1//===-- SOPInstructions.td - SOP Instruction Definitions ------------------===// 2// 3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4// See https://llvm.org/LICENSE.txt for license information. 5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6// 7//===----------------------------------------------------------------------===// 8 9def GPRIdxMode : CustomOperand<i32>; 10 11class SOP_Pseudo<string opName, dag outs, dag ins, string asmOps, 12 list<dag> pattern=[]> : 13 InstSI<outs, ins, "", pattern>, 14 SIMCInstr<opName, SIEncodingFamily.NONE> { 15 16 let isPseudo = 1; 17 let isCodeGenOnly = 1; 18 let Size = 4; 19 20 string Mnemonic = opName; 21 string AsmOperands = asmOps; 22 23 bits<1> has_sdst = 0; 24} 25 26//===----------------------------------------------------------------------===// 27// SOP1 Instructions 28//===----------------------------------------------------------------------===// 29 30class SOP1_Pseudo <string opName, dag outs, dag ins, 31 string asmOps, list<dag> pattern=[]> : 32 SOP_Pseudo<opName, outs, ins, " " # asmOps, pattern> { 33 34 let mayLoad = 0; 35 let mayStore = 0; 36 let hasSideEffects = 0; 37 let SALU = 1; 38 let SOP1 = 1; 39 let SchedRW = [WriteSALU]; 40 let UseNamedOperandTable = 1; 41 42 bits<1> has_src0 = 1; 43 let has_sdst = 1; 44} 45 46class SOP1_Real<bits<8> op, SOP1_Pseudo ps, string real_name = ps.Mnemonic> : 47 InstSI <ps.OutOperandList, ps.InOperandList, 48 real_name # ps.AsmOperands>, 49 Enc32 { 50 51 let SALU = 1; 52 let SOP1 = 1; 53 let isPseudo = 0; 54 let isCodeGenOnly = 0; 55 let Size = 4; 56 57 // copy relevant pseudo op flags 58 let SubtargetPredicate = ps.SubtargetPredicate; 59 let AsmMatchConverter = ps.AsmMatchConverter; 60 let SchedRW = ps.SchedRW; 61 let mayLoad = ps.mayLoad; 62 let mayStore = ps.mayStore; 63 64 // encoding 65 bits<7> sdst; 66 bits<8> src0; 67 68 let Inst{7-0} = !if(ps.has_src0, src0, ?); 69 let Inst{15-8} = op; 70 let Inst{22-16} = !if(ps.has_sdst, sdst, ?); 71 let Inst{31-23} = 0x17d; //encoding; 72} 73 74class SOP1_32 <string opName, list<dag> pattern=[], bit tied_in = 0> : SOP1_Pseudo < 75 opName, (outs SReg_32:$sdst), 76 !if(tied_in, (ins SSrc_b32:$src0, SReg_32:$sdst_in), 77 (ins SSrc_b32:$src0)), 78 "$sdst, $src0", pattern> { 79 let Constraints = !if(tied_in, "$sdst = $sdst_in", ""); 80} 81 82// Only register input allowed. 83class SOP1_32R <string opName, list<dag> pattern=[]> : SOP1_Pseudo < 84 opName, (outs SReg_32:$sdst), (ins SReg_32:$src0), 85 "$sdst, $src0", pattern>; 86 87// 32-bit input, no output. 88class SOP1_0_32 <string opName, list<dag> pattern = []> : SOP1_Pseudo < 89 opName, (outs), (ins SSrc_b32:$src0), 90 "$src0", pattern> { 91 let has_sdst = 0; 92} 93 94// Special case for movreld where sdst is treated as a use operand. 95class SOP1_32_movreld <string opName, list<dag> pattern=[]> : SOP1_Pseudo < 96 opName, (outs), (ins SReg_32:$sdst, SSrc_b32:$src0), 97 "$sdst, $src0", pattern>; 98 99// Special case for movreld where sdst is treated as a use operand. 100class SOP1_64_movreld <string opName, list<dag> pattern=[]> : SOP1_Pseudo < 101 opName, (outs), (ins SReg_64:$sdst, SSrc_b64:$src0), 102 "$sdst, $src0", pattern 103>; 104 105class SOP1_0_32R <string opName, list<dag> pattern = []> : SOP1_Pseudo < 106 opName, (outs), (ins SReg_32:$src0), 107 "$src0", pattern> { 108 let has_sdst = 0; 109} 110 111class SOP1_64 <string opName, list<dag> pattern=[]> : SOP1_Pseudo < 112 opName, (outs SReg_64:$sdst), (ins SSrc_b64:$src0), 113 "$sdst, $src0", pattern 114>; 115 116// Only register input allowed. 117class SOP1_64R <string opName, list<dag> pattern=[]> : SOP1_Pseudo < 118 opName, (outs SReg_64:$sdst), (ins SReg_64:$src0), 119 "$sdst, $src0", pattern 120>; 121 122// 64-bit input, 32-bit output. 123class SOP1_32_64 <string opName, list<dag> pattern=[]> : SOP1_Pseudo < 124 opName, (outs SReg_32:$sdst), (ins SSrc_b64:$src0), 125 "$sdst, $src0", pattern 126>; 127 128// 32-bit input, 64-bit output. 129class SOP1_64_32 <string opName, list<dag> pattern=[], bit tied_in = 0> : SOP1_Pseudo < 130 opName, (outs SReg_64:$sdst), 131 !if(tied_in, (ins SSrc_b32:$src0, SReg_64:$sdst_in), 132 (ins SSrc_b32:$src0)), 133 "$sdst, $src0", pattern> { 134 let Constraints = !if(tied_in, "$sdst = $sdst_in", ""); 135} 136 137// no input, 64-bit output. 138class SOP1_64_0 <string opName, list<dag> pattern=[]> : SOP1_Pseudo < 139 opName, (outs SReg_64:$sdst), (ins), "$sdst", pattern> { 140 let has_src0 = 0; 141} 142 143// 64-bit input, no output 144class SOP1_1 <string opName, list<dag> pattern=[]> : SOP1_Pseudo < 145 opName, (outs), (ins SReg_64:$src0), "$src0", pattern> { 146 let has_sdst = 0; 147} 148 149 150class UniformUnaryFrag<SDPatternOperator Op> : PatFrag < 151 (ops node:$src0), 152 (Op $src0), 153 [{ return !N->isDivergent(); }]> { 154 // This check is unnecessary as it's captured by the result register 155 // bank constraint. 156 // 157 // FIXME: Should add a way for the emitter to recognize this is a 158 // trivially true predicate to eliminate the check. 159 let GISelPredicateCode = [{return true;}]; 160} 161 162class UniformBinFrag<SDPatternOperator Op> : PatFrag < 163 (ops node:$src0, node:$src1), 164 (Op $src0, $src1), 165 [{ return !N->isDivergent(); }]> { 166 // This check is unnecessary as it's captured by the result register 167 // bank constraint. 168 // 169 // FIXME: Should add a way for the emitter to recognize this is a 170 // trivially true predicate to eliminate the check. 171 let GISelPredicateCode = [{return true;}]; 172} 173 174class UniformTernaryFrag<SDPatternOperator Op> : PatFrag < 175 (ops node:$src0, node:$src1, node:$src2), 176 (Op $src0, $src1, $src2), 177 [{ return !N->isDivergent(); }]> { 178 // This check is unnecessary as it's captured by the result register 179 // bank constraint. 180 // 181 // FIXME: Should add a way for the emitter to recognize this is a 182 // trivially true predicate to eliminate the check. 183 let GISelPredicateCode = [{return true;}]; 184} 185 186class DivergentBinFrag<SDPatternOperator Op> : PatFrag < 187 (ops node:$src0, node:$src1), 188 (Op $src0, $src1), 189 [{ return N->isDivergent(); }]> { 190 // This check is unnecessary as it's captured by the result register 191 // bank constraint. 192 // 193 // FIXME: Should add a way for the emitter to recognize this is a 194 // trivially true predicate to eliminate the check. 195 let GISelPredicateCode = [{return true;}]; 196} 197 198 199let isMoveImm = 1 in { 200 let isReMaterializable = 1, isAsCheapAsAMove = 1 in { 201 def S_MOV_B32 : SOP1_32 <"s_mov_b32">; 202 def S_MOV_B64 : SOP1_64 <"s_mov_b64">; 203 } // End isReMaterializable = 1 204 205 let Uses = [SCC] in { 206 def S_CMOV_B32 : SOP1_32 <"s_cmov_b32">; 207 def S_CMOV_B64 : SOP1_64 <"s_cmov_b64">; 208 } // End Uses = [SCC] 209} // End isMoveImm = 1 210 211let Defs = [SCC] in { 212 def S_NOT_B32 : SOP1_32 <"s_not_b32", 213 [(set i32:$sdst, (UniformUnaryFrag<not> i32:$src0))] 214 >; 215 216 def S_NOT_B64 : SOP1_64 <"s_not_b64", 217 [(set i64:$sdst, (UniformUnaryFrag<not> i64:$src0))] 218 >; 219 def S_WQM_B32 : SOP1_32 <"s_wqm_b32", 220 [(set i32:$sdst, (int_amdgcn_s_wqm i32:$src0))]>; 221 def S_WQM_B64 : SOP1_64 <"s_wqm_b64", 222 [(set i64:$sdst, (int_amdgcn_s_wqm i64:$src0))]>; 223} // End Defs = [SCC] 224 225 226let WaveSizePredicate = isWave32 in { 227def : GCNPat < 228 (int_amdgcn_wqm_vote i1:$src0), 229 (S_WQM_B32 SSrc_b32:$src0) 230>; 231} 232 233let WaveSizePredicate = isWave64 in { 234def : GCNPat < 235 (int_amdgcn_wqm_vote i1:$src0), 236 (S_WQM_B64 SSrc_b64:$src0) 237>; 238} 239 240let isReMaterializable = 1, isAsCheapAsAMove = 1 in { 241def S_BREV_B32 : SOP1_32 <"s_brev_b32", 242 [(set i32:$sdst, (UniformUnaryFrag<bitreverse> i32:$src0))] 243>; 244def S_BREV_B64 : SOP1_64 <"s_brev_b64", 245 [(set i64:$sdst, (UniformUnaryFrag<bitreverse> i64:$src0))] 246>; 247} // End isReMaterializable = 1, isAsCheapAsAMove = 1 248 249let Defs = [SCC] in { 250def S_BCNT0_I32_B32 : SOP1_32 <"s_bcnt0_i32_b32">; 251def S_BCNT0_I32_B64 : SOP1_32_64 <"s_bcnt0_i32_b64">; 252def S_BCNT1_I32_B32 : SOP1_32 <"s_bcnt1_i32_b32", 253 [(set i32:$sdst, (UniformUnaryFrag<ctpop> i32:$src0))] 254>; 255def S_BCNT1_I32_B64 : SOP1_32_64 <"s_bcnt1_i32_b64", 256 [(set i32:$sdst, (UniformUnaryFrag<ctpop> i64:$src0))] 257>; 258} // End Defs = [SCC] 259 260let isReMaterializable = 1 in { 261def S_FF0_I32_B32 : SOP1_32 <"s_ff0_i32_b32">; 262def S_FF0_I32_B64 : SOP1_32_64 <"s_ff0_i32_b64">; 263def S_FF1_I32_B64 : SOP1_32_64 <"s_ff1_i32_b64", 264 [(set i32:$sdst, (UniformUnaryFrag<AMDGPUffbl_b32> i64:$src0))] 265>; 266 267def S_FF1_I32_B32 : SOP1_32 <"s_ff1_i32_b32", 268 [(set i32:$sdst, (UniformUnaryFrag<AMDGPUffbl_b32> i32:$src0))] 269>; 270 271def S_FLBIT_I32_B32 : SOP1_32 <"s_flbit_i32_b32", 272 [(set i32:$sdst, (UniformUnaryFrag<AMDGPUffbh_u32> i32:$src0))] 273>; 274 275def S_FLBIT_I32_B64 : SOP1_32_64 <"s_flbit_i32_b64", 276 [(set i32:$sdst, (UniformUnaryFrag<AMDGPUffbh_u32> i64:$src0))] 277>; 278def S_FLBIT_I32 : SOP1_32 <"s_flbit_i32", 279 [(set i32:$sdst, (UniformUnaryFrag<AMDGPUffbh_i32> i32:$src0))] 280>; 281def S_FLBIT_I32_I64 : SOP1_32_64 <"s_flbit_i32_i64">; 282def S_SEXT_I32_I8 : SOP1_32 <"s_sext_i32_i8", 283 [(set i32:$sdst, (UniformSextInreg<i8> i32:$src0))] 284>; 285def S_SEXT_I32_I16 : SOP1_32 <"s_sext_i32_i16", 286 [(set i32:$sdst, (UniformSextInreg<i16> i32:$src0))] 287>; 288} // End isReMaterializable = 1 289 290def S_BITSET0_B32 : SOP1_32 <"s_bitset0_b32", [], 1>; 291def S_BITSET0_B64 : SOP1_64_32 <"s_bitset0_b64", [], 1>; 292def S_BITSET1_B32 : SOP1_32 <"s_bitset1_b32", [], 1>; 293def S_BITSET1_B64 : SOP1_64_32 <"s_bitset1_b64", [], 1>; 294 295def S_GETPC_B64 : SOP1_64_0 <"s_getpc_b64">; 296// PSEUDO includes a workaround for a hardware anomaly where some ASICs 297// zero-extend the result from 48 bits instead of sign-extending. 298let isReMaterializable = 1 in 299def S_GETPC_B64_pseudo : SOP1_64_0 <"s_getpc_b64", 300 [(set i64:$sdst, (int_amdgcn_s_getpc))] 301>; 302 303let isTerminator = 1, isBarrier = 1, SchedRW = [WriteBranch] in { 304 305let isBranch = 1, isIndirectBranch = 1 in { 306def S_SETPC_B64 : SOP1_1 <"s_setpc_b64">; 307} // End isBranch = 1, isIndirectBranch = 1 308 309let isReturn = 1 in { 310// Define variant marked as return rather than branch. 311def S_SETPC_B64_return : SOP1_1<"">; 312} 313} // End isTerminator = 1, isBarrier = 1 314 315let isCall = 1 in { 316def S_SWAPPC_B64 : SOP1_64 <"s_swappc_b64" 317>; 318} 319 320def S_RFE_B64 : SOP1_1 <"s_rfe_b64">; 321 322let hasSideEffects = 1, Uses = [EXEC], Defs = [EXEC, SCC] in { 323 324def S_AND_SAVEEXEC_B64 : SOP1_64 <"s_and_saveexec_b64">; 325def S_OR_SAVEEXEC_B64 : SOP1_64 <"s_or_saveexec_b64">; 326def S_XOR_SAVEEXEC_B64 : SOP1_64 <"s_xor_saveexec_b64">; 327def S_ANDN2_SAVEEXEC_B64 : SOP1_64 <"s_andn2_saveexec_b64">; 328def S_ORN2_SAVEEXEC_B64 : SOP1_64 <"s_orn2_saveexec_b64">; 329def S_NAND_SAVEEXEC_B64 : SOP1_64 <"s_nand_saveexec_b64">; 330def S_NOR_SAVEEXEC_B64 : SOP1_64 <"s_nor_saveexec_b64">; 331def S_XNOR_SAVEEXEC_B64 : SOP1_64 <"s_xnor_saveexec_b64">; 332 333} // End hasSideEffects = 1, Uses = [EXEC], Defs = [EXEC, SCC] 334 335def S_QUADMASK_B32 : SOP1_32 <"s_quadmask_b32", 336 [(set i32:$sdst, (int_amdgcn_s_quadmask i32:$src0))]>; 337def S_QUADMASK_B64 : SOP1_64 <"s_quadmask_b64", 338 [(set i64:$sdst, (int_amdgcn_s_quadmask i64:$src0))]>; 339 340let Uses = [M0] in { 341def S_MOVRELS_B32 : SOP1_32R <"s_movrels_b32">; 342def S_MOVRELS_B64 : SOP1_64R <"s_movrels_b64">; 343def S_MOVRELD_B32 : SOP1_32_movreld <"s_movreld_b32">; 344def S_MOVRELD_B64 : SOP1_64_movreld <"s_movreld_b64">; 345} // End Uses = [M0] 346 347let SubtargetPredicate = isGFX6GFX7GFX8GFX9 in { 348def S_CBRANCH_JOIN : SOP1_0_32R <"s_cbranch_join">; 349} // End SubtargetPredicate = isGFX6GFX7GFX8GFX9 350 351let Defs = [SCC] in { 352def S_ABS_I32 : SOP1_32 <"s_abs_i32", 353 [(set i32:$sdst, (UniformUnaryFrag<abs> i32:$src0))] 354 >; 355} // End Defs = [SCC] 356 357let SubtargetPredicate = HasVGPRIndexMode in { 358def S_SET_GPR_IDX_IDX : SOP1_0_32<"s_set_gpr_idx_idx"> { 359 let Uses = [M0, MODE]; 360 let Defs = [M0, MODE]; 361} 362} 363 364let SubtargetPredicate = isGFX9Plus in { 365 let hasSideEffects = 1, Defs = [EXEC, SCC], Uses = [EXEC] in { 366 def S_ANDN1_SAVEEXEC_B64 : SOP1_64<"s_andn1_saveexec_b64">; 367 def S_ORN1_SAVEEXEC_B64 : SOP1_64<"s_orn1_saveexec_b64">; 368 def S_ANDN1_WREXEC_B64 : SOP1_64<"s_andn1_wrexec_b64">; 369 def S_ANDN2_WREXEC_B64 : SOP1_64<"s_andn2_wrexec_b64">; 370 } // End hasSideEffects = 1, Defs = [EXEC, SCC], Uses = [EXEC] 371 372 let isReMaterializable = 1 in 373 def S_BITREPLICATE_B64_B32 : SOP1_64_32<"s_bitreplicate_b64_b32", 374 [(set i64:$sdst, (int_amdgcn_s_bitreplicate i32:$src0))]>; 375} // End SubtargetPredicate = isGFX9Plus 376 377let SubtargetPredicate = isGFX10Plus in { 378 let hasSideEffects = 1, Defs = [EXEC, SCC], Uses = [EXEC] in { 379 def S_AND_SAVEEXEC_B32 : SOP1_32<"s_and_saveexec_b32">; 380 def S_OR_SAVEEXEC_B32 : SOP1_32<"s_or_saveexec_b32">; 381 def S_XOR_SAVEEXEC_B32 : SOP1_32<"s_xor_saveexec_b32">; 382 def S_ANDN2_SAVEEXEC_B32 : SOP1_32<"s_andn2_saveexec_b32">; 383 def S_ORN2_SAVEEXEC_B32 : SOP1_32<"s_orn2_saveexec_b32">; 384 def S_NAND_SAVEEXEC_B32 : SOP1_32<"s_nand_saveexec_b32">; 385 def S_NOR_SAVEEXEC_B32 : SOP1_32<"s_nor_saveexec_b32">; 386 def S_XNOR_SAVEEXEC_B32 : SOP1_32<"s_xnor_saveexec_b32">; 387 def S_ANDN1_SAVEEXEC_B32 : SOP1_32<"s_andn1_saveexec_b32">; 388 def S_ORN1_SAVEEXEC_B32 : SOP1_32<"s_orn1_saveexec_b32">; 389 def S_ANDN1_WREXEC_B32 : SOP1_32<"s_andn1_wrexec_b32">; 390 def S_ANDN2_WREXEC_B32 : SOP1_32<"s_andn2_wrexec_b32">; 391 } // End hasSideEffects = 1, Defs = [EXEC, SCC], Uses = [EXEC] 392 393 let Uses = [M0] in { 394 def S_MOVRELSD_2_B32 : SOP1_32<"s_movrelsd_2_b32">; 395 } // End Uses = [M0] 396} // End SubtargetPredicate = isGFX10Plus 397 398let SubtargetPredicate = isGFX11Plus in { 399 let hasSideEffects = 1 in { 400 // For s_sendmsg_rtn_* the src0 field encodes the message type directly; it 401 // is not an SGPR number. 402 def S_SENDMSG_RTN_B32 : SOP1_Pseudo< 403 "s_sendmsg_rtn_b32", (outs SReg_32:$sdst), (ins SendMsg:$src0), 404 "$sdst, $src0", [(set i32:$sdst, (int_amdgcn_s_sendmsg_rtn timm:$src0))] 405 >; 406 def S_SENDMSG_RTN_B64 : SOP1_Pseudo< 407 "s_sendmsg_rtn_b64", (outs SReg_64:$sdst), (ins SendMsg:$src0), 408 "$sdst, $src0", [(set i64:$sdst, (int_amdgcn_s_sendmsg_rtn timm:$src0))] 409 >; 410 } 411} // End SubtargetPredicate = isGFX11Plus 412 413class SOP1_F32_Inst<string opName, SDPatternOperator Op, ValueType vt0=f32, 414 ValueType vt1=vt0> : 415 SOP1_32<opName, [(set vt0:$sdst, (UniformUnaryFrag<Op> vt1:$src0))]>; 416 417let SubtargetPredicate = HasSALUFloatInsts, Uses = [MODE], 418 SchedRW = [WriteSFPU], isReMaterializable = 1 in { 419 def S_CVT_F32_I32 : SOP1_F32_Inst<"s_cvt_f32_i32", sint_to_fp, f32, i32>; 420 def S_CVT_F32_U32 : SOP1_F32_Inst<"s_cvt_f32_u32", uint_to_fp, f32, i32>; 421 422 let mayRaiseFPException = 1 in { 423 def S_CVT_I32_F32 : SOP1_F32_Inst<"s_cvt_i32_f32", fp_to_sint, i32, f32>; 424 def S_CVT_U32_F32 : SOP1_F32_Inst<"s_cvt_u32_f32", fp_to_uint, i32, f32>; 425 def S_CVT_F32_F16 : SOP1_F32_Inst<"s_cvt_f32_f16", fpextend, f32, f16>; 426 def S_CVT_HI_F32_F16 : SOP1_32<"s_cvt_hi_f32_f16">; 427 428 def S_CEIL_F32 : SOP1_F32_Inst<"s_ceil_f32", fceil>; 429 def S_FLOOR_F32 : SOP1_F32_Inst<"s_floor_f32", ffloor>; 430 def S_TRUNC_F32 : SOP1_F32_Inst<"s_trunc_f32", ftrunc>; 431 def S_RNDNE_F32 : SOP1_F32_Inst<"s_rndne_f32", froundeven>; 432 433 let FPDPRounding = 1 in 434 def S_CVT_F16_F32 : SOP1_F32_Inst<"s_cvt_f16_f32", fpround, f16, f32>; 435 436 def S_CEIL_F16 : SOP1_F32_Inst<"s_ceil_f16", fceil, f16>; 437 def S_FLOOR_F16 : SOP1_F32_Inst<"s_floor_f16", ffloor, f16>; 438 def S_TRUNC_F16 : SOP1_F32_Inst<"s_trunc_f16", ftrunc, f16>; 439 def S_RNDNE_F16 : SOP1_F32_Inst<"s_rndne_f16", froundeven, f16>; 440 } // End mayRaiseFPException = 1 441} // End SubtargetPredicate = HasSALUFloatInsts, Uses = [MODE] 442 // SchedRW = [WriteSFPU], isReMaterializable = 1 443 444let hasSideEffects = 1 in { 445let has_sdst = 0 in { 446let Uses = [M0] in { 447def S_BARRIER_SIGNAL_M0 : SOP1_Pseudo <"s_barrier_signal m0", (outs), (ins), 448 "", [(int_amdgcn_s_barrier_signal_var M0)]>{ 449 let SchedRW = [WriteBarrier]; 450 let isConvergent = 1; 451} 452 453def S_BARRIER_SIGNAL_ISFIRST_M0 : SOP1_Pseudo <"s_barrier_signal_isfirst m0", (outs), (ins), 454 "", [(set SCC, (int_amdgcn_s_barrier_signal_isfirst_var M0))]>{ 455 let Defs = [SCC]; 456 let SchedRW = [WriteBarrier]; 457 let isConvergent = 1; 458} 459 460def S_BARRIER_INIT_M0 : SOP1_Pseudo <"s_barrier_init m0", (outs), (ins), 461 "", []>{ 462 let SchedRW = [WriteBarrier]; 463 let isConvergent = 1; 464} 465 466def S_BARRIER_INIT_IMM : SOP1_Pseudo <"s_barrier_init", (outs), 467 (ins SplitBarrier:$src0), "$src0", []>{ 468 let SchedRW = [WriteBarrier]; 469 let isConvergent = 1; 470} 471 472def S_BARRIER_JOIN_M0 : SOP1_Pseudo <"s_barrier_join m0", (outs), (ins), 473 "", []>{ 474 let SchedRW = [WriteBarrier]; 475 let isConvergent = 1; 476} 477 478def S_WAKEUP_BARRIER_M0 : SOP1_Pseudo <"s_wakeup_barrier m0", (outs), (ins), 479 "", []>{ 480 let SchedRW = [WriteBarrier]; 481 let isConvergent = 1; 482} 483} // End Uses = [M0] 484 485def S_BARRIER_SIGNAL_IMM : SOP1_Pseudo <"s_barrier_signal", (outs), 486 (ins SplitBarrier:$src0), "$src0", [(int_amdgcn_s_barrier_signal timm:$src0)]>{ 487 let SchedRW = [WriteBarrier]; 488 let isConvergent = 1; 489} 490 491def S_BARRIER_SIGNAL_ISFIRST_IMM : SOP1_Pseudo <"s_barrier_signal_isfirst", (outs), 492 (ins SplitBarrier:$src0), "$src0", [(set SCC, (int_amdgcn_s_barrier_signal_isfirst timm:$src0))]>{ 493 let Defs = [SCC]; 494 let SchedRW = [WriteBarrier]; 495 let isConvergent = 1; 496} 497 498def S_BARRIER_JOIN_IMM : SOP1_Pseudo <"s_barrier_join", (outs), 499 (ins SplitBarrier:$src0), "$src0", []>{ 500 let SchedRW = [WriteBarrier]; 501 let isConvergent = 1; 502} 503 504def S_WAKEUP_BARRIER_IMM : SOP1_Pseudo <"s_wakeup_barrier", (outs), 505 (ins SplitBarrier:$src0), "$src0", []>{ 506 let SchedRW = [WriteBarrier]; 507 let isConvergent = 1; 508} 509} // End has_sdst = 0 510 511def S_GET_BARRIER_STATE_IMM : SOP1_Pseudo <"s_get_barrier_state", (outs SSrc_b32:$sdst), 512 (ins SplitBarrier:$src0), "$sdst, $src0", []>{ 513 let SchedRW = [WriteBarrier]; 514 let isConvergent = 1; 515} 516 517def S_GET_BARRIER_STATE_M0 : SOP1_Pseudo <"s_get_barrier_state $sdst, m0", (outs SSrc_b32:$sdst), 518 (ins), "", []>{ 519 let Uses = [M0]; 520 let SchedRW = [WriteBarrier]; 521 let isConvergent = 1; 522} 523} // End hasSideEffects = 1 524 525//===----------------------------------------------------------------------===// 526// SOP2 Instructions 527//===----------------------------------------------------------------------===// 528 529class SOP2_Pseudo<string opName, dag outs, dag ins, 530 string asmOps, list<dag> pattern=[]> : 531 SOP_Pseudo<opName, outs, ins, " " # asmOps, pattern> { 532 533 let mayLoad = 0; 534 let mayStore = 0; 535 let hasSideEffects = 0; 536 let SALU = 1; 537 let SOP2 = 1; 538 let SchedRW = [WriteSALU]; 539 let UseNamedOperandTable = 1; 540 541 let has_sdst = 1; 542 543 // Pseudo instructions have no encodings, but adding this field here allows 544 // us to do: 545 // let sdst = xxx in { 546 // for multiclasses that include both real and pseudo instructions. 547 // field bits<7> sdst = 0; 548} 549 550class SOP2_Real<SOP_Pseudo ps, string real_name = ps.Mnemonic> : 551 InstSI <ps.OutOperandList, ps.InOperandList, 552 real_name # ps.AsmOperands> { 553 let SALU = 1; 554 let SOP2 = 1; 555 let isPseudo = 0; 556 let isCodeGenOnly = 0; 557 558 // copy relevant pseudo op flags 559 let SubtargetPredicate = ps.SubtargetPredicate; 560 let AsmMatchConverter = ps.AsmMatchConverter; 561 let UseNamedOperandTable = ps.UseNamedOperandTable; 562 let TSFlags = ps.TSFlags; 563 let SchedRW = ps.SchedRW; 564 let mayLoad = ps.mayLoad; 565 let mayStore = ps.mayStore; 566 let Constraints = ps.Constraints; 567 let DisableEncoding = ps.DisableEncoding; 568 569 // encoding 570 bits<7> sdst; 571 bits<8> src0; 572 bits<8> src1; 573 bits<32> imm; 574} 575 576class SOP2_Real32<bits<7> op, SOP_Pseudo ps, string real_name = ps.Mnemonic> : 577 SOP2_Real<ps, real_name>, Enc32 { 578 let Inst{7-0} = src0; 579 let Inst{15-8} = src1; 580 let Inst{22-16} = !if(ps.has_sdst, sdst, ?); 581 let Inst{29-23} = op; 582 let Inst{31-30} = 0x2; // encoding 583} 584 585class SOP2_Real64<bits<7> op, SOP_Pseudo ps, string real_name = ps.Mnemonic> : 586 SOP2_Real<ps, real_name>, Enc64 { 587 let Inst{7-0} = src0; 588 let Inst{15-8} = src1; 589 let Inst{22-16} = !if(ps.has_sdst, sdst, ?); 590 let Inst{29-23} = op; 591 let Inst{31-30} = 0x2; // encoding 592 let Inst{63-32} = imm; 593} 594 595class SOP2_F16 <string opName, list<dag> pattern=[]> : SOP2_Pseudo < 596 opName, (outs SReg_32:$sdst), (ins SSrc_f16:$src0, SSrc_f16:$src1), 597 "$sdst, $src0, $src1", pattern 598>; 599 600class SOP2_32 <string opName, list<dag> pattern=[]> : SOP2_Pseudo < 601 opName, (outs SReg_32:$sdst), (ins SSrc_b32:$src0, SSrc_b32:$src1), 602 "$sdst, $src0, $src1", pattern 603>; 604 605class SOP2_F32 <string opName, list<dag> pattern=[]> : SOP2_Pseudo < 606 opName, (outs SReg_32:$sdst), (ins SSrc_f32:$src0, SSrc_f32:$src1), 607 "$sdst, $src0, $src1", pattern 608>; 609 610class SOP2_64 <string opName, list<dag> pattern=[]> : SOP2_Pseudo < 611 opName, (outs SReg_64:$sdst), (ins SSrc_b64:$src0, SSrc_b64:$src1), 612 "$sdst, $src0, $src1", pattern 613>; 614 615class SOP2_64_32 <string opName, list<dag> pattern=[]> : SOP2_Pseudo < 616 opName, (outs SReg_64:$sdst), (ins SSrc_b64:$src0, SSrc_b32:$src1), 617 "$sdst, $src0, $src1", pattern 618>; 619 620class SOP2_64_32_32 <string opName, list<dag> pattern=[]> : SOP2_Pseudo < 621 opName, (outs SReg_64:$sdst), (ins SSrc_b32:$src0, SSrc_b32:$src1), 622 "$sdst, $src0, $src1", pattern 623>; 624 625 626let Defs = [SCC] in { // Carry out goes to SCC 627let isCommutable = 1 in { 628def S_ADD_U32 : SOP2_32 <"s_add_u32">; 629def S_ADD_I32 : SOP2_32 <"s_add_i32", 630 [(set i32:$sdst, (UniformBinFrag<add> SSrc_b32:$src0, SSrc_b32:$src1))] 631>; 632} // End isCommutable = 1 633 634def S_SUB_U32 : SOP2_32 <"s_sub_u32">; 635def S_SUB_I32 : SOP2_32 <"s_sub_i32", 636 [(set i32:$sdst, (UniformBinFrag<sub> SSrc_b32:$src0, SSrc_b32:$src1))] 637>; 638 639let Uses = [SCC] in { // Carry in comes from SCC 640let isCommutable = 1 in { 641def S_ADDC_U32 : SOP2_32 <"s_addc_u32", 642 [(set i32:$sdst, (UniformBinFrag<adde> (i32 SSrc_b32:$src0), (i32 SSrc_b32:$src1)))]>; 643} // End isCommutable = 1 644 645def S_SUBB_U32 : SOP2_32 <"s_subb_u32", 646 [(set i32:$sdst, (UniformBinFrag<sube> (i32 SSrc_b32:$src0), (i32 SSrc_b32:$src1)))]>; 647} // End Uses = [SCC] 648 649let isCommutable = 1 in { 650def S_MIN_I32 : SOP2_32 <"s_min_i32", 651 [(set i32:$sdst, (UniformBinFrag<smin> i32:$src0, i32:$src1))] 652>; 653def S_MIN_U32 : SOP2_32 <"s_min_u32", 654 [(set i32:$sdst, (UniformBinFrag<umin> i32:$src0, i32:$src1))] 655>; 656def S_MAX_I32 : SOP2_32 <"s_max_i32", 657 [(set i32:$sdst, (UniformBinFrag<smax> i32:$src0, i32:$src1))] 658>; 659def S_MAX_U32 : SOP2_32 <"s_max_u32", 660 [(set i32:$sdst, (UniformBinFrag<umax> i32:$src0, i32:$src1))] 661>; 662} // End isCommutable = 1 663} // End Defs = [SCC] 664 665let SubtargetPredicate = isGFX12Plus in { 666 def S_ADD_U64 : SOP2_64<"s_add_u64">{ 667 let isCommutable = 1; 668 } 669 670 def S_SUB_U64 : SOP2_64<"s_sub_u64">; 671 672 def S_MUL_U64 : SOP2_64 <"s_mul_u64", 673 [(set i64:$sdst, (UniformBinFrag<mul> i64:$src0, i64:$src1))]> { 674 let isCommutable = 1; 675 } 676 677 // The higher 32-bits of the inputs contain the sign extension bits. 678 def S_MUL_I64_I32_PSEUDO : SPseudoInstSI < 679 (outs SReg_64:$sdst), (ins SSrc_b64:$src0, SSrc_b64:$src1) 680 >; 681 682 // The higher 32-bits of the inputs are zero. 683 def S_MUL_U64_U32_PSEUDO : SPseudoInstSI < 684 (outs SReg_64:$sdst), (ins SSrc_b64:$src0, SSrc_b64:$src1) 685 >; 686 687} // End SubtargetPredicate = isGFX12Plus 688 689let Uses = [SCC] in { 690 def S_CSELECT_B32 : SOP2_32 <"s_cselect_b32">; 691 def S_CSELECT_B64 : SOP2_64 <"s_cselect_b64">; 692} // End Uses = [SCC] 693 694let Defs = [SCC] in { 695let isCommutable = 1 in { 696def S_AND_B32 : SOP2_32 <"s_and_b32", 697 [(set i32:$sdst, (UniformBinFrag<and> i32:$src0, i32:$src1))] 698>; 699 700def S_AND_B64 : SOP2_64 <"s_and_b64", 701 [(set i64:$sdst, (UniformBinFrag<and> i64:$src0, i64:$src1))] 702>; 703 704def S_OR_B32 : SOP2_32 <"s_or_b32", 705 [(set i32:$sdst, (UniformBinFrag<or> i32:$src0, i32:$src1))] 706>; 707 708def S_OR_B64 : SOP2_64 <"s_or_b64", 709 [(set i64:$sdst, (UniformBinFrag<or> i64:$src0, i64:$src1))] 710>; 711 712def S_XOR_B32 : SOP2_32 <"s_xor_b32", 713 [(set i32:$sdst, (UniformBinFrag<xor> i32:$src0, i32:$src1))] 714>; 715 716def S_XOR_B64 : SOP2_64 <"s_xor_b64", 717 [(set i64:$sdst, (UniformBinFrag<xor> i64:$src0, i64:$src1))] 718>; 719 720def S_XNOR_B32 : SOP2_32 <"s_xnor_b32", 721 [(set i32:$sdst, (UniformUnaryFrag<not> (xor_oneuse i32:$src0, i32:$src1)))] 722>; 723 724def S_XNOR_B64 : SOP2_64 <"s_xnor_b64", 725 [(set i64:$sdst, (UniformUnaryFrag<not> (xor_oneuse i64:$src0, i64:$src1)))] 726>; 727 728def S_NAND_B32 : SOP2_32 <"s_nand_b32", 729 [(set i32:$sdst, (UniformUnaryFrag<not> (and_oneuse i32:$src0, i32:$src1)))] 730>; 731 732def S_NAND_B64 : SOP2_64 <"s_nand_b64", 733 [(set i64:$sdst, (UniformUnaryFrag<not> (and_oneuse i64:$src0, i64:$src1)))] 734>; 735 736def S_NOR_B32 : SOP2_32 <"s_nor_b32", 737 [(set i32:$sdst, (UniformUnaryFrag<not> (or_oneuse i32:$src0, i32:$src1)))] 738>; 739 740def S_NOR_B64 : SOP2_64 <"s_nor_b64", 741 [(set i64:$sdst, (UniformUnaryFrag<not> (or_oneuse i64:$src0, i64:$src1)))] 742>; 743} // End isCommutable = 1 744 745// There are also separate patterns for types other than i32 746def S_ANDN2_B32 : SOP2_32 <"s_andn2_b32", 747 [(set i32:$sdst, (UniformBinFrag<and> i32:$src0, (UniformUnaryFrag<not> i32:$src1)))] 748>; 749 750def S_ANDN2_B64 : SOP2_64 <"s_andn2_b64", 751 [(set i64:$sdst, (UniformBinFrag<and> i64:$src0, (UniformUnaryFrag<not> i64:$src1)))] 752>; 753 754def S_ORN2_B32 : SOP2_32 <"s_orn2_b32", 755 [(set i32:$sdst, (UniformBinFrag<or> i32:$src0, (UniformUnaryFrag<not> i32:$src1)))] 756>; 757 758def S_ORN2_B64 : SOP2_64 <"s_orn2_b64", 759 [(set i64:$sdst, (UniformBinFrag<or> i64:$src0, (UniformUnaryFrag<not> i64:$src1)))] 760>; 761} // End Defs = [SCC] 762 763// Use added complexity so these patterns are preferred to the VALU patterns. 764let AddedComplexity = 1 in { 765 766let Defs = [SCC] in { 767// TODO: b64 versions require VOP3 change since v_lshlrev_b64 is VOP3 768def S_LSHL_B32 : SOP2_32 <"s_lshl_b32", 769 [(set SReg_32:$sdst, (UniformBinFrag<cshl_32> (i32 SSrc_b32:$src0), (i32 SSrc_b32:$src1)))] 770>; 771def S_LSHL_B64 : SOP2_64_32 <"s_lshl_b64", 772 [(set SReg_64:$sdst, (UniformBinFrag<cshl_64> (i64 SSrc_b64:$src0), (i32 SSrc_b32:$src1)))] 773>; 774def S_LSHR_B32 : SOP2_32 <"s_lshr_b32", 775 [(set SReg_32:$sdst, (UniformBinFrag<csrl_32> (i32 SSrc_b32:$src0), (i32 SSrc_b32:$src1)))] 776>; 777def S_LSHR_B64 : SOP2_64_32 <"s_lshr_b64", 778 [(set SReg_64:$sdst, (UniformBinFrag<csrl_64> (i64 SSrc_b64:$src0), (i32 SSrc_b32:$src1)))] 779>; 780def S_ASHR_I32 : SOP2_32 <"s_ashr_i32", 781 [(set SReg_32:$sdst, (UniformBinFrag<csra_32> (i32 SSrc_b32:$src0), (i32 SSrc_b32:$src1)))] 782>; 783def S_ASHR_I64 : SOP2_64_32 <"s_ashr_i64", 784 [(set SReg_64:$sdst, (UniformBinFrag<csra_64> (i64 SSrc_b64:$src0), (i32 SSrc_b32:$src1)))] 785>; 786} // End Defs = [SCC] 787 788let isReMaterializable = 1 in { 789def S_BFM_B32 : SOP2_32 <"s_bfm_b32", 790 [(set i32:$sdst, (UniformBinFrag<AMDGPUbfm> i32:$src0, i32:$src1))]>; 791def S_BFM_B64 : SOP2_64_32_32 <"s_bfm_b64">; 792 793def S_MUL_I32 : SOP2_32 <"s_mul_i32", 794 [(set i32:$sdst, (UniformBinFrag<mul> i32:$src0, i32:$src1))]> { 795 let isCommutable = 1; 796} 797} // End isReMaterializable = 1 798} // End AddedComplexity = 1 799 800let Defs = [SCC] in { 801def S_BFE_U32 : SOP2_32 <"s_bfe_u32">; 802def S_BFE_I32 : SOP2_32 <"s_bfe_i32">; 803def S_BFE_U64 : SOP2_64_32 <"s_bfe_u64">; 804def S_BFE_I64 : SOP2_64_32 <"s_bfe_i64">; 805} // End Defs = [SCC] 806 807def S_CBRANCH_G_FORK : SOP2_Pseudo < 808 "s_cbranch_g_fork", (outs), 809 (ins SCSrc_b64:$src0, SCSrc_b64:$src1), 810 "$src0, $src1" 811> { 812 let has_sdst = 0; 813 let SubtargetPredicate = isGFX6GFX7GFX8GFX9; 814} 815 816let Defs = [SCC] in { 817def S_ABSDIFF_I32 : SOP2_32 <"s_absdiff_i32">; 818} // End Defs = [SCC] 819 820let SubtargetPredicate = isGFX8GFX9 in { 821 def S_RFE_RESTORE_B64 : SOP2_Pseudo < 822 "s_rfe_restore_b64", (outs), 823 (ins SSrc_b64:$src0, SSrc_b32:$src1), 824 "$src0, $src1" 825 > { 826 let hasSideEffects = 1; 827 let has_sdst = 0; 828 } 829} 830 831let SubtargetPredicate = isGFX9Plus in { 832 let isReMaterializable = 1 in { 833 def S_PACK_LL_B32_B16 : SOP2_32<"s_pack_ll_b32_b16">; 834 def S_PACK_LH_B32_B16 : SOP2_32<"s_pack_lh_b32_b16">; 835 def S_PACK_HH_B32_B16 : SOP2_32<"s_pack_hh_b32_b16">; 836 } // End isReMaterializable = 1 837 838 let Defs = [SCC] in { 839 def S_LSHL1_ADD_U32 : SOP2_32<"s_lshl1_add_u32", 840 [(set i32:$sdst, (shl1_add SSrc_b32:$src0, SSrc_b32:$src1))] 841 >; 842 def S_LSHL2_ADD_U32 : SOP2_32<"s_lshl2_add_u32", 843 [(set i32:$sdst, (shl2_add SSrc_b32:$src0, SSrc_b32:$src1))] 844 >; 845 def S_LSHL3_ADD_U32 : SOP2_32<"s_lshl3_add_u32", 846 [(set i32:$sdst, (shl3_add SSrc_b32:$src0, SSrc_b32:$src1))] 847 >; 848 def S_LSHL4_ADD_U32 : SOP2_32<"s_lshl4_add_u32", 849 [(set i32:$sdst, (shl4_add SSrc_b32:$src0, SSrc_b32:$src1))] 850 >; 851 } // End Defs = [SCC] 852 853 let isCommutable = 1, isReMaterializable = 1 in { 854 def S_MUL_HI_U32 : SOP2_32<"s_mul_hi_u32", 855 [(set i32:$sdst, (UniformBinFrag<mulhu> SSrc_b32:$src0, SSrc_b32:$src1))]>; 856 def S_MUL_HI_I32 : SOP2_32<"s_mul_hi_i32", 857 [(set i32:$sdst, (UniformBinFrag<mulhs> SSrc_b32:$src0, SSrc_b32:$src1))]>; 858 } // End isCommutable = 1, isReMaterializable = 1 859} // End SubtargetPredicate = isGFX9Plus 860 861let SubtargetPredicate = isGFX11Plus in { 862 def S_PACK_HL_B32_B16 : SOP2_32<"s_pack_hl_b32_b16">; 863} // End SubtargetPredicate = isGFX11Plus 864 865class SOP2_F32_Inst<string opName, SDPatternOperator Op, ValueType dstVt=f32> : 866 SOP2_F32<opName, 867 [(set dstVt:$sdst, (UniformBinFrag<Op> SSrc_f32:$src0, SSrc_f32:$src1))]>; 868 869class SOP2_F16_Inst<string opName, SDPatternOperator Op> : 870 SOP2_F16<opName, 871 [(set f16:$sdst, (UniformBinFrag<Op> SSrc_f16:$src0, SSrc_f16:$src1))]>; 872 873let SubtargetPredicate = HasSALUFloatInsts, mayRaiseFPException = 1, 874 Uses = [MODE], SchedRW = [WriteSFPU] in { 875 let isReMaterializable = 1 in { 876 let isCommutable = 1 in { 877 def S_ADD_F32 : SOP2_F32_Inst<"s_add_f32", any_fadd>; 878 def S_MIN_F32 : SOP2_F32_Inst<"s_min_f32", fminnum_like>; 879 def S_MAX_F32 : SOP2_F32_Inst<"s_max_f32", fmaxnum_like>; 880 def S_MUL_F32 : SOP2_F32_Inst<"s_mul_f32", any_fmul>; 881 882 let FixedSize = 1 in 883 def S_FMAAK_F32 : SOP2_Pseudo< 884 "s_fmaak_f32", (outs SReg_32:$sdst), 885 (ins SSrc_f32_Deferred:$src0, SSrc_f32_Deferred:$src1, KImmFP32:$imm), 886 "$sdst, $src0, $src1, $imm" 887 >; 888 889 let FPDPRounding = 1 in { 890 def S_ADD_F16 : SOP2_F16_Inst<"s_add_f16", any_fadd>; 891 def S_MUL_F16 : SOP2_F16_Inst<"s_mul_f16", any_fmul>; 892 } // End FPDPRounding 893 894 def S_MIN_F16 : SOP2_F16_Inst<"s_min_f16", fminnum_like>; 895 def S_MAX_F16 : SOP2_F16_Inst<"s_max_f16", fmaxnum_like>; 896 } // End isCommutable = 1 897 898 let FPDPRounding = 1 in 899 def S_SUB_F16 : SOP2_F16_Inst<"s_sub_f16", any_fsub>; 900 901 def S_SUB_F32 : SOP2_F32_Inst<"s_sub_f32", any_fsub>; 902 def S_CVT_PK_RTZ_F16_F32 : SOP2_F32_Inst<"s_cvt_pk_rtz_f16_f32", 903 AMDGPUpkrtz_f16_f32, v2f16>; 904 905 let FixedSize = 1 in 906 def S_FMAMK_F32 : SOP2_Pseudo< 907 "s_fmamk_f32", (outs SReg_32:$sdst), 908 (ins SSrc_f32_Deferred:$src0, KImmFP32:$imm, SSrc_f32_Deferred:$src1), 909 "$sdst, $src0, $imm, $src1" 910 >; 911 } // End isReMaterializable = 1 912 913 let Constraints = "$sdst = $src2", DisableEncoding="$src2", 914 isCommutable = 1, AddedComplexity = 20 in { 915 def S_FMAC_F32 : SOP2_Pseudo< 916 "s_fmac_f32", (outs SReg_32:$sdst), 917 (ins SSrc_f32:$src0, SSrc_f32:$src1, SReg_32:$src2), 918 "$sdst, $src0, $src1", 919 [(set f32:$sdst, (UniformTernaryFrag<any_fma> SSrc_f32:$src0, SSrc_f32:$src1, SReg_32:$src2))] 920 >; 921 922 def S_FMAC_F16 : SOP2_Pseudo< 923 "s_fmac_f16", (outs SReg_32:$sdst), 924 (ins SSrc_f16:$src0, SSrc_f16:$src1, SReg_32:$src2), 925 "$sdst, $src0, $src1", 926 [(set f16:$sdst, (UniformTernaryFrag<any_fma> SSrc_f16:$src0, SSrc_f16:$src1, SReg_32:$src2))] 927 >; 928 } // End Constraints = "$sdst = $src2", DisableEncoding="$src2", 929 // isCommutable = 1, AddedComplexity = 20 930} // End SubtargetPredicate = HasSALUFloatInsts, mayRaiseFPException = 1, 931 // Uses = [MODE], SchedRW = [WriteSFPU] 932 933// On GFX12 MIN/MAX instructions do not read MODE register. 934let SubtargetPredicate = isGFX12Plus, mayRaiseFPException = 1, isCommutable = 1, 935 isReMaterializable = 1, SchedRW = [WriteSFPU] in { 936 def S_MINIMUM_F32 : SOP2_F32_Inst<"s_minimum_f32", fminimum>; 937 def S_MAXIMUM_F32 : SOP2_F32_Inst<"s_maximum_f32", fmaximum>; 938 def S_MINIMUM_F16 : SOP2_F16_Inst<"s_minimum_f16", fminimum>; 939 def S_MAXIMUM_F16 : SOP2_F16_Inst<"s_maximum_f16", fmaximum>; 940} 941 942//===----------------------------------------------------------------------===// 943// SOPK Instructions 944//===----------------------------------------------------------------------===// 945 946class SOPK_Pseudo <string opName, dag outs, dag ins, 947 string asmOps, list<dag> pattern=[]> : 948 SOP_Pseudo<opName, outs, ins, " " # asmOps, pattern> { 949 let mayLoad = 0; 950 let mayStore = 0; 951 let hasSideEffects = 0; 952 let SALU = 1; 953 let SOPK = 1; 954 let FixedSize = 1; 955 let SchedRW = [WriteSALU]; 956 let UseNamedOperandTable = 1; 957 958 let has_sdst = 1; 959} 960 961class SOPK_Real<SOPK_Pseudo ps, string real_name = ps.Mnemonic> : 962 InstSI <ps.OutOperandList, ps.InOperandList, 963 real_name # ps.AsmOperands> { 964 let SALU = 1; 965 let SOPK = 1; 966 let isPseudo = 0; 967 let isCodeGenOnly = 0; 968 let UseNamedOperandTable = 1; 969 970 // copy relevant pseudo op flags 971 let SubtargetPredicate = ps.SubtargetPredicate; 972 let AsmMatchConverter = ps.AsmMatchConverter; 973 let DisableEncoding = ps.DisableEncoding; 974 let Constraints = ps.Constraints; 975 let SchedRW = ps.SchedRW; 976 let mayLoad = ps.mayLoad; 977 let mayStore = ps.mayStore; 978 let isBranch = ps.isBranch; 979 let isCall = ps.isCall; 980 981 // encoding 982 bits<7> sdst; 983 bits<16> simm16; 984 bits<32> imm; 985} 986 987class SOPK_Real32<bits<5> op, SOPK_Pseudo ps, string real_name = ps.Mnemonic> : 988 SOPK_Real <ps, real_name>, 989 Enc32 { 990 let Inst{15-0} = simm16; 991 let Inst{22-16} = !if(ps.has_sdst, sdst, ?); 992 let Inst{27-23} = op; 993 let Inst{31-28} = 0xb; //encoding 994} 995 996class SOPK_Real64<bits<5> op, SOPK_Pseudo ps> : 997 SOPK_Real<ps>, 998 Enc64 { 999 let Inst{15-0} = simm16; 1000 let Inst{22-16} = !if(ps.has_sdst, sdst, ?); 1001 let Inst{27-23} = op; 1002 let Inst{31-28} = 0xb; //encoding 1003 let Inst{63-32} = imm; 1004} 1005 1006class SOPKInstTable <bit is_sopk, string cmpOp = ""> { 1007 bit IsSOPK = is_sopk; 1008 string BaseCmpOp = cmpOp; 1009} 1010 1011class SOPK_32 <string opName, list<dag> pattern=[]> : SOPK_Pseudo < 1012 opName, 1013 (outs SReg_32:$sdst), 1014 (ins s16imm:$simm16), 1015 "$sdst, $simm16", 1016 pattern>; 1017 1018class SOPK_32_BR <string opName, list<dag> pattern=[]> : SOPK_Pseudo < 1019 opName, 1020 (outs), 1021 (ins SOPPBrTarget:$simm16, SReg_32:$sdst), 1022 "$sdst, $simm16", 1023 pattern> { 1024 let Defs = [EXEC]; 1025 let Uses = [EXEC]; 1026 let isBranch = 1; 1027 let isTerminator = 1; 1028 let SchedRW = [WriteBranch]; 1029} 1030 1031class SOPK_SCC <string opName, string base_op, bit isSignExt> : SOPK_Pseudo < 1032 opName, 1033 (outs), 1034 !if(isSignExt, 1035 (ins SReg_32:$sdst, s16imm:$simm16), 1036 (ins SReg_32:$sdst, u16imm:$simm16)), 1037 "$sdst, $simm16">, 1038 SOPKInstTable<1, base_op>{ 1039 let Defs = [SCC]; 1040} 1041 1042class SOPK_32TIE <string opName, list<dag> pattern=[]> : SOPK_Pseudo < 1043 opName, 1044 (outs SReg_32:$sdst), 1045 (ins SReg_32:$src0, s16imm:$simm16), 1046 "$sdst, $simm16", 1047 pattern 1048>; 1049 1050let isReMaterializable = 1, isMoveImm = 1 in { 1051def S_MOVK_I32 : SOPK_32 <"s_movk_i32">; 1052} // End isReMaterializable = 1 1053let Uses = [SCC] in { 1054def S_CMOVK_I32 : SOPK_32 <"s_cmovk_i32">; 1055} 1056 1057let isCompare = 1 in { 1058 1059// This instruction is disabled for now until we can figure out how to teach 1060// the instruction selector to correctly use the S_CMP* vs V_CMP* 1061// instructions. 1062// 1063// When this instruction is enabled the code generator sometimes produces this 1064// invalid sequence: 1065// 1066// SCC = S_CMPK_EQ_I32 SGPR0, imm 1067// VCC = COPY SCC 1068// VGPR0 = V_CNDMASK VCC, VGPR0, VGPR1 1069// 1070// def S_CMPK_EQ_I32 : SOPK_SCC <"s_cmpk_eq_i32", 1071// [(set i1:$dst, (setcc i32:$src0, imm:$src1, SETEQ))] 1072// >; 1073 1074def S_CMPK_EQ_I32 : SOPK_SCC <"s_cmpk_eq_i32", "s_cmp_eq_i32", 1>; 1075def S_CMPK_LG_I32 : SOPK_SCC <"s_cmpk_lg_i32", "s_cmp_lg_i32", 1>; 1076def S_CMPK_GT_I32 : SOPK_SCC <"s_cmpk_gt_i32", "s_cmp_gt_i32", 1>; 1077def S_CMPK_GE_I32 : SOPK_SCC <"s_cmpk_ge_i32", "s_cmp_ge_i32", 1>; 1078def S_CMPK_LT_I32 : SOPK_SCC <"s_cmpk_lt_i32", "s_cmp_lt_i32", 1>; 1079def S_CMPK_LE_I32 : SOPK_SCC <"s_cmpk_le_i32", "s_cmp_le_i32", 1>; 1080 1081let SOPKZext = 1 in { 1082def S_CMPK_EQ_U32 : SOPK_SCC <"s_cmpk_eq_u32", "s_cmp_eq_u32", 0>; 1083def S_CMPK_LG_U32 : SOPK_SCC <"s_cmpk_lg_u32", "s_cmp_lg_u32", 0>; 1084def S_CMPK_GT_U32 : SOPK_SCC <"s_cmpk_gt_u32", "s_cmp_gt_u32", 0>; 1085def S_CMPK_GE_U32 : SOPK_SCC <"s_cmpk_ge_u32", "s_cmp_ge_u32", 0>; 1086def S_CMPK_LT_U32 : SOPK_SCC <"s_cmpk_lt_u32", "s_cmp_lt_u32", 0>; 1087def S_CMPK_LE_U32 : SOPK_SCC <"s_cmpk_le_u32", "s_cmp_le_u32", 0>; 1088} // End SOPKZext = 1 1089} // End isCompare = 1 1090 1091let isCommutable = 1, DisableEncoding = "$src0", 1092 Constraints = "$sdst = $src0" in { 1093 let Defs = [SCC] in 1094 def S_ADDK_I32 : SOPK_32TIE <"s_addk_i32">; 1095 def S_MULK_I32 : SOPK_32TIE <"s_mulk_i32">; 1096} 1097 1098let SubtargetPredicate = isGFX6GFX7GFX8GFX9 in 1099def S_CBRANCH_I_FORK : SOPK_Pseudo < 1100 "s_cbranch_i_fork", 1101 (outs), (ins SReg_64:$sdst, SOPPBrTarget:$simm16), 1102 "$sdst, $simm16" 1103>; 1104 1105// This is hasSideEffects to allow its use in readcyclecounter selection. 1106// FIXME: Need to truncate immediate to 16-bits. 1107// FIXME: Missing mode register use. Should have separate pseudos for 1108// known may read MODE and only read MODE. 1109def S_GETREG_B32 : SOPK_Pseudo < 1110 "s_getreg_b32", 1111 (outs SReg_32:$sdst), (ins hwreg:$simm16), 1112 "$sdst, $simm16", 1113 [(set i32:$sdst, (int_amdgcn_s_getreg (i32 timm:$simm16)))]> { 1114 let SOPKZext = 1; 1115 let hasSideEffects = 1; 1116} 1117 1118let Defs = [MODE], Uses = [MODE] in { 1119 1120// FIXME: Need to truncate immediate to 16-bits. 1121class S_SETREG_B32_Pseudo <list<dag> pattern=[]> : SOPK_Pseudo < 1122 "s_setreg_b32", 1123 (outs), (ins SReg_32:$sdst, hwreg:$simm16), 1124 "$simm16, $sdst", 1125 pattern>; 1126 1127def S_SETREG_B32 : S_SETREG_B32_Pseudo < 1128 [(int_amdgcn_s_setreg (i32 SIMM16bit:$simm16), i32:$sdst)]> { 1129 // Use custom inserter to optimize some cases to 1130 // S_DENORM_MODE/S_ROUND_MODE/S_SETREG_B32_mode. 1131 let usesCustomInserter = 1; 1132 let hasSideEffects = 1; 1133} 1134 1135// Variant of SETREG that is guaranteed to only touch FP bits in the MODE 1136// register, so doesn't have unmodeled side effects. 1137def S_SETREG_B32_mode : S_SETREG_B32_Pseudo { 1138 let hasSideEffects = 0; 1139} 1140 1141// FIXME: Not on SI? 1142//def S_GETREG_REGRD_B32 : SOPK_32 <sopk<0x14, 0x13>, "s_getreg_regrd_b32">; 1143 1144class S_SETREG_IMM32_B32_Pseudo : SOPK_Pseudo < 1145 "s_setreg_imm32_b32", 1146 (outs), (ins i32imm:$imm, hwreg:$simm16), 1147 "$simm16, $imm"> { 1148 let Size = 8; // Unlike every other SOPK instruction. 1149 let has_sdst = 0; 1150} 1151 1152def S_SETREG_IMM32_B32 : S_SETREG_IMM32_B32_Pseudo { 1153 let hasSideEffects = 1; 1154} 1155 1156// Variant of SETREG_IMM32 that is guaranteed to only touch FP bits in the MODE 1157// register, so doesn't have unmodeled side effects. 1158def S_SETREG_IMM32_B32_mode : S_SETREG_IMM32_B32_Pseudo { 1159 let hasSideEffects = 0; 1160} 1161 1162} // End Defs = [MODE], Uses = [MODE] 1163 1164class SOPK_WAITCNT<string opName, list<dag> pat=[]> : 1165 SOPK_Pseudo< 1166 opName, 1167 (outs), 1168 (ins SReg_32:$sdst, s16imm:$simm16), 1169 "$sdst, $simm16", 1170 pat> { 1171 let hasSideEffects = 1; 1172 let mayLoad = 1; 1173 let mayStore = 1; 1174 let has_sdst = 1; // First source takes place of sdst in encoding 1175} 1176 1177let SubtargetPredicate = isGFX9Plus in { 1178 def S_CALL_B64 : SOPK_Pseudo< 1179 "s_call_b64", 1180 (outs SReg_64:$sdst), 1181 (ins SOPPBrTarget:$simm16), 1182 "$sdst, $simm16"> { 1183 let isCall = 1; 1184 } 1185} // End SubtargetPredicate = isGFX9Plus 1186 1187let SubtargetPredicate = isGFX10Plus in { 1188 def S_VERSION : SOPK_Pseudo< 1189 "s_version", 1190 (outs), 1191 (ins s16imm:$simm16), 1192 "$simm16"> { 1193 let has_sdst = 0; 1194 } 1195} // End SubtargetPredicate = isGFX10Plus 1196 1197let SubtargetPredicate = isGFX10GFX11 in { 1198 def S_SUBVECTOR_LOOP_BEGIN : SOPK_32_BR<"s_subvector_loop_begin">; 1199 def S_SUBVECTOR_LOOP_END : SOPK_32_BR<"s_subvector_loop_end">; 1200 1201 def S_WAITCNT_VSCNT : SOPK_WAITCNT<"s_waitcnt_vscnt">; 1202 def S_WAITCNT_VMCNT : SOPK_WAITCNT<"s_waitcnt_vmcnt">; 1203 def S_WAITCNT_EXPCNT : SOPK_WAITCNT<"s_waitcnt_expcnt">; 1204 def S_WAITCNT_LGKMCNT : SOPK_WAITCNT<"s_waitcnt_lgkmcnt">; 1205} // End SubtargetPredicate = isGFX10GFX11 1206 1207//===----------------------------------------------------------------------===// 1208// SOPC Instructions 1209//===----------------------------------------------------------------------===// 1210 1211class SOPC_Pseudo<string opName, dag outs, dag ins, 1212 string asmOps, list<dag> pattern=[]> : 1213 SOP_Pseudo<opName, outs, ins, " " # asmOps, pattern> { 1214 let mayLoad = 0; 1215 let mayStore = 0; 1216 let hasSideEffects = 0; 1217 let SALU = 1; 1218 let SOPC = 1; 1219 let Defs = [SCC]; 1220 let SchedRW = [WriteSALU]; 1221 let UseNamedOperandTable = 1; 1222} 1223 1224class SOPC_Real<bits<7> op, SOPC_Pseudo ps> : 1225 InstSI <ps.OutOperandList, ps.InOperandList, 1226 ps.Mnemonic # ps.AsmOperands>, 1227 Enc32 { 1228 let SALU = 1; 1229 let SOPC = 1; 1230 let isPseudo = 0; 1231 let isCodeGenOnly = 0; 1232 1233 // copy relevant pseudo op flags 1234 let SubtargetPredicate = ps.SubtargetPredicate; 1235 let OtherPredicates = ps.OtherPredicates; 1236 let AsmMatchConverter = ps.AsmMatchConverter; 1237 let UseNamedOperandTable = ps.UseNamedOperandTable; 1238 let TSFlags = ps.TSFlags; 1239 let SchedRW = ps.SchedRW; 1240 let mayLoad = ps.mayLoad; 1241 let mayStore = ps.mayStore; 1242 1243 // encoding 1244 bits<8> src0; 1245 bits<8> src1; 1246 1247 let Inst{7-0} = src0; 1248 let Inst{15-8} = src1; 1249 let Inst{22-16} = op; 1250 let Inst{31-23} = 0x17e; 1251} 1252 1253class SOPC_Base <RegisterOperand rc0, RegisterOperand rc1, 1254 string opName, list<dag> pattern = []> : SOPC_Pseudo < 1255 opName, (outs), (ins rc0:$src0, rc1:$src1), 1256 "$src0, $src1", pattern > { 1257} 1258 1259class SOPC_Helper <RegisterOperand rc, ValueType vt, 1260 string opName, SDPatternOperator cond> : SOPC_Base < 1261 rc, rc, opName, 1262 [(set SCC, (UniformTernaryFrag<setcc> vt:$src0, vt:$src1, cond))] > { 1263} 1264 1265class SOPC_CMP_32<string opName, 1266 SDPatternOperator cond = COND_NULL, string revOp = opName> 1267 : SOPC_Helper<SSrc_b32, i32, opName, cond>, 1268 Commutable_REV<revOp, !eq(revOp, opName)>, 1269 SOPKInstTable<0, opName> { 1270 let isCompare = 1; 1271 let isCommutable = 1; 1272} 1273 1274class SOPC_CMP_F32<string opName, 1275 SDPatternOperator cond = COND_NULL, string revOp = opName> 1276 : SOPC_Helper<SSrc_b32, f32, opName, cond>, 1277 Commutable_REV<revOp, !eq(revOp, opName)>, 1278 SOPKInstTable<0, opName> { 1279 let isCompare = 1; 1280 let isCommutable = 1; 1281 let mayRaiseFPException = 1; 1282 let Uses = [MODE]; 1283 let SchedRW = [WriteSFPU]; 1284} 1285 1286class SOPC_CMP_F16<string opName, 1287 SDPatternOperator cond = COND_NULL, string revOp = opName> 1288 : SOPC_Helper<SSrc_b16, f16, opName, cond>, 1289 Commutable_REV<revOp, !eq(revOp, opName)>, 1290 SOPKInstTable<0, opName> { 1291 let isCompare = 1; 1292 let isCommutable = 1; 1293 let mayRaiseFPException = 1; 1294 let Uses = [MODE]; 1295 let SchedRW = [WriteSFPU]; 1296} 1297 1298class SOPC_CMP_64<string opName, 1299 SDPatternOperator cond = COND_NULL, string revOp = opName> 1300 : SOPC_Helper<SSrc_b64, i64, opName, cond>, 1301 Commutable_REV<revOp, !eq(revOp, opName)> { 1302 let isCompare = 1; 1303 let isCommutable = 1; 1304} 1305 1306class SOPC_32<string opName, list<dag> pattern = []> 1307 : SOPC_Base<SSrc_b32, SSrc_b32, opName, pattern>; 1308 1309class SOPC_64_32<string opName, list<dag> pattern = []> 1310 : SOPC_Base<SSrc_b64, SSrc_b32, opName, pattern>; 1311 1312def S_CMP_EQ_I32 : SOPC_CMP_32 <"s_cmp_eq_i32">; 1313def S_CMP_LG_I32 : SOPC_CMP_32 <"s_cmp_lg_i32">; 1314def S_CMP_GT_I32 : SOPC_CMP_32 <"s_cmp_gt_i32", COND_SGT>; 1315def S_CMP_GE_I32 : SOPC_CMP_32 <"s_cmp_ge_i32", COND_SGE>; 1316def S_CMP_LT_I32 : SOPC_CMP_32 <"s_cmp_lt_i32", COND_SLT, "s_cmp_gt_i32">; 1317def S_CMP_LE_I32 : SOPC_CMP_32 <"s_cmp_le_i32", COND_SLE, "s_cmp_ge_i32">; 1318def S_CMP_EQ_U32 : SOPC_CMP_32 <"s_cmp_eq_u32", COND_EQ>; 1319def S_CMP_LG_U32 : SOPC_CMP_32 <"s_cmp_lg_u32", COND_NE>; 1320def S_CMP_GT_U32 : SOPC_CMP_32 <"s_cmp_gt_u32", COND_UGT>; 1321def S_CMP_GE_U32 : SOPC_CMP_32 <"s_cmp_ge_u32", COND_UGE>; 1322def S_CMP_LT_U32 : SOPC_CMP_32 <"s_cmp_lt_u32", COND_ULT, "s_cmp_gt_u32">; 1323def S_CMP_LE_U32 : SOPC_CMP_32 <"s_cmp_le_u32", COND_ULE, "s_cmp_ge_u32">; 1324 1325def S_BITCMP0_B32 : SOPC_32 <"s_bitcmp0_b32">; 1326def S_BITCMP1_B32 : SOPC_32 <"s_bitcmp1_b32">; 1327def S_BITCMP0_B64 : SOPC_64_32 <"s_bitcmp0_b64">; 1328def S_BITCMP1_B64 : SOPC_64_32 <"s_bitcmp1_b64">; 1329let SubtargetPredicate = isGFX6GFX7GFX8GFX9 in 1330def S_SETVSKIP : SOPC_32 <"s_setvskip">; 1331 1332let SubtargetPredicate = isGFX8Plus in { 1333def S_CMP_EQ_U64 : SOPC_CMP_64 <"s_cmp_eq_u64", COND_EQ>; 1334def S_CMP_LG_U64 : SOPC_CMP_64 <"s_cmp_lg_u64", COND_NE>; 1335} // End SubtargetPredicate = isGFX8Plus 1336 1337let SubtargetPredicate = HasVGPRIndexMode in { 1338// Setting the GPR index mode is really writing the fields in the mode 1339// register. We don't want to add mode register uses to every 1340// instruction, and it's too complicated to deal with anyway. This is 1341// modeled just as a side effect. 1342def S_SET_GPR_IDX_ON : SOPC_Pseudo < 1343 "s_set_gpr_idx_on" , 1344 (outs), 1345 (ins SSrc_b32:$src0, GPRIdxMode:$src1), 1346 "$src0, $src1"> { 1347 let Defs = [M0, MODE]; // No scc def 1348 let Uses = [M0, MODE]; // Other bits of mode, m0 unmodified. 1349 let hasSideEffects = 1; // Sets mode.gpr_idx_en 1350 let FixedSize = 1; 1351} 1352} 1353 1354let SubtargetPredicate = HasSALUFloatInsts in { 1355 1356def S_CMP_LT_F32 : SOPC_CMP_F32<"s_cmp_lt_f32", COND_OLT, "s_cmp_gt_f32">; 1357def S_CMP_EQ_F32 : SOPC_CMP_F32<"s_cmp_eq_f32", COND_OEQ>; 1358def S_CMP_LE_F32 : SOPC_CMP_F32<"s_cmp_le_f32", COND_OLE, "s_cmp_ge_f32">; 1359def S_CMP_GT_F32 : SOPC_CMP_F32<"s_cmp_gt_f32", COND_OGT>; 1360def S_CMP_LG_F32 : SOPC_CMP_F32<"s_cmp_lg_f32", COND_ONE>; 1361def S_CMP_GE_F32 : SOPC_CMP_F32<"s_cmp_ge_f32", COND_OGE>; 1362def S_CMP_O_F32 : SOPC_CMP_F32<"s_cmp_o_f32", COND_O>; 1363def S_CMP_U_F32 : SOPC_CMP_F32<"s_cmp_u_f32", COND_UO>; 1364def S_CMP_NGE_F32 : SOPC_CMP_F32<"s_cmp_nge_f32", COND_ULT, "s_cmp_nle_f32">; 1365def S_CMP_NLG_F32 : SOPC_CMP_F32<"s_cmp_nlg_f32", COND_UEQ>; 1366def S_CMP_NGT_F32 : SOPC_CMP_F32<"s_cmp_ngt_f32", COND_ULE, "s_cmp_nlt_f32">; 1367def S_CMP_NLE_F32 : SOPC_CMP_F32<"s_cmp_nle_f32", COND_UGT>; 1368def S_CMP_NEQ_F32 : SOPC_CMP_F32<"s_cmp_neq_f32", COND_UNE>; 1369def S_CMP_NLT_F32 : SOPC_CMP_F32<"s_cmp_nlt_f32", COND_UGE>; 1370 1371def S_CMP_LT_F16 : SOPC_CMP_F16<"s_cmp_lt_f16", COND_OLT, "s_cmp_gt_f16">; 1372def S_CMP_EQ_F16 : SOPC_CMP_F16<"s_cmp_eq_f16", COND_OEQ>; 1373def S_CMP_LE_F16 : SOPC_CMP_F16<"s_cmp_le_f16", COND_OLE, "s_cmp_ge_f16">; 1374def S_CMP_GT_F16 : SOPC_CMP_F16<"s_cmp_gt_f16", COND_OGT>; 1375def S_CMP_LG_F16 : SOPC_CMP_F16<"s_cmp_lg_f16", COND_ONE>; 1376def S_CMP_GE_F16 : SOPC_CMP_F16<"s_cmp_ge_f16", COND_OGE>; 1377def S_CMP_O_F16 : SOPC_CMP_F16<"s_cmp_o_f16", COND_O>; 1378def S_CMP_U_F16 : SOPC_CMP_F16<"s_cmp_u_f16", COND_UO>; 1379def S_CMP_NGE_F16 : SOPC_CMP_F16<"s_cmp_nge_f16", COND_ULT, "s_cmp_nle_f16">; 1380def S_CMP_NLG_F16 : SOPC_CMP_F16<"s_cmp_nlg_f16", COND_UEQ>; 1381def S_CMP_NGT_F16 : SOPC_CMP_F16<"s_cmp_ngt_f16", COND_ULE, "s_cmp_nlt_f16">; 1382def S_CMP_NLE_F16 : SOPC_CMP_F16<"s_cmp_nle_f16", COND_UGT>; 1383def S_CMP_NEQ_F16 : SOPC_CMP_F16<"s_cmp_neq_f16", COND_UNE>; 1384def S_CMP_NLT_F16 : SOPC_CMP_F16<"s_cmp_nlt_f16", COND_UGE>; 1385 1386} // End SubtargetPredicate = HasSALUFloatInsts 1387 1388//===----------------------------------------------------------------------===// 1389// SOPP Instructions 1390//===----------------------------------------------------------------------===// 1391 1392class SOPP_Pseudo<string opName, dag ins, 1393 string asmOps = "", list<dag> pattern=[], 1394 string sep = !if(!empty(asmOps), "", " "), 1395 string keyName = opName> : 1396 SOP_Pseudo<opName, (outs), ins, sep # asmOps, pattern> { 1397 let mayLoad = 0; 1398 let mayStore = 0; 1399 let hasSideEffects = 0; 1400 let SALU = 1; 1401 let SOPP = 1; 1402 let FixedSize = 1; 1403 let SchedRW = [WriteSALU]; 1404 let UseNamedOperandTable = 1; 1405 bits <16> simm16; 1406 bits <1> fixed_imm = 0; 1407 string KeyName = keyName; 1408} 1409 1410class SOPPRelaxTable <bit isRelaxed, string keyName, string gfxip> { 1411 bit IsRelaxed = isRelaxed; 1412 string KeyName = keyName # gfxip; 1413} 1414 1415class SOPP_Real<SOPP_Pseudo ps, string real_name = ps.Mnemonic> : 1416 InstSI <ps.OutOperandList, ps.InOperandList, 1417 real_name # ps.AsmOperands> { 1418 let SALU = 1; 1419 let SOPP = 1; 1420 let isPseudo = 0; 1421 let isCodeGenOnly = 0; 1422 1423 // copy relevant pseudo op flags 1424 let SubtargetPredicate = ps.SubtargetPredicate; 1425 let OtherPredicates = ps.OtherPredicates; 1426 let AsmMatchConverter = ps.AsmMatchConverter; 1427 let UseNamedOperandTable = ps.UseNamedOperandTable; 1428 let TSFlags = ps.TSFlags; 1429 let SchedRW = ps.SchedRW; 1430 let mayLoad = ps.mayLoad; 1431 let mayStore = ps.mayStore; 1432 bits <16> simm16; 1433} 1434 1435class SOPP_Real_32 <bits<7> op, SOPP_Pseudo ps, string real_name = ps.Mnemonic> : SOPP_Real<ps, real_name>, 1436Enc32 { 1437 let Inst{15-0} = !if(ps.fixed_imm, ps.simm16, simm16); 1438 let Inst{22-16} = op; 1439 let Inst{31-23} = 0x17f; 1440} 1441 1442class SOPP_Real_64 <bits<7> op, SOPP_Pseudo ps, string real_name = ps.Mnemonic> : SOPP_Real<ps, real_name>, 1443Enc64 { 1444 // encoding 1445 let Inst{15-0} = !if(ps.fixed_imm, ps.simm16, simm16); 1446 let Inst{22-16} = op; 1447 let Inst{31-23} = 0x17f; 1448 //effectively a nop 1449 let Inst{47-32} = 0x0; 1450 let Inst{54-48} = 0x0; 1451 let Inst{63-55} = 0x17f; 1452} 1453 1454multiclass SOPP_With_Relaxation <string opName, dag ins, 1455 string asmOps, list<dag> pattern=[]> { 1456 def "" : SOPP_Pseudo <opName, ins, asmOps, pattern>; 1457 def _pad_s_nop : SOPP_Pseudo <opName # "_pad_s_nop", ins, asmOps, pattern, " ", opName>; 1458} 1459 1460def S_NOP : SOPP_Pseudo<"s_nop" , (ins i16imm:$simm16), "$simm16", 1461 [(int_amdgcn_s_nop timm:$simm16)]> { 1462 let hasSideEffects = 1; 1463} 1464 1465let isTerminator = 1 in { 1466def S_ENDPGM : SOPP_Pseudo<"s_endpgm", (ins Endpgm:$simm16), "$simm16", [], ""> { 1467 let isBarrier = 1; 1468 let isReturn = 1; 1469 let hasSideEffects = 1; 1470} 1471 1472def S_ENDPGM_SAVED : SOPP_Pseudo<"s_endpgm_saved", (ins)> { 1473 let SubtargetPredicate = isGFX8Plus; 1474 let simm16 = 0; 1475 let fixed_imm = 1; 1476 let isBarrier = 1; 1477 let isReturn = 1; 1478} 1479 1480let SubtargetPredicate = isGFX9GFX10 in { 1481 let isBarrier = 1, isReturn = 1, simm16 = 0, fixed_imm = 1 in { 1482 def S_ENDPGM_ORDERED_PS_DONE : 1483 SOPP_Pseudo<"s_endpgm_ordered_ps_done", (ins)>; 1484 } // End isBarrier = 1, isReturn = 1, simm16 = 0, fixed_imm = 1 1485} // End SubtargetPredicate = isGFX9GFX10 1486 1487let SubtargetPredicate = isGFX10Plus in { 1488 let isBarrier = 1, isReturn = 1, simm16 = 0, fixed_imm = 1 in { 1489 def S_CODE_END : 1490 SOPP_Pseudo<"s_code_end", (ins)>; 1491 } // End isBarrier = 1, isReturn = 1, simm16 = 0, fixed_imm = 1 1492} // End SubtargetPredicate = isGFX10Plus 1493 1494let isBranch = 1, SchedRW = [WriteBranch] in { 1495let isBarrier = 1 in { 1496defm S_BRANCH : SOPP_With_Relaxation< 1497 "s_branch" , (ins SOPPBrTarget:$simm16), "$simm16", 1498 [(br bb:$simm16)]>; 1499} 1500 1501let Uses = [SCC] in { 1502defm S_CBRANCH_SCC0 : SOPP_With_Relaxation< 1503 "s_cbranch_scc0" , (ins SOPPBrTarget:$simm16), 1504 "$simm16" 1505>; 1506defm S_CBRANCH_SCC1 : SOPP_With_Relaxation < 1507 "s_cbranch_scc1" , (ins SOPPBrTarget:$simm16), 1508 "$simm16" 1509>; 1510} // End Uses = [SCC] 1511 1512let Uses = [VCC] in { 1513defm S_CBRANCH_VCCZ : SOPP_With_Relaxation < 1514 "s_cbranch_vccz" , (ins SOPPBrTarget:$simm16), 1515 "$simm16" 1516>; 1517defm S_CBRANCH_VCCNZ : SOPP_With_Relaxation < 1518 "s_cbranch_vccnz" , (ins SOPPBrTarget:$simm16), 1519 "$simm16" 1520>; 1521} // End Uses = [VCC] 1522 1523let Uses = [EXEC] in { 1524defm S_CBRANCH_EXECZ : SOPP_With_Relaxation < 1525 "s_cbranch_execz" , (ins SOPPBrTarget:$simm16), 1526 "$simm16" 1527>; 1528defm S_CBRANCH_EXECNZ : SOPP_With_Relaxation < 1529 "s_cbranch_execnz" , (ins SOPPBrTarget:$simm16), 1530 "$simm16" 1531>; 1532} // End Uses = [EXEC] 1533 1534defm S_CBRANCH_CDBGSYS : SOPP_With_Relaxation < 1535 "s_cbranch_cdbgsys" , (ins SOPPBrTarget:$simm16), 1536 "$simm16" 1537>; 1538 1539defm S_CBRANCH_CDBGSYS_AND_USER : SOPP_With_Relaxation < 1540 "s_cbranch_cdbgsys_and_user" , (ins SOPPBrTarget:$simm16), 1541 "$simm16" 1542>; 1543 1544defm S_CBRANCH_CDBGSYS_OR_USER : SOPP_With_Relaxation < 1545 "s_cbranch_cdbgsys_or_user" , (ins SOPPBrTarget:$simm16), 1546 "$simm16" 1547>; 1548 1549defm S_CBRANCH_CDBGUSER : SOPP_With_Relaxation < 1550 "s_cbranch_cdbguser" , (ins SOPPBrTarget:$simm16), 1551 "$simm16" 1552>; 1553 1554} // End isBranch = 1 1555} // End isTerminator = 1 1556 1557let hasSideEffects = 1 in { 1558def S_BARRIER : SOPP_Pseudo <"s_barrier", (ins), "", 1559 [(int_amdgcn_s_barrier)]> { 1560 let SchedRW = [WriteBarrier]; 1561 let simm16 = 0; 1562 let fixed_imm = 1; 1563 let isConvergent = 1; 1564} 1565 1566def S_BARRIER_WAIT : SOPP_Pseudo <"s_barrier_wait", (ins i16imm:$simm16), "$simm16", 1567 [(int_amdgcn_s_barrier_wait timm:$simm16)]> { 1568 let SchedRW = [WriteBarrier]; 1569 let isConvergent = 1; 1570} 1571 1572def S_BARRIER_LEAVE : SOPP_Pseudo <"s_barrier_leave", (ins), "", 1573 [(set SCC, (int_amdgcn_s_barrier_leave))]> { 1574 let SchedRW = [WriteBarrier]; 1575 let simm16 = 0; 1576 let fixed_imm = 1; 1577 let isConvergent = 1; 1578 let Defs = [SCC]; 1579} 1580 1581def S_WAKEUP : SOPP_Pseudo <"s_wakeup", (ins) > { 1582 let SubtargetPredicate = isGFX8Plus; 1583 let simm16 = 0; 1584 let fixed_imm = 1; 1585 let mayLoad = 1; 1586 let mayStore = 1; 1587} 1588 1589def S_WAITCNT : SOPP_Pseudo <"s_waitcnt" , (ins SWaitCnt:$simm16), "$simm16", 1590 [(int_amdgcn_s_waitcnt timm:$simm16)]>; 1591 1592// "_soft" waitcnts are waitcnts that are either relaxed into their non-soft 1593// counterpart, or completely removed. 1594// 1595// These are inserted by SIMemoryLegalizer to resolve memory dependencies 1596// and later optimized by SIInsertWaitcnts 1597// For example, a S_WAITCNT_soft 0 can be completely removed in a function 1598// that doesn't access memory. 1599def S_WAITCNT_soft : SOPP_Pseudo <"s_soft_waitcnt" , (ins SWaitCnt:$simm16), "$simm16">; 1600def S_WAITCNT_VSCNT_soft : SOPK_WAITCNT<"s_soft_waitcnt_vscnt">; 1601let SubtargetPredicate = isGFX12Plus in { 1602 def S_WAIT_LOADCNT_soft : SOPP_Pseudo <"s_soft_wait_loadcnt", (ins s16imm:$simm16), "$simm16">; 1603 def S_WAIT_STORECNT_soft : SOPP_Pseudo <"s_soft_wait_storecnt", (ins s16imm:$simm16), "$simm16">; 1604 def S_WAIT_SAMPLECNT_soft : SOPP_Pseudo <"s_soft_wait_samplecnt", (ins s16imm:$simm16), "$simm16">; 1605 def S_WAIT_BVHCNT_soft : SOPP_Pseudo <"s_soft_wait_bvhcnt", (ins s16imm:$simm16), "$simm16">; 1606 def S_WAIT_DSCNT_soft : SOPP_Pseudo <"s_soft_wait_dscnt", (ins s16imm:$simm16), "$simm16">; 1607} 1608 1609def S_SETHALT : SOPP_Pseudo <"s_sethalt" , (ins i32imm:$simm16), "$simm16", 1610 [(int_amdgcn_s_sethalt timm:$simm16)]>; 1611def S_SETKILL : SOPP_Pseudo <"s_setkill" , (ins i16imm:$simm16), "$simm16">; 1612 1613// On SI the documentation says sleep for approximately 64 * low 2 1614// bits, consistent with the reported maximum of 448. On VI the 1615// maximum reported is 960 cycles, so 960 / 64 = 15 max, so is the 1616// maximum really 15 on VI? 1617def S_SLEEP : SOPP_Pseudo <"s_sleep", (ins i32imm:$simm16), 1618 "$simm16", [(int_amdgcn_s_sleep timm:$simm16)]> { 1619} 1620 1621def S_SLEEP_VAR : SOP1_0_32 <"s_sleep_var", [(int_amdgcn_s_sleep_var SSrc_b32:$src0)]> { 1622 let hasSideEffects = 1; 1623} 1624 1625def S_SETPRIO : SOPP_Pseudo <"s_setprio", (ins i16imm:$simm16), "$simm16", 1626 [(int_amdgcn_s_setprio timm:$simm16)]> { 1627} 1628 1629let Uses = [EXEC, M0] in { 1630def S_SENDMSG : SOPP_Pseudo <"s_sendmsg" , (ins SendMsg:$simm16), "$simm16", 1631 [(int_amdgcn_s_sendmsg (i32 timm:$simm16), M0)]> { 1632} 1633 1634def S_SENDMSGHALT : SOPP_Pseudo <"s_sendmsghalt" , (ins SendMsg:$simm16), "$simm16", 1635 [(int_amdgcn_s_sendmsghalt (i32 timm:$simm16), M0)]> { 1636} 1637 1638} // End Uses = [EXEC, M0] 1639 1640def S_TRAP : SOPP_Pseudo <"s_trap" , (ins i16imm:$simm16), "$simm16"> { 1641 let isTrap = 1; 1642} 1643 1644def S_ICACHE_INV : SOPP_Pseudo <"s_icache_inv", (ins)> { 1645 let simm16 = 0; 1646 let fixed_imm = 1; 1647} 1648def S_INCPERFLEVEL : SOPP_Pseudo <"s_incperflevel", (ins i32imm:$simm16), "$simm16", 1649 [(int_amdgcn_s_incperflevel timm:$simm16)]> { 1650} 1651def S_DECPERFLEVEL : SOPP_Pseudo <"s_decperflevel", (ins i32imm:$simm16), "$simm16", 1652 [(int_amdgcn_s_decperflevel timm:$simm16)]> { 1653} 1654 1655let Uses = [M0] in 1656def S_TTRACEDATA : SOPP_Pseudo <"s_ttracedata", (ins), "", 1657 [(int_amdgcn_s_ttracedata M0)]> { 1658 let simm16 = 0; 1659 let fixed_imm = 1; 1660} 1661 1662let SubtargetPredicate = HasVGPRIndexMode in { 1663def S_SET_GPR_IDX_OFF : SOPP_Pseudo<"s_set_gpr_idx_off", (ins) > { 1664 let simm16 = 0; 1665 let fixed_imm = 1; 1666 let Defs = [MODE]; 1667 let Uses = [MODE]; 1668} 1669} 1670} // End hasSideEffects 1671 1672let SubtargetPredicate = HasVGPRIndexMode in { 1673def S_SET_GPR_IDX_MODE : SOPP_Pseudo<"s_set_gpr_idx_mode", (ins GPRIdxMode:$simm16), 1674 "$simm16"> { 1675 let Defs = [M0, MODE]; 1676 let Uses = [MODE]; 1677} 1678} 1679 1680let SubtargetPredicate = isGFX10Plus in { 1681 def S_INST_PREFETCH : 1682 SOPP_Pseudo<"s_inst_prefetch", (ins s16imm:$simm16), "$simm16">; 1683 def S_CLAUSE : 1684 SOPP_Pseudo<"s_clause", (ins s16imm:$simm16), "$simm16">; 1685 def S_WAIT_IDLE : 1686 SOPP_Pseudo <"s_wait_idle", (ins)> { 1687 let simm16 = 0; 1688 let fixed_imm = 1; 1689 } 1690 def S_WAITCNT_DEPCTR : 1691 SOPP_Pseudo <"s_waitcnt_depctr" , (ins DepCtr:$simm16), "$simm16">; 1692 1693 let hasSideEffects = 0, Uses = [MODE], Defs = [MODE] in { 1694 def S_ROUND_MODE : 1695 SOPP_Pseudo<"s_round_mode", (ins s16imm:$simm16), "$simm16">; 1696 def S_DENORM_MODE : 1697 SOPP_Pseudo<"s_denorm_mode", (ins i32imm:$simm16), "$simm16", 1698 [(SIdenorm_mode (i32 timm:$simm16))]>; 1699 } 1700 1701 let hasSideEffects = 1 in 1702 def S_TTRACEDATA_IMM : 1703 SOPP_Pseudo<"s_ttracedata_imm", (ins s16imm:$simm16), "$simm16", 1704 [(int_amdgcn_s_ttracedata_imm timm:$simm16)]>; 1705} // End SubtargetPredicate = isGFX10Plus 1706 1707let SubtargetPredicate = isGFX11Plus in { 1708 def S_WAIT_EVENT : SOPP_Pseudo<"s_wait_event", (ins s16imm:$simm16), 1709 "$simm16"> { 1710 let hasSideEffects = 1; 1711 } 1712 def S_DELAY_ALU : SOPP_Pseudo<"s_delay_alu", (ins SDelayALU:$simm16), 1713 "$simm16">; 1714} // End SubtargetPredicate = isGFX11Plus 1715 1716let SubtargetPredicate = HasVGPRSingleUseHintInsts in { 1717 def S_SINGLEUSE_VDST : 1718 SOPP_Pseudo<"s_singleuse_vdst", (ins s16imm:$simm16), "$simm16">; 1719} // End SubtargetPredicate = HasVGPRSingeUseHintInsts 1720 1721let SubtargetPredicate = isGFX12Plus, hasSideEffects = 1 in { 1722 def S_WAIT_LOADCNT : 1723 SOPP_Pseudo<"s_wait_loadcnt", (ins s16imm:$simm16), "$simm16", 1724 [(int_amdgcn_s_wait_loadcnt timm:$simm16)]>; 1725 def S_WAIT_LOADCNT_DSCNT : 1726 SOPP_Pseudo<"s_wait_loadcnt_dscnt", (ins s16imm:$simm16), "$simm16">; 1727 def S_WAIT_STORECNT : 1728 SOPP_Pseudo<"s_wait_storecnt", (ins s16imm:$simm16), "$simm16", 1729 [(int_amdgcn_s_wait_storecnt timm:$simm16)]>; 1730 def S_WAIT_STORECNT_DSCNT : 1731 SOPP_Pseudo<"s_wait_storecnt_dscnt", (ins s16imm:$simm16), "$simm16">; 1732 def S_WAIT_SAMPLECNT : 1733 SOPP_Pseudo<"s_wait_samplecnt", (ins s16imm:$simm16), "$simm16", 1734 [(int_amdgcn_s_wait_samplecnt timm:$simm16)]>; 1735 def S_WAIT_BVHCNT : 1736 SOPP_Pseudo<"s_wait_bvhcnt", (ins s16imm:$simm16), "$simm16", 1737 [(int_amdgcn_s_wait_bvhcnt timm:$simm16)]>; 1738 def S_WAIT_EXPCNT : 1739 SOPP_Pseudo<"s_wait_expcnt", (ins s16imm:$simm16), "$simm16", 1740 [(int_amdgcn_s_wait_expcnt timm:$simm16)]>; 1741 def S_WAIT_DSCNT : 1742 SOPP_Pseudo<"s_wait_dscnt", (ins s16imm:$simm16), "$simm16", 1743 [(int_amdgcn_s_wait_dscnt timm:$simm16)]>; 1744 def S_WAIT_KMCNT : 1745 SOPP_Pseudo<"s_wait_kmcnt", (ins s16imm:$simm16), "$simm16", 1746 [(int_amdgcn_s_wait_kmcnt timm:$simm16)]>; 1747} // End SubtargetPredicate = isGFX12Plus, hasSideEffects = 1 1748 1749//===----------------------------------------------------------------------===// 1750// SOP1 Patterns 1751//===----------------------------------------------------------------------===// 1752 1753def : GCNPat < 1754 (AMDGPUendpgm), 1755 (S_ENDPGM (i16 0)) 1756>; 1757 1758def : GCNPat < 1759 (int_amdgcn_endpgm), 1760 (S_ENDPGM (i16 0)) 1761>; 1762 1763def : GCNPat < 1764 (i64 (UniformUnaryFrag<ctpop> i64:$src)), 1765 (i64 (REG_SEQUENCE SReg_64, 1766 (i32 (COPY_TO_REGCLASS (S_BCNT1_I32_B64 $src), SReg_32)), sub0, 1767 (S_MOV_B32 (i32 0)), sub1)) 1768>; 1769 1770def : GCNPat < 1771 (i32 (UniformBinFrag<smax> i32:$x, (i32 (ineg i32:$x)))), 1772 (S_ABS_I32 SReg_32:$x) 1773>; 1774 1775def : GCNPat < 1776 (i16 imm:$imm), 1777 (S_MOV_B32 imm:$imm) 1778>; 1779 1780// Same as a 32-bit inreg 1781def : GCNPat< 1782 (i32 (UniformUnaryFrag<sext> i16:$src)), 1783 (S_SEXT_I32_I16 $src) 1784>; 1785 1786let SubtargetPredicate = isNotGFX12Plus in 1787 def : GCNPat <(int_amdgcn_s_wait_event_export_ready), (S_WAIT_EVENT (i16 0))>; 1788let SubtargetPredicate = isGFX12Plus in 1789 def : GCNPat <(int_amdgcn_s_wait_event_export_ready), (S_WAIT_EVENT (i16 1))>; 1790 1791// The first 10 bits of the mode register are the core FP mode on all 1792// subtargets. 1793// 1794// The high bits include additional fields, intermixed with some 1795// non-floating point environment information. We extract the full 1796// register and clear non-relevant bits. 1797// 1798// EXCP_EN covers floating point exceptions, but also some other 1799// non-FP exceptions. 1800// 1801// Bits 12-18 cover the relevant exception mask on all subtargets. 1802// 1803// FIXME: Bit 18 is int_div0, should this be in the FP environment? I 1804// think the only source is v_rcp_iflag_i32. 1805// 1806// On GFX9+: 1807// Bit 23 is the additional FP16_OVFL mode. 1808// 1809// Bits 19, 20, and 21 cover non-FP exceptions and differ between 1810// gfx9/10/11, so we ignore them here. 1811 1812// TODO: Would it be cheaper to emit multiple s_getreg_b32 calls for 1813// the ranges and combine the results? 1814 1815defvar fp_round_mask = !add(!shl(1, 4), -1); 1816defvar fp_denorm_mask = !shl(!add(!shl(1, 4), -1), 4); 1817defvar dx10_clamp_mask = !shl(1, 8); 1818defvar ieee_mode_mask = !shl(1, 9); 1819 1820// Covers fp_round, fp_denorm, dx10_clamp, and IEEE bit. 1821defvar fpmode_mask = 1822 !or(fp_round_mask, fp_denorm_mask, dx10_clamp_mask, ieee_mode_mask); 1823 1824defvar fp_excp_en_mask = !shl(!add(!shl(1, 7), -1), 12); 1825defvar fp16_ovfl = !shl(1, 23); 1826defvar fpmode_mask_gfx6plus = !or(fpmode_mask, fp_excp_en_mask); 1827defvar fpmode_mask_gfx9plus = !or(fpmode_mask_gfx6plus, fp16_ovfl); 1828 1829class GetFPModePat<int fpmode_mask> : GCNPat< 1830 (i32 get_fpmode), 1831 (S_AND_B32 (i32 fpmode_mask), 1832 (S_GETREG_B32 getHwRegImm< 1833 HWREG.MODE, 0, 1834 !add(!logtwo(fpmode_mask), 1)>.ret)) 1835>; 1836 1837// TODO: Might be worth moving to custom lowering so the and is 1838// exposed to demanded bits optimizations. Most users probably only 1839// care about the rounding or denorm mode bits. We also can reduce the 1840// demanded read from the getreg immediate. 1841let SubtargetPredicate = isGFX9Plus in { 1842// Last bit = FP16_OVFL 1843def : GetFPModePat<fpmode_mask_gfx9plus>; 1844} 1845 1846// Last bit = EXCP_EN.int_div0 1847let SubtargetPredicate = isNotGFX9Plus in { 1848def : GetFPModePat<fpmode_mask_gfx6plus>; 1849} 1850 1851//===----------------------------------------------------------------------===// 1852// SOP2 Patterns 1853//===----------------------------------------------------------------------===// 1854 1855def UniformSelect : PatFrag< 1856 (ops node:$src0, node:$src1), 1857 (select SCC, $src0, $src1), 1858 [{ return !N->isDivergent(); }] 1859>; 1860 1861let AddedComplexity = 20 in { 1862 def : GCNPat< 1863 (i32 (UniformSelect i32:$src0, i32:$src1)), 1864 (S_CSELECT_B32 SSrc_b32:$src0, SSrc_b32:$src1) 1865 >; 1866 1867 // TODO: The predicate should not be necessary, but enabling this pattern for 1868 // all subtargets generates worse code in some cases. 1869 let OtherPredicates = [HasPseudoScalarTrans] in 1870 def : GCNPat< 1871 (f32 (UniformSelect f32:$src0, f32:$src1)), 1872 (S_CSELECT_B32 SSrc_b32:$src0, SSrc_b32:$src1) 1873 >; 1874} 1875 1876// V_ADD_I32_e32/S_ADD_U32 produces carry in VCC/SCC. For the vector 1877// case, the sgpr-copies pass will fix this to use the vector version. 1878def : GCNPat < 1879 (i32 (addc i32:$src0, i32:$src1)), 1880 (S_ADD_U32 $src0, $src1) 1881>; 1882 1883// FIXME: We need to use COPY_TO_REGCLASS to work-around the fact that 1884// REG_SEQUENCE patterns don't support instructions with multiple 1885// outputs. 1886def : GCNPat< 1887 (i64 (UniformUnaryFrag<zext> i16:$src)), 1888 (REG_SEQUENCE SReg_64, 1889 (i32 (COPY_TO_REGCLASS (S_AND_B32 $src, (S_MOV_B32 (i32 0xffff))), SGPR_32)), sub0, 1890 (S_MOV_B32 (i32 0)), sub1) 1891>; 1892 1893def : GCNPat < 1894 (i64 (UniformUnaryFrag<sext> i16:$src)), 1895 (REG_SEQUENCE SReg_64, (i32 (S_SEXT_I32_I16 $src)), sub0, 1896 (i32 (COPY_TO_REGCLASS (S_ASHR_I32 (i32 (S_SEXT_I32_I16 $src)), (S_MOV_B32 (i32 31))), SGPR_32)), sub1) 1897>; 1898 1899def : GCNPat< 1900 (i32 (UniformUnaryFrag<zext> i16:$src)), 1901 (S_AND_B32 (S_MOV_B32 (i32 0xffff)), $src) 1902>; 1903 1904// FIXME: ValueType should have isVector field 1905class ScalarNot2Pat<Instruction inst, SDPatternOperator op, ValueType vt, 1906 bit isVector = 1> : GCNPat< 1907 (UniformBinFrag<op> vt:$src0, (UniformUnaryFrag<!if(isVector, vnot, not)> vt:$src1)), 1908 (inst getSOPSrcForVT<vt>.ret:$src0, getSOPSrcForVT<vt>.ret:$src1) 1909>; 1910 1911// Match these for some more types 1912// TODO: i1 1913def : ScalarNot2Pat<S_ANDN2_B32, and, i16, 0>; 1914def : ScalarNot2Pat<S_ANDN2_B32, and, v2i16>; 1915def : ScalarNot2Pat<S_ANDN2_B64, and, v4i16>; 1916def : ScalarNot2Pat<S_ANDN2_B64, and, v2i32>; 1917 1918def : ScalarNot2Pat<S_ORN2_B32, or, i16, 0>; 1919def : ScalarNot2Pat<S_ORN2_B32, or, v2i16>; 1920def : ScalarNot2Pat<S_ORN2_B64, or, v4i16>; 1921def : ScalarNot2Pat<S_ORN2_B64, or, v2i32>; 1922 1923//===----------------------------------------------------------------------===// 1924// Target-specific instruction encodings. 1925//===----------------------------------------------------------------------===// 1926 1927class Select_gfx12<string opName> : SIMCInstr<opName, SIEncodingFamily.GFX12> { 1928 Predicate AssemblerPredicate = isGFX12Only; 1929 string DecoderNamespace = "GFX12"; 1930} 1931 1932class Select_gfx11<string opName> : SIMCInstr<opName, SIEncodingFamily.GFX11> { 1933 Predicate AssemblerPredicate = isGFX11Only; 1934 string DecoderNamespace = "GFX11"; 1935} 1936 1937class Select_gfx10<string opName> : SIMCInstr<opName, SIEncodingFamily.GFX10> { 1938 Predicate AssemblerPredicate = isGFX10Only; 1939 string DecoderNamespace = "GFX10"; 1940} 1941 1942class Select_vi<string opName> : SIMCInstr<opName, SIEncodingFamily.VI> { 1943 Predicate AssemblerPredicate = isGFX8GFX9; 1944 string DecoderNamespace = "GFX8"; 1945} 1946 1947class Select_gfx6_gfx7<string opName> : SIMCInstr<opName, SIEncodingFamily.SI> { 1948 Predicate AssemblerPredicate = isGFX6GFX7; 1949 string DecoderNamespace = "GFX6GFX7"; 1950} 1951 1952//===----------------------------------------------------------------------===// 1953// SOP1 - GFX11, GFX12 1954//===----------------------------------------------------------------------===// 1955 1956multiclass SOP1_Real_gfx12<bits<8> op> { 1957 def _gfx12 : SOP1_Real<op, !cast<SOP1_Pseudo>(NAME)>, 1958 Select_gfx12<!cast<SOP1_Pseudo>(NAME).Mnemonic>; 1959} 1960 1961multiclass SOP1_M0_Real_gfx12<bits<8> op> { 1962 def _gfx12 : SOP1_Real<op, !cast<SOP1_Pseudo>(NAME)>, 1963 Select_gfx12<!cast<SOP1_Pseudo>(NAME).Mnemonic> { 1964 let Inst{7-0} = M0_gfx11plus.HWEncoding{7-0}; // Set Src0 encoding to M0 1965 } 1966} 1967 1968multiclass SOP1_Real_gfx11<bits<8> op> { 1969 def _gfx11 : SOP1_Real<op, !cast<SOP1_Pseudo>(NAME)>, 1970 Select_gfx11<!cast<SOP1_Pseudo>(NAME).Mnemonic>; 1971} 1972 1973multiclass SOP1_Real_Renamed_gfx12<bits<8> op, SOP1_Pseudo backing_pseudo, string real_name> { 1974 def _gfx12 : SOP1_Real<op, backing_pseudo, real_name>, 1975 Select_gfx12<backing_pseudo.Mnemonic>, 1976 MnemonicAlias<backing_pseudo.Mnemonic, real_name>, Requires<[isGFX12Plus]>; 1977} 1978 1979multiclass SOP1_Real_Renamed_gfx11<bits<8> op, SOP1_Pseudo backing_pseudo, string real_name> { 1980 def _gfx11 : SOP1_Real<op, backing_pseudo, real_name>, 1981 Select_gfx11<backing_pseudo.Mnemonic>, 1982 MnemonicAlias<backing_pseudo.Mnemonic, real_name>, Requires<[isGFX11Only]>; 1983} 1984 1985multiclass SOP1_Real_gfx11_gfx12<bits<8> op> : 1986 SOP1_Real_gfx11<op>, SOP1_Real_gfx12<op>; 1987 1988multiclass SOP1_Real_Renamed_gfx11_gfx12<bits<8> op, SOP1_Pseudo backing_pseudo, string real_name> : 1989 SOP1_Real_Renamed_gfx11<op, backing_pseudo, real_name>, 1990 SOP1_Real_Renamed_gfx12<op, backing_pseudo, real_name>; 1991 1992defm S_MOV_B32 : SOP1_Real_gfx11_gfx12<0x000>; 1993defm S_MOV_B64 : SOP1_Real_gfx11_gfx12<0x001>; 1994defm S_CMOV_B32 : SOP1_Real_gfx11_gfx12<0x002>; 1995defm S_CMOV_B64 : SOP1_Real_gfx11_gfx12<0x003>; 1996defm S_BREV_B32 : SOP1_Real_gfx11_gfx12<0x004>; 1997defm S_BREV_B64 : SOP1_Real_gfx11_gfx12<0x005>; 1998defm S_CTZ_I32_B32 : SOP1_Real_Renamed_gfx11_gfx12<0x008, S_FF1_I32_B32, "s_ctz_i32_b32">; 1999defm S_CTZ_I32_B64 : SOP1_Real_Renamed_gfx11_gfx12<0x009, S_FF1_I32_B64, "s_ctz_i32_b64">; 2000defm S_CLZ_I32_U32 : SOP1_Real_Renamed_gfx11_gfx12<0x00a, S_FLBIT_I32_B32, "s_clz_i32_u32">; 2001defm S_CLZ_I32_U64 : SOP1_Real_Renamed_gfx11_gfx12<0x00b, S_FLBIT_I32_B64, "s_clz_i32_u64">; 2002defm S_CLS_I32 : SOP1_Real_Renamed_gfx11_gfx12<0x00c, S_FLBIT_I32, "s_cls_i32">; 2003defm S_CLS_I32_I64 : SOP1_Real_Renamed_gfx11_gfx12<0x00d, S_FLBIT_I32_I64, "s_cls_i32_i64">; 2004defm S_SEXT_I32_I8 : SOP1_Real_gfx11_gfx12<0x00e>; 2005defm S_SEXT_I32_I16 : SOP1_Real_gfx11_gfx12<0x00f>; 2006defm S_BITSET0_B32 : SOP1_Real_gfx11_gfx12<0x010>; 2007defm S_BITSET0_B64 : SOP1_Real_gfx11_gfx12<0x011>; 2008defm S_BITSET1_B32 : SOP1_Real_gfx11_gfx12<0x012>; 2009defm S_BITSET1_B64 : SOP1_Real_gfx11_gfx12<0x013>; 2010defm S_BITREPLICATE_B64_B32 : SOP1_Real_gfx11_gfx12<0x014>; 2011defm S_ABS_I32 : SOP1_Real_gfx11_gfx12<0x015>; 2012defm S_BCNT0_I32_B32 : SOP1_Real_gfx11_gfx12<0x016>; 2013defm S_BCNT0_I32_B64 : SOP1_Real_gfx11_gfx12<0x017>; 2014defm S_BCNT1_I32_B32 : SOP1_Real_gfx11_gfx12<0x018>; 2015defm S_BCNT1_I32_B64 : SOP1_Real_gfx11_gfx12<0x019>; 2016defm S_QUADMASK_B32 : SOP1_Real_gfx11_gfx12<0x01a>; 2017defm S_QUADMASK_B64 : SOP1_Real_gfx11_gfx12<0x01b>; 2018defm S_WQM_B32 : SOP1_Real_gfx11_gfx12<0x01c>; 2019defm S_WQM_B64 : SOP1_Real_gfx11_gfx12<0x01d>; 2020defm S_NOT_B32 : SOP1_Real_gfx11_gfx12<0x01e>; 2021defm S_NOT_B64 : SOP1_Real_gfx11_gfx12<0x01f>; 2022defm S_AND_SAVEEXEC_B32 : SOP1_Real_gfx11_gfx12<0x020>; 2023defm S_AND_SAVEEXEC_B64 : SOP1_Real_gfx11_gfx12<0x021>; 2024defm S_OR_SAVEEXEC_B32 : SOP1_Real_gfx11_gfx12<0x022>; 2025defm S_OR_SAVEEXEC_B64 : SOP1_Real_gfx11_gfx12<0x023>; 2026defm S_XOR_SAVEEXEC_B32 : SOP1_Real_gfx11_gfx12<0x024>; 2027defm S_XOR_SAVEEXEC_B64 : SOP1_Real_gfx11_gfx12<0x025>; 2028defm S_NAND_SAVEEXEC_B32 : SOP1_Real_gfx11_gfx12<0x026>; 2029defm S_NAND_SAVEEXEC_B64 : SOP1_Real_gfx11_gfx12<0x027>; 2030defm S_NOR_SAVEEXEC_B32 : SOP1_Real_gfx11_gfx12<0x028>; 2031defm S_NOR_SAVEEXEC_B64 : SOP1_Real_gfx11_gfx12<0x029>; 2032defm S_XNOR_SAVEEXEC_B32 : SOP1_Real_gfx11_gfx12<0x02a>; 2033/*defm S_XNOR_SAVEEXEC_B64 : SOP1_Real_gfx11_gfx12<0x02b>; //same as older arch, handled there*/ 2034defm S_AND_NOT0_SAVEEXEC_B32 : SOP1_Real_Renamed_gfx11_gfx12<0x02c, S_ANDN1_SAVEEXEC_B32, "s_and_not0_saveexec_b32">; 2035defm S_AND_NOT0_SAVEEXEC_B64 : SOP1_Real_Renamed_gfx11_gfx12<0x02d, S_ANDN1_SAVEEXEC_B64, "s_and_not0_saveexec_b64">; 2036defm S_OR_NOT0_SAVEEXEC_B32 : SOP1_Real_Renamed_gfx11_gfx12<0x02e, S_ORN1_SAVEEXEC_B32, "s_or_not0_saveexec_b32">; 2037defm S_OR_NOT0_SAVEEXEC_B64 : SOP1_Real_Renamed_gfx11_gfx12<0x02f, S_ORN1_SAVEEXEC_B64, "s_or_not0_saveexec_b64">; 2038defm S_AND_NOT1_SAVEEXEC_B32 : SOP1_Real_Renamed_gfx11_gfx12<0x030, S_ANDN2_SAVEEXEC_B32, "s_and_not1_saveexec_b32">; 2039defm S_AND_NOT1_SAVEEXEC_B64 : SOP1_Real_Renamed_gfx11_gfx12<0x031, S_ANDN2_SAVEEXEC_B64, "s_and_not1_saveexec_b64">; 2040defm S_OR_NOT1_SAVEEXEC_B32 : SOP1_Real_Renamed_gfx11_gfx12<0x032, S_ORN2_SAVEEXEC_B32, "s_or_not1_saveexec_b32">; 2041defm S_OR_NOT1_SAVEEXEC_B64 : SOP1_Real_Renamed_gfx11_gfx12<0x033, S_ORN2_SAVEEXEC_B64, "s_or_not1_saveexec_b64">; 2042defm S_AND_NOT0_WREXEC_B32 : SOP1_Real_Renamed_gfx11_gfx12<0x034, S_ANDN1_WREXEC_B32, "s_and_not0_wrexec_b32">; 2043defm S_AND_NOT0_WREXEC_B64 : SOP1_Real_Renamed_gfx11_gfx12<0x035, S_ANDN1_WREXEC_B64, "s_and_not0_wrexec_b64">; 2044defm S_AND_NOT1_WREXEC_B32 : SOP1_Real_Renamed_gfx11_gfx12<0x036, S_ANDN2_WREXEC_B32, "s_and_not1_wrexec_b32">; 2045defm S_AND_NOT1_WREXEC_B64 : SOP1_Real_Renamed_gfx11_gfx12<0x037, S_ANDN2_WREXEC_B64, "s_and_not1_wrexec_b64">; 2046defm S_MOVRELS_B32 : SOP1_Real_gfx11_gfx12<0x040>; 2047defm S_MOVRELS_B64 : SOP1_Real_gfx11_gfx12<0x041>; 2048defm S_MOVRELD_B32 : SOP1_Real_gfx11_gfx12<0x042>; 2049defm S_MOVRELD_B64 : SOP1_Real_gfx11_gfx12<0x043>; 2050defm S_MOVRELSD_2_B32 : SOP1_Real_gfx11_gfx12<0x044>; 2051defm S_GETPC_B64 : SOP1_Real_gfx11_gfx12<0x047>; 2052defm S_SETPC_B64 : SOP1_Real_gfx11_gfx12<0x048>; 2053defm S_SWAPPC_B64 : SOP1_Real_gfx11_gfx12<0x049>; 2054defm S_RFE_B64 : SOP1_Real_gfx11_gfx12<0x04a>; 2055defm S_SENDMSG_RTN_B32 : SOP1_Real_gfx11_gfx12<0x04c>; 2056defm S_SENDMSG_RTN_B64 : SOP1_Real_gfx11_gfx12<0x04d>; 2057defm S_BARRIER_SIGNAL_M0 : SOP1_M0_Real_gfx12<0x04e>; 2058defm S_BARRIER_SIGNAL_ISFIRST_M0 : SOP1_M0_Real_gfx12<0x04f>; 2059defm S_GET_BARRIER_STATE_M0 : SOP1_M0_Real_gfx12<0x050>; 2060defm S_BARRIER_INIT_M0 : SOP1_M0_Real_gfx12<0x051>; 2061defm S_BARRIER_JOIN_M0 : SOP1_M0_Real_gfx12<0x052>; 2062defm S_WAKEUP_BARRIER_M0 : SOP1_M0_Real_gfx12<0x057>; 2063defm S_BARRIER_SIGNAL_IMM : SOP1_Real_gfx12<0x04e>; 2064defm S_BARRIER_SIGNAL_ISFIRST_IMM : SOP1_Real_gfx12<0x04f>; 2065defm S_GET_BARRIER_STATE_IMM : SOP1_Real_gfx12<0x050>; 2066defm S_BARRIER_INIT_IMM : SOP1_Real_gfx12<0x051>; 2067defm S_BARRIER_JOIN_IMM : SOP1_Real_gfx12<0x052>; 2068defm S_WAKEUP_BARRIER_IMM : SOP1_Real_gfx12<0x057>; 2069defm S_SLEEP_VAR : SOP1_Real_gfx12<0x058>; 2070 2071//===----------------------------------------------------------------------===// 2072// SOP1 - GFX1150, GFX12 2073//===----------------------------------------------------------------------===// 2074 2075defm S_CEIL_F32 : SOP1_Real_gfx11_gfx12<0x060>; 2076defm S_FLOOR_F32 : SOP1_Real_gfx11_gfx12<0x061>; 2077defm S_TRUNC_F32 : SOP1_Real_gfx11_gfx12<0x062>; 2078defm S_RNDNE_F32 : SOP1_Real_gfx11_gfx12<0x063>; 2079defm S_CVT_F32_I32 : SOP1_Real_gfx11_gfx12<0x064>; 2080defm S_CVT_F32_U32 : SOP1_Real_gfx11_gfx12<0x065>; 2081defm S_CVT_I32_F32 : SOP1_Real_gfx11_gfx12<0x066>; 2082defm S_CVT_U32_F32 : SOP1_Real_gfx11_gfx12<0x067>; 2083defm S_CVT_F16_F32 : SOP1_Real_gfx11_gfx12<0x068>; 2084defm S_CVT_F32_F16 : SOP1_Real_gfx11_gfx12<0x069>; 2085defm S_CVT_HI_F32_F16 : SOP1_Real_gfx11_gfx12<0x06a>; 2086defm S_CEIL_F16 : SOP1_Real_gfx11_gfx12<0x06b>; 2087defm S_FLOOR_F16 : SOP1_Real_gfx11_gfx12<0x06c>; 2088defm S_TRUNC_F16 : SOP1_Real_gfx11_gfx12<0x06d>; 2089defm S_RNDNE_F16 : SOP1_Real_gfx11_gfx12<0x06e>; 2090 2091//===----------------------------------------------------------------------===// 2092// SOP1 - GFX10. 2093//===----------------------------------------------------------------------===// 2094 2095multiclass SOP1_Real_gfx10<bits<8> op> { 2096 defvar ps = !cast<SOP1_Pseudo>(NAME); 2097 def _gfx10 : SOP1_Real<op, ps>, 2098 Select_gfx10<ps.Mnemonic>; 2099} 2100 2101multiclass SOP1_Real_gfx10_gfx11_gfx12<bits<8> op> : 2102 SOP1_Real_gfx10<op>, SOP1_Real_gfx11_gfx12<op>; 2103 2104defm S_ANDN1_SAVEEXEC_B64 : SOP1_Real_gfx10<0x037>; 2105defm S_ORN1_SAVEEXEC_B64 : SOP1_Real_gfx10<0x038>; 2106defm S_ANDN1_WREXEC_B64 : SOP1_Real_gfx10<0x039>; 2107defm S_ANDN2_WREXEC_B64 : SOP1_Real_gfx10<0x03a>; 2108defm S_BITREPLICATE_B64_B32 : SOP1_Real_gfx10<0x03b>; 2109defm S_AND_SAVEEXEC_B32 : SOP1_Real_gfx10<0x03c>; 2110defm S_OR_SAVEEXEC_B32 : SOP1_Real_gfx10<0x03d>; 2111defm S_XOR_SAVEEXEC_B32 : SOP1_Real_gfx10<0x03e>; 2112defm S_ANDN2_SAVEEXEC_B32 : SOP1_Real_gfx10<0x03f>; 2113defm S_ORN2_SAVEEXEC_B32 : SOP1_Real_gfx10<0x040>; 2114defm S_NAND_SAVEEXEC_B32 : SOP1_Real_gfx10<0x041>; 2115defm S_NOR_SAVEEXEC_B32 : SOP1_Real_gfx10<0x042>; 2116defm S_XNOR_SAVEEXEC_B32 : SOP1_Real_gfx10<0x043>; 2117defm S_ANDN1_SAVEEXEC_B32 : SOP1_Real_gfx10<0x044>; 2118defm S_ORN1_SAVEEXEC_B32 : SOP1_Real_gfx10<0x045>; 2119defm S_ANDN1_WREXEC_B32 : SOP1_Real_gfx10<0x046>; 2120defm S_ANDN2_WREXEC_B32 : SOP1_Real_gfx10<0x047>; 2121defm S_MOVRELSD_2_B32 : SOP1_Real_gfx10<0x049>; 2122 2123//===----------------------------------------------------------------------===// 2124// SOP1 - GFX6, GFX7, GFX10, GFX11. 2125//===----------------------------------------------------------------------===// 2126 2127 2128multiclass SOP1_Real_gfx6_gfx7<bits<8> op> { 2129 defvar ps = !cast<SOP1_Pseudo>(NAME); 2130 def _gfx6_gfx7 : SOP1_Real<op, ps>, 2131 Select_gfx6_gfx7<ps.Mnemonic>; 2132} 2133 2134multiclass SOP1_Real_gfx6_gfx7_gfx10<bits<8> op> : 2135 SOP1_Real_gfx6_gfx7<op>, SOP1_Real_gfx10<op>; 2136 2137multiclass SOP1_Real_gfx6_gfx7_gfx10_gfx11_gfx12<bits<8> op> : 2138 SOP1_Real_gfx6_gfx7<op>, SOP1_Real_gfx10_gfx11_gfx12<op>; 2139 2140defm S_CBRANCH_JOIN : SOP1_Real_gfx6_gfx7<0x032>; 2141 2142defm S_MOV_B32 : SOP1_Real_gfx6_gfx7_gfx10<0x003>; 2143defm S_MOV_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x004>; 2144defm S_CMOV_B32 : SOP1_Real_gfx6_gfx7_gfx10<0x005>; 2145defm S_CMOV_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x006>; 2146defm S_NOT_B32 : SOP1_Real_gfx6_gfx7_gfx10<0x007>; 2147defm S_NOT_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x008>; 2148defm S_WQM_B32 : SOP1_Real_gfx6_gfx7_gfx10<0x009>; 2149defm S_WQM_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x00a>; 2150defm S_BREV_B32 : SOP1_Real_gfx6_gfx7_gfx10<0x00b>; 2151defm S_BREV_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x00c>; 2152defm S_BCNT0_I32_B32 : SOP1_Real_gfx6_gfx7_gfx10<0x00d>; 2153defm S_BCNT0_I32_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x00e>; 2154defm S_BCNT1_I32_B32 : SOP1_Real_gfx6_gfx7_gfx10<0x00f>; 2155defm S_BCNT1_I32_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x010>; 2156defm S_FF0_I32_B32 : SOP1_Real_gfx6_gfx7_gfx10<0x011>; 2157defm S_FF0_I32_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x012>; 2158defm S_FF1_I32_B32 : SOP1_Real_gfx6_gfx7_gfx10<0x013>; 2159defm S_FF1_I32_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x014>; 2160defm S_FLBIT_I32_B32 : SOP1_Real_gfx6_gfx7_gfx10<0x015>; 2161defm S_FLBIT_I32_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x016>; 2162defm S_FLBIT_I32 : SOP1_Real_gfx6_gfx7_gfx10<0x017>; 2163defm S_FLBIT_I32_I64 : SOP1_Real_gfx6_gfx7_gfx10<0x018>; 2164defm S_SEXT_I32_I8 : SOP1_Real_gfx6_gfx7_gfx10<0x019>; 2165defm S_SEXT_I32_I16 : SOP1_Real_gfx6_gfx7_gfx10<0x01a>; 2166defm S_BITSET0_B32 : SOP1_Real_gfx6_gfx7_gfx10<0x01b>; 2167defm S_BITSET0_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x01c>; 2168defm S_BITSET1_B32 : SOP1_Real_gfx6_gfx7_gfx10<0x01d>; 2169defm S_BITSET1_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x01e>; 2170defm S_GETPC_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x01f>; 2171defm S_SETPC_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x020>; 2172defm S_SWAPPC_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x021>; 2173defm S_RFE_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x022>; 2174defm S_AND_SAVEEXEC_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x024>; 2175defm S_OR_SAVEEXEC_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x025>; 2176defm S_XOR_SAVEEXEC_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x026>; 2177defm S_ANDN2_SAVEEXEC_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x027>; 2178defm S_ORN2_SAVEEXEC_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x028>; 2179defm S_NAND_SAVEEXEC_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x029>; 2180defm S_NOR_SAVEEXEC_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x02a>; 2181defm S_XNOR_SAVEEXEC_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx11_gfx12<0x02b>; 2182defm S_QUADMASK_B32 : SOP1_Real_gfx6_gfx7_gfx10<0x02c>; 2183defm S_QUADMASK_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x02d>; 2184defm S_MOVRELS_B32 : SOP1_Real_gfx6_gfx7_gfx10<0x02e>; 2185defm S_MOVRELS_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x02f>; 2186defm S_MOVRELD_B32 : SOP1_Real_gfx6_gfx7_gfx10<0x030>; 2187defm S_MOVRELD_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x031>; 2188defm S_ABS_I32 : SOP1_Real_gfx6_gfx7_gfx10<0x034>; 2189 2190//===----------------------------------------------------------------------===// 2191// SOP2 - GFX12 2192//===----------------------------------------------------------------------===// 2193 2194multiclass SOP2_Real_gfx12<bits<7> op> { 2195 def _gfx12 : SOP2_Real32<op, !cast<SOP2_Pseudo>(NAME)>, 2196 Select_gfx12<!cast<SOP2_Pseudo>(NAME).Mnemonic>; 2197} 2198 2199multiclass SOP2_Real_Renamed_gfx12<bits<7> op, SOP2_Pseudo backing_pseudo, string real_name> { 2200 def _gfx12 : SOP2_Real32<op, backing_pseudo, real_name>, 2201 Select_gfx12<backing_pseudo.Mnemonic>, 2202 MnemonicAlias<backing_pseudo.Mnemonic, real_name>, Requires<[isGFX12Plus]>; 2203} 2204 2205defm S_MIN_NUM_F32 : SOP2_Real_Renamed_gfx12<0x042, S_MIN_F32, "s_min_num_f32">; 2206defm S_MAX_NUM_F32 : SOP2_Real_Renamed_gfx12<0x043, S_MAX_F32, "s_max_num_f32">; 2207defm S_MIN_NUM_F16 : SOP2_Real_Renamed_gfx12<0x04b, S_MIN_F16, "s_min_num_f16">; 2208defm S_MAX_NUM_F16 : SOP2_Real_Renamed_gfx12<0x04c, S_MAX_F16, "s_max_num_f16">; 2209defm S_MINIMUM_F32 : SOP2_Real_gfx12<0x04f>; 2210defm S_MAXIMUM_F32 : SOP2_Real_gfx12<0x050>; 2211defm S_MINIMUM_F16 : SOP2_Real_gfx12<0x051>; 2212defm S_MAXIMUM_F16 : SOP2_Real_gfx12<0x052>; 2213 2214defm S_ADD_CO_U32 : SOP2_Real_Renamed_gfx12<0x000, S_ADD_U32, "s_add_co_u32">; 2215defm S_SUB_CO_U32 : SOP2_Real_Renamed_gfx12<0x001, S_SUB_U32, "s_sub_co_u32">; 2216defm S_ADD_CO_I32 : SOP2_Real_Renamed_gfx12<0x002, S_ADD_I32, "s_add_co_i32">; 2217defm S_SUB_CO_I32 : SOP2_Real_Renamed_gfx12<0x003, S_SUB_I32, "s_sub_co_i32">; 2218defm S_ADD_CO_CI_U32 : SOP2_Real_Renamed_gfx12<0x004, S_ADDC_U32, "s_add_co_ci_u32">; 2219defm S_SUB_CO_CI_U32 : SOP2_Real_Renamed_gfx12<0x005, S_SUBB_U32, "s_sub_co_ci_u32">; 2220 2221//===----------------------------------------------------------------------===// 2222// SOP2 - GFX11, GFX12. 2223//===----------------------------------------------------------------------===// 2224 2225multiclass SOP2_Real_gfx11<bits<7> op> { 2226 def _gfx11 : SOP2_Real32<op, !cast<SOP2_Pseudo>(NAME)>, 2227 Select_gfx11<!cast<SOP2_Pseudo>(NAME).Mnemonic>; 2228} 2229 2230multiclass SOP2_Real_Renamed_gfx11<bits<7> op, SOP2_Pseudo backing_pseudo, string real_name> { 2231 def _gfx11 : SOP2_Real32<op, backing_pseudo, real_name>, 2232 Select_gfx11<backing_pseudo.Mnemonic>, 2233 MnemonicAlias<backing_pseudo.Mnemonic, real_name>, Requires<[isGFX11Only]>; 2234} 2235 2236multiclass SOP2_Real_gfx11_gfx12<bits<7> op> : 2237 SOP2_Real_gfx11<op>, SOP2_Real_gfx12<op>; 2238 2239multiclass SOP2_Real_Renamed_gfx11_gfx12<bits<8> op, SOP2_Pseudo backing_pseudo, string real_name> : 2240 SOP2_Real_Renamed_gfx11<op, backing_pseudo, real_name>, 2241 SOP2_Real_Renamed_gfx12<op, backing_pseudo, real_name>; 2242 2243defm S_ABSDIFF_I32 : SOP2_Real_gfx11_gfx12<0x006>; 2244defm S_LSHL_B32 : SOP2_Real_gfx11_gfx12<0x008>; 2245defm S_LSHL_B64 : SOP2_Real_gfx11_gfx12<0x009>; 2246defm S_LSHR_B32 : SOP2_Real_gfx11_gfx12<0x00a>; 2247defm S_LSHR_B64 : SOP2_Real_gfx11_gfx12<0x00b>; 2248defm S_ASHR_I32 : SOP2_Real_gfx11_gfx12<0x00c>; 2249defm S_ASHR_I64 : SOP2_Real_gfx11_gfx12<0x00d>; 2250defm S_LSHL1_ADD_U32 : SOP2_Real_gfx11_gfx12<0x00e>; 2251defm S_LSHL2_ADD_U32 : SOP2_Real_gfx11_gfx12<0x00f>; 2252defm S_LSHL3_ADD_U32 : SOP2_Real_gfx11_gfx12<0x010>; 2253defm S_LSHL4_ADD_U32 : SOP2_Real_gfx11_gfx12<0x011>; 2254defm S_MIN_I32 : SOP2_Real_gfx11_gfx12<0x012>; 2255defm S_MIN_U32 : SOP2_Real_gfx11_gfx12<0x013>; 2256defm S_MAX_I32 : SOP2_Real_gfx11_gfx12<0x014>; 2257defm S_MAX_U32 : SOP2_Real_gfx11_gfx12<0x015>; 2258defm S_AND_B32 : SOP2_Real_gfx11_gfx12<0x016>; 2259defm S_AND_B64 : SOP2_Real_gfx11_gfx12<0x017>; 2260defm S_OR_B32 : SOP2_Real_gfx11_gfx12<0x018>; 2261defm S_OR_B64 : SOP2_Real_gfx11_gfx12<0x019>; 2262defm S_XOR_B32 : SOP2_Real_gfx11_gfx12<0x01a>; 2263defm S_XOR_B64 : SOP2_Real_gfx11_gfx12<0x01b>; 2264defm S_NAND_B32 : SOP2_Real_gfx11_gfx12<0x01c>; 2265defm S_NAND_B64 : SOP2_Real_gfx11_gfx12<0x01d>; 2266defm S_NOR_B32 : SOP2_Real_gfx11_gfx12<0x01e>; 2267defm S_NOR_B64 : SOP2_Real_gfx11_gfx12<0x01f>; 2268defm S_XNOR_B32 : SOP2_Real_gfx11_gfx12<0x020>; 2269defm S_XNOR_B64 : SOP2_Real_gfx11_gfx12<0x021>; 2270defm S_AND_NOT1_B32 : SOP2_Real_Renamed_gfx11_gfx12<0x022, S_ANDN2_B32, "s_and_not1_b32">; 2271defm S_AND_NOT1_B64 : SOP2_Real_Renamed_gfx11_gfx12<0x023, S_ANDN2_B64, "s_and_not1_b64">; 2272defm S_OR_NOT1_B32 : SOP2_Real_Renamed_gfx11_gfx12<0x024, S_ORN2_B32, "s_or_not1_b32">; 2273defm S_OR_NOT1_B64 : SOP2_Real_Renamed_gfx11_gfx12<0x025, S_ORN2_B64, "s_or_not1_b64">; 2274defm S_BFE_U32 : SOP2_Real_gfx11_gfx12<0x026>; 2275defm S_BFE_I32 : SOP2_Real_gfx11_gfx12<0x027>; 2276defm S_BFE_U64 : SOP2_Real_gfx11_gfx12<0x028>; 2277defm S_BFE_I64 : SOP2_Real_gfx11_gfx12<0x029>; 2278defm S_BFM_B32 : SOP2_Real_gfx11_gfx12<0x02a>; 2279defm S_BFM_B64 : SOP2_Real_gfx11_gfx12<0x02b>; 2280defm S_MUL_I32 : SOP2_Real_gfx11_gfx12<0x02c>; 2281defm S_MUL_HI_U32 : SOP2_Real_gfx11_gfx12<0x02d>; 2282defm S_MUL_HI_I32 : SOP2_Real_gfx11_gfx12<0x02e>; 2283defm S_CSELECT_B32 : SOP2_Real_gfx11_gfx12<0x030>; 2284defm S_CSELECT_B64 : SOP2_Real_gfx11_gfx12<0x031>; 2285defm S_PACK_HL_B32_B16 : SOP2_Real_gfx11_gfx12<0x035>; 2286defm S_ADD_NC_U64 : SOP2_Real_Renamed_gfx12<0x053, S_ADD_U64, "s_add_nc_u64">; 2287defm S_SUB_NC_U64 : SOP2_Real_Renamed_gfx12<0x054, S_SUB_U64, "s_sub_nc_u64">; 2288defm S_MUL_U64 : SOP2_Real_gfx12<0x055>; 2289 2290//===----------------------------------------------------------------------===// 2291// SOP2 - GFX1150, GFX12 2292//===----------------------------------------------------------------------===// 2293 2294multiclass SOP2_Real_FMAK_gfx12<bits<7> op> { 2295 def _gfx12 : SOP2_Real64<op, !cast<SOP2_Pseudo>(NAME)>, 2296 Select_gfx12<!cast<SOP2_Pseudo>(NAME).Mnemonic>; 2297} 2298 2299multiclass SOP2_Real_FMAK_gfx11<bits<7> op> { 2300 def _gfx11 : SOP2_Real64<op, !cast<SOP2_Pseudo>(NAME)>, 2301 Select_gfx11<!cast<SOP2_Pseudo>(NAME).Mnemonic>; 2302} 2303 2304multiclass SOP2_Real_FMAK_gfx11_gfx12<bits<7> op> : 2305 SOP2_Real_FMAK_gfx11<op>, SOP2_Real_FMAK_gfx12<op>; 2306 2307defm S_ADD_F32 : SOP2_Real_gfx11_gfx12<0x040>; 2308defm S_SUB_F32 : SOP2_Real_gfx11_gfx12<0x041>; 2309defm S_MUL_F32 : SOP2_Real_gfx11_gfx12<0x044>; 2310defm S_FMAAK_F32 : SOP2_Real_FMAK_gfx11_gfx12<0x045>; 2311defm S_FMAMK_F32 : SOP2_Real_FMAK_gfx11_gfx12<0x046>; 2312defm S_FMAC_F32 : SOP2_Real_gfx11_gfx12<0x047>; 2313defm S_CVT_PK_RTZ_F16_F32 : SOP2_Real_gfx11_gfx12<0x048>; 2314defm S_ADD_F16 : SOP2_Real_gfx11_gfx12<0x049>; 2315defm S_SUB_F16 : SOP2_Real_gfx11_gfx12<0x04a>; 2316defm S_MUL_F16 : SOP2_Real_gfx11_gfx12<0x04d>; 2317defm S_FMAC_F16 : SOP2_Real_gfx11_gfx12<0x04e>; 2318 2319//===----------------------------------------------------------------------===// 2320// SOP2 - GFX1150 2321//===----------------------------------------------------------------------===// 2322 2323defm S_MIN_F32 : SOP2_Real_gfx11<0x042>; 2324defm S_MAX_F32 : SOP2_Real_gfx11<0x043>; 2325defm S_MIN_F16 : SOP2_Real_gfx11<0x04b>; 2326defm S_MAX_F16 : SOP2_Real_gfx11<0x04c>; 2327 2328//===----------------------------------------------------------------------===// 2329// SOP2 - GFX10. 2330//===----------------------------------------------------------------------===// 2331 2332multiclass SOP2_Real_gfx10<bits<7> op> { 2333 defvar ps = !cast<SOP2_Pseudo>(NAME); 2334 def _gfx10 : SOP2_Real32<op, ps>, 2335 Select_gfx10<ps.Mnemonic>; 2336} 2337 2338multiclass SOP2_Real_gfx10_gfx11_gfx12<bits<7> op> : 2339 SOP2_Real_gfx10<op>, SOP2_Real_gfx11_gfx12<op>; 2340 2341defm S_LSHL1_ADD_U32 : SOP2_Real_gfx10<0x02e>; 2342defm S_LSHL2_ADD_U32 : SOP2_Real_gfx10<0x02f>; 2343defm S_LSHL3_ADD_U32 : SOP2_Real_gfx10<0x030>; 2344defm S_LSHL4_ADD_U32 : SOP2_Real_gfx10<0x031>; 2345defm S_PACK_LL_B32_B16 : SOP2_Real_gfx10_gfx11_gfx12<0x032>; 2346defm S_PACK_LH_B32_B16 : SOP2_Real_gfx10_gfx11_gfx12<0x033>; 2347defm S_PACK_HH_B32_B16 : SOP2_Real_gfx10_gfx11_gfx12<0x034>; 2348defm S_MUL_HI_U32 : SOP2_Real_gfx10<0x035>; 2349defm S_MUL_HI_I32 : SOP2_Real_gfx10<0x036>; 2350 2351//===----------------------------------------------------------------------===// 2352// SOP2 - GFX6, GFX7. 2353//===----------------------------------------------------------------------===// 2354 2355multiclass SOP2_Real_gfx6_gfx7<bits<7> op> { 2356 defvar ps = !cast<SOP_Pseudo>(NAME); 2357 def _gfx6_gfx7 : SOP2_Real32<op, ps>, 2358 Select_gfx6_gfx7<ps.Mnemonic>; 2359} 2360 2361multiclass SOP2_Real_gfx6_gfx7_gfx10<bits<7> op> : 2362 SOP2_Real_gfx6_gfx7<op>, SOP2_Real_gfx10<op>; 2363 2364multiclass SOP2_Real_gfx6_gfx7_gfx10_gfx11<bits<7> op> : 2365 SOP2_Real_gfx6_gfx7<op>, SOP2_Real_gfx10<op>, SOP2_Real_gfx11<op>; 2366 2367multiclass SOP2_Real_gfx6_gfx7_gfx10_gfx11_gfx12<bits<7> op> : 2368 SOP2_Real_gfx6_gfx7<op>, SOP2_Real_gfx10_gfx11_gfx12<op>; 2369 2370defm S_CBRANCH_G_FORK : SOP2_Real_gfx6_gfx7<0x02b>; 2371 2372defm S_ADD_U32 : SOP2_Real_gfx6_gfx7_gfx10_gfx11<0x000>; 2373defm S_SUB_U32 : SOP2_Real_gfx6_gfx7_gfx10_gfx11<0x001>; 2374defm S_ADD_I32 : SOP2_Real_gfx6_gfx7_gfx10_gfx11<0x002>; 2375defm S_SUB_I32 : SOP2_Real_gfx6_gfx7_gfx10_gfx11<0x003>; 2376defm S_ADDC_U32 : SOP2_Real_gfx6_gfx7_gfx10_gfx11<0x004>; 2377defm S_SUBB_U32 : SOP2_Real_gfx6_gfx7_gfx10_gfx11<0x005>; 2378defm S_MIN_I32 : SOP2_Real_gfx6_gfx7_gfx10<0x006>; 2379defm S_MIN_U32 : SOP2_Real_gfx6_gfx7_gfx10<0x007>; 2380defm S_MAX_I32 : SOP2_Real_gfx6_gfx7_gfx10<0x008>; 2381defm S_MAX_U32 : SOP2_Real_gfx6_gfx7_gfx10<0x009>; 2382defm S_CSELECT_B32 : SOP2_Real_gfx6_gfx7_gfx10<0x00a>; 2383defm S_CSELECT_B64 : SOP2_Real_gfx6_gfx7_gfx10<0x00b>; 2384defm S_AND_B32 : SOP2_Real_gfx6_gfx7_gfx10<0x00e>; 2385defm S_AND_B64 : SOP2_Real_gfx6_gfx7_gfx10<0x00f>; 2386defm S_OR_B32 : SOP2_Real_gfx6_gfx7_gfx10<0x010>; 2387defm S_OR_B64 : SOP2_Real_gfx6_gfx7_gfx10<0x011>; 2388defm S_XOR_B32 : SOP2_Real_gfx6_gfx7_gfx10<0x012>; 2389defm S_XOR_B64 : SOP2_Real_gfx6_gfx7_gfx10<0x013>; 2390defm S_ANDN2_B32 : SOP2_Real_gfx6_gfx7_gfx10<0x014>; 2391defm S_ANDN2_B64 : SOP2_Real_gfx6_gfx7_gfx10<0x015>; 2392defm S_ORN2_B32 : SOP2_Real_gfx6_gfx7_gfx10<0x016>; 2393defm S_ORN2_B64 : SOP2_Real_gfx6_gfx7_gfx10<0x017>; 2394defm S_NAND_B32 : SOP2_Real_gfx6_gfx7_gfx10<0x018>; 2395defm S_NAND_B64 : SOP2_Real_gfx6_gfx7_gfx10<0x019>; 2396defm S_NOR_B32 : SOP2_Real_gfx6_gfx7_gfx10<0x01a>; 2397defm S_NOR_B64 : SOP2_Real_gfx6_gfx7_gfx10<0x01b>; 2398defm S_XNOR_B32 : SOP2_Real_gfx6_gfx7_gfx10<0x01c>; 2399defm S_XNOR_B64 : SOP2_Real_gfx6_gfx7_gfx10<0x01d>; 2400defm S_LSHL_B32 : SOP2_Real_gfx6_gfx7_gfx10<0x01e>; 2401defm S_LSHL_B64 : SOP2_Real_gfx6_gfx7_gfx10<0x01f>; 2402defm S_LSHR_B32 : SOP2_Real_gfx6_gfx7_gfx10<0x020>; 2403defm S_LSHR_B64 : SOP2_Real_gfx6_gfx7_gfx10<0x021>; 2404defm S_ASHR_I32 : SOP2_Real_gfx6_gfx7_gfx10<0x022>; 2405defm S_ASHR_I64 : SOP2_Real_gfx6_gfx7_gfx10<0x023>; 2406defm S_BFM_B32 : SOP2_Real_gfx6_gfx7_gfx10<0x024>; 2407defm S_BFM_B64 : SOP2_Real_gfx6_gfx7_gfx10<0x025>; 2408defm S_MUL_I32 : SOP2_Real_gfx6_gfx7_gfx10<0x026>; 2409defm S_BFE_U32 : SOP2_Real_gfx6_gfx7_gfx10<0x027>; 2410defm S_BFE_I32 : SOP2_Real_gfx6_gfx7_gfx10<0x028>; 2411defm S_BFE_U64 : SOP2_Real_gfx6_gfx7_gfx10<0x029>; 2412defm S_BFE_I64 : SOP2_Real_gfx6_gfx7_gfx10<0x02a>; 2413defm S_ABSDIFF_I32 : SOP2_Real_gfx6_gfx7_gfx10<0x02c>; 2414 2415//===----------------------------------------------------------------------===// 2416// SOPK - GFX11, GFX12. 2417//===----------------------------------------------------------------------===// 2418 2419multiclass SOPK_Real32_gfx12<bits<5> op> { 2420 def _gfx12 : SOPK_Real32<op, !cast<SOPK_Pseudo>(NAME)>, 2421 Select_gfx12<!cast<SOPK_Pseudo>(NAME).Mnemonic>; 2422} 2423 2424multiclass SOPK_Real32_Renamed_gfx12<bits<5> op, SOPK_Pseudo backing_pseudo, string real_name> { 2425 def _gfx12 : SOPK_Real32<op, backing_pseudo, real_name>, 2426 Select_gfx12<backing_pseudo.Mnemonic>, 2427 MnemonicAlias<backing_pseudo.Mnemonic, real_name>, Requires<[isGFX12Plus]>; 2428} 2429 2430multiclass SOPK_Real32_gfx11<bits<5> op> { 2431 def _gfx11 : SOPK_Real32<op, !cast<SOPK_Pseudo>(NAME)>, 2432 Select_gfx11<!cast<SOPK_Pseudo>(NAME).Mnemonic>; 2433} 2434 2435multiclass SOPK_Real64_gfx12<bits<5> op> { 2436 def _gfx12 : SOPK_Real64<op, !cast<SOPK_Pseudo>(NAME)>, 2437 Select_gfx12<!cast<SOPK_Pseudo>(NAME).Mnemonic>; 2438} 2439 2440multiclass SOPK_Real64_gfx11<bits<5> op> { 2441 def _gfx11 : SOPK_Real64<op, !cast<SOPK_Pseudo>(NAME)>, 2442 Select_gfx11<!cast<SOPK_Pseudo>(NAME).Mnemonic>; 2443} 2444 2445multiclass SOPK_Real32_gfx11_gfx12<bits<5> op> : 2446 SOPK_Real32_gfx11<op>, SOPK_Real32_gfx12<op>; 2447 2448multiclass SOPK_Real64_gfx11_gfx12<bits<5> op> : 2449 SOPK_Real64_gfx11<op>, SOPK_Real64_gfx12<op>; 2450 2451defm S_ADDK_CO_I32 : SOPK_Real32_Renamed_gfx12<0x00f, S_ADDK_I32, "s_addk_co_i32">; 2452defm S_GETREG_B32 : SOPK_Real32_gfx11_gfx12<0x011>; 2453defm S_SETREG_B32 : SOPK_Real32_gfx11_gfx12<0x012>; 2454defm S_SETREG_IMM32_B32 : SOPK_Real64_gfx11_gfx12<0x013>; 2455defm S_CALL_B64 : SOPK_Real32_gfx11_gfx12<0x014>; 2456defm S_SUBVECTOR_LOOP_BEGIN : SOPK_Real32_gfx11<0x016>; 2457defm S_SUBVECTOR_LOOP_END : SOPK_Real32_gfx11<0x017>; 2458defm S_WAITCNT_VSCNT : SOPK_Real32_gfx11<0x018>; 2459defm S_WAITCNT_VMCNT : SOPK_Real32_gfx11<0x019>; 2460defm S_WAITCNT_EXPCNT : SOPK_Real32_gfx11<0x01a>; 2461defm S_WAITCNT_LGKMCNT : SOPK_Real32_gfx11<0x01b>; 2462 2463//===----------------------------------------------------------------------===// 2464// SOPK - GFX10. 2465//===----------------------------------------------------------------------===// 2466 2467multiclass SOPK_Real32_gfx10<bits<5> op> { 2468 defvar ps = !cast<SOPK_Pseudo>(NAME); 2469 def _gfx10 : SOPK_Real32<op, ps>, 2470 Select_gfx10<ps.Mnemonic>; 2471} 2472 2473multiclass SOPK_Real64_gfx10<bits<5> op> { 2474 defvar ps = !cast<SOPK_Pseudo>(NAME); 2475 def _gfx10 : SOPK_Real64<op, ps>, 2476 Select_gfx10<ps.Mnemonic>; 2477} 2478 2479multiclass SOPK_Real32_gfx10_gfx11<bits<5> op> : 2480 SOPK_Real32_gfx10<op>, SOPK_Real32_gfx11<op>; 2481 2482multiclass SOPK_Real32_gfx10_gfx11_gfx12<bits<5> op> : 2483 SOPK_Real32_gfx10<op>, SOPK_Real32_gfx11_gfx12<op>; 2484 2485defm S_VERSION : SOPK_Real32_gfx10_gfx11_gfx12<0x001>; 2486defm S_CALL_B64 : SOPK_Real32_gfx10<0x016>; 2487defm S_WAITCNT_VSCNT : SOPK_Real32_gfx10<0x017>; 2488defm S_WAITCNT_VMCNT : SOPK_Real32_gfx10<0x018>; 2489defm S_WAITCNT_EXPCNT : SOPK_Real32_gfx10<0x019>; 2490defm S_WAITCNT_LGKMCNT : SOPK_Real32_gfx10<0x01a>; 2491defm S_SUBVECTOR_LOOP_BEGIN : SOPK_Real32_gfx10<0x01b>; 2492defm S_SUBVECTOR_LOOP_END : SOPK_Real32_gfx10<0x01c>; 2493 2494//===----------------------------------------------------------------------===// 2495// SOPK - GFX6, GFX7. 2496//===----------------------------------------------------------------------===// 2497 2498multiclass SOPK_Real32_gfx6_gfx7<bits<5> op> { 2499 defvar ps = !cast<SOPK_Pseudo>(NAME); 2500 def _gfx6_gfx7 : SOPK_Real32<op, ps>, 2501 Select_gfx6_gfx7<ps.Mnemonic>; 2502} 2503 2504multiclass SOPK_Real64_gfx6_gfx7<bits<5> op> { 2505 defvar ps = !cast<SOPK_Pseudo>(NAME); 2506 def _gfx6_gfx7 : SOPK_Real64<op, ps>, 2507 Select_gfx6_gfx7<ps.Mnemonic>; 2508} 2509 2510multiclass SOPK_Real32_gfx6_gfx7_gfx10<bits<5> op> : 2511 SOPK_Real32_gfx6_gfx7<op>, SOPK_Real32_gfx10<op>; 2512 2513multiclass SOPK_Real64_gfx6_gfx7_gfx10<bits<5> op> : 2514 SOPK_Real64_gfx6_gfx7<op>, SOPK_Real64_gfx10<op>; 2515 2516multiclass SOPK_Real32_gfx6_gfx7_gfx10_gfx11<bits<5> op> : 2517 SOPK_Real32_gfx6_gfx7<op>, SOPK_Real32_gfx10_gfx11<op>; 2518 2519multiclass SOPK_Real32_gfx6_gfx7_gfx10_gfx11_gfx12<bits<5> op> : 2520 SOPK_Real32_gfx6_gfx7<op>, SOPK_Real32_gfx10_gfx11_gfx12<op>; 2521 2522defm S_CBRANCH_I_FORK : SOPK_Real32_gfx6_gfx7<0x011>; 2523 2524defm S_MOVK_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11_gfx12<0x000>; 2525defm S_CMOVK_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11_gfx12<0x002>; 2526defm S_CMPK_EQ_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x003>; 2527defm S_CMPK_LG_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x004>; 2528defm S_CMPK_GT_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x005>; 2529defm S_CMPK_GE_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x006>; 2530defm S_CMPK_LT_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x007>; 2531defm S_CMPK_LE_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x008>; 2532defm S_CMPK_EQ_U32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x009>; 2533defm S_CMPK_LG_U32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x00a>; 2534defm S_CMPK_GT_U32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x00b>; 2535defm S_CMPK_GE_U32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x00c>; 2536defm S_CMPK_LT_U32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x00d>; 2537defm S_CMPK_LE_U32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x00e>; 2538defm S_ADDK_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x00f>; 2539defm S_MULK_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11_gfx12<0x010>; 2540defm S_GETREG_B32 : SOPK_Real32_gfx6_gfx7_gfx10<0x012>; 2541defm S_SETREG_B32 : SOPK_Real32_gfx6_gfx7_gfx10<0x013>; 2542defm S_SETREG_IMM32_B32 : SOPK_Real64_gfx6_gfx7_gfx10<0x015>; 2543 2544//===----------------------------------------------------------------------===// 2545// SOPP - GFX12 only. 2546//===----------------------------------------------------------------------===// 2547 2548multiclass SOPP_Real_32_gfx12<bits<7> op> { 2549 def _gfx12 : SOPP_Real_32<op, !cast<SOPP_Pseudo>(NAME), !cast<SOPP_Pseudo>(NAME).Mnemonic>, 2550 Select_gfx12<!cast<SOPP_Pseudo>(NAME).Mnemonic>, 2551 SOPPRelaxTable<0, !cast<SOPP_Pseudo>(NAME).KeyName, "_gfx12">; 2552} 2553 2554multiclass SOPP_Real_32_Renamed_gfx12<bits<7> op, SOPP_Pseudo backing_pseudo, string real_name> { 2555 def _gfx12 : SOPP_Real_32<op, backing_pseudo, real_name>, 2556 Select_gfx12<backing_pseudo.Mnemonic>, 2557 MnemonicAlias<backing_pseudo.Mnemonic, real_name>, Requires<[isGFX12Plus]>; 2558} 2559 2560defm S_WAIT_ALU : SOPP_Real_32_Renamed_gfx12<0x008, S_WAITCNT_DEPCTR, "s_wait_alu">; 2561defm S_BARRIER_WAIT : SOPP_Real_32_gfx12<0x014>; 2562defm S_BARRIER_LEAVE : SOPP_Real_32_gfx12<0x015>; 2563defm S_WAIT_LOADCNT : SOPP_Real_32_gfx12<0x040>; 2564defm S_WAIT_STORECNT : SOPP_Real_32_gfx12<0x041>; 2565defm S_WAIT_SAMPLECNT : SOPP_Real_32_gfx12<0x042>; 2566defm S_WAIT_BVHCNT : SOPP_Real_32_gfx12<0x043>; 2567defm S_WAIT_EXPCNT : SOPP_Real_32_gfx12<0x044>; 2568defm S_WAIT_DSCNT : SOPP_Real_32_gfx12<0x046>; 2569defm S_WAIT_KMCNT : SOPP_Real_32_gfx12<0x047>; 2570defm S_WAIT_LOADCNT_DSCNT : SOPP_Real_32_gfx12<0x048>; 2571defm S_WAIT_STORECNT_DSCNT : SOPP_Real_32_gfx12<0x049>; 2572 2573//===----------------------------------------------------------------------===// 2574// SOPP - GFX11, GFX12. 2575//===----------------------------------------------------------------------===// 2576 2577 2578multiclass SOPP_Real_32_gfx11<bits<7> op> { 2579 def _gfx11 : SOPP_Real_32<op, !cast<SOPP_Pseudo>(NAME), !cast<SOPP_Pseudo>(NAME).Mnemonic>, 2580 Select_gfx11<!cast<SOPP_Pseudo>(NAME).Mnemonic>, 2581 SOPPRelaxTable<0, !cast<SOPP_Pseudo>(NAME).KeyName, "_gfx11">; 2582} 2583 2584multiclass SOPP_Real_64_gfx12<bits<7> op> { 2585 def _gfx12 : SOPP_Real_64<op, !cast<SOPP_Pseudo>(NAME), !cast<SOPP_Pseudo>(NAME).Mnemonic>, 2586 Select_gfx12<!cast<SOPP_Pseudo>(NAME).Mnemonic>, 2587 SOPPRelaxTable<1, !cast<SOPP_Pseudo>(NAME).KeyName, "_gfx12">; 2588} 2589 2590multiclass SOPP_Real_64_gfx11<bits<7> op> { 2591 def _gfx11 : SOPP_Real_64<op, !cast<SOPP_Pseudo>(NAME), !cast<SOPP_Pseudo>(NAME).Mnemonic>, 2592 Select_gfx11<!cast<SOPP_Pseudo>(NAME).Mnemonic>, 2593 SOPPRelaxTable<1, !cast<SOPP_Pseudo>(NAME).KeyName, "_gfx11">; 2594} 2595 2596multiclass SOPP_Real_32_Renamed_gfx11<bits<7> op, SOPP_Pseudo backing_pseudo, string real_name> { 2597 def _gfx11 : SOPP_Real_32<op, backing_pseudo, real_name>, 2598 Select_gfx11<backing_pseudo.Mnemonic>, 2599 MnemonicAlias<backing_pseudo.Mnemonic, real_name>, Requires<[isGFX11Only]>; 2600} 2601 2602multiclass SOPP_Real_32_gfx11_gfx12<bits<7> op> : 2603 SOPP_Real_32_gfx11<op>, SOPP_Real_32_gfx12<op>; 2604 2605multiclass SOPP_Real_64_gfx11_gfx12<bits<7> op> : 2606 SOPP_Real_64_gfx11<op>, SOPP_Real_64_gfx12<op>; 2607 2608multiclass SOPP_Real_32_Renamed_gfx11_gfx12<bits<7> op, SOPP_Pseudo backing_pseudo, string real_name> : 2609 SOPP_Real_32_Renamed_gfx11<op, backing_pseudo, real_name>, 2610 SOPP_Real_32_Renamed_gfx12<op, backing_pseudo, real_name>; 2611 2612multiclass SOPP_Real_With_Relaxation_gfx12<bits<7> op> { 2613 defm "" : SOPP_Real_32_gfx12<op>; 2614 defm _pad_s_nop : SOPP_Real_64_gfx12<op>; 2615} 2616 2617multiclass SOPP_Real_With_Relaxation_gfx11<bits<7> op> { 2618 defm "" : SOPP_Real_32_gfx11<op>; 2619 defm _pad_s_nop : SOPP_Real_64_gfx11<op>; 2620} 2621 2622multiclass SOPP_Real_With_Relaxation_gfx11_gfx12<bits<7>op> : 2623 SOPP_Real_With_Relaxation_gfx11<op>, SOPP_Real_With_Relaxation_gfx12<op>; 2624 2625defm S_SETKILL : SOPP_Real_32_gfx11_gfx12<0x001>; 2626defm S_SETHALT : SOPP_Real_32_gfx11_gfx12<0x002>; 2627defm S_SLEEP : SOPP_Real_32_gfx11_gfx12<0x003>; 2628defm S_SET_INST_PREFETCH_DISTANCE : SOPP_Real_32_Renamed_gfx11<0x004, S_INST_PREFETCH, "s_set_inst_prefetch_distance">; 2629defm S_CLAUSE : SOPP_Real_32_gfx11_gfx12<0x005>; 2630defm S_DELAY_ALU : SOPP_Real_32_gfx11_gfx12<0x007>; 2631defm S_WAITCNT_DEPCTR : SOPP_Real_32_gfx11<0x008>; 2632defm S_WAITCNT : SOPP_Real_32_gfx11_gfx12<0x009>; 2633defm S_WAIT_IDLE : SOPP_Real_32_gfx11_gfx12<0x00a>; 2634defm S_WAIT_EVENT : SOPP_Real_32_gfx11_gfx12<0x00b>; 2635defm S_TRAP : SOPP_Real_32_gfx11_gfx12<0x010>; 2636defm S_ROUND_MODE : SOPP_Real_32_gfx11_gfx12<0x011>; 2637defm S_DENORM_MODE : SOPP_Real_32_gfx11_gfx12<0x012>; 2638defm S_BRANCH : SOPP_Real_With_Relaxation_gfx11_gfx12<0x020>; 2639defm S_CBRANCH_SCC0 : SOPP_Real_With_Relaxation_gfx11_gfx12<0x021>; 2640defm S_CBRANCH_SCC1 : SOPP_Real_With_Relaxation_gfx11_gfx12<0x022>; 2641defm S_CBRANCH_VCCZ : SOPP_Real_With_Relaxation_gfx11_gfx12<0x023>; 2642defm S_CBRANCH_VCCNZ : SOPP_Real_With_Relaxation_gfx11_gfx12<0x024>; 2643defm S_CBRANCH_EXECZ : SOPP_Real_With_Relaxation_gfx11_gfx12<0x025>; 2644defm S_CBRANCH_EXECNZ : SOPP_Real_With_Relaxation_gfx11_gfx12<0x026>; 2645defm S_CBRANCH_CDBGSYS : SOPP_Real_With_Relaxation_gfx11<0x027>; 2646defm S_CBRANCH_CDBGUSER : SOPP_Real_With_Relaxation_gfx11<0x028>; 2647defm S_CBRANCH_CDBGSYS_OR_USER : SOPP_Real_With_Relaxation_gfx11<0x029>; 2648defm S_CBRANCH_CDBGSYS_AND_USER : SOPP_Real_With_Relaxation_gfx11<0x02a>; 2649defm S_ENDPGM : SOPP_Real_32_gfx11_gfx12<0x030>; 2650defm S_ENDPGM_SAVED : SOPP_Real_32_gfx11_gfx12<0x031>; 2651defm S_WAKEUP : SOPP_Real_32_gfx11_gfx12<0x034>; 2652defm S_SETPRIO : SOPP_Real_32_gfx11_gfx12<0x035>; 2653defm S_SENDMSG : SOPP_Real_32_gfx11_gfx12<0x036>; 2654defm S_SENDMSGHALT : SOPP_Real_32_gfx11_gfx12<0x037>; 2655defm S_INCPERFLEVEL : SOPP_Real_32_gfx11_gfx12<0x038>; 2656defm S_DECPERFLEVEL : SOPP_Real_32_gfx11_gfx12<0x039>; 2657defm S_TTRACEDATA : SOPP_Real_32_gfx11_gfx12<0x03a>; 2658defm S_TTRACEDATA_IMM : SOPP_Real_32_gfx11_gfx12<0x03b>; 2659defm S_ICACHE_INV : SOPP_Real_32_gfx11_gfx12<0x03c>; 2660defm S_BARRIER : SOPP_Real_32_gfx11_gfx12<0x03d>; 2661 2662//===----------------------------------------------------------------------===// 2663// SOPP - GFX1150, GFX12. 2664//===----------------------------------------------------------------------===// 2665 2666defm S_SINGLEUSE_VDST : SOPP_Real_32_gfx11_gfx12<0x013>; 2667 2668//===----------------------------------------------------------------------===// 2669// SOPP - GFX6, GFX7, GFX8, GFX9, GFX10 2670//===----------------------------------------------------------------------===// 2671 2672multiclass SOPP_Real_32_gfx6_gfx7<bits<7> op> { 2673 defvar ps = !cast<SOPP_Pseudo>(NAME); 2674 def _gfx6_gfx7 : SOPP_Real_32<op, ps, !cast<SOPP_Pseudo>(NAME).Mnemonic>, 2675 Select_gfx6_gfx7<ps.Mnemonic>, 2676 SOPPRelaxTable<0, ps.KeyName, "_gfx6_gfx7">; 2677} 2678 2679multiclass SOPP_Real_32_gfx8_gfx9<bits<7> op> { 2680 defvar ps = !cast<SOPP_Pseudo>(NAME); 2681 def _vi : SOPP_Real_32<op, ps>, 2682 Select_vi<ps.Mnemonic>, 2683 SOPPRelaxTable<0, ps.KeyName, "_vi">; 2684} 2685 2686multiclass SOPP_Real_32_gfx10<bits<7> op> { 2687 defvar ps = !cast<SOPP_Pseudo>(NAME); 2688 def _gfx10 : SOPP_Real_32<op, ps>, 2689 Select_gfx10<ps.Mnemonic>, 2690 SOPPRelaxTable<0, ps.KeyName, "_gfx10">; 2691} 2692 2693multiclass SOPP_Real_32_gfx8_gfx9_gfx10<bits<7> op> : 2694 SOPP_Real_32_gfx8_gfx9<op>, SOPP_Real_32_gfx10<op>; 2695 2696multiclass SOPP_Real_32_gfx6_gfx7_gfx8_gfx9<bits<7> op> : 2697 SOPP_Real_32_gfx6_gfx7<op>, SOPP_Real_32_gfx8_gfx9<op>; 2698 2699multiclass SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<bits<7> op> : 2700 SOPP_Real_32_gfx6_gfx7_gfx8_gfx9<op>, SOPP_Real_32_gfx10<op>; 2701 2702multiclass SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12<bits<7> op> : 2703 SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<op>, SOPP_Real_32_gfx11_gfx12<op>; 2704 2705multiclass SOPP_Real_32_gfx10_gfx11_gfx12<bits<7> op> : 2706 SOPP_Real_32_gfx10<op>, SOPP_Real_32_gfx11_gfx12<op>; 2707 2708//64 bit encodings, for Relaxation 2709multiclass SOPP_Real_64_gfx6_gfx7<bits<7> op> { 2710 defvar ps = !cast<SOPP_Pseudo>(NAME); 2711 def _gfx6_gfx7 : SOPP_Real_64<op, ps>, 2712 Select_gfx6_gfx7<ps.Mnemonic>, 2713 SOPPRelaxTable<1, ps.KeyName, "_gfx6_gfx7">; 2714} 2715 2716multiclass SOPP_Real_64_gfx8_gfx9<bits<7> op> { 2717 defvar ps = !cast<SOPP_Pseudo>(NAME); 2718 def _vi : SOPP_Real_64<op, ps>, 2719 Select_vi<ps.Mnemonic>, 2720 SOPPRelaxTable<1, ps.KeyName, "_vi">; 2721} 2722 2723multiclass SOPP_Real_64_gfx10<bits<7> op> { 2724 defvar ps = !cast<SOPP_Pseudo>(NAME); 2725 def _gfx10 : SOPP_Real_64<op, ps>, 2726 Select_gfx10<ps.Mnemonic>, 2727 SOPPRelaxTable<1, ps.KeyName, "_gfx10">; 2728} 2729 2730multiclass SOPP_Real_64_gfx8_gfx9_gfx10<bits<7> op> : 2731 SOPP_Real_64_gfx8_gfx9<op>, SOPP_Real_64_gfx10<op>; 2732 2733multiclass SOPP_Real_64_gfx6_gfx7_gfx8_gfx9<bits<7> op> : 2734 SOPP_Real_64_gfx6_gfx7<op>, SOPP_Real_64_gfx8_gfx9<op>; 2735 2736multiclass SOPP_Real_64_gfx6_gfx7_gfx8_gfx9_gfx10<bits<7> op> : 2737 SOPP_Real_64_gfx6_gfx7_gfx8_gfx9<op>, SOPP_Real_64_gfx10<op>; 2738 2739multiclass SOPP_Real_64_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12<bits<7> op> : 2740 SOPP_Real_64_gfx6_gfx7_gfx8_gfx9_gfx10<op>, SOPP_Real_64_gfx11_gfx12<op>; 2741 2742//relaxation for insts with no operands not implemented 2743multiclass SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10<bits<7> op> { 2744 defm "" : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<op>; 2745 defm _pad_s_nop : SOPP_Real_64_gfx6_gfx7_gfx8_gfx9_gfx10<op>; 2746} 2747 2748defm S_NOP : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12<0x000>; 2749defm S_ENDPGM : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<0x001>; 2750defm S_WAKEUP : SOPP_Real_32_gfx8_gfx9_gfx10<0x003>; 2751defm S_BARRIER : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<0x00a>; 2752defm S_WAITCNT : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<0x00c>; 2753defm S_SETHALT : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<0x00d>; 2754defm S_SETKILL : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<0x00b>; 2755defm S_SLEEP : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<0x00e>; 2756defm S_SETPRIO : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<0x00f>; 2757defm S_SENDMSG : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<0x010>; 2758defm S_SENDMSGHALT : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<0x011>; 2759defm S_TRAP : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<0x012>; 2760defm S_ICACHE_INV : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<0x013>; 2761defm S_INCPERFLEVEL : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<0x014>; 2762defm S_DECPERFLEVEL : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<0x015>; 2763defm S_TTRACEDATA : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<0x016>; 2764defm S_ENDPGM_SAVED : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<0x01B>; 2765defm S_SET_GPR_IDX_OFF : SOPP_Real_32_gfx8_gfx9<0x01c>; 2766defm S_SET_GPR_IDX_MODE : SOPP_Real_32_gfx8_gfx9<0x01d>; 2767defm S_ENDPGM_ORDERED_PS_DONE : SOPP_Real_32_gfx8_gfx9_gfx10<0x01e>; 2768defm S_CODE_END : SOPP_Real_32_gfx10_gfx11_gfx12<0x01f>; 2769defm S_INST_PREFETCH : SOPP_Real_32_gfx10<0x020>; 2770defm S_CLAUSE : SOPP_Real_32_gfx10<0x021>; 2771defm S_WAIT_IDLE : SOPP_Real_32_gfx10<0x022>; 2772defm S_WAITCNT_DEPCTR : SOPP_Real_32_gfx10<0x023>; 2773defm S_ROUND_MODE : SOPP_Real_32_gfx10<0x024>; 2774defm S_DENORM_MODE : SOPP_Real_32_gfx10<0x025>; 2775defm S_TTRACEDATA_IMM : SOPP_Real_32_gfx10<0x028>; 2776 2777let isBranch = 1 in { 2778defm S_BRANCH : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10<0x002>; 2779defm S_CBRANCH_SCC0 : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10<0x004>; 2780defm S_CBRANCH_SCC1 : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10<0x005>; 2781defm S_CBRANCH_VCCZ : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10<0x006>; 2782defm S_CBRANCH_VCCNZ : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10<0x007>; 2783defm S_CBRANCH_EXECZ : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10<0x008>; 2784defm S_CBRANCH_EXECNZ : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10<0x009>; 2785defm S_CBRANCH_CDBGSYS : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10<0x017>; 2786defm S_CBRANCH_CDBGUSER : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10<0x018>; 2787defm S_CBRANCH_CDBGSYS_OR_USER : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10<0x019>; 2788defm S_CBRANCH_CDBGSYS_AND_USER : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10<0x01A>; 2789} 2790 2791//===----------------------------------------------------------------------===// 2792// SOPC - GFX11, GFX12. 2793//===----------------------------------------------------------------------===// 2794 2795multiclass SOPC_Real_gfx12<bits<7> op> { 2796 def _gfx12 : SOPC_Real<op, !cast<SOPC_Pseudo>(NAME)>, 2797 Select_gfx12<!cast<SOPC_Pseudo>(NAME).Mnemonic>; 2798} 2799 2800multiclass SOPC_Real_gfx11<bits<7> op> { 2801 def _gfx11 : SOPC_Real<op, !cast<SOPC_Pseudo>(NAME)>, 2802 Select_gfx11<!cast<SOPC_Pseudo>(NAME).Mnemonic>; 2803} 2804 2805multiclass SOPC_Real_gfx11_gfx12<bits<7> op> : 2806 SOPC_Real_gfx11<op>, SOPC_Real_gfx12<op>; 2807 2808defm S_CMP_EQ_I32 : SOPC_Real_gfx11_gfx12<0x00>; 2809defm S_CMP_LG_I32 : SOPC_Real_gfx11_gfx12<0x01>; 2810defm S_CMP_GT_I32 : SOPC_Real_gfx11_gfx12<0x02>; 2811defm S_CMP_GE_I32 : SOPC_Real_gfx11_gfx12<0x03>; 2812defm S_CMP_LT_I32 : SOPC_Real_gfx11_gfx12<0x04>; 2813defm S_CMP_LE_I32 : SOPC_Real_gfx11_gfx12<0x05>; 2814defm S_CMP_EQ_U32 : SOPC_Real_gfx11_gfx12<0x06>; 2815defm S_CMP_LG_U32 : SOPC_Real_gfx11_gfx12<0x07>; 2816defm S_CMP_GT_U32 : SOPC_Real_gfx11_gfx12<0x08>; 2817defm S_CMP_GE_U32 : SOPC_Real_gfx11_gfx12<0x09>; 2818defm S_CMP_LT_U32 : SOPC_Real_gfx11_gfx12<0x0a>; 2819defm S_CMP_LE_U32 : SOPC_Real_gfx11_gfx12<0x0b>; 2820defm S_BITCMP0_B32 : SOPC_Real_gfx11_gfx12<0x0c>; 2821defm S_BITCMP1_B32 : SOPC_Real_gfx11_gfx12<0x0d>; 2822defm S_BITCMP0_B64 : SOPC_Real_gfx11_gfx12<0x0e>; 2823defm S_BITCMP1_B64 : SOPC_Real_gfx11_gfx12<0x0f>; 2824defm S_CMP_EQ_U64 : SOPC_Real_gfx11_gfx12<0x10>; 2825defm S_CMP_LG_U64 : SOPC_Real_gfx11_gfx12<0x11>; 2826 2827//===----------------------------------------------------------------------===// 2828// SOPC - GFX1150, GFX12 2829//===----------------------------------------------------------------------===// 2830 2831defm S_CMP_LT_F32 : SOPC_Real_gfx11_gfx12<0x41>; 2832defm S_CMP_EQ_F32 : SOPC_Real_gfx11_gfx12<0x42>; 2833defm S_CMP_LE_F32 : SOPC_Real_gfx11_gfx12<0x43>; 2834defm S_CMP_GT_F32 : SOPC_Real_gfx11_gfx12<0x44>; 2835defm S_CMP_LG_F32 : SOPC_Real_gfx11_gfx12<0x45>; 2836defm S_CMP_GE_F32 : SOPC_Real_gfx11_gfx12<0x46>; 2837defm S_CMP_O_F32 : SOPC_Real_gfx11_gfx12<0x47>; 2838defm S_CMP_U_F32 : SOPC_Real_gfx11_gfx12<0x48>; 2839defm S_CMP_NGE_F32 : SOPC_Real_gfx11_gfx12<0x49>; 2840defm S_CMP_NLG_F32 : SOPC_Real_gfx11_gfx12<0x4a>; 2841defm S_CMP_NGT_F32 : SOPC_Real_gfx11_gfx12<0x4b>; 2842defm S_CMP_NLE_F32 : SOPC_Real_gfx11_gfx12<0x4c>; 2843defm S_CMP_NEQ_F32 : SOPC_Real_gfx11_gfx12<0x4d>; 2844defm S_CMP_NLT_F32 : SOPC_Real_gfx11_gfx12<0x4e>; 2845 2846defm S_CMP_LT_F16 : SOPC_Real_gfx11_gfx12<0x51>; 2847defm S_CMP_EQ_F16 : SOPC_Real_gfx11_gfx12<0x52>; 2848defm S_CMP_LE_F16 : SOPC_Real_gfx11_gfx12<0x53>; 2849defm S_CMP_GT_F16 : SOPC_Real_gfx11_gfx12<0x54>; 2850defm S_CMP_LG_F16 : SOPC_Real_gfx11_gfx12<0x55>; 2851defm S_CMP_GE_F16 : SOPC_Real_gfx11_gfx12<0x56>; 2852defm S_CMP_O_F16 : SOPC_Real_gfx11_gfx12<0x57>; 2853defm S_CMP_U_F16 : SOPC_Real_gfx11_gfx12<0x58>; 2854defm S_CMP_NGE_F16 : SOPC_Real_gfx11_gfx12<0x59>; 2855defm S_CMP_NLG_F16 : SOPC_Real_gfx11_gfx12<0x5a>; 2856defm S_CMP_NGT_F16 : SOPC_Real_gfx11_gfx12<0x5b>; 2857defm S_CMP_NLE_F16 : SOPC_Real_gfx11_gfx12<0x5c>; 2858defm S_CMP_NEQ_F16 : SOPC_Real_gfx11_gfx12<0x5d>; 2859defm S_CMP_NLT_F16 : SOPC_Real_gfx11_gfx12<0x5e>; 2860 2861//===----------------------------------------------------------------------===// 2862// SOPC - GFX6, GFX7, GFX8, GFX9, GFX10 2863//===----------------------------------------------------------------------===// 2864 2865multiclass SOPC_Real_gfx6_gfx7<bits<7> op> { 2866 defvar ps = !cast<SOPC_Pseudo>(NAME); 2867 def _gfx6_gfx7 : SOPC_Real<op, ps>, 2868 Select_gfx6_gfx7<ps.Mnemonic>; 2869} 2870 2871multiclass SOPC_Real_gfx8_gfx9<bits<7> op> { 2872 defvar ps = !cast<SOPC_Pseudo>(NAME); 2873 def _vi : SOPC_Real<op, ps>, 2874 Select_vi<ps.Mnemonic>; 2875} 2876 2877multiclass SOPC_Real_gfx10<bits<7> op> { 2878 defvar ps = !cast<SOPC_Pseudo>(NAME); 2879 def _gfx10 : SOPC_Real<op, ps>, 2880 Select_gfx10<ps.Mnemonic>; 2881} 2882 2883multiclass SOPC_Real_gfx8_gfx9_gfx10<bits<7> op> : 2884 SOPC_Real_gfx8_gfx9<op>, SOPC_Real_gfx10<op>; 2885 2886multiclass SOPC_Real_gfx6_gfx7_gfx8_gfx9<bits<7> op> : 2887 SOPC_Real_gfx6_gfx7<op>, SOPC_Real_gfx8_gfx9<op>; 2888 2889multiclass SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10<bits<7> op> : 2890 SOPC_Real_gfx6_gfx7_gfx8_gfx9<op>, SOPC_Real_gfx10<op>; 2891 2892defm S_CMP_EQ_I32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10<0x00>; 2893defm S_CMP_LG_I32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10<0x01>; 2894defm S_CMP_GT_I32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10<0x02>; 2895defm S_CMP_GE_I32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10<0x03>; 2896defm S_CMP_LT_I32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10<0x04>; 2897defm S_CMP_LE_I32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10<0x05>; 2898defm S_CMP_EQ_U32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10<0x06>; 2899defm S_CMP_LG_U32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10<0x07>; 2900defm S_CMP_GT_U32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10<0x08>; 2901defm S_CMP_GE_U32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10<0x09>; 2902defm S_CMP_LT_U32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10<0x0a>; 2903defm S_CMP_LE_U32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10<0x0b>; 2904defm S_BITCMP0_B32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10<0x0c>; 2905defm S_BITCMP1_B32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10<0x0d>; 2906defm S_BITCMP0_B64 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10<0x0e>; 2907defm S_BITCMP1_B64 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10<0x0f>; 2908defm S_SETVSKIP : SOPC_Real_gfx6_gfx7_gfx8_gfx9<0x10>; 2909defm S_SET_GPR_IDX_ON : SOPC_Real_gfx8_gfx9<0x11>; 2910defm S_CMP_EQ_U64 : SOPC_Real_gfx8_gfx9_gfx10<0x12>; 2911defm S_CMP_LG_U64 : SOPC_Real_gfx8_gfx9_gfx10<0x13>; 2912 2913//===----------------------------------------------------------------------===// 2914// GFX8 (VI), GFX9. 2915//===----------------------------------------------------------------------===// 2916 2917class SOP1_Real_vi<bits<8> op, SOP1_Pseudo ps> : 2918 SOP1_Real<op, ps>, 2919 Select_vi<ps.Mnemonic>; 2920 2921class SOP2_Real_vi<bits<7> op, SOP2_Pseudo ps> : 2922 SOP2_Real32<op, ps>, 2923 Select_vi<ps.Mnemonic>; 2924 2925class SOPK_Real_vi<bits<5> op, SOPK_Pseudo ps> : 2926 SOPK_Real32<op, ps>, 2927 Select_vi<ps.Mnemonic>; 2928 2929def S_MOV_B32_vi : SOP1_Real_vi <0x00, S_MOV_B32>; 2930def S_MOV_B64_vi : SOP1_Real_vi <0x01, S_MOV_B64>; 2931def S_CMOV_B32_vi : SOP1_Real_vi <0x02, S_CMOV_B32>; 2932def S_CMOV_B64_vi : SOP1_Real_vi <0x03, S_CMOV_B64>; 2933def S_NOT_B32_vi : SOP1_Real_vi <0x04, S_NOT_B32>; 2934def S_NOT_B64_vi : SOP1_Real_vi <0x05, S_NOT_B64>; 2935def S_WQM_B32_vi : SOP1_Real_vi <0x06, S_WQM_B32>; 2936def S_WQM_B64_vi : SOP1_Real_vi <0x07, S_WQM_B64>; 2937def S_BREV_B32_vi : SOP1_Real_vi <0x08, S_BREV_B32>; 2938def S_BREV_B64_vi : SOP1_Real_vi <0x09, S_BREV_B64>; 2939def S_BCNT0_I32_B32_vi : SOP1_Real_vi <0x0a, S_BCNT0_I32_B32>; 2940def S_BCNT0_I32_B64_vi : SOP1_Real_vi <0x0b, S_BCNT0_I32_B64>; 2941def S_BCNT1_I32_B32_vi : SOP1_Real_vi <0x0c, S_BCNT1_I32_B32>; 2942def S_BCNT1_I32_B64_vi : SOP1_Real_vi <0x0d, S_BCNT1_I32_B64>; 2943def S_FF0_I32_B32_vi : SOP1_Real_vi <0x0e, S_FF0_I32_B32>; 2944def S_FF0_I32_B64_vi : SOP1_Real_vi <0x0f, S_FF0_I32_B64>; 2945def S_FF1_I32_B32_vi : SOP1_Real_vi <0x10, S_FF1_I32_B32>; 2946def S_FF1_I32_B64_vi : SOP1_Real_vi <0x11, S_FF1_I32_B64>; 2947def S_FLBIT_I32_B32_vi : SOP1_Real_vi <0x12, S_FLBIT_I32_B32>; 2948def S_FLBIT_I32_B64_vi : SOP1_Real_vi <0x13, S_FLBIT_I32_B64>; 2949def S_FLBIT_I32_vi : SOP1_Real_vi <0x14, S_FLBIT_I32>; 2950def S_FLBIT_I32_I64_vi : SOP1_Real_vi <0x15, S_FLBIT_I32_I64>; 2951def S_SEXT_I32_I8_vi : SOP1_Real_vi <0x16, S_SEXT_I32_I8>; 2952def S_SEXT_I32_I16_vi : SOP1_Real_vi <0x17, S_SEXT_I32_I16>; 2953def S_BITSET0_B32_vi : SOP1_Real_vi <0x18, S_BITSET0_B32>; 2954def S_BITSET0_B64_vi : SOP1_Real_vi <0x19, S_BITSET0_B64>; 2955def S_BITSET1_B32_vi : SOP1_Real_vi <0x1a, S_BITSET1_B32>; 2956def S_BITSET1_B64_vi : SOP1_Real_vi <0x1b, S_BITSET1_B64>; 2957def S_GETPC_B64_vi : SOP1_Real_vi <0x1c, S_GETPC_B64>; 2958def S_SETPC_B64_vi : SOP1_Real_vi <0x1d, S_SETPC_B64>; 2959def S_SWAPPC_B64_vi : SOP1_Real_vi <0x1e, S_SWAPPC_B64>; 2960def S_RFE_B64_vi : SOP1_Real_vi <0x1f, S_RFE_B64>; 2961def S_AND_SAVEEXEC_B64_vi : SOP1_Real_vi <0x20, S_AND_SAVEEXEC_B64>; 2962def S_OR_SAVEEXEC_B64_vi : SOP1_Real_vi <0x21, S_OR_SAVEEXEC_B64>; 2963def S_XOR_SAVEEXEC_B64_vi : SOP1_Real_vi <0x22, S_XOR_SAVEEXEC_B64>; 2964def S_ANDN2_SAVEEXEC_B64_vi: SOP1_Real_vi <0x23, S_ANDN2_SAVEEXEC_B64>; 2965def S_ORN2_SAVEEXEC_B64_vi : SOP1_Real_vi <0x24, S_ORN2_SAVEEXEC_B64>; 2966def S_NAND_SAVEEXEC_B64_vi : SOP1_Real_vi <0x25, S_NAND_SAVEEXEC_B64>; 2967def S_NOR_SAVEEXEC_B64_vi : SOP1_Real_vi <0x26, S_NOR_SAVEEXEC_B64>; 2968def S_XNOR_SAVEEXEC_B64_vi : SOP1_Real_vi <0x27, S_XNOR_SAVEEXEC_B64>; 2969def S_QUADMASK_B32_vi : SOP1_Real_vi <0x28, S_QUADMASK_B32>; 2970def S_QUADMASK_B64_vi : SOP1_Real_vi <0x29, S_QUADMASK_B64>; 2971def S_MOVRELS_B32_vi : SOP1_Real_vi <0x2a, S_MOVRELS_B32>; 2972def S_MOVRELS_B64_vi : SOP1_Real_vi <0x2b, S_MOVRELS_B64>; 2973def S_MOVRELD_B32_vi : SOP1_Real_vi <0x2c, S_MOVRELD_B32>; 2974def S_MOVRELD_B64_vi : SOP1_Real_vi <0x2d, S_MOVRELD_B64>; 2975def S_CBRANCH_JOIN_vi : SOP1_Real_vi <0x2e, S_CBRANCH_JOIN>; 2976def S_ABS_I32_vi : SOP1_Real_vi <0x30, S_ABS_I32>; 2977def S_SET_GPR_IDX_IDX_vi : SOP1_Real_vi <0x32, S_SET_GPR_IDX_IDX>; 2978 2979def S_ADD_U32_vi : SOP2_Real_vi <0x00, S_ADD_U32>; 2980def S_ADD_I32_vi : SOP2_Real_vi <0x02, S_ADD_I32>; 2981def S_SUB_U32_vi : SOP2_Real_vi <0x01, S_SUB_U32>; 2982def S_SUB_I32_vi : SOP2_Real_vi <0x03, S_SUB_I32>; 2983def S_ADDC_U32_vi : SOP2_Real_vi <0x04, S_ADDC_U32>; 2984def S_SUBB_U32_vi : SOP2_Real_vi <0x05, S_SUBB_U32>; 2985def S_MIN_I32_vi : SOP2_Real_vi <0x06, S_MIN_I32>; 2986def S_MIN_U32_vi : SOP2_Real_vi <0x07, S_MIN_U32>; 2987def S_MAX_I32_vi : SOP2_Real_vi <0x08, S_MAX_I32>; 2988def S_MAX_U32_vi : SOP2_Real_vi <0x09, S_MAX_U32>; 2989def S_CSELECT_B32_vi : SOP2_Real_vi <0x0a, S_CSELECT_B32>; 2990def S_CSELECT_B64_vi : SOP2_Real_vi <0x0b, S_CSELECT_B64>; 2991def S_AND_B32_vi : SOP2_Real_vi <0x0c, S_AND_B32>; 2992def S_AND_B64_vi : SOP2_Real_vi <0x0d, S_AND_B64>; 2993def S_OR_B32_vi : SOP2_Real_vi <0x0e, S_OR_B32>; 2994def S_OR_B64_vi : SOP2_Real_vi <0x0f, S_OR_B64>; 2995def S_XOR_B32_vi : SOP2_Real_vi <0x10, S_XOR_B32>; 2996def S_XOR_B64_vi : SOP2_Real_vi <0x11, S_XOR_B64>; 2997def S_ANDN2_B32_vi : SOP2_Real_vi <0x12, S_ANDN2_B32>; 2998def S_ANDN2_B64_vi : SOP2_Real_vi <0x13, S_ANDN2_B64>; 2999def S_ORN2_B32_vi : SOP2_Real_vi <0x14, S_ORN2_B32>; 3000def S_ORN2_B64_vi : SOP2_Real_vi <0x15, S_ORN2_B64>; 3001def S_NAND_B32_vi : SOP2_Real_vi <0x16, S_NAND_B32>; 3002def S_NAND_B64_vi : SOP2_Real_vi <0x17, S_NAND_B64>; 3003def S_NOR_B32_vi : SOP2_Real_vi <0x18, S_NOR_B32>; 3004def S_NOR_B64_vi : SOP2_Real_vi <0x19, S_NOR_B64>; 3005def S_XNOR_B32_vi : SOP2_Real_vi <0x1a, S_XNOR_B32>; 3006def S_XNOR_B64_vi : SOP2_Real_vi <0x1b, S_XNOR_B64>; 3007def S_LSHL_B32_vi : SOP2_Real_vi <0x1c, S_LSHL_B32>; 3008def S_LSHL_B64_vi : SOP2_Real_vi <0x1d, S_LSHL_B64>; 3009def S_LSHR_B32_vi : SOP2_Real_vi <0x1e, S_LSHR_B32>; 3010def S_LSHR_B64_vi : SOP2_Real_vi <0x1f, S_LSHR_B64>; 3011def S_ASHR_I32_vi : SOP2_Real_vi <0x20, S_ASHR_I32>; 3012def S_ASHR_I64_vi : SOP2_Real_vi <0x21, S_ASHR_I64>; 3013def S_BFM_B32_vi : SOP2_Real_vi <0x22, S_BFM_B32>; 3014def S_BFM_B64_vi : SOP2_Real_vi <0x23, S_BFM_B64>; 3015def S_MUL_I32_vi : SOP2_Real_vi <0x24, S_MUL_I32>; 3016def S_BFE_U32_vi : SOP2_Real_vi <0x25, S_BFE_U32>; 3017def S_BFE_I32_vi : SOP2_Real_vi <0x26, S_BFE_I32>; 3018def S_BFE_U64_vi : SOP2_Real_vi <0x27, S_BFE_U64>; 3019def S_BFE_I64_vi : SOP2_Real_vi <0x28, S_BFE_I64>; 3020def S_CBRANCH_G_FORK_vi : SOP2_Real_vi <0x29, S_CBRANCH_G_FORK>; 3021def S_ABSDIFF_I32_vi : SOP2_Real_vi <0x2a, S_ABSDIFF_I32>; 3022def S_PACK_LL_B32_B16_vi : SOP2_Real_vi <0x32, S_PACK_LL_B32_B16>; 3023def S_PACK_LH_B32_B16_vi : SOP2_Real_vi <0x33, S_PACK_LH_B32_B16>; 3024def S_PACK_HH_B32_B16_vi : SOP2_Real_vi <0x34, S_PACK_HH_B32_B16>; 3025def S_RFE_RESTORE_B64_vi : SOP2_Real_vi <0x2b, S_RFE_RESTORE_B64>; 3026 3027def S_MOVK_I32_vi : SOPK_Real_vi <0x00, S_MOVK_I32>; 3028def S_CMOVK_I32_vi : SOPK_Real_vi <0x01, S_CMOVK_I32>; 3029def S_CMPK_EQ_I32_vi : SOPK_Real_vi <0x02, S_CMPK_EQ_I32>; 3030def S_CMPK_LG_I32_vi : SOPK_Real_vi <0x03, S_CMPK_LG_I32>; 3031def S_CMPK_GT_I32_vi : SOPK_Real_vi <0x04, S_CMPK_GT_I32>; 3032def S_CMPK_GE_I32_vi : SOPK_Real_vi <0x05, S_CMPK_GE_I32>; 3033def S_CMPK_LT_I32_vi : SOPK_Real_vi <0x06, S_CMPK_LT_I32>; 3034def S_CMPK_LE_I32_vi : SOPK_Real_vi <0x07, S_CMPK_LE_I32>; 3035def S_CMPK_EQ_U32_vi : SOPK_Real_vi <0x08, S_CMPK_EQ_U32>; 3036def S_CMPK_LG_U32_vi : SOPK_Real_vi <0x09, S_CMPK_LG_U32>; 3037def S_CMPK_GT_U32_vi : SOPK_Real_vi <0x0A, S_CMPK_GT_U32>; 3038def S_CMPK_GE_U32_vi : SOPK_Real_vi <0x0B, S_CMPK_GE_U32>; 3039def S_CMPK_LT_U32_vi : SOPK_Real_vi <0x0C, S_CMPK_LT_U32>; 3040def S_CMPK_LE_U32_vi : SOPK_Real_vi <0x0D, S_CMPK_LE_U32>; 3041def S_ADDK_I32_vi : SOPK_Real_vi <0x0E, S_ADDK_I32>; 3042def S_MULK_I32_vi : SOPK_Real_vi <0x0F, S_MULK_I32>; 3043def S_CBRANCH_I_FORK_vi : SOPK_Real_vi <0x10, S_CBRANCH_I_FORK>; 3044def S_GETREG_B32_vi : SOPK_Real_vi <0x11, S_GETREG_B32>; 3045def S_SETREG_B32_vi : SOPK_Real_vi <0x12, S_SETREG_B32>; 3046//def S_GETREG_REGRD_B32_vi : SOPK_Real_vi <0x13, S_GETREG_REGRD_B32>; // see pseudo for comments 3047def S_SETREG_IMM32_B32_vi : SOPK_Real64<0x14, S_SETREG_IMM32_B32>, 3048 Select_vi<S_SETREG_IMM32_B32.Mnemonic>; 3049 3050def S_CALL_B64_vi : SOPK_Real_vi <0x15, S_CALL_B64>; 3051 3052//===----------------------------------------------------------------------===// 3053// SOP1 - GFX9. 3054//===----------------------------------------------------------------------===// 3055 3056def S_ANDN1_SAVEEXEC_B64_vi : SOP1_Real_vi<0x33, S_ANDN1_SAVEEXEC_B64>; 3057def S_ORN1_SAVEEXEC_B64_vi : SOP1_Real_vi<0x34, S_ORN1_SAVEEXEC_B64>; 3058def S_ANDN1_WREXEC_B64_vi : SOP1_Real_vi<0x35, S_ANDN1_WREXEC_B64>; 3059def S_ANDN2_WREXEC_B64_vi : SOP1_Real_vi<0x36, S_ANDN2_WREXEC_B64>; 3060def S_BITREPLICATE_B64_B32_vi : SOP1_Real_vi<0x37, S_BITREPLICATE_B64_B32>; 3061 3062//===----------------------------------------------------------------------===// 3063// SOP2 - GFX9. 3064//===----------------------------------------------------------------------===// 3065 3066def S_LSHL1_ADD_U32_vi : SOP2_Real_vi<0x2e, S_LSHL1_ADD_U32>; 3067def S_LSHL2_ADD_U32_vi : SOP2_Real_vi<0x2f, S_LSHL2_ADD_U32>; 3068def S_LSHL3_ADD_U32_vi : SOP2_Real_vi<0x30, S_LSHL3_ADD_U32>; 3069def S_LSHL4_ADD_U32_vi : SOP2_Real_vi<0x31, S_LSHL4_ADD_U32>; 3070def S_MUL_HI_U32_vi : SOP2_Real_vi<0x2c, S_MUL_HI_U32>; 3071def S_MUL_HI_I32_vi : SOP2_Real_vi<0x2d, S_MUL_HI_I32>; 3072