1 /* 2 * Copyright 2013 Intel Corporation 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice (including the next 12 * paragraph) shall be included in all copies or substantial portions of the 13 * Software. 14 * 15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 21 * IN THE SOFTWARE. 22 */ 23 24 #include <stdio.h> 25 #include <stdlib.h> 26 #include "gen_device_info.h" 27 #include "compiler/shader_enums.h" 28 29 static const struct gen_device_info gen_device_info_i965 = { 30 .gen = 4, 31 .has_negative_rhw_bug = true, 32 .num_slices = 1, 33 .max_vs_threads = 16, 34 .max_gs_threads = 2, 35 .max_wm_threads = 8 * 4, 36 .urb = { 37 .size = 256, 38 }, 39 }; 40 41 static const struct gen_device_info gen_device_info_g4x = { 42 .gen = 4, 43 .has_pln = true, 44 .has_compr4 = true, 45 .has_surface_tile_offset = true, 46 .is_g4x = true, 47 .num_slices = 1, 48 .max_vs_threads = 32, 49 .max_gs_threads = 2, 50 .max_wm_threads = 10 * 5, 51 .urb = { 52 .size = 384, 53 }, 54 }; 55 56 static const struct gen_device_info gen_device_info_ilk = { 57 .gen = 5, 58 .has_pln = true, 59 .has_compr4 = true, 60 .has_surface_tile_offset = true, 61 .num_slices = 1, 62 .max_vs_threads = 72, 63 .max_gs_threads = 32, 64 .max_wm_threads = 12 * 6, 65 .urb = { 66 .size = 1024, 67 }, 68 }; 69 70 static const struct gen_device_info gen_device_info_snb_gt1 = { 71 .gen = 6, 72 .gt = 1, 73 .has_hiz_and_separate_stencil = true, 74 .has_llc = true, 75 .has_pln = true, 76 .has_surface_tile_offset = true, 77 .needs_unlit_centroid_workaround = true, 78 .num_slices = 1, 79 .max_vs_threads = 24, 80 .max_gs_threads = 21, /* conservative; 24 if rendering disabled. */ 81 .max_wm_threads = 40, 82 .urb = { 83 .size = 32, 84 .min_entries = { 85 [MESA_SHADER_VERTEX] = 24, 86 }, 87 .max_entries = { 88 [MESA_SHADER_VERTEX] = 256, 89 [MESA_SHADER_GEOMETRY] = 256, 90 }, 91 }, 92 }; 93 94 static const struct gen_device_info gen_device_info_snb_gt2 = { 95 .gen = 6, 96 .gt = 2, 97 .has_hiz_and_separate_stencil = true, 98 .has_llc = true, 99 .has_pln = true, 100 .has_surface_tile_offset = true, 101 .needs_unlit_centroid_workaround = true, 102 .num_slices = 1, 103 .max_vs_threads = 60, 104 .max_gs_threads = 60, 105 .max_wm_threads = 80, 106 .urb = { 107 .size = 64, 108 .min_entries = { 109 [MESA_SHADER_VERTEX] = 24, 110 }, 111 .max_entries = { 112 [MESA_SHADER_VERTEX] = 256, 113 [MESA_SHADER_GEOMETRY] = 256, 114 }, 115 }, 116 }; 117 118 #define GEN7_FEATURES \ 119 .gen = 7, \ 120 .has_hiz_and_separate_stencil = true, \ 121 .must_use_separate_stencil = true, \ 122 .has_llc = true, \ 123 .has_pln = true, \ 124 .has_surface_tile_offset = true 125 126 static const struct gen_device_info gen_device_info_ivb_gt1 = { 127 GEN7_FEATURES, .is_ivybridge = true, .gt = 1, 128 .num_slices = 1, 129 .max_vs_threads = 36, 130 .max_tcs_threads = 36, 131 .max_tes_threads = 36, 132 .max_gs_threads = 36, 133 .max_wm_threads = 48, 134 .max_cs_threads = 36, 135 .urb = { 136 .size = 128, 137 .min_entries = { 138 [MESA_SHADER_VERTEX] = 32, 139 [MESA_SHADER_TESS_EVAL] = 10, 140 }, 141 .max_entries = { 142 [MESA_SHADER_VERTEX] = 512, 143 [MESA_SHADER_TESS_CTRL] = 32, 144 [MESA_SHADER_TESS_EVAL] = 288, 145 [MESA_SHADER_GEOMETRY] = 192, 146 }, 147 }, 148 }; 149 150 static const struct gen_device_info gen_device_info_ivb_gt2 = { 151 GEN7_FEATURES, .is_ivybridge = true, .gt = 2, 152 .num_slices = 1, 153 .max_vs_threads = 128, 154 .max_tcs_threads = 128, 155 .max_tes_threads = 128, 156 .max_gs_threads = 128, 157 .max_wm_threads = 172, 158 .max_cs_threads = 64, 159 .urb = { 160 .size = 256, 161 .min_entries = { 162 [MESA_SHADER_VERTEX] = 32, 163 [MESA_SHADER_TESS_EVAL] = 10, 164 }, 165 .max_entries = { 166 [MESA_SHADER_VERTEX] = 704, 167 [MESA_SHADER_TESS_CTRL] = 64, 168 [MESA_SHADER_TESS_EVAL] = 448, 169 [MESA_SHADER_GEOMETRY] = 320, 170 }, 171 }, 172 }; 173 174 static const struct gen_device_info gen_device_info_byt = { 175 GEN7_FEATURES, .is_baytrail = true, .gt = 1, 176 .num_slices = 1, 177 .has_llc = false, 178 .max_vs_threads = 36, 179 .max_tcs_threads = 36, 180 .max_tes_threads = 36, 181 .max_gs_threads = 36, 182 .max_wm_threads = 48, 183 .max_cs_threads = 32, 184 .urb = { 185 .size = 128, 186 .min_entries = { 187 [MESA_SHADER_VERTEX] = 32, 188 [MESA_SHADER_TESS_EVAL] = 10, 189 }, 190 .max_entries = { 191 [MESA_SHADER_VERTEX] = 512, 192 [MESA_SHADER_TESS_CTRL] = 32, 193 [MESA_SHADER_TESS_EVAL] = 288, 194 [MESA_SHADER_GEOMETRY] = 192, 195 }, 196 }, 197 }; 198 199 #define HSW_FEATURES \ 200 GEN7_FEATURES, \ 201 .is_haswell = true, \ 202 .supports_simd16_3src = true, \ 203 .has_resource_streamer = true 204 205 static const struct gen_device_info gen_device_info_hsw_gt1 = { 206 HSW_FEATURES, .gt = 1, 207 .num_slices = 1, 208 .max_vs_threads = 70, 209 .max_tcs_threads = 70, 210 .max_tes_threads = 70, 211 .max_gs_threads = 70, 212 .max_wm_threads = 102, 213 .max_cs_threads = 70, 214 .urb = { 215 .size = 128, 216 .min_entries = { 217 [MESA_SHADER_VERTEX] = 32, 218 [MESA_SHADER_TESS_EVAL] = 10, 219 }, 220 .max_entries = { 221 [MESA_SHADER_VERTEX] = 640, 222 [MESA_SHADER_TESS_CTRL] = 64, 223 [MESA_SHADER_TESS_EVAL] = 384, 224 [MESA_SHADER_GEOMETRY] = 256, 225 }, 226 }, 227 }; 228 229 static const struct gen_device_info gen_device_info_hsw_gt2 = { 230 HSW_FEATURES, .gt = 2, 231 .num_slices = 1, 232 .max_vs_threads = 280, 233 .max_tcs_threads = 256, 234 .max_tes_threads = 280, 235 .max_gs_threads = 256, 236 .max_wm_threads = 204, 237 .max_cs_threads = 70, 238 .urb = { 239 .size = 256, 240 .min_entries = { 241 [MESA_SHADER_VERTEX] = 64, 242 [MESA_SHADER_TESS_EVAL] = 10, 243 }, 244 .max_entries = { 245 [MESA_SHADER_VERTEX] = 1664, 246 [MESA_SHADER_TESS_CTRL] = 128, 247 [MESA_SHADER_TESS_EVAL] = 960, 248 [MESA_SHADER_GEOMETRY] = 640, 249 }, 250 }, 251 }; 252 253 static const struct gen_device_info gen_device_info_hsw_gt3 = { 254 HSW_FEATURES, .gt = 3, 255 .num_slices = 2, 256 .max_vs_threads = 280, 257 .max_tcs_threads = 256, 258 .max_tes_threads = 280, 259 .max_gs_threads = 256, 260 .max_wm_threads = 408, 261 .max_cs_threads = 70, 262 .urb = { 263 .size = 512, 264 .min_entries = { 265 [MESA_SHADER_VERTEX] = 64, 266 [MESA_SHADER_TESS_EVAL] = 10, 267 }, 268 .max_entries = { 269 [MESA_SHADER_VERTEX] = 1664, 270 [MESA_SHADER_TESS_CTRL] = 128, 271 [MESA_SHADER_TESS_EVAL] = 960, 272 [MESA_SHADER_GEOMETRY] = 640, 273 }, 274 }, 275 }; 276 277 #define GEN8_FEATURES \ 278 .gen = 8, \ 279 .has_hiz_and_separate_stencil = true, \ 280 .has_resource_streamer = true, \ 281 .must_use_separate_stencil = true, \ 282 .has_llc = true, \ 283 .has_pln = true, \ 284 .supports_simd16_3src = true, \ 285 .has_surface_tile_offset = true, \ 286 .max_vs_threads = 504, \ 287 .max_tcs_threads = 504, \ 288 .max_tes_threads = 504, \ 289 .max_gs_threads = 504, \ 290 .max_wm_threads = 384 291 292 static const struct gen_device_info gen_device_info_bdw_gt1 = { 293 GEN8_FEATURES, .gt = 1, 294 .num_slices = 1, 295 .max_cs_threads = 42, 296 .urb = { 297 .size = 192, 298 .min_entries = { 299 [MESA_SHADER_VERTEX] = 64, 300 [MESA_SHADER_TESS_EVAL] = 34, 301 }, 302 .max_entries = { 303 [MESA_SHADER_VERTEX] = 2560, 304 [MESA_SHADER_TESS_CTRL] = 504, 305 [MESA_SHADER_TESS_EVAL] = 1536, 306 [MESA_SHADER_GEOMETRY] = 960, 307 }, 308 } 309 }; 310 311 static const struct gen_device_info gen_device_info_bdw_gt2 = { 312 GEN8_FEATURES, .gt = 2, 313 .num_slices = 1, 314 .max_cs_threads = 56, 315 .urb = { 316 .size = 384, 317 .min_entries = { 318 [MESA_SHADER_VERTEX] = 64, 319 [MESA_SHADER_TESS_EVAL] = 34, 320 }, 321 .max_entries = { 322 [MESA_SHADER_VERTEX] = 2560, 323 [MESA_SHADER_TESS_CTRL] = 504, 324 [MESA_SHADER_TESS_EVAL] = 1536, 325 [MESA_SHADER_GEOMETRY] = 960, 326 }, 327 } 328 }; 329 330 static const struct gen_device_info gen_device_info_bdw_gt3 = { 331 GEN8_FEATURES, .gt = 3, 332 .num_slices = 2, 333 .max_cs_threads = 56, 334 .urb = { 335 .size = 384, 336 .min_entries = { 337 [MESA_SHADER_VERTEX] = 64, 338 [MESA_SHADER_TESS_EVAL] = 34, 339 }, 340 .max_entries = { 341 [MESA_SHADER_VERTEX] = 2560, 342 [MESA_SHADER_TESS_CTRL] = 504, 343 [MESA_SHADER_TESS_EVAL] = 1536, 344 [MESA_SHADER_GEOMETRY] = 960, 345 }, 346 } 347 }; 348 349 static const struct gen_device_info gen_device_info_chv = { 350 GEN8_FEATURES, .is_cherryview = 1, .gt = 1, 351 .has_llc = false, 352 .num_slices = 1, 353 .max_vs_threads = 80, 354 .max_tcs_threads = 80, 355 .max_tes_threads = 80, 356 .max_gs_threads = 80, 357 .max_wm_threads = 128, 358 .max_cs_threads = 6 * 7, 359 .urb = { 360 .size = 192, 361 .min_entries = { 362 [MESA_SHADER_VERTEX] = 34, 363 [MESA_SHADER_TESS_EVAL] = 34, 364 }, 365 .max_entries = { 366 [MESA_SHADER_VERTEX] = 640, 367 [MESA_SHADER_TESS_CTRL] = 80, 368 [MESA_SHADER_TESS_EVAL] = 384, 369 [MESA_SHADER_GEOMETRY] = 256, 370 }, 371 } 372 }; 373 374 #define GEN9_FEATURES \ 375 .gen = 9, \ 376 .has_hiz_and_separate_stencil = true, \ 377 .has_resource_streamer = true, \ 378 .must_use_separate_stencil = true, \ 379 .has_llc = true, \ 380 .has_pln = true, \ 381 .supports_simd16_3src = true, \ 382 .has_surface_tile_offset = true, \ 383 .max_vs_threads = 336, \ 384 .max_gs_threads = 336, \ 385 .max_tcs_threads = 336, \ 386 .max_tes_threads = 336, \ 387 .max_cs_threads = 56, \ 388 .urb = { \ 389 .size = 384, \ 390 .min_entries = { \ 391 [MESA_SHADER_VERTEX] = 64, \ 392 [MESA_SHADER_TESS_EVAL] = 34, \ 393 }, \ 394 .max_entries = { \ 395 [MESA_SHADER_VERTEX] = 1856, \ 396 [MESA_SHADER_TESS_CTRL] = 672, \ 397 [MESA_SHADER_TESS_EVAL] = 1120, \ 398 [MESA_SHADER_GEOMETRY] = 640, \ 399 }, \ 400 } 401 402 #define GEN9_LP_FEATURES \ 403 GEN9_FEATURES, \ 404 .is_broxton = 1, \ 405 .gt = 1, \ 406 .has_llc = false, \ 407 .num_slices = 1, \ 408 .max_vs_threads = 112, \ 409 .max_tcs_threads = 112, \ 410 .max_tes_threads = 112, \ 411 .max_gs_threads = 112, \ 412 .max_cs_threads = 6 * 6, \ 413 .urb = { \ 414 .size = 192, \ 415 .min_entries = { \ 416 [MESA_SHADER_VERTEX] = 34, \ 417 [MESA_SHADER_TESS_EVAL] = 34, \ 418 }, \ 419 .max_entries = { \ 420 [MESA_SHADER_VERTEX] = 704, \ 421 [MESA_SHADER_TESS_CTRL] = 256, \ 422 [MESA_SHADER_TESS_EVAL] = 416, \ 423 [MESA_SHADER_GEOMETRY] = 256, \ 424 }, \ 425 } 426 427 #define GEN9_LP_FEATURES_2X6 \ 428 GEN9_LP_FEATURES, \ 429 .max_vs_threads = 56, \ 430 .max_tcs_threads = 56, \ 431 .max_tes_threads = 56, \ 432 .max_gs_threads = 56, \ 433 .max_cs_threads = 6 * 6, \ 434 .urb = { \ 435 .size = 128, \ 436 .min_entries = { \ 437 [MESA_SHADER_VERTEX] = 34, \ 438 [MESA_SHADER_TESS_EVAL] = 34, \ 439 }, \ 440 .max_entries = { \ 441 [MESA_SHADER_VERTEX] = 352, \ 442 [MESA_SHADER_TESS_CTRL] = 128, \ 443 [MESA_SHADER_TESS_EVAL] = 208, \ 444 [MESA_SHADER_GEOMETRY] = 128, \ 445 }, \ 446 } 447 448 static const struct gen_device_info gen_device_info_skl_gt1 = { 449 GEN9_FEATURES, .gt = 1, 450 .num_slices = 1, 451 .urb.size = 192, 452 }; 453 454 static const struct gen_device_info gen_device_info_skl_gt2 = { 455 GEN9_FEATURES, .gt = 2, 456 .num_slices = 1, 457 }; 458 459 static const struct gen_device_info gen_device_info_skl_gt3 = { 460 GEN9_FEATURES, .gt = 3, 461 .num_slices = 2, 462 }; 463 464 static const struct gen_device_info gen_device_info_skl_gt4 = { 465 GEN9_FEATURES, .gt = 4, 466 .num_slices = 3, 467 /* From the "L3 Allocation and Programming" documentation: 468 * 469 * "URB is limited to 1008KB due to programming restrictions. This is not a 470 * restriction of the L3 implementation, but of the FF and other clients. 471 * Therefore, in a GT4 implementation it is possible for the programmed 472 * allocation of the L3 data array to provide 3*384KB=1152KB for URB, but 473 * only 1008KB of this will be used." 474 */ 475 .urb.size = 1008 / 3, 476 }; 477 478 static const struct gen_device_info gen_device_info_bxt = { 479 GEN9_LP_FEATURES 480 }; 481 482 static const struct gen_device_info gen_device_info_bxt_2x6 = { 483 GEN9_LP_FEATURES_2X6 484 }; 485 /* 486 * Note: for all KBL SKUs, the PRM says SKL for GS entries, not SKL+. 487 * There's no KBL entry. Using the default SKL (GEN9) GS entries value. 488 */ 489 490 static const struct gen_device_info gen_device_info_kbl_gt1 = { 491 GEN9_FEATURES, 492 .is_kabylake = true, 493 .gt = 1, 494 495 .max_cs_threads = 7 * 6, 496 .urb.size = 192, 497 .num_slices = 1, 498 }; 499 500 static const struct gen_device_info gen_device_info_kbl_gt1_5 = { 501 GEN9_FEATURES, 502 .is_kabylake = true, 503 .gt = 1, 504 505 .max_cs_threads = 7 * 6, 506 .num_slices = 1, 507 }; 508 509 static const struct gen_device_info gen_device_info_kbl_gt2 = { 510 GEN9_FEATURES, 511 .is_kabylake = true, 512 .gt = 2, 513 514 .num_slices = 1, 515 }; 516 517 static const struct gen_device_info gen_device_info_kbl_gt3 = { 518 GEN9_FEATURES, 519 .is_kabylake = true, 520 .gt = 3, 521 522 .num_slices = 2, 523 }; 524 525 static const struct gen_device_info gen_device_info_kbl_gt4 = { 526 GEN9_FEATURES, 527 .is_kabylake = true, 528 .gt = 4, 529 530 /* 531 * From the "L3 Allocation and Programming" documentation: 532 * 533 * "URB is limited to 1008KB due to programming restrictions. This 534 * is not a restriction of the L3 implementation, but of the FF and 535 * other clients. Therefore, in a GT4 implementation it is 536 * possible for the programmed allocation of the L3 data array to 537 * provide 3*384KB=1152KB for URB, but only 1008KB of this 538 * will be used." 539 */ 540 .urb.size = 1008 / 3, 541 .num_slices = 3, 542 }; 543 544 static const struct gen_device_info gen_device_info_glk = { 545 GEN9_LP_FEATURES 546 }; 547 548 static const struct gen_device_info gen_device_info_glk_2x6 = { 549 GEN9_LP_FEATURES_2X6 550 }; 551 552 bool 553 gen_get_device_info(int devid, struct gen_device_info *devinfo) 554 { 555 switch (devid) { 556 #undef CHIPSET 557 #define CHIPSET(id, family, name) \ 558 case id: *devinfo = gen_device_info_##family; break; 559 #include "pci_ids/i965_pci_ids.h" 560 default: 561 fprintf(stderr, "i965_dri.so does not support the 0x%x PCI ID.\n", devid); 562 return false; 563 } 564 565 /* From the Skylake PRM, 3DSTATE_PS::Scratch Space Base Pointer: 566 * 567 * "Scratch Space per slice is computed based on 4 sub-slices. SW must 568 * allocate scratch space enough so that each slice has 4 slices allowed." 569 * 570 * The equivalent internal documentation says that this programming note 571 * applies to all Gen9+ platforms. 572 * 573 * The hardware typically calculates the scratch space pointer by taking 574 * the base address, and adding per-thread-scratch-space * thread ID. 575 * Extra padding can be necessary depending how the thread IDs are 576 * calculated for a particular shader stage. 577 */ 578 if (devinfo->gen >= 9) { 579 devinfo->max_wm_threads = 64 /* threads-per-PSD */ 580 * devinfo->num_slices 581 * 4; /* effective subslices per slice */ 582 } 583 584 return true; 585 } 586 587 const char * 588 gen_get_device_name(int devid) 589 { 590 switch (devid) { 591 #undef CHIPSET 592 #define CHIPSET(id, family, name) case id: return name; 593 #include "pci_ids/i965_pci_ids.h" 594 default: 595 return NULL; 596 } 597 } 598