projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
radv: Fix descriptors for cube images with VK_IMAGE_USAGE_STORAGE_BIT
[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
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->size = 0;
70
71 memset(set_layout->binding, 0, size - sizeof(struct radv_descriptor_set_layout));
72
73 uint32_t buffer_count = 0;
74 uint32_t dynamic_offset_count = 0;
75
76 for (uint32_t j = 0; j < pCreateInfo->bindingCount; j++) {
77 const VkDescriptorSetLayoutBinding *binding = &pCreateInfo->pBindings[j];
78 uint32_t b = binding->binding;
79 uint32_t alignment;
80 unsigned binding_buffer_count = 0;
81
82 switch (binding->descriptorType) {
83 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
84 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
85 assert(!(pCreateInfo->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
86 set_layout->binding[b].dynamic_offset_count = 1;
87 set_layout->dynamic_shader_stages |= binding->stageFlags;
88 set_layout->binding[b].size = 0;
89 binding_buffer_count = 1;
90 alignment = 1;
91 break;
92 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
93 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
94 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
95 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
96 set_layout->binding[b].size = 16;
97 binding_buffer_count = 1;
98 alignment = 16;
99 break;
100 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
101 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
102 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
103 /* main descriptor + fmask descriptor */
104 set_layout->binding[b].size = 64;
105 binding_buffer_count = 1;
106 alignment = 32;
107 break;
108 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
109 /* main descriptor + fmask descriptor + sampler */
110 set_layout->binding[b].size = 96;
111 binding_buffer_count = 1;
112 alignment = 32;
113 break;
114 case VK_DESCRIPTOR_TYPE_SAMPLER:
115 set_layout->binding[b].size = 16;
116 alignment = 16;
117 break;
118 default:
119 unreachable("unknown descriptor type\n");
120 break;
121 }
122
123 set_layout->size = align(set_layout->size, alignment);
124 assert(binding->descriptorCount > 0);
125 set_layout->binding[b].type = binding->descriptorType;
126 set_layout->binding[b].array_size = binding->descriptorCount;
127 set_layout->binding[b].offset = set_layout->size;
128 set_layout->binding[b].buffer_offset = buffer_count;
129 set_layout->binding[b].dynamic_offset_offset = dynamic_offset_count;
130
131 if (binding->pImmutableSamplers) {
132 set_layout->binding[b].immutable_samplers_offset = samplers_offset;
133 set_layout->binding[b].immutable_samplers_equal = true;
134
135
136 for (uint32_t i = 0; i < binding->descriptorCount; i++)
137 memcpy(samplers + 4 * i, &radv_sampler_from_handle(binding->pImmutableSamplers[i])->state, 16);
138 for (uint32_t i = 1; i < binding->descriptorCount; i++)
139 if (memcmp(samplers + 4 * i, samplers, 16) != 0)
140 set_layout->binding[b].immutable_samplers_equal = false;
141
142 /* Don't reserve space for the samplers if they're not accessed. */
143 if (set_layout->binding[b].immutable_samplers_equal) {
144 if (binding->descriptorType == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER)
145 set_layout->binding[b].size -= 32;
146 else if (binding->descriptorType == VK_DESCRIPTOR_TYPE_SAMPLER)
147 set_layout->binding[b].size -= 16;
148 }
149 samplers += 4 * binding->descriptorCount;
150 samplers_offset += 4 * sizeof(uint32_t) * binding->descriptorCount;
151 }
152
153 set_layout->size += binding->descriptorCount * set_layout->binding[b].size;
154 buffer_count += binding->descriptorCount * binding_buffer_count;
155 dynamic_offset_count += binding->descriptorCount *
156 set_layout->binding[b].dynamic_offset_count;
157 set_layout->shader_stages |= binding->stageFlags;
158 }
159
160 set_layout->buffer_count = buffer_count;
161 set_layout->dynamic_offset_count = dynamic_offset_count;
162
163 *pSetLayout = radv_descriptor_set_layout_to_handle(set_layout);
164
165 return VK_SUCCESS;
166 }
167
168 void radv_DestroyDescriptorSetLayout(
169 VkDevice _device,
170 VkDescriptorSetLayout _set_layout,
171 const VkAllocationCallbacks* pAllocator)
172 {
173 RADV_FROM_HANDLE(radv_device, device, _device);
174 RADV_FROM_HANDLE(radv_descriptor_set_layout, set_layout, _set_layout);
175
176 if (!set_layout)
177 return;
178
179 vk_free2(&device->alloc, pAllocator, set_layout);
180 }
181
182 /*
183 * Pipeline layouts. These have nothing to do with the pipeline. They are
184 * just muttiple descriptor set layouts pasted together
185 */
186
187 VkResult radv_CreatePipelineLayout(
188 VkDevice _device,
189 const VkPipelineLayoutCreateInfo* pCreateInfo,
190 const VkAllocationCallbacks* pAllocator,
191 VkPipelineLayout* pPipelineLayout)
192 {
193 RADV_FROM_HANDLE(radv_device, device, _device);
194 struct radv_pipeline_layout *layout;
195 struct mesa_sha1 ctx;
196
197 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO);
198
199 layout = vk_alloc2(&device->alloc, pAllocator, sizeof(*layout), 8,
200 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
201 if (layout == NULL)
202 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
203
204 layout->num_sets = pCreateInfo->setLayoutCount;
205
206 unsigned dynamic_offset_count = 0;
207
208
209 _mesa_sha1_init(&ctx);
210 for (uint32_t set = 0; set < pCreateInfo->setLayoutCount; set++) {
211 RADV_FROM_HANDLE(radv_descriptor_set_layout, set_layout,
212 pCreateInfo->pSetLayouts[set]);
213 layout->set[set].layout = set_layout;
214
215 layout->set[set].dynamic_offset_start = dynamic_offset_count;
216 for (uint32_t b = 0; b < set_layout->binding_count; b++) {
217 dynamic_offset_count += set_layout->binding[b].array_size * set_layout->binding[b].dynamic_offset_count;
218 if (set_layout->binding[b].immutable_samplers_offset)
219 _mesa_sha1_update(&ctx, radv_immutable_samplers(set_layout, set_layout->binding + b),
220 set_layout->binding[b].array_size * 4 * sizeof(uint32_t));
221 }
222 _mesa_sha1_update(&ctx, set_layout->binding,
223 sizeof(set_layout->binding[0]) * set_layout->binding_count);
224 }
225
226 layout->dynamic_offset_count = dynamic_offset_count;
227 layout->push_constant_size = 0;
228 for (unsigned i = 0; i < pCreateInfo->pushConstantRangeCount; ++i) {
229 const VkPushConstantRange *range = pCreateInfo->pPushConstantRanges + i;
230 layout->push_constant_size = MAX2(layout->push_constant_size,
231 range->offset + range->size);
232 }
233
234 layout->push_constant_size = align(layout->push_constant_size, 16);
235 _mesa_sha1_update(&ctx, &layout->push_constant_size,
236 sizeof(layout->push_constant_size));
237 _mesa_sha1_final(&ctx, layout->sha1);
238 *pPipelineLayout = radv_pipeline_layout_to_handle(layout);
239
240 return VK_SUCCESS;
241 }
242
243 void radv_DestroyPipelineLayout(
244 VkDevice _device,
245 VkPipelineLayout _pipelineLayout,
246 const VkAllocationCallbacks* pAllocator)
247 {
248 RADV_FROM_HANDLE(radv_device, device, _device);
249 RADV_FROM_HANDLE(radv_pipeline_layout, pipeline_layout, _pipelineLayout);
250
251 if (!pipeline_layout)
252 return;
253 vk_free2(&device->alloc, pAllocator, pipeline_layout);
254 }
255
256 #define EMPTY 1
257
258 static VkResult
259 radv_descriptor_set_create(struct radv_device *device,
260 struct radv_descriptor_pool *pool,
261 const struct radv_descriptor_set_layout *layout,
262 struct radv_descriptor_set **out_set)
263 {
264 struct radv_descriptor_set *set;
265 unsigned range_offset = sizeof(struct radv_descriptor_set) +
266 sizeof(struct radeon_winsys_bo *) * layout->buffer_count;
267 unsigned mem_size = range_offset +
268 sizeof(struct radv_descriptor_range) * layout->dynamic_offset_count;
269
270 if (pool->host_memory_base) {
271 if (pool->host_memory_end - pool->host_memory_ptr < mem_size)
272 return vk_error(VK_ERROR_OUT_OF_POOL_MEMORY_KHR);
273
274 set = (struct radv_descriptor_set*)pool->host_memory_ptr;
275 pool->host_memory_ptr += mem_size;
276 } else {
277 set = vk_alloc2(&device->alloc, NULL, mem_size, 8,
278 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
279
280 if (!set)
281 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
282 }
283
284 memset(set, 0, mem_size);
285
286 if (layout->dynamic_offset_count) {
287 set->dynamic_descriptors = (struct radv_descriptor_range*)((uint8_t*)set + range_offset);
288 }
289
290 set->layout = layout;
291 if (layout->size) {
292 uint32_t layout_size = align_u32(layout->size, 32);
293 set->size = layout->size;
294
295 /* try to allocate linearly first, so that we don't spend
296 * time looking for gaps if the app only allocates &
297 * resets via the pool. */
298 if (pool->current_offset + layout_size <= pool->size) {
299 set->bo = pool->bo;
300 set->mapped_ptr = (uint32_t*)(pool->mapped_ptr + pool->current_offset);
301 set->va = device->ws->buffer_get_va(set->bo) + pool->current_offset;
302 pool->current_offset += layout_size;
303 list_addtail(&set->vram_list, &pool->vram_list);
304 } else if (!pool->host_memory_base) {
305 uint64_t offset = 0;
306 struct list_head *prev = &pool->vram_list;
307 struct radv_descriptor_set *cur;
308
309 assert(!pool->host_memory_base);
310 LIST_FOR_EACH_ENTRY(cur, &pool->vram_list, vram_list) {
311 uint64_t start = (uint8_t*)cur->mapped_ptr - pool->mapped_ptr;
312 if (start - offset >= layout_size)
313 break;
314
315 offset = start + cur->size;
316 prev = &cur->vram_list;
317 }
318
319 if (pool->size - offset < layout_size) {
320 vk_free2(&device->alloc, NULL, set);
321 return vk_error(VK_ERROR_OUT_OF_POOL_MEMORY_KHR);
322 }
323 set->bo = pool->bo;
324 set->mapped_ptr = (uint32_t*)(pool->mapped_ptr + offset);
325 set->va = device->ws->buffer_get_va(set->bo) + offset;
326 list_add(&set->vram_list, prev);
327 } else
328 return vk_error(VK_ERROR_OUT_OF_POOL_MEMORY_KHR);
329 }
330
331 for (unsigned i = 0; i < layout->binding_count; ++i) {
332 if (!layout->binding[i].immutable_samplers_offset ||
333 layout->binding[i].immutable_samplers_equal)
334 continue;
335
336 unsigned offset = layout->binding[i].offset / 4;
337 if (layout->binding[i].type == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER)
338 offset += 16;
339
340 const uint32_t *samplers = (const uint32_t*)((const char*)layout + layout->binding[i].immutable_samplers_offset);
341 for (unsigned j = 0; j < layout->binding[i].array_size; ++j) {
342 memcpy(set->mapped_ptr + offset, samplers + 4 * j, 16);
343 offset += layout->binding[i].size / 4;
344 }
345
346 }
347 *out_set = set;
348 return VK_SUCCESS;
349 }
350
351 static void
352 radv_descriptor_set_destroy(struct radv_device *device,
353 struct radv_descriptor_pool *pool,
354 struct radv_descriptor_set *set,
355 bool free_bo)
356 {
357 assert(!pool->host_memory_base);
358
359 if (free_bo && set->size)
360 list_del(&set->vram_list);
361 vk_free2(&device->alloc, NULL, set);
362 }
363
364 VkResult radv_CreateDescriptorPool(
365 VkDevice _device,
366 const VkDescriptorPoolCreateInfo* pCreateInfo,
367 const VkAllocationCallbacks* pAllocator,
368 VkDescriptorPool* pDescriptorPool)
369 {
370 RADV_FROM_HANDLE(radv_device, device, _device);
371 struct radv_descriptor_pool *pool;
372 int size = sizeof(struct radv_descriptor_pool);
373 uint64_t bo_size = 0, bo_count = 0, range_count = 0;
374
375
376 for (unsigned i = 0; i < pCreateInfo->poolSizeCount; ++i) {
377 if (pCreateInfo->pPoolSizes[i].type != VK_DESCRIPTOR_TYPE_SAMPLER)
378 bo_count += pCreateInfo->pPoolSizes[i].descriptorCount;
379
380 switch(pCreateInfo->pPoolSizes[i].type) {
381 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
382 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
383 range_count += pCreateInfo->pPoolSizes[i].descriptorCount;
384 break;
385 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
386 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
387 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
388 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
389 case VK_DESCRIPTOR_TYPE_SAMPLER:
390 /* 32 as we may need to align for images */
391 bo_size += 32 * pCreateInfo->pPoolSizes[i].descriptorCount;
392 break;
393 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
394 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
395 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
396 bo_size += 64 * pCreateInfo->pPoolSizes[i].descriptorCount;
397 break;
398 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
399 bo_size += 96 * pCreateInfo->pPoolSizes[i].descriptorCount;
400 break;
401 default:
402 unreachable("unknown descriptor type\n");
403 break;
404 }
405 }
406
407 if (!(pCreateInfo->flags & VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT)) {
408 uint64_t host_size = pCreateInfo->maxSets * sizeof(struct radv_descriptor_set);
409 host_size += sizeof(struct radeon_winsys_bo*) * bo_count;
410 host_size += sizeof(struct radv_descriptor_range) * range_count;
411 size += host_size;
412 }
413
414 pool = vk_alloc2(&device->alloc, pAllocator, size, 8,
415 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
416 if (!pool)
417 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
418
419 memset(pool, 0, sizeof(*pool));
420
421 if (!(pCreateInfo->flags & VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT)) {
422 pool->host_memory_base = (uint8_t*)pool + sizeof(struct radv_descriptor_pool);
423 pool->host_memory_ptr = pool->host_memory_base;
424 pool->host_memory_end = (uint8_t*)pool + size;
425 }
426
427 if (bo_size) {
428 pool->bo = device->ws->buffer_create(device->ws, bo_size,
429 32, RADEON_DOMAIN_VRAM, 0);
430 pool->mapped_ptr = (uint8_t*)device->ws->buffer_map(pool->bo);
431 }
432 pool->size = bo_size;
433
434 list_inithead(&pool->vram_list);
435 *pDescriptorPool = radv_descriptor_pool_to_handle(pool);
436 return VK_SUCCESS;
437 }
438
439 void radv_DestroyDescriptorPool(
440 VkDevice _device,
441 VkDescriptorPool _pool,
442 const VkAllocationCallbacks* pAllocator)
443 {
444 RADV_FROM_HANDLE(radv_device, device, _device);
445 RADV_FROM_HANDLE(radv_descriptor_pool, pool, _pool);
446
447 if (!pool)
448 return;
449
450 if (!pool->host_memory_base) {
451 list_for_each_entry_safe(struct radv_descriptor_set, set,
452 &pool->vram_list, vram_list) {
453 radv_descriptor_set_destroy(device, pool, set, false);
454 }
455 }
456
457 if (pool->bo)
458 device->ws->buffer_destroy(pool->bo);
459 vk_free2(&device->alloc, pAllocator, pool);
460 }
461
462 VkResult radv_ResetDescriptorPool(
463 VkDevice _device,
464 VkDescriptorPool descriptorPool,
465 VkDescriptorPoolResetFlags flags)
466 {
467 RADV_FROM_HANDLE(radv_device, device, _device);
468 RADV_FROM_HANDLE(radv_descriptor_pool, pool, descriptorPool);
469
470 if (!pool->host_memory_base) {
471 list_for_each_entry_safe(struct radv_descriptor_set, set,
472 &pool->vram_list, vram_list) {
473 radv_descriptor_set_destroy(device, pool, set, false);
474 }
475 }
476
477 list_inithead(&pool->vram_list);
478
479 pool->current_offset = 0;
480 pool->host_memory_ptr = pool->host_memory_base;
481
482 return VK_SUCCESS;
483 }
484
485 VkResult radv_AllocateDescriptorSets(
486 VkDevice _device,
487 const VkDescriptorSetAllocateInfo* pAllocateInfo,
488 VkDescriptorSet* pDescriptorSets)
489 {
490 RADV_FROM_HANDLE(radv_device, device, _device);
491 RADV_FROM_HANDLE(radv_descriptor_pool, pool, pAllocateInfo->descriptorPool);
492
493 VkResult result = VK_SUCCESS;
494 uint32_t i;
495 struct radv_descriptor_set *set;
496
497 /* allocate a set of buffers for each shader to contain descriptors */
498 for (i = 0; i < pAllocateInfo->descriptorSetCount; i++) {
499 RADV_FROM_HANDLE(radv_descriptor_set_layout, layout,
500 pAllocateInfo->pSetLayouts[i]);
501
502 assert(!(layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
503
504 result = radv_descriptor_set_create(device, pool, layout, &set);
505 if (result != VK_SUCCESS)
506 break;
507
508 pDescriptorSets[i] = radv_descriptor_set_to_handle(set);
509 }
510
511 if (result != VK_SUCCESS)
512 radv_FreeDescriptorSets(_device, pAllocateInfo->descriptorPool,
513 i, pDescriptorSets);
514 return result;
515 }
516
517 VkResult radv_FreeDescriptorSets(
518 VkDevice _device,
519 VkDescriptorPool descriptorPool,
520 uint32_t count,
521 const VkDescriptorSet* pDescriptorSets)
522 {
523 RADV_FROM_HANDLE(radv_device, device, _device);
524 RADV_FROM_HANDLE(radv_descriptor_pool, pool, descriptorPool);
525
526 for (uint32_t i = 0; i < count; i++) {
527 RADV_FROM_HANDLE(radv_descriptor_set, set, pDescriptorSets[i]);
528
529 if (set && !pool->host_memory_base)
530 radv_descriptor_set_destroy(device, pool, set, true);
531 }
532 return VK_SUCCESS;
533 }
534
535 static void write_texel_buffer_descriptor(struct radv_device *device,
536 struct radv_cmd_buffer *cmd_buffer,
537 unsigned *dst,
538 struct radeon_winsys_bo **buffer_list,
539 const VkBufferView _buffer_view)
540 {
541 RADV_FROM_HANDLE(radv_buffer_view, buffer_view, _buffer_view);
542
543 memcpy(dst, buffer_view->state, 4 * 4);
544
545 if (cmd_buffer)
546 device->ws->cs_add_buffer(cmd_buffer->cs, buffer_view->bo, 7);
547 else
548 *buffer_list = buffer_view->bo;
549 }
550
551 static void write_buffer_descriptor(struct radv_device *device,
552 struct radv_cmd_buffer *cmd_buffer,
553 unsigned *dst,
554 struct radeon_winsys_bo **buffer_list,
555 const VkDescriptorBufferInfo *buffer_info)
556 {
557 RADV_FROM_HANDLE(radv_buffer, buffer, buffer_info->buffer);
558 uint64_t va = device->ws->buffer_get_va(buffer->bo);
559 uint32_t range = buffer_info->range;
560
561 if (buffer_info->range == VK_WHOLE_SIZE)
562 range = buffer->size - buffer_info->offset;
563
564 va += buffer_info->offset + buffer->offset;
565 dst[0] = va;
566 dst[1] = S_008F04_BASE_ADDRESS_HI(va >> 32);
567 dst[2] = range;
568 dst[3] = S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X) |
569 S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y) |
570 S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z) |
571 S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W) |
572 S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT) |
573 S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32);
574
575 if (cmd_buffer)
576 device->ws->cs_add_buffer(cmd_buffer->cs, buffer->bo, 7);
577 else
578 *buffer_list = buffer->bo;
579 }
580
581 static void write_dynamic_buffer_descriptor(struct radv_device *device,
582 struct radv_descriptor_range *range,
583 struct radeon_winsys_bo **buffer_list,
584 const VkDescriptorBufferInfo *buffer_info)
585 {
586 RADV_FROM_HANDLE(radv_buffer, buffer, buffer_info->buffer);
587 uint64_t va = device->ws->buffer_get_va(buffer->bo);
588 unsigned size = buffer_info->range;
589
590 if (buffer_info->range == VK_WHOLE_SIZE)
591 size = buffer->size - buffer_info->offset;
592
593 va += buffer_info->offset + buffer->offset;
594 range->va = va;
595 range->size = size;
596
597 *buffer_list = buffer->bo;
598 }
599
600 static void
601 write_image_descriptor(struct radv_device *device,
602 struct radv_cmd_buffer *cmd_buffer,
603 unsigned *dst,
604 struct radeon_winsys_bo **buffer_list,
605 VkDescriptorType descriptor_type,
606 const VkDescriptorImageInfo *image_info)
607 {
608 RADV_FROM_HANDLE(radv_image_view, iview, image_info->imageView);
609
610 if (descriptor_type == VK_DESCRIPTOR_TYPE_STORAGE_IMAGE) {
611 memcpy(dst, iview->storage_descriptor, 8 * 4);
612 memcpy(dst + 8, iview->storage_fmask_descriptor, 8 * 4);
613 } else {
614 memcpy(dst, iview->descriptor, 8 * 4);
615 memcpy(dst + 8, iview->fmask_descriptor, 8 * 4);
616 }
617
618 if (cmd_buffer)
619 device->ws->cs_add_buffer(cmd_buffer->cs, iview->bo, 7);
620 else
621 *buffer_list = iview->bo;
622 }
623
624 static void
625 write_combined_image_sampler_descriptor(struct radv_device *device,
626 struct radv_cmd_buffer *cmd_buffer,
627 unsigned *dst,
628 struct radeon_winsys_bo **buffer_list,
629 VkDescriptorType descriptor_type,
630 const VkDescriptorImageInfo *image_info,
631 bool has_sampler)
632 {
633 RADV_FROM_HANDLE(radv_sampler, sampler, image_info->sampler);
634
635 write_image_descriptor(device, cmd_buffer, dst, buffer_list, descriptor_type, image_info);
636 /* copy over sampler state */
637 if (has_sampler)
638 memcpy(dst + 16, sampler->state, 16);
639 }
640
641 static void
642 write_sampler_descriptor(struct radv_device *device,
643 unsigned *dst,
644 const VkDescriptorImageInfo *image_info)
645 {
646 RADV_FROM_HANDLE(radv_sampler, sampler, image_info->sampler);
647
648 memcpy(dst, sampler->state, 16);
649 }
650
651 void radv_update_descriptor_sets(
652 struct radv_device* device,
653 struct radv_cmd_buffer* cmd_buffer,
654 VkDescriptorSet dstSetOverride,
655 uint32_t descriptorWriteCount,
656 const VkWriteDescriptorSet* pDescriptorWrites,
657 uint32_t descriptorCopyCount,
658 const VkCopyDescriptorSet* pDescriptorCopies)
659 {
660 uint32_t i, j;
661 for (i = 0; i < descriptorWriteCount; i++) {
662 const VkWriteDescriptorSet *writeset = &pDescriptorWrites[i];
663 RADV_FROM_HANDLE(radv_descriptor_set, set,
664 dstSetOverride ? dstSetOverride : writeset->dstSet);
665 const struct radv_descriptor_set_binding_layout *binding_layout =
666 set->layout->binding + writeset->dstBinding;
667 uint32_t *ptr = set->mapped_ptr;
668 struct radeon_winsys_bo **buffer_list = set->descriptors;
669 /* Immutable samplers are not copied into push descriptors when they are
670 * allocated, so if we are writing push descriptors we have to copy the
671 * immutable samplers into them now.
672 */
673 const bool copy_immutable_samplers = cmd_buffer &&
674 binding_layout->immutable_samplers_offset && !binding_layout->immutable_samplers_equal;
675 const uint32_t *samplers = radv_immutable_samplers(set->layout, binding_layout);
676
677 ptr += binding_layout->offset / 4;
678 ptr += binding_layout->size * writeset->dstArrayElement / 4;
679 buffer_list += binding_layout->buffer_offset;
680 buffer_list += writeset->dstArrayElement;
681 for (j = 0; j < writeset->descriptorCount; ++j) {
682 switch(writeset->descriptorType) {
683 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
684 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: {
685 unsigned idx = writeset->dstArrayElement + j;
686 idx += binding_layout->dynamic_offset_offset;
687 assert(!(set->layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
688 write_dynamic_buffer_descriptor(device, set->dynamic_descriptors + idx,
689 buffer_list, writeset->pBufferInfo + j);
690 break;
691 }
692 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
693 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
694 write_buffer_descriptor(device, cmd_buffer, ptr, buffer_list,
695 writeset->pBufferInfo + j);
696 break;
697 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
698 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
699 write_texel_buffer_descriptor(device, cmd_buffer, ptr, buffer_list,
700 writeset->pTexelBufferView[j]);
701 break;
702 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
703 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
704 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
705 write_image_descriptor(device, cmd_buffer, ptr, buffer_list,
706 writeset->descriptorType,
707 writeset->pImageInfo + j);
708 break;
709 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
710 write_combined_image_sampler_descriptor(device, cmd_buffer, ptr, buffer_list,
711 writeset->descriptorType,
712 writeset->pImageInfo + j,
713 !binding_layout->immutable_samplers_offset);
714 if (copy_immutable_samplers) {
715 const unsigned idx = writeset->dstArrayElement + j;
716 memcpy(ptr + 16, samplers + 4 * idx, 16);
717 }
718 break;
719 case VK_DESCRIPTOR_TYPE_SAMPLER:
720 if (!binding_layout->immutable_samplers_offset) {
721 write_sampler_descriptor(device, ptr,
722 writeset->pImageInfo + j);
723 } else if (copy_immutable_samplers) {
724 unsigned idx = writeset->dstArrayElement + j;
725 memcpy(ptr, samplers + 4 * idx, 16);
726 }
727 break;
728 default:
729 unreachable("unimplemented descriptor type");
730 break;
731 }
732 ptr += binding_layout->size / 4;
733 ++buffer_list;
734 }
735
736 }
737 if (descriptorCopyCount)
738 radv_finishme("copy descriptors");
739 }
740
741 void radv_UpdateDescriptorSets(
742 VkDevice _device,
743 uint32_t descriptorWriteCount,
744 const VkWriteDescriptorSet* pDescriptorWrites,
745 uint32_t descriptorCopyCount,
746 const VkCopyDescriptorSet* pDescriptorCopies)
747 {
748 RADV_FROM_HANDLE(radv_device, device, _device);
749
750 radv_update_descriptor_sets(device, NULL, VK_NULL_HANDLE, descriptorWriteCount, pDescriptorWrites,
751 descriptorCopyCount, pDescriptorCopies);
752 }
753
754 VkResult radv_CreateDescriptorUpdateTemplateKHR(VkDevice _device,
755 const VkDescriptorUpdateTemplateCreateInfoKHR *pCreateInfo,
756 const VkAllocationCallbacks *pAllocator,
757 VkDescriptorUpdateTemplateKHR *pDescriptorUpdateTemplate)
758 {
759 RADV_FROM_HANDLE(radv_device, device, _device);
760 RADV_FROM_HANDLE(radv_descriptor_set_layout, set_layout, pCreateInfo->descriptorSetLayout);
761 const uint32_t entry_count = pCreateInfo->descriptorUpdateEntryCount;
762 const size_t size = sizeof(struct radv_descriptor_update_template) +
763 sizeof(struct radv_descriptor_update_template_entry) * entry_count;
764 struct radv_descriptor_update_template *templ;
765 uint32_t i;
766
767 templ = vk_alloc2(&device->alloc, pAllocator, size, 8, VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
768 if (!templ)
769 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
770
771 templ->entry_count = entry_count;
772
773 for (i = 0; i < entry_count; i++) {
774 const VkDescriptorUpdateTemplateEntryKHR *entry = &pCreateInfo->pDescriptorUpdateEntries[i];
775 const struct radv_descriptor_set_binding_layout *binding_layout =
776 set_layout->binding + entry->dstBinding;
777 const uint32_t buffer_offset = binding_layout->buffer_offset + entry->dstArrayElement;
778 const uint32_t *immutable_samplers = NULL;
779 uint32_t dst_offset;
780 uint32_t dst_stride;
781
782 /* dst_offset is an offset into dynamic_descriptors when the descriptor
783 is dynamic, and an offset into mapped_ptr otherwise */
784 switch (entry->descriptorType) {
785 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
786 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
787 assert(pCreateInfo->templateType == VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_DESCRIPTOR_SET_KHR);
788 dst_offset = binding_layout->dynamic_offset_offset + entry->dstArrayElement;
789 dst_stride = 0; /* Not used */
790 break;
791 default:
792 switch (entry->descriptorType) {
793 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
794 case VK_DESCRIPTOR_TYPE_SAMPLER:
795 /* Immutable samplers are copied into push descriptors when they are pushed */
796 if (pCreateInfo->templateType == VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_PUSH_DESCRIPTORS_KHR &&
797 binding_layout->immutable_samplers_offset && !binding_layout->immutable_samplers_equal) {
798 immutable_samplers = radv_immutable_samplers(set_layout, binding_layout) + entry->dstArrayElement * 4;
799 }
800 break;
801 default:
802 break;
803 }
804 dst_offset = binding_layout->offset / 4 + binding_layout->size * entry->dstArrayElement / 4;
805 dst_stride = binding_layout->size / 4;
806 break;
807 }
808
809 templ->entry[i] = (struct radv_descriptor_update_template_entry) {
810 .descriptor_type = entry->descriptorType,
811 .descriptor_count = entry->descriptorCount,
812 .src_offset = entry->offset,
813 .src_stride = entry->stride,
814 .dst_offset = dst_offset,
815 .dst_stride = dst_stride,
816 .buffer_offset = buffer_offset,
817 .has_sampler = !binding_layout->immutable_samplers_offset,
818 .immutable_samplers = immutable_samplers
819 };
820 }
821
822 *pDescriptorUpdateTemplate = radv_descriptor_update_template_to_handle(templ);
823 return VK_SUCCESS;
824 }
825
826 void radv_DestroyDescriptorUpdateTemplateKHR(VkDevice _device,
827 VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate,
828 const VkAllocationCallbacks *pAllocator)
829 {
830 RADV_FROM_HANDLE(radv_device, device, _device);
831 RADV_FROM_HANDLE(radv_descriptor_update_template, templ, descriptorUpdateTemplate);
832
833 if (!templ)
834 return;
835
836 vk_free2(&device->alloc, pAllocator, templ);
837 }
838
839 void radv_update_descriptor_set_with_template(struct radv_device *device,
840 struct radv_cmd_buffer *cmd_buffer,
841 struct radv_descriptor_set *set,
842 VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate,
843 const void *pData)
844 {
845 RADV_FROM_HANDLE(radv_descriptor_update_template, templ, descriptorUpdateTemplate);
846 uint32_t i;
847
848 for (i = 0; i < templ->entry_count; ++i) {
849 struct radeon_winsys_bo **buffer_list = set->descriptors + templ->entry[i].buffer_offset;
850 uint32_t *pDst = set->mapped_ptr + templ->entry[i].dst_offset;
851 const uint8_t *pSrc = ((const uint8_t *) pData) + templ->entry[i].src_offset;
852 uint32_t j;
853
854 for (j = 0; j < templ->entry[i].descriptor_count; ++j) {
855 switch (templ->entry[i].descriptor_type) {
856 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
857 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: {
858 const unsigned idx = templ->entry[i].dst_offset + j;
859 assert(!(set->layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
860 write_dynamic_buffer_descriptor(device, set->dynamic_descriptors + idx,
861 buffer_list, (struct VkDescriptorBufferInfo *) pSrc);
862 break;
863 }
864 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
865 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
866 write_buffer_descriptor(device, cmd_buffer, pDst, buffer_list,
867 (struct VkDescriptorBufferInfo *) pSrc);
868 break;
869 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
870 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
871 write_texel_buffer_descriptor(device, cmd_buffer, pDst, buffer_list,
872 *(VkBufferView *) pSrc);
873 break;
874 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
875 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
876 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
877 write_image_descriptor(device, cmd_buffer, pDst, buffer_list,
878 templ->entry[i].descriptor_type,
879 (struct VkDescriptorImageInfo *) pSrc);
880 break;
881 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
882 write_combined_image_sampler_descriptor(device, cmd_buffer, pDst, buffer_list,
883 templ->entry[i].descriptor_type,
884 (struct VkDescriptorImageInfo *) pSrc,
885 templ->entry[i].has_sampler);
886 if (templ->entry[i].immutable_samplers)
887 memcpy(pDst + 16, templ->entry[i].immutable_samplers + 4 * j, 16);
888 break;
889 case VK_DESCRIPTOR_TYPE_SAMPLER:
890 if (templ->entry[i].has_sampler)
891 write_sampler_descriptor(device, pDst,
892 (struct VkDescriptorImageInfo *) pSrc);
893 else if (templ->entry[i].immutable_samplers)
894 memcpy(pDst, templ->entry[i].immutable_samplers + 4 * j, 16);
895 break;
896 default:
897 unreachable("unimplemented descriptor type");
898 break;
899 }
900 pSrc += templ->entry[i].src_stride;
901 pDst += templ->entry[i].dst_stride;
902 ++buffer_list;
903 }
904 }
905 }
906
907 void radv_UpdateDescriptorSetWithTemplateKHR(VkDevice _device,
908 VkDescriptorSet descriptorSet,
909 VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate,
910 const void *pData)
911 {
912 RADV_FROM_HANDLE(radv_device, device, _device);
913 RADV_FROM_HANDLE(radv_descriptor_set, set, descriptorSet);
914
915 radv_update_descriptor_set_with_template(device, NULL, set, descriptorUpdateTemplate, pData);
916 }