2 * Copyright © 2015 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 #include "util/blob.h"
25 #include "util/hash_table.h"
26 #include "util/debug.h"
27 #include "util/disk_cache.h"
28 #include "util/mesa-sha1.h"
29 #include "nir/nir_serialize.h"
30 #include "anv_private.h"
31 #include "nir/nir_xfb_info.h"
33 struct anv_shader_bin
*
34 anv_shader_bin_create(struct anv_device
*device
,
35 const void *key_data
, uint32_t key_size
,
36 const void *kernel_data
, uint32_t kernel_size
,
37 const void *constant_data
, uint32_t constant_data_size
,
38 const struct brw_stage_prog_data
*prog_data_in
,
39 uint32_t prog_data_size
, const void *prog_data_param_in
,
40 const struct brw_compile_stats
*stats
, uint32_t num_stats
,
41 const nir_xfb_info
*xfb_info_in
,
42 const struct anv_pipeline_bind_map
*bind_map
)
44 struct anv_shader_bin
*shader
;
45 struct anv_shader_bin_key
*key
;
46 struct brw_stage_prog_data
*prog_data
;
47 uint32_t *prog_data_param
;
48 nir_xfb_info
*xfb_info
;
49 struct anv_pipeline_binding
*surface_to_descriptor
, *sampler_to_descriptor
;
52 anv_multialloc_add(&ma
, &shader
, 1);
53 anv_multialloc_add_size(&ma
, &key
, sizeof(*key
) + key_size
);
54 anv_multialloc_add_size(&ma
, &prog_data
, prog_data_size
);
55 anv_multialloc_add(&ma
, &prog_data_param
, prog_data_in
->nr_params
);
57 uint32_t xfb_info_size
= nir_xfb_info_size(xfb_info_in
->output_count
);
58 anv_multialloc_add_size(&ma
, &xfb_info
, xfb_info_size
);
60 anv_multialloc_add(&ma
, &surface_to_descriptor
,
61 bind_map
->surface_count
);
62 anv_multialloc_add(&ma
, &sampler_to_descriptor
,
63 bind_map
->sampler_count
);
65 if (!anv_multialloc_alloc(&ma
, &device
->alloc
,
66 VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
))
72 memcpy(key
->data
, key_data
, key_size
);
76 anv_state_pool_alloc(&device
->instruction_state_pool
, kernel_size
, 64);
77 memcpy(shader
->kernel
.map
, kernel_data
, kernel_size
);
78 shader
->kernel_size
= kernel_size
;
80 if (constant_data_size
) {
81 shader
->constant_data
=
82 anv_state_pool_alloc(&device
->dynamic_state_pool
,
83 constant_data_size
, 32);
84 memcpy(shader
->constant_data
.map
, constant_data
, constant_data_size
);
86 shader
->constant_data
= ANV_STATE_NULL
;
88 shader
->constant_data_size
= constant_data_size
;
90 memcpy(prog_data
, prog_data_in
, prog_data_size
);
91 memcpy(prog_data_param
, prog_data_param_in
,
92 prog_data
->nr_params
* sizeof(*prog_data_param
));
93 prog_data
->param
= prog_data_param
;
94 shader
->prog_data
= prog_data
;
95 shader
->prog_data_size
= prog_data_size
;
97 assert(num_stats
<= ARRAY_SIZE(shader
->stats
));
98 typed_memcpy(shader
->stats
, stats
, num_stats
);
99 shader
->num_stats
= num_stats
;
102 *xfb_info
= *xfb_info_in
;
103 typed_memcpy(xfb_info
->outputs
, xfb_info_in
->outputs
,
104 xfb_info_in
->output_count
);
105 shader
->xfb_info
= xfb_info
;
107 shader
->xfb_info
= NULL
;
110 shader
->bind_map
= *bind_map
;
111 typed_memcpy(surface_to_descriptor
, bind_map
->surface_to_descriptor
,
112 bind_map
->surface_count
);
113 shader
->bind_map
.surface_to_descriptor
= surface_to_descriptor
;
114 typed_memcpy(sampler_to_descriptor
, bind_map
->sampler_to_descriptor
,
115 bind_map
->sampler_count
);
116 shader
->bind_map
.sampler_to_descriptor
= sampler_to_descriptor
;
122 anv_shader_bin_destroy(struct anv_device
*device
,
123 struct anv_shader_bin
*shader
)
125 assert(shader
->ref_cnt
== 0);
126 anv_state_pool_free(&device
->instruction_state_pool
, shader
->kernel
);
127 anv_state_pool_free(&device
->dynamic_state_pool
, shader
->constant_data
);
128 vk_free(&device
->alloc
, shader
);
132 anv_shader_bin_write_to_blob(const struct anv_shader_bin
*shader
,
135 blob_write_uint32(blob
, shader
->key
->size
);
136 blob_write_bytes(blob
, shader
->key
->data
, shader
->key
->size
);
138 blob_write_uint32(blob
, shader
->kernel_size
);
139 blob_write_bytes(blob
, shader
->kernel
.map
, shader
->kernel_size
);
141 blob_write_uint32(blob
, shader
->constant_data_size
);
142 blob_write_bytes(blob
, shader
->constant_data
.map
,
143 shader
->constant_data_size
);
145 blob_write_uint32(blob
, shader
->prog_data_size
);
146 blob_write_bytes(blob
, shader
->prog_data
, shader
->prog_data_size
);
147 blob_write_bytes(blob
, shader
->prog_data
->param
,
148 shader
->prog_data
->nr_params
*
149 sizeof(*shader
->prog_data
->param
));
151 blob_write_uint32(blob
, shader
->num_stats
);
152 blob_write_bytes(blob
, shader
->stats
,
153 shader
->num_stats
* sizeof(shader
->stats
[0]));
155 if (shader
->xfb_info
) {
156 uint32_t xfb_info_size
=
157 nir_xfb_info_size(shader
->xfb_info
->output_count
);
158 blob_write_uint32(blob
, xfb_info_size
);
159 blob_write_bytes(blob
, shader
->xfb_info
, xfb_info_size
);
161 blob_write_uint32(blob
, 0);
164 blob_write_bytes(blob
, shader
->bind_map
.surface_sha1
,
165 sizeof(shader
->bind_map
.surface_sha1
));
166 blob_write_bytes(blob
, shader
->bind_map
.sampler_sha1
,
167 sizeof(shader
->bind_map
.sampler_sha1
));
168 blob_write_bytes(blob
, shader
->bind_map
.push_sha1
,
169 sizeof(shader
->bind_map
.push_sha1
));
170 blob_write_uint32(blob
, shader
->bind_map
.surface_count
);
171 blob_write_uint32(blob
, shader
->bind_map
.sampler_count
);
172 blob_write_bytes(blob
, shader
->bind_map
.surface_to_descriptor
,
173 shader
->bind_map
.surface_count
*
174 sizeof(*shader
->bind_map
.surface_to_descriptor
));
175 blob_write_bytes(blob
, shader
->bind_map
.sampler_to_descriptor
,
176 shader
->bind_map
.sampler_count
*
177 sizeof(*shader
->bind_map
.sampler_to_descriptor
));
178 blob_write_bytes(blob
, shader
->bind_map
.push_ranges
,
179 sizeof(shader
->bind_map
.push_ranges
));
181 return !blob
->out_of_memory
;
184 static struct anv_shader_bin
*
185 anv_shader_bin_create_from_blob(struct anv_device
*device
,
186 struct blob_reader
*blob
)
188 uint32_t key_size
= blob_read_uint32(blob
);
189 const void *key_data
= blob_read_bytes(blob
, key_size
);
191 uint32_t kernel_size
= blob_read_uint32(blob
);
192 const void *kernel_data
= blob_read_bytes(blob
, kernel_size
);
194 uint32_t constant_data_size
= blob_read_uint32(blob
);
195 const void *constant_data
= blob_read_bytes(blob
, constant_data_size
);
197 uint32_t prog_data_size
= blob_read_uint32(blob
);
198 const struct brw_stage_prog_data
*prog_data
=
199 blob_read_bytes(blob
, prog_data_size
);
202 const void *prog_data_param
=
203 blob_read_bytes(blob
, prog_data
->nr_params
* sizeof(*prog_data
->param
));
205 uint32_t num_stats
= blob_read_uint32(blob
);
206 const struct brw_compile_stats
*stats
=
207 blob_read_bytes(blob
, num_stats
* sizeof(stats
[0]));
209 const nir_xfb_info
*xfb_info
= NULL
;
210 uint32_t xfb_size
= blob_read_uint32(blob
);
212 xfb_info
= blob_read_bytes(blob
, xfb_size
);
214 struct anv_pipeline_bind_map bind_map
;
215 blob_copy_bytes(blob
, bind_map
.surface_sha1
, sizeof(bind_map
.surface_sha1
));
216 blob_copy_bytes(blob
, bind_map
.sampler_sha1
, sizeof(bind_map
.sampler_sha1
));
217 blob_copy_bytes(blob
, bind_map
.push_sha1
, sizeof(bind_map
.push_sha1
));
218 bind_map
.surface_count
= blob_read_uint32(blob
);
219 bind_map
.sampler_count
= blob_read_uint32(blob
);
220 bind_map
.surface_to_descriptor
= (void *)
221 blob_read_bytes(blob
, bind_map
.surface_count
*
222 sizeof(*bind_map
.surface_to_descriptor
));
223 bind_map
.sampler_to_descriptor
= (void *)
224 blob_read_bytes(blob
, bind_map
.sampler_count
*
225 sizeof(*bind_map
.sampler_to_descriptor
));
226 blob_copy_bytes(blob
, bind_map
.push_ranges
, sizeof(bind_map
.push_ranges
));
231 return anv_shader_bin_create(device
,
233 kernel_data
, kernel_size
,
234 constant_data
, constant_data_size
,
235 prog_data
, prog_data_size
, prog_data_param
,
236 stats
, num_stats
, xfb_info
, &bind_map
);
241 * - Compact binding table layout so it's tight and not dependent on
242 * descriptor set layout.
244 * - Review prog_data struct for size and cacheability: struct
245 * brw_stage_prog_data has binding_table which uses a lot of uint32_t for 8
246 * bit quantities etc; use bit fields for all bools, eg dual_src_blend.
250 shader_bin_key_hash_func(const void *void_key
)
252 const struct anv_shader_bin_key
*key
= void_key
;
253 return _mesa_hash_data(key
->data
, key
->size
);
257 shader_bin_key_compare_func(const void *void_a
, const void *void_b
)
259 const struct anv_shader_bin_key
*a
= void_a
, *b
= void_b
;
260 if (a
->size
!= b
->size
)
263 return memcmp(a
->data
, b
->data
, a
->size
) == 0;
267 sha1_hash_func(const void *sha1
)
269 return _mesa_hash_data(sha1
, 20);
273 sha1_compare_func(const void *sha1_a
, const void *sha1_b
)
275 return memcmp(sha1_a
, sha1_b
, 20) == 0;
279 anv_pipeline_cache_init(struct anv_pipeline_cache
*cache
,
280 struct anv_device
*device
,
283 cache
->device
= device
;
284 pthread_mutex_init(&cache
->mutex
, NULL
);
287 cache
->cache
= _mesa_hash_table_create(NULL
, shader_bin_key_hash_func
,
288 shader_bin_key_compare_func
);
289 cache
->nir_cache
= _mesa_hash_table_create(NULL
, sha1_hash_func
,
293 cache
->nir_cache
= NULL
;
298 anv_pipeline_cache_finish(struct anv_pipeline_cache
*cache
)
300 pthread_mutex_destroy(&cache
->mutex
);
303 /* This is a bit unfortunate. In order to keep things from randomly
304 * going away, the shader cache has to hold a reference to all shader
305 * binaries it contains. We unref them when we destroy the cache.
307 hash_table_foreach(cache
->cache
, entry
)
308 anv_shader_bin_unref(cache
->device
, entry
->data
);
310 _mesa_hash_table_destroy(cache
->cache
, NULL
);
313 if (cache
->nir_cache
) {
314 hash_table_foreach(cache
->nir_cache
, entry
)
315 ralloc_free(entry
->data
);
317 _mesa_hash_table_destroy(cache
->nir_cache
, NULL
);
321 static struct anv_shader_bin
*
322 anv_pipeline_cache_search_locked(struct anv_pipeline_cache
*cache
,
323 const void *key_data
, uint32_t key_size
)
325 uint32_t vla
[1 + DIV_ROUND_UP(key_size
, sizeof(uint32_t))];
326 struct anv_shader_bin_key
*key
= (void *)vla
;
327 key
->size
= key_size
;
328 memcpy(key
->data
, key_data
, key_size
);
330 struct hash_entry
*entry
= _mesa_hash_table_search(cache
->cache
, key
);
337 struct anv_shader_bin
*
338 anv_pipeline_cache_search(struct anv_pipeline_cache
*cache
,
339 const void *key_data
, uint32_t key_size
)
344 pthread_mutex_lock(&cache
->mutex
);
346 struct anv_shader_bin
*shader
=
347 anv_pipeline_cache_search_locked(cache
, key_data
, key_size
);
349 pthread_mutex_unlock(&cache
->mutex
);
351 /* We increment refcount before handing it to the caller */
353 anv_shader_bin_ref(shader
);
359 anv_pipeline_cache_add_shader_bin(struct anv_pipeline_cache
*cache
,
360 struct anv_shader_bin
*bin
)
365 pthread_mutex_lock(&cache
->mutex
);
367 struct hash_entry
*entry
= _mesa_hash_table_search(cache
->cache
, bin
->key
);
369 /* Take a reference for the cache */
370 anv_shader_bin_ref(bin
);
371 _mesa_hash_table_insert(cache
->cache
, bin
->key
, bin
);
374 pthread_mutex_unlock(&cache
->mutex
);
377 static struct anv_shader_bin
*
378 anv_pipeline_cache_add_shader_locked(struct anv_pipeline_cache
*cache
,
379 const void *key_data
, uint32_t key_size
,
380 const void *kernel_data
,
381 uint32_t kernel_size
,
382 const void *constant_data
,
383 uint32_t constant_data_size
,
384 const struct brw_stage_prog_data
*prog_data
,
385 uint32_t prog_data_size
,
386 const void *prog_data_param
,
387 const struct brw_compile_stats
*stats
,
389 const nir_xfb_info
*xfb_info
,
390 const struct anv_pipeline_bind_map
*bind_map
)
392 struct anv_shader_bin
*shader
=
393 anv_pipeline_cache_search_locked(cache
, key_data
, key_size
);
397 struct anv_shader_bin
*bin
=
398 anv_shader_bin_create(cache
->device
, key_data
, key_size
,
399 kernel_data
, kernel_size
,
400 constant_data
, constant_data_size
,
401 prog_data
, prog_data_size
, prog_data_param
,
402 stats
, num_stats
, xfb_info
, bind_map
);
406 _mesa_hash_table_insert(cache
->cache
, bin
->key
, bin
);
411 struct anv_shader_bin
*
412 anv_pipeline_cache_upload_kernel(struct anv_pipeline_cache
*cache
,
413 const void *key_data
, uint32_t key_size
,
414 const void *kernel_data
, uint32_t kernel_size
,
415 const void *constant_data
,
416 uint32_t constant_data_size
,
417 const struct brw_stage_prog_data
*prog_data
,
418 uint32_t prog_data_size
,
419 const struct brw_compile_stats
*stats
,
421 const nir_xfb_info
*xfb_info
,
422 const struct anv_pipeline_bind_map
*bind_map
)
425 pthread_mutex_lock(&cache
->mutex
);
427 struct anv_shader_bin
*bin
=
428 anv_pipeline_cache_add_shader_locked(cache
, key_data
, key_size
,
429 kernel_data
, kernel_size
,
430 constant_data
, constant_data_size
,
431 prog_data
, prog_data_size
,
436 pthread_mutex_unlock(&cache
->mutex
);
438 /* We increment refcount before handing it to the caller */
440 anv_shader_bin_ref(bin
);
444 /* In this case, we're not caching it so the caller owns it entirely */
445 return anv_shader_bin_create(cache
->device
, key_data
, key_size
,
446 kernel_data
, kernel_size
,
447 constant_data
, constant_data_size
,
448 prog_data
, prog_data_size
,
455 struct cache_header
{
456 uint32_t header_size
;
457 uint32_t header_version
;
460 uint8_t uuid
[VK_UUID_SIZE
];
464 anv_pipeline_cache_load(struct anv_pipeline_cache
*cache
,
465 const void *data
, size_t size
)
467 struct anv_device
*device
= cache
->device
;
468 struct anv_physical_device
*pdevice
= device
->physical
;
470 if (cache
->cache
== NULL
)
473 struct blob_reader blob
;
474 blob_reader_init(&blob
, data
, size
);
476 struct cache_header header
;
477 blob_copy_bytes(&blob
, &header
, sizeof(header
));
478 uint32_t count
= blob_read_uint32(&blob
);
482 if (header
.header_size
< sizeof(header
))
484 if (header
.header_version
!= VK_PIPELINE_CACHE_HEADER_VERSION_ONE
)
486 if (header
.vendor_id
!= 0x8086)
488 if (header
.device_id
!= device
->info
.chipset_id
)
490 if (memcmp(header
.uuid
, pdevice
->pipeline_cache_uuid
, VK_UUID_SIZE
) != 0)
493 for (uint32_t i
= 0; i
< count
; i
++) {
494 struct anv_shader_bin
*bin
=
495 anv_shader_bin_create_from_blob(device
, &blob
);
498 _mesa_hash_table_insert(cache
->cache
, bin
->key
, bin
);
502 VkResult
anv_CreatePipelineCache(
504 const VkPipelineCacheCreateInfo
* pCreateInfo
,
505 const VkAllocationCallbacks
* pAllocator
,
506 VkPipelineCache
* pPipelineCache
)
508 ANV_FROM_HANDLE(anv_device
, device
, _device
);
509 struct anv_pipeline_cache
*cache
;
511 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO
);
512 assert(pCreateInfo
->flags
== 0);
514 cache
= vk_alloc2(&device
->alloc
, pAllocator
,
516 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
518 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
520 anv_pipeline_cache_init(cache
, device
,
521 device
->physical
->instance
->pipeline_cache_enabled
);
523 if (pCreateInfo
->initialDataSize
> 0)
524 anv_pipeline_cache_load(cache
,
525 pCreateInfo
->pInitialData
,
526 pCreateInfo
->initialDataSize
);
528 *pPipelineCache
= anv_pipeline_cache_to_handle(cache
);
533 void anv_DestroyPipelineCache(
535 VkPipelineCache _cache
,
536 const VkAllocationCallbacks
* pAllocator
)
538 ANV_FROM_HANDLE(anv_device
, device
, _device
);
539 ANV_FROM_HANDLE(anv_pipeline_cache
, cache
, _cache
);
544 anv_pipeline_cache_finish(cache
);
546 vk_free2(&device
->alloc
, pAllocator
, cache
);
549 VkResult
anv_GetPipelineCacheData(
551 VkPipelineCache _cache
,
555 ANV_FROM_HANDLE(anv_device
, device
, _device
);
556 ANV_FROM_HANDLE(anv_pipeline_cache
, cache
, _cache
);
560 blob_init_fixed(&blob
, pData
, *pDataSize
);
562 blob_init_fixed(&blob
, NULL
, SIZE_MAX
);
565 struct cache_header header
= {
566 .header_size
= sizeof(struct cache_header
),
567 .header_version
= VK_PIPELINE_CACHE_HEADER_VERSION_ONE
,
569 .device_id
= device
->info
.chipset_id
,
571 memcpy(header
.uuid
, device
->physical
->pipeline_cache_uuid
, VK_UUID_SIZE
);
572 blob_write_bytes(&blob
, &header
, sizeof(header
));
575 intptr_t count_offset
= blob_reserve_uint32(&blob
);
576 if (count_offset
< 0) {
579 return VK_INCOMPLETE
;
582 VkResult result
= VK_SUCCESS
;
584 hash_table_foreach(cache
->cache
, entry
) {
585 struct anv_shader_bin
*shader
= entry
->data
;
587 size_t save_size
= blob
.size
;
588 if (!anv_shader_bin_write_to_blob(shader
, &blob
)) {
589 /* If it fails reset to the previous size and bail */
590 blob
.size
= save_size
;
591 result
= VK_INCOMPLETE
;
599 blob_overwrite_uint32(&blob
, count_offset
, count
);
601 *pDataSize
= blob
.size
;
608 VkResult
anv_MergePipelineCaches(
610 VkPipelineCache destCache
,
611 uint32_t srcCacheCount
,
612 const VkPipelineCache
* pSrcCaches
)
614 ANV_FROM_HANDLE(anv_pipeline_cache
, dst
, destCache
);
619 for (uint32_t i
= 0; i
< srcCacheCount
; i
++) {
620 ANV_FROM_HANDLE(anv_pipeline_cache
, src
, pSrcCaches
[i
]);
624 hash_table_foreach(src
->cache
, entry
) {
625 struct anv_shader_bin
*bin
= entry
->data
;
628 if (_mesa_hash_table_search(dst
->cache
, bin
->key
))
631 anv_shader_bin_ref(bin
);
632 _mesa_hash_table_insert(dst
->cache
, bin
->key
, bin
);
639 struct anv_shader_bin
*
640 anv_device_search_for_kernel(struct anv_device
*device
,
641 struct anv_pipeline_cache
*cache
,
642 const void *key_data
, uint32_t key_size
,
643 bool *user_cache_hit
)
645 struct anv_shader_bin
*bin
;
647 *user_cache_hit
= false;
650 bin
= anv_pipeline_cache_search(cache
, key_data
, key_size
);
652 *user_cache_hit
= cache
!= &device
->default_pipeline_cache
;
657 #ifdef ENABLE_SHADER_CACHE
658 struct disk_cache
*disk_cache
= device
->physical
->disk_cache
;
659 if (disk_cache
&& device
->physical
->instance
->pipeline_cache_enabled
) {
661 disk_cache_compute_key(disk_cache
, key_data
, key_size
, cache_key
);
664 uint8_t *buffer
= disk_cache_get(disk_cache
, cache_key
, &buffer_size
);
666 struct blob_reader blob
;
667 blob_reader_init(&blob
, buffer
, buffer_size
);
668 bin
= anv_shader_bin_create_from_blob(device
, &blob
);
673 anv_pipeline_cache_add_shader_bin(cache
, bin
);
683 struct anv_shader_bin
*
684 anv_device_upload_kernel(struct anv_device
*device
,
685 struct anv_pipeline_cache
*cache
,
686 const void *key_data
, uint32_t key_size
,
687 const void *kernel_data
, uint32_t kernel_size
,
688 const void *constant_data
,
689 uint32_t constant_data_size
,
690 const struct brw_stage_prog_data
*prog_data
,
691 uint32_t prog_data_size
,
692 const struct brw_compile_stats
*stats
,
694 const nir_xfb_info
*xfb_info
,
695 const struct anv_pipeline_bind_map
*bind_map
)
697 struct anv_shader_bin
*bin
;
699 bin
= anv_pipeline_cache_upload_kernel(cache
, key_data
, key_size
,
700 kernel_data
, kernel_size
,
701 constant_data
, constant_data_size
,
702 prog_data
, prog_data_size
,
706 bin
= anv_shader_bin_create(device
, key_data
, key_size
,
707 kernel_data
, kernel_size
,
708 constant_data
, constant_data_size
,
709 prog_data
, prog_data_size
,
718 #ifdef ENABLE_SHADER_CACHE
719 struct disk_cache
*disk_cache
= device
->physical
->disk_cache
;
723 if (anv_shader_bin_write_to_blob(bin
, &binary
)) {
725 disk_cache_compute_key(disk_cache
, key_data
, key_size
, cache_key
);
727 disk_cache_put(disk_cache
, cache_key
, binary
.data
, binary
.size
, NULL
);
730 blob_finish(&binary
);
737 struct serialized_nir
{
738 unsigned char sha1_key
[20];
744 anv_device_search_for_nir(struct anv_device
*device
,
745 struct anv_pipeline_cache
*cache
,
746 const nir_shader_compiler_options
*nir_options
,
747 unsigned char sha1_key
[20],
750 if (cache
&& cache
->nir_cache
) {
751 const struct serialized_nir
*snir
= NULL
;
753 pthread_mutex_lock(&cache
->mutex
);
754 struct hash_entry
*entry
=
755 _mesa_hash_table_search(cache
->nir_cache
, sha1_key
);
758 pthread_mutex_unlock(&cache
->mutex
);
761 struct blob_reader blob
;
762 blob_reader_init(&blob
, snir
->data
, snir
->size
);
764 nir_shader
*nir
= nir_deserialize(mem_ctx
, nir_options
, &blob
);
777 anv_device_upload_nir(struct anv_device
*device
,
778 struct anv_pipeline_cache
*cache
,
779 const struct nir_shader
*nir
,
780 unsigned char sha1_key
[20])
782 if (cache
&& cache
->nir_cache
) {
783 pthread_mutex_lock(&cache
->mutex
);
784 struct hash_entry
*entry
=
785 _mesa_hash_table_search(cache
->nir_cache
, sha1_key
);
786 pthread_mutex_unlock(&cache
->mutex
);
793 nir_serialize(&blob
, nir
, false);
794 if (blob
.out_of_memory
) {
799 pthread_mutex_lock(&cache
->mutex
);
800 /* Because ralloc isn't thread-safe, we have to do all this inside the
801 * lock. We could unlock for the big memcpy but it's probably not worth
804 entry
= _mesa_hash_table_search(cache
->nir_cache
, sha1_key
);
807 pthread_mutex_unlock(&cache
->mutex
);
811 struct serialized_nir
*snir
=
812 ralloc_size(cache
->nir_cache
, sizeof(*snir
) + blob
.size
);
813 memcpy(snir
->sha1_key
, sha1_key
, 20);
814 snir
->size
= blob
.size
;
815 memcpy(snir
->data
, blob
.data
, blob
.size
);
819 _mesa_hash_table_insert(cache
->nir_cache
, snir
->sha1_key
, snir
);
821 pthread_mutex_unlock(&cache
->mutex
);