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"
73 #include "wsi_common.h"
76 #define MAX_VERTEX_ATTRIBS 32
78 #define MAX_VIEWPORTS 16
79 #define MAX_SCISSORS 16
80 #define MAX_PUSH_CONSTANTS_SIZE 128
81 #define MAX_DYNAMIC_BUFFERS 16
83 #define MAX_SAMPLES_LOG2 4 /* SKL supports 16 samples */
84 #define NUM_META_FS_KEYS 11
86 #define NUM_DEPTH_CLEAR_PIPELINES 3
88 #define radv_noreturn __attribute__((__noreturn__))
89 #define radv_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
91 static inline uint32_t
92 align_u32(uint32_t v
, uint32_t a
)
94 assert(a
!= 0 && a
== (a
& -a
));
95 return (v
+ a
- 1) & ~(a
- 1);
98 static inline uint32_t
99 align_u32_npot(uint32_t v
, uint32_t a
)
101 return (v
+ a
- 1) / a
* a
;
104 static inline uint64_t
105 align_u64(uint64_t v
, uint64_t a
)
107 assert(a
!= 0 && a
== (a
& -a
));
108 return (v
+ a
- 1) & ~(a
- 1);
111 static inline int32_t
112 align_i32(int32_t v
, int32_t a
)
114 assert(a
!= 0 && a
== (a
& -a
));
115 return (v
+ a
- 1) & ~(a
- 1);
118 /** Alignment must be a power of 2. */
120 radv_is_aligned(uintmax_t n
, uintmax_t a
)
122 assert(a
== (a
& -a
));
123 return (n
& (a
- 1)) == 0;
126 static inline uint32_t
127 round_up_u32(uint32_t v
, uint32_t a
)
129 return (v
+ a
- 1) / a
;
132 static inline uint64_t
133 round_up_u64(uint64_t v
, uint64_t a
)
135 return (v
+ a
- 1) / a
;
138 static inline uint32_t
139 radv_minify(uint32_t n
, uint32_t levels
)
141 if (unlikely(n
== 0))
144 return MAX2(n
>> levels
, 1);
147 radv_clamp_f(float f
, float min
, float max
)
160 radv_clear_mask(uint32_t *inout_mask
, uint32_t clear_mask
)
162 if (*inout_mask
& clear_mask
) {
163 *inout_mask
&= ~clear_mask
;
170 #define for_each_bit(b, dword) \
171 for (uint32_t __dword = (dword); \
172 (b) = __builtin_ffs(__dword) - 1, __dword; \
173 __dword &= ~(1 << (b)))
175 #define typed_memcpy(dest, src, count) ({ \
176 STATIC_ASSERT(sizeof(*src) == sizeof(*dest)); \
177 memcpy((dest), (src), (count) * sizeof(*(src))); \
180 #define zero(x) (memset(&(x), 0, sizeof(x)))
182 /* Define no kernel as 1, since that's an illegal offset for a kernel */
186 VkStructureType sType
;
190 /* Whenever we generate an error, pass it through this function. Useful for
191 * debugging, where we can break on it. Only call at error site, not when
192 * propagating errors. Might be useful to plug in a stack trace here.
195 VkResult
__vk_errorf(VkResult error
, const char *file
, int line
, const char *format
, ...);
198 #define vk_error(error) __vk_errorf(error, __FILE__, __LINE__, NULL);
199 #define vk_errorf(error, format, ...) __vk_errorf(error, __FILE__, __LINE__, format, ## __VA_ARGS__);
201 #define vk_error(error) error
202 #define vk_errorf(error, format, ...) error
205 void __radv_finishme(const char *file
, int line
, const char *format
, ...)
206 radv_printflike(3, 4);
207 void radv_loge(const char *format
, ...) radv_printflike(1, 2);
208 void radv_loge_v(const char *format
, va_list va
);
211 * Print a FINISHME message, including its source location.
213 #define radv_finishme(format, ...) \
215 static bool reported = false; \
217 __radv_finishme(__FILE__, __LINE__, format, ##__VA_ARGS__); \
222 /* A non-fatal assert. Useful for debugging. */
224 #define radv_assert(x) ({ \
225 if (unlikely(!(x))) \
226 fprintf(stderr, "%s:%d ASSERT: %s\n", __FILE__, __LINE__, #x); \
229 #define radv_assert(x)
232 void radv_abortf(const char *format
, ...) radv_noreturn
radv_printflike(1, 2);
233 void radv_abortfv(const char *format
, va_list va
) radv_noreturn
;
235 #define stub_return(v) \
237 radv_finishme("stub %s", __func__); \
243 radv_finishme("stub %s", __func__); \
247 void *radv_resolve_entrypoint(uint32_t index
);
248 void *radv_lookup_entrypoint(const char *name
);
250 extern struct radv_dispatch_table dtable
;
252 struct radv_physical_device
{
253 VK_LOADER_DATA _loader_data
;
255 struct radv_instance
* instance
;
257 struct radeon_winsys
*ws
;
258 struct radeon_info rad_info
;
262 uint64_t aperture_size
;
263 int cmd_parser_version
;
264 uint32_t pci_vendor_id
;
265 uint32_t pci_device_id
;
267 uint8_t uuid
[VK_UUID_SIZE
];
269 struct wsi_device wsi_device
;
272 struct radv_instance
{
273 VK_LOADER_DATA _loader_data
;
275 VkAllocationCallbacks alloc
;
278 int physicalDeviceCount
;
279 struct radv_physical_device physicalDevice
;
282 VkResult
radv_init_wsi(struct radv_physical_device
*physical_device
);
283 void radv_finish_wsi(struct radv_physical_device
*physical_device
);
287 struct radv_pipeline_cache
{
288 struct radv_device
* device
;
289 pthread_mutex_t mutex
;
293 uint32_t kernel_count
;
294 struct cache_entry
** hash_table
;
297 VkAllocationCallbacks alloc
;
301 radv_pipeline_cache_init(struct radv_pipeline_cache
*cache
,
302 struct radv_device
*device
);
304 radv_pipeline_cache_finish(struct radv_pipeline_cache
*cache
);
306 radv_pipeline_cache_load(struct radv_pipeline_cache
*cache
,
307 const void *data
, size_t size
);
309 struct radv_shader_variant
*
310 radv_create_shader_variant_from_pipeline_cache(struct radv_device
*device
,
311 struct radv_pipeline_cache
*cache
,
312 const unsigned char *sha1
);
314 struct radv_shader_variant
*
315 radv_pipeline_cache_insert_shader(struct radv_pipeline_cache
*cache
,
316 const unsigned char *sha1
,
317 struct radv_shader_variant
*variant
,
318 const void *code
, unsigned code_size
);
320 void radv_shader_variant_destroy(struct radv_device
*device
,
321 struct radv_shader_variant
*variant
);
323 struct radv_meta_state
{
324 VkAllocationCallbacks alloc
;
326 struct radv_pipeline_cache cache
;
329 * Use array element `i` for images with `2^i` samples.
332 VkRenderPass render_pass
[NUM_META_FS_KEYS
];
333 struct radv_pipeline
*color_pipelines
[NUM_META_FS_KEYS
];
335 VkRenderPass depth_only_rp
[NUM_DEPTH_CLEAR_PIPELINES
];
336 struct radv_pipeline
*depth_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
337 VkRenderPass stencil_only_rp
[NUM_DEPTH_CLEAR_PIPELINES
];
338 struct radv_pipeline
*stencil_only_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
339 VkRenderPass depthstencil_rp
[NUM_DEPTH_CLEAR_PIPELINES
];
340 struct radv_pipeline
*depthstencil_pipeline
[NUM_DEPTH_CLEAR_PIPELINES
];
341 } clear
[1 + MAX_SAMPLES_LOG2
];
344 VkRenderPass render_pass
[NUM_META_FS_KEYS
];
346 /** Pipeline that blits from a 1D image. */
347 VkPipeline pipeline_1d_src
[NUM_META_FS_KEYS
];
349 /** Pipeline that blits from a 2D image. */
350 VkPipeline pipeline_2d_src
[NUM_META_FS_KEYS
];
352 /** Pipeline that blits from a 3D image. */
353 VkPipeline pipeline_3d_src
[NUM_META_FS_KEYS
];
355 VkRenderPass depth_only_rp
;
356 VkPipeline depth_only_1d_pipeline
;
357 VkPipeline depth_only_2d_pipeline
;
358 VkPipeline depth_only_3d_pipeline
;
360 VkRenderPass stencil_only_rp
;
361 VkPipeline stencil_only_1d_pipeline
;
362 VkPipeline stencil_only_2d_pipeline
;
363 VkPipeline stencil_only_3d_pipeline
;
364 VkPipelineLayout pipeline_layout
;
365 VkDescriptorSetLayout ds_layout
;
369 VkRenderPass render_passes
[NUM_META_FS_KEYS
];
371 VkPipelineLayout p_layouts
[2];
372 VkDescriptorSetLayout ds_layouts
[2];
373 VkPipeline pipelines
[2][NUM_META_FS_KEYS
];
375 VkRenderPass depth_only_rp
;
376 VkPipeline depth_only_pipeline
[2];
378 VkRenderPass stencil_only_rp
;
379 VkPipeline stencil_only_pipeline
[2];
383 VkPipelineLayout img_p_layout
;
384 VkDescriptorSetLayout img_ds_layout
;
388 VkRenderPass render_pass
;
389 VkPipelineLayout img_p_layout
;
390 VkDescriptorSetLayout img_ds_layout
;
400 VkDescriptorSetLayout ds_layout
;
401 VkPipelineLayout p_layout
;
404 VkPipeline i_pipeline
;
405 } rc
[MAX_SAMPLES_LOG2
];
409 VkPipeline decompress_pipeline
;
410 VkPipeline resummarize_pipeline
;
415 VkPipeline cmask_eliminate_pipeline
;
416 VkPipeline fmask_decompress_pipeline
;
421 VkPipelineLayout fill_p_layout
;
422 VkPipelineLayout copy_p_layout
;
423 VkDescriptorSetLayout fill_ds_layout
;
424 VkDescriptorSetLayout copy_ds_layout
;
425 VkPipeline fill_pipeline
;
426 VkPipeline copy_pipeline
;
431 VK_LOADER_DATA _loader_data
;
433 struct radv_device
* device
;
435 struct radv_state_pool
* pool
;
439 VK_LOADER_DATA _loader_data
;
441 VkAllocationCallbacks alloc
;
443 struct radv_instance
* instance
;
444 struct radeon_winsys
*ws
;
445 struct radeon_winsys_ctx
*hw_ctx
;
447 struct radv_meta_state meta_state
;
448 struct radv_queue queue
;
449 struct radeon_winsys_cs
*empty_cs
;
451 bool allow_fast_clears
;
453 bool shader_stats_dump
;
455 /* MSAA sample locations.
456 * The first index is the sample index.
457 * The second index is the coordinate: X, Y. */
458 float sample_locations_1x
[1][2];
459 float sample_locations_2x
[2][2];
460 float sample_locations_4x
[4][2];
461 float sample_locations_8x
[8][2];
462 float sample_locations_16x
[16][2];
465 struct radv_device_memory
{
466 struct radeon_winsys_bo
*bo
;
468 VkDeviceSize map_size
;
473 struct radv_descriptor_range
{
478 struct radv_descriptor_set
{
479 const struct radv_descriptor_set_layout
*layout
;
480 struct list_head descriptor_pool
;
483 struct radv_buffer_view
*buffer_views
;
484 struct radeon_winsys_bo
*bo
;
486 uint32_t *mapped_ptr
;
487 struct radv_descriptor_range
*dynamic_descriptors
;
488 struct radeon_winsys_bo
*descriptors
[0];
491 struct radv_descriptor_pool_free_node
{
497 struct radv_descriptor_pool
{
498 struct list_head descriptor_sets
;
500 struct radeon_winsys_bo
*bo
;
502 uint64_t current_offset
;
508 struct radv_descriptor_pool_free_node free_nodes
[];
512 struct radv_device
* device
;
515 VkBufferUsageFlags usage
;
518 struct radeon_winsys_bo
* bo
;
523 enum radv_cmd_dirty_bits
{
524 RADV_CMD_DIRTY_DYNAMIC_VIEWPORT
= 1 << 0, /* VK_DYNAMIC_STATE_VIEWPORT */
525 RADV_CMD_DIRTY_DYNAMIC_SCISSOR
= 1 << 1, /* VK_DYNAMIC_STATE_SCISSOR */
526 RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH
= 1 << 2, /* VK_DYNAMIC_STATE_LINE_WIDTH */
527 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS
= 1 << 3, /* VK_DYNAMIC_STATE_DEPTH_BIAS */
528 RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS
= 1 << 4, /* VK_DYNAMIC_STATE_BLEND_CONSTANTS */
529 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS
= 1 << 5, /* VK_DYNAMIC_STATE_DEPTH_BOUNDS */
530 RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6, /* VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK */
531 RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7, /* VK_DYNAMIC_STATE_STENCIL_WRITE_MASK */
532 RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE
= 1 << 8, /* VK_DYNAMIC_STATE_STENCIL_REFERENCE */
533 RADV_CMD_DIRTY_DYNAMIC_ALL
= (1 << 9) - 1,
534 RADV_CMD_DIRTY_PIPELINE
= 1 << 9,
535 RADV_CMD_DIRTY_INDEX_BUFFER
= 1 << 10,
536 RADV_CMD_DIRTY_RENDER_TARGETS
= 1 << 11,
538 typedef uint32_t radv_cmd_dirty_mask_t
;
540 enum radv_cmd_flush_bits
{
541 RADV_CMD_FLAG_INV_ICACHE
= 1 << 0,
542 /* SMEM L1, other names: KCACHE, constant cache, DCACHE, data cache */
543 RADV_CMD_FLAG_INV_SMEM_L1
= 1 << 1,
544 /* VMEM L1 can optionally be bypassed (GLC=1). Other names: TC L1 */
545 RADV_CMD_FLAG_INV_VMEM_L1
= 1 << 2,
546 /* Used by everything except CB/DB, can be bypassed (SLC=1). Other names: TC L2 */
547 RADV_CMD_FLAG_INV_GLOBAL_L2
= 1 << 3,
548 /* Framebuffer caches */
549 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
= 1 << 4,
550 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
= 1 << 5,
551 RADV_CMD_FLAG_FLUSH_AND_INV_DB
= 1 << 6,
552 RADV_CMD_FLAG_FLUSH_AND_INV_CB
= 1 << 7,
553 /* Engine synchronization. */
554 RADV_CMD_FLAG_VS_PARTIAL_FLUSH
= 1 << 8,
555 RADV_CMD_FLAG_PS_PARTIAL_FLUSH
= 1 << 9,
556 RADV_CMD_FLAG_CS_PARTIAL_FLUSH
= 1 << 10,
557 RADV_CMD_FLAG_VGT_FLUSH
= 1 << 11,
559 RADV_CMD_FLUSH_AND_INV_FRAMEBUFFER
= (RADV_CMD_FLAG_FLUSH_AND_INV_CB
|
560 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
|
561 RADV_CMD_FLAG_FLUSH_AND_INV_DB
|
562 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
)
565 struct radv_vertex_binding
{
566 struct radv_buffer
* buffer
;
570 struct radv_dynamic_state
{
573 VkViewport viewports
[MAX_VIEWPORTS
];
578 VkRect2D scissors
[MAX_SCISSORS
];
589 float blend_constants
[4];
599 } stencil_compare_mask
;
604 } stencil_write_mask
;
612 extern const struct radv_dynamic_state default_dynamic_state
;
614 void radv_dynamic_state_copy(struct radv_dynamic_state
*dest
,
615 const struct radv_dynamic_state
*src
,
618 * Attachment state when recording a renderpass instance.
620 * The clear value is valid only if there exists a pending clear.
622 struct radv_attachment_state
{
623 VkImageAspectFlags pending_clear_aspects
;
624 VkClearValue clear_value
;
625 VkImageLayout current_layout
;
628 struct radv_cmd_state
{
630 bool vertex_descriptors_dirty
;
631 radv_cmd_dirty_mask_t dirty
;
633 struct radv_pipeline
* pipeline
;
634 struct radv_pipeline
* emitted_pipeline
;
635 struct radv_pipeline
* compute_pipeline
;
636 struct radv_pipeline
* emitted_compute_pipeline
;
637 struct radv_framebuffer
* framebuffer
;
638 struct radv_render_pass
* pass
;
639 const struct radv_subpass
* subpass
;
640 struct radv_dynamic_state dynamic
;
641 struct radv_vertex_binding vertex_bindings
[MAX_VBS
];
642 struct radv_descriptor_set
* descriptors
[MAX_SETS
];
643 struct radv_attachment_state
* attachments
;
644 VkRect2D render_area
;
645 struct radv_buffer
* index_buffer
;
647 uint32_t index_offset
;
648 uint32_t last_primitive_reset_index
;
649 enum radv_cmd_flush_bits flush_bits
;
650 unsigned active_occlusion_queries
;
652 uint32_t descriptors_dirty
;
654 struct radv_cmd_pool
{
655 VkAllocationCallbacks alloc
;
656 struct list_head cmd_buffers
;
659 struct radv_cmd_buffer_upload
{
663 struct radeon_winsys_bo
*upload_bo
;
664 struct list_head list
;
667 struct radv_cmd_buffer
{
668 VK_LOADER_DATA _loader_data
;
670 struct radv_device
* device
;
672 struct radv_cmd_pool
* pool
;
673 struct list_head pool_link
;
675 VkCommandBufferUsageFlags usage_flags
;
676 VkCommandBufferLevel level
;
677 struct radeon_winsys_cs
*cs
;
678 struct radv_cmd_state state
;
680 uint8_t push_constants
[MAX_PUSH_CONSTANTS_SIZE
];
681 uint32_t dynamic_buffers
[16 * MAX_DYNAMIC_BUFFERS
];
682 VkShaderStageFlags push_constant_stages
;
684 struct radv_cmd_buffer_upload upload
;
691 void si_init_config(struct radv_physical_device
*physical_device
,
692 struct radv_cmd_buffer
*cmd_buffer
);
693 void si_write_viewport(struct radeon_winsys_cs
*cs
, int first_vp
,
694 int count
, const VkViewport
*viewports
);
695 void si_write_scissors(struct radeon_winsys_cs
*cs
, int first
,
696 int count
, const VkRect2D
*scissors
);
697 uint32_t si_get_ia_multi_vgt_param(struct radv_cmd_buffer
*cmd_buffer
);
698 void si_emit_cache_flush(struct radv_cmd_buffer
*cmd_buffer
);
699 void si_cp_dma_buffer_copy(struct radv_cmd_buffer
*cmd_buffer
,
700 uint64_t src_va
, uint64_t dest_va
,
702 void si_cp_dma_clear_buffer(struct radv_cmd_buffer
*cmd_buffer
, uint64_t va
,
703 uint64_t size
, unsigned value
);
704 void radv_set_db_count_control(struct radv_cmd_buffer
*cmd_buffer
);
705 void radv_bind_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
706 struct radv_descriptor_set
*set
,
709 radv_cmd_buffer_upload_alloc(struct radv_cmd_buffer
*cmd_buffer
,
712 unsigned *out_offset
,
715 radv_cmd_buffer_set_subpass(struct radv_cmd_buffer
*cmd_buffer
,
716 const struct radv_subpass
*subpass
,
719 radv_cmd_buffer_upload_data(struct radv_cmd_buffer
*cmd_buffer
,
720 unsigned size
, unsigned alignmnet
,
721 const void *data
, unsigned *out_offset
);
723 radv_emit_framebuffer_state(struct radv_cmd_buffer
*cmd_buffer
);
724 void radv_cmd_buffer_clear_subpass(struct radv_cmd_buffer
*cmd_buffer
);
725 void radv_cmd_buffer_resolve_subpass(struct radv_cmd_buffer
*cmd_buffer
);
726 void radv_cayman_emit_msaa_sample_locs(struct radeon_winsys_cs
*cs
, int nr_samples
);
727 unsigned radv_cayman_get_maxdist(int log_samples
);
728 void radv_device_init_msaa(struct radv_device
*device
);
729 void radv_set_depth_clear_regs(struct radv_cmd_buffer
*cmd_buffer
,
730 struct radv_image
*image
,
731 VkClearDepthStencilValue ds_clear_value
,
732 VkImageAspectFlags aspects
);
733 void radv_set_color_clear_regs(struct radv_cmd_buffer
*cmd_buffer
,
734 struct radv_image
*image
,
736 uint32_t color_values
[2]);
737 void radv_fill_buffer(struct radv_cmd_buffer
*cmd_buffer
,
738 struct radeon_winsys_bo
*bo
,
739 uint64_t offset
, uint64_t size
, uint32_t value
);
742 * Takes x,y,z as exact numbers of invocations, instead of blocks.
744 * Limitations: Can't call normal dispatch functions without binding or rebinding
745 * the compute pipeline.
747 void radv_unaligned_dispatch(
748 struct radv_cmd_buffer
*cmd_buffer
,
754 struct radeon_winsys_bo
*bo
;
760 struct radv_shader_module
{
761 struct nir_shader
* nir
;
762 unsigned char sha1
[20];
767 union ac_shader_variant_key
;
770 radv_hash_shader(unsigned char *hash
, struct radv_shader_module
*module
,
771 const char *entrypoint
,
772 const VkSpecializationInfo
*spec_info
,
773 const struct radv_pipeline_layout
*layout
,
774 const union ac_shader_variant_key
*key
);
776 static inline gl_shader_stage
777 vk_to_mesa_shader_stage(VkShaderStageFlagBits vk_stage
)
779 assert(__builtin_popcount(vk_stage
) == 1);
780 return ffs(vk_stage
) - 1;
783 static inline VkShaderStageFlagBits
784 mesa_to_vk_shader_stage(gl_shader_stage mesa_stage
)
786 return (1 << mesa_stage
);
789 #define RADV_STAGE_MASK ((1 << MESA_SHADER_STAGES) - 1)
791 #define radv_foreach_stage(stage, stage_bits) \
792 for (gl_shader_stage stage, \
793 __tmp = (gl_shader_stage)((stage_bits) & RADV_STAGE_MASK); \
794 stage = __builtin_ffs(__tmp) - 1, __tmp; \
795 __tmp &= ~(1 << (stage)))
797 struct radv_shader_variant
{
800 struct radeon_winsys_bo
*bo
;
801 struct ac_shader_config config
;
802 struct ac_shader_variant_info info
;
808 struct radv_depth_stencil_state
{
809 uint32_t db_depth_control
;
810 uint32_t db_stencil_control
;
811 uint32_t db_render_control
;
812 uint32_t db_render_override2
;
815 struct radv_blend_state
{
816 uint32_t cb_color_control
;
817 uint32_t cb_target_mask
;
818 uint32_t sx_mrt0_blend_opt
[8];
819 uint32_t cb_blend_control
[8];
821 uint32_t spi_shader_col_format
;
822 uint32_t cb_shader_mask
;
823 uint32_t db_alpha_to_mask
;
826 unsigned radv_format_meta_fs_key(VkFormat format
);
828 struct radv_raster_state
{
829 uint32_t pa_cl_clip_cntl
;
830 uint32_t pa_cl_vs_out_cntl
;
831 uint32_t spi_interp_control
;
832 uint32_t pa_su_point_size
;
833 uint32_t pa_su_point_minmax
;
834 uint32_t pa_su_line_cntl
;
835 uint32_t pa_su_vtx_cntl
;
836 uint32_t pa_su_sc_mode_cntl
;
839 struct radv_multisample_state
{
841 uint32_t pa_sc_line_cntl
;
842 uint32_t pa_sc_mode_cntl_0
;
843 uint32_t pa_sc_mode_cntl_1
;
844 uint32_t pa_sc_aa_config
;
845 uint32_t pa_sc_aa_mask
[2];
846 unsigned num_samples
;
849 struct radv_pipeline
{
850 struct radv_device
* device
;
851 uint32_t dynamic_state_mask
;
852 struct radv_dynamic_state dynamic_state
;
854 struct radv_pipeline_layout
* layout
;
856 bool needs_data_cache
;
858 struct radv_shader_variant
* shaders
[MESA_SHADER_STAGES
];
859 VkShaderStageFlags active_stages
;
861 uint32_t va_rsrc_word3
[MAX_VERTEX_ATTRIBS
];
862 uint32_t va_format_size
[MAX_VERTEX_ATTRIBS
];
863 uint32_t va_binding
[MAX_VERTEX_ATTRIBS
];
864 uint32_t va_offset
[MAX_VERTEX_ATTRIBS
];
865 uint32_t num_vertex_attribs
;
866 uint32_t binding_stride
[MAX_VBS
];
870 struct radv_blend_state blend
;
871 struct radv_depth_stencil_state ds
;
872 struct radv_raster_state raster
;
873 struct radv_multisample_state ms
;
876 bool prim_restart_enable
;
881 struct radv_graphics_pipeline_create_info
{
884 bool db_stencil_clear
;
885 bool db_depth_disable_expclear
;
886 bool db_stencil_disable_expclear
;
887 bool db_flush_depth_inplace
;
888 bool db_flush_stencil_inplace
;
890 uint32_t custom_blend_mode
;
894 radv_pipeline_init(struct radv_pipeline
*pipeline
, struct radv_device
*device
,
895 struct radv_pipeline_cache
*cache
,
896 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
897 const struct radv_graphics_pipeline_create_info
*extra
,
898 const VkAllocationCallbacks
*alloc
);
901 radv_graphics_pipeline_create(VkDevice device
,
902 VkPipelineCache cache
,
903 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
904 const struct radv_graphics_pipeline_create_info
*extra
,
905 const VkAllocationCallbacks
*alloc
,
906 VkPipeline
*pPipeline
);
908 struct vk_format_description
;
909 uint32_t radv_translate_buffer_dataformat(const struct vk_format_description
*desc
,
911 uint32_t radv_translate_buffer_numformat(const struct vk_format_description
*desc
,
913 uint32_t radv_translate_colorformat(VkFormat format
);
914 uint32_t radv_translate_color_numformat(VkFormat format
,
915 const struct vk_format_description
*desc
,
917 uint32_t radv_colorformat_endian_swap(uint32_t colorformat
);
918 unsigned radv_translate_colorswap(VkFormat format
, bool do_endian_swap
);
919 uint32_t radv_translate_dbformat(VkFormat format
);
920 uint32_t radv_translate_tex_dataformat(VkFormat format
,
921 const struct vk_format_description
*desc
,
923 uint32_t radv_translate_tex_numformat(VkFormat format
,
924 const struct vk_format_description
*desc
,
926 bool radv_format_pack_clear_color(VkFormat format
,
927 uint32_t clear_vals
[2],
928 VkClearColorValue
*value
);
929 bool radv_is_colorbuffer_format_supported(VkFormat format
, bool *blendable
);
931 struct radv_fmask_info
{
935 unsigned pitch_in_pixels
;
936 unsigned bank_height
;
937 unsigned slice_tile_max
;
938 unsigned tile_mode_index
;
941 struct radv_cmask_info
{
945 unsigned slice_tile_max
;
946 unsigned base_address_reg
;
949 struct r600_htile_info
{
960 /* The original VkFormat provided by the client. This may not match any
961 * of the actual surface formats.
964 VkImageAspectFlags aspects
;
968 uint32_t samples
; /**< VkImageCreateInfo::samples */
969 VkImageUsageFlags usage
; /**< Superset of VkImageCreateInfo::usage. */
970 VkImageTiling tiling
; /** VkImageCreateInfo::tiling */
976 struct radeon_winsys_bo
*bo
;
979 struct radeon_surf surface
;
981 struct radv_fmask_info fmask
;
982 struct radv_cmask_info cmask
;
983 uint32_t clear_value_offset
;
985 /* Depth buffer compression and fast clear. */
986 struct r600_htile_info htile
;
989 bool radv_layout_has_htile(const struct radv_image
*image
,
990 VkImageLayout layout
);
991 bool radv_layout_is_htile_compressed(const struct radv_image
*image
,
992 VkImageLayout layout
);
993 bool radv_layout_can_expclear(const struct radv_image
*image
,
994 VkImageLayout layout
);
995 bool radv_layout_has_cmask(const struct radv_image
*image
,
996 VkImageLayout layout
);
997 static inline uint32_t
998 radv_get_layerCount(const struct radv_image
*image
,
999 const VkImageSubresourceRange
*range
)
1001 return range
->layerCount
== VK_REMAINING_ARRAY_LAYERS
?
1002 image
->array_size
- range
->baseArrayLayer
: range
->layerCount
;
1005 static inline uint32_t
1006 radv_get_levelCount(const struct radv_image
*image
,
1007 const VkImageSubresourceRange
*range
)
1009 return range
->levelCount
== VK_REMAINING_MIP_LEVELS
?
1010 image
->levels
- range
->baseMipLevel
: range
->levelCount
;
1013 struct radeon_bo_metadata
;
1015 radv_init_metadata(struct radv_device
*device
,
1016 struct radv_image
*image
,
1017 struct radeon_bo_metadata
*metadata
);
1019 struct radv_image_view
{
1020 struct radv_image
*image
; /**< VkImageViewCreateInfo::image */
1021 struct radeon_winsys_bo
*bo
;
1023 VkImageViewType type
;
1024 VkImageAspectFlags aspect_mask
;
1026 uint32_t base_layer
;
1027 uint32_t layer_count
;
1029 VkExtent3D extent
; /**< Extent of VkImageViewCreateInfo::baseMipLevel. */
1031 uint32_t descriptor
[8];
1032 uint32_t fmask_descriptor
[8];
1035 struct radv_image_create_info
{
1036 const VkImageCreateInfo
*vk_info
;
1041 VkResult
radv_image_create(VkDevice _device
,
1042 const struct radv_image_create_info
*info
,
1043 const VkAllocationCallbacks
* alloc
,
1046 void radv_image_view_init(struct radv_image_view
*view
,
1047 struct radv_device
*device
,
1048 const VkImageViewCreateInfo
* pCreateInfo
,
1049 struct radv_cmd_buffer
*cmd_buffer
,
1050 VkImageUsageFlags usage_mask
);
1051 void radv_image_set_optimal_micro_tile_mode(struct radv_device
*device
,
1052 struct radv_image
*image
, uint32_t micro_tile_mode
);
1053 struct radv_buffer_view
{
1054 struct radeon_winsys_bo
*bo
;
1056 uint64_t range
; /**< VkBufferViewCreateInfo::range */
1059 void radv_buffer_view_init(struct radv_buffer_view
*view
,
1060 struct radv_device
*device
,
1061 const VkBufferViewCreateInfo
* pCreateInfo
,
1062 struct radv_cmd_buffer
*cmd_buffer
);
1064 static inline struct VkExtent3D
1065 radv_sanitize_image_extent(const VkImageType imageType
,
1066 const struct VkExtent3D imageExtent
)
1068 switch (imageType
) {
1069 case VK_IMAGE_TYPE_1D
:
1070 return (VkExtent3D
) { imageExtent
.width
, 1, 1 };
1071 case VK_IMAGE_TYPE_2D
:
1072 return (VkExtent3D
) { imageExtent
.width
, imageExtent
.height
, 1 };
1073 case VK_IMAGE_TYPE_3D
:
1076 unreachable("invalid image type");
1080 static inline struct VkOffset3D
1081 radv_sanitize_image_offset(const VkImageType imageType
,
1082 const struct VkOffset3D imageOffset
)
1084 switch (imageType
) {
1085 case VK_IMAGE_TYPE_1D
:
1086 return (VkOffset3D
) { imageOffset
.x
, 0, 0 };
1087 case VK_IMAGE_TYPE_2D
:
1088 return (VkOffset3D
) { imageOffset
.x
, imageOffset
.y
, 0 };
1089 case VK_IMAGE_TYPE_3D
:
1092 unreachable("invalid image type");
1096 struct radv_sampler
{
1100 struct radv_color_buffer_info
{
1101 uint32_t cb_color_base
;
1102 uint32_t cb_color_pitch
;
1103 uint32_t cb_color_slice
;
1104 uint32_t cb_color_view
;
1105 uint32_t cb_color_info
;
1106 uint32_t cb_color_attrib
;
1107 uint32_t cb_dcc_control
;
1108 uint32_t cb_color_cmask
;
1109 uint32_t cb_color_cmask_slice
;
1110 uint32_t cb_color_fmask
;
1111 uint32_t cb_color_fmask_slice
;
1112 uint32_t cb_clear_value0
;
1113 uint32_t cb_clear_value1
;
1114 uint32_t cb_dcc_base
;
1115 uint32_t micro_tile_mode
;
1118 struct radv_ds_buffer_info
{
1119 uint32_t db_depth_info
;
1121 uint32_t db_stencil_info
;
1122 uint32_t db_z_read_base
;
1123 uint32_t db_stencil_read_base
;
1124 uint32_t db_z_write_base
;
1125 uint32_t db_stencil_write_base
;
1126 uint32_t db_depth_view
;
1127 uint32_t db_depth_size
;
1128 uint32_t db_depth_slice
;
1129 uint32_t db_htile_surface
;
1130 uint32_t db_htile_data_base
;
1131 uint32_t pa_su_poly_offset_db_fmt_cntl
;
1135 struct radv_attachment_info
{
1137 struct radv_color_buffer_info cb
;
1138 struct radv_ds_buffer_info ds
;
1140 struct radv_image_view
*attachment
;
1143 struct radv_framebuffer
{
1148 uint32_t attachment_count
;
1149 struct radv_attachment_info attachments
[0];
1152 struct radv_subpass_barrier
{
1153 VkPipelineStageFlags src_stage_mask
;
1154 VkAccessFlags src_access_mask
;
1155 VkAccessFlags dst_access_mask
;
1158 struct radv_subpass
{
1159 uint32_t input_count
;
1160 VkAttachmentReference
* input_attachments
;
1161 uint32_t color_count
;
1162 VkAttachmentReference
* color_attachments
;
1163 VkAttachmentReference
* resolve_attachments
;
1164 VkAttachmentReference depth_stencil_attachment
;
1166 /** Subpass has at least one resolve attachment */
1169 struct radv_subpass_barrier start_barrier
;
1172 struct radv_render_pass_attachment
{
1175 VkAttachmentLoadOp load_op
;
1176 VkAttachmentLoadOp stencil_load_op
;
1177 VkImageLayout initial_layout
;
1178 VkImageLayout final_layout
;
1181 struct radv_render_pass
{
1182 uint32_t attachment_count
;
1183 uint32_t subpass_count
;
1184 VkAttachmentReference
* subpass_attachments
;
1185 struct radv_render_pass_attachment
* attachments
;
1186 struct radv_subpass_barrier end_barrier
;
1187 struct radv_subpass subpasses
[0];
1190 VkResult
radv_device_init_meta(struct radv_device
*device
);
1191 void radv_device_finish_meta(struct radv_device
*device
);
1193 struct radv_query_pool
{
1194 struct radeon_winsys_bo
*bo
;
1196 uint32_t availability_offset
;
1202 radv_temp_descriptor_set_create(struct radv_device
*device
,
1203 struct radv_cmd_buffer
*cmd_buffer
,
1204 VkDescriptorSetLayout _layout
,
1205 VkDescriptorSet
*_set
);
1208 radv_temp_descriptor_set_destroy(struct radv_device
*device
,
1209 VkDescriptorSet _set
);
1210 void radv_initialise_cmask(struct radv_cmd_buffer
*cmd_buffer
,
1211 struct radv_image
*image
, uint32_t value
);
1212 void radv_initialize_dcc(struct radv_cmd_buffer
*cmd_buffer
,
1213 struct radv_image
*image
, uint32_t value
);
1216 struct radeon_winsys_fence
*fence
;
1221 #define RADV_DEFINE_HANDLE_CASTS(__radv_type, __VkType) \
1223 static inline struct __radv_type * \
1224 __radv_type ## _from_handle(__VkType _handle) \
1226 return (struct __radv_type *) _handle; \
1229 static inline __VkType \
1230 __radv_type ## _to_handle(struct __radv_type *_obj) \
1232 return (__VkType) _obj; \
1235 #define RADV_DEFINE_NONDISP_HANDLE_CASTS(__radv_type, __VkType) \
1237 static inline struct __radv_type * \
1238 __radv_type ## _from_handle(__VkType _handle) \
1240 return (struct __radv_type *)(uintptr_t) _handle; \
1243 static inline __VkType \
1244 __radv_type ## _to_handle(struct __radv_type *_obj) \
1246 return (__VkType)(uintptr_t) _obj; \
1249 #define RADV_FROM_HANDLE(__radv_type, __name, __handle) \
1250 struct __radv_type *__name = __radv_type ## _from_handle(__handle)
1252 RADV_DEFINE_HANDLE_CASTS(radv_cmd_buffer
, VkCommandBuffer
)
1253 RADV_DEFINE_HANDLE_CASTS(radv_device
, VkDevice
)
1254 RADV_DEFINE_HANDLE_CASTS(radv_instance
, VkInstance
)
1255 RADV_DEFINE_HANDLE_CASTS(radv_physical_device
, VkPhysicalDevice
)
1256 RADV_DEFINE_HANDLE_CASTS(radv_queue
, VkQueue
)
1258 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_cmd_pool
, VkCommandPool
)
1259 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer
, VkBuffer
)
1260 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_buffer_view
, VkBufferView
)
1261 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_pool
, VkDescriptorPool
)
1262 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set
, VkDescriptorSet
)
1263 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set_layout
, VkDescriptorSetLayout
)
1264 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_device_memory
, VkDeviceMemory
)
1265 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_fence
, VkFence
)
1266 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_event
, VkEvent
)
1267 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_framebuffer
, VkFramebuffer
)
1268 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image
, VkImage
)
1269 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_image_view
, VkImageView
);
1270 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_cache
, VkPipelineCache
)
1271 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline
, VkPipeline
)
1272 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_pipeline_layout
, VkPipelineLayout
)
1273 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_query_pool
, VkQueryPool
)
1274 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_render_pass
, VkRenderPass
)
1275 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_sampler
, VkSampler
)
1276 RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_shader_module
, VkShaderModule
)
1278 #define RADV_DEFINE_STRUCT_CASTS(__radv_type, __VkType) \
1280 static inline const __VkType * \
1281 __radv_type ## _to_ ## __VkType(const struct __radv_type *__radv_obj) \
1283 return (const __VkType *) __radv_obj; \
1286 #define RADV_COMMON_TO_STRUCT(__VkType, __vk_name, __common_name) \
1287 const __VkType *__vk_name = radv_common_to_ ## __VkType(__common_name)
1289 RADV_DEFINE_STRUCT_CASTS(radv_common
, VkMemoryBarrier
)
1290 RADV_DEFINE_STRUCT_CASTS(radv_common
, VkBufferMemoryBarrier
)
1291 RADV_DEFINE_STRUCT_CASTS(radv_common
, VkImageMemoryBarrier
)
1294 #endif /* RADV_PRIVATE_H */