projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
radv: make sizes & offsets 32 bit in radv_descriptor_update_template_entry.
[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 struct radv_cmd_buffer *cmd_buffer,
261 const struct radv_descriptor_set_layout *layout,
262 struct radv_descriptor_set **out_set)
263 {
264 struct radv_descriptor_set *set;
265 unsigned mem_size = sizeof(struct radv_descriptor_set) +
266 sizeof(struct radeon_winsys_bo *) * layout->buffer_count;
267 set = vk_alloc2(&device->alloc, NULL, mem_size, 8,
268 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
269
270 if (!set)
271 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
272
273 memset(set, 0, mem_size);
274
275 if (layout->dynamic_offset_count) {
276 unsigned size = sizeof(struct radv_descriptor_range) *
277 layout->dynamic_offset_count;
278 set->dynamic_descriptors = vk_alloc2(&device->alloc, NULL, size, 8,
279 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
280
281 if (!set->dynamic_descriptors) {
282 vk_free2(&device->alloc, NULL, set);
283 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
284 }
285 }
286
287 set->layout = layout;
288 if (layout->size) {
289 uint32_t layout_size = align_u32(layout->size, 32);
290 set->size = layout->size;
291 if (!cmd_buffer) {
292 /* try to allocate linearly first, so that we don't spend
293 * time looking for gaps if the app only allocates &
294 * resets via the pool. */
295 if (pool->current_offset + layout_size <= pool->size) {
296 set->bo = pool->bo;
297 set->mapped_ptr = (uint32_t*)(pool->mapped_ptr + pool->current_offset);
298 set->va = device->ws->buffer_get_va(set->bo) + pool->current_offset;
299 pool->current_offset += layout_size;
300 list_addtail(&set->vram_list, &pool->vram_list);
301 } else {
302 uint64_t offset = 0;
303 struct list_head *prev = &pool->vram_list;
304 struct radv_descriptor_set *cur;
305 LIST_FOR_EACH_ENTRY(cur, &pool->vram_list, vram_list) {
306 uint64_t start = (uint8_t*)cur->mapped_ptr - pool->mapped_ptr;
307 if (start - offset >= layout_size)
308 break;
309
310 offset = start + cur->size;
311 prev = &cur->vram_list;
312 }
313
314 if (pool->size - offset < layout_size) {
315 vk_free2(&device->alloc, NULL, set->dynamic_descriptors);
316 vk_free2(&device->alloc, NULL, set);
317 return vk_error(VK_ERROR_OUT_OF_POOL_MEMORY_KHR);
318 }
319 set->bo = pool->bo;
320 set->mapped_ptr = (uint32_t*)(pool->mapped_ptr + offset);
321 set->va = device->ws->buffer_get_va(set->bo) + offset;
322 list_add(&set->vram_list, prev);
323 }
324 } else {
325 unsigned bo_offset;
326 if (!radv_cmd_buffer_upload_alloc(cmd_buffer, set->size, 32,
327 &bo_offset,
328 (void**)&set->mapped_ptr)) {
329 vk_free2(&device->alloc, NULL, set->dynamic_descriptors);
330 vk_free2(&device->alloc, NULL, set);
331 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
332 }
333
334 set->va = device->ws->buffer_get_va(cmd_buffer->upload.upload_bo);
335 set->va += bo_offset;
336 }
337 }
338
339 for (unsigned i = 0; i < layout->binding_count; ++i) {
340 if (!layout->binding[i].immutable_samplers_offset ||
341 layout->binding[i].immutable_samplers_equal)
342 continue;
343
344 unsigned offset = layout->binding[i].offset / 4;
345 if (layout->binding[i].type == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER)
346 offset += 16;
347
348 const uint32_t *samplers = (const uint32_t*)((const char*)layout + layout->binding[i].immutable_samplers_offset);
349 for (unsigned j = 0; j < layout->binding[i].array_size; ++j) {
350 memcpy(set->mapped_ptr + offset, samplers + 4 * j, 16);
351 offset += layout->binding[i].size / 4;
352 }
353
354 }
355 *out_set = set;
356 return VK_SUCCESS;
357 }
358
359 static void
360 radv_descriptor_set_destroy(struct radv_device *device,
361 struct radv_descriptor_pool *pool,
362 struct radv_descriptor_set *set,
363 bool free_bo)
364 {
365 if (free_bo && set->size)
366 list_del(&set->vram_list);
367 if (set->dynamic_descriptors)
368 vk_free2(&device->alloc, NULL, set->dynamic_descriptors);
369 vk_free2(&device->alloc, NULL, set);
370 }
371
372 VkResult
373 radv_temp_descriptor_set_create(struct radv_device *device,
374 struct radv_cmd_buffer *cmd_buffer,
375 VkDescriptorSetLayout _layout,
376 VkDescriptorSet *_set)
377 {
378 RADV_FROM_HANDLE(radv_descriptor_set_layout, layout, _layout);
379 struct radv_descriptor_set *set = NULL;
380 VkResult ret;
381
382 ret = radv_descriptor_set_create(device, NULL, cmd_buffer, layout, &set);
383 *_set = radv_descriptor_set_to_handle(set);
384 return ret;
385 }
386
387 void
388 radv_temp_descriptor_set_destroy(struct radv_device *device,
389 VkDescriptorSet _set)
390 {
391 RADV_FROM_HANDLE(radv_descriptor_set, set, _set);
392
393 radv_descriptor_set_destroy(device, NULL, set, false);
394 }
395
396 VkResult radv_CreateDescriptorPool(
397 VkDevice _device,
398 const VkDescriptorPoolCreateInfo* pCreateInfo,
399 const VkAllocationCallbacks* pAllocator,
400 VkDescriptorPool* pDescriptorPool)
401 {
402 RADV_FROM_HANDLE(radv_device, device, _device);
403 struct radv_descriptor_pool *pool;
404 int size = sizeof(struct radv_descriptor_pool);
405 uint64_t bo_size = 0;
406 pool = vk_alloc2(&device->alloc, pAllocator, size, 8,
407 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
408 if (!pool)
409 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
410
411 memset(pool, 0, sizeof(*pool));
412
413 for (unsigned i = 0; i < pCreateInfo->poolSizeCount; ++i) {
414 switch(pCreateInfo->pPoolSizes[i].type) {
415 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
416 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
417 break;
418 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
419 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
420 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
421 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
422 case VK_DESCRIPTOR_TYPE_SAMPLER:
423 /* 32 as we may need to align for images */
424 bo_size += 32 * pCreateInfo->pPoolSizes[i].descriptorCount;
425 break;
426 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
427 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
428 case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
429 bo_size += 64 * pCreateInfo->pPoolSizes[i].descriptorCount;
430 break;
431 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
432 bo_size += 96 * pCreateInfo->pPoolSizes[i].descriptorCount;
433 break;
434 default:
435 unreachable("unknown descriptor type\n");
436 break;
437 }
438 }
439
440 if (bo_size) {
441 pool->bo = device->ws->buffer_create(device->ws, bo_size,
442 32, RADEON_DOMAIN_VRAM, 0);
443 pool->mapped_ptr = (uint8_t*)device->ws->buffer_map(pool->bo);
444 }
445 pool->size = bo_size;
446
447 list_inithead(&pool->vram_list);
448 *pDescriptorPool = radv_descriptor_pool_to_handle(pool);
449 return VK_SUCCESS;
450 }
451
452 void radv_DestroyDescriptorPool(
453 VkDevice _device,
454 VkDescriptorPool _pool,
455 const VkAllocationCallbacks* pAllocator)
456 {
457 RADV_FROM_HANDLE(radv_device, device, _device);
458 RADV_FROM_HANDLE(radv_descriptor_pool, pool, _pool);
459
460 if (!pool)
461 return;
462
463 list_for_each_entry_safe(struct radv_descriptor_set, set,
464 &pool->vram_list, vram_list) {
465 radv_descriptor_set_destroy(device, pool, set, false);
466 }
467
468 if (pool->bo)
469 device->ws->buffer_destroy(pool->bo);
470 vk_free2(&device->alloc, pAllocator, pool);
471 }
472
473 VkResult radv_ResetDescriptorPool(
474 VkDevice _device,
475 VkDescriptorPool descriptorPool,
476 VkDescriptorPoolResetFlags flags)
477 {
478 RADV_FROM_HANDLE(radv_device, device, _device);
479 RADV_FROM_HANDLE(radv_descriptor_pool, pool, descriptorPool);
480
481 list_for_each_entry_safe(struct radv_descriptor_set, set,
482 &pool->vram_list, vram_list) {
483 radv_descriptor_set_destroy(device, pool, set, false);
484 }
485
486 list_inithead(&pool->vram_list);
487
488 pool->current_offset = 0;
489
490 return VK_SUCCESS;
491 }
492
493 VkResult radv_AllocateDescriptorSets(
494 VkDevice _device,
495 const VkDescriptorSetAllocateInfo* pAllocateInfo,
496 VkDescriptorSet* pDescriptorSets)
497 {
498 RADV_FROM_HANDLE(radv_device, device, _device);
499 RADV_FROM_HANDLE(radv_descriptor_pool, pool, pAllocateInfo->descriptorPool);
500
501 VkResult result = VK_SUCCESS;
502 uint32_t i;
503 struct radv_descriptor_set *set;
504
505 /* allocate a set of buffers for each shader to contain descriptors */
506 for (i = 0; i < pAllocateInfo->descriptorSetCount; i++) {
507 RADV_FROM_HANDLE(radv_descriptor_set_layout, layout,
508 pAllocateInfo->pSetLayouts[i]);
509
510 assert(!(layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
511
512 result = radv_descriptor_set_create(device, pool, NULL, layout, &set);
513 if (result != VK_SUCCESS)
514 break;
515
516 pDescriptorSets[i] = radv_descriptor_set_to_handle(set);
517 }
518
519 if (result != VK_SUCCESS)
520 radv_FreeDescriptorSets(_device, pAllocateInfo->descriptorPool,
521 i, pDescriptorSets);
522 return result;
523 }
524
525 VkResult radv_FreeDescriptorSets(
526 VkDevice _device,
527 VkDescriptorPool descriptorPool,
528 uint32_t count,
529 const VkDescriptorSet* pDescriptorSets)
530 {
531 RADV_FROM_HANDLE(radv_device, device, _device);
532 RADV_FROM_HANDLE(radv_descriptor_pool, pool, descriptorPool);
533
534 for (uint32_t i = 0; i < count; i++) {
535 RADV_FROM_HANDLE(radv_descriptor_set, set, pDescriptorSets[i]);
536
537 if (set)
538 radv_descriptor_set_destroy(device, pool, set, true);
539 }
540 return VK_SUCCESS;
541 }
542
543 static void write_texel_buffer_descriptor(struct radv_device *device,
544 struct radv_cmd_buffer *cmd_buffer,
545 unsigned *dst,
546 struct radeon_winsys_bo **buffer_list,
547 const VkBufferView _buffer_view)
548 {
549 RADV_FROM_HANDLE(radv_buffer_view, buffer_view, _buffer_view);
550
551 memcpy(dst, buffer_view->state, 4 * 4);
552
553 if (cmd_buffer)
554 device->ws->cs_add_buffer(cmd_buffer->cs, buffer_view->bo, 7);
555 else
556 *buffer_list = buffer_view->bo;
557 }
558
559 static void write_buffer_descriptor(struct radv_device *device,
560 struct radv_cmd_buffer *cmd_buffer,
561 unsigned *dst,
562 struct radeon_winsys_bo **buffer_list,
563 const VkDescriptorBufferInfo *buffer_info)
564 {
565 RADV_FROM_HANDLE(radv_buffer, buffer, buffer_info->buffer);
566 uint64_t va = device->ws->buffer_get_va(buffer->bo);
567 uint32_t range = buffer_info->range;
568
569 if (buffer_info->range == VK_WHOLE_SIZE)
570 range = buffer->size - buffer_info->offset;
571
572 va += buffer_info->offset + buffer->offset;
573 dst[0] = va;
574 dst[1] = S_008F04_BASE_ADDRESS_HI(va >> 32);
575 dst[2] = range;
576 dst[3] = S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X) |
577 S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y) |
578 S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z) |
579 S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W) |
580 S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT) |
581 S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32);
582
583 if (cmd_buffer)
584 device->ws->cs_add_buffer(cmd_buffer->cs, buffer->bo, 7);
585 else
586 *buffer_list = buffer->bo;
587 }
588
589 static void write_dynamic_buffer_descriptor(struct radv_device *device,
590 struct radv_descriptor_range *range,
591 struct radeon_winsys_bo **buffer_list,
592 const VkDescriptorBufferInfo *buffer_info)
593 {
594 RADV_FROM_HANDLE(radv_buffer, buffer, buffer_info->buffer);
595 uint64_t va = device->ws->buffer_get_va(buffer->bo);
596 unsigned size = buffer_info->range;
597
598 if (buffer_info->range == VK_WHOLE_SIZE)
599 size = buffer->size - buffer_info->offset;
600
601 va += buffer_info->offset + buffer->offset;
602 range->va = va;
603 range->size = size;
604
605 *buffer_list = buffer->bo;
606 }
607
608 static void
609 write_image_descriptor(struct radv_device *device,
610 struct radv_cmd_buffer *cmd_buffer,
611 unsigned *dst,
612 struct radeon_winsys_bo **buffer_list,
613 const VkDescriptorImageInfo *image_info)
614 {
615 RADV_FROM_HANDLE(radv_image_view, iview, image_info->imageView);
616 memcpy(dst, iview->descriptor, 8 * 4);
617 memcpy(dst + 8, iview->fmask_descriptor, 8 * 4);
618
619 if (cmd_buffer)
620 device->ws->cs_add_buffer(cmd_buffer->cs, iview->bo, 7);
621 else
622 *buffer_list = iview->bo;
623 }
624
625 static void
626 write_combined_image_sampler_descriptor(struct radv_device *device,
627 struct radv_cmd_buffer *cmd_buffer,
628 unsigned *dst,
629 struct radeon_winsys_bo **buffer_list,
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, 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 += binding_layout->buffer_count * 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->pImageInfo + j);
707 break;
708 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
709 write_combined_image_sampler_descriptor(device, cmd_buffer, ptr, buffer_list,
710 writeset->pImageInfo + j,
711 !binding_layout->immutable_samplers_offset);
712 if (copy_immutable_samplers) {
713 const unsigned idx = writeset->dstArrayElement + j;
714 memcpy(ptr + 16, samplers + 4 * idx, 16);
715 }
716 break;
717 case VK_DESCRIPTOR_TYPE_SAMPLER:
718 if (!binding_layout->immutable_samplers_offset) {
719 write_sampler_descriptor(device, ptr,
720 writeset->pImageInfo + j);
721 } else if (copy_immutable_samplers) {
722 unsigned idx = writeset->dstArrayElement + j;
723 memcpy(ptr, samplers + 4 * idx, 16);
724 }
725 break;
726 default:
727 unreachable("unimplemented descriptor type");
728 break;
729 }
730 ptr += binding_layout->size / 4;
731 buffer_list += binding_layout->buffer_count;
732 }
733
734 }
735 if (descriptorCopyCount)
736 radv_finishme("copy descriptors");
737 }
738
739 void radv_UpdateDescriptorSets(
740 VkDevice _device,
741 uint32_t descriptorWriteCount,
742 const VkWriteDescriptorSet* pDescriptorWrites,
743 uint32_t descriptorCopyCount,
744 const VkCopyDescriptorSet* pDescriptorCopies)
745 {
746 RADV_FROM_HANDLE(radv_device, device, _device);
747
748 radv_update_descriptor_sets(device, NULL, VK_NULL_HANDLE, descriptorWriteCount, pDescriptorWrites,
749 descriptorCopyCount, pDescriptorCopies);
750 }
751
752 VkResult radv_CreateDescriptorUpdateTemplateKHR(VkDevice _device,
753 const VkDescriptorUpdateTemplateCreateInfoKHR *pCreateInfo,
754 const VkAllocationCallbacks *pAllocator,
755 VkDescriptorUpdateTemplateKHR *pDescriptorUpdateTemplate)
756 {
757 RADV_FROM_HANDLE(radv_device, device, _device);
758 RADV_FROM_HANDLE(radv_descriptor_set_layout, set_layout, pCreateInfo->descriptorSetLayout);
759 const uint32_t entry_count = pCreateInfo->descriptorUpdateEntryCount;
760 const size_t size = sizeof(struct radv_descriptor_update_template) +
761 sizeof(struct radv_descriptor_update_template_entry) * entry_count;
762 struct radv_descriptor_update_template *templ;
763 uint32_t i;
764
765 templ = vk_alloc2(&device->alloc, pAllocator, size, 8, VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
766 if (!templ)
767 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
768
769 templ->entry_count = entry_count;
770
771 for (i = 0; i < entry_count; i++) {
772 const VkDescriptorUpdateTemplateEntryKHR *entry = &pCreateInfo->pDescriptorUpdateEntries[i];
773 const struct radv_descriptor_set_binding_layout *binding_layout =
774 set_layout->binding + entry->dstBinding;
775 const uint32_t buffer_offset = binding_layout->buffer_offset +
776 binding_layout->buffer_count * entry->dstArrayElement;
777 const uint32_t *immutable_samplers = NULL;
778 uint32_t dst_offset;
779 uint32_t dst_stride;
780
781 /* dst_offset is an offset into dynamic_descriptors when the descriptor
782 is dynamic, and an offset into mapped_ptr otherwise */
783 switch (entry->descriptorType) {
784 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
785 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
786 assert(pCreateInfo->templateType == VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_DESCRIPTOR_SET_KHR);
787 dst_offset = binding_layout->dynamic_offset_offset + entry->dstArrayElement;
788 dst_stride = 0; /* Not used */
789 break;
790 default:
791 switch (entry->descriptorType) {
792 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
793 case VK_DESCRIPTOR_TYPE_SAMPLER:
794 /* Immutable samplers are copied into push descriptors when they are pushed */
795 if (pCreateInfo->templateType == VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_PUSH_DESCRIPTORS_KHR &&
796 binding_layout->immutable_samplers_offset && !binding_layout->immutable_samplers_equal) {
797 immutable_samplers = radv_immutable_samplers(set_layout, binding_layout) + entry->dstArrayElement * 4;
798 }
799 break;
800 default:
801 break;
802 }
803 dst_offset = binding_layout->offset / 4 + binding_layout->size * entry->dstArrayElement / 4;
804 dst_stride = binding_layout->size / 4;
805 break;
806 }
807
808 templ->entry[i] = (struct radv_descriptor_update_template_entry) {
809 .descriptor_type = entry->descriptorType,
810 .descriptor_count = entry->descriptorCount,
811 .src_offset = entry->offset,
812 .src_stride = entry->stride,
813 .dst_offset = dst_offset,
814 .dst_stride = dst_stride,
815 .buffer_offset = buffer_offset,
816 .buffer_count = binding_layout->buffer_count,
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 (struct VkDescriptorImageInfo *) pSrc);
879 break;
880 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
881 write_combined_image_sampler_descriptor(device, cmd_buffer, pDst, buffer_list,
882 (struct VkDescriptorImageInfo *) pSrc,
883 templ->entry[i].has_sampler);
884 if (templ->entry[i].immutable_samplers)
885 memcpy(pDst + 16, templ->entry[i].immutable_samplers + 4 * j, 16);
886 break;
887 case VK_DESCRIPTOR_TYPE_SAMPLER:
888 if (templ->entry[i].has_sampler)
889 write_sampler_descriptor(device, pDst,
890 (struct VkDescriptorImageInfo *) pSrc);
891 else if (templ->entry[i].immutable_samplers)
892 memcpy(pDst, templ->entry[i].immutable_samplers + 4 * j, 16);
893 break;
894 default:
895 unreachable("unimplemented descriptor type");
896 break;
897 }
898 pSrc += templ->entry[i].src_stride;
899 pDst += templ->entry[i].dst_stride;
900 buffer_list += templ->entry[i].buffer_count;
901 }
902 }
903 }
904
905 void radv_UpdateDescriptorSetWithTemplateKHR(VkDevice _device,
906 VkDescriptorSet descriptorSet,
907 VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate,
908 const void *pData)
909 {
910 RADV_FROM_HANDLE(radv_device, device, _device);
911 RADV_FROM_HANDLE(radv_descriptor_set, set, descriptorSet);
912
913 radv_update_descriptor_set_with_template(device, NULL, set, descriptorUpdateTemplate, pData);
914 }