1 //===-- SIDefines.h - SI Helper Macros ----------------------*- C++ -*-===// 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 /// \file 8 //===----------------------------------------------------------------------===// 9 10 #ifndef LLVM_LIB_TARGET_AMDGPU_SIDEFINES_H 11 #define LLVM_LIB_TARGET_AMDGPU_SIDEFINES_H 12 13 #include "llvm/MC/MCInstrDesc.h" 14 15 namespace llvm { 16 17 // This needs to be kept in sync with the field bits in SIRegisterClass. 18 enum SIRCFlags : uint8_t { 19 RegTupleAlignUnitsWidth = 2, 20 HasVGPRBit = RegTupleAlignUnitsWidth, 21 HasAGPRBit, 22 HasSGPRbit, 23 24 HasVGPR = 1 << HasVGPRBit, 25 HasAGPR = 1 << HasAGPRBit, 26 HasSGPR = 1 << HasSGPRbit, 27 28 RegTupleAlignUnitsMask = (1 << RegTupleAlignUnitsWidth) - 1, 29 RegKindMask = (HasVGPR | HasAGPR | HasSGPR) 30 }; // enum SIRCFlagsr 31 32 namespace SIEncodingFamily { 33 // This must be kept in sync with the SIEncodingFamily class in SIInstrInfo.td 34 // and the columns of the getMCOpcodeGen table. 35 enum { 36 SI = 0, 37 VI = 1, 38 SDWA = 2, 39 SDWA9 = 3, 40 GFX80 = 4, 41 GFX9 = 5, 42 GFX10 = 6, 43 SDWA10 = 7, 44 GFX90A = 8, 45 GFX940 = 9, 46 GFX11 = 10, 47 GFX12 = 11, 48 }; 49 } 50 51 namespace SIInstrFlags { 52 // This needs to be kept in sync with the field bits in InstSI. 53 enum : uint64_t { 54 // Low bits - basic encoding information. 55 SALU = 1 << 0, 56 VALU = 1 << 1, 57 58 // SALU instruction formats. 59 SOP1 = 1 << 2, 60 SOP2 = 1 << 3, 61 SOPC = 1 << 4, 62 SOPK = 1 << 5, 63 SOPP = 1 << 6, 64 65 // VALU instruction formats. 66 VOP1 = 1 << 7, 67 VOP2 = 1 << 8, 68 VOPC = 1 << 9, 69 70 // TODO: Should this be spilt into VOP3 a and b? 71 VOP3 = 1 << 10, 72 VOP3P = 1 << 12, 73 74 VINTRP = 1 << 13, 75 SDWA = 1 << 14, 76 DPP = 1 << 15, 77 TRANS = 1 << 16, 78 79 // Memory instruction formats. 80 MUBUF = 1 << 17, 81 MTBUF = 1 << 18, 82 SMRD = 1 << 19, 83 MIMG = 1 << 20, 84 VIMAGE = 1 << 21, 85 VSAMPLE = 1 << 22, 86 EXP = 1 << 23, 87 FLAT = 1 << 24, 88 DS = 1 << 25, 89 90 // Combined SGPR/VGPR Spill bit 91 // Logic to separate them out is done in isSGPRSpill and isVGPRSpill 92 Spill = 1 << 26, 93 94 // LDSDIR instruction format. 95 LDSDIR = 1 << 28, 96 97 // VINTERP instruction format. 98 VINTERP = 1 << 29, 99 100 // High bits - other information. 101 VM_CNT = UINT64_C(1) << 32, 102 EXP_CNT = UINT64_C(1) << 33, 103 LGKM_CNT = UINT64_C(1) << 34, 104 105 WQM = UINT64_C(1) << 35, 106 DisableWQM = UINT64_C(1) << 36, 107 Gather4 = UINT64_C(1) << 37, 108 109 // Reserved, must be 0. 110 Reserved0 = UINT64_C(1) << 38, 111 112 SCALAR_STORE = UINT64_C(1) << 39, 113 FIXED_SIZE = UINT64_C(1) << 40, 114 115 // Reserved, must be 0. 116 Reserved1 = UINT64_C(1) << 41, 117 118 VOP3_OPSEL = UINT64_C(1) << 42, 119 maybeAtomic = UINT64_C(1) << 43, 120 renamedInGFX9 = UINT64_C(1) << 44, 121 122 // Is a clamp on FP type. 123 FPClamp = UINT64_C(1) << 45, 124 125 // Is an integer clamp 126 IntClamp = UINT64_C(1) << 46, 127 128 // Clamps lo component of register. 129 ClampLo = UINT64_C(1) << 47, 130 131 // Clamps hi component of register. 132 // ClampLo and ClampHi set for packed clamp. 133 ClampHi = UINT64_C(1) << 48, 134 135 // Is a packed VOP3P instruction. 136 IsPacked = UINT64_C(1) << 49, 137 138 // Is a D16 buffer instruction. 139 D16Buf = UINT64_C(1) << 50, 140 141 // FLAT instruction accesses FLAT_GLBL segment. 142 FlatGlobal = UINT64_C(1) << 51, 143 144 // Uses floating point double precision rounding mode 145 FPDPRounding = UINT64_C(1) << 52, 146 147 // Instruction is FP atomic. 148 FPAtomic = UINT64_C(1) << 53, 149 150 // Is a MFMA instruction. 151 IsMAI = UINT64_C(1) << 54, 152 153 // Is a DOT instruction. 154 IsDOT = UINT64_C(1) << 55, 155 156 // FLAT instruction accesses FLAT_SCRATCH segment. 157 FlatScratch = UINT64_C(1) << 56, 158 159 // Atomic without return. 160 IsAtomicNoRet = UINT64_C(1) << 57, 161 162 // Atomic with return. 163 IsAtomicRet = UINT64_C(1) << 58, 164 165 // Is a WMMA instruction. 166 IsWMMA = UINT64_C(1) << 59, 167 168 // Whether tied sources will be read. 169 TiedSourceNotRead = UINT64_C(1) << 60, 170 171 // Is never uniform. 172 IsNeverUniform = UINT64_C(1) << 61, 173 174 // ds_gws_* instructions. 175 GWS = UINT64_C(1) << 62, 176 177 // Is a SWMMAC instruction. 178 IsSWMMAC = UINT64_C(1) << 63, 179 }; 180 181 // v_cmp_class_* etc. use a 10-bit mask for what operation is checked. 182 // The result is true if any of these tests are true. 183 enum ClassFlags : unsigned { 184 S_NAN = 1 << 0, // Signaling NaN 185 Q_NAN = 1 << 1, // Quiet NaN 186 N_INFINITY = 1 << 2, // Negative infinity 187 N_NORMAL = 1 << 3, // Negative normal 188 N_SUBNORMAL = 1 << 4, // Negative subnormal 189 N_ZERO = 1 << 5, // Negative zero 190 P_ZERO = 1 << 6, // Positive zero 191 P_SUBNORMAL = 1 << 7, // Positive subnormal 192 P_NORMAL = 1 << 8, // Positive normal 193 P_INFINITY = 1 << 9 // Positive infinity 194 }; 195 } 196 197 namespace AMDGPU { 198 enum OperandType : unsigned { 199 /// Operands with register or 32-bit immediate 200 OPERAND_REG_IMM_INT32 = MCOI::OPERAND_FIRST_TARGET, 201 OPERAND_REG_IMM_INT64, 202 OPERAND_REG_IMM_INT16, 203 OPERAND_REG_IMM_FP32, 204 OPERAND_REG_IMM_FP64, 205 OPERAND_REG_IMM_BF16, 206 OPERAND_REG_IMM_FP16, 207 OPERAND_REG_IMM_BF16_DEFERRED, 208 OPERAND_REG_IMM_FP16_DEFERRED, 209 OPERAND_REG_IMM_FP32_DEFERRED, 210 OPERAND_REG_IMM_V2BF16, 211 OPERAND_REG_IMM_V2FP16, 212 OPERAND_REG_IMM_V2INT16, 213 OPERAND_REG_IMM_V2INT32, 214 OPERAND_REG_IMM_V2FP32, 215 216 /// Operands with register or inline constant 217 OPERAND_REG_INLINE_C_INT16, 218 OPERAND_REG_INLINE_C_INT32, 219 OPERAND_REG_INLINE_C_INT64, 220 OPERAND_REG_INLINE_C_BF16, 221 OPERAND_REG_INLINE_C_FP16, 222 OPERAND_REG_INLINE_C_FP32, 223 OPERAND_REG_INLINE_C_FP64, 224 OPERAND_REG_INLINE_C_V2INT16, 225 OPERAND_REG_INLINE_C_V2BF16, 226 OPERAND_REG_INLINE_C_V2FP16, 227 OPERAND_REG_INLINE_C_V2INT32, 228 OPERAND_REG_INLINE_C_V2FP32, 229 230 // Operand for split barrier inline constant 231 OPERAND_INLINE_SPLIT_BARRIER_INT32, 232 233 /// Operand with 32-bit immediate that uses the constant bus. 234 OPERAND_KIMM32, 235 OPERAND_KIMM16, 236 237 /// Operands with an AccVGPR register or inline constant 238 OPERAND_REG_INLINE_AC_INT16, 239 OPERAND_REG_INLINE_AC_INT32, 240 OPERAND_REG_INLINE_AC_BF16, 241 OPERAND_REG_INLINE_AC_FP16, 242 OPERAND_REG_INLINE_AC_FP32, 243 OPERAND_REG_INLINE_AC_FP64, 244 OPERAND_REG_INLINE_AC_V2INT16, 245 OPERAND_REG_INLINE_AC_V2BF16, 246 OPERAND_REG_INLINE_AC_V2FP16, 247 OPERAND_REG_INLINE_AC_V2INT32, 248 OPERAND_REG_INLINE_AC_V2FP32, 249 250 // Operand for source modifiers for VOP instructions 251 OPERAND_INPUT_MODS, 252 253 // Operand for SDWA instructions 254 OPERAND_SDWA_VOPC_DST, 255 256 OPERAND_REG_IMM_FIRST = OPERAND_REG_IMM_INT32, 257 OPERAND_REG_IMM_LAST = OPERAND_REG_IMM_V2FP32, 258 259 OPERAND_REG_INLINE_C_FIRST = OPERAND_REG_INLINE_C_INT16, 260 OPERAND_REG_INLINE_C_LAST = OPERAND_REG_INLINE_AC_V2FP32, 261 262 OPERAND_REG_INLINE_AC_FIRST = OPERAND_REG_INLINE_AC_INT16, 263 OPERAND_REG_INLINE_AC_LAST = OPERAND_REG_INLINE_AC_V2FP32, 264 265 OPERAND_SRC_FIRST = OPERAND_REG_IMM_INT32, 266 OPERAND_SRC_LAST = OPERAND_REG_INLINE_C_LAST, 267 268 OPERAND_KIMM_FIRST = OPERAND_KIMM32, 269 OPERAND_KIMM_LAST = OPERAND_KIMM16 270 271 }; 272 273 // Should be in sync with the OperandSemantics defined in SIRegisterInfo.td 274 enum OperandSemantics : unsigned { 275 INT = 0, 276 FP16 = 1, 277 BF16 = 2, 278 FP32 = 3, 279 FP64 = 4, 280 }; 281 } 282 283 // Input operand modifiers bit-masks 284 // NEG and SEXT share same bit-mask because they can't be set simultaneously. 285 namespace SISrcMods { 286 enum : unsigned { 287 NONE = 0, 288 NEG = 1 << 0, // Floating-point negate modifier 289 ABS = 1 << 1, // Floating-point absolute modifier 290 SEXT = 1 << 0, // Integer sign-extend modifier 291 NEG_HI = ABS, // Floating-point negate high packed component modifier. 292 OP_SEL_0 = 1 << 2, 293 OP_SEL_1 = 1 << 3, 294 DST_OP_SEL = 1 << 3 // VOP3 dst op_sel (share mask with OP_SEL_1) 295 }; 296 } 297 298 namespace SIOutMods { 299 enum : unsigned { 300 NONE = 0, 301 MUL2 = 1, 302 MUL4 = 2, 303 DIV2 = 3 304 }; 305 } 306 307 namespace AMDGPU { 308 namespace VGPRIndexMode { 309 310 enum Id : unsigned { // id of symbolic names 311 ID_SRC0 = 0, 312 ID_SRC1, 313 ID_SRC2, 314 ID_DST, 315 316 ID_MIN = ID_SRC0, 317 ID_MAX = ID_DST 318 }; 319 320 enum EncBits : unsigned { 321 OFF = 0, 322 SRC0_ENABLE = 1 << ID_SRC0, 323 SRC1_ENABLE = 1 << ID_SRC1, 324 SRC2_ENABLE = 1 << ID_SRC2, 325 DST_ENABLE = 1 << ID_DST, 326 ENABLE_MASK = SRC0_ENABLE | SRC1_ENABLE | SRC2_ENABLE | DST_ENABLE, 327 UNDEF = 0xFFFF 328 }; 329 330 } // namespace VGPRIndexMode 331 } // namespace AMDGPU 332 333 namespace AMDGPUAsmVariants { 334 enum : unsigned { 335 DEFAULT = 0, 336 VOP3 = 1, 337 SDWA = 2, 338 SDWA9 = 3, 339 DPP = 4, 340 VOP3_DPP = 5 341 }; 342 } // namespace AMDGPUAsmVariants 343 344 namespace AMDGPU { 345 namespace EncValues { // Encoding values of enum9/8/7 operands 346 347 enum : unsigned { 348 SGPR_MIN = 0, 349 SGPR_MAX_SI = 101, 350 SGPR_MAX_GFX10 = 105, 351 TTMP_VI_MIN = 112, 352 TTMP_VI_MAX = 123, 353 TTMP_GFX9PLUS_MIN = 108, 354 TTMP_GFX9PLUS_MAX = 123, 355 INLINE_INTEGER_C_MIN = 128, 356 INLINE_INTEGER_C_POSITIVE_MAX = 192, // 64 357 INLINE_INTEGER_C_MAX = 208, 358 INLINE_FLOATING_C_MIN = 240, 359 INLINE_FLOATING_C_MAX = 248, 360 LITERAL_CONST = 255, 361 VGPR_MIN = 256, 362 VGPR_MAX = 511, 363 IS_VGPR = 256, // Indicates VGPR or AGPR 364 }; 365 366 } // namespace EncValues 367 368 // Register codes as defined in the TableGen's HWEncoding field. 369 namespace HWEncoding { 370 enum : unsigned { 371 REG_IDX_MASK = 0xff, 372 IS_VGPR = 1 << 8, 373 IS_AGPR = 1 << 9, 374 IS_HI16 = 1 << 10, 375 }; 376 } // namespace HWEncoding 377 378 namespace CPol { 379 380 enum CPol { 381 GLC = 1, 382 SLC = 2, 383 DLC = 4, 384 SCC = 16, 385 SC0 = GLC, 386 SC1 = SCC, 387 NT = SLC, 388 ALL_pregfx12 = GLC | SLC | DLC | SCC, 389 SWZ_pregfx12 = 8, 390 391 // Below are GFX12+ cache policy bits 392 393 // Temporal hint 394 TH = 0x7, // All TH bits 395 TH_RT = 0, // regular 396 TH_NT = 1, // non-temporal 397 TH_HT = 2, // high-temporal 398 TH_LU = 3, // last use 399 TH_RT_WB = 3, // regular (CU, SE), high-temporal with write-back (MALL) 400 TH_NT_RT = 4, // non-temporal (CU, SE), regular (MALL) 401 TH_RT_NT = 5, // regular (CU, SE), non-temporal (MALL) 402 TH_NT_HT = 6, // non-temporal (CU, SE), high-temporal (MALL) 403 TH_NT_WB = 7, // non-temporal (CU, SE), high-temporal with write-back (MALL) 404 TH_BYPASS = 3, // only to be used with scope = 3 405 406 TH_RESERVED = 7, // unused value for load insts 407 408 // Bits of TH for atomics 409 TH_ATOMIC_RETURN = GLC, // Returning vs non-returning 410 TH_ATOMIC_NT = SLC, // Non-temporal vs regular 411 TH_ATOMIC_CASCADE = 4, // Cascading vs regular 412 413 // Scope 414 SCOPE = 0x3 << 3, // All Scope bits 415 SCOPE_CU = 0 << 3, 416 SCOPE_SE = 1 << 3, 417 SCOPE_DEV = 2 << 3, 418 SCOPE_SYS = 3 << 3, 419 420 SWZ = 1 << 6, // Swizzle bit 421 422 ALL = TH | SCOPE, 423 424 // Helper bits 425 TH_TYPE_LOAD = 1 << 7, // TH_LOAD policy 426 TH_TYPE_STORE = 1 << 8, // TH_STORE policy 427 TH_TYPE_ATOMIC = 1 << 9, // TH_ATOMIC policy 428 TH_REAL_BYPASS = 1 << 10, // is TH=3 bypass policy or not 429 430 // Volatile (used to preserve/signal operation volatility for buffer 431 // operations not a real instruction bit) 432 VOLATILE = 1 << 31, 433 }; 434 435 } // namespace CPol 436 437 namespace SendMsg { // Encoding of SIMM16 used in s_sendmsg* insns. 438 439 enum Id { // Message ID, width(4) [3:0]. 440 ID_INTERRUPT = 1, 441 442 ID_GS_PreGFX11 = 2, // replaced in GFX11 443 ID_GS_DONE_PreGFX11 = 3, // replaced in GFX11 444 445 ID_HS_TESSFACTOR_GFX11Plus = 2, // reused in GFX11 446 ID_DEALLOC_VGPRS_GFX11Plus = 3, // reused in GFX11 447 448 ID_SAVEWAVE = 4, // added in GFX8, removed in GFX11 449 ID_STALL_WAVE_GEN = 5, // added in GFX9, removed in GFX12 450 ID_HALT_WAVES = 6, // added in GFX9, removed in GFX12 451 ID_ORDERED_PS_DONE = 7, // added in GFX9, removed in GFX11 452 ID_EARLY_PRIM_DEALLOC = 8, // added in GFX9, removed in GFX10 453 ID_GS_ALLOC_REQ = 9, // added in GFX9 454 ID_GET_DOORBELL = 10, // added in GFX9, removed in GFX11 455 ID_GET_DDID = 11, // added in GFX10, removed in GFX11 456 ID_SYSMSG = 15, 457 458 ID_RTN_GET_DOORBELL = 128, 459 ID_RTN_GET_DDID = 129, 460 ID_RTN_GET_TMA = 130, 461 ID_RTN_GET_REALTIME = 131, 462 ID_RTN_SAVE_WAVE = 132, 463 ID_RTN_GET_TBA = 133, 464 ID_RTN_GET_TBA_TO_PC = 134, 465 ID_RTN_GET_SE_AID_ID = 135, 466 467 ID_MASK_PreGFX11_ = 0xF, 468 ID_MASK_GFX11Plus_ = 0xFF 469 }; 470 471 enum Op { // Both GS and SYS operation IDs. 472 OP_SHIFT_ = 4, 473 OP_NONE_ = 0, 474 // Bits used for operation encoding 475 OP_WIDTH_ = 3, 476 OP_MASK_ = (((1 << OP_WIDTH_) - 1) << OP_SHIFT_), 477 // GS operations are encoded in bits 5:4 478 OP_GS_NOP = 0, 479 OP_GS_CUT = 1, 480 OP_GS_EMIT = 2, 481 OP_GS_EMIT_CUT = 3, 482 OP_GS_FIRST_ = OP_GS_NOP, 483 // SYS operations are encoded in bits 6:4 484 OP_SYS_ECC_ERR_INTERRUPT = 1, 485 OP_SYS_REG_RD = 2, 486 OP_SYS_HOST_TRAP_ACK = 3, 487 OP_SYS_TTRACE_PC = 4, 488 OP_SYS_FIRST_ = OP_SYS_ECC_ERR_INTERRUPT, 489 }; 490 491 enum StreamId : unsigned { // Stream ID, (2) [9:8]. 492 STREAM_ID_NONE_ = 0, 493 STREAM_ID_DEFAULT_ = 0, 494 STREAM_ID_LAST_ = 4, 495 STREAM_ID_FIRST_ = STREAM_ID_DEFAULT_, 496 STREAM_ID_SHIFT_ = 8, 497 STREAM_ID_WIDTH_= 2, 498 STREAM_ID_MASK_ = (((1 << STREAM_ID_WIDTH_) - 1) << STREAM_ID_SHIFT_) 499 }; 500 501 } // namespace SendMsg 502 503 namespace Hwreg { // Encoding of SIMM16 used in s_setreg/getreg* insns. 504 505 enum Id { // HwRegCode, (6) [5:0] 506 ID_MODE = 1, 507 ID_STATUS = 2, 508 ID_TRAPSTS = 3, 509 ID_HW_ID = 4, 510 ID_GPR_ALLOC = 5, 511 ID_LDS_ALLOC = 6, 512 ID_IB_STS = 7, 513 ID_PERF_SNAPSHOT_DATA_gfx12 = 10, 514 ID_PERF_SNAPSHOT_PC_LO_gfx12 = 11, 515 ID_PERF_SNAPSHOT_PC_HI_gfx12 = 12, 516 ID_MEM_BASES = 15, 517 ID_TBA_LO = 16, 518 ID_TBA_HI = 17, 519 ID_TMA_LO = 18, 520 ID_TMA_HI = 19, 521 ID_FLAT_SCR_LO = 20, 522 ID_FLAT_SCR_HI = 21, 523 ID_XNACK_MASK = 22, 524 ID_HW_ID1 = 23, 525 ID_HW_ID2 = 24, 526 ID_POPS_PACKER = 25, 527 ID_PERF_SNAPSHOT_DATA_gfx11 = 27, 528 ID_SHADER_CYCLES = 29, 529 ID_SHADER_CYCLES_HI = 30, 530 ID_DVGPR_ALLOC_LO = 31, 531 ID_DVGPR_ALLOC_HI = 32, 532 533 // Register numbers reused in GFX11 534 ID_PERF_SNAPSHOT_PC_LO_gfx11 = 18, 535 ID_PERF_SNAPSHOT_PC_HI_gfx11 = 19, 536 537 // Register numbers reused in GFX12+ 538 ID_STATE_PRIV = 4, 539 ID_PERF_SNAPSHOT_DATA1 = 15, 540 ID_PERF_SNAPSHOT_DATA2 = 16, 541 ID_EXCP_FLAG_PRIV = 17, 542 ID_EXCP_FLAG_USER = 18, 543 ID_TRAP_CTRL = 19, 544 545 // GFX940 specific registers 546 ID_XCC_ID = 20, 547 ID_SQ_PERF_SNAPSHOT_DATA = 21, 548 ID_SQ_PERF_SNAPSHOT_DATA1 = 22, 549 ID_SQ_PERF_SNAPSHOT_PC_LO = 23, 550 ID_SQ_PERF_SNAPSHOT_PC_HI = 24, 551 }; 552 553 enum Offset : unsigned { // Offset, (5) [10:6] 554 OFFSET_MEM_VIOL = 8, 555 }; 556 557 enum ModeRegisterMasks : uint32_t { 558 FP_ROUND_MASK = 0xf << 0, // Bits 0..3 559 FP_DENORM_MASK = 0xf << 4, // Bits 4..7 560 DX10_CLAMP_MASK = 1 << 8, 561 IEEE_MODE_MASK = 1 << 9, 562 LOD_CLAMP_MASK = 1 << 10, 563 DEBUG_MASK = 1 << 11, 564 565 // EXCP_EN fields. 566 EXCP_EN_INVALID_MASK = 1 << 12, 567 EXCP_EN_INPUT_DENORMAL_MASK = 1 << 13, 568 EXCP_EN_FLOAT_DIV0_MASK = 1 << 14, 569 EXCP_EN_OVERFLOW_MASK = 1 << 15, 570 EXCP_EN_UNDERFLOW_MASK = 1 << 16, 571 EXCP_EN_INEXACT_MASK = 1 << 17, 572 EXCP_EN_INT_DIV0_MASK = 1 << 18, 573 574 GPR_IDX_EN_MASK = 1 << 27, 575 VSKIP_MASK = 1 << 28, 576 CSP_MASK = 0x7u << 29 // Bits 29..31 577 }; 578 579 } // namespace Hwreg 580 581 namespace MTBUFFormat { 582 583 enum DataFormat : int64_t { 584 DFMT_INVALID = 0, 585 DFMT_8, 586 DFMT_16, 587 DFMT_8_8, 588 DFMT_32, 589 DFMT_16_16, 590 DFMT_10_11_11, 591 DFMT_11_11_10, 592 DFMT_10_10_10_2, 593 DFMT_2_10_10_10, 594 DFMT_8_8_8_8, 595 DFMT_32_32, 596 DFMT_16_16_16_16, 597 DFMT_32_32_32, 598 DFMT_32_32_32_32, 599 DFMT_RESERVED_15, 600 601 DFMT_MIN = DFMT_INVALID, 602 DFMT_MAX = DFMT_RESERVED_15, 603 604 DFMT_UNDEF = -1, 605 DFMT_DEFAULT = DFMT_8, 606 607 DFMT_SHIFT = 0, 608 DFMT_MASK = 0xF 609 }; 610 611 enum NumFormat : int64_t { 612 NFMT_UNORM = 0, 613 NFMT_SNORM, 614 NFMT_USCALED, 615 NFMT_SSCALED, 616 NFMT_UINT, 617 NFMT_SINT, 618 NFMT_RESERVED_6, // VI and GFX9 619 NFMT_SNORM_OGL = NFMT_RESERVED_6, // SI and CI only 620 NFMT_FLOAT, 621 622 NFMT_MIN = NFMT_UNORM, 623 NFMT_MAX = NFMT_FLOAT, 624 625 NFMT_UNDEF = -1, 626 NFMT_DEFAULT = NFMT_UNORM, 627 628 NFMT_SHIFT = 4, 629 NFMT_MASK = 7 630 }; 631 632 enum MergedFormat : int64_t { 633 DFMT_NFMT_UNDEF = -1, 634 DFMT_NFMT_DEFAULT = ((DFMT_DEFAULT & DFMT_MASK) << DFMT_SHIFT) | 635 ((NFMT_DEFAULT & NFMT_MASK) << NFMT_SHIFT), 636 637 638 DFMT_NFMT_MASK = (DFMT_MASK << DFMT_SHIFT) | (NFMT_MASK << NFMT_SHIFT), 639 640 DFMT_NFMT_MAX = DFMT_NFMT_MASK 641 }; 642 643 enum UnifiedFormatCommon : int64_t { 644 UFMT_MAX = 127, 645 UFMT_UNDEF = -1, 646 UFMT_DEFAULT = 1 647 }; 648 649 } // namespace MTBUFFormat 650 651 namespace UfmtGFX10 { 652 enum UnifiedFormat : int64_t { 653 UFMT_INVALID = 0, 654 655 UFMT_8_UNORM, 656 UFMT_8_SNORM, 657 UFMT_8_USCALED, 658 UFMT_8_SSCALED, 659 UFMT_8_UINT, 660 UFMT_8_SINT, 661 662 UFMT_16_UNORM, 663 UFMT_16_SNORM, 664 UFMT_16_USCALED, 665 UFMT_16_SSCALED, 666 UFMT_16_UINT, 667 UFMT_16_SINT, 668 UFMT_16_FLOAT, 669 670 UFMT_8_8_UNORM, 671 UFMT_8_8_SNORM, 672 UFMT_8_8_USCALED, 673 UFMT_8_8_SSCALED, 674 UFMT_8_8_UINT, 675 UFMT_8_8_SINT, 676 677 UFMT_32_UINT, 678 UFMT_32_SINT, 679 UFMT_32_FLOAT, 680 681 UFMT_16_16_UNORM, 682 UFMT_16_16_SNORM, 683 UFMT_16_16_USCALED, 684 UFMT_16_16_SSCALED, 685 UFMT_16_16_UINT, 686 UFMT_16_16_SINT, 687 UFMT_16_16_FLOAT, 688 689 UFMT_10_11_11_UNORM, 690 UFMT_10_11_11_SNORM, 691 UFMT_10_11_11_USCALED, 692 UFMT_10_11_11_SSCALED, 693 UFMT_10_11_11_UINT, 694 UFMT_10_11_11_SINT, 695 UFMT_10_11_11_FLOAT, 696 697 UFMT_11_11_10_UNORM, 698 UFMT_11_11_10_SNORM, 699 UFMT_11_11_10_USCALED, 700 UFMT_11_11_10_SSCALED, 701 UFMT_11_11_10_UINT, 702 UFMT_11_11_10_SINT, 703 UFMT_11_11_10_FLOAT, 704 705 UFMT_10_10_10_2_UNORM, 706 UFMT_10_10_10_2_SNORM, 707 UFMT_10_10_10_2_USCALED, 708 UFMT_10_10_10_2_SSCALED, 709 UFMT_10_10_10_2_UINT, 710 UFMT_10_10_10_2_SINT, 711 712 UFMT_2_10_10_10_UNORM, 713 UFMT_2_10_10_10_SNORM, 714 UFMT_2_10_10_10_USCALED, 715 UFMT_2_10_10_10_SSCALED, 716 UFMT_2_10_10_10_UINT, 717 UFMT_2_10_10_10_SINT, 718 719 UFMT_8_8_8_8_UNORM, 720 UFMT_8_8_8_8_SNORM, 721 UFMT_8_8_8_8_USCALED, 722 UFMT_8_8_8_8_SSCALED, 723 UFMT_8_8_8_8_UINT, 724 UFMT_8_8_8_8_SINT, 725 726 UFMT_32_32_UINT, 727 UFMT_32_32_SINT, 728 UFMT_32_32_FLOAT, 729 730 UFMT_16_16_16_16_UNORM, 731 UFMT_16_16_16_16_SNORM, 732 UFMT_16_16_16_16_USCALED, 733 UFMT_16_16_16_16_SSCALED, 734 UFMT_16_16_16_16_UINT, 735 UFMT_16_16_16_16_SINT, 736 UFMT_16_16_16_16_FLOAT, 737 738 UFMT_32_32_32_UINT, 739 UFMT_32_32_32_SINT, 740 UFMT_32_32_32_FLOAT, 741 UFMT_32_32_32_32_UINT, 742 UFMT_32_32_32_32_SINT, 743 UFMT_32_32_32_32_FLOAT, 744 745 UFMT_FIRST = UFMT_INVALID, 746 UFMT_LAST = UFMT_32_32_32_32_FLOAT, 747 }; 748 749 } // namespace UfmtGFX10 750 751 namespace UfmtGFX11 { 752 enum UnifiedFormat : int64_t { 753 UFMT_INVALID = 0, 754 755 UFMT_8_UNORM, 756 UFMT_8_SNORM, 757 UFMT_8_USCALED, 758 UFMT_8_SSCALED, 759 UFMT_8_UINT, 760 UFMT_8_SINT, 761 762 UFMT_16_UNORM, 763 UFMT_16_SNORM, 764 UFMT_16_USCALED, 765 UFMT_16_SSCALED, 766 UFMT_16_UINT, 767 UFMT_16_SINT, 768 UFMT_16_FLOAT, 769 770 UFMT_8_8_UNORM, 771 UFMT_8_8_SNORM, 772 UFMT_8_8_USCALED, 773 UFMT_8_8_SSCALED, 774 UFMT_8_8_UINT, 775 UFMT_8_8_SINT, 776 777 UFMT_32_UINT, 778 UFMT_32_SINT, 779 UFMT_32_FLOAT, 780 781 UFMT_16_16_UNORM, 782 UFMT_16_16_SNORM, 783 UFMT_16_16_USCALED, 784 UFMT_16_16_SSCALED, 785 UFMT_16_16_UINT, 786 UFMT_16_16_SINT, 787 UFMT_16_16_FLOAT, 788 789 UFMT_10_11_11_FLOAT, 790 791 UFMT_11_11_10_FLOAT, 792 793 UFMT_10_10_10_2_UNORM, 794 UFMT_10_10_10_2_SNORM, 795 UFMT_10_10_10_2_UINT, 796 UFMT_10_10_10_2_SINT, 797 798 UFMT_2_10_10_10_UNORM, 799 UFMT_2_10_10_10_SNORM, 800 UFMT_2_10_10_10_USCALED, 801 UFMT_2_10_10_10_SSCALED, 802 UFMT_2_10_10_10_UINT, 803 UFMT_2_10_10_10_SINT, 804 805 UFMT_8_8_8_8_UNORM, 806 UFMT_8_8_8_8_SNORM, 807 UFMT_8_8_8_8_USCALED, 808 UFMT_8_8_8_8_SSCALED, 809 UFMT_8_8_8_8_UINT, 810 UFMT_8_8_8_8_SINT, 811 812 UFMT_32_32_UINT, 813 UFMT_32_32_SINT, 814 UFMT_32_32_FLOAT, 815 816 UFMT_16_16_16_16_UNORM, 817 UFMT_16_16_16_16_SNORM, 818 UFMT_16_16_16_16_USCALED, 819 UFMT_16_16_16_16_SSCALED, 820 UFMT_16_16_16_16_UINT, 821 UFMT_16_16_16_16_SINT, 822 UFMT_16_16_16_16_FLOAT, 823 824 UFMT_32_32_32_UINT, 825 UFMT_32_32_32_SINT, 826 UFMT_32_32_32_FLOAT, 827 UFMT_32_32_32_32_UINT, 828 UFMT_32_32_32_32_SINT, 829 UFMT_32_32_32_32_FLOAT, 830 831 UFMT_FIRST = UFMT_INVALID, 832 UFMT_LAST = UFMT_32_32_32_32_FLOAT, 833 }; 834 835 } // namespace UfmtGFX11 836 837 namespace Swizzle { // Encoding of swizzle macro used in ds_swizzle_b32. 838 839 enum Id : unsigned { // id of symbolic names 840 ID_QUAD_PERM = 0, 841 ID_BITMASK_PERM, 842 ID_SWAP, 843 ID_REVERSE, 844 ID_BROADCAST, 845 ID_FFT, 846 ID_ROTATE 847 }; 848 849 // clang-format off 850 enum EncBits : unsigned { 851 852 // swizzle mode encodings 853 854 QUAD_PERM_ENC = 0x8000, 855 QUAD_PERM_ENC_MASK = 0xFF00, 856 857 BITMASK_PERM_ENC = 0x0000, 858 BITMASK_PERM_ENC_MASK = 0x8000, 859 860 FFT_MODE_ENC = 0xE000, 861 862 ROTATE_MODE_ENC = 0xC000, 863 FFT_ROTATE_MODE_MASK = 0xF000, 864 865 ROTATE_MODE_LO = 0xC000, 866 FFT_MODE_LO = 0xE000, 867 868 // QUAD_PERM encodings 869 870 LANE_MASK = 0x3, 871 LANE_MAX = LANE_MASK, 872 LANE_SHIFT = 2, 873 LANE_NUM = 4, 874 875 // BITMASK_PERM encodings 876 877 BITMASK_MASK = 0x1F, 878 BITMASK_MAX = BITMASK_MASK, 879 BITMASK_WIDTH = 5, 880 881 BITMASK_AND_SHIFT = 0, 882 BITMASK_OR_SHIFT = 5, 883 BITMASK_XOR_SHIFT = 10, 884 885 // FFT encodings 886 887 FFT_SWIZZLE_MASK = 0x1F, 888 FFT_SWIZZLE_MAX = 0x1F, 889 890 // ROTATE encodings 891 ROTATE_MAX_SIZE = 0x1F, 892 ROTATE_DIR_SHIFT = 10, // bit position of rotate direction 893 ROTATE_DIR_MASK = 0x1, 894 ROTATE_SIZE_SHIFT = 5, // bit position of rotate size 895 ROTATE_SIZE_MASK = ROTATE_MAX_SIZE, 896 }; 897 // clang-format on 898 899 } // namespace Swizzle 900 901 namespace SDWA { 902 903 enum SdwaSel : unsigned { 904 BYTE_0 = 0, 905 BYTE_1 = 1, 906 BYTE_2 = 2, 907 BYTE_3 = 3, 908 WORD_0 = 4, 909 WORD_1 = 5, 910 DWORD = 6, 911 }; 912 913 enum DstUnused : unsigned { 914 UNUSED_PAD = 0, 915 UNUSED_SEXT = 1, 916 UNUSED_PRESERVE = 2, 917 }; 918 919 enum SDWA9EncValues : unsigned { 920 SRC_SGPR_MASK = 0x100, 921 SRC_VGPR_MASK = 0xFF, 922 VOPC_DST_VCC_MASK = 0x80, 923 VOPC_DST_SGPR_MASK = 0x7F, 924 925 SRC_VGPR_MIN = 0, 926 SRC_VGPR_MAX = 255, 927 SRC_SGPR_MIN = 256, 928 SRC_SGPR_MAX_SI = 357, 929 SRC_SGPR_MAX_GFX10 = 361, 930 SRC_TTMP_MIN = 364, 931 SRC_TTMP_MAX = 379, 932 }; 933 934 } // namespace SDWA 935 936 namespace DPP { 937 938 // clang-format off 939 enum DppCtrl : unsigned { 940 QUAD_PERM_FIRST = 0, 941 QUAD_PERM_ID = 0xE4, // identity permutation 942 QUAD_PERM_LAST = 0xFF, 943 DPP_UNUSED1 = 0x100, 944 ROW_SHL0 = 0x100, 945 ROW_SHL_FIRST = 0x101, 946 ROW_SHL_LAST = 0x10F, 947 DPP_UNUSED2 = 0x110, 948 ROW_SHR0 = 0x110, 949 ROW_SHR_FIRST = 0x111, 950 ROW_SHR_LAST = 0x11F, 951 DPP_UNUSED3 = 0x120, 952 ROW_ROR0 = 0x120, 953 ROW_ROR_FIRST = 0x121, 954 ROW_ROR_LAST = 0x12F, 955 WAVE_SHL1 = 0x130, 956 DPP_UNUSED4_FIRST = 0x131, 957 DPP_UNUSED4_LAST = 0x133, 958 WAVE_ROL1 = 0x134, 959 DPP_UNUSED5_FIRST = 0x135, 960 DPP_UNUSED5_LAST = 0x137, 961 WAVE_SHR1 = 0x138, 962 DPP_UNUSED6_FIRST = 0x139, 963 DPP_UNUSED6_LAST = 0x13B, 964 WAVE_ROR1 = 0x13C, 965 DPP_UNUSED7_FIRST = 0x13D, 966 DPP_UNUSED7_LAST = 0x13F, 967 ROW_MIRROR = 0x140, 968 ROW_HALF_MIRROR = 0x141, 969 BCAST15 = 0x142, 970 BCAST31 = 0x143, 971 DPP_UNUSED8_FIRST = 0x144, 972 DPP_UNUSED8_LAST = 0x14F, 973 ROW_NEWBCAST_FIRST= 0x150, 974 ROW_NEWBCAST_LAST = 0x15F, 975 ROW_SHARE0 = 0x150, 976 ROW_SHARE_FIRST = 0x150, 977 ROW_SHARE_LAST = 0x15F, 978 ROW_XMASK0 = 0x160, 979 ROW_XMASK_FIRST = 0x160, 980 ROW_XMASK_LAST = 0x16F, 981 DPP_LAST = ROW_XMASK_LAST 982 }; 983 // clang-format on 984 985 enum DppFiMode { 986 DPP_FI_0 = 0, 987 DPP_FI_1 = 1, 988 DPP8_FI_0 = 0xE9, 989 DPP8_FI_1 = 0xEA, 990 }; 991 992 } // namespace DPP 993 994 namespace Exp { 995 996 enum Target : unsigned { 997 ET_MRT0 = 0, 998 ET_MRT7 = 7, 999 ET_MRTZ = 8, 1000 ET_NULL = 9, // Pre-GFX11 1001 ET_POS0 = 12, 1002 ET_POS3 = 15, 1003 ET_POS4 = 16, // GFX10+ 1004 ET_POS_LAST = ET_POS4, // Highest pos used on any subtarget 1005 ET_PRIM = 20, // GFX10+ 1006 ET_DUAL_SRC_BLEND0 = 21, // GFX11+ 1007 ET_DUAL_SRC_BLEND1 = 22, // GFX11+ 1008 ET_PARAM0 = 32, // Pre-GFX11 1009 ET_PARAM31 = 63, // Pre-GFX11 1010 1011 ET_NULL_MAX_IDX = 0, 1012 ET_MRTZ_MAX_IDX = 0, 1013 ET_PRIM_MAX_IDX = 0, 1014 ET_MRT_MAX_IDX = 7, 1015 ET_POS_MAX_IDX = 4, 1016 ET_DUAL_SRC_BLEND_MAX_IDX = 1, 1017 ET_PARAM_MAX_IDX = 31, 1018 1019 ET_INVALID = 255, 1020 }; 1021 1022 } // namespace Exp 1023 1024 namespace VOP3PEncoding { 1025 1026 enum OpSel : uint64_t { 1027 OP_SEL_HI_0 = UINT64_C(1) << 59, 1028 OP_SEL_HI_1 = UINT64_C(1) << 60, 1029 OP_SEL_HI_2 = UINT64_C(1) << 14, 1030 }; 1031 1032 } // namespace VOP3PEncoding 1033 1034 namespace ImplicitArg { 1035 // Implicit kernel argument offset for code object version 5. 1036 enum Offset_COV5 : unsigned { 1037 HOSTCALL_PTR_OFFSET = 80, 1038 MULTIGRID_SYNC_ARG_OFFSET = 88, 1039 HEAP_PTR_OFFSET = 96, 1040 1041 DEFAULT_QUEUE_OFFSET = 104, 1042 COMPLETION_ACTION_OFFSET = 112, 1043 1044 PRIVATE_BASE_OFFSET = 192, 1045 SHARED_BASE_OFFSET = 196, 1046 QUEUE_PTR_OFFSET = 200, 1047 }; 1048 1049 } // namespace ImplicitArg 1050 1051 namespace MFMAScaleFormats { 1052 // Enum value used in cbsz/blgp for F8F6F4 MFMA operations to select the matrix 1053 // format. 1054 enum MFMAScaleFormats { 1055 FP8_E4M3 = 0, 1056 FP8_E5M2 = 1, 1057 FP6_E2M3 = 2, 1058 FP6_E3M2 = 3, 1059 FP4_E2M1 = 4 1060 }; 1061 } // namespace MFMAScaleFormats 1062 1063 namespace VirtRegFlag { 1064 // Virtual register flags used for various target specific handlings during 1065 // codegen. 1066 enum Register_Flag : uint8_t { 1067 // Register operand in a whole-wave mode operation. 1068 WWM_REG = 1 << 0, 1069 }; 1070 1071 } // namespace VirtRegFlag 1072 1073 } // namespace AMDGPU 1074 1075 namespace AMDGPU { 1076 namespace Barrier { 1077 1078 enum Type { TRAP = -2, WORKGROUP = -1 }; 1079 1080 enum { 1081 BARRIER_SCOPE_WORKGROUP = 0, 1082 }; 1083 1084 } // namespace Barrier 1085 } // namespace AMDGPU 1086 1087 // clang-format off 1088 1089 #define R_00B028_SPI_SHADER_PGM_RSRC1_PS 0x00B028 1090 #define S_00B028_VGPRS(x) (((x) & 0x3F) << 0) 1091 #define S_00B028_SGPRS(x) (((x) & 0x0F) << 6) 1092 #define S_00B028_MEM_ORDERED(x) (((x) & 0x1) << 25) 1093 #define G_00B028_MEM_ORDERED(x) (((x) >> 25) & 0x1) 1094 #define C_00B028_MEM_ORDERED 0xFDFFFFFF 1095 1096 #define R_00B02C_SPI_SHADER_PGM_RSRC2_PS 0x00B02C 1097 #define S_00B02C_EXTRA_LDS_SIZE(x) (((x) & 0xFF) << 8) 1098 #define R_00B128_SPI_SHADER_PGM_RSRC1_VS 0x00B128 1099 #define S_00B128_MEM_ORDERED(x) (((x) & 0x1) << 27) 1100 #define G_00B128_MEM_ORDERED(x) (((x) >> 27) & 0x1) 1101 #define C_00B128_MEM_ORDERED 0xF7FFFFFF 1102 1103 #define R_00B228_SPI_SHADER_PGM_RSRC1_GS 0x00B228 1104 #define S_00B228_WGP_MODE(x) (((x) & 0x1) << 27) 1105 #define G_00B228_WGP_MODE(x) (((x) >> 27) & 0x1) 1106 #define C_00B228_WGP_MODE 0xF7FFFFFF 1107 #define S_00B228_MEM_ORDERED(x) (((x) & 0x1) << 25) 1108 #define G_00B228_MEM_ORDERED(x) (((x) >> 25) & 0x1) 1109 #define C_00B228_MEM_ORDERED 0xFDFFFFFF 1110 1111 #define R_00B328_SPI_SHADER_PGM_RSRC1_ES 0x00B328 1112 #define R_00B428_SPI_SHADER_PGM_RSRC1_HS 0x00B428 1113 #define S_00B428_WGP_MODE(x) (((x) & 0x1) << 26) 1114 #define G_00B428_WGP_MODE(x) (((x) >> 26) & 0x1) 1115 #define C_00B428_WGP_MODE 0xFBFFFFFF 1116 #define S_00B428_MEM_ORDERED(x) (((x) & 0x1) << 24) 1117 #define G_00B428_MEM_ORDERED(x) (((x) >> 24) & 0x1) 1118 #define C_00B428_MEM_ORDERED 0xFEFFFFFF 1119 1120 #define R_00B528_SPI_SHADER_PGM_RSRC1_LS 0x00B528 1121 1122 #define R_00B84C_COMPUTE_PGM_RSRC2 0x00B84C 1123 #define S_00B84C_SCRATCH_EN(x) (((x) & 0x1) << 0) 1124 #define G_00B84C_SCRATCH_EN(x) (((x) >> 0) & 0x1) 1125 #define C_00B84C_SCRATCH_EN 0xFFFFFFFE 1126 #define S_00B84C_USER_SGPR(x) (((x) & 0x1F) << 1) 1127 #define G_00B84C_USER_SGPR(x) (((x) >> 1) & 0x1F) 1128 #define C_00B84C_USER_SGPR 0xFFFFFFC1 1129 #define S_00B84C_TRAP_HANDLER(x) (((x) & 0x1) << 6) 1130 #define G_00B84C_TRAP_HANDLER(x) (((x) >> 6) & 0x1) 1131 #define C_00B84C_TRAP_HANDLER 0xFFFFFFBF 1132 #define S_00B84C_TGID_X_EN(x) (((x) & 0x1) << 7) 1133 #define G_00B84C_TGID_X_EN(x) (((x) >> 7) & 0x1) 1134 #define C_00B84C_TGID_X_EN 0xFFFFFF7F 1135 #define S_00B84C_TGID_Y_EN(x) (((x) & 0x1) << 8) 1136 #define G_00B84C_TGID_Y_EN(x) (((x) >> 8) & 0x1) 1137 #define C_00B84C_TGID_Y_EN 0xFFFFFEFF 1138 #define S_00B84C_TGID_Z_EN(x) (((x) & 0x1) << 9) 1139 #define G_00B84C_TGID_Z_EN(x) (((x) >> 9) & 0x1) 1140 #define C_00B84C_TGID_Z_EN 0xFFFFFDFF 1141 #define S_00B84C_TG_SIZE_EN(x) (((x) & 0x1) << 10) 1142 #define G_00B84C_TG_SIZE_EN(x) (((x) >> 10) & 0x1) 1143 #define C_00B84C_TG_SIZE_EN 0xFFFFFBFF 1144 #define S_00B84C_TIDIG_COMP_CNT(x) (((x) & 0x03) << 11) 1145 #define G_00B84C_TIDIG_COMP_CNT(x) (((x) >> 11) & 0x03) 1146 #define C_00B84C_TIDIG_COMP_CNT 0xFFFFE7FF 1147 /* CIK */ 1148 #define S_00B84C_EXCP_EN_MSB(x) (((x) & 0x03) << 13) 1149 #define G_00B84C_EXCP_EN_MSB(x) (((x) >> 13) & 0x03) 1150 #define C_00B84C_EXCP_EN_MSB 0xFFFF9FFF 1151 /* */ 1152 #define S_00B84C_LDS_SIZE(x) (((x) & 0x1FF) << 15) 1153 #define G_00B84C_LDS_SIZE(x) (((x) >> 15) & 0x1FF) 1154 #define C_00B84C_LDS_SIZE 0xFF007FFF 1155 #define S_00B84C_EXCP_EN(x) (((x) & 0x7F) << 24) 1156 #define G_00B84C_EXCP_EN(x) (((x) >> 24) & 0x7F) 1157 #define C_00B84C_EXCP_EN 0x80FFFFFF 1158 1159 #define R_0286CC_SPI_PS_INPUT_ENA 0x0286CC 1160 #define R_0286D0_SPI_PS_INPUT_ADDR 0x0286D0 1161 1162 #define R_00B848_COMPUTE_PGM_RSRC1 0x00B848 1163 #define S_00B848_VGPRS(x) (((x) & 0x3F) << 0) 1164 #define G_00B848_VGPRS(x) (((x) >> 0) & 0x3F) 1165 #define C_00B848_VGPRS 0xFFFFFFC0 1166 #define S_00B848_SGPRS(x) (((x) & 0x0F) << 6) 1167 #define G_00B848_SGPRS(x) (((x) >> 6) & 0x0F) 1168 #define C_00B848_SGPRS 0xFFFFFC3F 1169 #define S_00B848_PRIORITY(x) (((x) & 0x03) << 10) 1170 #define G_00B848_PRIORITY(x) (((x) >> 10) & 0x03) 1171 #define C_00B848_PRIORITY 0xFFFFF3FF 1172 #define S_00B848_FLOAT_MODE(x) (((x) & 0xFF) << 12) 1173 #define G_00B848_FLOAT_MODE(x) (((x) >> 12) & 0xFF) 1174 #define C_00B848_FLOAT_MODE 0xFFF00FFF 1175 #define S_00B848_PRIV(x) (((x) & 0x1) << 20) 1176 #define G_00B848_PRIV(x) (((x) >> 20) & 0x1) 1177 #define C_00B848_PRIV 0xFFEFFFFF 1178 #define S_00B848_DX10_CLAMP(x) (((x) & 0x1) << 21) 1179 #define G_00B848_DX10_CLAMP(x) (((x) >> 21) & 0x1) 1180 #define C_00B848_DX10_CLAMP 0xFFDFFFFF 1181 #define S_00B848_RR_WG_MODE(x) (((x) & 0x1) << 21) 1182 #define G_00B848_RR_WG_MODE(x) (((x) >> 21) & 0x1) 1183 #define C_00B848_RR_WG_MODE 0xFFDFFFFF 1184 #define S_00B848_DEBUG_MODE(x) (((x) & 0x1) << 22) 1185 #define G_00B848_DEBUG_MODE(x) (((x) >> 22) & 0x1) 1186 #define C_00B848_DEBUG_MODE 0xFFBFFFFF 1187 #define S_00B848_IEEE_MODE(x) (((x) & 0x1) << 23) 1188 #define G_00B848_IEEE_MODE(x) (((x) >> 23) & 0x1) 1189 #define C_00B848_IEEE_MODE 0xFF7FFFFF 1190 #define S_00B848_WGP_MODE(x) (((x) & 0x1) << 29) 1191 #define G_00B848_WGP_MODE(x) (((x) >> 29) & 0x1) 1192 #define C_00B848_WGP_MODE 0xDFFFFFFF 1193 #define S_00B848_MEM_ORDERED(x) (((x) & 0x1) << 30) 1194 #define G_00B848_MEM_ORDERED(x) (((x) >> 30) & 0x1) 1195 #define C_00B848_MEM_ORDERED 0xBFFFFFFF 1196 #define S_00B848_FWD_PROGRESS(x) (((x) & 0x1) << 31) 1197 #define G_00B848_FWD_PROGRESS(x) (((x) >> 31) & 0x1) 1198 #define C_00B848_FWD_PROGRESS 0x7FFFFFFF 1199 1200 // Helpers for setting FLOAT_MODE 1201 #define FP_ROUND_ROUND_TO_NEAREST 0 1202 #define FP_ROUND_ROUND_TO_INF 1 1203 #define FP_ROUND_ROUND_TO_NEGINF 2 1204 #define FP_ROUND_ROUND_TO_ZERO 3 1205 1206 // Bits 3:0 control rounding mode. 1:0 control single precision, 3:2 double 1207 // precision. 1208 #define FP_ROUND_MODE_SP(x) ((x) & 0x3) 1209 #define FP_ROUND_MODE_DP(x) (((x) & 0x3) << 2) 1210 1211 #define FP_DENORM_FLUSH_IN_FLUSH_OUT 0 1212 #define FP_DENORM_FLUSH_OUT 1 1213 #define FP_DENORM_FLUSH_IN 2 1214 #define FP_DENORM_FLUSH_NONE 3 1215 1216 1217 // Bits 7:4 control denormal handling. 5:4 control single precision, 6:7 double 1218 // precision. 1219 #define FP_DENORM_MODE_SP(x) (((x) & 0x3) << 4) 1220 #define FP_DENORM_MODE_DP(x) (((x) & 0x3) << 6) 1221 1222 #define R_00B860_COMPUTE_TMPRING_SIZE 0x00B860 1223 #define S_00B860_WAVESIZE_PreGFX11(x) (((x) & 0x1FFF) << 12) 1224 #define S_00B860_WAVESIZE_GFX11(x) (((x) & 0x7FFF) << 12) 1225 #define S_00B860_WAVESIZE_GFX12Plus(x) (((x) & 0x3FFFF) << 12) 1226 1227 #define R_0286E8_SPI_TMPRING_SIZE 0x0286E8 1228 #define S_0286E8_WAVESIZE_PreGFX11(x) (((x) & 0x1FFF) << 12) 1229 #define S_0286E8_WAVESIZE_GFX11(x) (((x) & 0x7FFF) << 12) 1230 #define S_0286E8_WAVESIZE_GFX12Plus(x) (((x) & 0x3FFFF) << 12) 1231 1232 #define R_028B54_VGT_SHADER_STAGES_EN 0x028B54 1233 #define S_028B54_HS_W32_EN(x) (((x) & 0x1) << 21) 1234 #define S_028B54_GS_W32_EN(x) (((x) & 0x1) << 22) 1235 #define S_028B54_VS_W32_EN(x) (((x) & 0x1) << 23) 1236 #define R_0286D8_SPI_PS_IN_CONTROL 0x0286D8 1237 #define S_0286D8_PS_W32_EN(x) (((x) & 0x1) << 15) 1238 #define R_00B800_COMPUTE_DISPATCH_INITIATOR 0x00B800 1239 #define S_00B800_CS_W32_EN(x) (((x) & 0x1) << 15) 1240 1241 #define R_SPILLED_SGPRS 0x4 1242 #define R_SPILLED_VGPRS 0x8 1243 1244 // clang-format on 1245 1246 } // End namespace llvm 1247 1248 #endif 1249