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
37 #define __gen_validate_value(x) VALGRIND_CHECK_MEM_IS_DEFINED(&(x), sizeof(x))
42 #include "brw_device_info.h"
43 #include "util/macros.h"
44 #include "util/list.h"
47 #include <vulkan/vulkan.h>
48 #include <vulkan/vulkan_intel.h>
49 #include <vulkan/vk_ext_khr_swapchain.h>
50 #include <vulkan/vk_ext_khr_device_swapchain.h>
52 #include "anv_entrypoints.h"
53 #include "brw_context.h"
60 #define ICD_LOADER_MAGIC 0x01CDC0DE
62 typedef union _VK_LOADER_DATA
{
63 uintptr_t loaderMagic
;
67 #define anv_noreturn __attribute__((__noreturn__))
68 #define anv_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
70 #define MIN(a, b) ((a) < (b) ? (a) : (b))
71 #define MAX(a, b) ((a) > (b) ? (a) : (b))
73 static inline uint32_t
74 align_u32(uint32_t v
, uint32_t a
)
76 return (v
+ a
- 1) & ~(a
- 1);
80 align_i32(int32_t v
, int32_t a
)
82 return (v
+ a
- 1) & ~(a
- 1);
85 /** Alignment must be a power of 2. */
87 anv_is_aligned(uintmax_t n
, uintmax_t a
)
89 assert(a
== (a
& -a
));
90 return (n
& (a
- 1)) == 0;
93 static inline uint32_t
94 anv_minify(uint32_t n
, uint32_t levels
)
99 return MAX(n
>> levels
, 1);
103 anv_clear_mask(uint32_t *inout_mask
, uint32_t clear_mask
)
105 if (*inout_mask
& clear_mask
) {
106 *inout_mask
&= ~clear_mask
;
113 #define for_each_bit(b, dword) \
114 for (uint32_t __dword = (dword); \
115 (b) = __builtin_ffs(__dword) - 1, __dword; \
116 __dword &= ~(1 << (b)))
118 #define typed_memcpy(dest, src, count) ({ \
119 static_assert(sizeof(*src) == sizeof(*dest), ""); \
120 memcpy((dest), (src), (count) * sizeof(*(src))); \
123 /* Define no kernel as 1, since that's an illegal offset for a kernel */
127 VkStructureType sType
;
131 /* Whenever we generate an error, pass it through this function. Useful for
132 * debugging, where we can break on it. Only call at error site, not when
133 * propagating errors. Might be useful to plug in a stack trace here.
136 VkResult
__vk_errorf(VkResult error
, const char *file
, int line
, const char *format
, ...);
139 #define vk_error(error) __vk_errorf(error, __FILE__, __LINE__, NULL);
140 #define vk_errorf(error, format, ...) __vk_errorf(error, __FILE__, __LINE__, format, ## __VA_ARGS__);
142 #define vk_error(error) error
143 #define vk_errorf(error, format, ...) error
146 void __anv_finishme(const char *file
, int line
, const char *format
, ...)
147 anv_printflike(3, 4);
148 void anv_loge(const char *format
, ...) anv_printflike(1, 2);
149 void anv_loge_v(const char *format
, va_list va
);
152 * Print a FINISHME message, including its source location.
154 #define anv_finishme(format, ...) \
155 __anv_finishme(__FILE__, __LINE__, format, ##__VA_ARGS__);
157 /* A non-fatal assert. Useful for debugging. */
159 #define anv_assert(x) ({ \
160 if (unlikely(!(x))) \
161 fprintf(stderr, "%s:%d ASSERT: %s\n", __FILE__, __LINE__, #x); \
164 #define anv_assert(x)
168 * If a block of code is annotated with anv_validate, then the block runs only
172 #define anv_validate if (1)
174 #define anv_validate if (0)
177 void anv_abortf(const char *format
, ...) anv_noreturn
anv_printflike(1, 2);
178 void anv_abortfv(const char *format
, va_list va
) anv_noreturn
;
180 #define stub_return(v) \
182 anv_finishme("stub %s", __func__); \
188 anv_finishme("stub %s", __func__); \
193 * A dynamically growable, circular buffer. Elements are added at head and
194 * removed from tail. head and tail are free-running uint32_t indices and we
195 * only compute the modulo with size when accessing the array. This way,
196 * number of bytes in the queue is always head - tail, even in case of
203 uint32_t element_size
;
208 int anv_vector_init(struct anv_vector
*queue
, uint32_t element_size
, uint32_t size
);
209 void *anv_vector_add(struct anv_vector
*queue
);
210 void *anv_vector_remove(struct anv_vector
*queue
);
213 anv_vector_length(struct anv_vector
*queue
)
215 return (queue
->head
- queue
->tail
) / queue
->element_size
;
219 anv_vector_head(struct anv_vector
*vector
)
221 assert(vector
->tail
< vector
->head
);
222 return (void *)((char *)vector
->data
+
223 ((vector
->head
- vector
->element_size
) &
224 (vector
->size
- 1)));
228 anv_vector_tail(struct anv_vector
*vector
)
230 return (void *)((char *)vector
->data
+ (vector
->tail
& (vector
->size
- 1)));
234 anv_vector_finish(struct anv_vector
*queue
)
239 #define anv_vector_foreach(elem, queue) \
240 static_assert(__builtin_types_compatible_p(__typeof__(queue), struct anv_vector *), ""); \
241 for (uint32_t __anv_vector_offset = (queue)->tail; \
242 elem = (queue)->data + (__anv_vector_offset & ((queue)->size - 1)), __anv_vector_offset < (queue)->head; \
243 __anv_vector_offset += (queue)->element_size)
248 /* Index into the current validation list. This is used by the
249 * validation list building alrogithm to track which buffers are already
250 * in the validation list so that we can ensure uniqueness.
254 /* Last known offset. This value is provided by the kernel when we
255 * execbuf and is used as the presumed offset for the next bunch of
264 /* Represents a lock-free linked list of "free" things. This is used by
265 * both the block pool and the state pools. Unfortunately, in order to
266 * solve the ABA problem, we can't use a single uint32_t head.
268 union anv_free_list
{
272 /* A simple count that is incremented every time the head changes. */
278 #define ANV_FREE_LIST_EMPTY ((union anv_free_list) { { 1, 0 } })
280 struct anv_block_state
{
290 struct anv_block_pool
{
291 struct anv_device
*device
;
295 /* The offset from the start of the bo to the "center" of the block
296 * pool. Pointers to allocated blocks are given by
297 * bo.map + center_bo_offset + offsets.
299 uint32_t center_bo_offset
;
301 /* Current memory map of the block pool. This pointer may or may not
302 * point to the actual beginning of the block pool memory. If
303 * anv_block_pool_alloc_back has ever been called, then this pointer
304 * will point to the "center" position of the buffer and all offsets
305 * (negative or positive) given out by the block pool alloc functions
306 * will be valid relative to this pointer.
308 * In particular, map == bo.map + center_offset
314 * Array of mmaps and gem handles owned by the block pool, reclaimed when
315 * the block pool is destroyed.
317 struct anv_vector mmap_cleanups
;
321 union anv_free_list free_list
;
322 struct anv_block_state state
;
324 union anv_free_list back_free_list
;
325 struct anv_block_state back_state
;
328 /* Block pools are backed by a fixed-size 2GB memfd */
329 #define BLOCK_POOL_MEMFD_SIZE (1ull << 32)
331 /* The center of the block pool is also the middle of the memfd. This may
332 * change in the future if we decide differently for some reason.
334 #define BLOCK_POOL_MEMFD_CENTER (BLOCK_POOL_MEMFD_SIZE / 2)
336 static inline uint32_t
337 anv_block_pool_size(struct anv_block_pool
*pool
)
339 return pool
->state
.end
+ pool
->back_state
.end
;
348 struct anv_fixed_size_state_pool
{
350 union anv_free_list free_list
;
351 struct anv_block_state block
;
354 #define ANV_MIN_STATE_SIZE_LOG2 6
355 #define ANV_MAX_STATE_SIZE_LOG2 10
357 #define ANV_STATE_BUCKETS (ANV_MAX_STATE_SIZE_LOG2 - ANV_MIN_STATE_SIZE_LOG2)
359 struct anv_state_pool
{
360 struct anv_block_pool
*block_pool
;
361 struct anv_fixed_size_state_pool buckets
[ANV_STATE_BUCKETS
];
364 struct anv_state_stream
{
365 struct anv_block_pool
*block_pool
;
367 uint32_t current_block
;
371 void anv_block_pool_init(struct anv_block_pool
*pool
,
372 struct anv_device
*device
, uint32_t block_size
);
373 void anv_block_pool_finish(struct anv_block_pool
*pool
);
374 int32_t anv_block_pool_alloc(struct anv_block_pool
*pool
);
375 int32_t anv_block_pool_alloc_back(struct anv_block_pool
*pool
);
376 void anv_block_pool_free(struct anv_block_pool
*pool
, int32_t offset
);
377 void anv_state_pool_init(struct anv_state_pool
*pool
,
378 struct anv_block_pool
*block_pool
);
379 void anv_state_pool_finish(struct anv_state_pool
*pool
);
380 struct anv_state
anv_state_pool_alloc(struct anv_state_pool
*pool
,
381 size_t state_size
, size_t alignment
);
382 void anv_state_pool_free(struct anv_state_pool
*pool
, struct anv_state state
);
383 void anv_state_stream_init(struct anv_state_stream
*stream
,
384 struct anv_block_pool
*block_pool
);
385 void anv_state_stream_finish(struct anv_state_stream
*stream
);
386 struct anv_state
anv_state_stream_alloc(struct anv_state_stream
*stream
,
387 uint32_t size
, uint32_t alignment
);
390 * Implements a pool of re-usable BOs. The interface is identical to that
391 * of block_pool except that each block is its own BO.
394 struct anv_device
*device
;
401 void anv_bo_pool_init(struct anv_bo_pool
*pool
,
402 struct anv_device
*device
, uint32_t block_size
);
403 void anv_bo_pool_finish(struct anv_bo_pool
*pool
);
404 VkResult
anv_bo_pool_alloc(struct anv_bo_pool
*pool
, struct anv_bo
*bo
);
405 void anv_bo_pool_free(struct anv_bo_pool
*pool
, const struct anv_bo
*bo
);
408 void *anv_resolve_entrypoint(uint32_t index
);
410 extern struct anv_dispatch_table dtable
;
412 #define ANV_CALL(func) ({ \
413 if (dtable.func == NULL) { \
414 size_t idx = offsetof(struct anv_dispatch_table, func) / sizeof(void *); \
415 dtable.entrypoints[idx] = anv_resolve_entrypoint(idx); \
421 struct anv_physical_device
{
422 VK_LOADER_DATA _loader_data
;
424 struct anv_instance
* instance
;
428 const struct brw_device_info
* info
;
429 uint64_t aperture_size
;
430 struct brw_compiler
* compiler
;
431 struct isl_device isl_dev
;
434 bool anv_is_scalar_shader_stage(const struct brw_compiler
*compiler
,
435 VkShaderStage stage
);
437 struct anv_instance
{
438 VK_LOADER_DATA _loader_data
;
440 void * pAllocUserData
;
441 PFN_vkAllocFunction pfnAlloc
;
442 PFN_vkFreeFunction pfnFree
;
444 int physicalDeviceCount
;
445 struct anv_physical_device physicalDevice
;
447 struct anv_wsi_implementation
* wsi_impl
[VK_PLATFORM_NUM_KHR
];
450 VkResult
anv_init_wsi(struct anv_instance
*instance
);
451 void anv_finish_wsi(struct anv_instance
*instance
);
453 struct anv_meta_state
{
455 struct anv_pipeline
*color_pipeline
;
456 struct anv_pipeline
*depth_only_pipeline
;
457 struct anv_pipeline
*stencil_only_pipeline
;
458 struct anv_pipeline
*depthstencil_pipeline
;
462 VkRenderPass render_pass
;
464 /** Pipeline that blits from a 2D image. */
465 VkPipeline pipeline_2d_src
;
467 /** Pipeline that blits from a 3D image. */
468 VkPipeline pipeline_3d_src
;
470 VkPipelineLayout pipeline_layout
;
471 VkDescriptorSetLayout ds_layout
;
476 VK_LOADER_DATA _loader_data
;
478 struct anv_device
* device
;
480 struct anv_state_pool
* pool
;
484 VK_LOADER_DATA _loader_data
;
486 struct anv_instance
* instance
;
488 struct brw_device_info info
;
489 struct isl_device isl_dev
;
493 struct anv_bo_pool batch_bo_pool
;
495 struct anv_block_pool dynamic_state_block_pool
;
496 struct anv_state_pool dynamic_state_pool
;
498 struct anv_block_pool instruction_block_pool
;
499 struct anv_block_pool surface_state_block_pool
;
500 struct anv_state_pool surface_state_pool
;
502 struct anv_bo workaround_bo
;
504 struct anv_meta_state meta_state
;
506 struct anv_state border_colors
;
508 struct anv_queue queue
;
510 struct anv_block_pool scratch_block_pool
;
512 pthread_mutex_t mutex
;
516 anv_instance_alloc(struct anv_instance
* instance
,
519 VkSystemAllocType allocType
);
522 anv_instance_free(struct anv_instance
* instance
,
526 anv_device_alloc(struct anv_device
* device
,
529 VkSystemAllocType allocType
);
532 anv_device_free(struct anv_device
* device
,
535 void* anv_gem_mmap(struct anv_device
*device
,
536 uint32_t gem_handle
, uint64_t offset
, uint64_t size
);
537 void anv_gem_munmap(void *p
, uint64_t size
);
538 uint32_t anv_gem_create(struct anv_device
*device
, size_t size
);
539 void anv_gem_close(struct anv_device
*device
, int gem_handle
);
540 int anv_gem_userptr(struct anv_device
*device
, void *mem
, size_t size
);
541 int anv_gem_wait(struct anv_device
*device
, int gem_handle
, int64_t *timeout_ns
);
542 int anv_gem_execbuffer(struct anv_device
*device
,
543 struct drm_i915_gem_execbuffer2
*execbuf
);
544 int anv_gem_set_tiling(struct anv_device
*device
, int gem_handle
,
545 uint32_t stride
, uint32_t tiling
);
546 int anv_gem_create_context(struct anv_device
*device
);
547 int anv_gem_destroy_context(struct anv_device
*device
, int context
);
548 int anv_gem_get_param(int fd
, uint32_t param
);
549 int anv_gem_get_aperture(int fd
, uint64_t *size
);
550 int anv_gem_handle_to_fd(struct anv_device
*device
, int gem_handle
);
551 int anv_gem_fd_to_handle(struct anv_device
*device
, int fd
);
552 int anv_gem_userptr(struct anv_device
*device
, void *mem
, size_t size
);
554 VkResult
anv_bo_init_new(struct anv_bo
*bo
, struct anv_device
*device
, uint64_t size
);
556 struct anv_reloc_list
{
559 struct drm_i915_gem_relocation_entry
* relocs
;
560 struct anv_bo
** reloc_bos
;
563 VkResult
anv_reloc_list_init(struct anv_reloc_list
*list
,
564 struct anv_device
*device
);
565 void anv_reloc_list_finish(struct anv_reloc_list
*list
,
566 struct anv_device
*device
);
568 uint64_t anv_reloc_list_add(struct anv_reloc_list
*list
,
569 struct anv_device
*device
,
570 uint32_t offset
, struct anv_bo
*target_bo
,
573 struct anv_batch_bo
{
574 /* Link in the anv_cmd_buffer.owned_batch_bos list */
575 struct list_head link
;
579 /* Bytes actually consumed in this batch BO */
582 /* Last seen surface state block pool bo offset */
583 uint32_t last_ss_pool_bo_offset
;
585 struct anv_reloc_list relocs
;
589 struct anv_device
* device
;
595 struct anv_reloc_list
* relocs
;
597 /* This callback is called (with the associated user data) in the event
598 * that the batch runs out of space.
600 VkResult (*extend_cb
)(struct anv_batch
*, void *);
604 void *anv_batch_emit_dwords(struct anv_batch
*batch
, int num_dwords
);
605 void anv_batch_emit_batch(struct anv_batch
*batch
, struct anv_batch
*other
);
606 uint64_t anv_batch_emit_reloc(struct anv_batch
*batch
,
607 void *location
, struct anv_bo
*bo
, uint32_t offset
);
614 #define __gen_address_type struct anv_address
615 #define __gen_user_data struct anv_batch
617 static inline uint64_t
618 __gen_combine_address(struct anv_batch
*batch
, void *location
,
619 const struct anv_address address
, uint32_t delta
)
621 if (address
.bo
== NULL
) {
622 return address
.offset
+ delta
;
624 assert(batch
->start
<= location
&& location
< batch
->end
);
626 return anv_batch_emit_reloc(batch
, location
, address
.bo
, address
.offset
+ delta
);
630 #include "gen7_pack.h"
631 #include "gen75_pack.h"
632 #undef GEN8_3DSTATE_MULTISAMPLE
633 #include "gen8_pack.h"
635 #define anv_batch_emit(batch, cmd, ...) do { \
636 void *__dst = anv_batch_emit_dwords(batch, cmd ## _length); \
637 struct cmd __template = { \
641 cmd ## _pack(batch, __dst, &__template); \
642 VG(VALGRIND_CHECK_MEM_IS_DEFINED(__dst, cmd ## _length * 4)); \
645 #define anv_batch_emitn(batch, n, cmd, ...) ({ \
646 void *__dst = anv_batch_emit_dwords(batch, n); \
647 struct cmd __template = { \
649 .DwordLength = n - cmd ## _length_bias, \
652 cmd ## _pack(batch, __dst, &__template); \
656 #define anv_batch_emit_merge(batch, dwords0, dwords1) \
660 assert(ARRAY_SIZE(dwords0) == ARRAY_SIZE(dwords1)); \
661 dw = anv_batch_emit_dwords((batch), ARRAY_SIZE(dwords0)); \
662 for (uint32_t i = 0; i < ARRAY_SIZE(dwords0); i++) \
663 dw[i] = (dwords0)[i] | (dwords1)[i]; \
664 VG(VALGRIND_CHECK_MEM_IS_DEFINED(dw, ARRAY_SIZE(dwords0) * 4));\
667 static const struct GEN7_MEMORY_OBJECT_CONTROL_STATE GEN7_MOCS
= {
668 .GraphicsDataTypeGFDT
= 0,
669 .LLCCacheabilityControlLLCCC
= 0,
670 .L3CacheabilityControlL3CC
= 1
673 #define GEN8_MOCS { \
674 .MemoryTypeLLCeLLCCacheabilityControl = WB, \
675 .TargetCache = L3DefertoPATforLLCeLLCselection, \
679 struct anv_device_memory
{
681 VkDeviceSize map_size
;
686 * Header for Vertex URB Entry (VUE)
688 struct anv_vue_header
{
690 uint32_t RTAIndex
; /* RenderTargetArrayIndex */
691 uint32_t ViewportIndex
;
695 struct anv_descriptor_set_binding_layout
{
696 /* Number of array elements in this binding */
699 /* Index into the flattend descriptor set */
700 uint16_t descriptor_index
;
702 /* Index into the dynamic state array for a dynamic buffer */
703 int16_t dynamic_offset_index
;
706 /* Index into the binding table for the associated surface */
707 int16_t surface_index
;
709 /* Index into the sampler table for the associated sampler */
710 int16_t sampler_index
;
711 } stage
[VK_SHADER_STAGE_NUM
];
713 /* Immutable samplers (or NULL if no immutable samplers) */
714 struct anv_sampler
**immutable_samplers
;
717 struct anv_descriptor_set_layout
{
718 /* Number of bindings in this descriptor set */
719 uint16_t binding_count
;
721 /* Total size of the descriptor set with room for all array entries */
724 /* Shader stages affected by this descriptor set */
725 uint16_t shader_stages
;
727 /* Number of dynamic offsets used by this descriptor set */
728 uint16_t dynamic_offset_count
;
730 /* Bindings in this descriptor set */
731 struct anv_descriptor_set_binding_layout binding
[0];
734 struct anv_descriptor
{
735 VkDescriptorType type
;
740 struct anv_image_view
*image_view
;
742 struct anv_sampler
*sampler
;
746 struct anv_buffer
*buffer
;
753 struct anv_descriptor_set
{
754 const struct anv_descriptor_set_layout
*layout
;
755 struct anv_descriptor descriptors
[0];
759 anv_descriptor_set_create(struct anv_device
*device
,
760 const struct anv_descriptor_set_layout
*layout
,
761 struct anv_descriptor_set
**out_set
);
764 anv_descriptor_set_destroy(struct anv_device
*device
,
765 struct anv_descriptor_set
*set
);
770 #define MAX_VIEWPORTS 16
771 #define MAX_SCISSORS 16
772 #define MAX_PUSH_CONSTANTS_SIZE 128
773 #define MAX_DYNAMIC_BUFFERS 16
776 struct anv_pipeline_binding
{
777 /* The descriptor set this surface corresponds to */
780 /* Offset into the descriptor set */
784 struct anv_pipeline_layout
{
786 struct anv_descriptor_set_layout
*layout
;
787 uint32_t dynamic_offset_start
;
789 uint32_t surface_start
;
790 uint32_t sampler_start
;
791 } stage
[VK_SHADER_STAGE_NUM
];
797 bool has_dynamic_offsets
;
798 uint32_t surface_count
;
799 struct anv_pipeline_binding
*surface_to_descriptor
;
800 uint32_t sampler_count
;
801 struct anv_pipeline_binding
*sampler_to_descriptor
;
802 } stage
[VK_SHADER_STAGE_NUM
];
804 struct anv_pipeline_binding entries
[0];
808 struct anv_device
* device
;
816 enum anv_cmd_dirty_bits
{
817 ANV_CMD_DIRTY_DYNAMIC_VIEWPORT
= 1 << 0, /* VK_DYNAMIC_STATE_VIEWPORT */
818 ANV_CMD_DIRTY_DYNAMIC_SCISSOR
= 1 << 1, /* VK_DYNAMIC_STATE_SCISSOR */
819 ANV_CMD_DIRTY_DYNAMIC_LINE_WIDTH
= 1 << 2, /* VK_DYNAMIC_STATE_LINE_WIDTH */
820 ANV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS
= 1 << 3, /* VK_DYNAMIC_STATE_DEPTH_BIAS */
821 ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS
= 1 << 4, /* VK_DYNAMIC_STATE_BLEND_CONSTANTS */
822 ANV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS
= 1 << 5, /* VK_DYNAMIC_STATE_DEPTH_BOUNDS */
823 ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6, /* VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK */
824 ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7, /* VK_DYNAMIC_STATE_STENCIL_WRITE_MASK */
825 ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE
= 1 << 8, /* VK_DYNAMIC_STATE_STENCIL_REFERENCE */
826 ANV_CMD_DIRTY_DYNAMIC_ALL
= (1 << 9) - 1,
827 ANV_CMD_DIRTY_PIPELINE
= 1 << 9,
828 ANV_CMD_DIRTY_INDEX_BUFFER
= 1 << 10,
830 typedef uint32_t anv_cmd_dirty_mask_t
;
832 struct anv_vertex_binding
{
833 struct anv_buffer
* buffer
;
837 struct anv_push_constants
{
838 /* Current allocated size of this push constants data structure.
839 * Because a decent chunk of it may not be used (images on SKL, for
840 * instance), we won't actually allocate the entire structure up-front.
844 /* Push constant data provided by the client through vkPushConstants */
845 uint8_t client_data
[MAX_PUSH_CONSTANTS_SIZE
];
847 /* Our hardware only provides zero-based vertex and instance id so, in
848 * order to satisfy the vulkan requirements, we may have to push one or
849 * both of these into the shader.
851 uint32_t base_vertex
;
852 uint32_t base_instance
;
854 /* Offsets and ranges for dynamically bound buffers */
858 } dynamic
[MAX_DYNAMIC_BUFFERS
];
860 /* Image data for image_load_store on pre-SKL */
861 struct brw_image_param images
[MAX_IMAGES
];
864 struct anv_dynamic_state
{
867 VkViewport viewports
[MAX_VIEWPORTS
];
872 VkRect2D scissors
[MAX_SCISSORS
];
883 float blend_constants
[4];
893 } stencil_compare_mask
;
898 } stencil_write_mask
;
906 extern const struct anv_dynamic_state default_dynamic_state
;
908 void anv_dynamic_state_copy(struct anv_dynamic_state
*dest
,
909 const struct anv_dynamic_state
*src
,
912 /** State required while building cmd buffer */
913 struct anv_cmd_state
{
914 uint32_t current_pipeline
;
916 anv_cmd_dirty_mask_t dirty
;
917 anv_cmd_dirty_mask_t compute_dirty
;
918 VkShaderStageFlags descriptors_dirty
;
919 VkShaderStageFlags push_constants_dirty
;
920 uint32_t scratch_size
;
921 struct anv_pipeline
* pipeline
;
922 struct anv_pipeline
* compute_pipeline
;
923 struct anv_framebuffer
* framebuffer
;
924 struct anv_render_pass
* pass
;
925 struct anv_subpass
* subpass
;
926 uint32_t state_vf
[GEN8_3DSTATE_VF_length
];
927 struct anv_vertex_binding vertex_bindings
[MAX_VBS
];
928 struct anv_descriptor_set
* descriptors
[MAX_SETS
];
929 struct anv_push_constants
* push_constants
[VK_SHADER_STAGE_NUM
];
930 struct anv_dynamic_state dynamic
;
933 struct anv_buffer
* index_buffer
;
934 uint32_t index_type
; /**< 3DSTATE_INDEX_BUFFER.IndexFormat */
935 uint32_t index_offset
;
939 struct anv_cmd_pool
{
940 struct list_head cmd_buffers
;
943 #define ANV_CMD_BUFFER_BATCH_SIZE 8192
945 enum anv_cmd_buffer_exec_mode
{
946 ANV_CMD_BUFFER_EXEC_MODE_PRIMARY
,
947 ANV_CMD_BUFFER_EXEC_MODE_EMIT
,
948 ANV_CMD_BUFFER_EXEC_MODE_CHAIN
,
949 ANV_CMD_BUFFER_EXEC_MODE_COPY_AND_CHAIN
,
952 struct anv_cmd_buffer
{
953 VK_LOADER_DATA _loader_data
;
955 struct anv_device
* device
;
957 struct list_head pool_link
;
959 struct anv_batch batch
;
961 /* Fields required for the actual chain of anv_batch_bo's.
963 * These fields are initialized by anv_cmd_buffer_init_batch_bo_chain().
965 struct list_head batch_bos
;
966 enum anv_cmd_buffer_exec_mode exec_mode
;
968 /* A vector of anv_batch_bo pointers for every batch or surface buffer
969 * referenced by this command buffer
971 * initialized by anv_cmd_buffer_init_batch_bo_chain()
973 struct anv_vector seen_bbos
;
975 /* A vector of int32_t's for every block of binding tables.
977 * initialized by anv_cmd_buffer_init_batch_bo_chain()
979 struct anv_vector bt_blocks
;
981 struct anv_reloc_list surface_relocs
;
983 /* Information needed for execbuf
985 * These fields are generated by anv_cmd_buffer_prepare_execbuf().
988 struct drm_i915_gem_execbuffer2 execbuf
;
990 struct drm_i915_gem_exec_object2
* objects
;
992 struct anv_bo
** bos
;
994 /* Allocated length of the 'objects' and 'bos' arrays */
995 uint32_t array_length
;
1000 /* Serial for tracking buffer completion */
1003 /* Stream objects for storing temporary data */
1004 struct anv_state_stream surface_state_stream
;
1005 struct anv_state_stream dynamic_state_stream
;
1007 VkCmdBufferOptimizeFlags opt_flags
;
1008 VkCmdBufferLevel level
;
1010 struct anv_cmd_state state
;
1013 VkResult
anv_cmd_buffer_init_batch_bo_chain(struct anv_cmd_buffer
*cmd_buffer
);
1014 void anv_cmd_buffer_fini_batch_bo_chain(struct anv_cmd_buffer
*cmd_buffer
);
1015 void anv_cmd_buffer_reset_batch_bo_chain(struct anv_cmd_buffer
*cmd_buffer
);
1016 void anv_cmd_buffer_end_batch_buffer(struct anv_cmd_buffer
*cmd_buffer
);
1017 void anv_cmd_buffer_add_secondary(struct anv_cmd_buffer
*primary
,
1018 struct anv_cmd_buffer
*secondary
);
1019 void anv_cmd_buffer_prepare_execbuf(struct anv_cmd_buffer
*cmd_buffer
);
1021 VkResult
anv_cmd_buffer_emit_binding_table(struct anv_cmd_buffer
*cmd_buffer
,
1022 unsigned stage
, struct anv_state
*bt_state
);
1023 VkResult
anv_cmd_buffer_emit_samplers(struct anv_cmd_buffer
*cmd_buffer
,
1024 unsigned stage
, struct anv_state
*state
);
1025 void anv_flush_descriptor_sets(struct anv_cmd_buffer
*cmd_buffer
);
1027 struct anv_state
anv_cmd_buffer_emit_dynamic(struct anv_cmd_buffer
*cmd_buffer
,
1028 uint32_t *a
, uint32_t dwords
,
1029 uint32_t alignment
);
1030 struct anv_state
anv_cmd_buffer_merge_dynamic(struct anv_cmd_buffer
*cmd_buffer
,
1031 uint32_t *a
, uint32_t *b
,
1032 uint32_t dwords
, uint32_t alignment
);
1033 void anv_cmd_buffer_begin_subpass(struct anv_cmd_buffer
*cmd_buffer
,
1034 struct anv_subpass
*subpass
);
1037 anv_cmd_buffer_surface_base_address(struct anv_cmd_buffer
*cmd_buffer
);
1039 anv_cmd_buffer_alloc_binding_table(struct anv_cmd_buffer
*cmd_buffer
,
1040 uint32_t entries
, uint32_t *state_offset
);
1042 anv_cmd_buffer_alloc_surface_state(struct anv_cmd_buffer
*cmd_buffer
);
1044 anv_cmd_buffer_alloc_dynamic_state(struct anv_cmd_buffer
*cmd_buffer
,
1045 uint32_t size
, uint32_t alignment
);
1048 anv_cmd_buffer_new_binding_table_block(struct anv_cmd_buffer
*cmd_buffer
);
1050 void anv_cmd_buffer_emit_viewport(struct anv_cmd_buffer
*cmd_buffer
);
1051 void anv_cmd_buffer_emit_scissor(struct anv_cmd_buffer
*cmd_buffer
);
1053 void gen7_cmd_buffer_emit_state_base_address(struct anv_cmd_buffer
*cmd_buffer
);
1054 void gen8_cmd_buffer_emit_state_base_address(struct anv_cmd_buffer
*cmd_buffer
);
1056 void anv_cmd_buffer_emit_state_base_address(struct anv_cmd_buffer
*cmd_buffer
);
1058 void gen7_cmd_buffer_begin_subpass(struct anv_cmd_buffer
*cmd_buffer
,
1059 struct anv_subpass
*subpass
);
1061 void gen8_cmd_buffer_begin_subpass(struct anv_cmd_buffer
*cmd_buffer
,
1062 struct anv_subpass
*subpass
);
1064 void anv_cmd_buffer_begin_subpass(struct anv_cmd_buffer
*cmd_buffer
,
1065 struct anv_subpass
*subpass
);
1068 anv_cmd_buffer_push_constants(struct anv_cmd_buffer
*cmd_buffer
,
1069 VkShaderStage stage
);
1071 void anv_cmd_buffer_clear_attachments(struct anv_cmd_buffer
*cmd_buffer
,
1072 struct anv_render_pass
*pass
,
1073 const VkClearValue
*clear_values
);
1074 const struct anv_image_view
*
1075 anv_cmd_buffer_get_depth_stencil_view(const struct anv_cmd_buffer
*cmd_buffer
);
1077 void anv_cmd_buffer_dump(struct anv_cmd_buffer
*cmd_buffer
);
1081 struct drm_i915_gem_execbuffer2 execbuf
;
1082 struct drm_i915_gem_exec_object2 exec2_objects
[1];
1088 struct anv_shader_module
{
1089 struct nir_shader
* nir
;
1096 struct anv_shader_module
* module
;
1100 struct anv_pipeline
{
1101 struct anv_device
* device
;
1102 struct anv_batch batch
;
1103 uint32_t batch_data
[512];
1104 struct anv_reloc_list batch_relocs
;
1105 uint32_t dynamic_state_mask
;
1106 struct anv_dynamic_state dynamic_state
;
1108 struct anv_pipeline_layout
* layout
;
1111 struct brw_vs_prog_data vs_prog_data
;
1112 struct brw_wm_prog_data wm_prog_data
;
1113 struct brw_gs_prog_data gs_prog_data
;
1114 struct brw_cs_prog_data cs_prog_data
;
1115 bool writes_point_size
;
1116 struct brw_stage_prog_data
* prog_data
[VK_SHADER_STAGE_NUM
];
1117 uint32_t scratch_start
[VK_SHADER_STAGE_NUM
];
1118 uint32_t total_scratch
;
1122 uint32_t nr_vs_entries
;
1125 uint32_t nr_gs_entries
;
1128 VkShaderStageFlags active_stages
;
1129 struct anv_state_stream program_stream
;
1130 struct anv_state blend_state
;
1137 uint32_t ps_grf_start0
;
1138 uint32_t ps_grf_start2
;
1140 uint32_t gs_vertex_count
;
1144 uint32_t binding_stride
[MAX_VBS
];
1145 bool instancing_enable
[MAX_VBS
];
1146 bool primitive_restart
;
1149 uint32_t cs_thread_width_max
;
1150 uint32_t cs_right_mask
;
1153 uint32_t sf
[GEN7_3DSTATE_SF_length
];
1154 uint32_t depth_stencil_state
[GEN7_DEPTH_STENCIL_STATE_length
];
1158 uint32_t sf
[GEN8_3DSTATE_SF_length
];
1159 uint32_t vf
[GEN8_3DSTATE_VF_length
];
1160 uint32_t raster
[GEN8_3DSTATE_RASTER_length
];
1161 uint32_t wm_depth_stencil
[GEN8_3DSTATE_WM_DEPTH_STENCIL_length
];
1165 struct anv_graphics_pipeline_create_info
{
1167 bool disable_viewport
;
1168 bool disable_scissor
;
1174 anv_pipeline_init(struct anv_pipeline
*pipeline
, struct anv_device
*device
,
1175 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
1176 const struct anv_graphics_pipeline_create_info
*extra
);
1179 anv_pipeline_compile_cs(struct anv_pipeline
*pipeline
,
1180 const VkComputePipelineCreateInfo
*info
,
1181 struct anv_shader
*shader
);
1184 anv_graphics_pipeline_create(VkDevice device
,
1185 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
1186 const struct anv_graphics_pipeline_create_info
*extra
,
1187 VkPipeline
*pPipeline
);
1190 gen7_graphics_pipeline_create(VkDevice _device
,
1191 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
1192 const struct anv_graphics_pipeline_create_info
*extra
,
1193 VkPipeline
*pPipeline
);
1196 gen8_graphics_pipeline_create(VkDevice _device
,
1197 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
1198 const struct anv_graphics_pipeline_create_info
*extra
,
1199 VkPipeline
*pPipeline
);
1201 gen7_compute_pipeline_create(VkDevice _device
,
1202 const VkComputePipelineCreateInfo
*pCreateInfo
,
1203 VkPipeline
*pPipeline
);
1206 gen8_compute_pipeline_create(VkDevice _device
,
1207 const VkComputePipelineCreateInfo
*pCreateInfo
,
1208 VkPipeline
*pPipeline
);
1211 const VkFormat vk_format
;
1213 enum isl_format surface_format
; /**< RENDER_SURFACE_STATE.SurfaceFormat */
1214 const struct isl_format_layout
*isl_layout
;
1215 uint8_t num_channels
;
1216 uint16_t depth_format
; /**< 3DSTATE_DEPTH_BUFFER.SurfaceFormat */
1221 * Stencil formats are often a special case. To reduce the number of lookups
1222 * into the VkFormat-to-anv_format translation table when working with
1223 * stencil, here is the handle to the table's entry for VK_FORMAT_S8_UINT.
1225 extern const struct anv_format
*const anv_format_s8_uint
;
1227 const struct anv_format
*
1228 anv_format_for_vk_format(VkFormat format
);
1231 anv_format_is_color(const struct anv_format
*format
)
1233 return !format
->depth_format
&& !format
->has_stencil
;
1237 anv_format_is_depth_or_stencil(const struct anv_format
*format
)
1239 return format
->depth_format
|| format
->has_stencil
;
1242 struct anv_image_view_info
{
1243 uint8_t surface_type
; /**< RENDER_SURFACE_STATE.SurfaceType */
1244 bool is_array
:1; /**< RENDER_SURFACE_STATE.SurfaceArray */
1245 bool is_cube
:1; /**< RENDER_SURFACE_STATE.CubeFaceEnable* */
1248 struct anv_image_view_info
1249 anv_image_view_info_for_vk_image_view_type(VkImageViewType type
);
1252 * A proxy for the color surfaces, depth surfaces, and stencil surfaces.
1254 struct anv_surface
{
1256 * Offset from VkImage's base address, as bound by vkBindImageMemory().
1260 uint32_t stride
; /**< RENDER_SURFACE_STATE.SurfacePitch */
1261 uint16_t qpitch
; /**< RENDER_SURFACE_STATE.QPitch */
1264 * \name Alignment of miptree images, in units of pixels.
1266 * These fields contain the real alignment values, not the values to be
1267 * given to the GPU. For example, if h_align is 4, then program the GPU
1271 uint8_t h_align
; /**< RENDER_SURFACE_STATE.SurfaceHorizontalAlignment */
1272 uint8_t v_align
; /**< RENDER_SURFACE_STATE.SurfaceVerticalAlignment */
1275 enum isl_tiling tiling
;
1280 const struct anv_format
*format
;
1283 uint32_t array_size
;
1284 VkImageUsageFlags usage
; /**< Superset of VkImageCreateInfo::usage. */
1289 /* Set when bound */
1291 VkDeviceSize offset
;
1293 uint8_t surface_type
; /**< RENDER_SURFACE_STATE.SurfaceType */
1295 bool needs_nonrt_surface_state
:1;
1296 bool needs_color_rt_surface_state
:1;
1301 * For each foo, anv_image::foo_surface is valid if and only if
1302 * anv_image::format has a foo aspect.
1304 * The hardware requires that the depth buffer and stencil buffer be
1305 * separate surfaces. From Vulkan's perspective, though, depth and stencil
1306 * reside in the same VkImage. To satisfy both the hardware and Vulkan, we
1307 * allocate the depth and stencil buffers as separate surfaces in the same
1311 struct anv_surface color_surface
;
1314 struct anv_surface depth_surface
;
1315 struct anv_surface stencil_surface
;
1320 struct anv_image_view
{
1321 const struct anv_image
*image
; /**< VkImageViewCreateInfo::image */
1322 const struct anv_format
*format
; /**< VkImageViewCreateInfo::format */
1324 uint32_t offset
; /**< Offset into bo. */
1325 VkExtent3D extent
; /**< Extent of VkImageViewCreateInfo::baseMipLevel. */
1327 /** RENDER_SURFACE_STATE when using image as a color render target. */
1328 struct anv_state color_rt_surface_state
;
1330 /** RENDER_SURFACE_STATE when using image as a non render target. */
1331 struct anv_state nonrt_surface_state
;
1334 struct anv_image_create_info
{
1335 const VkImageCreateInfo
*vk_info
;
1337 enum isl_tiling tiling
;
1341 VkResult
anv_image_create(VkDevice _device
,
1342 const struct anv_image_create_info
*info
,
1345 struct anv_surface
*
1346 anv_image_get_surface_for_aspect_mask(struct anv_image
*image
,
1347 VkImageAspectFlags aspect_mask
);
1349 void anv_image_view_init(struct anv_image_view
*view
,
1350 struct anv_device
*device
,
1351 const VkImageViewCreateInfo
* pCreateInfo
,
1352 struct anv_cmd_buffer
*cmd_buffer
);
1355 gen7_image_view_init(struct anv_image_view
*iview
,
1356 struct anv_device
*device
,
1357 const VkImageViewCreateInfo
* pCreateInfo
,
1358 struct anv_cmd_buffer
*cmd_buffer
);
1361 gen8_image_view_init(struct anv_image_view
*iview
,
1362 struct anv_device
*device
,
1363 const VkImageViewCreateInfo
* pCreateInfo
,
1364 struct anv_cmd_buffer
*cmd_buffer
);
1366 void anv_fill_buffer_surface_state(struct anv_device
*device
, void *state
,
1367 const struct anv_format
*format
,
1368 uint32_t offset
, uint32_t range
,
1371 void gen7_fill_buffer_surface_state(void *state
, const struct anv_format
*format
,
1372 uint32_t offset
, uint32_t range
,
1374 void gen8_fill_buffer_surface_state(void *state
, const struct anv_format
*format
,
1375 uint32_t offset
, uint32_t range
,
1378 struct anv_sampler
{
1382 struct anv_framebuffer
{
1387 uint32_t attachment_count
;
1388 const struct anv_image_view
* attachments
[0];
1391 struct anv_subpass
{
1392 uint32_t input_count
;
1393 uint32_t * input_attachments
;
1394 uint32_t color_count
;
1395 uint32_t * color_attachments
;
1396 uint32_t * resolve_attachments
;
1397 uint32_t depth_stencil_attachment
;
1400 struct anv_render_pass_attachment
{
1401 const struct anv_format
*format
;
1403 VkAttachmentLoadOp load_op
;
1404 VkAttachmentLoadOp stencil_load_op
;
1407 struct anv_render_pass
{
1408 uint32_t attachment_count
;
1409 uint32_t subpass_count
;
1410 struct anv_render_pass_attachment
* attachments
;
1411 struct anv_subpass subpasses
[0];
1414 extern struct anv_render_pass anv_meta_dummy_renderpass
;
1416 struct anv_query_pool_slot
{
1422 struct anv_query_pool
{
1428 void anv_device_init_meta(struct anv_device
*device
);
1429 void anv_device_finish_meta(struct anv_device
*device
);
1431 void *anv_lookup_entrypoint(const char *name
);
1433 void anv_dump_image_to_ppm(struct anv_device
*device
,
1434 struct anv_image
*image
, unsigned miplevel
,
1435 unsigned array_layer
, const char *filename
);
1437 #define ANV_DEFINE_HANDLE_CASTS(__anv_type, __VkType) \
1439 static inline struct __anv_type * \
1440 __anv_type ## _from_handle(__VkType _handle) \
1442 return (struct __anv_type *) _handle; \
1445 static inline __VkType \
1446 __anv_type ## _to_handle(struct __anv_type *_obj) \
1448 return (__VkType) _obj; \
1451 #define ANV_DEFINE_NONDISP_HANDLE_CASTS(__anv_type, __VkType) \
1453 static inline struct __anv_type * \
1454 __anv_type ## _from_handle(__VkType _handle) \
1456 return (struct __anv_type *) _handle.handle; \
1459 static inline __VkType \
1460 __anv_type ## _to_handle(struct __anv_type *_obj) \
1462 return (__VkType) { .handle = (uint64_t) _obj }; \
1465 #define ANV_FROM_HANDLE(__anv_type, __name, __handle) \
1466 struct __anv_type *__name = __anv_type ## _from_handle(__handle)
1468 ANV_DEFINE_HANDLE_CASTS(anv_cmd_buffer
, VkCmdBuffer
)
1469 ANV_DEFINE_HANDLE_CASTS(anv_device
, VkDevice
)
1470 ANV_DEFINE_HANDLE_CASTS(anv_instance
, VkInstance
)
1471 ANV_DEFINE_HANDLE_CASTS(anv_physical_device
, VkPhysicalDevice
)
1472 ANV_DEFINE_HANDLE_CASTS(anv_queue
, VkQueue
)
1474 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_cmd_pool
, VkCmdPool
)
1475 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_buffer
, VkBuffer
)
1476 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_descriptor_set
, VkDescriptorSet
)
1477 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_descriptor_set_layout
, VkDescriptorSetLayout
)
1478 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_device_memory
, VkDeviceMemory
)
1479 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_fence
, VkFence
)
1480 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_framebuffer
, VkFramebuffer
)
1481 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_image
, VkImage
)
1482 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_image_view
, VkImageView
);
1483 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_pipeline
, VkPipeline
)
1484 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_pipeline_layout
, VkPipelineLayout
)
1485 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_query_pool
, VkQueryPool
)
1486 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_render_pass
, VkRenderPass
)
1487 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_sampler
, VkSampler
)
1488 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_shader
, VkShader
)
1489 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_shader_module
, VkShaderModule
)
1491 #define ANV_DEFINE_STRUCT_CASTS(__anv_type, __VkType) \
1493 static inline const __VkType * \
1494 __anv_type ## _to_ ## __VkType(const struct __anv_type *__anv_obj) \
1496 return (const __VkType *) __anv_obj; \
1499 #define ANV_COMMON_TO_STRUCT(__VkType, __vk_name, __common_name) \
1500 const __VkType *__vk_name = anv_common_to_ ## __VkType(__common_name)
1502 ANV_DEFINE_STRUCT_CASTS(anv_common
, VkMemoryBarrier
)
1503 ANV_DEFINE_STRUCT_CASTS(anv_common
, VkBufferMemoryBarrier
)
1504 ANV_DEFINE_STRUCT_CASTS(anv_common
, VkImageMemoryBarrier
)