1 /* 2 * Copyright 2016 Red Hat. 3 * Copyright 2016 Bas Nieuwenhuizen 4 * 5 * Permission is hereby granted, free of charge, to any person obtaining a 6 * copy of this software and associated documentation files (the "Software"), 7 * to deal in the Software without restriction, including without limitation 8 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 9 * and/or sell copies of the Software, and to permit persons to whom the 10 * Software is furnished to do so, subject to the following conditions: 11 * 12 * The above copyright notice and this permission notice (including the next 13 * paragraph) shall be included in all copies or substantial portions of the 14 * Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 22 * IN THE SOFTWARE. 23 */ 24 #include <assert.h> 25 #include <stdbool.h> 26 #include <string.h> 27 #include <unistd.h> 28 #include <fcntl.h> 29 30 #include "util/mesa-sha1.h" 31 #include "radv_private.h" 32 #include "sid.h" 33 34 VkResult radv_CreateDescriptorSetLayout( 35 VkDevice _device, 36 const VkDescriptorSetLayoutCreateInfo* pCreateInfo, 37 const VkAllocationCallbacks* pAllocator, 38 VkDescriptorSetLayout* pSetLayout) 39 { 40 RADV_FROM_HANDLE(radv_device, device, _device); 41 struct radv_descriptor_set_layout *set_layout; 42 43 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO); 44 45 uint32_t max_binding = 0; 46 uint32_t immutable_sampler_count = 0; 47 for (uint32_t j = 0; j < pCreateInfo->bindingCount; j++) { 48 max_binding = MAX2(max_binding, pCreateInfo->pBindings[j].binding); 49 if (pCreateInfo->pBindings[j].pImmutableSamplers) 50 immutable_sampler_count += pCreateInfo->pBindings[j].descriptorCount; 51 } 52 53 uint32_t samplers_offset = sizeof(struct radv_descriptor_set_layout) + 54 (max_binding + 1) * sizeof(set_layout->binding[0]); 55 size_t size = samplers_offset + immutable_sampler_count * 4 * sizeof(uint32_t); 56 57 set_layout = vk_alloc2(&device->alloc, pAllocator, size, 8, 58 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT); 59 if (!set_layout) 60 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY); 61 62 set_layout->flags = pCreateInfo->flags; 63 64 /* We just allocate all the samplers at the end of the struct */ 65 uint32_t *samplers = (uint32_t*)&set_layout->binding[max_binding + 1]; 66 67 set_layout->binding_count = max_binding + 1; 68 set_layout->shader_stages = 0; 69 set_layout->dynamic_shader_stages = 0; 70 set_layout->has_immutable_samplers = false; 71 set_layout->size = 0; 72 73 memset(set_layout->binding, 0, size - sizeof(struct radv_descriptor_set_layout)); 74 75 uint32_t buffer_count = 0; 76 uint32_t dynamic_offset_count = 0; 77 78 for (uint32_t j = 0; j < pCreateInfo->bindingCount; j++) { 79 const VkDescriptorSetLayoutBinding *binding = &pCreateInfo->pBindings[j]; 80 uint32_t b = binding->binding; 81 uint32_t alignment; 82 unsigned binding_buffer_count = 0; 83 84 switch (binding->descriptorType) { 85 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC: 86 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: 87 assert(!(pCreateInfo->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR)); 88 set_layout->binding[b].dynamic_offset_count = 1; 89 set_layout->dynamic_shader_stages |= binding->stageFlags; 90 set_layout->binding[b].size = 0; 91 binding_buffer_count = 1; 92 alignment = 1; 93 break; 94 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER: 95 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER: 96 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER: 97 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER: 98 set_layout->binding[b].size = 16; 99 binding_buffer_count = 1; 100 alignment = 16; 101 break; 102 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE: 103 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE: 104 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT: 105 /* main descriptor + fmask descriptor */ 106 set_layout->binding[b].size = 64; 107 binding_buffer_count = 1; 108 alignment = 32; 109 break; 110 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER: 111 /* main descriptor + fmask descriptor + sampler */ 112 set_layout->binding[b].size = 96; 113 binding_buffer_count = 1; 114 alignment = 32; 115 break; 116 case VK_DESCRIPTOR_TYPE_SAMPLER: 117 set_layout->binding[b].size = 16; 118 alignment = 16; 119 break; 120 default: 121 unreachable("unknown descriptor type\n"); 122 break; 123 } 124 125 set_layout->size = align(set_layout->size, alignment); 126 assert(binding->descriptorCount > 0); 127 set_layout->binding[b].type = binding->descriptorType; 128 set_layout->binding[b].array_size = binding->descriptorCount; 129 set_layout->binding[b].offset = set_layout->size; 130 set_layout->binding[b].buffer_offset = buffer_count; 131 set_layout->binding[b].dynamic_offset_offset = dynamic_offset_count; 132 133 if (binding->pImmutableSamplers) { 134 set_layout->binding[b].immutable_samplers_offset = samplers_offset; 135 set_layout->binding[b].immutable_samplers_equal = true; 136 set_layout->has_immutable_samplers = true; 137 138 139 for (uint32_t i = 0; i < binding->descriptorCount; i++) 140 memcpy(samplers + 4 * i, &radv_sampler_from_handle(binding->pImmutableSamplers[i])->state, 16); 141 for (uint32_t i = 1; i < binding->descriptorCount; i++) 142 if (memcmp(samplers + 4 * i, samplers, 16) != 0) 143 set_layout->binding[b].immutable_samplers_equal = false; 144 145 /* Don't reserve space for the samplers if they're not accessed. */ 146 if (set_layout->binding[b].immutable_samplers_equal) { 147 if (binding->descriptorType == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER) 148 set_layout->binding[b].size -= 32; 149 else if (binding->descriptorType == VK_DESCRIPTOR_TYPE_SAMPLER) 150 set_layout->binding[b].size -= 16; 151 } 152 samplers += 4 * binding->descriptorCount; 153 samplers_offset += 4 * sizeof(uint32_t) * binding->descriptorCount; 154 } 155 156 set_layout->size += binding->descriptorCount * set_layout->binding[b].size; 157 buffer_count += binding->descriptorCount * binding_buffer_count; 158 dynamic_offset_count += binding->descriptorCount * 159 set_layout->binding[b].dynamic_offset_count; 160 set_layout->shader_stages |= binding->stageFlags; 161 } 162 163 set_layout->buffer_count = buffer_count; 164 set_layout->dynamic_offset_count = dynamic_offset_count; 165 166 *pSetLayout = radv_descriptor_set_layout_to_handle(set_layout); 167 168 return VK_SUCCESS; 169 } 170 171 void radv_DestroyDescriptorSetLayout( 172 VkDevice _device, 173 VkDescriptorSetLayout _set_layout, 174 const VkAllocationCallbacks* pAllocator) 175 { 176 RADV_FROM_HANDLE(radv_device, device, _device); 177 RADV_FROM_HANDLE(radv_descriptor_set_layout, set_layout, _set_layout); 178 179 if (!set_layout) 180 return; 181 182 vk_free2(&device->alloc, pAllocator, set_layout); 183 } 184 185 /* 186 * Pipeline layouts. These have nothing to do with the pipeline. They are 187 * just muttiple descriptor set layouts pasted together 188 */ 189 190 VkResult radv_CreatePipelineLayout( 191 VkDevice _device, 192 const VkPipelineLayoutCreateInfo* pCreateInfo, 193 const VkAllocationCallbacks* pAllocator, 194 VkPipelineLayout* pPipelineLayout) 195 { 196 RADV_FROM_HANDLE(radv_device, device, _device); 197 struct radv_pipeline_layout *layout; 198 struct mesa_sha1 ctx; 199 200 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO); 201 202 layout = vk_alloc2(&device->alloc, pAllocator, sizeof(*layout), 8, 203 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT); 204 if (layout == NULL) 205 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY); 206 207 layout->num_sets = pCreateInfo->setLayoutCount; 208 209 unsigned dynamic_offset_count = 0; 210 211 212 _mesa_sha1_init(&ctx); 213 for (uint32_t set = 0; set < pCreateInfo->setLayoutCount; set++) { 214 RADV_FROM_HANDLE(radv_descriptor_set_layout, set_layout, 215 pCreateInfo->pSetLayouts[set]); 216 layout->set[set].layout = set_layout; 217 218 layout->set[set].dynamic_offset_start = dynamic_offset_count; 219 for (uint32_t b = 0; b < set_layout->binding_count; b++) { 220 dynamic_offset_count += set_layout->binding[b].array_size * set_layout->binding[b].dynamic_offset_count; 221 if (set_layout->binding[b].immutable_samplers_offset) 222 _mesa_sha1_update(&ctx, radv_immutable_samplers(set_layout, set_layout->binding + b), 223 set_layout->binding[b].array_size * 4 * sizeof(uint32_t)); 224 } 225 _mesa_sha1_update(&ctx, set_layout->binding, 226 sizeof(set_layout->binding[0]) * set_layout->binding_count); 227 } 228 229 layout->dynamic_offset_count = dynamic_offset_count; 230 layout->push_constant_size = 0; 231 232 for (unsigned i = 0; i < pCreateInfo->pushConstantRangeCount; ++i) { 233 const VkPushConstantRange *range = pCreateInfo->pPushConstantRanges + i; 234 layout->push_constant_size = MAX2(layout->push_constant_size, 235 range->offset + range->size); 236 } 237 238 layout->push_constant_size = align(layout->push_constant_size, 16); 239 _mesa_sha1_update(&ctx, &layout->push_constant_size, 240 sizeof(layout->push_constant_size)); 241 _mesa_sha1_final(&ctx, layout->sha1); 242 *pPipelineLayout = radv_pipeline_layout_to_handle(layout); 243 244 return VK_SUCCESS; 245 } 246 247 void radv_DestroyPipelineLayout( 248 VkDevice _device, 249 VkPipelineLayout _pipelineLayout, 250 const VkAllocationCallbacks* pAllocator) 251 { 252 RADV_FROM_HANDLE(radv_device, device, _device); 253 RADV_FROM_HANDLE(radv_pipeline_layout, pipeline_layout, _pipelineLayout); 254 255 if (!pipeline_layout) 256 return; 257 vk_free2(&device->alloc, pAllocator, pipeline_layout); 258 } 259 260 #define EMPTY 1 261 262 static VkResult 263 radv_descriptor_set_create(struct radv_device *device, 264 struct radv_descriptor_pool *pool, 265 const struct radv_descriptor_set_layout *layout, 266 struct radv_descriptor_set **out_set) 267 { 268 struct radv_descriptor_set *set; 269 unsigned range_offset = sizeof(struct radv_descriptor_set) + 270 sizeof(struct radeon_winsys_bo *) * layout->buffer_count; 271 unsigned mem_size = range_offset + 272 sizeof(struct radv_descriptor_range) * layout->dynamic_offset_count; 273 274 if (pool->host_memory_base) { 275 if (pool->host_memory_end - pool->host_memory_ptr < mem_size) 276 return vk_error(VK_ERROR_OUT_OF_POOL_MEMORY_KHR); 277 278 set = (struct radv_descriptor_set*)pool->host_memory_ptr; 279 pool->host_memory_ptr += mem_size; 280 } else { 281 set = vk_alloc2(&device->alloc, NULL, mem_size, 8, 282 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT); 283 284 if (!set) 285 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY); 286 } 287 288 memset(set, 0, mem_size); 289 290 if (layout->dynamic_offset_count) { 291 set->dynamic_descriptors = (struct radv_descriptor_range*)((uint8_t*)set + range_offset); 292 } 293 294 set->layout = layout; 295 if (layout->size) { 296 uint32_t layout_size = align_u32(layout->size, 32); 297 set->size = layout->size; 298 299 if (!pool->host_memory_base && pool->entry_count == pool->max_entry_count) { 300 vk_free2(&device->alloc, NULL, set); 301 return vk_error(VK_ERROR_OUT_OF_POOL_MEMORY_KHR); 302 } 303 304 /* try to allocate linearly first, so that we don't spend 305 * time looking for gaps if the app only allocates & 306 * resets via the pool. */ 307 if (pool->current_offset + layout_size <= pool->size) { 308 set->bo = pool->bo; 309 set->mapped_ptr = (uint32_t*)(pool->mapped_ptr + pool->current_offset); 310 set->va = radv_buffer_get_va(set->bo) + pool->current_offset; 311 if (!pool->host_memory_base) { 312 pool->entries[pool->entry_count].offset = pool->current_offset; 313 pool->entries[pool->entry_count].size = layout_size; 314 pool->entries[pool->entry_count].set = set; 315 pool->entry_count++; 316 } 317 pool->current_offset += layout_size; 318 } else if (!pool->host_memory_base) { 319 uint64_t offset = 0; 320 int index; 321 322 for (index = 0; index < pool->entry_count; ++index) { 323 if (pool->entries[index].offset - offset >= layout_size) 324 break; 325 offset = pool->entries[index].offset + pool->entries[index].size; 326 } 327 328 if (pool->size - offset < layout_size) { 329 vk_free2(&device->alloc, NULL, set); 330 return vk_error(VK_ERROR_OUT_OF_POOL_MEMORY_KHR); 331 } 332 set->bo = pool->bo; 333 set->mapped_ptr = (uint32_t*)(pool->mapped_ptr + offset); 334 set->va = radv_buffer_get_va(set->bo) + offset; 335 memmove(&pool->entries[index + 1], &pool->entries[index], 336 sizeof(pool->entries[0]) * (pool->entry_count - index)); 337 pool->entries[index].offset = offset; 338 pool->entries[index].size = layout_size; 339 pool->entries[index].set = set; 340 pool->entry_count++; 341 } else 342 return vk_error(VK_ERROR_OUT_OF_POOL_MEMORY_KHR); 343 } 344 345 if (layout->has_immutable_samplers) { 346 for (unsigned i = 0; i < layout->binding_count; ++i) { 347 if (!layout->binding[i].immutable_samplers_offset || 348 layout->binding[i].immutable_samplers_equal) 349 continue; 350 351 unsigned offset = layout->binding[i].offset / 4; 352 if (layout->binding[i].type == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER) 353 offset += 16; 354 355 const uint32_t *samplers = (const uint32_t*)((const char*)layout + layout->binding[i].immutable_samplers_offset); 356 for (unsigned j = 0; j < layout->binding[i].array_size; ++j) { 357 memcpy(set->mapped_ptr + offset, samplers + 4 * j, 16); 358 offset += layout->binding[i].size / 4; 359 } 360 361 } 362 } 363 *out_set = set; 364 return VK_SUCCESS; 365 } 366 367 static void 368 radv_descriptor_set_destroy(struct radv_device *device, 369 struct radv_descriptor_pool *pool, 370 struct radv_descriptor_set *set, 371 bool free_bo) 372 { 373 assert(!pool->host_memory_base); 374 375 if (free_bo && set->size && !pool->host_memory_base) { 376 uint32_t offset = (uint8_t*)set->mapped_ptr - pool->mapped_ptr; 377 for (int i = 0; i < pool->entry_count; ++i) { 378 if (pool->entries[i].offset == offset) { 379 memmove(&pool->entries[i], &pool->entries[i+1], 380 sizeof(pool->entries[i]) * (pool->entry_count - i - 1)); 381 --pool->entry_count; 382 break; 383 } 384 } 385 } 386 vk_free2(&device->alloc, NULL, set); 387 } 388 389 VkResult radv_CreateDescriptorPool( 390 VkDevice _device, 391 const VkDescriptorPoolCreateInfo* pCreateInfo, 392 const VkAllocationCallbacks* pAllocator, 393 VkDescriptorPool* pDescriptorPool) 394 { 395 RADV_FROM_HANDLE(radv_device, device, _device); 396 struct radv_descriptor_pool *pool; 397 int size = sizeof(struct radv_descriptor_pool); 398 uint64_t bo_size = 0, bo_count = 0, range_count = 0; 399 400 401 for (unsigned i = 0; i < pCreateInfo->poolSizeCount; ++i) { 402 if (pCreateInfo->pPoolSizes[i].type != VK_DESCRIPTOR_TYPE_SAMPLER) 403 bo_count += pCreateInfo->pPoolSizes[i].descriptorCount; 404 405 switch(pCreateInfo->pPoolSizes[i].type) { 406 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC: 407 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: 408 range_count += pCreateInfo->pPoolSizes[i].descriptorCount; 409 break; 410 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER: 411 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER: 412 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER: 413 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER: 414 case VK_DESCRIPTOR_TYPE_SAMPLER: 415 /* 32 as we may need to align for images */ 416 bo_size += 32 * pCreateInfo->pPoolSizes[i].descriptorCount; 417 break; 418 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE: 419 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE: 420 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT: 421 bo_size += 64 * pCreateInfo->pPoolSizes[i].descriptorCount; 422 break; 423 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER: 424 bo_size += 96 * pCreateInfo->pPoolSizes[i].descriptorCount; 425 break; 426 default: 427 unreachable("unknown descriptor type\n"); 428 break; 429 } 430 } 431 432 if (!(pCreateInfo->flags & VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT)) { 433 uint64_t host_size = pCreateInfo->maxSets * sizeof(struct radv_descriptor_set); 434 host_size += sizeof(struct radeon_winsys_bo*) * bo_count; 435 host_size += sizeof(struct radv_descriptor_range) * range_count; 436 size += host_size; 437 } else { 438 size += sizeof(struct radv_descriptor_pool_entry) * pCreateInfo->maxSets; 439 } 440 441 pool = vk_alloc2(&device->alloc, pAllocator, size, 8, 442 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT); 443 if (!pool) 444 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY); 445 446 memset(pool, 0, sizeof(*pool)); 447 448 if (!(pCreateInfo->flags & VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT)) { 449 pool->host_memory_base = (uint8_t*)pool + sizeof(struct radv_descriptor_pool); 450 pool->host_memory_ptr = pool->host_memory_base; 451 pool->host_memory_end = (uint8_t*)pool + size; 452 } 453 454 if (bo_size) { 455 pool->bo = device->ws->buffer_create(device->ws, bo_size, 32, 456 RADEON_DOMAIN_VRAM, 457 RADEON_FLAG_NO_INTERPROCESS_SHARING | 458 RADEON_FLAG_READ_ONLY); 459 pool->mapped_ptr = (uint8_t*)device->ws->buffer_map(pool->bo); 460 } 461 pool->size = bo_size; 462 pool->max_entry_count = pCreateInfo->maxSets; 463 464 *pDescriptorPool = radv_descriptor_pool_to_handle(pool); 465 return VK_SUCCESS; 466 } 467 468 void radv_DestroyDescriptorPool( 469 VkDevice _device, 470 VkDescriptorPool _pool, 471 const VkAllocationCallbacks* pAllocator) 472 { 473 RADV_FROM_HANDLE(radv_device, device, _device); 474 RADV_FROM_HANDLE(radv_descriptor_pool, pool, _pool); 475 476 if (!pool) 477 return; 478 479 if (!pool->host_memory_base) { 480 for(int i = 0; i < pool->entry_count; ++i) { 481 radv_descriptor_set_destroy(device, pool, pool->entries[i].set, false); 482 } 483 } 484 485 if (pool->bo) 486 device->ws->buffer_destroy(pool->bo); 487 vk_free2(&device->alloc, pAllocator, pool); 488 } 489 490 VkResult radv_ResetDescriptorPool( 491 VkDevice _device, 492 VkDescriptorPool descriptorPool, 493 VkDescriptorPoolResetFlags flags) 494 { 495 RADV_FROM_HANDLE(radv_device, device, _device); 496 RADV_FROM_HANDLE(radv_descriptor_pool, pool, descriptorPool); 497 498 if (!pool->host_memory_base) { 499 for(int i = 0; i < pool->entry_count; ++i) { 500 radv_descriptor_set_destroy(device, pool, pool->entries[i].set, false); 501 } 502 pool->entry_count = 0; 503 } 504 505 pool->current_offset = 0; 506 pool->host_memory_ptr = pool->host_memory_base; 507 508 return VK_SUCCESS; 509 } 510 511 VkResult radv_AllocateDescriptorSets( 512 VkDevice _device, 513 const VkDescriptorSetAllocateInfo* pAllocateInfo, 514 VkDescriptorSet* pDescriptorSets) 515 { 516 RADV_FROM_HANDLE(radv_device, device, _device); 517 RADV_FROM_HANDLE(radv_descriptor_pool, pool, pAllocateInfo->descriptorPool); 518 519 VkResult result = VK_SUCCESS; 520 uint32_t i; 521 struct radv_descriptor_set *set; 522 523 /* allocate a set of buffers for each shader to contain descriptors */ 524 for (i = 0; i < pAllocateInfo->descriptorSetCount; i++) { 525 RADV_FROM_HANDLE(radv_descriptor_set_layout, layout, 526 pAllocateInfo->pSetLayouts[i]); 527 528 assert(!(layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR)); 529 530 result = radv_descriptor_set_create(device, pool, layout, &set); 531 if (result != VK_SUCCESS) 532 break; 533 534 pDescriptorSets[i] = radv_descriptor_set_to_handle(set); 535 } 536 537 if (result != VK_SUCCESS) 538 radv_FreeDescriptorSets(_device, pAllocateInfo->descriptorPool, 539 i, pDescriptorSets); 540 return result; 541 } 542 543 VkResult radv_FreeDescriptorSets( 544 VkDevice _device, 545 VkDescriptorPool descriptorPool, 546 uint32_t count, 547 const VkDescriptorSet* pDescriptorSets) 548 { 549 RADV_FROM_HANDLE(radv_device, device, _device); 550 RADV_FROM_HANDLE(radv_descriptor_pool, pool, descriptorPool); 551 552 for (uint32_t i = 0; i < count; i++) { 553 RADV_FROM_HANDLE(radv_descriptor_set, set, pDescriptorSets[i]); 554 555 if (set && !pool->host_memory_base) 556 radv_descriptor_set_destroy(device, pool, set, true); 557 } 558 return VK_SUCCESS; 559 } 560 561 static void write_texel_buffer_descriptor(struct radv_device *device, 562 struct radv_cmd_buffer *cmd_buffer, 563 unsigned *dst, 564 struct radeon_winsys_bo **buffer_list, 565 const VkBufferView _buffer_view) 566 { 567 RADV_FROM_HANDLE(radv_buffer_view, buffer_view, _buffer_view); 568 569 memcpy(dst, buffer_view->state, 4 * 4); 570 571 if (cmd_buffer) 572 radv_cs_add_buffer(device->ws, cmd_buffer->cs, buffer_view->bo, 7); 573 else 574 *buffer_list = buffer_view->bo; 575 } 576 577 static void write_buffer_descriptor(struct radv_device *device, 578 struct radv_cmd_buffer *cmd_buffer, 579 unsigned *dst, 580 struct radeon_winsys_bo **buffer_list, 581 const VkDescriptorBufferInfo *buffer_info) 582 { 583 RADV_FROM_HANDLE(radv_buffer, buffer, buffer_info->buffer); 584 uint64_t va = radv_buffer_get_va(buffer->bo); 585 uint32_t range = buffer_info->range; 586 587 if (buffer_info->range == VK_WHOLE_SIZE) 588 range = buffer->size - buffer_info->offset; 589 590 va += buffer_info->offset + buffer->offset; 591 dst[0] = va; 592 dst[1] = S_008F04_BASE_ADDRESS_HI(va >> 32); 593 dst[2] = range; 594 dst[3] = S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X) | 595 S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y) | 596 S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z) | 597 S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W) | 598 S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT) | 599 S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32); 600 601 if (cmd_buffer) 602 radv_cs_add_buffer(device->ws, cmd_buffer->cs, buffer->bo, 7); 603 else 604 *buffer_list = buffer->bo; 605 } 606 607 static void write_dynamic_buffer_descriptor(struct radv_device *device, 608 struct radv_descriptor_range *range, 609 struct radeon_winsys_bo **buffer_list, 610 const VkDescriptorBufferInfo *buffer_info) 611 { 612 RADV_FROM_HANDLE(radv_buffer, buffer, buffer_info->buffer); 613 uint64_t va = radv_buffer_get_va(buffer->bo); 614 unsigned size = buffer_info->range; 615 616 if (buffer_info->range == VK_WHOLE_SIZE) 617 size = buffer->size - buffer_info->offset; 618 619 va += buffer_info->offset + buffer->offset; 620 range->va = va; 621 range->size = size; 622 623 *buffer_list = buffer->bo; 624 } 625 626 static void 627 write_image_descriptor(struct radv_device *device, 628 struct radv_cmd_buffer *cmd_buffer, 629 unsigned *dst, 630 struct radeon_winsys_bo **buffer_list, 631 VkDescriptorType descriptor_type, 632 const VkDescriptorImageInfo *image_info) 633 { 634 RADV_FROM_HANDLE(radv_image_view, iview, image_info->imageView); 635 uint32_t *descriptor; 636 637 if (descriptor_type == VK_DESCRIPTOR_TYPE_STORAGE_IMAGE) { 638 descriptor = iview->storage_descriptor; 639 } else { 640 descriptor = iview->descriptor; 641 } 642 643 memcpy(dst, descriptor, 16 * 4); 644 645 if (cmd_buffer) 646 radv_cs_add_buffer(device->ws, cmd_buffer->cs, iview->bo, 7); 647 else 648 *buffer_list = iview->bo; 649 } 650 651 static void 652 write_combined_image_sampler_descriptor(struct radv_device *device, 653 struct radv_cmd_buffer *cmd_buffer, 654 unsigned *dst, 655 struct radeon_winsys_bo **buffer_list, 656 VkDescriptorType descriptor_type, 657 const VkDescriptorImageInfo *image_info, 658 bool has_sampler) 659 { 660 RADV_FROM_HANDLE(radv_sampler, sampler, image_info->sampler); 661 662 write_image_descriptor(device, cmd_buffer, dst, buffer_list, descriptor_type, image_info); 663 /* copy over sampler state */ 664 if (has_sampler) 665 memcpy(dst + 16, sampler->state, 16); 666 } 667 668 static void 669 write_sampler_descriptor(struct radv_device *device, 670 unsigned *dst, 671 const VkDescriptorImageInfo *image_info) 672 { 673 RADV_FROM_HANDLE(radv_sampler, sampler, image_info->sampler); 674 675 memcpy(dst, sampler->state, 16); 676 } 677 678 void radv_update_descriptor_sets( 679 struct radv_device* device, 680 struct radv_cmd_buffer* cmd_buffer, 681 VkDescriptorSet dstSetOverride, 682 uint32_t descriptorWriteCount, 683 const VkWriteDescriptorSet* pDescriptorWrites, 684 uint32_t descriptorCopyCount, 685 const VkCopyDescriptorSet* pDescriptorCopies) 686 { 687 uint32_t i, j; 688 for (i = 0; i < descriptorWriteCount; i++) { 689 const VkWriteDescriptorSet *writeset = &pDescriptorWrites[i]; 690 RADV_FROM_HANDLE(radv_descriptor_set, set, 691 dstSetOverride ? dstSetOverride : writeset->dstSet); 692 const struct radv_descriptor_set_binding_layout *binding_layout = 693 set->layout->binding + writeset->dstBinding; 694 uint32_t *ptr = set->mapped_ptr; 695 struct radeon_winsys_bo **buffer_list = set->descriptors; 696 /* Immutable samplers are not copied into push descriptors when they are 697 * allocated, so if we are writing push descriptors we have to copy the 698 * immutable samplers into them now. 699 */ 700 const bool copy_immutable_samplers = cmd_buffer && 701 binding_layout->immutable_samplers_offset && !binding_layout->immutable_samplers_equal; 702 const uint32_t *samplers = radv_immutable_samplers(set->layout, binding_layout); 703 704 ptr += binding_layout->offset / 4; 705 ptr += binding_layout->size * writeset->dstArrayElement / 4; 706 buffer_list += binding_layout->buffer_offset; 707 buffer_list += writeset->dstArrayElement; 708 for (j = 0; j < writeset->descriptorCount; ++j) { 709 switch(writeset->descriptorType) { 710 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC: 711 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: { 712 unsigned idx = writeset->dstArrayElement + j; 713 idx += binding_layout->dynamic_offset_offset; 714 assert(!(set->layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR)); 715 write_dynamic_buffer_descriptor(device, set->dynamic_descriptors + idx, 716 buffer_list, writeset->pBufferInfo + j); 717 break; 718 } 719 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER: 720 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER: 721 write_buffer_descriptor(device, cmd_buffer, ptr, buffer_list, 722 writeset->pBufferInfo + j); 723 break; 724 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER: 725 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER: 726 write_texel_buffer_descriptor(device, cmd_buffer, ptr, buffer_list, 727 writeset->pTexelBufferView[j]); 728 break; 729 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE: 730 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE: 731 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT: 732 write_image_descriptor(device, cmd_buffer, ptr, buffer_list, 733 writeset->descriptorType, 734 writeset->pImageInfo + j); 735 break; 736 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER: 737 write_combined_image_sampler_descriptor(device, cmd_buffer, ptr, buffer_list, 738 writeset->descriptorType, 739 writeset->pImageInfo + j, 740 !binding_layout->immutable_samplers_offset); 741 if (copy_immutable_samplers) { 742 const unsigned idx = writeset->dstArrayElement + j; 743 memcpy(ptr + 16, samplers + 4 * idx, 16); 744 } 745 break; 746 case VK_DESCRIPTOR_TYPE_SAMPLER: 747 if (!binding_layout->immutable_samplers_offset) { 748 write_sampler_descriptor(device, ptr, 749 writeset->pImageInfo + j); 750 } else if (copy_immutable_samplers) { 751 unsigned idx = writeset->dstArrayElement + j; 752 memcpy(ptr, samplers + 4 * idx, 16); 753 } 754 break; 755 default: 756 unreachable("unimplemented descriptor type"); 757 break; 758 } 759 ptr += binding_layout->size / 4; 760 ++buffer_list; 761 } 762 763 } 764 765 for (i = 0; i < descriptorCopyCount; i++) { 766 const VkCopyDescriptorSet *copyset = &pDescriptorCopies[i]; 767 RADV_FROM_HANDLE(radv_descriptor_set, src_set, 768 copyset->srcSet); 769 RADV_FROM_HANDLE(radv_descriptor_set, dst_set, 770 copyset->dstSet); 771 const struct radv_descriptor_set_binding_layout *src_binding_layout = 772 src_set->layout->binding + copyset->srcBinding; 773 const struct radv_descriptor_set_binding_layout *dst_binding_layout = 774 dst_set->layout->binding + copyset->dstBinding; 775 uint32_t *src_ptr = src_set->mapped_ptr; 776 uint32_t *dst_ptr = dst_set->mapped_ptr; 777 struct radeon_winsys_bo **src_buffer_list = src_set->descriptors; 778 struct radeon_winsys_bo **dst_buffer_list = dst_set->descriptors; 779 780 src_ptr += src_binding_layout->offset / 4; 781 dst_ptr += dst_binding_layout->offset / 4; 782 783 src_ptr += src_binding_layout->size * copyset->srcArrayElement / 4; 784 dst_ptr += dst_binding_layout->size * copyset->dstArrayElement / 4; 785 786 src_buffer_list += src_binding_layout->buffer_offset; 787 src_buffer_list += copyset->srcArrayElement; 788 789 dst_buffer_list += dst_binding_layout->buffer_offset; 790 dst_buffer_list += copyset->dstArrayElement; 791 792 for (j = 0; j < copyset->descriptorCount; ++j) { 793 switch (src_binding_layout->type) { 794 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC: 795 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: { 796 unsigned src_idx = copyset->srcArrayElement + j; 797 unsigned dst_idx = copyset->dstArrayElement + j; 798 struct radv_descriptor_range *src_range, *dst_range; 799 src_idx += src_binding_layout->dynamic_offset_offset; 800 dst_idx += dst_binding_layout->dynamic_offset_offset; 801 802 src_range = src_set->dynamic_descriptors + src_idx; 803 dst_range = dst_set->dynamic_descriptors + dst_idx; 804 *dst_range = *src_range; 805 break; 806 } 807 default: 808 memcpy(dst_ptr, src_ptr, src_binding_layout->size); 809 } 810 src_ptr += src_binding_layout->size / 4; 811 dst_ptr += dst_binding_layout->size / 4; 812 dst_buffer_list[j] = src_buffer_list[j]; 813 ++src_buffer_list; 814 ++dst_buffer_list; 815 } 816 } 817 } 818 819 void radv_UpdateDescriptorSets( 820 VkDevice _device, 821 uint32_t descriptorWriteCount, 822 const VkWriteDescriptorSet* pDescriptorWrites, 823 uint32_t descriptorCopyCount, 824 const VkCopyDescriptorSet* pDescriptorCopies) 825 { 826 RADV_FROM_HANDLE(radv_device, device, _device); 827 828 radv_update_descriptor_sets(device, NULL, VK_NULL_HANDLE, descriptorWriteCount, pDescriptorWrites, 829 descriptorCopyCount, pDescriptorCopies); 830 } 831 832 VkResult radv_CreateDescriptorUpdateTemplateKHR(VkDevice _device, 833 const VkDescriptorUpdateTemplateCreateInfoKHR *pCreateInfo, 834 const VkAllocationCallbacks *pAllocator, 835 VkDescriptorUpdateTemplateKHR *pDescriptorUpdateTemplate) 836 { 837 RADV_FROM_HANDLE(radv_device, device, _device); 838 RADV_FROM_HANDLE(radv_descriptor_set_layout, set_layout, pCreateInfo->descriptorSetLayout); 839 const uint32_t entry_count = pCreateInfo->descriptorUpdateEntryCount; 840 const size_t size = sizeof(struct radv_descriptor_update_template) + 841 sizeof(struct radv_descriptor_update_template_entry) * entry_count; 842 struct radv_descriptor_update_template *templ; 843 uint32_t i; 844 845 templ = vk_alloc2(&device->alloc, pAllocator, size, 8, VK_SYSTEM_ALLOCATION_SCOPE_OBJECT); 846 if (!templ) 847 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY); 848 849 templ->entry_count = entry_count; 850 851 for (i = 0; i < entry_count; i++) { 852 const VkDescriptorUpdateTemplateEntryKHR *entry = &pCreateInfo->pDescriptorUpdateEntries[i]; 853 const struct radv_descriptor_set_binding_layout *binding_layout = 854 set_layout->binding + entry->dstBinding; 855 const uint32_t buffer_offset = binding_layout->buffer_offset + entry->dstArrayElement; 856 const uint32_t *immutable_samplers = NULL; 857 uint32_t dst_offset; 858 uint32_t dst_stride; 859 860 /* dst_offset is an offset into dynamic_descriptors when the descriptor 861 is dynamic, and an offset into mapped_ptr otherwise */ 862 switch (entry->descriptorType) { 863 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC: 864 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: 865 assert(pCreateInfo->templateType == VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_DESCRIPTOR_SET_KHR); 866 dst_offset = binding_layout->dynamic_offset_offset + entry->dstArrayElement; 867 dst_stride = 0; /* Not used */ 868 break; 869 default: 870 switch (entry->descriptorType) { 871 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER: 872 case VK_DESCRIPTOR_TYPE_SAMPLER: 873 /* Immutable samplers are copied into push descriptors when they are pushed */ 874 if (pCreateInfo->templateType == VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_PUSH_DESCRIPTORS_KHR && 875 binding_layout->immutable_samplers_offset && !binding_layout->immutable_samplers_equal) { 876 immutable_samplers = radv_immutable_samplers(set_layout, binding_layout) + entry->dstArrayElement * 4; 877 } 878 break; 879 default: 880 break; 881 } 882 dst_offset = binding_layout->offset / 4 + binding_layout->size * entry->dstArrayElement / 4; 883 dst_stride = binding_layout->size / 4; 884 break; 885 } 886 887 templ->entry[i] = (struct radv_descriptor_update_template_entry) { 888 .descriptor_type = entry->descriptorType, 889 .descriptor_count = entry->descriptorCount, 890 .src_offset = entry->offset, 891 .src_stride = entry->stride, 892 .dst_offset = dst_offset, 893 .dst_stride = dst_stride, 894 .buffer_offset = buffer_offset, 895 .has_sampler = !binding_layout->immutable_samplers_offset, 896 .immutable_samplers = immutable_samplers 897 }; 898 } 899 900 *pDescriptorUpdateTemplate = radv_descriptor_update_template_to_handle(templ); 901 return VK_SUCCESS; 902 } 903 904 void radv_DestroyDescriptorUpdateTemplateKHR(VkDevice _device, 905 VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate, 906 const VkAllocationCallbacks *pAllocator) 907 { 908 RADV_FROM_HANDLE(radv_device, device, _device); 909 RADV_FROM_HANDLE(radv_descriptor_update_template, templ, descriptorUpdateTemplate); 910 911 if (!templ) 912 return; 913 914 vk_free2(&device->alloc, pAllocator, templ); 915 } 916 917 void radv_update_descriptor_set_with_template(struct radv_device *device, 918 struct radv_cmd_buffer *cmd_buffer, 919 struct radv_descriptor_set *set, 920 VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate, 921 const void *pData) 922 { 923 RADV_FROM_HANDLE(radv_descriptor_update_template, templ, descriptorUpdateTemplate); 924 uint32_t i; 925 926 for (i = 0; i < templ->entry_count; ++i) { 927 struct radeon_winsys_bo **buffer_list = set->descriptors + templ->entry[i].buffer_offset; 928 uint32_t *pDst = set->mapped_ptr + templ->entry[i].dst_offset; 929 const uint8_t *pSrc = ((const uint8_t *) pData) + templ->entry[i].src_offset; 930 uint32_t j; 931 932 for (j = 0; j < templ->entry[i].descriptor_count; ++j) { 933 switch (templ->entry[i].descriptor_type) { 934 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC: 935 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: { 936 const unsigned idx = templ->entry[i].dst_offset + j; 937 assert(!(set->layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR)); 938 write_dynamic_buffer_descriptor(device, set->dynamic_descriptors + idx, 939 buffer_list, (struct VkDescriptorBufferInfo *) pSrc); 940 break; 941 } 942 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER: 943 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER: 944 write_buffer_descriptor(device, cmd_buffer, pDst, buffer_list, 945 (struct VkDescriptorBufferInfo *) pSrc); 946 break; 947 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER: 948 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER: 949 write_texel_buffer_descriptor(device, cmd_buffer, pDst, buffer_list, 950 *(VkBufferView *) pSrc); 951 break; 952 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE: 953 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE: 954 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT: 955 write_image_descriptor(device, cmd_buffer, pDst, buffer_list, 956 templ->entry[i].descriptor_type, 957 (struct VkDescriptorImageInfo *) pSrc); 958 break; 959 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER: 960 write_combined_image_sampler_descriptor(device, cmd_buffer, pDst, buffer_list, 961 templ->entry[i].descriptor_type, 962 (struct VkDescriptorImageInfo *) pSrc, 963 templ->entry[i].has_sampler); 964 if (templ->entry[i].immutable_samplers) 965 memcpy(pDst + 16, templ->entry[i].immutable_samplers + 4 * j, 16); 966 break; 967 case VK_DESCRIPTOR_TYPE_SAMPLER: 968 if (templ->entry[i].has_sampler) 969 write_sampler_descriptor(device, pDst, 970 (struct VkDescriptorImageInfo *) pSrc); 971 else if (templ->entry[i].immutable_samplers) 972 memcpy(pDst, templ->entry[i].immutable_samplers + 4 * j, 16); 973 break; 974 default: 975 unreachable("unimplemented descriptor type"); 976 break; 977 } 978 pSrc += templ->entry[i].src_stride; 979 pDst += templ->entry[i].dst_stride; 980 ++buffer_list; 981 } 982 } 983 } 984 985 void radv_UpdateDescriptorSetWithTemplateKHR(VkDevice _device, 986 VkDescriptorSet descriptorSet, 987 VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate, 988 const void *pData) 989 { 990 RADV_FROM_HANDLE(radv_device, device, _device); 991 RADV_FROM_HANDLE(radv_descriptor_set, set, descriptorSet); 992 993 radv_update_descriptor_set_with_template(device, NULL, set, descriptorUpdateTemplate, pData); 994 } 995
