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
30 #include "anv_private.h"
32 #include "vk_format_info.h"
34 /** \file anv_cmd_buffer.c
36 * This file contains all of the stuff for emitting commands into a command
37 * buffer. This includes implementations of most of the vkCmd*
38 * entrypoints. This file is concerned entirely with state emission and
39 * not with the command buffer data structure itself. As far as this file
40 * is concerned, most of anv_cmd_buffer is magic.
43 /* TODO: These are taken from GLES. We should check the Vulkan spec */
44 const struct anv_dynamic_state default_dynamic_state
= {
57 .blend_constants
= { 0.0f
, 0.0f
, 0.0f
, 0.0f
},
62 .stencil_compare_mask
= {
66 .stencil_write_mask
= {
70 .stencil_reference
= {
77 anv_dynamic_state_copy(struct anv_dynamic_state
*dest
,
78 const struct anv_dynamic_state
*src
,
81 if (copy_mask
& (1 << VK_DYNAMIC_STATE_VIEWPORT
)) {
82 dest
->viewport
.count
= src
->viewport
.count
;
83 typed_memcpy(dest
->viewport
.viewports
, src
->viewport
.viewports
,
87 if (copy_mask
& (1 << VK_DYNAMIC_STATE_SCISSOR
)) {
88 dest
->scissor
.count
= src
->scissor
.count
;
89 typed_memcpy(dest
->scissor
.scissors
, src
->scissor
.scissors
,
93 if (copy_mask
& (1 << VK_DYNAMIC_STATE_LINE_WIDTH
))
94 dest
->line_width
= src
->line_width
;
96 if (copy_mask
& (1 << VK_DYNAMIC_STATE_DEPTH_BIAS
))
97 dest
->depth_bias
= src
->depth_bias
;
99 if (copy_mask
& (1 << VK_DYNAMIC_STATE_BLEND_CONSTANTS
))
100 typed_memcpy(dest
->blend_constants
, src
->blend_constants
, 4);
102 if (copy_mask
& (1 << VK_DYNAMIC_STATE_DEPTH_BOUNDS
))
103 dest
->depth_bounds
= src
->depth_bounds
;
105 if (copy_mask
& (1 << VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK
))
106 dest
->stencil_compare_mask
= src
->stencil_compare_mask
;
108 if (copy_mask
& (1 << VK_DYNAMIC_STATE_STENCIL_WRITE_MASK
))
109 dest
->stencil_write_mask
= src
->stencil_write_mask
;
111 if (copy_mask
& (1 << VK_DYNAMIC_STATE_STENCIL_REFERENCE
))
112 dest
->stencil_reference
= src
->stencil_reference
;
116 anv_cmd_state_reset(struct anv_cmd_buffer
*cmd_buffer
)
118 struct anv_cmd_state
*state
= &cmd_buffer
->state
;
120 memset(&state
->descriptors
, 0, sizeof(state
->descriptors
));
121 memset(&state
->push_constants
, 0, sizeof(state
->push_constants
));
122 memset(state
->binding_tables
, 0, sizeof(state
->binding_tables
));
123 memset(state
->samplers
, 0, sizeof(state
->samplers
));
125 /* 0 isn't a valid config. This ensures that we always configure L3$. */
126 cmd_buffer
->state
.current_l3_config
= 0;
130 state
->pending_pipe_bits
= 0;
131 state
->descriptors_dirty
= 0;
132 state
->push_constants_dirty
= 0;
133 state
->pipeline
= NULL
;
134 state
->push_constant_stages
= 0;
135 state
->restart_index
= UINT32_MAX
;
136 state
->dynamic
= default_dynamic_state
;
137 state
->need_query_wa
= true;
139 if (state
->attachments
!= NULL
) {
140 vk_free(&cmd_buffer
->pool
->alloc
, state
->attachments
);
141 state
->attachments
= NULL
;
144 state
->gen7
.index_buffer
= NULL
;
148 anv_cmd_buffer_ensure_push_constants_size(struct anv_cmd_buffer
*cmd_buffer
,
149 gl_shader_stage stage
, uint32_t size
)
151 struct anv_push_constants
**ptr
= &cmd_buffer
->state
.push_constants
[stage
];
154 *ptr
= vk_alloc(&cmd_buffer
->pool
->alloc
, size
, 8,
155 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
157 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
158 } else if ((*ptr
)->size
< size
) {
159 *ptr
= vk_realloc(&cmd_buffer
->pool
->alloc
, *ptr
, size
, 8,
160 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
162 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
169 static VkResult
anv_create_cmd_buffer(
170 struct anv_device
* device
,
171 struct anv_cmd_pool
* pool
,
172 VkCommandBufferLevel level
,
173 VkCommandBuffer
* pCommandBuffer
)
175 struct anv_cmd_buffer
*cmd_buffer
;
178 cmd_buffer
= vk_alloc(&pool
->alloc
, sizeof(*cmd_buffer
), 8,
179 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
180 if (cmd_buffer
== NULL
)
181 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
183 cmd_buffer
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
184 cmd_buffer
->device
= device
;
185 cmd_buffer
->pool
= pool
;
186 cmd_buffer
->level
= level
;
187 cmd_buffer
->state
.attachments
= NULL
;
189 result
= anv_cmd_buffer_init_batch_bo_chain(cmd_buffer
);
190 if (result
!= VK_SUCCESS
)
193 anv_state_stream_init(&cmd_buffer
->surface_state_stream
,
194 &device
->surface_state_block_pool
);
195 anv_state_stream_init(&cmd_buffer
->dynamic_state_stream
,
196 &device
->dynamic_state_block_pool
);
199 list_addtail(&cmd_buffer
->pool_link
, &pool
->cmd_buffers
);
201 /* Init the pool_link so we can safefly call list_del when we destroy
204 list_inithead(&cmd_buffer
->pool_link
);
207 *pCommandBuffer
= anv_cmd_buffer_to_handle(cmd_buffer
);
212 vk_free(&cmd_buffer
->pool
->alloc
, cmd_buffer
);
217 VkResult
anv_AllocateCommandBuffers(
219 const VkCommandBufferAllocateInfo
* pAllocateInfo
,
220 VkCommandBuffer
* pCommandBuffers
)
222 ANV_FROM_HANDLE(anv_device
, device
, _device
);
223 ANV_FROM_HANDLE(anv_cmd_pool
, pool
, pAllocateInfo
->commandPool
);
225 VkResult result
= VK_SUCCESS
;
228 for (i
= 0; i
< pAllocateInfo
->commandBufferCount
; i
++) {
229 result
= anv_create_cmd_buffer(device
, pool
, pAllocateInfo
->level
,
230 &pCommandBuffers
[i
]);
231 if (result
!= VK_SUCCESS
)
235 if (result
!= VK_SUCCESS
) {
236 anv_FreeCommandBuffers(_device
, pAllocateInfo
->commandPool
,
238 for (i
= 0; i
< pAllocateInfo
->commandBufferCount
; i
++)
239 pCommandBuffers
[i
] = VK_NULL_HANDLE
;
246 anv_cmd_buffer_destroy(struct anv_cmd_buffer
*cmd_buffer
)
248 list_del(&cmd_buffer
->pool_link
);
250 anv_cmd_buffer_fini_batch_bo_chain(cmd_buffer
);
252 anv_state_stream_finish(&cmd_buffer
->surface_state_stream
);
253 anv_state_stream_finish(&cmd_buffer
->dynamic_state_stream
);
255 vk_free(&cmd_buffer
->pool
->alloc
, cmd_buffer
->state
.attachments
);
256 vk_free(&cmd_buffer
->pool
->alloc
, cmd_buffer
);
259 void anv_FreeCommandBuffers(
261 VkCommandPool commandPool
,
262 uint32_t commandBufferCount
,
263 const VkCommandBuffer
* pCommandBuffers
)
265 for (uint32_t i
= 0; i
< commandBufferCount
; i
++) {
266 ANV_FROM_HANDLE(anv_cmd_buffer
, cmd_buffer
, pCommandBuffers
[i
]);
271 anv_cmd_buffer_destroy(cmd_buffer
);
276 anv_cmd_buffer_reset(struct anv_cmd_buffer
*cmd_buffer
)
278 cmd_buffer
->usage_flags
= 0;
279 cmd_buffer
->state
.current_pipeline
= UINT32_MAX
;
280 anv_cmd_buffer_reset_batch_bo_chain(cmd_buffer
);
281 anv_cmd_state_reset(cmd_buffer
);
283 anv_state_stream_finish(&cmd_buffer
->surface_state_stream
);
284 anv_state_stream_init(&cmd_buffer
->surface_state_stream
,
285 &cmd_buffer
->device
->surface_state_block_pool
);
287 anv_state_stream_finish(&cmd_buffer
->dynamic_state_stream
);
288 anv_state_stream_init(&cmd_buffer
->dynamic_state_stream
,
289 &cmd_buffer
->device
->dynamic_state_block_pool
);
293 VkResult
anv_ResetCommandBuffer(
294 VkCommandBuffer commandBuffer
,
295 VkCommandBufferResetFlags flags
)
297 ANV_FROM_HANDLE(anv_cmd_buffer
, cmd_buffer
, commandBuffer
);
298 return anv_cmd_buffer_reset(cmd_buffer
);
302 anv_cmd_buffer_emit_state_base_address(struct anv_cmd_buffer
*cmd_buffer
)
304 switch (cmd_buffer
->device
->info
.gen
) {
306 if (cmd_buffer
->device
->info
.is_haswell
)
307 return gen75_cmd_buffer_emit_state_base_address(cmd_buffer
);
309 return gen7_cmd_buffer_emit_state_base_address(cmd_buffer
);
311 return gen8_cmd_buffer_emit_state_base_address(cmd_buffer
);
313 return gen9_cmd_buffer_emit_state_base_address(cmd_buffer
);
315 unreachable("unsupported gen\n");
319 void anv_CmdBindPipeline(
320 VkCommandBuffer commandBuffer
,
321 VkPipelineBindPoint pipelineBindPoint
,
322 VkPipeline _pipeline
)
324 ANV_FROM_HANDLE(anv_cmd_buffer
, cmd_buffer
, commandBuffer
);
325 ANV_FROM_HANDLE(anv_pipeline
, pipeline
, _pipeline
);
327 switch (pipelineBindPoint
) {
328 case VK_PIPELINE_BIND_POINT_COMPUTE
:
329 cmd_buffer
->state
.compute_pipeline
= pipeline
;
330 cmd_buffer
->state
.compute_dirty
|= ANV_CMD_DIRTY_PIPELINE
;
331 cmd_buffer
->state
.push_constants_dirty
|= VK_SHADER_STAGE_COMPUTE_BIT
;
332 cmd_buffer
->state
.descriptors_dirty
|= VK_SHADER_STAGE_COMPUTE_BIT
;
335 case VK_PIPELINE_BIND_POINT_GRAPHICS
:
336 cmd_buffer
->state
.pipeline
= pipeline
;
337 cmd_buffer
->state
.vb_dirty
|= pipeline
->vb_used
;
338 cmd_buffer
->state
.dirty
|= ANV_CMD_DIRTY_PIPELINE
;
339 cmd_buffer
->state
.push_constants_dirty
|= pipeline
->active_stages
;
340 cmd_buffer
->state
.descriptors_dirty
|= pipeline
->active_stages
;
342 /* Apply the dynamic state from the pipeline */
343 cmd_buffer
->state
.dirty
|= pipeline
->dynamic_state_mask
;
344 anv_dynamic_state_copy(&cmd_buffer
->state
.dynamic
,
345 &pipeline
->dynamic_state
,
346 pipeline
->dynamic_state_mask
);
350 assert(!"invalid bind point");
355 void anv_CmdSetViewport(
356 VkCommandBuffer commandBuffer
,
357 uint32_t firstViewport
,
358 uint32_t viewportCount
,
359 const VkViewport
* pViewports
)
361 ANV_FROM_HANDLE(anv_cmd_buffer
, cmd_buffer
, commandBuffer
);
363 const uint32_t total_count
= firstViewport
+ viewportCount
;
364 if (cmd_buffer
->state
.dynamic
.viewport
.count
< total_count
)
365 cmd_buffer
->state
.dynamic
.viewport
.count
= total_count
;
367 memcpy(cmd_buffer
->state
.dynamic
.viewport
.viewports
+ firstViewport
,
368 pViewports
, viewportCount
* sizeof(*pViewports
));
370 cmd_buffer
->state
.dirty
|= ANV_CMD_DIRTY_DYNAMIC_VIEWPORT
;
373 void anv_CmdSetScissor(
374 VkCommandBuffer commandBuffer
,
375 uint32_t firstScissor
,
376 uint32_t scissorCount
,
377 const VkRect2D
* pScissors
)
379 ANV_FROM_HANDLE(anv_cmd_buffer
, cmd_buffer
, commandBuffer
);
381 const uint32_t total_count
= firstScissor
+ scissorCount
;
382 if (cmd_buffer
->state
.dynamic
.scissor
.count
< total_count
)
383 cmd_buffer
->state
.dynamic
.scissor
.count
= total_count
;
385 memcpy(cmd_buffer
->state
.dynamic
.scissor
.scissors
+ firstScissor
,
386 pScissors
, scissorCount
* sizeof(*pScissors
));
388 cmd_buffer
->state
.dirty
|= ANV_CMD_DIRTY_DYNAMIC_SCISSOR
;
391 void anv_CmdSetLineWidth(
392 VkCommandBuffer commandBuffer
,
395 ANV_FROM_HANDLE(anv_cmd_buffer
, cmd_buffer
, commandBuffer
);
397 cmd_buffer
->state
.dynamic
.line_width
= lineWidth
;
398 cmd_buffer
->state
.dirty
|= ANV_CMD_DIRTY_DYNAMIC_LINE_WIDTH
;
401 void anv_CmdSetDepthBias(
402 VkCommandBuffer commandBuffer
,
403 float depthBiasConstantFactor
,
404 float depthBiasClamp
,
405 float depthBiasSlopeFactor
)
407 ANV_FROM_HANDLE(anv_cmd_buffer
, cmd_buffer
, commandBuffer
);
409 cmd_buffer
->state
.dynamic
.depth_bias
.bias
= depthBiasConstantFactor
;
410 cmd_buffer
->state
.dynamic
.depth_bias
.clamp
= depthBiasClamp
;
411 cmd_buffer
->state
.dynamic
.depth_bias
.slope
= depthBiasSlopeFactor
;
413 cmd_buffer
->state
.dirty
|= ANV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS
;
416 void anv_CmdSetBlendConstants(
417 VkCommandBuffer commandBuffer
,
418 const float blendConstants
[4])
420 ANV_FROM_HANDLE(anv_cmd_buffer
, cmd_buffer
, commandBuffer
);
422 memcpy(cmd_buffer
->state
.dynamic
.blend_constants
,
423 blendConstants
, sizeof(float) * 4);
425 cmd_buffer
->state
.dirty
|= ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS
;
428 void anv_CmdSetDepthBounds(
429 VkCommandBuffer commandBuffer
,
430 float minDepthBounds
,
431 float maxDepthBounds
)
433 ANV_FROM_HANDLE(anv_cmd_buffer
, cmd_buffer
, commandBuffer
);
435 cmd_buffer
->state
.dynamic
.depth_bounds
.min
= minDepthBounds
;
436 cmd_buffer
->state
.dynamic
.depth_bounds
.max
= maxDepthBounds
;
438 cmd_buffer
->state
.dirty
|= ANV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS
;
441 void anv_CmdSetStencilCompareMask(
442 VkCommandBuffer commandBuffer
,
443 VkStencilFaceFlags faceMask
,
444 uint32_t compareMask
)
446 ANV_FROM_HANDLE(anv_cmd_buffer
, cmd_buffer
, commandBuffer
);
448 if (faceMask
& VK_STENCIL_FACE_FRONT_BIT
)
449 cmd_buffer
->state
.dynamic
.stencil_compare_mask
.front
= compareMask
;
450 if (faceMask
& VK_STENCIL_FACE_BACK_BIT
)
451 cmd_buffer
->state
.dynamic
.stencil_compare_mask
.back
= compareMask
;
453 cmd_buffer
->state
.dirty
|= ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK
;
456 void anv_CmdSetStencilWriteMask(
457 VkCommandBuffer commandBuffer
,
458 VkStencilFaceFlags faceMask
,
461 ANV_FROM_HANDLE(anv_cmd_buffer
, cmd_buffer
, commandBuffer
);
463 if (faceMask
& VK_STENCIL_FACE_FRONT_BIT
)
464 cmd_buffer
->state
.dynamic
.stencil_write_mask
.front
= writeMask
;
465 if (faceMask
& VK_STENCIL_FACE_BACK_BIT
)
466 cmd_buffer
->state
.dynamic
.stencil_write_mask
.back
= writeMask
;
468 cmd_buffer
->state
.dirty
|= ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK
;
471 void anv_CmdSetStencilReference(
472 VkCommandBuffer commandBuffer
,
473 VkStencilFaceFlags faceMask
,
476 ANV_FROM_HANDLE(anv_cmd_buffer
, cmd_buffer
, commandBuffer
);
478 if (faceMask
& VK_STENCIL_FACE_FRONT_BIT
)
479 cmd_buffer
->state
.dynamic
.stencil_reference
.front
= reference
;
480 if (faceMask
& VK_STENCIL_FACE_BACK_BIT
)
481 cmd_buffer
->state
.dynamic
.stencil_reference
.back
= reference
;
483 cmd_buffer
->state
.dirty
|= ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE
;
486 void anv_CmdBindDescriptorSets(
487 VkCommandBuffer commandBuffer
,
488 VkPipelineBindPoint pipelineBindPoint
,
489 VkPipelineLayout _layout
,
491 uint32_t descriptorSetCount
,
492 const VkDescriptorSet
* pDescriptorSets
,
493 uint32_t dynamicOffsetCount
,
494 const uint32_t* pDynamicOffsets
)
496 ANV_FROM_HANDLE(anv_cmd_buffer
, cmd_buffer
, commandBuffer
);
497 ANV_FROM_HANDLE(anv_pipeline_layout
, layout
, _layout
);
498 struct anv_descriptor_set_layout
*set_layout
;
500 assert(firstSet
+ descriptorSetCount
< MAX_SETS
);
502 for (uint32_t i
= 0; i
< descriptorSetCount
; i
++) {
503 ANV_FROM_HANDLE(anv_descriptor_set
, set
, pDescriptorSets
[i
]);
504 set_layout
= layout
->set
[firstSet
+ i
].layout
;
506 if (cmd_buffer
->state
.descriptors
[firstSet
+ i
] != set
) {
507 cmd_buffer
->state
.descriptors
[firstSet
+ i
] = set
;
508 cmd_buffer
->state
.descriptors_dirty
|= set_layout
->shader_stages
;
511 if (set_layout
->dynamic_offset_count
> 0) {
512 anv_foreach_stage(s
, set_layout
->shader_stages
) {
513 anv_cmd_buffer_ensure_push_constant_field(cmd_buffer
, s
, dynamic
);
515 struct anv_push_constants
*push
=
516 cmd_buffer
->state
.push_constants
[s
];
518 unsigned d
= layout
->set
[firstSet
+ i
].dynamic_offset_start
;
519 const uint32_t *offsets
= pDynamicOffsets
;
520 struct anv_descriptor
*desc
= set
->descriptors
;
522 for (unsigned b
= 0; b
< set_layout
->binding_count
; b
++) {
523 if (set_layout
->binding
[b
].dynamic_offset_index
< 0)
526 unsigned array_size
= set_layout
->binding
[b
].array_size
;
527 for (unsigned j
= 0; j
< array_size
; j
++) {
528 push
->dynamic
[d
].offset
= *(offsets
++);
529 push
->dynamic
[d
].range
= (desc
->buffer_view
) ?
530 desc
->buffer_view
->range
: 0;
536 cmd_buffer
->state
.push_constants_dirty
|= set_layout
->shader_stages
;
541 void anv_CmdBindVertexBuffers(
542 VkCommandBuffer commandBuffer
,
543 uint32_t firstBinding
,
544 uint32_t bindingCount
,
545 const VkBuffer
* pBuffers
,
546 const VkDeviceSize
* pOffsets
)
548 ANV_FROM_HANDLE(anv_cmd_buffer
, cmd_buffer
, commandBuffer
);
549 struct anv_vertex_binding
*vb
= cmd_buffer
->state
.vertex_bindings
;
551 /* We have to defer setting up vertex buffer since we need the buffer
552 * stride from the pipeline. */
554 assert(firstBinding
+ bindingCount
< MAX_VBS
);
555 for (uint32_t i
= 0; i
< bindingCount
; i
++) {
556 vb
[firstBinding
+ i
].buffer
= anv_buffer_from_handle(pBuffers
[i
]);
557 vb
[firstBinding
+ i
].offset
= pOffsets
[i
];
558 cmd_buffer
->state
.vb_dirty
|= 1 << (firstBinding
+ i
);
563 anv_isl_format_for_descriptor_type(VkDescriptorType type
)
566 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER
:
567 case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC
:
568 return ISL_FORMAT_R32G32B32A32_FLOAT
;
570 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER
:
571 case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC
:
572 return ISL_FORMAT_RAW
;
575 unreachable("Invalid descriptor type");
580 anv_cmd_buffer_emit_dynamic(struct anv_cmd_buffer
*cmd_buffer
,
581 const void *data
, uint32_t size
, uint32_t alignment
)
583 struct anv_state state
;
585 state
= anv_cmd_buffer_alloc_dynamic_state(cmd_buffer
, size
, alignment
);
586 memcpy(state
.map
, data
, size
);
588 if (!cmd_buffer
->device
->info
.has_llc
)
589 anv_state_clflush(state
);
591 VG(VALGRIND_CHECK_MEM_IS_DEFINED(state
.map
, size
));
597 anv_cmd_buffer_merge_dynamic(struct anv_cmd_buffer
*cmd_buffer
,
598 uint32_t *a
, uint32_t *b
,
599 uint32_t dwords
, uint32_t alignment
)
601 struct anv_state state
;
604 state
= anv_cmd_buffer_alloc_dynamic_state(cmd_buffer
,
605 dwords
* 4, alignment
);
607 for (uint32_t i
= 0; i
< dwords
; i
++)
610 if (!cmd_buffer
->device
->info
.has_llc
)
611 anv_state_clflush(state
);
613 VG(VALGRIND_CHECK_MEM_IS_DEFINED(p
, dwords
* 4));
619 anv_cmd_buffer_push_constants(struct anv_cmd_buffer
*cmd_buffer
,
620 gl_shader_stage stage
)
622 /* If we don't have this stage, bail. */
623 if (!anv_pipeline_has_stage(cmd_buffer
->state
.pipeline
, stage
))
624 return (struct anv_state
) { .offset
= 0 };
626 struct anv_push_constants
*data
=
627 cmd_buffer
->state
.push_constants
[stage
];
628 const struct brw_stage_prog_data
*prog_data
=
629 cmd_buffer
->state
.pipeline
->shaders
[stage
]->prog_data
;
631 /* If we don't actually have any push constants, bail. */
632 if (data
== NULL
|| prog_data
== NULL
|| prog_data
->nr_params
== 0)
633 return (struct anv_state
) { .offset
= 0 };
635 struct anv_state state
=
636 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer
,
637 prog_data
->nr_params
* sizeof(float),
638 32 /* bottom 5 bits MBZ */);
640 /* Walk through the param array and fill the buffer with data */
641 uint32_t *u32_map
= state
.map
;
642 for (unsigned i
= 0; i
< prog_data
->nr_params
; i
++) {
643 uint32_t offset
= (uintptr_t)prog_data
->param
[i
];
644 u32_map
[i
] = *(uint32_t *)((uint8_t *)data
+ offset
);
647 if (!cmd_buffer
->device
->info
.has_llc
)
648 anv_state_clflush(state
);
654 anv_cmd_buffer_cs_push_constants(struct anv_cmd_buffer
*cmd_buffer
)
656 struct anv_push_constants
*data
=
657 cmd_buffer
->state
.push_constants
[MESA_SHADER_COMPUTE
];
658 struct anv_pipeline
*pipeline
= cmd_buffer
->state
.compute_pipeline
;
659 const struct brw_cs_prog_data
*cs_prog_data
= get_cs_prog_data(pipeline
);
660 const struct brw_stage_prog_data
*prog_data
= &cs_prog_data
->base
;
662 /* If we don't actually have any push constants, bail. */
663 if (cs_prog_data
->push
.total
.size
== 0)
664 return (struct anv_state
) { .offset
= 0 };
666 const unsigned push_constant_alignment
=
667 cmd_buffer
->device
->info
.gen
< 8 ? 32 : 64;
668 const unsigned aligned_total_push_constants_size
=
669 ALIGN(cs_prog_data
->push
.total
.size
, push_constant_alignment
);
670 struct anv_state state
=
671 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer
,
672 aligned_total_push_constants_size
,
673 push_constant_alignment
);
675 /* Walk through the param array and fill the buffer with data */
676 uint32_t *u32_map
= state
.map
;
678 if (cs_prog_data
->push
.cross_thread
.size
> 0) {
679 assert(cs_prog_data
->thread_local_id_index
< 0 ||
680 cs_prog_data
->thread_local_id_index
>=
681 cs_prog_data
->push
.cross_thread
.dwords
);
683 i
< cs_prog_data
->push
.cross_thread
.dwords
;
685 uint32_t offset
= (uintptr_t)prog_data
->param
[i
];
686 u32_map
[i
] = *(uint32_t *)((uint8_t *)data
+ offset
);
690 if (cs_prog_data
->push
.per_thread
.size
> 0) {
691 for (unsigned t
= 0; t
< cs_prog_data
->threads
; t
++) {
693 8 * (cs_prog_data
->push
.per_thread
.regs
* t
+
694 cs_prog_data
->push
.cross_thread
.regs
);
695 unsigned src
= cs_prog_data
->push
.cross_thread
.dwords
;
696 for ( ; src
< prog_data
->nr_params
; src
++, dst
++) {
697 if (src
!= cs_prog_data
->thread_local_id_index
) {
698 uint32_t offset
= (uintptr_t)prog_data
->param
[src
];
699 u32_map
[dst
] = *(uint32_t *)((uint8_t *)data
+ offset
);
701 u32_map
[dst
] = t
* cs_prog_data
->simd_size
;
707 if (!cmd_buffer
->device
->info
.has_llc
)
708 anv_state_clflush(state
);
713 void anv_CmdPushConstants(
714 VkCommandBuffer commandBuffer
,
715 VkPipelineLayout layout
,
716 VkShaderStageFlags stageFlags
,
721 ANV_FROM_HANDLE(anv_cmd_buffer
, cmd_buffer
, commandBuffer
);
723 anv_foreach_stage(stage
, stageFlags
) {
724 anv_cmd_buffer_ensure_push_constant_field(cmd_buffer
, stage
, client_data
);
726 memcpy(cmd_buffer
->state
.push_constants
[stage
]->client_data
+ offset
,
730 cmd_buffer
->state
.push_constants_dirty
|= stageFlags
;
733 VkResult
anv_CreateCommandPool(
735 const VkCommandPoolCreateInfo
* pCreateInfo
,
736 const VkAllocationCallbacks
* pAllocator
,
737 VkCommandPool
* pCmdPool
)
739 ANV_FROM_HANDLE(anv_device
, device
, _device
);
740 struct anv_cmd_pool
*pool
;
742 pool
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*pool
), 8,
743 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
745 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
748 pool
->alloc
= *pAllocator
;
750 pool
->alloc
= device
->alloc
;
752 list_inithead(&pool
->cmd_buffers
);
754 *pCmdPool
= anv_cmd_pool_to_handle(pool
);
759 void anv_DestroyCommandPool(
761 VkCommandPool commandPool
,
762 const VkAllocationCallbacks
* pAllocator
)
764 ANV_FROM_HANDLE(anv_device
, device
, _device
);
765 ANV_FROM_HANDLE(anv_cmd_pool
, pool
, commandPool
);
770 list_for_each_entry_safe(struct anv_cmd_buffer
, cmd_buffer
,
771 &pool
->cmd_buffers
, pool_link
) {
772 anv_cmd_buffer_destroy(cmd_buffer
);
775 vk_free2(&device
->alloc
, pAllocator
, pool
);
778 VkResult
anv_ResetCommandPool(
780 VkCommandPool commandPool
,
781 VkCommandPoolResetFlags flags
)
783 ANV_FROM_HANDLE(anv_cmd_pool
, pool
, commandPool
);
785 list_for_each_entry(struct anv_cmd_buffer
, cmd_buffer
,
786 &pool
->cmd_buffers
, pool_link
) {
787 anv_cmd_buffer_reset(cmd_buffer
);
793 void anv_TrimCommandPoolKHR(
795 VkCommandPool commandPool
,
796 VkCommandPoolTrimFlagsKHR flags
)
798 /* Nothing for us to do here. Our pools stay pretty tidy. */
802 * Return NULL if the current subpass has no depthstencil attachment.
804 const struct anv_image_view
*
805 anv_cmd_buffer_get_depth_stencil_view(const struct anv_cmd_buffer
*cmd_buffer
)
807 const struct anv_subpass
*subpass
= cmd_buffer
->state
.subpass
;
808 const struct anv_framebuffer
*fb
= cmd_buffer
->state
.framebuffer
;
810 if (subpass
->depth_stencil_attachment
== VK_ATTACHMENT_UNUSED
)
813 const struct anv_image_view
*iview
=
814 fb
->attachments
[subpass
->depth_stencil_attachment
];
816 assert(iview
->aspect_mask
& (VK_IMAGE_ASPECT_DEPTH_BIT
|
817 VK_IMAGE_ASPECT_STENCIL_BIT
));