projects / mesa.git / blob
? search:


89945f177363aae033cfda1981e6a4fdf13d73a5
[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
81 switch (binding->descriptorType) {
82 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
83 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
84 assert(!(pCreateInfo->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
85 set_layout->binding[b].dynamic_offset_count = 1;
86 set_layout->dynamic_shader_stages |= binding->stageFlags;
87 set_layout->binding[b].size = 0;
88 set_layout->binding[b].buffer_count = 1;
89 alignment = 1;
90 break;
91 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
92 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
93 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
94 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
95 set_layout->binding[b].size = 16;
96 set_layout->binding[b].buffer_count = 1;
97 alignment = 16;
98 break;
99 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
100 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
101 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
102 /* main descriptor + fmask descriptor */
103 set_layout->binding[b].size = 64;
104 set_layout->binding[b].buffer_count = 1;
105 alignment = 32;
106 break;
107 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
108 /* main descriptor + fmask descriptor + sampler */
109 set_layout->binding[b].size = 96;
110 set_layout->binding[b].buffer_count = 1;
111 alignment = 32;
112 break;
113 case VK_DESCRIPTOR_TYPE_SAMPLER:
114 set_layout->binding[b].size = 16;
115 alignment = 16;
116 break;
117 default:
118 unreachable("unknown descriptor type\n");
119 break;
120 }
121
122 set_layout->size = align(set_layout->size, alignment);
123 assert(binding->descriptorCount > 0);
124 set_layout->binding[b].type = binding->descriptorType;
125 set_layout->binding[b].array_size = binding->descriptorCount;
126 set_layout->binding[b].offset = set_layout->size;
127 set_layout->binding[b].buffer_offset = buffer_count;
128 set_layout->binding[b].dynamic_offset_offset = dynamic_offset_count;
129
130 if (binding->pImmutableSamplers) {
131 set_layout->binding[b].immutable_samplers_offset = samplers_offset;
132 set_layout->binding[b].immutable_samplers_equal = true;
133
134
135 for (uint32_t i = 0; i < binding->descriptorCount; i++)
136 memcpy(samplers + 4 * i, &radv_sampler_from_handle(binding->pImmutableSamplers[i])->state, 16);
137 for (uint32_t i = 1; i < binding->descriptorCount; i++)
138 if (memcmp(samplers + 4 * i, samplers, 16) != 0)
139 set_layout->binding[b].immutable_samplers_equal = false;
140
141 /* Don't reserve space for the samplers if they're not accessed. */
142 if (set_layout->binding[b].immutable_samplers_equal) {
143 if (binding->descriptorType == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER)
144 set_layout->binding[b].size -= 32;
145 else if (binding->descriptorType == VK_DESCRIPTOR_TYPE_SAMPLER)
146 set_layout->binding[b].size -= 16;
147 }
148 samplers += 4 * binding->descriptorCount;
149 samplers_offset += 4 * sizeof(uint32_t) * binding->descriptorCount;
150 }
151
152 set_layout->size += binding->descriptorCount * set_layout->binding[b].size;
153 buffer_count += binding->descriptorCount * set_layout->binding[b].buffer_count;
154 dynamic_offset_count += binding->descriptorCount *
155 set_layout->binding[b].dynamic_offset_count;
156 set_layout->shader_stages |= binding->stageFlags;
157 }
158
159 set_layout->buffer_count = buffer_count;
160 set_layout->dynamic_offset_count = dynamic_offset_count;
161
162 *pSetLayout = radv_descriptor_set_layout_to_handle(set_layout);
163
164 return VK_SUCCESS;
165 }
166
167 void radv_DestroyDescriptorSetLayout(
168 VkDevice _device,
169 VkDescriptorSetLayout _set_layout,
170 const VkAllocationCallbacks* pAllocator)
171 {
172 RADV_FROM_HANDLE(radv_device, device, _device);
173 RADV_FROM_HANDLE(radv_descriptor_set_layout, set_layout, _set_layout);
174
175 if (!set_layout)
176 return;
177
178 vk_free2(&device->alloc, pAllocator, set_layout);
179 }
180
181 /*
182 * Pipeline layouts. These have nothing to do with the pipeline. They are
183 * just muttiple descriptor set layouts pasted together
184 */
185
186 VkResult radv_CreatePipelineLayout(
187 VkDevice _device,
188 const VkPipelineLayoutCreateInfo* pCreateInfo,
189 const VkAllocationCallbacks* pAllocator,
190 VkPipelineLayout* pPipelineLayout)
191 {
192 RADV_FROM_HANDLE(radv_device, device, _device);
193 struct radv_pipeline_layout *layout;
194 struct mesa_sha1 ctx;
195
196 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO);
197
198 layout = vk_alloc2(&device->alloc, pAllocator, sizeof(*layout), 8,
199 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
200 if (layout == NULL)
201 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
202
203 layout->num_sets = pCreateInfo->setLayoutCount;
204
205 unsigned dynamic_offset_count = 0;
206
207
208 _mesa_sha1_init(&ctx);
209 for (uint32_t set = 0; set < pCreateInfo->setLayoutCount; set++) {
210 RADV_FROM_HANDLE(radv_descriptor_set_layout, set_layout,
211 pCreateInfo->pSetLayouts[set]);
212 layout->set[set].layout = set_layout;
213
214 layout->set[set].dynamic_offset_start = dynamic_offset_count;
215 for (uint32_t b = 0; b < set_layout->binding_count; b++) {
216 dynamic_offset_count += set_layout->binding[b].array_size * set_layout->binding[b].dynamic_offset_count;
217 if (set_layout->binding[b].immutable_samplers_offset)
218 _mesa_sha1_update(&ctx, radv_immutable_samplers(set_layout, set_layout->binding + b),
219 set_layout->binding[b].array_size * 4 * sizeof(uint32_t));
220 }
221 _mesa_sha1_update(&ctx, set_layout->binding,
222 sizeof(set_layout->binding[0]) * set_layout->binding_count);
223 }
224
225 layout->dynamic_offset_count = dynamic_offset_count;
226 layout->push_constant_size = 0;
227 for (unsigned i = 0; i < pCreateInfo->pushConstantRangeCount; ++i) {
228 const VkPushConstantRange *range = pCreateInfo->pPushConstantRanges + i;
229 layout->push_constant_size = MAX2(layout->push_constant_size,
230 range->offset + range->size);
231 }
232
233 layout->push_constant_size = align(layout->push_constant_size, 16);
234 _mesa_sha1_update(&ctx, &layout->push_constant_size,
235 sizeof(layout->push_constant_size));
236 _mesa_sha1_final(&ctx, layout->sha1);
237 *pPipelineLayout = radv_pipeline_layout_to_handle(layout);
238
239 return VK_SUCCESS;
240 }
241
242 void radv_DestroyPipelineLayout(
243 VkDevice _device,
244 VkPipelineLayout _pipelineLayout,
245 const VkAllocationCallbacks* pAllocator)
246 {
247 RADV_FROM_HANDLE(radv_device, device, _device);
248 RADV_FROM_HANDLE(radv_pipeline_layout, pipeline_layout, _pipelineLayout);
249
250 if (!pipeline_layout)
251 return;
252 vk_free2(&device->alloc, pAllocator, pipeline_layout);
253 }
254
255 #define EMPTY 1
256
257 static VkResult
258 radv_descriptor_set_create(struct radv_device *device,
259 struct radv_descriptor_pool *pool,
260 const struct radv_descriptor_set_layout *layout,
261 struct radv_descriptor_set **out_set)
262 {
263 struct radv_descriptor_set *set;
264 unsigned range_offset = sizeof(struct radv_descriptor_set) +
265 sizeof(struct radeon_winsys_bo *) * layout->buffer_count;
266 unsigned mem_size = range_offset +
267 sizeof(struct radv_descriptor_range) * layout->dynamic_offset_count;
268
269 if (pool->host_memory_base) {
270 if (pool->host_memory_end - pool->host_memory_ptr < mem_size)
271 return vk_error(VK_ERROR_OUT_OF_POOL_MEMORY_KHR);
272
273 set = (struct radv_descriptor_set*)pool->host_memory_ptr;
274 pool->host_memory_ptr += mem_size;
275 } else {
276 set = vk_alloc2(&device->alloc, NULL, mem_size, 8,
277 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
278
279 if (!set)
280 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
281 }
282
283 memset(set, 0, mem_size);
284
285 if (layout->dynamic_offset_count) {
286 set->dynamic_descriptors = (struct radv_descriptor_range*)((uint8_t*)set + range_offset);
287 }
288
289 set->layout = layout;
290 if (layout->size) {
291 uint32_t layout_size = align_u32(layout->size, 32);
292 set->size = layout->size;
293
294 /* try to allocate linearly first, so that we don't spend
295 * time looking for gaps if the app only allocates &
296 * resets via the pool. */
297 if (pool->current_offset + layout_size <= pool->size) {
298 set->bo = pool->bo;
299 set->mapped_ptr = (uint32_t*)(pool->mapped_ptr + pool->current_offset);
300 set->va = device->ws->buffer_get_va(set->bo) + pool->current_offset;
301 pool->current_offset += layout_size;
302 list_addtail(&set->vram_list, &pool->vram_list);
303 } else {
304 uint64_t offset = 0;
305 struct list_head *prev = &pool->vram_list;
306 struct radv_descriptor_set *cur;
307
308 assert(!pool->host_memory_base);
309 LIST_FOR_EACH_ENTRY(cur, &pool->vram_list, vram_list) {
310 uint64_t start = (uint8_t*)cur->mapped_ptr - pool->mapped_ptr;
311 if (start - offset >= layout_size)
312 break;
313
314 offset = start + cur->size;
315 prev = &cur->vram_list;
316 }
317
318 if (pool->size - offset < layout_size) {
319 vk_free2(&device->alloc, NULL, set->dynamic_descriptors);
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 }
328 }
329
330 for (unsigned i = 0; i < layout->binding_count; ++i) {
331 if (!layout->binding[i].immutable_samplers_offset ||
332 layout->binding[i].immutable_samplers_equal)
333 continue;
334
335 unsigned offset = layout->binding[i].offset / 4;
336 if (layout->binding[i].type == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER)
337 offset += 16;
338
339 const uint32_t *samplers = (const uint32_t*)((const char*)layout + layout->binding[i].immutable_samplers_offset);
340 for (unsigned j = 0; j < layout->binding[i].array_size; ++j) {
341 memcpy(set->mapped_ptr + offset, samplers + 4 * j, 16);
342 offset += layout->binding[i].size / 4;
343 }
344
345 }
346 *out_set = set;
347 return VK_SUCCESS;
348 }
349
350 static void
351 radv_descriptor_set_destroy(struct radv_device *device,
352 struct radv_descriptor_pool *pool,
353 struct radv_descriptor_set *set,
354 bool free_bo)
355 {
356 assert(!pool->host_memory_base);
357
358 if (free_bo && set->size)
359 list_del(&set->vram_list);
360 vk_free2(&device->alloc, NULL, set);
361 }
362
363 VkResult radv_CreateDescriptorPool(
364 VkDevice _device,
365 const VkDescriptorPoolCreateInfo* pCreateInfo,
366 const VkAllocationCallbacks* pAllocator,
367 VkDescriptorPool* pDescriptorPool)
368 {
369 RADV_FROM_HANDLE(radv_device, device, _device);
370 struct radv_descriptor_pool *pool;
371 int size = sizeof(struct radv_descriptor_pool);
372 uint64_t bo_size = 0, bo_count = 0, range_count = 0;
373
374
375 for (unsigned i = 0; i < pCreateInfo->poolSizeCount; ++i) {
376 if (pCreateInfo->pPoolSizes[i].type != VK_DESCRIPTOR_TYPE_SAMPLER)
377 bo_count += pCreateInfo->pPoolSizes[i].descriptorCount;
378
379 switch(pCreateInfo->pPoolSizes[i].type) {
380 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
381 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
382 range_count += pCreateInfo->pPoolSizes[i].descriptorCount;
383 break;
384 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
385 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
386 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
387 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
388 case VK_DESCRIPTOR_TYPE_SAMPLER:
389 /* 32 as we may need to align for images */
390 bo_size += 32 * pCreateInfo->pPoolSizes[i].descriptorCount;
391 break;
392 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
393 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
394 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
395 bo_size += 64 * pCreateInfo->pPoolSizes[i].descriptorCount;
396 break;
397 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
398 bo_size += 96 * pCreateInfo->pPoolSizes[i].descriptorCount;
399 break;
400 default:
401 unreachable("unknown descriptor type\n");
402 break;
403 }
404 }
405
406 if (!(pCreateInfo->flags & VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT)) {
407 uint64_t host_size = pCreateInfo->maxSets * sizeof(struct radv_descriptor_set);
408 host_size += sizeof(struct radeon_winsys_bo*) * bo_count;
409 host_size += sizeof(struct radv_descriptor_range) * range_count;
410 size += host_size;
411 }
412
413 pool = vk_alloc2(&device->alloc, pAllocator, size, 8,
414 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
415 if (!pool)
416 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
417
418 memset(pool, 0, sizeof(*pool));
419
420 if (!(pCreateInfo->flags & VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT)) {
421 pool->host_memory_base = (uint8_t*)pool + sizeof(struct radv_descriptor_pool);
422 pool->host_memory_ptr = pool->host_memory_base;
423 pool->host_memory_end = (uint8_t*)pool + size;
424 }
425
426 if (bo_size) {
427 pool->bo = device->ws->buffer_create(device->ws, bo_size,
428 32, RADEON_DOMAIN_VRAM, 0);
429 pool->mapped_ptr = (uint8_t*)device->ws->buffer_map(pool->bo);
430 }
431 pool->size = bo_size;
432
433 list_inithead(&pool->vram_list);
434 *pDescriptorPool = radv_descriptor_pool_to_handle(pool);
435 return VK_SUCCESS;
436 }
437
438 void radv_DestroyDescriptorPool(
439 VkDevice _device,
440 VkDescriptorPool _pool,
441 const VkAllocationCallbacks* pAllocator)
442 {
443 RADV_FROM_HANDLE(radv_device, device, _device);
444 RADV_FROM_HANDLE(radv_descriptor_pool, pool, _pool);
445
446 if (!pool)
447 return;
448
449 if (!pool->host_memory_base) {
450 list_for_each_entry_safe(struct radv_descriptor_set, set,
451 &pool->vram_list, vram_list) {
452 radv_descriptor_set_destroy(device, pool, set, false);
453 }
454 }
455
456 if (pool->bo)
457 device->ws->buffer_destroy(pool->bo);
458 vk_free2(&device->alloc, pAllocator, pool);
459 }
460
461 VkResult radv_ResetDescriptorPool(
462 VkDevice _device,
463 VkDescriptorPool descriptorPool,
464 VkDescriptorPoolResetFlags flags)
465 {
466 RADV_FROM_HANDLE(radv_device, device, _device);
467 RADV_FROM_HANDLE(radv_descriptor_pool, pool, descriptorPool);
468
469 if (!pool->host_memory_base) {
470 list_for_each_entry_safe(struct radv_descriptor_set, set,
471 &pool->vram_list, vram_list) {
472 radv_descriptor_set_destroy(device, pool, set, false);
473 }
474 }
475
476 list_inithead(&pool->vram_list);
477
478 pool->current_offset = 0;
479 pool->host_memory_ptr = pool->host_memory_base;
480
481 return VK_SUCCESS;
482 }
483
484 VkResult radv_AllocateDescriptorSets(
485 VkDevice _device,
486 const VkDescriptorSetAllocateInfo* pAllocateInfo,
487 VkDescriptorSet* pDescriptorSets)
488 {
489 RADV_FROM_HANDLE(radv_device, device, _device);
490 RADV_FROM_HANDLE(radv_descriptor_pool, pool, pAllocateInfo->descriptorPool);
491
492 VkResult result = VK_SUCCESS;
493 uint32_t i;
494 struct radv_descriptor_set *set;
495
496 /* allocate a set of buffers for each shader to contain descriptors */
497 for (i = 0; i < pAllocateInfo->descriptorSetCount; i++) {
498 RADV_FROM_HANDLE(radv_descriptor_set_layout, layout,
499 pAllocateInfo->pSetLayouts[i]);
500
501 assert(!(layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
502
503 result = radv_descriptor_set_create(device, pool, layout, &set);
504 if (result != VK_SUCCESS)
505 break;
506
507 pDescriptorSets[i] = radv_descriptor_set_to_handle(set);
508 }
509
510 if (result != VK_SUCCESS)
511 radv_FreeDescriptorSets(_device, pAllocateInfo->descriptorPool,
512 i, pDescriptorSets);
513 return result;
514 }
515
516 VkResult radv_FreeDescriptorSets(
517 VkDevice _device,
518 VkDescriptorPool descriptorPool,
519 uint32_t count,
520 const VkDescriptorSet* pDescriptorSets)
521 {
522 RADV_FROM_HANDLE(radv_device, device, _device);
523 RADV_FROM_HANDLE(radv_descriptor_pool, pool, descriptorPool);
524
525 for (uint32_t i = 0; i < count; i++) {
526 RADV_FROM_HANDLE(radv_descriptor_set, set, pDescriptorSets[i]);
527
528 if (set)
529 radv_descriptor_set_destroy(device, pool, set, true);
530 }
531 return VK_SUCCESS;
532 }
533
534 static void write_texel_buffer_descriptor(struct radv_device *device,
535 struct radv_cmd_buffer *cmd_buffer,
536 unsigned *dst,
537 struct radeon_winsys_bo **buffer_list,
538 const VkBufferView _buffer_view)
539 {
540 RADV_FROM_HANDLE(radv_buffer_view, buffer_view, _buffer_view);
541
542 memcpy(dst, buffer_view->state, 4 * 4);
543
544 if (cmd_buffer)
545 device->ws->cs_add_buffer(cmd_buffer->cs, buffer_view->bo, 7);
546 else
547 *buffer_list = buffer_view->bo;
548 }
549
550 static void write_buffer_descriptor(struct radv_device *device,
551 struct radv_cmd_buffer *cmd_buffer,
552 unsigned *dst,
553 struct radeon_winsys_bo **buffer_list,
554 const VkDescriptorBufferInfo *buffer_info)
555 {
556 RADV_FROM_HANDLE(radv_buffer, buffer, buffer_info->buffer);
557 uint64_t va = device->ws->buffer_get_va(buffer->bo);
558 uint32_t range = buffer_info->range;
559
560 if (buffer_info->range == VK_WHOLE_SIZE)
561 range = buffer->size - buffer_info->offset;
562
563 va += buffer_info->offset + buffer->offset;
564 dst[0] = va;
565 dst[1] = S_008F04_BASE_ADDRESS_HI(va >> 32);
566 dst[2] = range;
567 dst[3] = S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X) |
568 S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y) |
569 S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z) |
570 S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W) |
571 S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT) |
572 S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32);
573
574 if (cmd_buffer)
575 device->ws->cs_add_buffer(cmd_buffer->cs, buffer->bo, 7);
576 else
577 *buffer_list = buffer->bo;
578 }
579
580 static void write_dynamic_buffer_descriptor(struct radv_device *device,
581 struct radv_descriptor_range *range,
582 struct radeon_winsys_bo **buffer_list,
583 const VkDescriptorBufferInfo *buffer_info)
584 {
585 RADV_FROM_HANDLE(radv_buffer, buffer, buffer_info->buffer);
586 uint64_t va = device->ws->buffer_get_va(buffer->bo);
587 unsigned size = buffer_info->range;
588
589 if (buffer_info->range == VK_WHOLE_SIZE)
590 size = buffer->size - buffer_info->offset;
591
592 va += buffer_info->offset + buffer->offset;
593 range->va = va;
594 range->size = size;
595
596 *buffer_list = buffer->bo;
597 }
598
599 static void
600 write_image_descriptor(struct radv_device *device,
601 struct radv_cmd_buffer *cmd_buffer,
602 unsigned *dst,
603 struct radeon_winsys_bo **buffer_list,
604 const VkDescriptorImageInfo *image_info)
605 {
606 RADV_FROM_HANDLE(radv_image_view, iview, image_info->imageView);
607 memcpy(dst, iview->descriptor, 8 * 4);
608 memcpy(dst + 8, iview->fmask_descriptor, 8 * 4);
609
610 if (cmd_buffer)
611 device->ws->cs_add_buffer(cmd_buffer->cs, iview->bo, 7);
612 else
613 *buffer_list = iview->bo;
614 }
615
616 static void
617 write_combined_image_sampler_descriptor(struct radv_device *device,
618 struct radv_cmd_buffer *cmd_buffer,
619 unsigned *dst,
620 struct radeon_winsys_bo **buffer_list,
621 const VkDescriptorImageInfo *image_info,
622 bool has_sampler)
623 {
624 RADV_FROM_HANDLE(radv_sampler, sampler, image_info->sampler);
625
626 write_image_descriptor(device, cmd_buffer, dst, buffer_list, image_info);
627 /* copy over sampler state */
628 if (has_sampler)
629 memcpy(dst + 16, sampler->state, 16);
630 }
631
632 static void
633 write_sampler_descriptor(struct radv_device *device,
634 unsigned *dst,
635 const VkDescriptorImageInfo *image_info)
636 {
637 RADV_FROM_HANDLE(radv_sampler, sampler, image_info->sampler);
638
639 memcpy(dst, sampler->state, 16);
640 }
641
642 void radv_update_descriptor_sets(
643 struct radv_device* device,
644 struct radv_cmd_buffer* cmd_buffer,
645 VkDescriptorSet dstSetOverride,
646 uint32_t descriptorWriteCount,
647 const VkWriteDescriptorSet* pDescriptorWrites,
648 uint32_t descriptorCopyCount,
649 const VkCopyDescriptorSet* pDescriptorCopies)
650 {
651 uint32_t i, j;
652 for (i = 0; i < descriptorWriteCount; i++) {
653 const VkWriteDescriptorSet *writeset = &pDescriptorWrites[i];
654 RADV_FROM_HANDLE(radv_descriptor_set, set,
655 dstSetOverride ? dstSetOverride : writeset->dstSet);
656 const struct radv_descriptor_set_binding_layout *binding_layout =
657 set->layout->binding + writeset->dstBinding;
658 uint32_t *ptr = set->mapped_ptr;
659 struct radeon_winsys_bo **buffer_list = set->descriptors;
660 /* Immutable samplers are not copied into push descriptors when they are
661 * allocated, so if we are writing push descriptors we have to copy the
662 * immutable samplers into them now.
663 */
664 const bool copy_immutable_samplers = cmd_buffer &&
665 binding_layout->immutable_samplers_offset && !binding_layout->immutable_samplers_equal;
666 const uint32_t *samplers = radv_immutable_samplers(set->layout, binding_layout);
667
668 ptr += binding_layout->offset / 4;
669 ptr += binding_layout->size * writeset->dstArrayElement / 4;
670 buffer_list += binding_layout->buffer_offset;
671 buffer_list += binding_layout->buffer_count * writeset->dstArrayElement;
672 for (j = 0; j < writeset->descriptorCount; ++j) {
673 switch(writeset->descriptorType) {
674 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
675 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: {
676 unsigned idx = writeset->dstArrayElement + j;
677 idx += binding_layout->dynamic_offset_offset;
678 assert(!(set->layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
679 write_dynamic_buffer_descriptor(device, set->dynamic_descriptors + idx,
680 buffer_list, writeset->pBufferInfo + j);
681 break;
682 }
683 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
684 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
685 write_buffer_descriptor(device, cmd_buffer, ptr, buffer_list,
686 writeset->pBufferInfo + j);
687 break;
688 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
689 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
690 write_texel_buffer_descriptor(device, cmd_buffer, ptr, buffer_list,
691 writeset->pTexelBufferView[j]);
692 break;
693 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
694 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
695 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
696 write_image_descriptor(device, cmd_buffer, ptr, buffer_list,
697 writeset->pImageInfo + j);
698 break;
699 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
700 write_combined_image_sampler_descriptor(device, cmd_buffer, ptr, buffer_list,
701 writeset->pImageInfo + j,
702 !binding_layout->immutable_samplers_offset);
703 if (copy_immutable_samplers) {
704 const unsigned idx = writeset->dstArrayElement + j;
705 memcpy(ptr + 16, samplers + 4 * idx, 16);
706 }
707 break;
708 case VK_DESCRIPTOR_TYPE_SAMPLER:
709 if (!binding_layout->immutable_samplers_offset) {
710 write_sampler_descriptor(device, ptr,
711 writeset->pImageInfo + j);
712 } else if (copy_immutable_samplers) {
713 unsigned idx = writeset->dstArrayElement + j;
714 memcpy(ptr, samplers + 4 * idx, 16);
715 }
716 break;
717 default:
718 unreachable("unimplemented descriptor type");
719 break;
720 }
721 ptr += binding_layout->size / 4;
722 buffer_list += binding_layout->buffer_count;
723 }
724
725 }
726 if (descriptorCopyCount)
727 radv_finishme("copy descriptors");
728 }
729
730 void radv_UpdateDescriptorSets(
731 VkDevice _device,
732 uint32_t descriptorWriteCount,
733 const VkWriteDescriptorSet* pDescriptorWrites,
734 uint32_t descriptorCopyCount,
735 const VkCopyDescriptorSet* pDescriptorCopies)
736 {
737 RADV_FROM_HANDLE(radv_device, device, _device);
738
739 radv_update_descriptor_sets(device, NULL, VK_NULL_HANDLE, descriptorWriteCount, pDescriptorWrites,
740 descriptorCopyCount, pDescriptorCopies);
741 }
742
743 VkResult radv_CreateDescriptorUpdateTemplateKHR(VkDevice _device,
744 const VkDescriptorUpdateTemplateCreateInfoKHR *pCreateInfo,
745 const VkAllocationCallbacks *pAllocator,
746 VkDescriptorUpdateTemplateKHR *pDescriptorUpdateTemplate)
747 {
748 RADV_FROM_HANDLE(radv_device, device, _device);
749 RADV_FROM_HANDLE(radv_descriptor_set_layout, set_layout, pCreateInfo->descriptorSetLayout);
750 const uint32_t entry_count = pCreateInfo->descriptorUpdateEntryCount;
751 const size_t size = sizeof(struct radv_descriptor_update_template) +
752 sizeof(struct radv_descriptor_update_template_entry) * entry_count;
753 struct radv_descriptor_update_template *templ;
754 uint32_t i;
755
756 templ = vk_alloc2(&device->alloc, pAllocator, size, 8, VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
757 if (!templ)
758 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
759
760 templ->entry_count = entry_count;
761
762 for (i = 0; i < entry_count; i++) {
763 const VkDescriptorUpdateTemplateEntryKHR *entry = &pCreateInfo->pDescriptorUpdateEntries[i];
764 const struct radv_descriptor_set_binding_layout *binding_layout =
765 set_layout->binding + entry->dstBinding;
766 const uint32_t buffer_offset = binding_layout->buffer_offset +
767 binding_layout->buffer_count * entry->dstArrayElement;
768 const uint32_t *immutable_samplers = NULL;
769 uint32_t dst_offset;
770 uint32_t dst_stride;
771
772 /* dst_offset is an offset into dynamic_descriptors when the descriptor
773 is dynamic, and an offset into mapped_ptr otherwise */
774 switch (entry->descriptorType) {
775 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
776 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
777 assert(pCreateInfo->templateType == VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_DESCRIPTOR_SET_KHR);
778 dst_offset = binding_layout->dynamic_offset_offset + entry->dstArrayElement;
779 dst_stride = 0; /* Not used */
780 break;
781 default:
782 switch (entry->descriptorType) {
783 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
784 case VK_DESCRIPTOR_TYPE_SAMPLER:
785 /* Immutable samplers are copied into push descriptors when they are pushed */
786 if (pCreateInfo->templateType == VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_PUSH_DESCRIPTORS_KHR &&
787 binding_layout->immutable_samplers_offset && !binding_layout->immutable_samplers_equal) {
788 immutable_samplers = radv_immutable_samplers(set_layout, binding_layout) + entry->dstArrayElement * 4;
789 }
790 break;
791 default:
792 break;
793 }
794 dst_offset = binding_layout->offset / 4 + binding_layout->size * entry->dstArrayElement / 4;
795 dst_stride = binding_layout->size / 4;
796 break;
797 }
798
799 templ->entry[i] = (struct radv_descriptor_update_template_entry) {
800 .descriptor_type = entry->descriptorType,
801 .descriptor_count = entry->descriptorCount,
802 .src_offset = entry->offset,
803 .src_stride = entry->stride,
804 .dst_offset = dst_offset,
805 .dst_stride = dst_stride,
806 .buffer_offset = buffer_offset,
807 .buffer_count = binding_layout->buffer_count,
808 .has_sampler = !binding_layout->immutable_samplers_offset,
809 .immutable_samplers = immutable_samplers
810 };
811 }
812
813 *pDescriptorUpdateTemplate = radv_descriptor_update_template_to_handle(templ);
814 return VK_SUCCESS;
815 }
816
817 void radv_DestroyDescriptorUpdateTemplateKHR(VkDevice _device,
818 VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate,
819 const VkAllocationCallbacks *pAllocator)
820 {
821 RADV_FROM_HANDLE(radv_device, device, _device);
822 RADV_FROM_HANDLE(radv_descriptor_update_template, templ, descriptorUpdateTemplate);
823
824 if (!templ)
825 return;
826
827 vk_free2(&device->alloc, pAllocator, templ);
828 }
829
830 void radv_update_descriptor_set_with_template(struct radv_device *device,
831 struct radv_cmd_buffer *cmd_buffer,
832 struct radv_descriptor_set *set,
833 VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate,
834 const void *pData)
835 {
836 RADV_FROM_HANDLE(radv_descriptor_update_template, templ, descriptorUpdateTemplate);
837 uint32_t i;
838
839 for (i = 0; i < templ->entry_count; ++i) {
840 struct radeon_winsys_bo **buffer_list = set->descriptors + templ->entry[i].buffer_offset;
841 uint32_t *pDst = set->mapped_ptr + templ->entry[i].dst_offset;
842 const uint8_t *pSrc = ((const uint8_t *) pData) + templ->entry[i].src_offset;
843 uint32_t j;
844
845 for (j = 0; j < templ->entry[i].descriptor_count; ++j) {
846 switch (templ->entry[i].descriptor_type) {
847 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
848 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: {
849 const unsigned idx = templ->entry[i].dst_offset + j;
850 assert(!(set->layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
851 write_dynamic_buffer_descriptor(device, set->dynamic_descriptors + idx,
852 buffer_list, (struct VkDescriptorBufferInfo *) pSrc);
853 break;
854 }
855 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
856 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
857 write_buffer_descriptor(device, cmd_buffer, pDst, buffer_list,
858 (struct VkDescriptorBufferInfo *) pSrc);
859 break;
860 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
861 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
862 write_texel_buffer_descriptor(device, cmd_buffer, pDst, buffer_list,
863 *(VkBufferView *) pSrc);
864 break;
865 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
866 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
867 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
868 write_image_descriptor(device, cmd_buffer, pDst, buffer_list,
869 (struct VkDescriptorImageInfo *) pSrc);
870 break;
871 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
872 write_combined_image_sampler_descriptor(device, cmd_buffer, pDst, buffer_list,
873 (struct VkDescriptorImageInfo *) pSrc,
874 templ->entry[i].has_sampler);
875 if (templ->entry[i].immutable_samplers)
876 memcpy(pDst + 16, templ->entry[i].immutable_samplers + 4 * j, 16);
877 break;
878 case VK_DESCRIPTOR_TYPE_SAMPLER:
879 if (templ->entry[i].has_sampler)
880 write_sampler_descriptor(device, pDst,
881 (struct VkDescriptorImageInfo *) pSrc);
882 else if (templ->entry[i].immutable_samplers)
883 memcpy(pDst, templ->entry[i].immutable_samplers + 4 * j, 16);
884 break;
885 default:
886 unreachable("unimplemented descriptor type");
887 break;
888 }
889 pSrc += templ->entry[i].src_stride;
890 pDst += templ->entry[i].dst_stride;
891 buffer_list += templ->entry[i].buffer_count;
892 }
893 }
894 }
895
896 void radv_UpdateDescriptorSetWithTemplateKHR(VkDevice _device,
897 VkDescriptorSet descriptorSet,
898 VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate,
899 const void *pData)
900 {
901 RADV_FROM_HANDLE(radv_device, device, _device);
902 RADV_FROM_HANDLE(radv_descriptor_set, set, descriptorSet);
903
904 radv_update_descriptor_set_with_template(device, NULL, set, descriptorUpdateTemplate, pData);
905 }