1//===-- AMDGPU.td - AMDGPU Tablegen files --------*- tablegen -*-===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===------------------------------------------------------------===// 9 10include "llvm/Target/Target.td" 11 12//===------------------------------------------------------------===// 13// Subtarget Features (device properties) 14//===------------------------------------------------------------===// 15 16def FeatureFP64 : SubtargetFeature<"fp64", 17 "FP64", 18 "true", 19 "Enable double precision operations" 20>; 21 22def FeatureFastFMAF32 : SubtargetFeature<"fast-fmaf", 23 "FastFMAF32", 24 "true", 25 "Assuming f32 fma is at least as fast as mul + add" 26>; 27 28def HalfRate64Ops : SubtargetFeature<"half-rate-64-ops", 29 "HalfRate64Ops", 30 "true", 31 "Most fp64 instructions are half rate instead of quarter" 32>; 33 34def FeatureR600ALUInst : SubtargetFeature<"R600ALUInst", 35 "R600ALUInst", 36 "false", 37 "Older version of ALU instructions encoding" 38>; 39 40def FeatureVertexCache : SubtargetFeature<"HasVertexCache", 41 "HasVertexCache", 42 "true", 43 "Specify use of dedicated vertex cache" 44>; 45 46def FeatureCaymanISA : SubtargetFeature<"caymanISA", 47 "CaymanISA", 48 "true", 49 "Use Cayman ISA" 50>; 51 52def FeatureCFALUBug : SubtargetFeature<"cfalubug", 53 "CFALUBug", 54 "true", 55 "GPU has CF_ALU bug" 56>; 57 58def FeatureFlatAddressSpace : SubtargetFeature<"flat-address-space", 59 "FlatAddressSpace", 60 "true", 61 "Support flat address space" 62>; 63 64def FeatureFlatInstOffsets : SubtargetFeature<"flat-inst-offsets", 65 "FlatInstOffsets", 66 "true", 67 "Flat instructions have immediate offset addressing mode" 68>; 69 70def FeatureFlatGlobalInsts : SubtargetFeature<"flat-global-insts", 71 "FlatGlobalInsts", 72 "true", 73 "Have global_* flat memory instructions" 74>; 75 76def FeatureFlatScratchInsts : SubtargetFeature<"flat-scratch-insts", 77 "FlatScratchInsts", 78 "true", 79 "Have scratch_* flat memory instructions" 80>; 81 82def FeatureUnalignedBufferAccess : SubtargetFeature<"unaligned-buffer-access", 83 "UnalignedBufferAccess", 84 "true", 85 "Support unaligned global loads and stores" 86>; 87 88def FeatureTrapHandler: SubtargetFeature<"trap-handler", 89 "TrapHandler", 90 "true", 91 "Trap handler support" 92>; 93 94def FeatureUnalignedScratchAccess : SubtargetFeature<"unaligned-scratch-access", 95 "UnalignedScratchAccess", 96 "true", 97 "Support unaligned scratch loads and stores" 98>; 99 100def FeatureApertureRegs : SubtargetFeature<"aperture-regs", 101 "HasApertureRegs", 102 "true", 103 "Has Memory Aperture Base and Size Registers" 104>; 105 106// XNACK is disabled if SH_MEM_CONFIG.ADDRESS_MODE = GPUVM on chips that support 107// XNACK. The current default kernel driver setting is: 108// - graphics ring: XNACK disabled 109// - compute ring: XNACK enabled 110// 111// If XNACK is enabled, the VMEM latency can be worse. 112// If XNACK is disabled, the 2 SGPRs can be used for general purposes. 113def FeatureXNACK : SubtargetFeature<"xnack", 114 "EnableXNACK", 115 "true", 116 "Enable XNACK support" 117>; 118 119def FeatureSGPRInitBug : SubtargetFeature<"sgpr-init-bug", 120 "SGPRInitBug", 121 "true", 122 "VI SGPR initilization bug requiring a fixed SGPR allocation size" 123>; 124 125class SubtargetFeatureFetchLimit <string Value> : 126 SubtargetFeature <"fetch"#Value, 127 "TexVTXClauseSize", 128 Value, 129 "Limit the maximum number of fetches in a clause to "#Value 130>; 131 132def FeatureFetchLimit8 : SubtargetFeatureFetchLimit <"8">; 133def FeatureFetchLimit16 : SubtargetFeatureFetchLimit <"16">; 134 135class SubtargetFeatureWavefrontSize <int Value> : SubtargetFeature< 136 "wavefrontsize"#Value, 137 "WavefrontSize", 138 !cast<string>(Value), 139 "The number of threads per wavefront" 140>; 141 142def FeatureWavefrontSize16 : SubtargetFeatureWavefrontSize<16>; 143def FeatureWavefrontSize32 : SubtargetFeatureWavefrontSize<32>; 144def FeatureWavefrontSize64 : SubtargetFeatureWavefrontSize<64>; 145 146class SubtargetFeatureLDSBankCount <int Value> : SubtargetFeature < 147 "ldsbankcount"#Value, 148 "LDSBankCount", 149 !cast<string>(Value), 150 "The number of LDS banks per compute unit." 151>; 152 153def FeatureLDSBankCount16 : SubtargetFeatureLDSBankCount<16>; 154def FeatureLDSBankCount32 : SubtargetFeatureLDSBankCount<32>; 155 156class SubtargetFeatureLocalMemorySize <int Value> : SubtargetFeature< 157 "localmemorysize"#Value, 158 "LocalMemorySize", 159 !cast<string>(Value), 160 "The size of local memory in bytes" 161>; 162 163def FeatureGCN : SubtargetFeature<"gcn", 164 "IsGCN", 165 "true", 166 "GCN or newer GPU" 167>; 168 169def FeatureGCN1Encoding : SubtargetFeature<"gcn1-encoding", 170 "GCN1Encoding", 171 "true", 172 "Encoding format for SI and CI" 173>; 174 175def FeatureGCN3Encoding : SubtargetFeature<"gcn3-encoding", 176 "GCN3Encoding", 177 "true", 178 "Encoding format for VI" 179>; 180 181def FeatureCIInsts : SubtargetFeature<"ci-insts", 182 "CIInsts", 183 "true", 184 "Additional intstructions for CI+" 185>; 186 187def FeatureGFX9Insts : SubtargetFeature<"gfx9-insts", 188 "GFX9Insts", 189 "true", 190 "Additional intstructions for GFX9+" 191>; 192 193def FeatureSMemRealTime : SubtargetFeature<"s-memrealtime", 194 "HasSMemRealTime", 195 "true", 196 "Has s_memrealtime instruction" 197>; 198 199def FeatureInv2PiInlineImm : SubtargetFeature<"inv-2pi-inline-imm", 200 "HasInv2PiInlineImm", 201 "true", 202 "Has 1 / (2 * pi) as inline immediate" 203>; 204 205def Feature16BitInsts : SubtargetFeature<"16-bit-insts", 206 "Has16BitInsts", 207 "true", 208 "Has i16/f16 instructions" 209>; 210 211def FeatureVOP3P : SubtargetFeature<"vop3p", 212 "HasVOP3PInsts", 213 "true", 214 "Has VOP3P packed instructions" 215>; 216 217def FeatureMovrel : SubtargetFeature<"movrel", 218 "HasMovrel", 219 "true", 220 "Has v_movrel*_b32 instructions" 221>; 222 223def FeatureVGPRIndexMode : SubtargetFeature<"vgpr-index-mode", 224 "HasVGPRIndexMode", 225 "true", 226 "Has VGPR mode register indexing" 227>; 228 229def FeatureScalarStores : SubtargetFeature<"scalar-stores", 230 "HasScalarStores", 231 "true", 232 "Has store scalar memory instructions" 233>; 234 235def FeatureSDWA : SubtargetFeature<"sdwa", 236 "HasSDWA", 237 "true", 238 "Support SDWA (Sub-DWORD Addressing) extension" 239>; 240 241def FeatureSDWAOmod : SubtargetFeature<"sdwa-omod", 242 "HasSDWAOmod", 243 "true", 244 "Support OMod with SDWA (Sub-DWORD Addressing) extension" 245>; 246 247def FeatureSDWAScalar : SubtargetFeature<"sdwa-scalar", 248 "HasSDWAScalar", 249 "true", 250 "Support scalar register with SDWA (Sub-DWORD Addressing) extension" 251>; 252 253def FeatureSDWASdst : SubtargetFeature<"sdwa-sdst", 254 "HasSDWASdst", 255 "true", 256 "Support scalar dst for VOPC with SDWA (Sub-DWORD Addressing) extension" 257>; 258 259def FeatureSDWAMac : SubtargetFeature<"sdwa-mav", 260 "HasSDWAMac", 261 "true", 262 "Support v_mac_f32/f16 with SDWA (Sub-DWORD Addressing) extension" 263>; 264 265def FeatureSDWAClampVOPC : SubtargetFeature<"sdwa-clamp-vopc", 266 "HasSDWAClampVOPC", 267 "true", 268 "Support clamp for VOPC with SDWA (Sub-DWORD Addressing) extension" 269>; 270 271def FeatureDPP : SubtargetFeature<"dpp", 272 "HasDPP", 273 "true", 274 "Support DPP (Data Parallel Primitives) extension" 275>; 276 277//===------------------------------------------------------------===// 278// Subtarget Features (options and debugging) 279//===------------------------------------------------------------===// 280 281// Some instructions do not support denormals despite this flag. Using 282// fp32 denormals also causes instructions to run at the double 283// precision rate for the device. 284def FeatureFP32Denormals : SubtargetFeature<"fp32-denormals", 285 "FP32Denormals", 286 "true", 287 "Enable single precision denormal handling" 288>; 289 290// Denormal handling for fp64 and fp16 is controlled by the same 291// config register when fp16 supported. 292// TODO: Do we need a separate f16 setting when not legal? 293def FeatureFP64FP16Denormals : SubtargetFeature<"fp64-fp16-denormals", 294 "FP64FP16Denormals", 295 "true", 296 "Enable double and half precision denormal handling", 297 [FeatureFP64] 298>; 299 300def FeatureFP64Denormals : SubtargetFeature<"fp64-denormals", 301 "FP64FP16Denormals", 302 "true", 303 "Enable double and half precision denormal handling", 304 [FeatureFP64, FeatureFP64FP16Denormals] 305>; 306 307def FeatureFP16Denormals : SubtargetFeature<"fp16-denormals", 308 "FP64FP16Denormals", 309 "true", 310 "Enable half precision denormal handling", 311 [FeatureFP64FP16Denormals] 312>; 313 314def FeatureDX10Clamp : SubtargetFeature<"dx10-clamp", 315 "DX10Clamp", 316 "true", 317 "clamp modifier clamps NaNs to 0.0" 318>; 319 320def FeatureFPExceptions : SubtargetFeature<"fp-exceptions", 321 "FPExceptions", 322 "true", 323 "Enable floating point exceptions" 324>; 325 326class FeatureMaxPrivateElementSize<int size> : SubtargetFeature< 327 "max-private-element-size-"#size, 328 "MaxPrivateElementSize", 329 !cast<string>(size), 330 "Maximum private access size may be "#size 331>; 332 333def FeatureMaxPrivateElementSize4 : FeatureMaxPrivateElementSize<4>; 334def FeatureMaxPrivateElementSize8 : FeatureMaxPrivateElementSize<8>; 335def FeatureMaxPrivateElementSize16 : FeatureMaxPrivateElementSize<16>; 336 337def FeatureVGPRSpilling : SubtargetFeature<"vgpr-spilling", 338 "EnableVGPRSpilling", 339 "true", 340 "Enable spilling of VGPRs to scratch memory" 341>; 342 343def FeatureDumpCode : SubtargetFeature <"DumpCode", 344 "DumpCode", 345 "true", 346 "Dump MachineInstrs in the CodeEmitter" 347>; 348 349def FeatureDumpCodeLower : SubtargetFeature <"dumpcode", 350 "DumpCode", 351 "true", 352 "Dump MachineInstrs in the CodeEmitter" 353>; 354 355def FeaturePromoteAlloca : SubtargetFeature <"promote-alloca", 356 "EnablePromoteAlloca", 357 "true", 358 "Enable promote alloca pass" 359>; 360 361// XXX - This should probably be removed once enabled by default 362def FeatureEnableLoadStoreOpt : SubtargetFeature <"load-store-opt", 363 "EnableLoadStoreOpt", 364 "true", 365 "Enable SI load/store optimizer pass" 366>; 367 368// Performance debugging feature. Allow using DS instruction immediate 369// offsets even if the base pointer can't be proven to be base. On SI, 370// base pointer values that won't give the same result as a 16-bit add 371// are not safe to fold, but this will override the conservative test 372// for the base pointer. 373def FeatureEnableUnsafeDSOffsetFolding : SubtargetFeature < 374 "unsafe-ds-offset-folding", 375 "EnableUnsafeDSOffsetFolding", 376 "true", 377 "Force using DS instruction immediate offsets on SI" 378>; 379 380def FeatureEnableSIScheduler : SubtargetFeature<"si-scheduler", 381 "EnableSIScheduler", 382 "true", 383 "Enable SI Machine Scheduler" 384>; 385 386// Unless +-flat-for-global is specified, turn on FlatForGlobal for 387// all OS-es on VI and newer hardware to avoid assertion failures due 388// to missing ADDR64 variants of MUBUF instructions. 389// FIXME: moveToVALU should be able to handle converting addr64 MUBUF 390// instructions. 391 392def FeatureFlatForGlobal : SubtargetFeature<"flat-for-global", 393 "FlatForGlobal", 394 "true", 395 "Force to generate flat instruction for global" 396>; 397 398def FeatureAutoWaitcntBeforeBarrier : SubtargetFeature < 399 "auto-waitcnt-before-barrier", 400 "AutoWaitcntBeforeBarrier", 401 "true", 402 "Hardware automatically inserts waitcnt before barrier" 403>; 404 405// Dummy feature used to disable assembler instructions. 406def FeatureDisable : SubtargetFeature<"", 407 "FeatureDisable","true", 408 "Dummy feature to disable assembler instructions" 409>; 410 411class SubtargetFeatureGeneration <string Value, 412 list<SubtargetFeature> Implies> : 413 SubtargetFeature <Value, "Gen", "AMDGPUSubtarget::"#Value, 414 Value#" GPU generation", Implies>; 415 416def FeatureLocalMemorySize0 : SubtargetFeatureLocalMemorySize<0>; 417def FeatureLocalMemorySize32768 : SubtargetFeatureLocalMemorySize<32768>; 418def FeatureLocalMemorySize65536 : SubtargetFeatureLocalMemorySize<65536>; 419 420def FeatureR600 : SubtargetFeatureGeneration<"R600", 421 [FeatureR600ALUInst, FeatureFetchLimit8, FeatureLocalMemorySize0] 422>; 423 424def FeatureR700 : SubtargetFeatureGeneration<"R700", 425 [FeatureFetchLimit16, FeatureLocalMemorySize0] 426>; 427 428def FeatureEvergreen : SubtargetFeatureGeneration<"EVERGREEN", 429 [FeatureFetchLimit16, FeatureLocalMemorySize32768] 430>; 431 432def FeatureNorthernIslands : SubtargetFeatureGeneration<"NORTHERN_ISLANDS", 433 [FeatureFetchLimit16, FeatureWavefrontSize64, 434 FeatureLocalMemorySize32768] 435>; 436 437def FeatureSouthernIslands : SubtargetFeatureGeneration<"SOUTHERN_ISLANDS", 438 [FeatureFP64, FeatureLocalMemorySize32768, 439 FeatureWavefrontSize64, FeatureGCN, FeatureGCN1Encoding, 440 FeatureLDSBankCount32, FeatureMovrel] 441>; 442 443def FeatureSeaIslands : SubtargetFeatureGeneration<"SEA_ISLANDS", 444 [FeatureFP64, FeatureLocalMemorySize65536, 445 FeatureWavefrontSize64, FeatureGCN, FeatureFlatAddressSpace, 446 FeatureGCN1Encoding, FeatureCIInsts, FeatureMovrel] 447>; 448 449def FeatureVolcanicIslands : SubtargetFeatureGeneration<"VOLCANIC_ISLANDS", 450 [FeatureFP64, FeatureLocalMemorySize65536, 451 FeatureWavefrontSize64, FeatureFlatAddressSpace, FeatureGCN, 452 FeatureGCN3Encoding, FeatureCIInsts, Feature16BitInsts, 453 FeatureSMemRealTime, FeatureVGPRIndexMode, FeatureMovrel, 454 FeatureScalarStores, FeatureInv2PiInlineImm, 455 FeatureSDWA, FeatureSDWAClampVOPC, FeatureSDWAMac, FeatureDPP 456 ] 457>; 458 459def FeatureGFX9 : SubtargetFeatureGeneration<"GFX9", 460 [FeatureFP64, FeatureLocalMemorySize65536, 461 FeatureWavefrontSize64, FeatureFlatAddressSpace, FeatureGCN, 462 FeatureGCN3Encoding, FeatureCIInsts, Feature16BitInsts, 463 FeatureSMemRealTime, FeatureScalarStores, FeatureInv2PiInlineImm, 464 FeatureApertureRegs, FeatureGFX9Insts, FeatureVOP3P, FeatureVGPRIndexMode, 465 FeatureFastFMAF32, FeatureDPP, 466 FeatureSDWA, FeatureSDWAOmod, FeatureSDWAScalar, FeatureSDWASdst, 467 FeatureFlatInstOffsets, FeatureFlatGlobalInsts, FeatureFlatScratchInsts 468 ] 469>; 470 471class SubtargetFeatureISAVersion <int Major, int Minor, int Stepping, 472 list<SubtargetFeature> Implies> 473 : SubtargetFeature < 474 "isaver"#Major#"."#Minor#"."#Stepping, 475 "IsaVersion", 476 "ISAVersion"#Major#"_"#Minor#"_"#Stepping, 477 "Instruction set version number", 478 Implies 479>; 480 481def FeatureISAVersion6_0_0 : SubtargetFeatureISAVersion <6,0,0, 482 [FeatureSouthernIslands, 483 FeatureFastFMAF32, 484 HalfRate64Ops, 485 FeatureLDSBankCount32]>; 486 487def FeatureISAVersion6_0_1 : SubtargetFeatureISAVersion <6,0,1, 488 [FeatureSouthernIslands, 489 FeatureLDSBankCount32]>; 490 491def FeatureISAVersion7_0_0 : SubtargetFeatureISAVersion <7,0,0, 492 [FeatureSeaIslands, 493 FeatureLDSBankCount32]>; 494 495def FeatureISAVersion7_0_1 : SubtargetFeatureISAVersion <7,0,1, 496 [FeatureSeaIslands, 497 HalfRate64Ops, 498 FeatureLDSBankCount32, 499 FeatureFastFMAF32]>; 500 501def FeatureISAVersion7_0_2 : SubtargetFeatureISAVersion <7,0,2, 502 [FeatureSeaIslands, 503 FeatureLDSBankCount16]>; 504 505def FeatureISAVersion7_0_3 : SubtargetFeatureISAVersion <7,0,3, 506 [FeatureSeaIslands, 507 FeatureLDSBankCount16]>; 508 509def FeatureISAVersion8_0_0 : SubtargetFeatureISAVersion <8,0,0, 510 [FeatureVolcanicIslands, 511 FeatureLDSBankCount32, 512 FeatureSGPRInitBug]>; 513 514def FeatureISAVersion8_0_1 : SubtargetFeatureISAVersion <8,0,1, 515 [FeatureVolcanicIslands, 516 FeatureLDSBankCount32, 517 FeatureXNACK]>; 518 519def FeatureISAVersion8_0_2 : SubtargetFeatureISAVersion <8,0,2, 520 [FeatureVolcanicIslands, 521 FeatureLDSBankCount32, 522 FeatureSGPRInitBug]>; 523 524def FeatureISAVersion8_0_3 : SubtargetFeatureISAVersion <8,0,3, 525 [FeatureVolcanicIslands, 526 FeatureLDSBankCount32]>; 527 528def FeatureISAVersion8_0_4 : SubtargetFeatureISAVersion <8,0,4, 529 [FeatureVolcanicIslands, 530 FeatureLDSBankCount32]>; 531 532def FeatureISAVersion8_1_0 : SubtargetFeatureISAVersion <8,1,0, 533 [FeatureVolcanicIslands, 534 FeatureLDSBankCount16, 535 FeatureXNACK]>; 536 537def FeatureISAVersion9_0_0 : SubtargetFeatureISAVersion <9,0,0, 538 [FeatureGFX9, 539 FeatureLDSBankCount32]>; 540 541def FeatureISAVersion9_0_1 : SubtargetFeatureISAVersion <9,0,1, 542 [FeatureGFX9, 543 FeatureLDSBankCount32, 544 FeatureXNACK]>; 545 546def FeatureISAVersion9_0_2 : SubtargetFeatureISAVersion <9,0,2, 547 [FeatureGFX9, 548 FeatureLDSBankCount32]>; 549 550def FeatureISAVersion9_0_3 : SubtargetFeatureISAVersion <9,0,3, 551 [FeatureGFX9, 552 FeatureLDSBankCount32, 553 FeatureXNACK]>; 554 555//===----------------------------------------------------------------------===// 556// Debugger related subtarget features. 557//===----------------------------------------------------------------------===// 558 559def FeatureDebuggerInsertNops : SubtargetFeature< 560 "amdgpu-debugger-insert-nops", 561 "DebuggerInsertNops", 562 "true", 563 "Insert one nop instruction for each high level source statement" 564>; 565 566def FeatureDebuggerReserveRegs : SubtargetFeature< 567 "amdgpu-debugger-reserve-regs", 568 "DebuggerReserveRegs", 569 "true", 570 "Reserve registers for debugger usage" 571>; 572 573def FeatureDebuggerEmitPrologue : SubtargetFeature< 574 "amdgpu-debugger-emit-prologue", 575 "DebuggerEmitPrologue", 576 "true", 577 "Emit debugger prologue" 578>; 579 580//===----------------------------------------------------------------------===// 581 582def AMDGPUInstrInfo : InstrInfo { 583 let guessInstructionProperties = 1; 584 let noNamedPositionallyEncodedOperands = 1; 585} 586 587def AMDGPUAsmParser : AsmParser { 588 // Some of the R600 registers have the same name, so this crashes. 589 // For example T0_XYZW and T0_XY both have the asm name T0. 590 let ShouldEmitMatchRegisterName = 0; 591} 592 593def AMDGPUAsmWriter : AsmWriter { 594 int PassSubtarget = 1; 595} 596 597def AMDGPUAsmVariants { 598 string Default = "Default"; 599 int Default_ID = 0; 600 string VOP3 = "VOP3"; 601 int VOP3_ID = 1; 602 string SDWA = "SDWA"; 603 int SDWA_ID = 2; 604 string SDWA9 = "SDWA9"; 605 int SDWA9_ID = 3; 606 string DPP = "DPP"; 607 int DPP_ID = 4; 608 string Disable = "Disable"; 609 int Disable_ID = 5; 610} 611 612def DefaultAMDGPUAsmParserVariant : AsmParserVariant { 613 let Variant = AMDGPUAsmVariants.Default_ID; 614 let Name = AMDGPUAsmVariants.Default; 615} 616 617def VOP3AsmParserVariant : AsmParserVariant { 618 let Variant = AMDGPUAsmVariants.VOP3_ID; 619 let Name = AMDGPUAsmVariants.VOP3; 620} 621 622def SDWAAsmParserVariant : AsmParserVariant { 623 let Variant = AMDGPUAsmVariants.SDWA_ID; 624 let Name = AMDGPUAsmVariants.SDWA; 625} 626 627def SDWA9AsmParserVariant : AsmParserVariant { 628 let Variant = AMDGPUAsmVariants.SDWA9_ID; 629 let Name = AMDGPUAsmVariants.SDWA9; 630} 631 632 633def DPPAsmParserVariant : AsmParserVariant { 634 let Variant = AMDGPUAsmVariants.DPP_ID; 635 let Name = AMDGPUAsmVariants.DPP; 636} 637 638def AMDGPU : Target { 639 // Pull in Instruction Info: 640 let InstructionSet = AMDGPUInstrInfo; 641 let AssemblyParsers = [AMDGPUAsmParser]; 642 let AssemblyParserVariants = [DefaultAMDGPUAsmParserVariant, 643 VOP3AsmParserVariant, 644 SDWAAsmParserVariant, 645 SDWA9AsmParserVariant, 646 DPPAsmParserVariant]; 647 let AssemblyWriters = [AMDGPUAsmWriter]; 648} 649 650// Dummy Instruction itineraries for pseudo instructions 651def ALU_NULL : FuncUnit; 652def NullALU : InstrItinClass; 653 654//===----------------------------------------------------------------------===// 655// Predicate helper class 656//===----------------------------------------------------------------------===// 657 658def TruePredicate : Predicate<"true">; 659 660def isSICI : Predicate< 661 "Subtarget->getGeneration() == AMDGPUSubtarget::SOUTHERN_ISLANDS ||" 662 "Subtarget->getGeneration() == AMDGPUSubtarget::SEA_ISLANDS" 663>, AssemblerPredicate<"FeatureGCN1Encoding">; 664 665def isVI : Predicate < 666 "Subtarget->getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS">, 667 AssemblerPredicate<"FeatureGCN3Encoding">; 668 669def isGFX9 : Predicate < 670 "Subtarget->getGeneration() >= AMDGPUSubtarget::GFX9">, 671 AssemblerPredicate<"FeatureGFX9Insts">; 672 673// TODO: Either the name to be changed or we simply use IsCI! 674def isCIVI : Predicate < 675 "Subtarget->getGeneration() >= AMDGPUSubtarget::SEA_ISLANDS">, 676 AssemblerPredicate<"FeatureCIInsts">; 677 678def HasFlatAddressSpace : Predicate<"Subtarget->hasFlatAddressSpace()">, 679 AssemblerPredicate<"FeatureFlatAddressSpace">; 680 681def HasFlatGlobalInsts : Predicate<"Subtarget->hasFlatGlobalInsts()">, 682 AssemblerPredicate<"FeatureFlatGlobalInsts">; 683 684def Has16BitInsts : Predicate<"Subtarget->has16BitInsts()">, 685 AssemblerPredicate<"Feature16BitInsts">; 686def HasVOP3PInsts : Predicate<"Subtarget->hasVOP3PInsts()">, 687 AssemblerPredicate<"FeatureVOP3P">; 688 689def HasSDWA : Predicate<"Subtarget->hasSDWA()">, 690 AssemblerPredicate<"FeatureSDWA,FeatureVolcanicIslands">; 691 692def HasSDWA9 : Predicate<"Subtarget->hasSDWA()">, 693 AssemblerPredicate<"FeatureSDWA,FeatureGFX9">; 694 695def HasDPP : Predicate<"Subtarget->hasDPP()">, 696 AssemblerPredicate<"FeatureDPP">; 697 698class PredicateControl { 699 Predicate SubtargetPredicate; 700 Predicate SIAssemblerPredicate = isSICI; 701 Predicate VIAssemblerPredicate = isVI; 702 list<Predicate> AssemblerPredicates = []; 703 Predicate AssemblerPredicate = TruePredicate; 704 list<Predicate> OtherPredicates = []; 705 list<Predicate> Predicates = !listconcat([SubtargetPredicate, AssemblerPredicate], 706 AssemblerPredicates, 707 OtherPredicates); 708} 709 710// Include AMDGPU TD files 711include "R600Schedule.td" 712include "SISchedule.td" 713include "Processors.td" 714include "AMDGPUInstrInfo.td" 715include "AMDGPUIntrinsics.td" 716include "AMDGPURegisterInfo.td" 717include "AMDGPURegisterBanks.td" 718include "AMDGPUInstructions.td" 719include "AMDGPUCallingConv.td" 720