1//===-- MIMGInstructions.td - MIMG 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 9// MIMG-specific encoding families to distinguish between semantically 10// equivalent machine instructions with different encoding. 11// 12// - MIMGEncGfx6: encoding introduced with gfx6 (obsoleted for atomics in gfx8) 13// - MIMGEncGfx8: encoding introduced with gfx8 for atomics 14// - MIMGEncGfx90a: encoding for gfx90a for atomics 15// - MIMGEncGfx10Default: gfx10 default (non-NSA) encoding 16// - MIMGEncGfx10NSA: gfx10 NSA encoding 17// - MIMGEncGfx11Default: gfx11 default (non-NSA) encoding 18// - MIMGEncGfx11NSA: gfx11 NSA encoding 19class MIMGEncoding; 20 21def MIMGEncGfx6 : MIMGEncoding; 22def MIMGEncGfx8 : MIMGEncoding; 23def MIMGEncGfx90a : MIMGEncoding; 24def MIMGEncGfx10Default : MIMGEncoding; 25def MIMGEncGfx10NSA : MIMGEncoding; 26def MIMGEncGfx11Default : MIMGEncoding; 27def MIMGEncGfx11NSA : MIMGEncoding; 28def MIMGEncGfx12 : MIMGEncoding; 29 30def MIMGEncoding : GenericEnum { 31 let FilterClass = "MIMGEncoding"; 32} 33 34// Represent an ISA-level opcode, independent of the encoding and the 35// vdata/vaddr size. 36class MIMGBaseOpcode : PredicateControl { 37 MIMGBaseOpcode BaseOpcode = !cast<MIMGBaseOpcode>(NAME); 38 bit Store = 0; 39 bit Atomic = 0; 40 bit AtomicX2 = 0; // (f)cmpswap 41 bit Sampler = 0; 42 bit Gather4 = 0; 43 bits<8> NumExtraArgs = 0; 44 bit Gradients = 0; 45 bit G16 = 0; 46 bit Coordinates = 1; 47 bit LodOrClampOrMip = 0; 48 bit HasD16 = 0; 49 bit IsAtomicRet = 0; 50 bit MSAA = 0; 51 bit BVH = 0; 52 bit A16 = 0; 53} 54 55def MIMGBaseOpcode : GenericEnum { 56 let FilterClass = "MIMGBaseOpcode"; 57} 58 59def MIMGBaseOpcodesTable : GenericTable { 60 let FilterClass = "MIMGBaseOpcode"; 61 let CppTypeName = "MIMGBaseOpcodeInfo"; 62 let Fields = ["BaseOpcode", "Store", "Atomic", "AtomicX2", "Sampler", 63 "Gather4", "NumExtraArgs", "Gradients", "G16", "Coordinates", 64 "LodOrClampOrMip", "HasD16", "MSAA", "BVH", "A16"]; 65 string TypeOf_BaseOpcode = "MIMGBaseOpcode"; 66 67 let PrimaryKey = ["BaseOpcode"]; 68 let PrimaryKeyName = "getMIMGBaseOpcodeInfo"; 69} 70 71def MIMGDim : GenericEnum { 72 let FilterClass = "AMDGPUDimProps"; 73} 74 75def MIMGDimInfoTable : GenericTable { 76 let FilterClass = "AMDGPUDimProps"; 77 let CppTypeName = "MIMGDimInfo"; 78 let Fields = ["Dim", "NumCoords", "NumGradients", "MSAA", "DA", "Encoding", "AsmSuffix"]; 79 string TypeOf_Dim = "MIMGDim"; 80 81 let PrimaryKey = ["Dim"]; 82 let PrimaryKeyName = "getMIMGDimInfo"; 83} 84 85def getMIMGDimInfoByEncoding : SearchIndex { 86 let Table = MIMGDimInfoTable; 87 let Key = ["Encoding"]; 88} 89 90def getMIMGDimInfoByAsmSuffix : SearchIndex { 91 let Table = MIMGDimInfoTable; 92 let Key = ["AsmSuffix"]; 93} 94 95def MIMG { 96 int NOP = -1; 97} 98 99class mimgopc <int gfx12, int gfx11, int gfx10m, int vi = gfx10m, int si = gfx10m> { 100 field bits<8> GFX12 = gfx12; 101 field bits<8> GFX11 = gfx11; 102 field bits<8> GFX10M = gfx10m; // GFX10minus for all but atomics 103 field bits<8> VI = vi; // VI is only used for atomic/sampler/gather instructions 104 field bits<8> SI = si; // SI is only used for atomic instructions 105 bit HAS_GFX12 = !ne(gfx12, MIMG.NOP); 106 bit HAS_GFX11 = !ne(gfx11, MIMG.NOP); 107 bit HAS_GFX10M = !ne(gfx10m, MIMG.NOP); 108 bit HAS_VI = !ne(vi, MIMG.NOP); 109 bit HAS_SI = !ne(si, MIMG.NOP); 110} 111 112class MIMGLZMapping<MIMGBaseOpcode l, MIMGBaseOpcode lz> { 113 MIMGBaseOpcode L = l; 114 MIMGBaseOpcode LZ = lz; 115} 116 117def MIMGLZMappingTable : GenericTable { 118 let FilterClass = "MIMGLZMapping"; 119 let CppTypeName = "MIMGLZMappingInfo"; 120 let Fields = ["L", "LZ"]; 121 string TypeOf_L = "MIMGBaseOpcode"; 122 string TypeOf_LZ = "MIMGBaseOpcode"; 123 124 let PrimaryKey = ["L"]; 125 let PrimaryKeyName = "getMIMGLZMappingInfo"; 126} 127 128class MIMGMIPMapping<MIMGBaseOpcode mip, MIMGBaseOpcode nonmip> { 129 MIMGBaseOpcode MIP = mip; 130 MIMGBaseOpcode NONMIP = nonmip; 131} 132 133def MIMGMIPMappingTable : GenericTable { 134 let FilterClass = "MIMGMIPMapping"; 135 let CppTypeName = "MIMGMIPMappingInfo"; 136 let Fields = ["MIP", "NONMIP"]; 137 string TypeOf_MIP = "MIMGBaseOpcode"; 138 string TypeOf_NONMIP = "MIMGBaseOpcode"; 139 140 let PrimaryKey = ["MIP"]; 141 let PrimaryKeyName = "getMIMGMIPMappingInfo"; 142} 143 144class MIMGBiasMapping<MIMGBaseOpcode bias, MIMGBaseOpcode nobias> { 145 MIMGBaseOpcode Bias = bias; 146 MIMGBaseOpcode NoBias = nobias; 147} 148 149def MIMGBiasMappingTable : GenericTable { 150 let FilterClass = "MIMGBiasMapping"; 151 let CppTypeName = "MIMGBiasMappingInfo"; 152 let Fields = ["Bias", "NoBias"]; 153 string TypeOf_Bias = "MIMGBaseOpcode"; 154 string TypeOf_NoBias = "MIMGBaseOpcode"; 155 156 let PrimaryKey = ["Bias"]; 157 let PrimaryKeyName = "getMIMGBiasMappingInfo"; 158} 159 160class MIMGOffsetMapping<MIMGBaseOpcode offset, MIMGBaseOpcode nooffset> { 161 MIMGBaseOpcode Offset = offset; 162 MIMGBaseOpcode NoOffset = nooffset; 163} 164 165def MIMGOffsetMappingTable : GenericTable { 166 let FilterClass = "MIMGOffsetMapping"; 167 let CppTypeName = "MIMGOffsetMappingInfo"; 168 let Fields = ["Offset", "NoOffset"]; 169 string TypeOf_Offset = "MIMGBaseOpcode"; 170 string TypeOf_NoOffset = "MIMGBaseOpcode"; 171 172 let PrimaryKey = ["Offset"]; 173 let PrimaryKeyName = "getMIMGOffsetMappingInfo"; 174} 175 176class MIMGG16Mapping<MIMGBaseOpcode g, MIMGBaseOpcode g16> { 177 MIMGBaseOpcode G = g; 178 MIMGBaseOpcode G16 = g16; 179} 180 181def MIMGG16MappingTable : GenericTable { 182 let FilterClass = "MIMGG16Mapping"; 183 let CppTypeName = "MIMGG16MappingInfo"; 184 let Fields = ["G", "G16"]; 185 string TypeOf_G = "MIMGBaseOpcode"; 186 string TypeOf_G16 = "MIMGBaseOpcode"; 187 188 let PrimaryKey = ["G"]; 189 let PrimaryKeyName = "getMIMGG16MappingInfo"; 190} 191 192class MIMG_Base <dag outs, string dns = ""> 193 : InstSI <outs, (ins), "", []> { 194 195 let VM_CNT = 1; 196 let EXP_CNT = 1; 197 let MIMG = 1; 198 let Uses = [EXEC]; 199 let mayLoad = 1; 200 let mayStore = 0; 201 let SchedRW = [WriteVMEM]; 202 let UseNamedOperandTable = 1; 203 let hasSideEffects = 0; // XXX ???? 204 205 let DecoderNamespace = dns; 206 let isAsmParserOnly = !eq(dns, ""); 207} 208 209class MIMG <dag outs, string dns = ""> 210 : MIMG_Base <outs, dns> { 211 212 let hasPostISelHook = 1; 213 214 Instruction Opcode = !cast<Instruction>(NAME); 215 MIMGBaseOpcode BaseOpcode; 216 MIMGEncoding MIMGEncoding; 217 bits<8> VDataDwords; 218 bits<8> VAddrDwords; 219 220 // If NSA is used this counts number of operands VAddrDwords is split into. 221 bits<8> VAddrOperands; 222} 223 224class VIMAGE <dag outs, string dns = ""> : MIMG<outs, dns> { 225 let MIMG = 0; 226 let VIMAGE = 1; 227} 228 229class VSAMPLE <dag outs, string dns = ""> : MIMG<outs, dns> { 230 let MIMG = 0; 231 let VSAMPLE = 1; 232} 233 234def MIMGInfoTable : GenericTable { 235 let FilterClass = "MIMG"; 236 let CppTypeName = "MIMGInfo"; 237 let Fields = ["Opcode", "BaseOpcode", "MIMGEncoding", "VDataDwords", 238 "VAddrDwords", "VAddrOperands"]; 239 string TypeOf_BaseOpcode = "MIMGBaseOpcode"; 240 string TypeOf_MIMGEncoding = "MIMGEncoding"; 241 242 let PrimaryKey = ["BaseOpcode", "MIMGEncoding", "VDataDwords", "VAddrDwords"]; 243 let PrimaryKeyName = "getMIMGOpcodeHelper"; 244} 245 246def getMIMGInfo : SearchIndex { 247 let Table = MIMGInfoTable; 248 let Key = ["Opcode"]; 249} 250 251class NSAHelper { 252 dag AddrIns; 253 string AddrAsm; 254 int NSA; 255} 256 257class MIMGNSAHelper<int num_addrs, 258 list<RegisterClass> addr_types=!listsplat(VGPR_32, num_addrs)> 259 : NSAHelper<> { 260 list<string> AddrAsmNames = !foreach(i, !range(num_addrs), "vaddr" # i); 261 let AddrIns = !dag(ins, addr_types, AddrAsmNames); 262 let AddrAsm = "[$" # !interleave(AddrAsmNames, ", $") # "]"; 263 264 let NSA = !if(!le(num_addrs, 1), ?, 265 !if(!le(num_addrs, 5), 1, 266 !if(!le(num_addrs, 9), 2, 267 !if(!le(num_addrs, 13), 3, ?)))); 268} 269 270class PartialNSAHelper<int num_addrs, int max_addr, RegisterClass LastAddrRC> 271 : NSAHelper<> { 272 273 list<RegisterClass> addr_types = 274 !if(!ge(num_addrs, max_addr), 275 !listconcat(!listsplat(VGPR_32, !sub(max_addr, 1)), [LastAddrRC]), 276 !listsplat(VGPR_32, num_addrs)); 277 278 int VAddrCount = !if(!gt(num_addrs, max_addr), max_addr, num_addrs); 279 list<string> AddrAsmNames = !foreach(i, !range(VAddrCount), "vaddr" # i); 280 281 let AddrIns = !dag(ins, addr_types, AddrAsmNames); 282 let AddrAsm = "[$" # !interleave(AddrAsmNames, ", $") # "]"; 283 let NSA = 1; 284} 285 286// Base class of all pre-gfx10 MIMG instructions. 287class MIMG_gfx6789<bits<8> op, dag outs, string dns = ""> 288 : MIMG<outs, dns>, MIMGe_gfx6789<op> { 289 let SubtargetPredicate = isGFX6GFX7GFX8GFX9NotGFX90A; 290 let AssemblerPredicate = isGFX6GFX7GFX8GFX9NotGFX90A; 291 292 let MIMGEncoding = MIMGEncGfx6; 293 let VAddrOperands = 1; 294 295 let d16 = !if(BaseOpcode.HasD16, ?, 0); 296} 297 298class MIMG_gfx90a<bits<8> op, dag outs, string dns = ""> 299 : MIMG<outs, dns>, MIMGe_gfx90a<op> { 300 let SubtargetPredicate = isGFX90APlus; 301 let AssemblerPredicate = isGFX90APlus; 302 303 let MIMGEncoding = MIMGEncGfx90a; 304 let VAddrOperands = 1; 305 306 let d16 = !if(BaseOpcode.HasD16, ?, 0); 307} 308 309// Base class of all non-NSA gfx10 MIMG instructions. 310class MIMG_gfx10<int op, dag outs, string dns = ""> 311 : MIMG<outs, dns>, MIMGe_gfx10<op> { 312 let SubtargetPredicate = isGFX10Only; 313 let AssemblerPredicate = isGFX10Only; 314 315 let MIMGEncoding = MIMGEncGfx10Default; 316 let VAddrOperands = 1; 317 318 let d16 = !if(BaseOpcode.HasD16, ?, 0); 319 let nsa = 0; 320} 321 322// Base class for all NSA MIMG instructions. 323// Note that 1-dword addresses always use non-NSA variants. 324class MIMG_nsa_gfx10<int op, dag outs, int num_addrs, string dns=""> 325 : MIMG<outs, dns>, MIMGe_gfx10<op> { 326 let SubtargetPredicate = isGFX10Only; 327 let AssemblerPredicate = isGFX10Only; 328 329 let MIMGEncoding = MIMGEncGfx10NSA; 330 let VAddrOperands = num_addrs; 331 332 MIMGNSAHelper nsah = MIMGNSAHelper<num_addrs>; 333 dag AddrIns = nsah.AddrIns; 334 string AddrAsm = nsah.AddrAsm; 335 336 let d16 = !if(BaseOpcode.HasD16, ?, 0); 337 let nsa = nsah.NSA; 338} 339 340// Base class of all non-NSA gfx11 MIMG instructions. 341class MIMG_gfx11<int op, dag outs, string dns = ""> 342 : MIMG<outs, dns>, MIMGe_gfx11<op> { 343 let SubtargetPredicate = isGFX11Only; 344 let AssemblerPredicate = isGFX11Only; 345 346 let MIMGEncoding = MIMGEncGfx11Default; 347 let VAddrOperands = 1; 348 349 let d16 = !if(BaseOpcode.HasD16, ?, 0); 350 let nsa = 0; 351} 352 353// Base class for all NSA MIMG instructions. 354// Note that 1-dword addresses always use non-NSA variants. 355class MIMG_nsa_gfx11<int op, dag outs, int num_addrs, string dns="", 356 list<RegisterClass> addr_types=[], 357 RegisterClass LastAddrRC = VGPR_32> 358 : MIMG<outs, dns>, MIMGe_gfx11<op> { 359 let SubtargetPredicate = isGFX11Only; 360 let AssemblerPredicate = isGFX11Only; 361 362 let MIMGEncoding = MIMGEncGfx11NSA; 363 let VAddrOperands = num_addrs; 364 365 NSAHelper nsah = !if(!empty(addr_types), 366 PartialNSAHelper<num_addrs, 5, LastAddrRC>, 367 MIMGNSAHelper<num_addrs, addr_types>); 368 dag AddrIns = nsah.AddrIns; 369 string AddrAsm = nsah.AddrAsm; 370 371 let d16 = !if(BaseOpcode.HasD16, ?, 0); 372 let nsa = nsah.NSA; 373} 374 375class VIMAGE_gfx12<int op, dag outs, int num_addrs, string dns="", 376 list<RegisterClass> addr_types=[]> 377 : VIMAGE<outs, dns>, VIMAGEe<op> { 378 let SubtargetPredicate = isGFX12Plus; 379 let AssemblerPredicate = isGFX12Plus; 380 381 let MIMGEncoding = MIMGEncGfx12; 382 let VAddrOperands = num_addrs; 383 384 MIMGNSAHelper nsah = !if(!empty(addr_types), 385 MIMGNSAHelper<num_addrs>, 386 MIMGNSAHelper<num_addrs, addr_types>); 387 dag AddrIns = nsah.AddrIns; 388 string AddrAsm = !if(!eq(num_addrs, 1), "$vaddr0", nsah.AddrAsm); 389 390 let d16 = !if(BaseOpcode.HasD16, ?, 0); 391 let vaddr1 = !if(!lt(num_addrs, 2), 0, ?); 392 let vaddr2 = !if(!lt(num_addrs, 3), 0, ?); 393 let vaddr3 = !if(!lt(num_addrs, 4), 0, ?); 394 let vaddr4 = !if(!lt(num_addrs, 5), 0, ?); 395} 396 397class VSAMPLE_gfx12<int op, dag outs, int num_addrs, string dns="", 398 RegisterClass Addr3RC> 399 : VSAMPLE<outs, dns>, VSAMPLEe<op> { 400 let SubtargetPredicate = isGFX12Plus; 401 let AssemblerPredicate = isGFX12Plus; 402 403 let MIMGEncoding = MIMGEncGfx12; 404 let VAddrOperands = num_addrs; 405 406 PartialNSAHelper nsah = PartialNSAHelper<num_addrs, 4, Addr3RC>; 407 408 dag AddrIns = nsah.AddrIns; 409 string AddrAsm = !if(!eq(num_addrs, 1), "$vaddr0", nsah.AddrAsm); 410 411 let d16 = !if(BaseOpcode.HasD16, ?, 0); 412 let vaddr1 = !if(!lt(num_addrs, 2), 0, ?); 413 let vaddr2 = !if(!lt(num_addrs, 3), 0, ?); 414 let vaddr3 = !if(!lt(num_addrs, 4), 0, ?); 415} 416 417class MIMG_NoSampler_Helper <mimgopc op, string asm, 418 RegisterClass dst_rc, 419 RegisterClass addr_rc, 420 string dns=""> 421 : MIMG_gfx6789 <op.GFX10M, (outs dst_rc:$vdata), dns> { 422 let InOperandList = !con((ins addr_rc:$vaddr, SReg_256:$srsrc, 423 DMask:$dmask, UNorm:$unorm, CPol:$cpol, 424 R128A16:$r128, TFE:$tfe, LWE:$lwe, DA:$da), 425 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 426 let AsmString = asm#" $vdata, $vaddr, $srsrc$dmask$unorm$cpol$r128$tfe$lwe$da" 427 #!if(BaseOpcode.HasD16, "$d16", ""); 428} 429 430class MIMG_NoSampler_Helper_gfx90a <mimgopc op, string asm, 431 RegisterClass dst_rc, 432 RegisterClass addr_rc, 433 string dns=""> 434 : MIMG_gfx90a <op.GFX10M, (outs getLdStRegisterOperand<dst_rc>.ret:$vdata), dns> { 435 let InOperandList = !con((ins addr_rc:$vaddr, SReg_256:$srsrc, 436 DMask:$dmask, UNorm:$unorm, CPol:$cpol, 437 R128A16:$r128, LWE:$lwe, DA:$da), 438 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 439 let AsmString = asm#" $vdata, $vaddr, $srsrc$dmask$unorm$cpol$r128$lwe$da" 440 #!if(BaseOpcode.HasD16, "$d16", ""); 441} 442 443class MIMG_NoSampler_gfx10<mimgopc op, string opcode, 444 RegisterClass DataRC, RegisterClass AddrRC, 445 string dns=""> 446 : MIMG_gfx10<op.GFX10M, (outs DataRC:$vdata), dns> { 447 let InOperandList = !con((ins AddrRC:$vaddr0, SReg_256:$srsrc, DMask:$dmask, 448 Dim:$dim, UNorm:$unorm, CPol:$cpol, 449 R128A16:$r128, A16:$a16, TFE:$tfe, LWE:$lwe), 450 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 451 let AsmString = opcode#" $vdata, $vaddr0, $srsrc$dmask$dim$unorm$cpol$r128$a16$tfe$lwe" 452 #!if(BaseOpcode.HasD16, "$d16", ""); 453} 454 455class MIMG_NoSampler_nsa_gfx10<mimgopc op, string opcode, 456 RegisterClass DataRC, int num_addrs, 457 string dns=""> 458 : MIMG_nsa_gfx10<op.GFX10M, (outs DataRC:$vdata), num_addrs, dns> { 459 let InOperandList = !con(AddrIns, 460 (ins SReg_256:$srsrc, DMask:$dmask, 461 Dim:$dim, UNorm:$unorm, CPol:$cpol, 462 R128A16:$r128, A16:$a16, TFE:$tfe, LWE:$lwe), 463 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 464 let AsmString = opcode#" $vdata, "#AddrAsm#", $srsrc$dmask$dim$unorm$cpol$r128$a16$tfe$lwe" 465 #!if(BaseOpcode.HasD16, "$d16", ""); 466} 467 468class MIMG_NoSampler_gfx11<mimgopc op, string opcode, 469 RegisterClass DataRC, RegisterClass AddrRC, 470 string dns=""> 471 : MIMG_gfx11<op.GFX11, (outs DataRC:$vdata), dns> { 472 let InOperandList = !con((ins AddrRC:$vaddr0, SReg_256:$srsrc, DMask:$dmask, 473 Dim:$dim, UNorm:$unorm, CPol:$cpol, 474 R128A16:$r128, A16:$a16, TFE:$tfe, LWE:$lwe), 475 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 476 let AsmString = opcode#" $vdata, $vaddr0, $srsrc$dmask$dim$unorm$cpol$r128$a16$tfe$lwe" 477 #!if(BaseOpcode.HasD16, "$d16", ""); 478} 479 480class MIMG_NoSampler_nsa_gfx11<mimgopc op, string opcode, 481 RegisterClass DataRC, int num_addrs, 482 string dns=""> 483 : MIMG_nsa_gfx11<op.GFX11, (outs DataRC:$vdata), num_addrs, dns> { 484 let InOperandList = !con(AddrIns, 485 (ins SReg_256:$srsrc, DMask:$dmask, 486 Dim:$dim, UNorm:$unorm, CPol:$cpol, 487 R128A16:$r128, A16:$a16, TFE:$tfe, LWE:$lwe), 488 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 489 let AsmString = opcode#" $vdata, "#AddrAsm#", $srsrc$dmask$dim$unorm$cpol$r128$a16$tfe$lwe" 490 #!if(BaseOpcode.HasD16, "$d16", ""); 491} 492 493class VIMAGE_NoSampler_gfx12<mimgopc op, string opcode, 494 RegisterClass DataRC, int num_addrs, 495 string dns=""> 496 : VIMAGE_gfx12<op.GFX11, (outs DataRC:$vdata), num_addrs, dns> { 497 let InOperandList = !con(AddrIns, 498 (ins SReg_256:$rsrc, DMask:$dmask, Dim:$dim, 499 CPol:$cpol, R128A16:$r128, A16:$a16, TFE:$tfe), 500 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 501 let AsmString = opcode#" $vdata, "#AddrAsm#", $rsrc$dmask$dim$cpol$r128$a16$tfe" 502 #!if(BaseOpcode.HasD16, "$d16", ""); 503} 504 505class VSAMPLE_Sampler_gfx12<mimgopc op, string opcode, RegisterClass DataRC, 506 int num_addrs, RegisterClass Addr3RC = VGPR_32, 507 string dns=""> 508 : VSAMPLE_gfx12<op.GFX12, (outs DataRC:$vdata), num_addrs, dns, Addr3RC> { 509 let InOperandList = !con(AddrIns, 510 (ins SReg_256:$rsrc), 511 !if(BaseOpcode.Sampler, (ins SReg_128:$samp), (ins)), 512 (ins DMask:$dmask, Dim:$dim, UNorm:$unorm, 513 CPol:$cpol, R128A16:$r128, A16:$a16, TFE:$tfe, 514 LWE:$lwe), 515 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 516 let AsmString = opcode#" $vdata, "#AddrAsm#", $rsrc" 517 #!if(BaseOpcode.Sampler, ", $samp", "") 518 #"$dmask$dim$unorm$cpol$r128$a16$tfe$lwe" 519 #!if(BaseOpcode.HasD16, "$d16", ""); 520} 521 522multiclass MIMG_NoSampler_Src_Helper <mimgopc op, string asm, 523 RegisterClass dst_rc, bit enableDisasm, 524 bit ExtendedImageInst = 1, 525 bit isVSample = 0> { 526 let VAddrDwords = 1 in { 527 let ssamp = 0 in { 528 if op.HAS_GFX10M then { 529 def _V1 : MIMG_NoSampler_Helper <op, asm, dst_rc, VGPR_32, 530 !if(enableDisasm, "AMDGPU", "")>; 531 if !not(ExtendedImageInst) then 532 def _V1_gfx90a : MIMG_NoSampler_Helper_gfx90a <op, asm, dst_rc, VGPR_32, 533 !if(enableDisasm, "GFX90A", "")>; 534 def _V1_gfx10 : MIMG_NoSampler_gfx10<op, asm, dst_rc, VGPR_32, 535 !if(enableDisasm, "AMDGPU", "")>; 536 } 537 if op.HAS_GFX11 then { 538 def _V1_gfx11 : MIMG_NoSampler_gfx11<op, asm, dst_rc, VGPR_32, 539 !if(enableDisasm, "AMDGPU", "")>; 540 } 541 } 542 if op.HAS_GFX12 then { 543 if isVSample then { 544 let samp = 0 in 545 def _V1_gfx12 : VSAMPLE_Sampler_gfx12<op, asm, dst_rc, 1>; 546 } 547 else { 548 def _V1_gfx12 : VIMAGE_NoSampler_gfx12<op, asm, dst_rc, 1, 549 !if(enableDisasm, "GFX12", "")>; 550 } 551 } 552 } 553 let VAddrDwords = 2 in { 554 let ssamp = 0 in { 555 if op.HAS_GFX10M then { 556 def _V2 : MIMG_NoSampler_Helper <op, asm, dst_rc, VReg_64>; 557 if !not(ExtendedImageInst) then 558 def _V2_gfx90a : MIMG_NoSampler_Helper_gfx90a <op, asm, dst_rc, VReg_64>; 559 def _V2_gfx10 : MIMG_NoSampler_gfx10<op, asm, dst_rc, VReg_64>; 560 def _V2_nsa_gfx10 : MIMG_NoSampler_nsa_gfx10<op, asm, dst_rc, 2>; 561 } 562 if op.HAS_GFX11 then { 563 def _V2_gfx11 : MIMG_NoSampler_gfx11<op, asm, dst_rc, VReg_64>; 564 def _V2_nsa_gfx11 : MIMG_NoSampler_nsa_gfx11<op, asm, dst_rc, 2>; 565 } 566 } 567 if op.HAS_GFX12 then { 568 if isVSample then { 569 let samp = 0 in 570 def _V2_gfx12 : VSAMPLE_Sampler_gfx12<op, asm, dst_rc, 2>; 571 } 572 else { 573 def _V2_gfx12 : VIMAGE_NoSampler_gfx12<op, asm, dst_rc, 2>; 574 } 575 } 576 } 577 let VAddrDwords = 3 in { 578 let ssamp = 0 in { 579 if op.HAS_GFX10M then { 580 def _V3 : MIMG_NoSampler_Helper <op, asm, dst_rc, VReg_96>; 581 if !not(ExtendedImageInst) then 582 def _V3_gfx90a : MIMG_NoSampler_Helper_gfx90a <op, asm, dst_rc, VReg_96>; 583 def _V3_gfx10 : MIMG_NoSampler_gfx10<op, asm, dst_rc, VReg_96>; 584 def _V3_nsa_gfx10 : MIMG_NoSampler_nsa_gfx10<op, asm, dst_rc, 3>; 585 } 586 if op.HAS_GFX11 then { 587 def _V3_gfx11 : MIMG_NoSampler_gfx11<op, asm, dst_rc, VReg_96>; 588 def _V3_nsa_gfx11 : MIMG_NoSampler_nsa_gfx11<op, asm, dst_rc, 3>; 589 } 590 } 591 if op.HAS_GFX12 then { 592 if isVSample then { 593 let samp = 0 in 594 def _V3_gfx12 : VSAMPLE_Sampler_gfx12<op, asm, dst_rc, 3>; 595 } 596 else { 597 def _V3_gfx12 : VIMAGE_NoSampler_gfx12<op, asm, dst_rc, 3>; 598 } 599 } 600 } 601 let VAddrDwords = 4 in { 602 let ssamp = 0 in { 603 if op.HAS_GFX10M then { 604 def _V4 : MIMG_NoSampler_Helper <op, asm, dst_rc, VReg_128>; 605 if !not(ExtendedImageInst) then 606 def _V4_gfx90a : MIMG_NoSampler_Helper_gfx90a <op, asm, dst_rc, VReg_128>; 607 def _V4_gfx10 : MIMG_NoSampler_gfx10<op, asm, dst_rc, VReg_128>; 608 def _V4_nsa_gfx10 : MIMG_NoSampler_nsa_gfx10<op, asm, dst_rc, 4, 609 !if(enableDisasm, "AMDGPU", "")>; 610 } 611 if op.HAS_GFX11 then { 612 def _V4_gfx11 : MIMG_NoSampler_gfx11<op, asm, dst_rc, VReg_128>; 613 def _V4_nsa_gfx11 : MIMG_NoSampler_nsa_gfx11<op, asm, dst_rc, 4, 614 !if(enableDisasm, "AMDGPU", "")>; 615 } 616 } 617 if op.HAS_GFX12 then { 618 if isVSample then { 619 let samp = 0 in 620 def _V4_gfx12 : VSAMPLE_Sampler_gfx12<op, asm, dst_rc, 4, VGPR_32, 621 !if(enableDisasm, "GFX12", "")>; 622 } 623 else { 624 def _V4_gfx12 : VIMAGE_NoSampler_gfx12<op, asm, dst_rc, 4, 625 !if(enableDisasm, "GFX12", "")>; 626 } 627 } 628 } 629} 630 631multiclass MIMG_NoSampler <mimgopc op, string asm, bit has_d16, bit mip = 0, 632 bit isResInfo = 0, 633 bit msaa = 0> { 634 def "" : MIMGBaseOpcode { 635 let Coordinates = !not(isResInfo); 636 let LodOrClampOrMip = mip; 637 let HasD16 = has_d16; 638 let MSAA = msaa; 639 } 640 641 let BaseOpcode = !cast<MIMGBaseOpcode>(NAME), 642 mayLoad = !not(isResInfo) in { 643 let VDataDwords = 1 in 644 defm _V1 : MIMG_NoSampler_Src_Helper <op, asm, VGPR_32, 1, msaa>; 645 let VDataDwords = 2 in 646 defm _V2 : MIMG_NoSampler_Src_Helper <op, asm, VReg_64, 0, msaa>; 647 let VDataDwords = 3 in 648 defm _V3 : MIMG_NoSampler_Src_Helper <op, asm, VReg_96, 0, msaa>; 649 let VDataDwords = 4 in 650 defm _V4 : MIMG_NoSampler_Src_Helper <op, asm, VReg_128, 0, msaa>; 651 let VDataDwords = 5 in 652 defm _V5 : MIMG_NoSampler_Src_Helper <op, asm, VReg_160, 0, msaa>; 653 } 654} 655 656class MIMG_Store_Helper <mimgopc op, string asm, 657 RegisterClass data_rc, 658 RegisterClass addr_rc, 659 string dns = ""> 660 : MIMG_gfx6789<op.GFX10M, (outs), dns> { 661 let InOperandList = !con((ins data_rc:$vdata, addr_rc:$vaddr, SReg_256:$srsrc, 662 DMask:$dmask, UNorm:$unorm, CPol:$cpol, 663 R128A16:$r128, TFE:$tfe, LWE:$lwe, DA:$da), 664 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 665 let AsmString = asm#" $vdata, $vaddr, $srsrc$dmask$unorm$cpol$r128$tfe$lwe$da" 666 #!if(BaseOpcode.HasD16, "$d16", ""); 667} 668 669class MIMG_Store_Helper_gfx90a <mimgopc op, string asm, 670 RegisterClass data_rc, 671 RegisterClass addr_rc, 672 string dns = ""> 673 : MIMG_gfx90a<op.GFX10M, (outs), dns> { 674 let InOperandList = !con((ins getLdStRegisterOperand<data_rc>.ret:$vdata, 675 addr_rc:$vaddr, SReg_256:$srsrc, 676 DMask:$dmask, UNorm:$unorm, CPol:$cpol, 677 R128A16:$r128, LWE:$lwe, DA:$da), 678 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 679 let AsmString = asm#" $vdata, $vaddr, $srsrc$dmask$unorm$cpol$r128$lwe$da" 680 #!if(BaseOpcode.HasD16, "$d16", ""); 681} 682 683class MIMG_Store_gfx10<mimgopc op, string opcode, 684 RegisterClass DataRC, RegisterClass AddrRC, 685 string dns=""> 686 : MIMG_gfx10<op.GFX10M, (outs), dns> { 687 let InOperandList = !con((ins DataRC:$vdata, AddrRC:$vaddr0, SReg_256:$srsrc, 688 DMask:$dmask, Dim:$dim, UNorm:$unorm, CPol:$cpol, 689 R128A16:$r128, A16:$a16, TFE:$tfe, LWE:$lwe), 690 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 691 let AsmString = opcode#" $vdata, $vaddr0, $srsrc$dmask$dim$unorm$cpol$r128$a16$tfe$lwe" 692 #!if(BaseOpcode.HasD16, "$d16", ""); 693} 694 695class MIMG_Store_nsa_gfx10<mimgopc op, string opcode, 696 RegisterClass DataRC, int num_addrs, 697 string dns=""> 698 : MIMG_nsa_gfx10<op.GFX10M, (outs), num_addrs, dns> { 699 let InOperandList = !con((ins DataRC:$vdata), 700 AddrIns, 701 (ins SReg_256:$srsrc, DMask:$dmask, 702 Dim:$dim, UNorm:$unorm, CPol:$cpol, 703 R128A16:$r128, A16:$a16, TFE:$tfe, LWE:$lwe), 704 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 705 let AsmString = opcode#" $vdata, "#AddrAsm#", $srsrc$dmask$dim$unorm$cpol$r128$a16$tfe$lwe" 706 #!if(BaseOpcode.HasD16, "$d16", ""); 707} 708 709class MIMG_Store_gfx11<mimgopc op, string opcode, 710 RegisterClass DataRC, RegisterClass AddrRC, 711 string dns=""> 712 : MIMG_gfx11<op.GFX11, (outs), dns> { 713 let InOperandList = !con((ins DataRC:$vdata, AddrRC:$vaddr0, SReg_256:$srsrc, 714 DMask:$dmask, Dim:$dim, UNorm:$unorm, CPol:$cpol, 715 R128A16:$r128, A16:$a16, TFE:$tfe, LWE:$lwe), 716 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 717 let AsmString = opcode#" $vdata, $vaddr0, $srsrc$dmask$dim$unorm$cpol$r128$a16$tfe$lwe" 718 #!if(BaseOpcode.HasD16, "$d16", ""); 719} 720 721class MIMG_Store_nsa_gfx11<mimgopc op, string opcode, 722 RegisterClass DataRC, int num_addrs, 723 string dns=""> 724 : MIMG_nsa_gfx11<op.GFX11, (outs), num_addrs, dns> { 725 let InOperandList = !con((ins DataRC:$vdata), 726 AddrIns, 727 (ins SReg_256:$srsrc, DMask:$dmask, 728 Dim:$dim, UNorm:$unorm, CPol:$cpol, 729 R128A16:$r128, A16:$a16, TFE:$tfe, LWE:$lwe), 730 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 731 let AsmString = opcode#" $vdata, "#AddrAsm#", $srsrc$dmask$dim$unorm$cpol$r128$a16$tfe$lwe" 732 #!if(BaseOpcode.HasD16, "$d16", ""); 733} 734 735class VIMAGE_Store_gfx12<mimgopc op, string opcode, 736 RegisterClass DataRC, int num_addrs, 737 string dns=""> 738 : VIMAGE_gfx12<op.GFX12, (outs), num_addrs, dns> { 739 let InOperandList = !con((ins DataRC:$vdata), 740 AddrIns, 741 (ins SReg_256:$rsrc, DMask:$dmask, Dim:$dim, 742 CPol:$cpol, R128A16:$r128, A16:$a16, TFE:$tfe), 743 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 744 let AsmString = opcode#" $vdata, "#AddrAsm#", $rsrc$dmask$dim$cpol$r128$a16$tfe" 745 #!if(BaseOpcode.HasD16, "$d16", ""); 746} 747 748multiclass MIMG_Store_Addr_Helper <mimgopc op, string asm, 749 RegisterClass data_rc, 750 bit enableDisasm> { 751 let mayLoad = 0, mayStore = 1, hasSideEffects = 0, hasPostISelHook = 0, 752 DisableWQM = 1 in { 753 let VAddrDwords = 1 in { 754 let ssamp = 0 in { 755 if op.HAS_GFX10M then { 756 def _V1 : MIMG_Store_Helper <op, asm, data_rc, VGPR_32, 757 !if(enableDisasm, "AMDGPU", "")>; 758 let hasPostISelHook = 1 in 759 def _V1_gfx90a : MIMG_Store_Helper_gfx90a <op, asm, data_rc, VGPR_32, 760 !if(enableDisasm, "GFX90A", "")>; 761 def _V1_gfx10 : MIMG_Store_gfx10 <op, asm, data_rc, VGPR_32, 762 !if(enableDisasm, "AMDGPU", "")>; 763 } 764 if op.HAS_GFX11 then { 765 def _V1_gfx11 : MIMG_Store_gfx11 <op, asm, data_rc, VGPR_32, 766 !if(enableDisasm, "AMDGPU", "")>; 767 } 768 } 769 if op.HAS_GFX12 then { 770 def _V1_gfx12 : VIMAGE_Store_gfx12 <op, asm, data_rc, 1, 771 !if(enableDisasm, "GFX12", "")>; 772 } 773 } 774 let VAddrDwords = 2 in { 775 let ssamp = 0 in { 776 if op.HAS_GFX10M then { 777 def _V2 : MIMG_Store_Helper <op, asm, data_rc, VReg_64>; 778 def _V2_gfx90a : MIMG_Store_Helper_gfx90a <op, asm, data_rc, VReg_64>; 779 def _V2_gfx10 : MIMG_Store_gfx10 <op, asm, data_rc, VReg_64>; 780 def _V2_nsa_gfx10 : MIMG_Store_nsa_gfx10 <op, asm, data_rc, 2>; 781 } 782 if op.HAS_GFX11 then { 783 def _V2_gfx11 : MIMG_Store_gfx11 <op, asm, data_rc, VReg_64>; 784 def _V2_nsa_gfx11 : MIMG_Store_nsa_gfx11 <op, asm, data_rc, 2>; 785 } 786 } 787 if op.HAS_GFX12 then { 788 def _V2_gfx12 : VIMAGE_Store_gfx12 <op, asm, data_rc, 2>; 789 } 790 } 791 let VAddrDwords = 3 in { 792 let ssamp = 0 in { 793 if op.HAS_GFX10M then { 794 def _V3 : MIMG_Store_Helper <op, asm, data_rc, VReg_96>; 795 def _V3_gfx90a : MIMG_Store_Helper_gfx90a <op, asm, data_rc, VReg_96>; 796 def _V3_gfx10 : MIMG_Store_gfx10 <op, asm, data_rc, VReg_96>; 797 def _V3_nsa_gfx10 : MIMG_Store_nsa_gfx10 <op, asm, data_rc, 3>; 798 } 799 if op.HAS_GFX11 then { 800 def _V3_gfx11 : MIMG_Store_gfx11 <op, asm, data_rc, VReg_96>; 801 def _V3_nsa_gfx11 : MIMG_Store_nsa_gfx11 <op, asm, data_rc, 3>; 802 } 803 } 804 if op.HAS_GFX12 then { 805 def _V3_gfx12 : VIMAGE_Store_gfx12 <op, asm, data_rc, 3>; 806 } 807 } 808 let VAddrDwords = 4 in { 809 let ssamp = 0 in { 810 if op.HAS_GFX10M then { 811 def _V4 : MIMG_Store_Helper <op, asm, data_rc, VReg_128>; 812 def _V4_gfx90a : MIMG_Store_Helper_gfx90a <op, asm, data_rc, VReg_128>; 813 def _V4_gfx10 : MIMG_Store_gfx10 <op, asm, data_rc, VReg_128>; 814 def _V4_nsa_gfx10 : MIMG_Store_nsa_gfx10 <op, asm, data_rc, 4, 815 !if(enableDisasm, "AMDGPU", "")>; 816 } 817 if op.HAS_GFX11 then { 818 def _V4_gfx11 : MIMG_Store_gfx11 <op, asm, data_rc, VReg_128>; 819 def _V4_nsa_gfx11 : MIMG_Store_nsa_gfx11 <op, asm, data_rc, 4, 820 !if(enableDisasm, "AMDGPU", "")>; 821 } 822 } 823 if op.HAS_GFX12 then { 824 def _V4_gfx12 : VIMAGE_Store_gfx12 <op, asm, data_rc, 4, 825 !if(enableDisasm, "GFX12", "")>; 826 } 827 } 828 } 829} 830 831multiclass MIMG_Store <mimgopc op, string asm, bit has_d16, bit mip = 0> { 832 def "" : MIMGBaseOpcode { 833 let Store = 1; 834 let LodOrClampOrMip = mip; 835 let HasD16 = has_d16; 836 } 837 838 let BaseOpcode = !cast<MIMGBaseOpcode>(NAME) in { 839 let VDataDwords = 1 in 840 defm _V1 : MIMG_Store_Addr_Helper <op, asm, VGPR_32, 1>; 841 let VDataDwords = 2 in 842 defm _V2 : MIMG_Store_Addr_Helper <op, asm, VReg_64, 0>; 843 let VDataDwords = 3 in 844 defm _V3 : MIMG_Store_Addr_Helper <op, asm, VReg_96, 0>; 845 let VDataDwords = 4 in 846 defm _V4 : MIMG_Store_Addr_Helper <op, asm, VReg_128, 0>; 847 let VDataDwords = 5 in 848 defm _V5 : MIMG_Store_Addr_Helper <op, asm, VReg_160, 0>; 849 } 850} 851 852class MIMG_Atomic_gfx6789_base <bits<8> op, string asm, RegisterClass data_rc, 853 RegisterClass addr_rc, string dns=""> 854 : MIMG_gfx6789 <op, (outs data_rc:$vdst), dns> { 855 let Constraints = "$vdst = $vdata"; 856 857 let InOperandList = (ins data_rc:$vdata, addr_rc:$vaddr, SReg_256:$srsrc, 858 DMask:$dmask, UNorm:$unorm, CPol:$cpol, 859 R128A16:$r128, TFE:$tfe, LWE:$lwe, DA:$da); 860 let AsmString = asm#" $vdst, $vaddr, $srsrc$dmask$unorm$cpol$r128$tfe$lwe$da"; 861} 862 863class MIMG_Atomic_gfx90a_base <bits<8> op, string asm, RegisterClass data_rc, 864 RegisterClass addr_rc, string dns=""> 865 : MIMG_gfx90a <op, (outs getLdStRegisterOperand<data_rc>.ret:$vdst), dns> { 866 let Constraints = "$vdst = $vdata"; 867 868 let InOperandList = (ins getLdStRegisterOperand<data_rc>.ret:$vdata, 869 addr_rc:$vaddr, SReg_256:$srsrc, 870 DMask:$dmask, UNorm:$unorm, CPol:$cpol, 871 R128A16:$r128, LWE:$lwe, DA:$da); 872 let AsmString = asm#" $vdst, $vaddr, $srsrc$dmask$unorm$cpol$r128$lwe$da"; 873} 874 875class MIMG_Atomic_si<mimgopc op, string asm, RegisterClass data_rc, 876 RegisterClass addr_rc, bit enableDasm = 0> 877 : MIMG_Atomic_gfx6789_base<op.SI, asm, data_rc, addr_rc, 878 !if(enableDasm, "GFX6GFX7", "")> { 879 let AssemblerPredicate = isGFX6GFX7; 880} 881 882class MIMG_Atomic_vi<mimgopc op, string asm, RegisterClass data_rc, 883 RegisterClass addr_rc, bit enableDasm = 0> 884 : MIMG_Atomic_gfx6789_base<op.VI, asm, data_rc, addr_rc, !if(enableDasm, "GFX8", "")> { 885 let AssemblerPredicate = isGFX8GFX9NotGFX90A; 886 let MIMGEncoding = MIMGEncGfx8; 887} 888 889class MIMG_Atomic_gfx90a<mimgopc op, string asm, RegisterClass data_rc, 890 RegisterClass addr_rc, bit enableDasm = 0> 891 : MIMG_Atomic_gfx90a_base<op.VI, asm, data_rc, addr_rc, !if(enableDasm, "GFX90A", "")> { 892 let AssemblerPredicate = isGFX90APlus; 893 let MIMGEncoding = MIMGEncGfx90a; 894} 895 896class MIMG_Atomic_gfx10<mimgopc op, string opcode, 897 RegisterClass DataRC, RegisterClass AddrRC, 898 bit enableDisasm = 0> 899 : MIMG_gfx10<!cast<int>(op.GFX10M), (outs DataRC:$vdst), 900 !if(enableDisasm, "AMDGPU", "")> { 901 let Constraints = "$vdst = $vdata"; 902 903 let InOperandList = (ins DataRC:$vdata, AddrRC:$vaddr0, SReg_256:$srsrc, 904 DMask:$dmask, Dim:$dim, UNorm:$unorm, CPol:$cpol, 905 R128A16:$r128, A16:$a16, TFE:$tfe, LWE:$lwe); 906 let AsmString = opcode#" $vdst, $vaddr0, $srsrc$dmask$dim$unorm$cpol$r128$a16$tfe$lwe"; 907} 908 909class MIMG_Atomic_nsa_gfx10<mimgopc op, string opcode, 910 RegisterClass DataRC, int num_addrs, 911 bit enableDisasm = 0> 912 : MIMG_nsa_gfx10<!cast<int>(op.GFX10M), (outs DataRC:$vdst), num_addrs, 913 !if(enableDisasm, "AMDGPU", "")> { 914 let Constraints = "$vdst = $vdata"; 915 916 let InOperandList = !con((ins DataRC:$vdata), 917 AddrIns, 918 (ins SReg_256:$srsrc, DMask:$dmask, 919 Dim:$dim, UNorm:$unorm, CPol:$cpol, 920 R128A16:$r128, A16:$a16, TFE:$tfe, LWE:$lwe)); 921 let AsmString = opcode#" $vdata, "#AddrAsm#", $srsrc$dmask$dim$unorm$cpol$r128$a16$tfe$lwe"; 922} 923 924class MIMG_Atomic_gfx11<mimgopc op, string opcode, 925 RegisterClass DataRC, RegisterClass AddrRC, 926 bit enableDisasm = 0> 927 : MIMG_gfx11<!cast<int>(op.GFX11), (outs DataRC:$vdst), 928 !if(enableDisasm, "AMDGPU", "")> { 929 let Constraints = "$vdst = $vdata"; 930 931 let InOperandList = (ins DataRC:$vdata, AddrRC:$vaddr0, SReg_256:$srsrc, 932 DMask:$dmask, Dim:$dim, UNorm:$unorm, CPol:$cpol, 933 R128A16:$r128, A16:$a16, TFE:$tfe, LWE:$lwe); 934 let AsmString = opcode#" $vdst, $vaddr0, $srsrc$dmask$dim$unorm$cpol$r128$a16$tfe$lwe"; 935} 936 937class MIMG_Atomic_nsa_gfx11<mimgopc op, string opcode, 938 RegisterClass DataRC, int num_addrs, 939 bit enableDisasm = 0> 940 : MIMG_nsa_gfx11<!cast<int>(op.GFX11), (outs DataRC:$vdst), num_addrs, 941 !if(enableDisasm, "AMDGPU", "")> { 942 let Constraints = "$vdst = $vdata"; 943 944 let InOperandList = !con((ins DataRC:$vdata), 945 AddrIns, 946 (ins SReg_256:$srsrc, DMask:$dmask, 947 Dim:$dim, UNorm:$unorm, CPol:$cpol, 948 R128A16:$r128, A16:$a16, TFE:$tfe, LWE:$lwe)); 949 let AsmString = opcode#" $vdata, "#AddrAsm#", $srsrc$dmask$dim$unorm$cpol$r128$a16$tfe$lwe"; 950} 951 952class VIMAGE_Atomic_gfx12<mimgopc op, string opcode, RegisterClass DataRC, 953 int num_addrs, bit enableDisasm = 0> 954 : VIMAGE_gfx12<!cast<int>(op.GFX12), (outs DataRC:$vdst), num_addrs, 955 !if(enableDisasm, "GFX12", "")> { 956 let Constraints = "$vdst = $vdata"; 957 958 let InOperandList = !con((ins DataRC:$vdata), 959 AddrIns, 960 (ins SReg_256:$rsrc, DMask:$dmask, Dim:$dim, 961 CPol:$cpol, R128A16:$r128, A16:$a16, TFE:$tfe)); 962 let AsmString = opcode#" $vdata, "#AddrAsm#", $rsrc$dmask$dim$cpol$r128$a16$tfe"; 963} 964 965class VIMAGE_Atomic_gfx12_Renamed<mimgopc op, string opcode, string renamed, 966 RegisterClass DataRC, int num_addrs, 967 bit enableDisasm = 0> 968 : VIMAGE_Atomic_gfx12<op, renamed, DataRC, num_addrs, enableDisasm>, 969 MnemonicAlias<opcode, renamed>, Requires<[isGFX12Plus]>; 970 971multiclass MIMG_Atomic_Addr_Helper_m <mimgopc op, string asm, 972 RegisterClass data_rc, 973 bit enableDasm = 0, 974 bit isFP = 0, 975 string renamed = ""> { 976 let hasSideEffects = 1, // FIXME: remove this 977 mayLoad = 1, mayStore = 1, hasPostISelHook = 0, DisableWQM = 1, 978 FPAtomic = isFP in { 979 let VAddrDwords = 1 in { 980 let ssamp = 0 in { 981 if op.HAS_SI then { 982 def _V1_si : MIMG_Atomic_si <op, asm, data_rc, VGPR_32, enableDasm>; 983 } 984 if op.HAS_VI then { 985 def _V1_vi : MIMG_Atomic_vi <op, asm, data_rc, VGPR_32, enableDasm>; 986 let hasPostISelHook = 1 in 987 def _V1_gfx90a : MIMG_Atomic_gfx90a <op, asm, data_rc, VGPR_32, enableDasm>; 988 } 989 if op.HAS_GFX10M then { 990 def _V1_gfx10 : MIMG_Atomic_gfx10 <op, asm, data_rc, VGPR_32, enableDasm>; 991 } 992 if op.HAS_GFX11 then { 993 def _V1_gfx11 : MIMG_Atomic_gfx11 <op, asm, data_rc, VGPR_32, enableDasm>; 994 } 995 } 996 if op.HAS_GFX12 then { 997 if !empty(renamed) then 998 def _V1_gfx12 : VIMAGE_Atomic_gfx12 <op, asm, data_rc, 1, enableDasm>; 999 else 1000 def _V1_gfx12 : VIMAGE_Atomic_gfx12_Renamed <op, asm, renamed, data_rc, 1, enableDasm>; 1001 } 1002 } 1003 let VAddrDwords = 2 in { 1004 let ssamp = 0 in { 1005 if op.HAS_SI then { 1006 def _V2_si : MIMG_Atomic_si <op, asm, data_rc, VReg_64, 0>; 1007 } 1008 if op.HAS_VI then { 1009 def _V2_vi : MIMG_Atomic_vi <op, asm, data_rc, VReg_64, 0>; 1010 def _V2_gfx90a : MIMG_Atomic_gfx90a <op, asm, data_rc, VReg_64, 0>; 1011 } 1012 if op.HAS_GFX10M then { 1013 def _V2_gfx10 : MIMG_Atomic_gfx10 <op, asm, data_rc, VReg_64, 0>; 1014 def _V2_nsa_gfx10 : MIMG_Atomic_nsa_gfx10 <op, asm, data_rc, 2, 0>; 1015 } 1016 if op.HAS_GFX11 then { 1017 def _V2_gfx11 : MIMG_Atomic_gfx11 <op, asm, data_rc, VReg_64, 0>; 1018 def _V2_nsa_gfx11 : MIMG_Atomic_nsa_gfx11 <op, asm, data_rc, 2, 0>; 1019 } 1020 } 1021 if op.HAS_GFX12 then { 1022 if !empty(renamed) then 1023 def _V2_gfx12 : VIMAGE_Atomic_gfx12 <op, asm, data_rc, 2, 0>; 1024 else 1025 def _V2_gfx12 : VIMAGE_Atomic_gfx12_Renamed <op, asm, renamed, data_rc, 2, 0>; 1026 } 1027 } 1028 let VAddrDwords = 3 in { 1029 let ssamp = 0 in { 1030 if op.HAS_SI then { 1031 def _V3_si : MIMG_Atomic_si <op, asm, data_rc, VReg_96, 0>; 1032 } 1033 if op.HAS_VI then { 1034 def _V3_vi : MIMG_Atomic_vi <op, asm, data_rc, VReg_96, 0>; 1035 def _V3_gfx90a : MIMG_Atomic_gfx90a <op, asm, data_rc, VReg_96, 0>; 1036 } 1037 if op.HAS_GFX10M then { 1038 def _V3_gfx10 : MIMG_Atomic_gfx10 <op, asm, data_rc, VReg_96, 0>; 1039 def _V3_nsa_gfx10 : MIMG_Atomic_nsa_gfx10 <op, asm, data_rc, 3, 0>; 1040 } 1041 if op.HAS_GFX11 then { 1042 def _V3_gfx11 : MIMG_Atomic_gfx11 <op, asm, data_rc, VReg_96, 0>; 1043 def _V3_nsa_gfx11 : MIMG_Atomic_nsa_gfx11 <op, asm, data_rc, 3, 0>; 1044 } 1045 } 1046 if op.HAS_GFX12 then { 1047 if !empty(renamed) then 1048 def _V3_gfx12 : VIMAGE_Atomic_gfx12 <op, asm, data_rc, 3, 0>; 1049 else 1050 def _V3_gfx12 : VIMAGE_Atomic_gfx12_Renamed <op, asm, renamed, data_rc, 3, 0>; 1051 } 1052 } 1053 let VAddrDwords = 4 in { 1054 let ssamp = 0 in { 1055 if op.HAS_SI then { 1056 def _V4_si : MIMG_Atomic_si <op, asm, data_rc, VReg_128, 0>; 1057 } 1058 if op.HAS_VI then { 1059 def _V4_vi : MIMG_Atomic_vi <op, asm, data_rc, VReg_128, 0>; 1060 def _V4_gfx90a : MIMG_Atomic_gfx90a <op, asm, data_rc, VReg_128, 0>; 1061 } 1062 if op.HAS_GFX10M then { 1063 def _V4_gfx10 : MIMG_Atomic_gfx10 <op, asm, data_rc, VReg_128, 0>; 1064 def _V4_nsa_gfx10 : MIMG_Atomic_nsa_gfx10 <op, asm, data_rc, 4, enableDasm>; 1065 } 1066 if op.HAS_GFX11 then { 1067 def _V4_gfx11 : MIMG_Atomic_gfx11 <op, asm, data_rc, VReg_128, 0>; 1068 def _V4_nsa_gfx11 : MIMG_Atomic_nsa_gfx11 <op, asm, data_rc, 4, enableDasm>; 1069 } 1070 } 1071 if op.HAS_GFX12 then { 1072 if !empty(renamed) then 1073 def _V4_gfx12 : VIMAGE_Atomic_gfx12 <op, asm, data_rc, 4, enableDasm>; 1074 else 1075 def _V4_gfx12 : VIMAGE_Atomic_gfx12_Renamed <op, asm, renamed, data_rc, 4, enableDasm>; 1076 } 1077 } 1078 } 1079} 1080 1081multiclass MIMG_Atomic <mimgopc op, string asm, bit isCmpSwap = 0, bit isFP = 0, 1082 string renamed = ""> { // 64-bit atomics 1083 let IsAtomicRet = 1 in { 1084 def "" : MIMGBaseOpcode { 1085 let Atomic = 1; 1086 let AtomicX2 = isCmpSwap; 1087 } 1088 1089 let BaseOpcode = !cast<MIMGBaseOpcode>(NAME) in { 1090 // _V* variants have different dst size, but the size is encoded implicitly, 1091 // using dmask and tfe. Only 32-bit variant is registered with disassembler. 1092 // Other variants are reconstructed by disassembler using dmask and tfe. 1093 let VDataDwords = !if(isCmpSwap, 2, 1) in 1094 defm _V1 : MIMG_Atomic_Addr_Helper_m <op, asm, !if(isCmpSwap, VReg_64, VGPR_32), 1, isFP, renamed>; 1095 let VDataDwords = !if(isCmpSwap, 4, 2) in 1096 defm _V2 : MIMG_Atomic_Addr_Helper_m <op, asm, !if(isCmpSwap, VReg_128, VReg_64), 0, isFP, renamed>; 1097 } 1098 } // End IsAtomicRet = 1 1099} 1100 1101multiclass MIMG_Atomic_Renamed <mimgopc op, string asm, string renamed, 1102 bit isCmpSwap = 0, bit isFP = 0> 1103 : MIMG_Atomic <op, asm, isCmpSwap, isFP, renamed>; 1104 1105class MIMG_Sampler_Helper <mimgopc op, string asm, RegisterClass dst_rc, 1106 RegisterClass src_rc, string dns=""> 1107 : MIMG_gfx6789 <op.VI, (outs dst_rc:$vdata), dns> { 1108 let InOperandList = !con((ins src_rc:$vaddr, SReg_256:$srsrc, SReg_128:$ssamp, 1109 DMask:$dmask, UNorm:$unorm, CPol:$cpol, 1110 R128A16:$r128, TFE:$tfe, LWE:$lwe, DA:$da), 1111 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 1112 let AsmString = asm#" $vdata, $vaddr, $srsrc, $ssamp$dmask$unorm$cpol$r128$tfe$lwe$da" 1113 #!if(BaseOpcode.HasD16, "$d16", ""); 1114} 1115 1116class MIMG_Sampler_gfx90a<mimgopc op, string asm, RegisterClass dst_rc, 1117 RegisterClass src_rc, string dns=""> 1118 : MIMG_gfx90a<op.GFX10M, (outs getLdStRegisterOperand<dst_rc>.ret:$vdata), dns> { 1119 let InOperandList = !con((ins src_rc:$vaddr, SReg_256:$srsrc, SReg_128:$ssamp, 1120 DMask:$dmask, UNorm:$unorm, CPol:$cpol, 1121 R128A16:$r128, LWE:$lwe, DA:$da), 1122 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 1123 let AsmString = asm#" $vdata, $vaddr, $srsrc, $ssamp$dmask$unorm$cpol$r128$lwe$da" 1124 #!if(BaseOpcode.HasD16, "$d16", ""); 1125} 1126 1127class MIMG_Sampler_gfx10<mimgopc op, string opcode, 1128 RegisterClass DataRC, RegisterClass AddrRC, 1129 string dns=""> 1130 : MIMG_gfx10<op.GFX10M, (outs DataRC:$vdata), dns> { 1131 let InOperandList = !con((ins AddrRC:$vaddr0, SReg_256:$srsrc, SReg_128:$ssamp, 1132 DMask:$dmask, Dim:$dim, UNorm:$unorm, CPol:$cpol, 1133 R128A16:$r128, A16:$a16, TFE:$tfe, LWE:$lwe), 1134 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 1135 let AsmString = opcode#" $vdata, $vaddr0, $srsrc, $ssamp$dmask$dim$unorm" 1136 #"$cpol$r128$a16$tfe$lwe" 1137 #!if(BaseOpcode.HasD16, "$d16", ""); 1138} 1139 1140class MIMG_Sampler_nsa_gfx10<mimgopc op, string opcode, 1141 RegisterClass DataRC, int num_addrs, 1142 string dns=""> 1143 : MIMG_nsa_gfx10<op.GFX10M, (outs DataRC:$vdata), num_addrs, dns> { 1144 let InOperandList = !con(AddrIns, 1145 (ins SReg_256:$srsrc, SReg_128:$ssamp, DMask:$dmask, 1146 Dim:$dim, UNorm:$unorm, CPol:$cpol, 1147 R128A16:$r128, A16:$a16, TFE:$tfe, LWE:$lwe), 1148 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 1149 let AsmString = opcode#" $vdata, "#AddrAsm#", $srsrc, $ssamp$dmask$dim$unorm" 1150 #"$cpol$r128$a16$tfe$lwe" 1151 #!if(BaseOpcode.HasD16, "$d16", ""); 1152} 1153 1154class MIMG_Sampler_gfx11<mimgopc op, string opcode, 1155 RegisterClass DataRC, RegisterClass AddrRC, 1156 string dns=""> 1157 : MIMG_gfx11<op.GFX11, (outs DataRC:$vdata), dns> { 1158 let InOperandList = !con((ins AddrRC:$vaddr0, SReg_256:$srsrc, SReg_128:$ssamp, 1159 DMask:$dmask, Dim:$dim, UNorm:$unorm, CPol:$cpol, 1160 R128A16:$r128, A16:$a16, TFE:$tfe, LWE:$lwe), 1161 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 1162 let AsmString = opcode#" $vdata, $vaddr0, $srsrc, $ssamp$dmask$dim$unorm" 1163 #"$cpol$r128$a16$tfe$lwe" 1164 #!if(BaseOpcode.HasD16, "$d16", ""); 1165} 1166 1167class MIMG_Sampler_nsa_gfx11<mimgopc op, string opcode, 1168 RegisterClass DataRC, int num_addrs, 1169 RegisterClass LastVAddrSize, string dns=""> 1170 : MIMG_nsa_gfx11<op.GFX11, (outs DataRC:$vdata), num_addrs, dns, [], 1171 LastVAddrSize> { 1172 let InOperandList = !con(AddrIns, 1173 (ins SReg_256:$srsrc, SReg_128:$ssamp, DMask:$dmask, 1174 Dim:$dim, UNorm:$unorm, CPol:$cpol, 1175 R128A16:$r128, A16:$a16, TFE:$tfe, LWE:$lwe), 1176 !if(BaseOpcode.HasD16, (ins D16:$d16), (ins))); 1177 let AsmString = opcode#" $vdata, "#AddrAsm#", $srsrc, $ssamp$dmask$dim$unorm" 1178 #"$cpol$r128$a16$tfe$lwe" 1179 #!if(BaseOpcode.HasD16, "$d16", ""); 1180} 1181 1182class MIMGAddrSize<int dw, bit enable_disasm, int AddrDW = dw> { 1183 int NumWords = dw; 1184 1185 RegisterClass RegClass = !if(!le(AddrDW, 0), ?, 1186 !if(!eq(AddrDW, 1), VGPR_32, 1187 !if(!eq(AddrDW, 2), VReg_64, 1188 !if(!eq(AddrDW, 3), VReg_96, 1189 !if(!eq(AddrDW, 4), VReg_128, 1190 !if(!eq(AddrDW, 5), VReg_160, 1191 !if(!eq(AddrDW, 6), VReg_192, 1192 !if(!eq(AddrDW, 7), VReg_224, 1193 !if(!eq(AddrDW, 8), VReg_256, 1194 !if(!eq(AddrDW, 9), VReg_288, 1195 !if(!eq(AddrDW, 10), VReg_320, 1196 !if(!eq(AddrDW, 11), VReg_352, 1197 !if(!eq(AddrDW, 12), VReg_384, 1198 !if(!le(AddrDW, 16), VReg_512, ?)))))))))))))); 1199 1200 // Whether the instruction variant with this vaddr size should be enabled for 1201 // the auto-generated disassembler. 1202 bit Disassemble = enable_disasm; 1203} 1204 1205// Returns the MIMGAddrSize with the size of last VAddr for partial NSA 1206class LastVAddrSize <int dw, int max_idx, bit enable_disasm> 1207 : MIMGAddrSize<dw, enable_disasm, 1208 !if(!gt(dw, max_idx), !sub(dw, max_idx), 0)>; 1209 1210// Return whether x is in lst. 1211class isIntInList<int x, list<int> lst> { 1212 bit ret = !foldl(0, lst, lhs, y, !or(lhs, !eq(x, y))); 1213} 1214 1215// Return whether a value inside the range [min, max] (endpoints inclusive) 1216// is in the given list. 1217class isRangeInList<int min, int max, list<int> lst> { 1218 bit ret = !foldl(0, lst, lhs, y, !or(lhs, !and(!le(min, y), !le(y, max)))); 1219} 1220 1221class MIMGAddrSizes_dw_range<list<int> range> { 1222 int Min = !head(range); 1223 int Max = !if(!empty(!tail(range)), Min, !head(!tail(range))); 1224} 1225 1226class MIMG_Sampler_AddrSizes<AMDGPUSampleVariant sample, bit isG16, 1227 int nsa_max_addr = 5, bit includeNSA1 = 0> { 1228 // List of all possible numbers of address words, taking all combinations of 1229 // A16 and image dimension into account (note: no MSAA, since this is for 1230 // sample/gather ops). 1231 list<int> AllNumAddrWords = 1232 !foreach(dw, !if(sample.Gradients, 1233 !if(isG16, 1234 !if(!eq(sample.LodOrClamp, ""), 1235 [2, 3, 4, 5, 6, 7], 1236 [2, 3, 4, 5, 6, 7, 8]), 1237 !if(!eq(sample.LodOrClamp, ""), 1238 [2, 3, 4, 5, 6, 7, 8, 9], 1239 [2, 3, 4, 5, 6, 7, 8, 9, 10])), 1240 !if(!eq(sample.LodOrClamp, ""), 1241 [1, 2, 3], 1242 [1, 2, 3, 4])), 1243 !add(dw, !size(sample.ExtraAddrArgs))); 1244 1245 // Generate machine instructions based on possible register classes for the 1246 // required numbers of address words. The disassembler defaults to the 1247 // smallest register class. 1248 list<MIMGAddrSize> MachineInstrs = 1249 !foldl([]<MIMGAddrSize>, 1250 !foreach(range, 1251 // V4 is generated for V3 and V4 1252 // V8 is generated for V5 through V8 1253 // V16 is generated for V13 through V16 1254 [[1],[2],[3],[3,4],[5],[6],[7],[5,8],[9],[10],[11],[12],[13,16]], 1255 MIMGAddrSizes_dw_range<range>), 1256 lhs, dw, 1257 !if(isRangeInList<dw.Min, dw.Max, AllNumAddrWords>.ret, 1258 !listconcat(lhs, [MIMGAddrSize<dw.Max, !empty(lhs)>]), 1259 lhs)); 1260 1261 // For NSA, generate machine instructions for all possible numbers of words 1262 // except 1 (which is already covered by the non-NSA case). 1263 // The disassembler defaults to the largest number of arguments among the 1264 // variants with the same number of NSA words, and custom code then derives 1265 // the exact variant based on the sample variant and the image dimension. 1266 list<MIMGAddrSize> NSAInstrs = 1267 !foldl([]<MIMGAddrSize>, [[12, 11, 10], [9, 8, 7, 6], [5, 4, 3, 2]], prev, nsa_group, 1268 !listconcat(prev, 1269 !foldl([]<MIMGAddrSize>, nsa_group, lhs, dw, 1270 !if(isIntInList<dw, AllNumAddrWords>.ret, 1271 !listconcat(lhs, [MIMGAddrSize<dw, !empty(lhs)>]), 1272 lhs)))); 1273 1274 // In NSA format if there is a requirement for more VGPRs than the format 1275 // supports, then the rest are sequential after the last one. Generate 1276 // machine instructions for all possible number of words. The disassembler 1277 // defaults to the largest number of arguments but no larger than max nsa 1278 // size. List is generated with the register class needed for last vaddr since 1279 // it is the only one that could have a register other than VGPR32. 1280 int EnableDisasmNum = !foldl(!head(AllNumAddrWords), !tail(AllNumAddrWords), 1281 acc, var, !if(!le(var, nsa_max_addr), var, acc)); 1282 list<int> PossibleVariants = 1283 !listconcat([12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2], !if(includeNSA1, [1], [])); 1284 list<LastVAddrSize> PartialNSAInstrs = 1285 !foldl([]<LastVAddrSize>, PossibleVariants, lhs, dw, 1286 !if(isIntInList<dw, AllNumAddrWords>.ret, 1287 !listconcat(lhs, [LastVAddrSize<dw, !sub(nsa_max_addr, 1), 1288 !eq(dw, EnableDisasmNum)>]), 1289 lhs)); 1290} 1291 1292multiclass MIMG_Sampler_Src_Helper <mimgopc op, string asm, 1293 AMDGPUSampleVariant sample, RegisterClass dst_rc, 1294 bit enableDisasm = 0, 1295 bit ExtendedImageInst = 1, bit isG16 = 0> { 1296 foreach addr = MIMG_Sampler_AddrSizes<sample, isG16>.MachineInstrs in { 1297 let VAddrDwords = addr.NumWords in { 1298 if op.HAS_GFX10M then { 1299 def _V # addr.NumWords 1300 : MIMG_Sampler_Helper <op, asm, dst_rc, addr.RegClass, 1301 !if(!and(enableDisasm, addr.Disassemble), "AMDGPU", "")>; 1302 if !not(ExtendedImageInst) then 1303 def _V # addr.NumWords # _gfx90a 1304 : MIMG_Sampler_gfx90a <op, asm, dst_rc, addr.RegClass, 1305 !if(!and(enableDisasm, addr.Disassemble), "GFX90A", "")>; 1306 def _V # addr.NumWords # _gfx10 1307 : MIMG_Sampler_gfx10 <op, asm, dst_rc, addr.RegClass, 1308 !if(!and(enableDisasm, addr.Disassemble), "AMDGPU", "")>; 1309 } 1310 if op.HAS_GFX11 then { 1311 def _V # addr.NumWords # _gfx11 1312 : MIMG_Sampler_gfx11 <op, asm, dst_rc, addr.RegClass, 1313 !if(!and(enableDisasm, addr.Disassemble), "AMDGPU", "")>; 1314 } 1315 } 1316 } 1317 1318 foreach addr = MIMG_Sampler_AddrSizes<sample, isG16>.NSAInstrs in { 1319 let VAddrDwords = addr.NumWords in { 1320 if op.HAS_GFX10M then { 1321 def _V # addr.NumWords # _nsa_gfx10 1322 : MIMG_Sampler_nsa_gfx10<op, asm, dst_rc, addr.NumWords, 1323 !if(!and(enableDisasm, addr.Disassemble), "AMDGPU", "")>; 1324 } 1325 } 1326 } 1327 1328 foreach addr = MIMG_Sampler_AddrSizes<sample, isG16, 5/*MaxNSASize*/>.PartialNSAInstrs in { 1329 let VAddrDwords = addr.NumWords in { 1330 if op.HAS_GFX11 then { 1331 def _V # addr.NumWords # _nsa_gfx11 1332 : MIMG_Sampler_nsa_gfx11<op, asm, dst_rc, addr.NumWords, addr.RegClass, 1333 !if(!and(enableDisasm, addr.Disassemble), "AMDGPU", "")>; 1334 } 1335 } 1336 } 1337 1338 foreach addr = MIMG_Sampler_AddrSizes<sample, isG16, 4/*MaxNSASize*/, 1>.PartialNSAInstrs in { 1339 let VAddrDwords = addr.NumWords in { 1340 if op.HAS_GFX12 then { 1341 def _V # addr.NumWords # _gfx12 1342 : VSAMPLE_Sampler_gfx12<op, asm, dst_rc, addr.NumWords, addr.RegClass, 1343 !if(!and(enableDisasm, addr.Disassemble), "GFX12", "")>; 1344 } 1345 } 1346 } 1347} 1348 1349class MIMG_Sampler_BaseOpcode<AMDGPUSampleVariant sample> 1350 : MIMGBaseOpcode { 1351 let Sampler = 1; 1352 let NumExtraArgs = !size(sample.ExtraAddrArgs); 1353 let Gradients = sample.Gradients; 1354 let LodOrClampOrMip = !ne(sample.LodOrClamp, ""); 1355} 1356 1357multiclass MIMG_Sampler <mimgopc op, AMDGPUSampleVariant sample, bit wqm = 0, 1358 bit isG16 = 0, bit isGetLod = 0, 1359 string asm = "image_sample"#sample.LowerCaseMod#!if(isG16, "_g16", ""), 1360 bit ExtendedImageInst = !ne(sample.LowerCaseMod, "")> { 1361 def "" : MIMG_Sampler_BaseOpcode<sample> { 1362 let HasD16 = !not(isGetLod); 1363 let G16 = isG16; 1364 } 1365 1366 let BaseOpcode = !cast<MIMGBaseOpcode>(NAME), WQM = wqm, 1367 mayLoad = !not(isGetLod) in { 1368 let VDataDwords = 1 in 1369 defm _V1 : MIMG_Sampler_Src_Helper<op, asm, sample, VGPR_32, 1, ExtendedImageInst, isG16>; 1370 let VDataDwords = 2 in 1371 defm _V2 : MIMG_Sampler_Src_Helper<op, asm, sample, VReg_64, 0, ExtendedImageInst, isG16>; 1372 let VDataDwords = 3 in 1373 defm _V3 : MIMG_Sampler_Src_Helper<op, asm, sample, VReg_96, 0, ExtendedImageInst, isG16>; 1374 let VDataDwords = 4 in 1375 defm _V4 : MIMG_Sampler_Src_Helper<op, asm, sample, VReg_128, 0, ExtendedImageInst, isG16>; 1376 let VDataDwords = 5 in 1377 defm _V5 : MIMG_Sampler_Src_Helper<op, asm, sample, VReg_160, 0, ExtendedImageInst, isG16>; 1378 } 1379} 1380 1381multiclass MIMG_Sampler_WQM <mimgopc op, AMDGPUSampleVariant sample> 1382 : MIMG_Sampler<op, sample, 1>; 1383 1384multiclass MIMG_Gather <mimgopc op, AMDGPUSampleVariant sample, bit wqm = 0, 1385 string asm = "image_gather4"#sample.LowerCaseMod> { 1386 def "" : MIMG_Sampler_BaseOpcode<sample> { 1387 let HasD16 = 1; 1388 let Gather4 = 1; 1389 } 1390 1391 let BaseOpcode = !cast<MIMGBaseOpcode>(NAME), WQM = wqm, 1392 Gather4 = 1 in { 1393 let VDataDwords = 2 in 1394 defm _V2 : MIMG_Sampler_Src_Helper<op, asm, sample, VReg_64>; /* for packed D16 only */ 1395 let VDataDwords = 4 in 1396 defm _V4 : MIMG_Sampler_Src_Helper<op, asm, sample, VReg_128, 1>; 1397 let VDataDwords = 5 in 1398 defm _V5 : MIMG_Sampler_Src_Helper<op, asm, sample, VReg_160>; 1399 } 1400} 1401 1402multiclass MIMG_Gather_WQM <mimgopc op, AMDGPUSampleVariant sample> 1403 : MIMG_Gather<op, sample, 1>; 1404 1405class MIMG_IntersectRay_Helper<bit Is64, bit IsA16> { 1406 int num_addrs = !if(Is64, !if(IsA16, 9, 12), !if(IsA16, 8, 11)); 1407 RegisterClass RegClass = MIMGAddrSize<num_addrs, 0>.RegClass; 1408 int VAddrDwords = !srl(RegClass.Size, 5); 1409 1410 int GFX11PlusNSAAddrs = !if(IsA16, 4, 5); 1411 RegisterClass node_ptr_type = !if(Is64, VReg_64, VGPR_32); 1412 list<RegisterClass> GFX11PlusAddrTypes = 1413 !if(IsA16, 1414 [node_ptr_type, VGPR_32, VReg_96, VReg_96], 1415 [node_ptr_type, VGPR_32, VReg_96, VReg_96, VReg_96]); 1416} 1417 1418class MIMG_IntersectRay_gfx10<mimgopc op, string opcode, RegisterClass AddrRC> 1419 : MIMG_gfx10<op.GFX10M, (outs VReg_128:$vdata), "AMDGPU"> { 1420 let InOperandList = (ins AddrRC:$vaddr0, SReg_128:$srsrc, A16:$a16); 1421 let AsmString = opcode#" $vdata, $vaddr0, $srsrc$a16"; 1422 1423 let nsa = 0; 1424} 1425 1426class MIMG_IntersectRay_nsa_gfx10<mimgopc op, string opcode, int num_addrs> 1427 : MIMG_nsa_gfx10<op.GFX10M, (outs VReg_128:$vdata), num_addrs, "AMDGPU"> { 1428 let InOperandList = !con(nsah.AddrIns, (ins SReg_128:$srsrc, A16:$a16)); 1429 let AsmString = opcode#" $vdata, "#nsah.AddrAsm#", $srsrc$a16"; 1430} 1431 1432class MIMG_IntersectRay_gfx11<mimgopc op, string opcode, RegisterClass AddrRC> 1433 : MIMG_gfx11<op.GFX11, (outs VReg_128:$vdata), "AMDGPU"> { 1434 let InOperandList = (ins AddrRC:$vaddr0, SReg_128:$srsrc, A16:$a16); 1435 let AsmString = opcode#" $vdata, $vaddr0, $srsrc$a16"; 1436 1437 let nsa = 0; 1438} 1439 1440class MIMG_IntersectRay_nsa_gfx11<mimgopc op, string opcode, int num_addrs, 1441 list<RegisterClass> addr_types> 1442 : MIMG_nsa_gfx11<op.GFX11, (outs VReg_128:$vdata), num_addrs, "AMDGPU", 1443 addr_types> { 1444 let InOperandList = !con(nsah.AddrIns, (ins SReg_128:$srsrc, A16:$a16)); 1445 let AsmString = opcode#" $vdata, "#nsah.AddrAsm#", $srsrc$a16"; 1446} 1447 1448class VIMAGE_IntersectRay_gfx12<mimgopc op, string opcode, int num_addrs, 1449 list<RegisterClass> addr_types> 1450 : VIMAGE_gfx12<op.GFX12, (outs VReg_128:$vdata), 1451 num_addrs, "GFX12", addr_types> { 1452 let InOperandList = !con(nsah.AddrIns, (ins SReg_128:$rsrc, A16:$a16)); 1453 let AsmString = opcode#" $vdata, "#nsah.AddrAsm#", $rsrc$a16"; 1454} 1455 1456multiclass MIMG_IntersectRay<mimgopc op, string opcode, bit Is64, bit IsA16> { 1457 defvar info = MIMG_IntersectRay_Helper<Is64, IsA16>; 1458 def "" : MIMGBaseOpcode { 1459 let BVH = 1; 1460 let A16 = IsA16; 1461 } 1462 let dmask = 0xf, 1463 d16 = 0, 1464 cpol = 0, 1465 tfe = 0, 1466 r128 = 1, 1467 dim = {0, 0, 0}, 1468 a16 = IsA16, 1469 d16 = 0, 1470 BaseOpcode = !cast<MIMGBaseOpcode>(NAME), 1471 VDataDwords = 4 in { 1472 let unorm = 1, 1473 lwe = 0, 1474 ssamp = 0 in { 1475 if op.HAS_GFX10M then 1476 def _sa_gfx10 : MIMG_IntersectRay_gfx10<op, opcode, info.RegClass> { 1477 let VAddrDwords = info.VAddrDwords; 1478 } 1479 if op.HAS_GFX11 then 1480 def _sa_gfx11 : MIMG_IntersectRay_gfx11<op, opcode, info.RegClass> { 1481 let VAddrDwords = info.VAddrDwords; 1482 } 1483 if op.HAS_GFX10M then 1484 def _nsa_gfx10 : MIMG_IntersectRay_nsa_gfx10<op, opcode, info.num_addrs> { 1485 let VAddrDwords = info.num_addrs; 1486 } 1487 if op.HAS_GFX11 then 1488 def _nsa_gfx11 : MIMG_IntersectRay_nsa_gfx11<op, opcode, 1489 info.GFX11PlusNSAAddrs, 1490 info.GFX11PlusAddrTypes> { 1491 let VAddrDwords = info.num_addrs; 1492 } 1493 } 1494 def _gfx12 : VIMAGE_IntersectRay_gfx12<op, opcode, info.GFX11PlusNSAAddrs, 1495 info.GFX11PlusAddrTypes> { 1496 let VAddrDwords = info.num_addrs; 1497 } 1498 } 1499} 1500 1501multiclass MIMG_MSAA_Load <mimgopc op, string asm> { 1502 def "" : MIMGBaseOpcode { 1503 let HasD16 = 1; 1504 let Gather4 = 1; /* for appropriate dmask handling */ 1505 let MSAA = 1; 1506 } 1507 1508 let BaseOpcode = !cast<MIMGBaseOpcode>(NAME), 1509 Gather4 = 1, hasPostISelHook = 0, mayLoad = 1 in { 1510 let VDataDwords = 2 in 1511 defm _V2 : MIMG_NoSampler_Src_Helper<op, asm, VReg_64, 0, 0, 1>; /* packed D16 */ 1512 let VDataDwords = 3 in 1513 defm _V3 : MIMG_NoSampler_Src_Helper<op, asm, VReg_96, 0, 0, 1>; /* packed D16 + tfe */ 1514 let VDataDwords = 4 in 1515 defm _V4 : MIMG_NoSampler_Src_Helper<op, asm, VReg_128, 1, 0, 1>; 1516 let VDataDwords = 5 in 1517 defm _V5 : MIMG_NoSampler_Src_Helper<op, asm, VReg_160, 0, 0, 1>; 1518 } 1519} 1520 1521//===----------------------------------------------------------------------===// 1522// MIMG Instructions 1523//===----------------------------------------------------------------------===// 1524let OtherPredicates = [HasImageInsts] in { 1525 1526defm IMAGE_LOAD : MIMG_NoSampler <mimgopc<0x00, 0x00, 0x00>, "image_load", 1>; 1527defm IMAGE_LOAD_MIP : MIMG_NoSampler <mimgopc<0x01, 0x01, 0x01>, "image_load_mip", 1, 1>; 1528defm IMAGE_LOAD_PCK : MIMG_NoSampler <mimgopc<0x02, 0x02, 0x02>, "image_load_pck", 0>; 1529defm IMAGE_LOAD_PCK_SGN : MIMG_NoSampler <mimgopc<0x03, 0x03, 0x03>, "image_load_pck_sgn", 0>; 1530defm IMAGE_LOAD_MIP_PCK : MIMG_NoSampler <mimgopc<0x04, 0x04, 0x04>, "image_load_mip_pck", 0, 1>; 1531defm IMAGE_LOAD_MIP_PCK_SGN : MIMG_NoSampler <mimgopc<0x05, 0x05, 0x05>, "image_load_mip_pck_sgn", 0, 1>; 1532defm IMAGE_STORE : MIMG_Store <mimgopc<0x06, 0x06, 0x08>, "image_store", 1>; 1533defm IMAGE_STORE_MIP : MIMG_Store <mimgopc<0x07, 0x07, 0x09>, "image_store_mip", 1, 1>; 1534defm IMAGE_STORE_PCK : MIMG_Store <mimgopc<0x08, 0x08, 0x0a>, "image_store_pck", 0>; 1535defm IMAGE_STORE_MIP_PCK : MIMG_Store <mimgopc<0x09, 0x09, 0x0b>, "image_store_mip_pck", 0, 1>; 1536 1537defm IMAGE_GET_RESINFO : MIMG_NoSampler <mimgopc<0x17, 0x17, 0x0e, 0x0e, 0x0e>, "image_get_resinfo", 0, 1, 1>; 1538 1539defm IMAGE_ATOMIC_SWAP : MIMG_Atomic <mimgopc<0x0a, 0x0a, 0x0f, 0x10, 0x0f>, "image_atomic_swap">; 1540defm IMAGE_ATOMIC_CMPSWAP : MIMG_Atomic <mimgopc<0x0b, 0x0b, 0x10, 0x11, 0x10>, "image_atomic_cmpswap", 1>; 1541defm IMAGE_ATOMIC_ADD : MIMG_Atomic_Renamed <mimgopc<0x0c, 0x0c, 0x11, 0x12, 0x11>, "image_atomic_add", "image_atomic_add_uint">; 1542defm IMAGE_ATOMIC_SUB : MIMG_Atomic_Renamed <mimgopc<0x0d, 0x0d, 0x12, 0x13, 0x12>, "image_atomic_sub", "image_atomic_sub_uint">; 1543defm IMAGE_ATOMIC_RSUB : MIMG_Atomic <mimgopc<MIMG.NOP, MIMG.NOP, MIMG.NOP, MIMG.NOP, 0x13>, "image_atomic_rsub">; 1544defm IMAGE_ATOMIC_SMIN : MIMG_Atomic_Renamed <mimgopc<0x0e, 0x0e, 0x14>, "image_atomic_smin", "image_atomic_min_int">; 1545defm IMAGE_ATOMIC_UMIN : MIMG_Atomic_Renamed <mimgopc<0x0f, 0x0f, 0x15>, "image_atomic_umin", "image_atomic_min_uint">; 1546defm IMAGE_ATOMIC_SMAX : MIMG_Atomic_Renamed <mimgopc<0x10, 0x10, 0x16>, "image_atomic_smax", "image_atomic_max_int">; 1547defm IMAGE_ATOMIC_UMAX : MIMG_Atomic_Renamed <mimgopc<0x11, 0x11, 0x17>, "image_atomic_umax", "image_atomic_max_uint">; 1548defm IMAGE_ATOMIC_AND : MIMG_Atomic <mimgopc<0x12, 0x12, 0x18>, "image_atomic_and">; 1549defm IMAGE_ATOMIC_OR : MIMG_Atomic <mimgopc<0x13, 0x13, 0x19>, "image_atomic_or">; 1550defm IMAGE_ATOMIC_XOR : MIMG_Atomic <mimgopc<0x14, 0x14, 0x1a>, "image_atomic_xor">; 1551defm IMAGE_ATOMIC_INC : MIMG_Atomic_Renamed <mimgopc<0x15, 0x15, 0x1b>, "image_atomic_inc", "image_atomic_inc_uint">; 1552defm IMAGE_ATOMIC_DEC : MIMG_Atomic_Renamed <mimgopc<0x16, 0x16, 0x1c>, "image_atomic_dec", "image_atomic_dec_uint">; 1553defm IMAGE_ATOMIC_FCMPSWAP : MIMG_Atomic <mimgopc<MIMG.NOP, MIMG.NOP, 0x1d, MIMG.NOP>, "image_atomic_fcmpswap", 1, 1>; 1554defm IMAGE_ATOMIC_FMIN : MIMG_Atomic <mimgopc<MIMG.NOP, MIMG.NOP, 0x1e, MIMG.NOP>, "image_atomic_fmin", 0, 1>; 1555defm IMAGE_ATOMIC_FMAX : MIMG_Atomic <mimgopc<MIMG.NOP, MIMG.NOP, 0x1f, MIMG.NOP>, "image_atomic_fmax", 0, 1>; 1556defm IMAGE_ATOMIC_PK_ADD_F16 : MIMG_Atomic <mimgopc<0x86, MIMG.NOP, MIMG.NOP, MIMG.NOP, MIMG.NOP>, "image_atomic_pk_add_f16", 0, 1>; 1557defm IMAGE_ATOMIC_PK_ADD_BF16 : MIMG_Atomic <mimgopc<0x87, MIMG.NOP, MIMG.NOP, MIMG.NOP, MIMG.NOP>, "image_atomic_pk_add_bf16", 0, 1>; 1558defm IMAGE_ATOMIC_ADD_FLT : MIMG_Atomic <mimgopc<0x83, MIMG.NOP, MIMG.NOP, MIMG.NOP>, "image_atomic_add_flt", 0, 1>; 1559defm IMAGE_ATOMIC_MIN_FLT : MIMG_Atomic <mimgopc<0x84, MIMG.NOP, MIMG.NOP, MIMG.NOP>, "image_atomic_min_num_flt", 0, 1, "image_atomic_min_flt">; 1560defm IMAGE_ATOMIC_MAX_FLT : MIMG_Atomic <mimgopc<0x85, MIMG.NOP, MIMG.NOP, MIMG.NOP>, "image_atomic_max_num_flt", 0, 1, "image_atomic_max_flt">; 1561 1562defm IMAGE_SAMPLE : MIMG_Sampler_WQM <mimgopc<0x1b, 0x1b, 0x20>, AMDGPUSample>; 1563let OtherPredicates = [HasExtendedImageInsts] in { 1564defm IMAGE_SAMPLE_CL : MIMG_Sampler_WQM <mimgopc<0x40, 0x40, 0x21>, AMDGPUSample_cl>; 1565defm IMAGE_SAMPLE_D : MIMG_Sampler <mimgopc<0x1c, 0x1c, 0x22>, AMDGPUSample_d>; 1566defm IMAGE_SAMPLE_D_CL : MIMG_Sampler <mimgopc<0x41, 0x41, 0x23>, AMDGPUSample_d_cl>; 1567defm IMAGE_SAMPLE_L : MIMG_Sampler <mimgopc<0x1d, 0x1d, 0x24>, AMDGPUSample_l>; 1568defm IMAGE_SAMPLE_B : MIMG_Sampler_WQM <mimgopc<0x1e, 0x1e, 0x25>, AMDGPUSample_b>; 1569defm IMAGE_SAMPLE_B_CL : MIMG_Sampler_WQM <mimgopc<0x42, 0x42, 0x26>, AMDGPUSample_b_cl>; 1570defm IMAGE_SAMPLE_LZ : MIMG_Sampler <mimgopc<0x1f, 0x1f, 0x27>, AMDGPUSample_lz>; 1571defm IMAGE_SAMPLE_C : MIMG_Sampler_WQM <mimgopc<0x20, 0x20, 0x28>, AMDGPUSample_c>; 1572defm IMAGE_SAMPLE_C_CL : MIMG_Sampler_WQM <mimgopc<0x43, 0x43, 0x29>, AMDGPUSample_c_cl>; 1573defm IMAGE_SAMPLE_C_D : MIMG_Sampler <mimgopc<0x21, 0x21, 0x2a>, AMDGPUSample_c_d>; 1574defm IMAGE_SAMPLE_C_D_CL : MIMG_Sampler <mimgopc<0x44, 0x44, 0x2b>, AMDGPUSample_c_d_cl>; 1575defm IMAGE_SAMPLE_C_L : MIMG_Sampler <mimgopc<0x22, 0x22, 0x2c>, AMDGPUSample_c_l>; 1576defm IMAGE_SAMPLE_C_B : MIMG_Sampler_WQM <mimgopc<0x23, 0x23, 0x2d>, AMDGPUSample_c_b>; 1577defm IMAGE_SAMPLE_C_B_CL : MIMG_Sampler_WQM <mimgopc<0x45, 0x45, 0x2e>, AMDGPUSample_c_b_cl>; 1578defm IMAGE_SAMPLE_C_LZ : MIMG_Sampler <mimgopc<0x24, 0x24, 0x2f>, AMDGPUSample_c_lz>; 1579defm IMAGE_SAMPLE_O : MIMG_Sampler_WQM <mimgopc<0x25, 0x25, 0x30>, AMDGPUSample_o>; 1580defm IMAGE_SAMPLE_CL_O : MIMG_Sampler_WQM <mimgopc<0x46, 0x46, 0x31>, AMDGPUSample_cl_o>; 1581defm IMAGE_SAMPLE_D_O : MIMG_Sampler <mimgopc<0x26, 0x26, 0x32>, AMDGPUSample_d_o>; 1582defm IMAGE_SAMPLE_D_CL_O : MIMG_Sampler <mimgopc<0x47, 0x47, 0x33>, AMDGPUSample_d_cl_o>; 1583defm IMAGE_SAMPLE_L_O : MIMG_Sampler <mimgopc<0x27, 0x27, 0x34>, AMDGPUSample_l_o>; 1584defm IMAGE_SAMPLE_B_O : MIMG_Sampler_WQM <mimgopc<0x28, 0x28, 0x35>, AMDGPUSample_b_o>; 1585defm IMAGE_SAMPLE_B_CL_O : MIMG_Sampler_WQM <mimgopc<0x48, 0x48, 0x36>, AMDGPUSample_b_cl_o>; 1586defm IMAGE_SAMPLE_LZ_O : MIMG_Sampler <mimgopc<0x29, 0x29, 0x37>, AMDGPUSample_lz_o>; 1587defm IMAGE_SAMPLE_C_O : MIMG_Sampler_WQM <mimgopc<0x2a, 0x2a, 0x38>, AMDGPUSample_c_o>; 1588defm IMAGE_SAMPLE_C_CL_O : MIMG_Sampler_WQM <mimgopc<0x49, 0x49, 0x39>, AMDGPUSample_c_cl_o>; 1589defm IMAGE_SAMPLE_C_D_O : MIMG_Sampler <mimgopc<0x2b, 0x2b, 0x3a>, AMDGPUSample_c_d_o>; 1590defm IMAGE_SAMPLE_C_D_CL_O : MIMG_Sampler <mimgopc<0x4a, 0x4a, 0x3b>, AMDGPUSample_c_d_cl_o>; 1591defm IMAGE_SAMPLE_C_L_O : MIMG_Sampler <mimgopc<0x2c, 0x2c, 0x3c>, AMDGPUSample_c_l_o>; 1592defm IMAGE_SAMPLE_C_B_CL_O : MIMG_Sampler_WQM <mimgopc<0x4b, 0x4b, 0x3e>, AMDGPUSample_c_b_cl_o>; 1593defm IMAGE_SAMPLE_C_B_O : MIMG_Sampler_WQM <mimgopc<0x2d, 0x2d, 0x3d>, AMDGPUSample_c_b_o>; 1594defm IMAGE_SAMPLE_C_LZ_O : MIMG_Sampler <mimgopc<0x2e, 0x2e, 0x3f>, AMDGPUSample_c_lz_o>; 1595defm IMAGE_GATHER4 : MIMG_Gather_WQM <mimgopc<0x2f, 0x2f, 0x40>, AMDGPUSample>; 1596defm IMAGE_GATHER4_CL : MIMG_Gather_WQM <mimgopc<0x60, 0x60, 0x41>, AMDGPUSample_cl>; 1597defm IMAGE_GATHER4_L : MIMG_Gather <mimgopc<0x30, 0x30, 0x44>, AMDGPUSample_l>; 1598defm IMAGE_GATHER4_B : MIMG_Gather_WQM <mimgopc<0x31, 0x31, 0x45>, AMDGPUSample_b>; 1599defm IMAGE_GATHER4_B_CL : MIMG_Gather_WQM <mimgopc<0x61, 0x61, 0x46>, AMDGPUSample_b_cl>; 1600defm IMAGE_GATHER4_LZ : MIMG_Gather <mimgopc<0x32, 0x32, 0x47>, AMDGPUSample_lz>; 1601defm IMAGE_GATHER4_C : MIMG_Gather_WQM <mimgopc<0x33, 0x33, 0x48>, AMDGPUSample_c>; 1602defm IMAGE_GATHER4_C_CL : MIMG_Gather_WQM <mimgopc<0x62, 0x62, 0x49>, AMDGPUSample_c_cl>; 1603defm IMAGE_GATHER4_C_L : MIMG_Gather <mimgopc<0x63, 0x63, 0x4c>, AMDGPUSample_c_l>; 1604defm IMAGE_GATHER4_C_B : MIMG_Gather_WQM <mimgopc<0x64, 0x64, 0x4d>, AMDGPUSample_c_b>; 1605defm IMAGE_GATHER4_C_B_CL : MIMG_Gather_WQM <mimgopc<0x65, 0x65, 0x4e>, AMDGPUSample_c_b_cl>; 1606defm IMAGE_GATHER4_C_LZ : MIMG_Gather <mimgopc<0x34, 0x34, 0x4f>, AMDGPUSample_c_lz>; 1607defm IMAGE_GATHER4_O : MIMG_Gather_WQM <mimgopc<0x35, 0x35, 0x50>, AMDGPUSample_o>; 1608defm IMAGE_GATHER4_CL_O : MIMG_Gather_WQM <mimgopc<MIMG.NOP, MIMG.NOP, 0x51>, AMDGPUSample_cl_o>; 1609defm IMAGE_GATHER4_L_O : MIMG_Gather <mimgopc<MIMG.NOP, MIMG.NOP, 0x54>, AMDGPUSample_l_o>; 1610defm IMAGE_GATHER4_B_O : MIMG_Gather_WQM <mimgopc<MIMG.NOP, MIMG.NOP, 0x55>, AMDGPUSample_b_o>; 1611defm IMAGE_GATHER4_B_CL_O : MIMG_Gather <mimgopc<MIMG.NOP, MIMG.NOP, 0x56>, AMDGPUSample_b_cl_o>; 1612defm IMAGE_GATHER4_LZ_O : MIMG_Gather <mimgopc<0x36, 0x36, 0x57>, AMDGPUSample_lz_o>; 1613defm IMAGE_GATHER4_C_O : MIMG_Gather_WQM <mimgopc<MIMG.NOP, MIMG.NOP, 0x58>, AMDGPUSample_c_o>; 1614defm IMAGE_GATHER4_C_CL_O : MIMG_Gather_WQM <mimgopc<MIMG.NOP, MIMG.NOP, 0x59>, AMDGPUSample_c_cl_o>; 1615defm IMAGE_GATHER4_C_L_O : MIMG_Gather <mimgopc<MIMG.NOP, MIMG.NOP, 0x5c>, AMDGPUSample_c_l_o>; 1616defm IMAGE_GATHER4_C_B_O : MIMG_Gather_WQM <mimgopc<MIMG.NOP, MIMG.NOP, 0x5d>, AMDGPUSample_c_b_o>; 1617defm IMAGE_GATHER4_C_B_CL_O : MIMG_Gather_WQM <mimgopc<MIMG.NOP, MIMG.NOP, 0x5e>, AMDGPUSample_c_b_cl_o>; 1618defm IMAGE_GATHER4_C_LZ_O : MIMG_Gather <mimgopc<0x37, 0x37, 0x5f>, AMDGPUSample_c_lz_o>; 1619 1620let SubtargetPredicate = isGFX9Plus in 1621defm IMAGE_GATHER4H : MIMG_Gather <mimgopc<0x90, 0x90, 0x61, 0x42>, AMDGPUSample, 1, "image_gather4h">; 1622 1623defm IMAGE_GET_LOD : MIMG_Sampler <mimgopc<0x38, 0x38, 0x60>, AMDGPUSample, 1, 0, 1, "image_get_lod">; 1624 1625defm IMAGE_SAMPLE_CD : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0x68>, AMDGPUSample_cd>; 1626defm IMAGE_SAMPLE_CD_CL : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0x69>, AMDGPUSample_cd_cl>; 1627defm IMAGE_SAMPLE_C_CD : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0x6a>, AMDGPUSample_c_cd>; 1628defm IMAGE_SAMPLE_C_CD_CL : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0x6b>, AMDGPUSample_c_cd_cl>; 1629defm IMAGE_SAMPLE_CD_O : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0x6c>, AMDGPUSample_cd_o>; 1630defm IMAGE_SAMPLE_CD_CL_O : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0x6d>, AMDGPUSample_cd_cl_o>; 1631defm IMAGE_SAMPLE_C_CD_O : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0x6e>, AMDGPUSample_c_cd_o>; 1632defm IMAGE_SAMPLE_C_CD_CL_O : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0x6f>, AMDGPUSample_c_cd_cl_o>; 1633} // End OtherPredicates = [HasExtendedImageInsts] 1634 1635let OtherPredicates = [HasExtendedImageInsts,HasG16] in { 1636defm IMAGE_SAMPLE_D_G16 : MIMG_Sampler <mimgopc<0x39, 0x39, 0xa2>, AMDGPUSample_d, 0, 1>; 1637defm IMAGE_SAMPLE_D_CL_G16 : MIMG_Sampler <mimgopc<0x5f, 0x5f, 0xa3>, AMDGPUSample_d_cl, 0, 1>; 1638defm IMAGE_SAMPLE_C_D_G16 : MIMG_Sampler <mimgopc<0x3a, 0x3a, 0xaa>, AMDGPUSample_c_d, 0, 1>; 1639defm IMAGE_SAMPLE_C_D_CL_G16 : MIMG_Sampler <mimgopc<0x54, 0x54, 0xab>, AMDGPUSample_c_d_cl, 0, 1>; 1640defm IMAGE_SAMPLE_D_O_G16 : MIMG_Sampler <mimgopc<0x3b, 0x3b, 0xb2>, AMDGPUSample_d_o, 0, 1>; 1641defm IMAGE_SAMPLE_D_CL_O_G16 : MIMG_Sampler <mimgopc<0x55, 0x55, 0xb3>, AMDGPUSample_d_cl_o, 0, 1>; 1642defm IMAGE_SAMPLE_C_D_O_G16 : MIMG_Sampler <mimgopc<0x3c, 0x3c, 0xba>, AMDGPUSample_c_d_o, 0, 1>; 1643defm IMAGE_SAMPLE_C_D_CL_O_G16 : MIMG_Sampler <mimgopc<0x56, 0x56, 0xbb>, AMDGPUSample_c_d_cl_o, 0, 1>; 1644defm IMAGE_SAMPLE_CD_G16 : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xe8>, AMDGPUSample_cd, 0, 1>; 1645defm IMAGE_SAMPLE_CD_CL_G16 : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xe9>, AMDGPUSample_cd_cl, 0, 1>; 1646defm IMAGE_SAMPLE_C_CD_G16 : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xea>, AMDGPUSample_c_cd, 0, 1>; 1647defm IMAGE_SAMPLE_C_CD_CL_G16 : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xeb>, AMDGPUSample_c_cd_cl, 0, 1>; 1648defm IMAGE_SAMPLE_CD_O_G16 : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xec>, AMDGPUSample_cd_o, 0, 1>; 1649defm IMAGE_SAMPLE_CD_CL_O_G16 : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xed>, AMDGPUSample_cd_cl_o, 0, 1>; 1650defm IMAGE_SAMPLE_C_CD_O_G16 : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xee>, AMDGPUSample_c_cd_o, 0, 1>; 1651defm IMAGE_SAMPLE_C_CD_CL_O_G16 : MIMG_Sampler <mimgopc<MIMG.NOP, MIMG.NOP, 0xef>, AMDGPUSample_c_cd_cl_o, 0, 1>; 1652} // End OtherPredicates = [HasExtendedImageInsts,HasG16] 1653 1654//def IMAGE_RSRC256 : MIMG_NoPattern_RSRC256 <"image_rsrc256", mimgopc<0x7e>>; 1655//def IMAGE_SAMPLER : MIMG_NoPattern_ <"image_sampler", mimgopc<0x7f>>; 1656 1657let SubtargetPredicate = isGFX10Only, OtherPredicates = [HasGFX10_AEncoding] in 1658defm IMAGE_MSAA_LOAD_X : MIMG_NoSampler <mimgopc<MIMG.NOP, MIMG.NOP, 0x80>, "image_msaa_load", 1, 0, 0, 1>; 1659 1660let OtherPredicates = [HasGFX10_AEncoding] in 1661defm IMAGE_MSAA_LOAD : MIMG_MSAA_Load <mimgopc<0x18, 0x18, MIMG.NOP>, "image_msaa_load">; 1662 1663let OtherPredicates = [HasGFX10_AEncoding] in { 1664defm IMAGE_BVH_INTERSECT_RAY : MIMG_IntersectRay<mimgopc<0x19, 0x19, 0xe6>, "image_bvh_intersect_ray", 0, 0>; 1665defm IMAGE_BVH_INTERSECT_RAY_a16 : MIMG_IntersectRay<mimgopc<0x19, 0x19, 0xe6>, "image_bvh_intersect_ray", 0, 1>; 1666defm IMAGE_BVH64_INTERSECT_RAY : MIMG_IntersectRay<mimgopc<0x1a, 0x1a, 0xe7>, "image_bvh64_intersect_ray", 1, 0>; 1667defm IMAGE_BVH64_INTERSECT_RAY_a16 : MIMG_IntersectRay<mimgopc<0x1a, 0x1a, 0xe7>, "image_bvh64_intersect_ray", 1, 1>; 1668} // End OtherPredicates = [HasGFX10_AEncoding] 1669 1670} // End let OtherPredicates = [HasImageInsts] 1671 1672/********** ========================================= **********/ 1673/********** Table of dimension-aware image intrinsics **********/ 1674/********** ========================================= **********/ 1675 1676class ImageDimIntrinsicInfo<AMDGPUImageDimIntrinsic I> { 1677 Intrinsic Intr = I; 1678 MIMGBaseOpcode BaseOpcode = !cast<MIMGBaseOpcode>(!strconcat("IMAGE_", I.P.OpMod)); 1679 AMDGPUDimProps Dim = I.P.Dim; 1680 AMDGPUImageDimIntrinsicEval DimEval = AMDGPUImageDimIntrinsicEval<I.P>; 1681 1682 bits<8> NumOffsetArgs = DimEval.NumOffsetArgs; 1683 bits<8> NumBiasArgs = DimEval.NumBiasArgs; 1684 bits<8> NumZCompareArgs = DimEval.NumZCompareArgs; 1685 bits<8> NumGradients = DimEval.NumGradientArgs; 1686 bits<8> NumDmask = DimEval.NumDmaskArgs; 1687 bits<8> NumData = DimEval.NumDataArgs; 1688 bits<8> NumVAddrs = DimEval.NumVAddrArgs; 1689 bits<8> NumArgs = !add(DimEval.CachePolicyArgIndex, 1); 1690 1691 bits<8> DMaskIndex = DimEval.DmaskArgIndex; 1692 bits<8> VAddrStart = DimEval.VAddrArgIndex; 1693 bits<8> OffsetIndex = DimEval.OffsetArgIndex; 1694 bits<8> BiasIndex = DimEval.BiasArgIndex; 1695 bits<8> ZCompareIndex = DimEval.ZCompareArgIndex; 1696 bits<8> GradientStart = DimEval.GradientArgIndex; 1697 bits<8> CoordStart = DimEval.CoordArgIndex; 1698 bits<8> LodIndex = DimEval.LodArgIndex; 1699 bits<8> MipIndex = DimEval.MipArgIndex; 1700 bits<8> VAddrEnd = !add(DimEval.VAddrArgIndex, DimEval.NumVAddrArgs); 1701 bits<8> RsrcIndex = DimEval.RsrcArgIndex; 1702 bits<8> SampIndex = DimEval.SampArgIndex; 1703 bits<8> UnormIndex = DimEval.UnormArgIndex; 1704 bits<8> TexFailCtrlIndex = DimEval.TexFailCtrlArgIndex; 1705 bits<8> CachePolicyIndex = DimEval.CachePolicyArgIndex; 1706 1707 bits<8> BiasTyArg = !add(I.P.NumRetAndDataAnyTypes, 1708 !if(!eq(NumOffsetArgs, 0), 0, I.P.ExtraAddrArgs[0].Type.isAny)); 1709 bits<8> GradientTyArg = !add(I.P.NumRetAndDataAnyTypes, 1710 !foldl(0, I.P.ExtraAddrArgs, cnt, arg, !add(cnt, arg.Type.isAny))); 1711 bits<8> CoordTyArg = !add(GradientTyArg, !if(I.P.Gradients, 1, 0)); 1712} 1713 1714def ImageDimIntrinsicTable : GenericTable { 1715 let FilterClass = "ImageDimIntrinsicInfo"; 1716 let Fields = ["Intr", "BaseOpcode", "Dim", "NumOffsetArgs", "NumBiasArgs", "NumZCompareArgs", "NumGradients", "NumDmask", "NumData", "NumVAddrs", "NumArgs", 1717 "DMaskIndex", "VAddrStart", "OffsetIndex", "BiasIndex", "ZCompareIndex", "GradientStart", "CoordStart", "LodIndex", "MipIndex", "VAddrEnd", 1718 "RsrcIndex", "SampIndex", "UnormIndex", "TexFailCtrlIndex", "CachePolicyIndex", 1719 "BiasTyArg", "GradientTyArg", "CoordTyArg"]; 1720 string TypeOf_BaseOpcode = "MIMGBaseOpcode"; 1721 string TypeOf_Dim = "MIMGDim"; 1722 1723 let PrimaryKey = ["Intr"]; 1724 let PrimaryKeyName = "getImageDimIntrinsicInfo"; 1725 let PrimaryKeyEarlyOut = 1; 1726} 1727 1728def getImageDimIntrinsicByBaseOpcode : SearchIndex { 1729 let Table = ImageDimIntrinsicTable; 1730 let Key = ["BaseOpcode", "Dim"]; 1731} 1732 1733foreach intr = !listconcat(AMDGPUImageDimIntrinsics, 1734 AMDGPUImageDimAtomicIntrinsics) in { 1735 def : ImageDimIntrinsicInfo<intr>; 1736} 1737 1738// L to LZ Optimization Mapping 1739def : MIMGLZMapping<IMAGE_SAMPLE_L, IMAGE_SAMPLE_LZ>; 1740def : MIMGLZMapping<IMAGE_SAMPLE_C_L, IMAGE_SAMPLE_C_LZ>; 1741def : MIMGLZMapping<IMAGE_SAMPLE_L_O, IMAGE_SAMPLE_LZ_O>; 1742def : MIMGLZMapping<IMAGE_SAMPLE_C_L_O, IMAGE_SAMPLE_C_LZ_O>; 1743def : MIMGLZMapping<IMAGE_GATHER4_L, IMAGE_GATHER4_LZ>; 1744def : MIMGLZMapping<IMAGE_GATHER4_C_L, IMAGE_GATHER4_C_LZ>; 1745def : MIMGLZMapping<IMAGE_GATHER4_L_O, IMAGE_GATHER4_LZ_O>; 1746def : MIMGLZMapping<IMAGE_GATHER4_C_L_O, IMAGE_GATHER4_C_LZ_O>; 1747 1748// MIP to NONMIP Optimization Mapping 1749def : MIMGMIPMapping<IMAGE_LOAD_MIP, IMAGE_LOAD>; 1750def : MIMGMIPMapping<IMAGE_STORE_MIP, IMAGE_STORE>; 1751 1752// Bias to NoBias Optimization Mapping 1753def : MIMGBiasMapping<IMAGE_SAMPLE_B, IMAGE_SAMPLE>; 1754def : MIMGBiasMapping<IMAGE_SAMPLE_B_CL, IMAGE_SAMPLE_CL>; 1755def : MIMGBiasMapping<IMAGE_SAMPLE_C_B, IMAGE_SAMPLE_C>; 1756def : MIMGBiasMapping<IMAGE_SAMPLE_C_B_CL, IMAGE_SAMPLE_C_CL>; 1757def : MIMGBiasMapping<IMAGE_SAMPLE_B_O, IMAGE_SAMPLE_O>; 1758def : MIMGBiasMapping<IMAGE_SAMPLE_B_CL_O, IMAGE_SAMPLE_CL_O>; 1759def : MIMGBiasMapping<IMAGE_SAMPLE_C_B_O, IMAGE_SAMPLE_C_O>; 1760def : MIMGBiasMapping<IMAGE_SAMPLE_C_B_CL_O, IMAGE_SAMPLE_C_CL_O>; 1761def : MIMGBiasMapping<IMAGE_GATHER4_B, IMAGE_GATHER4>; 1762def : MIMGBiasMapping<IMAGE_GATHER4_B_CL, IMAGE_GATHER4_CL>; 1763def : MIMGBiasMapping<IMAGE_GATHER4_C_B, IMAGE_GATHER4_C>; 1764def : MIMGBiasMapping<IMAGE_GATHER4_C_B_CL, IMAGE_GATHER4_C_CL>; 1765def : MIMGBiasMapping<IMAGE_GATHER4_B_O, IMAGE_GATHER4_O>; 1766def : MIMGBiasMapping<IMAGE_GATHER4_B_CL_O, IMAGE_GATHER4_CL_O>; 1767def : MIMGBiasMapping<IMAGE_GATHER4_C_B_O, IMAGE_GATHER4_C_O>; 1768def : MIMGBiasMapping<IMAGE_GATHER4_C_B_CL_O, IMAGE_GATHER4_C_CL_O>; 1769 1770// Offset to NoOffset Optimization Mapping 1771def : MIMGOffsetMapping<IMAGE_SAMPLE_O, IMAGE_SAMPLE>; 1772def : MIMGOffsetMapping<IMAGE_SAMPLE_CL_O, IMAGE_SAMPLE_CL>; 1773def : MIMGOffsetMapping<IMAGE_SAMPLE_D_O, IMAGE_SAMPLE_D>; 1774def : MIMGOffsetMapping<IMAGE_SAMPLE_D_CL_O, IMAGE_SAMPLE_D_CL>; 1775def : MIMGOffsetMapping<IMAGE_SAMPLE_D_O_G16, IMAGE_SAMPLE_D_G16>; 1776def : MIMGOffsetMapping<IMAGE_SAMPLE_D_CL_O_G16, IMAGE_SAMPLE_D_CL_G16>; 1777def : MIMGOffsetMapping<IMAGE_SAMPLE_L_O, IMAGE_SAMPLE_L>; 1778def : MIMGOffsetMapping<IMAGE_SAMPLE_B_O, IMAGE_SAMPLE_B>; 1779def : MIMGOffsetMapping<IMAGE_SAMPLE_B_CL_O, IMAGE_SAMPLE_B_CL>; 1780def : MIMGOffsetMapping<IMAGE_SAMPLE_LZ_O, IMAGE_SAMPLE_LZ>; 1781def : MIMGOffsetMapping<IMAGE_SAMPLE_C_O, IMAGE_SAMPLE_C>; 1782def : MIMGOffsetMapping<IMAGE_SAMPLE_C_CL_O, IMAGE_SAMPLE_C_CL>; 1783def : MIMGOffsetMapping<IMAGE_SAMPLE_C_D_O, IMAGE_SAMPLE_C_D>; 1784def : MIMGOffsetMapping<IMAGE_SAMPLE_C_D_CL_O, IMAGE_SAMPLE_C_D_CL>; 1785def : MIMGOffsetMapping<IMAGE_SAMPLE_C_D_O_G16, IMAGE_SAMPLE_C_D_G16>; 1786def : MIMGOffsetMapping<IMAGE_SAMPLE_C_D_CL_O_G16, IMAGE_SAMPLE_C_D_CL_G16>; 1787def : MIMGOffsetMapping<IMAGE_SAMPLE_C_L_O, IMAGE_SAMPLE_C_L>; 1788def : MIMGOffsetMapping<IMAGE_SAMPLE_C_B_CL_O, IMAGE_SAMPLE_C_B_CL>; 1789def : MIMGOffsetMapping<IMAGE_SAMPLE_C_B_O, IMAGE_SAMPLE_C_B>; 1790def : MIMGOffsetMapping<IMAGE_SAMPLE_C_LZ_O, IMAGE_SAMPLE_C_LZ>; 1791def : MIMGOffsetMapping<IMAGE_GATHER4_O, IMAGE_GATHER4>; 1792def : MIMGOffsetMapping<IMAGE_GATHER4_CL_O, IMAGE_GATHER4_CL>; 1793def : MIMGOffsetMapping<IMAGE_GATHER4_L_O, IMAGE_GATHER4_L>; 1794def : MIMGOffsetMapping<IMAGE_GATHER4_B_O, IMAGE_GATHER4_B>; 1795def : MIMGOffsetMapping<IMAGE_GATHER4_B_CL_O, IMAGE_GATHER4_B_CL>; 1796def : MIMGOffsetMapping<IMAGE_GATHER4_LZ_O, IMAGE_GATHER4_LZ>; 1797def : MIMGOffsetMapping<IMAGE_GATHER4_C_O, IMAGE_GATHER4_C>; 1798def : MIMGOffsetMapping<IMAGE_GATHER4_C_CL_O, IMAGE_GATHER4_C_CL>; 1799def : MIMGOffsetMapping<IMAGE_GATHER4_C_L_O, IMAGE_GATHER4_C_L>; 1800def : MIMGOffsetMapping<IMAGE_GATHER4_C_B_O, IMAGE_GATHER4_C_B>; 1801def : MIMGOffsetMapping<IMAGE_GATHER4_C_B_CL_O, IMAGE_GATHER4_C_B_CL>; 1802def : MIMGOffsetMapping<IMAGE_GATHER4_C_LZ_O, IMAGE_GATHER4_C_LZ>; 1803def : MIMGOffsetMapping<IMAGE_SAMPLE_CD_O, IMAGE_SAMPLE_CD>; 1804def : MIMGOffsetMapping<IMAGE_SAMPLE_CD_CL_O, IMAGE_SAMPLE_CD_CL>; 1805def : MIMGOffsetMapping<IMAGE_SAMPLE_C_CD_O, IMAGE_SAMPLE_C_CD>; 1806def : MIMGOffsetMapping<IMAGE_SAMPLE_C_CD_CL_O, IMAGE_SAMPLE_C_CD_CL>; 1807def : MIMGOffsetMapping<IMAGE_SAMPLE_CD_O_G16, IMAGE_SAMPLE_CD_G16>; 1808def : MIMGOffsetMapping<IMAGE_SAMPLE_CD_CL_O_G16, IMAGE_SAMPLE_CD_CL_G16>; 1809def : MIMGOffsetMapping<IMAGE_SAMPLE_C_CD_O_G16, IMAGE_SAMPLE_C_CD_G16>; 1810def : MIMGOffsetMapping<IMAGE_SAMPLE_C_CD_CL_O_G16, IMAGE_SAMPLE_C_CD_CL_G16>; 1811 1812// G to G16 Optimization Mapping 1813def : MIMGG16Mapping<IMAGE_SAMPLE_D, IMAGE_SAMPLE_D_G16>; 1814def : MIMGG16Mapping<IMAGE_SAMPLE_D_CL, IMAGE_SAMPLE_D_CL_G16>; 1815def : MIMGG16Mapping<IMAGE_SAMPLE_C_D, IMAGE_SAMPLE_C_D_G16>; 1816def : MIMGG16Mapping<IMAGE_SAMPLE_C_D_CL, IMAGE_SAMPLE_C_D_CL_G16>; 1817def : MIMGG16Mapping<IMAGE_SAMPLE_D_O, IMAGE_SAMPLE_D_O_G16>; 1818def : MIMGG16Mapping<IMAGE_SAMPLE_D_CL_O, IMAGE_SAMPLE_D_CL_O_G16>; 1819def : MIMGG16Mapping<IMAGE_SAMPLE_C_D_O, IMAGE_SAMPLE_C_D_O_G16>; 1820def : MIMGG16Mapping<IMAGE_SAMPLE_C_D_CL_O, IMAGE_SAMPLE_C_D_CL_O_G16>; 1821def : MIMGG16Mapping<IMAGE_SAMPLE_CD, IMAGE_SAMPLE_CD_G16>; 1822def : MIMGG16Mapping<IMAGE_SAMPLE_CD_CL, IMAGE_SAMPLE_CD_CL_G16>; 1823def : MIMGG16Mapping<IMAGE_SAMPLE_C_CD, IMAGE_SAMPLE_C_CD_G16>; 1824def : MIMGG16Mapping<IMAGE_SAMPLE_C_CD_CL, IMAGE_SAMPLE_C_CD_CL_G16>; 1825def : MIMGG16Mapping<IMAGE_SAMPLE_CD_O, IMAGE_SAMPLE_CD_O_G16>; 1826def : MIMGG16Mapping<IMAGE_SAMPLE_CD_CL_O, IMAGE_SAMPLE_CD_CL_O_G16>; 1827def : MIMGG16Mapping<IMAGE_SAMPLE_C_CD_O, IMAGE_SAMPLE_C_CD_O_G16>; 1828def : MIMGG16Mapping<IMAGE_SAMPLE_C_CD_CL_O, IMAGE_SAMPLE_C_CD_CL_O_G16>; 1829