1 /* -*- mesa-c++ -*- 2 * 3 * Copyright (c) 2018-2019 Collabora LTD 4 * 5 * Author: Gert Wollny <gert.wollny@collabora.com> 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a 8 * copy of this software and associated documentation files (the "Software"), 9 * to deal in the Software without restriction, including without limitation 10 * on the rights to use, copy, modify, merge, publish, distribute, sub 11 * license, and/or sell copies of the Software, and to permit persons to whom 12 * the Software is furnished to do so, subject to the following conditions: 13 * 14 * The above copyright notice and this permission notice (including the next 15 * paragraph) shall be included in all copies or substantial portions of the 16 * Software. 17 * 18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 20 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL 21 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, 22 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 23 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 24 * USE OR OTHER DEALINGS IN THE SOFTWARE. 25 */ 26 27 #ifndef r600_sfn_alu_defines_h 28 #define r600_sfn_alu_defines_h 29 30 #include "../r600_isa.h" 31 32 #include <bitset> 33 #include <map> 34 35 namespace r600 { 36 37 // We sacrifice 123 for dummy dests 38 static const int g_registers_end = 123; 39 static const int g_clause_local_start = 124; 40 static const int g_clause_local_end = 128; 41 42 /* ALU op2 instructions 17:7 top three bits always zero. */ 43 enum EAluOp { 44 op2_add = 0, 45 op2_mul = 1, 46 op2_mul_ieee = 2, 47 op2_max = 3, 48 op2_min = 4, 49 op2_max_dx10 = 5, 50 op2_min_dx10 = 6, 51 op2_sete = 8, 52 op2_setgt = 9, 53 op2_setge = 10, 54 op2_setne = 11, 55 op2_sete_dx10 = 12, 56 op2_setgt_dx10 = 13, 57 op2_setge_dx10 = 14, 58 op2_setne_dx10 = 15, 59 op1_fract = 16, 60 op1_trunc = 17, 61 op1_ceil = 18, 62 op1_rndne = 19, 63 op1_floor = 20, 64 op2_ashr_int = 21, 65 op2_lshr_int = 22, 66 op2_lshl_int = 23, 67 op1_mov = 25, 68 op0_nop = 26, 69 op2_mul_64 = 27, 70 op1_flt64_to_flt32 = 28, 71 op1_flt32_to_flt64 = 29, 72 op2_pred_setgt_uint = 30, 73 op2_pred_setge_uint = 31, 74 op2_pred_sete = 32, 75 op2_pred_setgt = 33, 76 op2_pred_setge = 34, 77 op2_pred_setne = 35, 78 op1_pred_set_inv = 36, 79 op2_pred_set_pop = 37, 80 op0_pred_set_clr = 38, 81 op1_pred_set_restore = 39, 82 op2_pred_sete_push = 40, 83 op2_pred_setgt_push = 41, 84 op2_pred_setge_push = 42, 85 op2_pred_setne_push = 43, 86 op2_kille = 44, 87 op2_killgt = 45, 88 op2_killge = 46, 89 op2_killne = 47, 90 op2_and_int = 48, 91 op2_or_int = 49, 92 op2_xor_int = 50, 93 op1_not_int = 51, 94 op2_add_int = 52, 95 op2_sub_int = 53, 96 op2_max_int = 54, 97 op2_min_int = 55, 98 op2_max_uint = 56, 99 op2_min_uint = 57, 100 op2_sete_int = 58, 101 op2_setgt_int = 59, 102 op2_setge_int = 60, 103 op2_setne_int = 61, 104 op2_setgt_uint = 62, 105 op2_setge_uint = 63, 106 op2_killgt_uint = 64, 107 op2_killge_uint = 65, 108 op2_prede_int = 66, 109 op2_pred_setgt_int = 67, 110 op2_pred_setge_int = 68, 111 op2_pred_setne_int = 69, 112 op2_kille_int = 70, 113 op2_killgt_int = 71, 114 op2_killge_int = 72, 115 op2_killne_int = 73, 116 op2_pred_sete_push_int = 74, 117 op2_pred_setgt_push_int = 75, 118 op2_pred_setge_push_int = 76, 119 op2_pred_setne_push_int = 77, 120 op2_pred_setlt_push_int = 78, 121 op2_pred_setle_push_int = 79, 122 op1_flt_to_int = 80, 123 op1_bfrev_int = 81, 124 op2_addc_uint = 82, 125 op2_subb_uint = 83, 126 op0_group_barrier = 84, 127 op0_group_seq_begin = 85, 128 op0_group_seq_end = 86, 129 op2_set_mode = 87, 130 op1_set_cf_idx0 = 88, 131 op1_set_cf_idx1 = 89, 132 op2_set_lds_size = 90, 133 op1_exp_ieee = 129, 134 op1_log_clamped = 130, 135 op1_log_ieee = 131, 136 op1_recip_clamped = 132, 137 op1_recip_ff = 133, 138 op1_recip_ieee = 134, 139 op1_recipsqrt_clamped = 135, 140 op1_recipsqrt_ff = 136, 141 op1_recipsqrt_ieee1 = 137, 142 op1_sqrt_ieee = 138, 143 op1_sin = 141, 144 op1_cos = 142, 145 op2_mullo_int = 143, 146 op2_mulhi_int = 144, 147 op2_mullo_uint = 145, 148 op2_mulhi_uint = 146, 149 op1_recip_int = 147, 150 op1_recip_uint = 148, 151 op1_recip_64 = 149, 152 op1_recip_clamped_64 = 150, 153 op1_recipsqrt_64 = 151, 154 op1_recipsqrt_clamped_64 = 152, 155 op1_sqrt_64 = 153, 156 op1_flt_to_uint = 154, 157 op1_int_to_flt = 155, 158 op1_uint_to_flt = 156, 159 op2_bfm_int = 160, 160 op1_flt32_to_flt16 = 162, 161 op1_flt16_to_flt32 = 163, 162 op1_ubyte0_flt = 164, 163 op1_ubyte1_flt = 165, 164 op1_ubyte2_flt = 166, 165 op1_ubyte3_flt = 167, 166 op1_bcnt_int = 170, 167 op1_ffbh_uint = 171, 168 op1_ffbl_int = 172, 169 op1_ffbh_int = 173, 170 op1_flt_to_uint4 = 174, 171 op2_dot_ieee = 175, 172 op1_flt_to_int_rpi = 176, 173 op1_flt_to_int_floor = 177, 174 op2_mulhi_uint24 = 178, 175 op1_mbcnt_32hi_int = 179, 176 op1_offset_to_flt = 180, 177 op2_mul_uint24 = 181, 178 op1_bcnt_accum_prev_int = 182, 179 op1_mbcnt_32lo_accum_prev_int = 183, 180 op2_sete_64 = 184, 181 op2_setne_64 = 185, 182 op2_setgt_64 = 186, 183 op2_setge_64 = 187, 184 op2_min_64 = 188, 185 op2_max_64 = 189, 186 op2_dot4 = 190, 187 op2_dot4_ieee = 191, 188 op2_cube = 192, 189 op1_max4 = 193, 190 op1_frexp_64 = 196, 191 op1_ldexp_64 = 197, 192 op1_fract_64 = 198, 193 op2_pred_setgt_64 = 199, 194 op2_pred_sete_64 = 198, 195 op2_pred_setge_64 = 201, 196 OP2V_MUL_64 = 202, 197 op2_add_64 = 203, 198 op1_mova_int = 204, 199 op1v_flt64_to_flt32 = 205, 200 op1v_flt32_to_flt64 = 206, 201 op2_sad_accum_prev_uint = 207, 202 op2_dot = 208, 203 op1_mul_prev = 209, 204 op1_mul_ieee_prev = 210, 205 op1_add_prev = 211, 206 op2_muladd_prev = 212, 207 op2_muladd_ieee_prev = 213, 208 op2_interp_xy = 214, 209 op2_interp_zw = 215, 210 op2_interp_x = 216, 211 op2_interp_z = 217, 212 op0_store_flags = 218, 213 op1_load_store_flags = 219, 214 op0_lds_1a = 220, 215 op0_lds_1a1d = 221, 216 op0_lds_2a = 223, 217 op1_interp_load_p0 = 224, 218 op1_interp_load_p10 = 125, 219 op1_interp_load_p20 = 126, 220 // op 3 all left shift 6 221 op3_bfe_uint = 4 << 6, 222 op3_bfe_int = 5 << 6, 223 op3_bfi_int = 6 << 6, 224 op3_fma = 7 << 6, 225 op3_cndne_64 = 9 << 6, 226 op3_fma_64 = 10 << 6, 227 op3_lerp_uint = 11 << 6, 228 op3_bit_align_int = 12 << 6, 229 op3_byte_align_int = 13 << 6, 230 op3_sad_accum_uint = 14 << 6, 231 op3_sad_accum_hi_uint = 15 << 6, 232 op3_muladd_uint24 = 16 << 6, 233 op3_lds_idx_op = 17 << 6, 234 op3_muladd = 20 << 6, 235 op3_muladd_m2 = 21 << 6, 236 op3_muladd_m4 = 22 << 6, 237 op3_muladd_d2 = 23 << 6, 238 op3_muladd_ieee = 24 << 6, 239 op3_cnde = 25 << 6, 240 op3_cndgt = 26 << 6, 241 op3_cndge = 27 << 6, 242 op3_cnde_int = 28 << 6, 243 op3_cndgt_int = 29 << 6, 244 op3_cndge_int = 30 << 6, 245 op3_mul_lit = 31 << 6, 246 op_invalid = 0xffff 247 }; 248 249 enum AluModifiers { 250 alu_src0_rel, 251 alu_src1_rel, 252 alu_src2_rel, 253 alu_dst_clamp, 254 alu_dst_rel, 255 alu_last_instr, 256 alu_update_exec, 257 alu_update_pred, 258 alu_write, 259 alu_op3, 260 alu_is_trans, 261 alu_is_cayman_trans, 262 alu_is_lds, 263 alu_lds_group_start, 264 alu_lds_group_end, 265 alu_lds_address, 266 alu_no_schedule_bias, 267 alu_64bit_op, 268 alu_flag_none, 269 alu_flag_count 270 }; 271 272 enum AluDstModifiers { 273 omod_off = 0, 274 omod_mul2 = 1, 275 omod_mul4 = 2, 276 omod_divl2 = 3 277 }; 278 279 enum AluPredSel { 280 pred_off = 0, 281 pred_zero = 2, 282 pred_one = 3 283 }; 284 285 enum AluBankSwizzle { 286 alu_vec_012 = 0, 287 sq_alu_scl_201 = 0, 288 alu_vec_021 = 1, 289 sq_alu_scl_122 = 1, 290 alu_vec_120 = 2, 291 sq_alu_scl_212 = 2, 292 alu_vec_102 = 3, 293 sq_alu_scl_221 = 3, 294 alu_vec_201 = 4, 295 sq_alu_scl_unknown = 4, 296 alu_vec_210 = 5, 297 alu_vec_unknown = 6 298 }; 299 300 inline AluBankSwizzle 301 operator++(AluBankSwizzle& x) 302 { 303 x = static_cast<AluBankSwizzle>(x + 1); 304 return x; 305 } 306 307 using AluOpFlags = std::bitset<alu_flag_count>; 308 309 struct AluOp { 310 static constexpr int x = 1; 311 static constexpr int y = 2; 312 static constexpr int z = 4; 313 static constexpr int w = 8; 314 static constexpr int v = 15; 315 static constexpr int t = 16; 316 static constexpr int a = 31; 317 AluOpAluOp318 AluOp(int ns, bool src_mod, bool clamp, bool fp64, uint8_t um_r600, 319 uint8_t um_r700, uint8_t um_eg, const char *n): 320 nsrc(ns), 321 can_srcmod(src_mod), 322 can_clamp(clamp), 323 is_fp64(fp64), 324 name(n) 325 { 326 unit_mask[0] = um_r600; 327 unit_mask[1] = um_r700; 328 unit_mask[2] = um_eg; 329 } 330 can_channelAluOp331 bool can_channel(int flags, r600_chip_class unit_type) const 332 { 333 assert(unit_type < 3); 334 return flags & unit_mask[unit_type]; 335 } 336 337 int nsrc : 4; 338 int can_srcmod : 1; 339 int can_clamp : 1; 340 int is_fp64 : 1; 341 uint8_t unit_mask[3]; 342 const char *name; 343 }; 344 345 extern const std::map<EAluOp, AluOp> alu_ops; 346 347 enum AluInlineConstants { 348 ALU_SRC_LDS_OQ_A = 219, 349 ALU_SRC_LDS_OQ_B = 220, 350 ALU_SRC_LDS_OQ_A_POP = 221, 351 ALU_SRC_LDS_OQ_B_POP = 222, 352 ALU_SRC_LDS_DIRECT_A = 223, 353 ALU_SRC_LDS_DIRECT_B = 224, 354 ALU_SRC_TIME_HI = 227, 355 ALU_SRC_TIME_LO = 228, 356 ALU_SRC_MASK_HI = 229, 357 ALU_SRC_MASK_LO = 230, 358 ALU_SRC_HW_WAVE_ID = 231, 359 ALU_SRC_SIMD_ID = 232, 360 ALU_SRC_SE_ID = 233, 361 ALU_SRC_HW_THREADGRP_ID = 234, 362 ALU_SRC_WAVE_ID_IN_GRP = 235, 363 ALU_SRC_NUM_THREADGRP_WAVES = 236, 364 ALU_SRC_HW_ALU_ODD = 237, 365 ALU_SRC_LOOP_IDX = 238, 366 ALU_SRC_PARAM_BASE_ADDR = 240, 367 ALU_SRC_NEW_PRIM_MASK = 241, 368 ALU_SRC_PRIM_MASK_HI = 242, 369 ALU_SRC_PRIM_MASK_LO = 243, 370 ALU_SRC_1_DBL_L = 244, 371 ALU_SRC_1_DBL_M = 245, 372 ALU_SRC_0_5_DBL_L = 246, 373 ALU_SRC_0_5_DBL_M = 247, 374 ALU_SRC_0 = 248, 375 ALU_SRC_1 = 249, 376 ALU_SRC_1_INT = 250, 377 ALU_SRC_M_1_INT = 251, 378 ALU_SRC_0_5 = 252, 379 ALU_SRC_LITERAL = 253, 380 ALU_SRC_PV = 254, 381 ALU_SRC_PS = 255, 382 ALU_SRC_PARAM_BASE = 0x1C0, 383 ALU_SRC_UNKNOWN 384 }; 385 386 struct AluInlineConstantDescr { 387 bool use_chan; 388 const char *descr; 389 }; 390 391 extern const std::map<AluInlineConstants, AluInlineConstantDescr> alu_src_const; 392 393 #define LDSOP2(X) LDS_##X = LDS_OP2_LDS_##X 394 395 enum ESDOp { 396 DS_OP_ADD = 0, 397 DS_OP_SUB = 1, 398 DS_OP_RSUB = 2, 399 DS_OP_INC = 3, 400 DS_OP_DEC = 4, 401 DS_OP_MIN_INT = 5, 402 DS_OP_MAX_INT = 6, 403 DS_OP_MIN_UINT = 7, 404 DS_OP_MAX_UINT = 8, 405 DS_OP_AND = 9, 406 DS_OP_OR = 10, 407 DS_OP_XOR = 11, 408 DS_OP_MSKOR = 12, 409 DS_OP_WRITE = 13, 410 DS_OP_WRITE_REL = 14, 411 DS_OP_WRITE2 = 15, 412 DS_OP_CMP_STORE = 16, 413 DS_OP_CMP_STORE_SPF = 17, 414 DS_OP_BYTE_WRITE = 18, 415 DS_OP_SHORT_WRITE = 19, 416 DS_OP_ADD_RET = 32, 417 DS_OP_SUB_RET = 33, 418 DS_OP_RSUB_RET = 34, 419 DS_OP_INC_RET = 35, 420 DS_OP_DEC_RET = 36, 421 DS_OP_MIN_INT_RET = 37, 422 DS_OP_MAX_INT_RET = 38, 423 DS_OP_MIN_UINT_RET = 39, 424 DS_OP_MAX_UINT_RET = 40, 425 DS_OP_AND_RET = 41, 426 DS_OP_OR_RET = 42, 427 DS_OP_XOR_RET = 43, 428 DS_OP_MSKOR_RET = 44, 429 DS_OP_XCHG_RET = 45, 430 DS_OP_XCHG_REL_RET = 46, 431 DS_OP_XCHG2_RET = 47, 432 DS_OP_CMP_XCHG_RET = 48, 433 DS_OP_CMP_XCHG_SPF_RET = 49, 434 DS_OP_READ_RET = 50, 435 DS_OP_READ_REL_RET = 51, 436 DS_OP_READ2_RET = 52, 437 DS_OP_READWRITE_RET = 53, 438 DS_OP_BYTE_READ_RET = 54, 439 DS_OP_UBYTE_READ_RET = 55, 440 DS_OP_SHORT_READ_RET = 56, 441 DS_OP_USHORT_READ_RET = 57, 442 DS_OP_ATOMIC_ORDERED_ALLOC_RET = 63, 443 DS_OP_INVALID = 64, 444 LDSOP2(ADD_RET), 445 LDSOP2(ADD), 446 LDSOP2(AND_RET), 447 LDSOP2(AND), 448 LDSOP2(WRITE), 449 LDSOP2(OR_RET), 450 LDSOP2(OR), 451 LDSOP2(MAX_INT_RET), 452 LDSOP2(MAX_INT), 453 LDSOP2(MAX_UINT_RET), 454 LDSOP2(MAX_UINT), 455 LDSOP2(MIN_INT_RET), 456 LDSOP2(MIN_INT), 457 LDSOP2(MIN_UINT_RET), 458 LDSOP2(MIN_UINT), 459 LDSOP2(XOR_RET), 460 LDSOP2(XOR), 461 LDSOP2(XCHG_RET), 462 LDS_CMP_XCHG_RET = LDS_OP3_LDS_CMP_XCHG_RET, 463 LDS_WRITE_REL = LDS_OP3_LDS_WRITE_REL 464 }; 465 466 #undef LDSOP2 467 468 struct LDSOp { 469 int nsrc; 470 const char *name; 471 }; 472 473 extern const std::map<ESDOp, LDSOp> lds_ops; 474 475 struct KCacheLine { 476 int bank{0}; 477 int addr{0}; 478 int len{0}; 479 int index_mode{0}; 480 enum KCacheLockMode { 481 free, 482 lock_1, 483 lock_2 484 } mode{free}; 485 }; 486 487 } // namespace r600 488 489 #endif // ALU_DEFINES_H 490