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
42 #define VG(x) ((void)0)
45 #include "c11/threads.h"
50 #include "compiler/shader_enums.h"
51 #include "util/bitscan.h"
52 #include "util/list.h"
53 #include "util/macros.h"
54 #include "util/rwlock.h"
55 #include "util/xmlconfig.h"
57 #include "vk_buffer.h"
58 #include "vk_command_buffer.h"
59 #include "vk_command_pool.h"
60 #include "vk_debug_report.h"
61 #include "vk_device.h"
62 #include "vk_format.h"
63 #include "vk_instance.h"
65 #include "vk_physical_device.h"
66 #include "vk_shader_module.h"
70 #include "vk_framebuffer.h"
72 #include "ac_binary.h"
73 #include "ac_gpu_info.h"
74 #include "ac_shader_util.h"
77 #include "ac_surface.h"
78 #include "radv_constants.h"
79 #include "radv_descriptor_set.h"
80 #include "radv_radeon_winsys.h"
81 #include "radv_shader.h"
82 #include "radv_shader_args.h"
85 #include "radix_sort/radix_sort_vk_devaddr.h"
87 /* Pre-declarations needed for WSI entrypoints */
90 typedef struct xcb_connection_t xcb_connection_t;
91 typedef uint32_t xcb_visualid_t;
92 typedef uint32_t xcb_window_t;
94 #include <vulkan/vk_android_native_buffer.h>
95 #include <vulkan/vk_icd.h>
96 #include <vulkan/vulkan.h>
97 #include <vulkan/vulkan_android.h>
99 #include "radv_entrypoints.h"
101 #include "wsi_common.h"
108 /* Helper to determine if we should compile
109 * any of the Android AHB support.
111 * To actually enable the ext we also need
112 * the necessary kernel support.
114 #if defined(ANDROID) && ANDROID_API_LEVEL >= 26
115 #define RADV_SUPPORT_ANDROID_HARDWARE_BUFFER 1
116 #include <vndk/hardware_buffer.h>
118 #define RADV_SUPPORT_ANDROID_HARDWARE_BUFFER 0
122 #define RADV_SUPPORT_CALIBRATED_TIMESTAMPS 0
124 #define RADV_SUPPORT_CALIBRATED_TIMESTAMPS 1
128 #define radv_printflike(a, b)
130 #define radv_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
133 static inline uint32_t
134 align_u32(uint32_t v, uint32_t a)
136 assert(a != 0 && a == (a & -a));
137 return (v + a - 1) & ~(a - 1);
140 static inline uint32_t
141 align_u32_npot(uint32_t v, uint32_t a)
143 return (v + a - 1) / a * a;
146 static inline uint64_t
147 align_u64(uint64_t v, uint64_t a)
149 assert(a != 0 && a == (a & -a));
150 return (v + a - 1) & ~(a - 1);
153 static inline int32_t
154 align_i32(int32_t v, int32_t a)
156 assert(a != 0 && a == (a & -a));
157 return (v + a - 1) & ~(a - 1);
160 /** Alignment must be a power of 2. */
162 radv_is_aligned(uintmax_t n, uintmax_t a)
164 assert(a == (a & -a));
165 return (n & (a - 1)) == 0;
168 static inline uint32_t
169 round_up_u32(uint32_t v, uint32_t a)
171 return (v + a - 1) / a;
174 static inline uint64_t
175 round_up_u64(uint64_t v, uint64_t a)
177 return (v + a - 1) / a;
180 static inline uint32_t
181 radv_minify(uint32_t n, uint32_t levels)
183 if (unlikely(n == 0))
186 return MAX2(n >> levels, 1);
189 radv_clamp_f(float f, float min, float max)
202 radv_clear_mask(uint32_t *inout_mask, uint32_t clear_mask)
204 if (*inout_mask & clear_mask) {
205 *inout_mask &= ~clear_mask;
213 radv_float_to_sfixed(float value, unsigned frac_bits)
215 return value * (1 << frac_bits);
218 static inline unsigned int
219 radv_float_to_ufixed(float value, unsigned frac_bits)
221 return value * (1 << frac_bits);
224 /* Whenever we generate an error, pass it through this function. Useful for
225 * debugging, where we can break on it. Only call at error site, not when
226 * propagating errors. Might be useful to plug in a stack trace here.
229 struct radv_image_view;
230 struct radv_instance;
232 /* A non-fatal assert. Useful for debugging. */
234 #define radv_assert(x) \
238 #define radv_assert(x) \
240 if (unlikely(!(x))) \
241 fprintf(stderr, "%s:%d ASSERT: %s\n", __FILE__, __LINE__, #x); \
245 int radv_get_instance_entrypoint_index(const char *name);
246 int radv_get_device_entrypoint_index(const char *name);
247 int radv_get_physical_device_entrypoint_index(const char *name);
249 const char *radv_get_instance_entry_name(int index);
250 const char *radv_get_physical_device_entry_name(int index);
251 const char *radv_get_device_entry_name(int index);
254 enum radv_queue_family {
258 RADV_MAX_QUEUE_FAMILIES,
259 RADV_QUEUE_FOREIGN = RADV_MAX_QUEUE_FAMILIES,
263 struct radv_perfcounter_desc;
265 struct radv_physical_device {
266 struct vk_physical_device vk;
268 /* Link in radv_instance::physical_devices */
269 struct list_head link;
271 struct radv_instance *instance;
273 struct radeon_winsys *ws;
274 struct radeon_info rad_info;
275 char name[VK_MAX_PHYSICAL_DEVICE_NAME_SIZE];
276 char marketing_name[VK_MAX_PHYSICAL_DEVICE_NAME_SIZE];
277 uint8_t driver_uuid[VK_UUID_SIZE];
278 uint8_t device_uuid[VK_UUID_SIZE];
279 uint8_t cache_uuid[VK_UUID_SIZE];
283 struct wsi_device wsi_device;
285 bool out_of_order_rast_allowed;
287 /* Whether DCC should be enabled for MSAA textures. */
288 bool dcc_msaa_allowed;
290 /* Whether to enable NGG. */
293 /* Whether to enable NGG culling. */
294 bool use_ngg_culling;
296 /* Whether to enable NGG streamout. */
297 bool use_ngg_streamout;
299 /* Number of threads per wave. */
300 uint8_t ps_wave_size;
301 uint8_t cs_wave_size;
302 uint8_t ge_wave_size;
303 uint8_t rt_wave_size;
305 /* Whether to use the LLVM compiler backend */
308 /* Whether to emulate ETC2 image support on HW without support. */
311 /* This is the drivers on-disk cache used as a fallback as opposed to
312 * the pipeline cache defined by apps.
314 struct disk_cache *disk_cache;
316 VkPhysicalDeviceMemoryProperties memory_properties;
317 enum radeon_bo_domain memory_domains[VK_MAX_MEMORY_TYPES];
318 enum radeon_bo_flag memory_flags[VK_MAX_MEMORY_TYPES];
323 drmPciBusInfo bus_info;
329 nir_shader_compiler_options nir_options[MESA_VULKAN_SHADER_STAGES];
331 enum radv_queue_family vk_queue_to_radv[RADV_MAX_QUEUE_FAMILIES];
334 uint32_t gs_table_depth;
336 struct ac_hs_info hs;
337 struct ac_task_info task_info;
339 /* Performance counters. */
340 struct ac_perfcounters ac_perfcounters;
342 uint32_t num_perfcounters;
343 struct radv_perfcounter_desc *perfcounters;
346 struct radv_instance {
347 struct vk_instance vk;
349 VkAllocationCallbacks alloc;
351 uint64_t debug_flags;
352 uint64_t perftest_flags;
354 bool physical_devices_enumerated;
355 struct list_head physical_devices;
357 struct driOptionCache dri_options;
358 struct driOptionCache available_dri_options;
361 * Workarounds for game bugs.
363 bool enable_mrt_output_nan_fixup;
364 bool disable_tc_compat_htile_in_general;
365 bool disable_shrink_image_store;
366 bool absolute_depth_bias;
367 bool report_apu_as_dgpu;
368 bool disable_aniso_single_level;
370 bool disable_sinking_load_input_fs;
373 VkResult radv_init_wsi(struct radv_physical_device *physical_device);
374 void radv_finish_wsi(struct radv_physical_device *physical_device);
378 struct radv_pipeline_cache {
379 struct vk_object_base base;
380 struct radv_device *device;
382 VkPipelineCacheCreateFlags flags;
386 uint32_t kernel_count;
387 struct cache_entry **hash_table;
390 VkAllocationCallbacks alloc;
393 struct radv_shader_binary;
395 struct radv_pipeline_shader_stack_size;
397 void radv_pipeline_cache_init(struct radv_pipeline_cache *cache, struct radv_device *device);
398 void radv_pipeline_cache_finish(struct radv_pipeline_cache *cache);
399 bool radv_pipeline_cache_load(struct radv_pipeline_cache *cache, const void *data, size_t size);
401 bool radv_create_shaders_from_pipeline_cache(
402 struct radv_device *device, struct radv_pipeline_cache *cache, const unsigned char *sha1,
403 struct radv_pipeline *pipeline, struct radv_pipeline_shader_stack_size **stack_sizes,
404 uint32_t *num_stack_sizes, bool *found_in_application_cache);
406 void radv_pipeline_cache_insert_shaders(
407 struct radv_device *device, struct radv_pipeline_cache *cache, const unsigned char *sha1,
408 struct radv_pipeline *pipeline, struct radv_shader_binary *const *binaries,
409 const struct radv_pipeline_shader_stack_size *stack_sizes, uint32_t num_stack_sizes);
411 VkResult radv_upload_shaders(struct radv_device *device, struct radv_pipeline *pipeline,
412 struct radv_shader_binary **binaries,
413 struct radv_shader_binary *gs_copy_binary);
415 enum radv_blit_ds_layout {
416 RADV_BLIT_DS_LAYOUT_TILE_ENABLE,
417 RADV_BLIT_DS_LAYOUT_TILE_DISABLE,
418 RADV_BLIT_DS_LAYOUT_COUNT,
421 static inline enum radv_blit_ds_layout
422 radv_meta_blit_ds_to_type(VkImageLayout layout)
424 return (layout == VK_IMAGE_LAYOUT_GENERAL) ? RADV_BLIT_DS_LAYOUT_TILE_DISABLE
425 : RADV_BLIT_DS_LAYOUT_TILE_ENABLE;
428 static inline VkImageLayout
429 radv_meta_blit_ds_to_layout(enum radv_blit_ds_layout ds_layout)
431 return ds_layout == RADV_BLIT_DS_LAYOUT_TILE_ENABLE ? VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
432 : VK_IMAGE_LAYOUT_GENERAL;
435 enum radv_meta_dst_layout {
436 RADV_META_DST_LAYOUT_GENERAL,
437 RADV_META_DST_LAYOUT_OPTIMAL,
438 RADV_META_DST_LAYOUT_COUNT,
441 static inline enum radv_meta_dst_layout
442 radv_meta_dst_layout_from_layout(VkImageLayout layout)
444 return (layout == VK_IMAGE_LAYOUT_GENERAL) ? RADV_META_DST_LAYOUT_GENERAL
445 : RADV_META_DST_LAYOUT_OPTIMAL;
448 static inline VkImageLayout
449 radv_meta_dst_layout_to_layout(enum radv_meta_dst_layout layout)
451 return layout == RADV_META_DST_LAYOUT_OPTIMAL ? VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
452 : VK_IMAGE_LAYOUT_GENERAL;
455 struct radv_meta_state {
456 VkAllocationCallbacks alloc;
458 struct radv_pipeline_cache cache;
461 * For on-demand pipeline creation, makes sure that
462 * only one thread tries to build a pipeline at the same time.
467 * Use array element `i` for images with `2^i` samples.
470 VkPipeline color_pipelines[NUM_META_FS_KEYS];
471 } color_clear[MAX_SAMPLES_LOG2][MAX_RTS];
474 VkPipeline depth_only_pipeline[NUM_DEPTH_CLEAR_PIPELINES];
475 VkPipeline stencil_only_pipeline[NUM_DEPTH_CLEAR_PIPELINES];
476 VkPipeline depthstencil_pipeline[NUM_DEPTH_CLEAR_PIPELINES];
478 VkPipeline depth_only_unrestricted_pipeline[NUM_DEPTH_CLEAR_PIPELINES];
479 VkPipeline stencil_only_unrestricted_pipeline[NUM_DEPTH_CLEAR_PIPELINES];
480 VkPipeline depthstencil_unrestricted_pipeline[NUM_DEPTH_CLEAR_PIPELINES];
481 } ds_clear[MAX_SAMPLES_LOG2];
483 VkPipelineLayout clear_color_p_layout;
484 VkPipelineLayout clear_depth_p_layout;
485 VkPipelineLayout clear_depth_unrestricted_p_layout;
487 /* Optimized compute fast HTILE clear for stencil or depth only. */
488 VkPipeline clear_htile_mask_pipeline;
489 VkPipelineLayout clear_htile_mask_p_layout;
490 VkDescriptorSetLayout clear_htile_mask_ds_layout;
492 /* Copy VRS into HTILE. */
493 VkPipeline copy_vrs_htile_pipeline;
494 VkPipelineLayout copy_vrs_htile_p_layout;
495 VkDescriptorSetLayout copy_vrs_htile_ds_layout;
497 /* Clear DCC with comp-to-single. */
498 VkPipeline clear_dcc_comp_to_single_pipeline[2]; /* 0: 1x, 1: 2x/4x/8x */
499 VkPipelineLayout clear_dcc_comp_to_single_p_layout;
500 VkDescriptorSetLayout clear_dcc_comp_to_single_ds_layout;
503 /** Pipeline that blits from a 1D image. */
504 VkPipeline pipeline_1d_src[NUM_META_FS_KEYS];
506 /** Pipeline that blits from a 2D image. */
507 VkPipeline pipeline_2d_src[NUM_META_FS_KEYS];
509 /** Pipeline that blits from a 3D image. */
510 VkPipeline pipeline_3d_src[NUM_META_FS_KEYS];
512 VkPipeline depth_only_1d_pipeline;
513 VkPipeline depth_only_2d_pipeline;
514 VkPipeline depth_only_3d_pipeline;
516 VkPipeline stencil_only_1d_pipeline;
517 VkPipeline stencil_only_2d_pipeline;
518 VkPipeline stencil_only_3d_pipeline;
519 VkPipelineLayout pipeline_layout;
520 VkDescriptorSetLayout ds_layout;
524 VkPipelineLayout p_layouts[5];
525 VkDescriptorSetLayout ds_layouts[5];
526 VkPipeline pipelines[5][NUM_META_FS_KEYS];
528 VkPipeline depth_only_pipeline[5];
530 VkPipeline stencil_only_pipeline[5];
531 } blit2d[MAX_SAMPLES_LOG2];
534 VkPipelineLayout img_p_layout;
535 VkDescriptorSetLayout img_ds_layout;
537 VkPipeline pipeline_3d;
540 VkPipelineLayout img_p_layout;
541 VkDescriptorSetLayout img_ds_layout;
543 VkPipeline pipeline_3d;
546 VkPipelineLayout img_p_layout;
547 VkDescriptorSetLayout img_ds_layout;
551 VkPipelineLayout img_p_layout;
552 VkDescriptorSetLayout img_ds_layout;
553 VkPipeline pipeline[MAX_SAMPLES_LOG2];
554 VkPipeline pipeline_3d;
557 VkPipelineLayout img_p_layout;
558 VkDescriptorSetLayout img_ds_layout;
562 VkPipelineLayout img_p_layout;
563 VkDescriptorSetLayout img_ds_layout;
564 VkPipeline pipeline[MAX_SAMPLES_LOG2];
565 VkPipeline pipeline_3d;
568 VkPipelineLayout img_p_layout;
569 VkDescriptorSetLayout img_ds_layout;
573 VkPipelineLayout p_layout;
574 VkDescriptorSetLayout ds_layout;
575 VkPipeline pipeline[MAX_SAMPLES_LOG2];
579 VkPipelineLayout p_layout;
580 VkPipeline pipeline[NUM_META_FS_KEYS];
584 VkDescriptorSetLayout ds_layout;
585 VkPipelineLayout p_layout;
588 VkPipeline i_pipeline;
589 VkPipeline srgb_pipeline;
590 } rc[MAX_SAMPLES_LOG2];
592 VkPipeline depth_zero_pipeline;
594 VkPipeline average_pipeline;
595 VkPipeline max_pipeline;
596 VkPipeline min_pipeline;
597 } depth[MAX_SAMPLES_LOG2];
599 VkPipeline stencil_zero_pipeline;
601 VkPipeline max_pipeline;
602 VkPipeline min_pipeline;
603 } stencil[MAX_SAMPLES_LOG2];
607 VkDescriptorSetLayout ds_layout;
608 VkPipelineLayout p_layout;
611 VkPipeline pipeline[NUM_META_FS_KEYS];
612 } rc[MAX_SAMPLES_LOG2];
614 VkPipeline depth_zero_pipeline;
616 VkPipeline average_pipeline;
617 VkPipeline max_pipeline;
618 VkPipeline min_pipeline;
619 } depth[MAX_SAMPLES_LOG2];
621 VkPipeline stencil_zero_pipeline;
623 VkPipeline max_pipeline;
624 VkPipeline min_pipeline;
625 } stencil[MAX_SAMPLES_LOG2];
629 VkPipelineLayout p_layout;
630 VkPipeline decompress_pipeline;
631 VkPipeline resummarize_pipeline;
632 } depth_decomp[MAX_SAMPLES_LOG2];
634 VkDescriptorSetLayout expand_depth_stencil_compute_ds_layout;
635 VkPipelineLayout expand_depth_stencil_compute_p_layout;
636 VkPipeline expand_depth_stencil_compute_pipeline;
639 VkPipelineLayout p_layout;
640 VkPipeline cmask_eliminate_pipeline;
641 VkPipeline fmask_decompress_pipeline;
642 VkPipeline dcc_decompress_pipeline;
644 VkDescriptorSetLayout dcc_decompress_compute_ds_layout;
645 VkPipelineLayout dcc_decompress_compute_p_layout;
646 VkPipeline dcc_decompress_compute_pipeline;
650 VkPipelineLayout fill_p_layout;
651 VkPipelineLayout copy_p_layout;
652 VkPipeline fill_pipeline;
653 VkPipeline copy_pipeline;
657 VkDescriptorSetLayout ds_layout;
658 VkPipelineLayout p_layout;
659 VkPipeline occlusion_query_pipeline;
660 VkPipeline pipeline_statistics_query_pipeline;
661 VkPipeline tfb_query_pipeline;
662 VkPipeline timestamp_query_pipeline;
663 VkPipeline pg_query_pipeline;
667 VkDescriptorSetLayout ds_layout;
668 VkPipelineLayout p_layout;
669 VkPipeline pipeline[MAX_SAMPLES_LOG2];
673 VkDescriptorSetLayout ds_layout;
674 VkPipelineLayout p_layout;
675 VkPipeline pipeline[32];
679 VkPipelineLayout leaf_p_layout;
680 VkPipeline leaf_pipeline;
681 VkPipelineLayout morton_p_layout;
682 VkPipeline morton_pipeline;
683 VkPipelineLayout internal_p_layout;
684 VkPipeline internal_pipeline;
685 VkPipelineLayout copy_p_layout;
686 VkPipeline copy_pipeline;
688 struct radix_sort_vk *radix_sort;
689 struct radix_sort_vk_sort_devaddr_info radix_sort_info;
690 } accel_struct_build;
693 VkDescriptorSetLayout ds_layout;
694 VkPipelineLayout p_layout;
699 #define RADV_NUM_HW_CTX (RADEON_CTX_PRIORITY_REALTIME + 1)
701 struct radv_deferred_queue_submission;
703 static inline enum radv_queue_family
704 vk_queue_to_radv(const struct radv_physical_device *phys_dev, int queue_family_index)
706 if (queue_family_index == VK_QUEUE_FAMILY_EXTERNAL ||
707 queue_family_index == VK_QUEUE_FAMILY_FOREIGN_EXT)
708 return RADV_QUEUE_FOREIGN;
709 if (queue_family_index == VK_QUEUE_FAMILY_IGNORED)
710 return RADV_QUEUE_IGNORED;
712 assert(queue_family_index < RADV_MAX_QUEUE_FAMILIES);
713 return phys_dev->vk_queue_to_radv[queue_family_index];
716 enum amd_ip_type radv_queue_family_to_ring(struct radv_physical_device *physical_device,
717 enum radv_queue_family f);
719 struct radv_queue_ring_info {
720 uint32_t scratch_size_per_wave;
721 uint32_t scratch_waves;
722 uint32_t compute_scratch_size_per_wave;
723 uint32_t compute_scratch_waves;
724 uint32_t esgs_ring_size;
725 uint32_t gsvs_ring_size;
728 bool mesh_scratch_ring;
731 bool sample_positions;
734 struct radv_queue_state {
735 enum radv_queue_family qf;
736 struct radv_queue_ring_info ring_info;
738 struct radeon_winsys_bo *scratch_bo;
739 struct radeon_winsys_bo *descriptor_bo;
740 struct radeon_winsys_bo *compute_scratch_bo;
741 struct radeon_winsys_bo *esgs_ring_bo;
742 struct radeon_winsys_bo *gsvs_ring_bo;
743 struct radeon_winsys_bo *tess_rings_bo;
744 struct radeon_winsys_bo *task_rings_bo;
745 struct radeon_winsys_bo *mesh_scratch_ring_bo;
746 struct radeon_winsys_bo *gds_bo;
747 struct radeon_winsys_bo *gds_oa_bo;
749 struct radeon_cmdbuf *initial_preamble_cs;
750 struct radeon_cmdbuf *initial_full_flush_preamble_cs;
751 struct radeon_cmdbuf *continue_preamble_cs;
756 struct radv_device *device;
757 struct radeon_winsys_ctx *hw_ctx;
758 enum radeon_ctx_priority priority;
759 struct radv_queue_state state;
762 #define RADV_BORDER_COLOR_COUNT 4096
763 #define RADV_BORDER_COLOR_BUFFER_SIZE (sizeof(VkClearColorValue) * RADV_BORDER_COLOR_COUNT)
765 struct radv_device_border_color_data {
766 bool used[RADV_BORDER_COLOR_COUNT];
768 struct radeon_winsys_bo *bo;
769 VkClearColorValue *colors_gpu_ptr;
771 /* Mutex is required to guarantee vkCreateSampler thread safety
772 * given that we are writing to a buffer and checking color occupation */
776 enum radv_force_vrs {
777 RADV_FORCE_VRS_1x1 = 0,
783 struct radv_notifier {
793 struct radv_instance *instance;
794 struct radeon_winsys *ws;
796 struct radeon_winsys_ctx *hw_ctx[RADV_NUM_HW_CTX];
797 struct radv_meta_state meta_state;
799 struct radv_queue *queues[RADV_MAX_QUEUE_FAMILIES];
800 int queue_count[RADV_MAX_QUEUE_FAMILIES];
803 uint32_t scratch_waves;
804 uint32_t dispatch_initiator;
806 /* MSAA sample locations.
807 * The first index is the sample index.
808 * The second index is the coordinate: X, Y. */
809 float sample_locations_1x[1][2];
810 float sample_locations_2x[2][2];
811 float sample_locations_4x[4][2];
812 float sample_locations_8x[8][2];
815 uint32_t gfx_init_size_dw;
816 struct radeon_winsys_bo *gfx_init;
818 struct radeon_winsys_bo *trace_bo;
819 uint32_t *trace_id_ptr;
821 /* Whether to keep shader debug info, for debugging. */
822 bool keep_shader_info;
824 struct radv_physical_device *physical_device;
826 /* Backup in-memory cache to be used if the app doesn't provide one */
827 struct radv_pipeline_cache *mem_cache;
830 * use different counters so MSAA MRTs get consecutive surface indices,
831 * even if MASK is allocated in between.
833 uint32_t image_mrt_offset_counter;
834 uint32_t fmask_mrt_offset_counter;
836 struct list_head shader_arenas;
837 unsigned shader_arena_shift;
838 uint8_t shader_free_list_mask;
839 struct list_head shader_free_lists[RADV_SHADER_ALLOC_NUM_FREE_LISTS];
840 struct list_head shader_block_obj_pool;
841 mtx_t shader_arena_mutex;
843 /* For detecting VM faults reported by dmesg. */
844 uint64_t dmesg_timestamp;
846 /* Whether the app has enabled the robustBufferAccess/robustBufferAccess2 features. */
847 bool robust_buffer_access;
848 bool robust_buffer_access2;
850 /* Whether to inline the compute dispatch size in user sgprs. */
851 bool load_grid_size_from_user_sgpr;
853 /* Whether the driver uses a global BO list. */
854 bool use_global_bo_list;
856 /* Whether attachment VRS is enabled. */
857 bool attachment_vrs_enabled;
859 /* Whether shader image 32-bit float atomics are enabled. */
860 bool image_float32_atomics;
862 /* Whether 2D views of 3D image is enabled. */
863 bool image_2d_view_of_3d;
865 /* Whether primitives generated query features are enabled. */
866 bool primitives_generated_query;
868 /* Whether anisotropy is forced with RADV_TEX_ANISO (-1 is disabled). */
871 struct radv_device_border_color_data border_color_data;
874 struct ac_thread_trace_data thread_trace;
877 struct ac_spm_trace_data spm_trace;
880 struct radv_trap_handler_shader *trap_handler_shader;
881 struct radeon_winsys_bo *tma_bo; /* Trap Memory Address */
884 /* Overallocation. */
885 bool overallocation_disallowed;
886 uint64_t allocated_memory_size[VK_MAX_MEMORY_HEAPS];
887 mtx_t overallocation_mutex;
889 /* RADV_FORCE_VRS. */
890 struct radv_notifier notifier;
891 enum radv_force_vrs force_vrs;
893 /* Depth image for VRS when not bound by the app. */
895 struct radv_image *image;
896 struct radv_buffer *buffer; /* HTILE */
897 struct radv_device_memory *mem;
900 struct u_rwlock vs_prologs_lock;
901 struct hash_table *vs_prologs;
903 /* Prime blit sdma queue */
904 struct radv_queue *private_sdma_queue;
906 struct radv_shader_part *simple_vs_prologs[MAX_VERTEX_ATTRIBS];
907 struct radv_shader_part *instance_rate_vs_prologs[816];
909 simple_mtx_t trace_mtx;
911 /* Whether per-vertex VRS is forced. */
912 bool force_vrs_enabled;
914 /* Whether shaders created through application entrypoints are considered internal. */
915 bool app_shaders_internal;
917 simple_mtx_t pstate_mtx;
920 /* BO to contain some performance counter helpers:
921 * - A lock for profiling cmdbuffers.
922 * - a temporary fence for the end query synchronization.
923 * - the pass to use for profiling. (as an array of bools)
925 struct radeon_winsys_bo *perf_counter_bo;
927 /* Interleaved lock/unlock commandbuffers for perfcounter passes. */
928 struct radeon_cmdbuf **perf_counter_lock_cs;
931 bool radv_device_acquire_performance_counters(struct radv_device *device);
932 void radv_device_release_performance_counters(struct radv_device *device);
934 struct radv_device_memory {
935 struct vk_object_base base;
936 struct radeon_winsys_bo *bo;
937 /* for dedicated allocations */
938 struct radv_image *image;
939 struct radv_buffer *buffer;
945 #if RADV_SUPPORT_ANDROID_HARDWARE_BUFFER
946 struct AHardwareBuffer *android_hardware_buffer;
950 void radv_device_memory_init(struct radv_device_memory *mem, struct radv_device *device,
951 struct radeon_winsys_bo *bo);
952 void radv_device_memory_finish(struct radv_device_memory *mem);
954 struct radv_descriptor_range {
959 struct radv_descriptor_set_header {
960 struct vk_object_base base;
961 struct radv_descriptor_set_layout *layout;
963 uint32_t buffer_count;
965 struct radeon_winsys_bo *bo;
967 uint32_t *mapped_ptr;
968 struct radv_descriptor_range *dynamic_descriptors;
971 struct radv_descriptor_set {
972 struct radv_descriptor_set_header header;
974 struct radeon_winsys_bo *descriptors[];
977 struct radv_push_descriptor_set {
978 struct radv_descriptor_set_header set;
982 struct radv_descriptor_pool_entry {
985 struct radv_descriptor_set *set;
988 struct radv_descriptor_pool {
989 struct vk_object_base base;
990 struct radeon_winsys_bo *bo;
993 uint64_t current_offset;
996 uint8_t *host_memory_base;
997 uint8_t *host_memory_ptr;
998 uint8_t *host_memory_end;
1000 uint32_t entry_count;
1001 uint32_t max_entry_count;
1002 struct radv_descriptor_pool_entry entries[0];
1005 struct radv_descriptor_update_template_entry {
1006 VkDescriptorType descriptor_type;
1008 /* The number of descriptors to update */
1009 uint32_t descriptor_count;
1011 /* Into mapped_ptr or dynamic_descriptors, in units of the respective array */
1012 uint32_t dst_offset;
1014 /* In dwords. Not valid/used for dynamic descriptors */
1015 uint32_t dst_stride;
1017 uint32_t buffer_offset;
1019 /* Only valid for combined image samplers and samplers */
1020 uint8_t has_sampler;
1021 uint8_t sampler_offset;
1027 /* For push descriptors */
1028 const uint32_t *immutable_samplers;
1031 struct radv_descriptor_update_template {
1032 struct vk_object_base base;
1033 uint32_t entry_count;
1034 VkPipelineBindPoint bind_point;
1035 struct radv_descriptor_update_template_entry entry[0];
1038 void radv_descriptor_set_layout_destroy(struct radv_device *device,
1039 struct radv_descriptor_set_layout *set_layout);
1042 radv_descriptor_set_layout_ref(struct radv_descriptor_set_layout *set_layout)
1044 assert(set_layout && set_layout->ref_cnt >= 1);
1045 p_atomic_inc(&set_layout->ref_cnt);
1049 radv_descriptor_set_layout_unref(struct radv_device *device,
1050 struct radv_descriptor_set_layout *set_layout)
1052 assert(set_layout && set_layout->ref_cnt >= 1);
1053 if (p_atomic_dec_zero(&set_layout->ref_cnt))
1054 radv_descriptor_set_layout_destroy(device, set_layout);
1057 struct radv_buffer {
1058 struct vk_buffer vk;
1060 /* Set when bound */
1061 struct radeon_winsys_bo *bo;
1062 VkDeviceSize offset;
1065 void radv_buffer_init(struct radv_buffer *buffer, struct radv_device *device,
1066 struct radeon_winsys_bo *bo, uint64_t size, uint64_t offset);
1067 void radv_buffer_finish(struct radv_buffer *buffer);
1069 enum radv_dynamic_state_bits {
1070 RADV_DYNAMIC_VIEWPORT = 1ull << 0,
1071 RADV_DYNAMIC_SCISSOR = 1ull << 1,
1072 RADV_DYNAMIC_LINE_WIDTH = 1ull << 2,
1073 RADV_DYNAMIC_DEPTH_BIAS = 1ull << 3,
1074 RADV_DYNAMIC_BLEND_CONSTANTS = 1ull << 4,
1075 RADV_DYNAMIC_DEPTH_BOUNDS = 1ull << 5,
1076 RADV_DYNAMIC_STENCIL_COMPARE_MASK = 1ull << 6,
1077 RADV_DYNAMIC_STENCIL_WRITE_MASK = 1ull << 7,
1078 RADV_DYNAMIC_STENCIL_REFERENCE = 1ull << 8,
1079 RADV_DYNAMIC_DISCARD_RECTANGLE = 1ull << 9,
1080 RADV_DYNAMIC_SAMPLE_LOCATIONS = 1ull << 10,
1081 RADV_DYNAMIC_LINE_STIPPLE = 1ull << 11,
1082 RADV_DYNAMIC_CULL_MODE = 1ull << 12,
1083 RADV_DYNAMIC_FRONT_FACE = 1ull << 13,
1084 RADV_DYNAMIC_PRIMITIVE_TOPOLOGY = 1ull << 14,
1085 RADV_DYNAMIC_DEPTH_TEST_ENABLE = 1ull << 15,
1086 RADV_DYNAMIC_DEPTH_WRITE_ENABLE = 1ull << 16,
1087 RADV_DYNAMIC_DEPTH_COMPARE_OP = 1ull << 17,
1088 RADV_DYNAMIC_DEPTH_BOUNDS_TEST_ENABLE = 1ull << 18,
1089 RADV_DYNAMIC_STENCIL_TEST_ENABLE = 1ull << 19,
1090 RADV_DYNAMIC_STENCIL_OP = 1ull << 20,
1091 RADV_DYNAMIC_VERTEX_INPUT_BINDING_STRIDE = 1ull << 21,
1092 RADV_DYNAMIC_FRAGMENT_SHADING_RATE = 1ull << 22,
1093 RADV_DYNAMIC_PATCH_CONTROL_POINTS = 1ull << 23,
1094 RADV_DYNAMIC_RASTERIZER_DISCARD_ENABLE = 1ull << 24,
1095 RADV_DYNAMIC_DEPTH_BIAS_ENABLE = 1ull << 25,
1096 RADV_DYNAMIC_LOGIC_OP = 1ull << 26,
1097 RADV_DYNAMIC_PRIMITIVE_RESTART_ENABLE = 1ull << 27,
1098 RADV_DYNAMIC_COLOR_WRITE_ENABLE = 1ull << 28,
1099 RADV_DYNAMIC_VERTEX_INPUT = 1ull << 29,
1100 RADV_DYNAMIC_ALL = (1ull << 30) - 1,
1103 enum radv_cmd_dirty_bits {
1104 /* Keep the dynamic state dirty bits in sync with
1105 * enum radv_dynamic_state_bits */
1106 RADV_CMD_DIRTY_DYNAMIC_VIEWPORT = 1ull << 0,
1107 RADV_CMD_DIRTY_DYNAMIC_SCISSOR = 1ull << 1,
1108 RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH = 1ull << 2,
1109 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS = 1ull << 3,
1110 RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS = 1ull << 4,
1111 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS = 1ull << 5,
1112 RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK = 1ull << 6,
1113 RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK = 1ull << 7,
1114 RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE = 1ull << 8,
1115 RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE = 1ull << 9,
1116 RADV_CMD_DIRTY_DYNAMIC_SAMPLE_LOCATIONS = 1ull << 10,
1117 RADV_CMD_DIRTY_DYNAMIC_LINE_STIPPLE = 1ull << 11,
1118 RADV_CMD_DIRTY_DYNAMIC_CULL_MODE = 1ull << 12,
1119 RADV_CMD_DIRTY_DYNAMIC_FRONT_FACE = 1ull << 13,
1120 RADV_CMD_DIRTY_DYNAMIC_PRIMITIVE_TOPOLOGY = 1ull << 14,
1121 RADV_CMD_DIRTY_DYNAMIC_DEPTH_TEST_ENABLE = 1ull << 15,
1122 RADV_CMD_DIRTY_DYNAMIC_DEPTH_WRITE_ENABLE = 1ull << 16,
1123 RADV_CMD_DIRTY_DYNAMIC_DEPTH_COMPARE_OP = 1ull << 17,
1124 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS_TEST_ENABLE = 1ull << 18,
1125 RADV_CMD_DIRTY_DYNAMIC_STENCIL_TEST_ENABLE = 1ull << 19,
1126 RADV_CMD_DIRTY_DYNAMIC_STENCIL_OP = 1ull << 20,
1127 RADV_CMD_DIRTY_DYNAMIC_VERTEX_INPUT_BINDING_STRIDE = 1ull << 21,
1128 RADV_CMD_DIRTY_DYNAMIC_FRAGMENT_SHADING_RATE = 1ull << 22,
1129 RADV_CMD_DIRTY_DYNAMIC_PATCH_CONTROL_POINTS = 1ull << 23,
1130 RADV_CMD_DIRTY_DYNAMIC_RASTERIZER_DISCARD_ENABLE = 1ull << 24,
1131 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS_ENABLE = 1ull << 25,
1132 RADV_CMD_DIRTY_DYNAMIC_LOGIC_OP = 1ull << 26,
1133 RADV_CMD_DIRTY_DYNAMIC_PRIMITIVE_RESTART_ENABLE = 1ull << 27,
1134 RADV_CMD_DIRTY_DYNAMIC_COLOR_WRITE_ENABLE = 1ull << 28,
1135 RADV_CMD_DIRTY_DYNAMIC_VERTEX_INPUT = 1ull << 29,
1136 RADV_CMD_DIRTY_DYNAMIC_ALL = (1ull << 30) - 1,
1137 RADV_CMD_DIRTY_PIPELINE = 1ull << 30,
1138 RADV_CMD_DIRTY_INDEX_BUFFER = 1ull << 31,
1139 RADV_CMD_DIRTY_FRAMEBUFFER = 1ull << 32,
1140 RADV_CMD_DIRTY_VERTEX_BUFFER = 1ull << 33,
1141 RADV_CMD_DIRTY_STREAMOUT_BUFFER = 1ull << 34,
1144 enum radv_cmd_flush_bits {
1145 /* Instruction cache. */
1146 RADV_CMD_FLAG_INV_ICACHE = 1 << 0,
1147 /* Scalar L1 cache. */
1148 RADV_CMD_FLAG_INV_SCACHE = 1 << 1,
1149 /* Vector L1 cache. */
1150 RADV_CMD_FLAG_INV_VCACHE = 1 << 2,
1151 /* L2 cache + L2 metadata cache writeback & invalidate.
1152 * GFX6-8: Used by shaders only. GFX9-10: Used by everything. */
1153 RADV_CMD_FLAG_INV_L2 = 1 << 3,
1154 /* L2 writeback (write dirty L2 lines to memory for non-L2 clients).
1155 * Only used for coherency with non-L2 clients like CB, DB, CP on GFX6-8.
1156 * GFX6-7 will do complete invalidation, because the writeback is unsupported. */
1157 RADV_CMD_FLAG_WB_L2 = 1 << 4,
1158 /* Invalidate the metadata cache. To be used when the DCC/HTILE metadata
1159 * changed and we want to read an image from shaders. */
1160 RADV_CMD_FLAG_INV_L2_METADATA = 1 << 5,
1161 /* Framebuffer caches */
1162 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META = 1 << 6,
1163 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META = 1 << 7,
1164 RADV_CMD_FLAG_FLUSH_AND_INV_DB = 1 << 8,
1165 RADV_CMD_FLAG_FLUSH_AND_INV_CB = 1 << 9,
1166 /* Engine synchronization. */
1167 RADV_CMD_FLAG_VS_PARTIAL_FLUSH = 1 << 10,
1168 RADV_CMD_FLAG_PS_PARTIAL_FLUSH = 1 << 11,
1169 RADV_CMD_FLAG_CS_PARTIAL_FLUSH = 1 << 12,
1170 RADV_CMD_FLAG_VGT_FLUSH = 1 << 13,
1171 /* Pipeline query controls. */
1172 RADV_CMD_FLAG_START_PIPELINE_STATS = 1 << 14,
1173 RADV_CMD_FLAG_STOP_PIPELINE_STATS = 1 << 15,
1174 RADV_CMD_FLAG_VGT_STREAMOUT_SYNC = 1 << 16,
1176 RADV_CMD_FLUSH_AND_INV_FRAMEBUFFER =
1177 (RADV_CMD_FLAG_FLUSH_AND_INV_CB | RADV_CMD_FLAG_FLUSH_AND_INV_CB_META |
1178 RADV_CMD_FLAG_FLUSH_AND_INV_DB | RADV_CMD_FLAG_FLUSH_AND_INV_DB_META)
1181 enum radv_nggc_settings {
1183 radv_nggc_front_face = 1 << 0,
1184 radv_nggc_back_face = 1 << 1,
1185 radv_nggc_face_is_ccw = 1 << 2,
1186 radv_nggc_small_primitives = 1 << 3,
1189 struct radv_vertex_binding {
1190 VkDeviceSize offset;
1192 VkDeviceSize stride;
1195 struct radv_streamout_binding {
1196 struct radv_buffer *buffer;
1197 VkDeviceSize offset;
1201 struct radv_streamout_state {
1202 /* Mask of bound streamout buffers. */
1203 uint8_t enabled_mask;
1205 /* State of VGT_STRMOUT_BUFFER_(CONFIG|END) */
1206 uint32_t hw_enabled_mask;
1208 /* State of VGT_STRMOUT_(CONFIG|EN) */
1209 bool streamout_enabled;
1212 struct radv_viewport_state {
1214 VkViewport viewports[MAX_VIEWPORTS];
1218 } xform[MAX_VIEWPORTS];
1221 struct radv_scissor_state {
1223 VkRect2D scissors[MAX_SCISSORS];
1226 struct radv_discard_rectangle_state {
1228 VkRect2D rectangles[MAX_DISCARD_RECTANGLES];
1231 struct radv_sample_locations_state {
1232 VkSampleCountFlagBits per_pixel;
1233 VkExtent2D grid_size;
1235 VkSampleLocationEXT locations[MAX_SAMPLE_LOCATIONS];
1238 struct radv_dynamic_state {
1240 * Bitmask of (1ull << VK_DYNAMIC_STATE_*).
1241 * Defines the set of saved dynamic state.
1245 struct radv_viewport_state viewport;
1247 struct radv_scissor_state scissor;
1257 float blend_constants[4];
1267 } stencil_compare_mask;
1272 } stencil_write_mask;
1276 VkStencilOp fail_op;
1277 VkStencilOp pass_op;
1278 VkStencilOp depth_fail_op;
1279 VkCompareOp compare_op;
1283 VkStencilOp fail_op;
1284 VkStencilOp pass_op;
1285 VkStencilOp depth_fail_op;
1286 VkCompareOp compare_op;
1293 } stencil_reference;
1295 struct radv_discard_rectangle_state discard_rectangle;
1297 struct radv_sample_locations_state sample_location;
1304 VkCullModeFlags cull_mode;
1305 VkFrontFace front_face;
1306 unsigned primitive_topology;
1308 bool depth_test_enable;
1309 bool depth_write_enable;
1310 VkCompareOp depth_compare_op;
1311 bool depth_bounds_test_enable;
1312 bool stencil_test_enable;
1316 VkFragmentShadingRateCombinerOpKHR combiner_ops[2];
1317 } fragment_shading_rate;
1319 bool depth_bias_enable;
1320 bool primitive_restart_enable;
1321 bool rasterizer_discard_enable;
1325 uint32_t color_write_enable;
1328 extern const struct radv_dynamic_state default_dynamic_state;
1330 const char *radv_get_debug_option_name(int id);
1332 const char *radv_get_perftest_option_name(int id);
1334 int radv_get_int_debug_option(const char *name, int default_value);
1336 struct radv_color_buffer_info {
1337 uint64_t cb_color_base;
1338 uint64_t cb_color_cmask;
1339 uint64_t cb_color_fmask;
1340 uint64_t cb_dcc_base;
1341 uint32_t cb_color_slice;
1342 uint32_t cb_color_view;
1343 uint32_t cb_color_info;
1344 uint32_t cb_color_attrib;
1345 uint32_t cb_color_attrib2; /* GFX9 and later */
1346 uint32_t cb_color_attrib3; /* GFX10 and later */
1347 uint32_t cb_dcc_control;
1348 uint32_t cb_color_cmask_slice;
1349 uint32_t cb_color_fmask_slice;
1351 uint32_t cb_color_pitch; // GFX6-GFX8
1352 uint32_t cb_mrt_epitch; // GFX9+
1356 struct radv_ds_buffer_info {
1357 uint64_t db_z_read_base;
1358 uint64_t db_stencil_read_base;
1359 uint64_t db_z_write_base;
1360 uint64_t db_stencil_write_base;
1361 uint64_t db_htile_data_base;
1362 uint32_t db_depth_info;
1364 uint32_t db_stencil_info;
1365 uint32_t db_depth_view;
1366 uint32_t db_depth_size;
1367 uint32_t db_depth_slice;
1368 uint32_t db_htile_surface;
1369 uint32_t pa_su_poly_offset_db_fmt_cntl;
1370 uint32_t db_z_info2; /* GFX9 only */
1371 uint32_t db_stencil_info2; /* GFX9 only */
1374 void radv_initialise_color_surface(struct radv_device *device, struct radv_color_buffer_info *cb,
1375 struct radv_image_view *iview);
1376 void radv_initialise_ds_surface(struct radv_device *device, struct radv_ds_buffer_info *ds,
1377 struct radv_image_view *iview);
1378 void radv_initialise_vrs_surface(struct radv_image *image, struct radv_buffer *htile_buffer,
1379 struct radv_ds_buffer_info *ds);
1382 * Attachment state when recording a renderpass instance.
1384 * The clear value is valid only if there exists a pending clear.
1386 struct radv_attachment_state {
1387 VkImageAspectFlags pending_clear_aspects;
1388 uint32_t cleared_views;
1389 VkClearValue clear_value;
1390 VkImageLayout current_layout;
1391 VkImageLayout current_stencil_layout;
1392 bool current_in_render_loop;
1393 struct radv_sample_locations_state sample_location;
1396 struct radv_color_buffer_info cb;
1397 struct radv_ds_buffer_info ds;
1399 struct radv_image_view *iview;
1402 struct radv_descriptor_state {
1403 struct radv_descriptor_set *sets[MAX_SETS];
1406 struct radv_push_descriptor_set push_set;
1408 uint32_t dynamic_buffers[4 * MAX_DYNAMIC_BUFFERS];
1411 struct radv_subpass_sample_locs_state {
1412 uint32_t subpass_idx;
1413 struct radv_sample_locations_state sample_location;
1416 enum rgp_flush_bits {
1417 RGP_FLUSH_WAIT_ON_EOP_TS = 0x1,
1418 RGP_FLUSH_VS_PARTIAL_FLUSH = 0x2,
1419 RGP_FLUSH_PS_PARTIAL_FLUSH = 0x4,
1420 RGP_FLUSH_CS_PARTIAL_FLUSH = 0x8,
1421 RGP_FLUSH_PFP_SYNC_ME = 0x10,
1422 RGP_FLUSH_SYNC_CP_DMA = 0x20,
1423 RGP_FLUSH_INVAL_VMEM_L0 = 0x40,
1424 RGP_FLUSH_INVAL_ICACHE = 0x80,
1425 RGP_FLUSH_INVAL_SMEM_L0 = 0x100,
1426 RGP_FLUSH_FLUSH_L2 = 0x200,
1427 RGP_FLUSH_INVAL_L2 = 0x400,
1428 RGP_FLUSH_FLUSH_CB = 0x800,
1429 RGP_FLUSH_INVAL_CB = 0x1000,
1430 RGP_FLUSH_FLUSH_DB = 0x2000,
1431 RGP_FLUSH_INVAL_DB = 0x4000,
1432 RGP_FLUSH_INVAL_L1 = 0x8000,
1435 struct radv_cmd_state {
1436 /* Vertex descriptors */
1442 uint32_t prefetch_L2_mask;
1444 struct radv_graphics_pipeline *graphics_pipeline;
1445 struct radv_graphics_pipeline *emitted_graphics_pipeline;
1446 struct radv_compute_pipeline *compute_pipeline;
1447 struct radv_compute_pipeline *emitted_compute_pipeline;
1448 struct radv_compute_pipeline *rt_pipeline; /* emitted = emitted_compute_pipeline */
1449 struct vk_framebuffer *framebuffer;
1450 struct radv_render_pass *pass;
1451 const struct radv_subpass *subpass;
1452 struct radv_dynamic_state dynamic;
1453 struct radv_vs_input_state dynamic_vs_input;
1454 struct radv_attachment_state *attachments;
1455 struct radv_streamout_state streamout;
1456 VkRect2D render_area;
1458 uint32_t num_subpass_sample_locs;
1459 struct radv_subpass_sample_locs_state *subpass_sample_locs;
1462 struct radv_buffer *index_buffer;
1463 uint64_t index_offset;
1464 uint32_t index_type;
1465 uint32_t max_index_count;
1467 int32_t last_index_type;
1469 int32_t last_primitive_reset_en;
1470 uint32_t last_primitive_reset_index;
1471 enum radv_cmd_flush_bits flush_bits;
1472 unsigned active_occlusion_queries;
1473 bool perfect_occlusion_queries_enabled;
1474 unsigned active_pipeline_queries;
1475 unsigned active_pipeline_gds_queries;
1476 bool prims_gen_query_enabled;
1478 uint32_t last_ia_multi_vgt_param;
1480 uint32_t last_num_instances;
1481 uint32_t last_first_instance;
1482 uint32_t last_vertex_offset;
1483 uint32_t last_drawid;
1484 uint32_t last_subpass_color_count;
1486 uint32_t last_sx_ps_downconvert;
1487 uint32_t last_sx_blend_opt_epsilon;
1488 uint32_t last_sx_blend_opt_control;
1490 /* Whether CP DMA is busy/idle. */
1493 /* Whether any images that are not L2 coherent are dirty from the CB. */
1494 bool rb_noncoherent_dirty;
1496 /* Conditional rendering info. */
1497 uint8_t predication_op; /* 32-bit or 64-bit predicate value */
1498 int predication_type; /* -1: disabled, 0: normal, 1: inverted */
1499 uint64_t predication_va;
1501 /* Inheritance info. */
1502 VkQueryPipelineStatisticFlags inherited_pipeline_statistics;
1504 bool context_roll_without_scissor_emitted;
1506 /* SQTT related state. */
1507 uint32_t current_event_type;
1508 uint32_t num_events;
1509 uint32_t num_layout_transitions;
1510 bool pending_sqtt_barrier_end;
1511 enum rgp_flush_bits sqtt_flush_bits;
1513 /* NGG culling state. */
1514 uint32_t last_nggc_settings;
1515 int8_t last_nggc_settings_sgpr_idx;
1516 bool last_nggc_skip;
1518 /* Mesh shading state. */
1521 uint8_t cb_mip[MAX_RTS];
1523 /* Whether DRAW_{INDEX}_INDIRECT_MULTI is emitted. */
1524 bool uses_draw_indirect_multi;
1526 uint32_t rt_stack_size;
1528 struct radv_shader_part *emitted_vs_prolog;
1529 uint32_t *emitted_vs_prolog_key;
1530 uint32_t emitted_vs_prolog_key_hash;
1531 uint32_t vbo_misaligned_mask;
1532 uint32_t vbo_bound_mask;
1534 /* Whether the cmdbuffer owns the current render pass rather than the app. */
1535 bool own_render_pass;
1537 /* Per-vertex VRS state. */
1538 uint32_t last_vrs_rates;
1539 int8_t last_vrs_rates_sgpr_idx;
1541 /* Whether to suspend streamout for internal driver operations. */
1542 bool suspend_streamout;
1544 /* Whether this commandbuffer uses performance counters. */
1545 bool uses_perf_counters;
1548 struct radv_cmd_pool {
1549 struct vk_command_pool vk;
1550 struct list_head cmd_buffers;
1551 struct list_head free_cmd_buffers;
1554 struct radv_cmd_buffer_upload {
1558 struct radeon_winsys_bo *upload_bo;
1559 struct list_head list;
1562 enum radv_cmd_buffer_status {
1563 RADV_CMD_BUFFER_STATUS_INVALID,
1564 RADV_CMD_BUFFER_STATUS_INITIAL,
1565 RADV_CMD_BUFFER_STATUS_RECORDING,
1566 RADV_CMD_BUFFER_STATUS_EXECUTABLE,
1567 RADV_CMD_BUFFER_STATUS_PENDING,
1570 struct radv_cmd_buffer {
1571 struct vk_command_buffer vk;
1573 struct radv_device *device;
1575 struct radv_cmd_pool *pool;
1576 struct list_head pool_link;
1578 VkCommandBufferUsageFlags usage_flags;
1579 enum radv_cmd_buffer_status status;
1580 struct radeon_cmdbuf *cs;
1581 struct radv_cmd_state state;
1582 struct radv_buffer *vertex_binding_buffers[MAX_VBS];
1583 struct radv_vertex_binding vertex_bindings[MAX_VBS];
1584 uint32_t used_vertex_bindings;
1585 struct radv_streamout_binding streamout_bindings[MAX_SO_BUFFERS];
1586 enum radv_queue_family qf;
1588 uint8_t push_constants[MAX_PUSH_CONSTANTS_SIZE];
1589 VkShaderStageFlags push_constant_stages;
1590 struct radv_descriptor_set_header meta_push_descriptors;
1592 struct radv_descriptor_state descriptors[MAX_BIND_POINTS];
1594 struct radv_cmd_buffer_upload upload;
1596 uint32_t scratch_size_per_wave_needed;
1597 uint32_t scratch_waves_wanted;
1598 uint32_t compute_scratch_size_per_wave_needed;
1599 uint32_t compute_scratch_waves_wanted;
1600 uint32_t esgs_ring_size_needed;
1601 uint32_t gsvs_ring_size_needed;
1602 bool tess_rings_needed;
1603 bool task_rings_needed;
1604 bool mesh_scratch_ring_needed;
1605 bool gds_needed; /* for GFX10 streamout and NGG GS queries */
1606 bool gds_oa_needed; /* for GFX10 streamout */
1607 bool sample_positions_needed;
1609 VkResult record_result;
1611 uint64_t gfx9_fence_va;
1612 uint32_t gfx9_fence_idx;
1613 uint64_t gfx9_eop_bug_va;
1616 * Whether a query pool has been resetted and we have to flush caches.
1618 bool pending_reset_query;
1621 * Bitmask of pending active query flushes.
1623 enum radv_cmd_flush_bits active_query_flush_bits;
1627 struct radv_image_view;
1629 bool radv_cmd_buffer_uses_mec(struct radv_cmd_buffer *cmd_buffer);
1631 bool radv_is_streamout_enabled(struct radv_cmd_buffer *cmd_buffer);
1632 void radv_emit_streamout_enable(struct radv_cmd_buffer *cmd_buffer);
1634 void si_emit_graphics(struct radv_device *device, struct radeon_cmdbuf *cs);
1635 void si_emit_compute(struct radv_device *device, struct radeon_cmdbuf *cs);
1637 void cik_create_gfx_config(struct radv_device *device);
1639 void si_write_scissors(struct radeon_cmdbuf *cs, int first, int count, const VkRect2D *scissors,
1640 const VkViewport *viewports, bool can_use_guardband);
1641 uint32_t si_get_ia_multi_vgt_param(struct radv_cmd_buffer *cmd_buffer, bool instanced_draw,
1642 bool indirect_draw, bool count_from_stream_output,
1643 uint32_t draw_vertex_count, unsigned topology,
1644 bool prim_restart_enable);
1645 void si_cs_emit_write_event_eop(struct radeon_cmdbuf *cs, enum amd_gfx_level gfx_level, bool is_mec,
1646 unsigned event, unsigned event_flags, unsigned dst_sel,
1647 unsigned data_sel, uint64_t va, uint32_t new_fence,
1648 uint64_t gfx9_eop_bug_va);
1650 void radv_cp_wait_mem(struct radeon_cmdbuf *cs, uint32_t op, uint64_t va, uint32_t ref,
1652 void si_cs_emit_cache_flush(struct radeon_cmdbuf *cs, enum amd_gfx_level gfx_level,
1653 uint32_t *fence_ptr, uint64_t va, bool is_mec,
1654 enum radv_cmd_flush_bits flush_bits,
1655 enum rgp_flush_bits *sqtt_flush_bits, uint64_t gfx9_eop_bug_va);
1656 void si_emit_cache_flush(struct radv_cmd_buffer *cmd_buffer);
1657 void si_emit_set_predication_state(struct radv_cmd_buffer *cmd_buffer, bool draw_visible,
1658 unsigned pred_op, uint64_t va);
1659 void si_cp_dma_buffer_copy(struct radv_cmd_buffer *cmd_buffer, uint64_t src_va, uint64_t dest_va,
1661 void si_cp_dma_prefetch(struct radv_cmd_buffer *cmd_buffer, uint64_t va, unsigned size);
1662 void si_cp_dma_clear_buffer(struct radv_cmd_buffer *cmd_buffer, uint64_t va, uint64_t size,
1664 void si_cp_dma_wait_for_idle(struct radv_cmd_buffer *cmd_buffer);
1666 void radv_set_db_count_control(struct radv_cmd_buffer *cmd_buffer, bool enable_occlusion_queries);
1668 unsigned radv_instance_rate_prolog_index(unsigned num_attributes, uint32_t instance_rate_inputs);
1669 uint32_t radv_hash_vs_prolog(const void *key_);
1670 bool radv_cmp_vs_prolog(const void *a_, const void *b_);
1672 bool radv_cmd_buffer_upload_alloc(struct radv_cmd_buffer *cmd_buffer, unsigned size,
1673 unsigned *out_offset, void **ptr);
1674 void radv_cmd_buffer_set_subpass(struct radv_cmd_buffer *cmd_buffer,
1675 const struct radv_subpass *subpass);
1676 void radv_cmd_buffer_restore_subpass(struct radv_cmd_buffer *cmd_buffer,
1677 const struct radv_subpass *subpass);
1678 bool radv_cmd_buffer_upload_data(struct radv_cmd_buffer *cmd_buffer, unsigned size,
1679 const void *data, unsigned *out_offset);
1681 void radv_cmd_buffer_clear_subpass(struct radv_cmd_buffer *cmd_buffer);
1682 void radv_cmd_buffer_resolve_subpass(struct radv_cmd_buffer *cmd_buffer);
1683 void radv_cmd_buffer_resolve_subpass_cs(struct radv_cmd_buffer *cmd_buffer);
1684 void radv_depth_stencil_resolve_subpass_cs(struct radv_cmd_buffer *cmd_buffer,
1685 VkImageAspectFlags aspects,
1686 VkResolveModeFlagBits resolve_mode);
1687 void radv_cmd_buffer_resolve_subpass_fs(struct radv_cmd_buffer *cmd_buffer);
1688 void radv_depth_stencil_resolve_subpass_fs(struct radv_cmd_buffer *cmd_buffer,
1689 VkImageAspectFlags aspects,
1690 VkResolveModeFlagBits resolve_mode);
1691 void radv_emit_default_sample_locations(struct radeon_cmdbuf *cs, int nr_samples);
1692 unsigned radv_get_default_max_sample_dist(int log_samples);
1693 void radv_device_init_msaa(struct radv_device *device);
1694 VkResult radv_device_init_vrs_state(struct radv_device *device);
1696 void radv_update_ds_clear_metadata(struct radv_cmd_buffer *cmd_buffer,
1697 const struct radv_image_view *iview,
1698 VkClearDepthStencilValue ds_clear_value,
1699 VkImageAspectFlags aspects);
1701 void radv_update_color_clear_metadata(struct radv_cmd_buffer *cmd_buffer,
1702 const struct radv_image_view *iview, int cb_idx,
1703 uint32_t color_values[2]);
1705 bool radv_image_use_dcc_image_stores(const struct radv_device *device,
1706 const struct radv_image *image);
1707 bool radv_image_use_dcc_predication(const struct radv_device *device,
1708 const struct radv_image *image);
1710 void radv_update_fce_metadata(struct radv_cmd_buffer *cmd_buffer, struct radv_image *image,
1711 const VkImageSubresourceRange *range, bool value);
1713 void radv_update_dcc_metadata(struct radv_cmd_buffer *cmd_buffer, struct radv_image *image,
1714 const VkImageSubresourceRange *range, bool value);
1715 enum radv_cmd_flush_bits radv_src_access_flush(struct radv_cmd_buffer *cmd_buffer,
1716 VkAccessFlags2 src_flags,
1717 const struct radv_image *image);
1718 enum radv_cmd_flush_bits radv_dst_access_flush(struct radv_cmd_buffer *cmd_buffer,
1719 VkAccessFlags2 dst_flags,
1720 const struct radv_image *image);
1721 uint32_t radv_fill_buffer(struct radv_cmd_buffer *cmd_buffer, const struct radv_image *image,
1722 struct radeon_winsys_bo *bo, uint64_t va, uint64_t size, uint32_t value);
1723 void radv_copy_buffer(struct radv_cmd_buffer *cmd_buffer, struct radeon_winsys_bo *src_bo,
1724 struct radeon_winsys_bo *dst_bo, uint64_t src_offset, uint64_t dst_offset,
1727 void radv_cmd_buffer_trace_emit(struct radv_cmd_buffer *cmd_buffer);
1728 bool radv_get_memory_fd(struct radv_device *device, struct radv_device_memory *memory, int *pFD);
1729 void radv_free_memory(struct radv_device *device, const VkAllocationCallbacks *pAllocator,
1730 struct radv_device_memory *mem);
1733 radv_emit_shader_pointer_head(struct radeon_cmdbuf *cs, unsigned sh_offset, unsigned pointer_count,
1734 bool use_32bit_pointers)
1736 radeon_emit(cs, PKT3(PKT3_SET_SH_REG, pointer_count * (use_32bit_pointers ? 1 : 2), 0));
1737 radeon_emit(cs, (sh_offset - SI_SH_REG_OFFSET) >> 2);
1741 radv_emit_shader_pointer_body(struct radv_device *device, struct radeon_cmdbuf *cs, uint64_t va,
1742 bool use_32bit_pointers)
1744 radeon_emit(cs, va);
1746 if (use_32bit_pointers) {
1747 assert(va == 0 || (va >> 32) == device->physical_device->rad_info.address32_hi);
1749 radeon_emit(cs, va >> 32);
1754 radv_emit_shader_pointer(struct radv_device *device, struct radeon_cmdbuf *cs, uint32_t sh_offset,
1755 uint64_t va, bool global)
1757 bool use_32bit_pointers = !global;
1759 radv_emit_shader_pointer_head(cs, sh_offset, 1, use_32bit_pointers);
1760 radv_emit_shader_pointer_body(device, cs, va, use_32bit_pointers);
1763 static inline struct radv_descriptor_state *
1764 radv_get_descriptors_state(struct radv_cmd_buffer *cmd_buffer, VkPipelineBindPoint bind_point)
1766 switch (bind_point) {
1767 case VK_PIPELINE_BIND_POINT_GRAPHICS:
1768 case VK_PIPELINE_BIND_POINT_COMPUTE:
1769 return &cmd_buffer->descriptors[bind_point];
1770 case VK_PIPELINE_BIND_POINT_RAY_TRACING_KHR:
1771 return &cmd_buffer->descriptors[2];
1773 unreachable("Unhandled bind point");
1778 radv_get_viewport_xform(const VkViewport *viewport, float scale[3], float translate[3]);
1781 * Takes x,y,z as exact numbers of invocations, instead of blocks.
1783 * Limitations: Can't call normal dispatch functions without binding or rebinding
1784 * the compute pipeline.
1786 void radv_unaligned_dispatch(struct radv_cmd_buffer *cmd_buffer, uint32_t x, uint32_t y,
1789 void radv_indirect_dispatch(struct radv_cmd_buffer *cmd_buffer, struct radeon_winsys_bo *bo,
1793 struct vk_object_base base;
1794 struct radeon_winsys_bo *bo;
1798 #define RADV_HASH_SHADER_CS_WAVE32 (1 << 1)
1799 #define RADV_HASH_SHADER_PS_WAVE32 (1 << 2)
1800 #define RADV_HASH_SHADER_GE_WAVE32 (1 << 3)
1801 #define RADV_HASH_SHADER_LLVM (1 << 4)
1802 #define RADV_HASH_SHADER_KEEP_STATISTICS (1 << 8)
1803 #define RADV_HASH_SHADER_USE_NGG_CULLING (1 << 13)
1804 #define RADV_HASH_SHADER_ROBUST_BUFFER_ACCESS (1 << 14)
1805 #define RADV_HASH_SHADER_ROBUST_BUFFER_ACCESS2 (1 << 15)
1806 #define RADV_HASH_SHADER_EMULATE_RT (1 << 16)
1807 #define RADV_HASH_SHADER_SPLIT_FMA (1 << 17)
1808 #define RADV_HASH_SHADER_RT_WAVE64 (1 << 18)
1810 struct radv_pipeline_key;
1812 void radv_pipeline_stage_init(const VkPipelineShaderStageCreateInfo *sinfo,
1813 struct radv_pipeline_stage *out_stage, gl_shader_stage stage);
1815 void radv_hash_shaders(unsigned char *hash, const struct radv_pipeline_stage *stages,
1816 const struct radv_pipeline_layout *layout,
1817 const struct radv_pipeline_key *key, uint32_t flags);
1819 void radv_hash_rt_shaders(unsigned char *hash, const VkRayTracingPipelineCreateInfoKHR *pCreateInfo,
1822 uint32_t radv_get_hash_flags(const struct radv_device *device, bool stats);
1824 bool radv_rt_pipeline_has_dynamic_stack_size(const VkRayTracingPipelineCreateInfoKHR *pCreateInfo);
1826 bool radv_enable_rt(const struct radv_physical_device *pdevice, bool rt_pipelines);
1828 bool radv_emulate_rt(const struct radv_physical_device *pdevice);
1831 RADV_RT_STAGE_BITS = (VK_SHADER_STAGE_RAYGEN_BIT_KHR | VK_SHADER_STAGE_ANY_HIT_BIT_KHR |
1832 VK_SHADER_STAGE_CLOSEST_HIT_BIT_KHR | VK_SHADER_STAGE_MISS_BIT_KHR |
1833 VK_SHADER_STAGE_INTERSECTION_BIT_KHR | VK_SHADER_STAGE_CALLABLE_BIT_KHR)
1836 #define RADV_STAGE_MASK ((1 << MESA_VULKAN_SHADER_STAGES) - 1)
1838 #define radv_foreach_stage(stage, stage_bits) \
1839 for (gl_shader_stage stage, __tmp = (gl_shader_stage)((stage_bits)&RADV_STAGE_MASK); \
1840 stage = ffs(__tmp) - 1, __tmp; __tmp &= ~(1 << (stage)))
1842 extern const VkFormat radv_fs_key_format_exemplars[NUM_META_FS_KEYS];
1843 unsigned radv_format_meta_fs_key(struct radv_device *device, VkFormat format);
1845 struct radv_multisample_state {
1847 uint32_t pa_sc_mode_cntl_0;
1848 uint32_t pa_sc_mode_cntl_1;
1849 uint32_t pa_sc_aa_config;
1850 uint32_t pa_sc_aa_mask[2];
1851 unsigned num_samples;
1854 struct radv_vrs_state {
1855 uint32_t pa_cl_vrs_cntl;
1858 struct radv_prim_vertex_count {
1863 struct radv_ia_multi_vgt_param_helpers {
1865 bool partial_es_wave;
1866 uint8_t primgroup_size;
1867 bool ia_switch_on_eoi;
1868 bool partial_vs_wave;
1871 struct radv_binning_state {
1872 uint32_t pa_sc_binner_cntl_0;
1875 #define SI_GS_PER_ES 128
1877 enum radv_pipeline_type {
1878 RADV_PIPELINE_GRAPHICS,
1879 /* Compute pipeline (incl raytracing pipeline) */
1880 RADV_PIPELINE_COMPUTE,
1881 /* Pipeline library. This can't actually run and merely is a partial pipeline. */
1882 RADV_PIPELINE_LIBRARY
1885 struct radv_pipeline_group_handle {
1886 uint32_t handles[2];
1889 struct radv_pipeline_shader_stack_size {
1890 uint32_t recursive_size;
1891 /* anyhit + intersection */
1892 uint32_t non_recursive_size;
1895 struct radv_pipeline_slab {
1898 union radv_shader_arena_block *alloc;
1901 void radv_pipeline_slab_destroy(struct radv_device *device, struct radv_pipeline_slab *slab);
1903 struct radv_vertex_input_info {
1904 uint32_t instance_rate_inputs;
1905 uint32_t instance_rate_divisors[MAX_VERTEX_ATTRIBS];
1906 uint8_t vertex_attribute_formats[MAX_VERTEX_ATTRIBS];
1907 uint32_t vertex_attribute_bindings[MAX_VERTEX_ATTRIBS];
1908 uint32_t vertex_attribute_offsets[MAX_VERTEX_ATTRIBS];
1909 uint32_t vertex_attribute_strides[MAX_VERTEX_ATTRIBS];
1910 uint8_t vertex_binding_align[MAX_VBS];
1911 enum radv_vs_input_alpha_adjust vertex_alpha_adjust[MAX_VERTEX_ATTRIBS];
1912 uint32_t vertex_post_shuffle;
1913 uint32_t binding_stride[MAX_VBS];
1914 uint8_t attrib_bindings[MAX_VERTEX_ATTRIBS];
1915 uint32_t attrib_ends[MAX_VERTEX_ATTRIBS];
1916 uint32_t attrib_index_offset[MAX_VERTEX_ATTRIBS];
1919 struct radv_input_assembly_info {
1920 uint8_t primitive_topology; /* VkPrimitiveTopology */
1921 bool primitive_restart_enable;
1924 struct radv_tessellation_info {
1925 uint8_t patch_control_points;
1926 VkTessellationDomainOrigin domain_origin;
1929 struct radv_viewport_info {
1930 bool negative_one_to_one;
1931 uint8_t viewport_count;
1932 uint8_t scissor_count;
1933 VkRect2D scissors[MAX_SCISSORS];
1934 VkViewport viewports[MAX_VIEWPORTS];
1937 struct radv_rasterization_info {
1938 bool discard_enable;
1939 VkFrontFace front_face;
1940 VkCullModeFlags cull_mode;
1941 uint8_t polygon_mode; /* VkPolygonMode */
1942 bool depth_bias_enable;
1943 bool depth_clamp_enable;
1945 float depth_bias_constant_factor;
1946 float depth_bias_clamp;
1947 float depth_bias_slope_factor;
1948 VkConservativeRasterizationModeEXT conservative_mode;
1949 bool provoking_vtx_last;
1950 bool stippled_line_enable;
1951 VkLineRasterizationModeEXT line_raster_mode;
1952 uint32_t line_stipple_factor;
1953 uint16_t line_stipple_pattern;
1954 bool depth_clip_disable;
1955 VkRasterizationOrderAMD order;
1958 struct radv_discard_rectangle_info {
1959 VkDiscardRectangleModeEXT mode;
1960 VkRect2D rects[MAX_DISCARD_RECTANGLES];
1964 struct radv_multisample_info {
1965 bool sample_shading_enable;
1966 bool alpha_to_coverage_enable;
1967 bool sample_locs_enable;
1968 VkSampleCountFlagBits raster_samples;
1969 float min_sample_shading;
1970 uint16_t sample_mask;
1971 uint8_t sample_locs_count;
1972 VkSampleCountFlagBits sample_locs_per_pixel;
1973 VkExtent2D sample_locs_grid_size;
1974 VkSampleLocationEXT sample_locs[MAX_SAMPLE_LOCATIONS];
1977 struct radv_stencil_op_info {
1978 VkStencilOp fail_op;
1979 VkStencilOp pass_op;
1980 VkStencilOp depth_fail_op;
1981 VkCompareOp compare_op;
1982 uint8_t compare_mask;
1987 struct radv_depth_stencil_info {
1988 bool stencil_test_enable;
1989 bool depth_test_enable;
1990 bool depth_write_enable;
1991 bool depth_bounds_test_enable;
1996 struct radv_stencil_op_info front;
1997 struct radv_stencil_op_info back;
1998 VkCompareOp depth_compare_op;
2001 struct radv_rendering_info {
2003 uint32_t color_att_count;
2004 VkFormat color_att_formats[MAX_RTS];
2005 VkFormat depth_att_format;
2006 VkFormat stencil_att_format;
2009 struct radv_color_blend_info {
2010 bool logic_op_enable;
2013 uint32_t color_write_enable;
2014 float blend_constants[4];
2016 uint8_t color_write_mask;
2018 uint16_t color_blend_op;
2019 uint16_t alpha_blend_op;
2020 uint16_t src_color_blend_factor;
2021 uint16_t dst_color_blend_factor;
2022 uint16_t src_alpha_blend_factor;
2023 uint16_t dst_alpha_blend_factor;
2027 struct radv_fragment_shading_rate_info {
2029 VkFragmentShadingRateCombinerOpKHR combiner_ops[2];
2032 struct radv_graphics_pipeline_info {
2033 struct radv_vertex_input_info vi;
2034 struct radv_input_assembly_info ia;
2036 struct radv_tessellation_info ts;
2037 struct radv_viewport_info vp;
2038 struct radv_rasterization_info rs;
2039 struct radv_discard_rectangle_info dr;
2041 struct radv_multisample_info ms;
2042 struct radv_depth_stencil_info ds;
2043 struct radv_rendering_info ri;
2044 struct radv_color_blend_info cb;
2046 struct radv_fragment_shading_rate_info fsr;
2048 /* VK_AMD_mixed_attachment_samples */
2049 uint8_t color_att_samples;
2050 uint8_t ds_att_samples;
2053 struct radv_pipeline {
2054 struct vk_object_base base;
2055 enum radv_pipeline_type type;
2057 struct radv_device *device;
2059 struct radv_pipeline_slab *slab;
2060 struct radeon_winsys_bo *slab_bo;
2062 bool need_indirect_descriptor_sets;
2063 struct radv_shader *shaders[MESA_VULKAN_SHADER_STAGES];
2064 struct radv_shader *gs_copy_shader;
2066 struct radeon_cmdbuf cs;
2067 uint32_t ctx_cs_hash;
2068 struct radeon_cmdbuf ctx_cs;
2070 uint32_t user_data_0[MESA_VULKAN_SHADER_STAGES];
2073 unsigned scratch_bytes_per_wave;
2075 /* Unique pipeline hash identifier. */
2076 uint64_t pipeline_hash;
2078 /* Pipeline layout info. */
2079 uint32_t push_constant_size;
2080 uint32_t dynamic_offset_count;
2083 struct radv_graphics_pipeline {
2084 struct radv_pipeline base;
2086 VkShaderStageFlags active_stages;
2088 struct radv_dynamic_state dynamic_state;
2090 uint64_t dynamic_states;
2091 struct radv_multisample_state ms;
2092 struct radv_binning_state binning;
2093 struct radv_vrs_state vrs;
2094 uint32_t spi_baryc_cntl;
2095 unsigned esgs_ring_size;
2096 unsigned gsvs_ring_size;
2097 uint32_t vtx_base_sgpr;
2098 struct radv_ia_multi_vgt_param_helpers ia_multi_vgt_param;
2099 uint8_t vtx_emit_num;
2100 uint64_t needed_dynamic_state;
2101 unsigned tess_patch_control_points;
2102 unsigned pa_su_sc_mode_cntl;
2103 unsigned db_depth_control;
2104 unsigned pa_cl_clip_cntl;
2105 unsigned cb_color_control;
2106 uint32_t binding_stride[MAX_VBS];
2107 uint8_t attrib_bindings[MAX_VERTEX_ATTRIBS];
2108 uint32_t attrib_ends[MAX_VERTEX_ATTRIBS];
2109 uint32_t attrib_index_offset[MAX_VERTEX_ATTRIBS];
2110 uint8_t last_vertex_attrib_bit;
2111 uint8_t next_vertex_stage : 8;
2112 uint32_t vb_desc_usage_mask;
2113 uint32_t vb_desc_alloc_size;
2115 /* Last pre-PS API stage */
2116 gl_shader_stage last_vgt_api_stage;
2118 /* Used for rbplus */
2119 uint32_t col_format;
2120 uint32_t cb_target_mask;
2122 bool disable_out_of_order_rast_for_occlusion;
2124 bool uses_baseinstance;
2125 bool can_use_guardband;
2126 bool uses_dynamic_stride;
2127 bool uses_conservative_overestimate;
2128 bool negative_one_to_one;
2129 bool use_per_attribute_vb_descs;
2130 bool can_use_simple_input;
2131 bool uses_user_sample_locations;
2133 /* Whether the pipeline forces per-vertex VRS (GFX10.3+). */
2134 bool force_vrs_per_vertex;
2136 /* Whether the pipeline uses NGG (GFX10+). */
2138 bool has_ngg_culling;
2140 /* Not NULL if graphics pipeline uses streamout. */
2141 struct radv_shader *streamout_shader;
2144 struct radv_compute_pipeline {
2145 struct radv_pipeline base;
2147 bool cs_regalloc_hang_bug;
2150 struct radv_pipeline_group_handle *rt_group_handles;
2151 struct radv_pipeline_shader_stack_size *rt_stack_sizes;
2152 bool dynamic_stack_size;
2153 uint32_t group_count;
2156 struct radv_library_pipeline {
2157 struct radv_pipeline base;
2159 unsigned stage_count;
2160 VkPipelineShaderStageCreateInfo *stages;
2161 unsigned group_count;
2162 VkRayTracingShaderGroupCreateInfoKHR *groups;
2163 VkPipelineShaderStageModuleIdentifierCreateInfoEXT *identifiers;
2165 uint8_t sha1[SHA1_DIGEST_LENGTH];
2169 #define RADV_DECL_PIPELINE_DOWNCAST(pipe_type, pipe_enum) \
2170 static inline struct radv_##pipe_type##_pipeline * \
2171 radv_pipeline_to_##pipe_type(struct radv_pipeline *pipeline) \
2173 assert(pipeline->type == pipe_enum); \
2174 return (struct radv_##pipe_type##_pipeline *) pipeline; \
2177 RADV_DECL_PIPELINE_DOWNCAST(graphics, RADV_PIPELINE_GRAPHICS)
2178 RADV_DECL_PIPELINE_DOWNCAST(compute, RADV_PIPELINE_COMPUTE)
2179 RADV_DECL_PIPELINE_DOWNCAST(library, RADV_PIPELINE_LIBRARY)
2181 struct radv_pipeline_stage {
2182 gl_shader_stage stage;
2185 const struct vk_object_base *object;
2188 unsigned char sha1[20];
2191 const char *entrypoint;
2192 const VkSpecializationInfo *spec_info;
2194 unsigned char shader_sha1[20];
2197 nir_shader *internal_nir; /* meta shaders */
2199 struct radv_shader_info info;
2200 struct radv_shader_args args;
2202 VkPipelineCreationFeedbackEXT feedback;
2206 radv_pipeline_has_stage(const struct radv_graphics_pipeline *pipeline, gl_shader_stage stage)
2208 return pipeline->base.shaders[stage];
2211 bool radv_pipeline_has_ngg_passthrough(const struct radv_graphics_pipeline *pipeline);
2213 bool radv_pipeline_has_gs_copy_shader(const struct radv_pipeline *pipeline);
2215 struct radv_userdata_info *radv_lookup_user_sgpr(struct radv_pipeline *pipeline,
2216 gl_shader_stage stage, int idx);
2218 struct radv_shader *radv_get_shader(const struct radv_pipeline *pipeline, gl_shader_stage stage);
2220 struct radv_graphics_pipeline_create_info {
2222 bool db_depth_clear;
2223 bool db_stencil_clear;
2224 bool depth_compress_disable;
2225 bool stencil_compress_disable;
2226 bool resummarize_enable;
2227 uint32_t custom_blend_mode;
2230 void radv_pipeline_init(struct radv_device *device, struct radv_pipeline *pipeline,
2231 enum radv_pipeline_type type);
2233 VkResult radv_graphics_pipeline_create(VkDevice device, VkPipelineCache cache,
2234 const VkGraphicsPipelineCreateInfo *pCreateInfo,
2235 const struct radv_graphics_pipeline_create_info *extra,
2236 const VkAllocationCallbacks *alloc, VkPipeline *pPipeline);
2238 VkResult radv_compute_pipeline_create(VkDevice _device, VkPipelineCache _cache,
2239 const VkComputePipelineCreateInfo *pCreateInfo,
2240 const VkAllocationCallbacks *pAllocator,
2241 const uint8_t *custom_hash,
2242 struct radv_pipeline_shader_stack_size *rt_stack_sizes,
2243 uint32_t rt_group_count, VkPipeline *pPipeline);
2245 void radv_pipeline_destroy(struct radv_device *device, struct radv_pipeline *pipeline,
2246 const VkAllocationCallbacks *allocator);
2248 struct radv_binning_settings {
2249 unsigned context_states_per_bin; /* allowed range: [1, 6] */
2250 unsigned persistent_states_per_bin; /* allowed range: [1, 32] */
2251 unsigned fpovs_per_batch; /* allowed range: [0, 255], 0 = unlimited */
2254 struct radv_binning_settings radv_get_binning_settings(const struct radv_physical_device *pdev);
2256 struct vk_format_description;
2257 uint32_t radv_translate_buffer_dataformat(const struct util_format_description *desc,
2258 int first_non_void);
2259 uint32_t radv_translate_buffer_numformat(const struct util_format_description *desc,
2260 int first_non_void);
2261 bool radv_is_buffer_format_supported(VkFormat format, bool *scaled);
2262 void radv_translate_vertex_format(const struct radv_physical_device *pdevice, VkFormat format,
2263 const struct util_format_description *desc, unsigned *dfmt,
2264 unsigned *nfmt, bool *post_shuffle,
2265 enum radv_vs_input_alpha_adjust *alpha_adjust);
2266 uint32_t radv_translate_colorformat(VkFormat format);
2267 uint32_t radv_translate_color_numformat(VkFormat format, const struct util_format_description *desc,
2268 int first_non_void);
2269 uint32_t radv_colorformat_endian_swap(uint32_t colorformat);
2270 unsigned radv_translate_colorswap(VkFormat format, bool do_endian_swap);
2271 uint32_t radv_translate_dbformat(VkFormat format);
2272 uint32_t radv_translate_tex_dataformat(VkFormat format, const struct util_format_description *desc,
2273 int first_non_void);
2274 uint32_t radv_translate_tex_numformat(VkFormat format, const struct util_format_description *desc,
2275 int first_non_void);
2276 bool radv_format_pack_clear_color(VkFormat format, uint32_t clear_vals[2],
2277 VkClearColorValue *value);
2278 bool radv_is_storage_image_format_supported(struct radv_physical_device *physical_device,
2280 bool radv_is_colorbuffer_format_supported(const struct radv_physical_device *pdevice,
2281 VkFormat format, bool *blendable);
2282 bool radv_dcc_formats_compatible(enum amd_gfx_level gfx_level, VkFormat format1, VkFormat format2,
2283 bool *sign_reinterpret);
2284 bool radv_is_atomic_format_supported(VkFormat format);
2285 bool radv_device_supports_etc(struct radv_physical_device *physical_device);
2287 static const VkImageUsageFlags RADV_IMAGE_USAGE_WRITE_BITS =
2288 VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT |
2289 VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | VK_IMAGE_USAGE_STORAGE_BIT;
2291 struct radv_image_plane {
2293 struct radeon_surf surface;
2299 struct ac_surf_info info;
2304 unsigned queue_family_mask;
2308 bool dcc_sign_reinterpret;
2309 bool support_comp_to_single;
2311 /* Set when bound */
2312 struct radeon_winsys_bo *bo;
2313 VkDeviceSize offset;
2314 bool tc_compatible_cmask;
2316 uint64_t clear_value_offset;
2317 uint64_t fce_pred_offset;
2318 uint64_t dcc_pred_offset;
2321 * Metadata for the TC-compat zrange workaround. If the 32-bit value
2322 * stored at this offset is UINT_MAX, the driver will emit
2323 * DB_Z_INFO.ZRANGE_PRECISION=0, otherwise it will skip the
2324 * SET_CONTEXT_REG packet.
2326 uint64_t tc_compat_zrange_offset;
2328 /* For VK_ANDROID_native_buffer, the WSI image owns the memory, */
2329 VkDeviceMemory owned_memory;
2331 unsigned plane_count;
2332 struct radv_image_plane planes[0];
2335 /* Whether the image has a htile that is known consistent with the contents of
2336 * the image and is allowed to be in compressed form.
2338 * If this is false reads that don't use the htile should be able to return
2341 bool radv_layout_is_htile_compressed(const struct radv_device *device,
2342 const struct radv_image *image, VkImageLayout layout,
2343 bool in_render_loop, unsigned queue_mask);
2345 bool radv_layout_can_fast_clear(const struct radv_device *device, const struct radv_image *image,
2346 unsigned level, VkImageLayout layout, bool in_render_loop,
2347 unsigned queue_mask);
2349 bool radv_layout_dcc_compressed(const struct radv_device *device, const struct radv_image *image,
2350 unsigned level, VkImageLayout layout, bool in_render_loop,
2351 unsigned queue_mask);
2353 bool radv_layout_fmask_compressed(const struct radv_device *device, const struct radv_image *image,
2354 VkImageLayout layout, unsigned queue_mask);
2357 * Return whether the image has CMASK metadata for color surfaces.
2360 radv_image_has_cmask(const struct radv_image *image)
2362 return image->planes[0].surface.cmask_offset;
2366 * Return whether the image has FMASK metadata for color surfaces.
2369 radv_image_has_fmask(const struct radv_image *image)
2371 return image->planes[0].surface.fmask_offset;
2375 * Return whether the image has DCC metadata for color surfaces.
2378 radv_image_has_dcc(const struct radv_image *image)
2380 return !(image->planes[0].surface.flags & RADEON_SURF_Z_OR_SBUFFER) &&
2381 image->planes[0].surface.meta_offset;
2385 * Return whether the image is TC-compatible CMASK.
2388 radv_image_is_tc_compat_cmask(const struct radv_image *image)
2390 return radv_image_has_fmask(image) && image->tc_compatible_cmask;
2394 * Return whether DCC metadata is enabled for a level.
2397 radv_dcc_enabled(const struct radv_image *image, unsigned level)
2399 return radv_image_has_dcc(image) && level < image->planes[0].surface.num_meta_levels;
2403 * Return whether the image has CB metadata.
2406 radv_image_has_CB_metadata(const struct radv_image *image)
2408 return radv_image_has_cmask(image) || radv_image_has_fmask(image) || radv_image_has_dcc(image);
2412 * Return whether the image has HTILE metadata for depth surfaces.
2415 radv_image_has_htile(const struct radv_image *image)
2417 return image->planes[0].surface.flags & RADEON_SURF_Z_OR_SBUFFER &&
2418 image->planes[0].surface.meta_size;
2422 * Return whether the image has VRS HTILE metadata for depth surfaces
2425 radv_image_has_vrs_htile(const struct radv_device *device, const struct radv_image *image)
2427 /* Any depth buffer can potentially use VRS. */
2428 return device->attachment_vrs_enabled && radv_image_has_htile(image) &&
2429 (image->vk.usage & VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT);
2433 * Return whether HTILE metadata is enabled for a level.
2436 radv_htile_enabled(const struct radv_image *image, unsigned level)
2438 return radv_image_has_htile(image) && level < image->planes[0].surface.num_meta_levels;
2442 * Return whether the image is TC-compatible HTILE.
2445 radv_image_is_tc_compat_htile(const struct radv_image *image)
2447 return radv_image_has_htile(image) &&
2448 (image->planes[0].surface.flags & RADEON_SURF_TC_COMPATIBLE_HTILE);
2452 * Return whether the entire HTILE buffer can be used for depth in order to
2453 * improve HiZ Z-Range precision.
2456 radv_image_tile_stencil_disabled(const struct radv_device *device, const struct radv_image *image)
2458 if (device->physical_device->rad_info.gfx_level >= GFX9) {
2459 return !vk_format_has_stencil(image->vk.format) && !radv_image_has_vrs_htile(device, image);
2461 /* Due to a hw bug, TILE_STENCIL_DISABLE must be set to 0 for
2462 * the TC-compat ZRANGE issue even if no stencil is used.
2464 return !vk_format_has_stencil(image->vk.format) && !radv_image_is_tc_compat_htile(image);
2469 radv_image_has_clear_value(const struct radv_image *image)
2471 return image->clear_value_offset != 0;
2474 static inline uint64_t
2475 radv_image_get_fast_clear_va(const struct radv_image *image, uint32_t base_level)
2477 assert(radv_image_has_clear_value(image));
2479 uint64_t va = radv_buffer_get_va(image->bo);
2480 va += image->offset + image->clear_value_offset + base_level * 8;
2484 static inline uint64_t
2485 radv_image_get_fce_pred_va(const struct radv_image *image, uint32_t base_level)
2487 assert(image->fce_pred_offset != 0);
2489 uint64_t va = radv_buffer_get_va(image->bo);
2490 va += image->offset + image->fce_pred_offset + base_level * 8;
2494 static inline uint64_t
2495 radv_image_get_dcc_pred_va(const struct radv_image *image, uint32_t base_level)
2497 assert(image->dcc_pred_offset != 0);
2499 uint64_t va = radv_buffer_get_va(image->bo);
2500 va += image->offset + image->dcc_pred_offset + base_level * 8;
2504 static inline uint64_t
2505 radv_get_tc_compat_zrange_va(const struct radv_image *image, uint32_t base_level)
2507 assert(image->tc_compat_zrange_offset != 0);
2509 uint64_t va = radv_buffer_get_va(image->bo);
2510 va += image->offset + image->tc_compat_zrange_offset + base_level * 4;
2514 static inline uint64_t
2515 radv_get_ds_clear_value_va(const struct radv_image *image, uint32_t base_level)
2517 assert(radv_image_has_clear_value(image));
2519 uint64_t va = radv_buffer_get_va(image->bo);
2520 va += image->offset + image->clear_value_offset + base_level * 8;
2524 static inline uint32_t
2525 radv_get_htile_initial_value(const struct radv_device *device, const struct radv_image *image)
2527 uint32_t initial_value;
2529 if (radv_image_tile_stencil_disabled(device, image)) {
2530 /* Z only (no stencil):
2533 * +---------+---------+-------+
2534 * | Max Z | Min Z | ZMask |
2536 initial_value = 0xfffc000f;
2540 * |31 12|11 10|9 8|7 6|5 4|3 0|
2541 * +-----------+-----+------+-----+-----+-------+
2542 * | Z Range | | SMem | SR1 | SR0 | ZMask |
2544 * SR0/SR1 contains the stencil test results. Initializing
2545 * SR0/SR1 to 0x3 means the stencil test result is unknown.
2547 * Z, stencil and 4 bit VRS encoding:
2548 * |31 12|11 10|9 8|7 6|5 4|3 0|
2549 * +-----------+------------+------+------------+-----+-------+
2550 * | Z Range | VRS y-rate | SMem | VRS x-rate | SR0 | ZMask |
2552 if (radv_image_has_vrs_htile(device, image)) {
2553 /* Initialize the VRS x-rate value at 0, so the hw interprets it as 1 sample. */
2554 initial_value = 0xfffff33f;
2556 initial_value = 0xfffff3ff;
2560 return initial_value;
2564 radv_image_get_iterate256(struct radv_device *device, struct radv_image *image)
2566 /* ITERATE_256 is required for depth or stencil MSAA images that are TC-compatible HTILE. */
2567 return device->physical_device->rad_info.gfx_level >= GFX10 &&
2569 (VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT)) &&
2570 radv_image_is_tc_compat_htile(image) && image->info.samples > 1;
2573 unsigned radv_image_queue_family_mask(const struct radv_image *image,
2574 enum radv_queue_family family,
2575 enum radv_queue_family queue_family);
2577 static inline uint32_t
2578 radv_get_layerCount(const struct radv_image *image, const VkImageSubresourceRange *range)
2580 return range->layerCount == VK_REMAINING_ARRAY_LAYERS
2581 ? image->info.array_size - range->baseArrayLayer
2582 : range->layerCount;
2585 static inline uint32_t
2586 radv_get_levelCount(const struct radv_image *image, const VkImageSubresourceRange *range)
2588 return range->levelCount == VK_REMAINING_MIP_LEVELS ? image->info.levels - range->baseMipLevel
2589 : range->levelCount;
2592 bool radv_image_is_renderable(struct radv_device *device, struct radv_image *image);
2594 struct radeon_bo_metadata;
2595 void radv_init_metadata(struct radv_device *device, struct radv_image *image,
2596 struct radeon_bo_metadata *metadata);
2598 void radv_image_override_offset_stride(struct radv_device *device, struct radv_image *image,
2599 uint64_t offset, uint32_t stride);
2601 union radv_descriptor {
2603 uint32_t plane0_descriptor[8];
2604 uint32_t fmask_descriptor[8];
2607 uint32_t plane_descriptors[3][8];
2611 struct radv_image_view {
2612 struct vk_image_view vk;
2613 struct radv_image *image; /**< VkImageViewCreateInfo::image */
2616 VkExtent3D extent; /**< Extent of VkImageViewCreateInfo::baseMipLevel. */
2618 /* Whether the image iview supports fast clear. */
2619 bool support_fast_clear;
2621 bool disable_dcc_mrt;
2623 union radv_descriptor descriptor;
2625 /* Descriptor for use as a storage image as opposed to a sampled image.
2626 * This has a few differences for cube maps (e.g. type).
2628 union radv_descriptor storage_descriptor;
2631 struct radv_image_create_info {
2632 const VkImageCreateInfo *vk_info;
2634 bool no_metadata_planes;
2635 bool prime_blit_src;
2636 const struct radeon_bo_metadata *bo_metadata;
2640 radv_image_create_layout(struct radv_device *device, struct radv_image_create_info create_info,
2641 const struct VkImageDrmFormatModifierExplicitCreateInfoEXT *mod_info,
2642 struct radv_image *image);
2644 VkResult radv_image_create(VkDevice _device, const struct radv_image_create_info *info,
2645 const VkAllocationCallbacks *alloc, VkImage *pImage);
2647 bool radv_are_formats_dcc_compatible(const struct radv_physical_device *pdev, const void *pNext,
2648 VkFormat format, VkImageCreateFlags flags,
2649 bool *sign_reinterpret);
2651 bool vi_alpha_is_on_msb(struct radv_device *device, VkFormat format);
2653 VkResult radv_image_from_gralloc(VkDevice device_h, const VkImageCreateInfo *base_info,
2654 const VkNativeBufferANDROID *gralloc_info,
2655 const VkAllocationCallbacks *alloc, VkImage *out_image_h);
2656 uint64_t radv_ahb_usage_from_vk_usage(const VkImageCreateFlags vk_create,
2657 const VkImageUsageFlags vk_usage);
2658 VkResult radv_import_ahb_memory(struct radv_device *device, struct radv_device_memory *mem,
2660 const VkImportAndroidHardwareBufferInfoANDROID *info);
2661 VkResult radv_create_ahb_memory(struct radv_device *device, struct radv_device_memory *mem,
2662 unsigned priority, const VkMemoryAllocateInfo *pAllocateInfo);
2664 VkFormat radv_select_android_external_format(const void *next, VkFormat default_format);
2666 bool radv_android_gralloc_supports_format(VkFormat format, VkImageUsageFlagBits usage);
2668 struct radv_image_view_extra_create_info {
2669 bool disable_compression;
2670 bool enable_compression;
2671 bool disable_dcc_mrt;
2672 bool from_client; /**< Set only if this came from vkCreateImage */
2675 void radv_image_view_init(struct radv_image_view *view, struct radv_device *device,
2676 const VkImageViewCreateInfo *pCreateInfo,
2677 VkImageCreateFlags img_create_flags,
2678 const struct radv_image_view_extra_create_info *extra_create_info);
2679 void radv_image_view_finish(struct radv_image_view *iview);
2681 VkFormat radv_get_aspect_format(struct radv_image *image, VkImageAspectFlags mask);
2683 struct radv_sampler_ycbcr_conversion_state {
2685 VkSamplerYcbcrModelConversion ycbcr_model;
2686 VkSamplerYcbcrRange ycbcr_range;
2687 VkComponentMapping components;
2688 VkChromaLocation chroma_offsets[2];
2689 VkFilter chroma_filter;
2692 struct radv_sampler_ycbcr_conversion {
2693 struct vk_object_base base;
2694 /* The state is hashed for the descriptor set layout. */
2695 struct radv_sampler_ycbcr_conversion_state state;
2698 struct radv_buffer_view {
2699 struct vk_object_base base;
2700 struct radeon_winsys_bo *bo;
2702 uint64_t range; /**< VkBufferViewCreateInfo::range */
2705 void radv_buffer_view_init(struct radv_buffer_view *view, struct radv_device *device,
2706 const VkBufferViewCreateInfo *pCreateInfo);
2707 void radv_buffer_view_finish(struct radv_buffer_view *view);
2710 radv_image_extent_compare(const struct radv_image *image, const VkExtent3D *extent)
2712 if (extent->width != image->info.width || extent->height != image->info.height ||
2713 extent->depth != image->info.depth)
2718 struct radv_sampler {
2719 struct vk_object_base base;
2721 struct radv_sampler_ycbcr_conversion *ycbcr_sampler;
2722 uint32_t border_color_slot;
2725 struct radv_subpass_barrier {
2726 VkPipelineStageFlags2 src_stage_mask;
2727 VkAccessFlags2 src_access_mask;
2728 VkAccessFlags2 dst_access_mask;
2731 void radv_emit_subpass_barrier(struct radv_cmd_buffer *cmd_buffer,
2732 const struct radv_subpass_barrier *barrier);
2734 struct radv_subpass_attachment {
2735 uint32_t attachment;
2736 VkImageLayout layout;
2737 VkImageLayout stencil_layout;
2738 bool in_render_loop;
2741 struct radv_subpass {
2742 uint32_t attachment_count;
2743 struct radv_subpass_attachment *attachments;
2745 uint32_t input_count;
2746 uint32_t color_count;
2747 struct radv_subpass_attachment *input_attachments;
2748 struct radv_subpass_attachment *color_attachments;
2749 struct radv_subpass_attachment *resolve_attachments;
2750 struct radv_subpass_attachment *depth_stencil_attachment;
2751 struct radv_subpass_attachment *ds_resolve_attachment;
2752 struct radv_subpass_attachment *vrs_attachment;
2753 VkResolveModeFlagBits depth_resolve_mode;
2754 VkResolveModeFlagBits stencil_resolve_mode;
2756 /** Subpass has at least one color resolve attachment */
2757 bool has_color_resolve;
2759 struct radv_subpass_barrier start_barrier;
2763 VkSampleCountFlagBits color_sample_count;
2764 VkSampleCountFlagBits depth_sample_count;
2765 VkSampleCountFlagBits max_sample_count;
2767 /* Whether the subpass has ingoing/outgoing external dependencies. */
2768 bool has_ingoing_dep;
2769 bool has_outgoing_dep;
2772 uint32_t radv_get_subpass_id(struct radv_cmd_buffer *cmd_buffer);
2774 struct radv_render_pass_attachment {
2777 VkAttachmentLoadOp load_op;
2778 VkAttachmentLoadOp stencil_load_op;
2779 VkImageLayout initial_layout;
2780 VkImageLayout final_layout;
2781 VkImageLayout stencil_initial_layout;
2782 VkImageLayout stencil_final_layout;
2784 /* The subpass id in which the attachment will be used first/last. */
2785 uint32_t first_subpass_idx;
2786 uint32_t last_subpass_idx;
2789 struct radv_render_pass {
2790 struct vk_object_base base;
2791 uint32_t attachment_count;
2792 uint32_t subpass_count;
2793 struct radv_subpass_attachment *subpass_attachments;
2794 struct radv_render_pass_attachment *attachments;
2795 struct radv_subpass_barrier end_barrier;
2796 struct radv_subpass subpasses[0];
2799 VkResult radv_device_init_meta(struct radv_device *device);
2800 void radv_device_finish_meta(struct radv_device *device);
2802 struct radv_query_pool {
2803 struct vk_object_base base;
2804 struct radeon_winsys_bo *bo;
2806 uint32_t availability_offset;
2810 uint32_t pipeline_stats_mask;
2811 bool uses_gds; /* For NGG GS on GFX10+ */
2814 bool radv_queue_internal_submit(struct radv_queue *queue, struct radeon_cmdbuf *cs);
2816 int radv_queue_init(struct radv_device *device, struct radv_queue *queue, int idx,
2817 const VkDeviceQueueCreateInfo *create_info,
2818 const VkDeviceQueueGlobalPriorityCreateInfoEXT *global_priority);
2820 void radv_set_descriptor_set(struct radv_cmd_buffer *cmd_buffer, VkPipelineBindPoint bind_point,
2821 struct radv_descriptor_set *set, unsigned idx);
2823 void radv_cmd_update_descriptor_sets(struct radv_device *device, struct radv_cmd_buffer *cmd_buffer,
2824 VkDescriptorSet overrideSet, uint32_t descriptorWriteCount,
2825 const VkWriteDescriptorSet *pDescriptorWrites,
2826 uint32_t descriptorCopyCount,
2827 const VkCopyDescriptorSet *pDescriptorCopies);
2829 void radv_cmd_update_descriptor_set_with_template(struct radv_device *device,
2830 struct radv_cmd_buffer *cmd_buffer,
2831 struct radv_descriptor_set *set,
2832 VkDescriptorUpdateTemplate descriptorUpdateTemplate,
2835 void radv_meta_push_descriptor_set(struct radv_cmd_buffer *cmd_buffer,
2836 VkPipelineBindPoint pipelineBindPoint, VkPipelineLayout _layout,
2837 uint32_t set, uint32_t descriptorWriteCount,
2838 const VkWriteDescriptorSet *pDescriptorWrites);
2840 uint32_t radv_init_dcc(struct radv_cmd_buffer *cmd_buffer, struct radv_image *image,
2841 const VkImageSubresourceRange *range, uint32_t value);
2843 uint32_t radv_init_fmask(struct radv_cmd_buffer *cmd_buffer, struct radv_image *image,
2844 const VkImageSubresourceRange *range);
2846 /* radv_nir_to_llvm.c */
2847 struct radv_shader_args;
2848 struct radv_nir_compiler_options;
2849 struct radv_shader_info;
2851 void llvm_compile_shader(const struct radv_nir_compiler_options *options,
2852 const struct radv_shader_info *info, unsigned shader_count,
2853 struct nir_shader *const *shaders, struct radv_shader_binary **binary,
2854 const struct radv_shader_args *args);
2856 /* radv_shader_info.h */
2857 struct radv_shader_info;
2859 void radv_nir_shader_info_pass(struct radv_device *device, const struct nir_shader *nir,
2860 const struct radv_pipeline_layout *layout,
2861 const struct radv_pipeline_key *pipeline_key,
2862 struct radv_shader_info *info);
2864 void radv_nir_shader_info_init(struct radv_shader_info *info);
2866 bool radv_thread_trace_init(struct radv_device *device);
2867 void radv_thread_trace_finish(struct radv_device *device);
2868 bool radv_begin_thread_trace(struct radv_queue *queue);
2869 bool radv_end_thread_trace(struct radv_queue *queue);
2870 bool radv_get_thread_trace(struct radv_queue *queue, struct ac_thread_trace *thread_trace);
2871 void radv_emit_thread_trace_userdata(struct radv_cmd_buffer *cmd_buffer, const void *data,
2872 uint32_t num_dwords);
2873 bool radv_is_instruction_timing_enabled(void);
2875 void radv_emit_inhibit_clockgating(struct radv_device *device, struct radeon_cmdbuf *cs,
2878 bool radv_sdma_copy_image(struct radv_cmd_buffer *cmd_buffer, struct radv_image *image,
2879 struct radv_buffer *buffer, const VkBufferImageCopy2 *region);
2881 /* radv_sqtt_layer_.c */
2882 struct radv_barrier_data {
2885 uint16_t depth_stencil_expand : 1;
2886 uint16_t htile_hiz_range_expand : 1;
2887 uint16_t depth_stencil_resummarize : 1;
2888 uint16_t dcc_decompress : 1;
2889 uint16_t fmask_decompress : 1;
2890 uint16_t fast_clear_eliminate : 1;
2891 uint16_t fmask_color_expand : 1;
2892 uint16_t init_mask_ram : 1;
2893 uint16_t reserved : 8;
2896 } layout_transitions;
2900 * Value for the reason field of an RGP barrier start marker originating from
2901 * the Vulkan client (does not include PAL-defined values). (Table 15)
2903 enum rgp_barrier_reason {
2904 RGP_BARRIER_UNKNOWN_REASON = 0xFFFFFFFF,
2906 /* External app-generated barrier reasons, i.e. API synchronization
2907 * commands Range of valid values: [0x00000001 ... 0x7FFFFFFF].
2909 RGP_BARRIER_EXTERNAL_CMD_PIPELINE_BARRIER = 0x00000001,
2910 RGP_BARRIER_EXTERNAL_RENDER_PASS_SYNC = 0x00000002,
2911 RGP_BARRIER_EXTERNAL_CMD_WAIT_EVENTS = 0x00000003,
2913 /* Internal barrier reasons, i.e. implicit synchronization inserted by
2914 * the Vulkan driver Range of valid values: [0xC0000000 ... 0xFFFFFFFE].
2916 RGP_BARRIER_INTERNAL_BASE = 0xC0000000,
2917 RGP_BARRIER_INTERNAL_PRE_RESET_QUERY_POOL_SYNC = RGP_BARRIER_INTERNAL_BASE + 0,
2918 RGP_BARRIER_INTERNAL_POST_RESET_QUERY_POOL_SYNC = RGP_BARRIER_INTERNAL_BASE + 1,
2919 RGP_BARRIER_INTERNAL_GPU_EVENT_RECYCLE_STALL = RGP_BARRIER_INTERNAL_BASE + 2,
2920 RGP_BARRIER_INTERNAL_PRE_COPY_QUERY_POOL_RESULTS_SYNC = RGP_BARRIER_INTERNAL_BASE + 3
2923 void radv_describe_begin_cmd_buffer(struct radv_cmd_buffer *cmd_buffer);
2924 void radv_describe_end_cmd_buffer(struct radv_cmd_buffer *cmd_buffer);
2925 void radv_describe_draw(struct radv_cmd_buffer *cmd_buffer);
2926 void radv_describe_dispatch(struct radv_cmd_buffer *cmd_buffer, int x, int y, int z);
2927 void radv_describe_begin_render_pass_clear(struct radv_cmd_buffer *cmd_buffer,
2928 VkImageAspectFlagBits aspects);
2929 void radv_describe_end_render_pass_clear(struct radv_cmd_buffer *cmd_buffer);
2930 void radv_describe_begin_render_pass_resolve(struct radv_cmd_buffer *cmd_buffer);
2931 void radv_describe_end_render_pass_resolve(struct radv_cmd_buffer *cmd_buffer);
2932 void radv_describe_barrier_start(struct radv_cmd_buffer *cmd_buffer,
2933 enum rgp_barrier_reason reason);
2934 void radv_describe_barrier_end(struct radv_cmd_buffer *cmd_buffer);
2935 void radv_describe_barrier_end_delayed(struct radv_cmd_buffer *cmd_buffer);
2936 void radv_describe_layout_transition(struct radv_cmd_buffer *cmd_buffer,
2937 const struct radv_barrier_data *barrier);
2939 uint64_t radv_get_current_time(void);
2941 static inline uint32_t
2942 si_conv_gl_prim_to_vertices(enum shader_prim gl_prim)
2945 case SHADER_PRIM_POINTS:
2947 case SHADER_PRIM_LINES:
2948 case SHADER_PRIM_LINE_STRIP:
2950 case SHADER_PRIM_TRIANGLES:
2951 case SHADER_PRIM_TRIANGLE_STRIP:
2953 case SHADER_PRIM_LINES_ADJACENCY:
2955 case SHADER_PRIM_TRIANGLES_ADJACENCY:
2957 case SHADER_PRIM_QUADS:
2958 return V_028A6C_TRISTRIP;
2965 static inline uint32_t
2966 si_conv_prim_to_gs_out(uint32_t topology)
2969 case V_008958_DI_PT_POINTLIST:
2970 case V_008958_DI_PT_PATCH:
2971 return V_028A6C_POINTLIST;
2972 case V_008958_DI_PT_LINELIST:
2973 case V_008958_DI_PT_LINESTRIP:
2974 case V_008958_DI_PT_LINELIST_ADJ:
2975 case V_008958_DI_PT_LINESTRIP_ADJ:
2976 return V_028A6C_LINESTRIP;
2977 case V_008958_DI_PT_TRILIST:
2978 case V_008958_DI_PT_TRISTRIP:
2979 case V_008958_DI_PT_TRIFAN:
2980 case V_008958_DI_PT_TRILIST_ADJ:
2981 case V_008958_DI_PT_TRISTRIP_ADJ:
2982 return V_028A6C_TRISTRIP;
2989 static inline uint32_t
2990 si_translate_prim(unsigned topology)
2993 case VK_PRIMITIVE_TOPOLOGY_POINT_LIST:
2994 return V_008958_DI_PT_POINTLIST;
2995 case VK_PRIMITIVE_TOPOLOGY_LINE_LIST:
2996 return V_008958_DI_PT_LINELIST;
2997 case VK_PRIMITIVE_TOPOLOGY_LINE_STRIP:
2998 return V_008958_DI_PT_LINESTRIP;
2999 case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST:
3000 return V_008958_DI_PT_TRILIST;
3001 case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP:
3002 return V_008958_DI_PT_TRISTRIP;
3003 case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN:
3004 return V_008958_DI_PT_TRIFAN;
3005 case VK_PRIMITIVE_TOPOLOGY_LINE_LIST_WITH_ADJACENCY:
3006 return V_008958_DI_PT_LINELIST_ADJ;
3007 case VK_PRIMITIVE_TOPOLOGY_LINE_STRIP_WITH_ADJACENCY:
3008 return V_008958_DI_PT_LINESTRIP_ADJ;
3009 case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_WITH_ADJACENCY:
3010 return V_008958_DI_PT_TRILIST_ADJ;
3011 case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP_WITH_ADJACENCY:
3012 return V_008958_DI_PT_TRISTRIP_ADJ;
3013 case VK_PRIMITIVE_TOPOLOGY_PATCH_LIST:
3014 return V_008958_DI_PT_PATCH;
3016 unreachable("unhandled primitive type");
3020 static inline uint32_t
3021 si_translate_stencil_op(enum VkStencilOp op)
3024 case VK_STENCIL_OP_KEEP:
3025 return V_02842C_STENCIL_KEEP;
3026 case VK_STENCIL_OP_ZERO:
3027 return V_02842C_STENCIL_ZERO;
3028 case VK_STENCIL_OP_REPLACE:
3029 return V_02842C_STENCIL_REPLACE_TEST;
3030 case VK_STENCIL_OP_INCREMENT_AND_CLAMP:
3031 return V_02842C_STENCIL_ADD_CLAMP;
3032 case VK_STENCIL_OP_DECREMENT_AND_CLAMP:
3033 return V_02842C_STENCIL_SUB_CLAMP;
3034 case VK_STENCIL_OP_INVERT:
3035 return V_02842C_STENCIL_INVERT;
3036 case VK_STENCIL_OP_INCREMENT_AND_WRAP:
3037 return V_02842C_STENCIL_ADD_WRAP;
3038 case VK_STENCIL_OP_DECREMENT_AND_WRAP:
3039 return V_02842C_STENCIL_SUB_WRAP;
3045 static inline uint32_t
3046 si_translate_blend_logic_op(VkLogicOp op)
3049 case VK_LOGIC_OP_CLEAR:
3050 return V_028808_ROP3_CLEAR;
3051 case VK_LOGIC_OP_AND:
3052 return V_028808_ROP3_AND;
3053 case VK_LOGIC_OP_AND_REVERSE:
3054 return V_028808_ROP3_AND_REVERSE;
3055 case VK_LOGIC_OP_COPY:
3056 return V_028808_ROP3_COPY;
3057 case VK_LOGIC_OP_AND_INVERTED:
3058 return V_028808_ROP3_AND_INVERTED;
3059 case VK_LOGIC_OP_NO_OP:
3060 return V_028808_ROP3_NO_OP;
3061 case VK_LOGIC_OP_XOR:
3062 return V_028808_ROP3_XOR;
3063 case VK_LOGIC_OP_OR:
3064 return V_028808_ROP3_OR;
3065 case VK_LOGIC_OP_NOR:
3066 return V_028808_ROP3_NOR;
3067 case VK_LOGIC_OP_EQUIVALENT:
3068 return V_028808_ROP3_EQUIVALENT;
3069 case VK_LOGIC_OP_INVERT:
3070 return V_028808_ROP3_INVERT;
3071 case VK_LOGIC_OP_OR_REVERSE:
3072 return V_028808_ROP3_OR_REVERSE;
3073 case VK_LOGIC_OP_COPY_INVERTED:
3074 return V_028808_ROP3_COPY_INVERTED;
3075 case VK_LOGIC_OP_OR_INVERTED:
3076 return V_028808_ROP3_OR_INVERTED;
3077 case VK_LOGIC_OP_NAND:
3078 return V_028808_ROP3_NAND;
3079 case VK_LOGIC_OP_SET:
3080 return V_028808_ROP3_SET;
3082 unreachable("Unhandled logic op");
3087 * Queue helper to get ring.
3088 * placed here as it needs queue + device structs.
3090 static inline enum amd_ip_type
3091 radv_queue_ring(struct radv_queue *queue)
3093 return radv_queue_family_to_ring(queue->device->physical_device, queue->state.qf);
3097 * Helper used for debugging compiler issues by enabling/disabling LLVM for a
3098 * specific shader stage (developers only).
3101 radv_use_llvm_for_stage(struct radv_device *device, UNUSED gl_shader_stage stage)
3103 return device->physical_device->use_llvm;
3107 radv_has_shader_buffer_float_minmax(const struct radv_physical_device *pdevice)
3109 return (pdevice->rad_info.gfx_level <= GFX7 && !pdevice->use_llvm) ||
3110 pdevice->rad_info.gfx_level >= GFX10;
3113 struct radv_acceleration_structure {
3114 struct vk_object_base base;
3116 struct radeon_winsys_bo *bo;
3117 uint64_t mem_offset;
3121 static inline uint64_t
3122 radv_accel_struct_get_va(const struct radv_acceleration_structure *accel)
3124 return radv_buffer_get_va(accel->bo) + accel->mem_offset;
3127 /* radv_perfcounter.c */
3128 void radv_perfcounter_emit_shaders(struct radeon_cmdbuf *cs, unsigned shaders);
3129 void radv_perfcounter_emit_spm_reset(struct radeon_cmdbuf *cs);
3130 void radv_perfcounter_emit_spm_start(struct radv_device *device, struct radeon_cmdbuf *cs,
3132 void radv_perfcounter_emit_spm_stop(struct radv_device *device, struct radeon_cmdbuf *cs,
3136 bool radv_spm_init(struct radv_device *device);
3137 void radv_spm_finish(struct radv_device *device);
3138 void radv_emit_spm_setup(struct radv_device *device, struct radeon_cmdbuf *cs);
3140 #define RADV_FROM_HANDLE(__radv_type, __name, __handle) \
3141 VK_FROM_HANDLE(__radv_type, __name, __handle)
3143 VK_DEFINE_HANDLE_CASTS(radv_cmd_buffer, vk.base, VkCommandBuffer,
3144 VK_OBJECT_TYPE_COMMAND_BUFFER)
3145 VK_DEFINE_HANDLE_CASTS(radv_device, vk.base, VkDevice, VK_OBJECT_TYPE_DEVICE)
3146 VK_DEFINE_HANDLE_CASTS(radv_instance, vk.base, VkInstance, VK_OBJECT_TYPE_INSTANCE)
3147 VK_DEFINE_HANDLE_CASTS(radv_physical_device, vk.base, VkPhysicalDevice,
3148 VK_OBJECT_TYPE_PHYSICAL_DEVICE)
3149 VK_DEFINE_HANDLE_CASTS(radv_queue, vk.base, VkQueue, VK_OBJECT_TYPE_QUEUE)
3150 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_acceleration_structure, base,
3151 VkAccelerationStructureKHR,
3152 VK_OBJECT_TYPE_ACCELERATION_STRUCTURE_KHR)
3153 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_cmd_pool, vk.base, VkCommandPool,
3154 VK_OBJECT_TYPE_COMMAND_POOL)
3155 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer, vk.base, VkBuffer, VK_OBJECT_TYPE_BUFFER)
3156 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer_view, base, VkBufferView,
3157 VK_OBJECT_TYPE_BUFFER_VIEW)
3158 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_pool, base, VkDescriptorPool,
3159 VK_OBJECT_TYPE_DESCRIPTOR_POOL)
3160 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set, header.base, VkDescriptorSet,
3161 VK_OBJECT_TYPE_DESCRIPTOR_SET)
3162 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set_layout, base,
3163 VkDescriptorSetLayout,
3164 VK_OBJECT_TYPE_DESCRIPTOR_SET_LAYOUT)
3165 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_update_template, base,
3166 VkDescriptorUpdateTemplate,
3167 VK_OBJECT_TYPE_DESCRIPTOR_UPDATE_TEMPLATE)
3168 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_device_memory, base, VkDeviceMemory,
3169 VK_OBJECT_TYPE_DEVICE_MEMORY)
3170 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_event, base, VkEvent, VK_OBJECT_TYPE_EVENT)
3171 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_image, vk.base, VkImage, VK_OBJECT_TYPE_IMAGE)
3172 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_image_view, vk.base, VkImageView,
3173 VK_OBJECT_TYPE_IMAGE_VIEW);
3174 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_cache, base, VkPipelineCache,
3175 VK_OBJECT_TYPE_PIPELINE_CACHE)
3176 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline, base, VkPipeline,
3177 VK_OBJECT_TYPE_PIPELINE)
3178 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_layout, base, VkPipelineLayout,
3179 VK_OBJECT_TYPE_PIPELINE_LAYOUT)
3180 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_query_pool, base, VkQueryPool,
3181 VK_OBJECT_TYPE_QUERY_POOL)
3182 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_render_pass, base, VkRenderPass,
3183 VK_OBJECT_TYPE_RENDER_PASS)
3184 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_sampler, base, VkSampler,
3185 VK_OBJECT_TYPE_SAMPLER)
3186 VK_DEFINE_NONDISP_HANDLE_CASTS(radv_sampler_ycbcr_conversion, base,
3187 VkSamplerYcbcrConversion,
3188 VK_OBJECT_TYPE_SAMPLER_YCBCR_CONVERSION)
3194 #endif /* RADV_PRIVATE_H */