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
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"
52 #include "main/macros.h"
54 #include "vk_debug_report.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_descriptor_set.h"
64 #include "radv_extensions.h"
67 #include <llvm-c/TargetMachine.h>
69 /* Pre-declarations needed for WSI entrypoints */
72 typedef struct xcb_connection_t xcb_connection_t
;
73 typedef uint32_t xcb_visualid_t
;
74 typedef uint32_t xcb_window_t
;
76 #include <vulkan/vulkan.h>
77 #include <vulkan/vulkan_intel.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 #define ATI_VENDOR_ID 0x1002
89 #define MAX_VERTEX_ATTRIBS 32
91 #define MAX_VIEWPORTS 16
92 #define MAX_SCISSORS 16
93 #define MAX_DISCARD_RECTANGLES 4
94 #define MAX_SAMPLE_LOCATIONS 32
95 #define MAX_PUSH_CONSTANTS_SIZE 128
96 #define MAX_PUSH_DESCRIPTORS 32
97 #define MAX_DYNAMIC_UNIFORM_BUFFERS 16
98 #define MAX_DYNAMIC_STORAGE_BUFFERS 8
99 #define MAX_DYNAMIC_BUFFERS (MAX_DYNAMIC_UNIFORM_BUFFERS + MAX_DYNAMIC_STORAGE_BUFFERS)
100 #define MAX_SAMPLES_LOG2 4
101 #define NUM_META_FS_KEYS 12
102 #define RADV_MAX_DRM_DEVICES 8
104 #define MAX_SO_STREAMS 4
105 #define MAX_SO_BUFFERS 4
106 #define MAX_SO_OUTPUTS 64
107 #define MAX_INLINE_UNIFORM_BLOCK_SIZE (4ull * 1024 * 1024)
108 #define MAX_INLINE_UNIFORM_BLOCK_COUNT 64
110 #define NUM_DEPTH_CLEAR_PIPELINES 3
113 * This is the point we switch from using CP to compute shader
114 * for certain buffer operations.
116 #define RADV_BUFFER_OPS_CS_THRESHOLD 4096
118 #define RADV_BUFFER_UPDATE_THRESHOLD 1024
122 RADV_MEM_HEAP_VRAM_CPU_ACCESS
,
129 RADV_MEM_TYPE_GTT_WRITE_COMBINE
,
130 RADV_MEM_TYPE_VRAM_CPU_ACCESS
,
131 RADV_MEM_TYPE_GTT_CACHED
,
135 #define radv_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
137 static inline uint32_t
138 align_u32(uint32_t v
, uint32_t a
)
140 assert(a
!= 0 && a
== (a
& -a
));
141 return (v
+ a
- 1) & ~(a
- 1);
144 static inline uint32_t
145 align_u32_npot(uint32_t v
, uint32_t a
)
147 return (v
+ a
- 1) / a
* a
;
150 static inline uint64_t
151 align_u64(uint64_t v
, uint64_t a
)
153 assert(a
!= 0 && a
== (a
& -a
));
154 return (v
+ a
- 1) & ~(a
- 1);
157 static inline int32_t
158 align_i32(int32_t v
, int32_t a
)
160 assert(a
!= 0 && a
== (a
& -a
));
161 return (v
+ a
- 1) & ~(a
- 1);
164 /** Alignment must be a power of 2. */
166 radv_is_aligned(uintmax_t n
, uintmax_t a
)
168 assert(a
== (a
& -a
));
169 return (n
& (a
- 1)) == 0;
172 static inline uint32_t
173 round_up_u32(uint32_t v
, uint32_t a
)
175 return (v
+ a
- 1) / a
;
178 static inline uint64_t
179 round_up_u64(uint64_t v
, uint64_t a
)
181 return (v
+ a
- 1) / a
;
184 static inline uint32_t
185 radv_minify(uint32_t n
, uint32_t levels
)
187 if (unlikely(n
== 0))
190 return MAX2(n
>> levels
, 1);
193 radv_clamp_f(float f
, float min
, float max
)
206 radv_clear_mask(uint32_t *inout_mask
, uint32_t clear_mask
)
208 if (*inout_mask
& clear_mask
) {
209 *inout_mask
&= ~clear_mask
;
216 #define for_each_bit(b, dword) \
217 for (uint32_t __dword = (dword); \
218 (b) = __builtin_ffs(__dword) - 1, __dword; \
219 __dword &= ~(1 << (b)))
221 #define typed_memcpy(dest, src, count) ({ \
222 STATIC_ASSERT(sizeof(*src) == sizeof(*dest)); \
223 memcpy((dest), (src), (count) * sizeof(*(src))); \
226 /* Whenever we generate an error, pass it through this function. Useful for
227 * debugging, where we can break on it. Only call at error site, not when
228 * propagating errors. Might be useful to plug in a stack trace here.
231 struct radv_instance
;
233 VkResult
__vk_errorf(struct radv_instance
*instance
, VkResult error
, const char *file
, int line
, const char *format
, ...);
235 #define vk_error(instance, error) __vk_errorf(instance, error, __FILE__, __LINE__, NULL);
236 #define vk_errorf(instance, error, format, ...) __vk_errorf(instance, error, __FILE__, __LINE__, format, ## __VA_ARGS__);
238 void __radv_finishme(const char *file
, int line
, const char *format
, ...)
239 radv_printflike(3, 4);
240 void radv_loge(const char *format
, ...) radv_printflike(1, 2);
241 void radv_loge_v(const char *format
, va_list va
);
242 void radv_logi(const char *format
, ...) radv_printflike(1, 2);
243 void radv_logi_v(const char *format
, va_list va
);
246 * Print a FINISHME message, including its source location.
248 #define radv_finishme(format, ...) \
250 static bool reported = false; \
252 __radv_finishme(__FILE__, __LINE__, format, ##__VA_ARGS__); \
257 /* A non-fatal assert. Useful for debugging. */
259 #define radv_assert(x) ({ \
260 if (unlikely(!(x))) \
261 fprintf(stderr, "%s:%d ASSERT: %s\n", __FILE__, __LINE__, #x); \
264 #define radv_assert(x)
267 #define stub_return(v) \
269 radv_finishme("stub %s", __func__); \
275 radv_finishme("stub %s", __func__); \
279 void *radv_lookup_entrypoint_unchecked(const char *name
);
280 void *radv_lookup_entrypoint_checked(const char *name
,
281 uint32_t core_version
,
282 const struct radv_instance_extension_table
*instance
,
283 const struct radv_device_extension_table
*device
);
284 void *radv_lookup_physical_device_entrypoint_checked(const char *name
,
285 uint32_t core_version
,
286 const struct radv_instance_extension_table
*instance
);
288 struct radv_physical_device
{
289 VK_LOADER_DATA _loader_data
;
291 struct radv_instance
* instance
;
293 struct radeon_winsys
*ws
;
294 struct radeon_info rad_info
;
295 char name
[VK_MAX_PHYSICAL_DEVICE_NAME_SIZE
];
296 uint8_t driver_uuid
[VK_UUID_SIZE
];
297 uint8_t device_uuid
[VK_UUID_SIZE
];
298 uint8_t cache_uuid
[VK_UUID_SIZE
];
302 struct wsi_device wsi_device
;
304 bool has_rbplus
; /* if RB+ register exist */
305 bool rbplus_allowed
; /* if RB+ is allowed */
306 bool has_clear_state
;
307 bool cpdma_prefetch_writes_memory
;
308 bool has_scissor_bug
;
310 bool has_out_of_order_rast
;
311 bool out_of_order_rast_allowed
;
313 /* Whether DCC should be enabled for MSAA textures. */
314 bool dcc_msaa_allowed
;
316 /* Whether LOAD_CONTEXT_REG packets are supported. */
317 bool has_load_ctx_reg_pkt
;
319 /* This is the drivers on-disk cache used as a fallback as opposed to
320 * the pipeline cache defined by apps.
322 struct disk_cache
* disk_cache
;
324 VkPhysicalDeviceMemoryProperties memory_properties
;
325 enum radv_mem_type mem_type_indices
[RADV_MEM_TYPE_COUNT
];
327 drmPciBusInfo bus_info
;
329 struct radv_device_extension_table supported_extensions
;
332 struct radv_instance
{
333 VK_LOADER_DATA _loader_data
;
335 VkAllocationCallbacks alloc
;
338 int physicalDeviceCount
;
339 struct radv_physical_device physicalDevices
[RADV_MAX_DRM_DEVICES
];
341 uint64_t debug_flags
;
342 uint64_t perftest_flags
;
344 struct vk_debug_report_instance debug_report_callbacks
;
346 struct radv_instance_extension_table enabled_extensions
;
348 struct driOptionCache dri_options
;
349 struct driOptionCache available_dri_options
;
352 VkResult
radv_init_wsi(struct radv_physical_device
*physical_device
);
353 void radv_finish_wsi(struct radv_physical_device
*physical_device
);
355 bool radv_instance_extension_supported(const char *name
);
356 uint32_t radv_physical_device_api_version(struct radv_physical_device
*dev
);
357 bool radv_physical_device_extension_supported(struct radv_physical_device
*dev
,
362 struct radv_pipeline_cache
{
363 struct radv_device
* device
;
364 pthread_mutex_t mutex
;
368 uint32_t kernel_count
;
369 struct cache_entry
** hash_table
;
372 VkAllocationCallbacks alloc
;
375 struct radv_pipeline_key
{
376 uint32_t instance_rate_inputs
;
377 uint32_t instance_rate_divisors
[MAX_VERTEX_ATTRIBS
];
378 uint8_t vertex_attribute_formats
[MAX_VERTEX_ATTRIBS
];
379 uint32_t vertex_attribute_bindings
[MAX_VERTEX_ATTRIBS
];
380 uint32_t vertex_attribute_offsets
[MAX_VERTEX_ATTRIBS
];
381 uint32_t vertex_attribute_strides
[MAX_VERTEX_ATTRIBS
];
382 uint64_t vertex_alpha_adjust
;
383 uint32_t vertex_post_shuffle
;
384 unsigned tess_input_vertices
;
388 uint8_t log2_ps_iter_samples
;
390 uint32_t has_multiview_view_index
: 1;
391 uint32_t optimisations_disabled
: 1;
395 radv_pipeline_cache_init(struct radv_pipeline_cache
*cache
,
396 struct radv_device
*device
);
398 radv_pipeline_cache_finish(struct radv_pipeline_cache
*cache
);
400 radv_pipeline_cache_load(struct radv_pipeline_cache
*cache
,
401 const void *data
, size_t size
);
403 struct radv_shader_variant
;
406 radv_create_shader_variants_from_pipeline_cache(struct radv_device
*device
,
407 struct radv_pipeline_cache
*cache
,
408 const unsigned char *sha1
,
409 struct radv_shader_variant
**variants
,
410 bool *found_in_application_cache
);
413 radv_pipeline_cache_insert_shaders(struct radv_device
*device
,
414 struct radv_pipeline_cache
*cache
,
415 const unsigned char *sha1
,
416 struct radv_shader_variant
**variants
,
417 const void *const *codes
,
418 const unsigned *code_sizes
);
420 enum radv_blit_ds_layout
{
421 RADV_BLIT_DS_LAYOUT_TILE_ENABLE
,
422 RADV_BLIT_DS_LAYOUT_TILE_DISABLE
,
423 RADV_BLIT_DS_LAYOUT_COUNT
,
426 static inline enum radv_blit_ds_layout
radv_meta_blit_ds_to_type(VkImageLayout layout
)
428 return (layout
== VK_IMAGE_LAYOUT_GENERAL
) ? RADV_BLIT_DS_LAYOUT_TILE_DISABLE
: RADV_BLIT_DS_LAYOUT_TILE_ENABLE
;
431 static inline VkImageLayout
radv_meta_blit_ds_to_layout(enum radv_blit_ds_layout ds_layout
)
433 return ds_layout
== RADV_BLIT_DS_LAYOUT_TILE_ENABLE
? VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
: VK_IMAGE_LAYOUT_GENERAL
;
436 enum radv_meta_dst_layout
{
437 RADV_META_DST_LAYOUT_GENERAL
,
438 RADV_META_DST_LAYOUT_OPTIMAL
,
439 RADV_META_DST_LAYOUT_COUNT
,
442 static inline enum radv_meta_dst_layout
radv_meta_dst_layout_from_layout(VkImageLayout layout
)
444 return (layout
== VK_IMAGE_LAYOUT_GENERAL
) ? RADV_META_DST_LAYOUT_GENERAL
: RADV_META_DST_LAYOUT_OPTIMAL
;
447 static inline VkImageLayout
radv_meta_dst_layout_to_layout(enum radv_meta_dst_layout layout
)
449 return layout
== RADV_META_DST_LAYOUT_OPTIMAL
? VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
: VK_IMAGE_LAYOUT_GENERAL
;
452 struct radv_meta_state
{
453 VkAllocationCallbacks alloc
;
455 struct radv_pipeline_cache cache
;
458 * For on-demand pipeline creation, makes sure that
459 * only one thread tries to build a pipeline at the same time.
464 * Use array element `i` for images with `2^i` samples.
467 VkRenderPass render_pass
[NUM_META_FS_KEYS
];
468 VkPipeline color_pipelines
[NUM_META_FS_KEYS
];
470 VkRenderPass depthstencil_rp
;
471 VkPipeline depth_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
472 VkPipeline stencil_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
473 VkPipeline depthstencil_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
474 } clear
[1 + MAX_SAMPLES_LOG2
];
476 VkPipelineLayout clear_color_p_layout
;
477 VkPipelineLayout clear_depth_p_layout
;
479 /* Optimized compute fast HTILE clear for stencil or depth only. */
480 VkPipeline clear_htile_mask_pipeline
;
481 VkPipelineLayout clear_htile_mask_p_layout
;
482 VkDescriptorSetLayout clear_htile_mask_ds_layout
;
485 VkRenderPass render_pass
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
487 /** Pipeline that blits from a 1D image. */
488 VkPipeline pipeline_1d_src
[NUM_META_FS_KEYS
];
490 /** Pipeline that blits from a 2D image. */
491 VkPipeline pipeline_2d_src
[NUM_META_FS_KEYS
];
493 /** Pipeline that blits from a 3D image. */
494 VkPipeline pipeline_3d_src
[NUM_META_FS_KEYS
];
496 VkRenderPass depth_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
497 VkPipeline depth_only_1d_pipeline
;
498 VkPipeline depth_only_2d_pipeline
;
499 VkPipeline depth_only_3d_pipeline
;
501 VkRenderPass stencil_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
502 VkPipeline stencil_only_1d_pipeline
;
503 VkPipeline stencil_only_2d_pipeline
;
504 VkPipeline stencil_only_3d_pipeline
;
505 VkPipelineLayout pipeline_layout
;
506 VkDescriptorSetLayout ds_layout
;
510 VkPipelineLayout p_layouts
[5];
511 VkDescriptorSetLayout ds_layouts
[5];
512 VkPipeline pipelines
[5][NUM_META_FS_KEYS
];
514 VkPipeline depth_only_pipeline
[5];
516 VkPipeline stencil_only_pipeline
[5];
517 } blit2d
[1 + MAX_SAMPLES_LOG2
];
519 VkRenderPass blit2d_render_passes
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
520 VkRenderPass blit2d_depth_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
521 VkRenderPass blit2d_stencil_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
524 VkPipelineLayout img_p_layout
;
525 VkDescriptorSetLayout img_ds_layout
;
527 VkPipeline pipeline_3d
;
530 VkPipelineLayout img_p_layout
;
531 VkDescriptorSetLayout img_ds_layout
;
533 VkPipeline pipeline_3d
;
536 VkPipelineLayout img_p_layout
;
537 VkDescriptorSetLayout img_ds_layout
;
541 VkPipelineLayout img_p_layout
;
542 VkDescriptorSetLayout img_ds_layout
;
544 VkPipeline pipeline_3d
;
547 VkPipelineLayout img_p_layout
;
548 VkDescriptorSetLayout img_ds_layout
;
552 VkPipelineLayout img_p_layout
;
553 VkDescriptorSetLayout img_ds_layout
;
555 VkPipeline pipeline_3d
;
558 VkPipelineLayout img_p_layout
;
559 VkDescriptorSetLayout img_ds_layout
;
564 VkPipelineLayout p_layout
;
565 VkPipeline pipeline
[NUM_META_FS_KEYS
];
566 VkRenderPass pass
[NUM_META_FS_KEYS
];
570 VkDescriptorSetLayout ds_layout
;
571 VkPipelineLayout p_layout
;
574 VkPipeline i_pipeline
;
575 VkPipeline srgb_pipeline
;
576 } rc
[MAX_SAMPLES_LOG2
];
580 VkDescriptorSetLayout ds_layout
;
581 VkPipelineLayout p_layout
;
584 VkRenderPass render_pass
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
585 VkPipeline pipeline
[NUM_META_FS_KEYS
];
586 } rc
[MAX_SAMPLES_LOG2
];
590 VkPipelineLayout p_layout
;
591 VkPipeline decompress_pipeline
;
592 VkPipeline resummarize_pipeline
;
594 } depth_decomp
[1 + MAX_SAMPLES_LOG2
];
597 VkPipelineLayout p_layout
;
598 VkPipeline cmask_eliminate_pipeline
;
599 VkPipeline fmask_decompress_pipeline
;
600 VkPipeline dcc_decompress_pipeline
;
603 VkDescriptorSetLayout dcc_decompress_compute_ds_layout
;
604 VkPipelineLayout dcc_decompress_compute_p_layout
;
605 VkPipeline dcc_decompress_compute_pipeline
;
609 VkPipelineLayout fill_p_layout
;
610 VkPipelineLayout copy_p_layout
;
611 VkDescriptorSetLayout fill_ds_layout
;
612 VkDescriptorSetLayout copy_ds_layout
;
613 VkPipeline fill_pipeline
;
614 VkPipeline copy_pipeline
;
618 VkDescriptorSetLayout ds_layout
;
619 VkPipelineLayout p_layout
;
620 VkPipeline occlusion_query_pipeline
;
621 VkPipeline pipeline_statistics_query_pipeline
;
622 VkPipeline tfb_query_pipeline
;
626 VkDescriptorSetLayout ds_layout
;
627 VkPipelineLayout p_layout
;
628 VkPipeline pipeline
[MAX_SAMPLES_LOG2
];
633 #define RADV_QUEUE_GENERAL 0
634 #define RADV_QUEUE_COMPUTE 1
635 #define RADV_QUEUE_TRANSFER 2
637 #define RADV_MAX_QUEUE_FAMILIES 3
639 enum ring_type
radv_queue_family_to_ring(int f
);
642 VK_LOADER_DATA _loader_data
;
643 struct radv_device
* device
;
644 struct radeon_winsys_ctx
*hw_ctx
;
645 enum radeon_ctx_priority priority
;
646 uint32_t queue_family_index
;
648 VkDeviceQueueCreateFlags flags
;
650 uint32_t scratch_size
;
651 uint32_t compute_scratch_size
;
652 uint32_t esgs_ring_size
;
653 uint32_t gsvs_ring_size
;
655 bool has_sample_positions
;
657 struct radeon_winsys_bo
*scratch_bo
;
658 struct radeon_winsys_bo
*descriptor_bo
;
659 struct radeon_winsys_bo
*compute_scratch_bo
;
660 struct radeon_winsys_bo
*esgs_ring_bo
;
661 struct radeon_winsys_bo
*gsvs_ring_bo
;
662 struct radeon_winsys_bo
*tess_rings_bo
;
663 struct radeon_cmdbuf
*initial_preamble_cs
;
664 struct radeon_cmdbuf
*initial_full_flush_preamble_cs
;
665 struct radeon_cmdbuf
*continue_preamble_cs
;
668 struct radv_bo_list
{
669 struct radv_winsys_bo_list list
;
671 pthread_mutex_t mutex
;
675 VK_LOADER_DATA _loader_data
;
677 VkAllocationCallbacks alloc
;
679 struct radv_instance
* instance
;
680 struct radeon_winsys
*ws
;
682 struct radv_meta_state meta_state
;
684 struct radv_queue
*queues
[RADV_MAX_QUEUE_FAMILIES
];
685 int queue_count
[RADV_MAX_QUEUE_FAMILIES
];
686 struct radeon_cmdbuf
*empty_cs
[RADV_MAX_QUEUE_FAMILIES
];
688 bool always_use_syncobj
;
689 bool has_distributed_tess
;
692 uint32_t tess_offchip_block_dw_size
;
693 uint32_t scratch_waves
;
694 uint32_t dispatch_initiator
;
696 uint32_t gs_table_depth
;
698 /* MSAA sample locations.
699 * The first index is the sample index.
700 * The second index is the coordinate: X, Y. */
701 float sample_locations_1x
[1][2];
702 float sample_locations_2x
[2][2];
703 float sample_locations_4x
[4][2];
704 float sample_locations_8x
[8][2];
707 uint32_t gfx_init_size_dw
;
708 struct radeon_winsys_bo
*gfx_init
;
710 struct radeon_winsys_bo
*trace_bo
;
711 uint32_t *trace_id_ptr
;
713 /* Whether to keep shader debug info, for tracing or VK_AMD_shader_info */
714 bool keep_shader_info
;
716 struct radv_physical_device
*physical_device
;
718 /* Backup in-memory cache to be used if the app doesn't provide one */
719 struct radv_pipeline_cache
* mem_cache
;
722 * use different counters so MSAA MRTs get consecutive surface indices,
723 * even if MASK is allocated in between.
725 uint32_t image_mrt_offset_counter
;
726 uint32_t fmask_mrt_offset_counter
;
727 struct list_head shader_slabs
;
728 mtx_t shader_slab_mutex
;
730 /* For detecting VM faults reported by dmesg. */
731 uint64_t dmesg_timestamp
;
733 struct radv_device_extension_table enabled_extensions
;
735 /* Whether the driver uses a global BO list. */
736 bool use_global_bo_list
;
738 struct radv_bo_list bo_list
;
740 /* Whether anisotropy is forced with RADV_TEX_ANISO (-1 is disabled). */
744 struct radv_device_memory
{
745 struct radeon_winsys_bo
*bo
;
746 /* for dedicated allocations */
747 struct radv_image
*image
;
748 struct radv_buffer
*buffer
;
750 VkDeviceSize map_size
;
756 struct radv_descriptor_range
{
761 struct radv_descriptor_set
{
762 const struct radv_descriptor_set_layout
*layout
;
765 struct radeon_winsys_bo
*bo
;
767 uint32_t *mapped_ptr
;
768 struct radv_descriptor_range
*dynamic_descriptors
;
770 struct radeon_winsys_bo
*descriptors
[0];
773 struct radv_push_descriptor_set
775 struct radv_descriptor_set set
;
779 struct radv_descriptor_pool_entry
{
782 struct radv_descriptor_set
*set
;
785 struct radv_descriptor_pool
{
786 struct radeon_winsys_bo
*bo
;
788 uint64_t current_offset
;
791 uint8_t *host_memory_base
;
792 uint8_t *host_memory_ptr
;
793 uint8_t *host_memory_end
;
795 uint32_t entry_count
;
796 uint32_t max_entry_count
;
797 struct radv_descriptor_pool_entry entries
[0];
800 struct radv_descriptor_update_template_entry
{
801 VkDescriptorType descriptor_type
;
803 /* The number of descriptors to update */
804 uint32_t descriptor_count
;
806 /* Into mapped_ptr or dynamic_descriptors, in units of the respective array */
809 /* In dwords. Not valid/used for dynamic descriptors */
812 uint32_t buffer_offset
;
814 /* Only valid for combined image samplers and samplers */
816 uint8_t sampler_offset
;
822 /* For push descriptors */
823 const uint32_t *immutable_samplers
;
826 struct radv_descriptor_update_template
{
827 uint32_t entry_count
;
828 VkPipelineBindPoint bind_point
;
829 struct radv_descriptor_update_template_entry entry
[0];
835 VkBufferUsageFlags usage
;
836 VkBufferCreateFlags flags
;
839 struct radeon_winsys_bo
* bo
;
845 enum radv_dynamic_state_bits
{
846 RADV_DYNAMIC_VIEWPORT
= 1 << 0,
847 RADV_DYNAMIC_SCISSOR
= 1 << 1,
848 RADV_DYNAMIC_LINE_WIDTH
= 1 << 2,
849 RADV_DYNAMIC_DEPTH_BIAS
= 1 << 3,
850 RADV_DYNAMIC_BLEND_CONSTANTS
= 1 << 4,
851 RADV_DYNAMIC_DEPTH_BOUNDS
= 1 << 5,
852 RADV_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6,
853 RADV_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7,
854 RADV_DYNAMIC_STENCIL_REFERENCE
= 1 << 8,
855 RADV_DYNAMIC_DISCARD_RECTANGLE
= 1 << 9,
856 RADV_DYNAMIC_SAMPLE_LOCATIONS
= 1 << 10,
857 RADV_DYNAMIC_ALL
= (1 << 11) - 1,
860 enum radv_cmd_dirty_bits
{
861 /* Keep the dynamic state dirty bits in sync with
862 * enum radv_dynamic_state_bits */
863 RADV_CMD_DIRTY_DYNAMIC_VIEWPORT
= 1 << 0,
864 RADV_CMD_DIRTY_DYNAMIC_SCISSOR
= 1 << 1,
865 RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH
= 1 << 2,
866 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS
= 1 << 3,
867 RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS
= 1 << 4,
868 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS
= 1 << 5,
869 RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6,
870 RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7,
871 RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE
= 1 << 8,
872 RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE
= 1 << 9,
873 RADV_CMD_DIRTY_DYNAMIC_SAMPLE_LOCATIONS
= 1 << 10,
874 RADV_CMD_DIRTY_DYNAMIC_ALL
= (1 << 11) - 1,
875 RADV_CMD_DIRTY_PIPELINE
= 1 << 11,
876 RADV_CMD_DIRTY_INDEX_BUFFER
= 1 << 12,
877 RADV_CMD_DIRTY_FRAMEBUFFER
= 1 << 13,
878 RADV_CMD_DIRTY_VERTEX_BUFFER
= 1 << 14,
879 RADV_CMD_DIRTY_STREAMOUT_BUFFER
= 1 << 15,
882 enum radv_cmd_flush_bits
{
883 RADV_CMD_FLAG_INV_ICACHE
= 1 << 0,
884 /* SMEM L1, other names: KCACHE, constant cache, DCACHE, data cache */
885 RADV_CMD_FLAG_INV_SMEM_L1
= 1 << 1,
886 /* VMEM L1 can optionally be bypassed (GLC=1). Other names: TC L1 */
887 RADV_CMD_FLAG_INV_VMEM_L1
= 1 << 2,
888 /* Used by everything except CB/DB, can be bypassed (SLC=1). Other names: TC L2 */
889 RADV_CMD_FLAG_INV_GLOBAL_L2
= 1 << 3,
890 /* Same as above, but only writes back and doesn't invalidate */
891 RADV_CMD_FLAG_WRITEBACK_GLOBAL_L2
= 1 << 4,
892 /* Framebuffer caches */
893 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
= 1 << 5,
894 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
= 1 << 6,
895 RADV_CMD_FLAG_FLUSH_AND_INV_DB
= 1 << 7,
896 RADV_CMD_FLAG_FLUSH_AND_INV_CB
= 1 << 8,
897 /* Engine synchronization. */
898 RADV_CMD_FLAG_VS_PARTIAL_FLUSH
= 1 << 9,
899 RADV_CMD_FLAG_PS_PARTIAL_FLUSH
= 1 << 10,
900 RADV_CMD_FLAG_CS_PARTIAL_FLUSH
= 1 << 11,
901 RADV_CMD_FLAG_VGT_FLUSH
= 1 << 12,
902 /* Pipeline query controls. */
903 RADV_CMD_FLAG_START_PIPELINE_STATS
= 1 << 13,
904 RADV_CMD_FLAG_STOP_PIPELINE_STATS
= 1 << 14,
905 RADV_CMD_FLAG_VGT_STREAMOUT_SYNC
= 1 << 15,
907 RADV_CMD_FLUSH_AND_INV_FRAMEBUFFER
= (RADV_CMD_FLAG_FLUSH_AND_INV_CB
|
908 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
|
909 RADV_CMD_FLAG_FLUSH_AND_INV_DB
|
910 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
)
913 struct radv_vertex_binding
{
914 struct radv_buffer
* buffer
;
918 struct radv_streamout_binding
{
919 struct radv_buffer
*buffer
;
924 struct radv_streamout_state
{
925 /* Mask of bound streamout buffers. */
926 uint8_t enabled_mask
;
928 /* External state that comes from the last vertex stage, it must be
929 * set explicitely when binding a new graphics pipeline.
931 uint16_t stride_in_dw
[MAX_SO_BUFFERS
];
932 uint32_t enabled_stream_buffers_mask
; /* stream0 buffers0-3 in 4 LSB */
934 /* State of VGT_STRMOUT_BUFFER_(CONFIG|END) */
935 uint32_t hw_enabled_mask
;
937 /* State of VGT_STRMOUT_(CONFIG|EN) */
938 bool streamout_enabled
;
941 struct radv_viewport_state
{
943 VkViewport viewports
[MAX_VIEWPORTS
];
946 struct radv_scissor_state
{
948 VkRect2D scissors
[MAX_SCISSORS
];
951 struct radv_discard_rectangle_state
{
953 VkRect2D rectangles
[MAX_DISCARD_RECTANGLES
];
956 struct radv_sample_locations_state
{
957 VkSampleCountFlagBits per_pixel
;
958 VkExtent2D grid_size
;
960 VkSampleLocationEXT locations
[MAX_SAMPLE_LOCATIONS
];
963 struct radv_dynamic_state
{
965 * Bitmask of (1 << VK_DYNAMIC_STATE_*).
966 * Defines the set of saved dynamic state.
970 struct radv_viewport_state viewport
;
972 struct radv_scissor_state scissor
;
982 float blend_constants
[4];
992 } stencil_compare_mask
;
997 } stencil_write_mask
;
1002 } stencil_reference
;
1004 struct radv_discard_rectangle_state discard_rectangle
;
1006 struct radv_sample_locations_state sample_location
;
1009 extern const struct radv_dynamic_state default_dynamic_state
;
1012 radv_get_debug_option_name(int id
);
1015 radv_get_perftest_option_name(int id
);
1018 * Attachment state when recording a renderpass instance.
1020 * The clear value is valid only if there exists a pending clear.
1022 struct radv_attachment_state
{
1023 VkImageAspectFlags pending_clear_aspects
;
1024 uint32_t cleared_views
;
1025 VkClearValue clear_value
;
1026 VkImageLayout current_layout
;
1029 struct radv_descriptor_state
{
1030 struct radv_descriptor_set
*sets
[MAX_SETS
];
1033 struct radv_push_descriptor_set push_set
;
1035 uint32_t dynamic_buffers
[4 * MAX_DYNAMIC_BUFFERS
];
1038 struct radv_cmd_state
{
1039 /* Vertex descriptors */
1046 uint32_t prefetch_L2_mask
;
1048 struct radv_pipeline
* pipeline
;
1049 struct radv_pipeline
* emitted_pipeline
;
1050 struct radv_pipeline
* compute_pipeline
;
1051 struct radv_pipeline
* emitted_compute_pipeline
;
1052 struct radv_framebuffer
* framebuffer
;
1053 struct radv_render_pass
* pass
;
1054 const struct radv_subpass
* subpass
;
1055 struct radv_dynamic_state dynamic
;
1056 struct radv_attachment_state
* attachments
;
1057 struct radv_streamout_state streamout
;
1058 VkRect2D render_area
;
1061 struct radv_buffer
*index_buffer
;
1062 uint64_t index_offset
;
1063 uint32_t index_type
;
1064 uint32_t max_index_count
;
1066 int32_t last_index_type
;
1068 int32_t last_primitive_reset_en
;
1069 uint32_t last_primitive_reset_index
;
1070 enum radv_cmd_flush_bits flush_bits
;
1071 unsigned active_occlusion_queries
;
1072 bool perfect_occlusion_queries_enabled
;
1073 unsigned active_pipeline_queries
;
1076 uint32_t last_ia_multi_vgt_param
;
1078 uint32_t last_num_instances
;
1079 uint32_t last_first_instance
;
1080 uint32_t last_vertex_offset
;
1082 /* Whether CP DMA is busy/idle. */
1085 /* Conditional rendering info. */
1086 int predication_type
; /* -1: disabled, 0: normal, 1: inverted */
1087 uint64_t predication_va
;
1089 bool context_roll_without_scissor_emitted
;
1092 struct radv_cmd_pool
{
1093 VkAllocationCallbacks alloc
;
1094 struct list_head cmd_buffers
;
1095 struct list_head free_cmd_buffers
;
1096 uint32_t queue_family_index
;
1099 struct radv_cmd_buffer_upload
{
1103 struct radeon_winsys_bo
*upload_bo
;
1104 struct list_head list
;
1107 enum radv_cmd_buffer_status
{
1108 RADV_CMD_BUFFER_STATUS_INVALID
,
1109 RADV_CMD_BUFFER_STATUS_INITIAL
,
1110 RADV_CMD_BUFFER_STATUS_RECORDING
,
1111 RADV_CMD_BUFFER_STATUS_EXECUTABLE
,
1112 RADV_CMD_BUFFER_STATUS_PENDING
,
1115 struct radv_cmd_buffer
{
1116 VK_LOADER_DATA _loader_data
;
1118 struct radv_device
* device
;
1120 struct radv_cmd_pool
* pool
;
1121 struct list_head pool_link
;
1123 VkCommandBufferUsageFlags usage_flags
;
1124 VkCommandBufferLevel level
;
1125 enum radv_cmd_buffer_status status
;
1126 struct radeon_cmdbuf
*cs
;
1127 struct radv_cmd_state state
;
1128 struct radv_vertex_binding vertex_bindings
[MAX_VBS
];
1129 struct radv_streamout_binding streamout_bindings
[MAX_SO_BUFFERS
];
1130 uint32_t queue_family_index
;
1132 uint8_t push_constants
[MAX_PUSH_CONSTANTS_SIZE
];
1133 VkShaderStageFlags push_constant_stages
;
1134 struct radv_descriptor_set meta_push_descriptors
;
1136 struct radv_descriptor_state descriptors
[VK_PIPELINE_BIND_POINT_RANGE_SIZE
];
1138 struct radv_cmd_buffer_upload upload
;
1140 uint32_t scratch_size_needed
;
1141 uint32_t compute_scratch_size_needed
;
1142 uint32_t esgs_ring_size_needed
;
1143 uint32_t gsvs_ring_size_needed
;
1144 bool tess_rings_needed
;
1145 bool sample_positions_needed
;
1147 VkResult record_result
;
1149 uint64_t gfx9_fence_va
;
1150 uint32_t gfx9_fence_idx
;
1151 uint64_t gfx9_eop_bug_va
;
1154 * Whether a query pool has been resetted and we have to flush caches.
1156 bool pending_reset_query
;
1159 * Bitmask of pending active query flushes.
1161 enum radv_cmd_flush_bits active_query_flush_bits
;
1166 bool radv_cmd_buffer_uses_mec(struct radv_cmd_buffer
*cmd_buffer
);
1168 void si_emit_graphics(struct radv_physical_device
*physical_device
,
1169 struct radeon_cmdbuf
*cs
);
1170 void si_emit_compute(struct radv_physical_device
*physical_device
,
1171 struct radeon_cmdbuf
*cs
);
1173 void cik_create_gfx_config(struct radv_device
*device
);
1175 void si_write_viewport(struct radeon_cmdbuf
*cs
, int first_vp
,
1176 int count
, const VkViewport
*viewports
);
1177 void si_write_scissors(struct radeon_cmdbuf
*cs
, int first
,
1178 int count
, const VkRect2D
*scissors
,
1179 const VkViewport
*viewports
, bool can_use_guardband
);
1180 uint32_t si_get_ia_multi_vgt_param(struct radv_cmd_buffer
*cmd_buffer
,
1181 bool instanced_draw
, bool indirect_draw
,
1182 bool count_from_stream_output
,
1183 uint32_t draw_vertex_count
);
1184 void si_cs_emit_write_event_eop(struct radeon_cmdbuf
*cs
,
1185 enum chip_class chip_class
,
1187 unsigned event
, unsigned event_flags
,
1191 uint64_t gfx9_eop_bug_va
);
1193 void radv_cp_wait_mem(struct radeon_cmdbuf
*cs
, uint32_t op
, uint64_t va
,
1194 uint32_t ref
, uint32_t mask
);
1195 void si_cs_emit_cache_flush(struct radeon_cmdbuf
*cs
,
1196 enum chip_class chip_class
,
1197 uint32_t *fence_ptr
, uint64_t va
,
1199 enum radv_cmd_flush_bits flush_bits
,
1200 uint64_t gfx9_eop_bug_va
);
1201 void si_emit_cache_flush(struct radv_cmd_buffer
*cmd_buffer
);
1202 void si_emit_set_predication_state(struct radv_cmd_buffer
*cmd_buffer
,
1203 bool inverted
, uint64_t va
);
1204 void si_cp_dma_buffer_copy(struct radv_cmd_buffer
*cmd_buffer
,
1205 uint64_t src_va
, uint64_t dest_va
,
1207 void si_cp_dma_prefetch(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
1209 void si_cp_dma_clear_buffer(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
1210 uint64_t size
, unsigned value
);
1211 void si_cp_dma_wait_for_idle(struct radv_cmd_buffer
*cmd_buffer
);
1213 void radv_set_db_count_control(struct radv_cmd_buffer
*cmd_buffer
);
1215 radv_cmd_buffer_upload_alloc(struct radv_cmd_buffer
*cmd_buffer
,
1218 unsigned *out_offset
,
1221 radv_cmd_buffer_set_subpass(struct radv_cmd_buffer
*cmd_buffer
,
1222 const struct radv_subpass
*subpass
);
1224 radv_cmd_buffer_upload_data(struct radv_cmd_buffer
*cmd_buffer
,
1225 unsigned size
, unsigned alignmnet
,
1226 const void *data
, unsigned *out_offset
);
1228 void radv_cmd_buffer_clear_subpass(struct radv_cmd_buffer
*cmd_buffer
);
1229 void radv_cmd_buffer_resolve_subpass(struct radv_cmd_buffer
*cmd_buffer
);
1230 void radv_cmd_buffer_resolve_subpass_cs(struct radv_cmd_buffer
*cmd_buffer
);
1231 void radv_cmd_buffer_resolve_subpass_fs(struct radv_cmd_buffer
*cmd_buffer
);
1232 void radv_emit_default_sample_locations(struct radeon_cmdbuf
*cs
, int nr_samples
);
1233 unsigned radv_get_default_max_sample_dist(int log_samples
);
1234 void radv_device_init_msaa(struct radv_device
*device
);
1236 void radv_update_ds_clear_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1237 struct radv_image
*image
,
1238 VkClearDepthStencilValue ds_clear_value
,
1239 VkImageAspectFlags aspects
);
1241 void radv_update_color_clear_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1242 struct radv_image
*image
,
1244 uint32_t color_values
[2]);
1246 void radv_update_fce_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1247 struct radv_image
*image
, bool value
);
1249 void radv_update_dcc_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1250 struct radv_image
*image
, bool value
);
1252 uint32_t radv_fill_buffer(struct radv_cmd_buffer
*cmd_buffer
,
1253 struct radeon_winsys_bo
*bo
,
1254 uint64_t offset
, uint64_t size
, uint32_t value
);
1255 void radv_cmd_buffer_trace_emit(struct radv_cmd_buffer
*cmd_buffer
);
1256 bool radv_get_memory_fd(struct radv_device
*device
,
1257 struct radv_device_memory
*memory
,
1261 radv_emit_shader_pointer_head(struct radeon_cmdbuf
*cs
,
1262 unsigned sh_offset
, unsigned pointer_count
,
1263 bool use_32bit_pointers
)
1265 radeon_emit(cs
, PKT3(PKT3_SET_SH_REG
, pointer_count
* (use_32bit_pointers
? 1 : 2), 0));
1266 radeon_emit(cs
, (sh_offset
- SI_SH_REG_OFFSET
) >> 2);
1270 radv_emit_shader_pointer_body(struct radv_device
*device
,
1271 struct radeon_cmdbuf
*cs
,
1272 uint64_t va
, bool use_32bit_pointers
)
1274 radeon_emit(cs
, va
);
1276 if (use_32bit_pointers
) {
1278 (va
>> 32) == device
->physical_device
->rad_info
.address32_hi
);
1280 radeon_emit(cs
, va
>> 32);
1285 radv_emit_shader_pointer(struct radv_device
*device
,
1286 struct radeon_cmdbuf
*cs
,
1287 uint32_t sh_offset
, uint64_t va
, bool global
)
1289 bool use_32bit_pointers
= !global
;
1291 radv_emit_shader_pointer_head(cs
, sh_offset
, 1, use_32bit_pointers
);
1292 radv_emit_shader_pointer_body(device
, cs
, va
, use_32bit_pointers
);
1295 static inline struct radv_descriptor_state
*
1296 radv_get_descriptors_state(struct radv_cmd_buffer
*cmd_buffer
,
1297 VkPipelineBindPoint bind_point
)
1299 assert(bind_point
== VK_PIPELINE_BIND_POINT_GRAPHICS
||
1300 bind_point
== VK_PIPELINE_BIND_POINT_COMPUTE
);
1301 return &cmd_buffer
->descriptors
[bind_point
];
1305 * Takes x,y,z as exact numbers of invocations, instead of blocks.
1307 * Limitations: Can't call normal dispatch functions without binding or rebinding
1308 * the compute pipeline.
1310 void radv_unaligned_dispatch(
1311 struct radv_cmd_buffer
*cmd_buffer
,
1317 struct radeon_winsys_bo
*bo
;
1321 struct radv_shader_module
;
1323 #define RADV_HASH_SHADER_IS_GEOM_COPY_SHADER (1 << 0)
1324 #define RADV_HASH_SHADER_SISCHED (1 << 1)
1325 #define RADV_HASH_SHADER_UNSAFE_MATH (1 << 2)
1327 radv_hash_shaders(unsigned char *hash
,
1328 const VkPipelineShaderStageCreateInfo
**stages
,
1329 const struct radv_pipeline_layout
*layout
,
1330 const struct radv_pipeline_key
*key
,
1333 static inline gl_shader_stage
1334 vk_to_mesa_shader_stage(VkShaderStageFlagBits vk_stage
)
1336 assert(__builtin_popcount(vk_stage
) == 1);
1337 return ffs(vk_stage
) - 1;
1340 static inline VkShaderStageFlagBits
1341 mesa_to_vk_shader_stage(gl_shader_stage mesa_stage
)
1343 return (1 << mesa_stage
);
1346 #define RADV_STAGE_MASK ((1 << MESA_SHADER_STAGES) - 1)
1348 #define radv_foreach_stage(stage, stage_bits) \
1349 for (gl_shader_stage stage, \
1350 __tmp = (gl_shader_stage)((stage_bits) & RADV_STAGE_MASK); \
1351 stage = __builtin_ffs(__tmp) - 1, __tmp; \
1352 __tmp &= ~(1 << (stage)))
1354 extern const VkFormat radv_fs_key_format_exemplars
[NUM_META_FS_KEYS
];
1355 unsigned radv_format_meta_fs_key(VkFormat format
);
1357 struct radv_multisample_state
{
1359 uint32_t pa_sc_line_cntl
;
1360 uint32_t pa_sc_mode_cntl_0
;
1361 uint32_t pa_sc_mode_cntl_1
;
1362 uint32_t pa_sc_aa_config
;
1363 uint32_t pa_sc_aa_mask
[2];
1364 unsigned num_samples
;
1367 struct radv_prim_vertex_count
{
1372 struct radv_vertex_elements_info
{
1373 uint32_t format_size
[MAX_VERTEX_ATTRIBS
];
1376 struct radv_ia_multi_vgt_param_helpers
{
1378 bool partial_es_wave
;
1379 uint8_t primgroup_size
;
1380 bool wd_switch_on_eop
;
1381 bool ia_switch_on_eoi
;
1382 bool partial_vs_wave
;
1385 #define SI_GS_PER_ES 128
1387 struct radv_pipeline
{
1388 struct radv_device
* device
;
1389 struct radv_dynamic_state dynamic_state
;
1391 struct radv_pipeline_layout
* layout
;
1393 bool need_indirect_descriptor_sets
;
1394 struct radv_shader_variant
* shaders
[MESA_SHADER_STAGES
];
1395 struct radv_shader_variant
*gs_copy_shader
;
1396 VkShaderStageFlags active_stages
;
1398 struct radeon_cmdbuf cs
;
1399 uint32_t ctx_cs_hash
;
1400 struct radeon_cmdbuf ctx_cs
;
1402 struct radv_vertex_elements_info vertex_elements
;
1404 uint32_t binding_stride
[MAX_VBS
];
1405 uint8_t num_vertex_bindings
;
1407 uint32_t user_data_0
[MESA_SHADER_STAGES
];
1410 struct radv_multisample_state ms
;
1411 uint32_t spi_baryc_cntl
;
1412 bool prim_restart_enable
;
1413 unsigned esgs_ring_size
;
1414 unsigned gsvs_ring_size
;
1415 uint32_t vtx_base_sgpr
;
1416 struct radv_ia_multi_vgt_param_helpers ia_multi_vgt_param
;
1417 uint8_t vtx_emit_num
;
1418 struct radv_prim_vertex_count prim_vertex_count
;
1419 bool can_use_guardband
;
1420 uint32_t needed_dynamic_state
;
1421 bool disable_out_of_order_rast_for_occlusion
;
1423 /* Used for rbplus */
1424 uint32_t col_format
;
1425 uint32_t cb_target_mask
;
1430 unsigned scratch_bytes_per_wave
;
1432 /* Not NULL if graphics pipeline uses streamout. */
1433 struct radv_shader_variant
*streamout_shader
;
1436 static inline bool radv_pipeline_has_gs(const struct radv_pipeline
*pipeline
)
1438 return pipeline
->shaders
[MESA_SHADER_GEOMETRY
] ? true : false;
1441 static inline bool radv_pipeline_has_tess(const struct radv_pipeline
*pipeline
)
1443 return pipeline
->shaders
[MESA_SHADER_TESS_CTRL
] ? true : false;
1446 struct radv_userdata_info
*radv_lookup_user_sgpr(struct radv_pipeline
*pipeline
,
1447 gl_shader_stage stage
,
1450 struct radv_shader_variant
*radv_get_shader(struct radv_pipeline
*pipeline
,
1451 gl_shader_stage stage
);
1453 struct radv_graphics_pipeline_create_info
{
1455 bool db_depth_clear
;
1456 bool db_stencil_clear
;
1457 bool db_depth_disable_expclear
;
1458 bool db_stencil_disable_expclear
;
1459 bool db_flush_depth_inplace
;
1460 bool db_flush_stencil_inplace
;
1461 bool db_resummarize
;
1462 uint32_t custom_blend_mode
;
1466 radv_graphics_pipeline_create(VkDevice device
,
1467 VkPipelineCache cache
,
1468 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
1469 const struct radv_graphics_pipeline_create_info
*extra
,
1470 const VkAllocationCallbacks
*alloc
,
1471 VkPipeline
*pPipeline
);
1473 struct vk_format_description
;
1474 uint32_t radv_translate_buffer_dataformat(const struct vk_format_description
*desc
,
1475 int first_non_void
);
1476 uint32_t radv_translate_buffer_numformat(const struct vk_format_description
*desc
,
1477 int first_non_void
);
1478 uint32_t radv_translate_colorformat(VkFormat format
);
1479 uint32_t radv_translate_color_numformat(VkFormat format
,
1480 const struct vk_format_description
*desc
,
1481 int first_non_void
);
1482 uint32_t radv_colorformat_endian_swap(uint32_t colorformat
);
1483 unsigned radv_translate_colorswap(VkFormat format
, bool do_endian_swap
);
1484 uint32_t radv_translate_dbformat(VkFormat format
);
1485 uint32_t radv_translate_tex_dataformat(VkFormat format
,
1486 const struct vk_format_description
*desc
,
1487 int first_non_void
);
1488 uint32_t radv_translate_tex_numformat(VkFormat format
,
1489 const struct vk_format_description
*desc
,
1490 int first_non_void
);
1491 bool radv_format_pack_clear_color(VkFormat format
,
1492 uint32_t clear_vals
[2],
1493 VkClearColorValue
*value
);
1494 bool radv_is_colorbuffer_format_supported(VkFormat format
, bool *blendable
);
1495 bool radv_dcc_formats_compatible(VkFormat format1
,
1497 bool radv_device_supports_etc(struct radv_physical_device
*physical_device
);
1499 struct radv_fmask_info
{
1503 unsigned pitch_in_pixels
;
1504 unsigned bank_height
;
1505 unsigned slice_tile_max
;
1506 unsigned tile_mode_index
;
1507 unsigned tile_swizzle
;
1510 struct radv_cmask_info
{
1514 unsigned slice_tile_max
;
1518 struct radv_image_plane
{
1520 struct radeon_surf surface
;
1526 /* The original VkFormat provided by the client. This may not match any
1527 * of the actual surface formats.
1530 VkImageAspectFlags aspects
;
1531 VkImageUsageFlags usage
; /**< Superset of VkImageCreateInfo::usage. */
1532 struct ac_surf_info info
;
1533 VkImageTiling tiling
; /** VkImageCreateInfo::tiling */
1534 VkImageCreateFlags flags
; /** VkImageCreateInfo::flags */
1539 unsigned queue_family_mask
;
1543 /* Set when bound */
1544 struct radeon_winsys_bo
*bo
;
1545 VkDeviceSize offset
;
1546 uint64_t dcc_offset
;
1547 uint64_t htile_offset
;
1548 bool tc_compatible_htile
;
1550 struct radv_fmask_info fmask
;
1551 struct radv_cmask_info cmask
;
1552 uint64_t clear_value_offset
;
1553 uint64_t fce_pred_offset
;
1554 uint64_t dcc_pred_offset
;
1557 * Metadata for the TC-compat zrange workaround. If the 32-bit value
1558 * stored at this offset is UINT_MAX, the driver will emit
1559 * DB_Z_INFO.ZRANGE_PRECISION=0, otherwise it will skip the
1560 * SET_CONTEXT_REG packet.
1562 uint64_t tc_compat_zrange_offset
;
1564 /* For VK_ANDROID_native_buffer, the WSI image owns the memory, */
1565 VkDeviceMemory owned_memory
;
1567 unsigned plane_count
;
1568 struct radv_image_plane planes
[0];
1571 /* Whether the image has a htile that is known consistent with the contents of
1573 bool radv_layout_has_htile(const struct radv_image
*image
,
1574 VkImageLayout layout
,
1575 unsigned queue_mask
);
1577 /* Whether the image has a htile that is known consistent with the contents of
1578 * the image and is allowed to be in compressed form.
1580 * If this is false reads that don't use the htile should be able to return
1583 bool radv_layout_is_htile_compressed(const struct radv_image
*image
,
1584 VkImageLayout layout
,
1585 unsigned queue_mask
);
1587 bool radv_layout_can_fast_clear(const struct radv_image
*image
,
1588 VkImageLayout layout
,
1589 unsigned queue_mask
);
1591 bool radv_layout_dcc_compressed(const struct radv_image
*image
,
1592 VkImageLayout layout
,
1593 unsigned queue_mask
);
1596 * Return whether the image has CMASK metadata for color surfaces.
1599 radv_image_has_cmask(const struct radv_image
*image
)
1601 return image
->cmask
.size
;
1605 * Return whether the image has FMASK metadata for color surfaces.
1608 radv_image_has_fmask(const struct radv_image
*image
)
1610 return image
->fmask
.size
;
1614 * Return whether the image has DCC metadata for color surfaces.
1617 radv_image_has_dcc(const struct radv_image
*image
)
1619 return image
->planes
[0].surface
.dcc_size
;
1623 * Return whether DCC metadata is enabled for a level.
1626 radv_dcc_enabled(const struct radv_image
*image
, unsigned level
)
1628 return radv_image_has_dcc(image
) &&
1629 level
< image
->planes
[0].surface
.num_dcc_levels
;
1633 * Return whether the image has CB metadata.
1636 radv_image_has_CB_metadata(const struct radv_image
*image
)
1638 return radv_image_has_cmask(image
) ||
1639 radv_image_has_fmask(image
) ||
1640 radv_image_has_dcc(image
);
1644 * Return whether the image has HTILE metadata for depth surfaces.
1647 radv_image_has_htile(const struct radv_image
*image
)
1649 return image
->planes
[0].surface
.htile_size
;
1653 * Return whether HTILE metadata is enabled for a level.
1656 radv_htile_enabled(const struct radv_image
*image
, unsigned level
)
1658 return radv_image_has_htile(image
) && level
== 0;
1662 * Return whether the image is TC-compatible HTILE.
1665 radv_image_is_tc_compat_htile(const struct radv_image
*image
)
1667 return radv_image_has_htile(image
) && image
->tc_compatible_htile
;
1670 unsigned radv_image_queue_family_mask(const struct radv_image
*image
, uint32_t family
, uint32_t queue_family
);
1672 static inline uint32_t
1673 radv_get_layerCount(const struct radv_image
*image
,
1674 const VkImageSubresourceRange
*range
)
1676 return range
->layerCount
== VK_REMAINING_ARRAY_LAYERS
?
1677 image
->info
.array_size
- range
->baseArrayLayer
: range
->layerCount
;
1680 static inline uint32_t
1681 radv_get_levelCount(const struct radv_image
*image
,
1682 const VkImageSubresourceRange
*range
)
1684 return range
->levelCount
== VK_REMAINING_MIP_LEVELS
?
1685 image
->info
.levels
- range
->baseMipLevel
: range
->levelCount
;
1688 struct radeon_bo_metadata
;
1690 radv_init_metadata(struct radv_device
*device
,
1691 struct radv_image
*image
,
1692 struct radeon_bo_metadata
*metadata
);
1695 radv_image_override_offset_stride(struct radv_device
*device
,
1696 struct radv_image
*image
,
1697 uint64_t offset
, uint32_t stride
);
1699 union radv_descriptor
{
1701 uint32_t plane0_descriptor
[8];
1702 uint32_t fmask_descriptor
[8];
1705 uint32_t plane_descriptors
[3][8];
1709 struct radv_image_view
{
1710 struct radv_image
*image
; /**< VkImageViewCreateInfo::image */
1711 struct radeon_winsys_bo
*bo
;
1713 VkImageViewType type
;
1714 VkImageAspectFlags aspect_mask
;
1717 bool multiple_planes
;
1718 uint32_t base_layer
;
1719 uint32_t layer_count
;
1721 uint32_t level_count
;
1722 VkExtent3D extent
; /**< Extent of VkImageViewCreateInfo::baseMipLevel. */
1724 union radv_descriptor descriptor
;
1726 /* Descriptor for use as a storage image as opposed to a sampled image.
1727 * This has a few differences for cube maps (e.g. type).
1729 union radv_descriptor storage_descriptor
;
1732 struct radv_image_create_info
{
1733 const VkImageCreateInfo
*vk_info
;
1735 bool no_metadata_planes
;
1738 VkResult
radv_image_create(VkDevice _device
,
1739 const struct radv_image_create_info
*info
,
1740 const VkAllocationCallbacks
* alloc
,
1744 radv_image_from_gralloc(VkDevice device_h
,
1745 const VkImageCreateInfo
*base_info
,
1746 const VkNativeBufferANDROID
*gralloc_info
,
1747 const VkAllocationCallbacks
*alloc
,
1748 VkImage
*out_image_h
);
1750 void radv_image_view_init(struct radv_image_view
*view
,
1751 struct radv_device
*device
,
1752 const VkImageViewCreateInfo
* pCreateInfo
);
1754 VkFormat
radv_get_aspect_format(struct radv_image
*image
, VkImageAspectFlags mask
);
1756 struct radv_sampler_ycbcr_conversion
{
1758 VkSamplerYcbcrModelConversion ycbcr_model
;
1759 VkSamplerYcbcrRange ycbcr_range
;
1760 VkComponentMapping components
;
1761 VkChromaLocation chroma_offsets
[2];
1762 VkFilter chroma_filter
;
1765 struct radv_buffer_view
{
1766 struct radeon_winsys_bo
*bo
;
1768 uint64_t range
; /**< VkBufferViewCreateInfo::range */
1771 void radv_buffer_view_init(struct radv_buffer_view
*view
,
1772 struct radv_device
*device
,
1773 const VkBufferViewCreateInfo
* pCreateInfo
);
1775 static inline struct VkExtent3D
1776 radv_sanitize_image_extent(const VkImageType imageType
,
1777 const struct VkExtent3D imageExtent
)
1779 switch (imageType
) {
1780 case VK_IMAGE_TYPE_1D
:
1781 return (VkExtent3D
) { imageExtent
.width
, 1, 1 };
1782 case VK_IMAGE_TYPE_2D
:
1783 return (VkExtent3D
) { imageExtent
.width
, imageExtent
.height
, 1 };
1784 case VK_IMAGE_TYPE_3D
:
1787 unreachable("invalid image type");
1791 static inline struct VkOffset3D
1792 radv_sanitize_image_offset(const VkImageType imageType
,
1793 const struct VkOffset3D imageOffset
)
1795 switch (imageType
) {
1796 case VK_IMAGE_TYPE_1D
:
1797 return (VkOffset3D
) { imageOffset
.x
, 0, 0 };
1798 case VK_IMAGE_TYPE_2D
:
1799 return (VkOffset3D
) { imageOffset
.x
, imageOffset
.y
, 0 };
1800 case VK_IMAGE_TYPE_3D
:
1803 unreachable("invalid image type");
1808 radv_image_extent_compare(const struct radv_image
*image
,
1809 const VkExtent3D
*extent
)
1811 if (extent
->width
!= image
->info
.width
||
1812 extent
->height
!= image
->info
.height
||
1813 extent
->depth
!= image
->info
.depth
)
1818 struct radv_sampler
{
1820 struct radv_sampler_ycbcr_conversion
*ycbcr_sampler
;
1823 struct radv_color_buffer_info
{
1824 uint64_t cb_color_base
;
1825 uint64_t cb_color_cmask
;
1826 uint64_t cb_color_fmask
;
1827 uint64_t cb_dcc_base
;
1828 uint32_t cb_color_slice
;
1829 uint32_t cb_color_view
;
1830 uint32_t cb_color_info
;
1831 uint32_t cb_color_attrib
;
1832 uint32_t cb_color_attrib2
;
1833 uint32_t cb_dcc_control
;
1834 uint32_t cb_color_cmask_slice
;
1835 uint32_t cb_color_fmask_slice
;
1837 uint32_t cb_color_pitch
; // GFX6-GFX8
1838 uint32_t cb_mrt_epitch
; // GFX9+
1842 struct radv_ds_buffer_info
{
1843 uint64_t db_z_read_base
;
1844 uint64_t db_stencil_read_base
;
1845 uint64_t db_z_write_base
;
1846 uint64_t db_stencil_write_base
;
1847 uint64_t db_htile_data_base
;
1848 uint32_t db_depth_info
;
1850 uint32_t db_stencil_info
;
1851 uint32_t db_depth_view
;
1852 uint32_t db_depth_size
;
1853 uint32_t db_depth_slice
;
1854 uint32_t db_htile_surface
;
1855 uint32_t pa_su_poly_offset_db_fmt_cntl
;
1856 uint32_t db_z_info2
;
1857 uint32_t db_stencil_info2
;
1861 struct radv_attachment_info
{
1863 struct radv_color_buffer_info cb
;
1864 struct radv_ds_buffer_info ds
;
1866 struct radv_image_view
*attachment
;
1869 struct radv_framebuffer
{
1874 uint32_t attachment_count
;
1875 struct radv_attachment_info attachments
[0];
1878 struct radv_subpass_barrier
{
1879 VkPipelineStageFlags src_stage_mask
;
1880 VkAccessFlags src_access_mask
;
1881 VkAccessFlags dst_access_mask
;
1884 void radv_subpass_barrier(struct radv_cmd_buffer
*cmd_buffer
,
1885 const struct radv_subpass_barrier
*barrier
);
1887 struct radv_subpass_attachment
{
1888 uint32_t attachment
;
1889 VkImageLayout layout
;
1892 struct radv_subpass
{
1893 uint32_t attachment_count
;
1894 struct radv_subpass_attachment
* attachments
;
1896 uint32_t input_count
;
1897 uint32_t color_count
;
1898 struct radv_subpass_attachment
* input_attachments
;
1899 struct radv_subpass_attachment
* color_attachments
;
1900 struct radv_subpass_attachment
* resolve_attachments
;
1901 struct radv_subpass_attachment
* depth_stencil_attachment
;
1903 /** Subpass has at least one resolve attachment */
1906 /** Subpass has at least one color attachment */
1909 struct radv_subpass_barrier start_barrier
;
1912 VkSampleCountFlagBits max_sample_count
;
1915 struct radv_render_pass_attachment
{
1918 VkAttachmentLoadOp load_op
;
1919 VkAttachmentLoadOp stencil_load_op
;
1920 VkImageLayout initial_layout
;
1921 VkImageLayout final_layout
;
1923 /* The subpass id in which the attachment will be used last. */
1924 uint32_t last_subpass_idx
;
1927 struct radv_render_pass
{
1928 uint32_t attachment_count
;
1929 uint32_t subpass_count
;
1930 struct radv_subpass_attachment
* subpass_attachments
;
1931 struct radv_render_pass_attachment
* attachments
;
1932 struct radv_subpass_barrier end_barrier
;
1933 struct radv_subpass subpasses
[0];
1936 VkResult
radv_device_init_meta(struct radv_device
*device
);
1937 void radv_device_finish_meta(struct radv_device
*device
);
1939 struct radv_query_pool
{
1940 struct radeon_winsys_bo
*bo
;
1942 uint32_t availability_offset
;
1946 uint32_t pipeline_stats_mask
;
1949 struct radv_semaphore
{
1950 /* use a winsys sem for non-exportable */
1951 struct radeon_winsys_sem
*sem
;
1953 uint32_t temp_syncobj
;
1956 void radv_set_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
1957 VkPipelineBindPoint bind_point
,
1958 struct radv_descriptor_set
*set
,
1962 radv_update_descriptor_sets(struct radv_device
*device
,
1963 struct radv_cmd_buffer
*cmd_buffer
,
1964 VkDescriptorSet overrideSet
,
1965 uint32_t descriptorWriteCount
,
1966 const VkWriteDescriptorSet
*pDescriptorWrites
,
1967 uint32_t descriptorCopyCount
,
1968 const VkCopyDescriptorSet
*pDescriptorCopies
);
1971 radv_update_descriptor_set_with_template(struct radv_device
*device
,
1972 struct radv_cmd_buffer
*cmd_buffer
,
1973 struct radv_descriptor_set
*set
,
1974 VkDescriptorUpdateTemplate descriptorUpdateTemplate
,
1977 void radv_meta_push_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
1978 VkPipelineBindPoint pipelineBindPoint
,
1979 VkPipelineLayout _layout
,
1981 uint32_t descriptorWriteCount
,
1982 const VkWriteDescriptorSet
*pDescriptorWrites
);
1984 void radv_initialize_dcc(struct radv_cmd_buffer
*cmd_buffer
,
1985 struct radv_image
*image
, uint32_t value
);
1987 void radv_initialize_fmask(struct radv_cmd_buffer
*cmd_buffer
,
1988 struct radv_image
*image
);
1991 struct radeon_winsys_fence
*fence
;
1992 struct wsi_fence
*fence_wsi
;
1995 uint32_t temp_syncobj
;
1998 /* radv_nir_to_llvm.c */
1999 struct radv_shader_variant_info
;
2000 struct radv_nir_compiler_options
;
2002 void radv_compile_gs_copy_shader(struct ac_llvm_compiler
*ac_llvm
,
2003 struct nir_shader
*geom_shader
,
2004 struct ac_shader_binary
*binary
,
2005 struct ac_shader_config
*config
,
2006 struct radv_shader_variant_info
*shader_info
,
2007 const struct radv_nir_compiler_options
*option
);
2009 void radv_compile_nir_shader(struct ac_llvm_compiler
*ac_llvm
,
2010 struct ac_shader_binary
*binary
,
2011 struct ac_shader_config
*config
,
2012 struct radv_shader_variant_info
*shader_info
,
2013 struct nir_shader
*const *nir
,
2015 const struct radv_nir_compiler_options
*options
);
2017 unsigned radv_nir_get_max_workgroup_size(enum chip_class chip_class
,
2018 const struct nir_shader
*nir
);
2020 /* radv_shader_info.h */
2021 struct radv_shader_info
;
2023 void radv_nir_shader_info_pass(const struct nir_shader
*nir
,
2024 const struct radv_nir_compiler_options
*options
,
2025 struct radv_shader_info
*info
);
2027 void radv_nir_shader_info_init(struct radv_shader_info
*info
);
2029 struct radeon_winsys_sem
;
2031 uint64_t radv_get_current_time(void);
2033 #define RADV_DEFINE_HANDLE_CASTS(__radv_type, __VkType) \
2035 static inline struct __radv_type * \
2036 __radv_type ## _from_handle(__VkType _handle) \
2038 return (struct __radv_type *) _handle; \
2041 static inline __VkType \
2042 __radv_type ## _to_handle(struct __radv_type *_obj) \
2044 return (__VkType) _obj; \
2047 #define RADV_DEFINE_NONDISP_HANDLE_CASTS(__radv_type, __VkType) \
2049 static inline struct __radv_type * \
2050 __radv_type ## _from_handle(__VkType _handle) \
2052 return (struct __radv_type *)(uintptr_t) _handle; \
2055 static inline __VkType \
2056 __radv_type ## _to_handle(struct __radv_type *_obj) \
2058 return (__VkType)(uintptr_t) _obj; \
2061 #define RADV_FROM_HANDLE(__radv_type, __name, __handle) \
2062 struct __radv_type *__name = __radv_type ## _from_handle(__handle)
2064 RADV_DEFINE_HANDLE_CASTS(radv_cmd_buffer
, VkCommandBuffer
)
2065 RADV_DEFINE_HANDLE_CASTS(radv_device
, VkDevice
)
2066 RADV_DEFINE_HANDLE_CASTS(radv_instance
, VkInstance
)
2067 RADV_DEFINE_HANDLE_CASTS(radv_physical_device
, VkPhysicalDevice
)
2068 RADV_DEFINE_HANDLE_CASTS(radv_queue
, VkQueue
)
2070 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_cmd_pool
, VkCommandPool
)
2071 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer
, VkBuffer
)
2072 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer_view
, VkBufferView
)
2073 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_pool
, VkDescriptorPool
)
2074 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set
, VkDescriptorSet
)
2075 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set_layout
, VkDescriptorSetLayout
)
2076 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_update_template
, VkDescriptorUpdateTemplate
)
2077 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_device_memory
, VkDeviceMemory
)
2078 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_fence
, VkFence
)
2079 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_event
, VkEvent
)
2080 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_framebuffer
, VkFramebuffer
)
2081 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image
, VkImage
)
2082 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image_view
, VkImageView
);
2083 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_cache
, VkPipelineCache
)
2084 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline
, VkPipeline
)
2085 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_layout
, VkPipelineLayout
)
2086 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_query_pool
, VkQueryPool
)
2087 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_render_pass
, VkRenderPass
)
2088 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_sampler
, VkSampler
)
2089 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_sampler_ycbcr_conversion
, VkSamplerYcbcrConversion
)
2090 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_shader_module
, VkShaderModule
)
2091 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_semaphore
, VkSemaphore
)
2093 #endif /* RADV_PRIVATE_H */