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
47 #include "compiler/shader_enums.h"
48 #include "util/macros.h"
49 #include "util/list.h"
50 #include "util/vk_alloc.h"
51 #include "main/macros.h"
53 #include "radv_radeon_winsys.h"
54 #include "ac_binary.h"
55 #include "ac_nir_to_llvm.h"
56 #include "radv_debug.h"
57 #include "radv_descriptor_set.h"
59 #include <llvm-c/TargetMachine.h>
61 /* Pre-declarations needed for WSI entrypoints */
64 typedef struct xcb_connection_t xcb_connection_t;
65 typedef uint32_t xcb_visualid_t;
66 typedef uint32_t xcb_window_t;
68 #include <vulkan/vulkan.h>
69 #include <vulkan/vulkan_intel.h>
70 #include <vulkan/vk_icd.h>
72 #include "radv_entrypoints.h"
74 #include "wsi_common.h"
77 #define MAX_VERTEX_ATTRIBS 32
79 #define MAX_VIEWPORTS 16
80 #define MAX_SCISSORS 16
81 #define MAX_PUSH_CONSTANTS_SIZE 128
82 #define MAX_DYNAMIC_BUFFERS 16
83 #define MAX_SAMPLES_LOG2 4
84 #define NUM_META_FS_KEYS 11
85 #define RADV_MAX_DRM_DEVICES 8
87 #define NUM_DEPTH_CLEAR_PIPELINES 3
91 RADV_MEM_HEAP_VRAM_CPU_ACCESS,
98 RADV_MEM_TYPE_GTT_WRITE_COMBINE,
99 RADV_MEM_TYPE_VRAM_CPU_ACCESS,
100 RADV_MEM_TYPE_GTT_CACHED,
104 #define radv_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
106 static inline uint32_t
107 align_u32(uint32_t v, uint32_t a)
109 assert(a != 0 && a == (a & -a));
110 return (v + a - 1) & ~(a - 1);
113 static inline uint32_t
114 align_u32_npot(uint32_t v, uint32_t a)
116 return (v + a - 1) / a * a;
119 static inline uint64_t
120 align_u64(uint64_t v, uint64_t a)
122 assert(a != 0 && a == (a & -a));
123 return (v + a - 1) & ~(a - 1);
126 static inline int32_t
127 align_i32(int32_t v, int32_t a)
129 assert(a != 0 && a == (a & -a));
130 return (v + a - 1) & ~(a - 1);
133 /** Alignment must be a power of 2. */
135 radv_is_aligned(uintmax_t n, uintmax_t a)
137 assert(a == (a & -a));
138 return (n & (a - 1)) == 0;
141 static inline uint32_t
142 round_up_u32(uint32_t v, uint32_t a)
144 return (v + a - 1) / a;
147 static inline uint64_t
148 round_up_u64(uint64_t v, uint64_t a)
150 return (v + a - 1) / a;
153 static inline uint32_t
154 radv_minify(uint32_t n, uint32_t levels)
156 if (unlikely(n == 0))
159 return MAX2(n >> levels, 1);
162 radv_clamp_f(float f, float min, float max)
175 radv_clear_mask(uint32_t *inout_mask, uint32_t clear_mask)
177 if (*inout_mask & clear_mask) {
178 *inout_mask &= ~clear_mask;
185 #define for_each_bit(b, dword) \
186 for (uint32_t __dword = (dword); \
187 (b) = __builtin_ffs(__dword) - 1, __dword; \
188 __dword &= ~(1 << (b)))
190 #define typed_memcpy(dest, src, count) ({ \
191 STATIC_ASSERT(sizeof(*src) == sizeof(*dest)); \
192 memcpy((dest), (src), (count) * sizeof(*(src))); \
195 #define zero(x) (memset(&(x), 0, sizeof(x)))
197 /* Whenever we generate an error, pass it through this function. Useful for
198 * debugging, where we can break on it. Only call at error site, not when
199 * propagating errors. Might be useful to plug in a stack trace here.
202 VkResult __vk_errorf(VkResult error, const char *file, int line, const char *format, ...);
205 #define vk_error(error) __vk_errorf(error, __FILE__, __LINE__, NULL);
206 #define vk_errorf(error, format, ...) __vk_errorf(error, __FILE__, __LINE__, format, ## __VA_ARGS__);
208 #define vk_error(error) error
209 #define vk_errorf(error, format, ...) error
212 void __radv_finishme(const char *file, int line, const char *format, ...)
213 radv_printflike(3, 4);
214 void radv_loge(const char *format, ...) radv_printflike(1, 2);
215 void radv_loge_v(const char *format, va_list va);
218 * Print a FINISHME message, including its source location.
220 #define radv_finishme(format, ...) \
222 static bool reported = false; \
224 __radv_finishme(__FILE__, __LINE__, format, ##__VA_ARGS__); \
229 /* A non-fatal assert. Useful for debugging. */
231 #define radv_assert(x) ({ \
232 if (unlikely(!(x))) \
233 fprintf(stderr, "%s:%d ASSERT: %s\n", __FILE__, __LINE__, #x); \
236 #define radv_assert(x)
239 #define stub_return(v) \
241 radv_finishme("stub %s", __func__); \
247 radv_finishme("stub %s", __func__); \
251 void *radv_lookup_entrypoint(const char *name);
253 struct radv_extensions {
254 VkExtensionProperties *ext_array;
258 struct radv_physical_device {
259 VK_LOADER_DATA _loader_data;
261 struct radv_instance * instance;
263 struct radeon_winsys *ws;
264 struct radeon_info rad_info;
267 uint8_t uuid[VK_UUID_SIZE];
270 struct wsi_device wsi_device;
271 struct radv_extensions extensions;
274 struct radv_instance {
275 VK_LOADER_DATA _loader_data;
277 VkAllocationCallbacks alloc;
280 int physicalDeviceCount;
281 struct radv_physical_device physicalDevices[RADV_MAX_DRM_DEVICES];
283 uint64_t debug_flags;
286 VkResult radv_init_wsi(struct radv_physical_device *physical_device);
287 void radv_finish_wsi(struct radv_physical_device *physical_device);
291 struct radv_pipeline_cache {
292 struct radv_device * device;
293 pthread_mutex_t mutex;
297 uint32_t kernel_count;
298 struct cache_entry ** hash_table;
301 VkAllocationCallbacks alloc;
305 radv_pipeline_cache_init(struct radv_pipeline_cache *cache,
306 struct radv_device *device);
308 radv_pipeline_cache_finish(struct radv_pipeline_cache *cache);
310 radv_pipeline_cache_load(struct radv_pipeline_cache *cache,
311 const void *data, size_t size);
313 struct radv_shader_variant *
314 radv_create_shader_variant_from_pipeline_cache(struct radv_device *device,
315 struct radv_pipeline_cache *cache,
316 const unsigned char *sha1);
318 struct radv_shader_variant *
319 radv_pipeline_cache_insert_shader(struct radv_pipeline_cache *cache,
320 const unsigned char *sha1,
321 struct radv_shader_variant *variant,
322 const void *code, unsigned code_size);
324 void radv_shader_variant_destroy(struct radv_device *device,
325 struct radv_shader_variant *variant);
327 struct radv_meta_state {
328 VkAllocationCallbacks alloc;
330 struct radv_pipeline_cache cache;
333 * Use array element `i` for images with `2^i` samples.
336 VkRenderPass render_pass[NUM_META_FS_KEYS];
337 struct radv_pipeline *color_pipelines[NUM_META_FS_KEYS];
339 VkRenderPass depthstencil_rp;
340 struct radv_pipeline *depth_only_pipeline[NUM_DEPTH_CLEAR_PIPELINES];
341 struct radv_pipeline *stencil_only_pipeline[NUM_DEPTH_CLEAR_PIPELINES];
342 struct radv_pipeline *depthstencil_pipeline[NUM_DEPTH_CLEAR_PIPELINES];
343 } clear[1 + MAX_SAMPLES_LOG2];
346 VkRenderPass render_pass[NUM_META_FS_KEYS];
348 /** Pipeline that blits from a 1D image. */
349 VkPipeline pipeline_1d_src[NUM_META_FS_KEYS];
351 /** Pipeline that blits from a 2D image. */
352 VkPipeline pipeline_2d_src[NUM_META_FS_KEYS];
354 /** Pipeline that blits from a 3D image. */
355 VkPipeline pipeline_3d_src[NUM_META_FS_KEYS];
357 VkRenderPass depth_only_rp;
358 VkPipeline depth_only_1d_pipeline;
359 VkPipeline depth_only_2d_pipeline;
360 VkPipeline depth_only_3d_pipeline;
362 VkRenderPass stencil_only_rp;
363 VkPipeline stencil_only_1d_pipeline;
364 VkPipeline stencil_only_2d_pipeline;
365 VkPipeline stencil_only_3d_pipeline;
366 VkPipelineLayout pipeline_layout;
367 VkDescriptorSetLayout ds_layout;
371 VkRenderPass render_passes[NUM_META_FS_KEYS];
373 VkPipelineLayout p_layouts[2];
374 VkDescriptorSetLayout ds_layouts[2];
375 VkPipeline pipelines[2][NUM_META_FS_KEYS];
377 VkRenderPass depth_only_rp;
378 VkPipeline depth_only_pipeline[2];
380 VkRenderPass stencil_only_rp;
381 VkPipeline stencil_only_pipeline[2];
385 VkPipelineLayout img_p_layout;
386 VkDescriptorSetLayout img_ds_layout;
390 VkRenderPass render_pass;
391 VkPipelineLayout img_p_layout;
392 VkDescriptorSetLayout img_ds_layout;
396 VkPipelineLayout img_p_layout;
397 VkDescriptorSetLayout img_ds_layout;
401 VkPipelineLayout img_p_layout;
402 VkDescriptorSetLayout img_ds_layout;
412 VkDescriptorSetLayout ds_layout;
413 VkPipelineLayout p_layout;
416 VkPipeline i_pipeline;
417 } rc[MAX_SAMPLES_LOG2];
421 VkPipeline decompress_pipeline;
422 VkPipeline resummarize_pipeline;
427 VkPipeline cmask_eliminate_pipeline;
428 VkPipeline fmask_decompress_pipeline;
433 VkPipelineLayout fill_p_layout;
434 VkPipelineLayout copy_p_layout;
435 VkDescriptorSetLayout fill_ds_layout;
436 VkDescriptorSetLayout copy_ds_layout;
437 VkPipeline fill_pipeline;
438 VkPipeline copy_pipeline;
443 #define RADV_QUEUE_GENERAL 0
444 #define RADV_QUEUE_COMPUTE 1
445 #define RADV_QUEUE_TRANSFER 2
447 #define RADV_MAX_QUEUE_FAMILIES 3
449 enum ring_type radv_queue_family_to_ring(int f);
452 VK_LOADER_DATA _loader_data;
453 struct radv_device * device;
454 struct radeon_winsys_ctx *hw_ctx;
455 int queue_family_index;
458 uint32_t scratch_size;
459 uint32_t compute_scratch_size;
460 uint32_t esgs_ring_size;
461 uint32_t gsvs_ring_size;
463 struct radeon_winsys_bo *scratch_bo;
464 struct radeon_winsys_bo *descriptor_bo;
465 struct radeon_winsys_bo *compute_scratch_bo;
466 struct radeon_winsys_bo *esgs_ring_bo;
467 struct radeon_winsys_bo *gsvs_ring_bo;
468 struct radeon_winsys_cs *initial_preamble_cs;
469 struct radeon_winsys_cs *continue_preamble_cs;
473 VK_LOADER_DATA _loader_data;
475 VkAllocationCallbacks alloc;
477 struct radv_instance * instance;
478 struct radeon_winsys *ws;
480 struct radv_meta_state meta_state;
482 struct radv_queue *queues[RADV_MAX_QUEUE_FAMILIES];
483 int queue_count[RADV_MAX_QUEUE_FAMILIES];
484 struct radeon_winsys_cs *empty_cs[RADV_MAX_QUEUE_FAMILIES];
485 struct radeon_winsys_cs *flush_cs[RADV_MAX_QUEUE_FAMILIES];
487 uint64_t debug_flags;
489 bool llvm_supports_spill;
490 uint32_t scratch_waves;
492 uint32_t gs_table_depth;
494 /* MSAA sample locations.
495 * The first index is the sample index.
496 * The second index is the coordinate: X, Y. */
497 float sample_locations_1x[1][2];
498 float sample_locations_2x[2][2];
499 float sample_locations_4x[4][2];
500 float sample_locations_8x[8][2];
501 float sample_locations_16x[16][2];
504 uint32_t gfx_init_size_dw;
505 struct radeon_winsys_bo *gfx_init;
507 struct radeon_winsys_bo *trace_bo;
508 uint32_t *trace_id_ptr;
510 struct radv_physical_device *physical_device;
512 /* Backup in-memory cache to be used if the app doesn't provide one */
513 struct radv_pipeline_cache * mem_cache;
516 struct radv_device_memory {
517 struct radeon_winsys_bo *bo;
518 /* for dedicated allocations */
519 struct radv_image *image;
520 struct radv_buffer *buffer;
522 VkDeviceSize map_size;
527 struct radv_descriptor_range {
532 struct radv_descriptor_set {
533 const struct radv_descriptor_set_layout *layout;
536 struct radeon_winsys_bo *bo;
538 uint32_t *mapped_ptr;
539 struct radv_descriptor_range *dynamic_descriptors;
541 struct list_head vram_list;
543 struct radeon_winsys_bo *descriptors[0];
546 struct radv_descriptor_pool {
547 struct radeon_winsys_bo *bo;
549 uint64_t current_offset;
552 struct list_head vram_list;
556 struct radv_device * device;
559 VkBufferUsageFlags usage;
562 struct radeon_winsys_bo * bo;
567 enum radv_cmd_dirty_bits {
568 RADV_CMD_DIRTY_DYNAMIC_VIEWPORT = 1 << 0, /* VK_DYNAMIC_STATE_VIEWPORT */
569 RADV_CMD_DIRTY_DYNAMIC_SCISSOR = 1 << 1, /* VK_DYNAMIC_STATE_SCISSOR */
570 RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH = 1 << 2, /* VK_DYNAMIC_STATE_LINE_WIDTH */
571 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS = 1 << 3, /* VK_DYNAMIC_STATE_DEPTH_BIAS */
572 RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS = 1 << 4, /* VK_DYNAMIC_STATE_BLEND_CONSTANTS */
573 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS = 1 << 5, /* VK_DYNAMIC_STATE_DEPTH_BOUNDS */
574 RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK = 1 << 6, /* VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK */
575 RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK = 1 << 7, /* VK_DYNAMIC_STATE_STENCIL_WRITE_MASK */
576 RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE = 1 << 8, /* VK_DYNAMIC_STATE_STENCIL_REFERENCE */
577 RADV_CMD_DIRTY_DYNAMIC_ALL = (1 << 9) - 1,
578 RADV_CMD_DIRTY_PIPELINE = 1 << 9,
579 RADV_CMD_DIRTY_INDEX_BUFFER = 1 << 10,
580 RADV_CMD_DIRTY_RENDER_TARGETS = 1 << 11,
582 typedef uint32_t radv_cmd_dirty_mask_t;
584 enum radv_cmd_flush_bits {
585 RADV_CMD_FLAG_INV_ICACHE = 1 << 0,
586 /* SMEM L1, other names: KCACHE, constant cache, DCACHE, data cache */
587 RADV_CMD_FLAG_INV_SMEM_L1 = 1 << 1,
588 /* VMEM L1 can optionally be bypassed (GLC=1). Other names: TC L1 */
589 RADV_CMD_FLAG_INV_VMEM_L1 = 1 << 2,
590 /* Used by everything except CB/DB, can be bypassed (SLC=1). Other names: TC L2 */
591 RADV_CMD_FLAG_INV_GLOBAL_L2 = 1 << 3,
592 /* Same as above, but only writes back and doesn't invalidate */
593 RADV_CMD_FLAG_WRITEBACK_GLOBAL_L2 = 1 << 4,
594 /* Framebuffer caches */
595 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META = 1 << 5,
596 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META = 1 << 6,
597 RADV_CMD_FLAG_FLUSH_AND_INV_DB = 1 << 7,
598 RADV_CMD_FLAG_FLUSH_AND_INV_CB = 1 << 8,
599 /* Engine synchronization. */
600 RADV_CMD_FLAG_VS_PARTIAL_FLUSH = 1 << 9,
601 RADV_CMD_FLAG_PS_PARTIAL_FLUSH = 1 << 10,
602 RADV_CMD_FLAG_CS_PARTIAL_FLUSH = 1 << 11,
603 RADV_CMD_FLAG_VGT_FLUSH = 1 << 12,
605 RADV_CMD_FLUSH_AND_INV_FRAMEBUFFER = (RADV_CMD_FLAG_FLUSH_AND_INV_CB |
606 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META |
607 RADV_CMD_FLAG_FLUSH_AND_INV_DB |
608 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META)
611 struct radv_vertex_binding {
612 struct radv_buffer * buffer;
616 struct radv_dynamic_state {
619 VkViewport viewports[MAX_VIEWPORTS];
624 VkRect2D scissors[MAX_SCISSORS];
635 float blend_constants[4];
645 } stencil_compare_mask;
650 } stencil_write_mask;
658 extern const struct radv_dynamic_state default_dynamic_state;
660 void radv_dynamic_state_copy(struct radv_dynamic_state *dest,
661 const struct radv_dynamic_state *src,
664 * Attachment state when recording a renderpass instance.
666 * The clear value is valid only if there exists a pending clear.
668 struct radv_attachment_state {
669 VkImageAspectFlags pending_clear_aspects;
670 VkClearValue clear_value;
671 VkImageLayout current_layout;
674 struct radv_cmd_state {
676 radv_cmd_dirty_mask_t dirty;
677 bool vertex_descriptors_dirty;
679 struct radv_pipeline * pipeline;
680 struct radv_pipeline * emitted_pipeline;
681 struct radv_pipeline * compute_pipeline;
682 struct radv_pipeline * emitted_compute_pipeline;
683 struct radv_framebuffer * framebuffer;
684 struct radv_render_pass * pass;
685 const struct radv_subpass * subpass;
686 struct radv_dynamic_state dynamic;
687 struct radv_vertex_binding vertex_bindings[MAX_VBS];
688 struct radv_descriptor_set * descriptors[MAX_SETS];
689 struct radv_attachment_state * attachments;
690 VkRect2D render_area;
691 struct radv_buffer * index_buffer;
693 uint32_t index_offset;
694 uint32_t last_primitive_reset_index;
695 enum radv_cmd_flush_bits flush_bits;
696 unsigned active_occlusion_queries;
698 uint32_t descriptors_dirty;
700 uint32_t last_ia_multi_vgt_param;
703 struct radv_cmd_pool {
704 VkAllocationCallbacks alloc;
705 struct list_head cmd_buffers;
706 struct list_head free_cmd_buffers;
707 uint32_t queue_family_index;
710 struct radv_cmd_buffer_upload {
714 struct radeon_winsys_bo *upload_bo;
715 struct list_head list;
718 struct radv_cmd_buffer {
719 VK_LOADER_DATA _loader_data;
721 struct radv_device * device;
723 struct radv_cmd_pool * pool;
724 struct list_head pool_link;
726 VkCommandBufferUsageFlags usage_flags;
727 VkCommandBufferLevel level;
728 struct radeon_winsys_cs *cs;
729 struct radv_cmd_state state;
730 uint32_t queue_family_index;
732 uint8_t push_constants[MAX_PUSH_CONSTANTS_SIZE];
733 uint32_t dynamic_buffers[4 * MAX_DYNAMIC_BUFFERS];
734 VkShaderStageFlags push_constant_stages;
736 struct radv_cmd_buffer_upload upload;
740 uint32_t scratch_size_needed;
741 uint32_t compute_scratch_size_needed;
742 uint32_t esgs_ring_size_needed;
743 uint32_t gsvs_ring_size_needed;
745 int ring_offsets_idx; /* just used for verification */
750 bool radv_cmd_buffer_uses_mec(struct radv_cmd_buffer *cmd_buffer);
752 void si_init_compute(struct radv_cmd_buffer *cmd_buffer);
753 void si_init_config(struct radv_cmd_buffer *cmd_buffer);
755 void cik_create_gfx_config(struct radv_device *device);
757 void si_write_viewport(struct radeon_winsys_cs *cs, int first_vp,
758 int count, const VkViewport *viewports);
759 void si_write_scissors(struct radeon_winsys_cs *cs, int first,
760 int count, const VkRect2D *scissors);
761 uint32_t si_get_ia_multi_vgt_param(struct radv_cmd_buffer *cmd_buffer,
762 bool instanced_draw, bool indirect_draw,
763 uint32_t draw_vertex_count);
764 void si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
765 enum chip_class chip_class,
767 enum radv_cmd_flush_bits flush_bits);
768 void si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
769 enum chip_class chip_class,
771 enum radv_cmd_flush_bits flush_bits);
772 void si_emit_cache_flush(struct radv_cmd_buffer *cmd_buffer);
773 void si_cp_dma_buffer_copy(struct radv_cmd_buffer *cmd_buffer,
774 uint64_t src_va, uint64_t dest_va,
776 void si_cp_dma_clear_buffer(struct radv_cmd_buffer *cmd_buffer, uint64_t va,
777 uint64_t size, unsigned value);
778 void radv_set_db_count_control(struct radv_cmd_buffer *cmd_buffer);
779 void radv_bind_descriptor_set(struct radv_cmd_buffer *cmd_buffer,
780 struct radv_descriptor_set *set,
783 radv_cmd_buffer_upload_alloc(struct radv_cmd_buffer *cmd_buffer,
786 unsigned *out_offset,
789 radv_cmd_buffer_set_subpass(struct radv_cmd_buffer *cmd_buffer,
790 const struct radv_subpass *subpass,
793 radv_cmd_buffer_upload_data(struct radv_cmd_buffer *cmd_buffer,
794 unsigned size, unsigned alignmnet,
795 const void *data, unsigned *out_offset);
797 radv_emit_framebuffer_state(struct radv_cmd_buffer *cmd_buffer);
798 void radv_cmd_buffer_clear_subpass(struct radv_cmd_buffer *cmd_buffer);
799 void radv_cmd_buffer_resolve_subpass(struct radv_cmd_buffer *cmd_buffer);
800 void radv_cayman_emit_msaa_sample_locs(struct radeon_winsys_cs *cs, int nr_samples);
801 unsigned radv_cayman_get_maxdist(int log_samples);
802 void radv_device_init_msaa(struct radv_device *device);
803 void radv_set_depth_clear_regs(struct radv_cmd_buffer *cmd_buffer,
804 struct radv_image *image,
805 VkClearDepthStencilValue ds_clear_value,
806 VkImageAspectFlags aspects);
807 void radv_set_color_clear_regs(struct radv_cmd_buffer *cmd_buffer,
808 struct radv_image *image,
810 uint32_t color_values[2]);
811 void radv_fill_buffer(struct radv_cmd_buffer *cmd_buffer,
812 struct radeon_winsys_bo *bo,
813 uint64_t offset, uint64_t size, uint32_t value);
814 void radv_cmd_buffer_trace_emit(struct radv_cmd_buffer *cmd_buffer);
815 bool radv_get_memory_fd(struct radv_device *device,
816 struct radv_device_memory *memory,
819 * Takes x,y,z as exact numbers of invocations, instead of blocks.
821 * Limitations: Can't call normal dispatch functions without binding or rebinding
822 * the compute pipeline.
824 void radv_unaligned_dispatch(
825 struct radv_cmd_buffer *cmd_buffer,
831 struct radeon_winsys_bo *bo;
837 struct radv_shader_module {
838 struct nir_shader * nir;
839 unsigned char sha1[20];
844 union ac_shader_variant_key;
847 radv_hash_shader(unsigned char *hash, struct radv_shader_module *module,
848 const char *entrypoint,
849 const VkSpecializationInfo *spec_info,
850 const struct radv_pipeline_layout *layout,
851 const union ac_shader_variant_key *key,
852 uint32_t is_geom_copy_shader);
854 static inline gl_shader_stage
855 vk_to_mesa_shader_stage(VkShaderStageFlagBits vk_stage)
857 assert(__builtin_popcount(vk_stage) == 1);
858 return ffs(vk_stage) - 1;
861 static inline VkShaderStageFlagBits
862 mesa_to_vk_shader_stage(gl_shader_stage mesa_stage)
864 return (1 << mesa_stage);
867 #define RADV_STAGE_MASK ((1 << MESA_SHADER_STAGES) - 1)
869 #define radv_foreach_stage(stage, stage_bits) \
870 for (gl_shader_stage stage, \
871 __tmp = (gl_shader_stage)((stage_bits) & RADV_STAGE_MASK); \
872 stage = __builtin_ffs(__tmp) - 1, __tmp; \
873 __tmp &= ~(1 << (stage)))
875 struct radv_shader_variant {
878 struct radeon_winsys_bo *bo;
879 struct ac_shader_config config;
880 struct ac_shader_variant_info info;
886 struct radv_depth_stencil_state {
887 uint32_t db_depth_control;
888 uint32_t db_stencil_control;
889 uint32_t db_render_control;
890 uint32_t db_render_override2;
893 struct radv_blend_state {
894 uint32_t cb_color_control;
895 uint32_t cb_target_mask;
896 uint32_t sx_mrt0_blend_opt[8];
897 uint32_t cb_blend_control[8];
899 uint32_t spi_shader_col_format;
900 uint32_t cb_shader_mask;
901 uint32_t db_alpha_to_mask;
904 unsigned radv_format_meta_fs_key(VkFormat format);
906 struct radv_raster_state {
907 uint32_t pa_cl_clip_cntl;
908 uint32_t pa_cl_vs_out_cntl;
909 uint32_t spi_interp_control;
910 uint32_t pa_su_point_size;
911 uint32_t pa_su_point_minmax;
912 uint32_t pa_su_line_cntl;
913 uint32_t pa_su_vtx_cntl;
914 uint32_t pa_su_sc_mode_cntl;
917 struct radv_multisample_state {
919 uint32_t pa_sc_line_cntl;
920 uint32_t pa_sc_mode_cntl_0;
921 uint32_t pa_sc_mode_cntl_1;
922 uint32_t pa_sc_aa_config;
923 uint32_t pa_sc_aa_mask[2];
924 unsigned num_samples;
927 struct radv_prim_vertex_count {
932 struct radv_pipeline {
933 struct radv_device * device;
934 uint32_t dynamic_state_mask;
935 struct radv_dynamic_state dynamic_state;
937 struct radv_pipeline_layout * layout;
939 bool needs_data_cache;
941 struct radv_shader_variant * shaders[MESA_SHADER_STAGES];
942 struct radv_shader_variant *gs_copy_shader;
943 VkShaderStageFlags active_stages;
945 uint32_t va_rsrc_word3[MAX_VERTEX_ATTRIBS];
946 uint32_t va_format_size[MAX_VERTEX_ATTRIBS];
947 uint32_t va_binding[MAX_VERTEX_ATTRIBS];
948 uint32_t va_offset[MAX_VERTEX_ATTRIBS];
949 uint32_t num_vertex_attribs;
950 uint32_t binding_stride[MAX_VBS];
954 struct radv_blend_state blend;
955 struct radv_depth_stencil_state ds;
956 struct radv_raster_state raster;
957 struct radv_multisample_state ms;
960 bool prim_restart_enable;
961 unsigned esgs_ring_size;
962 unsigned gsvs_ring_size;
963 struct radv_prim_vertex_count prim_vertex_count;
968 unsigned scratch_bytes_per_wave;
971 static inline bool radv_pipeline_has_gs(struct radv_pipeline *pipeline)
973 return pipeline->shaders[MESA_SHADER_GEOMETRY] ? true : false;
976 struct radv_graphics_pipeline_create_info {
979 bool db_stencil_clear;
980 bool db_depth_disable_expclear;
981 bool db_stencil_disable_expclear;
982 bool db_flush_depth_inplace;
983 bool db_flush_stencil_inplace;
985 uint32_t custom_blend_mode;
989 radv_pipeline_init(struct radv_pipeline *pipeline, struct radv_device *device,
990 struct radv_pipeline_cache *cache,
991 const VkGraphicsPipelineCreateInfo *pCreateInfo,
992 const struct radv_graphics_pipeline_create_info *extra,
993 const VkAllocationCallbacks *alloc);
996 radv_graphics_pipeline_create(VkDevice device,
997 VkPipelineCache cache,
998 const VkGraphicsPipelineCreateInfo *pCreateInfo,
999 const struct radv_graphics_pipeline_create_info *extra,
1000 const VkAllocationCallbacks *alloc,
1001 VkPipeline *pPipeline);
1003 struct vk_format_description;
1004 uint32_t radv_translate_buffer_dataformat(const struct vk_format_description *desc,
1005 int first_non_void);
1006 uint32_t radv_translate_buffer_numformat(const struct vk_format_description *desc,
1007 int first_non_void);
1008 uint32_t radv_translate_colorformat(VkFormat format);
1009 uint32_t radv_translate_color_numformat(VkFormat format,
1010 const struct vk_format_description *desc,
1011 int first_non_void);
1012 uint32_t radv_colorformat_endian_swap(uint32_t colorformat);
1013 unsigned radv_translate_colorswap(VkFormat format, bool do_endian_swap);
1014 uint32_t radv_translate_dbformat(VkFormat format);
1015 uint32_t radv_translate_tex_dataformat(VkFormat format,
1016 const struct vk_format_description *desc,
1017 int first_non_void);
1018 uint32_t radv_translate_tex_numformat(VkFormat format,
1019 const struct vk_format_description *desc,
1020 int first_non_void);
1021 bool radv_format_pack_clear_color(VkFormat format,
1022 uint32_t clear_vals[2],
1023 VkClearColorValue *value);
1024 bool radv_is_colorbuffer_format_supported(VkFormat format, bool *blendable);
1026 struct radv_fmask_info {
1030 unsigned pitch_in_pixels;
1031 unsigned bank_height;
1032 unsigned slice_tile_max;
1033 unsigned tile_mode_index;
1036 struct radv_cmask_info {
1040 unsigned slice_tile_max;
1041 unsigned base_address_reg;
1044 struct r600_htile_info {
1055 /* The original VkFormat provided by the client. This may not match any
1056 * of the actual surface formats.
1059 VkImageAspectFlags aspects;
1062 uint32_t array_size;
1063 uint32_t samples; /**< VkImageCreateInfo::samples */
1064 VkImageUsageFlags usage; /**< Superset of VkImageCreateInfo::usage. */
1065 VkImageTiling tiling; /** VkImageCreateInfo::tiling */
1071 unsigned queue_family_mask;
1073 /* Set when bound */
1074 struct radeon_winsys_bo *bo;
1075 VkDeviceSize offset;
1076 uint32_t dcc_offset;
1077 uint32_t htile_offset;
1078 struct radeon_surf surface;
1080 struct radv_fmask_info fmask;
1081 struct radv_cmask_info cmask;
1082 uint32_t clear_value_offset;
1085 bool radv_layout_has_htile(const struct radv_image *image,
1086 VkImageLayout layout);
1087 bool radv_layout_is_htile_compressed(const struct radv_image *image,
1088 VkImageLayout layout);
1089 bool radv_layout_can_expclear(const struct radv_image *image,
1090 VkImageLayout layout);
1091 bool radv_layout_can_fast_clear(const struct radv_image *image,
1092 VkImageLayout layout,
1093 unsigned queue_mask);
1096 unsigned radv_image_queue_family_mask(const struct radv_image *image, uint32_t family, uint32_t queue_family);
1098 static inline uint32_t
1099 radv_get_layerCount(const struct radv_image *image,
1100 const VkImageSubresourceRange *range)
1102 return range->layerCount == VK_REMAINING_ARRAY_LAYERS ?
1103 image->array_size - range->baseArrayLayer : range->layerCount;
1106 static inline uint32_t
1107 radv_get_levelCount(const struct radv_image *image,
1108 const VkImageSubresourceRange *range)
1110 return range->levelCount == VK_REMAINING_MIP_LEVELS ?
1111 image->levels - range->baseMipLevel : range->levelCount;
1114 struct radeon_bo_metadata;
1116 radv_init_metadata(struct radv_device *device,
1117 struct radv_image *image,
1118 struct radeon_bo_metadata *metadata);
1120 struct radv_image_view {
1121 struct radv_image *image; /**< VkImageViewCreateInfo::image */
1122 struct radeon_winsys_bo *bo;
1124 VkImageViewType type;
1125 VkImageAspectFlags aspect_mask;
1127 uint32_t base_layer;
1128 uint32_t layer_count;
1130 VkExtent3D extent; /**< Extent of VkImageViewCreateInfo::baseMipLevel. */
1132 uint32_t descriptor[8];
1133 uint32_t fmask_descriptor[8];
1136 struct radv_image_create_info {
1137 const VkImageCreateInfo *vk_info;
1142 VkResult radv_image_create(VkDevice _device,
1143 const struct radv_image_create_info *info,
1144 const VkAllocationCallbacks* alloc,
1147 void radv_image_view_init(struct radv_image_view *view,
1148 struct radv_device *device,
1149 const VkImageViewCreateInfo* pCreateInfo,
1150 struct radv_cmd_buffer *cmd_buffer,
1151 VkImageUsageFlags usage_mask);
1152 void radv_image_set_optimal_micro_tile_mode(struct radv_device *device,
1153 struct radv_image *image, uint32_t micro_tile_mode);
1154 struct radv_buffer_view {
1155 struct radeon_winsys_bo *bo;
1157 uint64_t range; /**< VkBufferViewCreateInfo::range */
1160 void radv_buffer_view_init(struct radv_buffer_view *view,
1161 struct radv_device *device,
1162 const VkBufferViewCreateInfo* pCreateInfo,
1163 struct radv_cmd_buffer *cmd_buffer);
1165 static inline struct VkExtent3D
1166 radv_sanitize_image_extent(const VkImageType imageType,
1167 const struct VkExtent3D imageExtent)
1169 switch (imageType) {
1170 case VK_IMAGE_TYPE_1D:
1171 return (VkExtent3D) { imageExtent.width, 1, 1 };
1172 case VK_IMAGE_TYPE_2D:
1173 return (VkExtent3D) { imageExtent.width, imageExtent.height, 1 };
1174 case VK_IMAGE_TYPE_3D:
1177 unreachable("invalid image type");
1181 static inline struct VkOffset3D
1182 radv_sanitize_image_offset(const VkImageType imageType,
1183 const struct VkOffset3D imageOffset)
1185 switch (imageType) {
1186 case VK_IMAGE_TYPE_1D:
1187 return (VkOffset3D) { imageOffset.x, 0, 0 };
1188 case VK_IMAGE_TYPE_2D:
1189 return (VkOffset3D) { imageOffset.x, imageOffset.y, 0 };
1190 case VK_IMAGE_TYPE_3D:
1193 unreachable("invalid image type");
1197 struct radv_sampler {
1201 struct radv_color_buffer_info {
1202 uint32_t cb_color_base;
1203 uint32_t cb_color_pitch;
1204 uint32_t cb_color_slice;
1205 uint32_t cb_color_view;
1206 uint32_t cb_color_info;
1207 uint32_t cb_color_attrib;
1208 uint32_t cb_dcc_control;
1209 uint32_t cb_color_cmask;
1210 uint32_t cb_color_cmask_slice;
1211 uint32_t cb_color_fmask;
1212 uint32_t cb_color_fmask_slice;
1213 uint32_t cb_clear_value0;
1214 uint32_t cb_clear_value1;
1215 uint32_t cb_dcc_base;
1216 uint32_t micro_tile_mode;
1219 struct radv_ds_buffer_info {
1220 uint32_t db_depth_info;
1222 uint32_t db_stencil_info;
1223 uint32_t db_z_read_base;
1224 uint32_t db_stencil_read_base;
1225 uint32_t db_z_write_base;
1226 uint32_t db_stencil_write_base;
1227 uint32_t db_depth_view;
1228 uint32_t db_depth_size;
1229 uint32_t db_depth_slice;
1230 uint32_t db_htile_surface;
1231 uint32_t db_htile_data_base;
1232 uint32_t pa_su_poly_offset_db_fmt_cntl;
1236 struct radv_attachment_info {
1238 struct radv_color_buffer_info cb;
1239 struct radv_ds_buffer_info ds;
1241 struct radv_image_view *attachment;
1244 struct radv_framebuffer {
1249 uint32_t attachment_count;
1250 struct radv_attachment_info attachments[0];
1253 struct radv_subpass_barrier {
1254 VkPipelineStageFlags src_stage_mask;
1255 VkAccessFlags src_access_mask;
1256 VkAccessFlags dst_access_mask;
1259 struct radv_subpass {
1260 uint32_t input_count;
1261 VkAttachmentReference * input_attachments;
1262 uint32_t color_count;
1263 VkAttachmentReference * color_attachments;
1264 VkAttachmentReference * resolve_attachments;
1265 VkAttachmentReference depth_stencil_attachment;
1267 /** Subpass has at least one resolve attachment */
1270 struct radv_subpass_barrier start_barrier;
1273 struct radv_render_pass_attachment {
1276 VkAttachmentLoadOp load_op;
1277 VkAttachmentLoadOp stencil_load_op;
1278 VkImageLayout initial_layout;
1279 VkImageLayout final_layout;
1282 struct radv_render_pass {
1283 uint32_t attachment_count;
1284 uint32_t subpass_count;
1285 VkAttachmentReference * subpass_attachments;
1286 struct radv_render_pass_attachment * attachments;
1287 struct radv_subpass_barrier end_barrier;
1288 struct radv_subpass subpasses[0];
1291 VkResult radv_device_init_meta(struct radv_device *device);
1292 void radv_device_finish_meta(struct radv_device *device);
1294 struct radv_query_pool {
1295 struct radeon_winsys_bo *bo;
1297 uint32_t availability_offset;
1303 radv_temp_descriptor_set_create(struct radv_device *device,
1304 struct radv_cmd_buffer *cmd_buffer,
1305 VkDescriptorSetLayout _layout,
1306 VkDescriptorSet *_set);
1309 radv_temp_descriptor_set_destroy(struct radv_device *device,
1310 VkDescriptorSet _set);
1311 void radv_initialise_cmask(struct radv_cmd_buffer *cmd_buffer,
1312 struct radv_image *image, uint32_t value);
1313 void radv_initialize_dcc(struct radv_cmd_buffer *cmd_buffer,
1314 struct radv_image *image, uint32_t value);
1317 struct radeon_winsys_fence *fence;
1322 struct radeon_winsys_sem;
1324 #define RADV_DEFINE_HANDLE_CASTS(__radv_type, __VkType) \
1326 static inline struct __radv_type * \
1327 __radv_type ## _from_handle(__VkType _handle) \
1329 return (struct __radv_type *) _handle; \
1332 static inline __VkType \
1333 __radv_type ## _to_handle(struct __radv_type *_obj) \
1335 return (__VkType) _obj; \
1338 #define RADV_DEFINE_NONDISP_HANDLE_CASTS(__radv_type, __VkType) \
1340 static inline struct __radv_type * \
1341 __radv_type ## _from_handle(__VkType _handle) \
1343 return (struct __radv_type *)(uintptr_t) _handle; \
1346 static inline __VkType \
1347 __radv_type ## _to_handle(struct __radv_type *_obj) \
1349 return (__VkType)(uintptr_t) _obj; \
1352 #define RADV_FROM_HANDLE(__radv_type, __name, __handle) \
1353 struct __radv_type *__name = __radv_type ## _from_handle(__handle)
1355 RADV_DEFINE_HANDLE_CASTS(radv_cmd_buffer, VkCommandBuffer)
1356 RADV_DEFINE_HANDLE_CASTS(radv_device, VkDevice)
1357 RADV_DEFINE_HANDLE_CASTS(radv_instance, VkInstance)
1358 RADV_DEFINE_HANDLE_CASTS(radv_physical_device, VkPhysicalDevice)
1359 RADV_DEFINE_HANDLE_CASTS(radv_queue, VkQueue)
1361 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_cmd_pool, VkCommandPool)
1362 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer, VkBuffer)
1363 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer_view, VkBufferView)
1364 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_pool, VkDescriptorPool)
1365 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set, VkDescriptorSet)
1366 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set_layout, VkDescriptorSetLayout)
1367 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_device_memory, VkDeviceMemory)
1368 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_fence, VkFence)
1369 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_event, VkEvent)
1370 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_framebuffer, VkFramebuffer)
1371 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image, VkImage)
1372 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image_view, VkImageView);
1373 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_cache, VkPipelineCache)
1374 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline, VkPipeline)
1375 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_layout, VkPipelineLayout)
1376 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_query_pool, VkQueryPool)
1377 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_render_pass, VkRenderPass)
1378 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_sampler, VkSampler)
1379 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_shader_module, VkShaderModule)
1380 RADV_DEFINE_NONDISP_HANDLE_CASTS(radeon_winsys_sem, VkSemaphore)
1382 #endif /* RADV_PRIVATE_H */