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 "radv_descriptor_set.h"
62 #include "radv_extensions.h"
65 #include <llvm-c/TargetMachine.h>
67 /* Pre-declarations needed for WSI entrypoints */
70 typedef struct xcb_connection_t xcb_connection_t
;
71 typedef uint32_t xcb_visualid_t
;
72 typedef uint32_t xcb_window_t
;
74 #include <vulkan/vulkan.h>
75 #include <vulkan/vulkan_intel.h>
76 #include <vulkan/vk_icd.h>
77 #include <vulkan/vk_android_native_buffer.h>
79 #include "radv_entrypoints.h"
81 #include "wsi_common.h"
83 #define ATI_VENDOR_ID 0x1002
86 #define MAX_VERTEX_ATTRIBS 32
88 #define MAX_VIEWPORTS 16
89 #define MAX_SCISSORS 16
90 #define MAX_DISCARD_RECTANGLES 4
91 #define MAX_PUSH_CONSTANTS_SIZE 128
92 #define MAX_PUSH_DESCRIPTORS 32
93 #define MAX_DYNAMIC_UNIFORM_BUFFERS 16
94 #define MAX_DYNAMIC_STORAGE_BUFFERS 8
95 #define MAX_DYNAMIC_BUFFERS (MAX_DYNAMIC_UNIFORM_BUFFERS + MAX_DYNAMIC_STORAGE_BUFFERS)
96 #define MAX_SAMPLES_LOG2 4
97 #define NUM_META_FS_KEYS 13
98 #define RADV_MAX_DRM_DEVICES 8
101 #define NUM_DEPTH_CLEAR_PIPELINES 3
104 * This is the point we switch from using CP to compute shader
105 * for certain buffer operations.
107 #define RADV_BUFFER_OPS_CS_THRESHOLD 4096
111 RADV_MEM_HEAP_VRAM_CPU_ACCESS
,
118 RADV_MEM_TYPE_GTT_WRITE_COMBINE
,
119 RADV_MEM_TYPE_VRAM_CPU_ACCESS
,
120 RADV_MEM_TYPE_GTT_CACHED
,
124 #define radv_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
126 static inline uint32_t
127 align_u32(uint32_t v
, uint32_t a
)
129 assert(a
!= 0 && a
== (a
& -a
));
130 return (v
+ a
- 1) & ~(a
- 1);
133 static inline uint32_t
134 align_u32_npot(uint32_t v
, uint32_t a
)
136 return (v
+ a
- 1) / a
* a
;
139 static inline uint64_t
140 align_u64(uint64_t v
, uint64_t a
)
142 assert(a
!= 0 && a
== (a
& -a
));
143 return (v
+ a
- 1) & ~(a
- 1);
146 static inline int32_t
147 align_i32(int32_t v
, int32_t a
)
149 assert(a
!= 0 && a
== (a
& -a
));
150 return (v
+ a
- 1) & ~(a
- 1);
153 /** Alignment must be a power of 2. */
155 radv_is_aligned(uintmax_t n
, uintmax_t a
)
157 assert(a
== (a
& -a
));
158 return (n
& (a
- 1)) == 0;
161 static inline uint32_t
162 round_up_u32(uint32_t v
, uint32_t a
)
164 return (v
+ a
- 1) / a
;
167 static inline uint64_t
168 round_up_u64(uint64_t v
, uint64_t a
)
170 return (v
+ a
- 1) / a
;
173 static inline uint32_t
174 radv_minify(uint32_t n
, uint32_t levels
)
176 if (unlikely(n
== 0))
179 return MAX2(n
>> levels
, 1);
182 radv_clamp_f(float f
, float min
, float max
)
195 radv_clear_mask(uint32_t *inout_mask
, uint32_t clear_mask
)
197 if (*inout_mask
& clear_mask
) {
198 *inout_mask
&= ~clear_mask
;
205 #define for_each_bit(b, dword) \
206 for (uint32_t __dword = (dword); \
207 (b) = __builtin_ffs(__dword) - 1, __dword; \
208 __dword &= ~(1 << (b)))
210 #define typed_memcpy(dest, src, count) ({ \
211 STATIC_ASSERT(sizeof(*src) == sizeof(*dest)); \
212 memcpy((dest), (src), (count) * sizeof(*(src))); \
215 /* Whenever we generate an error, pass it through this function. Useful for
216 * debugging, where we can break on it. Only call at error site, not when
217 * propagating errors. Might be useful to plug in a stack trace here.
220 struct radv_instance
;
222 VkResult
__vk_errorf(struct radv_instance
*instance
, VkResult error
, const char *file
, int line
, const char *format
, ...);
224 #define vk_error(instance, error) __vk_errorf(instance, error, __FILE__, __LINE__, NULL);
225 #define vk_errorf(instance, error, format, ...) __vk_errorf(instance, error, __FILE__, __LINE__, format, ## __VA_ARGS__);
227 void __radv_finishme(const char *file
, int line
, const char *format
, ...)
228 radv_printflike(3, 4);
229 void radv_loge(const char *format
, ...) radv_printflike(1, 2);
230 void radv_loge_v(const char *format
, va_list va
);
231 void radv_logi(const char *format
, ...) radv_printflike(1, 2);
232 void radv_logi_v(const char *format
, va_list va
);
235 * Print a FINISHME message, including its source location.
237 #define radv_finishme(format, ...) \
239 static bool reported = false; \
241 __radv_finishme(__FILE__, __LINE__, format, ##__VA_ARGS__); \
246 /* A non-fatal assert. Useful for debugging. */
248 #define radv_assert(x) ({ \
249 if (unlikely(!(x))) \
250 fprintf(stderr, "%s:%d ASSERT: %s\n", __FILE__, __LINE__, #x); \
253 #define radv_assert(x)
256 #define stub_return(v) \
258 radv_finishme("stub %s", __func__); \
264 radv_finishme("stub %s", __func__); \
268 void *radv_lookup_entrypoint_unchecked(const char *name
);
269 void *radv_lookup_entrypoint_checked(const char *name
,
270 uint32_t core_version
,
271 const struct radv_instance_extension_table
*instance
,
272 const struct radv_device_extension_table
*device
);
274 struct radv_physical_device
{
275 VK_LOADER_DATA _loader_data
;
277 struct radv_instance
* instance
;
279 struct radeon_winsys
*ws
;
280 struct radeon_info rad_info
;
282 char name
[VK_MAX_PHYSICAL_DEVICE_NAME_SIZE
];
283 uint8_t driver_uuid
[VK_UUID_SIZE
];
284 uint8_t device_uuid
[VK_UUID_SIZE
];
285 uint8_t cache_uuid
[VK_UUID_SIZE
];
288 struct wsi_device wsi_device
;
290 bool has_rbplus
; /* if RB+ register exist */
291 bool rbplus_allowed
; /* if RB+ is allowed */
292 bool has_clear_state
;
293 bool cpdma_prefetch_writes_memory
;
294 bool has_scissor_bug
;
296 bool has_out_of_order_rast
;
297 bool out_of_order_rast_allowed
;
299 /* Whether DCC should be enabled for MSAA textures. */
300 bool dcc_msaa_allowed
;
302 /* This is the drivers on-disk cache used as a fallback as opposed to
303 * the pipeline cache defined by apps.
305 struct disk_cache
* disk_cache
;
307 VkPhysicalDeviceMemoryProperties memory_properties
;
308 enum radv_mem_type mem_type_indices
[RADV_MEM_TYPE_COUNT
];
310 struct radv_device_extension_table supported_extensions
;
313 struct radv_instance
{
314 VK_LOADER_DATA _loader_data
;
316 VkAllocationCallbacks alloc
;
319 int physicalDeviceCount
;
320 struct radv_physical_device physicalDevices
[RADV_MAX_DRM_DEVICES
];
322 uint64_t debug_flags
;
323 uint64_t perftest_flags
;
325 struct vk_debug_report_instance debug_report_callbacks
;
327 struct radv_instance_extension_table enabled_extensions
;
330 VkResult
radv_init_wsi(struct radv_physical_device
*physical_device
);
331 void radv_finish_wsi(struct radv_physical_device
*physical_device
);
333 bool radv_instance_extension_supported(const char *name
);
334 uint32_t radv_physical_device_api_version(struct radv_physical_device
*dev
);
335 bool radv_physical_device_extension_supported(struct radv_physical_device
*dev
,
340 struct radv_pipeline_cache
{
341 struct radv_device
* device
;
342 pthread_mutex_t mutex
;
346 uint32_t kernel_count
;
347 struct cache_entry
** hash_table
;
350 VkAllocationCallbacks alloc
;
353 struct radv_pipeline_key
{
354 uint32_t instance_rate_inputs
;
355 uint32_t instance_rate_divisors
[MAX_VERTEX_ATTRIBS
];
356 uint64_t vertex_alpha_adjust
;
357 unsigned tess_input_vertices
;
361 uint8_t log2_ps_iter_samples
;
362 uint8_t log2_num_samples
;
363 uint32_t has_multiview_view_index
: 1;
364 uint32_t optimisations_disabled
: 1;
368 radv_pipeline_cache_init(struct radv_pipeline_cache
*cache
,
369 struct radv_device
*device
);
371 radv_pipeline_cache_finish(struct radv_pipeline_cache
*cache
);
373 radv_pipeline_cache_load(struct radv_pipeline_cache
*cache
,
374 const void *data
, size_t size
);
376 struct radv_shader_variant
;
379 radv_create_shader_variants_from_pipeline_cache(struct radv_device
*device
,
380 struct radv_pipeline_cache
*cache
,
381 const unsigned char *sha1
,
382 struct radv_shader_variant
**variants
);
385 radv_pipeline_cache_insert_shaders(struct radv_device
*device
,
386 struct radv_pipeline_cache
*cache
,
387 const unsigned char *sha1
,
388 struct radv_shader_variant
**variants
,
389 const void *const *codes
,
390 const unsigned *code_sizes
);
392 enum radv_blit_ds_layout
{
393 RADV_BLIT_DS_LAYOUT_TILE_ENABLE
,
394 RADV_BLIT_DS_LAYOUT_TILE_DISABLE
,
395 RADV_BLIT_DS_LAYOUT_COUNT
,
398 static inline enum radv_blit_ds_layout
radv_meta_blit_ds_to_type(VkImageLayout layout
)
400 return (layout
== VK_IMAGE_LAYOUT_GENERAL
) ? RADV_BLIT_DS_LAYOUT_TILE_DISABLE
: RADV_BLIT_DS_LAYOUT_TILE_ENABLE
;
403 static inline VkImageLayout
radv_meta_blit_ds_to_layout(enum radv_blit_ds_layout ds_layout
)
405 return ds_layout
== RADV_BLIT_DS_LAYOUT_TILE_ENABLE
? VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
: VK_IMAGE_LAYOUT_GENERAL
;
408 enum radv_meta_dst_layout
{
409 RADV_META_DST_LAYOUT_GENERAL
,
410 RADV_META_DST_LAYOUT_OPTIMAL
,
411 RADV_META_DST_LAYOUT_COUNT
,
414 static inline enum radv_meta_dst_layout
radv_meta_dst_layout_from_layout(VkImageLayout layout
)
416 return (layout
== VK_IMAGE_LAYOUT_GENERAL
) ? RADV_META_DST_LAYOUT_GENERAL
: RADV_META_DST_LAYOUT_OPTIMAL
;
419 static inline VkImageLayout
radv_meta_dst_layout_to_layout(enum radv_meta_dst_layout layout
)
421 return layout
== RADV_META_DST_LAYOUT_OPTIMAL
? VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
: VK_IMAGE_LAYOUT_GENERAL
;
424 struct radv_meta_state
{
425 VkAllocationCallbacks alloc
;
427 struct radv_pipeline_cache cache
;
430 * Use array element `i` for images with `2^i` samples.
433 VkRenderPass render_pass
[NUM_META_FS_KEYS
];
434 VkPipeline color_pipelines
[NUM_META_FS_KEYS
];
436 VkRenderPass depthstencil_rp
;
437 VkPipeline depth_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
438 VkPipeline stencil_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
439 VkPipeline depthstencil_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
440 } clear
[1 + MAX_SAMPLES_LOG2
];
442 VkPipelineLayout clear_color_p_layout
;
443 VkPipelineLayout clear_depth_p_layout
;
445 VkRenderPass render_pass
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
447 /** Pipeline that blits from a 1D image. */
448 VkPipeline pipeline_1d_src
[NUM_META_FS_KEYS
];
450 /** Pipeline that blits from a 2D image. */
451 VkPipeline pipeline_2d_src
[NUM_META_FS_KEYS
];
453 /** Pipeline that blits from a 3D image. */
454 VkPipeline pipeline_3d_src
[NUM_META_FS_KEYS
];
456 VkRenderPass depth_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
457 VkPipeline depth_only_1d_pipeline
;
458 VkPipeline depth_only_2d_pipeline
;
459 VkPipeline depth_only_3d_pipeline
;
461 VkRenderPass stencil_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
462 VkPipeline stencil_only_1d_pipeline
;
463 VkPipeline stencil_only_2d_pipeline
;
464 VkPipeline stencil_only_3d_pipeline
;
465 VkPipelineLayout pipeline_layout
;
466 VkDescriptorSetLayout ds_layout
;
470 VkPipelineLayout p_layouts
[5];
471 VkDescriptorSetLayout ds_layouts
[5];
472 VkPipeline pipelines
[5][NUM_META_FS_KEYS
];
474 VkPipeline depth_only_pipeline
[5];
476 VkPipeline stencil_only_pipeline
[5];
477 } blit2d
[1 + MAX_SAMPLES_LOG2
];
479 VkRenderPass blit2d_render_passes
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
480 VkRenderPass blit2d_depth_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
481 VkRenderPass blit2d_stencil_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
484 VkPipelineLayout img_p_layout
;
485 VkDescriptorSetLayout img_ds_layout
;
487 VkPipeline pipeline_3d
;
490 VkPipelineLayout img_p_layout
;
491 VkDescriptorSetLayout img_ds_layout
;
493 VkPipeline pipeline_3d
;
496 VkPipelineLayout img_p_layout
;
497 VkDescriptorSetLayout img_ds_layout
;
499 VkPipeline pipeline_3d
;
502 VkPipelineLayout img_p_layout
;
503 VkDescriptorSetLayout img_ds_layout
;
505 VkPipeline pipeline_3d
;
509 VkPipelineLayout p_layout
;
510 VkPipeline pipeline
[NUM_META_FS_KEYS
];
511 VkRenderPass pass
[NUM_META_FS_KEYS
];
515 VkDescriptorSetLayout ds_layout
;
516 VkPipelineLayout p_layout
;
519 VkPipeline i_pipeline
;
520 VkPipeline srgb_pipeline
;
521 } rc
[MAX_SAMPLES_LOG2
];
525 VkDescriptorSetLayout ds_layout
;
526 VkPipelineLayout p_layout
;
529 VkRenderPass render_pass
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
530 VkPipeline pipeline
[NUM_META_FS_KEYS
];
531 } rc
[MAX_SAMPLES_LOG2
];
535 VkPipelineLayout p_layout
;
536 VkPipeline decompress_pipeline
;
537 VkPipeline resummarize_pipeline
;
539 } depth_decomp
[1 + MAX_SAMPLES_LOG2
];
542 VkPipelineLayout p_layout
;
543 VkPipeline cmask_eliminate_pipeline
;
544 VkPipeline fmask_decompress_pipeline
;
545 VkPipeline dcc_decompress_pipeline
;
548 VkDescriptorSetLayout dcc_decompress_compute_ds_layout
;
549 VkPipelineLayout dcc_decompress_compute_p_layout
;
550 VkPipeline dcc_decompress_compute_pipeline
;
554 VkPipelineLayout fill_p_layout
;
555 VkPipelineLayout copy_p_layout
;
556 VkDescriptorSetLayout fill_ds_layout
;
557 VkDescriptorSetLayout copy_ds_layout
;
558 VkPipeline fill_pipeline
;
559 VkPipeline copy_pipeline
;
563 VkDescriptorSetLayout ds_layout
;
564 VkPipelineLayout p_layout
;
565 VkPipeline occlusion_query_pipeline
;
566 VkPipeline pipeline_statistics_query_pipeline
;
571 #define RADV_QUEUE_GENERAL 0
572 #define RADV_QUEUE_COMPUTE 1
573 #define RADV_QUEUE_TRANSFER 2
575 #define RADV_MAX_QUEUE_FAMILIES 3
577 enum ring_type
radv_queue_family_to_ring(int f
);
580 VK_LOADER_DATA _loader_data
;
581 struct radv_device
* device
;
582 struct radeon_winsys_ctx
*hw_ctx
;
583 enum radeon_ctx_priority priority
;
584 uint32_t queue_family_index
;
586 VkDeviceQueueCreateFlags flags
;
588 uint32_t scratch_size
;
589 uint32_t compute_scratch_size
;
590 uint32_t esgs_ring_size
;
591 uint32_t gsvs_ring_size
;
593 bool has_sample_positions
;
595 struct radeon_winsys_bo
*scratch_bo
;
596 struct radeon_winsys_bo
*descriptor_bo
;
597 struct radeon_winsys_bo
*compute_scratch_bo
;
598 struct radeon_winsys_bo
*esgs_ring_bo
;
599 struct radeon_winsys_bo
*gsvs_ring_bo
;
600 struct radeon_winsys_bo
*tess_rings_bo
;
601 struct radeon_winsys_cs
*initial_preamble_cs
;
602 struct radeon_winsys_cs
*initial_full_flush_preamble_cs
;
603 struct radeon_winsys_cs
*continue_preamble_cs
;
606 struct radv_bo_list
{
607 struct radv_winsys_bo_list list
;
609 pthread_mutex_t mutex
;
613 VK_LOADER_DATA _loader_data
;
615 VkAllocationCallbacks alloc
;
617 struct radv_instance
* instance
;
618 struct radeon_winsys
*ws
;
620 struct radv_meta_state meta_state
;
622 struct radv_queue
*queues
[RADV_MAX_QUEUE_FAMILIES
];
623 int queue_count
[RADV_MAX_QUEUE_FAMILIES
];
624 struct radeon_winsys_cs
*empty_cs
[RADV_MAX_QUEUE_FAMILIES
];
626 bool always_use_syncobj
;
627 bool has_distributed_tess
;
630 uint32_t tess_offchip_block_dw_size
;
631 uint32_t scratch_waves
;
632 uint32_t dispatch_initiator
;
634 uint32_t gs_table_depth
;
636 /* MSAA sample locations.
637 * The first index is the sample index.
638 * The second index is the coordinate: X, Y. */
639 float sample_locations_1x
[1][2];
640 float sample_locations_2x
[2][2];
641 float sample_locations_4x
[4][2];
642 float sample_locations_8x
[8][2];
643 float sample_locations_16x
[16][2];
646 uint32_t gfx_init_size_dw
;
647 struct radeon_winsys_bo
*gfx_init
;
649 struct radeon_winsys_bo
*trace_bo
;
650 uint32_t *trace_id_ptr
;
652 /* Whether to keep shader debug info, for tracing or VK_AMD_shader_info */
653 bool keep_shader_info
;
655 struct radv_physical_device
*physical_device
;
657 /* Backup in-memory cache to be used if the app doesn't provide one */
658 struct radv_pipeline_cache
* mem_cache
;
661 * use different counters so MSAA MRTs get consecutive surface indices,
662 * even if MASK is allocated in between.
664 uint32_t image_mrt_offset_counter
;
665 uint32_t fmask_mrt_offset_counter
;
666 struct list_head shader_slabs
;
667 mtx_t shader_slab_mutex
;
669 /* For detecting VM faults reported by dmesg. */
670 uint64_t dmesg_timestamp
;
672 struct radv_device_extension_table enabled_extensions
;
674 /* Whether the driver uses a global BO list. */
675 bool use_global_bo_list
;
677 struct radv_bo_list bo_list
;
680 struct radv_device_memory
{
681 struct radeon_winsys_bo
*bo
;
682 /* for dedicated allocations */
683 struct radv_image
*image
;
684 struct radv_buffer
*buffer
;
686 VkDeviceSize map_size
;
692 struct radv_descriptor_range
{
697 struct radv_descriptor_set
{
698 const struct radv_descriptor_set_layout
*layout
;
701 struct radeon_winsys_bo
*bo
;
703 uint32_t *mapped_ptr
;
704 struct radv_descriptor_range
*dynamic_descriptors
;
706 struct radeon_winsys_bo
*descriptors
[0];
709 struct radv_push_descriptor_set
711 struct radv_descriptor_set set
;
715 struct radv_descriptor_pool_entry
{
718 struct radv_descriptor_set
*set
;
721 struct radv_descriptor_pool
{
722 struct radeon_winsys_bo
*bo
;
724 uint64_t current_offset
;
727 uint8_t *host_memory_base
;
728 uint8_t *host_memory_ptr
;
729 uint8_t *host_memory_end
;
731 uint32_t entry_count
;
732 uint32_t max_entry_count
;
733 struct radv_descriptor_pool_entry entries
[0];
736 struct radv_descriptor_update_template_entry
{
737 VkDescriptorType descriptor_type
;
739 /* The number of descriptors to update */
740 uint32_t descriptor_count
;
742 /* Into mapped_ptr or dynamic_descriptors, in units of the respective array */
745 /* In dwords. Not valid/used for dynamic descriptors */
748 uint32_t buffer_offset
;
750 /* Only valid for combined image samplers and samplers */
751 uint16_t has_sampler
;
757 /* For push descriptors */
758 const uint32_t *immutable_samplers
;
761 struct radv_descriptor_update_template
{
762 uint32_t entry_count
;
763 VkPipelineBindPoint bind_point
;
764 struct radv_descriptor_update_template_entry entry
[0];
770 VkBufferUsageFlags usage
;
771 VkBufferCreateFlags flags
;
774 struct radeon_winsys_bo
* bo
;
780 enum radv_dynamic_state_bits
{
781 RADV_DYNAMIC_VIEWPORT
= 1 << 0,
782 RADV_DYNAMIC_SCISSOR
= 1 << 1,
783 RADV_DYNAMIC_LINE_WIDTH
= 1 << 2,
784 RADV_DYNAMIC_DEPTH_BIAS
= 1 << 3,
785 RADV_DYNAMIC_BLEND_CONSTANTS
= 1 << 4,
786 RADV_DYNAMIC_DEPTH_BOUNDS
= 1 << 5,
787 RADV_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6,
788 RADV_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7,
789 RADV_DYNAMIC_STENCIL_REFERENCE
= 1 << 8,
790 RADV_DYNAMIC_DISCARD_RECTANGLE
= 1 << 9,
791 RADV_DYNAMIC_ALL
= (1 << 10) - 1,
794 enum radv_cmd_dirty_bits
{
795 /* Keep the dynamic state dirty bits in sync with
796 * enum radv_dynamic_state_bits */
797 RADV_CMD_DIRTY_DYNAMIC_VIEWPORT
= 1 << 0,
798 RADV_CMD_DIRTY_DYNAMIC_SCISSOR
= 1 << 1,
799 RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH
= 1 << 2,
800 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS
= 1 << 3,
801 RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS
= 1 << 4,
802 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS
= 1 << 5,
803 RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6,
804 RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7,
805 RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE
= 1 << 8,
806 RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE
= 1 << 9,
807 RADV_CMD_DIRTY_DYNAMIC_ALL
= (1 << 10) - 1,
808 RADV_CMD_DIRTY_PIPELINE
= 1 << 10,
809 RADV_CMD_DIRTY_INDEX_BUFFER
= 1 << 11,
810 RADV_CMD_DIRTY_FRAMEBUFFER
= 1 << 12,
811 RADV_CMD_DIRTY_VERTEX_BUFFER
= 1 << 13,
814 enum radv_cmd_flush_bits
{
815 RADV_CMD_FLAG_INV_ICACHE
= 1 << 0,
816 /* SMEM L1, other names: KCACHE, constant cache, DCACHE, data cache */
817 RADV_CMD_FLAG_INV_SMEM_L1
= 1 << 1,
818 /* VMEM L1 can optionally be bypassed (GLC=1). Other names: TC L1 */
819 RADV_CMD_FLAG_INV_VMEM_L1
= 1 << 2,
820 /* Used by everything except CB/DB, can be bypassed (SLC=1). Other names: TC L2 */
821 RADV_CMD_FLAG_INV_GLOBAL_L2
= 1 << 3,
822 /* Same as above, but only writes back and doesn't invalidate */
823 RADV_CMD_FLAG_WRITEBACK_GLOBAL_L2
= 1 << 4,
824 /* Framebuffer caches */
825 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
= 1 << 5,
826 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
= 1 << 6,
827 RADV_CMD_FLAG_FLUSH_AND_INV_DB
= 1 << 7,
828 RADV_CMD_FLAG_FLUSH_AND_INV_CB
= 1 << 8,
829 /* Engine synchronization. */
830 RADV_CMD_FLAG_VS_PARTIAL_FLUSH
= 1 << 9,
831 RADV_CMD_FLAG_PS_PARTIAL_FLUSH
= 1 << 10,
832 RADV_CMD_FLAG_CS_PARTIAL_FLUSH
= 1 << 11,
833 RADV_CMD_FLAG_VGT_FLUSH
= 1 << 12,
835 RADV_CMD_FLUSH_AND_INV_FRAMEBUFFER
= (RADV_CMD_FLAG_FLUSH_AND_INV_CB
|
836 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
|
837 RADV_CMD_FLAG_FLUSH_AND_INV_DB
|
838 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
)
841 struct radv_vertex_binding
{
842 struct radv_buffer
* buffer
;
846 struct radv_viewport_state
{
848 VkViewport viewports
[MAX_VIEWPORTS
];
851 struct radv_scissor_state
{
853 VkRect2D scissors
[MAX_SCISSORS
];
856 struct radv_discard_rectangle_state
{
858 VkRect2D rectangles
[MAX_DISCARD_RECTANGLES
];
861 struct radv_dynamic_state
{
863 * Bitmask of (1 << VK_DYNAMIC_STATE_*).
864 * Defines the set of saved dynamic state.
868 struct radv_viewport_state viewport
;
870 struct radv_scissor_state scissor
;
880 float blend_constants
[4];
890 } stencil_compare_mask
;
895 } stencil_write_mask
;
902 struct radv_discard_rectangle_state discard_rectangle
;
905 extern const struct radv_dynamic_state default_dynamic_state
;
908 radv_get_debug_option_name(int id
);
911 radv_get_perftest_option_name(int id
);
914 * Attachment state when recording a renderpass instance.
916 * The clear value is valid only if there exists a pending clear.
918 struct radv_attachment_state
{
919 VkImageAspectFlags pending_clear_aspects
;
920 uint32_t cleared_views
;
921 VkClearValue clear_value
;
922 VkImageLayout current_layout
;
925 struct radv_descriptor_state
{
926 struct radv_descriptor_set
*sets
[MAX_SETS
];
929 struct radv_push_descriptor_set push_set
;
933 struct radv_cmd_state
{
934 /* Vertex descriptors */
941 uint32_t prefetch_L2_mask
;
943 struct radv_pipeline
* pipeline
;
944 struct radv_pipeline
* emitted_pipeline
;
945 struct radv_pipeline
* compute_pipeline
;
946 struct radv_pipeline
* emitted_compute_pipeline
;
947 struct radv_framebuffer
* framebuffer
;
948 struct radv_render_pass
* pass
;
949 const struct radv_subpass
* subpass
;
950 struct radv_dynamic_state dynamic
;
951 struct radv_attachment_state
* attachments
;
952 VkRect2D render_area
;
955 struct radv_buffer
*index_buffer
;
956 uint64_t index_offset
;
958 uint32_t max_index_count
;
960 int32_t last_index_type
;
962 int32_t last_primitive_reset_en
;
963 uint32_t last_primitive_reset_index
;
964 enum radv_cmd_flush_bits flush_bits
;
965 unsigned active_occlusion_queries
;
966 bool perfect_occlusion_queries_enabled
;
969 uint32_t last_ia_multi_vgt_param
;
971 uint32_t last_num_instances
;
972 uint32_t last_first_instance
;
973 uint32_t last_vertex_offset
;
976 struct radv_cmd_pool
{
977 VkAllocationCallbacks alloc
;
978 struct list_head cmd_buffers
;
979 struct list_head free_cmd_buffers
;
980 uint32_t queue_family_index
;
983 struct radv_cmd_buffer_upload
{
987 struct radeon_winsys_bo
*upload_bo
;
988 struct list_head list
;
991 enum radv_cmd_buffer_status
{
992 RADV_CMD_BUFFER_STATUS_INVALID
,
993 RADV_CMD_BUFFER_STATUS_INITIAL
,
994 RADV_CMD_BUFFER_STATUS_RECORDING
,
995 RADV_CMD_BUFFER_STATUS_EXECUTABLE
,
996 RADV_CMD_BUFFER_STATUS_PENDING
,
999 struct radv_cmd_buffer
{
1000 VK_LOADER_DATA _loader_data
;
1002 struct radv_device
* device
;
1004 struct radv_cmd_pool
* pool
;
1005 struct list_head pool_link
;
1007 VkCommandBufferUsageFlags usage_flags
;
1008 VkCommandBufferLevel level
;
1009 enum radv_cmd_buffer_status status
;
1010 struct radeon_winsys_cs
*cs
;
1011 struct radv_cmd_state state
;
1012 struct radv_vertex_binding vertex_bindings
[MAX_VBS
];
1013 uint32_t queue_family_index
;
1015 uint8_t push_constants
[MAX_PUSH_CONSTANTS_SIZE
];
1016 uint32_t dynamic_buffers
[4 * MAX_DYNAMIC_BUFFERS
];
1017 VkShaderStageFlags push_constant_stages
;
1018 struct radv_descriptor_set meta_push_descriptors
;
1020 struct radv_descriptor_state descriptors
[VK_PIPELINE_BIND_POINT_RANGE_SIZE
];
1022 struct radv_cmd_buffer_upload upload
;
1024 uint32_t scratch_size_needed
;
1025 uint32_t compute_scratch_size_needed
;
1026 uint32_t esgs_ring_size_needed
;
1027 uint32_t gsvs_ring_size_needed
;
1028 bool tess_rings_needed
;
1029 bool sample_positions_needed
;
1031 VkResult record_result
;
1033 int ring_offsets_idx
; /* just used for verification */
1034 uint32_t gfx9_fence_offset
;
1035 struct radeon_winsys_bo
*gfx9_fence_bo
;
1036 uint32_t gfx9_fence_idx
;
1039 * Whether a query pool has been resetted and we have to flush caches.
1041 bool pending_reset_query
;
1046 bool radv_cmd_buffer_uses_mec(struct radv_cmd_buffer
*cmd_buffer
);
1048 void si_init_compute(struct radv_cmd_buffer
*cmd_buffer
);
1049 void si_init_config(struct radv_cmd_buffer
*cmd_buffer
);
1051 void cik_create_gfx_config(struct radv_device
*device
);
1053 void si_write_viewport(struct radeon_winsys_cs
*cs
, int first_vp
,
1054 int count
, const VkViewport
*viewports
);
1055 void si_write_scissors(struct radeon_winsys_cs
*cs
, int first
,
1056 int count
, const VkRect2D
*scissors
,
1057 const VkViewport
*viewports
, bool can_use_guardband
);
1058 uint32_t si_get_ia_multi_vgt_param(struct radv_cmd_buffer
*cmd_buffer
,
1059 bool instanced_draw
, bool indirect_draw
,
1060 uint32_t draw_vertex_count
);
1061 void si_cs_emit_write_event_eop(struct radeon_winsys_cs
*cs
,
1063 enum chip_class chip_class
,
1065 unsigned event
, unsigned event_flags
,
1069 uint32_t new_fence
);
1071 void si_emit_wait_fence(struct radeon_winsys_cs
*cs
,
1073 uint64_t va
, uint32_t ref
,
1075 void si_cs_emit_cache_flush(struct radeon_winsys_cs
*cs
,
1076 enum chip_class chip_class
,
1077 uint32_t *fence_ptr
, uint64_t va
,
1079 enum radv_cmd_flush_bits flush_bits
);
1080 void si_emit_cache_flush(struct radv_cmd_buffer
*cmd_buffer
);
1081 void si_emit_set_predication_state(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
);
1082 void si_cp_dma_buffer_copy(struct radv_cmd_buffer
*cmd_buffer
,
1083 uint64_t src_va
, uint64_t dest_va
,
1085 void si_cp_dma_prefetch(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
1087 void si_cp_dma_clear_buffer(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
1088 uint64_t size
, unsigned value
);
1089 void radv_set_db_count_control(struct radv_cmd_buffer
*cmd_buffer
);
1091 radv_cmd_buffer_upload_alloc(struct radv_cmd_buffer
*cmd_buffer
,
1094 unsigned *out_offset
,
1097 radv_cmd_buffer_set_subpass(struct radv_cmd_buffer
*cmd_buffer
,
1098 const struct radv_subpass
*subpass
,
1101 radv_cmd_buffer_upload_data(struct radv_cmd_buffer
*cmd_buffer
,
1102 unsigned size
, unsigned alignmnet
,
1103 const void *data
, unsigned *out_offset
);
1105 void radv_cmd_buffer_clear_subpass(struct radv_cmd_buffer
*cmd_buffer
);
1106 void radv_cmd_buffer_resolve_subpass(struct radv_cmd_buffer
*cmd_buffer
);
1107 void radv_cmd_buffer_resolve_subpass_cs(struct radv_cmd_buffer
*cmd_buffer
);
1108 void radv_cmd_buffer_resolve_subpass_fs(struct radv_cmd_buffer
*cmd_buffer
);
1109 void radv_cayman_emit_msaa_sample_locs(struct radeon_winsys_cs
*cs
, int nr_samples
);
1110 unsigned radv_cayman_get_maxdist(int log_samples
);
1111 void radv_device_init_msaa(struct radv_device
*device
);
1112 void radv_set_depth_clear_regs(struct radv_cmd_buffer
*cmd_buffer
,
1113 struct radv_image
*image
,
1114 VkClearDepthStencilValue ds_clear_value
,
1115 VkImageAspectFlags aspects
);
1116 void radv_set_color_clear_regs(struct radv_cmd_buffer
*cmd_buffer
,
1117 struct radv_image
*image
,
1119 uint32_t color_values
[2]);
1120 void radv_set_dcc_need_cmask_elim_pred(struct radv_cmd_buffer
*cmd_buffer
,
1121 struct radv_image
*image
,
1123 uint32_t radv_fill_buffer(struct radv_cmd_buffer
*cmd_buffer
,
1124 struct radeon_winsys_bo
*bo
,
1125 uint64_t offset
, uint64_t size
, uint32_t value
);
1126 void radv_cmd_buffer_trace_emit(struct radv_cmd_buffer
*cmd_buffer
);
1127 bool radv_get_memory_fd(struct radv_device
*device
,
1128 struct radv_device_memory
*memory
,
1132 radv_emit_shader_pointer_head(struct radeon_winsys_cs
*cs
,
1133 unsigned sh_offset
, unsigned pointer_count
,
1134 bool use_32bit_pointers
)
1136 radeon_emit(cs
, PKT3(PKT3_SET_SH_REG
, pointer_count
* (use_32bit_pointers
? 1 : 2), 0));
1137 radeon_emit(cs
, (sh_offset
- SI_SH_REG_OFFSET
) >> 2);
1141 radv_emit_shader_pointer_body(struct radv_device
*device
,
1142 struct radeon_winsys_cs
*cs
,
1143 uint64_t va
, bool use_32bit_pointers
)
1145 radeon_emit(cs
, va
);
1147 if (use_32bit_pointers
) {
1149 (va
>> 32) == device
->physical_device
->rad_info
.address32_hi
);
1151 radeon_emit(cs
, va
>> 32);
1156 radv_emit_shader_pointer(struct radv_device
*device
,
1157 struct radeon_winsys_cs
*cs
,
1158 uint32_t sh_offset
, uint64_t va
, bool global
)
1160 bool use_32bit_pointers
= HAVE_32BIT_POINTERS
&& !global
;
1162 radv_emit_shader_pointer_head(cs
, sh_offset
, 1, use_32bit_pointers
);
1163 radv_emit_shader_pointer_body(device
, cs
, va
, use_32bit_pointers
);
1166 static inline struct radv_descriptor_state
*
1167 radv_get_descriptors_state(struct radv_cmd_buffer
*cmd_buffer
,
1168 VkPipelineBindPoint bind_point
)
1170 assert(bind_point
== VK_PIPELINE_BIND_POINT_GRAPHICS
||
1171 bind_point
== VK_PIPELINE_BIND_POINT_COMPUTE
);
1172 return &cmd_buffer
->descriptors
[bind_point
];
1176 * Takes x,y,z as exact numbers of invocations, instead of blocks.
1178 * Limitations: Can't call normal dispatch functions without binding or rebinding
1179 * the compute pipeline.
1181 void radv_unaligned_dispatch(
1182 struct radv_cmd_buffer
*cmd_buffer
,
1188 struct radeon_winsys_bo
*bo
;
1192 struct radv_shader_module
;
1194 #define RADV_HASH_SHADER_IS_GEOM_COPY_SHADER (1 << 0)
1195 #define RADV_HASH_SHADER_SISCHED (1 << 1)
1196 #define RADV_HASH_SHADER_UNSAFE_MATH (1 << 2)
1198 radv_hash_shaders(unsigned char *hash
,
1199 const VkPipelineShaderStageCreateInfo
**stages
,
1200 const struct radv_pipeline_layout
*layout
,
1201 const struct radv_pipeline_key
*key
,
1204 static inline gl_shader_stage
1205 vk_to_mesa_shader_stage(VkShaderStageFlagBits vk_stage
)
1207 assert(__builtin_popcount(vk_stage
) == 1);
1208 return ffs(vk_stage
) - 1;
1211 static inline VkShaderStageFlagBits
1212 mesa_to_vk_shader_stage(gl_shader_stage mesa_stage
)
1214 return (1 << mesa_stage
);
1217 #define RADV_STAGE_MASK ((1 << MESA_SHADER_STAGES) - 1)
1219 #define radv_foreach_stage(stage, stage_bits) \
1220 for (gl_shader_stage stage, \
1221 __tmp = (gl_shader_stage)((stage_bits) & RADV_STAGE_MASK); \
1222 stage = __builtin_ffs(__tmp) - 1, __tmp; \
1223 __tmp &= ~(1 << (stage)))
1225 unsigned radv_format_meta_fs_key(VkFormat format
);
1227 struct radv_multisample_state
{
1229 uint32_t pa_sc_line_cntl
;
1230 uint32_t pa_sc_mode_cntl_0
;
1231 uint32_t pa_sc_mode_cntl_1
;
1232 uint32_t pa_sc_aa_config
;
1233 uint32_t pa_sc_aa_mask
[2];
1234 unsigned num_samples
;
1237 struct radv_prim_vertex_count
{
1242 struct radv_vertex_elements_info
{
1243 uint32_t rsrc_word3
[MAX_VERTEX_ATTRIBS
];
1244 uint32_t format_size
[MAX_VERTEX_ATTRIBS
];
1245 uint32_t binding
[MAX_VERTEX_ATTRIBS
];
1246 uint32_t offset
[MAX_VERTEX_ATTRIBS
];
1250 struct radv_ia_multi_vgt_param_helpers
{
1252 bool partial_es_wave
;
1253 uint8_t primgroup_size
;
1254 bool wd_switch_on_eop
;
1255 bool ia_switch_on_eoi
;
1256 bool partial_vs_wave
;
1259 #define SI_GS_PER_ES 128
1261 struct radv_pipeline
{
1262 struct radv_device
* device
;
1263 struct radv_dynamic_state dynamic_state
;
1265 struct radv_pipeline_layout
* layout
;
1267 bool need_indirect_descriptor_sets
;
1268 struct radv_shader_variant
* shaders
[MESA_SHADER_STAGES
];
1269 struct radv_shader_variant
*gs_copy_shader
;
1270 VkShaderStageFlags active_stages
;
1272 struct radeon_winsys_cs cs
;
1274 struct radv_vertex_elements_info vertex_elements
;
1276 uint32_t binding_stride
[MAX_VBS
];
1278 uint32_t user_data_0
[MESA_SHADER_STAGES
];
1281 struct radv_multisample_state ms
;
1282 uint32_t spi_baryc_cntl
;
1283 bool prim_restart_enable
;
1284 unsigned esgs_ring_size
;
1285 unsigned gsvs_ring_size
;
1286 uint32_t vtx_base_sgpr
;
1287 struct radv_ia_multi_vgt_param_helpers ia_multi_vgt_param
;
1288 uint8_t vtx_emit_num
;
1289 struct radv_prim_vertex_count prim_vertex_count
;
1290 bool can_use_guardband
;
1291 uint32_t needed_dynamic_state
;
1292 bool disable_out_of_order_rast_for_occlusion
;
1294 /* Used for rbplus */
1295 uint32_t col_format
;
1296 uint32_t cb_target_mask
;
1301 unsigned scratch_bytes_per_wave
;
1304 static inline bool radv_pipeline_has_gs(const struct radv_pipeline
*pipeline
)
1306 return pipeline
->shaders
[MESA_SHADER_GEOMETRY
] ? true : false;
1309 static inline bool radv_pipeline_has_tess(const struct radv_pipeline
*pipeline
)
1311 return pipeline
->shaders
[MESA_SHADER_TESS_CTRL
] ? true : false;
1314 struct radv_userdata_info
*radv_lookup_user_sgpr(struct radv_pipeline
*pipeline
,
1315 gl_shader_stage stage
,
1318 struct radv_shader_variant
*radv_get_shader(struct radv_pipeline
*pipeline
,
1319 gl_shader_stage stage
);
1321 struct radv_graphics_pipeline_create_info
{
1323 bool db_depth_clear
;
1324 bool db_stencil_clear
;
1325 bool db_depth_disable_expclear
;
1326 bool db_stencil_disable_expclear
;
1327 bool db_flush_depth_inplace
;
1328 bool db_flush_stencil_inplace
;
1329 bool db_resummarize
;
1330 uint32_t custom_blend_mode
;
1334 radv_graphics_pipeline_create(VkDevice device
,
1335 VkPipelineCache cache
,
1336 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
1337 const struct radv_graphics_pipeline_create_info
*extra
,
1338 const VkAllocationCallbacks
*alloc
,
1339 VkPipeline
*pPipeline
);
1341 struct vk_format_description
;
1342 uint32_t radv_translate_buffer_dataformat(const struct vk_format_description
*desc
,
1343 int first_non_void
);
1344 uint32_t radv_translate_buffer_numformat(const struct vk_format_description
*desc
,
1345 int first_non_void
);
1346 uint32_t radv_translate_colorformat(VkFormat format
);
1347 uint32_t radv_translate_color_numformat(VkFormat format
,
1348 const struct vk_format_description
*desc
,
1349 int first_non_void
);
1350 uint32_t radv_colorformat_endian_swap(uint32_t colorformat
);
1351 unsigned radv_translate_colorswap(VkFormat format
, bool do_endian_swap
);
1352 uint32_t radv_translate_dbformat(VkFormat format
);
1353 uint32_t radv_translate_tex_dataformat(VkFormat format
,
1354 const struct vk_format_description
*desc
,
1355 int first_non_void
);
1356 uint32_t radv_translate_tex_numformat(VkFormat format
,
1357 const struct vk_format_description
*desc
,
1358 int first_non_void
);
1359 bool radv_format_pack_clear_color(VkFormat format
,
1360 uint32_t clear_vals
[2],
1361 VkClearColorValue
*value
);
1362 bool radv_is_colorbuffer_format_supported(VkFormat format
, bool *blendable
);
1363 bool radv_dcc_formats_compatible(VkFormat format1
,
1366 struct radv_fmask_info
{
1370 unsigned pitch_in_pixels
;
1371 unsigned bank_height
;
1372 unsigned slice_tile_max
;
1373 unsigned tile_mode_index
;
1374 unsigned tile_swizzle
;
1377 struct radv_cmask_info
{
1381 unsigned slice_tile_max
;
1386 /* The original VkFormat provided by the client. This may not match any
1387 * of the actual surface formats.
1390 VkImageAspectFlags aspects
;
1391 VkImageUsageFlags usage
; /**< Superset of VkImageCreateInfo::usage. */
1392 struct ac_surf_info info
;
1393 VkImageTiling tiling
; /** VkImageCreateInfo::tiling */
1394 VkImageCreateFlags flags
; /** VkImageCreateInfo::flags */
1399 unsigned queue_family_mask
;
1403 /* Set when bound */
1404 struct radeon_winsys_bo
*bo
;
1405 VkDeviceSize offset
;
1406 uint64_t dcc_offset
;
1407 uint64_t htile_offset
;
1408 bool tc_compatible_htile
;
1409 struct radeon_surf surface
;
1411 struct radv_fmask_info fmask
;
1412 struct radv_cmask_info cmask
;
1413 uint64_t clear_value_offset
;
1414 uint64_t dcc_pred_offset
;
1416 /* For VK_ANDROID_native_buffer, the WSI image owns the memory, */
1417 VkDeviceMemory owned_memory
;
1420 /* Whether the image has a htile that is known consistent with the contents of
1422 bool radv_layout_has_htile(const struct radv_image
*image
,
1423 VkImageLayout layout
,
1424 unsigned queue_mask
);
1426 /* Whether the image has a htile that is known consistent with the contents of
1427 * the image and is allowed to be in compressed form.
1429 * If this is false reads that don't use the htile should be able to return
1432 bool radv_layout_is_htile_compressed(const struct radv_image
*image
,
1433 VkImageLayout layout
,
1434 unsigned queue_mask
);
1436 bool radv_layout_can_fast_clear(const struct radv_image
*image
,
1437 VkImageLayout layout
,
1438 unsigned queue_mask
);
1440 bool radv_layout_dcc_compressed(const struct radv_image
*image
,
1441 VkImageLayout layout
,
1442 unsigned queue_mask
);
1445 * Return whether the image has CMASK metadata for color surfaces.
1448 radv_image_has_cmask(const struct radv_image
*image
)
1450 return image
->cmask
.size
;
1454 * Return whether the image has FMASK metadata for color surfaces.
1457 radv_image_has_fmask(const struct radv_image
*image
)
1459 return image
->fmask
.size
;
1463 * Return whether the image has DCC metadata for color surfaces.
1466 radv_image_has_dcc(const struct radv_image
*image
)
1468 return image
->surface
.dcc_size
;
1472 * Return whether DCC metadata is enabled for a level.
1475 radv_dcc_enabled(const struct radv_image
*image
, unsigned level
)
1477 return radv_image_has_dcc(image
) &&
1478 level
< image
->surface
.num_dcc_levels
;
1482 * Return whether the image has HTILE metadata for depth surfaces.
1485 radv_image_has_htile(const struct radv_image
*image
)
1487 return image
->surface
.htile_size
;
1491 * Return whether HTILE metadata is enabled for a level.
1494 radv_htile_enabled(const struct radv_image
*image
, unsigned level
)
1496 return radv_image_has_htile(image
) && level
== 0;
1500 * Return whether the image is TC-compatible HTILE.
1503 radv_image_is_tc_compat_htile(const struct radv_image
*image
)
1505 return radv_image_has_htile(image
) && image
->tc_compatible_htile
;
1508 unsigned radv_image_queue_family_mask(const struct radv_image
*image
, uint32_t family
, uint32_t queue_family
);
1510 static inline uint32_t
1511 radv_get_layerCount(const struct radv_image
*image
,
1512 const VkImageSubresourceRange
*range
)
1514 return range
->layerCount
== VK_REMAINING_ARRAY_LAYERS
?
1515 image
->info
.array_size
- range
->baseArrayLayer
: range
->layerCount
;
1518 static inline uint32_t
1519 radv_get_levelCount(const struct radv_image
*image
,
1520 const VkImageSubresourceRange
*range
)
1522 return range
->levelCount
== VK_REMAINING_MIP_LEVELS
?
1523 image
->info
.levels
- range
->baseMipLevel
: range
->levelCount
;
1526 struct radeon_bo_metadata
;
1528 radv_init_metadata(struct radv_device
*device
,
1529 struct radv_image
*image
,
1530 struct radeon_bo_metadata
*metadata
);
1532 struct radv_image_view
{
1533 struct radv_image
*image
; /**< VkImageViewCreateInfo::image */
1534 struct radeon_winsys_bo
*bo
;
1536 VkImageViewType type
;
1537 VkImageAspectFlags aspect_mask
;
1539 uint32_t base_layer
;
1540 uint32_t layer_count
;
1542 uint32_t level_count
;
1543 VkExtent3D extent
; /**< Extent of VkImageViewCreateInfo::baseMipLevel. */
1545 uint32_t descriptor
[16];
1547 /* Descriptor for use as a storage image as opposed to a sampled image.
1548 * This has a few differences for cube maps (e.g. type).
1550 uint32_t storage_descriptor
[16];
1553 struct radv_image_create_info
{
1554 const VkImageCreateInfo
*vk_info
;
1556 bool no_metadata_planes
;
1559 VkResult
radv_image_create(VkDevice _device
,
1560 const struct radv_image_create_info
*info
,
1561 const VkAllocationCallbacks
* alloc
,
1565 radv_image_from_gralloc(VkDevice device_h
,
1566 const VkImageCreateInfo
*base_info
,
1567 const VkNativeBufferANDROID
*gralloc_info
,
1568 const VkAllocationCallbacks
*alloc
,
1569 VkImage
*out_image_h
);
1571 void radv_image_view_init(struct radv_image_view
*view
,
1572 struct radv_device
*device
,
1573 const VkImageViewCreateInfo
* pCreateInfo
);
1575 struct radv_buffer_view
{
1576 struct radeon_winsys_bo
*bo
;
1578 uint64_t range
; /**< VkBufferViewCreateInfo::range */
1581 void radv_buffer_view_init(struct radv_buffer_view
*view
,
1582 struct radv_device
*device
,
1583 const VkBufferViewCreateInfo
* pCreateInfo
);
1585 static inline struct VkExtent3D
1586 radv_sanitize_image_extent(const VkImageType imageType
,
1587 const struct VkExtent3D imageExtent
)
1589 switch (imageType
) {
1590 case VK_IMAGE_TYPE_1D
:
1591 return (VkExtent3D
) { imageExtent
.width
, 1, 1 };
1592 case VK_IMAGE_TYPE_2D
:
1593 return (VkExtent3D
) { imageExtent
.width
, imageExtent
.height
, 1 };
1594 case VK_IMAGE_TYPE_3D
:
1597 unreachable("invalid image type");
1601 static inline struct VkOffset3D
1602 radv_sanitize_image_offset(const VkImageType imageType
,
1603 const struct VkOffset3D imageOffset
)
1605 switch (imageType
) {
1606 case VK_IMAGE_TYPE_1D
:
1607 return (VkOffset3D
) { imageOffset
.x
, 0, 0 };
1608 case VK_IMAGE_TYPE_2D
:
1609 return (VkOffset3D
) { imageOffset
.x
, imageOffset
.y
, 0 };
1610 case VK_IMAGE_TYPE_3D
:
1613 unreachable("invalid image type");
1618 radv_image_extent_compare(const struct radv_image
*image
,
1619 const VkExtent3D
*extent
)
1621 if (extent
->width
!= image
->info
.width
||
1622 extent
->height
!= image
->info
.height
||
1623 extent
->depth
!= image
->info
.depth
)
1628 struct radv_sampler
{
1632 struct radv_color_buffer_info
{
1633 uint64_t cb_color_base
;
1634 uint64_t cb_color_cmask
;
1635 uint64_t cb_color_fmask
;
1636 uint64_t cb_dcc_base
;
1637 uint32_t cb_color_pitch
;
1638 uint32_t cb_color_slice
;
1639 uint32_t cb_color_view
;
1640 uint32_t cb_color_info
;
1641 uint32_t cb_color_attrib
;
1642 uint32_t cb_color_attrib2
;
1643 uint32_t cb_dcc_control
;
1644 uint32_t cb_color_cmask_slice
;
1645 uint32_t cb_color_fmask_slice
;
1648 struct radv_ds_buffer_info
{
1649 uint64_t db_z_read_base
;
1650 uint64_t db_stencil_read_base
;
1651 uint64_t db_z_write_base
;
1652 uint64_t db_stencil_write_base
;
1653 uint64_t db_htile_data_base
;
1654 uint32_t db_depth_info
;
1656 uint32_t db_stencil_info
;
1657 uint32_t db_depth_view
;
1658 uint32_t db_depth_size
;
1659 uint32_t db_depth_slice
;
1660 uint32_t db_htile_surface
;
1661 uint32_t pa_su_poly_offset_db_fmt_cntl
;
1662 uint32_t db_z_info2
;
1663 uint32_t db_stencil_info2
;
1667 struct radv_attachment_info
{
1669 struct radv_color_buffer_info cb
;
1670 struct radv_ds_buffer_info ds
;
1672 struct radv_image_view
*attachment
;
1675 struct radv_framebuffer
{
1680 uint32_t attachment_count
;
1681 struct radv_attachment_info attachments
[0];
1684 struct radv_subpass_barrier
{
1685 VkPipelineStageFlags src_stage_mask
;
1686 VkAccessFlags src_access_mask
;
1687 VkAccessFlags dst_access_mask
;
1690 struct radv_subpass
{
1691 uint32_t input_count
;
1692 uint32_t color_count
;
1693 VkAttachmentReference
* input_attachments
;
1694 VkAttachmentReference
* color_attachments
;
1695 VkAttachmentReference
* resolve_attachments
;
1696 VkAttachmentReference depth_stencil_attachment
;
1698 /** Subpass has at least one resolve attachment */
1701 struct radv_subpass_barrier start_barrier
;
1704 VkSampleCountFlagBits max_sample_count
;
1707 struct radv_render_pass_attachment
{
1710 VkAttachmentLoadOp load_op
;
1711 VkAttachmentLoadOp stencil_load_op
;
1712 VkImageLayout initial_layout
;
1713 VkImageLayout final_layout
;
1717 struct radv_render_pass
{
1718 uint32_t attachment_count
;
1719 uint32_t subpass_count
;
1720 VkAttachmentReference
* subpass_attachments
;
1721 struct radv_render_pass_attachment
* attachments
;
1722 struct radv_subpass_barrier end_barrier
;
1723 struct radv_subpass subpasses
[0];
1726 VkResult
radv_device_init_meta(struct radv_device
*device
);
1727 void radv_device_finish_meta(struct radv_device
*device
);
1729 struct radv_query_pool
{
1730 struct radeon_winsys_bo
*bo
;
1732 uint32_t availability_offset
;
1736 uint32_t pipeline_stats_mask
;
1739 struct radv_semaphore
{
1740 /* use a winsys sem for non-exportable */
1741 struct radeon_winsys_sem
*sem
;
1743 uint32_t temp_syncobj
;
1746 void radv_set_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
1747 VkPipelineBindPoint bind_point
,
1748 struct radv_descriptor_set
*set
,
1752 radv_update_descriptor_sets(struct radv_device
*device
,
1753 struct radv_cmd_buffer
*cmd_buffer
,
1754 VkDescriptorSet overrideSet
,
1755 uint32_t descriptorWriteCount
,
1756 const VkWriteDescriptorSet
*pDescriptorWrites
,
1757 uint32_t descriptorCopyCount
,
1758 const VkCopyDescriptorSet
*pDescriptorCopies
);
1761 radv_update_descriptor_set_with_template(struct radv_device
*device
,
1762 struct radv_cmd_buffer
*cmd_buffer
,
1763 struct radv_descriptor_set
*set
,
1764 VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate
,
1767 void radv_meta_push_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
1768 VkPipelineBindPoint pipelineBindPoint
,
1769 VkPipelineLayout _layout
,
1771 uint32_t descriptorWriteCount
,
1772 const VkWriteDescriptorSet
*pDescriptorWrites
);
1774 void radv_initialize_dcc(struct radv_cmd_buffer
*cmd_buffer
,
1775 struct radv_image
*image
, uint32_t value
);
1778 struct radeon_winsys_fence
*fence
;
1783 uint32_t temp_syncobj
;
1786 /* radv_nir_to_llvm.c */
1787 struct radv_shader_variant_info
;
1788 struct radv_nir_compiler_options
;
1790 void radv_compile_gs_copy_shader(LLVMTargetMachineRef tm
,
1791 struct nir_shader
*geom_shader
,
1792 struct ac_shader_binary
*binary
,
1793 struct ac_shader_config
*config
,
1794 struct radv_shader_variant_info
*shader_info
,
1795 const struct radv_nir_compiler_options
*option
);
1797 void radv_compile_nir_shader(LLVMTargetMachineRef tm
,
1798 struct ac_shader_binary
*binary
,
1799 struct ac_shader_config
*config
,
1800 struct radv_shader_variant_info
*shader_info
,
1801 struct nir_shader
*const *nir
,
1803 const struct radv_nir_compiler_options
*options
);
1805 /* radv_shader_info.h */
1806 struct radv_shader_info
;
1808 void radv_nir_shader_info_pass(const struct nir_shader
*nir
,
1809 const struct radv_nir_compiler_options
*options
,
1810 struct radv_shader_info
*info
);
1812 struct radeon_winsys_sem
;
1814 #define RADV_DEFINE_HANDLE_CASTS(__radv_type, __VkType) \
1816 static inline struct __radv_type * \
1817 __radv_type ## _from_handle(__VkType _handle) \
1819 return (struct __radv_type *) _handle; \
1822 static inline __VkType \
1823 __radv_type ## _to_handle(struct __radv_type *_obj) \
1825 return (__VkType) _obj; \
1828 #define RADV_DEFINE_NONDISP_HANDLE_CASTS(__radv_type, __VkType) \
1830 static inline struct __radv_type * \
1831 __radv_type ## _from_handle(__VkType _handle) \
1833 return (struct __radv_type *)(uintptr_t) _handle; \
1836 static inline __VkType \
1837 __radv_type ## _to_handle(struct __radv_type *_obj) \
1839 return (__VkType)(uintptr_t) _obj; \
1842 #define RADV_FROM_HANDLE(__radv_type, __name, __handle) \
1843 struct __radv_type *__name = __radv_type ## _from_handle(__handle)
1845 RADV_DEFINE_HANDLE_CASTS(radv_cmd_buffer
, VkCommandBuffer
)
1846 RADV_DEFINE_HANDLE_CASTS(radv_device
, VkDevice
)
1847 RADV_DEFINE_HANDLE_CASTS(radv_instance
, VkInstance
)
1848 RADV_DEFINE_HANDLE_CASTS(radv_physical_device
, VkPhysicalDevice
)
1849 RADV_DEFINE_HANDLE_CASTS(radv_queue
, VkQueue
)
1851 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_cmd_pool
, VkCommandPool
)
1852 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer
, VkBuffer
)
1853 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer_view
, VkBufferView
)
1854 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_pool
, VkDescriptorPool
)
1855 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set
, VkDescriptorSet
)
1856 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set_layout
, VkDescriptorSetLayout
)
1857 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_update_template
, VkDescriptorUpdateTemplateKHR
)
1858 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_device_memory
, VkDeviceMemory
)
1859 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_fence
, VkFence
)
1860 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_event
, VkEvent
)
1861 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_framebuffer
, VkFramebuffer
)
1862 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image
, VkImage
)
1863 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image_view
, VkImageView
);
1864 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_cache
, VkPipelineCache
)
1865 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline
, VkPipeline
)
1866 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_layout
, VkPipelineLayout
)
1867 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_query_pool
, VkQueryPool
)
1868 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_render_pass
, VkRenderPass
)
1869 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_sampler
, VkSampler
)
1870 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_shader_module
, VkShaderModule
)
1871 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_semaphore
, VkSemaphore
)
1873 #endif /* RADV_PRIVATE_H */