1/* 2 * Copyright (c) 2012-2015 Etnaviv Project 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, sub license, 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 12 * next paragraph) shall be included in all copies or substantial portions 13 * of the 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 NON-INFRINGEMENT. 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 21 * DEALINGS IN THE SOFTWARE. 22 * 23 * Authors: 24 * Wladimir J. van der Laan <laanwj@gmail.com> 25 */ 26 27#include "etnaviv_shader.h" 28 29#include "etnaviv_compiler.h" 30#include "etnaviv_context.h" 31#include "etnaviv_debug.h" 32#include "etnaviv_disasm.h" 33#include "etnaviv_disk_cache.h" 34#include "etnaviv_screen.h" 35#include "etnaviv_util.h" 36 37#include "tgsi/tgsi_parse.h" 38#include "nir/tgsi_to_nir.h" 39#include "util/u_atomic.h" 40#include "util/u_cpu_detect.h" 41#include "util/u_math.h" 42#include "util/u_memory.h" 43 44/* Upload shader code to bo, if not already done */ 45static bool etna_icache_upload_shader(struct etna_context *ctx, struct etna_shader_variant *v) 46{ 47 if (v->bo) 48 return true; 49 v->bo = etna_bo_new(ctx->screen->dev, v->code_size*4, DRM_ETNA_GEM_CACHE_WC); 50 if (!v->bo) 51 return false; 52 53 void *buf = etna_bo_map(v->bo); 54 etna_bo_cpu_prep(v->bo, DRM_ETNA_PREP_WRITE); 55 memcpy(buf, v->code, v->code_size*4); 56 etna_bo_cpu_fini(v->bo); 57 DBG("Uploaded %s of %u words to bo %p", v->stage == MESA_SHADER_FRAGMENT ? "fs":"vs", v->code_size, v->bo); 58 return true; 59} 60 61extern const char *tgsi_swizzle_names[]; 62void 63etna_dump_shader(const struct etna_shader_variant *shader) 64{ 65 if (shader->stage == MESA_SHADER_VERTEX) 66 printf("VERT\n"); 67 else 68 printf("FRAG\n"); 69 70 etna_disasm(shader->code, shader->code_size, PRINT_RAW); 71 72 printf("num loops: %i\n", shader->num_loops); 73 printf("num temps: %i\n", shader->num_temps); 74 printf("immediates:\n"); 75 for (int idx = 0; idx < shader->uniforms.count; ++idx) { 76 printf(" [%i].%s = %f (0x%08x) (%d)\n", 77 idx / 4, 78 tgsi_swizzle_names[idx % 4], 79 *((float *)&shader->uniforms.data[idx]), 80 shader->uniforms.data[idx], 81 shader->uniforms.contents[idx]); 82 } 83 printf("inputs:\n"); 84 for (int idx = 0; idx < shader->infile.num_reg; ++idx) { 85 printf(" [%i] name=%s comps=%i\n", shader->infile.reg[idx].reg, 86 (shader->stage == MESA_SHADER_VERTEX) ? 87 gl_vert_attrib_name(shader->infile.reg[idx].slot) : 88 gl_varying_slot_name_for_stage(shader->infile.reg[idx].slot, shader->stage), 89 shader->infile.reg[idx].num_components); 90 } 91 printf("outputs:\n"); 92 for (int idx = 0; idx < shader->outfile.num_reg; ++idx) { 93 printf(" [%i] name=%s comps=%i\n", shader->outfile.reg[idx].reg, 94 (shader->stage == MESA_SHADER_VERTEX) ? 95 gl_varying_slot_name_for_stage(shader->outfile.reg[idx].slot, shader->stage) : 96 gl_frag_result_name(shader->outfile.reg[idx].slot), 97 shader->outfile.reg[idx].num_components); 98 } 99 printf("special:\n"); 100 if (shader->stage == MESA_SHADER_VERTEX) { 101 printf(" vs_pos_out_reg=%i\n", shader->vs_pos_out_reg); 102 printf(" vs_pointsize_out_reg=%i\n", shader->vs_pointsize_out_reg); 103 printf(" vs_load_balancing=0x%08x\n", shader->vs_load_balancing); 104 } else { 105 printf(" ps_color_out_reg=%i\n", shader->ps_color_out_reg); 106 printf(" ps_depth_out_reg=%i\n", shader->ps_depth_out_reg); 107 } 108 printf(" input_count_unk8=0x%08x\n", shader->input_count_unk8); 109} 110 111/* Link vs and fs together: fill in shader_state from vs and fs 112 * as this function is called every time a new fs or vs is bound, the goal is to 113 * do little processing as possible here, and to precompute as much as possible in 114 * the vs/fs shader_object. 115 * 116 * XXX we could cache the link result for a certain set of VS/PS; usually a pair 117 * of VS and PS will be used together anyway. 118 */ 119static bool 120etna_link_shaders(struct etna_context *ctx, struct compiled_shader_state *cs, 121 struct etna_shader_variant *vs, struct etna_shader_variant *fs) 122{ 123 struct etna_shader_link_info link = { }; 124 bool failed; 125 126 assert(vs->stage == MESA_SHADER_VERTEX); 127 assert(fs->stage == MESA_SHADER_FRAGMENT); 128 129#ifdef DEBUG 130 if (DBG_ENABLED(ETNA_DBG_DUMP_SHADERS)) { 131 etna_dump_shader(vs); 132 etna_dump_shader(fs); 133 } 134#endif 135 136 failed = etna_link_shader(&link, vs, fs); 137 138 if (failed) { 139 /* linking failed: some fs inputs do not have corresponding 140 * vs outputs */ 141 assert(0); 142 143 return false; 144 } 145 146 if (DBG_ENABLED(ETNA_DBG_LINKER_MSGS)) { 147 debug_printf("link result:\n"); 148 debug_printf(" vs -> fs comps use pa_attr\n"); 149 150 for (int idx = 0; idx < link.num_varyings; ++idx) 151 debug_printf(" t%-2u -> t%-2u %-5.*s %u,%u,%u,%u 0x%08x\n", 152 link.varyings[idx].reg, idx + 1, 153 link.varyings[idx].num_components, "xyzw", 154 link.varyings[idx].use[0], link.varyings[idx].use[1], 155 link.varyings[idx].use[2], link.varyings[idx].use[3], 156 link.varyings[idx].pa_attributes); 157 } 158 159 /* set last_varying_2x flag if the last varying has 1 or 2 components */ 160 bool last_varying_2x = false; 161 if (link.num_varyings > 0 && link.varyings[link.num_varyings - 1].num_components <= 2) 162 last_varying_2x = true; 163 164 cs->RA_CONTROL = VIVS_RA_CONTROL_UNK0 | 165 COND(last_varying_2x, VIVS_RA_CONTROL_LAST_VARYING_2X); 166 167 cs->PA_ATTRIBUTE_ELEMENT_COUNT = VIVS_PA_ATTRIBUTE_ELEMENT_COUNT_COUNT(link.num_varyings); 168 for (int idx = 0; idx < link.num_varyings; ++idx) 169 cs->PA_SHADER_ATTRIBUTES[idx] = link.varyings[idx].pa_attributes; 170 171 cs->VS_END_PC = vs->code_size / 4; 172 cs->VS_OUTPUT_COUNT = 1 + link.num_varyings; /* position + varyings */ 173 174 /* vs outputs (varyings) */ 175 DEFINE_ETNA_BITARRAY(vs_output, 16, 8) = {0}; 176 int varid = 0; 177 etna_bitarray_set(vs_output, 8, varid++, vs->vs_pos_out_reg); 178 for (int idx = 0; idx < link.num_varyings; ++idx) 179 etna_bitarray_set(vs_output, 8, varid++, link.varyings[idx].reg); 180 if (vs->vs_pointsize_out_reg >= 0) 181 etna_bitarray_set(vs_output, 8, varid++, vs->vs_pointsize_out_reg); /* pointsize is last */ 182 183 for (int idx = 0; idx < ARRAY_SIZE(cs->VS_OUTPUT); ++idx) 184 cs->VS_OUTPUT[idx] = vs_output[idx]; 185 186 if (vs->vs_pointsize_out_reg != -1) { 187 /* vertex shader outputs point coordinate, provide extra output and make 188 * sure PA config is 189 * not masked */ 190 cs->PA_CONFIG = ~0; 191 cs->VS_OUTPUT_COUNT_PSIZE = cs->VS_OUTPUT_COUNT + 1; 192 } else { 193 /* vertex shader does not output point coordinate, make sure thate 194 * POINT_SIZE_ENABLE is masked 195 * and no extra output is given */ 196 cs->PA_CONFIG = ~VIVS_PA_CONFIG_POINT_SIZE_ENABLE; 197 cs->VS_OUTPUT_COUNT_PSIZE = cs->VS_OUTPUT_COUNT; 198 } 199 200 /* if fragment shader doesn't read pointcoord, disable it */ 201 if (link.pcoord_varying_comp_ofs == -1) 202 cs->PA_CONFIG &= ~VIVS_PA_CONFIG_POINT_SPRITE_ENABLE; 203 204 cs->VS_LOAD_BALANCING = vs->vs_load_balancing; 205 cs->VS_START_PC = 0; 206 207 cs->PS_END_PC = fs->code_size / 4; 208 cs->PS_OUTPUT_REG = fs->ps_color_out_reg; 209 cs->PS_INPUT_COUNT = 210 VIVS_PS_INPUT_COUNT_COUNT(link.num_varyings + 1) | /* Number of inputs plus position */ 211 VIVS_PS_INPUT_COUNT_UNK8(fs->input_count_unk8); 212 cs->PS_TEMP_REGISTER_CONTROL = 213 VIVS_PS_TEMP_REGISTER_CONTROL_NUM_TEMPS(MAX2(fs->num_temps, link.num_varyings + 1)); 214 cs->PS_START_PC = 0; 215 216 /* Precompute PS_INPUT_COUNT and TEMP_REGISTER_CONTROL in the case of MSAA 217 * mode, avoids some fumbling in sync_context. */ 218 cs->PS_INPUT_COUNT_MSAA = 219 VIVS_PS_INPUT_COUNT_COUNT(link.num_varyings + 2) | /* MSAA adds another input */ 220 VIVS_PS_INPUT_COUNT_UNK8(fs->input_count_unk8); 221 cs->PS_TEMP_REGISTER_CONTROL_MSAA = 222 VIVS_PS_TEMP_REGISTER_CONTROL_NUM_TEMPS(MAX2(fs->num_temps, link.num_varyings + 2)); 223 224 uint32_t total_components = 0; 225 DEFINE_ETNA_BITARRAY(num_components, ETNA_NUM_VARYINGS, 4) = {0}; 226 DEFINE_ETNA_BITARRAY(component_use, 4 * ETNA_NUM_VARYINGS, 2) = {0}; 227 for (int idx = 0; idx < link.num_varyings; ++idx) { 228 const struct etna_varying *varying = &link.varyings[idx]; 229 230 etna_bitarray_set(num_components, 4, idx, varying->num_components); 231 for (int comp = 0; comp < varying->num_components; ++comp) { 232 etna_bitarray_set(component_use, 2, total_components, varying->use[comp]); 233 total_components += 1; 234 } 235 } 236 237 cs->GL_VARYING_TOTAL_COMPONENTS = 238 VIVS_GL_VARYING_TOTAL_COMPONENTS_NUM(align(total_components, 2)); 239 cs->GL_VARYING_NUM_COMPONENTS[0] = num_components[0]; 240 cs->GL_VARYING_NUM_COMPONENTS[1] = num_components[1]; 241 cs->GL_VARYING_COMPONENT_USE[0] = component_use[0]; 242 cs->GL_VARYING_COMPONENT_USE[1] = component_use[1]; 243 244 cs->GL_HALTI5_SH_SPECIALS = 245 0x7f7f0000 | /* unknown bits, probably other PS inputs */ 246 /* pointsize is last (see above) */ 247 VIVS_GL_HALTI5_SH_SPECIALS_VS_PSIZE_OUT((vs->vs_pointsize_out_reg != -1) ? 248 cs->VS_OUTPUT_COUNT * 4 : 0x00) | 249 VIVS_GL_HALTI5_SH_SPECIALS_PS_PCOORD_IN((link.pcoord_varying_comp_ofs != -1) ? 250 link.pcoord_varying_comp_ofs : 0x7f); 251 252 cs->writes_z = fs->ps_depth_out_reg >= 0; 253 cs->uses_discard = fs->uses_discard; 254 255 /* reference instruction memory */ 256 cs->vs_inst_mem_size = vs->code_size; 257 cs->VS_INST_MEM = vs->code; 258 259 cs->ps_inst_mem_size = fs->code_size; 260 cs->PS_INST_MEM = fs->code; 261 262 if (vs->needs_icache || fs->needs_icache) { 263 /* If either of the shaders needs ICACHE, we use it for both. It is 264 * either switched on or off for the entire shader processor. 265 */ 266 if (!etna_icache_upload_shader(ctx, vs) || 267 !etna_icache_upload_shader(ctx, fs)) { 268 assert(0); 269 return false; 270 } 271 272 cs->VS_INST_ADDR.bo = vs->bo; 273 cs->VS_INST_ADDR.offset = 0; 274 cs->VS_INST_ADDR.flags = ETNA_RELOC_READ; 275 cs->PS_INST_ADDR.bo = fs->bo; 276 cs->PS_INST_ADDR.offset = 0; 277 cs->PS_INST_ADDR.flags = ETNA_RELOC_READ; 278 } else { 279 /* clear relocs */ 280 memset(&cs->VS_INST_ADDR, 0, sizeof(cs->VS_INST_ADDR)); 281 memset(&cs->PS_INST_ADDR, 0, sizeof(cs->PS_INST_ADDR)); 282 } 283 284 return true; 285} 286 287bool 288etna_shader_link(struct etna_context *ctx) 289{ 290 if (!ctx->shader.vs || !ctx->shader.fs) 291 return false; 292 293 /* re-link vs and fs if needed */ 294 return etna_link_shaders(ctx, &ctx->shader_state, ctx->shader.vs, ctx->shader.fs); 295} 296 297void 298etna_destroy_shader(struct etna_shader_variant *shader) 299{ 300 assert(shader); 301 302 FREE(shader->code); 303 FREE(shader->uniforms.data); 304 FREE(shader->uniforms.contents); 305 FREE(shader); 306} 307 308static bool 309etna_shader_update_vs_inputs(struct compiled_shader_state *cs, 310 const struct etna_shader_variant *vs, 311 const struct compiled_vertex_elements_state *ves) 312{ 313 unsigned num_temps, cur_temp, num_vs_inputs; 314 315 if (!vs) 316 return false; 317 318 /* Number of vertex elements determines number of VS inputs. Otherwise, 319 * the GPU crashes. Allocate any unused vertex elements to VS temporary 320 * registers. */ 321 num_vs_inputs = MAX2(ves->num_elements, vs->infile.num_reg); 322 if (num_vs_inputs != ves->num_elements) { 323 BUG("Number of elements %u does not match the number of VS inputs %zu", 324 ves->num_elements, vs->infile.num_reg); 325 return false; 326 } 327 328 cur_temp = vs->num_temps; 329 num_temps = num_vs_inputs - vs->infile.num_reg + cur_temp; 330 331 cs->VS_INPUT_COUNT = VIVS_VS_INPUT_COUNT_COUNT(num_vs_inputs) | 332 VIVS_VS_INPUT_COUNT_UNK8(vs->input_count_unk8); 333 cs->VS_TEMP_REGISTER_CONTROL = 334 VIVS_VS_TEMP_REGISTER_CONTROL_NUM_TEMPS(num_temps); 335 336 /* vs inputs (attributes) */ 337 DEFINE_ETNA_BITARRAY(vs_input, 16, 8) = {0}; 338 for (int idx = 0; idx < num_vs_inputs; ++idx) { 339 if (idx < vs->infile.num_reg) 340 etna_bitarray_set(vs_input, 8, idx, vs->infile.reg[idx].reg); 341 else 342 etna_bitarray_set(vs_input, 8, idx, cur_temp++); 343 } 344 345 if (vs->vs_id_in_reg >= 0) { 346 cs->VS_INPUT_COUNT = VIVS_VS_INPUT_COUNT_COUNT(num_vs_inputs + 1) | 347 VIVS_VS_INPUT_COUNT_UNK8(vs->input_count_unk8) | 348 VIVS_VS_INPUT_COUNT_ID_ENABLE; 349 350 etna_bitarray_set(vs_input, 8, num_vs_inputs, vs->vs_id_in_reg); 351 352 cs->FE_HALTI5_ID_CONFIG = 353 VIVS_FE_HALTI5_ID_CONFIG_VERTEX_ID_ENABLE | 354 VIVS_FE_HALTI5_ID_CONFIG_INSTANCE_ID_ENABLE | 355 VIVS_FE_HALTI5_ID_CONFIG_VERTEX_ID_REG(vs->vs_id_in_reg * 4) | 356 VIVS_FE_HALTI5_ID_CONFIG_INSTANCE_ID_REG(vs->vs_id_in_reg * 4 + 1); 357 } 358 359 for (int idx = 0; idx < ARRAY_SIZE(cs->VS_INPUT); ++idx) 360 cs->VS_INPUT[idx] = vs_input[idx]; 361 362 return true; 363} 364 365static inline const char * 366etna_shader_stage(struct etna_shader_variant *shader) 367{ 368 switch (shader->stage) { 369 case MESA_SHADER_VERTEX: return "VERT"; 370 case MESA_SHADER_FRAGMENT: return "FRAG"; 371 case MESA_SHADER_COMPUTE: return "CL"; 372 default: 373 unreachable("invalid type"); 374 return NULL; 375 } 376} 377 378static void 379dump_shader_info(struct etna_shader_variant *v, struct util_debug_callback *debug) 380{ 381 if (!unlikely(etna_mesa_debug & ETNA_DBG_SHADERDB)) 382 return; 383 384 util_debug_message(debug, SHADER_INFO, 385 "%s shader: %u instructions, %u temps, " 386 "%u immediates, %u loops", 387 etna_shader_stage(v), 388 v->code_size, 389 v->num_temps, 390 v->uniforms.count, 391 v->num_loops); 392} 393 394bool 395etna_shader_update_vertex(struct etna_context *ctx) 396{ 397 return etna_shader_update_vs_inputs(&ctx->shader_state, ctx->shader.vs, 398 ctx->vertex_elements); 399} 400 401static struct etna_shader_variant * 402create_variant(struct etna_shader *shader, struct etna_shader_key key) 403{ 404 struct etna_shader_variant *v = CALLOC_STRUCT(etna_shader_variant); 405 int ret; 406 407 if (!v) 408 return NULL; 409 410 v->shader = shader; 411 v->key = key; 412 v->id = ++shader->variant_count; 413 414 if (etna_disk_cache_retrieve(shader->compiler, v)) 415 return v; 416 417 ret = etna_compile_shader(v); 418 if (!ret) { 419 debug_error("compile failed!"); 420 goto fail; 421 } 422 423 etna_disk_cache_store(shader->compiler, v); 424 425 return v; 426 427fail: 428 FREE(v); 429 return NULL; 430} 431 432struct etna_shader_variant * 433etna_shader_variant(struct etna_shader *shader, struct etna_shader_key key, 434 struct util_debug_callback *debug) 435{ 436 struct etna_shader_variant *v; 437 438 for (v = shader->variants; v; v = v->next) 439 if (etna_shader_key_equal(&key, &v->key)) 440 return v; 441 442 /* compile new variant if it doesn't exist already */ 443 v = create_variant(shader, key); 444 if (v) { 445 v->next = shader->variants; 446 shader->variants = v; 447 dump_shader_info(v, debug); 448 } 449 450 return v; 451} 452 453/** 454 * Should initial variants be compiled synchronously? 455 * 456 * The only case where pipe_debug_message() is used in the initial-variants 457 * path is with ETNA_MESA_DEBUG=shaderdb. So if either debug is disabled (ie. 458 * debug.debug_message==NULL), or shaderdb stats are not enabled, we can 459 * compile the initial shader variant asynchronously. 460 */ 461static inline bool 462initial_variants_synchronous(struct etna_context *ctx) 463{ 464 return unlikely(ctx->debug.debug_message) || (etna_mesa_debug & ETNA_DBG_SHADERDB); 465} 466 467static void 468create_initial_variants_async(void *job, void *gdata, int thread_index) 469{ 470 struct etna_shader *shader = job; 471 struct util_debug_callback debug = {}; 472 static struct etna_shader_key key; 473 474 etna_shader_variant(shader, key, &debug); 475} 476 477static void * 478etna_create_shader_state(struct pipe_context *pctx, 479 const struct pipe_shader_state *pss) 480{ 481 struct etna_context *ctx = etna_context(pctx); 482 struct etna_screen *screen = ctx->screen; 483 struct etna_compiler *compiler = screen->compiler; 484 struct etna_shader *shader = CALLOC_STRUCT(etna_shader); 485 486 if (!shader) 487 return NULL; 488 489 shader->id = p_atomic_inc_return(&compiler->shader_count); 490 shader->specs = &screen->specs; 491 shader->compiler = screen->compiler; 492 util_queue_fence_init(&shader->ready); 493 494 shader->nir = (pss->type == PIPE_SHADER_IR_NIR) ? pss->ir.nir : 495 tgsi_to_nir(pss->tokens, pctx->screen, false); 496 497 etna_disk_cache_init_shader_key(compiler, shader); 498 499 if (initial_variants_synchronous(ctx)) { 500 struct etna_shader_key key = {}; 501 etna_shader_variant(shader, key, &ctx->debug); 502 } else { 503 struct etna_screen *screen = ctx->screen; 504 util_queue_add_job(&screen->shader_compiler_queue, shader, &shader->ready, 505 create_initial_variants_async, NULL, 0); 506 } 507 508 return shader; 509} 510 511static void 512etna_delete_shader_state(struct pipe_context *pctx, void *ss) 513{ 514 struct etna_context *ctx = etna_context(pctx); 515 struct etna_screen *screen = ctx->screen; 516 struct etna_shader *shader = ss; 517 struct etna_shader_variant *v, *t; 518 519 util_queue_drop_job(&screen->shader_compiler_queue, &shader->ready); 520 521 v = shader->variants; 522 while (v) { 523 t = v; 524 v = v->next; 525 if (t->bo) 526 etna_bo_del(t->bo); 527 528 etna_destroy_shader(t); 529 } 530 531 tgsi_free_tokens(shader->tokens); 532 ralloc_free(shader->nir); 533 util_queue_fence_destroy(&shader->ready); 534 FREE(shader); 535} 536 537static void 538etna_bind_fs_state(struct pipe_context *pctx, void *hwcso) 539{ 540 struct etna_context *ctx = etna_context(pctx); 541 542 ctx->shader.bind_fs = hwcso; 543 ctx->dirty |= ETNA_DIRTY_SHADER; 544} 545 546static void 547etna_bind_vs_state(struct pipe_context *pctx, void *hwcso) 548{ 549 struct etna_context *ctx = etna_context(pctx); 550 551 ctx->shader.bind_vs = hwcso; 552 ctx->dirty |= ETNA_DIRTY_SHADER; 553} 554 555static void 556etna_set_max_shader_compiler_threads(struct pipe_screen *pscreen, 557 unsigned max_threads) 558{ 559 struct etna_screen *screen = etna_screen(pscreen); 560 561 util_queue_adjust_num_threads(&screen->shader_compiler_queue, max_threads); 562} 563 564static bool 565etna_is_parallel_shader_compilation_finished(struct pipe_screen *pscreen, 566 void *hwcso, 567 enum pipe_shader_type shader_type) 568{ 569 struct etna_shader *shader = (struct etna_shader *)hwcso; 570 571 return util_queue_fence_is_signalled(&shader->ready); 572} 573 574void 575etna_shader_init(struct pipe_context *pctx) 576{ 577 pctx->create_fs_state = etna_create_shader_state; 578 pctx->bind_fs_state = etna_bind_fs_state; 579 pctx->delete_fs_state = etna_delete_shader_state; 580 pctx->create_vs_state = etna_create_shader_state; 581 pctx->bind_vs_state = etna_bind_vs_state; 582 pctx->delete_vs_state = etna_delete_shader_state; 583} 584 585bool 586etna_shader_screen_init(struct pipe_screen *pscreen) 587{ 588 struct etna_screen *screen = etna_screen(pscreen); 589 unsigned num_threads = util_get_cpu_caps()->nr_cpus - 1; 590 591 /* Create at least one thread - even on single core CPU systems. */ 592 num_threads = MAX2(1, num_threads); 593 594 screen->compiler = etna_compiler_create(pscreen->get_name(pscreen), &screen->specs); 595 if (!screen->compiler) 596 return false; 597 598 pscreen->set_max_shader_compiler_threads = etna_set_max_shader_compiler_threads; 599 pscreen->is_parallel_shader_compilation_finished = etna_is_parallel_shader_compilation_finished; 600 601 return util_queue_init(&screen->shader_compiler_queue, "sh", 64, num_threads, 602 UTIL_QUEUE_INIT_RESIZE_IF_FULL | UTIL_QUEUE_INIT_SET_FULL_THREAD_AFFINITY, 603 NULL); 604} 605 606void 607etna_shader_screen_fini(struct pipe_screen *pscreen) 608{ 609 struct etna_screen *screen = etna_screen(pscreen); 610 611 util_queue_destroy(&screen->shader_compiler_queue); 612 etna_compiler_destroy(screen->compiler); 613} 614