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
;
609 #define RADV_QUEUE_GENERAL 0
610 #define RADV_QUEUE_COMPUTE 1
611 #define RADV_QUEUE_TRANSFER 2
613 #define RADV_MAX_QUEUE_FAMILIES 3
615 enum ring_type
radv_queue_family_to_ring(int f
);
618 VK_LOADER_DATA _loader_data
;
619 struct radv_device
* device
;
620 struct radeon_winsys_ctx
*hw_ctx
;
621 enum radeon_ctx_priority priority
;
622 uint32_t queue_family_index
;
624 VkDeviceQueueCreateFlags flags
;
626 uint32_t scratch_size
;
627 uint32_t compute_scratch_size
;
628 uint32_t esgs_ring_size
;
629 uint32_t gsvs_ring_size
;
631 bool has_sample_positions
;
633 struct radeon_winsys_bo
*scratch_bo
;
634 struct radeon_winsys_bo
*descriptor_bo
;
635 struct radeon_winsys_bo
*compute_scratch_bo
;
636 struct radeon_winsys_bo
*esgs_ring_bo
;
637 struct radeon_winsys_bo
*gsvs_ring_bo
;
638 struct radeon_winsys_bo
*tess_rings_bo
;
639 struct radeon_cmdbuf
*initial_preamble_cs
;
640 struct radeon_cmdbuf
*initial_full_flush_preamble_cs
;
641 struct radeon_cmdbuf
*continue_preamble_cs
;
644 struct radv_bo_list
{
645 struct radv_winsys_bo_list list
;
647 pthread_mutex_t mutex
;
651 VK_LOADER_DATA _loader_data
;
653 VkAllocationCallbacks alloc
;
655 struct radv_instance
* instance
;
656 struct radeon_winsys
*ws
;
658 struct radv_meta_state meta_state
;
660 struct radv_queue
*queues
[RADV_MAX_QUEUE_FAMILIES
];
661 int queue_count
[RADV_MAX_QUEUE_FAMILIES
];
662 struct radeon_cmdbuf
*empty_cs
[RADV_MAX_QUEUE_FAMILIES
];
664 bool always_use_syncobj
;
665 bool has_distributed_tess
;
668 uint32_t tess_offchip_block_dw_size
;
669 uint32_t scratch_waves
;
670 uint32_t dispatch_initiator
;
672 uint32_t gs_table_depth
;
674 /* MSAA sample locations.
675 * The first index is the sample index.
676 * The second index is the coordinate: X, Y. */
677 float sample_locations_1x
[1][2];
678 float sample_locations_2x
[2][2];
679 float sample_locations_4x
[4][2];
680 float sample_locations_8x
[8][2];
681 float sample_locations_16x
[16][2];
684 uint32_t gfx_init_size_dw
;
685 struct radeon_winsys_bo
*gfx_init
;
687 struct radeon_winsys_bo
*trace_bo
;
688 uint32_t *trace_id_ptr
;
690 /* Whether to keep shader debug info, for tracing or VK_AMD_shader_info */
691 bool keep_shader_info
;
693 struct radv_physical_device
*physical_device
;
695 /* Backup in-memory cache to be used if the app doesn't provide one */
696 struct radv_pipeline_cache
* mem_cache
;
699 * use different counters so MSAA MRTs get consecutive surface indices,
700 * even if MASK is allocated in between.
702 uint32_t image_mrt_offset_counter
;
703 uint32_t fmask_mrt_offset_counter
;
704 struct list_head shader_slabs
;
705 mtx_t shader_slab_mutex
;
707 /* For detecting VM faults reported by dmesg. */
708 uint64_t dmesg_timestamp
;
710 struct radv_device_extension_table enabled_extensions
;
712 /* Whether the driver uses a global BO list. */
713 bool use_global_bo_list
;
715 struct radv_bo_list bo_list
;
717 /* Whether anisotropy is forced with RADV_TEX_ANISO (-1 is disabled). */
721 struct radv_device_memory
{
722 struct radeon_winsys_bo
*bo
;
723 /* for dedicated allocations */
724 struct radv_image
*image
;
725 struct radv_buffer
*buffer
;
727 VkDeviceSize map_size
;
733 struct radv_descriptor_range
{
738 struct radv_descriptor_set
{
739 const struct radv_descriptor_set_layout
*layout
;
742 struct radeon_winsys_bo
*bo
;
744 uint32_t *mapped_ptr
;
745 struct radv_descriptor_range
*dynamic_descriptors
;
747 struct radeon_winsys_bo
*descriptors
[0];
750 struct radv_push_descriptor_set
752 struct radv_descriptor_set set
;
756 struct radv_descriptor_pool_entry
{
759 struct radv_descriptor_set
*set
;
762 struct radv_descriptor_pool
{
763 struct radeon_winsys_bo
*bo
;
765 uint64_t current_offset
;
768 uint8_t *host_memory_base
;
769 uint8_t *host_memory_ptr
;
770 uint8_t *host_memory_end
;
772 uint32_t entry_count
;
773 uint32_t max_entry_count
;
774 struct radv_descriptor_pool_entry entries
[0];
777 struct radv_descriptor_update_template_entry
{
778 VkDescriptorType descriptor_type
;
780 /* The number of descriptors to update */
781 uint32_t descriptor_count
;
783 /* Into mapped_ptr or dynamic_descriptors, in units of the respective array */
786 /* In dwords. Not valid/used for dynamic descriptors */
789 uint32_t buffer_offset
;
791 /* Only valid for combined image samplers and samplers */
792 uint16_t has_sampler
;
798 /* For push descriptors */
799 const uint32_t *immutable_samplers
;
802 struct radv_descriptor_update_template
{
803 uint32_t entry_count
;
804 VkPipelineBindPoint bind_point
;
805 struct radv_descriptor_update_template_entry entry
[0];
811 VkBufferUsageFlags usage
;
812 VkBufferCreateFlags flags
;
815 struct radeon_winsys_bo
* bo
;
821 enum radv_dynamic_state_bits
{
822 RADV_DYNAMIC_VIEWPORT
= 1 << 0,
823 RADV_DYNAMIC_SCISSOR
= 1 << 1,
824 RADV_DYNAMIC_LINE_WIDTH
= 1 << 2,
825 RADV_DYNAMIC_DEPTH_BIAS
= 1 << 3,
826 RADV_DYNAMIC_BLEND_CONSTANTS
= 1 << 4,
827 RADV_DYNAMIC_DEPTH_BOUNDS
= 1 << 5,
828 RADV_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6,
829 RADV_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7,
830 RADV_DYNAMIC_STENCIL_REFERENCE
= 1 << 8,
831 RADV_DYNAMIC_DISCARD_RECTANGLE
= 1 << 9,
832 RADV_DYNAMIC_ALL
= (1 << 10) - 1,
835 enum radv_cmd_dirty_bits
{
836 /* Keep the dynamic state dirty bits in sync with
837 * enum radv_dynamic_state_bits */
838 RADV_CMD_DIRTY_DYNAMIC_VIEWPORT
= 1 << 0,
839 RADV_CMD_DIRTY_DYNAMIC_SCISSOR
= 1 << 1,
840 RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH
= 1 << 2,
841 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS
= 1 << 3,
842 RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS
= 1 << 4,
843 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS
= 1 << 5,
844 RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6,
845 RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7,
846 RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE
= 1 << 8,
847 RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE
= 1 << 9,
848 RADV_CMD_DIRTY_DYNAMIC_ALL
= (1 << 10) - 1,
849 RADV_CMD_DIRTY_PIPELINE
= 1 << 10,
850 RADV_CMD_DIRTY_INDEX_BUFFER
= 1 << 11,
851 RADV_CMD_DIRTY_FRAMEBUFFER
= 1 << 12,
852 RADV_CMD_DIRTY_VERTEX_BUFFER
= 1 << 13,
853 RADV_CMD_DIRTY_STREAMOUT_BUFFER
= 1 << 14,
856 enum radv_cmd_flush_bits
{
857 RADV_CMD_FLAG_INV_ICACHE
= 1 << 0,
858 /* SMEM L1, other names: KCACHE, constant cache, DCACHE, data cache */
859 RADV_CMD_FLAG_INV_SMEM_L1
= 1 << 1,
860 /* VMEM L1 can optionally be bypassed (GLC=1). Other names: TC L1 */
861 RADV_CMD_FLAG_INV_VMEM_L1
= 1 << 2,
862 /* Used by everything except CB/DB, can be bypassed (SLC=1). Other names: TC L2 */
863 RADV_CMD_FLAG_INV_GLOBAL_L2
= 1 << 3,
864 /* Same as above, but only writes back and doesn't invalidate */
865 RADV_CMD_FLAG_WRITEBACK_GLOBAL_L2
= 1 << 4,
866 /* Framebuffer caches */
867 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
= 1 << 5,
868 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
= 1 << 6,
869 RADV_CMD_FLAG_FLUSH_AND_INV_DB
= 1 << 7,
870 RADV_CMD_FLAG_FLUSH_AND_INV_CB
= 1 << 8,
871 /* Engine synchronization. */
872 RADV_CMD_FLAG_VS_PARTIAL_FLUSH
= 1 << 9,
873 RADV_CMD_FLAG_PS_PARTIAL_FLUSH
= 1 << 10,
874 RADV_CMD_FLAG_CS_PARTIAL_FLUSH
= 1 << 11,
875 RADV_CMD_FLAG_VGT_FLUSH
= 1 << 12,
876 /* Pipeline query controls. */
877 RADV_CMD_FLAG_START_PIPELINE_STATS
= 1 << 13,
878 RADV_CMD_FLAG_STOP_PIPELINE_STATS
= 1 << 14,
879 RADV_CMD_FLAG_VGT_STREAMOUT_SYNC
= 1 << 15,
881 RADV_CMD_FLUSH_AND_INV_FRAMEBUFFER
= (RADV_CMD_FLAG_FLUSH_AND_INV_CB
|
882 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
|
883 RADV_CMD_FLAG_FLUSH_AND_INV_DB
|
884 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
)
887 struct radv_vertex_binding
{
888 struct radv_buffer
* buffer
;
892 struct radv_streamout_binding
{
893 struct radv_buffer
*buffer
;
898 struct radv_streamout_state
{
899 /* Mask of bound streamout buffers. */
900 uint8_t enabled_mask
;
902 /* External state that comes from the last vertex stage, it must be
903 * set explicitely when binding a new graphics pipeline.
905 uint16_t stride_in_dw
[MAX_SO_BUFFERS
];
906 uint32_t enabled_stream_buffers_mask
; /* stream0 buffers0-3 in 4 LSB */
908 /* State of VGT_STRMOUT_BUFFER_(CONFIG|END) */
909 uint32_t hw_enabled_mask
;
911 /* State of VGT_STRMOUT_(CONFIG|EN) */
912 bool streamout_enabled
;
915 struct radv_viewport_state
{
917 VkViewport viewports
[MAX_VIEWPORTS
];
920 struct radv_scissor_state
{
922 VkRect2D scissors
[MAX_SCISSORS
];
925 struct radv_discard_rectangle_state
{
927 VkRect2D rectangles
[MAX_DISCARD_RECTANGLES
];
930 struct radv_dynamic_state
{
932 * Bitmask of (1 << VK_DYNAMIC_STATE_*).
933 * Defines the set of saved dynamic state.
937 struct radv_viewport_state viewport
;
939 struct radv_scissor_state scissor
;
949 float blend_constants
[4];
959 } stencil_compare_mask
;
964 } stencil_write_mask
;
971 struct radv_discard_rectangle_state discard_rectangle
;
974 extern const struct radv_dynamic_state default_dynamic_state
;
977 radv_get_debug_option_name(int id
);
980 radv_get_perftest_option_name(int id
);
983 * Attachment state when recording a renderpass instance.
985 * The clear value is valid only if there exists a pending clear.
987 struct radv_attachment_state
{
988 VkImageAspectFlags pending_clear_aspects
;
989 uint32_t cleared_views
;
990 VkClearValue clear_value
;
991 VkImageLayout current_layout
;
994 struct radv_descriptor_state
{
995 struct radv_descriptor_set
*sets
[MAX_SETS
];
998 struct radv_push_descriptor_set push_set
;
1000 uint32_t dynamic_buffers
[4 * MAX_DYNAMIC_BUFFERS
];
1003 struct radv_cmd_state
{
1004 /* Vertex descriptors */
1011 uint32_t prefetch_L2_mask
;
1013 struct radv_pipeline
* pipeline
;
1014 struct radv_pipeline
* emitted_pipeline
;
1015 struct radv_pipeline
* compute_pipeline
;
1016 struct radv_pipeline
* emitted_compute_pipeline
;
1017 struct radv_framebuffer
* framebuffer
;
1018 struct radv_render_pass
* pass
;
1019 const struct radv_subpass
* subpass
;
1020 struct radv_dynamic_state dynamic
;
1021 struct radv_attachment_state
* attachments
;
1022 struct radv_streamout_state streamout
;
1023 VkRect2D render_area
;
1026 struct radv_buffer
*index_buffer
;
1027 uint64_t index_offset
;
1028 uint32_t index_type
;
1029 uint32_t max_index_count
;
1031 int32_t last_index_type
;
1033 int32_t last_primitive_reset_en
;
1034 uint32_t last_primitive_reset_index
;
1035 enum radv_cmd_flush_bits flush_bits
;
1036 unsigned active_occlusion_queries
;
1037 bool perfect_occlusion_queries_enabled
;
1038 unsigned active_pipeline_queries
;
1041 uint32_t last_ia_multi_vgt_param
;
1043 uint32_t last_num_instances
;
1044 uint32_t last_first_instance
;
1045 uint32_t last_vertex_offset
;
1047 /* Whether CP DMA is busy/idle. */
1050 /* Conditional rendering info. */
1051 int predication_type
; /* -1: disabled, 0: normal, 1: inverted */
1052 uint64_t predication_va
;
1055 struct radv_cmd_pool
{
1056 VkAllocationCallbacks alloc
;
1057 struct list_head cmd_buffers
;
1058 struct list_head free_cmd_buffers
;
1059 uint32_t queue_family_index
;
1062 struct radv_cmd_buffer_upload
{
1066 struct radeon_winsys_bo
*upload_bo
;
1067 struct list_head list
;
1070 enum radv_cmd_buffer_status
{
1071 RADV_CMD_BUFFER_STATUS_INVALID
,
1072 RADV_CMD_BUFFER_STATUS_INITIAL
,
1073 RADV_CMD_BUFFER_STATUS_RECORDING
,
1074 RADV_CMD_BUFFER_STATUS_EXECUTABLE
,
1075 RADV_CMD_BUFFER_STATUS_PENDING
,
1078 struct radv_cmd_buffer
{
1079 VK_LOADER_DATA _loader_data
;
1081 struct radv_device
* device
;
1083 struct radv_cmd_pool
* pool
;
1084 struct list_head pool_link
;
1086 VkCommandBufferUsageFlags usage_flags
;
1087 VkCommandBufferLevel level
;
1088 enum radv_cmd_buffer_status status
;
1089 struct radeon_cmdbuf
*cs
;
1090 struct radv_cmd_state state
;
1091 struct radv_vertex_binding vertex_bindings
[MAX_VBS
];
1092 struct radv_streamout_binding streamout_bindings
[MAX_SO_BUFFERS
];
1093 uint32_t queue_family_index
;
1095 uint8_t push_constants
[MAX_PUSH_CONSTANTS_SIZE
];
1096 VkShaderStageFlags push_constant_stages
;
1097 struct radv_descriptor_set meta_push_descriptors
;
1099 struct radv_descriptor_state descriptors
[VK_PIPELINE_BIND_POINT_RANGE_SIZE
];
1101 struct radv_cmd_buffer_upload upload
;
1103 uint32_t scratch_size_needed
;
1104 uint32_t compute_scratch_size_needed
;
1105 uint32_t esgs_ring_size_needed
;
1106 uint32_t gsvs_ring_size_needed
;
1107 bool tess_rings_needed
;
1108 bool sample_positions_needed
;
1110 VkResult record_result
;
1112 uint32_t gfx9_fence_offset
;
1113 struct radeon_winsys_bo
*gfx9_fence_bo
;
1114 uint32_t gfx9_fence_idx
;
1115 uint64_t gfx9_eop_bug_va
;
1118 * Whether a query pool has been resetted and we have to flush caches.
1120 bool pending_reset_query
;
1125 bool radv_cmd_buffer_uses_mec(struct radv_cmd_buffer
*cmd_buffer
);
1127 void si_emit_graphics(struct radv_physical_device
*physical_device
,
1128 struct radeon_cmdbuf
*cs
);
1129 void si_emit_compute(struct radv_physical_device
*physical_device
,
1130 struct radeon_cmdbuf
*cs
);
1132 void cik_create_gfx_config(struct radv_device
*device
);
1134 void si_write_viewport(struct radeon_cmdbuf
*cs
, int first_vp
,
1135 int count
, const VkViewport
*viewports
);
1136 void si_write_scissors(struct radeon_cmdbuf
*cs
, int first
,
1137 int count
, const VkRect2D
*scissors
,
1138 const VkViewport
*viewports
, bool can_use_guardband
);
1139 uint32_t si_get_ia_multi_vgt_param(struct radv_cmd_buffer
*cmd_buffer
,
1140 bool instanced_draw
, bool indirect_draw
,
1141 uint32_t draw_vertex_count
);
1142 void si_cs_emit_write_event_eop(struct radeon_cmdbuf
*cs
,
1143 enum chip_class chip_class
,
1145 unsigned event
, unsigned event_flags
,
1150 uint64_t gfx9_eop_bug_va
);
1152 void radv_cp_wait_mem(struct radeon_cmdbuf
*cs
, uint32_t op
, uint64_t va
,
1153 uint32_t ref
, uint32_t mask
);
1154 void si_cs_emit_cache_flush(struct radeon_cmdbuf
*cs
,
1155 enum chip_class chip_class
,
1156 uint32_t *fence_ptr
, uint64_t va
,
1158 enum radv_cmd_flush_bits flush_bits
,
1159 uint64_t gfx9_eop_bug_va
);
1160 void si_emit_cache_flush(struct radv_cmd_buffer
*cmd_buffer
);
1161 void si_emit_set_predication_state(struct radv_cmd_buffer
*cmd_buffer
,
1162 bool inverted
, uint64_t va
);
1163 void si_cp_dma_buffer_copy(struct radv_cmd_buffer
*cmd_buffer
,
1164 uint64_t src_va
, uint64_t dest_va
,
1166 void si_cp_dma_prefetch(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
1168 void si_cp_dma_clear_buffer(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
1169 uint64_t size
, unsigned value
);
1170 void si_cp_dma_wait_for_idle(struct radv_cmd_buffer
*cmd_buffer
);
1172 void radv_set_db_count_control(struct radv_cmd_buffer
*cmd_buffer
);
1174 radv_cmd_buffer_upload_alloc(struct radv_cmd_buffer
*cmd_buffer
,
1177 unsigned *out_offset
,
1180 radv_cmd_buffer_set_subpass(struct radv_cmd_buffer
*cmd_buffer
,
1181 const struct radv_subpass
*subpass
,
1184 radv_cmd_buffer_upload_data(struct radv_cmd_buffer
*cmd_buffer
,
1185 unsigned size
, unsigned alignmnet
,
1186 const void *data
, unsigned *out_offset
);
1188 void radv_cmd_buffer_clear_subpass(struct radv_cmd_buffer
*cmd_buffer
);
1189 void radv_cmd_buffer_resolve_subpass(struct radv_cmd_buffer
*cmd_buffer
);
1190 void radv_cmd_buffer_resolve_subpass_cs(struct radv_cmd_buffer
*cmd_buffer
);
1191 void radv_cmd_buffer_resolve_subpass_fs(struct radv_cmd_buffer
*cmd_buffer
);
1192 void radv_cayman_emit_msaa_sample_locs(struct radeon_cmdbuf
*cs
, int nr_samples
);
1193 unsigned radv_cayman_get_maxdist(int log_samples
);
1194 void radv_device_init_msaa(struct radv_device
*device
);
1196 void radv_update_ds_clear_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1197 struct radv_image
*image
,
1198 VkClearDepthStencilValue ds_clear_value
,
1199 VkImageAspectFlags aspects
);
1201 void radv_update_color_clear_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1202 struct radv_image
*image
,
1204 uint32_t color_values
[2]);
1206 void radv_update_fce_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1207 struct radv_image
*image
, bool value
);
1209 void radv_update_dcc_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1210 struct radv_image
*image
, bool value
);
1212 uint32_t radv_fill_buffer(struct radv_cmd_buffer
*cmd_buffer
,
1213 struct radeon_winsys_bo
*bo
,
1214 uint64_t offset
, uint64_t size
, uint32_t value
);
1215 void radv_cmd_buffer_trace_emit(struct radv_cmd_buffer
*cmd_buffer
);
1216 bool radv_get_memory_fd(struct radv_device
*device
,
1217 struct radv_device_memory
*memory
,
1221 radv_emit_shader_pointer_head(struct radeon_cmdbuf
*cs
,
1222 unsigned sh_offset
, unsigned pointer_count
,
1223 bool use_32bit_pointers
)
1225 radeon_emit(cs
, PKT3(PKT3_SET_SH_REG
, pointer_count
* (use_32bit_pointers
? 1 : 2), 0));
1226 radeon_emit(cs
, (sh_offset
- SI_SH_REG_OFFSET
) >> 2);
1230 radv_emit_shader_pointer_body(struct radv_device
*device
,
1231 struct radeon_cmdbuf
*cs
,
1232 uint64_t va
, bool use_32bit_pointers
)
1234 radeon_emit(cs
, va
);
1236 if (use_32bit_pointers
) {
1238 (va
>> 32) == device
->physical_device
->rad_info
.address32_hi
);
1240 radeon_emit(cs
, va
>> 32);
1245 radv_emit_shader_pointer(struct radv_device
*device
,
1246 struct radeon_cmdbuf
*cs
,
1247 uint32_t sh_offset
, uint64_t va
, bool global
)
1249 bool use_32bit_pointers
= !global
;
1251 radv_emit_shader_pointer_head(cs
, sh_offset
, 1, use_32bit_pointers
);
1252 radv_emit_shader_pointer_body(device
, cs
, va
, use_32bit_pointers
);
1255 static inline struct radv_descriptor_state
*
1256 radv_get_descriptors_state(struct radv_cmd_buffer
*cmd_buffer
,
1257 VkPipelineBindPoint bind_point
)
1259 assert(bind_point
== VK_PIPELINE_BIND_POINT_GRAPHICS
||
1260 bind_point
== VK_PIPELINE_BIND_POINT_COMPUTE
);
1261 return &cmd_buffer
->descriptors
[bind_point
];
1265 * Takes x,y,z as exact numbers of invocations, instead of blocks.
1267 * Limitations: Can't call normal dispatch functions without binding or rebinding
1268 * the compute pipeline.
1270 void radv_unaligned_dispatch(
1271 struct radv_cmd_buffer
*cmd_buffer
,
1277 struct radeon_winsys_bo
*bo
;
1281 struct radv_shader_module
;
1283 #define RADV_HASH_SHADER_IS_GEOM_COPY_SHADER (1 << 0)
1284 #define RADV_HASH_SHADER_SISCHED (1 << 1)
1285 #define RADV_HASH_SHADER_UNSAFE_MATH (1 << 2)
1287 radv_hash_shaders(unsigned char *hash
,
1288 const VkPipelineShaderStageCreateInfo
**stages
,
1289 const struct radv_pipeline_layout
*layout
,
1290 const struct radv_pipeline_key
*key
,
1293 static inline gl_shader_stage
1294 vk_to_mesa_shader_stage(VkShaderStageFlagBits vk_stage
)
1296 assert(__builtin_popcount(vk_stage
) == 1);
1297 return ffs(vk_stage
) - 1;
1300 static inline VkShaderStageFlagBits
1301 mesa_to_vk_shader_stage(gl_shader_stage mesa_stage
)
1303 return (1 << mesa_stage
);
1306 #define RADV_STAGE_MASK ((1 << MESA_SHADER_STAGES) - 1)
1308 #define radv_foreach_stage(stage, stage_bits) \
1309 for (gl_shader_stage stage, \
1310 __tmp = (gl_shader_stage)((stage_bits) & RADV_STAGE_MASK); \
1311 stage = __builtin_ffs(__tmp) - 1, __tmp; \
1312 __tmp &= ~(1 << (stage)))
1314 extern const VkFormat radv_fs_key_format_exemplars
[NUM_META_FS_KEYS
];
1315 unsigned radv_format_meta_fs_key(VkFormat format
);
1317 struct radv_multisample_state
{
1319 uint32_t pa_sc_line_cntl
;
1320 uint32_t pa_sc_mode_cntl_0
;
1321 uint32_t pa_sc_mode_cntl_1
;
1322 uint32_t pa_sc_aa_config
;
1323 uint32_t pa_sc_aa_mask
[2];
1324 unsigned num_samples
;
1327 struct radv_prim_vertex_count
{
1332 struct radv_vertex_elements_info
{
1333 uint32_t rsrc_word3
[MAX_VERTEX_ATTRIBS
];
1334 uint32_t format_size
[MAX_VERTEX_ATTRIBS
];
1335 uint32_t binding
[MAX_VERTEX_ATTRIBS
];
1336 uint32_t offset
[MAX_VERTEX_ATTRIBS
];
1340 struct radv_ia_multi_vgt_param_helpers
{
1342 bool partial_es_wave
;
1343 uint8_t primgroup_size
;
1344 bool wd_switch_on_eop
;
1345 bool ia_switch_on_eoi
;
1346 bool partial_vs_wave
;
1349 #define SI_GS_PER_ES 128
1351 struct radv_pipeline
{
1352 struct radv_device
* device
;
1353 struct radv_dynamic_state dynamic_state
;
1355 struct radv_pipeline_layout
* layout
;
1357 bool need_indirect_descriptor_sets
;
1358 struct radv_shader_variant
* shaders
[MESA_SHADER_STAGES
];
1359 struct radv_shader_variant
*gs_copy_shader
;
1360 VkShaderStageFlags active_stages
;
1362 struct radeon_cmdbuf cs
;
1364 struct radv_vertex_elements_info vertex_elements
;
1366 uint32_t binding_stride
[MAX_VBS
];
1368 uint32_t user_data_0
[MESA_SHADER_STAGES
];
1371 struct radv_multisample_state ms
;
1372 uint32_t spi_baryc_cntl
;
1373 bool prim_restart_enable
;
1374 unsigned esgs_ring_size
;
1375 unsigned gsvs_ring_size
;
1376 uint32_t vtx_base_sgpr
;
1377 struct radv_ia_multi_vgt_param_helpers ia_multi_vgt_param
;
1378 uint8_t vtx_emit_num
;
1379 struct radv_prim_vertex_count prim_vertex_count
;
1380 bool can_use_guardband
;
1381 uint32_t needed_dynamic_state
;
1382 bool disable_out_of_order_rast_for_occlusion
;
1384 /* Used for rbplus */
1385 uint32_t col_format
;
1386 uint32_t cb_target_mask
;
1391 unsigned scratch_bytes_per_wave
;
1393 /* Not NULL if graphics pipeline uses streamout. */
1394 struct radv_shader_variant
*streamout_shader
;
1397 static inline bool radv_pipeline_has_gs(const struct radv_pipeline
*pipeline
)
1399 return pipeline
->shaders
[MESA_SHADER_GEOMETRY
] ? true : false;
1402 static inline bool radv_pipeline_has_tess(const struct radv_pipeline
*pipeline
)
1404 return pipeline
->shaders
[MESA_SHADER_TESS_CTRL
] ? true : false;
1407 struct radv_userdata_info
*radv_lookup_user_sgpr(struct radv_pipeline
*pipeline
,
1408 gl_shader_stage stage
,
1411 struct radv_shader_variant
*radv_get_shader(struct radv_pipeline
*pipeline
,
1412 gl_shader_stage stage
);
1414 struct radv_graphics_pipeline_create_info
{
1416 bool db_depth_clear
;
1417 bool db_stencil_clear
;
1418 bool db_depth_disable_expclear
;
1419 bool db_stencil_disable_expclear
;
1420 bool db_flush_depth_inplace
;
1421 bool db_flush_stencil_inplace
;
1422 bool db_resummarize
;
1423 uint32_t custom_blend_mode
;
1427 radv_graphics_pipeline_create(VkDevice device
,
1428 VkPipelineCache cache
,
1429 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
1430 const struct radv_graphics_pipeline_create_info
*extra
,
1431 const VkAllocationCallbacks
*alloc
,
1432 VkPipeline
*pPipeline
);
1434 struct vk_format_description
;
1435 uint32_t radv_translate_buffer_dataformat(const struct vk_format_description
*desc
,
1436 int first_non_void
);
1437 uint32_t radv_translate_buffer_numformat(const struct vk_format_description
*desc
,
1438 int first_non_void
);
1439 uint32_t radv_translate_colorformat(VkFormat format
);
1440 uint32_t radv_translate_color_numformat(VkFormat format
,
1441 const struct vk_format_description
*desc
,
1442 int first_non_void
);
1443 uint32_t radv_colorformat_endian_swap(uint32_t colorformat
);
1444 unsigned radv_translate_colorswap(VkFormat format
, bool do_endian_swap
);
1445 uint32_t radv_translate_dbformat(VkFormat format
);
1446 uint32_t radv_translate_tex_dataformat(VkFormat format
,
1447 const struct vk_format_description
*desc
,
1448 int first_non_void
);
1449 uint32_t radv_translate_tex_numformat(VkFormat format
,
1450 const struct vk_format_description
*desc
,
1451 int first_non_void
);
1452 bool radv_format_pack_clear_color(VkFormat format
,
1453 uint32_t clear_vals
[2],
1454 VkClearColorValue
*value
);
1455 bool radv_is_colorbuffer_format_supported(VkFormat format
, bool *blendable
);
1456 bool radv_dcc_formats_compatible(VkFormat format1
,
1459 struct radv_fmask_info
{
1463 unsigned pitch_in_pixels
;
1464 unsigned bank_height
;
1465 unsigned slice_tile_max
;
1466 unsigned tile_mode_index
;
1467 unsigned tile_swizzle
;
1470 struct radv_cmask_info
{
1474 unsigned slice_tile_max
;
1479 /* The original VkFormat provided by the client. This may not match any
1480 * of the actual surface formats.
1483 VkImageAspectFlags aspects
;
1484 VkImageUsageFlags usage
; /**< Superset of VkImageCreateInfo::usage. */
1485 struct ac_surf_info info
;
1486 VkImageTiling tiling
; /** VkImageCreateInfo::tiling */
1487 VkImageCreateFlags flags
; /** VkImageCreateInfo::flags */
1492 unsigned queue_family_mask
;
1496 /* Set when bound */
1497 struct radeon_winsys_bo
*bo
;
1498 VkDeviceSize offset
;
1499 uint64_t dcc_offset
;
1500 uint64_t htile_offset
;
1501 bool tc_compatible_htile
;
1502 struct radeon_surf surface
;
1504 struct radv_fmask_info fmask
;
1505 struct radv_cmask_info cmask
;
1506 uint64_t clear_value_offset
;
1507 uint64_t fce_pred_offset
;
1508 uint64_t dcc_pred_offset
;
1511 * Metadata for the TC-compat zrange workaround. If the 32-bit value
1512 * stored at this offset is UINT_MAX, the driver will emit
1513 * DB_Z_INFO.ZRANGE_PRECISION=0, otherwise it will skip the
1514 * SET_CONTEXT_REG packet.
1516 uint64_t tc_compat_zrange_offset
;
1518 /* For VK_ANDROID_native_buffer, the WSI image owns the memory, */
1519 VkDeviceMemory owned_memory
;
1522 /* Whether the image has a htile that is known consistent with the contents of
1524 bool radv_layout_has_htile(const struct radv_image
*image
,
1525 VkImageLayout layout
,
1526 unsigned queue_mask
);
1528 /* Whether the image has a htile that is known consistent with the contents of
1529 * the image and is allowed to be in compressed form.
1531 * If this is false reads that don't use the htile should be able to return
1534 bool radv_layout_is_htile_compressed(const struct radv_image
*image
,
1535 VkImageLayout layout
,
1536 unsigned queue_mask
);
1538 bool radv_layout_can_fast_clear(const struct radv_image
*image
,
1539 VkImageLayout layout
,
1540 unsigned queue_mask
);
1542 bool radv_layout_dcc_compressed(const struct radv_image
*image
,
1543 VkImageLayout layout
,
1544 unsigned queue_mask
);
1547 * Return whether the image has CMASK metadata for color surfaces.
1550 radv_image_has_cmask(const struct radv_image
*image
)
1552 return image
->cmask
.size
;
1556 * Return whether the image has FMASK metadata for color surfaces.
1559 radv_image_has_fmask(const struct radv_image
*image
)
1561 return image
->fmask
.size
;
1565 * Return whether the image has DCC metadata for color surfaces.
1568 radv_image_has_dcc(const struct radv_image
*image
)
1570 return image
->surface
.dcc_size
;
1574 * Return whether DCC metadata is enabled for a level.
1577 radv_dcc_enabled(const struct radv_image
*image
, unsigned level
)
1579 return radv_image_has_dcc(image
) &&
1580 level
< image
->surface
.num_dcc_levels
;
1584 * Return whether the image has CB metadata.
1587 radv_image_has_CB_metadata(const struct radv_image
*image
)
1589 return radv_image_has_cmask(image
) ||
1590 radv_image_has_fmask(image
) ||
1591 radv_image_has_dcc(image
);
1595 * Return whether the image has HTILE metadata for depth surfaces.
1598 radv_image_has_htile(const struct radv_image
*image
)
1600 return image
->surface
.htile_size
;
1604 * Return whether HTILE metadata is enabled for a level.
1607 radv_htile_enabled(const struct radv_image
*image
, unsigned level
)
1609 return radv_image_has_htile(image
) && level
== 0;
1613 * Return whether the image is TC-compatible HTILE.
1616 radv_image_is_tc_compat_htile(const struct radv_image
*image
)
1618 return radv_image_has_htile(image
) && image
->tc_compatible_htile
;
1621 unsigned radv_image_queue_family_mask(const struct radv_image
*image
, uint32_t family
, uint32_t queue_family
);
1623 static inline uint32_t
1624 radv_get_layerCount(const struct radv_image
*image
,
1625 const VkImageSubresourceRange
*range
)
1627 return range
->layerCount
== VK_REMAINING_ARRAY_LAYERS
?
1628 image
->info
.array_size
- range
->baseArrayLayer
: range
->layerCount
;
1631 static inline uint32_t
1632 radv_get_levelCount(const struct radv_image
*image
,
1633 const VkImageSubresourceRange
*range
)
1635 return range
->levelCount
== VK_REMAINING_MIP_LEVELS
?
1636 image
->info
.levels
- range
->baseMipLevel
: range
->levelCount
;
1639 struct radeon_bo_metadata
;
1641 radv_init_metadata(struct radv_device
*device
,
1642 struct radv_image
*image
,
1643 struct radeon_bo_metadata
*metadata
);
1645 struct radv_image_view
{
1646 struct radv_image
*image
; /**< VkImageViewCreateInfo::image */
1647 struct radeon_winsys_bo
*bo
;
1649 VkImageViewType type
;
1650 VkImageAspectFlags aspect_mask
;
1652 uint32_t base_layer
;
1653 uint32_t layer_count
;
1655 uint32_t level_count
;
1656 VkExtent3D extent
; /**< Extent of VkImageViewCreateInfo::baseMipLevel. */
1658 uint32_t descriptor
[16];
1660 /* Descriptor for use as a storage image as opposed to a sampled image.
1661 * This has a few differences for cube maps (e.g. type).
1663 uint32_t storage_descriptor
[16];
1666 struct radv_image_create_info
{
1667 const VkImageCreateInfo
*vk_info
;
1669 bool no_metadata_planes
;
1672 VkResult
radv_image_create(VkDevice _device
,
1673 const struct radv_image_create_info
*info
,
1674 const VkAllocationCallbacks
* alloc
,
1678 radv_image_from_gralloc(VkDevice device_h
,
1679 const VkImageCreateInfo
*base_info
,
1680 const VkNativeBufferANDROID
*gralloc_info
,
1681 const VkAllocationCallbacks
*alloc
,
1682 VkImage
*out_image_h
);
1684 void radv_image_view_init(struct radv_image_view
*view
,
1685 struct radv_device
*device
,
1686 const VkImageViewCreateInfo
* pCreateInfo
);
1688 struct radv_buffer_view
{
1689 struct radeon_winsys_bo
*bo
;
1691 uint64_t range
; /**< VkBufferViewCreateInfo::range */
1694 void radv_buffer_view_init(struct radv_buffer_view
*view
,
1695 struct radv_device
*device
,
1696 const VkBufferViewCreateInfo
* pCreateInfo
);
1698 static inline struct VkExtent3D
1699 radv_sanitize_image_extent(const VkImageType imageType
,
1700 const struct VkExtent3D imageExtent
)
1702 switch (imageType
) {
1703 case VK_IMAGE_TYPE_1D
:
1704 return (VkExtent3D
) { imageExtent
.width
, 1, 1 };
1705 case VK_IMAGE_TYPE_2D
:
1706 return (VkExtent3D
) { imageExtent
.width
, imageExtent
.height
, 1 };
1707 case VK_IMAGE_TYPE_3D
:
1710 unreachable("invalid image type");
1714 static inline struct VkOffset3D
1715 radv_sanitize_image_offset(const VkImageType imageType
,
1716 const struct VkOffset3D imageOffset
)
1718 switch (imageType
) {
1719 case VK_IMAGE_TYPE_1D
:
1720 return (VkOffset3D
) { imageOffset
.x
, 0, 0 };
1721 case VK_IMAGE_TYPE_2D
:
1722 return (VkOffset3D
) { imageOffset
.x
, imageOffset
.y
, 0 };
1723 case VK_IMAGE_TYPE_3D
:
1726 unreachable("invalid image type");
1731 radv_image_extent_compare(const struct radv_image
*image
,
1732 const VkExtent3D
*extent
)
1734 if (extent
->width
!= image
->info
.width
||
1735 extent
->height
!= image
->info
.height
||
1736 extent
->depth
!= image
->info
.depth
)
1741 struct radv_sampler
{
1745 struct radv_color_buffer_info
{
1746 uint64_t cb_color_base
;
1747 uint64_t cb_color_cmask
;
1748 uint64_t cb_color_fmask
;
1749 uint64_t cb_dcc_base
;
1750 uint32_t cb_color_pitch
;
1751 uint32_t cb_color_slice
;
1752 uint32_t cb_color_view
;
1753 uint32_t cb_color_info
;
1754 uint32_t cb_color_attrib
;
1755 uint32_t cb_color_attrib2
;
1756 uint32_t cb_dcc_control
;
1757 uint32_t cb_color_cmask_slice
;
1758 uint32_t cb_color_fmask_slice
;
1761 struct radv_ds_buffer_info
{
1762 uint64_t db_z_read_base
;
1763 uint64_t db_stencil_read_base
;
1764 uint64_t db_z_write_base
;
1765 uint64_t db_stencil_write_base
;
1766 uint64_t db_htile_data_base
;
1767 uint32_t db_depth_info
;
1769 uint32_t db_stencil_info
;
1770 uint32_t db_depth_view
;
1771 uint32_t db_depth_size
;
1772 uint32_t db_depth_slice
;
1773 uint32_t db_htile_surface
;
1774 uint32_t pa_su_poly_offset_db_fmt_cntl
;
1775 uint32_t db_z_info2
;
1776 uint32_t db_stencil_info2
;
1780 struct radv_attachment_info
{
1782 struct radv_color_buffer_info cb
;
1783 struct radv_ds_buffer_info ds
;
1785 struct radv_image_view
*attachment
;
1788 struct radv_framebuffer
{
1793 uint32_t attachment_count
;
1794 struct radv_attachment_info attachments
[0];
1797 struct radv_subpass_barrier
{
1798 VkPipelineStageFlags src_stage_mask
;
1799 VkAccessFlags src_access_mask
;
1800 VkAccessFlags dst_access_mask
;
1803 void radv_subpass_barrier(struct radv_cmd_buffer
*cmd_buffer
,
1804 const struct radv_subpass_barrier
*barrier
);
1806 struct radv_subpass_attachment
{
1807 uint32_t attachment
;
1808 VkImageLayout layout
;
1811 struct radv_subpass
{
1812 uint32_t input_count
;
1813 uint32_t color_count
;
1814 struct radv_subpass_attachment
* input_attachments
;
1815 struct radv_subpass_attachment
* color_attachments
;
1816 struct radv_subpass_attachment
* resolve_attachments
;
1817 struct radv_subpass_attachment depth_stencil_attachment
;
1819 /** Subpass has at least one resolve attachment */
1822 struct radv_subpass_barrier start_barrier
;
1825 VkSampleCountFlagBits max_sample_count
;
1828 struct radv_render_pass_attachment
{
1831 VkAttachmentLoadOp load_op
;
1832 VkAttachmentLoadOp stencil_load_op
;
1833 VkImageLayout initial_layout
;
1834 VkImageLayout final_layout
;
1838 struct radv_render_pass
{
1839 uint32_t attachment_count
;
1840 uint32_t subpass_count
;
1841 struct radv_subpass_attachment
* subpass_attachments
;
1842 struct radv_render_pass_attachment
* attachments
;
1843 struct radv_subpass_barrier end_barrier
;
1844 struct radv_subpass subpasses
[0];
1847 VkResult
radv_device_init_meta(struct radv_device
*device
);
1848 void radv_device_finish_meta(struct radv_device
*device
);
1850 struct radv_query_pool
{
1851 struct radeon_winsys_bo
*bo
;
1853 uint32_t availability_offset
;
1857 uint32_t pipeline_stats_mask
;
1860 struct radv_semaphore
{
1861 /* use a winsys sem for non-exportable */
1862 struct radeon_winsys_sem
*sem
;
1864 uint32_t temp_syncobj
;
1867 void radv_set_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
1868 VkPipelineBindPoint bind_point
,
1869 struct radv_descriptor_set
*set
,
1873 radv_update_descriptor_sets(struct radv_device
*device
,
1874 struct radv_cmd_buffer
*cmd_buffer
,
1875 VkDescriptorSet overrideSet
,
1876 uint32_t descriptorWriteCount
,
1877 const VkWriteDescriptorSet
*pDescriptorWrites
,
1878 uint32_t descriptorCopyCount
,
1879 const VkCopyDescriptorSet
*pDescriptorCopies
);
1882 radv_update_descriptor_set_with_template(struct radv_device
*device
,
1883 struct radv_cmd_buffer
*cmd_buffer
,
1884 struct radv_descriptor_set
*set
,
1885 VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate
,
1888 void radv_meta_push_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
1889 VkPipelineBindPoint pipelineBindPoint
,
1890 VkPipelineLayout _layout
,
1892 uint32_t descriptorWriteCount
,
1893 const VkWriteDescriptorSet
*pDescriptorWrites
);
1895 void radv_initialize_dcc(struct radv_cmd_buffer
*cmd_buffer
,
1896 struct radv_image
*image
, uint32_t value
);
1899 struct radeon_winsys_fence
*fence
;
1900 struct wsi_fence
*fence_wsi
;
1905 uint32_t temp_syncobj
;
1908 /* radv_nir_to_llvm.c */
1909 struct radv_shader_variant_info
;
1910 struct radv_nir_compiler_options
;
1912 void radv_compile_gs_copy_shader(struct ac_llvm_compiler
*ac_llvm
,
1913 struct nir_shader
*geom_shader
,
1914 struct ac_shader_binary
*binary
,
1915 struct ac_shader_config
*config
,
1916 struct radv_shader_variant_info
*shader_info
,
1917 const struct radv_nir_compiler_options
*option
);
1919 void radv_compile_nir_shader(struct ac_llvm_compiler
*ac_llvm
,
1920 struct ac_shader_binary
*binary
,
1921 struct ac_shader_config
*config
,
1922 struct radv_shader_variant_info
*shader_info
,
1923 struct nir_shader
*const *nir
,
1925 const struct radv_nir_compiler_options
*options
);
1927 /* radv_shader_info.h */
1928 struct radv_shader_info
;
1930 void radv_nir_shader_info_pass(const struct nir_shader
*nir
,
1931 const struct radv_nir_compiler_options
*options
,
1932 struct radv_shader_info
*info
);
1934 struct radeon_winsys_sem
;
1936 #define RADV_DEFINE_HANDLE_CASTS(__radv_type, __VkType) \
1938 static inline struct __radv_type * \
1939 __radv_type ## _from_handle(__VkType _handle) \
1941 return (struct __radv_type *) _handle; \
1944 static inline __VkType \
1945 __radv_type ## _to_handle(struct __radv_type *_obj) \
1947 return (__VkType) _obj; \
1950 #define RADV_DEFINE_NONDISP_HANDLE_CASTS(__radv_type, __VkType) \
1952 static inline struct __radv_type * \
1953 __radv_type ## _from_handle(__VkType _handle) \
1955 return (struct __radv_type *)(uintptr_t) _handle; \
1958 static inline __VkType \
1959 __radv_type ## _to_handle(struct __radv_type *_obj) \
1961 return (__VkType)(uintptr_t) _obj; \
1964 #define RADV_FROM_HANDLE(__radv_type, __name, __handle) \
1965 struct __radv_type *__name = __radv_type ## _from_handle(__handle)
1967 RADV_DEFINE_HANDLE_CASTS(radv_cmd_buffer
, VkCommandBuffer
)
1968 RADV_DEFINE_HANDLE_CASTS(radv_device
, VkDevice
)
1969 RADV_DEFINE_HANDLE_CASTS(radv_instance
, VkInstance
)
1970 RADV_DEFINE_HANDLE_CASTS(radv_physical_device
, VkPhysicalDevice
)
1971 RADV_DEFINE_HANDLE_CASTS(radv_queue
, VkQueue
)
1973 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_cmd_pool
, VkCommandPool
)
1974 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer
, VkBuffer
)
1975 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer_view
, VkBufferView
)
1976 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_pool
, VkDescriptorPool
)
1977 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set
, VkDescriptorSet
)
1978 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set_layout
, VkDescriptorSetLayout
)
1979 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_update_template
, VkDescriptorUpdateTemplateKHR
)
1980 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_device_memory
, VkDeviceMemory
)
1981 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_fence
, VkFence
)
1982 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_event
, VkEvent
)
1983 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_framebuffer
, VkFramebuffer
)
1984 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image
, VkImage
)
1985 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image_view
, VkImageView
);
1986 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_cache
, VkPipelineCache
)
1987 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline
, VkPipeline
)
1988 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_layout
, VkPipelineLayout
)
1989 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_query_pool
, VkQueryPool
)
1990 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_render_pass
, VkRenderPass
)
1991 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_sampler
, VkSampler
)
1992 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_shader_module
, VkShaderModule
)
1993 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_semaphore
, VkSemaphore
)
1995 #endif /* RADV_PRIVATE_H */