1//===-- X86InstrMMX.td - Describe the MMX Instruction Set --*- tablegen -*-===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file 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. 13// 14// All instructions that use MMX should be in this file, even if they also use 15// SSE. 16// 17//===----------------------------------------------------------------------===// 18 19//===----------------------------------------------------------------------===// 20// MMX Multiclasses 21//===----------------------------------------------------------------------===// 22 23let Sched = WriteVecALU in { 24def MMX_INTALU_ITINS : OpndItins< 25 IIC_MMX_ALU_RR, IIC_MMX_ALU_RM 26>; 27 28def MMX_INTALUQ_ITINS : OpndItins< 29 IIC_MMX_ALUQ_RR, IIC_MMX_ALUQ_RM 30>; 31 32def MMX_PHADDSUBW : OpndItins< 33 IIC_MMX_PHADDSUBW_RR, IIC_MMX_PHADDSUBW_RM 34>; 35 36def MMX_PHADDSUBD : OpndItins< 37 IIC_MMX_PHADDSUBD_RR, IIC_MMX_PHADDSUBD_RM 38>; 39} 40 41let Sched = WriteVecLogic in 42def MMX_INTALU_ITINS_VECLOGICSCHED : OpndItins< 43 IIC_MMX_ALU_RR, IIC_MMX_ALU_RM 44>; 45 46let Sched = WriteVecIMul in 47def MMX_PMUL_ITINS : OpndItins< 48 IIC_MMX_PMUL, IIC_MMX_PMUL 49>; 50 51let Sched = WriteVecIMul in { 52def MMX_PSADBW_ITINS : OpndItins< 53 IIC_MMX_PSADBW, IIC_MMX_PSADBW 54>; 55 56def MMX_MISC_FUNC_ITINS : OpndItins< 57 IIC_MMX_MISC_FUNC_MEM, IIC_MMX_MISC_FUNC_REG 58>; 59} 60 61def MMX_SHIFT_ITINS : ShiftOpndItins< 62 IIC_MMX_SHIFT_RR, IIC_MMX_SHIFT_RM, IIC_MMX_SHIFT_RI 63>; 64 65let Sched = WriteShuffle in { 66def MMX_UNPCK_H_ITINS : OpndItins< 67 IIC_MMX_UNPCK_H_RR, IIC_MMX_UNPCK_H_RM 68>; 69 70def MMX_UNPCK_L_ITINS : OpndItins< 71 IIC_MMX_UNPCK_L, IIC_MMX_UNPCK_L 72>; 73 74def MMX_PCK_ITINS : OpndItins< 75 IIC_MMX_PCK_RR, IIC_MMX_PCK_RM 76>; 77 78def MMX_PSHUF_ITINS : OpndItins< 79 IIC_MMX_PSHUF, IIC_MMX_PSHUF 80>; 81} // Sched 82 83let Sched = WriteCvtF2I in { 84def MMX_CVT_PD_ITINS : OpndItins< 85 IIC_MMX_CVT_PD_RR, IIC_MMX_CVT_PD_RM 86>; 87 88def MMX_CVT_PS_ITINS : OpndItins< 89 IIC_MMX_CVT_PS_RR, IIC_MMX_CVT_PS_RM 90>; 91} 92 93let Constraints = "$src1 = $dst" in { 94 // MMXI_binop_rm_int - Simple MMX binary operator based on intrinsic. 95 // When this is cleaned up, remove the FIXME from X86RecognizableInstr.cpp. 96 multiclass MMXI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId, 97 OpndItins itins, bit Commutable = 0> { 98 def irr : MMXI<opc, MRMSrcReg, (outs VR64:$dst), 99 (ins VR64:$src1, VR64:$src2), 100 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), 101 [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))], itins.rr>, 102 Sched<[itins.Sched]> { 103 let isCommutable = Commutable; 104 } 105 def irm : MMXI<opc, MRMSrcMem, (outs VR64:$dst), 106 (ins VR64:$src1, i64mem:$src2), 107 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), 108 [(set VR64:$dst, (IntId VR64:$src1, 109 (bitconvert (load_mmx addr:$src2))))], 110 itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>; 111 } 112 113 multiclass MMXI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm, 114 string OpcodeStr, Intrinsic IntId, 115 Intrinsic IntId2, ShiftOpndItins itins> { 116 def rr : MMXI<opc, MRMSrcReg, (outs VR64:$dst), 117 (ins VR64:$src1, VR64:$src2), 118 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), 119 [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))], itins.rr>, 120 Sched<[WriteVecShift]>; 121 def rm : MMXI<opc, MRMSrcMem, (outs VR64:$dst), 122 (ins VR64:$src1, i64mem:$src2), 123 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), 124 [(set VR64:$dst, (IntId VR64:$src1, 125 (bitconvert (load_mmx addr:$src2))))], 126 itins.rm>, Sched<[WriteVecShiftLd, ReadAfterLd]>; 127 def ri : MMXIi8<opc2, ImmForm, (outs VR64:$dst), 128 (ins VR64:$src1, i32u8imm:$src2), 129 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), 130 [(set VR64:$dst, (IntId2 VR64:$src1, imm:$src2))], itins.ri>, 131 Sched<[WriteVecShift]>; 132 } 133} 134 135/// Unary MMX instructions requiring SSSE3. 136multiclass SS3I_unop_rm_int_mm<bits<8> opc, string OpcodeStr, 137 Intrinsic IntId64, OpndItins itins> { 138 def rr64 : MMXSS38I<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src), 139 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), 140 [(set VR64:$dst, (IntId64 VR64:$src))], itins.rr>, 141 Sched<[itins.Sched]>; 142 143 def rm64 : MMXSS38I<opc, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src), 144 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), 145 [(set VR64:$dst, 146 (IntId64 (bitconvert (memopmmx addr:$src))))], 147 itins.rm>, Sched<[itins.Sched.Folded]>; 148} 149 150/// Binary MMX instructions requiring SSSE3. 151let ImmT = NoImm, Constraints = "$src1 = $dst" in { 152multiclass SS3I_binop_rm_int_mm<bits<8> opc, string OpcodeStr, 153 Intrinsic IntId64, OpndItins itins> { 154 let isCommutable = 0 in 155 def rr64 : MMXSS38I<opc, MRMSrcReg, (outs VR64:$dst), 156 (ins VR64:$src1, VR64:$src2), 157 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), 158 [(set VR64:$dst, (IntId64 VR64:$src1, VR64:$src2))], itins.rr>, 159 Sched<[itins.Sched]>; 160 def rm64 : MMXSS38I<opc, MRMSrcMem, (outs VR64:$dst), 161 (ins VR64:$src1, i64mem:$src2), 162 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), 163 [(set VR64:$dst, 164 (IntId64 VR64:$src1, 165 (bitconvert (memopmmx addr:$src2))))], itins.rm>, 166 Sched<[itins.Sched.Folded, ReadAfterLd]>; 167} 168} 169 170/// PALIGN MMX instructions (require SSSE3). 171multiclass ssse3_palign_mm<string asm, Intrinsic IntId> { 172 def R64irr : MMXSS3AI<0x0F, MRMSrcReg, (outs VR64:$dst), 173 (ins VR64:$src1, VR64:$src2, u8imm:$src3), 174 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), 175 [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2, (i8 imm:$src3)))]>, 176 Sched<[WriteShuffle]>; 177 def R64irm : MMXSS3AI<0x0F, MRMSrcMem, (outs VR64:$dst), 178 (ins VR64:$src1, i64mem:$src2, u8imm:$src3), 179 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), 180 [(set VR64:$dst, (IntId VR64:$src1, 181 (bitconvert (load_mmx addr:$src2)), (i8 imm:$src3)))]>, 182 Sched<[WriteShuffleLd, ReadAfterLd]>; 183} 184 185multiclass sse12_cvt_pint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC, 186 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag, 187 string asm, OpndItins itins, Domain d> { 188 def irr : MMXPI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm, 189 [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr, d>, 190 Sched<[itins.Sched]>; 191 def irm : MMXPI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm, 192 [(set DstRC:$dst, (Int (ld_frag addr:$src)))], itins.rm, d>, 193 Sched<[itins.Sched.Folded]>; 194} 195 196multiclass sse12_cvt_pint_3addr<bits<8> opc, RegisterClass SrcRC, 197 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop, 198 PatFrag ld_frag, string asm, Domain d> { 199 def irr : MMXPI<opc, MRMSrcReg, (outs DstRC:$dst), 200 (ins DstRC:$src1, SrcRC:$src2), asm, 201 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))], 202 NoItinerary, d>, Sched<[WriteCvtI2F]>; 203 def irm : MMXPI<opc, MRMSrcMem, (outs DstRC:$dst), 204 (ins DstRC:$src1, x86memop:$src2), asm, 205 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))], 206 NoItinerary, d>, Sched<[WriteCvtI2FLd]>; 207} 208 209//===----------------------------------------------------------------------===// 210// MMX EMMS Instruction 211//===----------------------------------------------------------------------===// 212 213def MMX_EMMS : MMXI<0x77, RawFrm, (outs), (ins), "emms", 214 [(int_x86_mmx_emms)], IIC_MMX_EMMS>; 215 216//===----------------------------------------------------------------------===// 217// MMX Scalar Instructions 218//===----------------------------------------------------------------------===// 219 220// Data Transfer Instructions 221def MMX_MOVD64rr : MMXI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR32:$src), 222 "movd\t{$src, $dst|$dst, $src}", 223 [(set VR64:$dst, 224 (x86mmx (scalar_to_vector GR32:$src)))], 225 IIC_MMX_MOV_MM_RM>, Sched<[WriteMove]>; 226def MMX_MOVD64rm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst), (ins i32mem:$src), 227 "movd\t{$src, $dst|$dst, $src}", 228 [(set VR64:$dst, 229 (x86mmx (scalar_to_vector (loadi32 addr:$src))))], 230 IIC_MMX_MOV_MM_RM>, Sched<[WriteLoad]>; 231 232let Predicates = [HasMMX] in { 233 let AddedComplexity = 15 in 234 def : Pat<(x86mmx (MMX_X86movw2d GR32:$src)), 235 (MMX_MOVD64rr GR32:$src)>; 236 let AddedComplexity = 20 in 237 def : Pat<(x86mmx (MMX_X86movw2d (loadi32 addr:$src))), 238 (MMX_MOVD64rm addr:$src)>; 239} 240 241let mayStore = 1 in 242def MMX_MOVD64mr : MMXI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR64:$src), 243 "movd\t{$src, $dst|$dst, $src}", [], IIC_MMX_MOV_MM_RM>, 244 Sched<[WriteStore]>; 245 246def MMX_MOVD64grr : MMXI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR64:$src), 247 "movd\t{$src, $dst|$dst, $src}", 248 [(set GR32:$dst, 249 (MMX_X86movd2w (x86mmx VR64:$src)))], 250 IIC_MMX_MOV_REG_MM>, Sched<[WriteMove]>; 251 252def MMX_MOVD64to64rr : MMXRI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR64:$src), 253 "movd\t{$src, $dst|$dst, $src}", 254 [(set VR64:$dst, (bitconvert GR64:$src))], 255 IIC_MMX_MOV_MM_RM>, Sched<[WriteMove]>; 256 257let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0, mayLoad = 1 in 258def MMX_MOVD64to64rm : MMXRI<0x6E, MRMSrcMem, (outs VR64:$dst), 259 (ins i64mem:$src), "movd\t{$src, $dst|$dst, $src}", 260 [], IIC_MMX_MOVQ_RM>, Sched<[WriteLoad]>; 261 262// These are 64 bit moves, but since the OS X assembler doesn't 263// recognize a register-register movq, we write them as 264// movd. 265let SchedRW = [WriteMove] in { 266def MMX_MOVD64from64rr : MMXRI<0x7E, MRMDestReg, 267 (outs GR64:$dst), (ins VR64:$src), 268 "movd\t{$src, $dst|$dst, $src}", 269 [(set GR64:$dst, 270 (bitconvert VR64:$src))], IIC_MMX_MOV_REG_MM>; 271let hasSideEffects = 0 in 272def MMX_MOVQ64rr : MMXI<0x6F, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src), 273 "movq\t{$src, $dst|$dst, $src}", [], 274 IIC_MMX_MOVQ_RR>; 275let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in { 276def MMX_MOVQ64rr_REV : MMXI<0x7F, MRMDestReg, (outs VR64:$dst), (ins VR64:$src), 277 "movq\t{$src, $dst|$dst, $src}", [], 278 IIC_MMX_MOVQ_RR>; 279} 280} // SchedRW 281 282let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0, mayStore = 1 in 283def MMX_MOVD64from64rm : MMXRI<0x7E, MRMDestMem, 284 (outs i64mem:$dst), (ins VR64:$src), 285 "movd\t{$src, $dst|$dst, $src}", 286 [], IIC_MMX_MOV_REG_MM>, Sched<[WriteStore]>; 287 288let SchedRW = [WriteLoad] in { 289let canFoldAsLoad = 1 in 290def MMX_MOVQ64rm : MMXI<0x6F, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src), 291 "movq\t{$src, $dst|$dst, $src}", 292 [(set VR64:$dst, (load_mmx addr:$src))], 293 IIC_MMX_MOVQ_RM>; 294} // SchedRW 295let SchedRW = [WriteStore] in 296def MMX_MOVQ64mr : MMXI<0x7F, MRMDestMem, (outs), (ins i64mem:$dst, VR64:$src), 297 "movq\t{$src, $dst|$dst, $src}", 298 [(store (x86mmx VR64:$src), addr:$dst)], 299 IIC_MMX_MOVQ_RM>; 300 301let SchedRW = [WriteMove] in { 302def MMX_MOVDQ2Qrr : MMXSDIi8<0xD6, MRMSrcReg, (outs VR64:$dst), 303 (ins VR128:$src), "movdq2q\t{$src, $dst|$dst, $src}", 304 [(set VR64:$dst, 305 (x86mmx (bitconvert 306 (i64 (vector_extract (v2i64 VR128:$src), 307 (iPTR 0))))))], 308 IIC_MMX_MOVQ_RR>; 309 310def MMX_MOVQ2DQrr : MMXS2SIi8<0xD6, MRMSrcReg, (outs VR128:$dst), 311 (ins VR64:$src), "movq2dq\t{$src, $dst|$dst, $src}", 312 [(set VR128:$dst, 313 (v2i64 314 (scalar_to_vector 315 (i64 (bitconvert (x86mmx VR64:$src))))))], 316 IIC_MMX_MOVQ_RR>; 317 318let isCodeGenOnly = 1, hasSideEffects = 1 in { 319def MMX_MOVQ2FR64rr: MMXS2SIi8<0xD6, MRMSrcReg, (outs FR64:$dst), 320 (ins VR64:$src), "movq2dq\t{$src, $dst|$dst, $src}", 321 [], IIC_MMX_MOVQ_RR>; 322 323def MMX_MOVFR642Qrr: MMXSDIi8<0xD6, MRMSrcReg, (outs VR64:$dst), 324 (ins FR64:$src), "movdq2q\t{$src, $dst|$dst, $src}", 325 [], IIC_MMX_MOVQ_RR>; 326} 327} // SchedRW 328 329def MMX_MOVNTQmr : MMXI<0xE7, MRMDestMem, (outs), (ins i64mem:$dst, VR64:$src), 330 "movntq\t{$src, $dst|$dst, $src}", 331 [(int_x86_mmx_movnt_dq addr:$dst, VR64:$src)], 332 IIC_MMX_MOVQ_RM>, Sched<[WriteStore]>; 333 334let Predicates = [HasMMX] in { 335 let AddedComplexity = 15 in 336 // movd to MMX register zero-extends 337 def : Pat<(x86mmx (X86vzmovl (x86mmx (scalar_to_vector GR32:$src)))), 338 (MMX_MOVD64rr GR32:$src)>; 339 let AddedComplexity = 20 in 340 def : Pat<(x86mmx (X86vzmovl (x86mmx (scalar_to_vector (loadi32 addr:$src))))), 341 (MMX_MOVD64rm addr:$src)>; 342} 343 344// Arithmetic Instructions 345defm MMX_PABSB : SS3I_unop_rm_int_mm<0x1C, "pabsb", int_x86_ssse3_pabs_b, 346 MMX_INTALU_ITINS>; 347defm MMX_PABSW : SS3I_unop_rm_int_mm<0x1D, "pabsw", int_x86_ssse3_pabs_w, 348 MMX_INTALU_ITINS>; 349defm MMX_PABSD : SS3I_unop_rm_int_mm<0x1E, "pabsd", int_x86_ssse3_pabs_d, 350 MMX_INTALU_ITINS>; 351// -- Addition 352defm MMX_PADDB : MMXI_binop_rm_int<0xFC, "paddb", int_x86_mmx_padd_b, 353 MMX_INTALU_ITINS, 1>; 354defm MMX_PADDW : MMXI_binop_rm_int<0xFD, "paddw", int_x86_mmx_padd_w, 355 MMX_INTALU_ITINS, 1>; 356defm MMX_PADDD : MMXI_binop_rm_int<0xFE, "paddd", int_x86_mmx_padd_d, 357 MMX_INTALU_ITINS, 1>; 358defm MMX_PADDQ : MMXI_binop_rm_int<0xD4, "paddq", int_x86_mmx_padd_q, 359 MMX_INTALUQ_ITINS, 1>; 360defm MMX_PADDSB : MMXI_binop_rm_int<0xEC, "paddsb" , int_x86_mmx_padds_b, 361 MMX_INTALU_ITINS, 1>; 362defm MMX_PADDSW : MMXI_binop_rm_int<0xED, "paddsw" , int_x86_mmx_padds_w, 363 MMX_INTALU_ITINS, 1>; 364 365defm MMX_PADDUSB : MMXI_binop_rm_int<0xDC, "paddusb", int_x86_mmx_paddus_b, 366 MMX_INTALU_ITINS, 1>; 367defm MMX_PADDUSW : MMXI_binop_rm_int<0xDD, "paddusw", int_x86_mmx_paddus_w, 368 MMX_INTALU_ITINS, 1>; 369 370defm MMX_PHADDW : SS3I_binop_rm_int_mm<0x01, "phaddw", int_x86_ssse3_phadd_w, 371 MMX_PHADDSUBW>; 372defm MMX_PHADD : SS3I_binop_rm_int_mm<0x02, "phaddd", int_x86_ssse3_phadd_d, 373 MMX_PHADDSUBD>; 374defm MMX_PHADDSW : SS3I_binop_rm_int_mm<0x03, "phaddsw",int_x86_ssse3_phadd_sw, 375 MMX_PHADDSUBW>; 376 377 378// -- Subtraction 379defm MMX_PSUBB : MMXI_binop_rm_int<0xF8, "psubb", int_x86_mmx_psub_b, 380 MMX_INTALU_ITINS>; 381defm MMX_PSUBW : MMXI_binop_rm_int<0xF9, "psubw", int_x86_mmx_psub_w, 382 MMX_INTALU_ITINS>; 383defm MMX_PSUBD : MMXI_binop_rm_int<0xFA, "psubd", int_x86_mmx_psub_d, 384 MMX_INTALU_ITINS>; 385defm MMX_PSUBQ : MMXI_binop_rm_int<0xFB, "psubq", int_x86_mmx_psub_q, 386 MMX_INTALUQ_ITINS>; 387 388defm MMX_PSUBSB : MMXI_binop_rm_int<0xE8, "psubsb" , int_x86_mmx_psubs_b, 389 MMX_INTALU_ITINS>; 390defm MMX_PSUBSW : MMXI_binop_rm_int<0xE9, "psubsw" , int_x86_mmx_psubs_w, 391 MMX_INTALU_ITINS>; 392 393defm MMX_PSUBUSB : MMXI_binop_rm_int<0xD8, "psubusb", int_x86_mmx_psubus_b, 394 MMX_INTALU_ITINS>; 395defm MMX_PSUBUSW : MMXI_binop_rm_int<0xD9, "psubusw", int_x86_mmx_psubus_w, 396 MMX_INTALU_ITINS>; 397 398defm MMX_PHSUBW : SS3I_binop_rm_int_mm<0x05, "phsubw", int_x86_ssse3_phsub_w, 399 MMX_PHADDSUBW>; 400defm MMX_PHSUBD : SS3I_binop_rm_int_mm<0x06, "phsubd", int_x86_ssse3_phsub_d, 401 MMX_PHADDSUBD>; 402defm MMX_PHSUBSW : SS3I_binop_rm_int_mm<0x07, "phsubsw",int_x86_ssse3_phsub_sw, 403 MMX_PHADDSUBW>; 404 405// -- Multiplication 406defm MMX_PMULLW : MMXI_binop_rm_int<0xD5, "pmullw", int_x86_mmx_pmull_w, 407 MMX_PMUL_ITINS, 1>; 408 409defm MMX_PMULHW : MMXI_binop_rm_int<0xE5, "pmulhw", int_x86_mmx_pmulh_w, 410 MMX_PMUL_ITINS, 1>; 411defm MMX_PMULHUW : MMXI_binop_rm_int<0xE4, "pmulhuw", int_x86_mmx_pmulhu_w, 412 MMX_PMUL_ITINS, 1>; 413defm MMX_PMULUDQ : MMXI_binop_rm_int<0xF4, "pmuludq", int_x86_mmx_pmulu_dq, 414 MMX_PMUL_ITINS, 1>; 415let isCommutable = 1 in 416defm MMX_PMULHRSW : SS3I_binop_rm_int_mm<0x0B, "pmulhrsw", 417 int_x86_ssse3_pmul_hr_sw, MMX_PMUL_ITINS>; 418 419// -- Miscellanea 420defm MMX_PMADDWD : MMXI_binop_rm_int<0xF5, "pmaddwd", int_x86_mmx_pmadd_wd, 421 MMX_PMUL_ITINS, 1>; 422 423defm MMX_PMADDUBSW : SS3I_binop_rm_int_mm<0x04, "pmaddubsw", 424 int_x86_ssse3_pmadd_ub_sw, MMX_PMUL_ITINS>; 425defm MMX_PAVGB : MMXI_binop_rm_int<0xE0, "pavgb", int_x86_mmx_pavg_b, 426 MMX_MISC_FUNC_ITINS, 1>; 427defm MMX_PAVGW : MMXI_binop_rm_int<0xE3, "pavgw", int_x86_mmx_pavg_w, 428 MMX_MISC_FUNC_ITINS, 1>; 429 430defm MMX_PMINUB : MMXI_binop_rm_int<0xDA, "pminub", int_x86_mmx_pminu_b, 431 MMX_MISC_FUNC_ITINS, 1>; 432defm MMX_PMINSW : MMXI_binop_rm_int<0xEA, "pminsw", int_x86_mmx_pmins_w, 433 MMX_MISC_FUNC_ITINS, 1>; 434 435defm MMX_PMAXUB : MMXI_binop_rm_int<0xDE, "pmaxub", int_x86_mmx_pmaxu_b, 436 MMX_MISC_FUNC_ITINS, 1>; 437defm MMX_PMAXSW : MMXI_binop_rm_int<0xEE, "pmaxsw", int_x86_mmx_pmaxs_w, 438 MMX_MISC_FUNC_ITINS, 1>; 439 440defm MMX_PSADBW : MMXI_binop_rm_int<0xF6, "psadbw", int_x86_mmx_psad_bw, 441 MMX_PSADBW_ITINS, 1>; 442 443defm MMX_PSIGNB : SS3I_binop_rm_int_mm<0x08, "psignb", int_x86_ssse3_psign_b, 444 MMX_MISC_FUNC_ITINS>; 445defm MMX_PSIGNW : SS3I_binop_rm_int_mm<0x09, "psignw", int_x86_ssse3_psign_w, 446 MMX_MISC_FUNC_ITINS>; 447defm MMX_PSIGND : SS3I_binop_rm_int_mm<0x0A, "psignd", int_x86_ssse3_psign_d, 448 MMX_MISC_FUNC_ITINS>; 449let Constraints = "$src1 = $dst" in 450 defm MMX_PALIGN : ssse3_palign_mm<"palignr", int_x86_mmx_palignr_b>; 451 452// Logical Instructions 453defm MMX_PAND : MMXI_binop_rm_int<0xDB, "pand", int_x86_mmx_pand, 454 MMX_INTALU_ITINS_VECLOGICSCHED, 1>; 455defm MMX_POR : MMXI_binop_rm_int<0xEB, "por" , int_x86_mmx_por, 456 MMX_INTALU_ITINS_VECLOGICSCHED, 1>; 457defm MMX_PXOR : MMXI_binop_rm_int<0xEF, "pxor", int_x86_mmx_pxor, 458 MMX_INTALU_ITINS_VECLOGICSCHED, 1>; 459defm MMX_PANDN : MMXI_binop_rm_int<0xDF, "pandn", int_x86_mmx_pandn, 460 MMX_INTALU_ITINS_VECLOGICSCHED>; 461 462// Shift Instructions 463defm MMX_PSRLW : MMXI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw", 464 int_x86_mmx_psrl_w, int_x86_mmx_psrli_w, 465 MMX_SHIFT_ITINS>; 466defm MMX_PSRLD : MMXI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld", 467 int_x86_mmx_psrl_d, int_x86_mmx_psrli_d, 468 MMX_SHIFT_ITINS>; 469defm MMX_PSRLQ : MMXI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq", 470 int_x86_mmx_psrl_q, int_x86_mmx_psrli_q, 471 MMX_SHIFT_ITINS>; 472 473def : Pat<(int_x86_mmx_psrl_w VR64:$src1, (load_mvmmx addr:$src2)), 474 (MMX_PSRLWrm VR64:$src1, addr:$src2)>; 475def : Pat<(int_x86_mmx_psrl_d VR64:$src1, (load_mvmmx addr:$src2)), 476 (MMX_PSRLDrm VR64:$src1, addr:$src2)>; 477def : Pat<(int_x86_mmx_psrl_q VR64:$src1, (load_mvmmx addr:$src2)), 478 (MMX_PSRLQrm VR64:$src1, addr:$src2)>; 479 480defm MMX_PSLLW : MMXI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw", 481 int_x86_mmx_psll_w, int_x86_mmx_pslli_w, 482 MMX_SHIFT_ITINS>; 483defm MMX_PSLLD : MMXI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld", 484 int_x86_mmx_psll_d, int_x86_mmx_pslli_d, 485 MMX_SHIFT_ITINS>; 486defm MMX_PSLLQ : MMXI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq", 487 int_x86_mmx_psll_q, int_x86_mmx_pslli_q, 488 MMX_SHIFT_ITINS>; 489 490def : Pat<(int_x86_mmx_psll_w VR64:$src1, (load_mvmmx addr:$src2)), 491 (MMX_PSLLWrm VR64:$src1, addr:$src2)>; 492def : Pat<(int_x86_mmx_psll_d VR64:$src1, (load_mvmmx addr:$src2)), 493 (MMX_PSLLDrm VR64:$src1, addr:$src2)>; 494def : Pat<(int_x86_mmx_psll_q VR64:$src1, (load_mvmmx addr:$src2)), 495 (MMX_PSLLQrm VR64:$src1, addr:$src2)>; 496 497defm MMX_PSRAW : MMXI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw", 498 int_x86_mmx_psra_w, int_x86_mmx_psrai_w, 499 MMX_SHIFT_ITINS>; 500defm MMX_PSRAD : MMXI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad", 501 int_x86_mmx_psra_d, int_x86_mmx_psrai_d, 502 MMX_SHIFT_ITINS>; 503 504def : Pat<(int_x86_mmx_psra_w VR64:$src1, (load_mvmmx addr:$src2)), 505 (MMX_PSRAWrm VR64:$src1, addr:$src2)>; 506def : Pat<(int_x86_mmx_psra_d VR64:$src1, (load_mvmmx addr:$src2)), 507 (MMX_PSRADrm VR64:$src1, addr:$src2)>; 508 509// Comparison Instructions 510defm MMX_PCMPEQB : MMXI_binop_rm_int<0x74, "pcmpeqb", int_x86_mmx_pcmpeq_b, 511 MMX_INTALU_ITINS>; 512defm MMX_PCMPEQW : MMXI_binop_rm_int<0x75, "pcmpeqw", int_x86_mmx_pcmpeq_w, 513 MMX_INTALU_ITINS>; 514defm MMX_PCMPEQD : MMXI_binop_rm_int<0x76, "pcmpeqd", int_x86_mmx_pcmpeq_d, 515 MMX_INTALU_ITINS>; 516 517defm MMX_PCMPGTB : MMXI_binop_rm_int<0x64, "pcmpgtb", int_x86_mmx_pcmpgt_b, 518 MMX_INTALU_ITINS>; 519defm MMX_PCMPGTW : MMXI_binop_rm_int<0x65, "pcmpgtw", int_x86_mmx_pcmpgt_w, 520 MMX_INTALU_ITINS>; 521defm MMX_PCMPGTD : MMXI_binop_rm_int<0x66, "pcmpgtd", int_x86_mmx_pcmpgt_d, 522 MMX_INTALU_ITINS>; 523 524// -- Unpack Instructions 525defm MMX_PUNPCKHBW : MMXI_binop_rm_int<0x68, "punpckhbw", 526 int_x86_mmx_punpckhbw, 527 MMX_UNPCK_H_ITINS>; 528defm MMX_PUNPCKHWD : MMXI_binop_rm_int<0x69, "punpckhwd", 529 int_x86_mmx_punpckhwd, 530 MMX_UNPCK_H_ITINS>; 531defm MMX_PUNPCKHDQ : MMXI_binop_rm_int<0x6A, "punpckhdq", 532 int_x86_mmx_punpckhdq, 533 MMX_UNPCK_H_ITINS>; 534defm MMX_PUNPCKLBW : MMXI_binop_rm_int<0x60, "punpcklbw", 535 int_x86_mmx_punpcklbw, 536 MMX_UNPCK_L_ITINS>; 537defm MMX_PUNPCKLWD : MMXI_binop_rm_int<0x61, "punpcklwd", 538 int_x86_mmx_punpcklwd, 539 MMX_UNPCK_L_ITINS>; 540defm MMX_PUNPCKLDQ : MMXI_binop_rm_int<0x62, "punpckldq", 541 int_x86_mmx_punpckldq, 542 MMX_UNPCK_L_ITINS>; 543 544// -- Pack Instructions 545defm MMX_PACKSSWB : MMXI_binop_rm_int<0x63, "packsswb", int_x86_mmx_packsswb, 546 MMX_PCK_ITINS>; 547defm MMX_PACKSSDW : MMXI_binop_rm_int<0x6B, "packssdw", int_x86_mmx_packssdw, 548 MMX_PCK_ITINS>; 549defm MMX_PACKUSWB : MMXI_binop_rm_int<0x67, "packuswb", int_x86_mmx_packuswb, 550 MMX_PCK_ITINS>; 551 552// -- Shuffle Instructions 553defm MMX_PSHUFB : SS3I_binop_rm_int_mm<0x00, "pshufb", int_x86_ssse3_pshuf_b, 554 MMX_PSHUF_ITINS>; 555 556def MMX_PSHUFWri : MMXIi8<0x70, MRMSrcReg, 557 (outs VR64:$dst), (ins VR64:$src1, u8imm:$src2), 558 "pshufw\t{$src2, $src1, $dst|$dst, $src1, $src2}", 559 [(set VR64:$dst, 560 (int_x86_sse_pshuf_w VR64:$src1, imm:$src2))], 561 IIC_MMX_PSHUF>, Sched<[WriteShuffle]>; 562def MMX_PSHUFWmi : MMXIi8<0x70, MRMSrcMem, 563 (outs VR64:$dst), (ins i64mem:$src1, u8imm:$src2), 564 "pshufw\t{$src2, $src1, $dst|$dst, $src1, $src2}", 565 [(set VR64:$dst, 566 (int_x86_sse_pshuf_w (load_mmx addr:$src1), 567 imm:$src2))], 568 IIC_MMX_PSHUF>, Sched<[WriteShuffleLd]>; 569 570 571 572 573// -- Conversion Instructions 574defm MMX_CVTPS2PI : sse12_cvt_pint<0x2D, VR128, VR64, int_x86_sse_cvtps2pi, 575 f64mem, load, "cvtps2pi\t{$src, $dst|$dst, $src}", 576 MMX_CVT_PS_ITINS, SSEPackedSingle>, PS; 577defm MMX_CVTPD2PI : sse12_cvt_pint<0x2D, VR128, VR64, int_x86_sse_cvtpd2pi, 578 f128mem, memop, "cvtpd2pi\t{$src, $dst|$dst, $src}", 579 MMX_CVT_PD_ITINS, SSEPackedDouble>, PD; 580defm MMX_CVTTPS2PI : sse12_cvt_pint<0x2C, VR128, VR64, int_x86_sse_cvttps2pi, 581 f64mem, load, "cvttps2pi\t{$src, $dst|$dst, $src}", 582 MMX_CVT_PS_ITINS, SSEPackedSingle>, PS; 583defm MMX_CVTTPD2PI : sse12_cvt_pint<0x2C, VR128, VR64, int_x86_sse_cvttpd2pi, 584 f128mem, memop, "cvttpd2pi\t{$src, $dst|$dst, $src}", 585 MMX_CVT_PD_ITINS, SSEPackedDouble>, PD; 586defm MMX_CVTPI2PD : sse12_cvt_pint<0x2A, VR64, VR128, int_x86_sse_cvtpi2pd, 587 i64mem, load, "cvtpi2pd\t{$src, $dst|$dst, $src}", 588 MMX_CVT_PD_ITINS, SSEPackedDouble>, PD; 589let Constraints = "$src1 = $dst" in { 590 defm MMX_CVTPI2PS : sse12_cvt_pint_3addr<0x2A, VR64, VR128, 591 int_x86_sse_cvtpi2ps, 592 i64mem, load, "cvtpi2ps\t{$src2, $dst|$dst, $src2}", 593 SSEPackedSingle>, PS; 594} 595 596// Extract / Insert 597def MMX_PEXTRWirri: MMXIi8<0xC5, MRMSrcReg, 598 (outs GR32orGR64:$dst), (ins VR64:$src1, i32u8imm:$src2), 599 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}", 600 [(set GR32orGR64:$dst, (int_x86_mmx_pextr_w VR64:$src1, 601 imm:$src2))], 602 IIC_MMX_PEXTR>, Sched<[WriteShuffle]>; 603let Constraints = "$src1 = $dst" in { 604 def MMX_PINSRWirri : MMXIi8<0xC4, MRMSrcReg, 605 (outs VR64:$dst), 606 (ins VR64:$src1, GR32orGR64:$src2, i32u8imm:$src3), 607 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}", 608 [(set VR64:$dst, (int_x86_mmx_pinsr_w VR64:$src1, 609 GR32orGR64:$src2, imm:$src3))], 610 IIC_MMX_PINSRW>, Sched<[WriteShuffle]>; 611 612 def MMX_PINSRWirmi : MMXIi8<0xC4, MRMSrcMem, 613 (outs VR64:$dst), 614 (ins VR64:$src1, i16mem:$src2, i32u8imm:$src3), 615 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}", 616 [(set VR64:$dst, (int_x86_mmx_pinsr_w VR64:$src1, 617 (i32 (anyext (loadi16 addr:$src2))), 618 imm:$src3))], 619 IIC_MMX_PINSRW>, Sched<[WriteShuffleLd, ReadAfterLd]>; 620} 621 622// Mask creation 623def MMX_PMOVMSKBrr : MMXI<0xD7, MRMSrcReg, (outs GR32orGR64:$dst), 624 (ins VR64:$src), 625 "pmovmskb\t{$src, $dst|$dst, $src}", 626 [(set GR32orGR64:$dst, 627 (int_x86_mmx_pmovmskb VR64:$src))]>; 628 629 630// Low word of XMM to MMX. 631def MMX_X86movdq2q : SDNode<"X86ISD::MOVDQ2Q", SDTypeProfile<1, 1, 632 [SDTCisVT<0, x86mmx>, SDTCisVT<1, v2i64>]>>; 633 634def : Pat<(x86mmx (MMX_X86movdq2q VR128:$src)), 635 (x86mmx (MMX_MOVDQ2Qrr VR128:$src))>; 636 637def : Pat<(x86mmx (MMX_X86movdq2q (loadv2i64 addr:$src))), 638 (x86mmx (MMX_MOVQ64rm addr:$src))>; 639 640// Misc. 641let SchedRW = [WriteShuffle] in { 642let Uses = [EDI] in 643def MMX_MASKMOVQ : MMXI32<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask), 644 "maskmovq\t{$mask, $src|$src, $mask}", 645 [(int_x86_mmx_maskmovq VR64:$src, VR64:$mask, EDI)], 646 IIC_MMX_MASKMOV>; 647let Uses = [RDI] in 648def MMX_MASKMOVQ64: MMXI64<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask), 649 "maskmovq\t{$mask, $src|$src, $mask}", 650 [(int_x86_mmx_maskmovq VR64:$src, VR64:$mask, RDI)], 651 IIC_MMX_MASKMOV>; 652} 653 654// 64-bit bit convert. 655let Predicates = [HasSSE2] in { 656def : Pat<(x86mmx (bitconvert (i64 GR64:$src))), 657 (MMX_MOVD64to64rr GR64:$src)>; 658def : Pat<(i64 (bitconvert (x86mmx VR64:$src))), 659 (MMX_MOVD64from64rr VR64:$src)>; 660def : Pat<(f64 (bitconvert (x86mmx VR64:$src))), 661 (MMX_MOVQ2FR64rr VR64:$src)>; 662def : Pat<(x86mmx (bitconvert (f64 FR64:$src))), 663 (MMX_MOVFR642Qrr FR64:$src)>; 664} 665 666 667