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
102 #define MAX_SO_STREAMS 4
103 #define MAX_SO_BUFFERS 4
104 #define MAX_SO_OUTPUTS 64
106 #define NUM_DEPTH_CLEAR_PIPELINES 3
109 * This is the point we switch from using CP to compute shader
110 * for certain buffer operations.
112 #define RADV_BUFFER_OPS_CS_THRESHOLD 4096
114 #define RADV_BUFFER_UPDATE_THRESHOLD 1024
118 RADV_MEM_HEAP_VRAM_CPU_ACCESS
,
125 RADV_MEM_TYPE_GTT_WRITE_COMBINE
,
126 RADV_MEM_TYPE_VRAM_CPU_ACCESS
,
127 RADV_MEM_TYPE_GTT_CACHED
,
131 #define radv_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
133 static inline uint32_t
134 align_u32(uint32_t v
, uint32_t a
)
136 assert(a
!= 0 && a
== (a
& -a
));
137 return (v
+ a
- 1) & ~(a
- 1);
140 static inline uint32_t
141 align_u32_npot(uint32_t v
, uint32_t a
)
143 return (v
+ a
- 1) / a
* a
;
146 static inline uint64_t
147 align_u64(uint64_t v
, uint64_t a
)
149 assert(a
!= 0 && a
== (a
& -a
));
150 return (v
+ a
- 1) & ~(a
- 1);
153 static inline int32_t
154 align_i32(int32_t v
, int32_t a
)
156 assert(a
!= 0 && a
== (a
& -a
));
157 return (v
+ a
- 1) & ~(a
- 1);
160 /** Alignment must be a power of 2. */
162 radv_is_aligned(uintmax_t n
, uintmax_t a
)
164 assert(a
== (a
& -a
));
165 return (n
& (a
- 1)) == 0;
168 static inline uint32_t
169 round_up_u32(uint32_t v
, uint32_t a
)
171 return (v
+ a
- 1) / a
;
174 static inline uint64_t
175 round_up_u64(uint64_t v
, uint64_t a
)
177 return (v
+ a
- 1) / a
;
180 static inline uint32_t
181 radv_minify(uint32_t n
, uint32_t levels
)
183 if (unlikely(n
== 0))
186 return MAX2(n
>> levels
, 1);
189 radv_clamp_f(float f
, float min
, float max
)
202 radv_clear_mask(uint32_t *inout_mask
, uint32_t clear_mask
)
204 if (*inout_mask
& clear_mask
) {
205 *inout_mask
&= ~clear_mask
;
212 #define for_each_bit(b, dword) \
213 for (uint32_t __dword = (dword); \
214 (b) = __builtin_ffs(__dword) - 1, __dword; \
215 __dword &= ~(1 << (b)))
217 #define typed_memcpy(dest, src, count) ({ \
218 STATIC_ASSERT(sizeof(*src) == sizeof(*dest)); \
219 memcpy((dest), (src), (count) * sizeof(*(src))); \
222 /* Whenever we generate an error, pass it through this function. Useful for
223 * debugging, where we can break on it. Only call at error site, not when
224 * propagating errors. Might be useful to plug in a stack trace here.
227 struct radv_instance
;
229 VkResult
__vk_errorf(struct radv_instance
*instance
, VkResult error
, const char *file
, int line
, const char *format
, ...);
231 #define vk_error(instance, error) __vk_errorf(instance, error, __FILE__, __LINE__, NULL);
232 #define vk_errorf(instance, error, format, ...) __vk_errorf(instance, error, __FILE__, __LINE__, format, ## __VA_ARGS__);
234 void __radv_finishme(const char *file
, int line
, const char *format
, ...)
235 radv_printflike(3, 4);
236 void radv_loge(const char *format
, ...) radv_printflike(1, 2);
237 void radv_loge_v(const char *format
, va_list va
);
238 void radv_logi(const char *format
, ...) radv_printflike(1, 2);
239 void radv_logi_v(const char *format
, va_list va
);
242 * Print a FINISHME message, including its source location.
244 #define radv_finishme(format, ...) \
246 static bool reported = false; \
248 __radv_finishme(__FILE__, __LINE__, format, ##__VA_ARGS__); \
253 /* A non-fatal assert. Useful for debugging. */
255 #define radv_assert(x) ({ \
256 if (unlikely(!(x))) \
257 fprintf(stderr, "%s:%d ASSERT: %s\n", __FILE__, __LINE__, #x); \
260 #define radv_assert(x)
263 #define stub_return(v) \
265 radv_finishme("stub %s", __func__); \
271 radv_finishme("stub %s", __func__); \
275 void *radv_lookup_entrypoint_unchecked(const char *name
);
276 void *radv_lookup_entrypoint_checked(const char *name
,
277 uint32_t core_version
,
278 const struct radv_instance_extension_table
*instance
,
279 const struct radv_device_extension_table
*device
);
281 struct radv_physical_device
{
282 VK_LOADER_DATA _loader_data
;
284 struct radv_instance
* instance
;
286 struct radeon_winsys
*ws
;
287 struct radeon_info rad_info
;
289 char name
[VK_MAX_PHYSICAL_DEVICE_NAME_SIZE
];
290 uint8_t driver_uuid
[VK_UUID_SIZE
];
291 uint8_t device_uuid
[VK_UUID_SIZE
];
292 uint8_t cache_uuid
[VK_UUID_SIZE
];
296 struct wsi_device wsi_device
;
298 bool has_rbplus
; /* if RB+ register exist */
299 bool rbplus_allowed
; /* if RB+ is allowed */
300 bool has_clear_state
;
301 bool cpdma_prefetch_writes_memory
;
302 bool has_scissor_bug
;
304 bool has_out_of_order_rast
;
305 bool out_of_order_rast_allowed
;
307 /* Whether DCC should be enabled for MSAA textures. */
308 bool dcc_msaa_allowed
;
310 /* This is the drivers on-disk cache used as a fallback as opposed to
311 * the pipeline cache defined by apps.
313 struct disk_cache
* disk_cache
;
315 VkPhysicalDeviceMemoryProperties memory_properties
;
316 enum radv_mem_type mem_type_indices
[RADV_MEM_TYPE_COUNT
];
318 drmPciBusInfo bus_info
;
320 struct radv_device_extension_table supported_extensions
;
323 struct radv_instance
{
324 VK_LOADER_DATA _loader_data
;
326 VkAllocationCallbacks alloc
;
329 int physicalDeviceCount
;
330 struct radv_physical_device physicalDevices
[RADV_MAX_DRM_DEVICES
];
332 uint64_t debug_flags
;
333 uint64_t perftest_flags
;
335 struct vk_debug_report_instance debug_report_callbacks
;
337 struct radv_instance_extension_table enabled_extensions
;
340 VkResult
radv_init_wsi(struct radv_physical_device
*physical_device
);
341 void radv_finish_wsi(struct radv_physical_device
*physical_device
);
343 bool radv_instance_extension_supported(const char *name
);
344 uint32_t radv_physical_device_api_version(struct radv_physical_device
*dev
);
345 bool radv_physical_device_extension_supported(struct radv_physical_device
*dev
,
350 struct radv_pipeline_cache
{
351 struct radv_device
* device
;
352 pthread_mutex_t mutex
;
356 uint32_t kernel_count
;
357 struct cache_entry
** hash_table
;
360 VkAllocationCallbacks alloc
;
363 struct radv_pipeline_key
{
364 uint32_t instance_rate_inputs
;
365 uint32_t instance_rate_divisors
[MAX_VERTEX_ATTRIBS
];
366 uint64_t vertex_alpha_adjust
;
367 unsigned tess_input_vertices
;
371 uint8_t log2_ps_iter_samples
;
373 uint32_t has_multiview_view_index
: 1;
374 uint32_t optimisations_disabled
: 1;
378 radv_pipeline_cache_init(struct radv_pipeline_cache
*cache
,
379 struct radv_device
*device
);
381 radv_pipeline_cache_finish(struct radv_pipeline_cache
*cache
);
383 radv_pipeline_cache_load(struct radv_pipeline_cache
*cache
,
384 const void *data
, size_t size
);
386 struct radv_shader_variant
;
389 radv_create_shader_variants_from_pipeline_cache(struct radv_device
*device
,
390 struct radv_pipeline_cache
*cache
,
391 const unsigned char *sha1
,
392 struct radv_shader_variant
**variants
);
395 radv_pipeline_cache_insert_shaders(struct radv_device
*device
,
396 struct radv_pipeline_cache
*cache
,
397 const unsigned char *sha1
,
398 struct radv_shader_variant
**variants
,
399 const void *const *codes
,
400 const unsigned *code_sizes
);
402 enum radv_blit_ds_layout
{
403 RADV_BLIT_DS_LAYOUT_TILE_ENABLE
,
404 RADV_BLIT_DS_LAYOUT_TILE_DISABLE
,
405 RADV_BLIT_DS_LAYOUT_COUNT
,
408 static inline enum radv_blit_ds_layout
radv_meta_blit_ds_to_type(VkImageLayout layout
)
410 return (layout
== VK_IMAGE_LAYOUT_GENERAL
) ? RADV_BLIT_DS_LAYOUT_TILE_DISABLE
: RADV_BLIT_DS_LAYOUT_TILE_ENABLE
;
413 static inline VkImageLayout
radv_meta_blit_ds_to_layout(enum radv_blit_ds_layout ds_layout
)
415 return ds_layout
== RADV_BLIT_DS_LAYOUT_TILE_ENABLE
? VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
: VK_IMAGE_LAYOUT_GENERAL
;
418 enum radv_meta_dst_layout
{
419 RADV_META_DST_LAYOUT_GENERAL
,
420 RADV_META_DST_LAYOUT_OPTIMAL
,
421 RADV_META_DST_LAYOUT_COUNT
,
424 static inline enum radv_meta_dst_layout
radv_meta_dst_layout_from_layout(VkImageLayout layout
)
426 return (layout
== VK_IMAGE_LAYOUT_GENERAL
) ? RADV_META_DST_LAYOUT_GENERAL
: RADV_META_DST_LAYOUT_OPTIMAL
;
429 static inline VkImageLayout
radv_meta_dst_layout_to_layout(enum radv_meta_dst_layout layout
)
431 return layout
== RADV_META_DST_LAYOUT_OPTIMAL
? VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
: VK_IMAGE_LAYOUT_GENERAL
;
434 struct radv_meta_state
{
435 VkAllocationCallbacks alloc
;
437 struct radv_pipeline_cache cache
;
440 * For on-demand pipeline creation, makes sure that
441 * only one thread tries to build a pipeline at the same time.
446 * Use array element `i` for images with `2^i` samples.
449 VkRenderPass render_pass
[NUM_META_FS_KEYS
];
450 VkPipeline color_pipelines
[NUM_META_FS_KEYS
];
452 VkRenderPass depthstencil_rp
;
453 VkPipeline depth_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
454 VkPipeline stencil_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
455 VkPipeline depthstencil_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
456 } clear
[1 + MAX_SAMPLES_LOG2
];
458 VkPipelineLayout clear_color_p_layout
;
459 VkPipelineLayout clear_depth_p_layout
;
461 /* Optimized compute fast HTILE clear for stencil or depth only. */
462 VkPipeline clear_htile_mask_pipeline
;
463 VkPipelineLayout clear_htile_mask_p_layout
;
464 VkDescriptorSetLayout clear_htile_mask_ds_layout
;
467 VkRenderPass render_pass
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
469 /** Pipeline that blits from a 1D image. */
470 VkPipeline pipeline_1d_src
[NUM_META_FS_KEYS
];
472 /** Pipeline that blits from a 2D image. */
473 VkPipeline pipeline_2d_src
[NUM_META_FS_KEYS
];
475 /** Pipeline that blits from a 3D image. */
476 VkPipeline pipeline_3d_src
[NUM_META_FS_KEYS
];
478 VkRenderPass depth_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
479 VkPipeline depth_only_1d_pipeline
;
480 VkPipeline depth_only_2d_pipeline
;
481 VkPipeline depth_only_3d_pipeline
;
483 VkRenderPass stencil_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
484 VkPipeline stencil_only_1d_pipeline
;
485 VkPipeline stencil_only_2d_pipeline
;
486 VkPipeline stencil_only_3d_pipeline
;
487 VkPipelineLayout pipeline_layout
;
488 VkDescriptorSetLayout ds_layout
;
492 VkPipelineLayout p_layouts
[5];
493 VkDescriptorSetLayout ds_layouts
[5];
494 VkPipeline pipelines
[5][NUM_META_FS_KEYS
];
496 VkPipeline depth_only_pipeline
[5];
498 VkPipeline stencil_only_pipeline
[5];
499 } blit2d
[1 + MAX_SAMPLES_LOG2
];
501 VkRenderPass blit2d_render_passes
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
502 VkRenderPass blit2d_depth_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
503 VkRenderPass blit2d_stencil_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
506 VkPipelineLayout img_p_layout
;
507 VkDescriptorSetLayout img_ds_layout
;
509 VkPipeline pipeline_3d
;
512 VkPipelineLayout img_p_layout
;
513 VkDescriptorSetLayout img_ds_layout
;
515 VkPipeline pipeline_3d
;
518 VkPipelineLayout img_p_layout
;
519 VkDescriptorSetLayout img_ds_layout
;
523 VkPipelineLayout img_p_layout
;
524 VkDescriptorSetLayout img_ds_layout
;
526 VkPipeline pipeline_3d
;
529 VkPipelineLayout img_p_layout
;
530 VkDescriptorSetLayout img_ds_layout
;
534 VkPipelineLayout img_p_layout
;
535 VkDescriptorSetLayout img_ds_layout
;
537 VkPipeline pipeline_3d
;
540 VkPipelineLayout img_p_layout
;
541 VkDescriptorSetLayout img_ds_layout
;
546 VkPipelineLayout p_layout
;
547 VkPipeline pipeline
[NUM_META_FS_KEYS
];
548 VkRenderPass pass
[NUM_META_FS_KEYS
];
552 VkDescriptorSetLayout ds_layout
;
553 VkPipelineLayout p_layout
;
556 VkPipeline i_pipeline
;
557 VkPipeline srgb_pipeline
;
558 } rc
[MAX_SAMPLES_LOG2
];
562 VkDescriptorSetLayout ds_layout
;
563 VkPipelineLayout p_layout
;
566 VkRenderPass render_pass
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
567 VkPipeline pipeline
[NUM_META_FS_KEYS
];
568 } rc
[MAX_SAMPLES_LOG2
];
572 VkPipelineLayout p_layout
;
573 VkPipeline decompress_pipeline
;
574 VkPipeline resummarize_pipeline
;
576 } depth_decomp
[1 + MAX_SAMPLES_LOG2
];
579 VkPipelineLayout p_layout
;
580 VkPipeline cmask_eliminate_pipeline
;
581 VkPipeline fmask_decompress_pipeline
;
582 VkPipeline dcc_decompress_pipeline
;
585 VkDescriptorSetLayout dcc_decompress_compute_ds_layout
;
586 VkPipelineLayout dcc_decompress_compute_p_layout
;
587 VkPipeline dcc_decompress_compute_pipeline
;
591 VkPipelineLayout fill_p_layout
;
592 VkPipelineLayout copy_p_layout
;
593 VkDescriptorSetLayout fill_ds_layout
;
594 VkDescriptorSetLayout copy_ds_layout
;
595 VkPipeline fill_pipeline
;
596 VkPipeline copy_pipeline
;
600 VkDescriptorSetLayout ds_layout
;
601 VkPipelineLayout p_layout
;
602 VkPipeline occlusion_query_pipeline
;
603 VkPipeline pipeline_statistics_query_pipeline
;
604 VkPipeline tfb_query_pipeline
;
608 VkDescriptorSetLayout ds_layout
;
609 VkPipelineLayout p_layout
;
610 VkPipeline pipeline
[MAX_SAMPLES_LOG2
];
615 #define RADV_QUEUE_GENERAL 0
616 #define RADV_QUEUE_COMPUTE 1
617 #define RADV_QUEUE_TRANSFER 2
619 #define RADV_MAX_QUEUE_FAMILIES 3
621 enum ring_type
radv_queue_family_to_ring(int f
);
624 VK_LOADER_DATA _loader_data
;
625 struct radv_device
* device
;
626 struct radeon_winsys_ctx
*hw_ctx
;
627 enum radeon_ctx_priority priority
;
628 uint32_t queue_family_index
;
630 VkDeviceQueueCreateFlags flags
;
632 uint32_t scratch_size
;
633 uint32_t compute_scratch_size
;
634 uint32_t esgs_ring_size
;
635 uint32_t gsvs_ring_size
;
637 bool has_sample_positions
;
639 struct radeon_winsys_bo
*scratch_bo
;
640 struct radeon_winsys_bo
*descriptor_bo
;
641 struct radeon_winsys_bo
*compute_scratch_bo
;
642 struct radeon_winsys_bo
*esgs_ring_bo
;
643 struct radeon_winsys_bo
*gsvs_ring_bo
;
644 struct radeon_winsys_bo
*tess_rings_bo
;
645 struct radeon_cmdbuf
*initial_preamble_cs
;
646 struct radeon_cmdbuf
*initial_full_flush_preamble_cs
;
647 struct radeon_cmdbuf
*continue_preamble_cs
;
650 struct radv_bo_list
{
651 struct radv_winsys_bo_list list
;
653 pthread_mutex_t mutex
;
657 VK_LOADER_DATA _loader_data
;
659 VkAllocationCallbacks alloc
;
661 struct radv_instance
* instance
;
662 struct radeon_winsys
*ws
;
664 struct radv_meta_state meta_state
;
666 struct radv_queue
*queues
[RADV_MAX_QUEUE_FAMILIES
];
667 int queue_count
[RADV_MAX_QUEUE_FAMILIES
];
668 struct radeon_cmdbuf
*empty_cs
[RADV_MAX_QUEUE_FAMILIES
];
670 bool always_use_syncobj
;
671 bool has_distributed_tess
;
674 uint32_t tess_offchip_block_dw_size
;
675 uint32_t scratch_waves
;
676 uint32_t dispatch_initiator
;
678 uint32_t gs_table_depth
;
680 /* MSAA sample locations.
681 * The first index is the sample index.
682 * The second index is the coordinate: X, Y. */
683 float sample_locations_1x
[1][2];
684 float sample_locations_2x
[2][2];
685 float sample_locations_4x
[4][2];
686 float sample_locations_8x
[8][2];
687 float sample_locations_16x
[16][2];
690 uint32_t gfx_init_size_dw
;
691 struct radeon_winsys_bo
*gfx_init
;
693 struct radeon_winsys_bo
*trace_bo
;
694 uint32_t *trace_id_ptr
;
696 /* Whether to keep shader debug info, for tracing or VK_AMD_shader_info */
697 bool keep_shader_info
;
699 struct radv_physical_device
*physical_device
;
701 /* Backup in-memory cache to be used if the app doesn't provide one */
702 struct radv_pipeline_cache
* mem_cache
;
705 * use different counters so MSAA MRTs get consecutive surface indices,
706 * even if MASK is allocated in between.
708 uint32_t image_mrt_offset_counter
;
709 uint32_t fmask_mrt_offset_counter
;
710 struct list_head shader_slabs
;
711 mtx_t shader_slab_mutex
;
713 /* For detecting VM faults reported by dmesg. */
714 uint64_t dmesg_timestamp
;
716 struct radv_device_extension_table enabled_extensions
;
718 /* Whether the driver uses a global BO list. */
719 bool use_global_bo_list
;
721 struct radv_bo_list bo_list
;
723 /* Whether anisotropy is forced with RADV_TEX_ANISO (-1 is disabled). */
727 struct radv_device_memory
{
728 struct radeon_winsys_bo
*bo
;
729 /* for dedicated allocations */
730 struct radv_image
*image
;
731 struct radv_buffer
*buffer
;
733 VkDeviceSize map_size
;
739 struct radv_descriptor_range
{
744 struct radv_descriptor_set
{
745 const struct radv_descriptor_set_layout
*layout
;
748 struct radeon_winsys_bo
*bo
;
750 uint32_t *mapped_ptr
;
751 struct radv_descriptor_range
*dynamic_descriptors
;
753 struct radeon_winsys_bo
*descriptors
[0];
756 struct radv_push_descriptor_set
758 struct radv_descriptor_set set
;
762 struct radv_descriptor_pool_entry
{
765 struct radv_descriptor_set
*set
;
768 struct radv_descriptor_pool
{
769 struct radeon_winsys_bo
*bo
;
771 uint64_t current_offset
;
774 uint8_t *host_memory_base
;
775 uint8_t *host_memory_ptr
;
776 uint8_t *host_memory_end
;
778 uint32_t entry_count
;
779 uint32_t max_entry_count
;
780 struct radv_descriptor_pool_entry entries
[0];
783 struct radv_descriptor_update_template_entry
{
784 VkDescriptorType descriptor_type
;
786 /* The number of descriptors to update */
787 uint32_t descriptor_count
;
789 /* Into mapped_ptr or dynamic_descriptors, in units of the respective array */
792 /* In dwords. Not valid/used for dynamic descriptors */
795 uint32_t buffer_offset
;
797 /* Only valid for combined image samplers and samplers */
798 uint16_t has_sampler
;
804 /* For push descriptors */
805 const uint32_t *immutable_samplers
;
808 struct radv_descriptor_update_template
{
809 uint32_t entry_count
;
810 VkPipelineBindPoint bind_point
;
811 struct radv_descriptor_update_template_entry entry
[0];
817 VkBufferUsageFlags usage
;
818 VkBufferCreateFlags flags
;
821 struct radeon_winsys_bo
* bo
;
827 enum radv_dynamic_state_bits
{
828 RADV_DYNAMIC_VIEWPORT
= 1 << 0,
829 RADV_DYNAMIC_SCISSOR
= 1 << 1,
830 RADV_DYNAMIC_LINE_WIDTH
= 1 << 2,
831 RADV_DYNAMIC_DEPTH_BIAS
= 1 << 3,
832 RADV_DYNAMIC_BLEND_CONSTANTS
= 1 << 4,
833 RADV_DYNAMIC_DEPTH_BOUNDS
= 1 << 5,
834 RADV_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6,
835 RADV_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7,
836 RADV_DYNAMIC_STENCIL_REFERENCE
= 1 << 8,
837 RADV_DYNAMIC_DISCARD_RECTANGLE
= 1 << 9,
838 RADV_DYNAMIC_ALL
= (1 << 10) - 1,
841 enum radv_cmd_dirty_bits
{
842 /* Keep the dynamic state dirty bits in sync with
843 * enum radv_dynamic_state_bits */
844 RADV_CMD_DIRTY_DYNAMIC_VIEWPORT
= 1 << 0,
845 RADV_CMD_DIRTY_DYNAMIC_SCISSOR
= 1 << 1,
846 RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH
= 1 << 2,
847 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS
= 1 << 3,
848 RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS
= 1 << 4,
849 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS
= 1 << 5,
850 RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6,
851 RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7,
852 RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE
= 1 << 8,
853 RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE
= 1 << 9,
854 RADV_CMD_DIRTY_DYNAMIC_ALL
= (1 << 10) - 1,
855 RADV_CMD_DIRTY_PIPELINE
= 1 << 10,
856 RADV_CMD_DIRTY_INDEX_BUFFER
= 1 << 11,
857 RADV_CMD_DIRTY_FRAMEBUFFER
= 1 << 12,
858 RADV_CMD_DIRTY_VERTEX_BUFFER
= 1 << 13,
859 RADV_CMD_DIRTY_STREAMOUT_BUFFER
= 1 << 14,
862 enum radv_cmd_flush_bits
{
863 RADV_CMD_FLAG_INV_ICACHE
= 1 << 0,
864 /* SMEM L1, other names: KCACHE, constant cache, DCACHE, data cache */
865 RADV_CMD_FLAG_INV_SMEM_L1
= 1 << 1,
866 /* VMEM L1 can optionally be bypassed (GLC=1). Other names: TC L1 */
867 RADV_CMD_FLAG_INV_VMEM_L1
= 1 << 2,
868 /* Used by everything except CB/DB, can be bypassed (SLC=1). Other names: TC L2 */
869 RADV_CMD_FLAG_INV_GLOBAL_L2
= 1 << 3,
870 /* Same as above, but only writes back and doesn't invalidate */
871 RADV_CMD_FLAG_WRITEBACK_GLOBAL_L2
= 1 << 4,
872 /* Framebuffer caches */
873 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
= 1 << 5,
874 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
= 1 << 6,
875 RADV_CMD_FLAG_FLUSH_AND_INV_DB
= 1 << 7,
876 RADV_CMD_FLAG_FLUSH_AND_INV_CB
= 1 << 8,
877 /* Engine synchronization. */
878 RADV_CMD_FLAG_VS_PARTIAL_FLUSH
= 1 << 9,
879 RADV_CMD_FLAG_PS_PARTIAL_FLUSH
= 1 << 10,
880 RADV_CMD_FLAG_CS_PARTIAL_FLUSH
= 1 << 11,
881 RADV_CMD_FLAG_VGT_FLUSH
= 1 << 12,
882 /* Pipeline query controls. */
883 RADV_CMD_FLAG_START_PIPELINE_STATS
= 1 << 13,
884 RADV_CMD_FLAG_STOP_PIPELINE_STATS
= 1 << 14,
885 RADV_CMD_FLAG_VGT_STREAMOUT_SYNC
= 1 << 15,
887 RADV_CMD_FLUSH_AND_INV_FRAMEBUFFER
= (RADV_CMD_FLAG_FLUSH_AND_INV_CB
|
888 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
|
889 RADV_CMD_FLAG_FLUSH_AND_INV_DB
|
890 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
)
893 struct radv_vertex_binding
{
894 struct radv_buffer
* buffer
;
898 struct radv_streamout_binding
{
899 struct radv_buffer
*buffer
;
904 struct radv_streamout_state
{
905 /* Mask of bound streamout buffers. */
906 uint8_t enabled_mask
;
908 /* External state that comes from the last vertex stage, it must be
909 * set explicitely when binding a new graphics pipeline.
911 uint16_t stride_in_dw
[MAX_SO_BUFFERS
];
912 uint32_t enabled_stream_buffers_mask
; /* stream0 buffers0-3 in 4 LSB */
914 /* State of VGT_STRMOUT_BUFFER_(CONFIG|END) */
915 uint32_t hw_enabled_mask
;
917 /* State of VGT_STRMOUT_(CONFIG|EN) */
918 bool streamout_enabled
;
921 struct radv_viewport_state
{
923 VkViewport viewports
[MAX_VIEWPORTS
];
926 struct radv_scissor_state
{
928 VkRect2D scissors
[MAX_SCISSORS
];
931 struct radv_discard_rectangle_state
{
933 VkRect2D rectangles
[MAX_DISCARD_RECTANGLES
];
936 struct radv_dynamic_state
{
938 * Bitmask of (1 << VK_DYNAMIC_STATE_*).
939 * Defines the set of saved dynamic state.
943 struct radv_viewport_state viewport
;
945 struct radv_scissor_state scissor
;
955 float blend_constants
[4];
965 } stencil_compare_mask
;
970 } stencil_write_mask
;
977 struct radv_discard_rectangle_state discard_rectangle
;
980 extern const struct radv_dynamic_state default_dynamic_state
;
983 radv_get_debug_option_name(int id
);
986 radv_get_perftest_option_name(int id
);
989 * Attachment state when recording a renderpass instance.
991 * The clear value is valid only if there exists a pending clear.
993 struct radv_attachment_state
{
994 VkImageAspectFlags pending_clear_aspects
;
995 uint32_t cleared_views
;
996 VkClearValue clear_value
;
997 VkImageLayout current_layout
;
1000 struct radv_descriptor_state
{
1001 struct radv_descriptor_set
*sets
[MAX_SETS
];
1004 struct radv_push_descriptor_set push_set
;
1006 uint32_t dynamic_buffers
[4 * MAX_DYNAMIC_BUFFERS
];
1009 struct radv_cmd_state
{
1010 /* Vertex descriptors */
1017 uint32_t prefetch_L2_mask
;
1019 struct radv_pipeline
* pipeline
;
1020 struct radv_pipeline
* emitted_pipeline
;
1021 struct radv_pipeline
* compute_pipeline
;
1022 struct radv_pipeline
* emitted_compute_pipeline
;
1023 struct radv_framebuffer
* framebuffer
;
1024 struct radv_render_pass
* pass
;
1025 const struct radv_subpass
* subpass
;
1026 struct radv_dynamic_state dynamic
;
1027 struct radv_attachment_state
* attachments
;
1028 struct radv_streamout_state streamout
;
1029 VkRect2D render_area
;
1032 struct radv_buffer
*index_buffer
;
1033 uint64_t index_offset
;
1034 uint32_t index_type
;
1035 uint32_t max_index_count
;
1037 int32_t last_index_type
;
1039 int32_t last_primitive_reset_en
;
1040 uint32_t last_primitive_reset_index
;
1041 enum radv_cmd_flush_bits flush_bits
;
1042 unsigned active_occlusion_queries
;
1043 bool perfect_occlusion_queries_enabled
;
1044 unsigned active_pipeline_queries
;
1047 uint32_t last_ia_multi_vgt_param
;
1049 uint32_t last_num_instances
;
1050 uint32_t last_first_instance
;
1051 uint32_t last_vertex_offset
;
1053 /* Whether CP DMA is busy/idle. */
1056 /* Conditional rendering info. */
1057 int predication_type
; /* -1: disabled, 0: normal, 1: inverted */
1058 uint64_t predication_va
;
1061 struct radv_cmd_pool
{
1062 VkAllocationCallbacks alloc
;
1063 struct list_head cmd_buffers
;
1064 struct list_head free_cmd_buffers
;
1065 uint32_t queue_family_index
;
1068 struct radv_cmd_buffer_upload
{
1072 struct radeon_winsys_bo
*upload_bo
;
1073 struct list_head list
;
1076 enum radv_cmd_buffer_status
{
1077 RADV_CMD_BUFFER_STATUS_INVALID
,
1078 RADV_CMD_BUFFER_STATUS_INITIAL
,
1079 RADV_CMD_BUFFER_STATUS_RECORDING
,
1080 RADV_CMD_BUFFER_STATUS_EXECUTABLE
,
1081 RADV_CMD_BUFFER_STATUS_PENDING
,
1084 struct radv_cmd_buffer
{
1085 VK_LOADER_DATA _loader_data
;
1087 struct radv_device
* device
;
1089 struct radv_cmd_pool
* pool
;
1090 struct list_head pool_link
;
1092 VkCommandBufferUsageFlags usage_flags
;
1093 VkCommandBufferLevel level
;
1094 enum radv_cmd_buffer_status status
;
1095 struct radeon_cmdbuf
*cs
;
1096 struct radv_cmd_state state
;
1097 struct radv_vertex_binding vertex_bindings
[MAX_VBS
];
1098 struct radv_streamout_binding streamout_bindings
[MAX_SO_BUFFERS
];
1099 uint32_t queue_family_index
;
1101 uint8_t push_constants
[MAX_PUSH_CONSTANTS_SIZE
];
1102 VkShaderStageFlags push_constant_stages
;
1103 struct radv_descriptor_set meta_push_descriptors
;
1105 struct radv_descriptor_state descriptors
[VK_PIPELINE_BIND_POINT_RANGE_SIZE
];
1107 struct radv_cmd_buffer_upload upload
;
1109 uint32_t scratch_size_needed
;
1110 uint32_t compute_scratch_size_needed
;
1111 uint32_t esgs_ring_size_needed
;
1112 uint32_t gsvs_ring_size_needed
;
1113 bool tess_rings_needed
;
1114 bool sample_positions_needed
;
1116 VkResult record_result
;
1118 uint32_t gfx9_fence_offset
;
1119 struct radeon_winsys_bo
*gfx9_fence_bo
;
1120 uint32_t gfx9_fence_idx
;
1121 uint64_t gfx9_eop_bug_va
;
1124 * Whether a query pool has been resetted and we have to flush caches.
1126 bool pending_reset_query
;
1131 bool radv_cmd_buffer_uses_mec(struct radv_cmd_buffer
*cmd_buffer
);
1133 void si_emit_graphics(struct radv_physical_device
*physical_device
,
1134 struct radeon_cmdbuf
*cs
);
1135 void si_emit_compute(struct radv_physical_device
*physical_device
,
1136 struct radeon_cmdbuf
*cs
);
1138 void cik_create_gfx_config(struct radv_device
*device
);
1140 void si_write_viewport(struct radeon_cmdbuf
*cs
, int first_vp
,
1141 int count
, const VkViewport
*viewports
);
1142 void si_write_scissors(struct radeon_cmdbuf
*cs
, int first
,
1143 int count
, const VkRect2D
*scissors
,
1144 const VkViewport
*viewports
, bool can_use_guardband
);
1145 uint32_t si_get_ia_multi_vgt_param(struct radv_cmd_buffer
*cmd_buffer
,
1146 bool instanced_draw
, bool indirect_draw
,
1147 uint32_t draw_vertex_count
);
1148 void si_cs_emit_write_event_eop(struct radeon_cmdbuf
*cs
,
1149 enum chip_class chip_class
,
1151 unsigned event
, unsigned event_flags
,
1156 uint64_t gfx9_eop_bug_va
);
1158 void radv_cp_wait_mem(struct radeon_cmdbuf
*cs
, uint32_t op
, uint64_t va
,
1159 uint32_t ref
, uint32_t mask
);
1160 void si_cs_emit_cache_flush(struct radeon_cmdbuf
*cs
,
1161 enum chip_class chip_class
,
1162 uint32_t *fence_ptr
, uint64_t va
,
1164 enum radv_cmd_flush_bits flush_bits
,
1165 uint64_t gfx9_eop_bug_va
);
1166 void si_emit_cache_flush(struct radv_cmd_buffer
*cmd_buffer
);
1167 void si_emit_set_predication_state(struct radv_cmd_buffer
*cmd_buffer
,
1168 bool inverted
, uint64_t va
);
1169 void si_cp_dma_buffer_copy(struct radv_cmd_buffer
*cmd_buffer
,
1170 uint64_t src_va
, uint64_t dest_va
,
1172 void si_cp_dma_prefetch(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
1174 void si_cp_dma_clear_buffer(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
1175 uint64_t size
, unsigned value
);
1176 void si_cp_dma_wait_for_idle(struct radv_cmd_buffer
*cmd_buffer
);
1178 void radv_set_db_count_control(struct radv_cmd_buffer
*cmd_buffer
);
1180 radv_cmd_buffer_upload_alloc(struct radv_cmd_buffer
*cmd_buffer
,
1183 unsigned *out_offset
,
1186 radv_cmd_buffer_set_subpass(struct radv_cmd_buffer
*cmd_buffer
,
1187 const struct radv_subpass
*subpass
,
1190 radv_cmd_buffer_upload_data(struct radv_cmd_buffer
*cmd_buffer
,
1191 unsigned size
, unsigned alignmnet
,
1192 const void *data
, unsigned *out_offset
);
1194 void radv_cmd_buffer_clear_subpass(struct radv_cmd_buffer
*cmd_buffer
);
1195 void radv_cmd_buffer_resolve_subpass(struct radv_cmd_buffer
*cmd_buffer
);
1196 void radv_cmd_buffer_resolve_subpass_cs(struct radv_cmd_buffer
*cmd_buffer
);
1197 void radv_cmd_buffer_resolve_subpass_fs(struct radv_cmd_buffer
*cmd_buffer
);
1198 void radv_cayman_emit_msaa_sample_locs(struct radeon_cmdbuf
*cs
, int nr_samples
);
1199 unsigned radv_cayman_get_maxdist(int log_samples
);
1200 void radv_device_init_msaa(struct radv_device
*device
);
1202 void radv_update_ds_clear_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1203 struct radv_image
*image
,
1204 VkClearDepthStencilValue ds_clear_value
,
1205 VkImageAspectFlags aspects
);
1207 void radv_update_color_clear_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1208 struct radv_image
*image
,
1210 uint32_t color_values
[2]);
1212 void radv_update_fce_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1213 struct radv_image
*image
, bool value
);
1215 void radv_update_dcc_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1216 struct radv_image
*image
, bool value
);
1218 uint32_t radv_fill_buffer(struct radv_cmd_buffer
*cmd_buffer
,
1219 struct radeon_winsys_bo
*bo
,
1220 uint64_t offset
, uint64_t size
, uint32_t value
);
1221 void radv_cmd_buffer_trace_emit(struct radv_cmd_buffer
*cmd_buffer
);
1222 bool radv_get_memory_fd(struct radv_device
*device
,
1223 struct radv_device_memory
*memory
,
1227 radv_emit_shader_pointer_head(struct radeon_cmdbuf
*cs
,
1228 unsigned sh_offset
, unsigned pointer_count
,
1229 bool use_32bit_pointers
)
1231 radeon_emit(cs
, PKT3(PKT3_SET_SH_REG
, pointer_count
* (use_32bit_pointers
? 1 : 2), 0));
1232 radeon_emit(cs
, (sh_offset
- SI_SH_REG_OFFSET
) >> 2);
1236 radv_emit_shader_pointer_body(struct radv_device
*device
,
1237 struct radeon_cmdbuf
*cs
,
1238 uint64_t va
, bool use_32bit_pointers
)
1240 radeon_emit(cs
, va
);
1242 if (use_32bit_pointers
) {
1244 (va
>> 32) == device
->physical_device
->rad_info
.address32_hi
);
1246 radeon_emit(cs
, va
>> 32);
1251 radv_emit_shader_pointer(struct radv_device
*device
,
1252 struct radeon_cmdbuf
*cs
,
1253 uint32_t sh_offset
, uint64_t va
, bool global
)
1255 bool use_32bit_pointers
= !global
;
1257 radv_emit_shader_pointer_head(cs
, sh_offset
, 1, use_32bit_pointers
);
1258 radv_emit_shader_pointer_body(device
, cs
, va
, use_32bit_pointers
);
1261 static inline struct radv_descriptor_state
*
1262 radv_get_descriptors_state(struct radv_cmd_buffer
*cmd_buffer
,
1263 VkPipelineBindPoint bind_point
)
1265 assert(bind_point
== VK_PIPELINE_BIND_POINT_GRAPHICS
||
1266 bind_point
== VK_PIPELINE_BIND_POINT_COMPUTE
);
1267 return &cmd_buffer
->descriptors
[bind_point
];
1271 * Takes x,y,z as exact numbers of invocations, instead of blocks.
1273 * Limitations: Can't call normal dispatch functions without binding or rebinding
1274 * the compute pipeline.
1276 void radv_unaligned_dispatch(
1277 struct radv_cmd_buffer
*cmd_buffer
,
1283 struct radeon_winsys_bo
*bo
;
1287 struct radv_shader_module
;
1289 #define RADV_HASH_SHADER_IS_GEOM_COPY_SHADER (1 << 0)
1290 #define RADV_HASH_SHADER_SISCHED (1 << 1)
1291 #define RADV_HASH_SHADER_UNSAFE_MATH (1 << 2)
1293 radv_hash_shaders(unsigned char *hash
,
1294 const VkPipelineShaderStageCreateInfo
**stages
,
1295 const struct radv_pipeline_layout
*layout
,
1296 const struct radv_pipeline_key
*key
,
1299 static inline gl_shader_stage
1300 vk_to_mesa_shader_stage(VkShaderStageFlagBits vk_stage
)
1302 assert(__builtin_popcount(vk_stage
) == 1);
1303 return ffs(vk_stage
) - 1;
1306 static inline VkShaderStageFlagBits
1307 mesa_to_vk_shader_stage(gl_shader_stage mesa_stage
)
1309 return (1 << mesa_stage
);
1312 #define RADV_STAGE_MASK ((1 << MESA_SHADER_STAGES) - 1)
1314 #define radv_foreach_stage(stage, stage_bits) \
1315 for (gl_shader_stage stage, \
1316 __tmp = (gl_shader_stage)((stage_bits) & RADV_STAGE_MASK); \
1317 stage = __builtin_ffs(__tmp) - 1, __tmp; \
1318 __tmp &= ~(1 << (stage)))
1320 extern const VkFormat radv_fs_key_format_exemplars
[NUM_META_FS_KEYS
];
1321 unsigned radv_format_meta_fs_key(VkFormat format
);
1323 struct radv_multisample_state
{
1325 uint32_t pa_sc_line_cntl
;
1326 uint32_t pa_sc_mode_cntl_0
;
1327 uint32_t pa_sc_mode_cntl_1
;
1328 uint32_t pa_sc_aa_config
;
1329 uint32_t pa_sc_aa_mask
[2];
1330 unsigned num_samples
;
1333 struct radv_prim_vertex_count
{
1338 struct radv_vertex_elements_info
{
1339 uint32_t rsrc_word3
[MAX_VERTEX_ATTRIBS
];
1340 uint32_t format_size
[MAX_VERTEX_ATTRIBS
];
1341 uint32_t binding
[MAX_VERTEX_ATTRIBS
];
1342 uint32_t offset
[MAX_VERTEX_ATTRIBS
];
1346 struct radv_ia_multi_vgt_param_helpers
{
1348 bool partial_es_wave
;
1349 uint8_t primgroup_size
;
1350 bool wd_switch_on_eop
;
1351 bool ia_switch_on_eoi
;
1352 bool partial_vs_wave
;
1355 #define SI_GS_PER_ES 128
1357 struct radv_pipeline
{
1358 struct radv_device
* device
;
1359 struct radv_dynamic_state dynamic_state
;
1361 struct radv_pipeline_layout
* layout
;
1363 bool need_indirect_descriptor_sets
;
1364 struct radv_shader_variant
* shaders
[MESA_SHADER_STAGES
];
1365 struct radv_shader_variant
*gs_copy_shader
;
1366 VkShaderStageFlags active_stages
;
1368 struct radeon_cmdbuf cs
;
1370 struct radv_vertex_elements_info vertex_elements
;
1372 uint32_t binding_stride
[MAX_VBS
];
1374 uint32_t user_data_0
[MESA_SHADER_STAGES
];
1377 struct radv_multisample_state ms
;
1378 uint32_t spi_baryc_cntl
;
1379 bool prim_restart_enable
;
1380 unsigned esgs_ring_size
;
1381 unsigned gsvs_ring_size
;
1382 uint32_t vtx_base_sgpr
;
1383 struct radv_ia_multi_vgt_param_helpers ia_multi_vgt_param
;
1384 uint8_t vtx_emit_num
;
1385 struct radv_prim_vertex_count prim_vertex_count
;
1386 bool can_use_guardband
;
1387 uint32_t needed_dynamic_state
;
1388 bool disable_out_of_order_rast_for_occlusion
;
1390 /* Used for rbplus */
1391 uint32_t col_format
;
1392 uint32_t cb_target_mask
;
1397 unsigned scratch_bytes_per_wave
;
1399 /* Not NULL if graphics pipeline uses streamout. */
1400 struct radv_shader_variant
*streamout_shader
;
1403 static inline bool radv_pipeline_has_gs(const struct radv_pipeline
*pipeline
)
1405 return pipeline
->shaders
[MESA_SHADER_GEOMETRY
] ? true : false;
1408 static inline bool radv_pipeline_has_tess(const struct radv_pipeline
*pipeline
)
1410 return pipeline
->shaders
[MESA_SHADER_TESS_CTRL
] ? true : false;
1413 struct radv_userdata_info
*radv_lookup_user_sgpr(struct radv_pipeline
*pipeline
,
1414 gl_shader_stage stage
,
1417 struct radv_shader_variant
*radv_get_shader(struct radv_pipeline
*pipeline
,
1418 gl_shader_stage stage
);
1420 struct radv_graphics_pipeline_create_info
{
1422 bool db_depth_clear
;
1423 bool db_stencil_clear
;
1424 bool db_depth_disable_expclear
;
1425 bool db_stencil_disable_expclear
;
1426 bool db_flush_depth_inplace
;
1427 bool db_flush_stencil_inplace
;
1428 bool db_resummarize
;
1429 uint32_t custom_blend_mode
;
1433 radv_graphics_pipeline_create(VkDevice device
,
1434 VkPipelineCache cache
,
1435 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
1436 const struct radv_graphics_pipeline_create_info
*extra
,
1437 const VkAllocationCallbacks
*alloc
,
1438 VkPipeline
*pPipeline
);
1440 struct vk_format_description
;
1441 uint32_t radv_translate_buffer_dataformat(const struct vk_format_description
*desc
,
1442 int first_non_void
);
1443 uint32_t radv_translate_buffer_numformat(const struct vk_format_description
*desc
,
1444 int first_non_void
);
1445 uint32_t radv_translate_colorformat(VkFormat format
);
1446 uint32_t radv_translate_color_numformat(VkFormat format
,
1447 const struct vk_format_description
*desc
,
1448 int first_non_void
);
1449 uint32_t radv_colorformat_endian_swap(uint32_t colorformat
);
1450 unsigned radv_translate_colorswap(VkFormat format
, bool do_endian_swap
);
1451 uint32_t radv_translate_dbformat(VkFormat format
);
1452 uint32_t radv_translate_tex_dataformat(VkFormat format
,
1453 const struct vk_format_description
*desc
,
1454 int first_non_void
);
1455 uint32_t radv_translate_tex_numformat(VkFormat format
,
1456 const struct vk_format_description
*desc
,
1457 int first_non_void
);
1458 bool radv_format_pack_clear_color(VkFormat format
,
1459 uint32_t clear_vals
[2],
1460 VkClearColorValue
*value
);
1461 bool radv_is_colorbuffer_format_supported(VkFormat format
, bool *blendable
);
1462 bool radv_dcc_formats_compatible(VkFormat format1
,
1465 struct radv_fmask_info
{
1469 unsigned pitch_in_pixels
;
1470 unsigned bank_height
;
1471 unsigned slice_tile_max
;
1472 unsigned tile_mode_index
;
1473 unsigned tile_swizzle
;
1476 struct radv_cmask_info
{
1480 unsigned slice_tile_max
;
1485 /* The original VkFormat provided by the client. This may not match any
1486 * of the actual surface formats.
1489 VkImageAspectFlags aspects
;
1490 VkImageUsageFlags usage
; /**< Superset of VkImageCreateInfo::usage. */
1491 struct ac_surf_info info
;
1492 VkImageTiling tiling
; /** VkImageCreateInfo::tiling */
1493 VkImageCreateFlags flags
; /** VkImageCreateInfo::flags */
1498 unsigned queue_family_mask
;
1502 /* Set when bound */
1503 struct radeon_winsys_bo
*bo
;
1504 VkDeviceSize offset
;
1505 uint64_t dcc_offset
;
1506 uint64_t htile_offset
;
1507 bool tc_compatible_htile
;
1508 struct radeon_surf surface
;
1510 struct radv_fmask_info fmask
;
1511 struct radv_cmask_info cmask
;
1512 uint64_t clear_value_offset
;
1513 uint64_t fce_pred_offset
;
1514 uint64_t dcc_pred_offset
;
1517 * Metadata for the TC-compat zrange workaround. If the 32-bit value
1518 * stored at this offset is UINT_MAX, the driver will emit
1519 * DB_Z_INFO.ZRANGE_PRECISION=0, otherwise it will skip the
1520 * SET_CONTEXT_REG packet.
1522 uint64_t tc_compat_zrange_offset
;
1524 /* For VK_ANDROID_native_buffer, the WSI image owns the memory, */
1525 VkDeviceMemory owned_memory
;
1528 /* Whether the image has a htile that is known consistent with the contents of
1530 bool radv_layout_has_htile(const struct radv_image
*image
,
1531 VkImageLayout layout
,
1532 unsigned queue_mask
);
1534 /* Whether the image has a htile that is known consistent with the contents of
1535 * the image and is allowed to be in compressed form.
1537 * If this is false reads that don't use the htile should be able to return
1540 bool radv_layout_is_htile_compressed(const struct radv_image
*image
,
1541 VkImageLayout layout
,
1542 unsigned queue_mask
);
1544 bool radv_layout_can_fast_clear(const struct radv_image
*image
,
1545 VkImageLayout layout
,
1546 unsigned queue_mask
);
1548 bool radv_layout_dcc_compressed(const struct radv_image
*image
,
1549 VkImageLayout layout
,
1550 unsigned queue_mask
);
1553 * Return whether the image has CMASK metadata for color surfaces.
1556 radv_image_has_cmask(const struct radv_image
*image
)
1558 return image
->cmask
.size
;
1562 * Return whether the image has FMASK metadata for color surfaces.
1565 radv_image_has_fmask(const struct radv_image
*image
)
1567 return image
->fmask
.size
;
1571 * Return whether the image has DCC metadata for color surfaces.
1574 radv_image_has_dcc(const struct radv_image
*image
)
1576 return image
->surface
.dcc_size
;
1580 * Return whether DCC metadata is enabled for a level.
1583 radv_dcc_enabled(const struct radv_image
*image
, unsigned level
)
1585 return radv_image_has_dcc(image
) &&
1586 level
< image
->surface
.num_dcc_levels
;
1590 * Return whether the image has CB metadata.
1593 radv_image_has_CB_metadata(const struct radv_image
*image
)
1595 return radv_image_has_cmask(image
) ||
1596 radv_image_has_fmask(image
) ||
1597 radv_image_has_dcc(image
);
1601 * Return whether the image has HTILE metadata for depth surfaces.
1604 radv_image_has_htile(const struct radv_image
*image
)
1606 return image
->surface
.htile_size
;
1610 * Return whether HTILE metadata is enabled for a level.
1613 radv_htile_enabled(const struct radv_image
*image
, unsigned level
)
1615 return radv_image_has_htile(image
) && level
== 0;
1619 * Return whether the image is TC-compatible HTILE.
1622 radv_image_is_tc_compat_htile(const struct radv_image
*image
)
1624 return radv_image_has_htile(image
) && image
->tc_compatible_htile
;
1627 unsigned radv_image_queue_family_mask(const struct radv_image
*image
, uint32_t family
, uint32_t queue_family
);
1629 static inline uint32_t
1630 radv_get_layerCount(const struct radv_image
*image
,
1631 const VkImageSubresourceRange
*range
)
1633 return range
->layerCount
== VK_REMAINING_ARRAY_LAYERS
?
1634 image
->info
.array_size
- range
->baseArrayLayer
: range
->layerCount
;
1637 static inline uint32_t
1638 radv_get_levelCount(const struct radv_image
*image
,
1639 const VkImageSubresourceRange
*range
)
1641 return range
->levelCount
== VK_REMAINING_MIP_LEVELS
?
1642 image
->info
.levels
- range
->baseMipLevel
: range
->levelCount
;
1645 struct radeon_bo_metadata
;
1647 radv_init_metadata(struct radv_device
*device
,
1648 struct radv_image
*image
,
1649 struct radeon_bo_metadata
*metadata
);
1651 struct radv_image_view
{
1652 struct radv_image
*image
; /**< VkImageViewCreateInfo::image */
1653 struct radeon_winsys_bo
*bo
;
1655 VkImageViewType type
;
1656 VkImageAspectFlags aspect_mask
;
1658 uint32_t base_layer
;
1659 uint32_t layer_count
;
1661 uint32_t level_count
;
1662 VkExtent3D extent
; /**< Extent of VkImageViewCreateInfo::baseMipLevel. */
1664 uint32_t descriptor
[16];
1666 /* Descriptor for use as a storage image as opposed to a sampled image.
1667 * This has a few differences for cube maps (e.g. type).
1669 uint32_t storage_descriptor
[16];
1672 struct radv_image_create_info
{
1673 const VkImageCreateInfo
*vk_info
;
1675 bool no_metadata_planes
;
1678 VkResult
radv_image_create(VkDevice _device
,
1679 const struct radv_image_create_info
*info
,
1680 const VkAllocationCallbacks
* alloc
,
1684 radv_image_from_gralloc(VkDevice device_h
,
1685 const VkImageCreateInfo
*base_info
,
1686 const VkNativeBufferANDROID
*gralloc_info
,
1687 const VkAllocationCallbacks
*alloc
,
1688 VkImage
*out_image_h
);
1690 void radv_image_view_init(struct radv_image_view
*view
,
1691 struct radv_device
*device
,
1692 const VkImageViewCreateInfo
* pCreateInfo
);
1694 struct radv_buffer_view
{
1695 struct radeon_winsys_bo
*bo
;
1697 uint64_t range
; /**< VkBufferViewCreateInfo::range */
1700 void radv_buffer_view_init(struct radv_buffer_view
*view
,
1701 struct radv_device
*device
,
1702 const VkBufferViewCreateInfo
* pCreateInfo
);
1704 static inline struct VkExtent3D
1705 radv_sanitize_image_extent(const VkImageType imageType
,
1706 const struct VkExtent3D imageExtent
)
1708 switch (imageType
) {
1709 case VK_IMAGE_TYPE_1D
:
1710 return (VkExtent3D
) { imageExtent
.width
, 1, 1 };
1711 case VK_IMAGE_TYPE_2D
:
1712 return (VkExtent3D
) { imageExtent
.width
, imageExtent
.height
, 1 };
1713 case VK_IMAGE_TYPE_3D
:
1716 unreachable("invalid image type");
1720 static inline struct VkOffset3D
1721 radv_sanitize_image_offset(const VkImageType imageType
,
1722 const struct VkOffset3D imageOffset
)
1724 switch (imageType
) {
1725 case VK_IMAGE_TYPE_1D
:
1726 return (VkOffset3D
) { imageOffset
.x
, 0, 0 };
1727 case VK_IMAGE_TYPE_2D
:
1728 return (VkOffset3D
) { imageOffset
.x
, imageOffset
.y
, 0 };
1729 case VK_IMAGE_TYPE_3D
:
1732 unreachable("invalid image type");
1737 radv_image_extent_compare(const struct radv_image
*image
,
1738 const VkExtent3D
*extent
)
1740 if (extent
->width
!= image
->info
.width
||
1741 extent
->height
!= image
->info
.height
||
1742 extent
->depth
!= image
->info
.depth
)
1747 struct radv_sampler
{
1751 struct radv_color_buffer_info
{
1752 uint64_t cb_color_base
;
1753 uint64_t cb_color_cmask
;
1754 uint64_t cb_color_fmask
;
1755 uint64_t cb_dcc_base
;
1756 uint32_t cb_color_pitch
;
1757 uint32_t cb_color_slice
;
1758 uint32_t cb_color_view
;
1759 uint32_t cb_color_info
;
1760 uint32_t cb_color_attrib
;
1761 uint32_t cb_color_attrib2
;
1762 uint32_t cb_dcc_control
;
1763 uint32_t cb_color_cmask_slice
;
1764 uint32_t cb_color_fmask_slice
;
1767 struct radv_ds_buffer_info
{
1768 uint64_t db_z_read_base
;
1769 uint64_t db_stencil_read_base
;
1770 uint64_t db_z_write_base
;
1771 uint64_t db_stencil_write_base
;
1772 uint64_t db_htile_data_base
;
1773 uint32_t db_depth_info
;
1775 uint32_t db_stencil_info
;
1776 uint32_t db_depth_view
;
1777 uint32_t db_depth_size
;
1778 uint32_t db_depth_slice
;
1779 uint32_t db_htile_surface
;
1780 uint32_t pa_su_poly_offset_db_fmt_cntl
;
1781 uint32_t db_z_info2
;
1782 uint32_t db_stencil_info2
;
1786 struct radv_attachment_info
{
1788 struct radv_color_buffer_info cb
;
1789 struct radv_ds_buffer_info ds
;
1791 struct radv_image_view
*attachment
;
1794 struct radv_framebuffer
{
1799 uint32_t attachment_count
;
1800 struct radv_attachment_info attachments
[0];
1803 struct radv_subpass_barrier
{
1804 VkPipelineStageFlags src_stage_mask
;
1805 VkAccessFlags src_access_mask
;
1806 VkAccessFlags dst_access_mask
;
1809 void radv_subpass_barrier(struct radv_cmd_buffer
*cmd_buffer
,
1810 const struct radv_subpass_barrier
*barrier
);
1812 struct radv_subpass_attachment
{
1813 uint32_t attachment
;
1814 VkImageLayout layout
;
1817 struct radv_subpass
{
1818 uint32_t input_count
;
1819 uint32_t color_count
;
1820 struct radv_subpass_attachment
* input_attachments
;
1821 struct radv_subpass_attachment
* color_attachments
;
1822 struct radv_subpass_attachment
* resolve_attachments
;
1823 struct radv_subpass_attachment depth_stencil_attachment
;
1825 /** Subpass has at least one resolve attachment */
1828 struct radv_subpass_barrier start_barrier
;
1831 VkSampleCountFlagBits max_sample_count
;
1834 struct radv_render_pass_attachment
{
1837 VkAttachmentLoadOp load_op
;
1838 VkAttachmentLoadOp stencil_load_op
;
1839 VkImageLayout initial_layout
;
1840 VkImageLayout final_layout
;
1844 struct radv_render_pass
{
1845 uint32_t attachment_count
;
1846 uint32_t subpass_count
;
1847 struct radv_subpass_attachment
* subpass_attachments
;
1848 struct radv_render_pass_attachment
* attachments
;
1849 struct radv_subpass_barrier end_barrier
;
1850 struct radv_subpass subpasses
[0];
1853 VkResult
radv_device_init_meta(struct radv_device
*device
);
1854 void radv_device_finish_meta(struct radv_device
*device
);
1856 struct radv_query_pool
{
1857 struct radeon_winsys_bo
*bo
;
1859 uint32_t availability_offset
;
1863 uint32_t pipeline_stats_mask
;
1866 struct radv_semaphore
{
1867 /* use a winsys sem for non-exportable */
1868 struct radeon_winsys_sem
*sem
;
1870 uint32_t temp_syncobj
;
1873 void radv_set_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
1874 VkPipelineBindPoint bind_point
,
1875 struct radv_descriptor_set
*set
,
1879 radv_update_descriptor_sets(struct radv_device
*device
,
1880 struct radv_cmd_buffer
*cmd_buffer
,
1881 VkDescriptorSet overrideSet
,
1882 uint32_t descriptorWriteCount
,
1883 const VkWriteDescriptorSet
*pDescriptorWrites
,
1884 uint32_t descriptorCopyCount
,
1885 const VkCopyDescriptorSet
*pDescriptorCopies
);
1888 radv_update_descriptor_set_with_template(struct radv_device
*device
,
1889 struct radv_cmd_buffer
*cmd_buffer
,
1890 struct radv_descriptor_set
*set
,
1891 VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate
,
1894 void radv_meta_push_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
1895 VkPipelineBindPoint pipelineBindPoint
,
1896 VkPipelineLayout _layout
,
1898 uint32_t descriptorWriteCount
,
1899 const VkWriteDescriptorSet
*pDescriptorWrites
);
1901 void radv_initialize_dcc(struct radv_cmd_buffer
*cmd_buffer
,
1902 struct radv_image
*image
, uint32_t value
);
1904 void radv_initialize_fmask(struct radv_cmd_buffer
*cmd_buffer
,
1905 struct radv_image
*image
);
1908 struct radeon_winsys_fence
*fence
;
1909 struct wsi_fence
*fence_wsi
;
1914 uint32_t temp_syncobj
;
1917 /* radv_nir_to_llvm.c */
1918 struct radv_shader_variant_info
;
1919 struct radv_nir_compiler_options
;
1921 void radv_compile_gs_copy_shader(struct ac_llvm_compiler
*ac_llvm
,
1922 struct nir_shader
*geom_shader
,
1923 struct ac_shader_binary
*binary
,
1924 struct ac_shader_config
*config
,
1925 struct radv_shader_variant_info
*shader_info
,
1926 const struct radv_nir_compiler_options
*option
);
1928 void radv_compile_nir_shader(struct ac_llvm_compiler
*ac_llvm
,
1929 struct ac_shader_binary
*binary
,
1930 struct ac_shader_config
*config
,
1931 struct radv_shader_variant_info
*shader_info
,
1932 struct nir_shader
*const *nir
,
1934 const struct radv_nir_compiler_options
*options
);
1936 /* radv_shader_info.h */
1937 struct radv_shader_info
;
1939 void radv_nir_shader_info_pass(const struct nir_shader
*nir
,
1940 const struct radv_nir_compiler_options
*options
,
1941 struct radv_shader_info
*info
);
1943 struct radeon_winsys_sem
;
1945 #define RADV_DEFINE_HANDLE_CASTS(__radv_type, __VkType) \
1947 static inline struct __radv_type * \
1948 __radv_type ## _from_handle(__VkType _handle) \
1950 return (struct __radv_type *) _handle; \
1953 static inline __VkType \
1954 __radv_type ## _to_handle(struct __radv_type *_obj) \
1956 return (__VkType) _obj; \
1959 #define RADV_DEFINE_NONDISP_HANDLE_CASTS(__radv_type, __VkType) \
1961 static inline struct __radv_type * \
1962 __radv_type ## _from_handle(__VkType _handle) \
1964 return (struct __radv_type *)(uintptr_t) _handle; \
1967 static inline __VkType \
1968 __radv_type ## _to_handle(struct __radv_type *_obj) \
1970 return (__VkType)(uintptr_t) _obj; \
1973 #define RADV_FROM_HANDLE(__radv_type, __name, __handle) \
1974 struct __radv_type *__name = __radv_type ## _from_handle(__handle)
1976 RADV_DEFINE_HANDLE_CASTS(radv_cmd_buffer
, VkCommandBuffer
)
1977 RADV_DEFINE_HANDLE_CASTS(radv_device
, VkDevice
)
1978 RADV_DEFINE_HANDLE_CASTS(radv_instance
, VkInstance
)
1979 RADV_DEFINE_HANDLE_CASTS(radv_physical_device
, VkPhysicalDevice
)
1980 RADV_DEFINE_HANDLE_CASTS(radv_queue
, VkQueue
)
1982 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_cmd_pool
, VkCommandPool
)
1983 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer
, VkBuffer
)
1984 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer_view
, VkBufferView
)
1985 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_pool
, VkDescriptorPool
)
1986 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set
, VkDescriptorSet
)
1987 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set_layout
, VkDescriptorSetLayout
)
1988 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_update_template
, VkDescriptorUpdateTemplateKHR
)
1989 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_device_memory
, VkDeviceMemory
)
1990 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_fence
, VkFence
)
1991 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_event
, VkEvent
)
1992 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_framebuffer
, VkFramebuffer
)
1993 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image
, VkImage
)
1994 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image_view
, VkImageView
);
1995 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_cache
, VkPipelineCache
)
1996 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline
, VkPipeline
)
1997 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_layout
, VkPipelineLayout
)
1998 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_query_pool
, VkQueryPool
)
1999 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_render_pass
, VkRenderPass
)
2000 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_sampler
, VkSampler
)
2001 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_shader_module
, VkShaderModule
)
2002 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_semaphore
, VkSemaphore
)
2004 #endif /* RADV_PRIVATE_H */