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_descriptor_set.h"
58 #include <llvm-c/TargetMachine.h>
60 /* Pre-declarations needed for WSI entrypoints */
63 typedef struct xcb_connection_t xcb_connection_t
;
64 typedef uint32_t xcb_visualid_t
;
65 typedef uint32_t xcb_window_t
;
67 #include <vulkan/vulkan.h>
68 #include <vulkan/vulkan_intel.h>
69 #include <vulkan/vk_icd.h>
71 #include "radv_entrypoints.h"
75 #define MAX_VERTEX_ATTRIBS 32
77 #define MAX_VIEWPORTS 16
78 #define MAX_SCISSORS 16
79 #define MAX_PUSH_CONSTANTS_SIZE 128
80 #define MAX_DYNAMIC_BUFFERS 16
82 #define MAX_SAMPLES_LOG2 4 /* SKL supports 16 samples */
83 #define NUM_META_FS_KEYS 11
85 #define NUM_DEPTH_CLEAR_PIPELINES 3
87 #define radv_noreturn __attribute__((__noreturn__))
88 #define radv_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
90 static inline uint32_t
91 align_u32(uint32_t v
, uint32_t a
)
93 assert(a
!= 0 && a
== (a
& -a
));
94 return (v
+ a
- 1) & ~(a
- 1);
97 static inline uint32_t
98 align_u32_npot(uint32_t v
, uint32_t a
)
100 return (v
+ a
- 1) / a
* a
;
103 static inline uint64_t
104 align_u64(uint64_t v
, uint64_t a
)
106 assert(a
!= 0 && a
== (a
& -a
));
107 return (v
+ a
- 1) & ~(a
- 1);
110 static inline int32_t
111 align_i32(int32_t v
, int32_t a
)
113 assert(a
!= 0 && a
== (a
& -a
));
114 return (v
+ a
- 1) & ~(a
- 1);
117 /** Alignment must be a power of 2. */
119 radv_is_aligned(uintmax_t n
, uintmax_t a
)
121 assert(a
== (a
& -a
));
122 return (n
& (a
- 1)) == 0;
125 static inline uint32_t
126 round_up_u32(uint32_t v
, uint32_t a
)
128 return (v
+ a
- 1) / a
;
131 static inline uint64_t
132 round_up_u64(uint64_t v
, uint64_t a
)
134 return (v
+ a
- 1) / a
;
137 static inline uint32_t
138 radv_minify(uint32_t n
, uint32_t levels
)
140 if (unlikely(n
== 0))
143 return MAX2(n
>> levels
, 1);
146 radv_clamp_f(float f
, float min
, float max
)
159 radv_clear_mask(uint32_t *inout_mask
, uint32_t clear_mask
)
161 if (*inout_mask
& clear_mask
) {
162 *inout_mask
&= ~clear_mask
;
169 #define for_each_bit(b, dword) \
170 for (uint32_t __dword = (dword); \
171 (b) = __builtin_ffs(__dword) - 1, __dword; \
172 __dword &= ~(1 << (b)))
174 #define typed_memcpy(dest, src, count) ({ \
175 static_assert(sizeof(*src) == sizeof(*dest), ""); \
176 memcpy((dest), (src), (count) * sizeof(*(src))); \
179 #define zero(x) (memset(&(x), 0, sizeof(x)))
181 /* Define no kernel as 1, since that's an illegal offset for a kernel */
185 VkStructureType sType
;
189 /* Whenever we generate an error, pass it through this function. Useful for
190 * debugging, where we can break on it. Only call at error site, not when
191 * propagating errors. Might be useful to plug in a stack trace here.
194 VkResult
__vk_errorf(VkResult error
, const char *file
, int line
, const char *format
, ...);
197 #define vk_error(error) __vk_errorf(error, __FILE__, __LINE__, NULL);
198 #define vk_errorf(error, format, ...) __vk_errorf(error, __FILE__, __LINE__, format, ## __VA_ARGS__);
200 #define vk_error(error) error
201 #define vk_errorf(error, format, ...) error
204 void __radv_finishme(const char *file
, int line
, const char *format
, ...)
205 radv_printflike(3, 4);
206 void radv_loge(const char *format
, ...) radv_printflike(1, 2);
207 void radv_loge_v(const char *format
, va_list va
);
210 * Print a FINISHME message, including its source location.
212 #define radv_finishme(format, ...) \
213 __radv_finishme(__FILE__, __LINE__, format, ##__VA_ARGS__);
215 /* A non-fatal assert. Useful for debugging. */
217 #define radv_assert(x) ({ \
218 if (unlikely(!(x))) \
219 fprintf(stderr, "%s:%d ASSERT: %s\n", __FILE__, __LINE__, #x); \
222 #define radv_assert(x)
225 void radv_abortf(const char *format
, ...) radv_noreturn
radv_printflike(1, 2);
226 void radv_abortfv(const char *format
, va_list va
) radv_noreturn
;
228 #define stub_return(v) \
230 radv_finishme("stub %s", __func__); \
236 radv_finishme("stub %s", __func__); \
240 void *radv_resolve_entrypoint(uint32_t index
);
241 void *radv_lookup_entrypoint(const char *name
);
243 extern struct radv_dispatch_table dtable
;
245 struct radv_wsi_interaface
;
247 #define VK_ICD_WSI_PLATFORM_MAX 5
249 struct radv_physical_device
{
250 VK_LOADER_DATA _loader_data
;
252 struct radv_instance
* instance
;
254 struct radeon_winsys
*ws
;
255 struct radeon_info rad_info
;
259 uint64_t aperture_size
;
260 int cmd_parser_version
;
261 uint32_t pci_vendor_id
;
262 uint32_t pci_device_id
;
264 struct radv_wsi_interface
* wsi
[VK_ICD_WSI_PLATFORM_MAX
];
267 struct radv_instance
{
268 VK_LOADER_DATA _loader_data
;
270 VkAllocationCallbacks alloc
;
273 int physicalDeviceCount
;
274 struct radv_physical_device physicalDevice
;
277 VkResult
radv_init_wsi(struct radv_physical_device
*physical_device
);
278 void radv_finish_wsi(struct radv_physical_device
*physical_device
);
282 struct radv_pipeline_cache
{
283 struct radv_device
* device
;
284 pthread_mutex_t mutex
;
288 uint32_t kernel_count
;
289 struct cache_entry
** hash_table
;
292 VkAllocationCallbacks alloc
;
296 radv_pipeline_cache_init(struct radv_pipeline_cache
*cache
,
297 struct radv_device
*device
);
299 radv_pipeline_cache_finish(struct radv_pipeline_cache
*cache
);
301 radv_pipeline_cache_load(struct radv_pipeline_cache
*cache
,
302 const void *data
, size_t size
);
304 struct radv_shader_variant
*
305 radv_create_shader_variant_from_pipeline_cache(struct radv_device
*device
,
306 struct radv_pipeline_cache
*cache
,
307 const unsigned char *sha1
);
309 struct radv_shader_variant
*
310 radv_pipeline_cache_insert_shader(struct radv_pipeline_cache
*cache
,
311 const unsigned char *sha1
,
312 struct radv_shader_variant
*variant
,
313 const void *code
, unsigned code_size
);
315 void radv_shader_variant_destroy(struct radv_device
*device
,
316 struct radv_shader_variant
*variant
);
318 struct radv_meta_state
{
319 VkAllocationCallbacks alloc
;
321 struct radv_pipeline_cache cache
;
324 * Use array element `i` for images with `2^i` samples.
327 VkRenderPass render_pass
[NUM_META_FS_KEYS
];
328 struct radv_pipeline
*color_pipelines
[NUM_META_FS_KEYS
];
330 VkRenderPass depth_only_rp
[NUM_DEPTH_CLEAR_PIPELINES
];
331 struct radv_pipeline
*depth_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
332 VkRenderPass stencil_only_rp
[NUM_DEPTH_CLEAR_PIPELINES
];
333 struct radv_pipeline
*stencil_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
334 VkRenderPass depthstencil_rp
[NUM_DEPTH_CLEAR_PIPELINES
];
335 struct radv_pipeline
*depthstencil_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
336 } clear
[1 + MAX_SAMPLES_LOG2
];
339 VkRenderPass render_pass
[NUM_META_FS_KEYS
];
341 /** Pipeline that blits from a 1D image. */
342 VkPipeline pipeline_1d_src
[NUM_META_FS_KEYS
];
344 /** Pipeline that blits from a 2D image. */
345 VkPipeline pipeline_2d_src
[NUM_META_FS_KEYS
];
347 /** Pipeline that blits from a 3D image. */
348 VkPipeline pipeline_3d_src
[NUM_META_FS_KEYS
];
350 VkRenderPass depth_only_rp
;
351 VkPipeline depth_only_1d_pipeline
;
352 VkPipeline depth_only_2d_pipeline
;
353 VkPipeline depth_only_3d_pipeline
;
355 VkRenderPass stencil_only_rp
;
356 VkPipeline stencil_only_1d_pipeline
;
357 VkPipeline stencil_only_2d_pipeline
;
358 VkPipeline stencil_only_3d_pipeline
;
359 VkPipelineLayout pipeline_layout
;
360 VkDescriptorSetLayout ds_layout
;
364 VkRenderPass render_passes
[NUM_META_FS_KEYS
];
366 VkPipelineLayout p_layouts
[2];
367 VkDescriptorSetLayout ds_layouts
[2];
368 VkPipeline pipelines
[2][NUM_META_FS_KEYS
];
370 VkRenderPass depth_only_rp
;
371 VkPipeline depth_only_pipeline
[2];
373 VkRenderPass stencil_only_rp
;
374 VkPipeline stencil_only_pipeline
[2];
378 VkPipelineLayout img_p_layout
;
379 VkDescriptorSetLayout img_ds_layout
;
383 VkRenderPass render_pass
;
384 VkPipelineLayout img_p_layout
;
385 VkDescriptorSetLayout img_ds_layout
;
395 VkDescriptorSetLayout ds_layout
;
396 VkPipelineLayout p_layout
;
399 VkPipeline i_pipeline
;
400 } rc
[MAX_SAMPLES_LOG2
];
404 VkPipeline decompress_pipeline
;
405 VkPipeline resummarize_pipeline
;
410 VkPipeline cmask_eliminate_pipeline
;
411 VkPipeline fmask_decompress_pipeline
;
416 VkPipelineLayout fill_p_layout
;
417 VkPipelineLayout copy_p_layout
;
418 VkDescriptorSetLayout fill_ds_layout
;
419 VkDescriptorSetLayout copy_ds_layout
;
420 VkPipeline fill_pipeline
;
421 VkPipeline copy_pipeline
;
426 VK_LOADER_DATA _loader_data
;
428 struct radv_device
* device
;
430 struct radv_state_pool
* pool
;
434 VK_LOADER_DATA _loader_data
;
436 VkAllocationCallbacks alloc
;
438 struct radv_instance
* instance
;
439 struct radeon_winsys
*ws
;
440 struct radeon_winsys_ctx
*hw_ctx
;
442 struct radv_meta_state meta_state
;
443 struct radv_queue queue
;
444 struct radeon_winsys_cs
*empty_cs
;
446 bool allow_fast_clears
;
449 /* MSAA sample locations.
450 * The first index is the sample index.
451 * The second index is the coordinate: X, Y. */
452 float sample_locations_1x
[1][2];
453 float sample_locations_2x
[2][2];
454 float sample_locations_4x
[4][2];
455 float sample_locations_8x
[8][2];
456 float sample_locations_16x
[16][2];
459 void radv_device_get_cache_uuid(void *uuid
);
461 struct radv_device_memory
{
462 struct radeon_winsys_bo
*bo
;
464 VkDeviceSize map_size
;
469 struct radv_descriptor_range
{
474 struct radv_descriptor_set
{
475 const struct radv_descriptor_set_layout
*layout
;
476 struct list_head descriptor_pool
;
479 struct radv_buffer_view
*buffer_views
;
480 struct radeon_winsys_bo
*bo
;
482 uint32_t *mapped_ptr
;
483 struct radv_descriptor_range
*dynamic_descriptors
;
484 struct radeon_winsys_bo
*descriptors
[0];
487 struct radv_descriptor_pool_free_node
{
493 struct radv_descriptor_pool
{
494 struct list_head descriptor_sets
;
496 struct radeon_winsys_bo
*bo
;
498 uint64_t current_offset
;
504 struct radv_descriptor_pool_free_node free_nodes
[];
508 struct radv_device
* device
;
511 VkBufferUsageFlags usage
;
514 struct radeon_winsys_bo
* bo
;
519 enum radv_cmd_dirty_bits
{
520 RADV_CMD_DIRTY_DYNAMIC_VIEWPORT
= 1 << 0, /* VK_DYNAMIC_STATE_VIEWPORT */
521 RADV_CMD_DIRTY_DYNAMIC_SCISSOR
= 1 << 1, /* VK_DYNAMIC_STATE_SCISSOR */
522 RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH
= 1 << 2, /* VK_DYNAMIC_STATE_LINE_WIDTH */
523 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS
= 1 << 3, /* VK_DYNAMIC_STATE_DEPTH_BIAS */
524 RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS
= 1 << 4, /* VK_DYNAMIC_STATE_BLEND_CONSTANTS */
525 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS
= 1 << 5, /* VK_DYNAMIC_STATE_DEPTH_BOUNDS */
526 RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6, /* VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK */
527 RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7, /* VK_DYNAMIC_STATE_STENCIL_WRITE_MASK */
528 RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE
= 1 << 8, /* VK_DYNAMIC_STATE_STENCIL_REFERENCE */
529 RADV_CMD_DIRTY_DYNAMIC_ALL
= (1 << 9) - 1,
530 RADV_CMD_DIRTY_PIPELINE
= 1 << 9,
531 RADV_CMD_DIRTY_INDEX_BUFFER
= 1 << 10,
532 RADV_CMD_DIRTY_RENDER_TARGETS
= 1 << 11,
534 typedef uint32_t radv_cmd_dirty_mask_t
;
536 enum radv_cmd_flush_bits
{
537 RADV_CMD_FLAG_INV_ICACHE
= 1 << 0,
538 /* SMEM L1, other names: KCACHE, constant cache, DCACHE, data cache */
539 RADV_CMD_FLAG_INV_SMEM_L1
= 1 << 1,
540 /* VMEM L1 can optionally be bypassed (GLC=1). Other names: TC L1 */
541 RADV_CMD_FLAG_INV_VMEM_L1
= 1 << 2,
542 /* Used by everything except CB/DB, can be bypassed (SLC=1). Other names: TC L2 */
543 RADV_CMD_FLAG_INV_GLOBAL_L2
= 1 << 3,
544 /* Framebuffer caches */
545 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
= 1 << 4,
546 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
= 1 << 5,
547 RADV_CMD_FLAG_FLUSH_AND_INV_DB
= 1 << 6,
548 RADV_CMD_FLAG_FLUSH_AND_INV_CB
= 1 << 7,
549 /* Engine synchronization. */
550 RADV_CMD_FLAG_VS_PARTIAL_FLUSH
= 1 << 8,
551 RADV_CMD_FLAG_PS_PARTIAL_FLUSH
= 1 << 9,
552 RADV_CMD_FLAG_CS_PARTIAL_FLUSH
= 1 << 10,
553 RADV_CMD_FLAG_VGT_FLUSH
= 1 << 11,
555 RADV_CMD_FLUSH_AND_INV_FRAMEBUFFER
= (RADV_CMD_FLAG_FLUSH_AND_INV_CB
|
556 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
|
557 RADV_CMD_FLAG_FLUSH_AND_INV_DB
|
558 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
)
561 struct radv_vertex_binding
{
562 struct radv_buffer
* buffer
;
566 struct radv_dynamic_state
{
569 VkViewport viewports
[MAX_VIEWPORTS
];
574 VkRect2D scissors
[MAX_SCISSORS
];
585 float blend_constants
[4];
595 } stencil_compare_mask
;
600 } stencil_write_mask
;
608 extern const struct radv_dynamic_state default_dynamic_state
;
610 void radv_dynamic_state_copy(struct radv_dynamic_state
*dest
,
611 const struct radv_dynamic_state
*src
,
614 * Attachment state when recording a renderpass instance.
616 * The clear value is valid only if there exists a pending clear.
618 struct radv_attachment_state
{
619 VkImageAspectFlags pending_clear_aspects
;
620 VkClearValue clear_value
;
621 VkImageLayout current_layout
;
624 struct radv_cmd_state
{
626 bool vertex_descriptors_dirty
;
627 radv_cmd_dirty_mask_t dirty
;
629 struct radv_pipeline
* pipeline
;
630 struct radv_pipeline
* emitted_pipeline
;
631 struct radv_pipeline
* compute_pipeline
;
632 struct radv_pipeline
* emitted_compute_pipeline
;
633 struct radv_framebuffer
* framebuffer
;
634 struct radv_render_pass
* pass
;
635 const struct radv_subpass
* subpass
;
636 struct radv_dynamic_state dynamic
;
637 struct radv_vertex_binding vertex_bindings
[MAX_VBS
];
638 struct radv_descriptor_set
* descriptors
[MAX_SETS
];
639 struct radv_attachment_state
* attachments
;
640 VkRect2D render_area
;
641 struct radv_buffer
* index_buffer
;
643 uint32_t index_offset
;
644 uint32_t last_primitive_reset_index
;
645 enum radv_cmd_flush_bits flush_bits
;
646 unsigned active_occlusion_queries
;
649 struct radv_cmd_pool
{
650 VkAllocationCallbacks alloc
;
651 struct list_head cmd_buffers
;
654 struct radv_cmd_buffer_upload
{
658 struct radeon_winsys_bo
*upload_bo
;
659 struct list_head list
;
662 struct radv_cmd_buffer
{
663 VK_LOADER_DATA _loader_data
;
665 struct radv_device
* device
;
667 struct radv_cmd_pool
* pool
;
668 struct list_head pool_link
;
670 VkCommandBufferUsageFlags usage_flags
;
671 VkCommandBufferLevel level
;
672 struct radeon_winsys_cs
*cs
;
673 struct radv_cmd_state state
;
675 uint8_t push_constants
[MAX_PUSH_CONSTANTS_SIZE
];
676 uint32_t dynamic_buffers
[16 * MAX_DYNAMIC_BUFFERS
];
677 VkShaderStageFlags push_constant_stages
;
679 struct radv_cmd_buffer_upload upload
;
686 void si_init_config(struct radv_physical_device
*physical_device
,
687 struct radv_cmd_buffer
*cmd_buffer
);
688 void si_write_viewport(struct radeon_winsys_cs
*cs
, int first_vp
,
689 int count
, const VkViewport
*viewports
);
690 void si_write_scissors(struct radeon_winsys_cs
*cs
, int first
,
691 int count
, const VkRect2D
*scissors
);
692 uint32_t si_get_ia_multi_vgt_param(struct radv_cmd_buffer
*cmd_buffer
);
693 void si_emit_cache_flush(struct radv_cmd_buffer
*cmd_buffer
);
694 void si_cp_dma_buffer_copy(struct radv_cmd_buffer
*cmd_buffer
,
695 uint64_t src_va
, uint64_t dest_va
,
697 void si_cp_dma_clear_buffer(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
698 uint64_t size
, unsigned value
);
699 void radv_set_db_count_control(struct radv_cmd_buffer
*cmd_buffer
);
700 void radv_bind_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
701 struct radv_descriptor_set
*set
,
704 radv_cmd_buffer_upload_alloc(struct radv_cmd_buffer
*cmd_buffer
,
707 unsigned *out_offset
,
710 radv_cmd_buffer_set_subpass(struct radv_cmd_buffer
*cmd_buffer
,
711 const struct radv_subpass
*subpass
,
714 radv_cmd_buffer_upload_data(struct radv_cmd_buffer
*cmd_buffer
,
715 unsigned size
, unsigned alignmnet
,
716 const void *data
, unsigned *out_offset
);
718 radv_emit_framebuffer_state(struct radv_cmd_buffer
*cmd_buffer
);
719 void radv_cmd_buffer_clear_subpass(struct radv_cmd_buffer
*cmd_buffer
);
720 void radv_cmd_buffer_resolve_subpass(struct radv_cmd_buffer
*cmd_buffer
);
721 void radv_cayman_emit_msaa_sample_locs(struct radeon_winsys_cs
*cs
, int nr_samples
);
722 unsigned radv_cayman_get_maxdist(int log_samples
);
723 void radv_device_init_msaa(struct radv_device
*device
);
724 void radv_set_depth_clear_regs(struct radv_cmd_buffer
*cmd_buffer
,
725 struct radv_image
*image
,
726 VkClearDepthStencilValue ds_clear_value
,
727 VkImageAspectFlags aspects
);
728 void radv_set_color_clear_regs(struct radv_cmd_buffer
*cmd_buffer
,
729 struct radv_image
*image
,
731 uint32_t color_values
[2]);
732 void radv_fill_buffer(struct radv_cmd_buffer
*cmd_buffer
,
733 struct radeon_winsys_bo
*bo
,
734 uint64_t offset
, uint64_t size
, uint32_t value
);
737 * Takes x,y,z as exact numbers of invocations, instead of blocks.
739 * Limitations: Can't call normal dispatch functions without binding or rebinding
740 * the compute pipeline.
742 void radv_unaligned_dispatch(
743 struct radv_cmd_buffer
*cmd_buffer
,
749 struct radeon_winsys_bo
*bo
;
755 struct radv_shader_module
{
756 struct nir_shader
* nir
;
757 unsigned char sha1
[20];
762 union ac_shader_variant_key
;
765 radv_hash_shader(unsigned char *hash
, struct radv_shader_module
*module
,
766 const char *entrypoint
,
767 const VkSpecializationInfo
*spec_info
,
768 const struct radv_pipeline_layout
*layout
,
769 const union ac_shader_variant_key
*key
);
771 static inline gl_shader_stage
772 vk_to_mesa_shader_stage(VkShaderStageFlagBits vk_stage
)
774 assert(__builtin_popcount(vk_stage
) == 1);
775 return ffs(vk_stage
) - 1;
778 static inline VkShaderStageFlagBits
779 mesa_to_vk_shader_stage(gl_shader_stage mesa_stage
)
781 return (1 << mesa_stage
);
784 #define RADV_STAGE_MASK ((1 << MESA_SHADER_STAGES) - 1)
786 #define radv_foreach_stage(stage, stage_bits) \
787 for (gl_shader_stage stage, \
788 __tmp = (gl_shader_stage)((stage_bits) & RADV_STAGE_MASK); \
789 stage = __builtin_ffs(__tmp) - 1, __tmp; \
790 __tmp &= ~(1 << (stage)))
792 struct radv_shader_variant
{
795 struct radeon_winsys_bo
*bo
;
796 struct ac_shader_config config
;
797 struct ac_shader_variant_info info
;
802 struct radv_depth_stencil_state
{
803 uint32_t db_depth_control
;
804 uint32_t db_stencil_control
;
805 uint32_t db_render_control
;
806 uint32_t db_render_override2
;
809 struct radv_blend_state
{
810 uint32_t cb_color_control
;
811 uint32_t cb_target_mask
;
812 uint32_t sx_mrt0_blend_opt
[8];
813 uint32_t cb_blend_control
[8];
815 uint32_t spi_shader_col_format
;
816 uint32_t cb_shader_mask
;
817 uint32_t db_alpha_to_mask
;
820 unsigned radv_format_meta_fs_key(VkFormat format
);
822 struct radv_raster_state
{
823 uint32_t pa_cl_clip_cntl
;
824 uint32_t pa_cl_vs_out_cntl
;
825 uint32_t spi_interp_control
;
826 uint32_t pa_su_point_size
;
827 uint32_t pa_su_point_minmax
;
828 uint32_t pa_su_line_cntl
;
829 uint32_t pa_su_vtx_cntl
;
830 uint32_t pa_su_sc_mode_cntl
;
833 struct radv_multisample_state
{
835 uint32_t pa_sc_line_cntl
;
836 uint32_t pa_sc_mode_cntl_0
;
837 uint32_t pa_sc_mode_cntl_1
;
838 uint32_t pa_sc_aa_config
;
839 uint32_t pa_sc_aa_mask
[2];
840 unsigned num_samples
;
843 struct radv_pipeline
{
844 struct radv_device
* device
;
845 uint32_t dynamic_state_mask
;
846 struct radv_dynamic_state dynamic_state
;
848 struct radv_pipeline_layout
* layout
;
850 bool needs_data_cache
;
852 struct radv_shader_variant
* shaders
[MESA_SHADER_STAGES
];
853 VkShaderStageFlags active_stages
;
855 uint32_t va_rsrc_word3
[MAX_VERTEX_ATTRIBS
];
856 uint32_t va_format_size
[MAX_VERTEX_ATTRIBS
];
857 uint32_t va_binding
[MAX_VERTEX_ATTRIBS
];
858 uint32_t va_offset
[MAX_VERTEX_ATTRIBS
];
859 uint32_t num_vertex_attribs
;
860 uint32_t binding_stride
[MAX_VBS
];
864 struct radv_blend_state blend
;
865 struct radv_depth_stencil_state ds
;
866 struct radv_raster_state raster
;
867 struct radv_multisample_state ms
;
870 bool prim_restart_enable
;
875 struct radv_graphics_pipeline_create_info
{
878 bool db_stencil_clear
;
879 bool db_depth_disable_expclear
;
880 bool db_stencil_disable_expclear
;
881 bool db_flush_depth_inplace
;
882 bool db_flush_stencil_inplace
;
884 uint32_t custom_blend_mode
;
888 radv_pipeline_init(struct radv_pipeline
*pipeline
, struct radv_device
*device
,
889 struct radv_pipeline_cache
*cache
,
890 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
891 const struct radv_graphics_pipeline_create_info
*extra
,
892 const VkAllocationCallbacks
*alloc
);
895 radv_graphics_pipeline_create(VkDevice device
,
896 VkPipelineCache cache
,
897 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
898 const struct radv_graphics_pipeline_create_info
*extra
,
899 const VkAllocationCallbacks
*alloc
,
900 VkPipeline
*pPipeline
);
902 struct vk_format_description
;
903 uint32_t radv_translate_buffer_dataformat(const struct vk_format_description
*desc
,
905 uint32_t radv_translate_buffer_numformat(const struct vk_format_description
*desc
,
907 uint32_t radv_translate_colorformat(VkFormat format
);
908 uint32_t radv_translate_color_numformat(VkFormat format
,
909 const struct vk_format_description
*desc
,
911 uint32_t radv_colorformat_endian_swap(uint32_t colorformat
);
912 unsigned radv_translate_colorswap(VkFormat format
, bool do_endian_swap
);
913 uint32_t radv_translate_dbformat(VkFormat format
);
914 uint32_t radv_translate_tex_dataformat(VkFormat format
,
915 const struct vk_format_description
*desc
,
917 uint32_t radv_translate_tex_numformat(VkFormat format
,
918 const struct vk_format_description
*desc
,
920 bool radv_format_pack_clear_color(VkFormat format
,
921 uint32_t clear_vals
[2],
922 VkClearColorValue
*value
);
923 bool radv_is_colorbuffer_format_supported(VkFormat format
, bool *blendable
);
925 struct radv_fmask_info
{
929 unsigned pitch_in_pixels
;
930 unsigned bank_height
;
931 unsigned slice_tile_max
;
932 unsigned tile_mode_index
;
935 struct radv_cmask_info
{
943 unsigned slice_tile_max
;
944 unsigned base_address_reg
;
947 struct r600_htile_info
{
958 /* The original VkFormat provided by the client. This may not match any
959 * of the actual surface formats.
962 VkImageAspectFlags aspects
;
966 uint32_t samples
; /**< VkImageCreateInfo::samples */
967 VkImageUsageFlags usage
; /**< Superset of VkImageCreateInfo::usage. */
968 VkImageTiling tiling
; /** VkImageCreateInfo::tiling */
974 struct radeon_winsys_bo
*bo
;
977 struct radeon_surf surface
;
979 struct radv_fmask_info fmask
;
980 struct radv_cmask_info cmask
;
981 uint32_t clear_value_offset
;
983 /* Depth buffer compression and fast clear. */
984 struct r600_htile_info htile
;
987 bool radv_layout_has_htile(const struct radv_image
*image
,
988 VkImageLayout layout
);
989 bool radv_layout_is_htile_compressed(const struct radv_image
*image
,
990 VkImageLayout layout
);
991 bool radv_layout_can_expclear(const struct radv_image
*image
,
992 VkImageLayout layout
);
993 bool radv_layout_has_cmask(const struct radv_image
*image
,
994 VkImageLayout layout
);
995 static inline uint32_t
996 radv_get_layerCount(const struct radv_image
*image
,
997 const VkImageSubresourceRange
*range
)
999 return range
->layerCount
== VK_REMAINING_ARRAY_LAYERS
?
1000 image
->array_size
- range
->baseArrayLayer
: range
->layerCount
;
1003 static inline uint32_t
1004 radv_get_levelCount(const struct radv_image
*image
,
1005 const VkImageSubresourceRange
*range
)
1007 return range
->levelCount
== VK_REMAINING_MIP_LEVELS
?
1008 image
->levels
- range
->baseMipLevel
: range
->levelCount
;
1011 struct radeon_bo_metadata
;
1013 radv_init_metadata(struct radv_device
*device
,
1014 struct radv_image
*image
,
1015 struct radeon_bo_metadata
*metadata
);
1017 struct radv_image_view
{
1018 struct radv_image
*image
; /**< VkImageViewCreateInfo::image */
1019 struct radeon_winsys_bo
*bo
;
1021 VkImageViewType type
;
1022 VkImageAspectFlags aspect_mask
;
1024 uint32_t base_layer
;
1025 uint32_t layer_count
;
1027 VkExtent3D extent
; /**< Extent of VkImageViewCreateInfo::baseMipLevel. */
1029 uint32_t descriptor
[8];
1030 uint32_t fmask_descriptor
[8];
1033 struct radv_image_create_info
{
1034 const VkImageCreateInfo
*vk_info
;
1039 VkResult
radv_image_create(VkDevice _device
,
1040 const struct radv_image_create_info
*info
,
1041 const VkAllocationCallbacks
* alloc
,
1044 void radv_image_view_init(struct radv_image_view
*view
,
1045 struct radv_device
*device
,
1046 const VkImageViewCreateInfo
* pCreateInfo
,
1047 struct radv_cmd_buffer
*cmd_buffer
,
1048 VkImageUsageFlags usage_mask
);
1049 void radv_image_set_optimal_micro_tile_mode(struct radv_device
*device
,
1050 struct radv_image
*image
, uint32_t micro_tile_mode
);
1051 struct radv_buffer_view
{
1052 struct radeon_winsys_bo
*bo
;
1054 uint64_t range
; /**< VkBufferViewCreateInfo::range */
1057 void radv_buffer_view_init(struct radv_buffer_view
*view
,
1058 struct radv_device
*device
,
1059 const VkBufferViewCreateInfo
* pCreateInfo
,
1060 struct radv_cmd_buffer
*cmd_buffer
);
1062 static inline struct VkExtent3D
1063 radv_sanitize_image_extent(const VkImageType imageType
,
1064 const struct VkExtent3D imageExtent
)
1066 switch (imageType
) {
1067 case VK_IMAGE_TYPE_1D
:
1068 return (VkExtent3D
) { imageExtent
.width
, 1, 1 };
1069 case VK_IMAGE_TYPE_2D
:
1070 return (VkExtent3D
) { imageExtent
.width
, imageExtent
.height
, 1 };
1071 case VK_IMAGE_TYPE_3D
:
1074 unreachable("invalid image type");
1078 static inline struct VkOffset3D
1079 radv_sanitize_image_offset(const VkImageType imageType
,
1080 const struct VkOffset3D imageOffset
)
1082 switch (imageType
) {
1083 case VK_IMAGE_TYPE_1D
:
1084 return (VkOffset3D
) { imageOffset
.x
, 0, 0 };
1085 case VK_IMAGE_TYPE_2D
:
1086 return (VkOffset3D
) { imageOffset
.x
, imageOffset
.y
, 0 };
1087 case VK_IMAGE_TYPE_3D
:
1090 unreachable("invalid image type");
1094 struct radv_sampler
{
1098 struct radv_color_buffer_info
{
1099 uint32_t cb_color_base
;
1100 uint32_t cb_color_pitch
;
1101 uint32_t cb_color_slice
;
1102 uint32_t cb_color_view
;
1103 uint32_t cb_color_info
;
1104 uint32_t cb_color_attrib
;
1105 uint32_t cb_dcc_control
;
1106 uint32_t cb_color_cmask
;
1107 uint32_t cb_color_cmask_slice
;
1108 uint32_t cb_color_fmask
;
1109 uint32_t cb_color_fmask_slice
;
1110 uint32_t cb_clear_value0
;
1111 uint32_t cb_clear_value1
;
1112 uint32_t cb_dcc_base
;
1113 uint32_t micro_tile_mode
;
1116 struct radv_ds_buffer_info
{
1117 uint32_t db_depth_info
;
1119 uint32_t db_stencil_info
;
1120 uint32_t db_z_read_base
;
1121 uint32_t db_stencil_read_base
;
1122 uint32_t db_z_write_base
;
1123 uint32_t db_stencil_write_base
;
1124 uint32_t db_depth_view
;
1125 uint32_t db_depth_size
;
1126 uint32_t db_depth_slice
;
1127 uint32_t db_htile_surface
;
1128 uint32_t db_htile_data_base
;
1129 uint32_t pa_su_poly_offset_db_fmt_cntl
;
1133 struct radv_attachment_info
{
1135 struct radv_color_buffer_info cb
;
1136 struct radv_ds_buffer_info ds
;
1138 struct radv_image_view
*attachment
;
1141 struct radv_framebuffer
{
1146 uint32_t attachment_count
;
1147 struct radv_attachment_info attachments
[0];
1150 struct radv_subpass_barrier
{
1151 VkPipelineStageFlags src_stage_mask
;
1152 VkAccessFlags src_access_mask
;
1153 VkAccessFlags dst_access_mask
;
1156 struct radv_subpass
{
1157 uint32_t input_count
;
1158 VkAttachmentReference
* input_attachments
;
1159 uint32_t color_count
;
1160 VkAttachmentReference
* color_attachments
;
1161 VkAttachmentReference
* resolve_attachments
;
1162 VkAttachmentReference depth_stencil_attachment
;
1164 /** Subpass has at least one resolve attachment */
1167 struct radv_subpass_barrier start_barrier
;
1170 struct radv_render_pass_attachment
{
1173 VkAttachmentLoadOp load_op
;
1174 VkAttachmentLoadOp stencil_load_op
;
1175 VkImageLayout initial_layout
;
1176 VkImageLayout final_layout
;
1179 struct radv_render_pass
{
1180 uint32_t attachment_count
;
1181 uint32_t subpass_count
;
1182 VkAttachmentReference
* subpass_attachments
;
1183 struct radv_render_pass_attachment
* attachments
;
1184 struct radv_subpass_barrier end_barrier
;
1185 struct radv_subpass subpasses
[0];
1188 VkResult
radv_device_init_meta(struct radv_device
*device
);
1189 void radv_device_finish_meta(struct radv_device
*device
);
1191 struct radv_query_pool
{
1192 struct radeon_winsys_bo
*bo
;
1194 uint32_t availability_offset
;
1200 radv_temp_descriptor_set_create(struct radv_device
*device
,
1201 struct radv_cmd_buffer
*cmd_buffer
,
1202 VkDescriptorSetLayout _layout
,
1203 VkDescriptorSet
*_set
);
1206 radv_temp_descriptor_set_destroy(struct radv_device
*device
,
1207 VkDescriptorSet _set
);
1208 void radv_initialise_cmask(struct radv_cmd_buffer
*cmd_buffer
,
1209 struct radv_image
*image
, uint32_t value
);
1210 void radv_initialize_dcc(struct radv_cmd_buffer
*cmd_buffer
,
1211 struct radv_image
*image
, uint32_t value
);
1212 #define RADV_DEFINE_HANDLE_CASTS(__radv_type, __VkType) \
1214 static inline struct __radv_type * \
1215 __radv_type ## _from_handle(__VkType _handle) \
1217 return (struct __radv_type *) _handle; \
1220 static inline __VkType \
1221 __radv_type ## _to_handle(struct __radv_type *_obj) \
1223 return (__VkType) _obj; \
1226 #define RADV_DEFINE_NONDISP_HANDLE_CASTS(__radv_type, __VkType) \
1228 static inline struct __radv_type * \
1229 __radv_type ## _from_handle(__VkType _handle) \
1231 return (struct __radv_type *)(uintptr_t) _handle; \
1234 static inline __VkType \
1235 __radv_type ## _to_handle(struct __radv_type *_obj) \
1237 return (__VkType)(uintptr_t) _obj; \
1240 #define RADV_FROM_HANDLE(__radv_type, __name, __handle) \
1241 struct __radv_type *__name = __radv_type ## _from_handle(__handle)
1243 RADV_DEFINE_HANDLE_CASTS(radv_cmd_buffer
, VkCommandBuffer
)
1244 RADV_DEFINE_HANDLE_CASTS(radv_device
, VkDevice
)
1245 RADV_DEFINE_HANDLE_CASTS(radv_instance
, VkInstance
)
1246 RADV_DEFINE_HANDLE_CASTS(radv_physical_device
, VkPhysicalDevice
)
1247 RADV_DEFINE_HANDLE_CASTS(radv_queue
, VkQueue
)
1249 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_cmd_pool
, VkCommandPool
)
1250 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer
, VkBuffer
)
1251 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer_view
, VkBufferView
)
1252 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_pool
, VkDescriptorPool
)
1253 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set
, VkDescriptorSet
)
1254 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set_layout
, VkDescriptorSetLayout
)
1255 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_device_memory
, VkDeviceMemory
)
1256 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_fence
, VkFence
)
1257 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_event
, VkEvent
)
1258 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_framebuffer
, VkFramebuffer
)
1259 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image
, VkImage
)
1260 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image_view
, VkImageView
);
1261 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_cache
, VkPipelineCache
)
1262 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline
, VkPipeline
)
1263 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_layout
, VkPipelineLayout
)
1264 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_query_pool
, VkQueryPool
)
1265 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_render_pass
, VkRenderPass
)
1266 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_sampler
, VkSampler
)
1267 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_shader_module
, VkShaderModule
)
1269 #define RADV_DEFINE_STRUCT_CASTS(__radv_type, __VkType) \
1271 static inline const __VkType * \
1272 __radv_type ## _to_ ## __VkType(const struct __radv_type *__radv_obj) \
1274 return (const __VkType *) __radv_obj; \
1277 #define RADV_COMMON_TO_STRUCT(__VkType, __vk_name, __common_name) \
1278 const __VkType *__vk_name = radv_common_to_ ## __VkType(__common_name)
1280 RADV_DEFINE_STRUCT_CASTS(radv_common
, VkMemoryBarrier
)
1281 RADV_DEFINE_STRUCT_CASTS(radv_common
, VkBufferMemoryBarrier
)
1282 RADV_DEFINE_STRUCT_CASTS(radv_common
, VkImageMemoryBarrier
)
1285 #endif /* RADV_PRIVATE_H */