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
43 #define VG(x) ((void)0)
46 #include "c11/threads.h"
48 #include "compiler/shader_enums.h"
49 #include "util/macros.h"
50 #include "util/list.h"
51 #include "util/xmlconfig.h"
53 #include "vk_debug_report.h"
54 #include "vk_object.h"
56 #include "radv_radeon_winsys.h"
57 #include "ac_binary.h"
58 #include "ac_nir_to_llvm.h"
59 #include "ac_gpu_info.h"
60 #include "ac_surface.h"
61 #include "ac_llvm_build.h"
62 #include "ac_llvm_util.h"
63 #include "radv_constants.h"
64 #include "radv_descriptor_set.h"
65 #include "radv_extensions.h"
68 /* Pre-declarations needed for WSI entrypoints */
71 typedef struct xcb_connection_t xcb_connection_t
;
72 typedef uint32_t xcb_visualid_t
;
73 typedef uint32_t xcb_window_t
;
75 #include <vulkan/vulkan.h>
76 #include <vulkan/vulkan_intel.h>
77 #include <vulkan/vulkan_android.h>
78 #include <vulkan/vk_icd.h>
79 #include <vulkan/vk_android_native_buffer.h>
81 #include "radv_entrypoints.h"
83 #include "wsi_common.h"
84 #include "wsi_common_display.h"
86 /* Helper to determine if we should compile
87 * any of the Android AHB support.
89 * To actually enable the ext we also need
90 * the necessary kernel support.
92 #if defined(ANDROID) && ANDROID_API_LEVEL >= 26
93 #define RADV_SUPPORT_ANDROID_HARDWARE_BUFFER 1
95 #define RADV_SUPPORT_ANDROID_HARDWARE_BUFFER 0
98 #define radv_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
100 static inline uint32_t
101 align_u32(uint32_t v
, uint32_t a
)
103 assert(a
!= 0 && a
== (a
& -a
));
104 return (v
+ a
- 1) & ~(a
- 1);
107 static inline uint32_t
108 align_u32_npot(uint32_t v
, uint32_t a
)
110 return (v
+ a
- 1) / a
* a
;
113 static inline uint64_t
114 align_u64(uint64_t v
, uint64_t a
)
116 assert(a
!= 0 && a
== (a
& -a
));
117 return (v
+ a
- 1) & ~(a
- 1);
120 static inline int32_t
121 align_i32(int32_t v
, int32_t a
)
123 assert(a
!= 0 && a
== (a
& -a
));
124 return (v
+ a
- 1) & ~(a
- 1);
127 /** Alignment must be a power of 2. */
129 radv_is_aligned(uintmax_t n
, uintmax_t a
)
131 assert(a
== (a
& -a
));
132 return (n
& (a
- 1)) == 0;
135 static inline uint32_t
136 round_up_u32(uint32_t v
, uint32_t a
)
138 return (v
+ a
- 1) / a
;
141 static inline uint64_t
142 round_up_u64(uint64_t v
, uint64_t a
)
144 return (v
+ a
- 1) / a
;
147 static inline uint32_t
148 radv_minify(uint32_t n
, uint32_t levels
)
150 if (unlikely(n
== 0))
153 return MAX2(n
>> levels
, 1);
156 radv_clamp_f(float f
, float min
, float max
)
169 radv_clear_mask(uint32_t *inout_mask
, uint32_t clear_mask
)
171 if (*inout_mask
& clear_mask
) {
172 *inout_mask
&= ~clear_mask
;
179 #define for_each_bit(b, dword) \
180 for (uint32_t __dword = (dword); \
181 (b) = __builtin_ffs(__dword) - 1, __dword; \
182 __dword &= ~(1 << (b)))
184 #define typed_memcpy(dest, src, count) ({ \
185 STATIC_ASSERT(sizeof(*src) == sizeof(*dest)); \
186 memcpy((dest), (src), (count) * sizeof(*(src))); \
189 /* Whenever we generate an error, pass it through this function. Useful for
190 * debugging, where we can break on it. Only call at error site, not when
191 * propagating errors. Might be useful to plug in a stack trace here.
194 struct radv_image_view
;
195 struct radv_instance
;
197 VkResult
__vk_errorf(struct radv_instance
*instance
, VkResult error
, const char *file
, int line
, const char *format
, ...);
199 #define vk_error(instance, error) __vk_errorf(instance, error, __FILE__, __LINE__, NULL);
200 #define vk_errorf(instance, error, format, ...) __vk_errorf(instance, error, __FILE__, __LINE__, format, ## __VA_ARGS__);
202 void __radv_finishme(const char *file
, int line
, const char *format
, ...)
203 radv_printflike(3, 4);
204 void radv_loge(const char *format
, ...) radv_printflike(1, 2);
205 void radv_loge_v(const char *format
, va_list va
);
206 void radv_logi(const char *format
, ...) radv_printflike(1, 2);
207 void radv_logi_v(const char *format
, va_list va
);
210 * Print a FINISHME message, including its source location.
212 #define radv_finishme(format, ...) \
214 static bool reported = false; \
216 __radv_finishme(__FILE__, __LINE__, format, ##__VA_ARGS__); \
221 /* A non-fatal assert. Useful for debugging. */
223 #define radv_assert(x) ({ \
224 if (unlikely(!(x))) \
225 fprintf(stderr, "%s:%d ASSERT: %s\n", __FILE__, __LINE__, #x); \
228 #define radv_assert(x) do {} while(0)
231 #define stub_return(v) \
233 radv_finishme("stub %s", __func__); \
239 radv_finishme("stub %s", __func__); \
243 int radv_get_instance_entrypoint_index(const char *name
);
244 int radv_get_device_entrypoint_index(const char *name
);
245 int radv_get_physical_device_entrypoint_index(const char *name
);
247 const char *radv_get_instance_entry_name(int index
);
248 const char *radv_get_physical_device_entry_name(int index
);
249 const char *radv_get_device_entry_name(int index
);
251 bool radv_instance_entrypoint_is_enabled(int index
, uint32_t core_version
,
252 const struct radv_instance_extension_table
*instance
);
253 bool radv_physical_device_entrypoint_is_enabled(int index
, uint32_t core_version
,
254 const struct radv_instance_extension_table
*instance
);
255 bool radv_device_entrypoint_is_enabled(int index
, uint32_t core_version
,
256 const struct radv_instance_extension_table
*instance
,
257 const struct radv_device_extension_table
*device
);
259 void *radv_lookup_entrypoint(const char *name
);
261 struct radv_physical_device
{
262 VK_LOADER_DATA _loader_data
;
264 /* Link in radv_instance::physical_devices */
265 struct list_head link
;
267 struct radv_instance
* instance
;
269 struct radeon_winsys
*ws
;
270 struct radeon_info rad_info
;
271 char name
[VK_MAX_PHYSICAL_DEVICE_NAME_SIZE
];
272 uint8_t driver_uuid
[VK_UUID_SIZE
];
273 uint8_t device_uuid
[VK_UUID_SIZE
];
274 uint8_t cache_uuid
[VK_UUID_SIZE
];
278 struct wsi_device wsi_device
;
280 bool out_of_order_rast_allowed
;
282 /* Whether DCC should be enabled for MSAA textures. */
283 bool dcc_msaa_allowed
;
285 /* Whether to enable NGG. */
288 /* Whether to enable NGG GS. */
291 /* Whether to enable NGG streamout. */
292 bool use_ngg_streamout
;
294 /* Number of threads per wave. */
295 uint8_t ps_wave_size
;
296 uint8_t cs_wave_size
;
297 uint8_t ge_wave_size
;
299 /* Whether to use the LLVM compiler backend */
302 /* This is the drivers on-disk cache used as a fallback as opposed to
303 * the pipeline cache defined by apps.
305 struct disk_cache
* disk_cache
;
307 VkPhysicalDeviceMemoryProperties memory_properties
;
308 enum radeon_bo_domain memory_domains
[VK_MAX_MEMORY_TYPES
];
309 enum radeon_bo_flag memory_flags
[VK_MAX_MEMORY_TYPES
];
311 drmPciBusInfo bus_info
;
313 struct radv_device_extension_table supported_extensions
;
316 struct radv_instance
{
317 struct vk_object_base base
;
319 VkAllocationCallbacks alloc
;
324 uint32_t engineVersion
;
326 uint64_t debug_flags
;
327 uint64_t perftest_flags
;
329 struct vk_debug_report_instance debug_report_callbacks
;
331 struct radv_instance_extension_table enabled_extensions
;
332 struct radv_instance_dispatch_table dispatch
;
333 struct radv_physical_device_dispatch_table physical_device_dispatch
;
334 struct radv_device_dispatch_table device_dispatch
;
336 bool physical_devices_enumerated
;
337 struct list_head physical_devices
;
339 struct driOptionCache dri_options
;
340 struct driOptionCache available_dri_options
;
343 * Workarounds for game bugs.
345 bool enable_mrt_output_nan_fixup
;
348 VkResult
radv_init_wsi(struct radv_physical_device
*physical_device
);
349 void radv_finish_wsi(struct radv_physical_device
*physical_device
);
351 bool radv_instance_extension_supported(const char *name
);
352 uint32_t radv_physical_device_api_version(struct radv_physical_device
*dev
);
353 bool radv_physical_device_extension_supported(struct radv_physical_device
*dev
,
358 struct radv_pipeline_cache
{
359 struct vk_object_base base
;
360 struct radv_device
* device
;
361 pthread_mutex_t mutex
;
362 VkPipelineCacheCreateFlags flags
;
366 uint32_t kernel_count
;
367 struct cache_entry
** hash_table
;
370 VkAllocationCallbacks alloc
;
373 struct radv_pipeline_key
{
374 uint32_t instance_rate_inputs
;
375 uint32_t instance_rate_divisors
[MAX_VERTEX_ATTRIBS
];
376 uint8_t vertex_attribute_formats
[MAX_VERTEX_ATTRIBS
];
377 uint32_t vertex_attribute_bindings
[MAX_VERTEX_ATTRIBS
];
378 uint32_t vertex_attribute_offsets
[MAX_VERTEX_ATTRIBS
];
379 uint32_t vertex_attribute_strides
[MAX_VERTEX_ATTRIBS
];
380 uint64_t vertex_alpha_adjust
;
381 uint32_t vertex_post_shuffle
;
382 unsigned tess_input_vertices
;
386 uint8_t log2_ps_iter_samples
;
389 uint32_t has_multiview_view_index
: 1;
390 uint32_t optimisations_disabled
: 1;
393 /* Non-zero if a required subgroup size is specified via
394 * VK_EXT_subgroup_size_control.
396 uint8_t compute_subgroup_size
;
399 struct radv_shader_binary
;
400 struct radv_shader_variant
;
403 radv_pipeline_cache_init(struct radv_pipeline_cache
*cache
,
404 struct radv_device
*device
);
406 radv_pipeline_cache_finish(struct radv_pipeline_cache
*cache
);
408 radv_pipeline_cache_load(struct radv_pipeline_cache
*cache
,
409 const void *data
, size_t size
);
412 radv_create_shader_variants_from_pipeline_cache(struct radv_device
*device
,
413 struct radv_pipeline_cache
*cache
,
414 const unsigned char *sha1
,
415 struct radv_shader_variant
**variants
,
416 bool *found_in_application_cache
);
419 radv_pipeline_cache_insert_shaders(struct radv_device
*device
,
420 struct radv_pipeline_cache
*cache
,
421 const unsigned char *sha1
,
422 struct radv_shader_variant
**variants
,
423 struct radv_shader_binary
*const *binaries
);
425 enum radv_blit_ds_layout
{
426 RADV_BLIT_DS_LAYOUT_TILE_ENABLE
,
427 RADV_BLIT_DS_LAYOUT_TILE_DISABLE
,
428 RADV_BLIT_DS_LAYOUT_COUNT
,
431 static inline enum radv_blit_ds_layout
radv_meta_blit_ds_to_type(VkImageLayout layout
)
433 return (layout
== VK_IMAGE_LAYOUT_GENERAL
) ? RADV_BLIT_DS_LAYOUT_TILE_DISABLE
: RADV_BLIT_DS_LAYOUT_TILE_ENABLE
;
436 static inline VkImageLayout
radv_meta_blit_ds_to_layout(enum radv_blit_ds_layout ds_layout
)
438 return ds_layout
== RADV_BLIT_DS_LAYOUT_TILE_ENABLE
? VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
: VK_IMAGE_LAYOUT_GENERAL
;
441 enum radv_meta_dst_layout
{
442 RADV_META_DST_LAYOUT_GENERAL
,
443 RADV_META_DST_LAYOUT_OPTIMAL
,
444 RADV_META_DST_LAYOUT_COUNT
,
447 static inline enum radv_meta_dst_layout
radv_meta_dst_layout_from_layout(VkImageLayout layout
)
449 return (layout
== VK_IMAGE_LAYOUT_GENERAL
) ? RADV_META_DST_LAYOUT_GENERAL
: RADV_META_DST_LAYOUT_OPTIMAL
;
452 static inline VkImageLayout
radv_meta_dst_layout_to_layout(enum radv_meta_dst_layout layout
)
454 return layout
== RADV_META_DST_LAYOUT_OPTIMAL
? VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
: VK_IMAGE_LAYOUT_GENERAL
;
457 struct radv_meta_state
{
458 VkAllocationCallbacks alloc
;
460 struct radv_pipeline_cache cache
;
463 * For on-demand pipeline creation, makes sure that
464 * only one thread tries to build a pipeline at the same time.
469 * Use array element `i` for images with `2^i` samples.
472 VkRenderPass render_pass
[NUM_META_FS_KEYS
];
473 VkPipeline color_pipelines
[NUM_META_FS_KEYS
];
475 VkRenderPass depthstencil_rp
;
476 VkPipeline depth_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
477 VkPipeline stencil_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
478 VkPipeline depthstencil_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
480 VkPipeline depth_only_unrestricted_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
481 VkPipeline stencil_only_unrestricted_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
482 VkPipeline depthstencil_unrestricted_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
483 } clear
[MAX_SAMPLES_LOG2
];
485 VkPipelineLayout clear_color_p_layout
;
486 VkPipelineLayout clear_depth_p_layout
;
487 VkPipelineLayout clear_depth_unrestricted_p_layout
;
489 /* Optimized compute fast HTILE clear for stencil or depth only. */
490 VkPipeline clear_htile_mask_pipeline
;
491 VkPipelineLayout clear_htile_mask_p_layout
;
492 VkDescriptorSetLayout clear_htile_mask_ds_layout
;
495 VkRenderPass render_pass
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
497 /** Pipeline that blits from a 1D image. */
498 VkPipeline pipeline_1d_src
[NUM_META_FS_KEYS
];
500 /** Pipeline that blits from a 2D image. */
501 VkPipeline pipeline_2d_src
[NUM_META_FS_KEYS
];
503 /** Pipeline that blits from a 3D image. */
504 VkPipeline pipeline_3d_src
[NUM_META_FS_KEYS
];
506 VkRenderPass depth_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
507 VkPipeline depth_only_1d_pipeline
;
508 VkPipeline depth_only_2d_pipeline
;
509 VkPipeline depth_only_3d_pipeline
;
511 VkRenderPass stencil_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
512 VkPipeline stencil_only_1d_pipeline
;
513 VkPipeline stencil_only_2d_pipeline
;
514 VkPipeline stencil_only_3d_pipeline
;
515 VkPipelineLayout pipeline_layout
;
516 VkDescriptorSetLayout ds_layout
;
520 VkPipelineLayout p_layouts
[5];
521 VkDescriptorSetLayout ds_layouts
[5];
522 VkPipeline pipelines
[5][NUM_META_FS_KEYS
];
524 VkPipeline depth_only_pipeline
[5];
526 VkPipeline stencil_only_pipeline
[5];
527 } blit2d
[MAX_SAMPLES_LOG2
];
529 VkRenderPass blit2d_render_passes
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
530 VkRenderPass blit2d_depth_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
531 VkRenderPass blit2d_stencil_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
534 VkPipelineLayout img_p_layout
;
535 VkDescriptorSetLayout img_ds_layout
;
537 VkPipeline pipeline_3d
;
540 VkPipelineLayout img_p_layout
;
541 VkDescriptorSetLayout img_ds_layout
;
543 VkPipeline pipeline_3d
;
546 VkPipelineLayout img_p_layout
;
547 VkDescriptorSetLayout img_ds_layout
;
551 VkPipelineLayout img_p_layout
;
552 VkDescriptorSetLayout img_ds_layout
;
554 VkPipeline pipeline_3d
;
557 VkPipelineLayout img_p_layout
;
558 VkDescriptorSetLayout img_ds_layout
;
562 VkPipelineLayout img_p_layout
;
563 VkDescriptorSetLayout img_ds_layout
;
565 VkPipeline pipeline_3d
;
568 VkPipelineLayout img_p_layout
;
569 VkDescriptorSetLayout img_ds_layout
;
574 VkPipelineLayout p_layout
;
575 VkPipeline pipeline
[NUM_META_FS_KEYS
];
576 VkRenderPass pass
[NUM_META_FS_KEYS
];
580 VkDescriptorSetLayout ds_layout
;
581 VkPipelineLayout p_layout
;
584 VkPipeline i_pipeline
;
585 VkPipeline srgb_pipeline
;
586 } rc
[MAX_SAMPLES_LOG2
];
588 VkPipeline depth_zero_pipeline
;
590 VkPipeline average_pipeline
;
591 VkPipeline max_pipeline
;
592 VkPipeline min_pipeline
;
593 } depth
[MAX_SAMPLES_LOG2
];
595 VkPipeline stencil_zero_pipeline
;
597 VkPipeline max_pipeline
;
598 VkPipeline min_pipeline
;
599 } stencil
[MAX_SAMPLES_LOG2
];
603 VkDescriptorSetLayout ds_layout
;
604 VkPipelineLayout p_layout
;
607 VkRenderPass render_pass
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
608 VkPipeline pipeline
[NUM_META_FS_KEYS
];
609 } rc
[MAX_SAMPLES_LOG2
];
611 VkRenderPass depth_render_pass
;
612 VkPipeline depth_zero_pipeline
;
614 VkPipeline average_pipeline
;
615 VkPipeline max_pipeline
;
616 VkPipeline min_pipeline
;
617 } depth
[MAX_SAMPLES_LOG2
];
619 VkRenderPass stencil_render_pass
;
620 VkPipeline stencil_zero_pipeline
;
622 VkPipeline max_pipeline
;
623 VkPipeline min_pipeline
;
624 } stencil
[MAX_SAMPLES_LOG2
];
628 VkPipelineLayout p_layout
;
629 VkPipeline decompress_pipeline
[NUM_DEPTH_DECOMPRESS_PIPELINES
];
630 VkPipeline resummarize_pipeline
;
632 } depth_decomp
[MAX_SAMPLES_LOG2
];
635 VkPipelineLayout p_layout
;
636 VkPipeline cmask_eliminate_pipeline
;
637 VkPipeline fmask_decompress_pipeline
;
638 VkPipeline dcc_decompress_pipeline
;
641 VkDescriptorSetLayout dcc_decompress_compute_ds_layout
;
642 VkPipelineLayout dcc_decompress_compute_p_layout
;
643 VkPipeline dcc_decompress_compute_pipeline
;
647 VkPipelineLayout fill_p_layout
;
648 VkPipelineLayout copy_p_layout
;
649 VkDescriptorSetLayout fill_ds_layout
;
650 VkDescriptorSetLayout copy_ds_layout
;
651 VkPipeline fill_pipeline
;
652 VkPipeline copy_pipeline
;
656 VkDescriptorSetLayout ds_layout
;
657 VkPipelineLayout p_layout
;
658 VkPipeline occlusion_query_pipeline
;
659 VkPipeline pipeline_statistics_query_pipeline
;
660 VkPipeline tfb_query_pipeline
;
661 VkPipeline timestamp_query_pipeline
;
665 VkDescriptorSetLayout ds_layout
;
666 VkPipelineLayout p_layout
;
667 VkPipeline pipeline
[MAX_SAMPLES_LOG2
];
672 #define RADV_QUEUE_GENERAL 0
673 #define RADV_QUEUE_COMPUTE 1
674 #define RADV_QUEUE_TRANSFER 2
676 #define RADV_MAX_QUEUE_FAMILIES 3
678 enum ring_type
radv_queue_family_to_ring(int f
);
681 VK_LOADER_DATA _loader_data
;
682 struct radv_device
* device
;
683 struct radeon_winsys_ctx
*hw_ctx
;
684 enum radeon_ctx_priority priority
;
685 uint32_t queue_family_index
;
687 VkDeviceQueueCreateFlags flags
;
689 uint32_t scratch_size_per_wave
;
690 uint32_t scratch_waves
;
691 uint32_t compute_scratch_size_per_wave
;
692 uint32_t compute_scratch_waves
;
693 uint32_t esgs_ring_size
;
694 uint32_t gsvs_ring_size
;
698 bool has_sample_positions
;
700 struct radeon_winsys_bo
*scratch_bo
;
701 struct radeon_winsys_bo
*descriptor_bo
;
702 struct radeon_winsys_bo
*compute_scratch_bo
;
703 struct radeon_winsys_bo
*esgs_ring_bo
;
704 struct radeon_winsys_bo
*gsvs_ring_bo
;
705 struct radeon_winsys_bo
*tess_rings_bo
;
706 struct radeon_winsys_bo
*gds_bo
;
707 struct radeon_winsys_bo
*gds_oa_bo
;
708 struct radeon_cmdbuf
*initial_preamble_cs
;
709 struct radeon_cmdbuf
*initial_full_flush_preamble_cs
;
710 struct radeon_cmdbuf
*continue_preamble_cs
;
712 struct list_head pending_submissions
;
713 pthread_mutex_t pending_mutex
;
716 struct radv_bo_list
{
717 struct radv_winsys_bo_list list
;
719 pthread_mutex_t mutex
;
722 VkResult
radv_bo_list_add(struct radv_device
*device
,
723 struct radeon_winsys_bo
*bo
);
724 void radv_bo_list_remove(struct radv_device
*device
,
725 struct radeon_winsys_bo
*bo
);
727 #define RADV_BORDER_COLOR_COUNT 4096
728 #define RADV_BORDER_COLOR_BUFFER_SIZE (sizeof(VkClearColorValue) * RADV_BORDER_COLOR_COUNT)
730 struct radv_device_border_color_data
{
731 bool used
[RADV_BORDER_COLOR_COUNT
];
733 struct radeon_winsys_bo
*bo
;
734 VkClearColorValue
*colors_gpu_ptr
;
736 /* Mutex is required to guarantee vkCreateSampler thread safety
737 * given that we are writing to a buffer and checking color occupation */
738 pthread_mutex_t mutex
;
744 struct radv_instance
* instance
;
745 struct radeon_winsys
*ws
;
747 struct radv_meta_state meta_state
;
749 struct radv_queue
*queues
[RADV_MAX_QUEUE_FAMILIES
];
750 int queue_count
[RADV_MAX_QUEUE_FAMILIES
];
751 struct radeon_cmdbuf
*empty_cs
[RADV_MAX_QUEUE_FAMILIES
];
753 bool always_use_syncobj
;
756 uint32_t tess_offchip_block_dw_size
;
757 uint32_t scratch_waves
;
758 uint32_t dispatch_initiator
;
760 uint32_t gs_table_depth
;
762 /* MSAA sample locations.
763 * The first index is the sample index.
764 * The second index is the coordinate: X, Y. */
765 float sample_locations_1x
[1][2];
766 float sample_locations_2x
[2][2];
767 float sample_locations_4x
[4][2];
768 float sample_locations_8x
[8][2];
771 uint32_t gfx_init_size_dw
;
772 struct radeon_winsys_bo
*gfx_init
;
774 struct radeon_winsys_bo
*trace_bo
;
775 uint32_t *trace_id_ptr
;
777 /* Whether to keep shader debug info, for tracing or VK_AMD_shader_info */
778 bool keep_shader_info
;
780 struct radv_physical_device
*physical_device
;
782 /* Backup in-memory cache to be used if the app doesn't provide one */
783 struct radv_pipeline_cache
* mem_cache
;
786 * use different counters so MSAA MRTs get consecutive surface indices,
787 * even if MASK is allocated in between.
789 uint32_t image_mrt_offset_counter
;
790 uint32_t fmask_mrt_offset_counter
;
791 struct list_head shader_slabs
;
792 mtx_t shader_slab_mutex
;
794 /* For detecting VM faults reported by dmesg. */
795 uint64_t dmesg_timestamp
;
797 struct radv_device_extension_table enabled_extensions
;
798 struct radv_device_dispatch_table dispatch
;
800 /* Whether the app has enabled the robustBufferAccess feature. */
801 bool robust_buffer_access
;
803 /* Whether the driver uses a global BO list. */
804 bool use_global_bo_list
;
806 struct radv_bo_list bo_list
;
808 /* Whether anisotropy is forced with RADV_TEX_ANISO (-1 is disabled). */
811 struct radv_device_border_color_data border_color_data
;
813 /* Condition variable for legacy timelines, to notify waiters when a
814 * new point gets submitted. */
815 pthread_cond_t timeline_cond
;
818 struct radeon_cmdbuf
*thread_trace_start_cs
[2];
819 struct radeon_cmdbuf
*thread_trace_stop_cs
[2];
820 struct radeon_winsys_bo
*thread_trace_bo
;
821 void *thread_trace_ptr
;
822 uint32_t thread_trace_buffer_size
;
823 int thread_trace_start_frame
;
825 /* Overallocation. */
826 bool overallocation_disallowed
;
827 uint64_t allocated_memory_size
[VK_MAX_MEMORY_HEAPS
];
828 mtx_t overallocation_mutex
;
831 struct radv_device_memory
{
832 struct vk_object_base base
;
833 struct radeon_winsys_bo
*bo
;
834 /* for dedicated allocations */
835 struct radv_image
*image
;
836 struct radv_buffer
*buffer
;
842 #if RADV_SUPPORT_ANDROID_HARDWARE_BUFFER
843 struct AHardwareBuffer
* android_hardware_buffer
;
848 struct radv_descriptor_range
{
853 struct radv_descriptor_set
{
854 struct vk_object_base base
;
855 const struct radv_descriptor_set_layout
*layout
;
857 uint32_t buffer_count
;
859 struct radeon_winsys_bo
*bo
;
861 uint32_t *mapped_ptr
;
862 struct radv_descriptor_range
*dynamic_descriptors
;
864 struct radeon_winsys_bo
*descriptors
[0];
867 struct radv_push_descriptor_set
869 struct radv_descriptor_set set
;
873 struct radv_descriptor_pool_entry
{
876 struct radv_descriptor_set
*set
;
879 struct radv_descriptor_pool
{
880 struct vk_object_base base
;
881 struct radeon_winsys_bo
*bo
;
883 uint64_t current_offset
;
886 uint8_t *host_memory_base
;
887 uint8_t *host_memory_ptr
;
888 uint8_t *host_memory_end
;
890 uint32_t entry_count
;
891 uint32_t max_entry_count
;
892 struct radv_descriptor_pool_entry entries
[0];
895 struct radv_descriptor_update_template_entry
{
896 VkDescriptorType descriptor_type
;
898 /* The number of descriptors to update */
899 uint32_t descriptor_count
;
901 /* Into mapped_ptr or dynamic_descriptors, in units of the respective array */
904 /* In dwords. Not valid/used for dynamic descriptors */
907 uint32_t buffer_offset
;
909 /* Only valid for combined image samplers and samplers */
911 uint8_t sampler_offset
;
917 /* For push descriptors */
918 const uint32_t *immutable_samplers
;
921 struct radv_descriptor_update_template
{
922 struct vk_object_base base
;
923 uint32_t entry_count
;
924 VkPipelineBindPoint bind_point
;
925 struct radv_descriptor_update_template_entry entry
[0];
929 struct vk_object_base base
;
932 VkBufferUsageFlags usage
;
933 VkBufferCreateFlags flags
;
936 struct radeon_winsys_bo
* bo
;
942 enum radv_dynamic_state_bits
{
943 RADV_DYNAMIC_VIEWPORT
= 1 << 0,
944 RADV_DYNAMIC_SCISSOR
= 1 << 1,
945 RADV_DYNAMIC_LINE_WIDTH
= 1 << 2,
946 RADV_DYNAMIC_DEPTH_BIAS
= 1 << 3,
947 RADV_DYNAMIC_BLEND_CONSTANTS
= 1 << 4,
948 RADV_DYNAMIC_DEPTH_BOUNDS
= 1 << 5,
949 RADV_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6,
950 RADV_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7,
951 RADV_DYNAMIC_STENCIL_REFERENCE
= 1 << 8,
952 RADV_DYNAMIC_DISCARD_RECTANGLE
= 1 << 9,
953 RADV_DYNAMIC_SAMPLE_LOCATIONS
= 1 << 10,
954 RADV_DYNAMIC_LINE_STIPPLE
= 1 << 11,
955 RADV_DYNAMIC_CULL_MODE
= 1 << 12,
956 RADV_DYNAMIC_FRONT_FACE
= 1 << 13,
957 RADV_DYNAMIC_PRIMITIVE_TOPOLOGY
= 1 << 14,
958 RADV_DYNAMIC_DEPTH_TEST_ENABLE
= 1 << 15,
959 RADV_DYNAMIC_DEPTH_WRITE_ENABLE
= 1 << 16,
960 RADV_DYNAMIC_DEPTH_COMPARE_OP
= 1 << 17,
961 RADV_DYNAMIC_DEPTH_BOUNDS_TEST_ENABLE
= 1 << 18,
962 RADV_DYNAMIC_STENCIL_TEST_ENABLE
= 1 << 19,
963 RADV_DYNAMIC_STENCIL_OP
= 1 << 20,
964 RADV_DYNAMIC_VERTEX_INPUT_BINDING_STRIDE
= 1 << 21,
965 RADV_DYNAMIC_ALL
= (1 << 22) - 1,
968 enum radv_cmd_dirty_bits
{
969 /* Keep the dynamic state dirty bits in sync with
970 * enum radv_dynamic_state_bits */
971 RADV_CMD_DIRTY_DYNAMIC_VIEWPORT
= 1 << 0,
972 RADV_CMD_DIRTY_DYNAMIC_SCISSOR
= 1 << 1,
973 RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH
= 1 << 2,
974 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS
= 1 << 3,
975 RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS
= 1 << 4,
976 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS
= 1 << 5,
977 RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6,
978 RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7,
979 RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE
= 1 << 8,
980 RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE
= 1 << 9,
981 RADV_CMD_DIRTY_DYNAMIC_SAMPLE_LOCATIONS
= 1 << 10,
982 RADV_CMD_DIRTY_DYNAMIC_LINE_STIPPLE
= 1 << 11,
983 RADV_CMD_DIRTY_DYNAMIC_CULL_MODE
= 1 << 12,
984 RADV_CMD_DIRTY_DYNAMIC_FRONT_FACE
= 1 << 13,
985 RADV_CMD_DIRTY_DYNAMIC_PRIMITIVE_TOPOLOGY
= 1 << 14,
986 RADV_CMD_DIRTY_DYNAMIC_DEPTH_TEST_ENABLE
= 1 << 15,
987 RADV_CMD_DIRTY_DYNAMIC_DEPTH_WRITE_ENABLE
= 1 << 16,
988 RADV_CMD_DIRTY_DYNAMIC_DEPTH_COMPARE_OP
= 1 << 17,
989 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS_TEST_ENABLE
= 1 << 18,
990 RADV_CMD_DIRTY_DYNAMIC_STENCIL_TEST_ENABLE
= 1 << 19,
991 RADV_CMD_DIRTY_DYNAMIC_STENCIL_OP
= 1 << 20,
992 RADV_CMD_DIRTY_DYNAMIC_VERTEX_INPUT_BINDING_STRIDE
= 1 << 21,
993 RADV_CMD_DIRTY_DYNAMIC_ALL
= (1 << 22) - 1,
994 RADV_CMD_DIRTY_PIPELINE
= 1 << 22,
995 RADV_CMD_DIRTY_INDEX_BUFFER
= 1 << 23,
996 RADV_CMD_DIRTY_FRAMEBUFFER
= 1 << 24,
997 RADV_CMD_DIRTY_VERTEX_BUFFER
= 1 << 25,
998 RADV_CMD_DIRTY_STREAMOUT_BUFFER
= 1 << 26,
1001 enum radv_cmd_flush_bits
{
1002 /* Instruction cache. */
1003 RADV_CMD_FLAG_INV_ICACHE
= 1 << 0,
1004 /* Scalar L1 cache. */
1005 RADV_CMD_FLAG_INV_SCACHE
= 1 << 1,
1006 /* Vector L1 cache. */
1007 RADV_CMD_FLAG_INV_VCACHE
= 1 << 2,
1008 /* L2 cache + L2 metadata cache writeback & invalidate.
1009 * GFX6-8: Used by shaders only. GFX9-10: Used by everything. */
1010 RADV_CMD_FLAG_INV_L2
= 1 << 3,
1011 /* L2 writeback (write dirty L2 lines to memory for non-L2 clients).
1012 * Only used for coherency with non-L2 clients like CB, DB, CP on GFX6-8.
1013 * GFX6-7 will do complete invalidation, because the writeback is unsupported. */
1014 RADV_CMD_FLAG_WB_L2
= 1 << 4,
1015 /* Framebuffer caches */
1016 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
= 1 << 5,
1017 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
= 1 << 6,
1018 RADV_CMD_FLAG_FLUSH_AND_INV_DB
= 1 << 7,
1019 RADV_CMD_FLAG_FLUSH_AND_INV_CB
= 1 << 8,
1020 /* Engine synchronization. */
1021 RADV_CMD_FLAG_VS_PARTIAL_FLUSH
= 1 << 9,
1022 RADV_CMD_FLAG_PS_PARTIAL_FLUSH
= 1 << 10,
1023 RADV_CMD_FLAG_CS_PARTIAL_FLUSH
= 1 << 11,
1024 RADV_CMD_FLAG_VGT_FLUSH
= 1 << 12,
1025 /* Pipeline query controls. */
1026 RADV_CMD_FLAG_START_PIPELINE_STATS
= 1 << 13,
1027 RADV_CMD_FLAG_STOP_PIPELINE_STATS
= 1 << 14,
1028 RADV_CMD_FLAG_VGT_STREAMOUT_SYNC
= 1 << 15,
1030 RADV_CMD_FLUSH_AND_INV_FRAMEBUFFER
= (RADV_CMD_FLAG_FLUSH_AND_INV_CB
|
1031 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
|
1032 RADV_CMD_FLAG_FLUSH_AND_INV_DB
|
1033 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
)
1036 struct radv_vertex_binding
{
1037 struct radv_buffer
* buffer
;
1038 VkDeviceSize offset
;
1041 struct radv_streamout_binding
{
1042 struct radv_buffer
*buffer
;
1043 VkDeviceSize offset
;
1047 struct radv_streamout_state
{
1048 /* Mask of bound streamout buffers. */
1049 uint8_t enabled_mask
;
1051 /* External state that comes from the last vertex stage, it must be
1052 * set explicitely when binding a new graphics pipeline.
1054 uint16_t stride_in_dw
[MAX_SO_BUFFERS
];
1055 uint32_t enabled_stream_buffers_mask
; /* stream0 buffers0-3 in 4 LSB */
1057 /* State of VGT_STRMOUT_BUFFER_(CONFIG|END) */
1058 uint32_t hw_enabled_mask
;
1060 /* State of VGT_STRMOUT_(CONFIG|EN) */
1061 bool streamout_enabled
;
1064 struct radv_viewport_state
{
1066 VkViewport viewports
[MAX_VIEWPORTS
];
1069 struct radv_scissor_state
{
1071 VkRect2D scissors
[MAX_SCISSORS
];
1074 struct radv_discard_rectangle_state
{
1076 VkRect2D rectangles
[MAX_DISCARD_RECTANGLES
];
1079 struct radv_sample_locations_state
{
1080 VkSampleCountFlagBits per_pixel
;
1081 VkExtent2D grid_size
;
1083 VkSampleLocationEXT locations
[MAX_SAMPLE_LOCATIONS
];
1086 struct radv_dynamic_state
{
1088 * Bitmask of (1 << VK_DYNAMIC_STATE_*).
1089 * Defines the set of saved dynamic state.
1093 struct radv_viewport_state viewport
;
1095 struct radv_scissor_state scissor
;
1105 float blend_constants
[4];
1115 } stencil_compare_mask
;
1120 } stencil_write_mask
;
1125 } stencil_reference
;
1127 struct radv_discard_rectangle_state discard_rectangle
;
1129 struct radv_sample_locations_state sample_location
;
1136 VkCullModeFlags cull_mode
;
1137 VkFrontFace front_face
;
1140 extern const struct radv_dynamic_state default_dynamic_state
;
1143 radv_get_debug_option_name(int id
);
1146 radv_get_perftest_option_name(int id
);
1148 struct radv_color_buffer_info
{
1149 uint64_t cb_color_base
;
1150 uint64_t cb_color_cmask
;
1151 uint64_t cb_color_fmask
;
1152 uint64_t cb_dcc_base
;
1153 uint32_t cb_color_slice
;
1154 uint32_t cb_color_view
;
1155 uint32_t cb_color_info
;
1156 uint32_t cb_color_attrib
;
1157 uint32_t cb_color_attrib2
; /* GFX9 and later */
1158 uint32_t cb_color_attrib3
; /* GFX10 and later */
1159 uint32_t cb_dcc_control
;
1160 uint32_t cb_color_cmask_slice
;
1161 uint32_t cb_color_fmask_slice
;
1163 uint32_t cb_color_pitch
; // GFX6-GFX8
1164 uint32_t cb_mrt_epitch
; // GFX9+
1168 struct radv_ds_buffer_info
{
1169 uint64_t db_z_read_base
;
1170 uint64_t db_stencil_read_base
;
1171 uint64_t db_z_write_base
;
1172 uint64_t db_stencil_write_base
;
1173 uint64_t db_htile_data_base
;
1174 uint32_t db_depth_info
;
1176 uint32_t db_stencil_info
;
1177 uint32_t db_depth_view
;
1178 uint32_t db_depth_size
;
1179 uint32_t db_depth_slice
;
1180 uint32_t db_htile_surface
;
1181 uint32_t pa_su_poly_offset_db_fmt_cntl
;
1182 uint32_t db_z_info2
; /* GFX9 only */
1183 uint32_t db_stencil_info2
; /* GFX9 only */
1188 radv_initialise_color_surface(struct radv_device
*device
,
1189 struct radv_color_buffer_info
*cb
,
1190 struct radv_image_view
*iview
);
1192 radv_initialise_ds_surface(struct radv_device
*device
,
1193 struct radv_ds_buffer_info
*ds
,
1194 struct radv_image_view
*iview
);
1197 * Attachment state when recording a renderpass instance.
1199 * The clear value is valid only if there exists a pending clear.
1201 struct radv_attachment_state
{
1202 VkImageAspectFlags pending_clear_aspects
;
1203 uint32_t cleared_views
;
1204 VkClearValue clear_value
;
1205 VkImageLayout current_layout
;
1206 VkImageLayout current_stencil_layout
;
1207 bool current_in_render_loop
;
1208 struct radv_sample_locations_state sample_location
;
1211 struct radv_color_buffer_info cb
;
1212 struct radv_ds_buffer_info ds
;
1214 struct radv_image_view
*iview
;
1217 struct radv_descriptor_state
{
1218 struct radv_descriptor_set
*sets
[MAX_SETS
];
1221 struct radv_push_descriptor_set push_set
;
1223 uint32_t dynamic_buffers
[4 * MAX_DYNAMIC_BUFFERS
];
1226 struct radv_subpass_sample_locs_state
{
1227 uint32_t subpass_idx
;
1228 struct radv_sample_locations_state sample_location
;
1231 struct radv_cmd_state
{
1232 /* Vertex descriptors */
1239 uint32_t prefetch_L2_mask
;
1241 struct radv_pipeline
* pipeline
;
1242 struct radv_pipeline
* emitted_pipeline
;
1243 struct radv_pipeline
* compute_pipeline
;
1244 struct radv_pipeline
* emitted_compute_pipeline
;
1245 struct radv_framebuffer
* framebuffer
;
1246 struct radv_render_pass
* pass
;
1247 const struct radv_subpass
* subpass
;
1248 struct radv_dynamic_state dynamic
;
1249 struct radv_attachment_state
* attachments
;
1250 struct radv_streamout_state streamout
;
1251 VkRect2D render_area
;
1253 uint32_t num_subpass_sample_locs
;
1254 struct radv_subpass_sample_locs_state
* subpass_sample_locs
;
1257 struct radv_buffer
*index_buffer
;
1258 uint64_t index_offset
;
1259 uint32_t index_type
;
1260 uint32_t max_index_count
;
1262 int32_t last_index_type
;
1264 int32_t last_primitive_reset_en
;
1265 uint32_t last_primitive_reset_index
;
1266 enum radv_cmd_flush_bits flush_bits
;
1267 unsigned active_occlusion_queries
;
1268 bool perfect_occlusion_queries_enabled
;
1269 unsigned active_pipeline_queries
;
1270 unsigned active_pipeline_gds_queries
;
1273 uint32_t last_ia_multi_vgt_param
;
1275 uint32_t last_num_instances
;
1276 uint32_t last_first_instance
;
1277 uint32_t last_vertex_offset
;
1279 uint32_t last_sx_ps_downconvert
;
1280 uint32_t last_sx_blend_opt_epsilon
;
1281 uint32_t last_sx_blend_opt_control
;
1283 /* Whether CP DMA is busy/idle. */
1286 /* Conditional rendering info. */
1287 int predication_type
; /* -1: disabled, 0: normal, 1: inverted */
1288 uint64_t predication_va
;
1290 /* Inheritance info. */
1291 VkQueryPipelineStatisticFlags inherited_pipeline_statistics
;
1293 bool context_roll_without_scissor_emitted
;
1295 /* SQTT related state. */
1296 uint32_t current_event_type
;
1297 uint32_t num_events
;
1298 uint32_t num_layout_transitions
;
1301 struct radv_cmd_pool
{
1302 struct vk_object_base base
;
1303 VkAllocationCallbacks alloc
;
1304 struct list_head cmd_buffers
;
1305 struct list_head free_cmd_buffers
;
1306 uint32_t queue_family_index
;
1309 struct radv_cmd_buffer_upload
{
1313 struct radeon_winsys_bo
*upload_bo
;
1314 struct list_head list
;
1317 enum radv_cmd_buffer_status
{
1318 RADV_CMD_BUFFER_STATUS_INVALID
,
1319 RADV_CMD_BUFFER_STATUS_INITIAL
,
1320 RADV_CMD_BUFFER_STATUS_RECORDING
,
1321 RADV_CMD_BUFFER_STATUS_EXECUTABLE
,
1322 RADV_CMD_BUFFER_STATUS_PENDING
,
1325 struct radv_cmd_buffer
{
1326 struct vk_object_base base
;
1328 struct radv_device
* device
;
1330 struct radv_cmd_pool
* pool
;
1331 struct list_head pool_link
;
1333 VkCommandBufferUsageFlags usage_flags
;
1334 VkCommandBufferLevel level
;
1335 enum radv_cmd_buffer_status status
;
1336 struct radeon_cmdbuf
*cs
;
1337 struct radv_cmd_state state
;
1338 struct radv_vertex_binding vertex_bindings
[MAX_VBS
];
1339 struct radv_streamout_binding streamout_bindings
[MAX_SO_BUFFERS
];
1340 uint32_t queue_family_index
;
1342 uint8_t push_constants
[MAX_PUSH_CONSTANTS_SIZE
];
1343 VkShaderStageFlags push_constant_stages
;
1344 struct radv_descriptor_set meta_push_descriptors
;
1346 struct radv_descriptor_state descriptors
[MAX_BIND_POINTS
];
1348 struct radv_cmd_buffer_upload upload
;
1350 uint32_t scratch_size_per_wave_needed
;
1351 uint32_t scratch_waves_wanted
;
1352 uint32_t compute_scratch_size_per_wave_needed
;
1353 uint32_t compute_scratch_waves_wanted
;
1354 uint32_t esgs_ring_size_needed
;
1355 uint32_t gsvs_ring_size_needed
;
1356 bool tess_rings_needed
;
1357 bool gds_needed
; /* for GFX10 streamout and NGG GS queries */
1358 bool gds_oa_needed
; /* for GFX10 streamout */
1359 bool sample_positions_needed
;
1361 VkResult record_result
;
1363 uint64_t gfx9_fence_va
;
1364 uint32_t gfx9_fence_idx
;
1365 uint64_t gfx9_eop_bug_va
;
1368 * Whether a query pool has been resetted and we have to flush caches.
1370 bool pending_reset_query
;
1373 * Bitmask of pending active query flushes.
1375 enum radv_cmd_flush_bits active_query_flush_bits
;
1379 struct radv_image_view
;
1381 bool radv_cmd_buffer_uses_mec(struct radv_cmd_buffer
*cmd_buffer
);
1383 void si_emit_graphics(struct radv_device
*device
,
1384 struct radeon_cmdbuf
*cs
);
1385 void si_emit_compute(struct radv_physical_device
*physical_device
,
1386 struct radeon_cmdbuf
*cs
);
1388 void cik_create_gfx_config(struct radv_device
*device
);
1390 void si_write_viewport(struct radeon_cmdbuf
*cs
, int first_vp
,
1391 int count
, const VkViewport
*viewports
);
1392 void si_write_scissors(struct radeon_cmdbuf
*cs
, int first
,
1393 int count
, const VkRect2D
*scissors
,
1394 const VkViewport
*viewports
, bool can_use_guardband
);
1395 uint32_t si_get_ia_multi_vgt_param(struct radv_cmd_buffer
*cmd_buffer
,
1396 bool instanced_draw
, bool indirect_draw
,
1397 bool count_from_stream_output
,
1398 uint32_t draw_vertex_count
);
1399 void si_cs_emit_write_event_eop(struct radeon_cmdbuf
*cs
,
1400 enum chip_class chip_class
,
1402 unsigned event
, unsigned event_flags
,
1403 unsigned dst_sel
, unsigned data_sel
,
1406 uint64_t gfx9_eop_bug_va
);
1408 void radv_cp_wait_mem(struct radeon_cmdbuf
*cs
, uint32_t op
, uint64_t va
,
1409 uint32_t ref
, uint32_t mask
);
1410 void si_cs_emit_cache_flush(struct radeon_cmdbuf
*cs
,
1411 enum chip_class chip_class
,
1412 uint32_t *fence_ptr
, uint64_t va
,
1414 enum radv_cmd_flush_bits flush_bits
,
1415 uint64_t gfx9_eop_bug_va
);
1416 void si_emit_cache_flush(struct radv_cmd_buffer
*cmd_buffer
);
1417 void si_emit_set_predication_state(struct radv_cmd_buffer
*cmd_buffer
,
1418 bool inverted
, uint64_t va
);
1419 void si_cp_dma_buffer_copy(struct radv_cmd_buffer
*cmd_buffer
,
1420 uint64_t src_va
, uint64_t dest_va
,
1422 void si_cp_dma_prefetch(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
1424 void si_cp_dma_clear_buffer(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
1425 uint64_t size
, unsigned value
);
1426 void si_cp_dma_wait_for_idle(struct radv_cmd_buffer
*cmd_buffer
);
1428 void radv_set_db_count_control(struct radv_cmd_buffer
*cmd_buffer
);
1430 radv_cmd_buffer_upload_alloc(struct radv_cmd_buffer
*cmd_buffer
,
1433 unsigned *out_offset
,
1436 radv_cmd_buffer_set_subpass(struct radv_cmd_buffer
*cmd_buffer
,
1437 const struct radv_subpass
*subpass
);
1439 radv_cmd_buffer_upload_data(struct radv_cmd_buffer
*cmd_buffer
,
1440 unsigned size
, unsigned alignmnet
,
1441 const void *data
, unsigned *out_offset
);
1443 void radv_cmd_buffer_clear_subpass(struct radv_cmd_buffer
*cmd_buffer
);
1444 void radv_cmd_buffer_resolve_subpass(struct radv_cmd_buffer
*cmd_buffer
);
1445 void radv_cmd_buffer_resolve_subpass_cs(struct radv_cmd_buffer
*cmd_buffer
);
1446 void radv_depth_stencil_resolve_subpass_cs(struct radv_cmd_buffer
*cmd_buffer
,
1447 VkImageAspectFlags aspects
,
1448 VkResolveModeFlagBits resolve_mode
);
1449 void radv_cmd_buffer_resolve_subpass_fs(struct radv_cmd_buffer
*cmd_buffer
);
1450 void radv_depth_stencil_resolve_subpass_fs(struct radv_cmd_buffer
*cmd_buffer
,
1451 VkImageAspectFlags aspects
,
1452 VkResolveModeFlagBits resolve_mode
);
1453 void radv_emit_default_sample_locations(struct radeon_cmdbuf
*cs
, int nr_samples
);
1454 unsigned radv_get_default_max_sample_dist(int log_samples
);
1455 void radv_device_init_msaa(struct radv_device
*device
);
1457 void radv_update_ds_clear_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1458 const struct radv_image_view
*iview
,
1459 VkClearDepthStencilValue ds_clear_value
,
1460 VkImageAspectFlags aspects
);
1462 void radv_update_color_clear_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1463 const struct radv_image_view
*iview
,
1465 uint32_t color_values
[2]);
1467 void radv_update_fce_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1468 struct radv_image
*image
,
1469 const VkImageSubresourceRange
*range
, bool value
);
1471 void radv_update_dcc_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1472 struct radv_image
*image
,
1473 const VkImageSubresourceRange
*range
, bool value
);
1475 uint32_t radv_fill_buffer(struct radv_cmd_buffer
*cmd_buffer
,
1476 struct radeon_winsys_bo
*bo
,
1477 uint64_t offset
, uint64_t size
, uint32_t value
);
1478 void radv_cmd_buffer_trace_emit(struct radv_cmd_buffer
*cmd_buffer
);
1479 bool radv_get_memory_fd(struct radv_device
*device
,
1480 struct radv_device_memory
*memory
,
1484 radv_emit_shader_pointer_head(struct radeon_cmdbuf
*cs
,
1485 unsigned sh_offset
, unsigned pointer_count
,
1486 bool use_32bit_pointers
)
1488 radeon_emit(cs
, PKT3(PKT3_SET_SH_REG
, pointer_count
* (use_32bit_pointers
? 1 : 2), 0));
1489 radeon_emit(cs
, (sh_offset
- SI_SH_REG_OFFSET
) >> 2);
1493 radv_emit_shader_pointer_body(struct radv_device
*device
,
1494 struct radeon_cmdbuf
*cs
,
1495 uint64_t va
, bool use_32bit_pointers
)
1497 radeon_emit(cs
, va
);
1499 if (use_32bit_pointers
) {
1501 (va
>> 32) == device
->physical_device
->rad_info
.address32_hi
);
1503 radeon_emit(cs
, va
>> 32);
1508 radv_emit_shader_pointer(struct radv_device
*device
,
1509 struct radeon_cmdbuf
*cs
,
1510 uint32_t sh_offset
, uint64_t va
, bool global
)
1512 bool use_32bit_pointers
= !global
;
1514 radv_emit_shader_pointer_head(cs
, sh_offset
, 1, use_32bit_pointers
);
1515 radv_emit_shader_pointer_body(device
, cs
, va
, use_32bit_pointers
);
1518 static inline struct radv_descriptor_state
*
1519 radv_get_descriptors_state(struct radv_cmd_buffer
*cmd_buffer
,
1520 VkPipelineBindPoint bind_point
)
1522 assert(bind_point
== VK_PIPELINE_BIND_POINT_GRAPHICS
||
1523 bind_point
== VK_PIPELINE_BIND_POINT_COMPUTE
);
1524 return &cmd_buffer
->descriptors
[bind_point
];
1528 * Takes x,y,z as exact numbers of invocations, instead of blocks.
1530 * Limitations: Can't call normal dispatch functions without binding or rebinding
1531 * the compute pipeline.
1533 void radv_unaligned_dispatch(
1534 struct radv_cmd_buffer
*cmd_buffer
,
1540 struct vk_object_base base
;
1541 struct radeon_winsys_bo
*bo
;
1545 struct radv_shader_module
;
1547 #define RADV_HASH_SHADER_NO_NGG (1 << 0)
1548 #define RADV_HASH_SHADER_CS_WAVE32 (1 << 1)
1549 #define RADV_HASH_SHADER_PS_WAVE32 (1 << 2)
1550 #define RADV_HASH_SHADER_GE_WAVE32 (1 << 3)
1551 #define RADV_HASH_SHADER_LLVM (1 << 4)
1554 radv_hash_shaders(unsigned char *hash
,
1555 const VkPipelineShaderStageCreateInfo
**stages
,
1556 const struct radv_pipeline_layout
*layout
,
1557 const struct radv_pipeline_key
*key
,
1560 static inline gl_shader_stage
1561 vk_to_mesa_shader_stage(VkShaderStageFlagBits vk_stage
)
1563 assert(__builtin_popcount(vk_stage
) == 1);
1564 return ffs(vk_stage
) - 1;
1567 static inline VkShaderStageFlagBits
1568 mesa_to_vk_shader_stage(gl_shader_stage mesa_stage
)
1570 return (1 << mesa_stage
);
1573 #define RADV_STAGE_MASK ((1 << MESA_SHADER_STAGES) - 1)
1575 #define radv_foreach_stage(stage, stage_bits) \
1576 for (gl_shader_stage stage, \
1577 __tmp = (gl_shader_stage)((stage_bits) & RADV_STAGE_MASK); \
1578 stage = __builtin_ffs(__tmp) - 1, __tmp; \
1579 __tmp &= ~(1 << (stage)))
1581 extern const VkFormat radv_fs_key_format_exemplars
[NUM_META_FS_KEYS
];
1582 unsigned radv_format_meta_fs_key(VkFormat format
);
1584 struct radv_multisample_state
{
1586 uint32_t pa_sc_line_cntl
;
1587 uint32_t pa_sc_mode_cntl_0
;
1588 uint32_t pa_sc_mode_cntl_1
;
1589 uint32_t pa_sc_aa_config
;
1590 uint32_t pa_sc_aa_mask
[2];
1591 unsigned num_samples
;
1594 struct radv_prim_vertex_count
{
1599 struct radv_vertex_elements_info
{
1600 uint32_t format_size
[MAX_VERTEX_ATTRIBS
];
1603 struct radv_ia_multi_vgt_param_helpers
{
1605 bool partial_es_wave
;
1606 uint8_t primgroup_size
;
1607 bool ia_switch_on_eoi
;
1608 bool partial_vs_wave
;
1611 struct radv_binning_state
{
1612 uint32_t pa_sc_binner_cntl_0
;
1613 uint32_t db_dfsm_control
;
1616 #define SI_GS_PER_ES 128
1618 struct radv_pipeline
{
1619 struct vk_object_base base
;
1620 struct radv_device
* device
;
1621 struct radv_dynamic_state dynamic_state
;
1623 struct radv_pipeline_layout
* layout
;
1625 bool need_indirect_descriptor_sets
;
1626 struct radv_shader_variant
* shaders
[MESA_SHADER_STAGES
];
1627 struct radv_shader_variant
*gs_copy_shader
;
1628 VkShaderStageFlags active_stages
;
1630 struct radeon_cmdbuf cs
;
1631 uint32_t ctx_cs_hash
;
1632 struct radeon_cmdbuf ctx_cs
;
1634 struct radv_vertex_elements_info vertex_elements
;
1636 uint32_t binding_stride
[MAX_VBS
];
1637 uint8_t num_vertex_bindings
;
1639 uint32_t user_data_0
[MESA_SHADER_STAGES
];
1642 struct radv_multisample_state ms
;
1643 struct radv_binning_state binning
;
1644 uint32_t spi_baryc_cntl
;
1645 bool prim_restart_enable
;
1646 unsigned esgs_ring_size
;
1647 unsigned gsvs_ring_size
;
1648 uint32_t vtx_base_sgpr
;
1649 struct radv_ia_multi_vgt_param_helpers ia_multi_vgt_param
;
1650 uint8_t vtx_emit_num
;
1651 bool can_use_guardband
;
1652 uint32_t needed_dynamic_state
;
1653 bool disable_out_of_order_rast_for_occlusion
;
1655 unsigned tess_patch_control_points
;
1656 unsigned pa_su_sc_mode_cntl
;
1658 /* Used for rbplus */
1659 uint32_t col_format
;
1660 uint32_t cb_target_mask
;
1666 unsigned scratch_bytes_per_wave
;
1668 /* Not NULL if graphics pipeline uses streamout. */
1669 struct radv_shader_variant
*streamout_shader
;
1672 static inline bool radv_pipeline_has_gs(const struct radv_pipeline
*pipeline
)
1674 return pipeline
->shaders
[MESA_SHADER_GEOMETRY
] ? true : false;
1677 static inline bool radv_pipeline_has_tess(const struct radv_pipeline
*pipeline
)
1679 return pipeline
->shaders
[MESA_SHADER_TESS_CTRL
] ? true : false;
1682 bool radv_pipeline_has_ngg(const struct radv_pipeline
*pipeline
);
1684 bool radv_pipeline_has_ngg_passthrough(const struct radv_pipeline
*pipeline
);
1686 bool radv_pipeline_has_gs_copy_shader(const struct radv_pipeline
*pipeline
);
1688 struct radv_userdata_info
*radv_lookup_user_sgpr(struct radv_pipeline
*pipeline
,
1689 gl_shader_stage stage
,
1692 struct radv_shader_variant
*radv_get_shader(struct radv_pipeline
*pipeline
,
1693 gl_shader_stage stage
);
1695 struct radv_graphics_pipeline_create_info
{
1697 bool db_depth_clear
;
1698 bool db_stencil_clear
;
1699 bool db_depth_disable_expclear
;
1700 bool db_stencil_disable_expclear
;
1701 bool depth_compress_disable
;
1702 bool stencil_compress_disable
;
1703 bool resummarize_enable
;
1704 uint32_t custom_blend_mode
;
1708 radv_graphics_pipeline_create(VkDevice device
,
1709 VkPipelineCache cache
,
1710 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
1711 const struct radv_graphics_pipeline_create_info
*extra
,
1712 const VkAllocationCallbacks
*alloc
,
1713 VkPipeline
*pPipeline
);
1715 struct radv_binning_settings
{
1716 unsigned context_states_per_bin
; /* allowed range: [1, 6] */
1717 unsigned persistent_states_per_bin
; /* allowed range: [1, 32] */
1718 unsigned fpovs_per_batch
; /* allowed range: [0, 255], 0 = unlimited */
1721 struct radv_binning_settings
1722 radv_get_binning_settings(const struct radv_physical_device
*pdev
);
1724 struct vk_format_description
;
1725 uint32_t radv_translate_buffer_dataformat(const struct vk_format_description
*desc
,
1726 int first_non_void
);
1727 uint32_t radv_translate_buffer_numformat(const struct vk_format_description
*desc
,
1728 int first_non_void
);
1729 bool radv_is_buffer_format_supported(VkFormat format
, bool *scaled
);
1730 uint32_t radv_translate_colorformat(VkFormat format
);
1731 uint32_t radv_translate_color_numformat(VkFormat format
,
1732 const struct vk_format_description
*desc
,
1733 int first_non_void
);
1734 uint32_t radv_colorformat_endian_swap(uint32_t colorformat
);
1735 unsigned radv_translate_colorswap(VkFormat format
, bool do_endian_swap
);
1736 uint32_t radv_translate_dbformat(VkFormat format
);
1737 uint32_t radv_translate_tex_dataformat(VkFormat format
,
1738 const struct vk_format_description
*desc
,
1739 int first_non_void
);
1740 uint32_t radv_translate_tex_numformat(VkFormat format
,
1741 const struct vk_format_description
*desc
,
1742 int first_non_void
);
1743 bool radv_format_pack_clear_color(VkFormat format
,
1744 uint32_t clear_vals
[2],
1745 VkClearColorValue
*value
);
1746 bool radv_is_colorbuffer_format_supported(VkFormat format
, bool *blendable
);
1747 bool radv_dcc_formats_compatible(VkFormat format1
,
1749 bool radv_device_supports_etc(struct radv_physical_device
*physical_device
);
1751 struct radv_image_plane
{
1753 struct radeon_surf surface
;
1758 struct vk_object_base base
;
1760 /* The original VkFormat provided by the client. This may not match any
1761 * of the actual surface formats.
1764 VkImageAspectFlags aspects
;
1765 VkImageUsageFlags usage
; /**< Superset of VkImageCreateInfo::usage. */
1766 struct ac_surf_info info
;
1767 VkImageTiling tiling
; /** VkImageCreateInfo::tiling */
1768 VkImageCreateFlags flags
; /** VkImageCreateInfo::flags */
1773 unsigned queue_family_mask
;
1777 /* Set when bound */
1778 struct radeon_winsys_bo
*bo
;
1779 VkDeviceSize offset
;
1780 bool tc_compatible_htile
;
1781 bool tc_compatible_cmask
;
1783 uint64_t clear_value_offset
;
1784 uint64_t fce_pred_offset
;
1785 uint64_t dcc_pred_offset
;
1788 * Metadata for the TC-compat zrange workaround. If the 32-bit value
1789 * stored at this offset is UINT_MAX, the driver will emit
1790 * DB_Z_INFO.ZRANGE_PRECISION=0, otherwise it will skip the
1791 * SET_CONTEXT_REG packet.
1793 uint64_t tc_compat_zrange_offset
;
1795 /* For VK_ANDROID_native_buffer, the WSI image owns the memory, */
1796 VkDeviceMemory owned_memory
;
1798 unsigned plane_count
;
1799 struct radv_image_plane planes
[0];
1802 /* Whether the image has a htile that is known consistent with the contents of
1803 * the image and is allowed to be in compressed form.
1805 * If this is false reads that don't use the htile should be able to return
1808 bool radv_layout_is_htile_compressed(const struct radv_image
*image
,
1809 VkImageLayout layout
,
1810 bool in_render_loop
,
1811 unsigned queue_mask
);
1813 bool radv_layout_can_fast_clear(const struct radv_image
*image
,
1814 VkImageLayout layout
,
1815 bool in_render_loop
,
1816 unsigned queue_mask
);
1818 bool radv_layout_dcc_compressed(const struct radv_device
*device
,
1819 const struct radv_image
*image
,
1820 VkImageLayout layout
,
1821 bool in_render_loop
,
1822 unsigned queue_mask
);
1825 * Return whether the image has CMASK metadata for color surfaces.
1828 radv_image_has_cmask(const struct radv_image
*image
)
1830 return image
->planes
[0].surface
.cmask_offset
;
1834 * Return whether the image has FMASK metadata for color surfaces.
1837 radv_image_has_fmask(const struct radv_image
*image
)
1839 return image
->planes
[0].surface
.fmask_offset
;
1843 * Return whether the image has DCC metadata for color surfaces.
1846 radv_image_has_dcc(const struct radv_image
*image
)
1848 return image
->planes
[0].surface
.dcc_size
;
1852 * Return whether the image is TC-compatible CMASK.
1855 radv_image_is_tc_compat_cmask(const struct radv_image
*image
)
1857 return radv_image_has_fmask(image
) && image
->tc_compatible_cmask
;
1861 * Return whether DCC metadata is enabled for a level.
1864 radv_dcc_enabled(const struct radv_image
*image
, unsigned level
)
1866 return radv_image_has_dcc(image
) &&
1867 level
< image
->planes
[0].surface
.num_dcc_levels
;
1871 * Return whether the image has CB metadata.
1874 radv_image_has_CB_metadata(const struct radv_image
*image
)
1876 return radv_image_has_cmask(image
) ||
1877 radv_image_has_fmask(image
) ||
1878 radv_image_has_dcc(image
);
1882 * Return whether the image has HTILE metadata for depth surfaces.
1885 radv_image_has_htile(const struct radv_image
*image
)
1887 return image
->planes
[0].surface
.htile_size
;
1891 * Return whether HTILE metadata is enabled for a level.
1894 radv_htile_enabled(const struct radv_image
*image
, unsigned level
)
1896 return radv_image_has_htile(image
) && level
== 0;
1900 * Return whether the image is TC-compatible HTILE.
1903 radv_image_is_tc_compat_htile(const struct radv_image
*image
)
1905 return radv_image_has_htile(image
) && image
->tc_compatible_htile
;
1908 static inline uint64_t
1909 radv_image_get_fast_clear_va(const struct radv_image
*image
,
1910 uint32_t base_level
)
1912 uint64_t va
= radv_buffer_get_va(image
->bo
);
1913 va
+= image
->offset
+ image
->clear_value_offset
+ base_level
* 8;
1917 static inline uint64_t
1918 radv_image_get_fce_pred_va(const struct radv_image
*image
,
1919 uint32_t base_level
)
1921 uint64_t va
= radv_buffer_get_va(image
->bo
);
1922 va
+= image
->offset
+ image
->fce_pred_offset
+ base_level
* 8;
1926 static inline uint64_t
1927 radv_image_get_dcc_pred_va(const struct radv_image
*image
,
1928 uint32_t base_level
)
1930 uint64_t va
= radv_buffer_get_va(image
->bo
);
1931 va
+= image
->offset
+ image
->dcc_pred_offset
+ base_level
* 8;
1935 static inline uint64_t
1936 radv_get_tc_compat_zrange_va(const struct radv_image
*image
,
1937 uint32_t base_level
)
1939 uint64_t va
= radv_buffer_get_va(image
->bo
);
1940 va
+= image
->offset
+ image
->tc_compat_zrange_offset
+ base_level
* 4;
1944 static inline uint64_t
1945 radv_get_ds_clear_value_va(const struct radv_image
*image
,
1946 uint32_t base_level
)
1948 uint64_t va
= radv_buffer_get_va(image
->bo
);
1949 va
+= image
->offset
+ image
->clear_value_offset
+ base_level
* 8;
1953 unsigned radv_image_queue_family_mask(const struct radv_image
*image
, uint32_t family
, uint32_t queue_family
);
1955 static inline uint32_t
1956 radv_get_layerCount(const struct radv_image
*image
,
1957 const VkImageSubresourceRange
*range
)
1959 return range
->layerCount
== VK_REMAINING_ARRAY_LAYERS
?
1960 image
->info
.array_size
- range
->baseArrayLayer
: range
->layerCount
;
1963 static inline uint32_t
1964 radv_get_levelCount(const struct radv_image
*image
,
1965 const VkImageSubresourceRange
*range
)
1967 return range
->levelCount
== VK_REMAINING_MIP_LEVELS
?
1968 image
->info
.levels
- range
->baseMipLevel
: range
->levelCount
;
1971 struct radeon_bo_metadata
;
1973 radv_init_metadata(struct radv_device
*device
,
1974 struct radv_image
*image
,
1975 struct radeon_bo_metadata
*metadata
);
1978 radv_image_override_offset_stride(struct radv_device
*device
,
1979 struct radv_image
*image
,
1980 uint64_t offset
, uint32_t stride
);
1982 union radv_descriptor
{
1984 uint32_t plane0_descriptor
[8];
1985 uint32_t fmask_descriptor
[8];
1988 uint32_t plane_descriptors
[3][8];
1992 struct radv_image_view
{
1993 struct vk_object_base base
;
1994 struct radv_image
*image
; /**< VkImageViewCreateInfo::image */
1995 struct radeon_winsys_bo
*bo
;
1997 VkImageViewType type
;
1998 VkImageAspectFlags aspect_mask
;
2001 bool multiple_planes
;
2002 uint32_t base_layer
;
2003 uint32_t layer_count
;
2005 uint32_t level_count
;
2006 VkExtent3D extent
; /**< Extent of VkImageViewCreateInfo::baseMipLevel. */
2008 union radv_descriptor descriptor
;
2010 /* Descriptor for use as a storage image as opposed to a sampled image.
2011 * This has a few differences for cube maps (e.g. type).
2013 union radv_descriptor storage_descriptor
;
2016 struct radv_image_create_info
{
2017 const VkImageCreateInfo
*vk_info
;
2019 bool no_metadata_planes
;
2020 const struct radeon_bo_metadata
*bo_metadata
;
2024 radv_image_create_layout(struct radv_device
*device
,
2025 struct radv_image_create_info create_info
,
2026 struct radv_image
*image
);
2028 VkResult
radv_image_create(VkDevice _device
,
2029 const struct radv_image_create_info
*info
,
2030 const VkAllocationCallbacks
* alloc
,
2033 bool vi_alpha_is_on_msb(struct radv_device
*device
, VkFormat format
);
2036 radv_image_from_gralloc(VkDevice device_h
,
2037 const VkImageCreateInfo
*base_info
,
2038 const VkNativeBufferANDROID
*gralloc_info
,
2039 const VkAllocationCallbacks
*alloc
,
2040 VkImage
*out_image_h
);
2042 radv_ahb_usage_from_vk_usage(const VkImageCreateFlags vk_create
,
2043 const VkImageUsageFlags vk_usage
);
2045 radv_import_ahb_memory(struct radv_device
*device
,
2046 struct radv_device_memory
*mem
,
2048 const VkImportAndroidHardwareBufferInfoANDROID
*info
);
2050 radv_create_ahb_memory(struct radv_device
*device
,
2051 struct radv_device_memory
*mem
,
2053 const VkMemoryAllocateInfo
*pAllocateInfo
);
2056 radv_select_android_external_format(const void *next
, VkFormat default_format
);
2058 bool radv_android_gralloc_supports_format(VkFormat format
, VkImageUsageFlagBits usage
);
2060 struct radv_image_view_extra_create_info
{
2061 bool disable_compression
;
2064 void radv_image_view_init(struct radv_image_view
*view
,
2065 struct radv_device
*device
,
2066 const VkImageViewCreateInfo
*pCreateInfo
,
2067 const struct radv_image_view_extra_create_info
* extra_create_info
);
2069 VkFormat
radv_get_aspect_format(struct radv_image
*image
, VkImageAspectFlags mask
);
2071 struct radv_sampler_ycbcr_conversion
{
2072 struct vk_object_base base
;
2074 VkSamplerYcbcrModelConversion ycbcr_model
;
2075 VkSamplerYcbcrRange ycbcr_range
;
2076 VkComponentMapping components
;
2077 VkChromaLocation chroma_offsets
[2];
2078 VkFilter chroma_filter
;
2081 struct radv_buffer_view
{
2082 struct vk_object_base base
;
2083 struct radeon_winsys_bo
*bo
;
2085 uint64_t range
; /**< VkBufferViewCreateInfo::range */
2088 void radv_buffer_view_init(struct radv_buffer_view
*view
,
2089 struct radv_device
*device
,
2090 const VkBufferViewCreateInfo
* pCreateInfo
);
2092 static inline struct VkExtent3D
2093 radv_sanitize_image_extent(const VkImageType imageType
,
2094 const struct VkExtent3D imageExtent
)
2096 switch (imageType
) {
2097 case VK_IMAGE_TYPE_1D
:
2098 return (VkExtent3D
) { imageExtent
.width
, 1, 1 };
2099 case VK_IMAGE_TYPE_2D
:
2100 return (VkExtent3D
) { imageExtent
.width
, imageExtent
.height
, 1 };
2101 case VK_IMAGE_TYPE_3D
:
2104 unreachable("invalid image type");
2108 static inline struct VkOffset3D
2109 radv_sanitize_image_offset(const VkImageType imageType
,
2110 const struct VkOffset3D imageOffset
)
2112 switch (imageType
) {
2113 case VK_IMAGE_TYPE_1D
:
2114 return (VkOffset3D
) { imageOffset
.x
, 0, 0 };
2115 case VK_IMAGE_TYPE_2D
:
2116 return (VkOffset3D
) { imageOffset
.x
, imageOffset
.y
, 0 };
2117 case VK_IMAGE_TYPE_3D
:
2120 unreachable("invalid image type");
2125 radv_image_extent_compare(const struct radv_image
*image
,
2126 const VkExtent3D
*extent
)
2128 if (extent
->width
!= image
->info
.width
||
2129 extent
->height
!= image
->info
.height
||
2130 extent
->depth
!= image
->info
.depth
)
2135 struct radv_sampler
{
2136 struct vk_object_base base
;
2138 struct radv_sampler_ycbcr_conversion
*ycbcr_sampler
;
2139 uint32_t border_color_slot
;
2142 struct radv_framebuffer
{
2143 struct vk_object_base base
;
2148 uint32_t attachment_count
;
2149 struct radv_image_view
*attachments
[0];
2152 struct radv_subpass_barrier
{
2153 VkPipelineStageFlags src_stage_mask
;
2154 VkAccessFlags src_access_mask
;
2155 VkAccessFlags dst_access_mask
;
2158 void radv_subpass_barrier(struct radv_cmd_buffer
*cmd_buffer
,
2159 const struct radv_subpass_barrier
*barrier
);
2161 struct radv_subpass_attachment
{
2162 uint32_t attachment
;
2163 VkImageLayout layout
;
2164 VkImageLayout stencil_layout
;
2165 bool in_render_loop
;
2168 struct radv_subpass
{
2169 uint32_t attachment_count
;
2170 struct radv_subpass_attachment
* attachments
;
2172 uint32_t input_count
;
2173 uint32_t color_count
;
2174 struct radv_subpass_attachment
* input_attachments
;
2175 struct radv_subpass_attachment
* color_attachments
;
2176 struct radv_subpass_attachment
* resolve_attachments
;
2177 struct radv_subpass_attachment
* depth_stencil_attachment
;
2178 struct radv_subpass_attachment
* ds_resolve_attachment
;
2179 VkResolveModeFlagBits depth_resolve_mode
;
2180 VkResolveModeFlagBits stencil_resolve_mode
;
2182 /** Subpass has at least one color resolve attachment */
2183 bool has_color_resolve
;
2185 /** Subpass has at least one color attachment */
2188 struct radv_subpass_barrier start_barrier
;
2192 VkSampleCountFlagBits color_sample_count
;
2193 VkSampleCountFlagBits depth_sample_count
;
2194 VkSampleCountFlagBits max_sample_count
;
2198 radv_get_subpass_id(struct radv_cmd_buffer
*cmd_buffer
);
2200 struct radv_render_pass_attachment
{
2203 VkAttachmentLoadOp load_op
;
2204 VkAttachmentLoadOp stencil_load_op
;
2205 VkImageLayout initial_layout
;
2206 VkImageLayout final_layout
;
2207 VkImageLayout stencil_initial_layout
;
2208 VkImageLayout stencil_final_layout
;
2210 /* The subpass id in which the attachment will be used first/last. */
2211 uint32_t first_subpass_idx
;
2212 uint32_t last_subpass_idx
;
2215 struct radv_render_pass
{
2216 struct vk_object_base base
;
2217 uint32_t attachment_count
;
2218 uint32_t subpass_count
;
2219 struct radv_subpass_attachment
* subpass_attachments
;
2220 struct radv_render_pass_attachment
* attachments
;
2221 struct radv_subpass_barrier end_barrier
;
2222 struct radv_subpass subpasses
[0];
2225 VkResult
radv_device_init_meta(struct radv_device
*device
);
2226 void radv_device_finish_meta(struct radv_device
*device
);
2228 struct radv_query_pool
{
2229 struct vk_object_base base
;
2230 struct radeon_winsys_bo
*bo
;
2232 uint32_t availability_offset
;
2236 uint32_t pipeline_stats_mask
;
2240 RADV_SEMAPHORE_NONE
,
2241 RADV_SEMAPHORE_WINSYS
,
2242 RADV_SEMAPHORE_SYNCOBJ
,
2243 RADV_SEMAPHORE_TIMELINE
,
2244 } radv_semaphore_kind
;
2246 struct radv_deferred_queue_submission
;
2248 struct radv_timeline_waiter
{
2249 struct list_head list
;
2250 struct radv_deferred_queue_submission
*submission
;
2254 struct radv_timeline_point
{
2255 struct list_head list
;
2260 /* Separate from the list to accomodate CPU wait being async, as well
2261 * as prevent point deletion during submission. */
2262 unsigned wait_count
;
2265 struct radv_timeline
{
2266 /* Using a pthread mutex to be compatible with condition variables. */
2267 pthread_mutex_t mutex
;
2269 uint64_t highest_signaled
;
2270 uint64_t highest_submitted
;
2272 struct list_head points
;
2274 /* Keep free points on hand so we do not have to recreate syncobjs all
2276 struct list_head free_points
;
2278 /* Submissions that are deferred waiting for a specific value to be
2280 struct list_head waiters
;
2283 struct radv_semaphore_part
{
2284 radv_semaphore_kind kind
;
2287 struct radeon_winsys_sem
*ws_sem
;
2288 struct radv_timeline timeline
;
2292 struct radv_semaphore
{
2293 struct vk_object_base base
;
2294 struct radv_semaphore_part permanent
;
2295 struct radv_semaphore_part temporary
;
2298 bool radv_queue_internal_submit(struct radv_queue
*queue
,
2299 struct radeon_cmdbuf
*cs
);
2301 void radv_set_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
2302 VkPipelineBindPoint bind_point
,
2303 struct radv_descriptor_set
*set
,
2307 radv_update_descriptor_sets(struct radv_device
*device
,
2308 struct radv_cmd_buffer
*cmd_buffer
,
2309 VkDescriptorSet overrideSet
,
2310 uint32_t descriptorWriteCount
,
2311 const VkWriteDescriptorSet
*pDescriptorWrites
,
2312 uint32_t descriptorCopyCount
,
2313 const VkCopyDescriptorSet
*pDescriptorCopies
);
2316 radv_update_descriptor_set_with_template(struct radv_device
*device
,
2317 struct radv_cmd_buffer
*cmd_buffer
,
2318 struct radv_descriptor_set
*set
,
2319 VkDescriptorUpdateTemplate descriptorUpdateTemplate
,
2322 void radv_meta_push_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
2323 VkPipelineBindPoint pipelineBindPoint
,
2324 VkPipelineLayout _layout
,
2326 uint32_t descriptorWriteCount
,
2327 const VkWriteDescriptorSet
*pDescriptorWrites
);
2329 void radv_initialize_dcc(struct radv_cmd_buffer
*cmd_buffer
,
2330 struct radv_image
*image
,
2331 const VkImageSubresourceRange
*range
, uint32_t value
);
2333 void radv_initialize_fmask(struct radv_cmd_buffer
*cmd_buffer
,
2334 struct radv_image
*image
,
2335 const VkImageSubresourceRange
*range
);
2338 struct vk_object_base base
;
2339 struct radeon_winsys_fence
*fence
;
2340 struct wsi_fence
*fence_wsi
;
2343 uint32_t temp_syncobj
;
2346 /* radv_nir_to_llvm.c */
2347 struct radv_shader_args
;
2349 void llvm_compile_shader(struct radv_device
*device
,
2350 unsigned shader_count
,
2351 struct nir_shader
*const *shaders
,
2352 struct radv_shader_binary
**binary
,
2353 struct radv_shader_args
*args
);
2355 unsigned radv_nir_get_max_workgroup_size(enum chip_class chip_class
,
2356 gl_shader_stage stage
,
2357 const struct nir_shader
*nir
);
2359 /* radv_shader_info.h */
2360 struct radv_shader_info
;
2361 struct radv_shader_variant_key
;
2363 void radv_nir_shader_info_pass(const struct nir_shader
*nir
,
2364 const struct radv_pipeline_layout
*layout
,
2365 const struct radv_shader_variant_key
*key
,
2366 struct radv_shader_info
*info
,
2369 void radv_nir_shader_info_init(struct radv_shader_info
*info
);
2372 struct radv_thread_trace_info
{
2373 uint32_t cur_offset
;
2374 uint32_t trace_status
;
2376 uint32_t gfx9_write_counter
;
2377 uint32_t gfx10_dropped_cntr
;
2381 struct radv_thread_trace_se
{
2382 struct radv_thread_trace_info info
;
2384 uint32_t shader_engine
;
2385 uint32_t compute_unit
;
2388 struct radv_thread_trace
{
2389 uint32_t num_traces
;
2390 struct radv_thread_trace_se traces
[4];
2393 bool radv_thread_trace_init(struct radv_device
*device
);
2394 void radv_thread_trace_finish(struct radv_device
*device
);
2395 bool radv_begin_thread_trace(struct radv_queue
*queue
);
2396 bool radv_end_thread_trace(struct radv_queue
*queue
);
2397 bool radv_get_thread_trace(struct radv_queue
*queue
,
2398 struct radv_thread_trace
*thread_trace
);
2399 void radv_emit_thread_trace_userdata(struct radeon_cmdbuf
*cs
,
2400 const void *data
, uint32_t num_dwords
);
2403 int radv_dump_thread_trace(struct radv_device
*device
,
2404 const struct radv_thread_trace
*trace
);
2406 /* radv_sqtt_layer_.c */
2407 struct radv_barrier_data
{
2410 uint16_t depth_stencil_expand
: 1;
2411 uint16_t htile_hiz_range_expand
: 1;
2412 uint16_t depth_stencil_resummarize
: 1;
2413 uint16_t dcc_decompress
: 1;
2414 uint16_t fmask_decompress
: 1;
2415 uint16_t fast_clear_eliminate
: 1;
2416 uint16_t fmask_color_expand
: 1;
2417 uint16_t init_mask_ram
: 1;
2418 uint16_t reserved
: 8;
2421 } layout_transitions
;
2425 * Value for the reason field of an RGP barrier start marker originating from
2426 * the Vulkan client (does not include PAL-defined values). (Table 15)
2428 enum rgp_barrier_reason
{
2429 RGP_BARRIER_UNKNOWN_REASON
= 0xFFFFFFFF,
2431 /* External app-generated barrier reasons, i.e. API synchronization
2432 * commands Range of valid values: [0x00000001 ... 0x7FFFFFFF].
2434 RGP_BARRIER_EXTERNAL_CMD_PIPELINE_BARRIER
= 0x00000001,
2435 RGP_BARRIER_EXTERNAL_RENDER_PASS_SYNC
= 0x00000002,
2436 RGP_BARRIER_EXTERNAL_CMD_WAIT_EVENTS
= 0x00000003,
2438 /* Internal barrier reasons, i.e. implicit synchronization inserted by
2439 * the Vulkan driver Range of valid values: [0xC0000000 ... 0xFFFFFFFE].
2441 RGP_BARRIER_INTERNAL_BASE
= 0xC0000000,
2442 RGP_BARRIER_INTERNAL_PRE_RESET_QUERY_POOL_SYNC
= RGP_BARRIER_INTERNAL_BASE
+ 0,
2443 RGP_BARRIER_INTERNAL_POST_RESET_QUERY_POOL_SYNC
= RGP_BARRIER_INTERNAL_BASE
+ 1,
2444 RGP_BARRIER_INTERNAL_GPU_EVENT_RECYCLE_STALL
= RGP_BARRIER_INTERNAL_BASE
+ 2,
2445 RGP_BARRIER_INTERNAL_PRE_COPY_QUERY_POOL_RESULTS_SYNC
= RGP_BARRIER_INTERNAL_BASE
+ 3
2448 void radv_describe_begin_cmd_buffer(struct radv_cmd_buffer
*cmd_buffer
);
2449 void radv_describe_end_cmd_buffer(struct radv_cmd_buffer
*cmd_buffer
);
2450 void radv_describe_draw(struct radv_cmd_buffer
*cmd_buffer
);
2451 void radv_describe_dispatch(struct radv_cmd_buffer
*cmd_buffer
, int x
, int y
, int z
);
2452 void radv_describe_begin_render_pass_clear(struct radv_cmd_buffer
*cmd_buffer
,
2453 VkImageAspectFlagBits aspects
);
2454 void radv_describe_end_render_pass_clear(struct radv_cmd_buffer
*cmd_buffer
);
2455 void radv_describe_barrier_start(struct radv_cmd_buffer
*cmd_buffer
,
2456 enum rgp_barrier_reason reason
);
2457 void radv_describe_barrier_end(struct radv_cmd_buffer
*cmd_buffer
);
2458 void radv_describe_layout_transition(struct radv_cmd_buffer
*cmd_buffer
,
2459 const struct radv_barrier_data
*barrier
);
2461 struct radeon_winsys_sem
;
2463 uint64_t radv_get_current_time(void);
2465 static inline uint32_t
2466 si_conv_gl_prim_to_vertices(unsigned gl_prim
)
2469 case 0: /* GL_POINTS */
2471 case 1: /* GL_LINES */
2472 case 3: /* GL_LINE_STRIP */
2474 case 4: /* GL_TRIANGLES */
2475 case 5: /* GL_TRIANGLE_STRIP */
2477 case 0xA: /* GL_LINE_STRIP_ADJACENCY_ARB */
2479 case 0xc: /* GL_TRIANGLES_ADJACENCY_ARB */
2481 case 7: /* GL_QUADS */
2482 return V_028A6C_OUTPRIM_TYPE_TRISTRIP
;
2489 void radv_cmd_buffer_begin_render_pass(struct radv_cmd_buffer
*cmd_buffer
,
2490 const VkRenderPassBeginInfo
*pRenderPassBegin
);
2491 void radv_cmd_buffer_end_render_pass(struct radv_cmd_buffer
*cmd_buffer
);
2493 #define RADV_DEFINE_HANDLE_CASTS(__radv_type, __VkType) \
2495 static inline struct __radv_type * \
2496 __radv_type ## _from_handle(__VkType _handle) \
2498 return (struct __radv_type *) _handle; \
2501 static inline __VkType \
2502 __radv_type ## _to_handle(struct __radv_type *_obj) \
2504 return (__VkType) _obj; \
2507 #define RADV_DEFINE_NONDISP_HANDLE_CASTS(__radv_type, __VkType) \
2509 static inline struct __radv_type * \
2510 __radv_type ## _from_handle(__VkType _handle) \
2512 return (struct __radv_type *)(uintptr_t) _handle; \
2515 static inline __VkType \
2516 __radv_type ## _to_handle(struct __radv_type *_obj) \
2518 return (__VkType)(uintptr_t) _obj; \
2521 #define RADV_FROM_HANDLE(__radv_type, __name, __handle) \
2522 struct __radv_type *__name = __radv_type ## _from_handle(__handle)
2524 RADV_DEFINE_HANDLE_CASTS(radv_cmd_buffer
, VkCommandBuffer
)
2525 RADV_DEFINE_HANDLE_CASTS(radv_device
, VkDevice
)
2526 RADV_DEFINE_HANDLE_CASTS(radv_instance
, VkInstance
)
2527 RADV_DEFINE_HANDLE_CASTS(radv_physical_device
, VkPhysicalDevice
)
2528 RADV_DEFINE_HANDLE_CASTS(radv_queue
, VkQueue
)
2530 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_cmd_pool
, VkCommandPool
)
2531 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer
, VkBuffer
)
2532 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer_view
, VkBufferView
)
2533 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_pool
, VkDescriptorPool
)
2534 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set
, VkDescriptorSet
)
2535 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set_layout
, VkDescriptorSetLayout
)
2536 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_update_template
, VkDescriptorUpdateTemplate
)
2537 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_device_memory
, VkDeviceMemory
)
2538 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_fence
, VkFence
)
2539 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_event
, VkEvent
)
2540 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_framebuffer
, VkFramebuffer
)
2541 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image
, VkImage
)
2542 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image_view
, VkImageView
);
2543 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_cache
, VkPipelineCache
)
2544 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline
, VkPipeline
)
2545 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_layout
, VkPipelineLayout
)
2546 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_query_pool
, VkQueryPool
)
2547 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_render_pass
, VkRenderPass
)
2548 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_sampler
, VkSampler
)
2549 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_sampler_ycbcr_conversion
, VkSamplerYcbcrConversion
)
2550 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_shader_module
, VkShaderModule
)
2551 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_semaphore
, VkSemaphore
)
2553 #endif /* RADV_PRIVATE_H */