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 "util/xmlconfig.h"
52 #include "main/macros.h"
54 #include "vk_debug_report.h"
56 #include "radv_radeon_winsys.h"
57 #include "ac_binary.h"
58 #include "ac_nir_to_llvm.h"
59 #include "ac_gpu_info.h"
60 #include "ac_surface.h"
61 #include "ac_llvm_build.h"
62 #include "ac_llvm_util.h"
63 #include "radv_constants.h"
64 #include "radv_descriptor_set.h"
65 #include "radv_extensions.h"
68 #include <llvm-c/TargetMachine.h>
70 /* Pre-declarations needed for WSI entrypoints */
73 typedef struct xcb_connection_t xcb_connection_t
;
74 typedef uint32_t xcb_visualid_t
;
75 typedef uint32_t xcb_window_t
;
77 #include <vulkan/vulkan.h>
78 #include <vulkan/vulkan_intel.h>
79 #include <vulkan/vk_icd.h>
80 #include <vulkan/vk_android_native_buffer.h>
82 #include "radv_entrypoints.h"
84 #include "wsi_common.h"
85 #include "wsi_common_display.h"
88 unsigned img_format
:9;
90 /* Various formats are only supported with workarounds for vertex fetch,
91 * and some 32_32_32 formats are supported natively, but only for buffers
92 * (possibly with some image support, actually, but no filtering). */
96 #include "gfx10_format_table.h"
100 RADV_MEM_HEAP_VRAM_CPU_ACCESS
,
107 RADV_MEM_TYPE_GTT_WRITE_COMBINE
,
108 RADV_MEM_TYPE_VRAM_CPU_ACCESS
,
109 RADV_MEM_TYPE_GTT_CACHED
,
113 #define radv_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
115 static inline uint32_t
116 align_u32(uint32_t v
, uint32_t a
)
118 assert(a
!= 0 && a
== (a
& -a
));
119 return (v
+ a
- 1) & ~(a
- 1);
122 static inline uint32_t
123 align_u32_npot(uint32_t v
, uint32_t a
)
125 return (v
+ a
- 1) / a
* a
;
128 static inline uint64_t
129 align_u64(uint64_t v
, uint64_t a
)
131 assert(a
!= 0 && a
== (a
& -a
));
132 return (v
+ a
- 1) & ~(a
- 1);
135 static inline int32_t
136 align_i32(int32_t v
, int32_t a
)
138 assert(a
!= 0 && a
== (a
& -a
));
139 return (v
+ a
- 1) & ~(a
- 1);
142 /** Alignment must be a power of 2. */
144 radv_is_aligned(uintmax_t n
, uintmax_t a
)
146 assert(a
== (a
& -a
));
147 return (n
& (a
- 1)) == 0;
150 static inline uint32_t
151 round_up_u32(uint32_t v
, uint32_t a
)
153 return (v
+ a
- 1) / a
;
156 static inline uint64_t
157 round_up_u64(uint64_t v
, uint64_t a
)
159 return (v
+ a
- 1) / a
;
162 static inline uint32_t
163 radv_minify(uint32_t n
, uint32_t levels
)
165 if (unlikely(n
== 0))
168 return MAX2(n
>> levels
, 1);
171 radv_clamp_f(float f
, float min
, float max
)
184 radv_clear_mask(uint32_t *inout_mask
, uint32_t clear_mask
)
186 if (*inout_mask
& clear_mask
) {
187 *inout_mask
&= ~clear_mask
;
194 #define for_each_bit(b, dword) \
195 for (uint32_t __dword = (dword); \
196 (b) = __builtin_ffs(__dword) - 1, __dword; \
197 __dword &= ~(1 << (b)))
199 #define typed_memcpy(dest, src, count) ({ \
200 STATIC_ASSERT(sizeof(*src) == sizeof(*dest)); \
201 memcpy((dest), (src), (count) * sizeof(*(src))); \
204 /* Whenever we generate an error, pass it through this function. Useful for
205 * debugging, where we can break on it. Only call at error site, not when
206 * propagating errors. Might be useful to plug in a stack trace here.
209 struct radv_image_view
;
210 struct radv_instance
;
212 VkResult
__vk_errorf(struct radv_instance
*instance
, VkResult error
, const char *file
, int line
, const char *format
, ...);
214 #define vk_error(instance, error) __vk_errorf(instance, error, __FILE__, __LINE__, NULL);
215 #define vk_errorf(instance, error, format, ...) __vk_errorf(instance, error, __FILE__, __LINE__, format, ## __VA_ARGS__);
217 void __radv_finishme(const char *file
, int line
, const char *format
, ...)
218 radv_printflike(3, 4);
219 void radv_loge(const char *format
, ...) radv_printflike(1, 2);
220 void radv_loge_v(const char *format
, va_list va
);
221 void radv_logi(const char *format
, ...) radv_printflike(1, 2);
222 void radv_logi_v(const char *format
, va_list va
);
225 * Print a FINISHME message, including its source location.
227 #define radv_finishme(format, ...) \
229 static bool reported = false; \
231 __radv_finishme(__FILE__, __LINE__, format, ##__VA_ARGS__); \
236 /* A non-fatal assert. Useful for debugging. */
238 #define radv_assert(x) ({ \
239 if (unlikely(!(x))) \
240 fprintf(stderr, "%s:%d ASSERT: %s\n", __FILE__, __LINE__, #x); \
243 #define radv_assert(x)
246 #define stub_return(v) \
248 radv_finishme("stub %s", __func__); \
254 radv_finishme("stub %s", __func__); \
258 void *radv_lookup_entrypoint_unchecked(const char *name
);
259 void *radv_lookup_entrypoint_checked(const char *name
,
260 uint32_t core_version
,
261 const struct radv_instance_extension_table
*instance
,
262 const struct radv_device_extension_table
*device
);
263 void *radv_lookup_physical_device_entrypoint_checked(const char *name
,
264 uint32_t core_version
,
265 const struct radv_instance_extension_table
*instance
);
267 struct radv_physical_device
{
268 VK_LOADER_DATA _loader_data
;
270 struct radv_instance
* instance
;
272 struct radeon_winsys
*ws
;
273 struct radeon_info rad_info
;
274 char name
[VK_MAX_PHYSICAL_DEVICE_NAME_SIZE
];
275 uint8_t driver_uuid
[VK_UUID_SIZE
];
276 uint8_t device_uuid
[VK_UUID_SIZE
];
277 uint8_t cache_uuid
[VK_UUID_SIZE
];
281 struct wsi_device wsi_device
;
283 bool has_rbplus
; /* if RB+ register exist */
284 bool rbplus_allowed
; /* if RB+ is allowed */
285 bool has_clear_state
;
286 bool cpdma_prefetch_writes_memory
;
287 bool has_scissor_bug
;
288 bool has_tc_compat_zrange_bug
;
290 bool has_out_of_order_rast
;
291 bool out_of_order_rast_allowed
;
293 /* Whether DCC should be enabled for MSAA textures. */
294 bool dcc_msaa_allowed
;
296 /* Whether LOAD_CONTEXT_REG packets are supported. */
297 bool has_load_ctx_reg_pkt
;
299 /* Whether to enable the AMD_shader_ballot extension */
300 bool use_shader_ballot
;
302 /* Whether DISABLE_CONSTANT_ENCODE_REG is supported. */
303 bool has_dcc_constant_encode
;
305 /* Number of threads per wave. */
306 uint8_t ps_wave_size
;
307 uint8_t cs_wave_size
;
308 uint8_t ge_wave_size
;
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
;
339 struct driOptionCache dri_options
;
340 struct driOptionCache available_dri_options
;
343 VkResult
radv_init_wsi(struct radv_physical_device
*physical_device
);
344 void radv_finish_wsi(struct radv_physical_device
*physical_device
);
346 bool radv_instance_extension_supported(const char *name
);
347 uint32_t radv_physical_device_api_version(struct radv_physical_device
*dev
);
348 bool radv_physical_device_extension_supported(struct radv_physical_device
*dev
,
353 struct radv_pipeline_cache
{
354 struct radv_device
* device
;
355 pthread_mutex_t mutex
;
359 uint32_t kernel_count
;
360 struct cache_entry
** hash_table
;
363 VkAllocationCallbacks alloc
;
366 struct radv_pipeline_key
{
367 uint32_t instance_rate_inputs
;
368 uint32_t instance_rate_divisors
[MAX_VERTEX_ATTRIBS
];
369 uint8_t vertex_attribute_formats
[MAX_VERTEX_ATTRIBS
];
370 uint32_t vertex_attribute_bindings
[MAX_VERTEX_ATTRIBS
];
371 uint32_t vertex_attribute_offsets
[MAX_VERTEX_ATTRIBS
];
372 uint32_t vertex_attribute_strides
[MAX_VERTEX_ATTRIBS
];
373 uint64_t vertex_alpha_adjust
;
374 uint32_t vertex_post_shuffle
;
375 unsigned tess_input_vertices
;
379 uint8_t log2_ps_iter_samples
;
381 uint32_t has_multiview_view_index
: 1;
382 uint32_t optimisations_disabled
: 1;
385 struct radv_shader_binary
;
386 struct radv_shader_variant
;
389 radv_pipeline_cache_init(struct radv_pipeline_cache
*cache
,
390 struct radv_device
*device
);
392 radv_pipeline_cache_finish(struct radv_pipeline_cache
*cache
);
394 radv_pipeline_cache_load(struct radv_pipeline_cache
*cache
,
395 const void *data
, size_t size
);
398 radv_create_shader_variants_from_pipeline_cache(struct radv_device
*device
,
399 struct radv_pipeline_cache
*cache
,
400 const unsigned char *sha1
,
401 struct radv_shader_variant
**variants
,
402 bool *found_in_application_cache
);
405 radv_pipeline_cache_insert_shaders(struct radv_device
*device
,
406 struct radv_pipeline_cache
*cache
,
407 const unsigned char *sha1
,
408 struct radv_shader_variant
**variants
,
409 struct radv_shader_binary
*const *binaries
);
411 enum radv_blit_ds_layout
{
412 RADV_BLIT_DS_LAYOUT_TILE_ENABLE
,
413 RADV_BLIT_DS_LAYOUT_TILE_DISABLE
,
414 RADV_BLIT_DS_LAYOUT_COUNT
,
417 static inline enum radv_blit_ds_layout
radv_meta_blit_ds_to_type(VkImageLayout layout
)
419 return (layout
== VK_IMAGE_LAYOUT_GENERAL
) ? RADV_BLIT_DS_LAYOUT_TILE_DISABLE
: RADV_BLIT_DS_LAYOUT_TILE_ENABLE
;
422 static inline VkImageLayout
radv_meta_blit_ds_to_layout(enum radv_blit_ds_layout ds_layout
)
424 return ds_layout
== RADV_BLIT_DS_LAYOUT_TILE_ENABLE
? VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
: VK_IMAGE_LAYOUT_GENERAL
;
427 enum radv_meta_dst_layout
{
428 RADV_META_DST_LAYOUT_GENERAL
,
429 RADV_META_DST_LAYOUT_OPTIMAL
,
430 RADV_META_DST_LAYOUT_COUNT
,
433 static inline enum radv_meta_dst_layout
radv_meta_dst_layout_from_layout(VkImageLayout layout
)
435 return (layout
== VK_IMAGE_LAYOUT_GENERAL
) ? RADV_META_DST_LAYOUT_GENERAL
: RADV_META_DST_LAYOUT_OPTIMAL
;
438 static inline VkImageLayout
radv_meta_dst_layout_to_layout(enum radv_meta_dst_layout layout
)
440 return layout
== RADV_META_DST_LAYOUT_OPTIMAL
? VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
: VK_IMAGE_LAYOUT_GENERAL
;
443 struct radv_meta_state
{
444 VkAllocationCallbacks alloc
;
446 struct radv_pipeline_cache cache
;
449 * For on-demand pipeline creation, makes sure that
450 * only one thread tries to build a pipeline at the same time.
455 * Use array element `i` for images with `2^i` samples.
458 VkRenderPass render_pass
[NUM_META_FS_KEYS
];
459 VkPipeline color_pipelines
[NUM_META_FS_KEYS
];
461 VkRenderPass depthstencil_rp
;
462 VkPipeline depth_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
463 VkPipeline stencil_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
464 VkPipeline depthstencil_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
465 } clear
[1 + MAX_SAMPLES_LOG2
];
467 VkPipelineLayout clear_color_p_layout
;
468 VkPipelineLayout clear_depth_p_layout
;
470 /* Optimized compute fast HTILE clear for stencil or depth only. */
471 VkPipeline clear_htile_mask_pipeline
;
472 VkPipelineLayout clear_htile_mask_p_layout
;
473 VkDescriptorSetLayout clear_htile_mask_ds_layout
;
476 VkRenderPass render_pass
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
478 /** Pipeline that blits from a 1D image. */
479 VkPipeline pipeline_1d_src
[NUM_META_FS_KEYS
];
481 /** Pipeline that blits from a 2D image. */
482 VkPipeline pipeline_2d_src
[NUM_META_FS_KEYS
];
484 /** Pipeline that blits from a 3D image. */
485 VkPipeline pipeline_3d_src
[NUM_META_FS_KEYS
];
487 VkRenderPass depth_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
488 VkPipeline depth_only_1d_pipeline
;
489 VkPipeline depth_only_2d_pipeline
;
490 VkPipeline depth_only_3d_pipeline
;
492 VkRenderPass stencil_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
493 VkPipeline stencil_only_1d_pipeline
;
494 VkPipeline stencil_only_2d_pipeline
;
495 VkPipeline stencil_only_3d_pipeline
;
496 VkPipelineLayout pipeline_layout
;
497 VkDescriptorSetLayout ds_layout
;
501 VkPipelineLayout p_layouts
[5];
502 VkDescriptorSetLayout ds_layouts
[5];
503 VkPipeline pipelines
[5][NUM_META_FS_KEYS
];
505 VkPipeline depth_only_pipeline
[5];
507 VkPipeline stencil_only_pipeline
[5];
508 } blit2d
[1 + MAX_SAMPLES_LOG2
];
510 VkRenderPass blit2d_render_passes
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
511 VkRenderPass blit2d_depth_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
512 VkRenderPass blit2d_stencil_only_rp
[RADV_BLIT_DS_LAYOUT_COUNT
];
515 VkPipelineLayout img_p_layout
;
516 VkDescriptorSetLayout img_ds_layout
;
518 VkPipeline pipeline_3d
;
521 VkPipelineLayout img_p_layout
;
522 VkDescriptorSetLayout img_ds_layout
;
524 VkPipeline pipeline_3d
;
527 VkPipelineLayout img_p_layout
;
528 VkDescriptorSetLayout img_ds_layout
;
532 VkPipelineLayout img_p_layout
;
533 VkDescriptorSetLayout img_ds_layout
;
535 VkPipeline pipeline_3d
;
538 VkPipelineLayout img_p_layout
;
539 VkDescriptorSetLayout img_ds_layout
;
543 VkPipelineLayout img_p_layout
;
544 VkDescriptorSetLayout img_ds_layout
;
546 VkPipeline pipeline_3d
;
549 VkPipelineLayout img_p_layout
;
550 VkDescriptorSetLayout img_ds_layout
;
555 VkPipelineLayout p_layout
;
556 VkPipeline pipeline
[NUM_META_FS_KEYS
];
557 VkRenderPass pass
[NUM_META_FS_KEYS
];
561 VkDescriptorSetLayout ds_layout
;
562 VkPipelineLayout p_layout
;
565 VkPipeline i_pipeline
;
566 VkPipeline srgb_pipeline
;
567 } rc
[MAX_SAMPLES_LOG2
];
569 VkPipeline depth_zero_pipeline
;
571 VkPipeline average_pipeline
;
572 VkPipeline max_pipeline
;
573 VkPipeline min_pipeline
;
574 } depth
[MAX_SAMPLES_LOG2
];
576 VkPipeline stencil_zero_pipeline
;
578 VkPipeline max_pipeline
;
579 VkPipeline min_pipeline
;
580 } stencil
[MAX_SAMPLES_LOG2
];
584 VkDescriptorSetLayout ds_layout
;
585 VkPipelineLayout p_layout
;
588 VkRenderPass render_pass
[NUM_META_FS_KEYS
][RADV_META_DST_LAYOUT_COUNT
];
589 VkPipeline pipeline
[NUM_META_FS_KEYS
];
590 } rc
[MAX_SAMPLES_LOG2
];
592 VkRenderPass depth_render_pass
;
593 VkPipeline depth_zero_pipeline
;
595 VkPipeline average_pipeline
;
596 VkPipeline max_pipeline
;
597 VkPipeline min_pipeline
;
598 } depth
[MAX_SAMPLES_LOG2
];
600 VkRenderPass stencil_render_pass
;
601 VkPipeline stencil_zero_pipeline
;
603 VkPipeline max_pipeline
;
604 VkPipeline min_pipeline
;
605 } stencil
[MAX_SAMPLES_LOG2
];
609 VkPipelineLayout p_layout
;
610 VkPipeline decompress_pipeline
;
611 VkPipeline resummarize_pipeline
;
613 } depth_decomp
[1 + MAX_SAMPLES_LOG2
];
616 VkPipelineLayout p_layout
;
617 VkPipeline cmask_eliminate_pipeline
;
618 VkPipeline fmask_decompress_pipeline
;
619 VkPipeline dcc_decompress_pipeline
;
622 VkDescriptorSetLayout dcc_decompress_compute_ds_layout
;
623 VkPipelineLayout dcc_decompress_compute_p_layout
;
624 VkPipeline dcc_decompress_compute_pipeline
;
628 VkPipelineLayout fill_p_layout
;
629 VkPipelineLayout copy_p_layout
;
630 VkDescriptorSetLayout fill_ds_layout
;
631 VkDescriptorSetLayout copy_ds_layout
;
632 VkPipeline fill_pipeline
;
633 VkPipeline copy_pipeline
;
637 VkDescriptorSetLayout ds_layout
;
638 VkPipelineLayout p_layout
;
639 VkPipeline occlusion_query_pipeline
;
640 VkPipeline pipeline_statistics_query_pipeline
;
641 VkPipeline tfb_query_pipeline
;
645 VkDescriptorSetLayout ds_layout
;
646 VkPipelineLayout p_layout
;
647 VkPipeline pipeline
[MAX_SAMPLES_LOG2
];
652 #define RADV_QUEUE_GENERAL 0
653 #define RADV_QUEUE_COMPUTE 1
654 #define RADV_QUEUE_TRANSFER 2
656 #define RADV_MAX_QUEUE_FAMILIES 3
658 enum ring_type
radv_queue_family_to_ring(int f
);
661 VK_LOADER_DATA _loader_data
;
662 struct radv_device
* device
;
663 struct radeon_winsys_ctx
*hw_ctx
;
664 enum radeon_ctx_priority priority
;
665 uint32_t queue_family_index
;
667 VkDeviceQueueCreateFlags flags
;
669 uint32_t scratch_size
;
670 uint32_t compute_scratch_size
;
671 uint32_t esgs_ring_size
;
672 uint32_t gsvs_ring_size
;
674 bool has_sample_positions
;
676 struct radeon_winsys_bo
*scratch_bo
;
677 struct radeon_winsys_bo
*descriptor_bo
;
678 struct radeon_winsys_bo
*compute_scratch_bo
;
679 struct radeon_winsys_bo
*esgs_ring_bo
;
680 struct radeon_winsys_bo
*gsvs_ring_bo
;
681 struct radeon_winsys_bo
*tess_rings_bo
;
682 struct radeon_cmdbuf
*initial_preamble_cs
;
683 struct radeon_cmdbuf
*initial_full_flush_preamble_cs
;
684 struct radeon_cmdbuf
*continue_preamble_cs
;
687 struct radv_bo_list
{
688 struct radv_winsys_bo_list list
;
690 pthread_mutex_t mutex
;
694 VK_LOADER_DATA _loader_data
;
696 VkAllocationCallbacks alloc
;
698 struct radv_instance
* instance
;
699 struct radeon_winsys
*ws
;
701 struct radv_meta_state meta_state
;
703 struct radv_queue
*queues
[RADV_MAX_QUEUE_FAMILIES
];
704 int queue_count
[RADV_MAX_QUEUE_FAMILIES
];
705 struct radeon_cmdbuf
*empty_cs
[RADV_MAX_QUEUE_FAMILIES
];
707 bool always_use_syncobj
;
708 bool has_distributed_tess
;
711 uint32_t tess_offchip_block_dw_size
;
712 uint32_t scratch_waves
;
713 uint32_t dispatch_initiator
;
715 uint32_t gs_table_depth
;
717 /* MSAA sample locations.
718 * The first index is the sample index.
719 * The second index is the coordinate: X, Y. */
720 float sample_locations_1x
[1][2];
721 float sample_locations_2x
[2][2];
722 float sample_locations_4x
[4][2];
723 float sample_locations_8x
[8][2];
726 uint32_t gfx_init_size_dw
;
727 struct radeon_winsys_bo
*gfx_init
;
729 struct radeon_winsys_bo
*trace_bo
;
730 uint32_t *trace_id_ptr
;
732 /* Whether to keep shader debug info, for tracing or VK_AMD_shader_info */
733 bool keep_shader_info
;
735 struct radv_physical_device
*physical_device
;
737 /* Backup in-memory cache to be used if the app doesn't provide one */
738 struct radv_pipeline_cache
* mem_cache
;
741 * use different counters so MSAA MRTs get consecutive surface indices,
742 * even if MASK is allocated in between.
744 uint32_t image_mrt_offset_counter
;
745 uint32_t fmask_mrt_offset_counter
;
746 struct list_head shader_slabs
;
747 mtx_t shader_slab_mutex
;
749 /* For detecting VM faults reported by dmesg. */
750 uint64_t dmesg_timestamp
;
752 struct radv_device_extension_table enabled_extensions
;
754 /* Whether the app has enabled the robustBufferAccess feature. */
755 bool robust_buffer_access
;
757 /* Whether the driver uses a global BO list. */
758 bool use_global_bo_list
;
760 struct radv_bo_list bo_list
;
762 /* Whether anisotropy is forced with RADV_TEX_ANISO (-1 is disabled). */
766 struct radv_device_memory
{
767 struct radeon_winsys_bo
*bo
;
768 /* for dedicated allocations */
769 struct radv_image
*image
;
770 struct radv_buffer
*buffer
;
772 VkDeviceSize map_size
;
778 struct radv_descriptor_range
{
783 struct radv_descriptor_set
{
784 const struct radv_descriptor_set_layout
*layout
;
787 struct radeon_winsys_bo
*bo
;
789 uint32_t *mapped_ptr
;
790 struct radv_descriptor_range
*dynamic_descriptors
;
792 struct radeon_winsys_bo
*descriptors
[0];
795 struct radv_push_descriptor_set
797 struct radv_descriptor_set set
;
801 struct radv_descriptor_pool_entry
{
804 struct radv_descriptor_set
*set
;
807 struct radv_descriptor_pool
{
808 struct radeon_winsys_bo
*bo
;
810 uint64_t current_offset
;
813 uint8_t *host_memory_base
;
814 uint8_t *host_memory_ptr
;
815 uint8_t *host_memory_end
;
817 uint32_t entry_count
;
818 uint32_t max_entry_count
;
819 struct radv_descriptor_pool_entry entries
[0];
822 struct radv_descriptor_update_template_entry
{
823 VkDescriptorType descriptor_type
;
825 /* The number of descriptors to update */
826 uint32_t descriptor_count
;
828 /* Into mapped_ptr or dynamic_descriptors, in units of the respective array */
831 /* In dwords. Not valid/used for dynamic descriptors */
834 uint32_t buffer_offset
;
836 /* Only valid for combined image samplers and samplers */
838 uint8_t sampler_offset
;
844 /* For push descriptors */
845 const uint32_t *immutable_samplers
;
848 struct radv_descriptor_update_template
{
849 uint32_t entry_count
;
850 VkPipelineBindPoint bind_point
;
851 struct radv_descriptor_update_template_entry entry
[0];
857 VkBufferUsageFlags usage
;
858 VkBufferCreateFlags flags
;
861 struct radeon_winsys_bo
* bo
;
867 enum radv_dynamic_state_bits
{
868 RADV_DYNAMIC_VIEWPORT
= 1 << 0,
869 RADV_DYNAMIC_SCISSOR
= 1 << 1,
870 RADV_DYNAMIC_LINE_WIDTH
= 1 << 2,
871 RADV_DYNAMIC_DEPTH_BIAS
= 1 << 3,
872 RADV_DYNAMIC_BLEND_CONSTANTS
= 1 << 4,
873 RADV_DYNAMIC_DEPTH_BOUNDS
= 1 << 5,
874 RADV_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6,
875 RADV_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7,
876 RADV_DYNAMIC_STENCIL_REFERENCE
= 1 << 8,
877 RADV_DYNAMIC_DISCARD_RECTANGLE
= 1 << 9,
878 RADV_DYNAMIC_SAMPLE_LOCATIONS
= 1 << 10,
879 RADV_DYNAMIC_ALL
= (1 << 11) - 1,
882 enum radv_cmd_dirty_bits
{
883 /* Keep the dynamic state dirty bits in sync with
884 * enum radv_dynamic_state_bits */
885 RADV_CMD_DIRTY_DYNAMIC_VIEWPORT
= 1 << 0,
886 RADV_CMD_DIRTY_DYNAMIC_SCISSOR
= 1 << 1,
887 RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH
= 1 << 2,
888 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS
= 1 << 3,
889 RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS
= 1 << 4,
890 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS
= 1 << 5,
891 RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6,
892 RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7,
893 RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE
= 1 << 8,
894 RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE
= 1 << 9,
895 RADV_CMD_DIRTY_DYNAMIC_SAMPLE_LOCATIONS
= 1 << 10,
896 RADV_CMD_DIRTY_DYNAMIC_ALL
= (1 << 11) - 1,
897 RADV_CMD_DIRTY_PIPELINE
= 1 << 11,
898 RADV_CMD_DIRTY_INDEX_BUFFER
= 1 << 12,
899 RADV_CMD_DIRTY_FRAMEBUFFER
= 1 << 13,
900 RADV_CMD_DIRTY_VERTEX_BUFFER
= 1 << 14,
901 RADV_CMD_DIRTY_STREAMOUT_BUFFER
= 1 << 15,
904 enum radv_cmd_flush_bits
{
905 /* Instruction cache. */
906 RADV_CMD_FLAG_INV_ICACHE
= 1 << 0,
907 /* Scalar L1 cache. */
908 RADV_CMD_FLAG_INV_SCACHE
= 1 << 1,
909 /* Vector L1 cache. */
910 RADV_CMD_FLAG_INV_VCACHE
= 1 << 2,
911 /* L2 cache + L2 metadata cache writeback & invalidate.
912 * GFX6-8: Used by shaders only. GFX9-10: Used by everything. */
913 RADV_CMD_FLAG_INV_L2
= 1 << 3,
914 /* L2 writeback (write dirty L2 lines to memory for non-L2 clients).
915 * Only used for coherency with non-L2 clients like CB, DB, CP on GFX6-8.
916 * GFX6-7 will do complete invalidation, because the writeback is unsupported. */
917 RADV_CMD_FLAG_WB_L2
= 1 << 4,
918 /* Framebuffer caches */
919 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
= 1 << 5,
920 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
= 1 << 6,
921 RADV_CMD_FLAG_FLUSH_AND_INV_DB
= 1 << 7,
922 RADV_CMD_FLAG_FLUSH_AND_INV_CB
= 1 << 8,
923 /* Engine synchronization. */
924 RADV_CMD_FLAG_VS_PARTIAL_FLUSH
= 1 << 9,
925 RADV_CMD_FLAG_PS_PARTIAL_FLUSH
= 1 << 10,
926 RADV_CMD_FLAG_CS_PARTIAL_FLUSH
= 1 << 11,
927 RADV_CMD_FLAG_VGT_FLUSH
= 1 << 12,
928 /* Pipeline query controls. */
929 RADV_CMD_FLAG_START_PIPELINE_STATS
= 1 << 13,
930 RADV_CMD_FLAG_STOP_PIPELINE_STATS
= 1 << 14,
931 RADV_CMD_FLAG_VGT_STREAMOUT_SYNC
= 1 << 15,
933 RADV_CMD_FLUSH_AND_INV_FRAMEBUFFER
= (RADV_CMD_FLAG_FLUSH_AND_INV_CB
|
934 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
|
935 RADV_CMD_FLAG_FLUSH_AND_INV_DB
|
936 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
)
939 struct radv_vertex_binding
{
940 struct radv_buffer
* buffer
;
944 struct radv_streamout_binding
{
945 struct radv_buffer
*buffer
;
950 struct radv_streamout_state
{
951 /* Mask of bound streamout buffers. */
952 uint8_t enabled_mask
;
954 /* External state that comes from the last vertex stage, it must be
955 * set explicitely when binding a new graphics pipeline.
957 uint16_t stride_in_dw
[MAX_SO_BUFFERS
];
958 uint32_t enabled_stream_buffers_mask
; /* stream0 buffers0-3 in 4 LSB */
960 /* State of VGT_STRMOUT_BUFFER_(CONFIG|END) */
961 uint32_t hw_enabled_mask
;
963 /* State of VGT_STRMOUT_(CONFIG|EN) */
964 bool streamout_enabled
;
967 struct radv_viewport_state
{
969 VkViewport viewports
[MAX_VIEWPORTS
];
972 struct radv_scissor_state
{
974 VkRect2D scissors
[MAX_SCISSORS
];
977 struct radv_discard_rectangle_state
{
979 VkRect2D rectangles
[MAX_DISCARD_RECTANGLES
];
982 struct radv_sample_locations_state
{
983 VkSampleCountFlagBits per_pixel
;
984 VkExtent2D grid_size
;
986 VkSampleLocationEXT locations
[MAX_SAMPLE_LOCATIONS
];
989 struct radv_dynamic_state
{
991 * Bitmask of (1 << VK_DYNAMIC_STATE_*).
992 * Defines the set of saved dynamic state.
996 struct radv_viewport_state viewport
;
998 struct radv_scissor_state scissor
;
1008 float blend_constants
[4];
1018 } stencil_compare_mask
;
1023 } stencil_write_mask
;
1028 } stencil_reference
;
1030 struct radv_discard_rectangle_state discard_rectangle
;
1032 struct radv_sample_locations_state sample_location
;
1035 extern const struct radv_dynamic_state default_dynamic_state
;
1038 radv_get_debug_option_name(int id
);
1041 radv_get_perftest_option_name(int id
);
1043 struct radv_color_buffer_info
{
1044 uint64_t cb_color_base
;
1045 uint64_t cb_color_cmask
;
1046 uint64_t cb_color_fmask
;
1047 uint64_t cb_dcc_base
;
1048 uint32_t cb_color_slice
;
1049 uint32_t cb_color_view
;
1050 uint32_t cb_color_info
;
1051 uint32_t cb_color_attrib
;
1052 uint32_t cb_color_attrib2
; /* GFX9 and later */
1053 uint32_t cb_color_attrib3
; /* GFX10 and later */
1054 uint32_t cb_dcc_control
;
1055 uint32_t cb_color_cmask_slice
;
1056 uint32_t cb_color_fmask_slice
;
1058 uint32_t cb_color_pitch
; // GFX6-GFX8
1059 uint32_t cb_mrt_epitch
; // GFX9+
1063 struct radv_ds_buffer_info
{
1064 uint64_t db_z_read_base
;
1065 uint64_t db_stencil_read_base
;
1066 uint64_t db_z_write_base
;
1067 uint64_t db_stencil_write_base
;
1068 uint64_t db_htile_data_base
;
1069 uint32_t db_depth_info
;
1071 uint32_t db_stencil_info
;
1072 uint32_t db_depth_view
;
1073 uint32_t db_depth_size
;
1074 uint32_t db_depth_slice
;
1075 uint32_t db_htile_surface
;
1076 uint32_t pa_su_poly_offset_db_fmt_cntl
;
1077 uint32_t db_z_info2
; /* GFX9 only */
1078 uint32_t db_stencil_info2
; /* GFX9 only */
1083 radv_initialise_color_surface(struct radv_device
*device
,
1084 struct radv_color_buffer_info
*cb
,
1085 struct radv_image_view
*iview
);
1087 radv_initialise_ds_surface(struct radv_device
*device
,
1088 struct radv_ds_buffer_info
*ds
,
1089 struct radv_image_view
*iview
);
1092 * Attachment state when recording a renderpass instance.
1094 * The clear value is valid only if there exists a pending clear.
1096 struct radv_attachment_state
{
1097 VkImageAspectFlags pending_clear_aspects
;
1098 uint32_t cleared_views
;
1099 VkClearValue clear_value
;
1100 VkImageLayout current_layout
;
1101 bool current_in_render_loop
;
1102 struct radv_sample_locations_state sample_location
;
1105 struct radv_color_buffer_info cb
;
1106 struct radv_ds_buffer_info ds
;
1108 struct radv_image_view
*iview
;
1111 struct radv_descriptor_state
{
1112 struct radv_descriptor_set
*sets
[MAX_SETS
];
1115 struct radv_push_descriptor_set push_set
;
1117 uint32_t dynamic_buffers
[4 * MAX_DYNAMIC_BUFFERS
];
1120 struct radv_subpass_sample_locs_state
{
1121 uint32_t subpass_idx
;
1122 struct radv_sample_locations_state sample_location
;
1125 struct radv_cmd_state
{
1126 /* Vertex descriptors */
1133 uint32_t prefetch_L2_mask
;
1135 struct radv_pipeline
* pipeline
;
1136 struct radv_pipeline
* emitted_pipeline
;
1137 struct radv_pipeline
* compute_pipeline
;
1138 struct radv_pipeline
* emitted_compute_pipeline
;
1139 struct radv_framebuffer
* framebuffer
;
1140 struct radv_render_pass
* pass
;
1141 const struct radv_subpass
* subpass
;
1142 struct radv_dynamic_state dynamic
;
1143 struct radv_attachment_state
* attachments
;
1144 struct radv_streamout_state streamout
;
1145 VkRect2D render_area
;
1147 uint32_t num_subpass_sample_locs
;
1148 struct radv_subpass_sample_locs_state
* subpass_sample_locs
;
1151 struct radv_buffer
*index_buffer
;
1152 uint64_t index_offset
;
1153 uint32_t index_type
;
1154 uint32_t max_index_count
;
1156 int32_t last_index_type
;
1158 int32_t last_primitive_reset_en
;
1159 uint32_t last_primitive_reset_index
;
1160 enum radv_cmd_flush_bits flush_bits
;
1161 unsigned active_occlusion_queries
;
1162 bool perfect_occlusion_queries_enabled
;
1163 unsigned active_pipeline_queries
;
1166 uint32_t last_ia_multi_vgt_param
;
1168 uint32_t last_num_instances
;
1169 uint32_t last_first_instance
;
1170 uint32_t last_vertex_offset
;
1172 /* Whether CP DMA is busy/idle. */
1175 /* Conditional rendering info. */
1176 int predication_type
; /* -1: disabled, 0: normal, 1: inverted */
1177 uint64_t predication_va
;
1179 bool context_roll_without_scissor_emitted
;
1182 struct radv_cmd_pool
{
1183 VkAllocationCallbacks alloc
;
1184 struct list_head cmd_buffers
;
1185 struct list_head free_cmd_buffers
;
1186 uint32_t queue_family_index
;
1189 struct radv_cmd_buffer_upload
{
1193 struct radeon_winsys_bo
*upload_bo
;
1194 struct list_head list
;
1197 enum radv_cmd_buffer_status
{
1198 RADV_CMD_BUFFER_STATUS_INVALID
,
1199 RADV_CMD_BUFFER_STATUS_INITIAL
,
1200 RADV_CMD_BUFFER_STATUS_RECORDING
,
1201 RADV_CMD_BUFFER_STATUS_EXECUTABLE
,
1202 RADV_CMD_BUFFER_STATUS_PENDING
,
1205 struct radv_cmd_buffer
{
1206 VK_LOADER_DATA _loader_data
;
1208 struct radv_device
* device
;
1210 struct radv_cmd_pool
* pool
;
1211 struct list_head pool_link
;
1213 VkCommandBufferUsageFlags usage_flags
;
1214 VkCommandBufferLevel level
;
1215 enum radv_cmd_buffer_status status
;
1216 struct radeon_cmdbuf
*cs
;
1217 struct radv_cmd_state state
;
1218 struct radv_vertex_binding vertex_bindings
[MAX_VBS
];
1219 struct radv_streamout_binding streamout_bindings
[MAX_SO_BUFFERS
];
1220 uint32_t queue_family_index
;
1222 uint8_t push_constants
[MAX_PUSH_CONSTANTS_SIZE
];
1223 VkShaderStageFlags push_constant_stages
;
1224 struct radv_descriptor_set meta_push_descriptors
;
1226 struct radv_descriptor_state descriptors
[VK_PIPELINE_BIND_POINT_RANGE_SIZE
];
1228 struct radv_cmd_buffer_upload upload
;
1230 uint32_t scratch_size_needed
;
1231 uint32_t compute_scratch_size_needed
;
1232 uint32_t esgs_ring_size_needed
;
1233 uint32_t gsvs_ring_size_needed
;
1234 bool tess_rings_needed
;
1235 bool sample_positions_needed
;
1237 VkResult record_result
;
1239 uint64_t gfx9_fence_va
;
1240 uint32_t gfx9_fence_idx
;
1241 uint64_t gfx9_eop_bug_va
;
1244 * Whether a query pool has been resetted and we have to flush caches.
1246 bool pending_reset_query
;
1249 * Bitmask of pending active query flushes.
1251 enum radv_cmd_flush_bits active_query_flush_bits
;
1255 struct radv_image_view
;
1257 bool radv_cmd_buffer_uses_mec(struct radv_cmd_buffer
*cmd_buffer
);
1259 void si_emit_graphics(struct radv_physical_device
*physical_device
,
1260 struct radeon_cmdbuf
*cs
);
1261 void si_emit_compute(struct radv_physical_device
*physical_device
,
1262 struct radeon_cmdbuf
*cs
);
1264 void cik_create_gfx_config(struct radv_device
*device
);
1266 void si_write_viewport(struct radeon_cmdbuf
*cs
, int first_vp
,
1267 int count
, const VkViewport
*viewports
);
1268 void si_write_scissors(struct radeon_cmdbuf
*cs
, int first
,
1269 int count
, const VkRect2D
*scissors
,
1270 const VkViewport
*viewports
, bool can_use_guardband
);
1271 uint32_t si_get_ia_multi_vgt_param(struct radv_cmd_buffer
*cmd_buffer
,
1272 bool instanced_draw
, bool indirect_draw
,
1273 bool count_from_stream_output
,
1274 uint32_t draw_vertex_count
);
1275 void si_cs_emit_write_event_eop(struct radeon_cmdbuf
*cs
,
1276 enum chip_class chip_class
,
1278 unsigned event
, unsigned event_flags
,
1279 unsigned dst_sel
, unsigned data_sel
,
1282 uint64_t gfx9_eop_bug_va
);
1284 void radv_cp_wait_mem(struct radeon_cmdbuf
*cs
, uint32_t op
, uint64_t va
,
1285 uint32_t ref
, uint32_t mask
);
1286 void si_cs_emit_cache_flush(struct radeon_cmdbuf
*cs
,
1287 enum chip_class chip_class
,
1288 uint32_t *fence_ptr
, uint64_t va
,
1290 enum radv_cmd_flush_bits flush_bits
,
1291 uint64_t gfx9_eop_bug_va
);
1292 void si_emit_cache_flush(struct radv_cmd_buffer
*cmd_buffer
);
1293 void si_emit_set_predication_state(struct radv_cmd_buffer
*cmd_buffer
,
1294 bool inverted
, uint64_t va
);
1295 void si_cp_dma_buffer_copy(struct radv_cmd_buffer
*cmd_buffer
,
1296 uint64_t src_va
, uint64_t dest_va
,
1298 void si_cp_dma_prefetch(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
1300 void si_cp_dma_clear_buffer(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
1301 uint64_t size
, unsigned value
);
1302 void si_cp_dma_wait_for_idle(struct radv_cmd_buffer
*cmd_buffer
);
1304 void radv_set_db_count_control(struct radv_cmd_buffer
*cmd_buffer
);
1306 radv_cmd_buffer_upload_alloc(struct radv_cmd_buffer
*cmd_buffer
,
1309 unsigned *out_offset
,
1312 radv_cmd_buffer_set_subpass(struct radv_cmd_buffer
*cmd_buffer
,
1313 const struct radv_subpass
*subpass
);
1315 radv_cmd_buffer_upload_data(struct radv_cmd_buffer
*cmd_buffer
,
1316 unsigned size
, unsigned alignmnet
,
1317 const void *data
, unsigned *out_offset
);
1319 void radv_cmd_buffer_clear_subpass(struct radv_cmd_buffer
*cmd_buffer
);
1320 void radv_cmd_buffer_resolve_subpass(struct radv_cmd_buffer
*cmd_buffer
);
1321 void radv_cmd_buffer_resolve_subpass_cs(struct radv_cmd_buffer
*cmd_buffer
);
1322 void radv_depth_stencil_resolve_subpass_cs(struct radv_cmd_buffer
*cmd_buffer
,
1323 VkImageAspectFlags aspects
,
1324 VkResolveModeFlagBitsKHR resolve_mode
);
1325 void radv_cmd_buffer_resolve_subpass_fs(struct radv_cmd_buffer
*cmd_buffer
);
1326 void radv_depth_stencil_resolve_subpass_fs(struct radv_cmd_buffer
*cmd_buffer
,
1327 VkImageAspectFlags aspects
,
1328 VkResolveModeFlagBitsKHR resolve_mode
);
1329 void radv_emit_default_sample_locations(struct radeon_cmdbuf
*cs
, int nr_samples
);
1330 unsigned radv_get_default_max_sample_dist(int log_samples
);
1331 void radv_device_init_msaa(struct radv_device
*device
);
1333 void radv_update_ds_clear_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1334 const struct radv_image_view
*iview
,
1335 VkClearDepthStencilValue ds_clear_value
,
1336 VkImageAspectFlags aspects
);
1338 void radv_update_color_clear_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1339 const struct radv_image_view
*iview
,
1341 uint32_t color_values
[2]);
1343 void radv_update_fce_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1344 struct radv_image
*image
,
1345 const VkImageSubresourceRange
*range
, bool value
);
1347 void radv_update_dcc_metadata(struct radv_cmd_buffer
*cmd_buffer
,
1348 struct radv_image
*image
,
1349 const VkImageSubresourceRange
*range
, bool value
);
1351 uint32_t radv_fill_buffer(struct radv_cmd_buffer
*cmd_buffer
,
1352 struct radeon_winsys_bo
*bo
,
1353 uint64_t offset
, uint64_t size
, uint32_t value
);
1354 void radv_cmd_buffer_trace_emit(struct radv_cmd_buffer
*cmd_buffer
);
1355 bool radv_get_memory_fd(struct radv_device
*device
,
1356 struct radv_device_memory
*memory
,
1360 radv_emit_shader_pointer_head(struct radeon_cmdbuf
*cs
,
1361 unsigned sh_offset
, unsigned pointer_count
,
1362 bool use_32bit_pointers
)
1364 radeon_emit(cs
, PKT3(PKT3_SET_SH_REG
, pointer_count
* (use_32bit_pointers
? 1 : 2), 0));
1365 radeon_emit(cs
, (sh_offset
- SI_SH_REG_OFFSET
) >> 2);
1369 radv_emit_shader_pointer_body(struct radv_device
*device
,
1370 struct radeon_cmdbuf
*cs
,
1371 uint64_t va
, bool use_32bit_pointers
)
1373 radeon_emit(cs
, va
);
1375 if (use_32bit_pointers
) {
1377 (va
>> 32) == device
->physical_device
->rad_info
.address32_hi
);
1379 radeon_emit(cs
, va
>> 32);
1384 radv_emit_shader_pointer(struct radv_device
*device
,
1385 struct radeon_cmdbuf
*cs
,
1386 uint32_t sh_offset
, uint64_t va
, bool global
)
1388 bool use_32bit_pointers
= !global
;
1390 radv_emit_shader_pointer_head(cs
, sh_offset
, 1, use_32bit_pointers
);
1391 radv_emit_shader_pointer_body(device
, cs
, va
, use_32bit_pointers
);
1394 static inline struct radv_descriptor_state
*
1395 radv_get_descriptors_state(struct radv_cmd_buffer
*cmd_buffer
,
1396 VkPipelineBindPoint bind_point
)
1398 assert(bind_point
== VK_PIPELINE_BIND_POINT_GRAPHICS
||
1399 bind_point
== VK_PIPELINE_BIND_POINT_COMPUTE
);
1400 return &cmd_buffer
->descriptors
[bind_point
];
1404 * Takes x,y,z as exact numbers of invocations, instead of blocks.
1406 * Limitations: Can't call normal dispatch functions without binding or rebinding
1407 * the compute pipeline.
1409 void radv_unaligned_dispatch(
1410 struct radv_cmd_buffer
*cmd_buffer
,
1416 struct radeon_winsys_bo
*bo
;
1420 struct radv_shader_module
;
1422 #define RADV_HASH_SHADER_IS_GEOM_COPY_SHADER (1 << 0)
1423 #define RADV_HASH_SHADER_SISCHED (1 << 1)
1424 #define RADV_HASH_SHADER_UNSAFE_MATH (1 << 2)
1425 #define RADV_HASH_SHADER_NO_NGG (1 << 3)
1426 #define RADV_HASH_SHADER_CS_WAVE32 (1 << 4)
1427 #define RADV_HASH_SHADER_PS_WAVE32 (1 << 5)
1428 #define RADV_HASH_SHADER_GE_WAVE32 (1 << 6)
1431 radv_hash_shaders(unsigned char *hash
,
1432 const VkPipelineShaderStageCreateInfo
**stages
,
1433 const struct radv_pipeline_layout
*layout
,
1434 const struct radv_pipeline_key
*key
,
1437 static inline gl_shader_stage
1438 vk_to_mesa_shader_stage(VkShaderStageFlagBits vk_stage
)
1440 assert(__builtin_popcount(vk_stage
) == 1);
1441 return ffs(vk_stage
) - 1;
1444 static inline VkShaderStageFlagBits
1445 mesa_to_vk_shader_stage(gl_shader_stage mesa_stage
)
1447 return (1 << mesa_stage
);
1450 #define RADV_STAGE_MASK ((1 << MESA_SHADER_STAGES) - 1)
1452 #define radv_foreach_stage(stage, stage_bits) \
1453 for (gl_shader_stage stage, \
1454 __tmp = (gl_shader_stage)((stage_bits) & RADV_STAGE_MASK); \
1455 stage = __builtin_ffs(__tmp) - 1, __tmp; \
1456 __tmp &= ~(1 << (stage)))
1458 extern const VkFormat radv_fs_key_format_exemplars
[NUM_META_FS_KEYS
];
1459 unsigned radv_format_meta_fs_key(VkFormat format
);
1461 struct radv_multisample_state
{
1463 uint32_t pa_sc_line_cntl
;
1464 uint32_t pa_sc_mode_cntl_0
;
1465 uint32_t pa_sc_mode_cntl_1
;
1466 uint32_t pa_sc_aa_config
;
1467 uint32_t pa_sc_aa_mask
[2];
1468 unsigned num_samples
;
1471 struct radv_prim_vertex_count
{
1476 struct radv_vertex_elements_info
{
1477 uint32_t format_size
[MAX_VERTEX_ATTRIBS
];
1480 struct radv_ia_multi_vgt_param_helpers
{
1482 bool partial_es_wave
;
1483 uint8_t primgroup_size
;
1484 bool wd_switch_on_eop
;
1485 bool ia_switch_on_eoi
;
1486 bool partial_vs_wave
;
1489 struct radv_binning_state
{
1490 uint32_t pa_sc_binner_cntl_0
;
1491 uint32_t db_dfsm_control
;
1494 #define SI_GS_PER_ES 128
1496 struct radv_pipeline
{
1497 struct radv_device
* device
;
1498 struct radv_dynamic_state dynamic_state
;
1500 struct radv_pipeline_layout
* layout
;
1502 bool need_indirect_descriptor_sets
;
1503 struct radv_shader_variant
* shaders
[MESA_SHADER_STAGES
];
1504 struct radv_shader_variant
*gs_copy_shader
;
1505 VkShaderStageFlags active_stages
;
1507 struct radeon_cmdbuf cs
;
1508 uint32_t ctx_cs_hash
;
1509 struct radeon_cmdbuf ctx_cs
;
1511 struct radv_vertex_elements_info vertex_elements
;
1513 uint32_t binding_stride
[MAX_VBS
];
1514 uint8_t num_vertex_bindings
;
1516 uint32_t user_data_0
[MESA_SHADER_STAGES
];
1519 struct radv_multisample_state ms
;
1520 struct radv_binning_state binning
;
1521 uint32_t spi_baryc_cntl
;
1522 bool prim_restart_enable
;
1523 unsigned esgs_ring_size
;
1524 unsigned gsvs_ring_size
;
1525 uint32_t vtx_base_sgpr
;
1526 struct radv_ia_multi_vgt_param_helpers ia_multi_vgt_param
;
1527 uint8_t vtx_emit_num
;
1528 struct radv_prim_vertex_count prim_vertex_count
;
1529 bool can_use_guardband
;
1530 uint32_t needed_dynamic_state
;
1531 bool disable_out_of_order_rast_for_occlusion
;
1533 /* Used for rbplus */
1534 uint32_t col_format
;
1535 uint32_t cb_target_mask
;
1540 unsigned scratch_bytes_per_wave
;
1542 /* Not NULL if graphics pipeline uses streamout. */
1543 struct radv_shader_variant
*streamout_shader
;
1546 static inline bool radv_pipeline_has_gs(const struct radv_pipeline
*pipeline
)
1548 return pipeline
->shaders
[MESA_SHADER_GEOMETRY
] ? true : false;
1551 static inline bool radv_pipeline_has_tess(const struct radv_pipeline
*pipeline
)
1553 return pipeline
->shaders
[MESA_SHADER_TESS_CTRL
] ? true : false;
1556 bool radv_pipeline_has_ngg(const struct radv_pipeline
*pipeline
);
1558 bool radv_pipeline_has_gs_copy_shader(const struct radv_pipeline
*pipeline
);
1560 struct radv_userdata_info
*radv_lookup_user_sgpr(struct radv_pipeline
*pipeline
,
1561 gl_shader_stage stage
,
1564 struct radv_shader_variant
*radv_get_shader(struct radv_pipeline
*pipeline
,
1565 gl_shader_stage stage
);
1567 struct radv_graphics_pipeline_create_info
{
1569 bool db_depth_clear
;
1570 bool db_stencil_clear
;
1571 bool db_depth_disable_expclear
;
1572 bool db_stencil_disable_expclear
;
1573 bool db_flush_depth_inplace
;
1574 bool db_flush_stencil_inplace
;
1575 bool db_resummarize
;
1576 uint32_t custom_blend_mode
;
1580 radv_graphics_pipeline_create(VkDevice device
,
1581 VkPipelineCache cache
,
1582 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
1583 const struct radv_graphics_pipeline_create_info
*extra
,
1584 const VkAllocationCallbacks
*alloc
,
1585 VkPipeline
*pPipeline
);
1587 struct vk_format_description
;
1588 uint32_t radv_translate_buffer_dataformat(const struct vk_format_description
*desc
,
1589 int first_non_void
);
1590 uint32_t radv_translate_buffer_numformat(const struct vk_format_description
*desc
,
1591 int first_non_void
);
1592 bool radv_is_buffer_format_supported(VkFormat format
, bool *scaled
);
1593 uint32_t radv_translate_colorformat(VkFormat format
);
1594 uint32_t radv_translate_color_numformat(VkFormat format
,
1595 const struct vk_format_description
*desc
,
1596 int first_non_void
);
1597 uint32_t radv_colorformat_endian_swap(uint32_t colorformat
);
1598 unsigned radv_translate_colorswap(VkFormat format
, bool do_endian_swap
);
1599 uint32_t radv_translate_dbformat(VkFormat format
);
1600 uint32_t radv_translate_tex_dataformat(VkFormat format
,
1601 const struct vk_format_description
*desc
,
1602 int first_non_void
);
1603 uint32_t radv_translate_tex_numformat(VkFormat format
,
1604 const struct vk_format_description
*desc
,
1605 int first_non_void
);
1606 bool radv_format_pack_clear_color(VkFormat format
,
1607 uint32_t clear_vals
[2],
1608 VkClearColorValue
*value
);
1609 bool radv_is_colorbuffer_format_supported(VkFormat format
, bool *blendable
);
1610 bool radv_dcc_formats_compatible(VkFormat format1
,
1612 bool radv_device_supports_etc(struct radv_physical_device
*physical_device
);
1614 struct radv_image_plane
{
1616 struct radeon_surf surface
;
1622 /* The original VkFormat provided by the client. This may not match any
1623 * of the actual surface formats.
1626 VkImageAspectFlags aspects
;
1627 VkImageUsageFlags usage
; /**< Superset of VkImageCreateInfo::usage. */
1628 struct ac_surf_info info
;
1629 VkImageTiling tiling
; /** VkImageCreateInfo::tiling */
1630 VkImageCreateFlags flags
; /** VkImageCreateInfo::flags */
1635 unsigned queue_family_mask
;
1639 /* Set when bound */
1640 struct radeon_winsys_bo
*bo
;
1641 VkDeviceSize offset
;
1642 uint64_t dcc_offset
;
1643 uint64_t htile_offset
;
1644 bool tc_compatible_htile
;
1645 bool tc_compatible_cmask
;
1647 uint64_t cmask_offset
;
1648 uint64_t fmask_offset
;
1649 uint64_t clear_value_offset
;
1650 uint64_t fce_pred_offset
;
1651 uint64_t dcc_pred_offset
;
1654 * Metadata for the TC-compat zrange workaround. If the 32-bit value
1655 * stored at this offset is UINT_MAX, the driver will emit
1656 * DB_Z_INFO.ZRANGE_PRECISION=0, otherwise it will skip the
1657 * SET_CONTEXT_REG packet.
1659 uint64_t tc_compat_zrange_offset
;
1661 /* For VK_ANDROID_native_buffer, the WSI image owns the memory, */
1662 VkDeviceMemory owned_memory
;
1664 unsigned plane_count
;
1665 struct radv_image_plane planes
[0];
1668 /* Whether the image has a htile that is known consistent with the contents of
1670 bool radv_layout_has_htile(const struct radv_image
*image
,
1671 VkImageLayout layout
,
1672 bool in_render_loop
,
1673 unsigned queue_mask
);
1675 /* Whether the image has a htile that is known consistent with the contents of
1676 * the image and is allowed to be in compressed form.
1678 * If this is false reads that don't use the htile should be able to return
1681 bool radv_layout_is_htile_compressed(const struct radv_image
*image
,
1682 VkImageLayout layout
,
1683 bool in_render_loop
,
1684 unsigned queue_mask
);
1686 bool radv_layout_can_fast_clear(const struct radv_image
*image
,
1687 VkImageLayout layout
,
1688 bool in_render_loop
,
1689 unsigned queue_mask
);
1691 bool radv_layout_dcc_compressed(const struct radv_device
*device
,
1692 const struct radv_image
*image
,
1693 VkImageLayout layout
,
1694 bool in_render_loop
,
1695 unsigned queue_mask
);
1698 * Return whether the image has CMASK metadata for color surfaces.
1701 radv_image_has_cmask(const struct radv_image
*image
)
1703 return image
->cmask_offset
;
1707 * Return whether the image has FMASK metadata for color surfaces.
1710 radv_image_has_fmask(const struct radv_image
*image
)
1712 return image
->fmask_offset
;
1716 * Return whether the image has DCC metadata for color surfaces.
1719 radv_image_has_dcc(const struct radv_image
*image
)
1721 return image
->planes
[0].surface
.dcc_size
;
1725 * Return whether the image is TC-compatible CMASK.
1728 radv_image_is_tc_compat_cmask(const struct radv_image
*image
)
1730 return radv_image_has_fmask(image
) && image
->tc_compatible_cmask
;
1734 * Return whether DCC metadata is enabled for a level.
1737 radv_dcc_enabled(const struct radv_image
*image
, unsigned level
)
1739 return radv_image_has_dcc(image
) &&
1740 level
< image
->planes
[0].surface
.num_dcc_levels
;
1744 * Return whether the image has CB metadata.
1747 radv_image_has_CB_metadata(const struct radv_image
*image
)
1749 return radv_image_has_cmask(image
) ||
1750 radv_image_has_fmask(image
) ||
1751 radv_image_has_dcc(image
);
1755 * Return whether the image has HTILE metadata for depth surfaces.
1758 radv_image_has_htile(const struct radv_image
*image
)
1760 return image
->planes
[0].surface
.htile_size
;
1764 * Return whether HTILE metadata is enabled for a level.
1767 radv_htile_enabled(const struct radv_image
*image
, unsigned level
)
1769 return radv_image_has_htile(image
) && level
== 0;
1773 * Return whether the image is TC-compatible HTILE.
1776 radv_image_is_tc_compat_htile(const struct radv_image
*image
)
1778 return radv_image_has_htile(image
) && image
->tc_compatible_htile
;
1781 static inline uint64_t
1782 radv_image_get_fast_clear_va(const struct radv_image
*image
,
1783 uint32_t base_level
)
1785 uint64_t va
= radv_buffer_get_va(image
->bo
);
1786 va
+= image
->offset
+ image
->clear_value_offset
+ base_level
* 8;
1790 static inline uint64_t
1791 radv_image_get_fce_pred_va(const struct radv_image
*image
,
1792 uint32_t base_level
)
1794 uint64_t va
= radv_buffer_get_va(image
->bo
);
1795 va
+= image
->offset
+ image
->fce_pred_offset
+ base_level
* 8;
1799 static inline uint64_t
1800 radv_image_get_dcc_pred_va(const struct radv_image
*image
,
1801 uint32_t base_level
)
1803 uint64_t va
= radv_buffer_get_va(image
->bo
);
1804 va
+= image
->offset
+ image
->dcc_pred_offset
+ base_level
* 8;
1808 static inline uint64_t
1809 radv_get_tc_compat_zrange_va(const struct radv_image
*image
,
1810 uint32_t base_level
)
1812 uint64_t va
= radv_buffer_get_va(image
->bo
);
1813 va
+= image
->offset
+ image
->tc_compat_zrange_offset
+ base_level
* 4;
1817 static inline uint64_t
1818 radv_get_ds_clear_value_va(const struct radv_image
*image
,
1819 uint32_t base_level
)
1821 uint64_t va
= radv_buffer_get_va(image
->bo
);
1822 va
+= image
->offset
+ image
->clear_value_offset
+ base_level
* 8;
1826 unsigned radv_image_queue_family_mask(const struct radv_image
*image
, uint32_t family
, uint32_t queue_family
);
1828 static inline uint32_t
1829 radv_get_layerCount(const struct radv_image
*image
,
1830 const VkImageSubresourceRange
*range
)
1832 return range
->layerCount
== VK_REMAINING_ARRAY_LAYERS
?
1833 image
->info
.array_size
- range
->baseArrayLayer
: range
->layerCount
;
1836 static inline uint32_t
1837 radv_get_levelCount(const struct radv_image
*image
,
1838 const VkImageSubresourceRange
*range
)
1840 return range
->levelCount
== VK_REMAINING_MIP_LEVELS
?
1841 image
->info
.levels
- range
->baseMipLevel
: range
->levelCount
;
1844 struct radeon_bo_metadata
;
1846 radv_init_metadata(struct radv_device
*device
,
1847 struct radv_image
*image
,
1848 struct radeon_bo_metadata
*metadata
);
1851 radv_image_override_offset_stride(struct radv_device
*device
,
1852 struct radv_image
*image
,
1853 uint64_t offset
, uint32_t stride
);
1855 union radv_descriptor
{
1857 uint32_t plane0_descriptor
[8];
1858 uint32_t fmask_descriptor
[8];
1861 uint32_t plane_descriptors
[3][8];
1865 struct radv_image_view
{
1866 struct radv_image
*image
; /**< VkImageViewCreateInfo::image */
1867 struct radeon_winsys_bo
*bo
;
1869 VkImageViewType type
;
1870 VkImageAspectFlags aspect_mask
;
1873 bool multiple_planes
;
1874 uint32_t base_layer
;
1875 uint32_t layer_count
;
1877 uint32_t level_count
;
1878 VkExtent3D extent
; /**< Extent of VkImageViewCreateInfo::baseMipLevel. */
1880 union radv_descriptor descriptor
;
1882 /* Descriptor for use as a storage image as opposed to a sampled image.
1883 * This has a few differences for cube maps (e.g. type).
1885 union radv_descriptor storage_descriptor
;
1888 struct radv_image_create_info
{
1889 const VkImageCreateInfo
*vk_info
;
1891 bool no_metadata_planes
;
1892 const struct radeon_bo_metadata
*bo_metadata
;
1895 VkResult
radv_image_create(VkDevice _device
,
1896 const struct radv_image_create_info
*info
,
1897 const VkAllocationCallbacks
* alloc
,
1901 radv_image_from_gralloc(VkDevice device_h
,
1902 const VkImageCreateInfo
*base_info
,
1903 const VkNativeBufferANDROID
*gralloc_info
,
1904 const VkAllocationCallbacks
*alloc
,
1905 VkImage
*out_image_h
);
1907 struct radv_image_view_extra_create_info
{
1908 bool disable_compression
;
1911 void radv_image_view_init(struct radv_image_view
*view
,
1912 struct radv_device
*device
,
1913 const VkImageViewCreateInfo
*pCreateInfo
,
1914 const struct radv_image_view_extra_create_info
* extra_create_info
);
1916 VkFormat
radv_get_aspect_format(struct radv_image
*image
, VkImageAspectFlags mask
);
1918 struct radv_sampler_ycbcr_conversion
{
1920 VkSamplerYcbcrModelConversion ycbcr_model
;
1921 VkSamplerYcbcrRange ycbcr_range
;
1922 VkComponentMapping components
;
1923 VkChromaLocation chroma_offsets
[2];
1924 VkFilter chroma_filter
;
1927 struct radv_buffer_view
{
1928 struct radeon_winsys_bo
*bo
;
1930 uint64_t range
; /**< VkBufferViewCreateInfo::range */
1933 void radv_buffer_view_init(struct radv_buffer_view
*view
,
1934 struct radv_device
*device
,
1935 const VkBufferViewCreateInfo
* pCreateInfo
);
1937 static inline struct VkExtent3D
1938 radv_sanitize_image_extent(const VkImageType imageType
,
1939 const struct VkExtent3D imageExtent
)
1941 switch (imageType
) {
1942 case VK_IMAGE_TYPE_1D
:
1943 return (VkExtent3D
) { imageExtent
.width
, 1, 1 };
1944 case VK_IMAGE_TYPE_2D
:
1945 return (VkExtent3D
) { imageExtent
.width
, imageExtent
.height
, 1 };
1946 case VK_IMAGE_TYPE_3D
:
1949 unreachable("invalid image type");
1953 static inline struct VkOffset3D
1954 radv_sanitize_image_offset(const VkImageType imageType
,
1955 const struct VkOffset3D imageOffset
)
1957 switch (imageType
) {
1958 case VK_IMAGE_TYPE_1D
:
1959 return (VkOffset3D
) { imageOffset
.x
, 0, 0 };
1960 case VK_IMAGE_TYPE_2D
:
1961 return (VkOffset3D
) { imageOffset
.x
, imageOffset
.y
, 0 };
1962 case VK_IMAGE_TYPE_3D
:
1965 unreachable("invalid image type");
1970 radv_image_extent_compare(const struct radv_image
*image
,
1971 const VkExtent3D
*extent
)
1973 if (extent
->width
!= image
->info
.width
||
1974 extent
->height
!= image
->info
.height
||
1975 extent
->depth
!= image
->info
.depth
)
1980 struct radv_sampler
{
1982 struct radv_sampler_ycbcr_conversion
*ycbcr_sampler
;
1985 struct radv_framebuffer
{
1990 uint32_t attachment_count
;
1991 struct radv_image_view
*attachments
[0];
1994 struct radv_subpass_barrier
{
1995 VkPipelineStageFlags src_stage_mask
;
1996 VkAccessFlags src_access_mask
;
1997 VkAccessFlags dst_access_mask
;
2000 void radv_subpass_barrier(struct radv_cmd_buffer
*cmd_buffer
,
2001 const struct radv_subpass_barrier
*barrier
);
2003 struct radv_subpass_attachment
{
2004 uint32_t attachment
;
2005 VkImageLayout layout
;
2006 bool in_render_loop
;
2009 struct radv_subpass
{
2010 uint32_t attachment_count
;
2011 struct radv_subpass_attachment
* attachments
;
2013 uint32_t input_count
;
2014 uint32_t color_count
;
2015 struct radv_subpass_attachment
* input_attachments
;
2016 struct radv_subpass_attachment
* color_attachments
;
2017 struct radv_subpass_attachment
* resolve_attachments
;
2018 struct radv_subpass_attachment
* depth_stencil_attachment
;
2019 struct radv_subpass_attachment
* ds_resolve_attachment
;
2020 VkResolveModeFlagBitsKHR depth_resolve_mode
;
2021 VkResolveModeFlagBitsKHR stencil_resolve_mode
;
2023 /** Subpass has at least one color resolve attachment */
2024 bool has_color_resolve
;
2026 /** Subpass has at least one color attachment */
2029 struct radv_subpass_barrier start_barrier
;
2032 VkSampleCountFlagBits max_sample_count
;
2036 radv_get_subpass_id(struct radv_cmd_buffer
*cmd_buffer
);
2038 struct radv_render_pass_attachment
{
2041 VkAttachmentLoadOp load_op
;
2042 VkAttachmentLoadOp stencil_load_op
;
2043 VkImageLayout initial_layout
;
2044 VkImageLayout final_layout
;
2046 /* The subpass id in which the attachment will be used first/last. */
2047 uint32_t first_subpass_idx
;
2048 uint32_t last_subpass_idx
;
2051 struct radv_render_pass
{
2052 uint32_t attachment_count
;
2053 uint32_t subpass_count
;
2054 struct radv_subpass_attachment
* subpass_attachments
;
2055 struct radv_render_pass_attachment
* attachments
;
2056 struct radv_subpass_barrier end_barrier
;
2057 struct radv_subpass subpasses
[0];
2060 VkResult
radv_device_init_meta(struct radv_device
*device
);
2061 void radv_device_finish_meta(struct radv_device
*device
);
2063 struct radv_query_pool
{
2064 struct radeon_winsys_bo
*bo
;
2066 uint32_t availability_offset
;
2070 uint32_t pipeline_stats_mask
;
2073 struct radv_semaphore
{
2074 /* use a winsys sem for non-exportable */
2075 struct radeon_winsys_sem
*sem
;
2077 uint32_t temp_syncobj
;
2080 void radv_set_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
2081 VkPipelineBindPoint bind_point
,
2082 struct radv_descriptor_set
*set
,
2086 radv_update_descriptor_sets(struct radv_device
*device
,
2087 struct radv_cmd_buffer
*cmd_buffer
,
2088 VkDescriptorSet overrideSet
,
2089 uint32_t descriptorWriteCount
,
2090 const VkWriteDescriptorSet
*pDescriptorWrites
,
2091 uint32_t descriptorCopyCount
,
2092 const VkCopyDescriptorSet
*pDescriptorCopies
);
2095 radv_update_descriptor_set_with_template(struct radv_device
*device
,
2096 struct radv_cmd_buffer
*cmd_buffer
,
2097 struct radv_descriptor_set
*set
,
2098 VkDescriptorUpdateTemplate descriptorUpdateTemplate
,
2101 void radv_meta_push_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
2102 VkPipelineBindPoint pipelineBindPoint
,
2103 VkPipelineLayout _layout
,
2105 uint32_t descriptorWriteCount
,
2106 const VkWriteDescriptorSet
*pDescriptorWrites
);
2108 void radv_initialize_dcc(struct radv_cmd_buffer
*cmd_buffer
,
2109 struct radv_image
*image
,
2110 const VkImageSubresourceRange
*range
, uint32_t value
);
2112 void radv_initialize_fmask(struct radv_cmd_buffer
*cmd_buffer
,
2113 struct radv_image
*image
,
2114 const VkImageSubresourceRange
*range
);
2117 struct radeon_winsys_fence
*fence
;
2118 struct wsi_fence
*fence_wsi
;
2121 uint32_t temp_syncobj
;
2124 /* radv_nir_to_llvm.c */
2125 struct radv_shader_variant_info
;
2126 struct radv_nir_compiler_options
;
2128 void radv_compile_gs_copy_shader(struct ac_llvm_compiler
*ac_llvm
,
2129 struct nir_shader
*geom_shader
,
2130 struct radv_shader_binary
**rbinary
,
2131 struct radv_shader_variant_info
*shader_info
,
2132 const struct radv_nir_compiler_options
*option
);
2134 void radv_compile_nir_shader(struct ac_llvm_compiler
*ac_llvm
,
2135 struct radv_shader_binary
**rbinary
,
2136 struct radv_shader_variant_info
*shader_info
,
2137 struct nir_shader
*const *nir
,
2139 const struct radv_nir_compiler_options
*options
);
2141 unsigned radv_nir_get_max_workgroup_size(enum chip_class chip_class
,
2142 gl_shader_stage stage
,
2143 const struct nir_shader
*nir
);
2145 /* radv_shader_info.h */
2146 struct radv_shader_info
;
2148 void radv_nir_shader_info_pass(const struct nir_shader
*nir
,
2149 const struct radv_nir_compiler_options
*options
,
2150 struct radv_shader_info
*info
);
2152 void radv_nir_shader_info_init(struct radv_shader_info
*info
);
2154 struct radeon_winsys_sem
;
2156 uint64_t radv_get_current_time(void);
2158 static inline uint32_t
2159 si_conv_gl_prim_to_vertices(unsigned gl_prim
)
2162 case 0: /* GL_POINTS */
2164 case 1: /* GL_LINES */
2165 case 3: /* GL_LINE_STRIP */
2167 case 4: /* GL_TRIANGLES */
2168 case 5: /* GL_TRIANGLE_STRIP */
2170 case 0xA: /* GL_LINE_STRIP_ADJACENCY_ARB */
2172 case 0xc: /* GL_TRIANGLES_ADJACENCY_ARB */
2174 case 7: /* GL_QUADS */
2175 return V_028A6C_OUTPRIM_TYPE_TRISTRIP
;
2182 #define RADV_DEFINE_HANDLE_CASTS(__radv_type, __VkType) \
2184 static inline struct __radv_type * \
2185 __radv_type ## _from_handle(__VkType _handle) \
2187 return (struct __radv_type *) _handle; \
2190 static inline __VkType \
2191 __radv_type ## _to_handle(struct __radv_type *_obj) \
2193 return (__VkType) _obj; \
2196 #define RADV_DEFINE_NONDISP_HANDLE_CASTS(__radv_type, __VkType) \
2198 static inline struct __radv_type * \
2199 __radv_type ## _from_handle(__VkType _handle) \
2201 return (struct __radv_type *)(uintptr_t) _handle; \
2204 static inline __VkType \
2205 __radv_type ## _to_handle(struct __radv_type *_obj) \
2207 return (__VkType)(uintptr_t) _obj; \
2210 #define RADV_FROM_HANDLE(__radv_type, __name, __handle) \
2211 struct __radv_type *__name = __radv_type ## _from_handle(__handle)
2213 RADV_DEFINE_HANDLE_CASTS(radv_cmd_buffer
, VkCommandBuffer
)
2214 RADV_DEFINE_HANDLE_CASTS(radv_device
, VkDevice
)
2215 RADV_DEFINE_HANDLE_CASTS(radv_instance
, VkInstance
)
2216 RADV_DEFINE_HANDLE_CASTS(radv_physical_device
, VkPhysicalDevice
)
2217 RADV_DEFINE_HANDLE_CASTS(radv_queue
, VkQueue
)
2219 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_cmd_pool
, VkCommandPool
)
2220 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer
, VkBuffer
)
2221 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer_view
, VkBufferView
)
2222 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_pool
, VkDescriptorPool
)
2223 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set
, VkDescriptorSet
)
2224 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set_layout
, VkDescriptorSetLayout
)
2225 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_update_template
, VkDescriptorUpdateTemplate
)
2226 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_device_memory
, VkDeviceMemory
)
2227 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_fence
, VkFence
)
2228 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_event
, VkEvent
)
2229 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_framebuffer
, VkFramebuffer
)
2230 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image
, VkImage
)
2231 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image_view
, VkImageView
);
2232 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_cache
, VkPipelineCache
)
2233 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline
, VkPipeline
)
2234 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_layout
, VkPipelineLayout
)
2235 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_query_pool
, VkQueryPool
)
2236 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_render_pass
, VkRenderPass
)
2237 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_sampler
, VkSampler
)
2238 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_sampler_ycbcr_conversion
, VkSamplerYcbcrConversion
)
2239 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_shader_module
, VkShaderModule
)
2240 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_semaphore
, VkSemaphore
)
2242 #endif /* RADV_PRIVATE_H */