src/freedreno/vulkan/tu_descriptor_set.c

   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
  22  * DEALINGS IN THE SOFTWARE.
  23  */
  24
  25 /**
  26  * @file
  27  *
  28  * The texture and sampler descriptors are laid out in a single global space
  29  * across all shader stages, for both simplicity of implementation and because
  30  * that seems to be how things have to be structured for border color
  31  * handling.
  32  *
  33  * Each shader stage will declare its texture/sampler count based on the last
  34  * descriptor set it uses.  At draw emit time (though it really should be
  35  * CmdBind time), we upload the descriptor sets used by each shader stage to
  36  * their stage.
  37  */
  38
  39 #include "tu_private.h"
  40
  41 #include <assert.h>
  42 #include <fcntl.h>
  43 #include <stdbool.h>
  44 #include <string.h>
  45 #include <unistd.h>
  46
  47 #include "util/mesa-sha1.h"
  48 #include "vk_util.h"
  49
  50 static int
  51 binding_compare(const void *av, const void *bv)
  52 {
  53    const VkDescriptorSetLayoutBinding *a =
  54       (const VkDescriptorSetLayoutBinding *) av;
  55    const VkDescriptorSetLayoutBinding *b =
  56       (const VkDescriptorSetLayoutBinding *) bv;
  57
  58    return (a->binding < b->binding) ? -1 : (a->binding > b->binding) ? 1 : 0;
  59 }
  60
  61 static VkDescriptorSetLayoutBinding *
  62 create_sorted_bindings(const VkDescriptorSetLayoutBinding *bindings,
  63                        unsigned count)
  64 {
  65    VkDescriptorSetLayoutBinding *sorted_bindings =
  66       malloc(count * sizeof(VkDescriptorSetLayoutBinding));
  67    if (!sorted_bindings)
  68       return NULL;
  69
  70    memcpy(sorted_bindings, bindings,
  71           count * sizeof(VkDescriptorSetLayoutBinding));
  72
  73    qsort(sorted_bindings, count, sizeof(VkDescriptorSetLayoutBinding),
  74          binding_compare);
  75
  76    return sorted_bindings;
  77 }
  78
  79 static uint32_t
  80 descriptor_size(enum VkDescriptorType type)
  81 {
  82    switch (type) {
  83    case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
  84    case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
  85       return 0;
  86    case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
  87    case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
  88    case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
  89    case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
  90       /* 64bit pointer */
  91       return 8;
  92    case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
  93    case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
  94    case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
  95       return A6XX_TEX_CONST_DWORDS*4;
  96    case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
  97       /* texture const + tu_sampler struct (includes border color) */
  98       return A6XX_TEX_CONST_DWORDS*4 + sizeof(struct tu_sampler);
  99    case VK_DESCRIPTOR_TYPE_SAMPLER:
 100       return sizeof(struct tu_sampler);
 101    default:
 102       unreachable("unknown descriptor type\n");
 103       return 0;
 104    }
 105 }
 106
 107 VkResult
 108 tu_CreateDescriptorSetLayout(
 109    VkDevice _device,
 110    const VkDescriptorSetLayoutCreateInfo *pCreateInfo,
 111    const VkAllocationCallbacks *pAllocator,
 112    VkDescriptorSetLayout *pSetLayout)
 113 {
 114    TU_FROM_HANDLE(tu_device, device, _device);
 115    struct tu_descriptor_set_layout *set_layout;
 116
 117    assert(pCreateInfo->sType ==
 118           VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO);
 119    const VkDescriptorSetLayoutBindingFlagsCreateInfoEXT *variable_flags =
 120       vk_find_struct_const(
 121          pCreateInfo->pNext,
 122          DESCRIPTOR_SET_LAYOUT_BINDING_FLAGS_CREATE_INFO_EXT);
 123
 124    uint32_t max_binding = 0;
 125    uint32_t immutable_sampler_count = 0;
 126    for (uint32_t j = 0; j < pCreateInfo->bindingCount; j++) {
 127       max_binding = MAX2(max_binding, pCreateInfo->pBindings[j].binding);
 128       if ((pCreateInfo->pBindings[j].descriptorType == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER ||
 129            pCreateInfo->pBindings[j].descriptorType == VK_DESCRIPTOR_TYPE_SAMPLER) &&
 130            pCreateInfo->pBindings[j].pImmutableSamplers) {
 131          immutable_sampler_count += pCreateInfo->pBindings[j].descriptorCount;
 132       }
 133    }
 134
 135    uint32_t samplers_offset = sizeof(struct tu_descriptor_set_layout) +
 136       (max_binding + 1) * sizeof(set_layout->binding[0]);
 137    uint32_t size = samplers_offset + immutable_sampler_count * sizeof(struct tu_sampler);
 138
 139    set_layout = vk_alloc2(&device->alloc, pAllocator, size, 8,
 140                           VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 141    if (!set_layout)
 142       return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 143
 144    set_layout->flags = pCreateInfo->flags;
 145
 146    /* We just allocate all the samplers at the end of the struct */
 147    struct tu_sampler *samplers = (void*) &set_layout->binding[max_binding + 1];
 148
 149    VkDescriptorSetLayoutBinding *bindings = create_sorted_bindings(
 150       pCreateInfo->pBindings, pCreateInfo->bindingCount);
 151    if (!bindings) {
 152       vk_free2(&device->alloc, pAllocator, set_layout);
 153       return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 154    }
 155
 156    set_layout->binding_count = max_binding + 1;
 157    set_layout->shader_stages = 0;
 158    set_layout->dynamic_shader_stages = 0;
 159    set_layout->has_immutable_samplers = false;
 160    set_layout->size = 0;
 161
 162    memset(set_layout->binding, 0,
 163           size - sizeof(struct tu_descriptor_set_layout));
 164
 165    uint32_t buffer_count = 0;
 166    uint32_t dynamic_offset_count = 0;
 167
 168    for (uint32_t j = 0; j < pCreateInfo->bindingCount; j++) {
 169       const VkDescriptorSetLayoutBinding *binding = bindings + j;
 170       uint32_t b = binding->binding;
 171       uint32_t alignment = 4;
 172       unsigned binding_buffer_count = 1;
 173
 174       switch (binding->descriptorType) {
 175       case VK_DESCRIPTOR_TYPE_SAMPLER:
 176          binding_buffer_count = 0;
 177          break;
 178       case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
 179       case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
 180          assert(!(pCreateInfo->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
 181          set_layout->binding[b].dynamic_offset_count = 1;
 182          break;
 183       default:
 184          break;
 185       }
 186
 187       set_layout->size = align(set_layout->size, alignment);
 188       set_layout->binding[b].type = binding->descriptorType;
 189       set_layout->binding[b].array_size = binding->descriptorCount;
 190       set_layout->binding[b].offset = set_layout->size;
 191       set_layout->binding[b].buffer_offset = buffer_count;
 192       set_layout->binding[b].dynamic_offset_offset = dynamic_offset_count;
 193       set_layout->binding[b].size = descriptor_size(binding->descriptorType);
 194
 195       if (variable_flags && binding->binding < variable_flags->bindingCount &&
 196           (variable_flags->pBindingFlags[binding->binding] &
 197            VK_DESCRIPTOR_BINDING_VARIABLE_DESCRIPTOR_COUNT_BIT_EXT)) {
 198          assert(!binding->pImmutableSamplers); /* Terribly ill defined  how
 199                                                   many samplers are valid */
 200          assert(binding->binding == max_binding);
 201
 202          set_layout->has_variable_descriptors = true;
 203       }
 204
 205       if ((binding->descriptorType == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER ||
 206            binding->descriptorType == VK_DESCRIPTOR_TYPE_SAMPLER) &&
 207           binding->pImmutableSamplers) {
 208          set_layout->binding[b].immutable_samplers_offset = samplers_offset;
 209          set_layout->has_immutable_samplers = true;
 210
 211          for (uint32_t i = 0; i < binding->descriptorCount; i++)
 212             samplers[i] = *tu_sampler_from_handle(binding->pImmutableSamplers[i]);
 213
 214          samplers += binding->descriptorCount;
 215          samplers_offset += sizeof(struct tu_sampler) * binding->descriptorCount;
 216       }
 217
 218       set_layout->size +=
 219          binding->descriptorCount * set_layout->binding[b].size;
 220       buffer_count += binding->descriptorCount * binding_buffer_count;
 221       dynamic_offset_count += binding->descriptorCount *
 222                               set_layout->binding[b].dynamic_offset_count;
 223       set_layout->shader_stages |= binding->stageFlags;
 224    }
 225
 226    free(bindings);
 227
 228    set_layout->buffer_count = buffer_count;
 229    set_layout->dynamic_offset_count = dynamic_offset_count;
 230
 231    *pSetLayout = tu_descriptor_set_layout_to_handle(set_layout);
 232
 233    return VK_SUCCESS;
 234 }
 235
 236 void
 237 tu_DestroyDescriptorSetLayout(VkDevice _device,
 238                               VkDescriptorSetLayout _set_layout,
 239                               const VkAllocationCallbacks *pAllocator)
 240 {
 241    TU_FROM_HANDLE(tu_device, device, _device);
 242    TU_FROM_HANDLE(tu_descriptor_set_layout, set_layout, _set_layout);
 243
 244    if (!set_layout)
 245       return;
 246
 247    vk_free2(&device->alloc, pAllocator, set_layout);
 248 }
 249
 250 void
 251 tu_GetDescriptorSetLayoutSupport(
 252    VkDevice device,
 253    const VkDescriptorSetLayoutCreateInfo *pCreateInfo,
 254    VkDescriptorSetLayoutSupport *pSupport)
 255 {
 256    VkDescriptorSetLayoutBinding *bindings = create_sorted_bindings(
 257       pCreateInfo->pBindings, pCreateInfo->bindingCount);
 258    if (!bindings) {
 259       pSupport->supported = false;
 260       return;
 261    }
 262
 263    const VkDescriptorSetLayoutBindingFlagsCreateInfoEXT *variable_flags =
 264       vk_find_struct_const(
 265          pCreateInfo->pNext,
 266          DESCRIPTOR_SET_LAYOUT_BINDING_FLAGS_CREATE_INFO_EXT);
 267    VkDescriptorSetVariableDescriptorCountLayoutSupportEXT *variable_count =
 268       vk_find_struct(
 269          (void *) pCreateInfo->pNext,
 270          DESCRIPTOR_SET_VARIABLE_DESCRIPTOR_COUNT_LAYOUT_SUPPORT_EXT);
 271    if (variable_count) {
 272       variable_count->maxVariableDescriptorCount = 0;
 273    }
 274
 275    bool supported = true;
 276    uint64_t size = 0;
 277    for (uint32_t i = 0; i < pCreateInfo->bindingCount; i++) {
 278       const VkDescriptorSetLayoutBinding *binding = bindings + i;
 279
 280       uint64_t descriptor_sz = descriptor_size(binding->descriptorType);
 281       uint64_t descriptor_alignment = 8;
 282
 283       if (size && !align_u64(size, descriptor_alignment)) {
 284          supported = false;
 285       }
 286       size = align_u64(size, descriptor_alignment);
 287
 288       uint64_t max_count = UINT64_MAX;
 289       if (descriptor_sz)
 290          max_count = (UINT64_MAX - size) / descriptor_sz;
 291
 292       if (max_count < binding->descriptorCount) {
 293          supported = false;
 294       }
 295       if (variable_flags && binding->binding < variable_flags->bindingCount &&
 296           variable_count &&
 297           (variable_flags->pBindingFlags[binding->binding] &
 298            VK_DESCRIPTOR_BINDING_VARIABLE_DESCRIPTOR_COUNT_BIT_EXT)) {
 299          variable_count->maxVariableDescriptorCount =
 300             MIN2(UINT32_MAX, max_count);
 301       }
 302       size += binding->descriptorCount * descriptor_sz;
 303    }
 304
 305    free(bindings);
 306
 307    pSupport->supported = supported;
 308 }
 309
 310 /*
 311  * Pipeline layouts.  These have nothing to do with the pipeline.  They are
 312  * just multiple descriptor set layouts pasted together.
 313  */
 314
 315 VkResult
 316 tu_CreatePipelineLayout(VkDevice _device,
 317                         const VkPipelineLayoutCreateInfo *pCreateInfo,
 318                         const VkAllocationCallbacks *pAllocator,
 319                         VkPipelineLayout *pPipelineLayout)
 320 {
 321    TU_FROM_HANDLE(tu_device, device, _device);
 322    struct tu_pipeline_layout *layout;
 323    struct mesa_sha1 ctx;
 324
 325    assert(pCreateInfo->sType ==
 326           VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO);
 327
 328    layout = vk_alloc2(&device->alloc, pAllocator, sizeof(*layout), 8,
 329                       VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 330    if (layout == NULL)
 331       return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 332
 333    layout->num_sets = pCreateInfo->setLayoutCount;
 334
 335    unsigned dynamic_offset_count = 0;
 336
 337    _mesa_sha1_init(&ctx);
 338    for (uint32_t set = 0; set < pCreateInfo->setLayoutCount; set++) {
 339       TU_FROM_HANDLE(tu_descriptor_set_layout, set_layout,
 340                      pCreateInfo->pSetLayouts[set]);
 341       layout->set[set].layout = set_layout;
 342
 343       layout->set[set].dynamic_offset_start = dynamic_offset_count;
 344       for (uint32_t b = 0; b < set_layout->binding_count; b++) {
 345          dynamic_offset_count += set_layout->binding[b].array_size *
 346                                  set_layout->binding[b].dynamic_offset_count;
 347          if (set_layout->binding[b].immutable_samplers_offset)
 348             _mesa_sha1_update(
 349                &ctx,
 350                tu_immutable_samplers(set_layout, set_layout->binding + b),
 351                set_layout->binding[b].array_size * 4 * sizeof(uint32_t));
 352       }
 353       _mesa_sha1_update(
 354          &ctx, set_layout->binding,
 355          sizeof(set_layout->binding[0]) * set_layout->binding_count);
 356    }
 357
 358    layout->dynamic_offset_count = dynamic_offset_count;
 359    layout->push_constant_size = 0;
 360
 361    for (unsigned i = 0; i < pCreateInfo->pushConstantRangeCount; ++i) {
 362       const VkPushConstantRange *range = pCreateInfo->pPushConstantRanges + i;
 363       layout->push_constant_size =
 364          MAX2(layout->push_constant_size, range->offset + range->size);
 365    }
 366
 367    layout->push_constant_size = align(layout->push_constant_size, 16);
 368    _mesa_sha1_update(&ctx, &layout->push_constant_size,
 369                      sizeof(layout->push_constant_size));
 370    _mesa_sha1_final(&ctx, layout->sha1);
 371    *pPipelineLayout = tu_pipeline_layout_to_handle(layout);
 372
 373    return VK_SUCCESS;
 374 }
 375
 376 void
 377 tu_DestroyPipelineLayout(VkDevice _device,
 378                          VkPipelineLayout _pipelineLayout,
 379                          const VkAllocationCallbacks *pAllocator)
 380 {
 381    TU_FROM_HANDLE(tu_device, device, _device);
 382    TU_FROM_HANDLE(tu_pipeline_layout, pipeline_layout, _pipelineLayout);
 383
 384    if (!pipeline_layout)
 385       return;
 386    vk_free2(&device->alloc, pAllocator, pipeline_layout);
 387 }
 388
 389 #define EMPTY 1
 390
 391 static VkResult
 392 tu_descriptor_set_create(struct tu_device *device,
 393             struct tu_descriptor_pool *pool,
 394             const struct tu_descriptor_set_layout *layout,
 395             const uint32_t *variable_count,
 396             struct tu_descriptor_set **out_set)
 397 {
 398    struct tu_descriptor_set *set;
 399    uint32_t buffer_count = layout->buffer_count;
 400    if (variable_count) {
 401       unsigned stride = 1;
 402       if (layout->binding[layout->binding_count - 1].type == VK_DESCRIPTOR_TYPE_SAMPLER ||
 403           layout->binding[layout->binding_count - 1].type == VK_DESCRIPTOR_TYPE_INLINE_UNIFORM_BLOCK_EXT)
 404          stride = 0;
 405       buffer_count = layout->binding[layout->binding_count - 1].buffer_offset +
 406                      *variable_count * stride;
 407    }
 408    unsigned range_offset = sizeof(struct tu_descriptor_set) +
 409       sizeof(struct tu_bo *) * buffer_count;
 410    unsigned mem_size = range_offset +
 411       sizeof(struct tu_descriptor_range) * layout->dynamic_offset_count;
 412
 413    if (pool->host_memory_base) {
 414       if (pool->host_memory_end - pool->host_memory_ptr < mem_size)
 415          return vk_error(device->instance, VK_ERROR_OUT_OF_POOL_MEMORY);
 416
 417       set = (struct tu_descriptor_set*)pool->host_memory_ptr;
 418       pool->host_memory_ptr += mem_size;
 419    } else {
 420       set = vk_alloc2(&device->alloc, NULL, mem_size, 8,
 421                       VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 422
 423       if (!set)
 424          return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 425    }
 426
 427    memset(set, 0, mem_size);
 428
 429    if (layout->dynamic_offset_count) {
 430       set->dynamic_descriptors = (struct tu_descriptor_range*)((uint8_t*)set + range_offset);
 431    }
 432
 433    set->layout = layout;
 434    uint32_t layout_size = layout->size;
 435    if (variable_count) {
 436       assert(layout->has_variable_descriptors);
 437       uint32_t stride = layout->binding[layout->binding_count - 1].size;
 438       if (layout->binding[layout->binding_count - 1].type == VK_DESCRIPTOR_TYPE_INLINE_UNIFORM_BLOCK_EXT)
 439          stride = 1;
 440
 441       layout_size = layout->binding[layout->binding_count - 1].offset +
 442                     *variable_count * stride;
 443    }
 444
 445    if (layout_size) {
 446       set->size = layout_size;
 447
 448       if (!pool->host_memory_base && pool->entry_count == pool->max_entry_count) {
 449          vk_free2(&device->alloc, NULL, set);
 450          return vk_error(device->instance, VK_ERROR_OUT_OF_POOL_MEMORY);
 451       }
 452
 453       /* try to allocate linearly first, so that we don't spend
 454        * time looking for gaps if the app only allocates &
 455        * resets via the pool. */
 456       if (pool->current_offset + layout_size <= pool->size) {
 457          set->mapped_ptr = (uint32_t*)(pool->bo.map + pool->current_offset);
 458          set->va = pool->bo.iova + pool->current_offset;
 459          if (!pool->host_memory_base) {
 460             pool->entries[pool->entry_count].offset = pool->current_offset;
 461             pool->entries[pool->entry_count].size = layout_size;
 462             pool->entries[pool->entry_count].set = set;
 463             pool->entry_count++;
 464          }
 465          pool->current_offset += layout_size;
 466       } else if (!pool->host_memory_base) {
 467          uint64_t offset = 0;
 468          int index;
 469
 470          for (index = 0; index < pool->entry_count; ++index) {
 471             if (pool->entries[index].offset - offset >= layout_size)
 472                break;
 473             offset = pool->entries[index].offset + pool->entries[index].size;
 474          }
 475
 476          if (pool->size - offset < layout_size) {
 477             vk_free2(&device->alloc, NULL, set);
 478             return vk_error(device->instance, VK_ERROR_OUT_OF_POOL_MEMORY);
 479          }
 480
 481          set->mapped_ptr = (uint32_t*)(pool->bo.map + offset);
 482          set->va = pool->bo.iova + offset;
 483          memmove(&pool->entries[index + 1], &pool->entries[index],
 484             sizeof(pool->entries[0]) * (pool->entry_count - index));
 485          pool->entries[index].offset = offset;
 486          pool->entries[index].size = layout_size;
 487          pool->entries[index].set = set;
 488          pool->entry_count++;
 489       } else
 490          return vk_error(device->instance, VK_ERROR_OUT_OF_POOL_MEMORY);
 491    }
 492
 493    *out_set = set;
 494    return VK_SUCCESS;
 495 }
 496
 497 static void
 498 tu_descriptor_set_destroy(struct tu_device *device,
 499              struct tu_descriptor_pool *pool,
 500              struct tu_descriptor_set *set,
 501              bool free_bo)
 502 {
 503    assert(!pool->host_memory_base);
 504
 505    if (free_bo && set->size && !pool->host_memory_base) {
 506       uint32_t offset = (uint8_t*)set->mapped_ptr - (uint8_t*)pool->bo.map;
 507       for (int i = 0; i < pool->entry_count; ++i) {
 508          if (pool->entries[i].offset == offset) {
 509             memmove(&pool->entries[i], &pool->entries[i+1],
 510                sizeof(pool->entries[i]) * (pool->entry_count - i - 1));
 511             --pool->entry_count;
 512             break;
 513          }
 514       }
 515    }
 516    vk_free2(&device->alloc, NULL, set);
 517 }
 518
 519 VkResult
 520 tu_CreateDescriptorPool(VkDevice _device,
 521                         const VkDescriptorPoolCreateInfo *pCreateInfo,
 522                         const VkAllocationCallbacks *pAllocator,
 523                         VkDescriptorPool *pDescriptorPool)
 524 {
 525    TU_FROM_HANDLE(tu_device, device, _device);
 526    struct tu_descriptor_pool *pool;
 527    uint64_t size = sizeof(struct tu_descriptor_pool);
 528    uint64_t bo_size = 0, bo_count = 0, range_count = 0;
 529
 530    for (unsigned i = 0; i < pCreateInfo->poolSizeCount; ++i) {
 531       if (pCreateInfo->pPoolSizes[i].type != VK_DESCRIPTOR_TYPE_SAMPLER)
 532          bo_count += pCreateInfo->pPoolSizes[i].descriptorCount;
 533
 534       switch(pCreateInfo->pPoolSizes[i].type) {
 535       case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
 536       case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
 537          range_count += pCreateInfo->pPoolSizes[i].descriptorCount;
 538       default:
 539          break;
 540       }
 541
 542       bo_size += descriptor_size(pCreateInfo->pPoolSizes[i].type) *
 543                            pCreateInfo->pPoolSizes[i].descriptorCount;
 544    }
 545
 546    if (!(pCreateInfo->flags & VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT)) {
 547       uint64_t host_size = pCreateInfo->maxSets * sizeof(struct tu_descriptor_set);
 548       host_size += sizeof(struct tu_bo*) * bo_count;
 549       host_size += sizeof(struct tu_descriptor_range) * range_count;
 550       size += host_size;
 551    } else {
 552       size += sizeof(struct tu_descriptor_pool_entry) * pCreateInfo->maxSets;
 553    }
 554
 555    pool = vk_alloc2(&device->alloc, pAllocator, size, 8,
 556                     VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 557    if (!pool)
 558       return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 559
 560    memset(pool, 0, sizeof(*pool));
 561
 562    if (!(pCreateInfo->flags & VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT)) {
 563       pool->host_memory_base = (uint8_t*)pool + sizeof(struct tu_descriptor_pool);
 564       pool->host_memory_ptr = pool->host_memory_base;
 565       pool->host_memory_end = (uint8_t*)pool + size;
 566    }
 567
 568    if (bo_size) {
 569       VkResult ret;
 570
 571       ret = tu_bo_init_new(device, &pool->bo, bo_size);
 572       assert(ret == VK_SUCCESS);
 573
 574       ret = tu_bo_map(device, &pool->bo);
 575       assert(ret == VK_SUCCESS);
 576    }
 577    pool->size = bo_size;
 578    pool->max_entry_count = pCreateInfo->maxSets;
 579
 580    *pDescriptorPool = tu_descriptor_pool_to_handle(pool);
 581    return VK_SUCCESS;
 582 }
 583
 584 void
 585 tu_DestroyDescriptorPool(VkDevice _device,
 586                          VkDescriptorPool _pool,
 587                          const VkAllocationCallbacks *pAllocator)
 588 {
 589    TU_FROM_HANDLE(tu_device, device, _device);
 590    TU_FROM_HANDLE(tu_descriptor_pool, pool, _pool);
 591
 592    if (!pool)
 593       return;
 594
 595    if (!pool->host_memory_base) {
 596       for(int i = 0; i < pool->entry_count; ++i) {
 597          tu_descriptor_set_destroy(device, pool, pool->entries[i].set, false);
 598       }
 599    }
 600
 601    if (pool->size)
 602       tu_bo_finish(device, &pool->bo);
 603    vk_free2(&device->alloc, pAllocator, pool);
 604 }
 605
 606 VkResult
 607 tu_ResetDescriptorPool(VkDevice _device,
 608                        VkDescriptorPool descriptorPool,
 609                        VkDescriptorPoolResetFlags flags)
 610 {
 611    TU_FROM_HANDLE(tu_device, device, _device);
 612    TU_FROM_HANDLE(tu_descriptor_pool, pool, descriptorPool);
 613
 614    if (!pool->host_memory_base) {
 615       for(int i = 0; i < pool->entry_count; ++i) {
 616          tu_descriptor_set_destroy(device, pool, pool->entries[i].set, false);
 617       }
 618       pool->entry_count = 0;
 619    }
 620
 621    pool->current_offset = 0;
 622    pool->host_memory_ptr = pool->host_memory_base;
 623
 624    return VK_SUCCESS;
 625 }
 626
 627 VkResult
 628 tu_AllocateDescriptorSets(VkDevice _device,
 629                           const VkDescriptorSetAllocateInfo *pAllocateInfo,
 630                           VkDescriptorSet *pDescriptorSets)
 631 {
 632    TU_FROM_HANDLE(tu_device, device, _device);
 633    TU_FROM_HANDLE(tu_descriptor_pool, pool, pAllocateInfo->descriptorPool);
 634
 635    VkResult result = VK_SUCCESS;
 636    uint32_t i;
 637    struct tu_descriptor_set *set = NULL;
 638
 639    const VkDescriptorSetVariableDescriptorCountAllocateInfoEXT *variable_counts =
 640       vk_find_struct_const(pAllocateInfo->pNext, DESCRIPTOR_SET_VARIABLE_DESCRIPTOR_COUNT_ALLOCATE_INFO_EXT);
 641    const uint32_t zero = 0;
 642
 643    /* allocate a set of buffers for each shader to contain descriptors */
 644    for (i = 0; i < pAllocateInfo->descriptorSetCount; i++) {
 645       TU_FROM_HANDLE(tu_descriptor_set_layout, layout,
 646              pAllocateInfo->pSetLayouts[i]);
 647
 648       const uint32_t *variable_count = NULL;
 649       if (variable_counts) {
 650          if (i < variable_counts->descriptorSetCount)
 651             variable_count = variable_counts->pDescriptorCounts + i;
 652          else
 653             variable_count = &zero;
 654       }
 655
 656       assert(!(layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
 657
 658       result = tu_descriptor_set_create(device, pool, layout, variable_count, &set);
 659       if (result != VK_SUCCESS)
 660          break;
 661
 662       pDescriptorSets[i] = tu_descriptor_set_to_handle(set);
 663    }
 664
 665    if (result != VK_SUCCESS) {
 666       tu_FreeDescriptorSets(_device, pAllocateInfo->descriptorPool,
 667                i, pDescriptorSets);
 668       for (i = 0; i < pAllocateInfo->descriptorSetCount; i++) {
 669          pDescriptorSets[i] = VK_NULL_HANDLE;
 670       }
 671    }
 672    return result;
 673 }
 674
 675 VkResult
 676 tu_FreeDescriptorSets(VkDevice _device,
 677                       VkDescriptorPool descriptorPool,
 678                       uint32_t count,
 679                       const VkDescriptorSet *pDescriptorSets)
 680 {
 681    TU_FROM_HANDLE(tu_device, device, _device);
 682    TU_FROM_HANDLE(tu_descriptor_pool, pool, descriptorPool);
 683
 684    for (uint32_t i = 0; i < count; i++) {
 685       TU_FROM_HANDLE(tu_descriptor_set, set, pDescriptorSets[i]);
 686
 687       if (set && !pool->host_memory_base)
 688          tu_descriptor_set_destroy(device, pool, set, true);
 689    }
 690    return VK_SUCCESS;
 691 }
 692
 693 static void write_texel_buffer_descriptor(struct tu_device *device,
 694                                           struct tu_cmd_buffer *cmd_buffer,
 695                                           unsigned *dst,
 696                                           struct tu_bo **buffer_list,
 697                                           const VkBufferView _buffer_view)
 698 {
 699    tu_finishme("texel buffer descriptor");
 700 }
 701
 702 static void write_buffer_descriptor(struct tu_device *device,
 703                                     struct tu_cmd_buffer *cmd_buffer,
 704                                     unsigned *dst,
 705                                     struct tu_bo **buffer_list,
 706                                     const VkDescriptorBufferInfo *buffer_info)
 707 {
 708    TU_FROM_HANDLE(tu_buffer, buffer, buffer_info->buffer);
 709
 710    uint64_t va = tu_buffer_iova(buffer) + buffer_info->offset;
 711    dst[0] = va;
 712    dst[1] = va >> 32;
 713
 714    if (cmd_buffer)
 715       tu_bo_list_add(&cmd_buffer->bo_list, buffer->bo, MSM_SUBMIT_BO_READ);
 716    else
 717       *buffer_list = buffer->bo;
 718 }
 719
 720 static void write_dynamic_buffer_descriptor(struct tu_device *device,
 721                                             struct tu_descriptor_range *range,
 722                                             struct tu_bo **buffer_list,
 723                                             const VkDescriptorBufferInfo *buffer_info)
 724 {
 725    TU_FROM_HANDLE(tu_buffer, buffer, buffer_info->buffer);
 726    uint64_t va = tu_buffer_iova(buffer) + buffer_info->offset;
 727    unsigned size = buffer_info->range;
 728
 729    if (buffer_info->range == VK_WHOLE_SIZE)
 730       size = buffer->size - buffer_info->offset;
 731
 732    range->va = va;
 733    range->size = size;
 734
 735    *buffer_list = buffer->bo;
 736 }
 737
 738 static void
 739 write_image_descriptor(struct tu_device *device,
 740              struct tu_cmd_buffer *cmd_buffer,
 741              unsigned *dst,
 742              struct tu_bo **buffer_list,
 743              VkDescriptorType descriptor_type,
 744              const VkDescriptorImageInfo *image_info)
 745 {
 746    TU_FROM_HANDLE(tu_image_view, iview, image_info->imageView);
 747    uint32_t *descriptor;
 748
 749    if (descriptor_type == VK_DESCRIPTOR_TYPE_STORAGE_IMAGE) {
 750       descriptor = iview->storage_descriptor;
 751    } else {
 752       descriptor = iview->descriptor;
 753    }
 754
 755    memcpy(dst, descriptor, sizeof(iview->descriptor));
 756
 757    if (cmd_buffer)
 758       tu_bo_list_add(&cmd_buffer->bo_list, iview->image->bo, MSM_SUBMIT_BO_READ);
 759    else
 760       *buffer_list = iview->image->bo;
 761 }
 762
 763 static void
 764 write_combined_image_sampler_descriptor(struct tu_device *device,
 765                struct tu_cmd_buffer *cmd_buffer,
 766                unsigned sampler_offset,
 767                unsigned *dst,
 768                struct tu_bo **buffer_list,
 769                VkDescriptorType descriptor_type,
 770                const VkDescriptorImageInfo *image_info,
 771                bool has_sampler)
 772 {
 773    TU_FROM_HANDLE(tu_sampler, sampler, image_info->sampler);
 774
 775    write_image_descriptor(device, cmd_buffer, dst, buffer_list,
 776                           descriptor_type, image_info);
 777    /* copy over sampler state */
 778    if (has_sampler) {
 779       memcpy(dst + sampler_offset / sizeof(*dst), sampler, sizeof(*sampler));
 780    }
 781 }
 782
 783 static void
 784 write_sampler_descriptor(struct tu_device *device,
 785                          unsigned *dst,
 786                          const VkDescriptorImageInfo *image_info)
 787 {
 788    TU_FROM_HANDLE(tu_sampler, sampler, image_info->sampler);
 789
 790    memcpy(dst, sampler, sizeof(*sampler));
 791 }
 792
 793 void
 794 tu_update_descriptor_sets(struct tu_device *device,
 795                           struct tu_cmd_buffer *cmd_buffer,
 796                           VkDescriptorSet dstSetOverride,
 797                           uint32_t descriptorWriteCount,
 798                           const VkWriteDescriptorSet *pDescriptorWrites,
 799                           uint32_t descriptorCopyCount,
 800                           const VkCopyDescriptorSet *pDescriptorCopies)
 801 {
 802    uint32_t i, j;
 803    for (i = 0; i < descriptorWriteCount; i++) {
 804       const VkWriteDescriptorSet *writeset = &pDescriptorWrites[i];
 805       TU_FROM_HANDLE(tu_descriptor_set, set,
 806                        dstSetOverride ? dstSetOverride : writeset->dstSet);
 807       const struct tu_descriptor_set_binding_layout *binding_layout =
 808          set->layout->binding + writeset->dstBinding;
 809       uint32_t *ptr = set->mapped_ptr;
 810       struct tu_bo **buffer_list = set->descriptors;
 811
 812       ptr += binding_layout->offset / 4;
 813
 814       ptr += binding_layout->size * writeset->dstArrayElement / 4;
 815       buffer_list += binding_layout->buffer_offset;
 816       buffer_list += writeset->dstArrayElement;
 817       for (j = 0; j < writeset->descriptorCount; ++j) {
 818          switch(writeset->descriptorType) {
 819          case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
 820          case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: {
 821             unsigned idx = writeset->dstArrayElement + j;
 822             idx += binding_layout->dynamic_offset_offset;
 823             assert(!(set->layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
 824             write_dynamic_buffer_descriptor(device, set->dynamic_descriptors + idx,
 825                         buffer_list, writeset->pBufferInfo + j);
 826             break;
 827          }
 828
 829          case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
 830          case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
 831             write_buffer_descriptor(device, cmd_buffer, ptr, buffer_list,
 832                      writeset->pBufferInfo + j);
 833             break;
 834          case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
 835          case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
 836             write_texel_buffer_descriptor(device, cmd_buffer, ptr, buffer_list,
 837                      writeset->pTexelBufferView[j]);
 838             break;
 839          case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
 840          case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
 841          case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
 842             write_image_descriptor(device, cmd_buffer, ptr, buffer_list,
 843                                    writeset->descriptorType,
 844                                    writeset->pImageInfo + j);
 845             break;
 846          case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
 847             write_combined_image_sampler_descriptor(device, cmd_buffer,
 848                                                     A6XX_TEX_CONST_DWORDS*4,
 849                                                     ptr, buffer_list,
 850                                                     writeset->descriptorType,
 851                                                     writeset->pImageInfo + j,
 852                                                     !binding_layout->immutable_samplers_offset);
 853             break;
 854          case VK_DESCRIPTOR_TYPE_SAMPLER:
 855             write_sampler_descriptor(device, ptr, writeset->pImageInfo + j);
 856             break;
 857          default:
 858             unreachable("unimplemented descriptor type");
 859             break;
 860          }
 861          ptr += binding_layout->size / 4;
 862          ++buffer_list;
 863       }
 864    }
 865
 866    for (i = 0; i < descriptorCopyCount; i++) {
 867       const VkCopyDescriptorSet *copyset = &pDescriptorCopies[i];
 868       TU_FROM_HANDLE(tu_descriptor_set, src_set,
 869                        copyset->srcSet);
 870       TU_FROM_HANDLE(tu_descriptor_set, dst_set,
 871                        copyset->dstSet);
 872       const struct tu_descriptor_set_binding_layout *src_binding_layout =
 873          src_set->layout->binding + copyset->srcBinding;
 874       const struct tu_descriptor_set_binding_layout *dst_binding_layout =
 875          dst_set->layout->binding + copyset->dstBinding;
 876       uint32_t *src_ptr = src_set->mapped_ptr;
 877       uint32_t *dst_ptr = dst_set->mapped_ptr;
 878       struct tu_bo **src_buffer_list = src_set->descriptors;
 879       struct tu_bo **dst_buffer_list = dst_set->descriptors;
 880
 881       src_ptr += src_binding_layout->offset / 4;
 882       dst_ptr += dst_binding_layout->offset / 4;
 883
 884       src_ptr += src_binding_layout->size * copyset->srcArrayElement / 4;
 885       dst_ptr += dst_binding_layout->size * copyset->dstArrayElement / 4;
 886
 887       src_buffer_list += src_binding_layout->buffer_offset;
 888       src_buffer_list += copyset->srcArrayElement;
 889
 890       dst_buffer_list += dst_binding_layout->buffer_offset;
 891       dst_buffer_list += copyset->dstArrayElement;
 892
 893       for (j = 0; j < copyset->descriptorCount; ++j) {
 894          switch (src_binding_layout->type) {
 895          case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
 896          case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: {
 897             unsigned src_idx = copyset->srcArrayElement + j;
 898             unsigned dst_idx = copyset->dstArrayElement + j;
 899             struct tu_descriptor_range *src_range, *dst_range;
 900             src_idx += src_binding_layout->dynamic_offset_offset;
 901             dst_idx += dst_binding_layout->dynamic_offset_offset;
 902
 903             src_range = src_set->dynamic_descriptors + src_idx;
 904             dst_range = dst_set->dynamic_descriptors + dst_idx;
 905             *dst_range = *src_range;
 906             break;
 907          }
 908          default:
 909             memcpy(dst_ptr, src_ptr, src_binding_layout->size);
 910          }
 911          src_ptr += src_binding_layout->size / 4;
 912          dst_ptr += dst_binding_layout->size / 4;
 913
 914          if (src_binding_layout->type != VK_DESCRIPTOR_TYPE_SAMPLER) {
 915             /* Sampler descriptors don't have a buffer list. */
 916             dst_buffer_list[j] = src_buffer_list[j];
 917          }
 918       }
 919    }
 920 }
 921
 922 void
 923 tu_UpdateDescriptorSets(VkDevice _device,
 924                         uint32_t descriptorWriteCount,
 925                         const VkWriteDescriptorSet *pDescriptorWrites,
 926                         uint32_t descriptorCopyCount,
 927                         const VkCopyDescriptorSet *pDescriptorCopies)
 928 {
 929    TU_FROM_HANDLE(tu_device, device, _device);
 930
 931    tu_update_descriptor_sets(device, NULL, VK_NULL_HANDLE,
 932                              descriptorWriteCount, pDescriptorWrites,
 933                              descriptorCopyCount, pDescriptorCopies);
 934 }
 935
 936 VkResult
 937 tu_CreateDescriptorUpdateTemplate(
 938    VkDevice _device,
 939    const VkDescriptorUpdateTemplateCreateInfo *pCreateInfo,
 940    const VkAllocationCallbacks *pAllocator,
 941    VkDescriptorUpdateTemplate *pDescriptorUpdateTemplate)
 942 {
 943    TU_FROM_HANDLE(tu_device, device, _device);
 944    TU_FROM_HANDLE(tu_descriptor_set_layout, set_layout,
 945                   pCreateInfo->descriptorSetLayout);
 946    const uint32_t entry_count = pCreateInfo->descriptorUpdateEntryCount;
 947    const size_t size =
 948       sizeof(struct tu_descriptor_update_template) +
 949       sizeof(struct tu_descriptor_update_template_entry) * entry_count;
 950    struct tu_descriptor_update_template *templ;
 951
 952    templ = vk_alloc2(&device->alloc, pAllocator, size, 8,
 953                      VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 954    if (!templ)
 955       return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 956
 957    *pDescriptorUpdateTemplate =
 958       tu_descriptor_update_template_to_handle(templ);
 959
 960    tu_use_args(set_layout);
 961    tu_stub();
 962    return VK_SUCCESS;
 963 }
 964
 965 void
 966 tu_DestroyDescriptorUpdateTemplate(
 967    VkDevice _device,
 968    VkDescriptorUpdateTemplate descriptorUpdateTemplate,
 969    const VkAllocationCallbacks *pAllocator)
 970 {
 971    TU_FROM_HANDLE(tu_device, device, _device);
 972    TU_FROM_HANDLE(tu_descriptor_update_template, templ,
 973                   descriptorUpdateTemplate);
 974
 975    if (!templ)
 976       return;
 977
 978    vk_free2(&device->alloc, pAllocator, templ);
 979 }
 980
 981 void
 982 tu_update_descriptor_set_with_template(
 983    struct tu_device *device,
 984    struct tu_cmd_buffer *cmd_buffer,
 985    struct tu_descriptor_set *set,
 986    VkDescriptorUpdateTemplate descriptorUpdateTemplate,
 987    const void *pData)
 988 {
 989    TU_FROM_HANDLE(tu_descriptor_update_template, templ,
 990                   descriptorUpdateTemplate);
 991    tu_use_args(templ);
 992 }
 993
 994 void
 995 tu_UpdateDescriptorSetWithTemplate(
 996    VkDevice _device,
 997    VkDescriptorSet descriptorSet,
 998    VkDescriptorUpdateTemplate descriptorUpdateTemplate,
 999    const void *pData)
1000 {
1001    TU_FROM_HANDLE(tu_device, device, _device);
1002    TU_FROM_HANDLE(tu_descriptor_set, set, descriptorSet);
1003
1004    tu_update_descriptor_set_with_template(device, NULL, set,
1005                                           descriptorUpdateTemplate, pData);
1006 }
1007
1008 VkResult
1009 tu_CreateSamplerYcbcrConversion(
1010    VkDevice device,
1011    const VkSamplerYcbcrConversionCreateInfo *pCreateInfo,
1012    const VkAllocationCallbacks *pAllocator,
1013    VkSamplerYcbcrConversion *pYcbcrConversion)
1014 {
1015    *pYcbcrConversion = VK_NULL_HANDLE;
1016    return VK_SUCCESS;
1017 }
1018
1019 void
1020 tu_DestroySamplerYcbcrConversion(VkDevice device,
1021                                  VkSamplerYcbcrConversion ycbcrConversion,
1022                                  const VkAllocationCallbacks *pAllocator)
1023 {
1024    /* Do nothing. */
1025 }