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 // For vector registers. 20 HasVGPR = 1 << 0, 21 HasAGPR = 1 << 1, 22 HasSGPR = 1 << 2 23 }; // enum SIRCFlags 24 25 namespace SIInstrFlags { 26 // This needs to be kept in sync with the field bits in InstSI. 27 enum : uint64_t { 28 // Low bits - basic encoding information. 29 SALU = 1 << 0, 30 VALU = 1 << 1, 31 32 // SALU instruction formats. 33 SOP1 = 1 << 2, 34 SOP2 = 1 << 3, 35 SOPC = 1 << 4, 36 SOPK = 1 << 5, 37 SOPP = 1 << 6, 38 39 // VALU instruction formats. 40 VOP1 = 1 << 7, 41 VOP2 = 1 << 8, 42 VOPC = 1 << 9, 43 44 // TODO: Should this be spilt into VOP3 a and b? 45 VOP3 = 1 << 10, 46 VOP3P = 1 << 12, 47 48 VINTRP = 1 << 13, 49 SDWA = 1 << 14, 50 DPP = 1 << 15, 51 TRANS = 1 << 16, 52 53 // Memory instruction formats. 54 MUBUF = 1 << 17, 55 MTBUF = 1 << 18, 56 SMRD = 1 << 19, 57 MIMG = 1 << 20, 58 EXP = 1 << 21, 59 FLAT = 1 << 22, 60 DS = 1 << 23, 61 62 // Pseudo instruction formats. 63 VGPRSpill = 1 << 24, 64 SGPRSpill = 1 << 25, 65 66 // LDSDIR instruction format. 67 LDSDIR = 1 << 26, 68 69 // VINTERP instruction format. 70 VINTERP = 1 << 27, 71 72 // High bits - other information. 73 VM_CNT = UINT64_C(1) << 32, 74 EXP_CNT = UINT64_C(1) << 33, 75 LGKM_CNT = UINT64_C(1) << 34, 76 77 WQM = UINT64_C(1) << 35, 78 DisableWQM = UINT64_C(1) << 36, 79 Gather4 = UINT64_C(1) << 37, 80 SOPK_ZEXT = UINT64_C(1) << 38, 81 SCALAR_STORE = UINT64_C(1) << 39, 82 FIXED_SIZE = UINT64_C(1) << 40, 83 VOPAsmPrefer32Bit = UINT64_C(1) << 41, 84 VOP3_OPSEL = UINT64_C(1) << 42, 85 maybeAtomic = UINT64_C(1) << 43, 86 renamedInGFX9 = UINT64_C(1) << 44, 87 88 // Is a clamp on FP type. 89 FPClamp = UINT64_C(1) << 45, 90 91 // Is an integer clamp 92 IntClamp = UINT64_C(1) << 46, 93 94 // Clamps lo component of register. 95 ClampLo = UINT64_C(1) << 47, 96 97 // Clamps hi component of register. 98 // ClampLo and ClampHi set for packed clamp. 99 ClampHi = UINT64_C(1) << 48, 100 101 // Is a packed VOP3P instruction. 102 IsPacked = UINT64_C(1) << 49, 103 104 // Is a D16 buffer instruction. 105 D16Buf = UINT64_C(1) << 50, 106 107 // FLAT instruction accesses FLAT_GLBL segment. 108 FlatGlobal = UINT64_C(1) << 51, 109 110 // Uses floating point double precision rounding mode 111 FPDPRounding = UINT64_C(1) << 52, 112 113 // Instruction is FP atomic. 114 FPAtomic = UINT64_C(1) << 53, 115 116 // Is a MFMA instruction. 117 IsMAI = UINT64_C(1) << 54, 118 119 // Is a DOT instruction. 120 IsDOT = UINT64_C(1) << 55, 121 122 // FLAT instruction accesses FLAT_SCRATCH segment. 123 FlatScratch = UINT64_C(1) << 56, 124 125 // Atomic without return. 126 IsAtomicNoRet = UINT64_C(1) << 57, 127 128 // Atomic with return. 129 IsAtomicRet = UINT64_C(1) << 58, 130 131 // Is a WMMA instruction. 132 IsWMMA = UINT64_C(1) << 59, 133 134 // Whether tied sources will be read. 135 TiedSourceNotRead = UINT64_C(1) << 60, 136 }; 137 138 // v_cmp_class_* etc. use a 10-bit mask for what operation is checked. 139 // The result is true if any of these tests are true. 140 enum ClassFlags : unsigned { 141 S_NAN = 1 << 0, // Signaling NaN 142 Q_NAN = 1 << 1, // Quiet NaN 143 N_INFINITY = 1 << 2, // Negative infinity 144 N_NORMAL = 1 << 3, // Negative normal 145 N_SUBNORMAL = 1 << 4, // Negative subnormal 146 N_ZERO = 1 << 5, // Negative zero 147 P_ZERO = 1 << 6, // Positive zero 148 P_SUBNORMAL = 1 << 7, // Positive subnormal 149 P_NORMAL = 1 << 8, // Positive normal 150 P_INFINITY = 1 << 9 // Positive infinity 151 }; 152 } 153 154 namespace AMDGPU { 155 enum OperandType : unsigned { 156 /// Operands with register or 32-bit immediate 157 OPERAND_REG_IMM_INT32 = MCOI::OPERAND_FIRST_TARGET, 158 OPERAND_REG_IMM_INT64, 159 OPERAND_REG_IMM_INT16, 160 OPERAND_REG_IMM_FP32, 161 OPERAND_REG_IMM_FP64, 162 OPERAND_REG_IMM_FP16, 163 OPERAND_REG_IMM_FP16_DEFERRED, 164 OPERAND_REG_IMM_FP32_DEFERRED, 165 OPERAND_REG_IMM_V2FP16, 166 OPERAND_REG_IMM_V2INT16, 167 OPERAND_REG_IMM_V2INT32, 168 OPERAND_REG_IMM_V2FP32, 169 170 /// Operands with register or inline constant 171 OPERAND_REG_INLINE_C_INT16, 172 OPERAND_REG_INLINE_C_INT32, 173 OPERAND_REG_INLINE_C_INT64, 174 OPERAND_REG_INLINE_C_FP16, 175 OPERAND_REG_INLINE_C_FP32, 176 OPERAND_REG_INLINE_C_FP64, 177 OPERAND_REG_INLINE_C_V2INT16, 178 OPERAND_REG_INLINE_C_V2FP16, 179 OPERAND_REG_INLINE_C_V2INT32, 180 OPERAND_REG_INLINE_C_V2FP32, 181 182 /// Operand with 32-bit immediate that uses the constant bus. 183 OPERAND_KIMM32, 184 OPERAND_KIMM16, 185 186 /// Operands with an AccVGPR register or inline constant 187 OPERAND_REG_INLINE_AC_INT16, 188 OPERAND_REG_INLINE_AC_INT32, 189 OPERAND_REG_INLINE_AC_FP16, 190 OPERAND_REG_INLINE_AC_FP32, 191 OPERAND_REG_INLINE_AC_FP64, 192 OPERAND_REG_INLINE_AC_V2INT16, 193 OPERAND_REG_INLINE_AC_V2FP16, 194 OPERAND_REG_INLINE_AC_V2INT32, 195 OPERAND_REG_INLINE_AC_V2FP32, 196 197 // Operand for source modifiers for VOP instructions 198 OPERAND_INPUT_MODS, 199 200 // Operand for SDWA instructions 201 OPERAND_SDWA_VOPC_DST, 202 203 OPERAND_REG_IMM_FIRST = OPERAND_REG_IMM_INT32, 204 OPERAND_REG_IMM_LAST = OPERAND_REG_IMM_V2FP32, 205 206 OPERAND_REG_INLINE_C_FIRST = OPERAND_REG_INLINE_C_INT16, 207 OPERAND_REG_INLINE_C_LAST = OPERAND_REG_INLINE_AC_V2FP32, 208 209 OPERAND_REG_INLINE_AC_FIRST = OPERAND_REG_INLINE_AC_INT16, 210 OPERAND_REG_INLINE_AC_LAST = OPERAND_REG_INLINE_AC_V2FP32, 211 212 OPERAND_SRC_FIRST = OPERAND_REG_IMM_INT32, 213 OPERAND_SRC_LAST = OPERAND_REG_INLINE_C_LAST, 214 215 OPERAND_KIMM_FIRST = OPERAND_KIMM32, 216 OPERAND_KIMM_LAST = OPERAND_KIMM16 217 218 }; 219 } 220 221 // Input operand modifiers bit-masks 222 // NEG and SEXT share same bit-mask because they can't be set simultaneously. 223 namespace SISrcMods { 224 enum : unsigned { 225 NEG = 1 << 0, // Floating-point negate modifier 226 ABS = 1 << 1, // Floating-point absolute modifier 227 SEXT = 1 << 0, // Integer sign-extend modifier 228 NEG_HI = ABS, // Floating-point negate high packed component modifier. 229 OP_SEL_0 = 1 << 2, 230 OP_SEL_1 = 1 << 3, 231 DST_OP_SEL = 1 << 3 // VOP3 dst op_sel (share mask with OP_SEL_1) 232 }; 233 } 234 235 namespace SIOutMods { 236 enum : unsigned { 237 NONE = 0, 238 MUL2 = 1, 239 MUL4 = 2, 240 DIV2 = 3 241 }; 242 } 243 244 namespace AMDGPU { 245 namespace VGPRIndexMode { 246 247 enum Id : unsigned { // id of symbolic names 248 ID_SRC0 = 0, 249 ID_SRC1, 250 ID_SRC2, 251 ID_DST, 252 253 ID_MIN = ID_SRC0, 254 ID_MAX = ID_DST 255 }; 256 257 enum EncBits : unsigned { 258 OFF = 0, 259 SRC0_ENABLE = 1 << ID_SRC0, 260 SRC1_ENABLE = 1 << ID_SRC1, 261 SRC2_ENABLE = 1 << ID_SRC2, 262 DST_ENABLE = 1 << ID_DST, 263 ENABLE_MASK = SRC0_ENABLE | SRC1_ENABLE | SRC2_ENABLE | DST_ENABLE, 264 UNDEF = 0xFFFF 265 }; 266 267 } // namespace VGPRIndexMode 268 } // namespace AMDGPU 269 270 namespace AMDGPUAsmVariants { 271 enum : unsigned { 272 DEFAULT = 0, 273 VOP3 = 1, 274 SDWA = 2, 275 SDWA9 = 3, 276 DPP = 4, 277 VOP3_DPP = 5 278 }; 279 } // namespace AMDGPUAsmVariants 280 281 namespace AMDGPU { 282 namespace EncValues { // Encoding values of enum9/8/7 operands 283 284 enum : unsigned { 285 SGPR_MIN = 0, 286 SGPR_MAX_SI = 101, 287 SGPR_MAX_GFX10 = 105, 288 TTMP_VI_MIN = 112, 289 TTMP_VI_MAX = 123, 290 TTMP_GFX9PLUS_MIN = 108, 291 TTMP_GFX9PLUS_MAX = 123, 292 INLINE_INTEGER_C_MIN = 128, 293 INLINE_INTEGER_C_POSITIVE_MAX = 192, // 64 294 INLINE_INTEGER_C_MAX = 208, 295 INLINE_FLOATING_C_MIN = 240, 296 INLINE_FLOATING_C_MAX = 248, 297 LITERAL_CONST = 255, 298 VGPR_MIN = 256, 299 VGPR_MAX = 511, 300 IS_VGPR = 256 // Indicates VGPR or AGPR 301 }; 302 303 } // namespace EncValues 304 } // namespace AMDGPU 305 306 namespace AMDGPU { 307 namespace CPol { 308 309 enum CPol { 310 GLC = 1, 311 SLC = 2, 312 DLC = 4, 313 SCC = 16, 314 SC0 = GLC, 315 SC1 = SCC, 316 NT = SLC, 317 ALL = GLC | SLC | DLC | SCC 318 }; 319 320 } // namespace CPol 321 322 namespace SendMsg { // Encoding of SIMM16 used in s_sendmsg* insns. 323 324 enum Id { // Message ID, width(4) [3:0]. 325 ID_INTERRUPT = 1, 326 327 ID_GS_PreGFX11 = 2, // replaced in GFX11 328 ID_GS_DONE_PreGFX11 = 3, // replaced in GFX11 329 330 ID_HS_TESSFACTOR_GFX11Plus = 2, // reused in GFX11 331 ID_DEALLOC_VGPRS_GFX11Plus = 3, // reused in GFX11 332 333 ID_SAVEWAVE = 4, // added in GFX8, removed in GFX11 334 ID_STALL_WAVE_GEN = 5, // added in GFX9 335 ID_HALT_WAVES = 6, // added in GFX9 336 ID_ORDERED_PS_DONE = 7, // added in GFX9 337 ID_EARLY_PRIM_DEALLOC = 8, // added in GFX9, removed in GFX10 338 ID_GS_ALLOC_REQ = 9, // added in GFX9 339 ID_GET_DOORBELL = 10, // added in GFX9, removed in GFX11 340 ID_GET_DDID = 11, // added in GFX10, removed in GFX11 341 ID_SYSMSG = 15, 342 343 ID_RTN_GET_DOORBELL = 128, 344 ID_RTN_GET_DDID = 129, 345 ID_RTN_GET_TMA = 130, 346 ID_RTN_GET_REALTIME = 131, 347 ID_RTN_SAVE_WAVE = 132, 348 ID_RTN_GET_TBA = 133, 349 350 ID_MASK_PreGFX11_ = 0xF, 351 ID_MASK_GFX11Plus_ = 0xFF 352 }; 353 354 enum Op { // Both GS and SYS operation IDs. 355 OP_UNKNOWN_ = -1, 356 OP_SHIFT_ = 4, 357 OP_NONE_ = 0, 358 // Bits used for operation encoding 359 OP_WIDTH_ = 3, 360 OP_MASK_ = (((1 << OP_WIDTH_) - 1) << OP_SHIFT_), 361 // GS operations are encoded in bits 5:4 362 OP_GS_NOP = 0, 363 OP_GS_CUT = 1, 364 OP_GS_EMIT = 2, 365 OP_GS_EMIT_CUT = 3, 366 OP_GS_LAST_, 367 OP_GS_FIRST_ = OP_GS_NOP, 368 // SYS operations are encoded in bits 6:4 369 OP_SYS_ECC_ERR_INTERRUPT = 1, 370 OP_SYS_REG_RD = 2, 371 OP_SYS_HOST_TRAP_ACK = 3, 372 OP_SYS_TTRACE_PC = 4, 373 OP_SYS_LAST_, 374 OP_SYS_FIRST_ = OP_SYS_ECC_ERR_INTERRUPT, 375 }; 376 377 enum StreamId : unsigned { // Stream ID, (2) [9:8]. 378 STREAM_ID_NONE_ = 0, 379 STREAM_ID_DEFAULT_ = 0, 380 STREAM_ID_LAST_ = 4, 381 STREAM_ID_FIRST_ = STREAM_ID_DEFAULT_, 382 STREAM_ID_SHIFT_ = 8, 383 STREAM_ID_WIDTH_= 2, 384 STREAM_ID_MASK_ = (((1 << STREAM_ID_WIDTH_) - 1) << STREAM_ID_SHIFT_) 385 }; 386 387 } // namespace SendMsg 388 389 namespace Hwreg { // Encoding of SIMM16 used in s_setreg/getreg* insns. 390 391 enum Id { // HwRegCode, (6) [5:0] 392 ID_MODE = 1, 393 ID_STATUS = 2, 394 ID_TRAPSTS = 3, 395 ID_HW_ID = 4, 396 ID_GPR_ALLOC = 5, 397 ID_LDS_ALLOC = 6, 398 ID_IB_STS = 7, 399 ID_MEM_BASES = 15, 400 ID_TBA_LO = 16, 401 ID_TBA_HI = 17, 402 ID_TMA_LO = 18, 403 ID_TMA_HI = 19, 404 ID_XCC_ID = 20, 405 ID_SQ_PERF_SNAPSHOT_DATA = 21, 406 ID_SQ_PERF_SNAPSHOT_DATA1 = 22, 407 ID_SQ_PERF_SNAPSHOT_PC_LO = 23, 408 ID_SQ_PERF_SNAPSHOT_PC_HI = 24, 409 ID_FLAT_SCR_LO = 20, 410 ID_FLAT_SCR_HI = 21, 411 ID_XNACK_MASK = 22, 412 ID_HW_ID1 = 23, 413 ID_HW_ID2 = 24, 414 ID_POPS_PACKER = 25, 415 ID_SHADER_CYCLES = 29, 416 417 ID_SHIFT_ = 0, 418 ID_WIDTH_ = 6, 419 ID_MASK_ = (((1 << ID_WIDTH_) - 1) << ID_SHIFT_) 420 }; 421 422 enum Offset : unsigned { // Offset, (5) [10:6] 423 OFFSET_DEFAULT_ = 0, 424 OFFSET_SHIFT_ = 6, 425 OFFSET_WIDTH_ = 5, 426 OFFSET_MASK_ = (((1 << OFFSET_WIDTH_) - 1) << OFFSET_SHIFT_), 427 428 OFFSET_MEM_VIOL = 8, 429 }; 430 431 enum WidthMinusOne : unsigned { // WidthMinusOne, (5) [15:11] 432 WIDTH_M1_DEFAULT_ = 31, 433 WIDTH_M1_SHIFT_ = 11, 434 WIDTH_M1_WIDTH_ = 5, 435 WIDTH_M1_MASK_ = (((1 << WIDTH_M1_WIDTH_) - 1) << WIDTH_M1_SHIFT_), 436 }; 437 438 // Some values from WidthMinusOne mapped into Width domain. 439 enum Width : unsigned { 440 WIDTH_DEFAULT_ = WIDTH_M1_DEFAULT_ + 1, 441 }; 442 443 enum ModeRegisterMasks : uint32_t { 444 FP_ROUND_MASK = 0xf << 0, // Bits 0..3 445 FP_DENORM_MASK = 0xf << 4, // Bits 4..7 446 DX10_CLAMP_MASK = 1 << 8, 447 IEEE_MODE_MASK = 1 << 9, 448 LOD_CLAMP_MASK = 1 << 10, 449 DEBUG_MASK = 1 << 11, 450 451 // EXCP_EN fields. 452 EXCP_EN_INVALID_MASK = 1 << 12, 453 EXCP_EN_INPUT_DENORMAL_MASK = 1 << 13, 454 EXCP_EN_FLOAT_DIV0_MASK = 1 << 14, 455 EXCP_EN_OVERFLOW_MASK = 1 << 15, 456 EXCP_EN_UNDERFLOW_MASK = 1 << 16, 457 EXCP_EN_INEXACT_MASK = 1 << 17, 458 EXCP_EN_INT_DIV0_MASK = 1 << 18, 459 460 GPR_IDX_EN_MASK = 1 << 27, 461 VSKIP_MASK = 1 << 28, 462 CSP_MASK = 0x7u << 29 // Bits 29..31 463 }; 464 465 } // namespace Hwreg 466 467 namespace MTBUFFormat { 468 469 enum DataFormat : int64_t { 470 DFMT_INVALID = 0, 471 DFMT_8, 472 DFMT_16, 473 DFMT_8_8, 474 DFMT_32, 475 DFMT_16_16, 476 DFMT_10_11_11, 477 DFMT_11_11_10, 478 DFMT_10_10_10_2, 479 DFMT_2_10_10_10, 480 DFMT_8_8_8_8, 481 DFMT_32_32, 482 DFMT_16_16_16_16, 483 DFMT_32_32_32, 484 DFMT_32_32_32_32, 485 DFMT_RESERVED_15, 486 487 DFMT_MIN = DFMT_INVALID, 488 DFMT_MAX = DFMT_RESERVED_15, 489 490 DFMT_UNDEF = -1, 491 DFMT_DEFAULT = DFMT_8, 492 493 DFMT_SHIFT = 0, 494 DFMT_MASK = 0xF 495 }; 496 497 enum NumFormat : int64_t { 498 NFMT_UNORM = 0, 499 NFMT_SNORM, 500 NFMT_USCALED, 501 NFMT_SSCALED, 502 NFMT_UINT, 503 NFMT_SINT, 504 NFMT_RESERVED_6, // VI and GFX9 505 NFMT_SNORM_OGL = NFMT_RESERVED_6, // SI and CI only 506 NFMT_FLOAT, 507 508 NFMT_MIN = NFMT_UNORM, 509 NFMT_MAX = NFMT_FLOAT, 510 511 NFMT_UNDEF = -1, 512 NFMT_DEFAULT = NFMT_UNORM, 513 514 NFMT_SHIFT = 4, 515 NFMT_MASK = 7 516 }; 517 518 enum MergedFormat : int64_t { 519 DFMT_NFMT_UNDEF = -1, 520 DFMT_NFMT_DEFAULT = ((DFMT_DEFAULT & DFMT_MASK) << DFMT_SHIFT) | 521 ((NFMT_DEFAULT & NFMT_MASK) << NFMT_SHIFT), 522 523 524 DFMT_NFMT_MASK = (DFMT_MASK << DFMT_SHIFT) | (NFMT_MASK << NFMT_SHIFT), 525 526 DFMT_NFMT_MAX = DFMT_NFMT_MASK 527 }; 528 529 enum UnifiedFormatCommon : int64_t { 530 UFMT_MAX = 127, 531 UFMT_UNDEF = -1, 532 UFMT_DEFAULT = 1 533 }; 534 535 } // namespace MTBUFFormat 536 537 namespace UfmtGFX10 { 538 enum UnifiedFormat : int64_t { 539 UFMT_INVALID = 0, 540 541 UFMT_8_UNORM, 542 UFMT_8_SNORM, 543 UFMT_8_USCALED, 544 UFMT_8_SSCALED, 545 UFMT_8_UINT, 546 UFMT_8_SINT, 547 548 UFMT_16_UNORM, 549 UFMT_16_SNORM, 550 UFMT_16_USCALED, 551 UFMT_16_SSCALED, 552 UFMT_16_UINT, 553 UFMT_16_SINT, 554 UFMT_16_FLOAT, 555 556 UFMT_8_8_UNORM, 557 UFMT_8_8_SNORM, 558 UFMT_8_8_USCALED, 559 UFMT_8_8_SSCALED, 560 UFMT_8_8_UINT, 561 UFMT_8_8_SINT, 562 563 UFMT_32_UINT, 564 UFMT_32_SINT, 565 UFMT_32_FLOAT, 566 567 UFMT_16_16_UNORM, 568 UFMT_16_16_SNORM, 569 UFMT_16_16_USCALED, 570 UFMT_16_16_SSCALED, 571 UFMT_16_16_UINT, 572 UFMT_16_16_SINT, 573 UFMT_16_16_FLOAT, 574 575 UFMT_10_11_11_UNORM, 576 UFMT_10_11_11_SNORM, 577 UFMT_10_11_11_USCALED, 578 UFMT_10_11_11_SSCALED, 579 UFMT_10_11_11_UINT, 580 UFMT_10_11_11_SINT, 581 UFMT_10_11_11_FLOAT, 582 583 UFMT_11_11_10_UNORM, 584 UFMT_11_11_10_SNORM, 585 UFMT_11_11_10_USCALED, 586 UFMT_11_11_10_SSCALED, 587 UFMT_11_11_10_UINT, 588 UFMT_11_11_10_SINT, 589 UFMT_11_11_10_FLOAT, 590 591 UFMT_10_10_10_2_UNORM, 592 UFMT_10_10_10_2_SNORM, 593 UFMT_10_10_10_2_USCALED, 594 UFMT_10_10_10_2_SSCALED, 595 UFMT_10_10_10_2_UINT, 596 UFMT_10_10_10_2_SINT, 597 598 UFMT_2_10_10_10_UNORM, 599 UFMT_2_10_10_10_SNORM, 600 UFMT_2_10_10_10_USCALED, 601 UFMT_2_10_10_10_SSCALED, 602 UFMT_2_10_10_10_UINT, 603 UFMT_2_10_10_10_SINT, 604 605 UFMT_8_8_8_8_UNORM, 606 UFMT_8_8_8_8_SNORM, 607 UFMT_8_8_8_8_USCALED, 608 UFMT_8_8_8_8_SSCALED, 609 UFMT_8_8_8_8_UINT, 610 UFMT_8_8_8_8_SINT, 611 612 UFMT_32_32_UINT, 613 UFMT_32_32_SINT, 614 UFMT_32_32_FLOAT, 615 616 UFMT_16_16_16_16_UNORM, 617 UFMT_16_16_16_16_SNORM, 618 UFMT_16_16_16_16_USCALED, 619 UFMT_16_16_16_16_SSCALED, 620 UFMT_16_16_16_16_UINT, 621 UFMT_16_16_16_16_SINT, 622 UFMT_16_16_16_16_FLOAT, 623 624 UFMT_32_32_32_UINT, 625 UFMT_32_32_32_SINT, 626 UFMT_32_32_32_FLOAT, 627 UFMT_32_32_32_32_UINT, 628 UFMT_32_32_32_32_SINT, 629 UFMT_32_32_32_32_FLOAT, 630 631 UFMT_FIRST = UFMT_INVALID, 632 UFMT_LAST = UFMT_32_32_32_32_FLOAT, 633 }; 634 635 } // namespace UfmtGFX10 636 637 namespace UfmtGFX11 { 638 enum UnifiedFormat : int64_t { 639 UFMT_INVALID = 0, 640 641 UFMT_8_UNORM, 642 UFMT_8_SNORM, 643 UFMT_8_USCALED, 644 UFMT_8_SSCALED, 645 UFMT_8_UINT, 646 UFMT_8_SINT, 647 648 UFMT_16_UNORM, 649 UFMT_16_SNORM, 650 UFMT_16_USCALED, 651 UFMT_16_SSCALED, 652 UFMT_16_UINT, 653 UFMT_16_SINT, 654 UFMT_16_FLOAT, 655 656 UFMT_8_8_UNORM, 657 UFMT_8_8_SNORM, 658 UFMT_8_8_USCALED, 659 UFMT_8_8_SSCALED, 660 UFMT_8_8_UINT, 661 UFMT_8_8_SINT, 662 663 UFMT_32_UINT, 664 UFMT_32_SINT, 665 UFMT_32_FLOAT, 666 667 UFMT_16_16_UNORM, 668 UFMT_16_16_SNORM, 669 UFMT_16_16_USCALED, 670 UFMT_16_16_SSCALED, 671 UFMT_16_16_UINT, 672 UFMT_16_16_SINT, 673 UFMT_16_16_FLOAT, 674 675 UFMT_10_11_11_FLOAT, 676 677 UFMT_11_11_10_FLOAT, 678 679 UFMT_10_10_10_2_UNORM, 680 UFMT_10_10_10_2_SNORM, 681 UFMT_10_10_10_2_UINT, 682 UFMT_10_10_10_2_SINT, 683 684 UFMT_2_10_10_10_UNORM, 685 UFMT_2_10_10_10_SNORM, 686 UFMT_2_10_10_10_USCALED, 687 UFMT_2_10_10_10_SSCALED, 688 UFMT_2_10_10_10_UINT, 689 UFMT_2_10_10_10_SINT, 690 691 UFMT_8_8_8_8_UNORM, 692 UFMT_8_8_8_8_SNORM, 693 UFMT_8_8_8_8_USCALED, 694 UFMT_8_8_8_8_SSCALED, 695 UFMT_8_8_8_8_UINT, 696 UFMT_8_8_8_8_SINT, 697 698 UFMT_32_32_UINT, 699 UFMT_32_32_SINT, 700 UFMT_32_32_FLOAT, 701 702 UFMT_16_16_16_16_UNORM, 703 UFMT_16_16_16_16_SNORM, 704 UFMT_16_16_16_16_USCALED, 705 UFMT_16_16_16_16_SSCALED, 706 UFMT_16_16_16_16_UINT, 707 UFMT_16_16_16_16_SINT, 708 UFMT_16_16_16_16_FLOAT, 709 710 UFMT_32_32_32_UINT, 711 UFMT_32_32_32_SINT, 712 UFMT_32_32_32_FLOAT, 713 UFMT_32_32_32_32_UINT, 714 UFMT_32_32_32_32_SINT, 715 UFMT_32_32_32_32_FLOAT, 716 717 UFMT_FIRST = UFMT_INVALID, 718 UFMT_LAST = UFMT_32_32_32_32_FLOAT, 719 }; 720 721 } // namespace UfmtGFX11 722 723 namespace Swizzle { // Encoding of swizzle macro used in ds_swizzle_b32. 724 725 enum Id : unsigned { // id of symbolic names 726 ID_QUAD_PERM = 0, 727 ID_BITMASK_PERM, 728 ID_SWAP, 729 ID_REVERSE, 730 ID_BROADCAST 731 }; 732 733 enum EncBits : unsigned { 734 735 // swizzle mode encodings 736 737 QUAD_PERM_ENC = 0x8000, 738 QUAD_PERM_ENC_MASK = 0xFF00, 739 740 BITMASK_PERM_ENC = 0x0000, 741 BITMASK_PERM_ENC_MASK = 0x8000, 742 743 // QUAD_PERM encodings 744 745 LANE_MASK = 0x3, 746 LANE_MAX = LANE_MASK, 747 LANE_SHIFT = 2, 748 LANE_NUM = 4, 749 750 // BITMASK_PERM encodings 751 752 BITMASK_MASK = 0x1F, 753 BITMASK_MAX = BITMASK_MASK, 754 BITMASK_WIDTH = 5, 755 756 BITMASK_AND_SHIFT = 0, 757 BITMASK_OR_SHIFT = 5, 758 BITMASK_XOR_SHIFT = 10 759 }; 760 761 } // namespace Swizzle 762 763 namespace SDWA { 764 765 enum SdwaSel : unsigned { 766 BYTE_0 = 0, 767 BYTE_1 = 1, 768 BYTE_2 = 2, 769 BYTE_3 = 3, 770 WORD_0 = 4, 771 WORD_1 = 5, 772 DWORD = 6, 773 }; 774 775 enum DstUnused : unsigned { 776 UNUSED_PAD = 0, 777 UNUSED_SEXT = 1, 778 UNUSED_PRESERVE = 2, 779 }; 780 781 enum SDWA9EncValues : unsigned { 782 SRC_SGPR_MASK = 0x100, 783 SRC_VGPR_MASK = 0xFF, 784 VOPC_DST_VCC_MASK = 0x80, 785 VOPC_DST_SGPR_MASK = 0x7F, 786 787 SRC_VGPR_MIN = 0, 788 SRC_VGPR_MAX = 255, 789 SRC_SGPR_MIN = 256, 790 SRC_SGPR_MAX_SI = 357, 791 SRC_SGPR_MAX_GFX10 = 361, 792 SRC_TTMP_MIN = 364, 793 SRC_TTMP_MAX = 379, 794 }; 795 796 } // namespace SDWA 797 798 namespace DPP { 799 800 // clang-format off 801 enum DppCtrl : unsigned { 802 QUAD_PERM_FIRST = 0, 803 QUAD_PERM_ID = 0xE4, // identity permutation 804 QUAD_PERM_LAST = 0xFF, 805 DPP_UNUSED1 = 0x100, 806 ROW_SHL0 = 0x100, 807 ROW_SHL_FIRST = 0x101, 808 ROW_SHL_LAST = 0x10F, 809 DPP_UNUSED2 = 0x110, 810 ROW_SHR0 = 0x110, 811 ROW_SHR_FIRST = 0x111, 812 ROW_SHR_LAST = 0x11F, 813 DPP_UNUSED3 = 0x120, 814 ROW_ROR0 = 0x120, 815 ROW_ROR_FIRST = 0x121, 816 ROW_ROR_LAST = 0x12F, 817 WAVE_SHL1 = 0x130, 818 DPP_UNUSED4_FIRST = 0x131, 819 DPP_UNUSED4_LAST = 0x133, 820 WAVE_ROL1 = 0x134, 821 DPP_UNUSED5_FIRST = 0x135, 822 DPP_UNUSED5_LAST = 0x137, 823 WAVE_SHR1 = 0x138, 824 DPP_UNUSED6_FIRST = 0x139, 825 DPP_UNUSED6_LAST = 0x13B, 826 WAVE_ROR1 = 0x13C, 827 DPP_UNUSED7_FIRST = 0x13D, 828 DPP_UNUSED7_LAST = 0x13F, 829 ROW_MIRROR = 0x140, 830 ROW_HALF_MIRROR = 0x141, 831 BCAST15 = 0x142, 832 BCAST31 = 0x143, 833 DPP_UNUSED8_FIRST = 0x144, 834 DPP_UNUSED8_LAST = 0x14F, 835 ROW_NEWBCAST_FIRST= 0x150, 836 ROW_NEWBCAST_LAST = 0x15F, 837 ROW_SHARE0 = 0x150, 838 ROW_SHARE_FIRST = 0x150, 839 ROW_SHARE_LAST = 0x15F, 840 ROW_XMASK0 = 0x160, 841 ROW_XMASK_FIRST = 0x160, 842 ROW_XMASK_LAST = 0x16F, 843 DPP_LAST = ROW_XMASK_LAST 844 }; 845 // clang-format on 846 847 enum DppFiMode { 848 DPP_FI_0 = 0, 849 DPP_FI_1 = 1, 850 DPP8_FI_0 = 0xE9, 851 DPP8_FI_1 = 0xEA, 852 }; 853 854 } // namespace DPP 855 856 namespace Exp { 857 858 enum Target : unsigned { 859 ET_MRT0 = 0, 860 ET_MRT7 = 7, 861 ET_MRTZ = 8, 862 ET_NULL = 9, // Pre-GFX11 863 ET_POS0 = 12, 864 ET_POS3 = 15, 865 ET_POS4 = 16, // GFX10+ 866 ET_POS_LAST = ET_POS4, // Highest pos used on any subtarget 867 ET_PRIM = 20, // GFX10+ 868 ET_DUAL_SRC_BLEND0 = 21, // GFX11+ 869 ET_DUAL_SRC_BLEND1 = 22, // GFX11+ 870 ET_PARAM0 = 32, // Pre-GFX11 871 ET_PARAM31 = 63, // Pre-GFX11 872 873 ET_NULL_MAX_IDX = 0, 874 ET_MRTZ_MAX_IDX = 0, 875 ET_PRIM_MAX_IDX = 0, 876 ET_MRT_MAX_IDX = 7, 877 ET_POS_MAX_IDX = 4, 878 ET_DUAL_SRC_BLEND_MAX_IDX = 1, 879 ET_PARAM_MAX_IDX = 31, 880 881 ET_INVALID = 255, 882 }; 883 884 } // namespace Exp 885 886 namespace VOP3PEncoding { 887 888 enum OpSel : uint64_t { 889 OP_SEL_HI_0 = UINT64_C(1) << 59, 890 OP_SEL_HI_1 = UINT64_C(1) << 60, 891 OP_SEL_HI_2 = UINT64_C(1) << 14, 892 }; 893 894 } // namespace VOP3PEncoding 895 896 namespace ImplicitArg { 897 // Implicit kernel argument offset for code object version 5. 898 enum Offset_COV5 : unsigned { 899 HOSTCALL_PTR_OFFSET = 80, 900 MULTIGRID_SYNC_ARG_OFFSET = 88, 901 HEAP_PTR_OFFSET = 96, 902 903 DEFAULT_QUEUE_OFFSET = 104, 904 COMPLETION_ACTION_OFFSET = 112, 905 906 PRIVATE_BASE_OFFSET = 192, 907 SHARED_BASE_OFFSET = 196, 908 QUEUE_PTR_OFFSET = 200, 909 }; 910 911 } // namespace ImplicitArg 912 } // namespace AMDGPU 913 914 #define R_00B028_SPI_SHADER_PGM_RSRC1_PS 0x00B028 915 #define S_00B028_VGPRS(x) (((x) & 0x3F) << 0) 916 #define S_00B028_SGPRS(x) (((x) & 0x0F) << 6) 917 #define S_00B028_MEM_ORDERED(x) (((x) & 0x1) << 25) 918 #define G_00B028_MEM_ORDERED(x) (((x) >> 25) & 0x1) 919 #define C_00B028_MEM_ORDERED 0xFDFFFFFF 920 921 #define R_00B02C_SPI_SHADER_PGM_RSRC2_PS 0x00B02C 922 #define S_00B02C_EXTRA_LDS_SIZE(x) (((x) & 0xFF) << 8) 923 #define R_00B128_SPI_SHADER_PGM_RSRC1_VS 0x00B128 924 #define S_00B128_MEM_ORDERED(x) (((x) & 0x1) << 27) 925 #define G_00B128_MEM_ORDERED(x) (((x) >> 27) & 0x1) 926 #define C_00B128_MEM_ORDERED 0xF7FFFFFF 927 928 #define R_00B228_SPI_SHADER_PGM_RSRC1_GS 0x00B228 929 #define S_00B228_WGP_MODE(x) (((x) & 0x1) << 27) 930 #define G_00B228_WGP_MODE(x) (((x) >> 27) & 0x1) 931 #define C_00B228_WGP_MODE 0xF7FFFFFF 932 #define S_00B228_MEM_ORDERED(x) (((x) & 0x1) << 25) 933 #define G_00B228_MEM_ORDERED(x) (((x) >> 25) & 0x1) 934 #define C_00B228_MEM_ORDERED 0xFDFFFFFF 935 936 #define R_00B328_SPI_SHADER_PGM_RSRC1_ES 0x00B328 937 #define R_00B428_SPI_SHADER_PGM_RSRC1_HS 0x00B428 938 #define S_00B428_WGP_MODE(x) (((x) & 0x1) << 26) 939 #define G_00B428_WGP_MODE(x) (((x) >> 26) & 0x1) 940 #define C_00B428_WGP_MODE 0xFBFFFFFF 941 #define S_00B428_MEM_ORDERED(x) (((x) & 0x1) << 24) 942 #define G_00B428_MEM_ORDERED(x) (((x) >> 24) & 0x1) 943 #define C_00B428_MEM_ORDERED 0xFEFFFFFF 944 945 #define R_00B528_SPI_SHADER_PGM_RSRC1_LS 0x00B528 946 947 #define R_00B84C_COMPUTE_PGM_RSRC2 0x00B84C 948 #define S_00B84C_SCRATCH_EN(x) (((x) & 0x1) << 0) 949 #define G_00B84C_SCRATCH_EN(x) (((x) >> 0) & 0x1) 950 #define C_00B84C_SCRATCH_EN 0xFFFFFFFE 951 #define S_00B84C_USER_SGPR(x) (((x) & 0x1F) << 1) 952 #define G_00B84C_USER_SGPR(x) (((x) >> 1) & 0x1F) 953 #define C_00B84C_USER_SGPR 0xFFFFFFC1 954 #define S_00B84C_TRAP_HANDLER(x) (((x) & 0x1) << 6) 955 #define G_00B84C_TRAP_HANDLER(x) (((x) >> 6) & 0x1) 956 #define C_00B84C_TRAP_HANDLER 0xFFFFFFBF 957 #define S_00B84C_TGID_X_EN(x) (((x) & 0x1) << 7) 958 #define G_00B84C_TGID_X_EN(x) (((x) >> 7) & 0x1) 959 #define C_00B84C_TGID_X_EN 0xFFFFFF7F 960 #define S_00B84C_TGID_Y_EN(x) (((x) & 0x1) << 8) 961 #define G_00B84C_TGID_Y_EN(x) (((x) >> 8) & 0x1) 962 #define C_00B84C_TGID_Y_EN 0xFFFFFEFF 963 #define S_00B84C_TGID_Z_EN(x) (((x) & 0x1) << 9) 964 #define G_00B84C_TGID_Z_EN(x) (((x) >> 9) & 0x1) 965 #define C_00B84C_TGID_Z_EN 0xFFFFFDFF 966 #define S_00B84C_TG_SIZE_EN(x) (((x) & 0x1) << 10) 967 #define G_00B84C_TG_SIZE_EN(x) (((x) >> 10) & 0x1) 968 #define C_00B84C_TG_SIZE_EN 0xFFFFFBFF 969 #define S_00B84C_TIDIG_COMP_CNT(x) (((x) & 0x03) << 11) 970 #define G_00B84C_TIDIG_COMP_CNT(x) (((x) >> 11) & 0x03) 971 #define C_00B84C_TIDIG_COMP_CNT 0xFFFFE7FF 972 /* CIK */ 973 #define S_00B84C_EXCP_EN_MSB(x) (((x) & 0x03) << 13) 974 #define G_00B84C_EXCP_EN_MSB(x) (((x) >> 13) & 0x03) 975 #define C_00B84C_EXCP_EN_MSB 0xFFFF9FFF 976 /* */ 977 #define S_00B84C_LDS_SIZE(x) (((x) & 0x1FF) << 15) 978 #define G_00B84C_LDS_SIZE(x) (((x) >> 15) & 0x1FF) 979 #define C_00B84C_LDS_SIZE 0xFF007FFF 980 #define S_00B84C_EXCP_EN(x) (((x) & 0x7F) << 24) 981 #define G_00B84C_EXCP_EN(x) (((x) >> 24) & 0x7F) 982 #define C_00B84C_EXCP_EN 983 984 #define R_0286CC_SPI_PS_INPUT_ENA 0x0286CC 985 #define R_0286D0_SPI_PS_INPUT_ADDR 0x0286D0 986 987 #define R_00B848_COMPUTE_PGM_RSRC1 0x00B848 988 #define S_00B848_VGPRS(x) (((x) & 0x3F) << 0) 989 #define G_00B848_VGPRS(x) (((x) >> 0) & 0x3F) 990 #define C_00B848_VGPRS 0xFFFFFFC0 991 #define S_00B848_SGPRS(x) (((x) & 0x0F) << 6) 992 #define G_00B848_SGPRS(x) (((x) >> 6) & 0x0F) 993 #define C_00B848_SGPRS 0xFFFFFC3F 994 #define S_00B848_PRIORITY(x) (((x) & 0x03) << 10) 995 #define G_00B848_PRIORITY(x) (((x) >> 10) & 0x03) 996 #define C_00B848_PRIORITY 0xFFFFF3FF 997 #define S_00B848_FLOAT_MODE(x) (((x) & 0xFF) << 12) 998 #define G_00B848_FLOAT_MODE(x) (((x) >> 12) & 0xFF) 999 #define C_00B848_FLOAT_MODE 0xFFF00FFF 1000 #define S_00B848_PRIV(x) (((x) & 0x1) << 20) 1001 #define G_00B848_PRIV(x) (((x) >> 20) & 0x1) 1002 #define C_00B848_PRIV 0xFFEFFFFF 1003 #define S_00B848_DX10_CLAMP(x) (((x) & 0x1) << 21) 1004 #define G_00B848_DX10_CLAMP(x) (((x) >> 21) & 0x1) 1005 #define C_00B848_DX10_CLAMP 0xFFDFFFFF 1006 #define S_00B848_DEBUG_MODE(x) (((x) & 0x1) << 22) 1007 #define G_00B848_DEBUG_MODE(x) (((x) >> 22) & 0x1) 1008 #define C_00B848_DEBUG_MODE 0xFFBFFFFF 1009 #define S_00B848_IEEE_MODE(x) (((x) & 0x1) << 23) 1010 #define G_00B848_IEEE_MODE(x) (((x) >> 23) & 0x1) 1011 #define C_00B848_IEEE_MODE 0xFF7FFFFF 1012 #define S_00B848_WGP_MODE(x) (((x) & 0x1) << 29) 1013 #define G_00B848_WGP_MODE(x) (((x) >> 29) & 0x1) 1014 #define C_00B848_WGP_MODE 0xDFFFFFFF 1015 #define S_00B848_MEM_ORDERED(x) (((x) & 0x1) << 30) 1016 #define G_00B848_MEM_ORDERED(x) (((x) >> 30) & 0x1) 1017 #define C_00B848_MEM_ORDERED 0xBFFFFFFF 1018 #define S_00B848_FWD_PROGRESS(x) (((x) & 0x1) << 31) 1019 #define G_00B848_FWD_PROGRESS(x) (((x) >> 31) & 0x1) 1020 #define C_00B848_FWD_PROGRESS 0x7FFFFFFF 1021 1022 1023 // Helpers for setting FLOAT_MODE 1024 #define FP_ROUND_ROUND_TO_NEAREST 0 1025 #define FP_ROUND_ROUND_TO_INF 1 1026 #define FP_ROUND_ROUND_TO_NEGINF 2 1027 #define FP_ROUND_ROUND_TO_ZERO 3 1028 1029 // Bits 3:0 control rounding mode. 1:0 control single precision, 3:2 double 1030 // precision. 1031 #define FP_ROUND_MODE_SP(x) ((x) & 0x3) 1032 #define FP_ROUND_MODE_DP(x) (((x) & 0x3) << 2) 1033 1034 #define FP_DENORM_FLUSH_IN_FLUSH_OUT 0 1035 #define FP_DENORM_FLUSH_OUT 1 1036 #define FP_DENORM_FLUSH_IN 2 1037 #define FP_DENORM_FLUSH_NONE 3 1038 1039 1040 // Bits 7:4 control denormal handling. 5:4 control single precision, 6:7 double 1041 // precision. 1042 #define FP_DENORM_MODE_SP(x) (((x) & 0x3) << 4) 1043 #define FP_DENORM_MODE_DP(x) (((x) & 0x3) << 6) 1044 1045 #define R_00B860_COMPUTE_TMPRING_SIZE 0x00B860 1046 #define S_00B860_WAVESIZE_PreGFX11(x) (((x) & 0x1FFF) << 12) 1047 #define S_00B860_WAVESIZE_GFX11Plus(x) (((x) & 0x7FFF) << 12) 1048 1049 #define R_0286E8_SPI_TMPRING_SIZE 0x0286E8 1050 #define S_0286E8_WAVESIZE_PreGFX11(x) (((x) & 0x1FFF) << 12) 1051 #define S_0286E8_WAVESIZE_GFX11Plus(x) (((x) & 0x7FFF) << 12) 1052 1053 #define R_028B54_VGT_SHADER_STAGES_EN 0x028B54 1054 #define S_028B54_HS_W32_EN(x) (((x) & 0x1) << 21) 1055 #define S_028B54_GS_W32_EN(x) (((x) & 0x1) << 22) 1056 #define S_028B54_VS_W32_EN(x) (((x) & 0x1) << 23) 1057 #define R_0286D8_SPI_PS_IN_CONTROL 0x0286D8 1058 #define S_0286D8_PS_W32_EN(x) (((x) & 0x1) << 15) 1059 #define R_00B800_COMPUTE_DISPATCH_INITIATOR 0x00B800 1060 #define S_00B800_CS_W32_EN(x) (((x) & 0x1) << 15) 1061 1062 #define R_SPILLED_SGPRS 0x4 1063 #define R_SPILLED_VGPRS 0x8 1064 } // End namespace llvm 1065 1066 #endif 1067