1 //===-- X86InstrMMX.td - Describe the MMX Instruction Set --*- 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 MMX instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 // All instructions that use MMX should be in this file, even if they also use
17 //===----------------------------------------------------------------------===//
19 //===----------------------------------------------------------------------===//
21 //===----------------------------------------------------------------------===//
23 let Sched = WriteVecALU in {
24 def MMX_INTALU_ITINS : OpndItins<
25 IIC_MMX_ALU_RR, IIC_MMX_ALU_RM
28 def MMX_INTALUQ_ITINS : OpndItins<
29 IIC_MMX_ALUQ_RR, IIC_MMX_ALUQ_RM
32 def MMX_PHADDSUBW : OpndItins<
33 IIC_MMX_PHADDSUBW_RR, IIC_MMX_PHADDSUBW_RM
36 def MMX_PHADDSUBD : OpndItins<
37 IIC_MMX_PHADDSUBD_RR, IIC_MMX_PHADDSUBD_RM
41 let Sched = WriteVecIMul in
42 def MMX_PMUL_ITINS : OpndItins<
43 IIC_MMX_PMUL, IIC_MMX_PMUL
46 let Sched = WriteVecALU in {
47 def MMX_PSADBW_ITINS : OpndItins<
48 IIC_MMX_PSADBW, IIC_MMX_PSADBW
51 def MMX_MISC_FUNC_ITINS : OpndItins<
52 IIC_MMX_MISC_FUNC_MEM, IIC_MMX_MISC_FUNC_REG
56 def MMX_SHIFT_ITINS : ShiftOpndItins<
57 IIC_MMX_SHIFT_RR, IIC_MMX_SHIFT_RM, IIC_MMX_SHIFT_RI
60 let Sched = WriteShuffle in {
61 def MMX_UNPCK_H_ITINS : OpndItins<
62 IIC_MMX_UNPCK_H_RR, IIC_MMX_UNPCK_H_RM
65 def MMX_UNPCK_L_ITINS : OpndItins<
66 IIC_MMX_UNPCK_L, IIC_MMX_UNPCK_L
69 def MMX_PCK_ITINS : OpndItins<
70 IIC_MMX_PCK_RR, IIC_MMX_PCK_RM
73 def MMX_PSHUF_ITINS : OpndItins<
74 IIC_MMX_PSHUF, IIC_MMX_PSHUF
78 let Sched = WriteCvtF2I in {
79 def MMX_CVT_PD_ITINS : OpndItins<
80 IIC_MMX_CVT_PD_RR, IIC_MMX_CVT_PD_RM
83 def MMX_CVT_PS_ITINS : OpndItins<
84 IIC_MMX_CVT_PS_RR, IIC_MMX_CVT_PS_RM
88 let Constraints = "$src1 = $dst" in {
89 // MMXI_binop_rm_int - Simple MMX binary operator based on intrinsic.
90 // When this is cleaned up, remove the FIXME from X86RecognizableInstr.cpp.
91 multiclass MMXI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
92 OpndItins itins, bit Commutable = 0> {
93 def irr : MMXI<opc, MRMSrcReg, (outs VR64:$dst),
94 (ins VR64:$src1, VR64:$src2),
95 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
96 [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))], itins.rr>,
97 Sched<[itins.Sched]> {
98 let isCommutable = Commutable;
100 def irm : MMXI<opc, MRMSrcMem, (outs VR64:$dst),
101 (ins VR64:$src1, i64mem:$src2),
102 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
103 [(set VR64:$dst, (IntId VR64:$src1,
104 (bitconvert (load_mmx addr:$src2))))],
105 itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
108 multiclass MMXI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
109 string OpcodeStr, Intrinsic IntId,
110 Intrinsic IntId2, ShiftOpndItins itins> {
111 def rr : MMXI<opc, MRMSrcReg, (outs VR64:$dst),
112 (ins VR64:$src1, VR64:$src2),
113 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
114 [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))], itins.rr>,
115 Sched<[WriteVecShift]>;
116 def rm : MMXI<opc, MRMSrcMem, (outs VR64:$dst),
117 (ins VR64:$src1, i64mem:$src2),
118 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
119 [(set VR64:$dst, (IntId VR64:$src1,
120 (bitconvert (load_mmx addr:$src2))))],
121 itins.rm>, Sched<[WriteVecShiftLd, ReadAfterLd]>;
122 def ri : MMXIi8<opc2, ImmForm, (outs VR64:$dst),
123 (ins VR64:$src1, i32i8imm:$src2),
124 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
125 [(set VR64:$dst, (IntId2 VR64:$src1, (i32 imm:$src2)))], itins.ri>,
126 Sched<[WriteVecShift]>;
130 /// Unary MMX instructions requiring SSSE3.
131 multiclass SS3I_unop_rm_int_mm<bits<8> opc, string OpcodeStr,
132 Intrinsic IntId64, OpndItins itins> {
133 def rr64 : MMXSS38I<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
134 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
135 [(set VR64:$dst, (IntId64 VR64:$src))], itins.rr>,
136 Sched<[itins.Sched]>;
138 def rm64 : MMXSS38I<opc, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
139 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
141 (IntId64 (bitconvert (memopmmx addr:$src))))],
142 itins.rm>, Sched<[itins.Sched.Folded]>;
145 /// Binary MMX instructions requiring SSSE3.
146 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
147 multiclass SS3I_binop_rm_int_mm<bits<8> opc, string OpcodeStr,
148 Intrinsic IntId64, OpndItins itins> {
149 let isCommutable = 0 in
150 def rr64 : MMXSS38I<opc, MRMSrcReg, (outs VR64:$dst),
151 (ins VR64:$src1, VR64:$src2),
152 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
153 [(set VR64:$dst, (IntId64 VR64:$src1, VR64:$src2))], itins.rr>,
154 Sched<[itins.Sched]>;
155 def rm64 : MMXSS38I<opc, MRMSrcMem, (outs VR64:$dst),
156 (ins VR64:$src1, i64mem:$src2),
157 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
160 (bitconvert (memopmmx addr:$src2))))], itins.rm>,
161 Sched<[itins.Sched.Folded, ReadAfterLd]>;
165 /// PALIGN MMX instructions (require SSSE3).
166 multiclass ssse3_palign_mm<string asm, Intrinsic IntId> {
167 def R64irr : MMXSS3AI<0x0F, MRMSrcReg, (outs VR64:$dst),
168 (ins VR64:$src1, VR64:$src2, i8imm:$src3),
169 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
170 [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2, (i8 imm:$src3)))]>;
171 def R64irm : MMXSS3AI<0x0F, MRMSrcMem, (outs VR64:$dst),
172 (ins VR64:$src1, i64mem:$src2, i8imm:$src3),
173 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
174 [(set VR64:$dst, (IntId VR64:$src1,
175 (bitconvert (load_mmx addr:$src2)), (i8 imm:$src3)))]>;
178 multiclass sse12_cvt_pint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
179 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
180 string asm, OpndItins itins, Domain d> {
181 def irr : MMXPI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
182 [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr, d>,
183 Sched<[itins.Sched]>;
184 def irm : MMXPI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
185 [(set DstRC:$dst, (Int (ld_frag addr:$src)))], itins.rm, d>,
186 Sched<[itins.Sched.Folded]>;
189 multiclass sse12_cvt_pint_3addr<bits<8> opc, RegisterClass SrcRC,
190 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
191 PatFrag ld_frag, string asm, Domain d> {
192 def irr : MMXPI<opc, MRMSrcReg, (outs DstRC:$dst),
193 (ins DstRC:$src1, SrcRC:$src2), asm,
194 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
196 def irm : MMXPI<opc, MRMSrcMem, (outs DstRC:$dst),
197 (ins DstRC:$src1, x86memop:$src2), asm,
198 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
202 //===----------------------------------------------------------------------===//
203 // MMX EMMS Instruction
204 //===----------------------------------------------------------------------===//
206 def MMX_EMMS : MMXI<0x77, RawFrm, (outs), (ins), "emms",
207 [(int_x86_mmx_emms)], IIC_MMX_EMMS>;
209 //===----------------------------------------------------------------------===//
210 // MMX Scalar Instructions
211 //===----------------------------------------------------------------------===//
213 // Data Transfer Instructions
214 def MMX_MOVD64rr : MMXI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR32:$src),
215 "movd\t{$src, $dst|$dst, $src}",
217 (x86mmx (scalar_to_vector GR32:$src)))],
218 IIC_MMX_MOV_MM_RM>, Sched<[WriteMove]>;
219 let canFoldAsLoad = 1 in
220 def MMX_MOVD64rm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst), (ins i32mem:$src),
221 "movd\t{$src, $dst|$dst, $src}",
223 (x86mmx (scalar_to_vector (loadi32 addr:$src))))],
224 IIC_MMX_MOV_MM_RM>, Sched<[WriteLoad]>;
226 def MMX_MOVD64mr : MMXI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR64:$src),
227 "movd\t{$src, $dst|$dst, $src}", [], IIC_MMX_MOV_MM_RM>,
230 // Low word of MMX to GPR.
231 def MMX_X86movd2w : SDNode<"X86ISD::MMX_MOVD2W", SDTypeProfile<1, 1,
232 [SDTCisVT<0, i32>, SDTCisVT<1, x86mmx>]>>;
233 def MMX_MOVD64grr : MMXI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR64:$src),
234 "movd\t{$src, $dst|$dst, $src}",
236 (MMX_X86movd2w (x86mmx VR64:$src)))],
237 IIC_MMX_MOV_REG_MM>, Sched<[WriteMove]>;
239 def MMX_MOVD64to64rr : MMXRI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR64:$src),
240 "movd\t{$src, $dst|$dst, $src}",
241 [(set VR64:$dst, (bitconvert GR64:$src))],
242 IIC_MMX_MOV_MM_RM>, Sched<[WriteMove]>;
244 // These are 64 bit moves, but since the OS X assembler doesn't
245 // recognize a register-register movq, we write them as
247 let SchedRW = [WriteMove] in {
248 def MMX_MOVD64from64rr : MMXRI<0x7E, MRMDestReg,
249 (outs GR64:$dst), (ins VR64:$src),
250 "movd\t{$src, $dst|$dst, $src}",
252 (bitconvert VR64:$src))], IIC_MMX_MOV_REG_MM>;
253 let neverHasSideEffects = 1 in
254 def MMX_MOVQ64rr : MMXI<0x6F, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
255 "movq\t{$src, $dst|$dst, $src}", [],
259 let SchedRW = [WriteLoad] in {
260 let canFoldAsLoad = 1 in
261 def MMX_MOVQ64rm : MMXI<0x6F, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
262 "movq\t{$src, $dst|$dst, $src}",
263 [(set VR64:$dst, (load_mmx addr:$src))],
265 def MMX_MOVQ64mr : MMXI<0x7F, MRMDestMem, (outs), (ins i64mem:$dst, VR64:$src),
266 "movq\t{$src, $dst|$dst, $src}",
267 [(store (x86mmx VR64:$src), addr:$dst)],
271 let SchedRW = [WriteMove] in {
272 def MMX_MOVDQ2Qrr : MMXSDIi8<0xD6, MRMSrcReg, (outs VR64:$dst),
273 (ins VR128:$src), "movdq2q\t{$src, $dst|$dst, $src}",
276 (i64 (vector_extract (v2i64 VR128:$src),
280 def MMX_MOVQ2DQrr : MMXS2SIi8<0xD6, MRMSrcReg, (outs VR128:$dst),
281 (ins VR64:$src), "movq2dq\t{$src, $dst|$dst, $src}",
285 (i64 (bitconvert (x86mmx VR64:$src))))))],
288 let isCodeGenOnly = 1, hasSideEffects = 1 in {
289 def MMX_MOVQ2FR64rr: MMXS2SIi8<0xD6, MRMSrcReg, (outs FR64:$dst),
290 (ins VR64:$src), "movq2dq\t{$src, $dst|$dst, $src}",
291 [], IIC_MMX_MOVQ_RR>;
293 def MMX_MOVFR642Qrr: MMXSDIi8<0xD6, MRMSrcReg, (outs VR64:$dst),
294 (ins FR64:$src), "movdq2q\t{$src, $dst|$dst, $src}",
295 [], IIC_MMX_MOVQ_RR>;
299 def MMX_MOVNTQmr : MMXI<0xE7, MRMDestMem, (outs), (ins i64mem:$dst, VR64:$src),
300 "movntq\t{$src, $dst|$dst, $src}",
301 [(int_x86_mmx_movnt_dq addr:$dst, VR64:$src)],
302 IIC_MMX_MOVQ_RM>, Sched<[WriteStore]>;
304 let Predicates = [HasMMX] in {
305 let AddedComplexity = 15 in
306 // movd to MMX register zero-extends
307 def : Pat<(x86mmx (X86vzmovl (x86mmx (scalar_to_vector GR32:$src)))),
308 (MMX_MOVD64rr GR32:$src)>;
309 let AddedComplexity = 20 in
310 def : Pat<(x86mmx (X86vzmovl (x86mmx (scalar_to_vector (loadi32 addr:$src))))),
311 (MMX_MOVD64rm addr:$src)>;
314 // Arithmetic Instructions
315 defm MMX_PABSB : SS3I_unop_rm_int_mm<0x1C, "pabsb", int_x86_ssse3_pabs_b,
317 defm MMX_PABSW : SS3I_unop_rm_int_mm<0x1D, "pabsw", int_x86_ssse3_pabs_w,
319 defm MMX_PABSD : SS3I_unop_rm_int_mm<0x1E, "pabsd", int_x86_ssse3_pabs_d,
322 defm MMX_PADDB : MMXI_binop_rm_int<0xFC, "paddb", int_x86_mmx_padd_b,
323 MMX_INTALU_ITINS, 1>;
324 defm MMX_PADDW : MMXI_binop_rm_int<0xFD, "paddw", int_x86_mmx_padd_w,
325 MMX_INTALU_ITINS, 1>;
326 defm MMX_PADDD : MMXI_binop_rm_int<0xFE, "paddd", int_x86_mmx_padd_d,
327 MMX_INTALU_ITINS, 1>;
328 defm MMX_PADDQ : MMXI_binop_rm_int<0xD4, "paddq", int_x86_mmx_padd_q,
329 MMX_INTALUQ_ITINS, 1>;
330 defm MMX_PADDSB : MMXI_binop_rm_int<0xEC, "paddsb" , int_x86_mmx_padds_b,
331 MMX_INTALU_ITINS, 1>;
332 defm MMX_PADDSW : MMXI_binop_rm_int<0xED, "paddsw" , int_x86_mmx_padds_w,
333 MMX_INTALU_ITINS, 1>;
335 defm MMX_PADDUSB : MMXI_binop_rm_int<0xDC, "paddusb", int_x86_mmx_paddus_b,
336 MMX_INTALU_ITINS, 1>;
337 defm MMX_PADDUSW : MMXI_binop_rm_int<0xDD, "paddusw", int_x86_mmx_paddus_w,
338 MMX_INTALU_ITINS, 1>;
340 defm MMX_PHADDW : SS3I_binop_rm_int_mm<0x01, "phaddw", int_x86_ssse3_phadd_w,
342 defm MMX_PHADD : SS3I_binop_rm_int_mm<0x02, "phaddd", int_x86_ssse3_phadd_d,
344 defm MMX_PHADDSW : SS3I_binop_rm_int_mm<0x03, "phaddsw",int_x86_ssse3_phadd_sw,
349 defm MMX_PSUBB : MMXI_binop_rm_int<0xF8, "psubb", int_x86_mmx_psub_b,
351 defm MMX_PSUBW : MMXI_binop_rm_int<0xF9, "psubw", int_x86_mmx_psub_w,
353 defm MMX_PSUBD : MMXI_binop_rm_int<0xFA, "psubd", int_x86_mmx_psub_d,
355 defm MMX_PSUBQ : MMXI_binop_rm_int<0xFB, "psubq", int_x86_mmx_psub_q,
358 defm MMX_PSUBSB : MMXI_binop_rm_int<0xE8, "psubsb" , int_x86_mmx_psubs_b,
360 defm MMX_PSUBSW : MMXI_binop_rm_int<0xE9, "psubsw" , int_x86_mmx_psubs_w,
363 defm MMX_PSUBUSB : MMXI_binop_rm_int<0xD8, "psubusb", int_x86_mmx_psubus_b,
365 defm MMX_PSUBUSW : MMXI_binop_rm_int<0xD9, "psubusw", int_x86_mmx_psubus_w,
368 defm MMX_PHSUBW : SS3I_binop_rm_int_mm<0x05, "phsubw", int_x86_ssse3_phsub_w,
370 defm MMX_PHSUBD : SS3I_binop_rm_int_mm<0x06, "phsubd", int_x86_ssse3_phsub_d,
372 defm MMX_PHSUBSW : SS3I_binop_rm_int_mm<0x07, "phsubsw",int_x86_ssse3_phsub_sw,
376 defm MMX_PMULLW : MMXI_binop_rm_int<0xD5, "pmullw", int_x86_mmx_pmull_w,
379 defm MMX_PMULHW : MMXI_binop_rm_int<0xE5, "pmulhw", int_x86_mmx_pmulh_w,
381 defm MMX_PMULHUW : MMXI_binop_rm_int<0xE4, "pmulhuw", int_x86_mmx_pmulhu_w,
383 defm MMX_PMULUDQ : MMXI_binop_rm_int<0xF4, "pmuludq", int_x86_mmx_pmulu_dq,
385 let isCommutable = 1 in
386 defm MMX_PMULHRSW : SS3I_binop_rm_int_mm<0x0B, "pmulhrsw",
387 int_x86_ssse3_pmul_hr_sw, MMX_PMUL_ITINS>;
390 defm MMX_PMADDWD : MMXI_binop_rm_int<0xF5, "pmaddwd", int_x86_mmx_pmadd_wd,
393 defm MMX_PMADDUBSW : SS3I_binop_rm_int_mm<0x04, "pmaddubsw",
394 int_x86_ssse3_pmadd_ub_sw, MMX_PMUL_ITINS>;
395 defm MMX_PAVGB : MMXI_binop_rm_int<0xE0, "pavgb", int_x86_mmx_pavg_b,
396 MMX_MISC_FUNC_ITINS, 1>;
397 defm MMX_PAVGW : MMXI_binop_rm_int<0xE3, "pavgw", int_x86_mmx_pavg_w,
398 MMX_MISC_FUNC_ITINS, 1>;
400 defm MMX_PMINUB : MMXI_binop_rm_int<0xDA, "pminub", int_x86_mmx_pminu_b,
401 MMX_MISC_FUNC_ITINS, 1>;
402 defm MMX_PMINSW : MMXI_binop_rm_int<0xEA, "pminsw", int_x86_mmx_pmins_w,
403 MMX_MISC_FUNC_ITINS, 1>;
405 defm MMX_PMAXUB : MMXI_binop_rm_int<0xDE, "pmaxub", int_x86_mmx_pmaxu_b,
406 MMX_MISC_FUNC_ITINS, 1>;
407 defm MMX_PMAXSW : MMXI_binop_rm_int<0xEE, "pmaxsw", int_x86_mmx_pmaxs_w,
408 MMX_MISC_FUNC_ITINS, 1>;
410 defm MMX_PSADBW : MMXI_binop_rm_int<0xF6, "psadbw", int_x86_mmx_psad_bw,
411 MMX_PSADBW_ITINS, 1>;
413 defm MMX_PSIGNB : SS3I_binop_rm_int_mm<0x08, "psignb", int_x86_ssse3_psign_b,
414 MMX_MISC_FUNC_ITINS>;
415 defm MMX_PSIGNW : SS3I_binop_rm_int_mm<0x09, "psignw", int_x86_ssse3_psign_w,
416 MMX_MISC_FUNC_ITINS>;
417 defm MMX_PSIGND : SS3I_binop_rm_int_mm<0x0A, "psignd", int_x86_ssse3_psign_d,
418 MMX_MISC_FUNC_ITINS>;
419 let Constraints = "$src1 = $dst" in
420 defm MMX_PALIGN : ssse3_palign_mm<"palignr", int_x86_mmx_palignr_b>;
422 // Logical Instructions
423 defm MMX_PAND : MMXI_binop_rm_int<0xDB, "pand", int_x86_mmx_pand,
424 MMX_INTALU_ITINS, 1>;
425 defm MMX_POR : MMXI_binop_rm_int<0xEB, "por" , int_x86_mmx_por,
426 MMX_INTALU_ITINS, 1>;
427 defm MMX_PXOR : MMXI_binop_rm_int<0xEF, "pxor", int_x86_mmx_pxor,
428 MMX_INTALU_ITINS, 1>;
429 defm MMX_PANDN : MMXI_binop_rm_int<0xDF, "pandn", int_x86_mmx_pandn,
432 // Shift Instructions
433 defm MMX_PSRLW : MMXI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
434 int_x86_mmx_psrl_w, int_x86_mmx_psrli_w,
436 defm MMX_PSRLD : MMXI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
437 int_x86_mmx_psrl_d, int_x86_mmx_psrli_d,
439 defm MMX_PSRLQ : MMXI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
440 int_x86_mmx_psrl_q, int_x86_mmx_psrli_q,
443 defm MMX_PSLLW : MMXI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
444 int_x86_mmx_psll_w, int_x86_mmx_pslli_w,
446 defm MMX_PSLLD : MMXI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
447 int_x86_mmx_psll_d, int_x86_mmx_pslli_d,
449 defm MMX_PSLLQ : MMXI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
450 int_x86_mmx_psll_q, int_x86_mmx_pslli_q,
453 defm MMX_PSRAW : MMXI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
454 int_x86_mmx_psra_w, int_x86_mmx_psrai_w,
456 defm MMX_PSRAD : MMXI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
457 int_x86_mmx_psra_d, int_x86_mmx_psrai_d,
460 // Comparison Instructions
461 defm MMX_PCMPEQB : MMXI_binop_rm_int<0x74, "pcmpeqb", int_x86_mmx_pcmpeq_b,
463 defm MMX_PCMPEQW : MMXI_binop_rm_int<0x75, "pcmpeqw", int_x86_mmx_pcmpeq_w,
465 defm MMX_PCMPEQD : MMXI_binop_rm_int<0x76, "pcmpeqd", int_x86_mmx_pcmpeq_d,
468 defm MMX_PCMPGTB : MMXI_binop_rm_int<0x64, "pcmpgtb", int_x86_mmx_pcmpgt_b,
470 defm MMX_PCMPGTW : MMXI_binop_rm_int<0x65, "pcmpgtw", int_x86_mmx_pcmpgt_w,
472 defm MMX_PCMPGTD : MMXI_binop_rm_int<0x66, "pcmpgtd", int_x86_mmx_pcmpgt_d,
475 // -- Unpack Instructions
476 defm MMX_PUNPCKHBW : MMXI_binop_rm_int<0x68, "punpckhbw",
477 int_x86_mmx_punpckhbw,
479 defm MMX_PUNPCKHWD : MMXI_binop_rm_int<0x69, "punpckhwd",
480 int_x86_mmx_punpckhwd,
482 defm MMX_PUNPCKHDQ : MMXI_binop_rm_int<0x6A, "punpckhdq",
483 int_x86_mmx_punpckhdq,
485 defm MMX_PUNPCKLBW : MMXI_binop_rm_int<0x60, "punpcklbw",
486 int_x86_mmx_punpcklbw,
488 defm MMX_PUNPCKLWD : MMXI_binop_rm_int<0x61, "punpcklwd",
489 int_x86_mmx_punpcklwd,
491 defm MMX_PUNPCKLDQ : MMXI_binop_rm_int<0x62, "punpckldq",
492 int_x86_mmx_punpckldq,
495 // -- Pack Instructions
496 defm MMX_PACKSSWB : MMXI_binop_rm_int<0x63, "packsswb", int_x86_mmx_packsswb,
498 defm MMX_PACKSSDW : MMXI_binop_rm_int<0x6B, "packssdw", int_x86_mmx_packssdw,
500 defm MMX_PACKUSWB : MMXI_binop_rm_int<0x67, "packuswb", int_x86_mmx_packuswb,
503 // -- Shuffle Instructions
504 defm MMX_PSHUFB : SS3I_binop_rm_int_mm<0x00, "pshufb", int_x86_ssse3_pshuf_b,
507 def MMX_PSHUFWri : MMXIi8<0x70, MRMSrcReg,
508 (outs VR64:$dst), (ins VR64:$src1, i8imm:$src2),
509 "pshufw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
511 (int_x86_sse_pshuf_w VR64:$src1, imm:$src2))],
512 IIC_MMX_PSHUF>, Sched<[WriteShuffle]>;
513 def MMX_PSHUFWmi : MMXIi8<0x70, MRMSrcMem,
514 (outs VR64:$dst), (ins i64mem:$src1, i8imm:$src2),
515 "pshufw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
517 (int_x86_sse_pshuf_w (load_mmx addr:$src1),
519 IIC_MMX_PSHUF>, Sched<[WriteShuffleLd]>;
524 // -- Conversion Instructions
525 defm MMX_CVTPS2PI : sse12_cvt_pint<0x2D, VR128, VR64, int_x86_sse_cvtps2pi,
526 f64mem, load, "cvtps2pi\t{$src, $dst|$dst, $src}",
527 MMX_CVT_PS_ITINS, SSEPackedSingle>, PS;
528 defm MMX_CVTPD2PI : sse12_cvt_pint<0x2D, VR128, VR64, int_x86_sse_cvtpd2pi,
529 f128mem, memop, "cvtpd2pi\t{$src, $dst|$dst, $src}",
530 MMX_CVT_PD_ITINS, SSEPackedDouble>, PD;
531 defm MMX_CVTTPS2PI : sse12_cvt_pint<0x2C, VR128, VR64, int_x86_sse_cvttps2pi,
532 f64mem, load, "cvttps2pi\t{$src, $dst|$dst, $src}",
533 MMX_CVT_PS_ITINS, SSEPackedSingle>, PS;
534 defm MMX_CVTTPD2PI : sse12_cvt_pint<0x2C, VR128, VR64, int_x86_sse_cvttpd2pi,
535 f128mem, memop, "cvttpd2pi\t{$src, $dst|$dst, $src}",
536 MMX_CVT_PD_ITINS, SSEPackedDouble>, PD;
537 defm MMX_CVTPI2PD : sse12_cvt_pint<0x2A, VR64, VR128, int_x86_sse_cvtpi2pd,
538 i64mem, load, "cvtpi2pd\t{$src, $dst|$dst, $src}",
539 MMX_CVT_PD_ITINS, SSEPackedDouble>, PD;
540 let Constraints = "$src1 = $dst" in {
541 defm MMX_CVTPI2PS : sse12_cvt_pint_3addr<0x2A, VR64, VR128,
542 int_x86_sse_cvtpi2ps,
543 i64mem, load, "cvtpi2ps\t{$src2, $dst|$dst, $src2}",
544 SSEPackedSingle>, PS;
548 def MMX_PEXTRWirri: MMXIi8<0xC5, MRMSrcReg,
549 (outs GR32orGR64:$dst), (ins VR64:$src1, i32i8imm:$src2),
550 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
551 [(set GR32orGR64:$dst, (int_x86_mmx_pextr_w VR64:$src1,
553 IIC_MMX_PEXTR>, Sched<[WriteShuffle]>;
554 let Constraints = "$src1 = $dst" in {
555 def MMX_PINSRWirri : MMXIi8<0xC4, MRMSrcReg,
557 (ins VR64:$src1, GR32orGR64:$src2, i32i8imm:$src3),
558 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
559 [(set VR64:$dst, (int_x86_mmx_pinsr_w VR64:$src1,
560 GR32orGR64:$src2, (iPTR imm:$src3)))],
561 IIC_MMX_PINSRW>, Sched<[WriteShuffle]>;
563 def MMX_PINSRWirmi : MMXIi8<0xC4, MRMSrcMem,
565 (ins VR64:$src1, i16mem:$src2, i32i8imm:$src3),
566 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
567 [(set VR64:$dst, (int_x86_mmx_pinsr_w VR64:$src1,
568 (i32 (anyext (loadi16 addr:$src2))),
570 IIC_MMX_PINSRW>, Sched<[WriteShuffleLd, ReadAfterLd]>;
574 def MMX_PMOVMSKBrr : MMXI<0xD7, MRMSrcReg, (outs GR32orGR64:$dst),
576 "pmovmskb\t{$src, $dst|$dst, $src}",
577 [(set GR32orGR64:$dst,
578 (int_x86_mmx_pmovmskb VR64:$src))]>;
581 // Low word of XMM to MMX.
582 def MMX_X86movdq2q : SDNode<"X86ISD::MOVDQ2Q", SDTypeProfile<1, 1,
583 [SDTCisVT<0, x86mmx>, SDTCisVT<1, v2i64>]>>;
585 def : Pat<(x86mmx (MMX_X86movdq2q VR128:$src)),
586 (x86mmx (MMX_MOVDQ2Qrr VR128:$src))>;
588 def : Pat<(x86mmx (MMX_X86movdq2q (loadv2i64 addr:$src))),
589 (x86mmx (MMX_MOVQ64rm addr:$src))>;
592 let SchedRW = [WriteShuffle] in {
594 def MMX_MASKMOVQ : MMXI32<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask),
595 "maskmovq\t{$mask, $src|$src, $mask}",
596 [(int_x86_mmx_maskmovq VR64:$src, VR64:$mask, EDI)],
599 def MMX_MASKMOVQ64: MMXI64<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask),
600 "maskmovq\t{$mask, $src|$src, $mask}",
601 [(int_x86_mmx_maskmovq VR64:$src, VR64:$mask, RDI)],
605 // 64-bit bit convert.
606 let Predicates = [HasSSE2] in {
607 def : Pat<(x86mmx (bitconvert (i64 GR64:$src))),
608 (MMX_MOVD64to64rr GR64:$src)>;
609 def : Pat<(i64 (bitconvert (x86mmx VR64:$src))),
610 (MMX_MOVD64from64rr VR64:$src)>;
611 def : Pat<(f64 (bitconvert (x86mmx VR64:$src))),
612 (MMX_MOVQ2FR64rr VR64:$src)>;
613 def : Pat<(x86mmx (bitconvert (f64 FR64:$src))),
614 (MMX_MOVFR642Qrr FR64:$src)>;