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 "main/macros.h"
53 #include "vk_debug_report.h"
55 #include "radv_radeon_winsys.h"
56 #include "ac_binary.h"
57 #include "ac_nir_to_llvm.h"
58 #include "ac_gpu_info.h"
59 #include "ac_surface.h"
60 #include "ac_llvm_build.h"
61 #include "ac_llvm_util.h"
62 #include "radv_descriptor_set.h"
63 #include "radv_extensions.h"
66 #include <llvm-c/TargetMachine.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/vk_icd.h>
78 #include <vulkan/vk_android_native_buffer.h>
80 #include "radv_entrypoints.h"
82 #include "wsi_common.h"
83 #include "wsi_common_display.h"
85 #define ATI_VENDOR_ID 0x1002
88 #define MAX_VERTEX_ATTRIBS 32
90 #define MAX_VIEWPORTS 16
91 #define MAX_SCISSORS 16
92 #define MAX_DISCARD_RECTANGLES 4
93 #define MAX_PUSH_CONSTANTS_SIZE 128
94 #define MAX_PUSH_DESCRIPTORS 32
95 #define MAX_DYNAMIC_UNIFORM_BUFFERS 16
96 #define MAX_DYNAMIC_STORAGE_BUFFERS 8
97 #define MAX_DYNAMIC_BUFFERS (MAX_DYNAMIC_UNIFORM_BUFFERS + MAX_DYNAMIC_STORAGE_BUFFERS)
98 #define MAX_SAMPLES_LOG2 4
99 #define NUM_META_FS_KEYS 12
100 #define RADV_MAX_DRM_DEVICES 8
103 #define NUM_DEPTH_CLEAR_PIPELINES 3
106 * This is the point we switch from using CP to compute shader
107 * for certain buffer operations.
109 #define RADV_BUFFER_OPS_CS_THRESHOLD 4096
111 #define RADV_BUFFER_UPDATE_THRESHOLD 1024
115 RADV_MEM_HEAP_VRAM_CPU_ACCESS
,
122 RADV_MEM_TYPE_GTT_WRITE_COMBINE
,
123 RADV_MEM_TYPE_VRAM_CPU_ACCESS
,
124 RADV_MEM_TYPE_GTT_CACHED
,
128 #define radv_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
130 static inline uint32_t
131 align_u32(uint32_t v
, uint32_t a
)
133 assert(a
!= 0 && a
== (a
& -a
));
134 return (v
+ a
- 1) & ~(a
- 1);
137 static inline uint32_t
138 align_u32_npot(uint32_t v
, uint32_t a
)
140 return (v
+ a
- 1) / a
* a
;
143 static inline uint64_t
144 align_u64(uint64_t v
, uint64_t a
)
146 assert(a
!= 0 && a
== (a
& -a
));
147 return (v
+ a
- 1) & ~(a
- 1);
150 static inline int32_t
151 align_i32(int32_t v
, int32_t a
)
153 assert(a
!= 0 && a
== (a
& -a
));
154 return (v
+ a
- 1) & ~(a
- 1);
157 /** Alignment must be a power of 2. */
159 radv_is_aligned(uintmax_t n
, uintmax_t a
)
161 assert(a
== (a
& -a
));
162 return (n
& (a
- 1)) == 0;
165 static inline uint32_t
166 round_up_u32(uint32_t v
, uint32_t a
)
168 return (v
+ a
- 1) / a
;
171 static inline uint64_t
172 round_up_u64(uint64_t v
, uint64_t a
)
174 return (v
+ a
- 1) / a
;
177 static inline uint32_t
178 radv_minify(uint32_t n
, uint32_t levels
)
180 if (unlikely(n
== 0))
183 return MAX2(n
>> levels
, 1);
186 radv_clamp_f(float f
, float min
, float max
)
199 radv_clear_mask(uint32_t *inout_mask
, uint32_t clear_mask
)
201 if (*inout_mask
& clear_mask
) {
202 *inout_mask
&= ~clear_mask
;
209 #define for_each_bit(b, dword) \
210 for (uint32_t __dword = (dword); \
211 (b) = __builtin_ffs(__dword) - 1, __dword; \
212 __dword &= ~(1 << (b)))
214 #define typed_memcpy(dest, src, count) ({ \
215 STATIC_ASSERT(sizeof(*src) == sizeof(*dest)); \
216 memcpy((dest), (src), (count) * sizeof(*(src))); \
219 /* Whenever we generate an error, pass it through this function. Useful for
220 * debugging, where we can break on it. Only call at error site, not when
221 * propagating errors. Might be useful to plug in a stack trace here.
224 struct radv_instance
;
226 VkResult
__vk_errorf(struct radv_instance
*instance
, VkResult error
, const char *file
, int line
, const char *format
, ...);
228 #define vk_error(instance, error) __vk_errorf(instance, error, __FILE__, __LINE__, NULL);
229 #define vk_errorf(instance, error, format, ...) __vk_errorf(instance, error, __FILE__, __LINE__, format, ## __VA_ARGS__);
231 void __radv_finishme(const char *file
, int line
, const char *format
, ...)
232 radv_printflike(3, 4);
233 void radv_loge(const char *format
, ...) radv_printflike(1, 2);
234 void radv_loge_v(const char *format
, va_list va
);
235 void radv_logi(const char *format
, ...) radv_printflike(1, 2);
236 void radv_logi_v(const char *format
, va_list va
);
239 * Print a FINISHME message, including its source location.
241 #define radv_finishme(format, ...) \
243 static bool reported = false; \
245 __radv_finishme(__FILE__, __LINE__, format, ##__VA_ARGS__); \
250 /* A non-fatal assert. Useful for debugging. */
252 #define radv_assert(x) ({ \
253 if (unlikely(!(x))) \
254 fprintf(stderr, "%s:%d ASSERT: %s\n", __FILE__, __LINE__, #x); \
257 #define radv_assert(x)
260 #define stub_return(v) \
262 radv_finishme("stub %s", __func__); \
268 radv_finishme("stub %s", __func__); \
272 void *radv_lookup_entrypoint_unchecked(const char *name
);
273 void *radv_lookup_entrypoint_checked(const char *name
,
274 uint32_t core_version
,
275 const struct radv_instance_extension_table
*instance
,
276 const struct radv_device_extension_table
*device
);
278 struct radv_physical_device
{
279 VK_LOADER_DATA _loader_data
;
281 struct radv_instance
* instance
;
283 struct radeon_winsys
*ws
;
284 struct radeon_info rad_info
;
286 char name
[VK_MAX_PHYSICAL_DEVICE_NAME_SIZE
];
287 uint8_t driver_uuid
[VK_UUID_SIZE
];
288 uint8_t device_uuid
[VK_UUID_SIZE
];
289 uint8_t cache_uuid
[VK_UUID_SIZE
];
293 struct wsi_device wsi_device
;
295 bool has_rbplus
; /* if RB+ register exist */
296 bool rbplus_allowed
; /* if RB+ is allowed */
297 bool has_clear_state
;
298 bool cpdma_prefetch_writes_memory
;
299 bool has_scissor_bug
;
301 bool has_out_of_order_rast
;
302 bool out_of_order_rast_allowed
;
304 /* Whether DCC should be enabled for MSAA textures. */
305 bool dcc_msaa_allowed
;
307 /* This is the drivers on-disk cache used as a fallback as opposed to
308 * the pipeline cache defined by apps.
310 struct disk_cache
* disk_cache
;
312 VkPhysicalDeviceMemoryProperties memory_properties
;
313 enum radv_mem_type mem_type_indices
[RADV_MEM_TYPE_COUNT
];
315 drmPciBusInfo bus_info
;
317 struct radv_device_extension_table supported_extensions
;
320 struct radv_instance
{
321 VK_LOADER_DATA _loader_data
;
323 VkAllocationCallbacks alloc
;
326 int physicalDeviceCount
;
327 struct radv_physical_device physicalDevices
[RADV_MAX_DRM_DEVICES
];
329 uint64_t debug_flags
;
330 uint64_t perftest_flags
;
332 struct vk_debug_report_instance debug_report_callbacks
;
334 struct radv_instance_extension_table enabled_extensions
;
337 VkResult
radv_init_wsi(struct radv_physical_device
*physical_device
);
338 void radv_finish_wsi(struct radv_physical_device
*physical_device
);
340 bool radv_instance_extension_supported(const char *name
);
341 uint32_t radv_physical_device_api_version(struct radv_physical_device
*dev
);
342 bool radv_physical_device_extension_supported(struct radv_physical_device
*dev
,
347 struct radv_pipeline_cache
{
348 struct radv_device
* device
;
349 pthread_mutex_t mutex
;
353 uint32_t kernel_count
;
354 struct cache_entry
** hash_table
;
357 VkAllocationCallbacks alloc
;
360 struct radv_pipeline_key
{
361 uint32_t instance_rate_inputs
;
362 uint32_t instance_rate_divisors
[MAX_VERTEX_ATTRIBS
];
363 uint64_t vertex_alpha_adjust
;
364 unsigned tess_input_vertices
;
368 uint8_t log2_ps_iter_samples
;
370 uint32_t has_multiview_view_index
: 1;
371 uint32_t optimisations_disabled
: 1;
375 radv_pipeline_cache_init(struct radv_pipeline_cache
*cache
,
376 struct radv_device
*device
);
378 radv_pipeline_cache_finish(struct radv_pipeline_cache
*cache
);
380 radv_pipeline_cache_load(struct radv_pipeline_cache
*cache
,
381 const void *data
, size_t size
);
383 struct radv_shader_variant
;
386 radv_create_shader_variants_from_pipeline_cache(struct radv_device
*device
,
387 struct radv_pipeline_cache
*cache
,
388 const unsigned char *sha1
,
389 struct radv_shader_variant
**variants
);
392 radv_pipeline_cache_insert_shaders(struct radv_device
*device
,
393 struct radv_pipeline_cache
*cache
,
394 const unsigned char *sha1
,
395 struct radv_shader_variant
**variants
,
396 const void *const *codes
,
397 const unsigned *code_sizes
);
399 enum radv_blit_ds_layout
{
400 RADV_BLIT_DS_LAYOUT_TILE_ENABLE
,
401 RADV_BLIT_DS_LAYOUT_TILE_DISABLE
,
402 RADV_BLIT_DS_LAYOUT_COUNT
,
405 static inline enum radv_blit_ds_layout
radv_meta_blit_ds_to_type(VkImageLayout layout
)
407 return (layout
== VK_IMAGE_LAYOUT_GENERAL
) ? RADV_BLIT_DS_LAYOUT_TILE_DISABLE
: RADV_BLIT_DS_LAYOUT_TILE_ENABLE
;
410 static inline VkImageLayout
radv_meta_blit_ds_to_layout(enum radv_blit_ds_layout ds_layout
)
412 return ds_layout
== RADV_BLIT_DS_LAYOUT_TILE_ENABLE
? VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
: VK_IMAGE_LAYOUT_GENERAL
;
415 enum radv_meta_dst_layout
{
416 RADV_META_DST_LAYOUT_GENERAL
,
417 RADV_META_DST_LAYOUT_OPTIMAL
,
418 RADV_META_DST_LAYOUT_COUNT
,
421 static inline enum radv_meta_dst_layout
radv_meta_dst_layout_from_layout(VkImageLayout layout
)
423 return (layout
== VK_IMAGE_LAYOUT_GENERAL
) ? RADV_META_DST_LAYOUT_GENERAL
: RADV_META_DST_LAYOUT_OPTIMAL
;
426 static inline VkImageLayout
radv_meta_dst_layout_to_layout(enum radv_meta_dst_layout layout
)
428 return layout
== RADV_META_DST_LAYOUT_OPTIMAL
? VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
: VK_IMAGE_LAYOUT_GENERAL
;
431 struct radv_meta_state
{
432 VkAllocationCallbacks alloc
;
434 struct radv_pipeline_cache cache
;
437 * For on-demand pipeline creation, makes sure that
438 * only one thread tries to build a pipeline at the same time.
443 * Use array element `i` for images with `2^i` samples.
446 VkRenderPass render_pass
[NUM_META_FS_KEYS
];
447 VkPipeline color_pipelines
[NUM_META_FS_KEYS
];
449 VkRenderPass depthstencil_rp
;
450 VkPipeline depth_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
451 VkPipeline stencil_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
452 VkPipeline depthstencil_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
453 } clear
[1 + MAX_SAMPLES_LOG2
];
455 VkPipelineLayout clear_color_p_layout
;
456 VkPipelineLayout clear_depth_p_layout
;
458 VkRenderPass render_pass
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
460 /** Pipeline that blits from a 1D image. */
461 VkPipeline pipeline_1d_src
[NUM_META_FS_KEYS
];
463 /** Pipeline that blits from a 2D image. */
464 VkPipeline pipeline_2d_src
[NUM_META_FS_KEYS
];
466 /** Pipeline that blits from a 3D image. */
467 VkPipeline pipeline_3d_src
[NUM_META_FS_KEYS
];
469 VkRenderPass depth_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
470 VkPipeline depth_only_1d_pipeline
;
471 VkPipeline depth_only_2d_pipeline
;
472 VkPipeline depth_only_3d_pipeline
;
474 VkRenderPass stencil_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
475 VkPipeline stencil_only_1d_pipeline
;
476 VkPipeline stencil_only_2d_pipeline
;
477 VkPipeline stencil_only_3d_pipeline
;
478 VkPipelineLayout pipeline_layout
;
479 VkDescriptorSetLayout ds_layout
;
483 VkPipelineLayout p_layouts
[5];
484 VkDescriptorSetLayout ds_layouts
[5];
485 VkPipeline pipelines
[5][NUM_META_FS_KEYS
];
487 VkPipeline depth_only_pipeline
[5];
489 VkPipeline stencil_only_pipeline
[5];
490 } blit2d
[1 + MAX_SAMPLES_LOG2
];
492 VkRenderPass blit2d_render_passes
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
493 VkRenderPass blit2d_depth_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
494 VkRenderPass blit2d_stencil_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
497 VkPipelineLayout img_p_layout
;
498 VkDescriptorSetLayout img_ds_layout
;
500 VkPipeline pipeline_3d
;
503 VkPipelineLayout img_p_layout
;
504 VkDescriptorSetLayout img_ds_layout
;
506 VkPipeline pipeline_3d
;
509 VkPipelineLayout img_p_layout
;
510 VkDescriptorSetLayout img_ds_layout
;
514 VkPipelineLayout img_p_layout
;
515 VkDescriptorSetLayout img_ds_layout
;
517 VkPipeline pipeline_3d
;
520 VkPipelineLayout img_p_layout
;
521 VkDescriptorSetLayout img_ds_layout
;
525 VkPipelineLayout img_p_layout
;
526 VkDescriptorSetLayout img_ds_layout
;
528 VkPipeline pipeline_3d
;
531 VkPipelineLayout img_p_layout
;
532 VkDescriptorSetLayout img_ds_layout
;
537 VkPipelineLayout p_layout
;
538 VkPipeline pipeline
[NUM_META_FS_KEYS
];
539 VkRenderPass pass
[NUM_META_FS_KEYS
];
543 VkDescriptorSetLayout ds_layout
;
544 VkPipelineLayout p_layout
;
547 VkPipeline i_pipeline
;
548 VkPipeline srgb_pipeline
;
549 } rc
[MAX_SAMPLES_LOG2
];
553 VkDescriptorSetLayout ds_layout
;
554 VkPipelineLayout p_layout
;
557 VkRenderPass render_pass
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
558 VkPipeline pipeline
[NUM_META_FS_KEYS
];
559 } rc
[MAX_SAMPLES_LOG2
];
563 VkPipelineLayout p_layout
;
564 VkPipeline decompress_pipeline
;
565 VkPipeline resummarize_pipeline
;
567 } depth_decomp
[1 + MAX_SAMPLES_LOG2
];
570 VkPipelineLayout p_layout
;
571 VkPipeline cmask_eliminate_pipeline
;
572 VkPipeline fmask_decompress_pipeline
;
573 VkPipeline dcc_decompress_pipeline
;
576 VkDescriptorSetLayout dcc_decompress_compute_ds_layout
;
577 VkPipelineLayout dcc_decompress_compute_p_layout
;
578 VkPipeline dcc_decompress_compute_pipeline
;
582 VkPipelineLayout fill_p_layout
;
583 VkPipelineLayout copy_p_layout
;
584 VkDescriptorSetLayout fill_ds_layout
;
585 VkDescriptorSetLayout copy_ds_layout
;
586 VkPipeline fill_pipeline
;
587 VkPipeline copy_pipeline
;
591 VkDescriptorSetLayout ds_layout
;
592 VkPipelineLayout p_layout
;
593 VkPipeline occlusion_query_pipeline
;
594 VkPipeline pipeline_statistics_query_pipeline
;
599 #define RADV_QUEUE_GENERAL 0
600 #define RADV_QUEUE_COMPUTE 1
601 #define RADV_QUEUE_TRANSFER 2
603 #define RADV_MAX_QUEUE_FAMILIES 3
605 enum ring_type
radv_queue_family_to_ring(int f
);
608 VK_LOADER_DATA _loader_data
;
609 struct radv_device
* device
;
610 struct radeon_winsys_ctx
*hw_ctx
;
611 enum radeon_ctx_priority priority
;
612 uint32_t queue_family_index
;
614 VkDeviceQueueCreateFlags flags
;
616 uint32_t scratch_size
;
617 uint32_t compute_scratch_size
;
618 uint32_t esgs_ring_size
;
619 uint32_t gsvs_ring_size
;
621 bool has_sample_positions
;
623 struct radeon_winsys_bo
*scratch_bo
;
624 struct radeon_winsys_bo
*descriptor_bo
;
625 struct radeon_winsys_bo
*compute_scratch_bo
;
626 struct radeon_winsys_bo
*esgs_ring_bo
;
627 struct radeon_winsys_bo
*gsvs_ring_bo
;
628 struct radeon_winsys_bo
*tess_rings_bo
;
629 struct radeon_cmdbuf
*initial_preamble_cs
;
630 struct radeon_cmdbuf
*initial_full_flush_preamble_cs
;
631 struct radeon_cmdbuf
*continue_preamble_cs
;
634 struct radv_bo_list
{
635 struct radv_winsys_bo_list list
;
637 pthread_mutex_t mutex
;
641 VK_LOADER_DATA _loader_data
;
643 VkAllocationCallbacks alloc
;
645 struct radv_instance
* instance
;
646 struct radeon_winsys
*ws
;
648 struct radv_meta_state meta_state
;
650 struct radv_queue
*queues
[RADV_MAX_QUEUE_FAMILIES
];
651 int queue_count
[RADV_MAX_QUEUE_FAMILIES
];
652 struct radeon_cmdbuf
*empty_cs
[RADV_MAX_QUEUE_FAMILIES
];
654 bool always_use_syncobj
;
655 bool has_distributed_tess
;
658 uint32_t tess_offchip_block_dw_size
;
659 uint32_t scratch_waves
;
660 uint32_t dispatch_initiator
;
662 uint32_t gs_table_depth
;
664 /* MSAA sample locations.
665 * The first index is the sample index.
666 * The second index is the coordinate: X, Y. */
667 float sample_locations_1x
[1][2];
668 float sample_locations_2x
[2][2];
669 float sample_locations_4x
[4][2];
670 float sample_locations_8x
[8][2];
671 float sample_locations_16x
[16][2];
674 uint32_t gfx_init_size_dw
;
675 struct radeon_winsys_bo
*gfx_init
;
677 struct radeon_winsys_bo
*trace_bo
;
678 uint32_t *trace_id_ptr
;
680 /* Whether to keep shader debug info, for tracing or VK_AMD_shader_info */
681 bool keep_shader_info
;
683 struct radv_physical_device
*physical_device
;
685 /* Backup in-memory cache to be used if the app doesn't provide one */
686 struct radv_pipeline_cache
* mem_cache
;
689 * use different counters so MSAA MRTs get consecutive surface indices,
690 * even if MASK is allocated in between.
692 uint32_t image_mrt_offset_counter
;
693 uint32_t fmask_mrt_offset_counter
;
694 struct list_head shader_slabs
;
695 mtx_t shader_slab_mutex
;
697 /* For detecting VM faults reported by dmesg. */
698 uint64_t dmesg_timestamp
;
700 struct radv_device_extension_table enabled_extensions
;
702 /* Whether the driver uses a global BO list. */
703 bool use_global_bo_list
;
705 struct radv_bo_list bo_list
;
707 /* Whether anisotropy is forced with RADV_TEX_ANISO (-1 is disabled). */
711 struct radv_device_memory
{
712 struct radeon_winsys_bo
*bo
;
713 /* for dedicated allocations */
714 struct radv_image
*image
;
715 struct radv_buffer
*buffer
;
717 VkDeviceSize map_size
;
723 struct radv_descriptor_range
{
728 struct radv_descriptor_set
{
729 const struct radv_descriptor_set_layout
*layout
;
732 struct radeon_winsys_bo
*bo
;
734 uint32_t *mapped_ptr
;
735 struct radv_descriptor_range
*dynamic_descriptors
;
737 struct radeon_winsys_bo
*descriptors
[0];
740 struct radv_push_descriptor_set
742 struct radv_descriptor_set set
;
746 struct radv_descriptor_pool_entry
{
749 struct radv_descriptor_set
*set
;
752 struct radv_descriptor_pool
{
753 struct radeon_winsys_bo
*bo
;
755 uint64_t current_offset
;
758 uint8_t *host_memory_base
;
759 uint8_t *host_memory_ptr
;
760 uint8_t *host_memory_end
;
762 uint32_t entry_count
;
763 uint32_t max_entry_count
;
764 struct radv_descriptor_pool_entry entries
[0];
767 struct radv_descriptor_update_template_entry
{
768 VkDescriptorType descriptor_type
;
770 /* The number of descriptors to update */
771 uint32_t descriptor_count
;
773 /* Into mapped_ptr or dynamic_descriptors, in units of the respective array */
776 /* In dwords. Not valid/used for dynamic descriptors */
779 uint32_t buffer_offset
;
781 /* Only valid for combined image samplers and samplers */
782 uint16_t has_sampler
;
788 /* For push descriptors */
789 const uint32_t *immutable_samplers
;
792 struct radv_descriptor_update_template
{
793 uint32_t entry_count
;
794 VkPipelineBindPoint bind_point
;
795 struct radv_descriptor_update_template_entry entry
[0];
801 VkBufferUsageFlags usage
;
802 VkBufferCreateFlags flags
;
805 struct radeon_winsys_bo
* bo
;
811 enum radv_dynamic_state_bits
{
812 RADV_DYNAMIC_VIEWPORT
= 1 << 0,
813 RADV_DYNAMIC_SCISSOR
= 1 << 1,
814 RADV_DYNAMIC_LINE_WIDTH
= 1 << 2,
815 RADV_DYNAMIC_DEPTH_BIAS
= 1 << 3,
816 RADV_DYNAMIC_BLEND_CONSTANTS
= 1 << 4,
817 RADV_DYNAMIC_DEPTH_BOUNDS
= 1 << 5,
818 RADV_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6,
819 RADV_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7,
820 RADV_DYNAMIC_STENCIL_REFERENCE
= 1 << 8,
821 RADV_DYNAMIC_DISCARD_RECTANGLE
= 1 << 9,
822 RADV_DYNAMIC_ALL
= (1 << 10) - 1,
825 enum radv_cmd_dirty_bits
{
826 /* Keep the dynamic state dirty bits in sync with
827 * enum radv_dynamic_state_bits */
828 RADV_CMD_DIRTY_DYNAMIC_VIEWPORT
= 1 << 0,
829 RADV_CMD_DIRTY_DYNAMIC_SCISSOR
= 1 << 1,
830 RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH
= 1 << 2,
831 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS
= 1 << 3,
832 RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS
= 1 << 4,
833 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS
= 1 << 5,
834 RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6,
835 RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7,
836 RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE
= 1 << 8,
837 RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE
= 1 << 9,
838 RADV_CMD_DIRTY_DYNAMIC_ALL
= (1 << 10) - 1,
839 RADV_CMD_DIRTY_PIPELINE
= 1 << 10,
840 RADV_CMD_DIRTY_INDEX_BUFFER
= 1 << 11,
841 RADV_CMD_DIRTY_FRAMEBUFFER
= 1 << 12,
842 RADV_CMD_DIRTY_VERTEX_BUFFER
= 1 << 13,
845 enum radv_cmd_flush_bits
{
846 RADV_CMD_FLAG_INV_ICACHE
= 1 << 0,
847 /* SMEM L1, other names: KCACHE, constant cache, DCACHE, data cache */
848 RADV_CMD_FLAG_INV_SMEM_L1
= 1 << 1,
849 /* VMEM L1 can optionally be bypassed (GLC=1). Other names: TC L1 */
850 RADV_CMD_FLAG_INV_VMEM_L1
= 1 << 2,
851 /* Used by everything except CB/DB, can be bypassed (SLC=1). Other names: TC L2 */
852 RADV_CMD_FLAG_INV_GLOBAL_L2
= 1 << 3,
853 /* Same as above, but only writes back and doesn't invalidate */
854 RADV_CMD_FLAG_WRITEBACK_GLOBAL_L2
= 1 << 4,
855 /* Framebuffer caches */
856 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
= 1 << 5,
857 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
= 1 << 6,
858 RADV_CMD_FLAG_FLUSH_AND_INV_DB
= 1 << 7,
859 RADV_CMD_FLAG_FLUSH_AND_INV_CB
= 1 << 8,
860 /* Engine synchronization. */
861 RADV_CMD_FLAG_VS_PARTIAL_FLUSH
= 1 << 9,
862 RADV_CMD_FLAG_PS_PARTIAL_FLUSH
= 1 << 10,
863 RADV_CMD_FLAG_CS_PARTIAL_FLUSH
= 1 << 11,
864 RADV_CMD_FLAG_VGT_FLUSH
= 1 << 12,
865 /* Pipeline query controls. */
866 RADV_CMD_FLAG_START_PIPELINE_STATS
= 1 << 13,
867 RADV_CMD_FLAG_STOP_PIPELINE_STATS
= 1 << 14,
869 RADV_CMD_FLUSH_AND_INV_FRAMEBUFFER
= (RADV_CMD_FLAG_FLUSH_AND_INV_CB
|
870 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
|
871 RADV_CMD_FLAG_FLUSH_AND_INV_DB
|
872 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
)
875 struct radv_vertex_binding
{
876 struct radv_buffer
* buffer
;
880 struct radv_viewport_state
{
882 VkViewport viewports
[MAX_VIEWPORTS
];
885 struct radv_scissor_state
{
887 VkRect2D scissors
[MAX_SCISSORS
];
890 struct radv_discard_rectangle_state
{
892 VkRect2D rectangles
[MAX_DISCARD_RECTANGLES
];
895 struct radv_dynamic_state
{
897 * Bitmask of (1 << VK_DYNAMIC_STATE_*).
898 * Defines the set of saved dynamic state.
902 struct radv_viewport_state viewport
;
904 struct radv_scissor_state scissor
;
914 float blend_constants
[4];
924 } stencil_compare_mask
;
929 } stencil_write_mask
;
936 struct radv_discard_rectangle_state discard_rectangle
;
939 extern const struct radv_dynamic_state default_dynamic_state
;
942 radv_get_debug_option_name(int id
);
945 radv_get_perftest_option_name(int id
);
948 * Attachment state when recording a renderpass instance.
950 * The clear value is valid only if there exists a pending clear.
952 struct radv_attachment_state
{
953 VkImageAspectFlags pending_clear_aspects
;
954 uint32_t cleared_views
;
955 VkClearValue clear_value
;
956 VkImageLayout current_layout
;
959 struct radv_descriptor_state
{
960 struct radv_descriptor_set
*sets
[MAX_SETS
];
963 struct radv_push_descriptor_set push_set
;
965 uint32_t dynamic_buffers
[4 * MAX_DYNAMIC_BUFFERS
];
968 struct radv_cmd_state
{
969 /* Vertex descriptors */
976 uint32_t prefetch_L2_mask
;
978 struct radv_pipeline
* pipeline
;
979 struct radv_pipeline
* emitted_pipeline
;
980 struct radv_pipeline
* compute_pipeline
;
981 struct radv_pipeline
* emitted_compute_pipeline
;
982 struct radv_framebuffer
* framebuffer
;
983 struct radv_render_pass
* pass
;
984 const struct radv_subpass
* subpass
;
985 struct radv_dynamic_state dynamic
;
986 struct radv_attachment_state
* attachments
;
987 VkRect2D render_area
;
990 struct radv_buffer
*index_buffer
;
991 uint64_t index_offset
;
993 uint32_t max_index_count
;
995 int32_t last_index_type
;
997 int32_t last_primitive_reset_en
;
998 uint32_t last_primitive_reset_index
;
999 enum radv_cmd_flush_bits flush_bits
;
1000 unsigned active_occlusion_queries
;
1001 bool perfect_occlusion_queries_enabled
;
1002 unsigned active_pipeline_queries
;
1005 uint32_t last_ia_multi_vgt_param
;
1007 uint32_t last_num_instances
;
1008 uint32_t last_first_instance
;
1009 uint32_t last_vertex_offset
;
1011 /* Whether CP DMA is busy/idle. */
1014 /* Conditional rendering info. */
1015 int predication_type
; /* -1: disabled, 0: normal, 1: inverted */
1016 uint64_t predication_va
;
1019 struct radv_cmd_pool
{
1020 VkAllocationCallbacks alloc
;
1021 struct list_head cmd_buffers
;
1022 struct list_head free_cmd_buffers
;
1023 uint32_t queue_family_index
;
1026 struct radv_cmd_buffer_upload
{
1030 struct radeon_winsys_bo
*upload_bo
;
1031 struct list_head list
;
1034 enum radv_cmd_buffer_status
{
1035 RADV_CMD_BUFFER_STATUS_INVALID
,
1036 RADV_CMD_BUFFER_STATUS_INITIAL
,
1037 RADV_CMD_BUFFER_STATUS_RECORDING
,
1038 RADV_CMD_BUFFER_STATUS_EXECUTABLE
,
1039 RADV_CMD_BUFFER_STATUS_PENDING
,
1042 struct radv_cmd_buffer
{
1043 VK_LOADER_DATA _loader_data
;
1045 struct radv_device
* device
;
1047 struct radv_cmd_pool
* pool
;
1048 struct list_head pool_link
;
1050 VkCommandBufferUsageFlags usage_flags
;
1051 VkCommandBufferLevel level
;
1052 enum radv_cmd_buffer_status status
;
1053 struct radeon_cmdbuf
*cs
;
1054 struct radv_cmd_state state
;
1055 struct radv_vertex_binding vertex_bindings
[MAX_VBS
];
1056 uint32_t queue_family_index
;
1058 uint8_t push_constants
[MAX_PUSH_CONSTANTS_SIZE
];
1059 VkShaderStageFlags push_constant_stages
;
1060 struct radv_descriptor_set meta_push_descriptors
;
1062 struct radv_descriptor_state descriptors
[VK_PIPELINE_BIND_POINT_RANGE_SIZE
];
1064 struct radv_cmd_buffer_upload upload
;
1066 uint32_t scratch_size_needed
;
1067 uint32_t compute_scratch_size_needed
;
1068 uint32_t esgs_ring_size_needed
;
1069 uint32_t gsvs_ring_size_needed
;
1070 bool tess_rings_needed
;
1071 bool sample_positions_needed
;
1073 VkResult record_result
;
1075 uint32_t gfx9_fence_offset
;
1076 struct radeon_winsys_bo
*gfx9_fence_bo
;
1077 uint32_t gfx9_fence_idx
;
1078 uint64_t gfx9_eop_bug_va
;
1081 * Whether a query pool has been resetted and we have to flush caches.
1083 bool pending_reset_query
;
1088 bool radv_cmd_buffer_uses_mec(struct radv_cmd_buffer
*cmd_buffer
);
1090 void si_emit_graphics(struct radv_physical_device
*physical_device
,
1091 struct radeon_cmdbuf
*cs
);
1092 void si_emit_compute(struct radv_physical_device
*physical_device
,
1093 struct radeon_cmdbuf
*cs
);
1095 void cik_create_gfx_config(struct radv_device
*device
);
1097 void si_write_viewport(struct radeon_cmdbuf
*cs
, int first_vp
,
1098 int count
, const VkViewport
*viewports
);
1099 void si_write_scissors(struct radeon_cmdbuf
*cs
, int first
,
1100 int count
, const VkRect2D
*scissors
,
1101 const VkViewport
*viewports
, bool can_use_guardband
);
1102 uint32_t si_get_ia_multi_vgt_param(struct radv_cmd_buffer
*cmd_buffer
,
1103 bool instanced_draw
, bool indirect_draw
,
1104 uint32_t draw_vertex_count
);
1105 void si_cs_emit_write_event_eop(struct radeon_cmdbuf
*cs
,
1106 enum chip_class chip_class
,
1108 unsigned event
, unsigned event_flags
,
1113 uint64_t gfx9_eop_bug_va
);
1115 void si_emit_wait_fence(struct radeon_cmdbuf
*cs
,
1116 uint64_t va
, uint32_t ref
,
1118 void si_cs_emit_cache_flush(struct radeon_cmdbuf
*cs
,
1119 enum chip_class chip_class
,
1120 uint32_t *fence_ptr
, uint64_t va
,
1122 enum radv_cmd_flush_bits flush_bits
,
1123 uint64_t gfx9_eop_bug_va
);
1124 void si_emit_cache_flush(struct radv_cmd_buffer
*cmd_buffer
);
1125 void si_emit_set_predication_state(struct radv_cmd_buffer
*cmd_buffer
,
1126 bool inverted
, uint64_t va
);
1127 void si_cp_dma_buffer_copy(struct radv_cmd_buffer
*cmd_buffer
,
1128 uint64_t src_va
, uint64_t dest_va
,
1130 void si_cp_dma_prefetch(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
1132 void si_cp_dma_clear_buffer(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
1133 uint64_t size
, unsigned value
);
1134 void si_cp_dma_wait_for_idle(struct radv_cmd_buffer
*cmd_buffer
);
1136 void radv_set_db_count_control(struct radv_cmd_buffer
*cmd_buffer
);
1138 radv_cmd_buffer_upload_alloc(struct radv_cmd_buffer
*cmd_buffer
,
1141 unsigned *out_offset
,
1144 radv_cmd_buffer_set_subpass(struct radv_cmd_buffer
*cmd_buffer
,
1145 const struct radv_subpass
*subpass
,
1148 radv_cmd_buffer_upload_data(struct radv_cmd_buffer
*cmd_buffer
,
1149 unsigned size
, unsigned alignmnet
,
1150 const void *data
, unsigned *out_offset
);
1152 void radv_cmd_buffer_clear_subpass(struct radv_cmd_buffer
*cmd_buffer
);
1153 void radv_cmd_buffer_resolve_subpass(struct radv_cmd_buffer
*cmd_buffer
);
1154 void radv_cmd_buffer_resolve_subpass_cs(struct radv_cmd_buffer
*cmd_buffer
);
1155 void radv_cmd_buffer_resolve_subpass_fs(struct radv_cmd_buffer
*cmd_buffer
);
1156 void radv_cayman_emit_msaa_sample_locs(struct radeon_cmdbuf
*cs
, int nr_samples
);
1157 unsigned radv_cayman_get_maxdist(int log_samples
);
1158 void radv_device_init_msaa(struct radv_device
*device
);
1160 void radv_update_ds_clear_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1161 struct radv_image
*image
,
1162 VkClearDepthStencilValue ds_clear_value
,
1163 VkImageAspectFlags aspects
);
1165 void radv_update_color_clear_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1166 struct radv_image
*image
,
1168 uint32_t color_values
[2]);
1170 void radv_set_dcc_need_cmask_elim_pred(struct radv_cmd_buffer
*cmd_buffer
,
1171 struct radv_image
*image
,
1173 uint32_t radv_fill_buffer(struct radv_cmd_buffer
*cmd_buffer
,
1174 struct radeon_winsys_bo
*bo
,
1175 uint64_t offset
, uint64_t size
, uint32_t value
);
1176 void radv_cmd_buffer_trace_emit(struct radv_cmd_buffer
*cmd_buffer
);
1177 bool radv_get_memory_fd(struct radv_device
*device
,
1178 struct radv_device_memory
*memory
,
1182 radv_emit_shader_pointer_head(struct radeon_cmdbuf
*cs
,
1183 unsigned sh_offset
, unsigned pointer_count
,
1184 bool use_32bit_pointers
)
1186 radeon_emit(cs
, PKT3(PKT3_SET_SH_REG
, pointer_count
* (use_32bit_pointers
? 1 : 2), 0));
1187 radeon_emit(cs
, (sh_offset
- SI_SH_REG_OFFSET
) >> 2);
1191 radv_emit_shader_pointer_body(struct radv_device
*device
,
1192 struct radeon_cmdbuf
*cs
,
1193 uint64_t va
, bool use_32bit_pointers
)
1195 radeon_emit(cs
, va
);
1197 if (use_32bit_pointers
) {
1199 (va
>> 32) == device
->physical_device
->rad_info
.address32_hi
);
1201 radeon_emit(cs
, va
>> 32);
1206 radv_emit_shader_pointer(struct radv_device
*device
,
1207 struct radeon_cmdbuf
*cs
,
1208 uint32_t sh_offset
, uint64_t va
, bool global
)
1210 bool use_32bit_pointers
= HAVE_32BIT_POINTERS
&& !global
;
1212 radv_emit_shader_pointer_head(cs
, sh_offset
, 1, use_32bit_pointers
);
1213 radv_emit_shader_pointer_body(device
, cs
, va
, use_32bit_pointers
);
1216 static inline struct radv_descriptor_state
*
1217 radv_get_descriptors_state(struct radv_cmd_buffer
*cmd_buffer
,
1218 VkPipelineBindPoint bind_point
)
1220 assert(bind_point
== VK_PIPELINE_BIND_POINT_GRAPHICS
||
1221 bind_point
== VK_PIPELINE_BIND_POINT_COMPUTE
);
1222 return &cmd_buffer
->descriptors
[bind_point
];
1226 * Takes x,y,z as exact numbers of invocations, instead of blocks.
1228 * Limitations: Can't call normal dispatch functions without binding or rebinding
1229 * the compute pipeline.
1231 void radv_unaligned_dispatch(
1232 struct radv_cmd_buffer
*cmd_buffer
,
1238 struct radeon_winsys_bo
*bo
;
1242 struct radv_shader_module
;
1244 #define RADV_HASH_SHADER_IS_GEOM_COPY_SHADER (1 << 0)
1245 #define RADV_HASH_SHADER_SISCHED (1 << 1)
1246 #define RADV_HASH_SHADER_UNSAFE_MATH (1 << 2)
1248 radv_hash_shaders(unsigned char *hash
,
1249 const VkPipelineShaderStageCreateInfo
**stages
,
1250 const struct radv_pipeline_layout
*layout
,
1251 const struct radv_pipeline_key
*key
,
1254 static inline gl_shader_stage
1255 vk_to_mesa_shader_stage(VkShaderStageFlagBits vk_stage
)
1257 assert(__builtin_popcount(vk_stage
) == 1);
1258 return ffs(vk_stage
) - 1;
1261 static inline VkShaderStageFlagBits
1262 mesa_to_vk_shader_stage(gl_shader_stage mesa_stage
)
1264 return (1 << mesa_stage
);
1267 #define RADV_STAGE_MASK ((1 << MESA_SHADER_STAGES) - 1)
1269 #define radv_foreach_stage(stage, stage_bits) \
1270 for (gl_shader_stage stage, \
1271 __tmp = (gl_shader_stage)((stage_bits) & RADV_STAGE_MASK); \
1272 stage = __builtin_ffs(__tmp) - 1, __tmp; \
1273 __tmp &= ~(1 << (stage)))
1275 extern const VkFormat radv_fs_key_format_exemplars
[NUM_META_FS_KEYS
];
1276 unsigned radv_format_meta_fs_key(VkFormat format
);
1278 struct radv_multisample_state
{
1280 uint32_t pa_sc_line_cntl
;
1281 uint32_t pa_sc_mode_cntl_0
;
1282 uint32_t pa_sc_mode_cntl_1
;
1283 uint32_t pa_sc_aa_config
;
1284 uint32_t pa_sc_aa_mask
[2];
1285 unsigned num_samples
;
1288 struct radv_prim_vertex_count
{
1293 struct radv_vertex_elements_info
{
1294 uint32_t rsrc_word3
[MAX_VERTEX_ATTRIBS
];
1295 uint32_t format_size
[MAX_VERTEX_ATTRIBS
];
1296 uint32_t binding
[MAX_VERTEX_ATTRIBS
];
1297 uint32_t offset
[MAX_VERTEX_ATTRIBS
];
1301 struct radv_ia_multi_vgt_param_helpers
{
1303 bool partial_es_wave
;
1304 uint8_t primgroup_size
;
1305 bool wd_switch_on_eop
;
1306 bool ia_switch_on_eoi
;
1307 bool partial_vs_wave
;
1310 #define SI_GS_PER_ES 128
1312 struct radv_pipeline
{
1313 struct radv_device
* device
;
1314 struct radv_dynamic_state dynamic_state
;
1316 struct radv_pipeline_layout
* layout
;
1318 bool need_indirect_descriptor_sets
;
1319 struct radv_shader_variant
* shaders
[MESA_SHADER_STAGES
];
1320 struct radv_shader_variant
*gs_copy_shader
;
1321 VkShaderStageFlags active_stages
;
1323 struct radeon_cmdbuf cs
;
1325 struct radv_vertex_elements_info vertex_elements
;
1327 uint32_t binding_stride
[MAX_VBS
];
1329 uint32_t user_data_0
[MESA_SHADER_STAGES
];
1332 struct radv_multisample_state ms
;
1333 uint32_t spi_baryc_cntl
;
1334 bool prim_restart_enable
;
1335 unsigned esgs_ring_size
;
1336 unsigned gsvs_ring_size
;
1337 uint32_t vtx_base_sgpr
;
1338 struct radv_ia_multi_vgt_param_helpers ia_multi_vgt_param
;
1339 uint8_t vtx_emit_num
;
1340 struct radv_prim_vertex_count prim_vertex_count
;
1341 bool can_use_guardband
;
1342 uint32_t needed_dynamic_state
;
1343 bool disable_out_of_order_rast_for_occlusion
;
1345 /* Used for rbplus */
1346 uint32_t col_format
;
1347 uint32_t cb_target_mask
;
1352 unsigned scratch_bytes_per_wave
;
1355 static inline bool radv_pipeline_has_gs(const struct radv_pipeline
*pipeline
)
1357 return pipeline
->shaders
[MESA_SHADER_GEOMETRY
] ? true : false;
1360 static inline bool radv_pipeline_has_tess(const struct radv_pipeline
*pipeline
)
1362 return pipeline
->shaders
[MESA_SHADER_TESS_CTRL
] ? true : false;
1365 struct radv_userdata_info
*radv_lookup_user_sgpr(struct radv_pipeline
*pipeline
,
1366 gl_shader_stage stage
,
1369 struct radv_shader_variant
*radv_get_shader(struct radv_pipeline
*pipeline
,
1370 gl_shader_stage stage
);
1372 struct radv_graphics_pipeline_create_info
{
1374 bool db_depth_clear
;
1375 bool db_stencil_clear
;
1376 bool db_depth_disable_expclear
;
1377 bool db_stencil_disable_expclear
;
1378 bool db_flush_depth_inplace
;
1379 bool db_flush_stencil_inplace
;
1380 bool db_resummarize
;
1381 uint32_t custom_blend_mode
;
1385 radv_graphics_pipeline_create(VkDevice device
,
1386 VkPipelineCache cache
,
1387 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
1388 const struct radv_graphics_pipeline_create_info
*extra
,
1389 const VkAllocationCallbacks
*alloc
,
1390 VkPipeline
*pPipeline
);
1392 struct vk_format_description
;
1393 uint32_t radv_translate_buffer_dataformat(const struct vk_format_description
*desc
,
1394 int first_non_void
);
1395 uint32_t radv_translate_buffer_numformat(const struct vk_format_description
*desc
,
1396 int first_non_void
);
1397 uint32_t radv_translate_colorformat(VkFormat format
);
1398 uint32_t radv_translate_color_numformat(VkFormat format
,
1399 const struct vk_format_description
*desc
,
1400 int first_non_void
);
1401 uint32_t radv_colorformat_endian_swap(uint32_t colorformat
);
1402 unsigned radv_translate_colorswap(VkFormat format
, bool do_endian_swap
);
1403 uint32_t radv_translate_dbformat(VkFormat format
);
1404 uint32_t radv_translate_tex_dataformat(VkFormat format
,
1405 const struct vk_format_description
*desc
,
1406 int first_non_void
);
1407 uint32_t radv_translate_tex_numformat(VkFormat format
,
1408 const struct vk_format_description
*desc
,
1409 int first_non_void
);
1410 bool radv_format_pack_clear_color(VkFormat format
,
1411 uint32_t clear_vals
[2],
1412 VkClearColorValue
*value
);
1413 bool radv_is_colorbuffer_format_supported(VkFormat format
, bool *blendable
);
1414 bool radv_dcc_formats_compatible(VkFormat format1
,
1417 struct radv_fmask_info
{
1421 unsigned pitch_in_pixels
;
1422 unsigned bank_height
;
1423 unsigned slice_tile_max
;
1424 unsigned tile_mode_index
;
1425 unsigned tile_swizzle
;
1428 struct radv_cmask_info
{
1432 unsigned slice_tile_max
;
1437 /* The original VkFormat provided by the client. This may not match any
1438 * of the actual surface formats.
1441 VkImageAspectFlags aspects
;
1442 VkImageUsageFlags usage
; /**< Superset of VkImageCreateInfo::usage. */
1443 struct ac_surf_info info
;
1444 VkImageTiling tiling
; /** VkImageCreateInfo::tiling */
1445 VkImageCreateFlags flags
; /** VkImageCreateInfo::flags */
1450 unsigned queue_family_mask
;
1454 /* Set when bound */
1455 struct radeon_winsys_bo
*bo
;
1456 VkDeviceSize offset
;
1457 uint64_t dcc_offset
;
1458 uint64_t htile_offset
;
1459 bool tc_compatible_htile
;
1460 struct radeon_surf surface
;
1462 struct radv_fmask_info fmask
;
1463 struct radv_cmask_info cmask
;
1464 uint64_t clear_value_offset
;
1465 uint64_t dcc_pred_offset
;
1467 /* For VK_ANDROID_native_buffer, the WSI image owns the memory, */
1468 VkDeviceMemory owned_memory
;
1471 /* Whether the image has a htile that is known consistent with the contents of
1473 bool radv_layout_has_htile(const struct radv_image
*image
,
1474 VkImageLayout layout
,
1475 unsigned queue_mask
);
1477 /* Whether the image has a htile that is known consistent with the contents of
1478 * the image and is allowed to be in compressed form.
1480 * If this is false reads that don't use the htile should be able to return
1483 bool radv_layout_is_htile_compressed(const struct radv_image
*image
,
1484 VkImageLayout layout
,
1485 unsigned queue_mask
);
1487 bool radv_layout_can_fast_clear(const struct radv_image
*image
,
1488 VkImageLayout layout
,
1489 unsigned queue_mask
);
1491 bool radv_layout_dcc_compressed(const struct radv_image
*image
,
1492 VkImageLayout layout
,
1493 unsigned queue_mask
);
1496 * Return whether the image has CMASK metadata for color surfaces.
1499 radv_image_has_cmask(const struct radv_image
*image
)
1501 return image
->cmask
.size
;
1505 * Return whether the image has FMASK metadata for color surfaces.
1508 radv_image_has_fmask(const struct radv_image
*image
)
1510 return image
->fmask
.size
;
1514 * Return whether the image has DCC metadata for color surfaces.
1517 radv_image_has_dcc(const struct radv_image
*image
)
1519 return image
->surface
.dcc_size
;
1523 * Return whether DCC metadata is enabled for a level.
1526 radv_dcc_enabled(const struct radv_image
*image
, unsigned level
)
1528 return radv_image_has_dcc(image
) &&
1529 level
< image
->surface
.num_dcc_levels
;
1533 * Return whether the image has CB metadata.
1536 radv_image_has_CB_metadata(const struct radv_image
*image
)
1538 return radv_image_has_cmask(image
) ||
1539 radv_image_has_fmask(image
) ||
1540 radv_image_has_dcc(image
);
1544 * Return whether the image has HTILE metadata for depth surfaces.
1547 radv_image_has_htile(const struct radv_image
*image
)
1549 return image
->surface
.htile_size
;
1553 * Return whether HTILE metadata is enabled for a level.
1556 radv_htile_enabled(const struct radv_image
*image
, unsigned level
)
1558 return radv_image_has_htile(image
) && level
== 0;
1562 * Return whether the image is TC-compatible HTILE.
1565 radv_image_is_tc_compat_htile(const struct radv_image
*image
)
1567 return radv_image_has_htile(image
) && image
->tc_compatible_htile
;
1570 unsigned radv_image_queue_family_mask(const struct radv_image
*image
, uint32_t family
, uint32_t queue_family
);
1572 static inline uint32_t
1573 radv_get_layerCount(const struct radv_image
*image
,
1574 const VkImageSubresourceRange
*range
)
1576 return range
->layerCount
== VK_REMAINING_ARRAY_LAYERS
?
1577 image
->info
.array_size
- range
->baseArrayLayer
: range
->layerCount
;
1580 static inline uint32_t
1581 radv_get_levelCount(const struct radv_image
*image
,
1582 const VkImageSubresourceRange
*range
)
1584 return range
->levelCount
== VK_REMAINING_MIP_LEVELS
?
1585 image
->info
.levels
- range
->baseMipLevel
: range
->levelCount
;
1588 struct radeon_bo_metadata
;
1590 radv_init_metadata(struct radv_device
*device
,
1591 struct radv_image
*image
,
1592 struct radeon_bo_metadata
*metadata
);
1594 struct radv_image_view
{
1595 struct radv_image
*image
; /**< VkImageViewCreateInfo::image */
1596 struct radeon_winsys_bo
*bo
;
1598 VkImageViewType type
;
1599 VkImageAspectFlags aspect_mask
;
1601 uint32_t base_layer
;
1602 uint32_t layer_count
;
1604 uint32_t level_count
;
1605 VkExtent3D extent
; /**< Extent of VkImageViewCreateInfo::baseMipLevel. */
1607 uint32_t descriptor
[16];
1609 /* Descriptor for use as a storage image as opposed to a sampled image.
1610 * This has a few differences for cube maps (e.g. type).
1612 uint32_t storage_descriptor
[16];
1615 struct radv_image_create_info
{
1616 const VkImageCreateInfo
*vk_info
;
1618 bool no_metadata_planes
;
1621 VkResult
radv_image_create(VkDevice _device
,
1622 const struct radv_image_create_info
*info
,
1623 const VkAllocationCallbacks
* alloc
,
1627 radv_image_from_gralloc(VkDevice device_h
,
1628 const VkImageCreateInfo
*base_info
,
1629 const VkNativeBufferANDROID
*gralloc_info
,
1630 const VkAllocationCallbacks
*alloc
,
1631 VkImage
*out_image_h
);
1633 void radv_image_view_init(struct radv_image_view
*view
,
1634 struct radv_device
*device
,
1635 const VkImageViewCreateInfo
* pCreateInfo
);
1637 struct radv_buffer_view
{
1638 struct radeon_winsys_bo
*bo
;
1640 uint64_t range
; /**< VkBufferViewCreateInfo::range */
1643 void radv_buffer_view_init(struct radv_buffer_view
*view
,
1644 struct radv_device
*device
,
1645 const VkBufferViewCreateInfo
* pCreateInfo
);
1647 static inline struct VkExtent3D
1648 radv_sanitize_image_extent(const VkImageType imageType
,
1649 const struct VkExtent3D imageExtent
)
1651 switch (imageType
) {
1652 case VK_IMAGE_TYPE_1D
:
1653 return (VkExtent3D
) { imageExtent
.width
, 1, 1 };
1654 case VK_IMAGE_TYPE_2D
:
1655 return (VkExtent3D
) { imageExtent
.width
, imageExtent
.height
, 1 };
1656 case VK_IMAGE_TYPE_3D
:
1659 unreachable("invalid image type");
1663 static inline struct VkOffset3D
1664 radv_sanitize_image_offset(const VkImageType imageType
,
1665 const struct VkOffset3D imageOffset
)
1667 switch (imageType
) {
1668 case VK_IMAGE_TYPE_1D
:
1669 return (VkOffset3D
) { imageOffset
.x
, 0, 0 };
1670 case VK_IMAGE_TYPE_2D
:
1671 return (VkOffset3D
) { imageOffset
.x
, imageOffset
.y
, 0 };
1672 case VK_IMAGE_TYPE_3D
:
1675 unreachable("invalid image type");
1680 radv_image_extent_compare(const struct radv_image
*image
,
1681 const VkExtent3D
*extent
)
1683 if (extent
->width
!= image
->info
.width
||
1684 extent
->height
!= image
->info
.height
||
1685 extent
->depth
!= image
->info
.depth
)
1690 struct radv_sampler
{
1694 struct radv_color_buffer_info
{
1695 uint64_t cb_color_base
;
1696 uint64_t cb_color_cmask
;
1697 uint64_t cb_color_fmask
;
1698 uint64_t cb_dcc_base
;
1699 uint32_t cb_color_pitch
;
1700 uint32_t cb_color_slice
;
1701 uint32_t cb_color_view
;
1702 uint32_t cb_color_info
;
1703 uint32_t cb_color_attrib
;
1704 uint32_t cb_color_attrib2
;
1705 uint32_t cb_dcc_control
;
1706 uint32_t cb_color_cmask_slice
;
1707 uint32_t cb_color_fmask_slice
;
1710 struct radv_ds_buffer_info
{
1711 uint64_t db_z_read_base
;
1712 uint64_t db_stencil_read_base
;
1713 uint64_t db_z_write_base
;
1714 uint64_t db_stencil_write_base
;
1715 uint64_t db_htile_data_base
;
1716 uint32_t db_depth_info
;
1718 uint32_t db_stencil_info
;
1719 uint32_t db_depth_view
;
1720 uint32_t db_depth_size
;
1721 uint32_t db_depth_slice
;
1722 uint32_t db_htile_surface
;
1723 uint32_t pa_su_poly_offset_db_fmt_cntl
;
1724 uint32_t db_z_info2
;
1725 uint32_t db_stencil_info2
;
1729 struct radv_attachment_info
{
1731 struct radv_color_buffer_info cb
;
1732 struct radv_ds_buffer_info ds
;
1734 struct radv_image_view
*attachment
;
1737 struct radv_framebuffer
{
1742 uint32_t attachment_count
;
1743 struct radv_attachment_info attachments
[0];
1746 struct radv_subpass_barrier
{
1747 VkPipelineStageFlags src_stage_mask
;
1748 VkAccessFlags src_access_mask
;
1749 VkAccessFlags dst_access_mask
;
1752 void radv_subpass_barrier(struct radv_cmd_buffer
*cmd_buffer
,
1753 const struct radv_subpass_barrier
*barrier
);
1755 struct radv_subpass_attachment
{
1756 uint32_t attachment
;
1757 VkImageLayout layout
;
1760 struct radv_subpass
{
1761 uint32_t input_count
;
1762 uint32_t color_count
;
1763 struct radv_subpass_attachment
* input_attachments
;
1764 struct radv_subpass_attachment
* color_attachments
;
1765 struct radv_subpass_attachment
* resolve_attachments
;
1766 struct radv_subpass_attachment depth_stencil_attachment
;
1768 /** Subpass has at least one resolve attachment */
1771 struct radv_subpass_barrier start_barrier
;
1774 VkSampleCountFlagBits max_sample_count
;
1777 struct radv_render_pass_attachment
{
1780 VkAttachmentLoadOp load_op
;
1781 VkAttachmentLoadOp stencil_load_op
;
1782 VkImageLayout initial_layout
;
1783 VkImageLayout final_layout
;
1787 struct radv_render_pass
{
1788 uint32_t attachment_count
;
1789 uint32_t subpass_count
;
1790 struct radv_subpass_attachment
* subpass_attachments
;
1791 struct radv_render_pass_attachment
* attachments
;
1792 struct radv_subpass_barrier end_barrier
;
1793 struct radv_subpass subpasses
[0];
1796 VkResult
radv_device_init_meta(struct radv_device
*device
);
1797 void radv_device_finish_meta(struct radv_device
*device
);
1799 struct radv_query_pool
{
1800 struct radeon_winsys_bo
*bo
;
1802 uint32_t availability_offset
;
1806 uint32_t pipeline_stats_mask
;
1809 struct radv_semaphore
{
1810 /* use a winsys sem for non-exportable */
1811 struct radeon_winsys_sem
*sem
;
1813 uint32_t temp_syncobj
;
1816 void radv_set_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
1817 VkPipelineBindPoint bind_point
,
1818 struct radv_descriptor_set
*set
,
1822 radv_update_descriptor_sets(struct radv_device
*device
,
1823 struct radv_cmd_buffer
*cmd_buffer
,
1824 VkDescriptorSet overrideSet
,
1825 uint32_t descriptorWriteCount
,
1826 const VkWriteDescriptorSet
*pDescriptorWrites
,
1827 uint32_t descriptorCopyCount
,
1828 const VkCopyDescriptorSet
*pDescriptorCopies
);
1831 radv_update_descriptor_set_with_template(struct radv_device
*device
,
1832 struct radv_cmd_buffer
*cmd_buffer
,
1833 struct radv_descriptor_set
*set
,
1834 VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate
,
1837 void radv_meta_push_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
1838 VkPipelineBindPoint pipelineBindPoint
,
1839 VkPipelineLayout _layout
,
1841 uint32_t descriptorWriteCount
,
1842 const VkWriteDescriptorSet
*pDescriptorWrites
);
1844 void radv_initialize_dcc(struct radv_cmd_buffer
*cmd_buffer
,
1845 struct radv_image
*image
, uint32_t value
);
1848 struct radeon_winsys_fence
*fence
;
1849 struct wsi_fence
*fence_wsi
;
1854 uint32_t temp_syncobj
;
1857 /* radv_nir_to_llvm.c */
1858 struct radv_shader_variant_info
;
1859 struct radv_nir_compiler_options
;
1861 void radv_compile_gs_copy_shader(struct ac_llvm_compiler
*ac_llvm
,
1862 struct nir_shader
*geom_shader
,
1863 struct ac_shader_binary
*binary
,
1864 struct ac_shader_config
*config
,
1865 struct radv_shader_variant_info
*shader_info
,
1866 const struct radv_nir_compiler_options
*option
);
1868 void radv_compile_nir_shader(struct ac_llvm_compiler
*ac_llvm
,
1869 struct ac_shader_binary
*binary
,
1870 struct ac_shader_config
*config
,
1871 struct radv_shader_variant_info
*shader_info
,
1872 struct nir_shader
*const *nir
,
1874 const struct radv_nir_compiler_options
*options
);
1876 /* radv_shader_info.h */
1877 struct radv_shader_info
;
1879 void radv_nir_shader_info_pass(const struct nir_shader
*nir
,
1880 const struct radv_nir_compiler_options
*options
,
1881 struct radv_shader_info
*info
);
1883 struct radeon_winsys_sem
;
1885 #define RADV_DEFINE_HANDLE_CASTS(__radv_type, __VkType) \
1887 static inline struct __radv_type * \
1888 __radv_type ## _from_handle(__VkType _handle) \
1890 return (struct __radv_type *) _handle; \
1893 static inline __VkType \
1894 __radv_type ## _to_handle(struct __radv_type *_obj) \
1896 return (__VkType) _obj; \
1899 #define RADV_DEFINE_NONDISP_HANDLE_CASTS(__radv_type, __VkType) \
1901 static inline struct __radv_type * \
1902 __radv_type ## _from_handle(__VkType _handle) \
1904 return (struct __radv_type *)(uintptr_t) _handle; \
1907 static inline __VkType \
1908 __radv_type ## _to_handle(struct __radv_type *_obj) \
1910 return (__VkType)(uintptr_t) _obj; \
1913 #define RADV_FROM_HANDLE(__radv_type, __name, __handle) \
1914 struct __radv_type *__name = __radv_type ## _from_handle(__handle)
1916 RADV_DEFINE_HANDLE_CASTS(radv_cmd_buffer
, VkCommandBuffer
)
1917 RADV_DEFINE_HANDLE_CASTS(radv_device
, VkDevice
)
1918 RADV_DEFINE_HANDLE_CASTS(radv_instance
, VkInstance
)
1919 RADV_DEFINE_HANDLE_CASTS(radv_physical_device
, VkPhysicalDevice
)
1920 RADV_DEFINE_HANDLE_CASTS(radv_queue
, VkQueue
)
1922 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_cmd_pool
, VkCommandPool
)
1923 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer
, VkBuffer
)
1924 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer_view
, VkBufferView
)
1925 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_pool
, VkDescriptorPool
)
1926 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set
, VkDescriptorSet
)
1927 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set_layout
, VkDescriptorSetLayout
)
1928 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_update_template
, VkDescriptorUpdateTemplateKHR
)
1929 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_device_memory
, VkDeviceMemory
)
1930 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_fence
, VkFence
)
1931 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_event
, VkEvent
)
1932 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_framebuffer
, VkFramebuffer
)
1933 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image
, VkImage
)
1934 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image_view
, VkImageView
);
1935 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_cache
, VkPipelineCache
)
1936 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline
, VkPipeline
)
1937 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_layout
, VkPipelineLayout
)
1938 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_query_pool
, VkQueryPool
)
1939 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_render_pass
, VkRenderPass
)
1940 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_sampler
, VkSampler
)
1941 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_shader_module
, VkShaderModule
)
1942 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_semaphore
, VkSemaphore
)
1944 #endif /* RADV_PRIVATE_H */