD64mrCL addr:$dst, GR64:$src2)>;
1832 def : Pat<(shld GR64:$src1, (i8 imm:$amt1), GR64:$src2, (i8 imm:$amt2)),
1833 (SHLD64rri8 GR64:$src1, GR64:$src2, (i8 imm:$amt1))>;
1835 def : Pat<(store (shld (loadi64 addr:$dst), (i8 imm:$amt1),
1836 GR64:$src2, (i8 imm:$amt2)), addr:$dst),
1837 (SHLD64mri8 addr:$dst, GR64:$src2, (i8 imm:$amt1))>;
1839 // X86 specific add which produces a flag.
1840 def : Pat<(addc GR64:$src1, GR64:$src2),
1841 (ADD64rr GR64:$src1, GR64:$src2)>;
1842 def : Pat<(addc GR64:$src1, (load addr:$src2)),
1843 (ADD64rm GR64:$src1, addr:$src2)>;
1844 def : Pat<(addc GR64:$src1, i64immSExt8:$src2),
1845 (ADD64ri8 GR64:$src1, i64immSExt8:$src2)>;
1846 def : Pat<(addc GR64:$src1, i64immSExt32:$src2),
1847 (ADD64ri32 GR64:$src1, imm:$src2)>;
1849 def : Pat<(subc GR64:$src1, GR64:$src2),
1850 (SUB64rr GR64:$src1, GR64:$src2)>;
1851 def : Pat<(subc GR64:$src1, (load addr:$src2)),
1852 (SUB64rm GR64:$src1, addr:$src2)>;
1853 def : Pat<(subc GR64:$src1, i64immSExt8:$src2),
1854 (SUB64ri8 GR64:$src1, i64immSExt8:$src2)>;
1855 def : Pat<(subc GR64:$src1, imm:$src2),
1856 (SUB64ri32 GR64:$src1, i64immSExt32:$src2)>;
1858 //===----------------------------------------------------------------------===//
1859 // EFLAGS-defining Patterns
1860 //===----------------------------------------------------------------------===//
1862 // Register-Register Addition with EFLAGS result
1863 def : Pat<(parallel (X86add_flag GR64:$src1, GR64:$src2),
1865 (ADD64rr GR64:$src1, GR64:$src2)>;
1867 // Register-Integer Addition with EFLAGS result
1868 def : Pat<(parallel (X86add_flag GR64:$src1, i64immSExt8:$src2),
1870 (ADD64ri8 GR64:$src1, i64immSExt8:$src2)>;
1871 def : Pat<(parallel (X86add_flag GR64:$src1, i64immSExt32:$src2),
1873 (ADD64ri32 GR64:$src1, i64immSExt32:$src2)>;
1875 // Register-Memory Addition with EFLAGS result
1876 def : Pat<(parallel (X86add_flag GR64:$src1, (loadi64 addr:$src2)),
1878 (ADD64rm GR64:$src1, addr:$src2)>;
1880 // Memory-Register Addition with EFLAGS result
1881 def : Pat<(parallel (store (X86add_flag (loadi64 addr:$dst), GR64:$src2),
1884 (ADD64mr addr:$dst, GR64:$src2)>;
1885 def : Pat<(parallel (store (X86add_flag (loadi64 addr:$dst), i64immSExt8:$src2),
1888 (ADD64mi8 addr:$dst, i64immSExt8:$src2)>;
1889 def : Pat<(parallel (store (X86add_flag (loadi64 addr:$dst), i64immSExt32:$src2),
1892 (ADD64mi32 addr:$dst, i64immSExt32:$src2)>;
1894 // Register-Register Subtraction with EFLAGS result
1895 def : Pat<(parallel (X86sub_flag GR64:$src1, GR64:$src2),
1897 (SUB64rr GR64:$src1, GR64:$src2)>;
1899 // Register-Memory Subtraction with EFLAGS result
1900 def : Pat<(parallel (X86sub_flag GR64:$src1, (loadi64 addr:$src2)),
1902 (SUB64rm GR64:$src1, addr:$src2)>;
1904 // Register-Integer Subtraction with EFLAGS result
1905 def : Pat<(parallel (X86sub_flag GR64:$src1, i64immSExt8:$src2),
1907 (SUB64ri8 GR64:$src1, i64immSExt8:$src2)>;
1908 def : Pat<(parallel (X86sub_flag GR64:$src1, i64immSExt32:$src2),
1910 (SUB64ri32 GR64:$src1, i64immSExt32:$src2)>;
1912 // Memory-Register Subtraction with EFLAGS result
1913 def : Pat<(parallel (store (X86sub_flag (loadi64 addr:$dst), GR64:$src2),
1916 (SUB64mr addr:$dst, GR64:$src2)>;
1918 // Memory-Integer Subtraction with EFLAGS result
1919 def : Pat<(parallel (store (X86sub_flag (loadi64 addr:$dst), i64immSExt8:$src2),
1922 (SUB64mi8 addr:$dst, i64immSExt8:$src2)>;
1923 def : Pat<(parallel (store (X86sub_flag (loadi64 addr:$dst), i64immSExt32:$src2),
1926 (SUB64mi32 addr:$dst, i64immSExt32:$src2)>;
1928 // Register-Register Signed Integer Multiplication with EFLAGS result
1929 def : Pat<(parallel (X86smul_flag GR64:$src1, GR64:$src2),
1931 (IMUL64rr GR64:$src1, GR64:$src2)>;
1933 // Register-Memory Signed Integer Multiplication with EFLAGS result
1934 def : Pat<(parallel (X86smul_flag GR64:$src1, (loadi64 addr:$src2)),
1936 (IMUL64rm GR64:$src1, addr:$src2)>;
1938 // Register-Integer Signed Integer Multiplication with EFLAGS result
1939 def : Pat<(parallel (X86smul_flag GR64:$src1, i64immSExt8:$src2),
1941 (IMUL64rri8 GR64:$src1, i64immSExt8:$src2)>;
1942 def : Pat<(parallel (X86smul_flag GR64:$src1, i64immSExt32:$src2),
1944 (IMUL64rri32 GR64:$src1, i64immSExt32:$src2)>;
1946 // Memory-Integer Signed Integer Multiplication with EFLAGS result
1947 def : Pat<(parallel (X86smul_flag (loadi64 addr:$src1), i64immSExt8:$src2),
1949 (IMUL64rmi8 addr:$src1, i64immSExt8:$src2)>;
1950 def : Pat<(parallel (X86smul_flag (loadi64 addr:$src1), i64immSExt32:$src2),
1952 (IMUL64rmi32 addr:$src1, i64immSExt32:$src2)>;
1954 // INC and DEC with EFLAGS result. Note that these do not set CF.
1955 def : Pat<(parallel (X86inc_flag GR16:$src), (implicit EFLAGS)),
1956 (INC64_16r GR16:$src)>, Requires<[In64BitMode]>;
1957 def : Pat<(parallel (store (i16 (X86inc_flag (loadi16 addr:$dst))), addr:$dst),
1959 (INC64_16m addr:$dst)>, Requires<[In64BitMode]>;
1960 def : Pat<(parallel (X86dec_flag GR16:$src), (implicit EFLAGS)),
1961 (DEC64_16r GR16:$src)>, Requires<[In64BitMode]>;
1962 def : Pat<(parallel (store (i16 (X86dec_flag (loadi16 addr:$dst))), addr:$dst),
1964 (DEC64_16m addr:$dst)>, Requires<[In64BitMode]>;
1966 def : Pat<(parallel (X86inc_flag GR32:$src), (implicit EFLAGS)),
1967 (INC64_32r GR32:$src)>, Requires<[In64BitMode]>;
1968 def : Pat<(parallel (store (i32 (X86inc_flag (loadi32 addr:$dst))), addr:$dst),
1970 (INC64_32m addr:$dst)>, Requires<[In64BitMode]>;
1971 def : Pat<(parallel (X86dec_flag GR32:$src), (implicit EFLAGS)),
1972 (DEC64_32r GR32:$src)>, Requires<[In64BitMode]>;
1973 def : Pat<(parallel (store (i32 (X86dec_flag (loadi32 addr:$dst))), addr:$dst),
1975 (DEC64_32m addr:$dst)>, Requires<[In64BitMode]>;
1977 def : Pat<(parallel (X86inc_flag GR64:$src), (implicit EFLAGS)),
1978 (INC64r GR64:$src)>;
1979 def : Pat<(parallel (store (i64 (X86inc_flag (loadi64 addr:$dst))), addr:$dst),
1981 (INC64m addr:$dst)>;
1982 def : Pat<(parallel (X86dec_flag GR64:$src), (implicit EFLAGS)),
1983 (DEC64r GR64:$src)>;
1984 def : Pat<(parallel (store (i64 (X86dec_flag (loadi64 addr:$dst))), addr:$dst),
1986 (DEC64m addr:$dst)>;
1988 //===----------------------------------------------------------------------===//
1989 // X86-64 SSE Instructions
1990 //===----------------------------------------------------------------------===//
1992 // Move instructions...
1994 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
1995 "mov{d|q}\t{$src, $dst|$dst, $src}",
1997 (v2i64 (scalar_to_vector GR64:$src)))]>;
1998 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
1999 "mov{d|q}\t{$src, $dst|$dst, $src}",
2000 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
2003 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
2004 "mov{d|q}\t{$src, $dst|$dst, $src}",
2005 [(set FR64:$dst, (bitconvert GR64:$src))]>;
2006 def MOV64toSDrm : RPDI<0x6E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
2007 "movq\t{$src, $dst|$dst, $src}",
2008 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
2010 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
2011 "mov{d|q}\t{$src, $dst|$dst, $src}",
2012 [(set GR64:$dst, (bitconvert FR64:$src))]>;
2013 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
2014 "movq\t{$src, $dst|$dst, $src}",
2015 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
2017 //===----------------------------------------------------------------------===//
2018 // X86-64 SSE4.1 Instructions
2019 //===----------------------------------------------------------------------===//
2021 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
2022 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
2023 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
2024 (ins VR128:$src1, i32i8imm:$src2),
2025 !strconcat(OpcodeStr,
2026 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2028 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
2029 def mr : SS4AIi8<opc, MRMDestMem, (outs),
2030 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
2031 !strconcat(OpcodeStr,
2032 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2033 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
2034 addr:$dst)]>, OpSize, REX_W;
2037 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
2039 let isTwoAddress = 1 in {
2040 multiclass SS41I_insert64<bits<8> opc, string OpcodeStr> {
2041 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
2042 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
2043 !strconcat(OpcodeStr,
2044 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
2046 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
2048 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
2049 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
2050 !strconcat(OpcodeStr,
2051 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
2053 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
2054 imm:$src3)))]>, OpSize, REX_W;
2058 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">;
2060 // -disable-16bit support.
2061 def : Pat<(truncstorei16 (i64 imm:$src), addr:$dst),
2062 (MOV16mi addr:$dst, imm:$src)>;
2063 def : Pat<(truncstorei16 GR64:$src, addr:$dst),
2064 (MOV16mr addr:$dst, (EXTRACT_SUBREG GR64:$src, x86_subreg_16bit))>;
2065 def : Pat<(i64 (sextloadi16 addr:$dst)),
2066 (MOVSX64rm16 addr:$dst)>;
2067 def : Pat<(i64 (zextloadi16 addr:$dst)),
2068 (MOVZX64rm16 addr:$dst)>;
2069 def : Pat<(i64 (extloadi16 addr:$dst)),
2070 (MOVZX64rm16 addr:$dst)>;