projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
radv: Don't use a virtual function for getting the buffer virtual address.
[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->dynamic_shader_stages = 0;
70 set_layout->has_immutable_samplers = false;
71 set_layout->size = 0;
72
73 memset(set_layout->binding, 0, size - sizeof(struct radv_descriptor_set_layout));
74
75 uint32_t buffer_count = 0;
76 uint32_t dynamic_offset_count = 0;
77
78 for (uint32_t j = 0; j < pCreateInfo->bindingCount; j++) {
79 const VkDescriptorSetLayoutBinding *binding = &pCreateInfo->pBindings[j];
80 uint32_t b = binding->binding;
81 uint32_t alignment;
82 unsigned binding_buffer_count = 0;
83
84 switch (binding->descriptorType) {
85 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
86 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
87 assert(!(pCreateInfo->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
88 set_layout->binding[b].dynamic_offset_count = 1;
89 set_layout->dynamic_shader_stages |= binding->stageFlags;
90 set_layout->binding[b].size = 0;
91 binding_buffer_count = 1;
92 alignment = 1;
93 break;
94 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
95 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
96 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
97 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
98 set_layout->binding[b].size = 16;
99 binding_buffer_count = 1;
100 alignment = 16;
101 break;
102 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
103 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
104 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
105 /* main descriptor + fmask descriptor */
106 set_layout->binding[b].size = 64;
107 binding_buffer_count = 1;
108 alignment = 32;
109 break;
110 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
111 /* main descriptor + fmask descriptor + sampler */
112 set_layout->binding[b].size = 96;
113 binding_buffer_count = 1;
114 alignment = 32;
115 break;
116 case VK_DESCRIPTOR_TYPE_SAMPLER:
117 set_layout->binding[b].size = 16;
118 alignment = 16;
119 break;
120 default:
121 unreachable("unknown descriptor type\n");
122 break;
123 }
124
125 set_layout->size = align(set_layout->size, alignment);
126 assert(binding->descriptorCount > 0);
127 set_layout->binding[b].type = binding->descriptorType;
128 set_layout->binding[b].array_size = binding->descriptorCount;
129 set_layout->binding[b].offset = set_layout->size;
130 set_layout->binding[b].buffer_offset = buffer_count;
131 set_layout->binding[b].dynamic_offset_offset = dynamic_offset_count;
132
133 if (binding->pImmutableSamplers) {
134 set_layout->binding[b].immutable_samplers_offset = samplers_offset;
135 set_layout->binding[b].immutable_samplers_equal = true;
136 set_layout->has_immutable_samplers = true;
137
138
139 for (uint32_t i = 0; i < binding->descriptorCount; i++)
140 memcpy(samplers + 4 * i, &radv_sampler_from_handle(binding->pImmutableSamplers[i])->state, 16);
141 for (uint32_t i = 1; i < binding->descriptorCount; i++)
142 if (memcmp(samplers + 4 * i, samplers, 16) != 0)
143 set_layout->binding[b].immutable_samplers_equal = false;
144
145 /* Don't reserve space for the samplers if they're not accessed. */
146 if (set_layout->binding[b].immutable_samplers_equal) {
147 if (binding->descriptorType == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER)
148 set_layout->binding[b].size -= 32;
149 else if (binding->descriptorType == VK_DESCRIPTOR_TYPE_SAMPLER)
150 set_layout->binding[b].size -= 16;
151 }
152 samplers += 4 * binding->descriptorCount;
153 samplers_offset += 4 * sizeof(uint32_t) * binding->descriptorCount;
154 }
155
156 set_layout->size += binding->descriptorCount * set_layout->binding[b].size;
157 buffer_count += binding->descriptorCount * binding_buffer_count;
158 dynamic_offset_count += binding->descriptorCount *
159 set_layout->binding[b].dynamic_offset_count;
160 set_layout->shader_stages |= binding->stageFlags;
161 }
162
163 set_layout->buffer_count = buffer_count;
164 set_layout->dynamic_offset_count = dynamic_offset_count;
165
166 *pSetLayout = radv_descriptor_set_layout_to_handle(set_layout);
167
168 return VK_SUCCESS;
169 }
170
171 void radv_DestroyDescriptorSetLayout(
172 VkDevice _device,
173 VkDescriptorSetLayout _set_layout,
174 const VkAllocationCallbacks* pAllocator)
175 {
176 RADV_FROM_HANDLE(radv_device, device, _device);
177 RADV_FROM_HANDLE(radv_descriptor_set_layout, set_layout, _set_layout);
178
179 if (!set_layout)
180 return;
181
182 vk_free2(&device->alloc, pAllocator, set_layout);
183 }
184
185 /*
186 * Pipeline layouts. These have nothing to do with the pipeline. They are
187 * just muttiple descriptor set layouts pasted together
188 */
189
190 VkResult radv_CreatePipelineLayout(
191 VkDevice _device,
192 const VkPipelineLayoutCreateInfo* pCreateInfo,
193 const VkAllocationCallbacks* pAllocator,
194 VkPipelineLayout* pPipelineLayout)
195 {
196 RADV_FROM_HANDLE(radv_device, device, _device);
197 struct radv_pipeline_layout *layout;
198 struct mesa_sha1 ctx;
199
200 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO);
201
202 layout = vk_alloc2(&device->alloc, pAllocator, sizeof(*layout), 8,
203 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
204 if (layout == NULL)
205 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
206
207 layout->num_sets = pCreateInfo->setLayoutCount;
208
209 unsigned dynamic_offset_count = 0;
210
211
212 _mesa_sha1_init(&ctx);
213 for (uint32_t set = 0; set < pCreateInfo->setLayoutCount; set++) {
214 RADV_FROM_HANDLE(radv_descriptor_set_layout, set_layout,
215 pCreateInfo->pSetLayouts[set]);
216 layout->set[set].layout = set_layout;
217
218 layout->set[set].dynamic_offset_start = dynamic_offset_count;
219 for (uint32_t b = 0; b < set_layout->binding_count; b++) {
220 dynamic_offset_count += set_layout->binding[b].array_size * set_layout->binding[b].dynamic_offset_count;
221 if (set_layout->binding[b].immutable_samplers_offset)
222 _mesa_sha1_update(&ctx, radv_immutable_samplers(set_layout, set_layout->binding + b),
223 set_layout->binding[b].array_size * 4 * sizeof(uint32_t));
224 }
225 _mesa_sha1_update(&ctx, set_layout->binding,
226 sizeof(set_layout->binding[0]) * set_layout->binding_count);
227 }
228
229 layout->dynamic_offset_count = dynamic_offset_count;
230 layout->push_constant_size = 0;
231 for (unsigned i = 0; i < pCreateInfo->pushConstantRangeCount; ++i) {
232 const VkPushConstantRange *range = pCreateInfo->pPushConstantRanges + i;
233 layout->push_constant_size = MAX2(layout->push_constant_size,
234 range->offset + range->size);
235 }
236
237 layout->push_constant_size = align(layout->push_constant_size, 16);
238 _mesa_sha1_update(&ctx, &layout->push_constant_size,
239 sizeof(layout->push_constant_size));
240 _mesa_sha1_final(&ctx, layout->sha1);
241 *pPipelineLayout = radv_pipeline_layout_to_handle(layout);
242
243 return VK_SUCCESS;
244 }
245
246 void radv_DestroyPipelineLayout(
247 VkDevice _device,
248 VkPipelineLayout _pipelineLayout,
249 const VkAllocationCallbacks* pAllocator)
250 {
251 RADV_FROM_HANDLE(radv_device, device, _device);
252 RADV_FROM_HANDLE(radv_pipeline_layout, pipeline_layout, _pipelineLayout);
253
254 if (!pipeline_layout)
255 return;
256 vk_free2(&device->alloc, pAllocator, pipeline_layout);
257 }
258
259 #define EMPTY 1
260
261 static VkResult
262 radv_descriptor_set_create(struct radv_device *device,
263 struct radv_descriptor_pool *pool,
264 const struct radv_descriptor_set_layout *layout,
265 struct radv_descriptor_set **out_set)
266 {
267 struct radv_descriptor_set *set;
268 unsigned range_offset = sizeof(struct radv_descriptor_set) +
269 sizeof(struct radeon_winsys_bo *) * layout->buffer_count;
270 unsigned mem_size = range_offset +
271 sizeof(struct radv_descriptor_range) * layout->dynamic_offset_count;
272
273 if (pool->host_memory_base) {
274 if (pool->host_memory_end - pool->host_memory_ptr < mem_size)
275 return vk_error(VK_ERROR_OUT_OF_POOL_MEMORY_KHR);
276
277 set = (struct radv_descriptor_set*)pool->host_memory_ptr;
278 pool->host_memory_ptr += mem_size;
279 } else {
280 set = vk_alloc2(&device->alloc, NULL, mem_size, 8,
281 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
282
283 if (!set)
284 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
285 }
286
287 memset(set, 0, mem_size);
288
289 if (layout->dynamic_offset_count) {
290 set->dynamic_descriptors = (struct radv_descriptor_range*)((uint8_t*)set + range_offset);
291 }
292
293 set->layout = layout;
294 if (layout->size) {
295 uint32_t layout_size = align_u32(layout->size, 32);
296 set->size = layout->size;
297
298 /* try to allocate linearly first, so that we don't spend
299 * time looking for gaps if the app only allocates &
300 * resets via the pool. */
301 if (pool->current_offset + layout_size <= pool->size) {
302 set->bo = pool->bo;
303 set->mapped_ptr = (uint32_t*)(pool->mapped_ptr + pool->current_offset);
304 set->va = radv_buffer_get_va(set->bo) + pool->current_offset;
305 pool->current_offset += layout_size;
306 list_addtail(&set->vram_list, &pool->vram_list);
307 } else if (!pool->host_memory_base) {
308 uint64_t offset = 0;
309 struct list_head *prev = &pool->vram_list;
310 struct radv_descriptor_set *cur;
311
312 assert(!pool->host_memory_base);
313 LIST_FOR_EACH_ENTRY(cur, &pool->vram_list, vram_list) {
314 uint64_t start = (uint8_t*)cur->mapped_ptr - pool->mapped_ptr;
315 if (start - offset >= layout_size)
316 break;
317
318 offset = start + cur->size;
319 prev = &cur->vram_list;
320 }
321
322 if (pool->size - offset < layout_size) {
323 vk_free2(&device->alloc, NULL, set);
324 return vk_error(VK_ERROR_OUT_OF_POOL_MEMORY_KHR);
325 }
326 set->bo = pool->bo;
327 set->mapped_ptr = (uint32_t*)(pool->mapped_ptr + offset);
328 set->va = radv_buffer_get_va(set->bo) + offset;
329 list_add(&set->vram_list, prev);
330 } else
331 return vk_error(VK_ERROR_OUT_OF_POOL_MEMORY_KHR);
332 }
333
334 if (layout->has_immutable_samplers) {
335 for (unsigned i = 0; i < layout->binding_count; ++i) {
336 if (!layout->binding[i].immutable_samplers_offset ||
337 layout->binding[i].immutable_samplers_equal)
338 continue;
339
340 unsigned offset = layout->binding[i].offset / 4;
341 if (layout->binding[i].type == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER)
342 offset += 16;
343
344 const uint32_t *samplers = (const uint32_t*)((const char*)layout + layout->binding[i].immutable_samplers_offset);
345 for (unsigned j = 0; j < layout->binding[i].array_size; ++j) {
346 memcpy(set->mapped_ptr + offset, samplers + 4 * j, 16);
347 offset += layout->binding[i].size / 4;
348 }
349
350 }
351 }
352 *out_set = set;
353 return VK_SUCCESS;
354 }
355
356 static void
357 radv_descriptor_set_destroy(struct radv_device *device,
358 struct radv_descriptor_pool *pool,
359 struct radv_descriptor_set *set,
360 bool free_bo)
361 {
362 assert(!pool->host_memory_base);
363
364 if (free_bo && set->size)
365 list_del(&set->vram_list);
366 vk_free2(&device->alloc, NULL, set);
367 }
368
369 VkResult radv_CreateDescriptorPool(
370 VkDevice _device,
371 const VkDescriptorPoolCreateInfo* pCreateInfo,
372 const VkAllocationCallbacks* pAllocator,
373 VkDescriptorPool* pDescriptorPool)
374 {
375 RADV_FROM_HANDLE(radv_device, device, _device);
376 struct radv_descriptor_pool *pool;
377 int size = sizeof(struct radv_descriptor_pool);
378 uint64_t bo_size = 0, bo_count = 0, range_count = 0;
379
380
381 for (unsigned i = 0; i < pCreateInfo->poolSizeCount; ++i) {
382 if (pCreateInfo->pPoolSizes[i].type != VK_DESCRIPTOR_TYPE_SAMPLER)
383 bo_count += pCreateInfo->pPoolSizes[i].descriptorCount;
384
385 switch(pCreateInfo->pPoolSizes[i].type) {
386 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
387 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
388 range_count += pCreateInfo->pPoolSizes[i].descriptorCount;
389 break;
390 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
391 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
392 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
393 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
394 case VK_DESCRIPTOR_TYPE_SAMPLER:
395 /* 32 as we may need to align for images */
396 bo_size += 32 * pCreateInfo->pPoolSizes[i].descriptorCount;
397 break;
398 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
399 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
400 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
401 bo_size += 64 * pCreateInfo->pPoolSizes[i].descriptorCount;
402 break;
403 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
404 bo_size += 96 * pCreateInfo->pPoolSizes[i].descriptorCount;
405 break;
406 default:
407 unreachable("unknown descriptor type\n");
408 break;
409 }
410 }
411
412 if (!(pCreateInfo->flags & VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT)) {
413 uint64_t host_size = pCreateInfo->maxSets * sizeof(struct radv_descriptor_set);
414 host_size += sizeof(struct radeon_winsys_bo*) * bo_count;
415 host_size += sizeof(struct radv_descriptor_range) * range_count;
416 size += host_size;
417 }
418
419 pool = vk_alloc2(&device->alloc, pAllocator, size, 8,
420 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
421 if (!pool)
422 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
423
424 memset(pool, 0, sizeof(*pool));
425
426 if (!(pCreateInfo->flags & VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT)) {
427 pool->host_memory_base = (uint8_t*)pool + sizeof(struct radv_descriptor_pool);
428 pool->host_memory_ptr = pool->host_memory_base;
429 pool->host_memory_end = (uint8_t*)pool + size;
430 }
431
432 if (bo_size) {
433 pool->bo = device->ws->buffer_create(device->ws, bo_size,
434 32, RADEON_DOMAIN_VRAM, 0);
435 pool->mapped_ptr = (uint8_t*)device->ws->buffer_map(pool->bo);
436 }
437 pool->size = bo_size;
438
439 list_inithead(&pool->vram_list);
440 *pDescriptorPool = radv_descriptor_pool_to_handle(pool);
441 return VK_SUCCESS;
442 }
443
444 void radv_DestroyDescriptorPool(
445 VkDevice _device,
446 VkDescriptorPool _pool,
447 const VkAllocationCallbacks* pAllocator)
448 {
449 RADV_FROM_HANDLE(radv_device, device, _device);
450 RADV_FROM_HANDLE(radv_descriptor_pool, pool, _pool);
451
452 if (!pool)
453 return;
454
455 if (!pool->host_memory_base) {
456 list_for_each_entry_safe(struct radv_descriptor_set, set,
457 &pool->vram_list, vram_list) {
458 radv_descriptor_set_destroy(device, pool, set, false);
459 }
460 }
461
462 if (pool->bo)
463 device->ws->buffer_destroy(pool->bo);
464 vk_free2(&device->alloc, pAllocator, pool);
465 }
466
467 VkResult radv_ResetDescriptorPool(
468 VkDevice _device,
469 VkDescriptorPool descriptorPool,
470 VkDescriptorPoolResetFlags flags)
471 {
472 RADV_FROM_HANDLE(radv_device, device, _device);
473 RADV_FROM_HANDLE(radv_descriptor_pool, pool, descriptorPool);
474
475 if (!pool->host_memory_base) {
476 list_for_each_entry_safe(struct radv_descriptor_set, set,
477 &pool->vram_list, vram_list) {
478 radv_descriptor_set_destroy(device, pool, set, false);
479 }
480 }
481
482 list_inithead(&pool->vram_list);
483
484 pool->current_offset = 0;
485 pool->host_memory_ptr = pool->host_memory_base;
486
487 return VK_SUCCESS;
488 }
489
490 VkResult radv_AllocateDescriptorSets(
491 VkDevice _device,
492 const VkDescriptorSetAllocateInfo* pAllocateInfo,
493 VkDescriptorSet* pDescriptorSets)
494 {
495 RADV_FROM_HANDLE(radv_device, device, _device);
496 RADV_FROM_HANDLE(radv_descriptor_pool, pool, pAllocateInfo->descriptorPool);
497
498 VkResult result = VK_SUCCESS;
499 uint32_t i;
500 struct radv_descriptor_set *set;
501
502 /* allocate a set of buffers for each shader to contain descriptors */
503 for (i = 0; i < pAllocateInfo->descriptorSetCount; i++) {
504 RADV_FROM_HANDLE(radv_descriptor_set_layout, layout,
505 pAllocateInfo->pSetLayouts[i]);
506
507 assert(!(layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
508
509 result = radv_descriptor_set_create(device, pool, layout, &set);
510 if (result != VK_SUCCESS)
511 break;
512
513 pDescriptorSets[i] = radv_descriptor_set_to_handle(set);
514 }
515
516 if (result != VK_SUCCESS)
517 radv_FreeDescriptorSets(_device, pAllocateInfo->descriptorPool,
518 i, pDescriptorSets);
519 return result;
520 }
521
522 VkResult radv_FreeDescriptorSets(
523 VkDevice _device,
524 VkDescriptorPool descriptorPool,
525 uint32_t count,
526 const VkDescriptorSet* pDescriptorSets)
527 {
528 RADV_FROM_HANDLE(radv_device, device, _device);
529 RADV_FROM_HANDLE(radv_descriptor_pool, pool, descriptorPool);
530
531 for (uint32_t i = 0; i < count; i++) {
532 RADV_FROM_HANDLE(radv_descriptor_set, set, pDescriptorSets[i]);
533
534 if (set && !pool->host_memory_base)
535 radv_descriptor_set_destroy(device, pool, set, true);
536 }
537 return VK_SUCCESS;
538 }
539
540 static void write_texel_buffer_descriptor(struct radv_device *device,
541 struct radv_cmd_buffer *cmd_buffer,
542 unsigned *dst,
543 struct radeon_winsys_bo **buffer_list,
544 const VkBufferView _buffer_view)
545 {
546 RADV_FROM_HANDLE(radv_buffer_view, buffer_view, _buffer_view);
547
548 memcpy(dst, buffer_view->state, 4 * 4);
549
550 if (cmd_buffer)
551 device->ws->cs_add_buffer(cmd_buffer->cs, buffer_view->bo, 7);
552 else
553 *buffer_list = buffer_view->bo;
554 }
555
556 static void write_buffer_descriptor(struct radv_device *device,
557 struct radv_cmd_buffer *cmd_buffer,
558 unsigned *dst,
559 struct radeon_winsys_bo **buffer_list,
560 const VkDescriptorBufferInfo *buffer_info)
561 {
562 RADV_FROM_HANDLE(radv_buffer, buffer, buffer_info->buffer);
563 uint64_t va = radv_buffer_get_va(buffer->bo);
564 uint32_t range = buffer_info->range;
565
566 if (buffer_info->range == VK_WHOLE_SIZE)
567 range = buffer->size - buffer_info->offset;
568
569 va += buffer_info->offset + buffer->offset;
570 dst[0] = va;
571 dst[1] = S_008F04_BASE_ADDRESS_HI(va >> 32);
572 dst[2] = range;
573 dst[3] = S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X) |
574 S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y) |
575 S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z) |
576 S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W) |
577 S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT) |
578 S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32);
579
580 if (cmd_buffer)
581 device->ws->cs_add_buffer(cmd_buffer->cs, buffer->bo, 7);
582 else
583 *buffer_list = buffer->bo;
584 }
585
586 static void write_dynamic_buffer_descriptor(struct radv_device *device,
587 struct radv_descriptor_range *range,
588 struct radeon_winsys_bo **buffer_list,
589 const VkDescriptorBufferInfo *buffer_info)
590 {
591 RADV_FROM_HANDLE(radv_buffer, buffer, buffer_info->buffer);
592 uint64_t va = radv_buffer_get_va(buffer->bo);
593 unsigned size = buffer_info->range;
594
595 if (buffer_info->range == VK_WHOLE_SIZE)
596 size = buffer->size - buffer_info->offset;
597
598 va += buffer_info->offset + buffer->offset;
599 range->va = va;
600 range->size = size;
601
602 *buffer_list = buffer->bo;
603 }
604
605 static void
606 write_image_descriptor(struct radv_device *device,
607 struct radv_cmd_buffer *cmd_buffer,
608 unsigned *dst,
609 struct radeon_winsys_bo **buffer_list,
610 VkDescriptorType descriptor_type,
611 const VkDescriptorImageInfo *image_info)
612 {
613 RADV_FROM_HANDLE(radv_image_view, iview, image_info->imageView);
614
615 if (descriptor_type == VK_DESCRIPTOR_TYPE_STORAGE_IMAGE) {
616 memcpy(dst, iview->storage_descriptor, 8 * 4);
617 memcpy(dst + 8, iview->storage_fmask_descriptor, 8 * 4);
618 } else {
619 memcpy(dst, iview->descriptor, 8 * 4);
620 memcpy(dst + 8, iview->fmask_descriptor, 8 * 4);
621 }
622
623 if (cmd_buffer)
624 device->ws->cs_add_buffer(cmd_buffer->cs, iview->bo, 7);
625 else
626 *buffer_list = iview->bo;
627 }
628
629 static void
630 write_combined_image_sampler_descriptor(struct radv_device *device,
631 struct radv_cmd_buffer *cmd_buffer,
632 unsigned *dst,
633 struct radeon_winsys_bo **buffer_list,
634 VkDescriptorType descriptor_type,
635 const VkDescriptorImageInfo *image_info,
636 bool has_sampler)
637 {
638 RADV_FROM_HANDLE(radv_sampler, sampler, image_info->sampler);
639
640 write_image_descriptor(device, cmd_buffer, dst, buffer_list, descriptor_type, image_info);
641 /* copy over sampler state */
642 if (has_sampler)
643 memcpy(dst + 16, sampler->state, 16);
644 }
645
646 static void
647 write_sampler_descriptor(struct radv_device *device,
648 unsigned *dst,
649 const VkDescriptorImageInfo *image_info)
650 {
651 RADV_FROM_HANDLE(radv_sampler, sampler, image_info->sampler);
652
653 memcpy(dst, sampler->state, 16);
654 }
655
656 void radv_update_descriptor_sets(
657 struct radv_device* device,
658 struct radv_cmd_buffer* cmd_buffer,
659 VkDescriptorSet dstSetOverride,
660 uint32_t descriptorWriteCount,
661 const VkWriteDescriptorSet* pDescriptorWrites,
662 uint32_t descriptorCopyCount,
663 const VkCopyDescriptorSet* pDescriptorCopies)
664 {
665 uint32_t i, j;
666 for (i = 0; i < descriptorWriteCount; i++) {
667 const VkWriteDescriptorSet *writeset = &pDescriptorWrites[i];
668 RADV_FROM_HANDLE(radv_descriptor_set, set,
669 dstSetOverride ? dstSetOverride : writeset->dstSet);
670 const struct radv_descriptor_set_binding_layout *binding_layout =
671 set->layout->binding + writeset->dstBinding;
672 uint32_t *ptr = set->mapped_ptr;
673 struct radeon_winsys_bo **buffer_list = set->descriptors;
674 /* Immutable samplers are not copied into push descriptors when they are
675 * allocated, so if we are writing push descriptors we have to copy the
676 * immutable samplers into them now.
677 */
678 const bool copy_immutable_samplers = cmd_buffer &&
679 binding_layout->immutable_samplers_offset && !binding_layout->immutable_samplers_equal;
680 const uint32_t *samplers = radv_immutable_samplers(set->layout, binding_layout);
681
682 ptr += binding_layout->offset / 4;
683 ptr += binding_layout->size * writeset->dstArrayElement / 4;
684 buffer_list += binding_layout->buffer_offset;
685 buffer_list += writeset->dstArrayElement;
686 for (j = 0; j < writeset->descriptorCount; ++j) {
687 switch(writeset->descriptorType) {
688 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
689 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: {
690 unsigned idx = writeset->dstArrayElement + j;
691 idx += binding_layout->dynamic_offset_offset;
692 assert(!(set->layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
693 write_dynamic_buffer_descriptor(device, set->dynamic_descriptors + idx,
694 buffer_list, writeset->pBufferInfo + j);
695 break;
696 }
697 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
698 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
699 write_buffer_descriptor(device, cmd_buffer, ptr, buffer_list,
700 writeset->pBufferInfo + j);
701 break;
702 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
703 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
704 write_texel_buffer_descriptor(device, cmd_buffer, ptr, buffer_list,
705 writeset->pTexelBufferView[j]);
706 break;
707 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
708 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
709 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
710 write_image_descriptor(device, cmd_buffer, ptr, buffer_list,
711 writeset->descriptorType,
712 writeset->pImageInfo + j);
713 break;
714 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
715 write_combined_image_sampler_descriptor(device, cmd_buffer, ptr, buffer_list,
716 writeset->descriptorType,
717 writeset->pImageInfo + j,
718 !binding_layout->immutable_samplers_offset);
719 if (copy_immutable_samplers) {
720 const unsigned idx = writeset->dstArrayElement + j;
721 memcpy(ptr + 16, samplers + 4 * idx, 16);
722 }
723 break;
724 case VK_DESCRIPTOR_TYPE_SAMPLER:
725 if (!binding_layout->immutable_samplers_offset) {
726 write_sampler_descriptor(device, ptr,
727 writeset->pImageInfo + j);
728 } else if (copy_immutable_samplers) {
729 unsigned idx = writeset->dstArrayElement + j;
730 memcpy(ptr, samplers + 4 * idx, 16);
731 }
732 break;
733 default:
734 unreachable("unimplemented descriptor type");
735 break;
736 }
737 ptr += binding_layout->size / 4;
738 ++buffer_list;
739 }
740
741 }
742 if (descriptorCopyCount)
743 radv_finishme("copy descriptors");
744 }
745
746 void radv_UpdateDescriptorSets(
747 VkDevice _device,
748 uint32_t descriptorWriteCount,
749 const VkWriteDescriptorSet* pDescriptorWrites,
750 uint32_t descriptorCopyCount,
751 const VkCopyDescriptorSet* pDescriptorCopies)
752 {
753 RADV_FROM_HANDLE(radv_device, device, _device);
754
755 radv_update_descriptor_sets(device, NULL, VK_NULL_HANDLE, descriptorWriteCount, pDescriptorWrites,
756 descriptorCopyCount, pDescriptorCopies);
757 }
758
759 VkResult radv_CreateDescriptorUpdateTemplateKHR(VkDevice _device,
760 const VkDescriptorUpdateTemplateCreateInfoKHR *pCreateInfo,
761 const VkAllocationCallbacks *pAllocator,
762 VkDescriptorUpdateTemplateKHR *pDescriptorUpdateTemplate)
763 {
764 RADV_FROM_HANDLE(radv_device, device, _device);
765 RADV_FROM_HANDLE(radv_descriptor_set_layout, set_layout, pCreateInfo->descriptorSetLayout);
766 const uint32_t entry_count = pCreateInfo->descriptorUpdateEntryCount;
767 const size_t size = sizeof(struct radv_descriptor_update_template) +
768 sizeof(struct radv_descriptor_update_template_entry) * entry_count;
769 struct radv_descriptor_update_template *templ;
770 uint32_t i;
771
772 templ = vk_alloc2(&device->alloc, pAllocator, size, 8, VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
773 if (!templ)
774 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
775
776 templ->entry_count = entry_count;
777
778 for (i = 0; i < entry_count; i++) {
779 const VkDescriptorUpdateTemplateEntryKHR *entry = &pCreateInfo->pDescriptorUpdateEntries[i];
780 const struct radv_descriptor_set_binding_layout *binding_layout =
781 set_layout->binding + entry->dstBinding;
782 const uint32_t buffer_offset = binding_layout->buffer_offset + entry->dstArrayElement;
783 const uint32_t *immutable_samplers = NULL;
784 uint32_t dst_offset;
785 uint32_t dst_stride;
786
787 /* dst_offset is an offset into dynamic_descriptors when the descriptor
788 is dynamic, and an offset into mapped_ptr otherwise */
789 switch (entry->descriptorType) {
790 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
791 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
792 assert(pCreateInfo->templateType == VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_DESCRIPTOR_SET_KHR);
793 dst_offset = binding_layout->dynamic_offset_offset + entry->dstArrayElement;
794 dst_stride = 0; /* Not used */
795 break;
796 default:
797 switch (entry->descriptorType) {
798 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
799 case VK_DESCRIPTOR_TYPE_SAMPLER:
800 /* Immutable samplers are copied into push descriptors when they are pushed */
801 if (pCreateInfo->templateType == VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_PUSH_DESCRIPTORS_KHR &&
802 binding_layout->immutable_samplers_offset && !binding_layout->immutable_samplers_equal) {
803 immutable_samplers = radv_immutable_samplers(set_layout, binding_layout) + entry->dstArrayElement * 4;
804 }
805 break;
806 default:
807 break;
808 }
809 dst_offset = binding_layout->offset / 4 + binding_layout->size * entry->dstArrayElement / 4;
810 dst_stride = binding_layout->size / 4;
811 break;
812 }
813
814 templ->entry[i] = (struct radv_descriptor_update_template_entry) {
815 .descriptor_type = entry->descriptorType,
816 .descriptor_count = entry->descriptorCount,
817 .src_offset = entry->offset,
818 .src_stride = entry->stride,
819 .dst_offset = dst_offset,
820 .dst_stride = dst_stride,
821 .buffer_offset = buffer_offset,
822 .has_sampler = !binding_layout->immutable_samplers_offset,
823 .immutable_samplers = immutable_samplers
824 };
825 }
826
827 *pDescriptorUpdateTemplate = radv_descriptor_update_template_to_handle(templ);
828 return VK_SUCCESS;
829 }
830
831 void radv_DestroyDescriptorUpdateTemplateKHR(VkDevice _device,
832 VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate,
833 const VkAllocationCallbacks *pAllocator)
834 {
835 RADV_FROM_HANDLE(radv_device, device, _device);
836 RADV_FROM_HANDLE(radv_descriptor_update_template, templ, descriptorUpdateTemplate);
837
838 if (!templ)
839 return;
840
841 vk_free2(&device->alloc, pAllocator, templ);
842 }
843
844 void radv_update_descriptor_set_with_template(struct radv_device *device,
845 struct radv_cmd_buffer *cmd_buffer,
846 struct radv_descriptor_set *set,
847 VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate,
848 const void *pData)
849 {
850 RADV_FROM_HANDLE(radv_descriptor_update_template, templ, descriptorUpdateTemplate);
851 uint32_t i;
852
853 for (i = 0; i < templ->entry_count; ++i) {
854 struct radeon_winsys_bo **buffer_list = set->descriptors + templ->entry[i].buffer_offset;
855 uint32_t *pDst = set->mapped_ptr + templ->entry[i].dst_offset;
856 const uint8_t *pSrc = ((const uint8_t *) pData) + templ->entry[i].src_offset;
857 uint32_t j;
858
859 for (j = 0; j < templ->entry[i].descriptor_count; ++j) {
860 switch (templ->entry[i].descriptor_type) {
861 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
862 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: {
863 const unsigned idx = templ->entry[i].dst_offset + j;
864 assert(!(set->layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
865 write_dynamic_buffer_descriptor(device, set->dynamic_descriptors + idx,
866 buffer_list, (struct VkDescriptorBufferInfo *) pSrc);
867 break;
868 }
869 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
870 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
871 write_buffer_descriptor(device, cmd_buffer, pDst, buffer_list,
872 (struct VkDescriptorBufferInfo *) pSrc);
873 break;
874 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
875 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
876 write_texel_buffer_descriptor(device, cmd_buffer, pDst, buffer_list,
877 *(VkBufferView *) pSrc);
878 break;
879 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
880 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
881 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
882 write_image_descriptor(device, cmd_buffer, pDst, buffer_list,
883 templ->entry[i].descriptor_type,
884 (struct VkDescriptorImageInfo *) pSrc);
885 break;
886 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
887 write_combined_image_sampler_descriptor(device, cmd_buffer, pDst, buffer_list,
888 templ->entry[i].descriptor_type,
889 (struct VkDescriptorImageInfo *) pSrc,
890 templ->entry[i].has_sampler);
891 if (templ->entry[i].immutable_samplers)
892 memcpy(pDst + 16, templ->entry[i].immutable_samplers + 4 * j, 16);
893 break;
894 case VK_DESCRIPTOR_TYPE_SAMPLER:
895 if (templ->entry[i].has_sampler)
896 write_sampler_descriptor(device, pDst,
897 (struct VkDescriptorImageInfo *) pSrc);
898 else if (templ->entry[i].immutable_samplers)
899 memcpy(pDst, templ->entry[i].immutable_samplers + 4 * j, 16);
900 break;
901 default:
902 unreachable("unimplemented descriptor type");
903 break;
904 }
905 pSrc += templ->entry[i].src_stride;
906 pDst += templ->entry[i].dst_stride;
907 ++buffer_list;
908 }
909 }
910 }
911
912 void radv_UpdateDescriptorSetWithTemplateKHR(VkDevice _device,
913 VkDescriptorSet descriptorSet,
914 VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate,
915 const void *pData)
916 {
917 RADV_FROM_HANDLE(radv_device, device, _device);
918 RADV_FROM_HANDLE(radv_descriptor_set, set, descriptorSet);
919
920 radv_update_descriptor_set_with_template(device, NULL, set, descriptorUpdateTemplate, pData);
921 }