2 * Copyright © 2016 Red Hat.
3 * Copyright © 2016 Bas Nieuwenhuizen
5 * based in part on anv driver which is:
6 * Copyright © 2015 Intel Corporation
8 * Permission is hereby granted, free of charge, to any person obtaining a
9 * copy of this software and associated documentation files (the "Software"),
10 * to deal in the Software without restriction, including without limitation
11 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
12 * and/or sell copies of the Software, and to permit persons to whom the
13 * Software is furnished to do so, subject to the following conditions:
15 * The above copyright notice and this permission notice (including the next
16 * paragraph) shall be included in all copies or substantial portions of the
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
20 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
21 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
22 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
23 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
24 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
28 #ifndef RADV_PRIVATE_H
29 #define RADV_PRIVATE_H
47 #include "compiler/shader_enums.h"
48 #include "util/macros.h"
49 #include "util/list.h"
50 #include "util/vk_alloc.h"
51 #include "main/macros.h"
53 #include "radv_radeon_winsys.h"
54 #include "ac_binary.h"
55 #include "ac_nir_to_llvm.h"
56 #include "radv_descriptor_set.h"
58 #include <llvm-c/TargetMachine.h>
60 /* Pre-declarations needed for WSI entrypoints */
63 typedef struct xcb_connection_t xcb_connection_t
;
64 typedef uint32_t xcb_visualid_t
;
65 typedef uint32_t xcb_window_t
;
67 #include <vulkan/vulkan.h>
68 #include <vulkan/vulkan_intel.h>
69 #include <vulkan/vk_icd.h>
71 #include "radv_entrypoints.h"
73 #include "wsi_common.h"
76 #define MAX_VERTEX_ATTRIBS 32
78 #define MAX_VIEWPORTS 16
79 #define MAX_SCISSORS 16
80 #define MAX_PUSH_CONSTANTS_SIZE 128
81 #define MAX_DYNAMIC_BUFFERS 16
82 #define MAX_SAMPLES_LOG2 4
83 #define NUM_META_FS_KEYS 11
84 #define RADV_MAX_DRM_DEVICES 8
86 #define NUM_DEPTH_CLEAR_PIPELINES 3
90 RADV_MEM_HEAP_VRAM_CPU_ACCESS
,
97 RADV_MEM_TYPE_GTT_WRITE_COMBINE
,
98 RADV_MEM_TYPE_VRAM_CPU_ACCESS
,
99 RADV_MEM_TYPE_GTT_CACHED
,
105 RADV_DEBUG_NO_FAST_CLEARS
= 0x1,
106 RADV_DEBUG_NO_DCC
= 0x2,
107 RADV_DEBUG_DUMP_SHADERS
= 0x4,
108 RADV_DEBUG_NO_CACHE
= 0x8,
109 RADV_DEBUG_DUMP_SHADER_STATS
= 0x10,
110 RADV_DEBUG_NO_HIZ
= 0x20,
111 RADV_DEBUG_NO_COMPUTE_QUEUE
= 0x40,
112 RADV_DEBUG_UNSAFE_MATH
= 0x80,
115 #define radv_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
117 static inline uint32_t
118 align_u32(uint32_t v
, uint32_t a
)
120 assert(a
!= 0 && a
== (a
& -a
));
121 return (v
+ a
- 1) & ~(a
- 1);
124 static inline uint32_t
125 align_u32_npot(uint32_t v
, uint32_t a
)
127 return (v
+ a
- 1) / a
* a
;
130 static inline uint64_t
131 align_u64(uint64_t v
, uint64_t a
)
133 assert(a
!= 0 && a
== (a
& -a
));
134 return (v
+ a
- 1) & ~(a
- 1);
137 static inline int32_t
138 align_i32(int32_t v
, int32_t a
)
140 assert(a
!= 0 && a
== (a
& -a
));
141 return (v
+ a
- 1) & ~(a
- 1);
144 /** Alignment must be a power of 2. */
146 radv_is_aligned(uintmax_t n
, uintmax_t a
)
148 assert(a
== (a
& -a
));
149 return (n
& (a
- 1)) == 0;
152 static inline uint32_t
153 round_up_u32(uint32_t v
, uint32_t a
)
155 return (v
+ a
- 1) / a
;
158 static inline uint64_t
159 round_up_u64(uint64_t v
, uint64_t a
)
161 return (v
+ a
- 1) / a
;
164 static inline uint32_t
165 radv_minify(uint32_t n
, uint32_t levels
)
167 if (unlikely(n
== 0))
170 return MAX2(n
>> levels
, 1);
173 radv_clamp_f(float f
, float min
, float max
)
186 radv_clear_mask(uint32_t *inout_mask
, uint32_t clear_mask
)
188 if (*inout_mask
& clear_mask
) {
189 *inout_mask
&= ~clear_mask
;
196 #define for_each_bit(b, dword) \
197 for (uint32_t __dword = (dword); \
198 (b) = __builtin_ffs(__dword) - 1, __dword; \
199 __dword &= ~(1 << (b)))
201 #define typed_memcpy(dest, src, count) ({ \
202 STATIC_ASSERT(sizeof(*src) == sizeof(*dest)); \
203 memcpy((dest), (src), (count) * sizeof(*(src))); \
206 #define zero(x) (memset(&(x), 0, sizeof(x)))
208 /* Whenever we generate an error, pass it through this function. Useful for
209 * debugging, where we can break on it. Only call at error site, not when
210 * propagating errors. Might be useful to plug in a stack trace here.
213 VkResult
__vk_errorf(VkResult error
, const char *file
, int line
, const char *format
, ...);
216 #define vk_error(error) __vk_errorf(error, __FILE__, __LINE__, NULL);
217 #define vk_errorf(error, format, ...) __vk_errorf(error, __FILE__, __LINE__, format, ## __VA_ARGS__);
219 #define vk_error(error) error
220 #define vk_errorf(error, format, ...) error
223 void __radv_finishme(const char *file
, int line
, const char *format
, ...)
224 radv_printflike(3, 4);
225 void radv_loge(const char *format
, ...) radv_printflike(1, 2);
226 void radv_loge_v(const char *format
, va_list va
);
229 * Print a FINISHME message, including its source location.
231 #define radv_finishme(format, ...) \
233 static bool reported = false; \
235 __radv_finishme(__FILE__, __LINE__, format, ##__VA_ARGS__); \
240 /* A non-fatal assert. Useful for debugging. */
242 #define radv_assert(x) ({ \
243 if (unlikely(!(x))) \
244 fprintf(stderr, "%s:%d ASSERT: %s\n", __FILE__, __LINE__, #x); \
247 #define radv_assert(x)
250 #define stub_return(v) \
252 radv_finishme("stub %s", __func__); \
258 radv_finishme("stub %s", __func__); \
262 void *radv_lookup_entrypoint(const char *name
);
264 struct radv_extensions
{
265 VkExtensionProperties
*ext_array
;
269 struct radv_physical_device
{
270 VK_LOADER_DATA _loader_data
;
272 struct radv_instance
* instance
;
274 struct radeon_winsys
*ws
;
275 struct radeon_info rad_info
;
278 uint8_t uuid
[VK_UUID_SIZE
];
281 struct wsi_device wsi_device
;
282 struct radv_extensions extensions
;
285 struct radv_instance
{
286 VK_LOADER_DATA _loader_data
;
288 VkAllocationCallbacks alloc
;
291 int physicalDeviceCount
;
292 struct radv_physical_device physicalDevices
[RADV_MAX_DRM_DEVICES
];
294 uint64_t debug_flags
;
297 VkResult
radv_init_wsi(struct radv_physical_device
*physical_device
);
298 void radv_finish_wsi(struct radv_physical_device
*physical_device
);
302 struct radv_pipeline_cache
{
303 struct radv_device
* device
;
304 pthread_mutex_t mutex
;
308 uint32_t kernel_count
;
309 struct cache_entry
** hash_table
;
312 VkAllocationCallbacks alloc
;
316 radv_pipeline_cache_init(struct radv_pipeline_cache
*cache
,
317 struct radv_device
*device
);
319 radv_pipeline_cache_finish(struct radv_pipeline_cache
*cache
);
321 radv_pipeline_cache_load(struct radv_pipeline_cache
*cache
,
322 const void *data
, size_t size
);
324 struct radv_shader_variant
*
325 radv_create_shader_variant_from_pipeline_cache(struct radv_device
*device
,
326 struct radv_pipeline_cache
*cache
,
327 const unsigned char *sha1
);
329 struct radv_shader_variant
*
330 radv_pipeline_cache_insert_shader(struct radv_pipeline_cache
*cache
,
331 const unsigned char *sha1
,
332 struct radv_shader_variant
*variant
,
333 const void *code
, unsigned code_size
);
335 void radv_shader_variant_destroy(struct radv_device
*device
,
336 struct radv_shader_variant
*variant
);
338 struct radv_meta_state
{
339 VkAllocationCallbacks alloc
;
341 struct radv_pipeline_cache cache
;
344 * Use array element `i` for images with `2^i` samples.
347 VkRenderPass render_pass
[NUM_META_FS_KEYS
];
348 struct radv_pipeline
*color_pipelines
[NUM_META_FS_KEYS
];
350 VkRenderPass depthstencil_rp
;
351 struct radv_pipeline
*depth_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
352 struct radv_pipeline
*stencil_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
353 struct radv_pipeline
*depthstencil_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
354 } clear
[1 + MAX_SAMPLES_LOG2
];
357 VkRenderPass render_pass
[NUM_META_FS_KEYS
];
359 /** Pipeline that blits from a 1D image. */
360 VkPipeline pipeline_1d_src
[NUM_META_FS_KEYS
];
362 /** Pipeline that blits from a 2D image. */
363 VkPipeline pipeline_2d_src
[NUM_META_FS_KEYS
];
365 /** Pipeline that blits from a 3D image. */
366 VkPipeline pipeline_3d_src
[NUM_META_FS_KEYS
];
368 VkRenderPass depth_only_rp
;
369 VkPipeline depth_only_1d_pipeline
;
370 VkPipeline depth_only_2d_pipeline
;
371 VkPipeline depth_only_3d_pipeline
;
373 VkRenderPass stencil_only_rp
;
374 VkPipeline stencil_only_1d_pipeline
;
375 VkPipeline stencil_only_2d_pipeline
;
376 VkPipeline stencil_only_3d_pipeline
;
377 VkPipelineLayout pipeline_layout
;
378 VkDescriptorSetLayout ds_layout
;
382 VkRenderPass render_passes
[NUM_META_FS_KEYS
];
384 VkPipelineLayout p_layouts
[2];
385 VkDescriptorSetLayout ds_layouts
[2];
386 VkPipeline pipelines
[2][NUM_META_FS_KEYS
];
388 VkRenderPass depth_only_rp
;
389 VkPipeline depth_only_pipeline
[2];
391 VkRenderPass stencil_only_rp
;
392 VkPipeline stencil_only_pipeline
[2];
396 VkPipelineLayout img_p_layout
;
397 VkDescriptorSetLayout img_ds_layout
;
401 VkRenderPass render_pass
;
402 VkPipelineLayout img_p_layout
;
403 VkDescriptorSetLayout img_ds_layout
;
407 VkPipelineLayout img_p_layout
;
408 VkDescriptorSetLayout img_ds_layout
;
412 VkPipelineLayout img_p_layout
;
413 VkDescriptorSetLayout img_ds_layout
;
423 VkDescriptorSetLayout ds_layout
;
424 VkPipelineLayout p_layout
;
427 VkPipeline i_pipeline
;
428 } rc
[MAX_SAMPLES_LOG2
];
432 VkPipeline decompress_pipeline
;
433 VkPipeline resummarize_pipeline
;
438 VkPipeline cmask_eliminate_pipeline
;
439 VkPipeline fmask_decompress_pipeline
;
444 VkPipelineLayout fill_p_layout
;
445 VkPipelineLayout copy_p_layout
;
446 VkDescriptorSetLayout fill_ds_layout
;
447 VkDescriptorSetLayout copy_ds_layout
;
448 VkPipeline fill_pipeline
;
449 VkPipeline copy_pipeline
;
454 #define RADV_QUEUE_GENERAL 0
455 #define RADV_QUEUE_COMPUTE 1
456 #define RADV_QUEUE_TRANSFER 2
458 #define RADV_MAX_QUEUE_FAMILIES 3
460 enum ring_type
radv_queue_family_to_ring(int f
);
463 VK_LOADER_DATA _loader_data
;
464 struct radv_device
* device
;
465 struct radeon_winsys_ctx
*hw_ctx
;
466 int queue_family_index
;
469 uint32_t scratch_size
;
470 uint32_t compute_scratch_size
;
471 uint32_t esgs_ring_size
;
472 uint32_t gsvs_ring_size
;
474 struct radeon_winsys_bo
*scratch_bo
;
475 struct radeon_winsys_bo
*descriptor_bo
;
476 struct radeon_winsys_bo
*compute_scratch_bo
;
477 struct radeon_winsys_bo
*esgs_ring_bo
;
478 struct radeon_winsys_bo
*gsvs_ring_bo
;
479 struct radeon_winsys_cs
*initial_preamble_cs
;
480 struct radeon_winsys_cs
*continue_preamble_cs
;
484 VK_LOADER_DATA _loader_data
;
486 VkAllocationCallbacks alloc
;
488 struct radv_instance
* instance
;
489 struct radeon_winsys
*ws
;
491 struct radv_meta_state meta_state
;
493 struct radv_queue
*queues
[RADV_MAX_QUEUE_FAMILIES
];
494 int queue_count
[RADV_MAX_QUEUE_FAMILIES
];
495 struct radeon_winsys_cs
*empty_cs
[RADV_MAX_QUEUE_FAMILIES
];
496 struct radeon_winsys_cs
*flush_cs
[RADV_MAX_QUEUE_FAMILIES
];
498 uint64_t debug_flags
;
500 bool llvm_supports_spill
;
501 uint32_t scratch_waves
;
503 uint32_t gs_table_depth
;
505 /* MSAA sample locations.
506 * The first index is the sample index.
507 * The second index is the coordinate: X, Y. */
508 float sample_locations_1x
[1][2];
509 float sample_locations_2x
[2][2];
510 float sample_locations_4x
[4][2];
511 float sample_locations_8x
[8][2];
512 float sample_locations_16x
[16][2];
515 uint32_t gfx_init_size_dw
;
516 struct radeon_winsys_bo
*gfx_init
;
518 struct radeon_winsys_bo
*trace_bo
;
519 uint32_t *trace_id_ptr
;
521 struct radv_physical_device
*physical_device
;
524 struct radv_device_memory
{
525 struct radeon_winsys_bo
*bo
;
526 /* for dedicated allocations */
527 struct radv_image
*image
;
528 struct radv_buffer
*buffer
;
530 VkDeviceSize map_size
;
535 struct radv_descriptor_range
{
540 struct radv_descriptor_set
{
541 const struct radv_descriptor_set_layout
*layout
;
544 struct radv_buffer_view
*buffer_views
;
545 struct radeon_winsys_bo
*bo
;
547 uint32_t *mapped_ptr
;
548 struct radv_descriptor_range
*dynamic_descriptors
;
550 struct list_head vram_list
;
552 struct radeon_winsys_bo
*descriptors
[0];
555 struct radv_descriptor_pool
{
556 struct radeon_winsys_bo
*bo
;
558 uint64_t current_offset
;
561 struct list_head vram_list
;
565 struct radv_device
* device
;
568 VkBufferUsageFlags usage
;
571 struct radeon_winsys_bo
* bo
;
576 enum radv_cmd_dirty_bits
{
577 RADV_CMD_DIRTY_DYNAMIC_VIEWPORT
= 1 << 0, /* VK_DYNAMIC_STATE_VIEWPORT */
578 RADV_CMD_DIRTY_DYNAMIC_SCISSOR
= 1 << 1, /* VK_DYNAMIC_STATE_SCISSOR */
579 RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH
= 1 << 2, /* VK_DYNAMIC_STATE_LINE_WIDTH */
580 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS
= 1 << 3, /* VK_DYNAMIC_STATE_DEPTH_BIAS */
581 RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS
= 1 << 4, /* VK_DYNAMIC_STATE_BLEND_CONSTANTS */
582 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS
= 1 << 5, /* VK_DYNAMIC_STATE_DEPTH_BOUNDS */
583 RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6, /* VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK */
584 RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7, /* VK_DYNAMIC_STATE_STENCIL_WRITE_MASK */
585 RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE
= 1 << 8, /* VK_DYNAMIC_STATE_STENCIL_REFERENCE */
586 RADV_CMD_DIRTY_DYNAMIC_ALL
= (1 << 9) - 1,
587 RADV_CMD_DIRTY_PIPELINE
= 1 << 9,
588 RADV_CMD_DIRTY_INDEX_BUFFER
= 1 << 10,
589 RADV_CMD_DIRTY_RENDER_TARGETS
= 1 << 11,
591 typedef uint32_t radv_cmd_dirty_mask_t
;
593 enum radv_cmd_flush_bits
{
594 RADV_CMD_FLAG_INV_ICACHE
= 1 << 0,
595 /* SMEM L1, other names: KCACHE, constant cache, DCACHE, data cache */
596 RADV_CMD_FLAG_INV_SMEM_L1
= 1 << 1,
597 /* VMEM L1 can optionally be bypassed (GLC=1). Other names: TC L1 */
598 RADV_CMD_FLAG_INV_VMEM_L1
= 1 << 2,
599 /* Used by everything except CB/DB, can be bypassed (SLC=1). Other names: TC L2 */
600 RADV_CMD_FLAG_INV_GLOBAL_L2
= 1 << 3,
601 /* Framebuffer caches */
602 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
= 1 << 4,
603 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
= 1 << 5,
604 RADV_CMD_FLAG_FLUSH_AND_INV_DB
= 1 << 6,
605 RADV_CMD_FLAG_FLUSH_AND_INV_CB
= 1 << 7,
606 /* Engine synchronization. */
607 RADV_CMD_FLAG_VS_PARTIAL_FLUSH
= 1 << 8,
608 RADV_CMD_FLAG_PS_PARTIAL_FLUSH
= 1 << 9,
609 RADV_CMD_FLAG_CS_PARTIAL_FLUSH
= 1 << 10,
610 RADV_CMD_FLAG_VGT_FLUSH
= 1 << 11,
612 RADV_CMD_FLUSH_AND_INV_FRAMEBUFFER
= (RADV_CMD_FLAG_FLUSH_AND_INV_CB
|
613 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
|
614 RADV_CMD_FLAG_FLUSH_AND_INV_DB
|
615 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
)
618 struct radv_vertex_binding
{
619 struct radv_buffer
* buffer
;
623 struct radv_dynamic_state
{
626 VkViewport viewports
[MAX_VIEWPORTS
];
631 VkRect2D scissors
[MAX_SCISSORS
];
642 float blend_constants
[4];
652 } stencil_compare_mask
;
657 } stencil_write_mask
;
665 extern const struct radv_dynamic_state default_dynamic_state
;
667 void radv_dynamic_state_copy(struct radv_dynamic_state
*dest
,
668 const struct radv_dynamic_state
*src
,
671 * Attachment state when recording a renderpass instance.
673 * The clear value is valid only if there exists a pending clear.
675 struct radv_attachment_state
{
676 VkImageAspectFlags pending_clear_aspects
;
677 VkClearValue clear_value
;
678 VkImageLayout current_layout
;
681 struct radv_cmd_state
{
683 radv_cmd_dirty_mask_t dirty
;
684 bool vertex_descriptors_dirty
;
686 struct radv_pipeline
* pipeline
;
687 struct radv_pipeline
* emitted_pipeline
;
688 struct radv_pipeline
* compute_pipeline
;
689 struct radv_pipeline
* emitted_compute_pipeline
;
690 struct radv_framebuffer
* framebuffer
;
691 struct radv_render_pass
* pass
;
692 const struct radv_subpass
* subpass
;
693 struct radv_dynamic_state dynamic
;
694 struct radv_vertex_binding vertex_bindings
[MAX_VBS
];
695 struct radv_descriptor_set
* descriptors
[MAX_SETS
];
696 struct radv_attachment_state
* attachments
;
697 VkRect2D render_area
;
698 struct radv_buffer
* index_buffer
;
700 uint32_t index_offset
;
701 uint32_t last_primitive_reset_index
;
702 enum radv_cmd_flush_bits flush_bits
;
703 unsigned active_occlusion_queries
;
705 uint32_t descriptors_dirty
;
707 uint32_t last_ia_multi_vgt_param
;
710 struct radv_cmd_pool
{
711 VkAllocationCallbacks alloc
;
712 struct list_head cmd_buffers
;
713 uint32_t queue_family_index
;
716 struct radv_cmd_buffer_upload
{
720 struct radeon_winsys_bo
*upload_bo
;
721 struct list_head list
;
724 struct radv_cmd_buffer
{
725 VK_LOADER_DATA _loader_data
;
727 struct radv_device
* device
;
729 struct radv_cmd_pool
* pool
;
730 struct list_head pool_link
;
732 VkCommandBufferUsageFlags usage_flags
;
733 VkCommandBufferLevel level
;
734 struct radeon_winsys_cs
*cs
;
735 struct radv_cmd_state state
;
736 uint32_t queue_family_index
;
738 uint8_t push_constants
[MAX_PUSH_CONSTANTS_SIZE
];
739 uint32_t dynamic_buffers
[16 * MAX_DYNAMIC_BUFFERS
];
740 VkShaderStageFlags push_constant_stages
;
742 struct radv_cmd_buffer_upload upload
;
746 uint32_t scratch_size_needed
;
747 uint32_t compute_scratch_size_needed
;
748 uint32_t esgs_ring_size_needed
;
749 uint32_t gsvs_ring_size_needed
;
751 int ring_offsets_idx
; /* just used for verification */
756 bool radv_cmd_buffer_uses_mec(struct radv_cmd_buffer
*cmd_buffer
);
758 void si_init_compute(struct radv_cmd_buffer
*cmd_buffer
);
759 void si_init_config(struct radv_cmd_buffer
*cmd_buffer
);
761 void cik_create_gfx_config(struct radv_device
*device
);
763 void si_write_viewport(struct radeon_winsys_cs
*cs
, int first_vp
,
764 int count
, const VkViewport
*viewports
);
765 void si_write_scissors(struct radeon_winsys_cs
*cs
, int first
,
766 int count
, const VkRect2D
*scissors
);
767 uint32_t si_get_ia_multi_vgt_param(struct radv_cmd_buffer
*cmd_buffer
,
768 bool instanced_or_indirect_draw
, uint32_t draw_vertex_count
);
769 void si_cs_emit_cache_flush(struct radeon_winsys_cs
*cs
,
770 enum chip_class chip_class
,
772 enum radv_cmd_flush_bits flush_bits
);
773 void si_cs_emit_cache_flush(struct radeon_winsys_cs
*cs
,
774 enum chip_class chip_class
,
776 enum radv_cmd_flush_bits flush_bits
);
777 void si_emit_cache_flush(struct radv_cmd_buffer
*cmd_buffer
);
778 void si_cp_dma_buffer_copy(struct radv_cmd_buffer
*cmd_buffer
,
779 uint64_t src_va
, uint64_t dest_va
,
781 void si_cp_dma_clear_buffer(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
782 uint64_t size
, unsigned value
);
783 void radv_set_db_count_control(struct radv_cmd_buffer
*cmd_buffer
);
784 void radv_bind_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
785 struct radv_descriptor_set
*set
,
788 radv_cmd_buffer_upload_alloc(struct radv_cmd_buffer
*cmd_buffer
,
791 unsigned *out_offset
,
794 radv_cmd_buffer_set_subpass(struct radv_cmd_buffer
*cmd_buffer
,
795 const struct radv_subpass
*subpass
,
798 radv_cmd_buffer_upload_data(struct radv_cmd_buffer
*cmd_buffer
,
799 unsigned size
, unsigned alignmnet
,
800 const void *data
, unsigned *out_offset
);
802 radv_emit_framebuffer_state(struct radv_cmd_buffer
*cmd_buffer
);
803 void radv_cmd_buffer_clear_subpass(struct radv_cmd_buffer
*cmd_buffer
);
804 void radv_cmd_buffer_resolve_subpass(struct radv_cmd_buffer
*cmd_buffer
);
805 void radv_cayman_emit_msaa_sample_locs(struct radeon_winsys_cs
*cs
, int nr_samples
);
806 unsigned radv_cayman_get_maxdist(int log_samples
);
807 void radv_device_init_msaa(struct radv_device
*device
);
808 void radv_set_depth_clear_regs(struct radv_cmd_buffer
*cmd_buffer
,
809 struct radv_image
*image
,
810 VkClearDepthStencilValue ds_clear_value
,
811 VkImageAspectFlags aspects
);
812 void radv_set_color_clear_regs(struct radv_cmd_buffer
*cmd_buffer
,
813 struct radv_image
*image
,
815 uint32_t color_values
[2]);
816 void radv_fill_buffer(struct radv_cmd_buffer
*cmd_buffer
,
817 struct radeon_winsys_bo
*bo
,
818 uint64_t offset
, uint64_t size
, uint32_t value
);
819 void radv_cmd_buffer_trace_emit(struct radv_cmd_buffer
*cmd_buffer
);
820 bool radv_get_memory_fd(struct radv_device
*device
,
821 struct radv_device_memory
*memory
,
824 * Takes x,y,z as exact numbers of invocations, instead of blocks.
826 * Limitations: Can't call normal dispatch functions without binding or rebinding
827 * the compute pipeline.
829 void radv_unaligned_dispatch(
830 struct radv_cmd_buffer
*cmd_buffer
,
836 struct radeon_winsys_bo
*bo
;
842 struct radv_shader_module
{
843 struct nir_shader
* nir
;
844 unsigned char sha1
[20];
849 union ac_shader_variant_key
;
852 radv_hash_shader(unsigned char *hash
, struct radv_shader_module
*module
,
853 const char *entrypoint
,
854 const VkSpecializationInfo
*spec_info
,
855 const struct radv_pipeline_layout
*layout
,
856 const union ac_shader_variant_key
*key
,
857 uint32_t is_geom_copy_shader
);
859 static inline gl_shader_stage
860 vk_to_mesa_shader_stage(VkShaderStageFlagBits vk_stage
)
862 assert(__builtin_popcount(vk_stage
) == 1);
863 return ffs(vk_stage
) - 1;
866 static inline VkShaderStageFlagBits
867 mesa_to_vk_shader_stage(gl_shader_stage mesa_stage
)
869 return (1 << mesa_stage
);
872 #define RADV_STAGE_MASK ((1 << MESA_SHADER_STAGES) - 1)
874 #define radv_foreach_stage(stage, stage_bits) \
875 for (gl_shader_stage stage, \
876 __tmp = (gl_shader_stage)((stage_bits) & RADV_STAGE_MASK); \
877 stage = __builtin_ffs(__tmp) - 1, __tmp; \
878 __tmp &= ~(1 << (stage)))
880 struct radv_shader_variant
{
883 struct radeon_winsys_bo
*bo
;
884 struct ac_shader_config config
;
885 struct ac_shader_variant_info info
;
891 struct radv_depth_stencil_state
{
892 uint32_t db_depth_control
;
893 uint32_t db_stencil_control
;
894 uint32_t db_render_control
;
895 uint32_t db_render_override2
;
898 struct radv_blend_state
{
899 uint32_t cb_color_control
;
900 uint32_t cb_target_mask
;
901 uint32_t sx_mrt0_blend_opt
[8];
902 uint32_t cb_blend_control
[8];
904 uint32_t spi_shader_col_format
;
905 uint32_t cb_shader_mask
;
906 uint32_t db_alpha_to_mask
;
909 unsigned radv_format_meta_fs_key(VkFormat format
);
911 struct radv_raster_state
{
912 uint32_t pa_cl_clip_cntl
;
913 uint32_t pa_cl_vs_out_cntl
;
914 uint32_t spi_interp_control
;
915 uint32_t pa_su_point_size
;
916 uint32_t pa_su_point_minmax
;
917 uint32_t pa_su_line_cntl
;
918 uint32_t pa_su_vtx_cntl
;
919 uint32_t pa_su_sc_mode_cntl
;
922 struct radv_multisample_state
{
924 uint32_t pa_sc_line_cntl
;
925 uint32_t pa_sc_mode_cntl_0
;
926 uint32_t pa_sc_mode_cntl_1
;
927 uint32_t pa_sc_aa_config
;
928 uint32_t pa_sc_aa_mask
[2];
929 unsigned num_samples
;
932 struct radv_prim_vertex_count
{
937 struct radv_pipeline
{
938 struct radv_device
* device
;
939 uint32_t dynamic_state_mask
;
940 struct radv_dynamic_state dynamic_state
;
942 struct radv_pipeline_layout
* layout
;
944 bool needs_data_cache
;
946 struct radv_shader_variant
* shaders
[MESA_SHADER_STAGES
];
947 struct radv_shader_variant
*gs_copy_shader
;
948 VkShaderStageFlags active_stages
;
950 uint32_t va_rsrc_word3
[MAX_VERTEX_ATTRIBS
];
951 uint32_t va_format_size
[MAX_VERTEX_ATTRIBS
];
952 uint32_t va_binding
[MAX_VERTEX_ATTRIBS
];
953 uint32_t va_offset
[MAX_VERTEX_ATTRIBS
];
954 uint32_t num_vertex_attribs
;
955 uint32_t binding_stride
[MAX_VBS
];
959 struct radv_blend_state blend
;
960 struct radv_depth_stencil_state ds
;
961 struct radv_raster_state raster
;
962 struct radv_multisample_state ms
;
965 bool prim_restart_enable
;
966 unsigned esgs_ring_size
;
967 unsigned gsvs_ring_size
;
968 struct radv_prim_vertex_count prim_vertex_count
;
973 unsigned scratch_bytes_per_wave
;
976 static inline bool radv_pipeline_has_gs(struct radv_pipeline
*pipeline
)
978 return pipeline
->shaders
[MESA_SHADER_GEOMETRY
] ? true : false;
981 struct radv_graphics_pipeline_create_info
{
984 bool db_stencil_clear
;
985 bool db_depth_disable_expclear
;
986 bool db_stencil_disable_expclear
;
987 bool db_flush_depth_inplace
;
988 bool db_flush_stencil_inplace
;
990 uint32_t custom_blend_mode
;
994 radv_pipeline_init(struct radv_pipeline
*pipeline
, struct radv_device
*device
,
995 struct radv_pipeline_cache
*cache
,
996 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
997 const struct radv_graphics_pipeline_create_info
*extra
,
998 const VkAllocationCallbacks
*alloc
);
1001 radv_graphics_pipeline_create(VkDevice device
,
1002 VkPipelineCache cache
,
1003 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
1004 const struct radv_graphics_pipeline_create_info
*extra
,
1005 const VkAllocationCallbacks
*alloc
,
1006 VkPipeline
*pPipeline
);
1008 struct vk_format_description
;
1009 uint32_t radv_translate_buffer_dataformat(const struct vk_format_description
*desc
,
1010 int first_non_void
);
1011 uint32_t radv_translate_buffer_numformat(const struct vk_format_description
*desc
,
1012 int first_non_void
);
1013 uint32_t radv_translate_colorformat(VkFormat format
);
1014 uint32_t radv_translate_color_numformat(VkFormat format
,
1015 const struct vk_format_description
*desc
,
1016 int first_non_void
);
1017 uint32_t radv_colorformat_endian_swap(uint32_t colorformat
);
1018 unsigned radv_translate_colorswap(VkFormat format
, bool do_endian_swap
);
1019 uint32_t radv_translate_dbformat(VkFormat format
);
1020 uint32_t radv_translate_tex_dataformat(VkFormat format
,
1021 const struct vk_format_description
*desc
,
1022 int first_non_void
);
1023 uint32_t radv_translate_tex_numformat(VkFormat format
,
1024 const struct vk_format_description
*desc
,
1025 int first_non_void
);
1026 bool radv_format_pack_clear_color(VkFormat format
,
1027 uint32_t clear_vals
[2],
1028 VkClearColorValue
*value
);
1029 bool radv_is_colorbuffer_format_supported(VkFormat format
, bool *blendable
);
1031 struct radv_fmask_info
{
1035 unsigned pitch_in_pixels
;
1036 unsigned bank_height
;
1037 unsigned slice_tile_max
;
1038 unsigned tile_mode_index
;
1041 struct radv_cmask_info
{
1045 unsigned slice_tile_max
;
1046 unsigned base_address_reg
;
1049 struct r600_htile_info
{
1060 /* The original VkFormat provided by the client. This may not match any
1061 * of the actual surface formats.
1064 VkImageAspectFlags aspects
;
1067 uint32_t array_size
;
1068 uint32_t samples
; /**< VkImageCreateInfo::samples */
1069 VkImageUsageFlags usage
; /**< Superset of VkImageCreateInfo::usage. */
1070 VkImageTiling tiling
; /** VkImageCreateInfo::tiling */
1076 unsigned queue_family_mask
;
1078 /* Set when bound */
1079 struct radeon_winsys_bo
*bo
;
1080 VkDeviceSize offset
;
1081 uint32_t dcc_offset
;
1082 struct radeon_surf surface
;
1084 struct radv_fmask_info fmask
;
1085 struct radv_cmask_info cmask
;
1086 uint32_t clear_value_offset
;
1088 /* Depth buffer compression and fast clear. */
1089 struct r600_htile_info htile
;
1092 bool radv_layout_has_htile(const struct radv_image
*image
,
1093 VkImageLayout layout
);
1094 bool radv_layout_is_htile_compressed(const struct radv_image
*image
,
1095 VkImageLayout layout
);
1096 bool radv_layout_can_expclear(const struct radv_image
*image
,
1097 VkImageLayout layout
);
1098 bool radv_layout_can_fast_clear(const struct radv_image
*image
,
1099 VkImageLayout layout
,
1100 unsigned queue_mask
);
1103 unsigned radv_image_queue_family_mask(const struct radv_image
*image
, uint32_t family
, uint32_t queue_family
);
1105 static inline uint32_t
1106 radv_get_layerCount(const struct radv_image
*image
,
1107 const VkImageSubresourceRange
*range
)
1109 return range
->layerCount
== VK_REMAINING_ARRAY_LAYERS
?
1110 image
->array_size
- range
->baseArrayLayer
: range
->layerCount
;
1113 static inline uint32_t
1114 radv_get_levelCount(const struct radv_image
*image
,
1115 const VkImageSubresourceRange
*range
)
1117 return range
->levelCount
== VK_REMAINING_MIP_LEVELS
?
1118 image
->levels
- range
->baseMipLevel
: range
->levelCount
;
1121 struct radeon_bo_metadata
;
1123 radv_init_metadata(struct radv_device
*device
,
1124 struct radv_image
*image
,
1125 struct radeon_bo_metadata
*metadata
);
1127 struct radv_image_view
{
1128 struct radv_image
*image
; /**< VkImageViewCreateInfo::image */
1129 struct radeon_winsys_bo
*bo
;
1131 VkImageViewType type
;
1132 VkImageAspectFlags aspect_mask
;
1134 uint32_t base_layer
;
1135 uint32_t layer_count
;
1137 VkExtent3D extent
; /**< Extent of VkImageViewCreateInfo::baseMipLevel. */
1139 uint32_t descriptor
[8];
1140 uint32_t fmask_descriptor
[8];
1143 struct radv_image_create_info
{
1144 const VkImageCreateInfo
*vk_info
;
1149 VkResult
radv_image_create(VkDevice _device
,
1150 const struct radv_image_create_info
*info
,
1151 const VkAllocationCallbacks
* alloc
,
1154 void radv_image_view_init(struct radv_image_view
*view
,
1155 struct radv_device
*device
,
1156 const VkImageViewCreateInfo
* pCreateInfo
,
1157 struct radv_cmd_buffer
*cmd_buffer
,
1158 VkImageUsageFlags usage_mask
);
1159 void radv_image_set_optimal_micro_tile_mode(struct radv_device
*device
,
1160 struct radv_image
*image
, uint32_t micro_tile_mode
);
1161 struct radv_buffer_view
{
1162 struct radeon_winsys_bo
*bo
;
1164 uint64_t range
; /**< VkBufferViewCreateInfo::range */
1167 void radv_buffer_view_init(struct radv_buffer_view
*view
,
1168 struct radv_device
*device
,
1169 const VkBufferViewCreateInfo
* pCreateInfo
,
1170 struct radv_cmd_buffer
*cmd_buffer
);
1172 static inline struct VkExtent3D
1173 radv_sanitize_image_extent(const VkImageType imageType
,
1174 const struct VkExtent3D imageExtent
)
1176 switch (imageType
) {
1177 case VK_IMAGE_TYPE_1D
:
1178 return (VkExtent3D
) { imageExtent
.width
, 1, 1 };
1179 case VK_IMAGE_TYPE_2D
:
1180 return (VkExtent3D
) { imageExtent
.width
, imageExtent
.height
, 1 };
1181 case VK_IMAGE_TYPE_3D
:
1184 unreachable("invalid image type");
1188 static inline struct VkOffset3D
1189 radv_sanitize_image_offset(const VkImageType imageType
,
1190 const struct VkOffset3D imageOffset
)
1192 switch (imageType
) {
1193 case VK_IMAGE_TYPE_1D
:
1194 return (VkOffset3D
) { imageOffset
.x
, 0, 0 };
1195 case VK_IMAGE_TYPE_2D
:
1196 return (VkOffset3D
) { imageOffset
.x
, imageOffset
.y
, 0 };
1197 case VK_IMAGE_TYPE_3D
:
1200 unreachable("invalid image type");
1204 struct radv_sampler
{
1208 struct radv_color_buffer_info
{
1209 uint32_t cb_color_base
;
1210 uint32_t cb_color_pitch
;
1211 uint32_t cb_color_slice
;
1212 uint32_t cb_color_view
;
1213 uint32_t cb_color_info
;
1214 uint32_t cb_color_attrib
;
1215 uint32_t cb_dcc_control
;
1216 uint32_t cb_color_cmask
;
1217 uint32_t cb_color_cmask_slice
;
1218 uint32_t cb_color_fmask
;
1219 uint32_t cb_color_fmask_slice
;
1220 uint32_t cb_clear_value0
;
1221 uint32_t cb_clear_value1
;
1222 uint32_t cb_dcc_base
;
1223 uint32_t micro_tile_mode
;
1226 struct radv_ds_buffer_info
{
1227 uint32_t db_depth_info
;
1229 uint32_t db_stencil_info
;
1230 uint32_t db_z_read_base
;
1231 uint32_t db_stencil_read_base
;
1232 uint32_t db_z_write_base
;
1233 uint32_t db_stencil_write_base
;
1234 uint32_t db_depth_view
;
1235 uint32_t db_depth_size
;
1236 uint32_t db_depth_slice
;
1237 uint32_t db_htile_surface
;
1238 uint32_t db_htile_data_base
;
1239 uint32_t pa_su_poly_offset_db_fmt_cntl
;
1243 struct radv_attachment_info
{
1245 struct radv_color_buffer_info cb
;
1246 struct radv_ds_buffer_info ds
;
1248 struct radv_image_view
*attachment
;
1251 struct radv_framebuffer
{
1256 uint32_t attachment_count
;
1257 struct radv_attachment_info attachments
[0];
1260 struct radv_subpass_barrier
{
1261 VkPipelineStageFlags src_stage_mask
;
1262 VkAccessFlags src_access_mask
;
1263 VkAccessFlags dst_access_mask
;
1266 struct radv_subpass
{
1267 uint32_t input_count
;
1268 VkAttachmentReference
* input_attachments
;
1269 uint32_t color_count
;
1270 VkAttachmentReference
* color_attachments
;
1271 VkAttachmentReference
* resolve_attachments
;
1272 VkAttachmentReference depth_stencil_attachment
;
1274 /** Subpass has at least one resolve attachment */
1277 struct radv_subpass_barrier start_barrier
;
1280 struct radv_render_pass_attachment
{
1283 VkAttachmentLoadOp load_op
;
1284 VkAttachmentLoadOp stencil_load_op
;
1285 VkImageLayout initial_layout
;
1286 VkImageLayout final_layout
;
1289 struct radv_render_pass
{
1290 uint32_t attachment_count
;
1291 uint32_t subpass_count
;
1292 VkAttachmentReference
* subpass_attachments
;
1293 struct radv_render_pass_attachment
* attachments
;
1294 struct radv_subpass_barrier end_barrier
;
1295 struct radv_subpass subpasses
[0];
1298 VkResult
radv_device_init_meta(struct radv_device
*device
);
1299 void radv_device_finish_meta(struct radv_device
*device
);
1301 struct radv_query_pool
{
1302 struct radeon_winsys_bo
*bo
;
1304 uint32_t availability_offset
;
1310 radv_temp_descriptor_set_create(struct radv_device
*device
,
1311 struct radv_cmd_buffer
*cmd_buffer
,
1312 VkDescriptorSetLayout _layout
,
1313 VkDescriptorSet
*_set
);
1316 radv_temp_descriptor_set_destroy(struct radv_device
*device
,
1317 VkDescriptorSet _set
);
1318 void radv_initialise_cmask(struct radv_cmd_buffer
*cmd_buffer
,
1319 struct radv_image
*image
, uint32_t value
);
1320 void radv_initialize_dcc(struct radv_cmd_buffer
*cmd_buffer
,
1321 struct radv_image
*image
, uint32_t value
);
1324 struct radeon_winsys_fence
*fence
;
1329 #define RADV_DEFINE_HANDLE_CASTS(__radv_type, __VkType) \
1331 static inline struct __radv_type * \
1332 __radv_type ## _from_handle(__VkType _handle) \
1334 return (struct __radv_type *) _handle; \
1337 static inline __VkType \
1338 __radv_type ## _to_handle(struct __radv_type *_obj) \
1340 return (__VkType) _obj; \
1343 #define RADV_DEFINE_NONDISP_HANDLE_CASTS(__radv_type, __VkType) \
1345 static inline struct __radv_type * \
1346 __radv_type ## _from_handle(__VkType _handle) \
1348 return (struct __radv_type *)(uintptr_t) _handle; \
1351 static inline __VkType \
1352 __radv_type ## _to_handle(struct __radv_type *_obj) \
1354 return (__VkType)(uintptr_t) _obj; \
1357 #define RADV_FROM_HANDLE(__radv_type, __name, __handle) \
1358 struct __radv_type *__name = __radv_type ## _from_handle(__handle)
1360 RADV_DEFINE_HANDLE_CASTS(radv_cmd_buffer
, VkCommandBuffer
)
1361 RADV_DEFINE_HANDLE_CASTS(radv_device
, VkDevice
)
1362 RADV_DEFINE_HANDLE_CASTS(radv_instance
, VkInstance
)
1363 RADV_DEFINE_HANDLE_CASTS(radv_physical_device
, VkPhysicalDevice
)
1364 RADV_DEFINE_HANDLE_CASTS(radv_queue
, VkQueue
)
1366 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_cmd_pool
, VkCommandPool
)
1367 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer
, VkBuffer
)
1368 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer_view
, VkBufferView
)
1369 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_pool
, VkDescriptorPool
)
1370 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set
, VkDescriptorSet
)
1371 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set_layout
, VkDescriptorSetLayout
)
1372 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_device_memory
, VkDeviceMemory
)
1373 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_fence
, VkFence
)
1374 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_event
, VkEvent
)
1375 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_framebuffer
, VkFramebuffer
)
1376 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image
, VkImage
)
1377 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image_view
, VkImageView
);
1378 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_cache
, VkPipelineCache
)
1379 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline
, VkPipeline
)
1380 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_layout
, VkPipelineLayout
)
1381 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_query_pool
, VkQueryPool
)
1382 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_render_pass
, VkRenderPass
)
1383 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_sampler
, VkSampler
)
1384 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_shader_module
, VkShaderModule
)
1386 #endif /* RADV_PRIVATE_H */