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
33 anv_env_get_int(const char *name
)
35 const char *val
= getenv(name
);
40 return strtol(val
, NULL
, 0);
44 fill_physical_device(struct anv_physical_device
*device
,
45 struct anv_instance
*instance
,
50 fd
= open("/dev/dri/renderD128", O_RDWR
| O_CLOEXEC
);
52 return vk_error(VK_ERROR_UNAVAILABLE
);
54 device
->instance
= instance
;
57 device
->chipset_id
= anv_env_get_int("INTEL_DEVID_OVERRIDE");
58 device
->no_hw
= false;
59 if (device
->chipset_id
) {
60 /* INTEL_DEVID_OVERRIDE implies INTEL_NO_HW. */
63 device
->chipset_id
= anv_gem_get_param(fd
, I915_PARAM_CHIPSET_ID
);
65 if (!device
->chipset_id
)
68 device
->name
= brw_get_device_name(device
->chipset_id
);
69 device
->info
= brw_get_device_info(device
->chipset_id
, -1);
73 if (!anv_gem_get_param(fd
, I915_PARAM_HAS_WAIT_TIMEOUT
))
76 if (!anv_gem_get_param(fd
, I915_PARAM_HAS_EXECBUF2
))
79 if (!anv_gem_get_param(fd
, I915_PARAM_HAS_LLC
))
82 if (!anv_gem_get_param(fd
, I915_PARAM_HAS_EXEC_CONSTANTS
))
92 return vk_error(VK_ERROR_UNAVAILABLE
);
95 static void *default_alloc(
99 VkSystemAllocType allocType
)
104 static void default_free(
111 static const VkAllocCallbacks default_alloc_callbacks
= {
113 .pfnAlloc
= default_alloc
,
114 .pfnFree
= default_free
117 VkResult
anv_CreateInstance(
118 const VkInstanceCreateInfo
* pCreateInfo
,
119 VkInstance
* pInstance
)
121 struct anv_instance
*instance
;
122 const VkAllocCallbacks
*alloc_callbacks
= &default_alloc_callbacks
;
123 void *user_data
= NULL
;
126 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO
);
128 if (pCreateInfo
->pAllocCb
) {
129 alloc_callbacks
= pCreateInfo
->pAllocCb
;
130 user_data
= pCreateInfo
->pAllocCb
->pUserData
;
132 instance
= alloc_callbacks
->pfnAlloc(user_data
, sizeof(*instance
), 8,
133 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
135 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
137 instance
->pAllocUserData
= alloc_callbacks
->pUserData
;
138 instance
->pfnAlloc
= alloc_callbacks
->pfnAlloc
;
139 instance
->pfnFree
= alloc_callbacks
->pfnFree
;
140 instance
->apiVersion
= pCreateInfo
->pAppInfo
->apiVersion
;
142 instance
->physicalDeviceCount
= 0;
143 result
= fill_physical_device(&instance
->physicalDevice
,
144 instance
, "/dev/dri/renderD128");
146 if (result
!= VK_SUCCESS
)
149 instance
->physicalDeviceCount
++;
150 *pInstance
= (VkInstance
) instance
;
155 VkResult
anv_DestroyInstance(
156 VkInstance _instance
)
158 struct anv_instance
*instance
= (struct anv_instance
*) _instance
;
160 instance
->pfnFree(instance
->pAllocUserData
, instance
);
165 VkResult
anv_EnumeratePhysicalDevices(
166 VkInstance _instance
,
167 uint32_t* pPhysicalDeviceCount
,
168 VkPhysicalDevice
* pPhysicalDevices
)
170 struct anv_instance
*instance
= (struct anv_instance
*) _instance
;
172 if (*pPhysicalDeviceCount
>= 1)
173 pPhysicalDevices
[0] = (VkPhysicalDevice
) &instance
->physicalDevice
;
174 *pPhysicalDeviceCount
= instance
->physicalDeviceCount
;
179 VkResult
anv_GetPhysicalDeviceInfo(
180 VkPhysicalDevice physicalDevice
,
181 VkPhysicalDeviceInfoType infoType
,
185 struct anv_physical_device
*device
= (struct anv_physical_device
*) physicalDevice
;
186 VkPhysicalDeviceProperties
*properties
;
187 VkPhysicalDevicePerformance
*performance
;
188 VkPhysicalDeviceQueueProperties
*queue_properties
;
189 VkPhysicalDeviceMemoryProperties
*memory_properties
;
190 VkDisplayPropertiesWSI
*display_properties
;
191 uint64_t ns_per_tick
= 80;
193 switch ((uint32_t) infoType
) {
194 case VK_PHYSICAL_DEVICE_INFO_TYPE_PROPERTIES
:
197 *pDataSize
= sizeof(*properties
);
201 properties
->apiVersion
= 1;
202 properties
->driverVersion
= 1;
203 properties
->vendorId
= 0x8086;
204 properties
->deviceId
= device
->chipset_id
;
205 properties
->deviceType
= VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU
;
206 strcpy(properties
->deviceName
, device
->name
);
207 properties
->maxInlineMemoryUpdateSize
= 0;
208 properties
->maxBoundDescriptorSets
= MAX_SETS
;
209 properties
->maxThreadGroupSize
= 512;
210 properties
->timestampFrequency
= 1000 * 1000 * 1000 / ns_per_tick
;
211 properties
->multiColorAttachmentClears
= true;
212 properties
->maxDescriptorSets
= 8;
213 properties
->maxViewports
= 16;
214 properties
->maxColorAttachments
= 8;
217 case VK_PHYSICAL_DEVICE_INFO_TYPE_PERFORMANCE
:
220 *pDataSize
= sizeof(*performance
);
224 performance
->maxDeviceClock
= 1.0;
225 performance
->aluPerClock
= 1.0;
226 performance
->texPerClock
= 1.0;
227 performance
->primsPerClock
= 1.0;
228 performance
->pixelsPerClock
= 1.0;
231 case VK_PHYSICAL_DEVICE_INFO_TYPE_QUEUE_PROPERTIES
:
232 queue_properties
= pData
;
234 *pDataSize
= sizeof(*queue_properties
);
238 queue_properties
->queueFlags
= 0;
239 queue_properties
->queueCount
= 1;
240 queue_properties
->maxAtomicCounters
= 0;
241 queue_properties
->supportsTimestamps
= true;
242 queue_properties
->maxMemReferences
= 256;
245 case VK_PHYSICAL_DEVICE_INFO_TYPE_MEMORY_PROPERTIES
:
246 memory_properties
= pData
;
248 *pDataSize
= sizeof(*memory_properties
);
252 memory_properties
->supportsMigration
= false;
253 memory_properties
->supportsPinning
= false;
256 case VK_PHYSICAL_DEVICE_INFO_TYPE_DISPLAY_PROPERTIES_WSI
:
257 anv_finishme("VK_PHYSICAL_DEVICE_INFO_TYPE_DISPLAY_PROPERTIES_WSI");
259 *pDataSize
= sizeof(*display_properties
);
263 display_properties
= pData
;
264 display_properties
->display
= 0;
265 display_properties
->physicalResolution
= (VkExtent2D
) { 0, 0 };
268 case VK_PHYSICAL_DEVICE_INFO_TYPE_QUEUE_PRESENT_PROPERTIES_WSI
:
269 anv_finishme("VK_PHYSICAL_DEVICE_INFO_TYPE_QUEUE_PRESENT_PROPERTIES_WSI");
274 return VK_UNSUPPORTED
;
279 void * vkGetProcAddr(
280 VkPhysicalDevice physicalDevice
,
283 return anv_lookup_entrypoint(pName
);
287 parse_debug_flags(struct anv_device
*device
)
289 const char *debug
, *p
, *end
;
291 debug
= getenv("INTEL_DEBUG");
292 device
->dump_aub
= false;
294 for (p
= debug
; *p
; p
= end
+ 1) {
295 end
= strchrnul(p
, ',');
296 if (end
- p
== 3 && memcmp(p
, "aub", 3) == 0)
297 device
->dump_aub
= true;
298 if (end
- p
== 5 && memcmp(p
, "no_hw", 5) == 0)
299 device
->no_hw
= true;
306 static const uint32_t BATCH_SIZE
= 8192;
308 VkResult
anv_CreateDevice(
309 VkPhysicalDevice _physicalDevice
,
310 const VkDeviceCreateInfo
* pCreateInfo
,
313 struct anv_physical_device
*physicalDevice
=
314 (struct anv_physical_device
*) _physicalDevice
;
315 struct anv_instance
*instance
= physicalDevice
->instance
;
316 struct anv_device
*device
;
318 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO
);
320 device
= instance
->pfnAlloc(instance
->pAllocUserData
,
322 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
324 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
326 device
->no_hw
= physicalDevice
->no_hw
;
327 parse_debug_flags(device
);
329 device
->instance
= physicalDevice
->instance
;
330 device
->fd
= open("/dev/dri/renderD128", O_RDWR
| O_CLOEXEC
);
331 if (device
->fd
== -1)
334 device
->context_id
= anv_gem_create_context(device
);
335 if (device
->context_id
== -1)
338 anv_bo_pool_init(&device
->batch_bo_pool
, device
, BATCH_SIZE
);
340 anv_block_pool_init(&device
->dynamic_state_block_pool
, device
, 2048);
342 anv_state_pool_init(&device
->dynamic_state_pool
,
343 &device
->dynamic_state_block_pool
);
345 anv_block_pool_init(&device
->instruction_block_pool
, device
, 2048);
346 anv_block_pool_init(&device
->surface_state_block_pool
, device
, 2048);
349 /* Binding table pointers are only 16 bits so we have to make sure that
350 * they get allocated at the beginning of the surface state BO. To
351 * handle this, we create a separate block pool that works out of the
352 * first 64 KB of the surface state BO.
354 anv_block_pool_init_slave(&device
->binding_table_block_pool
,
355 &device
->surface_state_block_pool
, 32);
357 anv_state_pool_init(&device
->surface_state_pool
,
358 &device
->surface_state_block_pool
);
360 device
->compiler
= anv_compiler_create(device
->fd
);
361 device
->aub_writer
= NULL
;
363 device
->info
= *physicalDevice
->info
;
365 pthread_mutex_init(&device
->mutex
, NULL
);
367 anv_device_init_meta(device
);
369 *pDevice
= (VkDevice
) device
;
376 anv_device_free(device
, device
);
378 return vk_error(VK_ERROR_UNAVAILABLE
);
381 VkResult
anv_DestroyDevice(
384 struct anv_device
*device
= (struct anv_device
*) _device
;
386 anv_compiler_destroy(device
->compiler
);
389 anv_bo_pool_finish(&device
->batch_bo_pool
);
390 anv_block_pool_finish(&device
->dynamic_state_block_pool
);
391 anv_block_pool_finish(&device
->instruction_block_pool
);
392 anv_block_pool_finish(&device
->surface_state_block_pool
);
396 if (device
->aub_writer
)
397 anv_aub_writer_destroy(device
->aub_writer
);
399 anv_device_free(device
, device
);
404 VkResult
anv_GetGlobalExtensionInfo(
405 VkExtensionInfoType infoType
,
406 uint32_t extensionIndex
,
410 static const VkExtensionProperties extensions
[] = {
412 .extName
= "VK_WSI_LunarG",
416 uint32_t count
= ARRAY_SIZE(extensions
);
419 case VK_EXTENSION_INFO_TYPE_COUNT
:
420 memcpy(pData
, &count
, sizeof(count
));
421 *pDataSize
= sizeof(count
);
424 case VK_EXTENSION_INFO_TYPE_PROPERTIES
:
425 if (extensionIndex
>= count
)
426 return vk_error(VK_ERROR_INVALID_EXTENSION
);
428 memcpy(pData
, &extensions
[extensionIndex
], sizeof(extensions
[0]));
429 *pDataSize
= sizeof(extensions
[0]);
433 return VK_UNSUPPORTED
;
437 VkResult
anv_GetPhysicalDeviceExtensionInfo(
438 VkPhysicalDevice physicalDevice
,
439 VkExtensionInfoType infoType
,
440 uint32_t extensionIndex
,
447 case VK_EXTENSION_INFO_TYPE_COUNT
:
456 case VK_EXTENSION_INFO_TYPE_PROPERTIES
:
457 return vk_error(VK_ERROR_INVALID_EXTENSION
);
460 return VK_UNSUPPORTED
;
464 VkResult
anv_EnumerateLayers(
465 VkPhysicalDevice physicalDevice
,
466 size_t maxStringSize
,
468 char* const* pOutLayers
,
476 VkResult
anv_GetDeviceQueue(
478 uint32_t queueNodeIndex
,
482 struct anv_device
*device
= (struct anv_device
*) _device
;
483 struct anv_queue
*queue
;
485 /* FIXME: Should allocate these at device create time. */
487 queue
= anv_device_alloc(device
, sizeof(*queue
), 8,
488 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
490 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
492 queue
->device
= device
;
493 queue
->pool
= &device
->surface_state_pool
;
495 queue
->completed_serial
= anv_state_pool_alloc(queue
->pool
, 4, 4);
496 *(uint32_t *)queue
->completed_serial
.map
= 0;
497 queue
->next_serial
= 1;
499 *pQueue
= (VkQueue
) queue
;
505 anv_reloc_list_init(struct anv_reloc_list
*list
, struct anv_device
*device
)
507 list
->num_relocs
= 0;
508 list
->array_length
= 256;
510 anv_device_alloc(device
, list
->array_length
* sizeof(*list
->relocs
), 8,
511 VK_SYSTEM_ALLOC_TYPE_INTERNAL
);
513 if (list
->relocs
== NULL
)
514 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
517 anv_device_alloc(device
, list
->array_length
* sizeof(*list
->reloc_bos
), 8,
518 VK_SYSTEM_ALLOC_TYPE_INTERNAL
);
520 if (list
->relocs
== NULL
) {
521 anv_device_free(device
, list
->relocs
);
522 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
529 anv_reloc_list_finish(struct anv_reloc_list
*list
, struct anv_device
*device
)
531 anv_device_free(device
, list
->relocs
);
532 anv_device_free(device
, list
->reloc_bos
);
536 anv_reloc_list_grow(struct anv_reloc_list
*list
, struct anv_device
*device
,
537 size_t num_additional_relocs
)
539 if (list
->num_relocs
+ num_additional_relocs
<= list
->array_length
)
542 size_t new_length
= list
->array_length
* 2;
543 while (new_length
< list
->num_relocs
+ num_additional_relocs
)
546 struct drm_i915_gem_relocation_entry
*new_relocs
=
547 anv_device_alloc(device
, new_length
* sizeof(*list
->relocs
), 8,
548 VK_SYSTEM_ALLOC_TYPE_INTERNAL
);
549 if (new_relocs
== NULL
)
550 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
552 struct anv_bo
**new_reloc_bos
=
553 anv_device_alloc(device
, new_length
* sizeof(*list
->reloc_bos
), 8,
554 VK_SYSTEM_ALLOC_TYPE_INTERNAL
);
555 if (new_relocs
== NULL
) {
556 anv_device_free(device
, new_relocs
);
557 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
560 memcpy(new_relocs
, list
->relocs
, list
->num_relocs
* sizeof(*list
->relocs
));
561 memcpy(new_reloc_bos
, list
->reloc_bos
,
562 list
->num_relocs
* sizeof(*list
->reloc_bos
));
564 anv_device_free(device
, list
->relocs
);
565 anv_device_free(device
, list
->reloc_bos
);
567 list
->relocs
= new_relocs
;
568 list
->reloc_bos
= new_reloc_bos
;
574 anv_batch_bo_create(struct anv_device
*device
, struct anv_batch_bo
**bbo_out
)
578 struct anv_batch_bo
*bbo
=
579 anv_device_alloc(device
, sizeof(*bbo
), 8, VK_SYSTEM_ALLOC_TYPE_INTERNAL
);
581 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
584 bbo
->prev_batch_bo
= NULL
;
586 result
= anv_bo_pool_alloc(&device
->batch_bo_pool
, &bbo
->bo
);
587 if (result
!= VK_SUCCESS
) {
588 anv_device_free(device
, bbo
);
598 anv_batch_bo_start(struct anv_batch_bo
*bbo
, struct anv_batch
*batch
,
599 size_t batch_padding
)
601 batch
->next
= batch
->start
= bbo
->bo
.map
;
602 batch
->end
= bbo
->bo
.map
+ bbo
->bo
.size
- batch_padding
;
603 bbo
->first_reloc
= batch
->relocs
.num_relocs
;
607 anv_batch_bo_finish(struct anv_batch_bo
*bbo
, struct anv_batch
*batch
)
609 assert(batch
->start
== bbo
->bo
.map
);
610 bbo
->length
= batch
->next
- batch
->start
;
611 bbo
->num_relocs
= batch
->relocs
.num_relocs
- bbo
->first_reloc
;
615 anv_batch_bo_destroy(struct anv_batch_bo
*bbo
, struct anv_device
*device
)
617 anv_bo_pool_free(&device
->batch_bo_pool
, &bbo
->bo
);
618 anv_device_free(device
, bbo
);
622 anv_batch_emit_dwords(struct anv_batch
*batch
, int num_dwords
)
624 if (batch
->next
+ num_dwords
* 4 > batch
->end
)
625 batch
->extend_cb(batch
, batch
->user_data
);
627 void *p
= batch
->next
;
629 batch
->next
+= num_dwords
* 4;
630 assert(batch
->next
<= batch
->end
);
636 anv_reloc_list_append(struct anv_reloc_list
*list
, struct anv_device
*device
,
637 struct anv_reloc_list
*other
, uint32_t offset
)
639 anv_reloc_list_grow(list
, device
, other
->num_relocs
);
640 /* TODO: Handle failure */
642 memcpy(&list
->relocs
[list
->num_relocs
], &other
->relocs
[0],
643 other
->num_relocs
* sizeof(other
->relocs
[0]));
644 memcpy(&list
->reloc_bos
[list
->num_relocs
], &other
->reloc_bos
[0],
645 other
->num_relocs
* sizeof(other
->reloc_bos
[0]));
647 for (uint32_t i
= 0; i
< other
->num_relocs
; i
++)
648 list
->relocs
[i
+ list
->num_relocs
].offset
+= offset
;
650 list
->num_relocs
+= other
->num_relocs
;
654 anv_reloc_list_add(struct anv_reloc_list
*list
, struct anv_device
*device
,
655 uint32_t offset
, struct anv_bo
*target_bo
, uint32_t delta
)
657 struct drm_i915_gem_relocation_entry
*entry
;
660 anv_reloc_list_grow(list
, device
, 1);
661 /* TODO: Handle failure */
663 /* XXX: Can we use I915_EXEC_HANDLE_LUT? */
664 index
= list
->num_relocs
++;
665 list
->reloc_bos
[index
] = target_bo
;
666 entry
= &list
->relocs
[index
];
667 entry
->target_handle
= target_bo
->gem_handle
;
668 entry
->delta
= delta
;
669 entry
->offset
= offset
;
670 entry
->presumed_offset
= target_bo
->offset
;
671 entry
->read_domains
= 0;
672 entry
->write_domain
= 0;
674 return target_bo
->offset
+ delta
;
678 anv_batch_emit_batch(struct anv_batch
*batch
, struct anv_batch
*other
)
680 uint32_t size
, offset
;
682 size
= other
->next
- other
->start
;
683 assert(size
% 4 == 0);
685 if (batch
->next
+ size
> batch
->end
)
686 batch
->extend_cb(batch
, batch
->user_data
);
688 assert(batch
->next
+ size
<= batch
->end
);
690 memcpy(batch
->next
, other
->start
, size
);
692 offset
= batch
->next
- batch
->start
;
693 anv_reloc_list_append(&batch
->relocs
, batch
->device
,
694 &other
->relocs
, offset
);
700 anv_batch_emit_reloc(struct anv_batch
*batch
,
701 void *location
, struct anv_bo
*bo
, uint32_t delta
)
703 return anv_reloc_list_add(&batch
->relocs
, batch
->device
,
704 location
- batch
->start
, bo
, delta
);
707 VkResult
anv_QueueSubmit(
709 uint32_t cmdBufferCount
,
710 const VkCmdBuffer
* pCmdBuffers
,
713 struct anv_queue
*queue
= (struct anv_queue
*) _queue
;
714 struct anv_device
*device
= queue
->device
;
715 struct anv_fence
*fence
= (struct anv_fence
*) _fence
;
718 for (uint32_t i
= 0; i
< cmdBufferCount
; i
++) {
719 struct anv_cmd_buffer
*cmd_buffer
=
720 (struct anv_cmd_buffer
*) pCmdBuffers
[i
];
722 if (device
->dump_aub
)
723 anv_cmd_buffer_dump(cmd_buffer
);
725 if (!device
->no_hw
) {
726 ret
= anv_gem_execbuffer(device
, &cmd_buffer
->execbuf
);
728 return vk_error(VK_ERROR_UNKNOWN
);
731 ret
= anv_gem_execbuffer(device
, &fence
->execbuf
);
733 return vk_error(VK_ERROR_UNKNOWN
);
736 for (uint32_t i
= 0; i
< cmd_buffer
->bo_count
; i
++)
737 cmd_buffer
->exec2_bos
[i
]->offset
= cmd_buffer
->exec2_objects
[i
].offset
;
739 *(uint32_t *)queue
->completed_serial
.map
= cmd_buffer
->serial
;
746 VkResult
anv_QueueAddMemReferences(
749 const VkDeviceMemory
* pMems
)
754 VkResult
anv_QueueRemoveMemReferences(
757 const VkDeviceMemory
* pMems
)
762 VkResult
anv_QueueWaitIdle(
765 struct anv_queue
*queue
= (struct anv_queue
*) _queue
;
767 return vkDeviceWaitIdle((VkDevice
) queue
->device
);
770 VkResult
anv_DeviceWaitIdle(
773 struct anv_device
*device
= (struct anv_device
*) _device
;
774 struct anv_state state
;
775 struct anv_batch batch
;
776 struct drm_i915_gem_execbuffer2 execbuf
;
777 struct drm_i915_gem_exec_object2 exec2_objects
[1];
778 struct anv_bo
*bo
= NULL
;
783 state
= anv_state_pool_alloc(&device
->dynamic_state_pool
, 32, 32);
784 bo
= &device
->dynamic_state_pool
.block_pool
->bo
;
785 batch
.start
= batch
.next
= state
.map
;
786 batch
.end
= state
.map
+ 32;
787 anv_batch_emit(&batch
, GEN8_MI_BATCH_BUFFER_END
);
788 anv_batch_emit(&batch
, GEN8_MI_NOOP
);
790 exec2_objects
[0].handle
= bo
->gem_handle
;
791 exec2_objects
[0].relocation_count
= 0;
792 exec2_objects
[0].relocs_ptr
= 0;
793 exec2_objects
[0].alignment
= 0;
794 exec2_objects
[0].offset
= bo
->offset
;
795 exec2_objects
[0].flags
= 0;
796 exec2_objects
[0].rsvd1
= 0;
797 exec2_objects
[0].rsvd2
= 0;
799 execbuf
.buffers_ptr
= (uintptr_t) exec2_objects
;
800 execbuf
.buffer_count
= 1;
801 execbuf
.batch_start_offset
= state
.offset
;
802 execbuf
.batch_len
= batch
.next
- state
.map
;
803 execbuf
.cliprects_ptr
= 0;
804 execbuf
.num_cliprects
= 0;
809 I915_EXEC_HANDLE_LUT
| I915_EXEC_NO_RELOC
| I915_EXEC_RENDER
;
810 execbuf
.rsvd1
= device
->context_id
;
813 if (!device
->no_hw
) {
814 ret
= anv_gem_execbuffer(device
, &execbuf
);
816 result
= vk_error(VK_ERROR_UNKNOWN
);
821 ret
= anv_gem_wait(device
, bo
->gem_handle
, &timeout
);
823 result
= vk_error(VK_ERROR_UNKNOWN
);
828 anv_state_pool_free(&device
->dynamic_state_pool
, state
);
833 anv_state_pool_free(&device
->dynamic_state_pool
, state
);
839 anv_device_alloc(struct anv_device
* device
,
842 VkSystemAllocType allocType
)
844 return device
->instance
->pfnAlloc(device
->instance
->pAllocUserData
,
851 anv_device_free(struct anv_device
* device
,
854 return device
->instance
->pfnFree(device
->instance
->pAllocUserData
,
859 anv_bo_init_new(struct anv_bo
*bo
, struct anv_device
*device
, uint64_t size
)
861 bo
->gem_handle
= anv_gem_create(device
, size
);
863 return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY
);
873 VkResult
anv_AllocMemory(
875 const VkMemoryAllocInfo
* pAllocInfo
,
876 VkDeviceMemory
* pMem
)
878 struct anv_device
*device
= (struct anv_device
*) _device
;
879 struct anv_device_memory
*mem
;
882 assert(pAllocInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_ALLOC_INFO
);
884 mem
= anv_device_alloc(device
, sizeof(*mem
), 8,
885 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
887 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
889 result
= anv_bo_init_new(&mem
->bo
, device
, pAllocInfo
->allocationSize
);
890 if (result
!= VK_SUCCESS
)
893 *pMem
= (VkDeviceMemory
) mem
;
898 anv_device_free(device
, mem
);
903 VkResult
anv_FreeMemory(
907 struct anv_device
*device
= (struct anv_device
*) _device
;
908 struct anv_device_memory
*mem
= (struct anv_device_memory
*) _mem
;
911 anv_gem_munmap(mem
->bo
.map
, mem
->bo
.size
);
913 if (mem
->bo
.gem_handle
!= 0)
914 anv_gem_close(device
, mem
->bo
.gem_handle
);
916 anv_device_free(device
, mem
);
921 VkResult
anv_SetMemoryPriority(
924 VkMemoryPriority priority
)
929 VkResult
anv_MapMemory(
934 VkMemoryMapFlags flags
,
937 struct anv_device
*device
= (struct anv_device
*) _device
;
938 struct anv_device_memory
*mem
= (struct anv_device_memory
*) _mem
;
940 /* FIXME: Is this supposed to be thread safe? Since vkUnmapMemory() only
941 * takes a VkDeviceMemory pointer, it seems like only one map of the memory
942 * at a time is valid. We could just mmap up front and return an offset
943 * pointer here, but that may exhaust virtual memory on 32 bit
946 mem
->map
= anv_gem_mmap(device
, mem
->bo
.gem_handle
, offset
, size
);
947 mem
->map_size
= size
;
954 VkResult
anv_UnmapMemory(
958 struct anv_device_memory
*mem
= (struct anv_device_memory
*) _mem
;
960 anv_gem_munmap(mem
->map
, mem
->map_size
);
965 VkResult
anv_FlushMappedMemory(
971 /* clflush here for !llc platforms */
976 VkResult
anv_PinSystemMemory(
980 VkDeviceMemory
* pMem
)
985 VkResult
anv_GetMultiDeviceCompatibility(
986 VkPhysicalDevice physicalDevice0
,
987 VkPhysicalDevice physicalDevice1
,
988 VkPhysicalDeviceCompatibilityInfo
* pInfo
)
990 return VK_UNSUPPORTED
;
993 VkResult
anv_OpenSharedMemory(
995 const VkMemoryOpenInfo
* pOpenInfo
,
996 VkDeviceMemory
* pMem
)
998 return VK_UNSUPPORTED
;
1001 VkResult
anv_OpenSharedSemaphore(
1003 const VkSemaphoreOpenInfo
* pOpenInfo
,
1004 VkSemaphore
* pSemaphore
)
1006 return VK_UNSUPPORTED
;
1009 VkResult
anv_OpenPeerMemory(
1011 const VkPeerMemoryOpenInfo
* pOpenInfo
,
1012 VkDeviceMemory
* pMem
)
1014 return VK_UNSUPPORTED
;
1017 VkResult
anv_OpenPeerImage(
1019 const VkPeerImageOpenInfo
* pOpenInfo
,
1021 VkDeviceMemory
* pMem
)
1023 return VK_UNSUPPORTED
;
1026 VkResult
anv_DestroyObject(
1028 VkObjectType objType
,
1031 struct anv_device
*device
= (struct anv_device
*) _device
;
1032 struct anv_object
*object
= (struct anv_object
*) _object
;
1035 case VK_OBJECT_TYPE_INSTANCE
:
1036 return anv_DestroyInstance((VkInstance
) _object
);
1038 case VK_OBJECT_TYPE_PHYSICAL_DEVICE
:
1039 /* We don't want to actually destroy physical devices */
1042 case VK_OBJECT_TYPE_DEVICE
:
1043 assert(_device
== (VkDevice
) _object
);
1044 return anv_DestroyDevice((VkDevice
) _object
);
1046 case VK_OBJECT_TYPE_QUEUE
:
1050 case VK_OBJECT_TYPE_DEVICE_MEMORY
:
1051 return anv_FreeMemory(_device
, (VkDeviceMemory
) _object
);
1053 case VK_OBJECT_TYPE_DESCRIPTOR_POOL
:
1054 /* These are just dummys anyway, so we don't need to destroy them */
1057 case VK_OBJECT_TYPE_BUFFER
:
1058 case VK_OBJECT_TYPE_BUFFER_VIEW
:
1059 case VK_OBJECT_TYPE_IMAGE
:
1060 case VK_OBJECT_TYPE_IMAGE_VIEW
:
1061 case VK_OBJECT_TYPE_COLOR_ATTACHMENT_VIEW
:
1062 case VK_OBJECT_TYPE_DEPTH_STENCIL_VIEW
:
1063 case VK_OBJECT_TYPE_SHADER
:
1064 case VK_OBJECT_TYPE_PIPELINE_LAYOUT
:
1065 case VK_OBJECT_TYPE_SAMPLER
:
1066 case VK_OBJECT_TYPE_DESCRIPTOR_SET
:
1067 case VK_OBJECT_TYPE_DESCRIPTOR_SET_LAYOUT
:
1068 case VK_OBJECT_TYPE_DYNAMIC_RS_STATE
:
1069 case VK_OBJECT_TYPE_DYNAMIC_CB_STATE
:
1070 case VK_OBJECT_TYPE_DYNAMIC_DS_STATE
:
1071 case VK_OBJECT_TYPE_RENDER_PASS
:
1072 /* These are trivially destroyable */
1073 anv_device_free(device
, (void *) _object
);
1076 case VK_OBJECT_TYPE_COMMAND_BUFFER
:
1077 case VK_OBJECT_TYPE_PIPELINE
:
1078 case VK_OBJECT_TYPE_DYNAMIC_VP_STATE
:
1079 case VK_OBJECT_TYPE_FENCE
:
1080 case VK_OBJECT_TYPE_QUERY_POOL
:
1081 case VK_OBJECT_TYPE_FRAMEBUFFER
:
1082 (object
->destructor
)(device
, object
, objType
);
1085 case VK_OBJECT_TYPE_SEMAPHORE
:
1086 case VK_OBJECT_TYPE_EVENT
:
1087 stub_return(VK_UNSUPPORTED
);
1090 unreachable("Invalid object type");
1095 fill_memory_requirements(
1096 VkObjectType objType
,
1098 VkMemoryRequirements
* memory_requirements
)
1100 struct anv_buffer
*buffer
;
1101 struct anv_image
*image
;
1103 memory_requirements
->memPropsAllowed
=
1104 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
1105 VK_MEMORY_PROPERTY_HOST_DEVICE_COHERENT_BIT
|
1106 /* VK_MEMORY_PROPERTY_HOST_UNCACHED_BIT | */
1107 VK_MEMORY_PROPERTY_HOST_WRITE_COMBINED_BIT
|
1108 VK_MEMORY_PROPERTY_PREFER_HOST_LOCAL
|
1109 VK_MEMORY_PROPERTY_SHAREABLE_BIT
;
1111 memory_requirements
->memPropsRequired
= 0;
1114 case VK_OBJECT_TYPE_BUFFER
:
1115 buffer
= (struct anv_buffer
*) object
;
1116 memory_requirements
->size
= buffer
->size
;
1117 memory_requirements
->alignment
= 16;
1119 case VK_OBJECT_TYPE_IMAGE
:
1120 image
= (struct anv_image
*) object
;
1121 memory_requirements
->size
= image
->size
;
1122 memory_requirements
->alignment
= image
->alignment
;
1125 memory_requirements
->size
= 0;
1131 get_allocation_count(VkObjectType objType
)
1134 case VK_OBJECT_TYPE_BUFFER
:
1135 case VK_OBJECT_TYPE_IMAGE
:
1142 VkResult
anv_GetObjectInfo(
1144 VkObjectType objType
,
1146 VkObjectInfoType infoType
,
1150 VkMemoryRequirements memory_requirements
;
1154 case VK_OBJECT_INFO_TYPE_MEMORY_REQUIREMENTS
:
1155 *pDataSize
= sizeof(memory_requirements
);
1159 fill_memory_requirements(objType
, object
, pData
);
1162 case VK_OBJECT_INFO_TYPE_MEMORY_ALLOCATION_COUNT
:
1163 *pDataSize
= sizeof(count
);
1168 *count
= get_allocation_count(objType
);
1172 return VK_UNSUPPORTED
;
1177 VkResult
anv_QueueBindObjectMemory(
1179 VkObjectType objType
,
1181 uint32_t allocationIdx
,
1182 VkDeviceMemory _mem
,
1183 VkDeviceSize memOffset
)
1185 struct anv_buffer
*buffer
;
1186 struct anv_image
*image
;
1187 struct anv_device_memory
*mem
= (struct anv_device_memory
*) _mem
;
1190 case VK_OBJECT_TYPE_BUFFER
:
1191 buffer
= (struct anv_buffer
*) object
;
1192 buffer
->bo
= &mem
->bo
;
1193 buffer
->offset
= memOffset
;
1195 case VK_OBJECT_TYPE_IMAGE
:
1196 image
= (struct anv_image
*) object
;
1197 image
->bo
= &mem
->bo
;
1198 image
->offset
= memOffset
;
1207 VkResult
anv_QueueBindObjectMemoryRange(
1209 VkObjectType objType
,
1211 uint32_t allocationIdx
,
1212 VkDeviceSize rangeOffset
,
1213 VkDeviceSize rangeSize
,
1215 VkDeviceSize memOffset
)
1217 stub_return(VK_UNSUPPORTED
);
1220 VkResult
anv_QueueBindImageMemoryRange(
1223 uint32_t allocationIdx
,
1224 const VkImageMemoryBindInfo
* pBindInfo
,
1226 VkDeviceSize memOffset
)
1228 stub_return(VK_UNSUPPORTED
);
1232 anv_fence_destroy(struct anv_device
*device
,
1233 struct anv_object
*object
,
1234 VkObjectType obj_type
)
1236 struct anv_fence
*fence
= (struct anv_fence
*) object
;
1238 assert(obj_type
== VK_OBJECT_TYPE_FENCE
);
1240 anv_gem_munmap(fence
->bo
.map
, fence
->bo
.size
);
1241 anv_gem_close(device
, fence
->bo
.gem_handle
);
1242 anv_device_free(device
, fence
);
1245 VkResult
anv_CreateFence(
1247 const VkFenceCreateInfo
* pCreateInfo
,
1250 struct anv_device
*device
= (struct anv_device
*) _device
;
1251 struct anv_fence
*fence
;
1252 struct anv_batch batch
;
1255 const uint32_t fence_size
= 128;
1257 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_FENCE_CREATE_INFO
);
1259 fence
= anv_device_alloc(device
, sizeof(*fence
), 8,
1260 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
1262 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1264 result
= anv_bo_init_new(&fence
->bo
, device
, fence_size
);
1265 if (result
!= VK_SUCCESS
)
1268 fence
->base
.destructor
= anv_fence_destroy
;
1271 anv_gem_mmap(device
, fence
->bo
.gem_handle
, 0, fence
->bo
.size
);
1272 batch
.next
= batch
.start
= fence
->bo
.map
;
1273 batch
.end
= fence
->bo
.map
+ fence
->bo
.size
;
1274 anv_batch_emit(&batch
, GEN8_MI_BATCH_BUFFER_END
);
1275 anv_batch_emit(&batch
, GEN8_MI_NOOP
);
1277 fence
->exec2_objects
[0].handle
= fence
->bo
.gem_handle
;
1278 fence
->exec2_objects
[0].relocation_count
= 0;
1279 fence
->exec2_objects
[0].relocs_ptr
= 0;
1280 fence
->exec2_objects
[0].alignment
= 0;
1281 fence
->exec2_objects
[0].offset
= fence
->bo
.offset
;
1282 fence
->exec2_objects
[0].flags
= 0;
1283 fence
->exec2_objects
[0].rsvd1
= 0;
1284 fence
->exec2_objects
[0].rsvd2
= 0;
1286 fence
->execbuf
.buffers_ptr
= (uintptr_t) fence
->exec2_objects
;
1287 fence
->execbuf
.buffer_count
= 1;
1288 fence
->execbuf
.batch_start_offset
= 0;
1289 fence
->execbuf
.batch_len
= batch
.next
- fence
->bo
.map
;
1290 fence
->execbuf
.cliprects_ptr
= 0;
1291 fence
->execbuf
.num_cliprects
= 0;
1292 fence
->execbuf
.DR1
= 0;
1293 fence
->execbuf
.DR4
= 0;
1295 fence
->execbuf
.flags
=
1296 I915_EXEC_HANDLE_LUT
| I915_EXEC_NO_RELOC
| I915_EXEC_RENDER
;
1297 fence
->execbuf
.rsvd1
= device
->context_id
;
1298 fence
->execbuf
.rsvd2
= 0;
1300 *pFence
= (VkQueryPool
) fence
;
1305 anv_device_free(device
, fence
);
1310 VkResult
anv_ResetFences(
1312 uint32_t fenceCount
,
1315 struct anv_fence
**fences
= (struct anv_fence
**) pFences
;
1317 for (uint32_t i
; i
< fenceCount
; i
++)
1318 fences
[i
]->ready
= false;
1323 VkResult
anv_GetFenceStatus(
1327 struct anv_device
*device
= (struct anv_device
*) _device
;
1328 struct anv_fence
*fence
= (struct anv_fence
*) _fence
;
1335 ret
= anv_gem_wait(device
, fence
->bo
.gem_handle
, &t
);
1337 fence
->ready
= true;
1341 return VK_NOT_READY
;
1344 VkResult
anv_WaitForFences(
1346 uint32_t fenceCount
,
1347 const VkFence
* pFences
,
1351 struct anv_device
*device
= (struct anv_device
*) _device
;
1352 struct anv_fence
**fences
= (struct anv_fence
**) pFences
;
1353 int64_t t
= timeout
;
1356 /* FIXME: handle !waitAll */
1358 for (uint32_t i
= 0; i
< fenceCount
; i
++) {
1359 ret
= anv_gem_wait(device
, fences
[i
]->bo
.gem_handle
, &t
);
1360 if (ret
== -1 && errno
== ETIME
)
1363 return vk_error(VK_ERROR_UNKNOWN
);
1369 // Queue semaphore functions
1371 VkResult
anv_CreateSemaphore(
1373 const VkSemaphoreCreateInfo
* pCreateInfo
,
1374 VkSemaphore
* pSemaphore
)
1376 stub_return(VK_UNSUPPORTED
);
1379 VkResult
anv_QueueSignalSemaphore(
1381 VkSemaphore semaphore
)
1383 stub_return(VK_UNSUPPORTED
);
1386 VkResult
anv_QueueWaitSemaphore(
1388 VkSemaphore semaphore
)
1390 stub_return(VK_UNSUPPORTED
);
1395 VkResult
anv_CreateEvent(
1397 const VkEventCreateInfo
* pCreateInfo
,
1400 stub_return(VK_UNSUPPORTED
);
1403 VkResult
anv_GetEventStatus(
1407 stub_return(VK_UNSUPPORTED
);
1410 VkResult
anv_SetEvent(
1414 stub_return(VK_UNSUPPORTED
);
1417 VkResult
anv_ResetEvent(
1421 stub_return(VK_UNSUPPORTED
);
1427 anv_query_pool_destroy(struct anv_device
*device
,
1428 struct anv_object
*object
,
1429 VkObjectType obj_type
)
1431 struct anv_query_pool
*pool
= (struct anv_query_pool
*) object
;
1433 assert(obj_type
== VK_OBJECT_TYPE_QUERY_POOL
);
1435 anv_gem_munmap(pool
->bo
.map
, pool
->bo
.size
);
1436 anv_gem_close(device
, pool
->bo
.gem_handle
);
1437 anv_device_free(device
, pool
);
1440 VkResult
anv_CreateQueryPool(
1442 const VkQueryPoolCreateInfo
* pCreateInfo
,
1443 VkQueryPool
* pQueryPool
)
1445 struct anv_device
*device
= (struct anv_device
*) _device
;
1446 struct anv_query_pool
*pool
;
1450 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_QUERY_POOL_CREATE_INFO
);
1452 switch (pCreateInfo
->queryType
) {
1453 case VK_QUERY_TYPE_OCCLUSION
:
1455 case VK_QUERY_TYPE_PIPELINE_STATISTICS
:
1456 return VK_UNSUPPORTED
;
1461 pool
= anv_device_alloc(device
, sizeof(*pool
), 8,
1462 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
1464 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1466 pool
->base
.destructor
= anv_query_pool_destroy
;
1468 pool
->type
= pCreateInfo
->queryType
;
1469 size
= pCreateInfo
->slots
* sizeof(struct anv_query_pool_slot
);
1470 result
= anv_bo_init_new(&pool
->bo
, device
, size
);
1471 if (result
!= VK_SUCCESS
)
1474 pool
->bo
.map
= anv_gem_mmap(device
, pool
->bo
.gem_handle
, 0, size
);
1476 *pQueryPool
= (VkQueryPool
) pool
;
1481 anv_device_free(device
, pool
);
1486 VkResult
anv_GetQueryPoolResults(
1488 VkQueryPool queryPool
,
1489 uint32_t startQuery
,
1490 uint32_t queryCount
,
1493 VkQueryResultFlags flags
)
1495 struct anv_device
*device
= (struct anv_device
*) _device
;
1496 struct anv_query_pool
*pool
= (struct anv_query_pool
*) queryPool
;
1497 struct anv_query_pool_slot
*slot
= pool
->bo
.map
;
1498 int64_t timeout
= INT64_MAX
;
1499 uint32_t *dst32
= pData
;
1500 uint64_t *dst64
= pData
;
1504 if (flags
& VK_QUERY_RESULT_WITH_AVAILABILITY_BIT
) {
1505 /* Where is the availabilty info supposed to go? */
1506 anv_finishme("VK_QUERY_RESULT_WITH_AVAILABILITY_BIT");
1507 return VK_UNSUPPORTED
;
1510 assert(pool
->type
== VK_QUERY_TYPE_OCCLUSION
);
1512 if (flags
& VK_QUERY_RESULT_64_BIT
)
1513 *pDataSize
= queryCount
* sizeof(uint64_t);
1515 *pDataSize
= queryCount
* sizeof(uint32_t);
1520 if (flags
& VK_QUERY_RESULT_WAIT_BIT
) {
1521 ret
= anv_gem_wait(device
, pool
->bo
.gem_handle
, &timeout
);
1523 return vk_error(VK_ERROR_UNKNOWN
);
1526 for (uint32_t i
= 0; i
< queryCount
; i
++) {
1527 result
= slot
[startQuery
+ i
].end
- slot
[startQuery
+ i
].begin
;
1528 if (flags
& VK_QUERY_RESULT_64_BIT
) {
1531 if (result
> UINT32_MAX
)
1532 result
= UINT32_MAX
;
1542 VkResult
anv_CreateBuffer(
1544 const VkBufferCreateInfo
* pCreateInfo
,
1547 struct anv_device
*device
= (struct anv_device
*) _device
;
1548 struct anv_buffer
*buffer
;
1550 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO
);
1552 buffer
= anv_device_alloc(device
, sizeof(*buffer
), 8,
1553 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
1555 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1557 buffer
->size
= pCreateInfo
->size
;
1561 *pBuffer
= (VkBuffer
) buffer
;
1566 // Buffer view functions
1568 VkResult
anv_CreateBufferView(
1570 const VkBufferViewCreateInfo
* pCreateInfo
,
1571 VkBufferView
* pView
)
1573 struct anv_device
*device
= (struct anv_device
*) _device
;
1574 struct anv_buffer
*buffer
= (struct anv_buffer
*) pCreateInfo
->buffer
;
1575 struct anv_surface_view
*view
;
1576 const struct anv_format
*format
;
1578 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO
);
1580 view
= anv_device_alloc(device
, sizeof(*view
), 8,
1581 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
1583 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1585 view
->bo
= buffer
->bo
;
1586 view
->offset
= buffer
->offset
+ pCreateInfo
->offset
;
1587 view
->surface_state
=
1588 anv_state_pool_alloc(&device
->surface_state_pool
, 64, 64);
1589 view
->format
= pCreateInfo
->format
;
1591 format
= anv_format_for_vk_format(pCreateInfo
->format
);
1592 /* This assumes RGBA float format. */
1593 uint32_t stride
= 4;
1594 uint32_t num_elements
= pCreateInfo
->range
/ stride
;
1595 struct GEN8_RENDER_SURFACE_STATE surface_state
= {
1596 .SurfaceType
= SURFTYPE_BUFFER
,
1597 .SurfaceArray
= false,
1598 .SurfaceFormat
= format
->format
,
1599 .SurfaceVerticalAlignment
= VALIGN4
,
1600 .SurfaceHorizontalAlignment
= HALIGN4
,
1602 .VerticalLineStride
= 0,
1603 .VerticalLineStrideOffset
= 0,
1604 .SamplerL2BypassModeDisable
= true,
1605 .RenderCacheReadWriteMode
= WriteOnlyCache
,
1606 .MemoryObjectControlState
= GEN8_MOCS
,
1609 .Height
= (num_elements
>> 7) & 0x3fff,
1610 .Width
= num_elements
& 0x7f,
1611 .Depth
= (num_elements
>> 21) & 0x3f,
1612 .SurfacePitch
= stride
- 1,
1613 .MinimumArrayElement
= 0,
1614 .NumberofMultisamples
= MULTISAMPLECOUNT_1
,
1619 .AuxiliarySurfaceMode
= AUX_NONE
,
1621 .GreenClearColor
= 0,
1622 .BlueClearColor
= 0,
1623 .AlphaClearColor
= 0,
1624 .ShaderChannelSelectRed
= SCS_RED
,
1625 .ShaderChannelSelectGreen
= SCS_GREEN
,
1626 .ShaderChannelSelectBlue
= SCS_BLUE
,
1627 .ShaderChannelSelectAlpha
= SCS_ALPHA
,
1628 .ResourceMinLOD
= 0,
1629 /* FIXME: We assume that the image must be bound at this time. */
1630 .SurfaceBaseAddress
= { NULL
, view
->offset
},
1633 GEN8_RENDER_SURFACE_STATE_pack(NULL
, view
->surface_state
.map
, &surface_state
);
1635 *pView
= (VkImageView
) view
;
1640 // Sampler functions
1642 VkResult
anv_CreateSampler(
1644 const VkSamplerCreateInfo
* pCreateInfo
,
1645 VkSampler
* pSampler
)
1647 struct anv_device
*device
= (struct anv_device
*) _device
;
1648 struct anv_sampler
*sampler
;
1650 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO
);
1652 sampler
= anv_device_alloc(device
, sizeof(*sampler
), 8,
1653 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
1655 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1657 static const uint32_t vk_to_gen_tex_filter
[] = {
1658 [VK_TEX_FILTER_NEAREST
] = MAPFILTER_NEAREST
,
1659 [VK_TEX_FILTER_LINEAR
] = MAPFILTER_LINEAR
1662 static const uint32_t vk_to_gen_mipmap_mode
[] = {
1663 [VK_TEX_MIPMAP_MODE_BASE
] = MIPFILTER_NONE
,
1664 [VK_TEX_MIPMAP_MODE_NEAREST
] = MIPFILTER_NEAREST
,
1665 [VK_TEX_MIPMAP_MODE_LINEAR
] = MIPFILTER_LINEAR
1668 static const uint32_t vk_to_gen_tex_address
[] = {
1669 [VK_TEX_ADDRESS_WRAP
] = TCM_WRAP
,
1670 [VK_TEX_ADDRESS_MIRROR
] = TCM_MIRROR
,
1671 [VK_TEX_ADDRESS_CLAMP
] = TCM_CLAMP
,
1672 [VK_TEX_ADDRESS_MIRROR_ONCE
] = TCM_MIRROR_ONCE
,
1673 [VK_TEX_ADDRESS_CLAMP_BORDER
] = TCM_CLAMP_BORDER
,
1676 static const uint32_t vk_to_gen_compare_op
[] = {
1677 [VK_COMPARE_OP_NEVER
] = PREFILTEROPNEVER
,
1678 [VK_COMPARE_OP_LESS
] = PREFILTEROPLESS
,
1679 [VK_COMPARE_OP_EQUAL
] = PREFILTEROPEQUAL
,
1680 [VK_COMPARE_OP_LESS_EQUAL
] = PREFILTEROPLEQUAL
,
1681 [VK_COMPARE_OP_GREATER
] = PREFILTEROPGREATER
,
1682 [VK_COMPARE_OP_NOT_EQUAL
] = PREFILTEROPNOTEQUAL
,
1683 [VK_COMPARE_OP_GREATER_EQUAL
] = PREFILTEROPGEQUAL
,
1684 [VK_COMPARE_OP_ALWAYS
] = PREFILTEROPALWAYS
,
1687 if (pCreateInfo
->maxAnisotropy
> 0)
1688 anv_finishme("missing support for anisotropic filtering");
1690 struct GEN8_SAMPLER_STATE sampler_state
= {
1691 .SamplerDisable
= false,
1692 .TextureBorderColorMode
= DX10OGL
,
1693 .LODPreClampMode
= 0,
1695 .MipModeFilter
= vk_to_gen_mipmap_mode
[pCreateInfo
->mipMode
],
1696 .MagModeFilter
= vk_to_gen_tex_filter
[pCreateInfo
->magFilter
],
1697 .MinModeFilter
= vk_to_gen_tex_filter
[pCreateInfo
->minFilter
],
1698 .TextureLODBias
= pCreateInfo
->mipLodBias
* 256,
1699 .AnisotropicAlgorithm
= EWAApproximation
,
1700 .MinLOD
= pCreateInfo
->minLod
* 256,
1701 .MaxLOD
= pCreateInfo
->maxLod
* 256,
1702 .ChromaKeyEnable
= 0,
1703 .ChromaKeyIndex
= 0,
1705 .ShadowFunction
= vk_to_gen_compare_op
[pCreateInfo
->compareOp
],
1706 .CubeSurfaceControlMode
= 0,
1707 .IndirectStatePointer
= 0,
1708 .LODClampMagnificationMode
= MIPNONE
,
1709 .MaximumAnisotropy
= 0,
1710 .RAddressMinFilterRoundingEnable
= 0,
1711 .RAddressMagFilterRoundingEnable
= 0,
1712 .VAddressMinFilterRoundingEnable
= 0,
1713 .VAddressMagFilterRoundingEnable
= 0,
1714 .UAddressMinFilterRoundingEnable
= 0,
1715 .UAddressMagFilterRoundingEnable
= 0,
1716 .TrilinearFilterQuality
= 0,
1717 .NonnormalizedCoordinateEnable
= 0,
1718 .TCXAddressControlMode
= vk_to_gen_tex_address
[pCreateInfo
->addressU
],
1719 .TCYAddressControlMode
= vk_to_gen_tex_address
[pCreateInfo
->addressV
],
1720 .TCZAddressControlMode
= vk_to_gen_tex_address
[pCreateInfo
->addressW
],
1723 GEN8_SAMPLER_STATE_pack(NULL
, sampler
->state
, &sampler_state
);
1725 *pSampler
= (VkSampler
) sampler
;
1730 // Descriptor set functions
1732 VkResult
anv_CreateDescriptorSetLayout(
1734 const VkDescriptorSetLayoutCreateInfo
* pCreateInfo
,
1735 VkDescriptorSetLayout
* pSetLayout
)
1737 struct anv_device
*device
= (struct anv_device
*) _device
;
1738 struct anv_descriptor_set_layout
*set_layout
;
1740 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO
);
1742 uint32_t sampler_count
[VK_NUM_SHADER_STAGE
] = { 0, };
1743 uint32_t surface_count
[VK_NUM_SHADER_STAGE
] = { 0, };
1744 uint32_t num_dynamic_buffers
= 0;
1748 for (uint32_t i
= 0; i
< pCreateInfo
->count
; i
++) {
1749 switch (pCreateInfo
->pBinding
[i
].descriptorType
) {
1750 case VK_DESCRIPTOR_TYPE_SAMPLER
:
1751 for_each_bit(s
, pCreateInfo
->pBinding
[i
].stageFlags
)
1752 sampler_count
[s
] += pCreateInfo
->pBinding
[i
].count
;
1755 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER
:
1756 for_each_bit(s
, pCreateInfo
->pBinding
[i
].stageFlags
)
1757 sampler_count
[s
] += pCreateInfo
->pBinding
[i
].count
;
1761 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE
:
1762 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE
:
1763 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER
:
1764 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER
:
1765 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER
:
1766 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER
:
1767 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC
:
1768 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC
:
1769 for_each_bit(s
, pCreateInfo
->pBinding
[i
].stageFlags
)
1770 surface_count
[s
] += pCreateInfo
->pBinding
[i
].count
;
1776 count
+= pCreateInfo
->pBinding
[i
].count
;
1779 for (uint32_t i
= 0; i
< pCreateInfo
->count
; i
++) {
1780 switch (pCreateInfo
->pBinding
[i
].descriptorType
) {
1781 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC
:
1782 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC
:
1783 num_dynamic_buffers
++;
1790 uint32_t sampler_total
= 0;
1791 uint32_t surface_total
= 0;
1792 for (uint32_t s
= 0; s
< VK_NUM_SHADER_STAGE
; s
++) {
1793 sampler_total
+= sampler_count
[s
];
1794 surface_total
+= surface_count
[s
];
1797 size_t size
= sizeof(*set_layout
) +
1798 (sampler_total
+ surface_total
) * sizeof(uint32_t);
1799 set_layout
= anv_device_alloc(device
, size
, 8,
1800 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
1802 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1804 set_layout
->num_dynamic_buffers
= num_dynamic_buffers
;
1805 set_layout
->count
= count
;
1807 uint32_t *p
= set_layout
->entries
;
1808 uint32_t *sampler
[VK_NUM_SHADER_STAGE
];
1809 uint32_t *surface
[VK_NUM_SHADER_STAGE
];
1810 for (uint32_t s
= 0; s
< VK_NUM_SHADER_STAGE
; s
++) {
1811 set_layout
->stage
[s
].surface_count
= surface_count
[s
];
1812 set_layout
->stage
[s
].surface_start
= surface
[s
] = p
;
1813 p
+= surface_count
[s
];
1814 set_layout
->stage
[s
].sampler_count
= sampler_count
[s
];
1815 set_layout
->stage
[s
].sampler_start
= sampler
[s
] = p
;
1816 p
+= sampler_count
[s
];
1819 uint32_t descriptor
= 0;
1820 for (uint32_t i
= 0; i
< pCreateInfo
->count
; i
++) {
1821 switch (pCreateInfo
->pBinding
[i
].descriptorType
) {
1822 case VK_DESCRIPTOR_TYPE_SAMPLER
:
1823 for_each_bit(s
, pCreateInfo
->pBinding
[i
].stageFlags
)
1824 for (uint32_t j
= 0; j
< pCreateInfo
->pBinding
[i
].count
; j
++)
1825 *(sampler
[s
])++ = descriptor
+ j
;
1828 case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER
:
1829 for_each_bit(s
, pCreateInfo
->pBinding
[i
].stageFlags
)
1830 for (uint32_t j
= 0; j
< pCreateInfo
->pBinding
[i
].count
; j
++)
1831 *(sampler
[s
])++ = descriptor
+ j
;
1835 case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE
:
1836 case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE
:
1837 case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER
:
1838 case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER
:
1839 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER
:
1840 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER
:
1841 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC
:
1842 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC
:
1843 for_each_bit(s
, pCreateInfo
->pBinding
[i
].stageFlags
)
1844 for (uint32_t j
= 0; j
< pCreateInfo
->pBinding
[i
].count
; j
++) {
1845 *(surface
[s
])++ = descriptor
+ j
;
1851 descriptor
+= pCreateInfo
->pBinding
[i
].count
;
1854 *pSetLayout
= (VkDescriptorSetLayout
) set_layout
;
1859 VkResult
anv_BeginDescriptorPoolUpdate(
1861 VkDescriptorUpdateMode updateMode
)
1866 VkResult
anv_EndDescriptorPoolUpdate(
1873 VkResult
anv_CreateDescriptorPool(
1875 VkDescriptorPoolUsage poolUsage
,
1877 const VkDescriptorPoolCreateInfo
* pCreateInfo
,
1878 VkDescriptorPool
* pDescriptorPool
)
1880 *pDescriptorPool
= 1;
1885 VkResult
anv_ResetDescriptorPool(
1887 VkDescriptorPool descriptorPool
)
1892 VkResult
anv_AllocDescriptorSets(
1894 VkDescriptorPool descriptorPool
,
1895 VkDescriptorSetUsage setUsage
,
1897 const VkDescriptorSetLayout
* pSetLayouts
,
1898 VkDescriptorSet
* pDescriptorSets
,
1901 struct anv_device
*device
= (struct anv_device
*) _device
;
1902 const struct anv_descriptor_set_layout
*layout
;
1903 struct anv_descriptor_set
*set
;
1906 for (uint32_t i
= 0; i
< count
; i
++) {
1907 layout
= (struct anv_descriptor_set_layout
*) pSetLayouts
[i
];
1908 size
= sizeof(*set
) + layout
->count
* sizeof(set
->descriptors
[0]);
1909 set
= anv_device_alloc(device
, size
, 8,
1910 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
1913 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
1916 /* Descriptor sets may not be 100% filled out so we need to memset to
1917 * ensure that we can properly detect and handle holes.
1919 memset(set
, 0, size
);
1921 pDescriptorSets
[i
] = (VkDescriptorSet
) set
;
1929 void anv_ClearDescriptorSets(
1931 VkDescriptorPool descriptorPool
,
1933 const VkDescriptorSet
* pDescriptorSets
)
1937 void anv_UpdateDescriptors(
1939 VkDescriptorSet descriptorSet
,
1940 uint32_t updateCount
,
1941 const void** ppUpdateArray
)
1943 struct anv_descriptor_set
*set
= (struct anv_descriptor_set
*) descriptorSet
;
1944 VkUpdateSamplers
*update_samplers
;
1945 VkUpdateSamplerTextures
*update_sampler_textures
;
1946 VkUpdateImages
*update_images
;
1947 VkUpdateBuffers
*update_buffers
;
1948 VkUpdateAsCopy
*update_as_copy
;
1950 for (uint32_t i
= 0; i
< updateCount
; i
++) {
1951 const struct anv_common
*common
= ppUpdateArray
[i
];
1953 switch (common
->sType
) {
1954 case VK_STRUCTURE_TYPE_UPDATE_SAMPLERS
:
1955 update_samplers
= (VkUpdateSamplers
*) common
;
1957 for (uint32_t j
= 0; j
< update_samplers
->count
; j
++) {
1958 set
->descriptors
[update_samplers
->binding
+ j
].sampler
=
1959 (struct anv_sampler
*) update_samplers
->pSamplers
[j
];
1963 case VK_STRUCTURE_TYPE_UPDATE_SAMPLER_TEXTURES
:
1964 /* FIXME: Shouldn't this be *_UPDATE_SAMPLER_IMAGES? */
1965 update_sampler_textures
= (VkUpdateSamplerTextures
*) common
;
1967 for (uint32_t j
= 0; j
< update_sampler_textures
->count
; j
++) {
1968 set
->descriptors
[update_sampler_textures
->binding
+ j
].view
=
1969 (struct anv_surface_view
*)
1970 update_sampler_textures
->pSamplerImageViews
[j
].pImageView
->view
;
1971 set
->descriptors
[update_sampler_textures
->binding
+ j
].sampler
=
1972 (struct anv_sampler
*)
1973 update_sampler_textures
->pSamplerImageViews
[j
].sampler
;
1977 case VK_STRUCTURE_TYPE_UPDATE_IMAGES
:
1978 update_images
= (VkUpdateImages
*) common
;
1980 for (uint32_t j
= 0; j
< update_images
->count
; j
++) {
1981 set
->descriptors
[update_images
->binding
+ j
].view
=
1982 (struct anv_surface_view
*) update_images
->pImageViews
[j
].view
;
1986 case VK_STRUCTURE_TYPE_UPDATE_BUFFERS
:
1987 update_buffers
= (VkUpdateBuffers
*) common
;
1989 for (uint32_t j
= 0; j
< update_buffers
->count
; j
++) {
1990 set
->descriptors
[update_buffers
->binding
+ j
].view
=
1991 (struct anv_surface_view
*) update_buffers
->pBufferViews
[j
].view
;
1993 /* FIXME: descriptor arrays? */
1996 case VK_STRUCTURE_TYPE_UPDATE_AS_COPY
:
1997 update_as_copy
= (VkUpdateAsCopy
*) common
;
1998 (void) update_as_copy
;
2007 // State object functions
2009 static inline int64_t
2010 clamp_int64(int64_t x
, int64_t min
, int64_t max
)
2021 anv_dynamic_vp_state_destroy(struct anv_device
*device
,
2022 struct anv_object
*object
,
2023 VkObjectType obj_type
)
2025 struct anv_dynamic_vp_state
*state
= (void *)object
;
2027 assert(obj_type
== VK_OBJECT_TYPE_DYNAMIC_VP_STATE
);
2029 anv_state_pool_free(&device
->dynamic_state_pool
, state
->sf_clip_vp
);
2030 anv_state_pool_free(&device
->dynamic_state_pool
, state
->cc_vp
);
2031 anv_state_pool_free(&device
->dynamic_state_pool
, state
->scissor
);
2033 anv_device_free(device
, state
);
2036 VkResult
anv_CreateDynamicViewportState(
2038 const VkDynamicVpStateCreateInfo
* pCreateInfo
,
2039 VkDynamicVpState
* pState
)
2041 struct anv_device
*device
= (struct anv_device
*) _device
;
2042 struct anv_dynamic_vp_state
*state
;
2044 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_DYNAMIC_VP_STATE_CREATE_INFO
);
2046 state
= anv_device_alloc(device
, sizeof(*state
), 8,
2047 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
2049 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
2051 state
->base
.destructor
= anv_dynamic_vp_state_destroy
;
2053 unsigned count
= pCreateInfo
->viewportAndScissorCount
;
2054 state
->sf_clip_vp
= anv_state_pool_alloc(&device
->dynamic_state_pool
,
2056 state
->cc_vp
= anv_state_pool_alloc(&device
->dynamic_state_pool
,
2058 state
->scissor
= anv_state_pool_alloc(&device
->dynamic_state_pool
,
2061 for (uint32_t i
= 0; i
< pCreateInfo
->viewportAndScissorCount
; i
++) {
2062 const VkViewport
*vp
= &pCreateInfo
->pViewports
[i
];
2063 const VkRect
*s
= &pCreateInfo
->pScissors
[i
];
2065 struct GEN8_SF_CLIP_VIEWPORT sf_clip_viewport
= {
2066 .ViewportMatrixElementm00
= vp
->width
/ 2,
2067 .ViewportMatrixElementm11
= vp
->height
/ 2,
2068 .ViewportMatrixElementm22
= (vp
->maxDepth
- vp
->minDepth
) / 2,
2069 .ViewportMatrixElementm30
= vp
->originX
+ vp
->width
/ 2,
2070 .ViewportMatrixElementm31
= vp
->originY
+ vp
->height
/ 2,
2071 .ViewportMatrixElementm32
= (vp
->maxDepth
+ vp
->minDepth
) / 2,
2072 .XMinClipGuardband
= -1.0f
,
2073 .XMaxClipGuardband
= 1.0f
,
2074 .YMinClipGuardband
= -1.0f
,
2075 .YMaxClipGuardband
= 1.0f
,
2076 .XMinViewPort
= vp
->originX
,
2077 .XMaxViewPort
= vp
->originX
+ vp
->width
- 1,
2078 .YMinViewPort
= vp
->originY
,
2079 .YMaxViewPort
= vp
->originY
+ vp
->height
- 1,
2082 struct GEN8_CC_VIEWPORT cc_viewport
= {
2083 .MinimumDepth
= vp
->minDepth
,
2084 .MaximumDepth
= vp
->maxDepth
2087 /* Since xmax and ymax are inclusive, we have to have xmax < xmin or
2088 * ymax < ymin for empty clips. In case clip x, y, width height are all
2089 * 0, the clamps below produce 0 for xmin, ymin, xmax, ymax, which isn't
2090 * what we want. Just special case empty clips and produce a canonical
2092 static const struct GEN8_SCISSOR_RECT empty_scissor
= {
2093 .ScissorRectangleYMin
= 1,
2094 .ScissorRectangleXMin
= 1,
2095 .ScissorRectangleYMax
= 0,
2096 .ScissorRectangleXMax
= 0
2099 const int max
= 0xffff;
2100 struct GEN8_SCISSOR_RECT scissor
= {
2101 /* Do this math using int64_t so overflow gets clamped correctly. */
2102 .ScissorRectangleYMin
= clamp_int64(s
->offset
.y
, 0, max
),
2103 .ScissorRectangleXMin
= clamp_int64(s
->offset
.x
, 0, max
),
2104 .ScissorRectangleYMax
= clamp_int64((uint64_t) s
->offset
.y
+ s
->extent
.height
- 1, 0, max
),
2105 .ScissorRectangleXMax
= clamp_int64((uint64_t) s
->offset
.x
+ s
->extent
.width
- 1, 0, max
)
2108 GEN8_SF_CLIP_VIEWPORT_pack(NULL
, state
->sf_clip_vp
.map
+ i
* 64, &sf_clip_viewport
);
2109 GEN8_CC_VIEWPORT_pack(NULL
, state
->cc_vp
.map
+ i
* 32, &cc_viewport
);
2111 if (s
->extent
.width
<= 0 || s
->extent
.height
<= 0) {
2112 GEN8_SCISSOR_RECT_pack(NULL
, state
->scissor
.map
+ i
* 32, &empty_scissor
);
2114 GEN8_SCISSOR_RECT_pack(NULL
, state
->scissor
.map
+ i
* 32, &scissor
);
2118 *pState
= (VkDynamicVpState
) state
;
2123 VkResult
anv_CreateDynamicRasterState(
2125 const VkDynamicRsStateCreateInfo
* pCreateInfo
,
2126 VkDynamicRsState
* pState
)
2128 struct anv_device
*device
= (struct anv_device
*) _device
;
2129 struct anv_dynamic_rs_state
*state
;
2131 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_DYNAMIC_RS_STATE_CREATE_INFO
);
2133 state
= anv_device_alloc(device
, sizeof(*state
), 8,
2134 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
2136 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
2139 * float pointFadeThreshold;
2140 * // optional (GL45) - Size of point fade threshold
2143 struct GEN8_3DSTATE_SF sf
= {
2144 GEN8_3DSTATE_SF_header
,
2145 .LineWidth
= pCreateInfo
->lineWidth
,
2146 .PointWidth
= pCreateInfo
->pointSize
,
2149 GEN8_3DSTATE_SF_pack(NULL
, state
->state_sf
, &sf
);
2151 bool enable_bias
= pCreateInfo
->depthBias
!= 0.0f
||
2152 pCreateInfo
->slopeScaledDepthBias
!= 0.0f
;
2153 struct GEN8_3DSTATE_RASTER raster
= {
2154 .GlobalDepthOffsetEnableSolid
= enable_bias
,
2155 .GlobalDepthOffsetEnableWireframe
= enable_bias
,
2156 .GlobalDepthOffsetEnablePoint
= enable_bias
,
2157 .GlobalDepthOffsetConstant
= pCreateInfo
->depthBias
,
2158 .GlobalDepthOffsetScale
= pCreateInfo
->slopeScaledDepthBias
,
2159 .GlobalDepthOffsetClamp
= pCreateInfo
->depthBiasClamp
2162 GEN8_3DSTATE_RASTER_pack(NULL
, state
->state_raster
, &raster
);
2164 *pState
= (VkDynamicRsState
) state
;
2169 VkResult
anv_CreateDynamicColorBlendState(
2171 const VkDynamicCbStateCreateInfo
* pCreateInfo
,
2172 VkDynamicCbState
* pState
)
2174 struct anv_device
*device
= (struct anv_device
*) _device
;
2175 struct anv_dynamic_cb_state
*state
;
2177 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_DYNAMIC_CB_STATE_CREATE_INFO
);
2179 state
= anv_device_alloc(device
, sizeof(*state
), 8,
2180 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
2182 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
2184 struct GEN8_COLOR_CALC_STATE color_calc_state
= {
2185 .BlendConstantColorRed
= pCreateInfo
->blendConst
[0],
2186 .BlendConstantColorGreen
= pCreateInfo
->blendConst
[1],
2187 .BlendConstantColorBlue
= pCreateInfo
->blendConst
[2],
2188 .BlendConstantColorAlpha
= pCreateInfo
->blendConst
[3]
2191 GEN8_COLOR_CALC_STATE_pack(NULL
, state
->state_color_calc
, &color_calc_state
);
2193 *pState
= (VkDynamicCbState
) state
;
2198 VkResult
anv_CreateDynamicDepthStencilState(
2200 const VkDynamicDsStateCreateInfo
* pCreateInfo
,
2201 VkDynamicDsState
* pState
)
2203 struct anv_device
*device
= (struct anv_device
*) _device
;
2204 struct anv_dynamic_ds_state
*state
;
2206 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_DYNAMIC_DS_STATE_CREATE_INFO
);
2208 state
= anv_device_alloc(device
, sizeof(*state
), 8,
2209 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
2211 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
2213 struct GEN8_3DSTATE_WM_DEPTH_STENCIL wm_depth_stencil
= {
2214 GEN8_3DSTATE_WM_DEPTH_STENCIL_header
,
2216 /* Is this what we need to do? */
2217 .StencilBufferWriteEnable
= pCreateInfo
->stencilWriteMask
!= 0,
2219 .StencilTestMask
= pCreateInfo
->stencilReadMask
,
2220 .StencilWriteMask
= pCreateInfo
->stencilWriteMask
,
2222 .BackfaceStencilTestMask
= pCreateInfo
->stencilReadMask
,
2223 .BackfaceStencilWriteMask
= pCreateInfo
->stencilWriteMask
,
2226 GEN8_3DSTATE_WM_DEPTH_STENCIL_pack(NULL
, state
->state_wm_depth_stencil
,
2229 struct GEN8_COLOR_CALC_STATE color_calc_state
= {
2230 .StencilReferenceValue
= pCreateInfo
->stencilFrontRef
,
2231 .BackFaceStencilReferenceValue
= pCreateInfo
->stencilBackRef
2234 GEN8_COLOR_CALC_STATE_pack(NULL
, state
->state_color_calc
, &color_calc_state
);
2236 *pState
= (VkDynamicDsState
) state
;
2241 // Command buffer functions
2244 anv_cmd_buffer_destroy(struct anv_device
*device
,
2245 struct anv_object
*object
,
2246 VkObjectType obj_type
)
2248 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*) object
;
2250 assert(obj_type
== VK_OBJECT_TYPE_COMMAND_BUFFER
);
2252 /* Destroy all of the batch buffers */
2253 struct anv_batch_bo
*bbo
= cmd_buffer
->last_batch_bo
;
2254 while (bbo
->prev_batch_bo
) {
2255 struct anv_batch_bo
*prev
= bbo
->prev_batch_bo
;
2256 anv_batch_bo_destroy(bbo
, cmd_buffer
->device
);
2260 anv_bo_pool_free(&device
->batch_bo_pool
, &cmd_buffer
->surface_bo
);
2261 anv_reloc_list_finish(&cmd_buffer
->surface_relocs
, device
);
2262 anv_state_stream_finish(&cmd_buffer
->surface_state_stream
);
2263 anv_state_stream_finish(&cmd_buffer
->dynamic_state_stream
);
2264 anv_state_stream_finish(&cmd_buffer
->binding_table_state_stream
);
2265 anv_reloc_list_finish(&cmd_buffer
->batch
.relocs
, device
);
2266 anv_device_free(device
, cmd_buffer
->exec2_objects
);
2267 anv_device_free(device
, cmd_buffer
->exec2_bos
);
2268 anv_device_free(device
, cmd_buffer
);
2272 anv_cmd_buffer_chain_batch(struct anv_batch
*batch
, void *_data
)
2274 struct anv_cmd_buffer
*cmd_buffer
= _data
;
2276 struct anv_batch_bo
*new_bbo
, *old_bbo
= cmd_buffer
->last_batch_bo
;
2278 VkResult result
= anv_batch_bo_create(cmd_buffer
->device
, &new_bbo
);
2279 if (result
!= VK_SUCCESS
)
2282 /* We set the end of the batch a little short so we would be sure we
2283 * have room for the chaining command. Since we're about to emit the
2284 * chaining command, let's set it back where it should go.
2286 batch
->end
+= GEN8_MI_BATCH_BUFFER_START_length
* 4;
2287 assert(batch
->end
== old_bbo
->bo
.map
+ old_bbo
->bo
.size
);
2289 anv_batch_emit(batch
, GEN8_MI_BATCH_BUFFER_START
,
2290 GEN8_MI_BATCH_BUFFER_START_header
,
2291 ._2ndLevelBatchBuffer
= _1stlevelbatch
,
2292 .AddressSpaceIndicator
= ASI_PPGTT
,
2293 .BatchBufferStartAddress
= { &new_bbo
->bo
, 0 },
2296 /* Pad out to a 2-dword aligned boundary with zeros */
2297 if ((uintptr_t)batch
->next
% 8 != 0) {
2298 *(uint32_t *)batch
->next
= 0;
2302 anv_batch_bo_finish(cmd_buffer
->last_batch_bo
, batch
);
2304 new_bbo
->prev_batch_bo
= old_bbo
;
2305 cmd_buffer
->last_batch_bo
= new_bbo
;
2307 anv_batch_bo_start(new_bbo
, batch
, GEN8_MI_BATCH_BUFFER_START_length
* 4);
2312 VkResult
anv_CreateCommandBuffer(
2314 const VkCmdBufferCreateInfo
* pCreateInfo
,
2315 VkCmdBuffer
* pCmdBuffer
)
2317 struct anv_device
*device
= (struct anv_device
*) _device
;
2318 struct anv_cmd_buffer
*cmd_buffer
;
2321 cmd_buffer
= anv_device_alloc(device
, sizeof(*cmd_buffer
), 8,
2322 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
2323 if (cmd_buffer
== NULL
)
2324 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
2326 cmd_buffer
->base
.destructor
= anv_cmd_buffer_destroy
;
2328 cmd_buffer
->device
= device
;
2329 cmd_buffer
->rs_state
= NULL
;
2330 cmd_buffer
->vp_state
= NULL
;
2331 memset(&cmd_buffer
->default_bindings
, 0, sizeof(cmd_buffer
->default_bindings
));
2332 cmd_buffer
->bindings
= &cmd_buffer
->default_bindings
;
2334 result
= anv_batch_bo_create(device
, &cmd_buffer
->last_batch_bo
);
2335 if (result
!= VK_SUCCESS
)
2338 result
= anv_reloc_list_init(&cmd_buffer
->batch
.relocs
, device
);
2339 if (result
!= VK_SUCCESS
)
2342 cmd_buffer
->batch
.device
= device
;
2343 cmd_buffer
->batch
.extend_cb
= anv_cmd_buffer_chain_batch
;
2344 cmd_buffer
->batch
.user_data
= cmd_buffer
;
2346 anv_batch_bo_start(cmd_buffer
->last_batch_bo
, &cmd_buffer
->batch
,
2347 GEN8_MI_BATCH_BUFFER_START_length
* 4);
2349 result
= anv_bo_pool_alloc(&device
->batch_bo_pool
, &cmd_buffer
->surface_bo
);
2350 if (result
!= VK_SUCCESS
)
2351 goto fail_batch_relocs
;
2353 /* Start surface_next at 1 so surface offset 0 is invalid. */
2354 cmd_buffer
->surface_next
= 1;
2355 anv_reloc_list_init(&cmd_buffer
->surface_relocs
, device
);
2357 cmd_buffer
->exec2_objects
= NULL
;
2358 cmd_buffer
->exec2_bos
= NULL
;
2359 cmd_buffer
->exec2_array_length
= 0;
2361 anv_state_stream_init(&cmd_buffer
->binding_table_state_stream
,
2362 &device
->binding_table_block_pool
);
2363 anv_state_stream_init(&cmd_buffer
->surface_state_stream
,
2364 &device
->surface_state_block_pool
);
2365 anv_state_stream_init(&cmd_buffer
->dynamic_state_stream
,
2366 &device
->dynamic_state_block_pool
);
2368 cmd_buffer
->dirty
= 0;
2369 cmd_buffer
->vb_dirty
= 0;
2370 cmd_buffer
->pipeline
= NULL
;
2371 cmd_buffer
->vp_state
= NULL
;
2372 cmd_buffer
->rs_state
= NULL
;
2373 cmd_buffer
->ds_state
= NULL
;
2375 *pCmdBuffer
= (VkCmdBuffer
) cmd_buffer
;
2380 anv_reloc_list_finish(&cmd_buffer
->batch
.relocs
, device
);
2382 anv_batch_bo_destroy(cmd_buffer
->last_batch_bo
, device
);
2384 anv_device_free(device
, cmd_buffer
);
2390 anv_cmd_buffer_emit_state_base_address(struct anv_cmd_buffer
*cmd_buffer
)
2392 struct anv_device
*device
= cmd_buffer
->device
;
2394 anv_batch_emit(&cmd_buffer
->batch
, GEN8_STATE_BASE_ADDRESS
,
2395 .GeneralStateBaseAddress
= { NULL
, 0 },
2396 .GeneralStateMemoryObjectControlState
= GEN8_MOCS
,
2397 .GeneralStateBaseAddressModifyEnable
= true,
2398 .GeneralStateBufferSize
= 0xfffff,
2399 .GeneralStateBufferSizeModifyEnable
= true,
2401 .SurfaceStateBaseAddress
= { &cmd_buffer
->surface_bo
, 0 },
2402 .SurfaceStateMemoryObjectControlState
= GEN8_MOCS
,
2403 .SurfaceStateBaseAddressModifyEnable
= true,
2405 .DynamicStateBaseAddress
= { &device
->dynamic_state_block_pool
.bo
, 0 },
2406 .DynamicStateMemoryObjectControlState
= GEN8_MOCS
,
2407 .DynamicStateBaseAddressModifyEnable
= true,
2408 .DynamicStateBufferSize
= 0xfffff,
2409 .DynamicStateBufferSizeModifyEnable
= true,
2411 .IndirectObjectBaseAddress
= { NULL
, 0 },
2412 .IndirectObjectMemoryObjectControlState
= GEN8_MOCS
,
2413 .IndirectObjectBaseAddressModifyEnable
= true,
2414 .IndirectObjectBufferSize
= 0xfffff,
2415 .IndirectObjectBufferSizeModifyEnable
= true,
2417 .InstructionBaseAddress
= { &device
->instruction_block_pool
.bo
, 0 },
2418 .InstructionMemoryObjectControlState
= GEN8_MOCS
,
2419 .InstructionBaseAddressModifyEnable
= true,
2420 .InstructionBufferSize
= 0xfffff,
2421 .InstructionBuffersizeModifyEnable
= true);
2424 VkResult
anv_BeginCommandBuffer(
2425 VkCmdBuffer cmdBuffer
,
2426 const VkCmdBufferBeginInfo
* pBeginInfo
)
2428 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*) cmdBuffer
;
2430 anv_batch_emit(&cmd_buffer
->batch
, GEN8_PIPELINE_SELECT
,
2431 .PipelineSelection
= _3D
);
2432 anv_batch_emit(&cmd_buffer
->batch
, GEN8_STATE_SIP
);
2434 anv_cmd_buffer_emit_state_base_address(cmd_buffer
);
2436 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_VF_STATISTICS
,
2437 .StatisticsEnable
= true);
2438 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_HS
, .Enable
= false);
2439 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_TE
, .TEEnable
= false);
2440 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_DS
, .FunctionEnable
= false);
2441 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_STREAMOUT
, .SOFunctionEnable
= false);
2443 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_VS
,
2444 .ConstantBufferOffset
= 0,
2445 .ConstantBufferSize
= 4);
2446 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_GS
,
2447 .ConstantBufferOffset
= 4,
2448 .ConstantBufferSize
= 4);
2449 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_PS
,
2450 .ConstantBufferOffset
= 8,
2451 .ConstantBufferSize
= 4);
2453 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_WM_CHROMAKEY
,
2454 .ChromaKeyKillEnable
= false);
2455 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_SBE_SWIZ
);
2456 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_AA_LINE_PARAMETERS
);
2462 anv_cmd_buffer_add_bo(struct anv_cmd_buffer
*cmd_buffer
,
2464 struct drm_i915_gem_relocation_entry
*relocs
,
2467 struct drm_i915_gem_exec_object2
*obj
;
2469 if (bo
->index
< cmd_buffer
->bo_count
&&
2470 cmd_buffer
->exec2_bos
[bo
->index
] == bo
)
2473 if (cmd_buffer
->bo_count
>= cmd_buffer
->exec2_array_length
) {
2474 uint32_t new_len
= cmd_buffer
->exec2_objects
?
2475 cmd_buffer
->exec2_array_length
* 2 : 64;
2477 struct drm_i915_gem_exec_object2
*new_objects
=
2478 anv_device_alloc(cmd_buffer
->device
, new_len
* sizeof(*new_objects
),
2479 8, VK_SYSTEM_ALLOC_TYPE_INTERNAL
);
2480 if (new_objects
== NULL
)
2481 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
2483 struct anv_bo
**new_bos
=
2484 anv_device_alloc(cmd_buffer
->device
, new_len
* sizeof(*new_bos
),
2485 8, VK_SYSTEM_ALLOC_TYPE_INTERNAL
);
2486 if (new_objects
== NULL
) {
2487 anv_device_free(cmd_buffer
->device
, new_objects
);
2488 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
2491 if (cmd_buffer
->exec2_objects
) {
2492 memcpy(new_objects
, cmd_buffer
->exec2_objects
,
2493 cmd_buffer
->bo_count
* sizeof(*new_objects
));
2494 memcpy(new_bos
, cmd_buffer
->exec2_bos
,
2495 cmd_buffer
->bo_count
* sizeof(*new_bos
));
2498 cmd_buffer
->exec2_objects
= new_objects
;
2499 cmd_buffer
->exec2_bos
= new_bos
;
2500 cmd_buffer
->exec2_array_length
= new_len
;
2503 assert(cmd_buffer
->bo_count
< cmd_buffer
->exec2_array_length
);
2505 bo
->index
= cmd_buffer
->bo_count
++;
2506 obj
= &cmd_buffer
->exec2_objects
[bo
->index
];
2507 cmd_buffer
->exec2_bos
[bo
->index
] = bo
;
2509 obj
->handle
= bo
->gem_handle
;
2510 obj
->relocation_count
= 0;
2511 obj
->relocs_ptr
= 0;
2513 obj
->offset
= bo
->offset
;
2519 obj
->relocation_count
= num_relocs
;
2520 obj
->relocs_ptr
= (uintptr_t) relocs
;
2527 anv_cmd_buffer_add_validate_bos(struct anv_cmd_buffer
*cmd_buffer
,
2528 struct anv_reloc_list
*list
)
2530 for (size_t i
= 0; i
< list
->num_relocs
; i
++)
2531 anv_cmd_buffer_add_bo(cmd_buffer
, list
->reloc_bos
[i
], NULL
, 0);
2535 anv_cmd_buffer_process_relocs(struct anv_cmd_buffer
*cmd_buffer
,
2536 struct anv_reloc_list
*list
)
2540 /* If the kernel supports I915_EXEC_NO_RELOC, it will compare offset in
2541 * struct drm_i915_gem_exec_object2 against the bos current offset and if
2542 * all bos haven't moved it will skip relocation processing alltogether.
2543 * If I915_EXEC_NO_RELOC is not supported, the kernel ignores the incoming
2544 * value of offset so we can set it either way. For that to work we need
2545 * to make sure all relocs use the same presumed offset.
2548 for (size_t i
= 0; i
< list
->num_relocs
; i
++) {
2549 bo
= list
->reloc_bos
[i
];
2550 if (bo
->offset
!= list
->relocs
[i
].presumed_offset
)
2551 cmd_buffer
->need_reloc
= true;
2553 list
->relocs
[i
].target_handle
= bo
->index
;
2557 VkResult
anv_EndCommandBuffer(
2558 VkCmdBuffer cmdBuffer
)
2560 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*) cmdBuffer
;
2561 struct anv_device
*device
= cmd_buffer
->device
;
2562 struct anv_batch
*batch
= &cmd_buffer
->batch
;
2564 anv_batch_emit(batch
, GEN8_MI_BATCH_BUFFER_END
);
2566 /* Round batch up to an even number of dwords. */
2567 if ((batch
->next
- batch
->start
) & 4)
2568 anv_batch_emit(batch
, GEN8_MI_NOOP
);
2570 anv_batch_bo_finish(cmd_buffer
->last_batch_bo
, &cmd_buffer
->batch
);
2572 cmd_buffer
->bo_count
= 0;
2573 cmd_buffer
->need_reloc
= false;
2575 /* Lock for access to bo->index. */
2576 pthread_mutex_lock(&device
->mutex
);
2578 /* Add block pool bos first so we can add them with their relocs. */
2579 anv_cmd_buffer_add_bo(cmd_buffer
, &cmd_buffer
->surface_bo
,
2580 cmd_buffer
->surface_relocs
.relocs
,
2581 cmd_buffer
->surface_relocs
.num_relocs
);
2583 /* Add all of the BOs referenced by surface state */
2584 anv_cmd_buffer_add_validate_bos(cmd_buffer
, &cmd_buffer
->surface_relocs
);
2586 /* Add all but the first batch BO */
2587 struct anv_batch_bo
*batch_bo
= cmd_buffer
->last_batch_bo
;
2588 while (batch_bo
->prev_batch_bo
) {
2589 anv_cmd_buffer_add_bo(cmd_buffer
, &batch_bo
->bo
,
2590 &batch
->relocs
.relocs
[batch_bo
->first_reloc
],
2591 batch_bo
->num_relocs
);
2592 batch_bo
= batch_bo
->prev_batch_bo
;
2595 /* Add everything referenced by the batches */
2596 anv_cmd_buffer_add_validate_bos(cmd_buffer
, &batch
->relocs
);
2598 /* Add the first batch bo last */
2599 assert(batch_bo
->prev_batch_bo
== NULL
&& batch_bo
->first_reloc
== 0);
2600 anv_cmd_buffer_add_bo(cmd_buffer
, &batch_bo
->bo
,
2601 &batch
->relocs
.relocs
[batch_bo
->first_reloc
],
2602 batch_bo
->num_relocs
);
2603 assert(batch_bo
->bo
.index
== cmd_buffer
->bo_count
- 1);
2605 anv_cmd_buffer_process_relocs(cmd_buffer
, &cmd_buffer
->surface_relocs
);
2606 anv_cmd_buffer_process_relocs(cmd_buffer
, &batch
->relocs
);
2608 cmd_buffer
->execbuf
.buffers_ptr
= (uintptr_t) cmd_buffer
->exec2_objects
;
2609 cmd_buffer
->execbuf
.buffer_count
= cmd_buffer
->bo_count
;
2610 cmd_buffer
->execbuf
.batch_start_offset
= 0;
2611 cmd_buffer
->execbuf
.batch_len
= batch
->next
- batch
->start
;
2612 cmd_buffer
->execbuf
.cliprects_ptr
= 0;
2613 cmd_buffer
->execbuf
.num_cliprects
= 0;
2614 cmd_buffer
->execbuf
.DR1
= 0;
2615 cmd_buffer
->execbuf
.DR4
= 0;
2617 cmd_buffer
->execbuf
.flags
= I915_EXEC_HANDLE_LUT
;
2618 if (!cmd_buffer
->need_reloc
)
2619 cmd_buffer
->execbuf
.flags
|= I915_EXEC_NO_RELOC
;
2620 cmd_buffer
->execbuf
.flags
|= I915_EXEC_RENDER
;
2621 cmd_buffer
->execbuf
.rsvd1
= device
->context_id
;
2622 cmd_buffer
->execbuf
.rsvd2
= 0;
2624 pthread_mutex_unlock(&device
->mutex
);
2629 VkResult
anv_ResetCommandBuffer(
2630 VkCmdBuffer cmdBuffer
)
2632 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*) cmdBuffer
;
2634 /* Delete all but the first batch bo */
2635 while (cmd_buffer
->last_batch_bo
->prev_batch_bo
) {
2636 struct anv_batch_bo
*prev
= cmd_buffer
->last_batch_bo
->prev_batch_bo
;
2637 anv_batch_bo_destroy(cmd_buffer
->last_batch_bo
, cmd_buffer
->device
);
2638 cmd_buffer
->last_batch_bo
= prev
;
2640 assert(cmd_buffer
->last_batch_bo
->prev_batch_bo
== NULL
);
2642 cmd_buffer
->batch
.relocs
.num_relocs
= 0;
2643 anv_batch_bo_start(cmd_buffer
->last_batch_bo
, &cmd_buffer
->batch
,
2644 GEN8_MI_BATCH_BUFFER_START_length
* 4);
2646 cmd_buffer
->surface_next
= 0;
2647 cmd_buffer
->surface_relocs
.num_relocs
= 0;
2652 // Command buffer building functions
2654 void anv_CmdBindPipeline(
2655 VkCmdBuffer cmdBuffer
,
2656 VkPipelineBindPoint pipelineBindPoint
,
2657 VkPipeline _pipeline
)
2659 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*) cmdBuffer
;
2660 struct anv_pipeline
*pipeline
= (struct anv_pipeline
*) _pipeline
;
2662 cmd_buffer
->pipeline
= pipeline
;
2663 cmd_buffer
->vb_dirty
|= pipeline
->vb_used
;
2664 cmd_buffer
->dirty
|= ANV_CMD_BUFFER_PIPELINE_DIRTY
;
2667 void anv_CmdBindDynamicStateObject(
2668 VkCmdBuffer cmdBuffer
,
2669 VkStateBindPoint stateBindPoint
,
2670 VkDynamicStateObject dynamicState
)
2672 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*) cmdBuffer
;
2673 struct anv_dynamic_vp_state
*vp_state
;
2675 switch (stateBindPoint
) {
2676 case VK_STATE_BIND_POINT_VIEWPORT
:
2677 vp_state
= (struct anv_dynamic_vp_state
*) dynamicState
;
2678 /* We emit state immediately, but set cmd_buffer->vp_state to indicate
2679 * that vp state has been set in this command buffer. */
2680 cmd_buffer
->vp_state
= vp_state
;
2681 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_SCISSOR_STATE_POINTERS
,
2682 .ScissorRectPointer
= vp_state
->scissor
.offset
);
2683 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_VIEWPORT_STATE_POINTERS_CC
,
2684 .CCViewportPointer
= vp_state
->cc_vp
.offset
);
2685 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP
,
2686 .SFClipViewportPointer
= vp_state
->sf_clip_vp
.offset
);
2688 case VK_STATE_BIND_POINT_RASTER
:
2689 cmd_buffer
->rs_state
= (struct anv_dynamic_rs_state
*) dynamicState
;
2690 cmd_buffer
->dirty
|= ANV_CMD_BUFFER_RS_DIRTY
;
2692 case VK_STATE_BIND_POINT_COLOR_BLEND
:
2693 cmd_buffer
->cb_state
= (struct anv_dynamic_cb_state
*) dynamicState
;
2694 cmd_buffer
->dirty
|= ANV_CMD_BUFFER_CB_DIRTY
;
2696 case VK_STATE_BIND_POINT_DEPTH_STENCIL
:
2697 cmd_buffer
->ds_state
= (struct anv_dynamic_ds_state
*) dynamicState
;
2698 cmd_buffer
->dirty
|= ANV_CMD_BUFFER_DS_DIRTY
;
2705 static struct anv_state
2706 anv_cmd_buffer_alloc_surface_state(struct anv_cmd_buffer
*cmd_buffer
,
2707 uint32_t size
, uint32_t alignment
)
2709 struct anv_state state
;
2711 state
.offset
= ALIGN_U32(cmd_buffer
->surface_next
, alignment
);
2712 state
.map
= cmd_buffer
->surface_bo
.map
+ state
.offset
;
2713 state
.alloc_size
= size
;
2714 cmd_buffer
->surface_next
= state
.offset
+ size
;
2716 assert(state
.offset
+ size
< cmd_buffer
->surface_bo
.size
);
2721 void anv_CmdBindDescriptorSets(
2722 VkCmdBuffer cmdBuffer
,
2723 VkPipelineBindPoint pipelineBindPoint
,
2726 const VkDescriptorSet
* pDescriptorSets
,
2727 uint32_t dynamicOffsetCount
,
2728 const uint32_t* pDynamicOffsets
)
2730 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*) cmdBuffer
;
2731 struct anv_pipeline_layout
*layout
= cmd_buffer
->pipeline
->layout
;
2732 struct anv_bindings
*bindings
= cmd_buffer
->bindings
;
2734 uint32_t offset
= 0;
2735 for (uint32_t i
= 0; i
< setCount
; i
++) {
2736 struct anv_descriptor_set
*set
=
2737 (struct anv_descriptor_set
*) pDescriptorSets
[i
];
2738 struct anv_descriptor_set_layout
*set_layout
= layout
->set
[firstSet
+ i
].layout
;
2740 for (uint32_t s
= 0; s
< VK_NUM_SHADER_STAGE
; s
++) {
2741 uint32_t *surface_to_desc
= set_layout
->stage
[s
].surface_start
;
2742 uint32_t *sampler_to_desc
= set_layout
->stage
[s
].sampler_start
;
2743 uint32_t bias
= s
== VK_SHADER_STAGE_FRAGMENT
? MAX_RTS
: 0;
2746 start
= bias
+ layout
->set
[firstSet
+ i
].surface_start
[s
];
2747 for (uint32_t b
= 0; b
< set_layout
->stage
[s
].surface_count
; b
++) {
2748 struct anv_surface_view
*view
= set
->descriptors
[surface_to_desc
[b
]].view
;
2752 struct anv_state state
=
2753 anv_cmd_buffer_alloc_surface_state(cmd_buffer
, 64, 64);
2754 memcpy(state
.map
, view
->surface_state
.map
, 64);
2756 /* The address goes in dwords 8 and 9 of the SURFACE_STATE */
2757 *(uint64_t *)(state
.map
+ 8 * 4) =
2758 anv_reloc_list_add(&cmd_buffer
->surface_relocs
,
2760 state
.offset
+ 8 * 4,
2761 view
->bo
, view
->offset
);
2763 bindings
->descriptors
[s
].surfaces
[start
+ b
] = state
.offset
;
2766 start
= layout
->set
[firstSet
+ i
].sampler_start
[s
];
2767 for (uint32_t b
= 0; b
< set_layout
->stage
[s
].sampler_count
; b
++) {
2768 struct anv_sampler
*sampler
= set
->descriptors
[sampler_to_desc
[b
]].sampler
;
2772 memcpy(&bindings
->descriptors
[s
].samplers
[start
+ b
],
2773 sampler
->state
, sizeof(sampler
->state
));
2777 offset
+= layout
->set
[firstSet
+ i
].layout
->num_dynamic_buffers
;
2780 cmd_buffer
->dirty
|= ANV_CMD_BUFFER_DESCRIPTOR_SET_DIRTY
;
2783 void anv_CmdBindIndexBuffer(
2784 VkCmdBuffer cmdBuffer
,
2786 VkDeviceSize offset
,
2787 VkIndexType indexType
)
2789 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*) cmdBuffer
;
2790 struct anv_buffer
*buffer
= (struct anv_buffer
*) _buffer
;
2792 static const uint32_t vk_to_gen_index_type
[] = {
2793 [VK_INDEX_TYPE_UINT8
] = INDEX_BYTE
,
2794 [VK_INDEX_TYPE_UINT16
] = INDEX_WORD
,
2795 [VK_INDEX_TYPE_UINT32
] = INDEX_DWORD
,
2798 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_INDEX_BUFFER
,
2799 .IndexFormat
= vk_to_gen_index_type
[indexType
],
2800 .MemoryObjectControlState
= GEN8_MOCS
,
2801 .BufferStartingAddress
= { buffer
->bo
, buffer
->offset
+ offset
},
2802 .BufferSize
= buffer
->size
- offset
);
2805 void anv_CmdBindVertexBuffers(
2806 VkCmdBuffer cmdBuffer
,
2807 uint32_t startBinding
,
2808 uint32_t bindingCount
,
2809 const VkBuffer
* pBuffers
,
2810 const VkDeviceSize
* pOffsets
)
2812 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*) cmdBuffer
;
2813 struct anv_bindings
*bindings
= cmd_buffer
->bindings
;
2815 /* We have to defer setting up vertex buffer since we need the buffer
2816 * stride from the pipeline. */
2818 for (uint32_t i
= 0; i
< bindingCount
; i
++) {
2819 bindings
->vb
[startBinding
+ i
].buffer
= (struct anv_buffer
*) pBuffers
[i
];
2820 bindings
->vb
[startBinding
+ i
].offset
= pOffsets
[i
];
2821 cmd_buffer
->vb_dirty
|= 1 << (startBinding
+ i
);
2826 flush_descriptor_sets(struct anv_cmd_buffer
*cmd_buffer
)
2828 struct anv_pipeline_layout
*layout
= cmd_buffer
->pipeline
->layout
;
2829 struct anv_bindings
*bindings
= cmd_buffer
->bindings
;
2830 uint32_t layers
= cmd_buffer
->framebuffer
->layers
;
2832 for (uint32_t s
= 0; s
< VK_NUM_SHADER_STAGE
; s
++) {
2835 if (s
== VK_SHADER_STAGE_FRAGMENT
) {
2837 layers
= cmd_buffer
->framebuffer
->layers
;
2843 /* This is a little awkward: layout can be NULL but we still have to
2844 * allocate and set a binding table for the PS stage for render
2846 uint32_t surface_count
= layout
? layout
->stage
[s
].surface_count
: 0;
2848 if (layers
+ surface_count
> 0) {
2849 struct anv_state state
;
2852 size
= (bias
+ surface_count
) * sizeof(uint32_t);
2853 state
= anv_cmd_buffer_alloc_surface_state(cmd_buffer
, size
, 32);
2854 memcpy(state
.map
, bindings
->descriptors
[s
].surfaces
, size
);
2856 static const uint32_t binding_table_opcodes
[] = {
2857 [VK_SHADER_STAGE_VERTEX
] = 38,
2858 [VK_SHADER_STAGE_TESS_CONTROL
] = 39,
2859 [VK_SHADER_STAGE_TESS_EVALUATION
] = 40,
2860 [VK_SHADER_STAGE_GEOMETRY
] = 41,
2861 [VK_SHADER_STAGE_FRAGMENT
] = 42,
2862 [VK_SHADER_STAGE_COMPUTE
] = 0,
2865 anv_batch_emit(&cmd_buffer
->batch
,
2866 GEN8_3DSTATE_BINDING_TABLE_POINTERS_VS
,
2867 ._3DCommandSubOpcode
= binding_table_opcodes
[s
],
2868 .PointertoVSBindingTable
= state
.offset
);
2871 if (layout
&& layout
->stage
[s
].sampler_count
> 0) {
2872 struct anv_state state
;
2875 size
= layout
->stage
[s
].sampler_count
* 16;
2876 state
= anv_state_stream_alloc(&cmd_buffer
->dynamic_state_stream
, size
, 32);
2877 memcpy(state
.map
, bindings
->descriptors
[s
].samplers
, size
);
2879 static const uint32_t sampler_state_opcodes
[] = {
2880 [VK_SHADER_STAGE_VERTEX
] = 43,
2881 [VK_SHADER_STAGE_TESS_CONTROL
] = 44, /* HS */
2882 [VK_SHADER_STAGE_TESS_EVALUATION
] = 45, /* DS */
2883 [VK_SHADER_STAGE_GEOMETRY
] = 46,
2884 [VK_SHADER_STAGE_FRAGMENT
] = 47,
2885 [VK_SHADER_STAGE_COMPUTE
] = 0,
2888 anv_batch_emit(&cmd_buffer
->batch
,
2889 GEN8_3DSTATE_SAMPLER_STATE_POINTERS_VS
,
2890 ._3DCommandSubOpcode
= sampler_state_opcodes
[s
],
2891 .PointertoVSSamplerState
= state
.offset
);
2896 static struct anv_state
2897 anv_cmd_buffer_emit_dynamic(struct anv_cmd_buffer
*cmd_buffer
,
2898 uint32_t *a
, uint32_t dwords
, uint32_t alignment
)
2900 struct anv_device
*device
= cmd_buffer
->device
;
2901 struct anv_state state
;
2903 state
= anv_state_pool_alloc(&device
->dynamic_state_pool
, dwords
* 4, alignment
);
2904 memcpy(state
.map
, a
, dwords
* 4);
2909 static struct anv_state
2910 anv_cmd_buffer_merge_dynamic(struct anv_cmd_buffer
*cmd_buffer
,
2911 uint32_t *a
, uint32_t *b
, uint32_t dwords
, uint32_t alignment
)
2913 struct anv_device
*device
= cmd_buffer
->device
;
2914 struct anv_state state
;
2917 state
= anv_state_pool_alloc(&device
->dynamic_state_pool
, dwords
* 4, alignment
);
2919 for (uint32_t i
= 0; i
< dwords
; i
++)
2926 anv_cmd_buffer_flush_state(struct anv_cmd_buffer
*cmd_buffer
)
2928 struct anv_pipeline
*pipeline
= cmd_buffer
->pipeline
;
2929 struct anv_bindings
*bindings
= cmd_buffer
->bindings
;
2932 uint32_t vb_emit
= cmd_buffer
->vb_dirty
& pipeline
->vb_used
;
2933 const uint32_t num_buffers
= __builtin_popcount(vb_emit
);
2934 const uint32_t num_dwords
= 1 + num_buffers
* 4;
2937 p
= anv_batch_emitn(&cmd_buffer
->batch
, num_dwords
,
2938 GEN8_3DSTATE_VERTEX_BUFFERS
);
2940 for_each_bit(vb
, vb_emit
) {
2941 struct anv_buffer
*buffer
= bindings
->vb
[vb
].buffer
;
2942 uint32_t offset
= bindings
->vb
[vb
].offset
;
2944 struct GEN8_VERTEX_BUFFER_STATE state
= {
2945 .VertexBufferIndex
= vb
,
2946 .MemoryObjectControlState
= GEN8_MOCS
,
2947 .AddressModifyEnable
= true,
2948 .BufferPitch
= pipeline
->binding_stride
[vb
],
2949 .BufferStartingAddress
= { buffer
->bo
, buffer
->offset
+ offset
},
2950 .BufferSize
= buffer
->size
- offset
2953 GEN8_VERTEX_BUFFER_STATE_pack(&cmd_buffer
->batch
, &p
[1 + i
* 4], &state
);
2958 if (cmd_buffer
->dirty
& ANV_CMD_BUFFER_PIPELINE_DIRTY
)
2959 anv_batch_emit_batch(&cmd_buffer
->batch
, &pipeline
->batch
);
2961 if (cmd_buffer
->dirty
& ANV_CMD_BUFFER_DESCRIPTOR_SET_DIRTY
)
2962 flush_descriptor_sets(cmd_buffer
);
2964 if (cmd_buffer
->dirty
& (ANV_CMD_BUFFER_PIPELINE_DIRTY
| ANV_CMD_BUFFER_RS_DIRTY
)) {
2965 anv_batch_emit_merge(&cmd_buffer
->batch
,
2966 cmd_buffer
->rs_state
->state_sf
, pipeline
->state_sf
);
2967 anv_batch_emit_merge(&cmd_buffer
->batch
,
2968 cmd_buffer
->rs_state
->state_raster
, pipeline
->state_raster
);
2971 if (cmd_buffer
->ds_state
&&
2972 (cmd_buffer
->dirty
& (ANV_CMD_BUFFER_PIPELINE_DIRTY
| ANV_CMD_BUFFER_DS_DIRTY
)))
2973 anv_batch_emit_merge(&cmd_buffer
->batch
,
2974 cmd_buffer
->ds_state
->state_wm_depth_stencil
,
2975 pipeline
->state_wm_depth_stencil
);
2977 if (cmd_buffer
->dirty
& (ANV_CMD_BUFFER_CB_DIRTY
| ANV_CMD_BUFFER_DS_DIRTY
)) {
2978 struct anv_state state
;
2979 if (cmd_buffer
->ds_state
== NULL
)
2980 state
= anv_cmd_buffer_emit_dynamic(cmd_buffer
,
2981 cmd_buffer
->cb_state
->state_color_calc
,
2982 GEN8_COLOR_CALC_STATE_length
, 32);
2983 else if (cmd_buffer
->cb_state
== NULL
)
2984 state
= anv_cmd_buffer_emit_dynamic(cmd_buffer
,
2985 cmd_buffer
->ds_state
->state_color_calc
,
2986 GEN8_COLOR_CALC_STATE_length
, 32);
2988 state
= anv_cmd_buffer_merge_dynamic(cmd_buffer
,
2989 cmd_buffer
->ds_state
->state_color_calc
,
2990 cmd_buffer
->cb_state
->state_color_calc
,
2991 GEN8_COLOR_CALC_STATE_length
, 32);
2993 anv_batch_emit(&cmd_buffer
->batch
,
2994 GEN8_3DSTATE_CC_STATE_POINTERS
,
2995 .ColorCalcStatePointer
= state
.offset
,
2996 .ColorCalcStatePointerValid
= true);
2999 cmd_buffer
->vb_dirty
&= ~vb_emit
;
3000 cmd_buffer
->dirty
= 0;
3004 VkCmdBuffer cmdBuffer
,
3005 uint32_t firstVertex
,
3006 uint32_t vertexCount
,
3007 uint32_t firstInstance
,
3008 uint32_t instanceCount
)
3010 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*) cmdBuffer
;
3012 anv_cmd_buffer_flush_state(cmd_buffer
);
3014 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DPRIMITIVE
,
3015 .VertexAccessType
= SEQUENTIAL
,
3016 .VertexCountPerInstance
= vertexCount
,
3017 .StartVertexLocation
= firstVertex
,
3018 .InstanceCount
= instanceCount
,
3019 .StartInstanceLocation
= firstInstance
,
3020 .BaseVertexLocation
= 0);
3023 void anv_CmdDrawIndexed(
3024 VkCmdBuffer cmdBuffer
,
3025 uint32_t firstIndex
,
3026 uint32_t indexCount
,
3027 int32_t vertexOffset
,
3028 uint32_t firstInstance
,
3029 uint32_t instanceCount
)
3031 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*) cmdBuffer
;
3033 anv_cmd_buffer_flush_state(cmd_buffer
);
3035 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DPRIMITIVE
,
3036 .VertexAccessType
= RANDOM
,
3037 .VertexCountPerInstance
= indexCount
,
3038 .StartVertexLocation
= firstIndex
,
3039 .InstanceCount
= instanceCount
,
3040 .StartInstanceLocation
= firstInstance
,
3041 .BaseVertexLocation
= 0);
3045 anv_batch_lrm(struct anv_batch
*batch
,
3046 uint32_t reg
, struct anv_bo
*bo
, uint32_t offset
)
3048 anv_batch_emit(batch
, GEN8_MI_LOAD_REGISTER_MEM
,
3049 .RegisterAddress
= reg
,
3050 .MemoryAddress
= { bo
, offset
});
3054 anv_batch_lri(struct anv_batch
*batch
, uint32_t reg
, uint32_t imm
)
3056 anv_batch_emit(batch
, GEN8_MI_LOAD_REGISTER_IMM
,
3057 .RegisterOffset
= reg
,
3061 /* Auto-Draw / Indirect Registers */
3062 #define GEN7_3DPRIM_END_OFFSET 0x2420
3063 #define GEN7_3DPRIM_START_VERTEX 0x2430
3064 #define GEN7_3DPRIM_VERTEX_COUNT 0x2434
3065 #define GEN7_3DPRIM_INSTANCE_COUNT 0x2438
3066 #define GEN7_3DPRIM_START_INSTANCE 0x243C
3067 #define GEN7_3DPRIM_BASE_VERTEX 0x2440
3069 void anv_CmdDrawIndirect(
3070 VkCmdBuffer cmdBuffer
,
3072 VkDeviceSize offset
,
3076 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*) cmdBuffer
;
3077 struct anv_buffer
*buffer
= (struct anv_buffer
*) _buffer
;
3078 struct anv_bo
*bo
= buffer
->bo
;
3079 uint32_t bo_offset
= buffer
->offset
+ offset
;
3081 anv_cmd_buffer_flush_state(cmd_buffer
);
3083 anv_batch_lrm(&cmd_buffer
->batch
, GEN7_3DPRIM_VERTEX_COUNT
, bo
, bo_offset
);
3084 anv_batch_lrm(&cmd_buffer
->batch
, GEN7_3DPRIM_INSTANCE_COUNT
, bo
, bo_offset
+ 4);
3085 anv_batch_lrm(&cmd_buffer
->batch
, GEN7_3DPRIM_START_VERTEX
, bo
, bo_offset
+ 8);
3086 anv_batch_lrm(&cmd_buffer
->batch
, GEN7_3DPRIM_START_INSTANCE
, bo
, bo_offset
+ 12);
3087 anv_batch_lri(&cmd_buffer
->batch
, GEN7_3DPRIM_BASE_VERTEX
, 0);
3089 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DPRIMITIVE
,
3090 .IndirectParameterEnable
= true,
3091 .VertexAccessType
= SEQUENTIAL
);
3094 void anv_CmdDrawIndexedIndirect(
3095 VkCmdBuffer cmdBuffer
,
3097 VkDeviceSize offset
,
3101 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*) cmdBuffer
;
3102 struct anv_buffer
*buffer
= (struct anv_buffer
*) _buffer
;
3103 struct anv_bo
*bo
= buffer
->bo
;
3104 uint32_t bo_offset
= buffer
->offset
+ offset
;
3106 anv_cmd_buffer_flush_state(cmd_buffer
);
3108 anv_batch_lrm(&cmd_buffer
->batch
, GEN7_3DPRIM_VERTEX_COUNT
, bo
, bo_offset
);
3109 anv_batch_lrm(&cmd_buffer
->batch
, GEN7_3DPRIM_INSTANCE_COUNT
, bo
, bo_offset
+ 4);
3110 anv_batch_lrm(&cmd_buffer
->batch
, GEN7_3DPRIM_START_VERTEX
, bo
, bo_offset
+ 8);
3111 anv_batch_lrm(&cmd_buffer
->batch
, GEN7_3DPRIM_BASE_VERTEX
, bo
, bo_offset
+ 12);
3112 anv_batch_lrm(&cmd_buffer
->batch
, GEN7_3DPRIM_START_INSTANCE
, bo
, bo_offset
+ 16);
3114 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DPRIMITIVE
,
3115 .IndirectParameterEnable
= true,
3116 .VertexAccessType
= RANDOM
);
3119 void anv_CmdDispatch(
3120 VkCmdBuffer cmdBuffer
,
3128 void anv_CmdDispatchIndirect(
3129 VkCmdBuffer cmdBuffer
,
3131 VkDeviceSize offset
)
3136 void anv_CmdSetEvent(
3137 VkCmdBuffer cmdBuffer
,
3139 VkPipeEvent pipeEvent
)
3144 void anv_CmdResetEvent(
3145 VkCmdBuffer cmdBuffer
,
3147 VkPipeEvent pipeEvent
)
3152 void anv_CmdWaitEvents(
3153 VkCmdBuffer cmdBuffer
,
3154 VkWaitEvent waitEvent
,
3155 uint32_t eventCount
,
3156 const VkEvent
* pEvents
,
3157 uint32_t memBarrierCount
,
3158 const void** ppMemBarriers
)
3163 void anv_CmdPipelineBarrier(
3164 VkCmdBuffer cmdBuffer
,
3165 VkWaitEvent waitEvent
,
3166 uint32_t pipeEventCount
,
3167 const VkPipeEvent
* pPipeEvents
,
3168 uint32_t memBarrierCount
,
3169 const void** ppMemBarriers
)
3175 anv_batch_emit_ps_depth_count(struct anv_batch
*batch
,
3176 struct anv_bo
*bo
, uint32_t offset
)
3178 anv_batch_emit(batch
, GEN8_PIPE_CONTROL
,
3179 .DestinationAddressType
= DAT_PPGTT
,
3180 .PostSyncOperation
= WritePSDepthCount
,
3181 .Address
= { bo
, offset
}); /* FIXME: This is only lower 32 bits */
3184 void anv_CmdBeginQuery(
3185 VkCmdBuffer cmdBuffer
,
3186 VkQueryPool queryPool
,
3188 VkQueryControlFlags flags
)
3190 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*) cmdBuffer
;
3191 struct anv_query_pool
*pool
= (struct anv_query_pool
*) queryPool
;
3193 switch (pool
->type
) {
3194 case VK_QUERY_TYPE_OCCLUSION
:
3195 anv_batch_emit_ps_depth_count(&cmd_buffer
->batch
, &pool
->bo
,
3196 slot
* sizeof(struct anv_query_pool_slot
));
3199 case VK_QUERY_TYPE_PIPELINE_STATISTICS
:
3205 void anv_CmdEndQuery(
3206 VkCmdBuffer cmdBuffer
,
3207 VkQueryPool queryPool
,
3210 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*) cmdBuffer
;
3211 struct anv_query_pool
*pool
= (struct anv_query_pool
*) queryPool
;
3213 switch (pool
->type
) {
3214 case VK_QUERY_TYPE_OCCLUSION
:
3215 anv_batch_emit_ps_depth_count(&cmd_buffer
->batch
, &pool
->bo
,
3216 slot
* sizeof(struct anv_query_pool_slot
) + 8);
3219 case VK_QUERY_TYPE_PIPELINE_STATISTICS
:
3225 void anv_CmdResetQueryPool(
3226 VkCmdBuffer cmdBuffer
,
3227 VkQueryPool queryPool
,
3228 uint32_t startQuery
,
3229 uint32_t queryCount
)
3234 #define TIMESTAMP 0x2358
3236 void anv_CmdWriteTimestamp(
3237 VkCmdBuffer cmdBuffer
,
3238 VkTimestampType timestampType
,
3239 VkBuffer destBuffer
,
3240 VkDeviceSize destOffset
)
3242 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*) cmdBuffer
;
3243 struct anv_buffer
*buffer
= (struct anv_buffer
*) destBuffer
;
3244 struct anv_bo
*bo
= buffer
->bo
;
3246 switch (timestampType
) {
3247 case VK_TIMESTAMP_TYPE_TOP
:
3248 anv_batch_emit(&cmd_buffer
->batch
, GEN8_MI_STORE_REGISTER_MEM
,
3249 .RegisterAddress
= TIMESTAMP
,
3250 .MemoryAddress
= { bo
, buffer
->offset
+ destOffset
});
3251 anv_batch_emit(&cmd_buffer
->batch
, GEN8_MI_STORE_REGISTER_MEM
,
3252 .RegisterAddress
= TIMESTAMP
+ 4,
3253 .MemoryAddress
= { bo
, buffer
->offset
+ destOffset
+ 4 });
3256 case VK_TIMESTAMP_TYPE_BOTTOM
:
3257 anv_batch_emit(&cmd_buffer
->batch
, GEN8_PIPE_CONTROL
,
3258 .DestinationAddressType
= DAT_PPGTT
,
3259 .PostSyncOperation
= WriteTimestamp
,
3260 .Address
= /* FIXME: This is only lower 32 bits */
3261 { bo
, buffer
->offset
+ destOffset
});
3269 #define alu_opcode(v) __gen_field((v), 20, 31)
3270 #define alu_operand1(v) __gen_field((v), 10, 19)
3271 #define alu_operand2(v) __gen_field((v), 0, 9)
3272 #define alu(opcode, operand1, operand2) \
3273 alu_opcode(opcode) | alu_operand1(operand1) | alu_operand2(operand2)
3275 #define OPCODE_NOOP 0x000
3276 #define OPCODE_LOAD 0x080
3277 #define OPCODE_LOADINV 0x480
3278 #define OPCODE_LOAD0 0x081
3279 #define OPCODE_LOAD1 0x481
3280 #define OPCODE_ADD 0x100
3281 #define OPCODE_SUB 0x101
3282 #define OPCODE_AND 0x102
3283 #define OPCODE_OR 0x103
3284 #define OPCODE_XOR 0x104
3285 #define OPCODE_STORE 0x180
3286 #define OPCODE_STOREINV 0x580
3288 #define OPERAND_R0 0x00
3289 #define OPERAND_R1 0x01
3290 #define OPERAND_R2 0x02
3291 #define OPERAND_R3 0x03
3292 #define OPERAND_R4 0x04
3293 #define OPERAND_SRCA 0x20
3294 #define OPERAND_SRCB 0x21
3295 #define OPERAND_ACCU 0x31
3296 #define OPERAND_ZF 0x32
3297 #define OPERAND_CF 0x33
3299 #define CS_GPR(n) (0x2600 + (n) * 8)
3302 emit_load_alu_reg_u64(struct anv_batch
*batch
, uint32_t reg
,
3303 struct anv_bo
*bo
, uint32_t offset
)
3305 anv_batch_emit(batch
, GEN8_MI_LOAD_REGISTER_MEM
,
3306 .RegisterAddress
= reg
,
3307 .MemoryAddress
= { bo
, offset
});
3308 anv_batch_emit(batch
, GEN8_MI_LOAD_REGISTER_MEM
,
3309 .RegisterAddress
= reg
+ 4,
3310 .MemoryAddress
= { bo
, offset
+ 4 });
3313 void anv_CmdCopyQueryPoolResults(
3314 VkCmdBuffer cmdBuffer
,
3315 VkQueryPool queryPool
,
3316 uint32_t startQuery
,
3317 uint32_t queryCount
,
3318 VkBuffer destBuffer
,
3319 VkDeviceSize destOffset
,
3320 VkDeviceSize destStride
,
3321 VkQueryResultFlags flags
)
3323 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*) cmdBuffer
;
3324 struct anv_query_pool
*pool
= (struct anv_query_pool
*) queryPool
;
3325 struct anv_buffer
*buffer
= (struct anv_buffer
*) destBuffer
;
3326 uint32_t slot_offset
, dst_offset
;
3328 if (flags
& VK_QUERY_RESULT_WITH_AVAILABILITY_BIT
) {
3329 /* Where is the availabilty info supposed to go? */
3330 anv_finishme("VK_QUERY_RESULT_WITH_AVAILABILITY_BIT");
3334 assert(pool
->type
== VK_QUERY_TYPE_OCCLUSION
);
3336 /* FIXME: If we're not waiting, should we just do this on the CPU? */
3337 if (flags
& VK_QUERY_RESULT_WAIT_BIT
)
3338 anv_batch_emit(&cmd_buffer
->batch
, GEN8_PIPE_CONTROL
,
3339 .CommandStreamerStallEnable
= true,
3340 .StallAtPixelScoreboard
= true);
3342 dst_offset
= buffer
->offset
+ destOffset
;
3343 for (uint32_t i
= 0; i
< queryCount
; i
++) {
3345 slot_offset
= (startQuery
+ i
) * sizeof(struct anv_query_pool_slot
);
3347 emit_load_alu_reg_u64(&cmd_buffer
->batch
, CS_GPR(0), &pool
->bo
, slot_offset
);
3348 emit_load_alu_reg_u64(&cmd_buffer
->batch
, CS_GPR(1), &pool
->bo
, slot_offset
+ 8);
3350 /* FIXME: We need to clamp the result for 32 bit. */
3352 uint32_t *dw
= anv_batch_emitn(&cmd_buffer
->batch
, 5, GEN8_MI_MATH
);
3353 dw
[1] = alu(OPCODE_LOAD
, OPERAND_SRCA
, OPERAND_R1
);
3354 dw
[2] = alu(OPCODE_LOAD
, OPERAND_SRCB
, OPERAND_R0
);
3355 dw
[3] = alu(OPCODE_SUB
, 0, 0);
3356 dw
[4] = alu(OPCODE_STORE
, OPERAND_R2
, OPERAND_ACCU
);
3358 anv_batch_emit(&cmd_buffer
->batch
, GEN8_MI_STORE_REGISTER_MEM
,
3359 .RegisterAddress
= CS_GPR(2),
3360 /* FIXME: This is only lower 32 bits */
3361 .MemoryAddress
= { buffer
->bo
, dst_offset
});
3363 if (flags
& VK_QUERY_RESULT_64_BIT
)
3364 anv_batch_emit(&cmd_buffer
->batch
, GEN8_MI_STORE_REGISTER_MEM
,
3365 .RegisterAddress
= CS_GPR(2) + 4,
3366 /* FIXME: This is only lower 32 bits */
3367 .MemoryAddress
= { buffer
->bo
, dst_offset
+ 4 });
3369 dst_offset
+= destStride
;
3373 void anv_CmdInitAtomicCounters(
3374 VkCmdBuffer cmdBuffer
,
3375 VkPipelineBindPoint pipelineBindPoint
,
3376 uint32_t startCounter
,
3377 uint32_t counterCount
,
3378 const uint32_t* pData
)
3383 void anv_CmdLoadAtomicCounters(
3384 VkCmdBuffer cmdBuffer
,
3385 VkPipelineBindPoint pipelineBindPoint
,
3386 uint32_t startCounter
,
3387 uint32_t counterCount
,
3389 VkDeviceSize srcOffset
)
3394 void anv_CmdSaveAtomicCounters(
3395 VkCmdBuffer cmdBuffer
,
3396 VkPipelineBindPoint pipelineBindPoint
,
3397 uint32_t startCounter
,
3398 uint32_t counterCount
,
3399 VkBuffer destBuffer
,
3400 VkDeviceSize destOffset
)
3406 anv_framebuffer_destroy(struct anv_device
*device
,
3407 struct anv_object
*object
,
3408 VkObjectType obj_type
)
3410 struct anv_framebuffer
*fb
= (struct anv_framebuffer
*)object
;
3412 assert(obj_type
== VK_OBJECT_TYPE_FRAMEBUFFER
);
3414 anv_DestroyObject((VkDevice
) device
,
3415 VK_OBJECT_TYPE_DYNAMIC_VP_STATE
,
3418 anv_device_free(device
, fb
);
3421 VkResult
anv_CreateFramebuffer(
3423 const VkFramebufferCreateInfo
* pCreateInfo
,
3424 VkFramebuffer
* pFramebuffer
)
3426 struct anv_device
*device
= (struct anv_device
*) _device
;
3427 struct anv_framebuffer
*framebuffer
;
3429 static const struct anv_depth_stencil_view null_view
=
3430 { .depth_format
= D16_UNORM
, .depth_stride
= 0, .stencil_stride
= 0 };
3432 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO
);
3434 framebuffer
= anv_device_alloc(device
, sizeof(*framebuffer
), 8,
3435 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
3436 if (framebuffer
== NULL
)
3437 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
3439 framebuffer
->base
.destructor
= anv_framebuffer_destroy
;
3441 framebuffer
->color_attachment_count
= pCreateInfo
->colorAttachmentCount
;
3442 for (uint32_t i
= 0; i
< pCreateInfo
->colorAttachmentCount
; i
++) {
3443 framebuffer
->color_attachments
[i
] =
3444 (struct anv_surface_view
*) pCreateInfo
->pColorAttachments
[i
].view
;
3447 if (pCreateInfo
->pDepthStencilAttachment
) {
3448 framebuffer
->depth_stencil
=
3449 (struct anv_depth_stencil_view
*) pCreateInfo
->pDepthStencilAttachment
->view
;
3451 framebuffer
->depth_stencil
= &null_view
;
3454 framebuffer
->sample_count
= pCreateInfo
->sampleCount
;
3455 framebuffer
->width
= pCreateInfo
->width
;
3456 framebuffer
->height
= pCreateInfo
->height
;
3457 framebuffer
->layers
= pCreateInfo
->layers
;
3459 vkCreateDynamicViewportState((VkDevice
) device
,
3460 &(VkDynamicVpStateCreateInfo
) {
3461 .sType
= VK_STRUCTURE_TYPE_DYNAMIC_VP_STATE_CREATE_INFO
,
3462 .viewportAndScissorCount
= 1,
3463 .pViewports
= (VkViewport
[]) {
3467 .width
= pCreateInfo
->width
,
3468 .height
= pCreateInfo
->height
,
3473 .pScissors
= (VkRect
[]) {
3475 { pCreateInfo
->width
, pCreateInfo
->height
} },
3478 &framebuffer
->vp_state
);
3480 *pFramebuffer
= (VkFramebuffer
) framebuffer
;
3485 VkResult
anv_CreateRenderPass(
3487 const VkRenderPassCreateInfo
* pCreateInfo
,
3488 VkRenderPass
* pRenderPass
)
3490 struct anv_device
*device
= (struct anv_device
*) _device
;
3491 struct anv_render_pass
*pass
;
3494 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO
);
3496 size
= sizeof(*pass
) +
3497 pCreateInfo
->layers
* sizeof(struct anv_render_pass_layer
);
3498 pass
= anv_device_alloc(device
, size
, 8,
3499 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
3501 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
3503 pass
->render_area
= pCreateInfo
->renderArea
;
3505 pass
->num_layers
= pCreateInfo
->layers
;
3507 pass
->num_clear_layers
= 0;
3508 for (uint32_t i
= 0; i
< pCreateInfo
->layers
; i
++) {
3509 pass
->layers
[i
].color_load_op
= pCreateInfo
->pColorLoadOps
[i
];
3510 pass
->layers
[i
].clear_color
= pCreateInfo
->pColorLoadClearValues
[i
];
3511 if (pass
->layers
[i
].color_load_op
== VK_ATTACHMENT_LOAD_OP_CLEAR
)
3512 pass
->num_clear_layers
++;
3515 *pRenderPass
= (VkRenderPass
) pass
;
3521 anv_cmd_buffer_fill_render_targets(struct anv_cmd_buffer
*cmd_buffer
)
3523 struct anv_framebuffer
*framebuffer
= cmd_buffer
->framebuffer
;
3524 struct anv_bindings
*bindings
= cmd_buffer
->bindings
;
3526 for (uint32_t i
= 0; i
< framebuffer
->color_attachment_count
; i
++) {
3527 const struct anv_surface_view
*view
= framebuffer
->color_attachments
[i
];
3529 struct anv_state state
=
3530 anv_cmd_buffer_alloc_surface_state(cmd_buffer
, 64, 64);
3531 memcpy(state
.map
, view
->surface_state
.map
, 64);
3533 /* The address goes in dwords 8 and 9 of the SURFACE_STATE */
3534 *(uint64_t *)(state
.map
+ 8 * 4) =
3535 anv_reloc_list_add(&cmd_buffer
->surface_relocs
,
3537 state
.offset
+ 8 * 4,
3538 view
->bo
, view
->offset
);
3540 bindings
->descriptors
[VK_SHADER_STAGE_FRAGMENT
].surfaces
[i
] = state
.offset
;
3542 cmd_buffer
->dirty
|= ANV_CMD_BUFFER_DESCRIPTOR_SET_DIRTY
;
3546 anv_cmd_buffer_emit_depth_stencil(struct anv_cmd_buffer
*cmd_buffer
,
3547 struct anv_render_pass
*pass
)
3549 const struct anv_depth_stencil_view
*view
=
3550 cmd_buffer
->framebuffer
->depth_stencil
;
3552 /* FIXME: Implement the PMA stall W/A */
3554 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_DEPTH_BUFFER
,
3555 .SurfaceType
= SURFTYPE_2D
,
3556 .DepthWriteEnable
= view
->depth_stride
> 0,
3557 .StencilWriteEnable
= view
->stencil_stride
> 0,
3558 .HierarchicalDepthBufferEnable
= false,
3559 .SurfaceFormat
= view
->depth_format
,
3560 .SurfacePitch
= view
->depth_stride
> 0 ? view
->depth_stride
- 1 : 0,
3561 .SurfaceBaseAddress
= { view
->bo
, view
->depth_offset
},
3562 .Height
= pass
->render_area
.extent
.height
- 1,
3563 .Width
= pass
->render_area
.extent
.width
- 1,
3566 .MinimumArrayElement
= 0,
3567 .DepthBufferObjectControlState
= GEN8_MOCS
,
3568 .RenderTargetViewExtent
= 1 - 1,
3569 .SurfaceQPitch
= 0);
3571 /* Disable hierarchial depth buffers. */
3572 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_HIER_DEPTH_BUFFER
);
3574 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_STENCIL_BUFFER
,
3575 .StencilBufferEnable
= view
->stencil_stride
> 0,
3576 .StencilBufferObjectControlState
= GEN8_MOCS
,
3577 .SurfacePitch
= view
->stencil_stride
> 0 ? view
->stencil_stride
- 1 : 0,
3578 .SurfaceBaseAddress
= { view
->bo
, view
->stencil_offset
},
3579 .SurfaceQPitch
= 0);
3581 /* Clear the clear params. */
3582 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_CLEAR_PARAMS
);
3585 void anv_CmdBeginRenderPass(
3586 VkCmdBuffer cmdBuffer
,
3587 const VkRenderPassBegin
* pRenderPassBegin
)
3589 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*) cmdBuffer
;
3590 struct anv_render_pass
*pass
= (struct anv_render_pass
*) pRenderPassBegin
->renderPass
;
3591 struct anv_framebuffer
*framebuffer
=
3592 (struct anv_framebuffer
*) pRenderPassBegin
->framebuffer
;
3594 cmd_buffer
->framebuffer
= framebuffer
;
3596 anv_batch_emit(&cmd_buffer
->batch
, GEN8_3DSTATE_DRAWING_RECTANGLE
,
3597 .ClippedDrawingRectangleYMin
= pass
->render_area
.offset
.y
,
3598 .ClippedDrawingRectangleXMin
= pass
->render_area
.offset
.x
,
3599 .ClippedDrawingRectangleYMax
=
3600 pass
->render_area
.offset
.y
+ pass
->render_area
.extent
.height
- 1,
3601 .ClippedDrawingRectangleXMax
=
3602 pass
->render_area
.offset
.x
+ pass
->render_area
.extent
.width
- 1,
3603 .DrawingRectangleOriginY
= 0,
3604 .DrawingRectangleOriginX
= 0);
3606 anv_cmd_buffer_fill_render_targets(cmd_buffer
);
3608 anv_cmd_buffer_emit_depth_stencil(cmd_buffer
, pass
);
3610 anv_cmd_buffer_clear(cmd_buffer
, pass
);
3613 void anv_CmdEndRenderPass(
3614 VkCmdBuffer cmdBuffer
,
3615 VkRenderPass renderPass
)
3617 /* Emit a flushing pipe control at the end of a pass. This is kind of a
3618 * hack but it ensures that render targets always actually get written.
3619 * Eventually, we should do flushing based on image format transitions
3620 * or something of that nature.
3622 struct anv_cmd_buffer
*cmd_buffer
= (struct anv_cmd_buffer
*)cmdBuffer
;
3623 anv_batch_emit(&cmd_buffer
->batch
, GEN8_PIPE_CONTROL
,
3624 .PostSyncOperation
= NoWrite
,
3625 .RenderTargetCacheFlushEnable
= true,
3626 .InstructionCacheInvalidateEnable
= true,
3627 .DepthCacheFlushEnable
= true,
3628 .VFCacheInvalidationEnable
= true,
3629 .TextureCacheInvalidationEnable
= true,
3630 .CommandStreamerStallEnable
= true);
3633 void vkCmdDbgMarkerBegin(
3634 VkCmdBuffer cmdBuffer
,
3635 const char* pMarker
)
3636 __attribute__ ((visibility ("default")));
3638 void vkCmdDbgMarkerEnd(
3639 VkCmdBuffer cmdBuffer
)
3640 __attribute__ ((visibility ("default")));
3642 VkResult
vkDbgSetObjectTag(
3647 __attribute__ ((visibility ("default")));
3650 void vkCmdDbgMarkerBegin(
3651 VkCmdBuffer cmdBuffer
,
3652 const char* pMarker
)
3656 void vkCmdDbgMarkerEnd(
3657 VkCmdBuffer cmdBuffer
)
3661 VkResult
vkDbgSetObjectTag(