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
113 RADV_MEM_HEAP_VRAM_CPU_ACCESS
,
120 RADV_MEM_TYPE_GTT_WRITE_COMBINE
,
121 RADV_MEM_TYPE_VRAM_CPU_ACCESS
,
122 RADV_MEM_TYPE_GTT_CACHED
,
126 #define radv_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
128 static inline uint32_t
129 align_u32(uint32_t v
, uint32_t a
)
131 assert(a
!= 0 && a
== (a
& -a
));
132 return (v
+ a
- 1) & ~(a
- 1);
135 static inline uint32_t
136 align_u32_npot(uint32_t v
, uint32_t a
)
138 return (v
+ a
- 1) / a
* a
;
141 static inline uint64_t
142 align_u64(uint64_t v
, uint64_t a
)
144 assert(a
!= 0 && a
== (a
& -a
));
145 return (v
+ a
- 1) & ~(a
- 1);
148 static inline int32_t
149 align_i32(int32_t v
, int32_t a
)
151 assert(a
!= 0 && a
== (a
& -a
));
152 return (v
+ a
- 1) & ~(a
- 1);
155 /** Alignment must be a power of 2. */
157 radv_is_aligned(uintmax_t n
, uintmax_t a
)
159 assert(a
== (a
& -a
));
160 return (n
& (a
- 1)) == 0;
163 static inline uint32_t
164 round_up_u32(uint32_t v
, uint32_t a
)
166 return (v
+ a
- 1) / a
;
169 static inline uint64_t
170 round_up_u64(uint64_t v
, uint64_t a
)
172 return (v
+ a
- 1) / a
;
175 static inline uint32_t
176 radv_minify(uint32_t n
, uint32_t levels
)
178 if (unlikely(n
== 0))
181 return MAX2(n
>> levels
, 1);
184 radv_clamp_f(float f
, float min
, float max
)
197 radv_clear_mask(uint32_t *inout_mask
, uint32_t clear_mask
)
199 if (*inout_mask
& clear_mask
) {
200 *inout_mask
&= ~clear_mask
;
207 #define for_each_bit(b, dword) \
208 for (uint32_t __dword = (dword); \
209 (b) = __builtin_ffs(__dword) - 1, __dword; \
210 __dword &= ~(1 << (b)))
212 #define typed_memcpy(dest, src, count) ({ \
213 STATIC_ASSERT(sizeof(*src) == sizeof(*dest)); \
214 memcpy((dest), (src), (count) * sizeof(*(src))); \
217 /* Whenever we generate an error, pass it through this function. Useful for
218 * debugging, where we can break on it. Only call at error site, not when
219 * propagating errors. Might be useful to plug in a stack trace here.
222 struct radv_instance
;
224 VkResult
__vk_errorf(struct radv_instance
*instance
, VkResult error
, const char *file
, int line
, const char *format
, ...);
226 #define vk_error(instance, error) __vk_errorf(instance, error, __FILE__, __LINE__, NULL);
227 #define vk_errorf(instance, error, format, ...) __vk_errorf(instance, error, __FILE__, __LINE__, format, ## __VA_ARGS__);
229 void __radv_finishme(const char *file
, int line
, const char *format
, ...)
230 radv_printflike(3, 4);
231 void radv_loge(const char *format
, ...) radv_printflike(1, 2);
232 void radv_loge_v(const char *format
, va_list va
);
233 void radv_logi(const char *format
, ...) radv_printflike(1, 2);
234 void radv_logi_v(const char *format
, va_list va
);
237 * Print a FINISHME message, including its source location.
239 #define radv_finishme(format, ...) \
241 static bool reported = false; \
243 __radv_finishme(__FILE__, __LINE__, format, ##__VA_ARGS__); \
248 /* A non-fatal assert. Useful for debugging. */
250 #define radv_assert(x) ({ \
251 if (unlikely(!(x))) \
252 fprintf(stderr, "%s:%d ASSERT: %s\n", __FILE__, __LINE__, #x); \
255 #define radv_assert(x)
258 #define stub_return(v) \
260 radv_finishme("stub %s", __func__); \
266 radv_finishme("stub %s", __func__); \
270 void *radv_lookup_entrypoint_unchecked(const char *name
);
271 void *radv_lookup_entrypoint_checked(const char *name
,
272 uint32_t core_version
,
273 const struct radv_instance_extension_table
*instance
,
274 const struct radv_device_extension_table
*device
);
276 struct radv_physical_device
{
277 VK_LOADER_DATA _loader_data
;
279 struct radv_instance
* instance
;
281 struct radeon_winsys
*ws
;
282 struct radeon_info rad_info
;
284 char name
[VK_MAX_PHYSICAL_DEVICE_NAME_SIZE
];
285 uint8_t driver_uuid
[VK_UUID_SIZE
];
286 uint8_t device_uuid
[VK_UUID_SIZE
];
287 uint8_t cache_uuid
[VK_UUID_SIZE
];
291 struct wsi_device wsi_device
;
293 bool has_rbplus
; /* if RB+ register exist */
294 bool rbplus_allowed
; /* if RB+ is allowed */
295 bool has_clear_state
;
296 bool cpdma_prefetch_writes_memory
;
297 bool has_scissor_bug
;
299 bool has_out_of_order_rast
;
300 bool out_of_order_rast_allowed
;
302 /* Whether DCC should be enabled for MSAA textures. */
303 bool dcc_msaa_allowed
;
305 /* This is the drivers on-disk cache used as a fallback as opposed to
306 * the pipeline cache defined by apps.
308 struct disk_cache
* disk_cache
;
310 VkPhysicalDeviceMemoryProperties memory_properties
;
311 enum radv_mem_type mem_type_indices
[RADV_MEM_TYPE_COUNT
];
313 struct radv_device_extension_table supported_extensions
;
316 struct radv_instance
{
317 VK_LOADER_DATA _loader_data
;
319 VkAllocationCallbacks alloc
;
322 int physicalDeviceCount
;
323 struct radv_physical_device physicalDevices
[RADV_MAX_DRM_DEVICES
];
325 uint64_t debug_flags
;
326 uint64_t perftest_flags
;
328 struct vk_debug_report_instance debug_report_callbacks
;
330 struct radv_instance_extension_table enabled_extensions
;
333 VkResult
radv_init_wsi(struct radv_physical_device
*physical_device
);
334 void radv_finish_wsi(struct radv_physical_device
*physical_device
);
336 bool radv_instance_extension_supported(const char *name
);
337 uint32_t radv_physical_device_api_version(struct radv_physical_device
*dev
);
338 bool radv_physical_device_extension_supported(struct radv_physical_device
*dev
,
343 struct radv_pipeline_cache
{
344 struct radv_device
* device
;
345 pthread_mutex_t mutex
;
349 uint32_t kernel_count
;
350 struct cache_entry
** hash_table
;
353 VkAllocationCallbacks alloc
;
356 struct radv_pipeline_key
{
357 uint32_t instance_rate_inputs
;
358 uint32_t instance_rate_divisors
[MAX_VERTEX_ATTRIBS
];
359 uint64_t vertex_alpha_adjust
;
360 unsigned tess_input_vertices
;
364 uint8_t log2_ps_iter_samples
;
366 uint32_t has_multiview_view_index
: 1;
367 uint32_t optimisations_disabled
: 1;
371 radv_pipeline_cache_init(struct radv_pipeline_cache
*cache
,
372 struct radv_device
*device
);
374 radv_pipeline_cache_finish(struct radv_pipeline_cache
*cache
);
376 radv_pipeline_cache_load(struct radv_pipeline_cache
*cache
,
377 const void *data
, size_t size
);
379 struct radv_shader_variant
;
382 radv_create_shader_variants_from_pipeline_cache(struct radv_device
*device
,
383 struct radv_pipeline_cache
*cache
,
384 const unsigned char *sha1
,
385 struct radv_shader_variant
**variants
);
388 radv_pipeline_cache_insert_shaders(struct radv_device
*device
,
389 struct radv_pipeline_cache
*cache
,
390 const unsigned char *sha1
,
391 struct radv_shader_variant
**variants
,
392 const void *const *codes
,
393 const unsigned *code_sizes
);
395 enum radv_blit_ds_layout
{
396 RADV_BLIT_DS_LAYOUT_TILE_ENABLE
,
397 RADV_BLIT_DS_LAYOUT_TILE_DISABLE
,
398 RADV_BLIT_DS_LAYOUT_COUNT
,
401 static inline enum radv_blit_ds_layout
radv_meta_blit_ds_to_type(VkImageLayout layout
)
403 return (layout
== VK_IMAGE_LAYOUT_GENERAL
) ? RADV_BLIT_DS_LAYOUT_TILE_DISABLE
: RADV_BLIT_DS_LAYOUT_TILE_ENABLE
;
406 static inline VkImageLayout
radv_meta_blit_ds_to_layout(enum radv_blit_ds_layout ds_layout
)
408 return ds_layout
== RADV_BLIT_DS_LAYOUT_TILE_ENABLE
? VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
: VK_IMAGE_LAYOUT_GENERAL
;
411 enum radv_meta_dst_layout
{
412 RADV_META_DST_LAYOUT_GENERAL
,
413 RADV_META_DST_LAYOUT_OPTIMAL
,
414 RADV_META_DST_LAYOUT_COUNT
,
417 static inline enum radv_meta_dst_layout
radv_meta_dst_layout_from_layout(VkImageLayout layout
)
419 return (layout
== VK_IMAGE_LAYOUT_GENERAL
) ? RADV_META_DST_LAYOUT_GENERAL
: RADV_META_DST_LAYOUT_OPTIMAL
;
422 static inline VkImageLayout
radv_meta_dst_layout_to_layout(enum radv_meta_dst_layout layout
)
424 return layout
== RADV_META_DST_LAYOUT_OPTIMAL
? VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
: VK_IMAGE_LAYOUT_GENERAL
;
427 struct radv_meta_state
{
428 VkAllocationCallbacks alloc
;
430 struct radv_pipeline_cache cache
;
433 * For on-demand pipeline creation, makes sure that
434 * only one thread tries to build a pipeline at the same time.
439 * Use array element `i` for images with `2^i` samples.
442 VkRenderPass render_pass
[NUM_META_FS_KEYS
];
443 VkPipeline color_pipelines
[NUM_META_FS_KEYS
];
445 VkRenderPass depthstencil_rp
;
446 VkPipeline depth_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
447 VkPipeline stencil_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
448 VkPipeline depthstencil_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
449 } clear
[1 + MAX_SAMPLES_LOG2
];
451 VkPipelineLayout clear_color_p_layout
;
452 VkPipelineLayout clear_depth_p_layout
;
454 VkRenderPass render_pass
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
456 /** Pipeline that blits from a 1D image. */
457 VkPipeline pipeline_1d_src
[NUM_META_FS_KEYS
];
459 /** Pipeline that blits from a 2D image. */
460 VkPipeline pipeline_2d_src
[NUM_META_FS_KEYS
];
462 /** Pipeline that blits from a 3D image. */
463 VkPipeline pipeline_3d_src
[NUM_META_FS_KEYS
];
465 VkRenderPass depth_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
466 VkPipeline depth_only_1d_pipeline
;
467 VkPipeline depth_only_2d_pipeline
;
468 VkPipeline depth_only_3d_pipeline
;
470 VkRenderPass stencil_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
471 VkPipeline stencil_only_1d_pipeline
;
472 VkPipeline stencil_only_2d_pipeline
;
473 VkPipeline stencil_only_3d_pipeline
;
474 VkPipelineLayout pipeline_layout
;
475 VkDescriptorSetLayout ds_layout
;
479 VkPipelineLayout p_layouts
[5];
480 VkDescriptorSetLayout ds_layouts
[5];
481 VkPipeline pipelines
[5][NUM_META_FS_KEYS
];
483 VkPipeline depth_only_pipeline
[5];
485 VkPipeline stencil_only_pipeline
[5];
486 } blit2d
[1 + MAX_SAMPLES_LOG2
];
488 VkRenderPass blit2d_render_passes
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
489 VkRenderPass blit2d_depth_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
490 VkRenderPass blit2d_stencil_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
493 VkPipelineLayout img_p_layout
;
494 VkDescriptorSetLayout img_ds_layout
;
496 VkPipeline pipeline_3d
;
499 VkPipelineLayout img_p_layout
;
500 VkDescriptorSetLayout img_ds_layout
;
502 VkPipeline pipeline_3d
;
505 VkPipelineLayout img_p_layout
;
506 VkDescriptorSetLayout img_ds_layout
;
508 VkPipeline pipeline_3d
;
511 VkPipelineLayout img_p_layout
;
512 VkDescriptorSetLayout img_ds_layout
;
514 VkPipeline pipeline_3d
;
518 VkPipelineLayout p_layout
;
519 VkPipeline pipeline
[NUM_META_FS_KEYS
];
520 VkRenderPass pass
[NUM_META_FS_KEYS
];
524 VkDescriptorSetLayout ds_layout
;
525 VkPipelineLayout p_layout
;
528 VkPipeline i_pipeline
;
529 VkPipeline srgb_pipeline
;
530 } rc
[MAX_SAMPLES_LOG2
];
534 VkDescriptorSetLayout ds_layout
;
535 VkPipelineLayout p_layout
;
538 VkRenderPass render_pass
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
539 VkPipeline pipeline
[NUM_META_FS_KEYS
];
540 } rc
[MAX_SAMPLES_LOG2
];
544 VkPipelineLayout p_layout
;
545 VkPipeline decompress_pipeline
;
546 VkPipeline resummarize_pipeline
;
548 } depth_decomp
[1 + MAX_SAMPLES_LOG2
];
551 VkPipelineLayout p_layout
;
552 VkPipeline cmask_eliminate_pipeline
;
553 VkPipeline fmask_decompress_pipeline
;
554 VkPipeline dcc_decompress_pipeline
;
557 VkDescriptorSetLayout dcc_decompress_compute_ds_layout
;
558 VkPipelineLayout dcc_decompress_compute_p_layout
;
559 VkPipeline dcc_decompress_compute_pipeline
;
563 VkPipelineLayout fill_p_layout
;
564 VkPipelineLayout copy_p_layout
;
565 VkDescriptorSetLayout fill_ds_layout
;
566 VkDescriptorSetLayout copy_ds_layout
;
567 VkPipeline fill_pipeline
;
568 VkPipeline copy_pipeline
;
572 VkDescriptorSetLayout ds_layout
;
573 VkPipelineLayout p_layout
;
574 VkPipeline occlusion_query_pipeline
;
575 VkPipeline pipeline_statistics_query_pipeline
;
580 #define RADV_QUEUE_GENERAL 0
581 #define RADV_QUEUE_COMPUTE 1
582 #define RADV_QUEUE_TRANSFER 2
584 #define RADV_MAX_QUEUE_FAMILIES 3
586 enum ring_type
radv_queue_family_to_ring(int f
);
589 VK_LOADER_DATA _loader_data
;
590 struct radv_device
* device
;
591 struct radeon_winsys_ctx
*hw_ctx
;
592 enum radeon_ctx_priority priority
;
593 uint32_t queue_family_index
;
595 VkDeviceQueueCreateFlags flags
;
597 uint32_t scratch_size
;
598 uint32_t compute_scratch_size
;
599 uint32_t esgs_ring_size
;
600 uint32_t gsvs_ring_size
;
602 bool has_sample_positions
;
604 struct radeon_winsys_bo
*scratch_bo
;
605 struct radeon_winsys_bo
*descriptor_bo
;
606 struct radeon_winsys_bo
*compute_scratch_bo
;
607 struct radeon_winsys_bo
*esgs_ring_bo
;
608 struct radeon_winsys_bo
*gsvs_ring_bo
;
609 struct radeon_winsys_bo
*tess_rings_bo
;
610 struct radeon_cmdbuf
*initial_preamble_cs
;
611 struct radeon_cmdbuf
*initial_full_flush_preamble_cs
;
612 struct radeon_cmdbuf
*continue_preamble_cs
;
615 struct radv_bo_list
{
616 struct radv_winsys_bo_list list
;
618 pthread_mutex_t mutex
;
622 VK_LOADER_DATA _loader_data
;
624 VkAllocationCallbacks alloc
;
626 struct radv_instance
* instance
;
627 struct radeon_winsys
*ws
;
629 struct radv_meta_state meta_state
;
631 struct radv_queue
*queues
[RADV_MAX_QUEUE_FAMILIES
];
632 int queue_count
[RADV_MAX_QUEUE_FAMILIES
];
633 struct radeon_cmdbuf
*empty_cs
[RADV_MAX_QUEUE_FAMILIES
];
635 bool always_use_syncobj
;
636 bool has_distributed_tess
;
639 uint32_t tess_offchip_block_dw_size
;
640 uint32_t scratch_waves
;
641 uint32_t dispatch_initiator
;
643 uint32_t gs_table_depth
;
645 /* MSAA sample locations.
646 * The first index is the sample index.
647 * The second index is the coordinate: X, Y. */
648 float sample_locations_1x
[1][2];
649 float sample_locations_2x
[2][2];
650 float sample_locations_4x
[4][2];
651 float sample_locations_8x
[8][2];
652 float sample_locations_16x
[16][2];
655 uint32_t gfx_init_size_dw
;
656 struct radeon_winsys_bo
*gfx_init
;
658 struct radeon_winsys_bo
*trace_bo
;
659 uint32_t *trace_id_ptr
;
661 /* Whether to keep shader debug info, for tracing or VK_AMD_shader_info */
662 bool keep_shader_info
;
664 struct radv_physical_device
*physical_device
;
666 /* Backup in-memory cache to be used if the app doesn't provide one */
667 struct radv_pipeline_cache
* mem_cache
;
670 * use different counters so MSAA MRTs get consecutive surface indices,
671 * even if MASK is allocated in between.
673 uint32_t image_mrt_offset_counter
;
674 uint32_t fmask_mrt_offset_counter
;
675 struct list_head shader_slabs
;
676 mtx_t shader_slab_mutex
;
678 /* For detecting VM faults reported by dmesg. */
679 uint64_t dmesg_timestamp
;
681 struct radv_device_extension_table enabled_extensions
;
683 /* Whether the driver uses a global BO list. */
684 bool use_global_bo_list
;
686 struct radv_bo_list bo_list
;
688 /* Whether anisotropy is forced with RADV_TEX_ANISO (-1 is disabled). */
692 struct radv_device_memory
{
693 struct radeon_winsys_bo
*bo
;
694 /* for dedicated allocations */
695 struct radv_image
*image
;
696 struct radv_buffer
*buffer
;
698 VkDeviceSize map_size
;
704 struct radv_descriptor_range
{
709 struct radv_descriptor_set
{
710 const struct radv_descriptor_set_layout
*layout
;
713 struct radeon_winsys_bo
*bo
;
715 uint32_t *mapped_ptr
;
716 struct radv_descriptor_range
*dynamic_descriptors
;
718 struct radeon_winsys_bo
*descriptors
[0];
721 struct radv_push_descriptor_set
723 struct radv_descriptor_set set
;
727 struct radv_descriptor_pool_entry
{
730 struct radv_descriptor_set
*set
;
733 struct radv_descriptor_pool
{
734 struct radeon_winsys_bo
*bo
;
736 uint64_t current_offset
;
739 uint8_t *host_memory_base
;
740 uint8_t *host_memory_ptr
;
741 uint8_t *host_memory_end
;
743 uint32_t entry_count
;
744 uint32_t max_entry_count
;
745 struct radv_descriptor_pool_entry entries
[0];
748 struct radv_descriptor_update_template_entry
{
749 VkDescriptorType descriptor_type
;
751 /* The number of descriptors to update */
752 uint32_t descriptor_count
;
754 /* Into mapped_ptr or dynamic_descriptors, in units of the respective array */
757 /* In dwords. Not valid/used for dynamic descriptors */
760 uint32_t buffer_offset
;
762 /* Only valid for combined image samplers and samplers */
763 uint16_t has_sampler
;
769 /* For push descriptors */
770 const uint32_t *immutable_samplers
;
773 struct radv_descriptor_update_template
{
774 uint32_t entry_count
;
775 VkPipelineBindPoint bind_point
;
776 struct radv_descriptor_update_template_entry entry
[0];
782 VkBufferUsageFlags usage
;
783 VkBufferCreateFlags flags
;
786 struct radeon_winsys_bo
* bo
;
792 enum radv_dynamic_state_bits
{
793 RADV_DYNAMIC_VIEWPORT
= 1 << 0,
794 RADV_DYNAMIC_SCISSOR
= 1 << 1,
795 RADV_DYNAMIC_LINE_WIDTH
= 1 << 2,
796 RADV_DYNAMIC_DEPTH_BIAS
= 1 << 3,
797 RADV_DYNAMIC_BLEND_CONSTANTS
= 1 << 4,
798 RADV_DYNAMIC_DEPTH_BOUNDS
= 1 << 5,
799 RADV_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6,
800 RADV_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7,
801 RADV_DYNAMIC_STENCIL_REFERENCE
= 1 << 8,
802 RADV_DYNAMIC_DISCARD_RECTANGLE
= 1 << 9,
803 RADV_DYNAMIC_ALL
= (1 << 10) - 1,
806 enum radv_cmd_dirty_bits
{
807 /* Keep the dynamic state dirty bits in sync with
808 * enum radv_dynamic_state_bits */
809 RADV_CMD_DIRTY_DYNAMIC_VIEWPORT
= 1 << 0,
810 RADV_CMD_DIRTY_DYNAMIC_SCISSOR
= 1 << 1,
811 RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH
= 1 << 2,
812 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS
= 1 << 3,
813 RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS
= 1 << 4,
814 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS
= 1 << 5,
815 RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6,
816 RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7,
817 RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE
= 1 << 8,
818 RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE
= 1 << 9,
819 RADV_CMD_DIRTY_DYNAMIC_ALL
= (1 << 10) - 1,
820 RADV_CMD_DIRTY_PIPELINE
= 1 << 10,
821 RADV_CMD_DIRTY_INDEX_BUFFER
= 1 << 11,
822 RADV_CMD_DIRTY_FRAMEBUFFER
= 1 << 12,
823 RADV_CMD_DIRTY_VERTEX_BUFFER
= 1 << 13,
826 enum radv_cmd_flush_bits
{
827 RADV_CMD_FLAG_INV_ICACHE
= 1 << 0,
828 /* SMEM L1, other names: KCACHE, constant cache, DCACHE, data cache */
829 RADV_CMD_FLAG_INV_SMEM_L1
= 1 << 1,
830 /* VMEM L1 can optionally be bypassed (GLC=1). Other names: TC L1 */
831 RADV_CMD_FLAG_INV_VMEM_L1
= 1 << 2,
832 /* Used by everything except CB/DB, can be bypassed (SLC=1). Other names: TC L2 */
833 RADV_CMD_FLAG_INV_GLOBAL_L2
= 1 << 3,
834 /* Same as above, but only writes back and doesn't invalidate */
835 RADV_CMD_FLAG_WRITEBACK_GLOBAL_L2
= 1 << 4,
836 /* Framebuffer caches */
837 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
= 1 << 5,
838 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
= 1 << 6,
839 RADV_CMD_FLAG_FLUSH_AND_INV_DB
= 1 << 7,
840 RADV_CMD_FLAG_FLUSH_AND_INV_CB
= 1 << 8,
841 /* Engine synchronization. */
842 RADV_CMD_FLAG_VS_PARTIAL_FLUSH
= 1 << 9,
843 RADV_CMD_FLAG_PS_PARTIAL_FLUSH
= 1 << 10,
844 RADV_CMD_FLAG_CS_PARTIAL_FLUSH
= 1 << 11,
845 RADV_CMD_FLAG_VGT_FLUSH
= 1 << 12,
846 /* Pipeline query controls. */
847 RADV_CMD_FLAG_START_PIPELINE_STATS
= 1 << 13,
848 RADV_CMD_FLAG_STOP_PIPELINE_STATS
= 1 << 14,
850 RADV_CMD_FLUSH_AND_INV_FRAMEBUFFER
= (RADV_CMD_FLAG_FLUSH_AND_INV_CB
|
851 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
|
852 RADV_CMD_FLAG_FLUSH_AND_INV_DB
|
853 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
)
856 struct radv_vertex_binding
{
857 struct radv_buffer
* buffer
;
861 struct radv_viewport_state
{
863 VkViewport viewports
[MAX_VIEWPORTS
];
866 struct radv_scissor_state
{
868 VkRect2D scissors
[MAX_SCISSORS
];
871 struct radv_discard_rectangle_state
{
873 VkRect2D rectangles
[MAX_DISCARD_RECTANGLES
];
876 struct radv_dynamic_state
{
878 * Bitmask of (1 << VK_DYNAMIC_STATE_*).
879 * Defines the set of saved dynamic state.
883 struct radv_viewport_state viewport
;
885 struct radv_scissor_state scissor
;
895 float blend_constants
[4];
905 } stencil_compare_mask
;
910 } stencil_write_mask
;
917 struct radv_discard_rectangle_state discard_rectangle
;
920 extern const struct radv_dynamic_state default_dynamic_state
;
923 radv_get_debug_option_name(int id
);
926 radv_get_perftest_option_name(int id
);
929 * Attachment state when recording a renderpass instance.
931 * The clear value is valid only if there exists a pending clear.
933 struct radv_attachment_state
{
934 VkImageAspectFlags pending_clear_aspects
;
935 uint32_t cleared_views
;
936 VkClearValue clear_value
;
937 VkImageLayout current_layout
;
940 struct radv_descriptor_state
{
941 struct radv_descriptor_set
*sets
[MAX_SETS
];
944 struct radv_push_descriptor_set push_set
;
946 uint32_t dynamic_buffers
[4 * MAX_DYNAMIC_BUFFERS
];
949 struct radv_cmd_state
{
950 /* Vertex descriptors */
957 uint32_t prefetch_L2_mask
;
959 struct radv_pipeline
* pipeline
;
960 struct radv_pipeline
* emitted_pipeline
;
961 struct radv_pipeline
* compute_pipeline
;
962 struct radv_pipeline
* emitted_compute_pipeline
;
963 struct radv_framebuffer
* framebuffer
;
964 struct radv_render_pass
* pass
;
965 const struct radv_subpass
* subpass
;
966 struct radv_dynamic_state dynamic
;
967 struct radv_attachment_state
* attachments
;
968 VkRect2D render_area
;
971 struct radv_buffer
*index_buffer
;
972 uint64_t index_offset
;
974 uint32_t max_index_count
;
976 int32_t last_index_type
;
978 int32_t last_primitive_reset_en
;
979 uint32_t last_primitive_reset_index
;
980 enum radv_cmd_flush_bits flush_bits
;
981 unsigned active_occlusion_queries
;
982 bool perfect_occlusion_queries_enabled
;
983 unsigned active_pipeline_queries
;
986 uint32_t last_ia_multi_vgt_param
;
988 uint32_t last_num_instances
;
989 uint32_t last_first_instance
;
990 uint32_t last_vertex_offset
;
992 /* Whether CP DMA is busy/idle. */
995 /* Conditional rendering info. */
996 int predication_type
; /* -1: disabled, 0: normal, 1: inverted */
997 uint64_t predication_va
;
1000 struct radv_cmd_pool
{
1001 VkAllocationCallbacks alloc
;
1002 struct list_head cmd_buffers
;
1003 struct list_head free_cmd_buffers
;
1004 uint32_t queue_family_index
;
1007 struct radv_cmd_buffer_upload
{
1011 struct radeon_winsys_bo
*upload_bo
;
1012 struct list_head list
;
1015 enum radv_cmd_buffer_status
{
1016 RADV_CMD_BUFFER_STATUS_INVALID
,
1017 RADV_CMD_BUFFER_STATUS_INITIAL
,
1018 RADV_CMD_BUFFER_STATUS_RECORDING
,
1019 RADV_CMD_BUFFER_STATUS_EXECUTABLE
,
1020 RADV_CMD_BUFFER_STATUS_PENDING
,
1023 struct radv_cmd_buffer
{
1024 VK_LOADER_DATA _loader_data
;
1026 struct radv_device
* device
;
1028 struct radv_cmd_pool
* pool
;
1029 struct list_head pool_link
;
1031 VkCommandBufferUsageFlags usage_flags
;
1032 VkCommandBufferLevel level
;
1033 enum radv_cmd_buffer_status status
;
1034 struct radeon_cmdbuf
*cs
;
1035 struct radv_cmd_state state
;
1036 struct radv_vertex_binding vertex_bindings
[MAX_VBS
];
1037 uint32_t queue_family_index
;
1039 uint8_t push_constants
[MAX_PUSH_CONSTANTS_SIZE
];
1040 VkShaderStageFlags push_constant_stages
;
1041 struct radv_descriptor_set meta_push_descriptors
;
1043 struct radv_descriptor_state descriptors
[VK_PIPELINE_BIND_POINT_RANGE_SIZE
];
1045 struct radv_cmd_buffer_upload upload
;
1047 uint32_t scratch_size_needed
;
1048 uint32_t compute_scratch_size_needed
;
1049 uint32_t esgs_ring_size_needed
;
1050 uint32_t gsvs_ring_size_needed
;
1051 bool tess_rings_needed
;
1052 bool sample_positions_needed
;
1054 VkResult record_result
;
1056 uint32_t gfx9_fence_offset
;
1057 struct radeon_winsys_bo
*gfx9_fence_bo
;
1058 uint32_t gfx9_fence_idx
;
1059 uint64_t gfx9_eop_bug_va
;
1062 * Whether a query pool has been resetted and we have to flush caches.
1064 bool pending_reset_query
;
1069 bool radv_cmd_buffer_uses_mec(struct radv_cmd_buffer
*cmd_buffer
);
1071 void si_emit_graphics(struct radv_physical_device
*physical_device
,
1072 struct radeon_cmdbuf
*cs
);
1073 void si_emit_compute(struct radv_physical_device
*physical_device
,
1074 struct radeon_cmdbuf
*cs
);
1076 void cik_create_gfx_config(struct radv_device
*device
);
1078 void si_write_viewport(struct radeon_cmdbuf
*cs
, int first_vp
,
1079 int count
, const VkViewport
*viewports
);
1080 void si_write_scissors(struct radeon_cmdbuf
*cs
, int first
,
1081 int count
, const VkRect2D
*scissors
,
1082 const VkViewport
*viewports
, bool can_use_guardband
);
1083 uint32_t si_get_ia_multi_vgt_param(struct radv_cmd_buffer
*cmd_buffer
,
1084 bool instanced_draw
, bool indirect_draw
,
1085 uint32_t draw_vertex_count
);
1086 void si_cs_emit_write_event_eop(struct radeon_cmdbuf
*cs
,
1087 enum chip_class chip_class
,
1089 unsigned event
, unsigned event_flags
,
1094 uint64_t gfx9_eop_bug_va
);
1096 void si_emit_wait_fence(struct radeon_cmdbuf
*cs
,
1097 uint64_t va
, uint32_t ref
,
1099 void si_cs_emit_cache_flush(struct radeon_cmdbuf
*cs
,
1100 enum chip_class chip_class
,
1101 uint32_t *fence_ptr
, uint64_t va
,
1103 enum radv_cmd_flush_bits flush_bits
,
1104 uint64_t gfx9_eop_bug_va
);
1105 void si_emit_cache_flush(struct radv_cmd_buffer
*cmd_buffer
);
1106 void si_emit_set_predication_state(struct radv_cmd_buffer
*cmd_buffer
,
1107 bool inverted
, uint64_t va
);
1108 void si_cp_dma_buffer_copy(struct radv_cmd_buffer
*cmd_buffer
,
1109 uint64_t src_va
, uint64_t dest_va
,
1111 void si_cp_dma_prefetch(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
1113 void si_cp_dma_clear_buffer(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
1114 uint64_t size
, unsigned value
);
1115 void si_cp_dma_wait_for_idle(struct radv_cmd_buffer
*cmd_buffer
);
1117 void radv_set_db_count_control(struct radv_cmd_buffer
*cmd_buffer
);
1119 radv_cmd_buffer_upload_alloc(struct radv_cmd_buffer
*cmd_buffer
,
1122 unsigned *out_offset
,
1125 radv_cmd_buffer_set_subpass(struct radv_cmd_buffer
*cmd_buffer
,
1126 const struct radv_subpass
*subpass
,
1129 radv_cmd_buffer_upload_data(struct radv_cmd_buffer
*cmd_buffer
,
1130 unsigned size
, unsigned alignmnet
,
1131 const void *data
, unsigned *out_offset
);
1133 void radv_cmd_buffer_clear_subpass(struct radv_cmd_buffer
*cmd_buffer
);
1134 void radv_cmd_buffer_resolve_subpass(struct radv_cmd_buffer
*cmd_buffer
);
1135 void radv_cmd_buffer_resolve_subpass_cs(struct radv_cmd_buffer
*cmd_buffer
);
1136 void radv_cmd_buffer_resolve_subpass_fs(struct radv_cmd_buffer
*cmd_buffer
);
1137 void radv_cayman_emit_msaa_sample_locs(struct radeon_cmdbuf
*cs
, int nr_samples
);
1138 unsigned radv_cayman_get_maxdist(int log_samples
);
1139 void radv_device_init_msaa(struct radv_device
*device
);
1141 void radv_update_ds_clear_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1142 struct radv_image
*image
,
1143 VkClearDepthStencilValue ds_clear_value
,
1144 VkImageAspectFlags aspects
);
1146 void radv_update_color_clear_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1147 struct radv_image
*image
,
1149 uint32_t color_values
[2]);
1151 void radv_set_dcc_need_cmask_elim_pred(struct radv_cmd_buffer
*cmd_buffer
,
1152 struct radv_image
*image
,
1154 uint32_t radv_fill_buffer(struct radv_cmd_buffer
*cmd_buffer
,
1155 struct radeon_winsys_bo
*bo
,
1156 uint64_t offset
, uint64_t size
, uint32_t value
);
1157 void radv_cmd_buffer_trace_emit(struct radv_cmd_buffer
*cmd_buffer
);
1158 bool radv_get_memory_fd(struct radv_device
*device
,
1159 struct radv_device_memory
*memory
,
1163 radv_emit_shader_pointer_head(struct radeon_cmdbuf
*cs
,
1164 unsigned sh_offset
, unsigned pointer_count
,
1165 bool use_32bit_pointers
)
1167 radeon_emit(cs
, PKT3(PKT3_SET_SH_REG
, pointer_count
* (use_32bit_pointers
? 1 : 2), 0));
1168 radeon_emit(cs
, (sh_offset
- SI_SH_REG_OFFSET
) >> 2);
1172 radv_emit_shader_pointer_body(struct radv_device
*device
,
1173 struct radeon_cmdbuf
*cs
,
1174 uint64_t va
, bool use_32bit_pointers
)
1176 radeon_emit(cs
, va
);
1178 if (use_32bit_pointers
) {
1180 (va
>> 32) == device
->physical_device
->rad_info
.address32_hi
);
1182 radeon_emit(cs
, va
>> 32);
1187 radv_emit_shader_pointer(struct radv_device
*device
,
1188 struct radeon_cmdbuf
*cs
,
1189 uint32_t sh_offset
, uint64_t va
, bool global
)
1191 bool use_32bit_pointers
= HAVE_32BIT_POINTERS
&& !global
;
1193 radv_emit_shader_pointer_head(cs
, sh_offset
, 1, use_32bit_pointers
);
1194 radv_emit_shader_pointer_body(device
, cs
, va
, use_32bit_pointers
);
1197 static inline struct radv_descriptor_state
*
1198 radv_get_descriptors_state(struct radv_cmd_buffer
*cmd_buffer
,
1199 VkPipelineBindPoint bind_point
)
1201 assert(bind_point
== VK_PIPELINE_BIND_POINT_GRAPHICS
||
1202 bind_point
== VK_PIPELINE_BIND_POINT_COMPUTE
);
1203 return &cmd_buffer
->descriptors
[bind_point
];
1207 * Takes x,y,z as exact numbers of invocations, instead of blocks.
1209 * Limitations: Can't call normal dispatch functions without binding or rebinding
1210 * the compute pipeline.
1212 void radv_unaligned_dispatch(
1213 struct radv_cmd_buffer
*cmd_buffer
,
1219 struct radeon_winsys_bo
*bo
;
1223 struct radv_shader_module
;
1225 #define RADV_HASH_SHADER_IS_GEOM_COPY_SHADER (1 << 0)
1226 #define RADV_HASH_SHADER_SISCHED (1 << 1)
1227 #define RADV_HASH_SHADER_UNSAFE_MATH (1 << 2)
1229 radv_hash_shaders(unsigned char *hash
,
1230 const VkPipelineShaderStageCreateInfo
**stages
,
1231 const struct radv_pipeline_layout
*layout
,
1232 const struct radv_pipeline_key
*key
,
1235 static inline gl_shader_stage
1236 vk_to_mesa_shader_stage(VkShaderStageFlagBits vk_stage
)
1238 assert(__builtin_popcount(vk_stage
) == 1);
1239 return ffs(vk_stage
) - 1;
1242 static inline VkShaderStageFlagBits
1243 mesa_to_vk_shader_stage(gl_shader_stage mesa_stage
)
1245 return (1 << mesa_stage
);
1248 #define RADV_STAGE_MASK ((1 << MESA_SHADER_STAGES) - 1)
1250 #define radv_foreach_stage(stage, stage_bits) \
1251 for (gl_shader_stage stage, \
1252 __tmp = (gl_shader_stage)((stage_bits) & RADV_STAGE_MASK); \
1253 stage = __builtin_ffs(__tmp) - 1, __tmp; \
1254 __tmp &= ~(1 << (stage)))
1256 extern const VkFormat radv_fs_key_format_exemplars
[NUM_META_FS_KEYS
];
1257 unsigned radv_format_meta_fs_key(VkFormat format
);
1259 struct radv_multisample_state
{
1261 uint32_t pa_sc_line_cntl
;
1262 uint32_t pa_sc_mode_cntl_0
;
1263 uint32_t pa_sc_mode_cntl_1
;
1264 uint32_t pa_sc_aa_config
;
1265 uint32_t pa_sc_aa_mask
[2];
1266 unsigned num_samples
;
1269 struct radv_prim_vertex_count
{
1274 struct radv_vertex_elements_info
{
1275 uint32_t rsrc_word3
[MAX_VERTEX_ATTRIBS
];
1276 uint32_t format_size
[MAX_VERTEX_ATTRIBS
];
1277 uint32_t binding
[MAX_VERTEX_ATTRIBS
];
1278 uint32_t offset
[MAX_VERTEX_ATTRIBS
];
1282 struct radv_ia_multi_vgt_param_helpers
{
1284 bool partial_es_wave
;
1285 uint8_t primgroup_size
;
1286 bool wd_switch_on_eop
;
1287 bool ia_switch_on_eoi
;
1288 bool partial_vs_wave
;
1291 #define SI_GS_PER_ES 128
1293 struct radv_pipeline
{
1294 struct radv_device
* device
;
1295 struct radv_dynamic_state dynamic_state
;
1297 struct radv_pipeline_layout
* layout
;
1299 bool need_indirect_descriptor_sets
;
1300 struct radv_shader_variant
* shaders
[MESA_SHADER_STAGES
];
1301 struct radv_shader_variant
*gs_copy_shader
;
1302 VkShaderStageFlags active_stages
;
1304 struct radeon_cmdbuf cs
;
1306 struct radv_vertex_elements_info vertex_elements
;
1308 uint32_t binding_stride
[MAX_VBS
];
1310 uint32_t user_data_0
[MESA_SHADER_STAGES
];
1313 struct radv_multisample_state ms
;
1314 uint32_t spi_baryc_cntl
;
1315 bool prim_restart_enable
;
1316 unsigned esgs_ring_size
;
1317 unsigned gsvs_ring_size
;
1318 uint32_t vtx_base_sgpr
;
1319 struct radv_ia_multi_vgt_param_helpers ia_multi_vgt_param
;
1320 uint8_t vtx_emit_num
;
1321 struct radv_prim_vertex_count prim_vertex_count
;
1322 bool can_use_guardband
;
1323 uint32_t needed_dynamic_state
;
1324 bool disable_out_of_order_rast_for_occlusion
;
1326 /* Used for rbplus */
1327 uint32_t col_format
;
1328 uint32_t cb_target_mask
;
1333 unsigned scratch_bytes_per_wave
;
1336 static inline bool radv_pipeline_has_gs(const struct radv_pipeline
*pipeline
)
1338 return pipeline
->shaders
[MESA_SHADER_GEOMETRY
] ? true : false;
1341 static inline bool radv_pipeline_has_tess(const struct radv_pipeline
*pipeline
)
1343 return pipeline
->shaders
[MESA_SHADER_TESS_CTRL
] ? true : false;
1346 struct radv_userdata_info
*radv_lookup_user_sgpr(struct radv_pipeline
*pipeline
,
1347 gl_shader_stage stage
,
1350 struct radv_shader_variant
*radv_get_shader(struct radv_pipeline
*pipeline
,
1351 gl_shader_stage stage
);
1353 struct radv_graphics_pipeline_create_info
{
1355 bool db_depth_clear
;
1356 bool db_stencil_clear
;
1357 bool db_depth_disable_expclear
;
1358 bool db_stencil_disable_expclear
;
1359 bool db_flush_depth_inplace
;
1360 bool db_flush_stencil_inplace
;
1361 bool db_resummarize
;
1362 uint32_t custom_blend_mode
;
1366 radv_graphics_pipeline_create(VkDevice device
,
1367 VkPipelineCache cache
,
1368 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
1369 const struct radv_graphics_pipeline_create_info
*extra
,
1370 const VkAllocationCallbacks
*alloc
,
1371 VkPipeline
*pPipeline
);
1373 struct vk_format_description
;
1374 uint32_t radv_translate_buffer_dataformat(const struct vk_format_description
*desc
,
1375 int first_non_void
);
1376 uint32_t radv_translate_buffer_numformat(const struct vk_format_description
*desc
,
1377 int first_non_void
);
1378 uint32_t radv_translate_colorformat(VkFormat format
);
1379 uint32_t radv_translate_color_numformat(VkFormat format
,
1380 const struct vk_format_description
*desc
,
1381 int first_non_void
);
1382 uint32_t radv_colorformat_endian_swap(uint32_t colorformat
);
1383 unsigned radv_translate_colorswap(VkFormat format
, bool do_endian_swap
);
1384 uint32_t radv_translate_dbformat(VkFormat format
);
1385 uint32_t radv_translate_tex_dataformat(VkFormat format
,
1386 const struct vk_format_description
*desc
,
1387 int first_non_void
);
1388 uint32_t radv_translate_tex_numformat(VkFormat format
,
1389 const struct vk_format_description
*desc
,
1390 int first_non_void
);
1391 bool radv_format_pack_clear_color(VkFormat format
,
1392 uint32_t clear_vals
[2],
1393 VkClearColorValue
*value
);
1394 bool radv_is_colorbuffer_format_supported(VkFormat format
, bool *blendable
);
1395 bool radv_dcc_formats_compatible(VkFormat format1
,
1398 struct radv_fmask_info
{
1402 unsigned pitch_in_pixels
;
1403 unsigned bank_height
;
1404 unsigned slice_tile_max
;
1405 unsigned tile_mode_index
;
1406 unsigned tile_swizzle
;
1409 struct radv_cmask_info
{
1413 unsigned slice_tile_max
;
1418 /* The original VkFormat provided by the client. This may not match any
1419 * of the actual surface formats.
1422 VkImageAspectFlags aspects
;
1423 VkImageUsageFlags usage
; /**< Superset of VkImageCreateInfo::usage. */
1424 struct ac_surf_info info
;
1425 VkImageTiling tiling
; /** VkImageCreateInfo::tiling */
1426 VkImageCreateFlags flags
; /** VkImageCreateInfo::flags */
1431 unsigned queue_family_mask
;
1435 /* Set when bound */
1436 struct radeon_winsys_bo
*bo
;
1437 VkDeviceSize offset
;
1438 uint64_t dcc_offset
;
1439 uint64_t htile_offset
;
1440 bool tc_compatible_htile
;
1441 struct radeon_surf surface
;
1443 struct radv_fmask_info fmask
;
1444 struct radv_cmask_info cmask
;
1445 uint64_t clear_value_offset
;
1446 uint64_t dcc_pred_offset
;
1448 /* For VK_ANDROID_native_buffer, the WSI image owns the memory, */
1449 VkDeviceMemory owned_memory
;
1452 /* Whether the image has a htile that is known consistent with the contents of
1454 bool radv_layout_has_htile(const struct radv_image
*image
,
1455 VkImageLayout layout
,
1456 unsigned queue_mask
);
1458 /* Whether the image has a htile that is known consistent with the contents of
1459 * the image and is allowed to be in compressed form.
1461 * If this is false reads that don't use the htile should be able to return
1464 bool radv_layout_is_htile_compressed(const struct radv_image
*image
,
1465 VkImageLayout layout
,
1466 unsigned queue_mask
);
1468 bool radv_layout_can_fast_clear(const struct radv_image
*image
,
1469 VkImageLayout layout
,
1470 unsigned queue_mask
);
1472 bool radv_layout_dcc_compressed(const struct radv_image
*image
,
1473 VkImageLayout layout
,
1474 unsigned queue_mask
);
1477 * Return whether the image has CMASK metadata for color surfaces.
1480 radv_image_has_cmask(const struct radv_image
*image
)
1482 return image
->cmask
.size
;
1486 * Return whether the image has FMASK metadata for color surfaces.
1489 radv_image_has_fmask(const struct radv_image
*image
)
1491 return image
->fmask
.size
;
1495 * Return whether the image has DCC metadata for color surfaces.
1498 radv_image_has_dcc(const struct radv_image
*image
)
1500 return image
->surface
.dcc_size
;
1504 * Return whether DCC metadata is enabled for a level.
1507 radv_dcc_enabled(const struct radv_image
*image
, unsigned level
)
1509 return radv_image_has_dcc(image
) &&
1510 level
< image
->surface
.num_dcc_levels
;
1514 * Return whether the image has CB metadata.
1517 radv_image_has_CB_metadata(const struct radv_image
*image
)
1519 return radv_image_has_cmask(image
) ||
1520 radv_image_has_fmask(image
) ||
1521 radv_image_has_dcc(image
);
1525 * Return whether the image has HTILE metadata for depth surfaces.
1528 radv_image_has_htile(const struct radv_image
*image
)
1530 return image
->surface
.htile_size
;
1534 * Return whether HTILE metadata is enabled for a level.
1537 radv_htile_enabled(const struct radv_image
*image
, unsigned level
)
1539 return radv_image_has_htile(image
) && level
== 0;
1543 * Return whether the image is TC-compatible HTILE.
1546 radv_image_is_tc_compat_htile(const struct radv_image
*image
)
1548 return radv_image_has_htile(image
) && image
->tc_compatible_htile
;
1551 unsigned radv_image_queue_family_mask(const struct radv_image
*image
, uint32_t family
, uint32_t queue_family
);
1553 static inline uint32_t
1554 radv_get_layerCount(const struct radv_image
*image
,
1555 const VkImageSubresourceRange
*range
)
1557 return range
->layerCount
== VK_REMAINING_ARRAY_LAYERS
?
1558 image
->info
.array_size
- range
->baseArrayLayer
: range
->layerCount
;
1561 static inline uint32_t
1562 radv_get_levelCount(const struct radv_image
*image
,
1563 const VkImageSubresourceRange
*range
)
1565 return range
->levelCount
== VK_REMAINING_MIP_LEVELS
?
1566 image
->info
.levels
- range
->baseMipLevel
: range
->levelCount
;
1569 struct radeon_bo_metadata
;
1571 radv_init_metadata(struct radv_device
*device
,
1572 struct radv_image
*image
,
1573 struct radeon_bo_metadata
*metadata
);
1575 struct radv_image_view
{
1576 struct radv_image
*image
; /**< VkImageViewCreateInfo::image */
1577 struct radeon_winsys_bo
*bo
;
1579 VkImageViewType type
;
1580 VkImageAspectFlags aspect_mask
;
1582 uint32_t base_layer
;
1583 uint32_t layer_count
;
1585 uint32_t level_count
;
1586 VkExtent3D extent
; /**< Extent of VkImageViewCreateInfo::baseMipLevel. */
1588 uint32_t descriptor
[16];
1590 /* Descriptor for use as a storage image as opposed to a sampled image.
1591 * This has a few differences for cube maps (e.g. type).
1593 uint32_t storage_descriptor
[16];
1596 struct radv_image_create_info
{
1597 const VkImageCreateInfo
*vk_info
;
1599 bool no_metadata_planes
;
1602 VkResult
radv_image_create(VkDevice _device
,
1603 const struct radv_image_create_info
*info
,
1604 const VkAllocationCallbacks
* alloc
,
1608 radv_image_from_gralloc(VkDevice device_h
,
1609 const VkImageCreateInfo
*base_info
,
1610 const VkNativeBufferANDROID
*gralloc_info
,
1611 const VkAllocationCallbacks
*alloc
,
1612 VkImage
*out_image_h
);
1614 void radv_image_view_init(struct radv_image_view
*view
,
1615 struct radv_device
*device
,
1616 const VkImageViewCreateInfo
* pCreateInfo
);
1618 struct radv_buffer_view
{
1619 struct radeon_winsys_bo
*bo
;
1621 uint64_t range
; /**< VkBufferViewCreateInfo::range */
1624 void radv_buffer_view_init(struct radv_buffer_view
*view
,
1625 struct radv_device
*device
,
1626 const VkBufferViewCreateInfo
* pCreateInfo
);
1628 static inline struct VkExtent3D
1629 radv_sanitize_image_extent(const VkImageType imageType
,
1630 const struct VkExtent3D imageExtent
)
1632 switch (imageType
) {
1633 case VK_IMAGE_TYPE_1D
:
1634 return (VkExtent3D
) { imageExtent
.width
, 1, 1 };
1635 case VK_IMAGE_TYPE_2D
:
1636 return (VkExtent3D
) { imageExtent
.width
, imageExtent
.height
, 1 };
1637 case VK_IMAGE_TYPE_3D
:
1640 unreachable("invalid image type");
1644 static inline struct VkOffset3D
1645 radv_sanitize_image_offset(const VkImageType imageType
,
1646 const struct VkOffset3D imageOffset
)
1648 switch (imageType
) {
1649 case VK_IMAGE_TYPE_1D
:
1650 return (VkOffset3D
) { imageOffset
.x
, 0, 0 };
1651 case VK_IMAGE_TYPE_2D
:
1652 return (VkOffset3D
) { imageOffset
.x
, imageOffset
.y
, 0 };
1653 case VK_IMAGE_TYPE_3D
:
1656 unreachable("invalid image type");
1661 radv_image_extent_compare(const struct radv_image
*image
,
1662 const VkExtent3D
*extent
)
1664 if (extent
->width
!= image
->info
.width
||
1665 extent
->height
!= image
->info
.height
||
1666 extent
->depth
!= image
->info
.depth
)
1671 struct radv_sampler
{
1675 struct radv_color_buffer_info
{
1676 uint64_t cb_color_base
;
1677 uint64_t cb_color_cmask
;
1678 uint64_t cb_color_fmask
;
1679 uint64_t cb_dcc_base
;
1680 uint32_t cb_color_pitch
;
1681 uint32_t cb_color_slice
;
1682 uint32_t cb_color_view
;
1683 uint32_t cb_color_info
;
1684 uint32_t cb_color_attrib
;
1685 uint32_t cb_color_attrib2
;
1686 uint32_t cb_dcc_control
;
1687 uint32_t cb_color_cmask_slice
;
1688 uint32_t cb_color_fmask_slice
;
1691 struct radv_ds_buffer_info
{
1692 uint64_t db_z_read_base
;
1693 uint64_t db_stencil_read_base
;
1694 uint64_t db_z_write_base
;
1695 uint64_t db_stencil_write_base
;
1696 uint64_t db_htile_data_base
;
1697 uint32_t db_depth_info
;
1699 uint32_t db_stencil_info
;
1700 uint32_t db_depth_view
;
1701 uint32_t db_depth_size
;
1702 uint32_t db_depth_slice
;
1703 uint32_t db_htile_surface
;
1704 uint32_t pa_su_poly_offset_db_fmt_cntl
;
1705 uint32_t db_z_info2
;
1706 uint32_t db_stencil_info2
;
1710 struct radv_attachment_info
{
1712 struct radv_color_buffer_info cb
;
1713 struct radv_ds_buffer_info ds
;
1715 struct radv_image_view
*attachment
;
1718 struct radv_framebuffer
{
1723 uint32_t attachment_count
;
1724 struct radv_attachment_info attachments
[0];
1727 struct radv_subpass_barrier
{
1728 VkPipelineStageFlags src_stage_mask
;
1729 VkAccessFlags src_access_mask
;
1730 VkAccessFlags dst_access_mask
;
1733 void radv_subpass_barrier(struct radv_cmd_buffer
*cmd_buffer
,
1734 const struct radv_subpass_barrier
*barrier
);
1736 struct radv_subpass_attachment
{
1737 uint32_t attachment
;
1738 VkImageLayout layout
;
1741 struct radv_subpass
{
1742 uint32_t input_count
;
1743 uint32_t color_count
;
1744 struct radv_subpass_attachment
* input_attachments
;
1745 struct radv_subpass_attachment
* color_attachments
;
1746 struct radv_subpass_attachment
* resolve_attachments
;
1747 struct radv_subpass_attachment depth_stencil_attachment
;
1749 /** Subpass has at least one resolve attachment */
1752 struct radv_subpass_barrier start_barrier
;
1755 VkSampleCountFlagBits max_sample_count
;
1758 struct radv_render_pass_attachment
{
1761 VkAttachmentLoadOp load_op
;
1762 VkAttachmentLoadOp stencil_load_op
;
1763 VkImageLayout initial_layout
;
1764 VkImageLayout final_layout
;
1768 struct radv_render_pass
{
1769 uint32_t attachment_count
;
1770 uint32_t subpass_count
;
1771 struct radv_subpass_attachment
* subpass_attachments
;
1772 struct radv_render_pass_attachment
* attachments
;
1773 struct radv_subpass_barrier end_barrier
;
1774 struct radv_subpass subpasses
[0];
1777 VkResult
radv_device_init_meta(struct radv_device
*device
);
1778 void radv_device_finish_meta(struct radv_device
*device
);
1780 struct radv_query_pool
{
1781 struct radeon_winsys_bo
*bo
;
1783 uint32_t availability_offset
;
1787 uint32_t pipeline_stats_mask
;
1790 struct radv_semaphore
{
1791 /* use a winsys sem for non-exportable */
1792 struct radeon_winsys_sem
*sem
;
1794 uint32_t temp_syncobj
;
1797 void radv_set_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
1798 VkPipelineBindPoint bind_point
,
1799 struct radv_descriptor_set
*set
,
1803 radv_update_descriptor_sets(struct radv_device
*device
,
1804 struct radv_cmd_buffer
*cmd_buffer
,
1805 VkDescriptorSet overrideSet
,
1806 uint32_t descriptorWriteCount
,
1807 const VkWriteDescriptorSet
*pDescriptorWrites
,
1808 uint32_t descriptorCopyCount
,
1809 const VkCopyDescriptorSet
*pDescriptorCopies
);
1812 radv_update_descriptor_set_with_template(struct radv_device
*device
,
1813 struct radv_cmd_buffer
*cmd_buffer
,
1814 struct radv_descriptor_set
*set
,
1815 VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate
,
1818 void radv_meta_push_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
1819 VkPipelineBindPoint pipelineBindPoint
,
1820 VkPipelineLayout _layout
,
1822 uint32_t descriptorWriteCount
,
1823 const VkWriteDescriptorSet
*pDescriptorWrites
);
1825 void radv_initialize_dcc(struct radv_cmd_buffer
*cmd_buffer
,
1826 struct radv_image
*image
, uint32_t value
);
1829 struct radeon_winsys_fence
*fence
;
1830 struct wsi_fence
*fence_wsi
;
1835 uint32_t temp_syncobj
;
1838 /* radv_nir_to_llvm.c */
1839 struct radv_shader_variant_info
;
1840 struct radv_nir_compiler_options
;
1842 void radv_compile_gs_copy_shader(struct ac_llvm_compiler
*ac_llvm
,
1843 struct nir_shader
*geom_shader
,
1844 struct ac_shader_binary
*binary
,
1845 struct ac_shader_config
*config
,
1846 struct radv_shader_variant_info
*shader_info
,
1847 const struct radv_nir_compiler_options
*option
);
1849 void radv_compile_nir_shader(struct ac_llvm_compiler
*ac_llvm
,
1850 struct ac_shader_binary
*binary
,
1851 struct ac_shader_config
*config
,
1852 struct radv_shader_variant_info
*shader_info
,
1853 struct nir_shader
*const *nir
,
1855 const struct radv_nir_compiler_options
*options
);
1857 /* radv_shader_info.h */
1858 struct radv_shader_info
;
1860 void radv_nir_shader_info_pass(const struct nir_shader
*nir
,
1861 const struct radv_nir_compiler_options
*options
,
1862 struct radv_shader_info
*info
);
1864 struct radeon_winsys_sem
;
1866 #define RADV_DEFINE_HANDLE_CASTS(__radv_type, __VkType) \
1868 static inline struct __radv_type * \
1869 __radv_type ## _from_handle(__VkType _handle) \
1871 return (struct __radv_type *) _handle; \
1874 static inline __VkType \
1875 __radv_type ## _to_handle(struct __radv_type *_obj) \
1877 return (__VkType) _obj; \
1880 #define RADV_DEFINE_NONDISP_HANDLE_CASTS(__radv_type, __VkType) \
1882 static inline struct __radv_type * \
1883 __radv_type ## _from_handle(__VkType _handle) \
1885 return (struct __radv_type *)(uintptr_t) _handle; \
1888 static inline __VkType \
1889 __radv_type ## _to_handle(struct __radv_type *_obj) \
1891 return (__VkType)(uintptr_t) _obj; \
1894 #define RADV_FROM_HANDLE(__radv_type, __name, __handle) \
1895 struct __radv_type *__name = __radv_type ## _from_handle(__handle)
1897 RADV_DEFINE_HANDLE_CASTS(radv_cmd_buffer
, VkCommandBuffer
)
1898 RADV_DEFINE_HANDLE_CASTS(radv_device
, VkDevice
)
1899 RADV_DEFINE_HANDLE_CASTS(radv_instance
, VkInstance
)
1900 RADV_DEFINE_HANDLE_CASTS(radv_physical_device
, VkPhysicalDevice
)
1901 RADV_DEFINE_HANDLE_CASTS(radv_queue
, VkQueue
)
1903 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_cmd_pool
, VkCommandPool
)
1904 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer
, VkBuffer
)
1905 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer_view
, VkBufferView
)
1906 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_pool
, VkDescriptorPool
)
1907 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set
, VkDescriptorSet
)
1908 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set_layout
, VkDescriptorSetLayout
)
1909 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_update_template
, VkDescriptorUpdateTemplateKHR
)
1910 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_device_memory
, VkDeviceMemory
)
1911 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_fence
, VkFence
)
1912 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_event
, VkEvent
)
1913 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_framebuffer
, VkFramebuffer
)
1914 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image
, VkImage
)
1915 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image_view
, VkImageView
);
1916 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_cache
, VkPipelineCache
)
1917 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline
, VkPipeline
)
1918 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_layout
, VkPipelineLayout
)
1919 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_query_pool
, VkQueryPool
)
1920 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_render_pass
, VkRenderPass
)
1921 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_sampler
, VkSampler
)
1922 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_shader_module
, VkShaderModule
)
1923 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_semaphore
, VkSemaphore
)
1925 #endif /* RADV_PRIVATE_H */