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