projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
radv: load the fast color clear values from the base level
[mesa.git] / src / amd / vulkan / radv_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 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 #include "vk_format.h"
34 #include "vk_util.h"
35
36
37 static bool has_equal_immutable_samplers(const VkSampler *samplers, uint32_t count)
38 {
39 if (!samplers)
40 return false;
41 for(uint32_t i = 1; i < count; ++i) {
42 if (memcmp(radv_sampler_from_handle(samplers[0])->state,
43 radv_sampler_from_handle(samplers[i])->state, 16)) {
44 return false;
45 }
46 }
47 return true;
48 }
49
50 static int binding_compare(const void* av, const void *bv)
51 {
52 const VkDescriptorSetLayoutBinding *a = (const VkDescriptorSetLayoutBinding*)av;
53 const VkDescriptorSetLayoutBinding *b = (const VkDescriptorSetLayoutBinding*)bv;
54
55 return (a->binding < b->binding) ? -1 : (a->binding > b->binding) ? 1 : 0;
56 }
57
58 static VkDescriptorSetLayoutBinding *
59 create_sorted_bindings(const VkDescriptorSetLayoutBinding *bindings, unsigned count) {
60 VkDescriptorSetLayoutBinding *sorted_bindings = malloc(count * sizeof(VkDescriptorSetLayoutBinding));
61 if (!sorted_bindings)
62 return NULL;
63
64 memcpy(sorted_bindings, bindings, count * sizeof(VkDescriptorSetLayoutBinding));
65
66 qsort(sorted_bindings, count, sizeof(VkDescriptorSetLayoutBinding), binding_compare);
67
68 return sorted_bindings;
69 }
70
71 VkResult radv_CreateDescriptorSetLayout(
72 VkDevice _device,
73 const VkDescriptorSetLayoutCreateInfo* pCreateInfo,
74 const VkAllocationCallbacks* pAllocator,
75 VkDescriptorSetLayout* pSetLayout)
76 {
77 RADV_FROM_HANDLE(radv_device, device, _device);
78 struct radv_descriptor_set_layout *set_layout;
79
80 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO);
81 const VkDescriptorSetLayoutBindingFlagsCreateInfoEXT *variable_flags =
82 vk_find_struct_const(pCreateInfo->pNext, DESCRIPTOR_SET_LAYOUT_BINDING_FLAGS_CREATE_INFO_EXT);
83
84 uint32_t max_binding = 0;
85 uint32_t immutable_sampler_count = 0;
86 uint32_t ycbcr_sampler_count = 0;
87 for (uint32_t j = 0; j < pCreateInfo->bindingCount; j++) {
88 max_binding = MAX2(max_binding, pCreateInfo->pBindings[j].binding);
89 if ((pCreateInfo->pBindings[j].descriptorType == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER ||
90 pCreateInfo->pBindings[j].descriptorType == VK_DESCRIPTOR_TYPE_SAMPLER) &&
91 pCreateInfo->pBindings[j].pImmutableSamplers) {
92 immutable_sampler_count += pCreateInfo->pBindings[j].descriptorCount;
93
94 bool has_ycbcr_sampler = false;
95 for (unsigned i = 0; i < pCreateInfo->pBindings[j].descriptorCount; ++i) {
96 if (radv_sampler_from_handle(pCreateInfo->pBindings[j].pImmutableSamplers[i])->ycbcr_sampler)
97 has_ycbcr_sampler = true;
98 }
99
100 if (has_ycbcr_sampler)
101 ycbcr_sampler_count += pCreateInfo->pBindings[j].descriptorCount;
102 }
103 }
104
105 uint32_t samplers_offset = sizeof(struct radv_descriptor_set_layout) +
106 (max_binding + 1) * sizeof(set_layout->binding[0]);
107 size_t size = samplers_offset + immutable_sampler_count * 4 * sizeof(uint32_t);
108 if (ycbcr_sampler_count > 0) {
109 size += ycbcr_sampler_count * sizeof(struct radv_sampler_ycbcr_conversion) + (max_binding + 1) * sizeof(uint32_t);
110 }
111
112 set_layout = vk_zalloc2(&device->alloc, pAllocator, size, 8,
113 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
114 if (!set_layout)
115 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
116
117 set_layout->flags = pCreateInfo->flags;
118 set_layout->layout_size = size;
119
120 /* We just allocate all the samplers at the end of the struct */
121 uint32_t *samplers = (uint32_t*)&set_layout->binding[max_binding + 1];
122 struct radv_sampler_ycbcr_conversion *ycbcr_samplers = NULL;
123 uint32_t *ycbcr_sampler_offsets = NULL;
124
125 if (ycbcr_sampler_count > 0) {
126 ycbcr_sampler_offsets = samplers + 4 * immutable_sampler_count;
127 set_layout->ycbcr_sampler_offsets_offset = (char*)ycbcr_sampler_offsets - (char*)set_layout;
128 ycbcr_samplers = (struct radv_sampler_ycbcr_conversion *)(ycbcr_sampler_offsets + max_binding + 1);
129 } else
130 set_layout->ycbcr_sampler_offsets_offset = 0;
131
132 VkDescriptorSetLayoutBinding *bindings = create_sorted_bindings(pCreateInfo->pBindings,
133 pCreateInfo->bindingCount);
134 if (!bindings) {
135 vk_free2(&device->alloc, pAllocator, set_layout);
136 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
137 }
138
139 set_layout->binding_count = max_binding + 1;
140 set_layout->shader_stages = 0;
141 set_layout->dynamic_shader_stages = 0;
142 set_layout->has_immutable_samplers = false;
143 set_layout->size = 0;
144
145 memset(set_layout->binding, 0, size - sizeof(struct radv_descriptor_set_layout));
146
147 uint32_t buffer_count = 0;
148 uint32_t dynamic_offset_count = 0;
149
150 for (uint32_t j = 0; j < pCreateInfo->bindingCount; j++) {
151 const VkDescriptorSetLayoutBinding *binding = bindings + j;
152 uint32_t b = binding->binding;
153 uint32_t alignment;
154 unsigned binding_buffer_count = 0;
155 uint32_t descriptor_count = binding->descriptorCount;
156 bool has_ycbcr_sampler = false;
157
158 /* main image + fmask */
159 uint32_t max_sampled_image_descriptors = 2;
160
161 if (binding->descriptorType == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER &&
162 binding->pImmutableSamplers) {
163 for (unsigned i = 0; i < binding->descriptorCount; ++i) {
164 struct radv_sampler_ycbcr_conversion *conversion =
165 radv_sampler_from_handle(binding->pImmutableSamplers[i])->ycbcr_sampler;
166
167 if (conversion) {
168 has_ycbcr_sampler = true;
169 max_sampled_image_descriptors = MAX2(max_sampled_image_descriptors,
170 vk_format_get_plane_count(conversion->format));
171 }
172 }
173 }
174
175 switch (binding->descriptorType) {
176 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
177 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
178 assert(!(pCreateInfo->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
179 set_layout->binding[b].dynamic_offset_count = 1;
180 set_layout->dynamic_shader_stages |= binding->stageFlags;
181 set_layout->binding[b].size = 0;
182 binding_buffer_count = 1;
183 alignment = 1;
184 break;
185 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
186 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
187 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
188 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
189 set_layout->binding[b].size = 16;
190 binding_buffer_count = 1;
191 alignment = 16;
192 break;
193 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
194 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
195 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
196 /* main descriptor + fmask descriptor */
197 set_layout->binding[b].size = 64;
198 binding_buffer_count = 1;
199 alignment = 32;
200 break;
201 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
202 /* main descriptor + fmask descriptor + sampler */
203 set_layout->binding[b].size = 96;
204 binding_buffer_count = 1;
205 alignment = 32;
206 break;
207 case VK_DESCRIPTOR_TYPE_SAMPLER:
208 set_layout->binding[b].size = 16;
209 alignment = 16;
210 break;
211 case VK_DESCRIPTOR_TYPE_INLINE_UNIFORM_BLOCK_EXT:
212 alignment = 16;
213 set_layout->binding[b].size = descriptor_count;
214 descriptor_count = 1;
215 break;
216 default:
217 unreachable("unknown descriptor type\n");
218 break;
219 }
220
221 set_layout->size = align(set_layout->size, alignment);
222 set_layout->binding[b].type = binding->descriptorType;
223 set_layout->binding[b].array_size = descriptor_count;
224 set_layout->binding[b].offset = set_layout->size;
225 set_layout->binding[b].buffer_offset = buffer_count;
226 set_layout->binding[b].dynamic_offset_offset = dynamic_offset_count;
227
228 if (variable_flags && binding->binding < variable_flags->bindingCount &&
229 (variable_flags->pBindingFlags[binding->binding] & VK_DESCRIPTOR_BINDING_VARIABLE_DESCRIPTOR_COUNT_BIT_EXT)) {
230 assert(!binding->pImmutableSamplers); /* Terribly ill defined how many samplers are valid */
231 assert(binding->binding == max_binding);
232
233 set_layout->has_variable_descriptors = true;
234 }
235
236 if ((binding->descriptorType == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER ||
237 binding->descriptorType == VK_DESCRIPTOR_TYPE_SAMPLER) &&
238 binding->pImmutableSamplers) {
239 set_layout->binding[b].immutable_samplers_offset = samplers_offset;
240 set_layout->binding[b].immutable_samplers_equal =
241 has_equal_immutable_samplers(binding->pImmutableSamplers, binding->descriptorCount);
242 set_layout->has_immutable_samplers = true;
243
244
245 for (uint32_t i = 0; i < binding->descriptorCount; i++)
246 memcpy(samplers + 4 * i, &radv_sampler_from_handle(binding->pImmutableSamplers[i])->state, 16);
247
248 /* Don't reserve space for the samplers if they're not accessed. */
249 if (set_layout->binding[b].immutable_samplers_equal) {
250 if (binding->descriptorType == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER &&
251 max_sampled_image_descriptors <= 2)
252 set_layout->binding[b].size -= 32;
253 else if (binding->descriptorType == VK_DESCRIPTOR_TYPE_SAMPLER)
254 set_layout->binding[b].size -= 16;
255 }
256 samplers += 4 * binding->descriptorCount;
257 samplers_offset += 4 * sizeof(uint32_t) * binding->descriptorCount;
258
259 if (has_ycbcr_sampler) {
260 ycbcr_sampler_offsets[b] = (const char*)ycbcr_samplers - (const char*)set_layout;
261 for (uint32_t i = 0; i < binding->descriptorCount; i++) {
262 if (radv_sampler_from_handle(binding->pImmutableSamplers[i])->ycbcr_sampler)
263 ycbcr_samplers[i] = *radv_sampler_from_handle(binding->pImmutableSamplers[i])->ycbcr_sampler;
264 else
265 ycbcr_samplers[i].format = VK_FORMAT_UNDEFINED;
266 }
267 ycbcr_samplers += binding->descriptorCount;
268 }
269 }
270
271 set_layout->size += descriptor_count * set_layout->binding[b].size;
272 buffer_count += descriptor_count * binding_buffer_count;
273 dynamic_offset_count += descriptor_count *
274 set_layout->binding[b].dynamic_offset_count;
275 set_layout->shader_stages |= binding->stageFlags;
276 }
277
278 free(bindings);
279
280 set_layout->buffer_count = buffer_count;
281 set_layout->dynamic_offset_count = dynamic_offset_count;
282
283 *pSetLayout = radv_descriptor_set_layout_to_handle(set_layout);
284
285 return VK_SUCCESS;
286 }
287
288 void radv_DestroyDescriptorSetLayout(
289 VkDevice _device,
290 VkDescriptorSetLayout _set_layout,
291 const VkAllocationCallbacks* pAllocator)
292 {
293 RADV_FROM_HANDLE(radv_device, device, _device);
294 RADV_FROM_HANDLE(radv_descriptor_set_layout, set_layout, _set_layout);
295
296 if (!set_layout)
297 return;
298
299 vk_free2(&device->alloc, pAllocator, set_layout);
300 }
301
302 void radv_GetDescriptorSetLayoutSupport(VkDevice device,
303 const VkDescriptorSetLayoutCreateInfo* pCreateInfo,
304 VkDescriptorSetLayoutSupport* pSupport)
305 {
306 VkDescriptorSetLayoutBinding *bindings = create_sorted_bindings(pCreateInfo->pBindings,
307 pCreateInfo->bindingCount);
308 if (!bindings) {
309 pSupport->supported = false;
310 return;
311 }
312
313 const VkDescriptorSetLayoutBindingFlagsCreateInfoEXT *variable_flags =
314 vk_find_struct_const(pCreateInfo->pNext, DESCRIPTOR_SET_LAYOUT_BINDING_FLAGS_CREATE_INFO_EXT);
315 VkDescriptorSetVariableDescriptorCountLayoutSupportEXT *variable_count =
316 vk_find_struct((void*)pCreateInfo->pNext, DESCRIPTOR_SET_VARIABLE_DESCRIPTOR_COUNT_LAYOUT_SUPPORT_EXT);
317 if (variable_count) {
318 variable_count->maxVariableDescriptorCount = 0;
319 }
320
321 bool supported = true;
322 uint64_t size = 0;
323 for (uint32_t i = 0; i < pCreateInfo->bindingCount; i++) {
324 const VkDescriptorSetLayoutBinding *binding = bindings + i;
325
326 uint64_t descriptor_size = 0;
327 uint64_t descriptor_alignment = 1;
328 uint32_t descriptor_count = binding->descriptorCount;
329 switch (binding->descriptorType) {
330 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
331 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
332 break;
333 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
334 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
335 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
336 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
337 descriptor_size = 16;
338 descriptor_alignment = 16;
339 break;
340 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
341 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
342 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
343 descriptor_size = 64;
344 descriptor_alignment = 32;
345 break;
346 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
347 if (!has_equal_immutable_samplers(binding->pImmutableSamplers, descriptor_count)) {
348 descriptor_size = 64;
349 } else {
350 descriptor_size = 96;
351 }
352 descriptor_alignment = 32;
353 break;
354 case VK_DESCRIPTOR_TYPE_SAMPLER:
355 if (!has_equal_immutable_samplers(binding->pImmutableSamplers, descriptor_count)) {
356 descriptor_size = 16;
357 descriptor_alignment = 16;
358 }
359 break;
360 case VK_DESCRIPTOR_TYPE_INLINE_UNIFORM_BLOCK_EXT:
361 descriptor_alignment = 16;
362 descriptor_size = descriptor_count;
363 descriptor_count = 1;
364 break;
365 default:
366 unreachable("unknown descriptor type\n");
367 break;
368 }
369
370 if (size && !align_u64(size, descriptor_alignment)) {
371 supported = false;
372 }
373 size = align_u64(size, descriptor_alignment);
374
375 uint64_t max_count = INT32_MAX;
376 if (binding->descriptorType == VK_DESCRIPTOR_TYPE_INLINE_UNIFORM_BLOCK_EXT)
377 max_count = INT32_MAX - size;
378 else if (descriptor_size)
379 max_count = (INT32_MAX - size) / descriptor_size;
380
381 if (max_count < descriptor_count) {
382 supported = false;
383 }
384 if (variable_flags && binding->binding <variable_flags->bindingCount && variable_count &&
385 (variable_flags->pBindingFlags[binding->binding] & VK_DESCRIPTOR_BINDING_VARIABLE_DESCRIPTOR_COUNT_BIT_EXT)) {
386 variable_count->maxVariableDescriptorCount = MIN2(UINT32_MAX, max_count);
387 }
388 size += descriptor_count * descriptor_size;
389 }
390
391 free(bindings);
392
393 pSupport->supported = supported;
394 }
395
396 /*
397 * Pipeline layouts. These have nothing to do with the pipeline. They are
398 * just multiple descriptor set layouts pasted together.
399 */
400
401 VkResult radv_CreatePipelineLayout(
402 VkDevice _device,
403 const VkPipelineLayoutCreateInfo* pCreateInfo,
404 const VkAllocationCallbacks* pAllocator,
405 VkPipelineLayout* pPipelineLayout)
406 {
407 RADV_FROM_HANDLE(radv_device, device, _device);
408 struct radv_pipeline_layout *layout;
409 struct mesa_sha1 ctx;
410
411 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO);
412
413 layout = vk_alloc2(&device->alloc, pAllocator, sizeof(*layout), 8,
414 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
415 if (layout == NULL)
416 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
417
418 layout->num_sets = pCreateInfo->setLayoutCount;
419
420 unsigned dynamic_offset_count = 0;
421 uint16_t dynamic_shader_stages = 0;
422
423
424 _mesa_sha1_init(&ctx);
425 for (uint32_t set = 0; set < pCreateInfo->setLayoutCount; set++) {
426 RADV_FROM_HANDLE(radv_descriptor_set_layout, set_layout,
427 pCreateInfo->pSetLayouts[set]);
428 layout->set[set].layout = set_layout;
429
430 layout->set[set].dynamic_offset_start = dynamic_offset_count;
431 for (uint32_t b = 0; b < set_layout->binding_count; b++) {
432 dynamic_offset_count += set_layout->binding[b].array_size * set_layout->binding[b].dynamic_offset_count;
433 dynamic_shader_stages |= set_layout->dynamic_shader_stages;
434 }
435 _mesa_sha1_update(&ctx, set_layout, set_layout->layout_size);
436 }
437
438 layout->dynamic_offset_count = dynamic_offset_count;
439 layout->dynamic_shader_stages = dynamic_shader_stages;
440 layout->push_constant_size = 0;
441
442 for (unsigned i = 0; i < pCreateInfo->pushConstantRangeCount; ++i) {
443 const VkPushConstantRange *range = pCreateInfo->pPushConstantRanges + i;
444 layout->push_constant_size = MAX2(layout->push_constant_size,
445 range->offset + range->size);
446 }
447
448 layout->push_constant_size = align(layout->push_constant_size, 16);
449 _mesa_sha1_update(&ctx, &layout->push_constant_size,
450 sizeof(layout->push_constant_size));
451 _mesa_sha1_final(&ctx, layout->sha1);
452 *pPipelineLayout = radv_pipeline_layout_to_handle(layout);
453
454 return VK_SUCCESS;
455 }
456
457 void radv_DestroyPipelineLayout(
458 VkDevice _device,
459 VkPipelineLayout _pipelineLayout,
460 const VkAllocationCallbacks* pAllocator)
461 {
462 RADV_FROM_HANDLE(radv_device, device, _device);
463 RADV_FROM_HANDLE(radv_pipeline_layout, pipeline_layout, _pipelineLayout);
464
465 if (!pipeline_layout)
466 return;
467 vk_free2(&device->alloc, pAllocator, pipeline_layout);
468 }
469
470 #define EMPTY 1
471
472 static VkResult
473 radv_descriptor_set_create(struct radv_device *device,
474 struct radv_descriptor_pool *pool,
475 const struct radv_descriptor_set_layout *layout,
476 const uint32_t *variable_count,
477 struct radv_descriptor_set **out_set)
478 {
479 struct radv_descriptor_set *set;
480 unsigned range_offset = sizeof(struct radv_descriptor_set) +
481 sizeof(struct radeon_winsys_bo *) * layout->buffer_count;
482 unsigned mem_size = range_offset +
483 sizeof(struct radv_descriptor_range) * layout->dynamic_offset_count;
484
485 if (pool->host_memory_base) {
486 if (pool->host_memory_end - pool->host_memory_ptr < mem_size)
487 return vk_error(device->instance, VK_ERROR_OUT_OF_POOL_MEMORY);
488
489 set = (struct radv_descriptor_set*)pool->host_memory_ptr;
490 pool->host_memory_ptr += mem_size;
491 } else {
492 set = vk_alloc2(&device->alloc, NULL, mem_size, 8,
493 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
494
495 if (!set)
496 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
497 }
498
499 memset(set, 0, mem_size);
500
501 if (layout->dynamic_offset_count) {
502 set->dynamic_descriptors = (struct radv_descriptor_range*)((uint8_t*)set + range_offset);
503 }
504
505 set->layout = layout;
506 uint32_t layout_size = align_u32(layout->size, 32);
507 if (layout_size) {
508 set->size = layout_size;
509
510 if (!pool->host_memory_base && pool->entry_count == pool->max_entry_count) {
511 vk_free2(&device->alloc, NULL, set);
512 return vk_error(device->instance, VK_ERROR_OUT_OF_POOL_MEMORY);
513 }
514
515 /* try to allocate linearly first, so that we don't spend
516 * time looking for gaps if the app only allocates &
517 * resets via the pool. */
518 if (pool->current_offset + layout_size <= pool->size) {
519 set->bo = pool->bo;
520 set->mapped_ptr = (uint32_t*)(pool->mapped_ptr + pool->current_offset);
521 set->va = radv_buffer_get_va(set->bo) + pool->current_offset;
522 if (!pool->host_memory_base) {
523 pool->entries[pool->entry_count].offset = pool->current_offset;
524 pool->entries[pool->entry_count].size = layout_size;
525 pool->entries[pool->entry_count].set = set;
526 pool->entry_count++;
527 }
528 pool->current_offset += layout_size;
529 } else if (!pool->host_memory_base) {
530 uint64_t offset = 0;
531 int index;
532
533 for (index = 0; index < pool->entry_count; ++index) {
534 if (pool->entries[index].offset - offset >= layout_size)
535 break;
536 offset = pool->entries[index].offset + pool->entries[index].size;
537 }
538
539 if (pool->size - offset < layout_size) {
540 vk_free2(&device->alloc, NULL, set);
541 return vk_error(device->instance, VK_ERROR_OUT_OF_POOL_MEMORY);
542 }
543 set->bo = pool->bo;
544 set->mapped_ptr = (uint32_t*)(pool->mapped_ptr + offset);
545 set->va = radv_buffer_get_va(set->bo) + offset;
546 memmove(&pool->entries[index + 1], &pool->entries[index],
547 sizeof(pool->entries[0]) * (pool->entry_count - index));
548 pool->entries[index].offset = offset;
549 pool->entries[index].size = layout_size;
550 pool->entries[index].set = set;
551 pool->entry_count++;
552 } else
553 return vk_error(device->instance, VK_ERROR_OUT_OF_POOL_MEMORY);
554 }
555
556 if (layout->has_immutable_samplers) {
557 for (unsigned i = 0; i < layout->binding_count; ++i) {
558 if (!layout->binding[i].immutable_samplers_offset ||
559 layout->binding[i].immutable_samplers_equal)
560 continue;
561
562 unsigned offset = layout->binding[i].offset / 4;
563 if (layout->binding[i].type == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER)
564 offset += radv_combined_image_descriptor_sampler_offset(layout->binding + i) / 4;
565
566 const uint32_t *samplers = (const uint32_t*)((const char*)layout + layout->binding[i].immutable_samplers_offset);
567 for (unsigned j = 0; j < layout->binding[i].array_size; ++j) {
568 memcpy(set->mapped_ptr + offset, samplers + 4 * j, 16);
569 offset += layout->binding[i].size / 4;
570 }
571
572 }
573 }
574 *out_set = set;
575 return VK_SUCCESS;
576 }
577
578 static void
579 radv_descriptor_set_destroy(struct radv_device *device,
580 struct radv_descriptor_pool *pool,
581 struct radv_descriptor_set *set,
582 bool free_bo)
583 {
584 assert(!pool->host_memory_base);
585
586 if (free_bo && set->size && !pool->host_memory_base) {
587 uint32_t offset = (uint8_t*)set->mapped_ptr - pool->mapped_ptr;
588 for (int i = 0; i < pool->entry_count; ++i) {
589 if (pool->entries[i].offset == offset) {
590 memmove(&pool->entries[i], &pool->entries[i+1],
591 sizeof(pool->entries[i]) * (pool->entry_count - i - 1));
592 --pool->entry_count;
593 break;
594 }
595 }
596 }
597 vk_free2(&device->alloc, NULL, set);
598 }
599
600 VkResult radv_CreateDescriptorPool(
601 VkDevice _device,
602 const VkDescriptorPoolCreateInfo* pCreateInfo,
603 const VkAllocationCallbacks* pAllocator,
604 VkDescriptorPool* pDescriptorPool)
605 {
606 RADV_FROM_HANDLE(radv_device, device, _device);
607 struct radv_descriptor_pool *pool;
608 uint64_t size = sizeof(struct radv_descriptor_pool);
609 uint64_t bo_size = 0, bo_count = 0, range_count = 0;
610
611 vk_foreach_struct(ext, pCreateInfo->pNext) {
612 switch (ext->sType) {
613 case VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_INLINE_UNIFORM_BLOCK_CREATE_INFO_EXT: {
614 const struct VkDescriptorPoolInlineUniformBlockCreateInfoEXT *info =
615 (const struct VkDescriptorPoolInlineUniformBlockCreateInfoEXT*)ext;
616 /* the sizes are 4 aligned, and we need to align to at
617 * most 32, which needs at most 28 bytes extra per
618 * binding. */
619 bo_size += 28llu * info->maxInlineUniformBlockBindings;
620 break;
621 }
622 default:
623 break;
624 }
625 }
626
627 for (unsigned i = 0; i < pCreateInfo->poolSizeCount; ++i) {
628 if (pCreateInfo->pPoolSizes[i].type != VK_DESCRIPTOR_TYPE_SAMPLER)
629 bo_count += pCreateInfo->pPoolSizes[i].descriptorCount;
630
631 switch(pCreateInfo->pPoolSizes[i].type) {
632 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
633 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
634 range_count += pCreateInfo->pPoolSizes[i].descriptorCount;
635 break;
636 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
637 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
638 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
639 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
640 case VK_DESCRIPTOR_TYPE_SAMPLER:
641 /* 32 as we may need to align for images */
642 bo_size += 32 * pCreateInfo->pPoolSizes[i].descriptorCount;
643 break;
644 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
645 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
646 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
647 bo_size += 64 * pCreateInfo->pPoolSizes[i].descriptorCount;
648 break;
649 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
650 bo_size += 96 * pCreateInfo->pPoolSizes[i].descriptorCount;
651 break;
652 case VK_DESCRIPTOR_TYPE_INLINE_UNIFORM_BLOCK_EXT:
653 bo_size += pCreateInfo->pPoolSizes[i].descriptorCount;
654 break;
655 default:
656 unreachable("unknown descriptor type\n");
657 break;
658 }
659 }
660
661 if (!(pCreateInfo->flags & VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT)) {
662 uint64_t host_size = pCreateInfo->maxSets * sizeof(struct radv_descriptor_set);
663 host_size += sizeof(struct radeon_winsys_bo*) * bo_count;
664 host_size += sizeof(struct radv_descriptor_range) * range_count;
665 size += host_size;
666 } else {
667 size += sizeof(struct radv_descriptor_pool_entry) * pCreateInfo->maxSets;
668 }
669
670 pool = vk_alloc2(&device->alloc, pAllocator, size, 8,
671 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
672 if (!pool)
673 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
674
675 memset(pool, 0, sizeof(*pool));
676
677 if (!(pCreateInfo->flags & VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT)) {
678 pool->host_memory_base = (uint8_t*)pool + sizeof(struct radv_descriptor_pool);
679 pool->host_memory_ptr = pool->host_memory_base;
680 pool->host_memory_end = (uint8_t*)pool + size;
681 }
682
683 if (bo_size) {
684 pool->bo = device->ws->buffer_create(device->ws, bo_size, 32,
685 RADEON_DOMAIN_VRAM,
686 RADEON_FLAG_NO_INTERPROCESS_SHARING |
687 RADEON_FLAG_READ_ONLY |
688 RADEON_FLAG_32BIT,
689 RADV_BO_PRIORITY_DESCRIPTOR);
690 pool->mapped_ptr = (uint8_t*)device->ws->buffer_map(pool->bo);
691 }
692 pool->size = bo_size;
693 pool->max_entry_count = pCreateInfo->maxSets;
694
695 *pDescriptorPool = radv_descriptor_pool_to_handle(pool);
696 return VK_SUCCESS;
697 }
698
699 void radv_DestroyDescriptorPool(
700 VkDevice _device,
701 VkDescriptorPool _pool,
702 const VkAllocationCallbacks* pAllocator)
703 {
704 RADV_FROM_HANDLE(radv_device, device, _device);
705 RADV_FROM_HANDLE(radv_descriptor_pool, pool, _pool);
706
707 if (!pool)
708 return;
709
710 if (!pool->host_memory_base) {
711 for(int i = 0; i < pool->entry_count; ++i) {
712 radv_descriptor_set_destroy(device, pool, pool->entries[i].set, false);
713 }
714 }
715
716 if (pool->bo)
717 device->ws->buffer_destroy(pool->bo);
718 vk_free2(&device->alloc, pAllocator, pool);
719 }
720
721 VkResult radv_ResetDescriptorPool(
722 VkDevice _device,
723 VkDescriptorPool descriptorPool,
724 VkDescriptorPoolResetFlags flags)
725 {
726 RADV_FROM_HANDLE(radv_device, device, _device);
727 RADV_FROM_HANDLE(radv_descriptor_pool, pool, descriptorPool);
728
729 if (!pool->host_memory_base) {
730 for(int i = 0; i < pool->entry_count; ++i) {
731 radv_descriptor_set_destroy(device, pool, pool->entries[i].set, false);
732 }
733 pool->entry_count = 0;
734 }
735
736 pool->current_offset = 0;
737 pool->host_memory_ptr = pool->host_memory_base;
738
739 return VK_SUCCESS;
740 }
741
742 VkResult radv_AllocateDescriptorSets(
743 VkDevice _device,
744 const VkDescriptorSetAllocateInfo* pAllocateInfo,
745 VkDescriptorSet* pDescriptorSets)
746 {
747 RADV_FROM_HANDLE(radv_device, device, _device);
748 RADV_FROM_HANDLE(radv_descriptor_pool, pool, pAllocateInfo->descriptorPool);
749
750 VkResult result = VK_SUCCESS;
751 uint32_t i;
752 struct radv_descriptor_set *set = NULL;
753
754 const VkDescriptorSetVariableDescriptorCountAllocateInfoEXT *variable_counts =
755 vk_find_struct_const(pAllocateInfo->pNext, DESCRIPTOR_SET_VARIABLE_DESCRIPTOR_COUNT_ALLOCATE_INFO_EXT);
756 const uint32_t zero = 0;
757
758 /* allocate a set of buffers for each shader to contain descriptors */
759 for (i = 0; i < pAllocateInfo->descriptorSetCount; i++) {
760 RADV_FROM_HANDLE(radv_descriptor_set_layout, layout,
761 pAllocateInfo->pSetLayouts[i]);
762
763 const uint32_t *variable_count = NULL;
764 if (variable_counts) {
765 if (i < variable_counts->descriptorSetCount)
766 variable_count = variable_counts->pDescriptorCounts + i;
767 else
768 variable_count = &zero;
769 }
770
771 assert(!(layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
772
773 result = radv_descriptor_set_create(device, pool, layout, variable_count, &set);
774 if (result != VK_SUCCESS)
775 break;
776
777 pDescriptorSets[i] = radv_descriptor_set_to_handle(set);
778 }
779
780 if (result != VK_SUCCESS)
781 radv_FreeDescriptorSets(_device, pAllocateInfo->descriptorPool,
782 i, pDescriptorSets);
783 return result;
784 }
785
786 VkResult radv_FreeDescriptorSets(
787 VkDevice _device,
788 VkDescriptorPool descriptorPool,
789 uint32_t count,
790 const VkDescriptorSet* pDescriptorSets)
791 {
792 RADV_FROM_HANDLE(radv_device, device, _device);
793 RADV_FROM_HANDLE(radv_descriptor_pool, pool, descriptorPool);
794
795 for (uint32_t i = 0; i < count; i++) {
796 RADV_FROM_HANDLE(radv_descriptor_set, set, pDescriptorSets[i]);
797
798 if (set && !pool->host_memory_base)
799 radv_descriptor_set_destroy(device, pool, set, true);
800 }
801 return VK_SUCCESS;
802 }
803
804 static void write_texel_buffer_descriptor(struct radv_device *device,
805 struct radv_cmd_buffer *cmd_buffer,
806 unsigned *dst,
807 struct radeon_winsys_bo **buffer_list,
808 const VkBufferView _buffer_view)
809 {
810 RADV_FROM_HANDLE(radv_buffer_view, buffer_view, _buffer_view);
811
812 memcpy(dst, buffer_view->state, 4 * 4);
813
814 if (cmd_buffer)
815 radv_cs_add_buffer(device->ws, cmd_buffer->cs, buffer_view->bo);
816 else
817 *buffer_list = buffer_view->bo;
818 }
819
820 static void write_buffer_descriptor(struct radv_device *device,
821 struct radv_cmd_buffer *cmd_buffer,
822 unsigned *dst,
823 struct radeon_winsys_bo **buffer_list,
824 const VkDescriptorBufferInfo *buffer_info)
825 {
826 RADV_FROM_HANDLE(radv_buffer, buffer, buffer_info->buffer);
827 uint64_t va = radv_buffer_get_va(buffer->bo);
828 uint32_t range = buffer_info->range;
829
830 if (buffer_info->range == VK_WHOLE_SIZE)
831 range = buffer->size - buffer_info->offset;
832
833 va += buffer_info->offset + buffer->offset;
834 dst[0] = va;
835 dst[1] = S_008F04_BASE_ADDRESS_HI(va >> 32);
836 dst[2] = range;
837 dst[3] = S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X) |
838 S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y) |
839 S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z) |
840 S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W) |
841 S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT) |
842 S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32);
843
844 if (cmd_buffer)
845 radv_cs_add_buffer(device->ws, cmd_buffer->cs, buffer->bo);
846 else
847 *buffer_list = buffer->bo;
848 }
849
850 static void write_block_descriptor(struct radv_device *device,
851 struct radv_cmd_buffer *cmd_buffer,
852 void *dst,
853 const VkWriteDescriptorSet *writeset)
854 {
855 const VkWriteDescriptorSetInlineUniformBlockEXT *inline_ub =
856 vk_find_struct_const(writeset->pNext, WRITE_DESCRIPTOR_SET_INLINE_UNIFORM_BLOCK_EXT);
857
858 memcpy(dst, inline_ub->pData, inline_ub->dataSize);
859 }
860
861 static void write_dynamic_buffer_descriptor(struct radv_device *device,
862 struct radv_descriptor_range *range,
863 struct radeon_winsys_bo **buffer_list,
864 const VkDescriptorBufferInfo *buffer_info)
865 {
866 RADV_FROM_HANDLE(radv_buffer, buffer, buffer_info->buffer);
867 uint64_t va = radv_buffer_get_va(buffer->bo);
868 unsigned size = buffer_info->range;
869
870 if (buffer_info->range == VK_WHOLE_SIZE)
871 size = buffer->size - buffer_info->offset;
872
873 va += buffer_info->offset + buffer->offset;
874 range->va = va;
875 range->size = size;
876
877 *buffer_list = buffer->bo;
878 }
879
880 static void
881 write_image_descriptor(struct radv_device *device,
882 struct radv_cmd_buffer *cmd_buffer,
883 unsigned size, unsigned *dst,
884 struct radeon_winsys_bo **buffer_list,
885 VkDescriptorType descriptor_type,
886 const VkDescriptorImageInfo *image_info)
887 {
888 RADV_FROM_HANDLE(radv_image_view, iview, image_info->imageView);
889 union radv_descriptor *descriptor;
890
891 if (descriptor_type == VK_DESCRIPTOR_TYPE_STORAGE_IMAGE) {
892 descriptor = &iview->storage_descriptor;
893 } else {
894 descriptor = &iview->descriptor;
895 }
896
897 memcpy(dst, descriptor, size);
898
899 if (cmd_buffer)
900 radv_cs_add_buffer(device->ws, cmd_buffer->cs, iview->bo);
901 else
902 *buffer_list = iview->bo;
903 }
904
905 static void
906 write_combined_image_sampler_descriptor(struct radv_device *device,
907 struct radv_cmd_buffer *cmd_buffer,
908 unsigned sampler_offset,
909 unsigned *dst,
910 struct radeon_winsys_bo **buffer_list,
911 VkDescriptorType descriptor_type,
912 const VkDescriptorImageInfo *image_info,
913 bool has_sampler)
914 {
915 RADV_FROM_HANDLE(radv_sampler, sampler, image_info->sampler);
916
917 write_image_descriptor(device, cmd_buffer, sampler_offset, dst, buffer_list,
918 descriptor_type, image_info);
919 /* copy over sampler state */
920 if (has_sampler) {
921 memcpy(dst + sampler_offset / sizeof(*dst), sampler->state, 16);
922 }
923 }
924
925 static void
926 write_sampler_descriptor(struct radv_device *device,
927 unsigned *dst,
928 const VkDescriptorImageInfo *image_info)
929 {
930 RADV_FROM_HANDLE(radv_sampler, sampler, image_info->sampler);
931
932 memcpy(dst, sampler->state, 16);
933 }
934
935 void radv_update_descriptor_sets(
936 struct radv_device* device,
937 struct radv_cmd_buffer* cmd_buffer,
938 VkDescriptorSet dstSetOverride,
939 uint32_t descriptorWriteCount,
940 const VkWriteDescriptorSet* pDescriptorWrites,
941 uint32_t descriptorCopyCount,
942 const VkCopyDescriptorSet* pDescriptorCopies)
943 {
944 uint32_t i, j;
945 for (i = 0; i < descriptorWriteCount; i++) {
946 const VkWriteDescriptorSet *writeset = &pDescriptorWrites[i];
947 RADV_FROM_HANDLE(radv_descriptor_set, set,
948 dstSetOverride ? dstSetOverride : writeset->dstSet);
949 const struct radv_descriptor_set_binding_layout *binding_layout =
950 set->layout->binding + writeset->dstBinding;
951 uint32_t *ptr = set->mapped_ptr;
952 struct radeon_winsys_bo **buffer_list = set->descriptors;
953 /* Immutable samplers are not copied into push descriptors when they are
954 * allocated, so if we are writing push descriptors we have to copy the
955 * immutable samplers into them now.
956 */
957 const bool copy_immutable_samplers = cmd_buffer &&
958 binding_layout->immutable_samplers_offset && !binding_layout->immutable_samplers_equal;
959 const uint32_t *samplers = radv_immutable_samplers(set->layout, binding_layout);
960
961 ptr += binding_layout->offset / 4;
962
963 if (writeset->descriptorType == VK_DESCRIPTOR_TYPE_INLINE_UNIFORM_BLOCK_EXT) {
964 write_block_descriptor(device, cmd_buffer, (uint8_t*)ptr + writeset->dstArrayElement, writeset);
965 continue;
966 }
967
968 ptr += binding_layout->size * writeset->dstArrayElement / 4;
969 buffer_list += binding_layout->buffer_offset;
970 buffer_list += writeset->dstArrayElement;
971 for (j = 0; j < writeset->descriptorCount; ++j) {
972 switch(writeset->descriptorType) {
973 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
974 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: {
975 unsigned idx = writeset->dstArrayElement + j;
976 idx += binding_layout->dynamic_offset_offset;
977 assert(!(set->layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
978 write_dynamic_buffer_descriptor(device, set->dynamic_descriptors + idx,
979 buffer_list, writeset->pBufferInfo + j);
980 break;
981 }
982 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
983 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
984 write_buffer_descriptor(device, cmd_buffer, ptr, buffer_list,
985 writeset->pBufferInfo + j);
986 break;
987 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
988 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
989 write_texel_buffer_descriptor(device, cmd_buffer, ptr, buffer_list,
990 writeset->pTexelBufferView[j]);
991 break;
992 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
993 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
994 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
995 write_image_descriptor(device, cmd_buffer, 64, ptr, buffer_list,
996 writeset->descriptorType,
997 writeset->pImageInfo + j);
998 break;
999 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER: {
1000 unsigned sampler_offset = radv_combined_image_descriptor_sampler_offset(binding_layout);
1001 write_combined_image_sampler_descriptor(device, cmd_buffer, sampler_offset,
1002 ptr, buffer_list,
1003 writeset->descriptorType,
1004 writeset->pImageInfo + j,
1005 !binding_layout->immutable_samplers_offset);
1006 if (copy_immutable_samplers) {
1007 const unsigned idx = writeset->dstArrayElement + j;
1008 memcpy((char*)ptr + sampler_offset, samplers + 4 * idx, 16);
1009 }
1010 break;
1011 }
1012 case VK_DESCRIPTOR_TYPE_SAMPLER:
1013 if (!binding_layout->immutable_samplers_offset) {
1014 write_sampler_descriptor(device, ptr,
1015 writeset->pImageInfo + j);
1016 } else if (copy_immutable_samplers) {
1017 unsigned idx = writeset->dstArrayElement + j;
1018 memcpy(ptr, samplers + 4 * idx, 16);
1019 }
1020 break;
1021 default:
1022 unreachable("unimplemented descriptor type");
1023 break;
1024 }
1025 ptr += binding_layout->size / 4;
1026 ++buffer_list;
1027 }
1028
1029 }
1030
1031 for (i = 0; i < descriptorCopyCount; i++) {
1032 const VkCopyDescriptorSet *copyset = &pDescriptorCopies[i];
1033 RADV_FROM_HANDLE(radv_descriptor_set, src_set,
1034 copyset->srcSet);
1035 RADV_FROM_HANDLE(radv_descriptor_set, dst_set,
1036 copyset->dstSet);
1037 const struct radv_descriptor_set_binding_layout *src_binding_layout =
1038 src_set->layout->binding + copyset->srcBinding;
1039 const struct radv_descriptor_set_binding_layout *dst_binding_layout =
1040 dst_set->layout->binding + copyset->dstBinding;
1041 uint32_t *src_ptr = src_set->mapped_ptr;
1042 uint32_t *dst_ptr = dst_set->mapped_ptr;
1043 struct radeon_winsys_bo **src_buffer_list = src_set->descriptors;
1044 struct radeon_winsys_bo **dst_buffer_list = dst_set->descriptors;
1045
1046 src_ptr += src_binding_layout->offset / 4;
1047 dst_ptr += dst_binding_layout->offset / 4;
1048
1049 src_ptr += src_binding_layout->size * copyset->srcArrayElement / 4;
1050 dst_ptr += dst_binding_layout->size * copyset->dstArrayElement / 4;
1051
1052 src_buffer_list += src_binding_layout->buffer_offset;
1053 src_buffer_list += copyset->srcArrayElement;
1054
1055 dst_buffer_list += dst_binding_layout->buffer_offset;
1056 dst_buffer_list += copyset->dstArrayElement;
1057
1058 for (j = 0; j < copyset->descriptorCount; ++j) {
1059 switch (src_binding_layout->type) {
1060 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
1061 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: {
1062 unsigned src_idx = copyset->srcArrayElement + j;
1063 unsigned dst_idx = copyset->dstArrayElement + j;
1064 struct radv_descriptor_range *src_range, *dst_range;
1065 src_idx += src_binding_layout->dynamic_offset_offset;
1066 dst_idx += dst_binding_layout->dynamic_offset_offset;
1067
1068 src_range = src_set->dynamic_descriptors + src_idx;
1069 dst_range = dst_set->dynamic_descriptors + dst_idx;
1070 *dst_range = *src_range;
1071 break;
1072 }
1073 default:
1074 memcpy(dst_ptr, src_ptr, src_binding_layout->size);
1075 }
1076 src_ptr += src_binding_layout->size / 4;
1077 dst_ptr += dst_binding_layout->size / 4;
1078
1079 if (src_binding_layout->type != VK_DESCRIPTOR_TYPE_SAMPLER) {
1080 /* Sampler descriptors don't have a buffer list. */
1081 dst_buffer_list[j] = src_buffer_list[j];
1082 }
1083 }
1084 }
1085 }
1086
1087 void radv_UpdateDescriptorSets(
1088 VkDevice _device,
1089 uint32_t descriptorWriteCount,
1090 const VkWriteDescriptorSet* pDescriptorWrites,
1091 uint32_t descriptorCopyCount,
1092 const VkCopyDescriptorSet* pDescriptorCopies)
1093 {
1094 RADV_FROM_HANDLE(radv_device, device, _device);
1095
1096 radv_update_descriptor_sets(device, NULL, VK_NULL_HANDLE, descriptorWriteCount, pDescriptorWrites,
1097 descriptorCopyCount, pDescriptorCopies);
1098 }
1099
1100 VkResult radv_CreateDescriptorUpdateTemplate(VkDevice _device,
1101 const VkDescriptorUpdateTemplateCreateInfo *pCreateInfo,
1102 const VkAllocationCallbacks *pAllocator,
1103 VkDescriptorUpdateTemplate *pDescriptorUpdateTemplate)
1104 {
1105 RADV_FROM_HANDLE(radv_device, device, _device);
1106 RADV_FROM_HANDLE(radv_descriptor_set_layout, set_layout, pCreateInfo->descriptorSetLayout);
1107 const uint32_t entry_count = pCreateInfo->descriptorUpdateEntryCount;
1108 const size_t size = sizeof(struct radv_descriptor_update_template) +
1109 sizeof(struct radv_descriptor_update_template_entry) * entry_count;
1110 struct radv_descriptor_update_template *templ;
1111 uint32_t i;
1112
1113 templ = vk_alloc2(&device->alloc, pAllocator, size, 8, VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
1114 if (!templ)
1115 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
1116
1117 templ->entry_count = entry_count;
1118 templ->bind_point = pCreateInfo->pipelineBindPoint;
1119
1120 for (i = 0; i < entry_count; i++) {
1121 const VkDescriptorUpdateTemplateEntry *entry = &pCreateInfo->pDescriptorUpdateEntries[i];
1122 const struct radv_descriptor_set_binding_layout *binding_layout =
1123 set_layout->binding + entry->dstBinding;
1124 const uint32_t buffer_offset = binding_layout->buffer_offset + entry->dstArrayElement;
1125 const uint32_t *immutable_samplers = NULL;
1126 uint32_t dst_offset;
1127 uint32_t dst_stride;
1128
1129 /* dst_offset is an offset into dynamic_descriptors when the descriptor
1130 is dynamic, and an offset into mapped_ptr otherwise */
1131 switch (entry->descriptorType) {
1132 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
1133 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
1134 assert(pCreateInfo->templateType == VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_DESCRIPTOR_SET);
1135 dst_offset = binding_layout->dynamic_offset_offset + entry->dstArrayElement;
1136 dst_stride = 0; /* Not used */
1137 break;
1138 default:
1139 switch (entry->descriptorType) {
1140 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
1141 case VK_DESCRIPTOR_TYPE_SAMPLER:
1142 /* Immutable samplers are copied into push descriptors when they are pushed */
1143 if (pCreateInfo->templateType == VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_PUSH_DESCRIPTORS_KHR &&
1144 binding_layout->immutable_samplers_offset && !binding_layout->immutable_samplers_equal) {
1145 immutable_samplers = radv_immutable_samplers(set_layout, binding_layout) + entry->dstArrayElement * 4;
1146 }
1147 break;
1148 default:
1149 break;
1150 }
1151 dst_offset = binding_layout->offset / 4;
1152 if (entry->descriptorType == VK_DESCRIPTOR_TYPE_INLINE_UNIFORM_BLOCK_EXT)
1153 dst_offset += entry->dstArrayElement / 4;
1154 else
1155 dst_offset += binding_layout->size * entry->dstArrayElement / 4;
1156
1157 dst_stride = binding_layout->size / 4;
1158 break;
1159 }
1160
1161 templ->entry[i] = (struct radv_descriptor_update_template_entry) {
1162 .descriptor_type = entry->descriptorType,
1163 .descriptor_count = entry->descriptorCount,
1164 .src_offset = entry->offset,
1165 .src_stride = entry->stride,
1166 .dst_offset = dst_offset,
1167 .dst_stride = dst_stride,
1168 .buffer_offset = buffer_offset,
1169 .has_sampler = !binding_layout->immutable_samplers_offset,
1170 .sampler_offset = radv_combined_image_descriptor_sampler_offset(binding_layout),
1171 .immutable_samplers = immutable_samplers
1172 };
1173 }
1174
1175 *pDescriptorUpdateTemplate = radv_descriptor_update_template_to_handle(templ);
1176 return VK_SUCCESS;
1177 }
1178
1179 void radv_DestroyDescriptorUpdateTemplate(VkDevice _device,
1180 VkDescriptorUpdateTemplate descriptorUpdateTemplate,
1181 const VkAllocationCallbacks *pAllocator)
1182 {
1183 RADV_FROM_HANDLE(radv_device, device, _device);
1184 RADV_FROM_HANDLE(radv_descriptor_update_template, templ, descriptorUpdateTemplate);
1185
1186 if (!templ)
1187 return;
1188
1189 vk_free2(&device->alloc, pAllocator, templ);
1190 }
1191
1192 void radv_update_descriptor_set_with_template(struct radv_device *device,
1193 struct radv_cmd_buffer *cmd_buffer,
1194 struct radv_descriptor_set *set,
1195 VkDescriptorUpdateTemplate descriptorUpdateTemplate,
1196 const void *pData)
1197 {
1198 RADV_FROM_HANDLE(radv_descriptor_update_template, templ, descriptorUpdateTemplate);
1199 uint32_t i;
1200
1201 for (i = 0; i < templ->entry_count; ++i) {
1202 struct radeon_winsys_bo **buffer_list = set->descriptors + templ->entry[i].buffer_offset;
1203 uint32_t *pDst = set->mapped_ptr + templ->entry[i].dst_offset;
1204 const uint8_t *pSrc = ((const uint8_t *) pData) + templ->entry[i].src_offset;
1205 uint32_t j;
1206
1207 if (templ->entry[i].descriptor_type == VK_DESCRIPTOR_TYPE_INLINE_UNIFORM_BLOCK_EXT) {
1208 memcpy((uint8_t*)pDst, pSrc, templ->entry[i].descriptor_count);
1209 continue;
1210 }
1211
1212 for (j = 0; j < templ->entry[i].descriptor_count; ++j) {
1213 switch (templ->entry[i].descriptor_type) {
1214 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
1215 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: {
1216 const unsigned idx = templ->entry[i].dst_offset + j;
1217 assert(!(set->layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
1218 write_dynamic_buffer_descriptor(device, set->dynamic_descriptors + idx,
1219 buffer_list, (struct VkDescriptorBufferInfo *) pSrc);
1220 break;
1221 }
1222 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
1223 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
1224 write_buffer_descriptor(device, cmd_buffer, pDst, buffer_list,
1225 (struct VkDescriptorBufferInfo *) pSrc);
1226 break;
1227 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
1228 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
1229 write_texel_buffer_descriptor(device, cmd_buffer, pDst, buffer_list,
1230 *(VkBufferView *) pSrc);
1231 break;
1232 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
1233 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
1234 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
1235 write_image_descriptor(device, cmd_buffer, 64, pDst, buffer_list,
1236 templ->entry[i].descriptor_type,
1237 (struct VkDescriptorImageInfo *) pSrc);
1238 break;
1239 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
1240 write_combined_image_sampler_descriptor(device, cmd_buffer, templ->entry[i].sampler_offset,
1241 pDst, buffer_list, templ->entry[i].descriptor_type,
1242 (struct VkDescriptorImageInfo *) pSrc,
1243 templ->entry[i].has_sampler);
1244 if (templ->entry[i].immutable_samplers) {
1245 memcpy((char*)pDst + templ->entry[i].sampler_offset, templ->entry[i].immutable_samplers + 4 * j, 16);
1246 }
1247 break;
1248 case VK_DESCRIPTOR_TYPE_SAMPLER:
1249 if (templ->entry[i].has_sampler)
1250 write_sampler_descriptor(device, pDst,
1251 (struct VkDescriptorImageInfo *) pSrc);
1252 else if (templ->entry[i].immutable_samplers)
1253 memcpy(pDst, templ->entry[i].immutable_samplers + 4 * j, 16);
1254 break;
1255 default:
1256 unreachable("unimplemented descriptor type");
1257 break;
1258 }
1259 pSrc += templ->entry[i].src_stride;
1260 pDst += templ->entry[i].dst_stride;
1261 ++buffer_list;
1262 }
1263 }
1264 }
1265
1266 void radv_UpdateDescriptorSetWithTemplate(VkDevice _device,
1267 VkDescriptorSet descriptorSet,
1268 VkDescriptorUpdateTemplate descriptorUpdateTemplate,
1269 const void *pData)
1270 {
1271 RADV_FROM_HANDLE(radv_device, device, _device);
1272 RADV_FROM_HANDLE(radv_descriptor_set, set, descriptorSet);
1273
1274 radv_update_descriptor_set_with_template(device, NULL, set, descriptorUpdateTemplate, pData);
1275 }
1276
1277
1278 VkResult radv_CreateSamplerYcbcrConversion(VkDevice _device,
1279 const VkSamplerYcbcrConversionCreateInfo* pCreateInfo,
1280 const VkAllocationCallbacks* pAllocator,
1281 VkSamplerYcbcrConversion* pYcbcrConversion)
1282 {
1283 RADV_FROM_HANDLE(radv_device, device, _device);
1284 struct radv_sampler_ycbcr_conversion *conversion = NULL;
1285
1286 conversion = vk_zalloc2(&device->alloc, pAllocator, sizeof(*conversion), 8,
1287 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
1288
1289 if (conversion == NULL)
1290 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
1291
1292 conversion->format = pCreateInfo->format;
1293 conversion->ycbcr_model = pCreateInfo->ycbcrModel;
1294 conversion->ycbcr_range = pCreateInfo->ycbcrRange;
1295 conversion->components = pCreateInfo->components;
1296 conversion->chroma_offsets[0] = pCreateInfo->xChromaOffset;
1297 conversion->chroma_offsets[1] = pCreateInfo->yChromaOffset;
1298 conversion->chroma_filter = pCreateInfo->chromaFilter;
1299
1300 *pYcbcrConversion = radv_sampler_ycbcr_conversion_to_handle(conversion);
1301 return VK_SUCCESS;
1302 }
1303
1304
1305 void radv_DestroySamplerYcbcrConversion(VkDevice _device,
1306 VkSamplerYcbcrConversion ycbcrConversion,
1307 const VkAllocationCallbacks* pAllocator)
1308 {
1309 RADV_FROM_HANDLE(radv_device, device, _device);
1310 RADV_FROM_HANDLE(radv_sampler_ycbcr_conversion, ycbcr_conversion, ycbcrConversion);
1311
1312 if (ycbcr_conversion)
1313 vk_free2(&device->alloc, pAllocator, ycbcr_conversion);
1314 }