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
25 * DEALINGS IN THE SOFTWARE.
43 #define VG(x) ((void)0)
46 #include "c11/threads.h"
47 #include "main/macros.h"
48 #include "util/list.h"
49 #include "util/macros.h"
51 #include "vk_debug_report.h"
52 #include "wsi_common.h"
54 #include "drm-uapi/msm_drm.h"
55 #include "ir3/ir3_compiler.h"
56 #include "ir3/ir3_shader.h"
58 #include "adreno_common.xml.h"
59 #include "adreno_pm4.xml.h"
61 #include "fdl/freedreno_layout.h"
63 #include "tu_descriptor_set.h"
64 #include "tu_extensions.h"
66 /* Pre-declarations needed for WSI entrypoints */
69 typedef struct xcb_connection_t xcb_connection_t
;
70 typedef uint32_t xcb_visualid_t
;
71 typedef uint32_t xcb_window_t
;
73 #include <vulkan/vk_android_native_buffer.h>
74 #include <vulkan/vk_icd.h>
75 #include <vulkan/vulkan.h>
76 #include <vulkan/vulkan_intel.h>
78 #include "tu_entrypoints.h"
81 #define MAX_VERTEX_ATTRIBS 32
83 #define MAX_VSC_PIPES 32
84 #define MAX_VIEWPORTS 1
85 #define MAX_SCISSORS 16
86 #define MAX_DISCARD_RECTANGLES 4
87 #define MAX_PUSH_CONSTANTS_SIZE 128
88 #define MAX_PUSH_DESCRIPTORS 32
89 #define MAX_DYNAMIC_UNIFORM_BUFFERS 16
90 #define MAX_DYNAMIC_STORAGE_BUFFERS 8
91 #define MAX_DYNAMIC_BUFFERS \
92 (MAX_DYNAMIC_UNIFORM_BUFFERS + MAX_DYNAMIC_STORAGE_BUFFERS)
93 #define MAX_SAMPLES_LOG2 4
94 #define NUM_META_FS_KEYS 13
95 #define TU_MAX_DRM_DEVICES 8
97 /* The Qualcomm driver exposes 0x20000058 */
98 #define MAX_STORAGE_BUFFER_RANGE 0x20000000
100 #define NUM_DEPTH_CLEAR_PIPELINES 3
103 * This is the point we switch from using CP to compute shader
104 * for certain buffer operations.
106 #define TU_BUFFER_OPS_CS_THRESHOLD 4096
108 #define A6XX_TEX_CONST_DWORDS 16
109 #define A6XX_TEX_SAMP_DWORDS 4
114 TU_MEM_HEAP_VRAM_CPU_ACCESS
,
122 TU_MEM_TYPE_GTT_WRITE_COMBINE
,
123 TU_MEM_TYPE_VRAM_CPU_ACCESS
,
124 TU_MEM_TYPE_GTT_CACHED
,
128 #define tu_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
130 static inline uint32_t
131 align_u32(uint32_t v
, uint32_t a
)
133 assert(a
!= 0 && a
== (a
& -a
));
134 return (v
+ a
- 1) & ~(a
- 1);
137 static inline uint32_t
138 align_u32_npot(uint32_t v
, uint32_t a
)
140 return (v
+ a
- 1) / a
* a
;
143 static inline uint64_t
144 align_u64(uint64_t v
, uint64_t a
)
146 assert(a
!= 0 && a
== (a
& -a
));
147 return (v
+ a
- 1) & ~(a
- 1);
150 static inline int32_t
151 align_i32(int32_t v
, int32_t a
)
153 assert(a
!= 0 && a
== (a
& -a
));
154 return (v
+ a
- 1) & ~(a
- 1);
157 /** Alignment must be a power of 2. */
159 tu_is_aligned(uintmax_t n
, uintmax_t a
)
161 assert(a
== (a
& -a
));
162 return (n
& (a
- 1)) == 0;
165 static inline uint32_t
166 round_up_u32(uint32_t v
, uint32_t a
)
168 return (v
+ a
- 1) / a
;
171 static inline uint64_t
172 round_up_u64(uint64_t v
, uint64_t a
)
174 return (v
+ a
- 1) / a
;
177 static inline uint32_t
178 tu_minify(uint32_t n
, uint32_t levels
)
180 if (unlikely(n
== 0))
183 return MAX2(n
>> levels
, 1);
186 tu_clamp_f(float f
, float min
, float max
)
199 tu_clear_mask(uint32_t *inout_mask
, uint32_t clear_mask
)
201 if (*inout_mask
& clear_mask
) {
202 *inout_mask
&= ~clear_mask
;
209 #define for_each_bit(b, dword) \
210 for (uint32_t __dword = (dword); \
211 (b) = __builtin_ffs(__dword) - 1, __dword; __dword &= ~(1 << (b)))
213 #define typed_memcpy(dest, src, count) \
215 STATIC_ASSERT(sizeof(*src) == sizeof(*dest)); \
216 memcpy((dest), (src), (count) * sizeof(*(src))); \
219 #define COND(bool, val) ((bool) ? (val) : 0)
221 /* Whenever we generate an error, pass it through this function. Useful for
222 * debugging, where we can break on it. Only call at error site, not when
223 * propagating errors. Might be useful to plug in a stack trace here.
229 __vk_errorf(struct tu_instance
*instance
,
236 #define vk_error(instance, error) \
237 __vk_errorf(instance, error, __FILE__, __LINE__, NULL);
238 #define vk_errorf(instance, error, format, ...) \
239 __vk_errorf(instance, error, __FILE__, __LINE__, format, ##__VA_ARGS__);
242 __tu_finishme(const char *file
, int line
, const char *format
, ...)
245 tu_loge(const char *format
, ...) tu_printflike(1, 2);
247 tu_loge_v(const char *format
, va_list va
);
249 tu_logi(const char *format
, ...) tu_printflike(1, 2);
251 tu_logi_v(const char *format
, va_list va
);
254 * Print a FINISHME message, including its source location.
256 #define tu_finishme(format, ...) \
258 static bool reported = false; \
260 __tu_finishme(__FILE__, __LINE__, format, ##__VA_ARGS__); \
265 /* A non-fatal assert. Useful for debugging. */
267 #define tu_assert(x) \
269 if (unlikely(!(x))) \
270 fprintf(stderr, "%s:%d ASSERT: %s\n", __FILE__, __LINE__, #x); \
276 /* Suppress -Wunused in stub functions */
277 #define tu_use_args(...) __tu_use_args(0, ##__VA_ARGS__)
279 __tu_use_args(int ignore
, ...)
285 tu_finishme("stub %s", __func__); \
289 tu_lookup_entrypoint_unchecked(const char *name
);
291 tu_lookup_entrypoint_checked(
293 uint32_t core_version
,
294 const struct tu_instance_extension_table
*instance
,
295 const struct tu_device_extension_table
*device
);
297 struct tu_physical_device
299 VK_LOADER_DATA _loader_data
;
301 struct tu_instance
*instance
;
304 char name
[VK_MAX_PHYSICAL_DEVICE_NAME_SIZE
];
305 uint8_t driver_uuid
[VK_UUID_SIZE
];
306 uint8_t device_uuid
[VK_UUID_SIZE
];
307 uint8_t cache_uuid
[VK_UUID_SIZE
];
309 struct wsi_device wsi_device
;
317 uint32_t ccu_offset_gmem
;
318 uint32_t ccu_offset_bypass
;
319 #define GMEM_ALIGN_W 16
320 #define GMEM_ALIGN_H 4
323 uint32_t RB_UNKNOWN_8E04_blit
; /* for CP_BLIT's */
324 uint32_t PC_UNKNOWN_9805
;
325 uint32_t SP_UNKNOWN_A0F8
;
328 /* This is the drivers on-disk cache used as a fallback as opposed to
329 * the pipeline cache defined by apps.
331 struct disk_cache
*disk_cache
;
333 struct tu_device_extension_table supported_extensions
;
338 TU_DEBUG_STARTUP
= 1 << 0,
339 TU_DEBUG_NIR
= 1 << 1,
340 TU_DEBUG_IR3
= 1 << 2,
341 TU_DEBUG_NOBIN
= 1 << 3,
342 TU_DEBUG_SYSMEM
= 1 << 4,
343 TU_DEBUG_FORCEBIN
= 1 << 5,
348 VK_LOADER_DATA _loader_data
;
350 VkAllocationCallbacks alloc
;
352 uint32_t api_version
;
353 int physical_device_count
;
354 struct tu_physical_device physical_devices
[TU_MAX_DRM_DEVICES
];
356 enum tu_debug_flags debug_flags
;
358 struct vk_debug_report_instance debug_report_callbacks
;
360 struct tu_instance_extension_table enabled_extensions
;
364 tu_wsi_init(struct tu_physical_device
*physical_device
);
366 tu_wsi_finish(struct tu_physical_device
*physical_device
);
369 tu_instance_extension_supported(const char *name
);
371 tu_physical_device_api_version(struct tu_physical_device
*dev
);
373 tu_physical_device_extension_supported(struct tu_physical_device
*dev
,
378 struct tu_pipeline_cache
380 struct tu_device
*device
;
381 pthread_mutex_t mutex
;
385 uint32_t kernel_count
;
386 struct cache_entry
**hash_table
;
389 VkAllocationCallbacks alloc
;
392 struct tu_pipeline_key
397 tu_pipeline_cache_init(struct tu_pipeline_cache
*cache
,
398 struct tu_device
*device
);
400 tu_pipeline_cache_finish(struct tu_pipeline_cache
*cache
);
402 tu_pipeline_cache_load(struct tu_pipeline_cache
*cache
,
406 struct tu_shader_variant
;
409 tu_create_shader_variants_from_pipeline_cache(
410 struct tu_device
*device
,
411 struct tu_pipeline_cache
*cache
,
412 const unsigned char *sha1
,
413 struct tu_shader_variant
**variants
);
416 tu_pipeline_cache_insert_shaders(struct tu_device
*device
,
417 struct tu_pipeline_cache
*cache
,
418 const unsigned char *sha1
,
419 struct tu_shader_variant
**variants
,
420 const void *const *codes
,
421 const unsigned *code_sizes
);
425 VkAllocationCallbacks alloc
;
427 struct tu_pipeline_cache cache
;
431 #define TU_QUEUE_GENERAL 0
433 #define TU_MAX_QUEUE_FAMILIES 1
437 struct wsi_fence
*fence_wsi
;
443 tu_fence_init(struct tu_fence
*fence
, bool signaled
);
445 tu_fence_finish(struct tu_fence
*fence
);
447 tu_fence_update_fd(struct tu_fence
*fence
, int fd
);
449 tu_fence_copy(struct tu_fence
*fence
, const struct tu_fence
*src
);
451 tu_fence_signal(struct tu_fence
*fence
);
453 tu_fence_wait_idle(struct tu_fence
*fence
);
457 VK_LOADER_DATA _loader_data
;
458 struct tu_device
*device
;
459 uint32_t queue_family_index
;
461 VkDeviceQueueCreateFlags flags
;
463 uint32_t msm_queue_id
;
464 struct tu_fence submit_fence
;
477 VK_LOADER_DATA _loader_data
;
479 VkAllocationCallbacks alloc
;
481 struct tu_instance
*instance
;
483 struct tu_meta_state meta_state
;
485 struct tu_queue
*queues
[TU_MAX_QUEUE_FAMILIES
];
486 int queue_count
[TU_MAX_QUEUE_FAMILIES
];
488 struct tu_physical_device
*physical_device
;
490 struct ir3_compiler
*compiler
;
492 /* Backup in-memory cache to be used if the app doesn't provide one */
493 struct tu_pipeline_cache
*mem_cache
;
495 struct tu_bo vsc_data
;
496 struct tu_bo vsc_data2
;
497 uint32_t vsc_data_pitch
;
498 uint32_t vsc_data2_pitch
;
500 struct tu_bo border_color
;
502 struct list_head shader_slabs
;
503 mtx_t shader_slab_mutex
;
505 struct tu_device_extension_table enabled_extensions
;
509 tu_bo_init_new(struct tu_device
*dev
, struct tu_bo
*bo
, uint64_t size
);
511 tu_bo_init_dmabuf(struct tu_device
*dev
,
516 tu_bo_export_dmabuf(struct tu_device
*dev
, struct tu_bo
*bo
);
518 tu_bo_finish(struct tu_device
*dev
, struct tu_bo
*bo
);
520 tu_bo_map(struct tu_device
*dev
, struct tu_bo
*bo
);
525 const struct tu_bo
*bo
;
531 struct ts_cs_memory
{
540 * A command stream in TU_CS_MODE_GROW mode grows automatically whenever it
541 * is full. tu_cs_begin must be called before command packet emission and
542 * tu_cs_end must be called after.
544 * This mode may create multiple entries internally. The entries must be
545 * submitted together.
550 * A command stream in TU_CS_MODE_EXTERNAL mode wraps an external,
551 * fixed-size buffer. tu_cs_begin and tu_cs_end are optional and have no
554 * This mode does not create any entry or any BO.
559 * A command stream in TU_CS_MODE_SUB_STREAM mode does not support direct
560 * command packet emission. tu_cs_begin_sub_stream must be called to get a
561 * sub-stream to emit comamnd packets to. When done with the sub-stream,
562 * tu_cs_end_sub_stream must be called.
564 * This mode does not create any entry internally.
566 TU_CS_MODE_SUB_STREAM
,
573 uint32_t *reserved_end
;
576 struct tu_device
*device
;
577 enum tu_cs_mode mode
;
578 uint32_t next_bo_size
;
580 struct tu_cs_entry
*entries
;
581 uint32_t entry_count
;
582 uint32_t entry_capacity
;
586 uint32_t bo_capacity
;
588 /* state for cond_exec_start/cond_exec_end */
590 uint32_t *cond_dwords
;
593 struct tu_device_memory
598 /* for dedicated allocations */
599 struct tu_image
*image
;
600 struct tu_buffer
*buffer
;
607 struct tu_descriptor_range
613 struct tu_descriptor_set
615 const struct tu_descriptor_set_layout
*layout
;
619 uint32_t *mapped_ptr
;
620 struct tu_descriptor_range
*dynamic_descriptors
;
622 struct tu_bo
*descriptors
[0];
625 struct tu_push_descriptor_set
627 struct tu_descriptor_set set
;
631 struct tu_descriptor_pool_entry
635 struct tu_descriptor_set
*set
;
638 struct tu_descriptor_pool
641 uint64_t current_offset
;
644 uint8_t *host_memory_base
;
645 uint8_t *host_memory_ptr
;
646 uint8_t *host_memory_end
;
648 uint32_t entry_count
;
649 uint32_t max_entry_count
;
650 struct tu_descriptor_pool_entry entries
[0];
653 struct tu_descriptor_update_template_entry
655 VkDescriptorType descriptor_type
;
657 /* The number of descriptors to update */
658 uint32_t descriptor_count
;
660 /* Into mapped_ptr or dynamic_descriptors, in units of the respective array
664 /* In dwords. Not valid/used for dynamic descriptors */
667 uint32_t buffer_offset
;
669 /* Only valid for combined image samplers and samplers */
670 uint16_t has_sampler
;
676 /* For push descriptors */
677 const uint32_t *immutable_samplers
;
680 struct tu_descriptor_update_template
682 uint32_t entry_count
;
683 VkPipelineBindPoint bind_point
;
684 struct tu_descriptor_update_template_entry entry
[0];
691 VkBufferUsageFlags usage
;
692 VkBufferCreateFlags flags
;
695 VkDeviceSize bo_offset
;
698 static inline uint64_t
699 tu_buffer_iova(struct tu_buffer
*buffer
)
701 return buffer
->bo
->iova
+ buffer
->bo_offset
;
704 enum tu_dynamic_state_bits
706 TU_DYNAMIC_VIEWPORT
= 1 << 0,
707 TU_DYNAMIC_SCISSOR
= 1 << 1,
708 TU_DYNAMIC_LINE_WIDTH
= 1 << 2,
709 TU_DYNAMIC_DEPTH_BIAS
= 1 << 3,
710 TU_DYNAMIC_BLEND_CONSTANTS
= 1 << 4,
711 TU_DYNAMIC_DEPTH_BOUNDS
= 1 << 5,
712 TU_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6,
713 TU_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7,
714 TU_DYNAMIC_STENCIL_REFERENCE
= 1 << 8,
715 TU_DYNAMIC_DISCARD_RECTANGLE
= 1 << 9,
716 TU_DYNAMIC_ALL
= (1 << 10) - 1,
719 struct tu_vertex_binding
721 struct tu_buffer
*buffer
;
725 struct tu_viewport_state
728 VkViewport viewports
[MAX_VIEWPORTS
];
731 struct tu_scissor_state
734 VkRect2D scissors
[MAX_SCISSORS
];
737 struct tu_discard_rectangle_state
740 VkRect2D rectangles
[MAX_DISCARD_RECTANGLES
];
743 struct tu_dynamic_state
746 * Bitmask of (1 << VK_DYNAMIC_STATE_*).
747 * Defines the set of saved dynamic state.
751 struct tu_viewport_state viewport
;
753 struct tu_scissor_state scissor
;
764 float blend_constants
[4];
776 } stencil_compare_mask
;
782 } stencil_write_mask
;
790 struct tu_discard_rectangle_state discard_rectangle
;
793 extern const struct tu_dynamic_state default_dynamic_state
;
796 tu_get_debug_option_name(int id
);
799 tu_get_perftest_option_name(int id
);
801 struct tu_descriptor_state
803 struct tu_descriptor_set
*sets
[MAX_SETS
];
805 struct tu_push_descriptor_set push_set
;
807 uint64_t dynamic_buffers
[MAX_DYNAMIC_BUFFERS
];
818 struct tu_tiling_config
820 VkRect2D render_area
;
822 /* position and size of the first tile */
824 /* number of tiles */
825 VkExtent2D tile_count
;
827 /* size of the first VSC pipe */
829 /* number of VSC pipes */
830 VkExtent2D pipe_count
;
832 /* pipe register values */
833 uint32_t pipe_config
[MAX_VSC_PIPES
];
834 uint32_t pipe_sizes
[MAX_VSC_PIPES
];
836 /* Whether sysmem rendering must be used */
840 enum tu_cmd_dirty_bits
842 TU_CMD_DIRTY_PIPELINE
= 1 << 0,
843 TU_CMD_DIRTY_COMPUTE_PIPELINE
= 1 << 1,
844 TU_CMD_DIRTY_VERTEX_BUFFERS
= 1 << 2,
845 TU_CMD_DIRTY_DESCRIPTOR_SETS
= 1 << 3,
846 TU_CMD_DIRTY_PUSH_CONSTANTS
= 1 << 4,
847 TU_CMD_DIRTY_STREAMOUT_BUFFERS
= 1 << 5,
849 TU_CMD_DIRTY_DYNAMIC_LINE_WIDTH
= 1 << 16,
850 TU_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 17,
851 TU_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 18,
852 TU_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE
= 1 << 19,
853 TU_CMD_DIRTY_DYNAMIC_VIEWPORT
= 1 << 20,
854 TU_CMD_DIRTY_DYNAMIC_SCISSOR
= 1 << 21,
857 struct tu_streamout_state
{
858 uint16_t stride
[IR3_MAX_SO_BUFFERS
];
859 uint32_t ncomp
[IR3_MAX_SO_BUFFERS
];
860 uint32_t prog
[IR3_MAX_SO_OUTPUTS
* 2];
862 uint32_t vpc_so_buf_cntl
;
869 struct tu_pipeline
*pipeline
;
870 struct tu_pipeline
*compute_pipeline
;
875 struct tu_buffer
*buffers
[MAX_VBS
];
876 VkDeviceSize offsets
[MAX_VBS
];
879 struct tu_dynamic_state dynamic
;
881 /* Stream output buffers */
884 struct tu_buffer
*buffers
[IR3_MAX_SO_BUFFERS
];
885 VkDeviceSize offsets
[IR3_MAX_SO_BUFFERS
];
886 VkDeviceSize sizes
[IR3_MAX_SO_BUFFERS
];
889 uint8_t streamout_reset
;
890 uint8_t streamout_enabled
;
893 struct tu_buffer
*index_buffer
;
894 uint64_t index_offset
;
896 uint32_t max_index_count
;
899 const struct tu_render_pass
*pass
;
900 const struct tu_subpass
*subpass
;
901 const struct tu_framebuffer
*framebuffer
;
903 struct tu_tiling_config tiling_config
;
905 struct tu_cs_entry tile_store_ib
;
910 VkAllocationCallbacks alloc
;
911 struct list_head cmd_buffers
;
912 struct list_head free_cmd_buffers
;
913 uint32_t queue_family_index
;
916 struct tu_cmd_buffer_upload
921 struct list_head list
;
924 enum tu_cmd_buffer_status
926 TU_CMD_BUFFER_STATUS_INVALID
,
927 TU_CMD_BUFFER_STATUS_INITIAL
,
928 TU_CMD_BUFFER_STATUS_RECORDING
,
929 TU_CMD_BUFFER_STATUS_EXECUTABLE
,
930 TU_CMD_BUFFER_STATUS_PENDING
,
937 struct drm_msm_gem_submit_bo
*bo_infos
;
940 #define TU_BO_LIST_FAILED (~0)
943 tu_bo_list_init(struct tu_bo_list
*list
);
945 tu_bo_list_destroy(struct tu_bo_list
*list
);
947 tu_bo_list_reset(struct tu_bo_list
*list
);
949 tu_bo_list_add(struct tu_bo_list
*list
,
950 const struct tu_bo
*bo
,
953 tu_bo_list_merge(struct tu_bo_list
*list
, const struct tu_bo_list
*other
);
955 /* This struct defines the layout of the scratch_bo */
958 uint32_t seqno
; /* seqno for async CP_EVENT_WRITE, etc */
960 volatile uint32_t vsc_overflow
;
962 /* flag set from cmdstream when VSC overflow detected: */
963 uint32_t vsc_scratch
;
968 /* scratch space for VPC_SO[i].FLUSH_BASE_LO/HI, start on 32 byte boundary. */
975 #define ctrl_offset(member) offsetof(struct tu6_control, member)
979 VK_LOADER_DATA _loader_data
;
981 struct tu_device
*device
;
983 struct tu_cmd_pool
*pool
;
984 struct list_head pool_link
;
986 VkCommandBufferUsageFlags usage_flags
;
987 VkCommandBufferLevel level
;
988 enum tu_cmd_buffer_status status
;
990 struct tu_cmd_state state
;
991 struct tu_vertex_binding vertex_bindings
[MAX_VBS
];
992 uint32_t queue_family_index
;
994 uint32_t push_constants
[MAX_PUSH_CONSTANTS_SIZE
/ 4];
995 VkShaderStageFlags push_constant_stages
;
996 struct tu_descriptor_set meta_push_descriptors
;
998 struct tu_descriptor_state descriptors
[VK_PIPELINE_BIND_POINT_RANGE_SIZE
];
1000 struct tu_cmd_buffer_upload upload
;
1002 VkResult record_result
;
1004 struct tu_bo_list bo_list
;
1006 struct tu_cs draw_cs
;
1007 struct tu_cs draw_epilogue_cs
;
1008 struct tu_cs sub_cs
;
1010 struct tu_bo scratch_bo
;
1011 uint32_t scratch_seqno
;
1013 struct tu_bo vsc_data
;
1014 struct tu_bo vsc_data2
;
1015 uint32_t vsc_data_pitch
;
1016 uint32_t vsc_data2_pitch
;
1022 /* Temporary struct for tracking a register state to be written, used by
1023 * a6xx-pack.h and tu_cs_emit_regs()
1025 struct tu_reg_value
{
1036 tu6_emit_event_write(struct tu_cmd_buffer
*cmd
,
1038 enum vgt_event_type event
,
1042 tu_get_memory_fd(struct tu_device
*device
,
1043 struct tu_device_memory
*memory
,
1046 static inline struct tu_descriptor_state
*
1047 tu_get_descriptors_state(struct tu_cmd_buffer
*cmd_buffer
,
1048 VkPipelineBindPoint bind_point
)
1050 return &cmd_buffer
->descriptors
[bind_point
];
1054 * Takes x,y,z as exact numbers of invocations, instead of blocks.
1056 * Limitations: Can't call normal dispatch functions without binding or
1058 * the compute pipeline.
1061 tu_unaligned_dispatch(struct tu_cmd_buffer
*cmd_buffer
,
1071 struct tu_shader_module
;
1073 #define TU_HASH_SHADER_IS_GEOM_COPY_SHADER (1 << 0)
1074 #define TU_HASH_SHADER_SISCHED (1 << 1)
1075 #define TU_HASH_SHADER_UNSAFE_MATH (1 << 2)
1077 tu_hash_shaders(unsigned char *hash
,
1078 const VkPipelineShaderStageCreateInfo
**stages
,
1079 const struct tu_pipeline_layout
*layout
,
1080 const struct tu_pipeline_key
*key
,
1083 static inline gl_shader_stage
1084 vk_to_mesa_shader_stage(VkShaderStageFlagBits vk_stage
)
1086 assert(__builtin_popcount(vk_stage
) == 1);
1087 return ffs(vk_stage
) - 1;
1090 static inline VkShaderStageFlagBits
1091 mesa_to_vk_shader_stage(gl_shader_stage mesa_stage
)
1093 return (1 << mesa_stage
);
1096 #define TU_STAGE_MASK ((1 << MESA_SHADER_STAGES) - 1)
1098 #define tu_foreach_stage(stage, stage_bits) \
1099 for (gl_shader_stage stage, \
1100 __tmp = (gl_shader_stage)((stage_bits) &TU_STAGE_MASK); \
1101 stage = __builtin_ffs(__tmp) - 1, __tmp; __tmp &= ~(1 << (stage)))
1103 struct tu_shader_module
1105 unsigned char sha1
[20];
1108 const uint32_t *code
[0];
1111 struct tu_shader_compile_options
1113 struct ir3_shader_key key
;
1116 bool include_binning_pass
;
1119 struct tu_descriptor_map
1121 /* TODO: avoid fixed size array/justify the size */
1122 unsigned num
; /* number of array entries */
1123 unsigned num_desc
; /* Number of descriptors (sum of array_size[]) */
1127 int array_size
[128];
1132 struct ir3_shader ir3_shader
;
1134 struct tu_descriptor_map texture_map
;
1135 struct tu_descriptor_map sampler_map
;
1136 struct tu_descriptor_map ubo_map
;
1137 struct tu_descriptor_map ssbo_map
;
1138 struct tu_descriptor_map image_map
;
1140 /* This may be true for vertex shaders. When true, variants[1] is the
1141 * binning variant and binning_binary is non-NULL.
1143 bool has_binning_pass
;
1146 void *binning_binary
;
1148 struct ir3_shader_variant variants
[0];
1152 tu_shader_create(struct tu_device
*dev
,
1153 gl_shader_stage stage
,
1154 const VkPipelineShaderStageCreateInfo
*stage_info
,
1155 struct tu_pipeline_layout
*layout
,
1156 const VkAllocationCallbacks
*alloc
);
1159 tu_shader_destroy(struct tu_device
*dev
,
1160 struct tu_shader
*shader
,
1161 const VkAllocationCallbacks
*alloc
);
1164 tu_shader_compile_options_init(
1165 struct tu_shader_compile_options
*options
,
1166 const VkGraphicsPipelineCreateInfo
*pipeline_info
);
1169 tu_shader_compile(struct tu_device
*dev
,
1170 struct tu_shader
*shader
,
1171 const struct tu_shader
*next_stage
,
1172 const struct tu_shader_compile_options
*options
,
1173 const VkAllocationCallbacks
*alloc
);
1175 struct tu_program_descriptor_linkage
1177 struct ir3_ubo_analysis_state ubo_state
;
1178 struct ir3_const_state const_state
;
1182 struct tu_descriptor_map texture_map
;
1183 struct tu_descriptor_map sampler_map
;
1184 struct tu_descriptor_map ubo_map
;
1185 struct tu_descriptor_map ssbo_map
;
1186 struct tu_descriptor_map image_map
;
1193 struct tu_dynamic_state dynamic_state
;
1195 struct tu_pipeline_layout
*layout
;
1197 bool need_indirect_descriptor_sets
;
1198 VkShaderStageFlags active_stages
;
1200 struct tu_streamout_state streamout
;
1204 struct tu_bo binary_bo
;
1205 struct tu_cs_entry state_ib
;
1206 struct tu_cs_entry binning_state_ib
;
1208 struct tu_program_descriptor_linkage link
[MESA_SHADER_STAGES
];
1213 uint8_t bindings
[MAX_VERTEX_ATTRIBS
];
1216 uint8_t binning_bindings
[MAX_VERTEX_ATTRIBS
];
1217 uint32_t binning_count
;
1219 struct tu_cs_entry state_ib
;
1220 struct tu_cs_entry binning_state_ib
;
1225 enum pc_di_primtype primtype
;
1226 bool primitive_restart
;
1231 struct tu_cs_entry state_ib
;
1236 uint32_t gras_su_cntl
;
1237 struct tu_cs_entry state_ib
;
1242 struct tu_cs_entry state_ib
;
1247 struct tu_cs_entry state_ib
;
1252 uint32_t local_size
[3];
1257 tu6_emit_viewport(struct tu_cs
*cs
, const VkViewport
*viewport
);
1260 tu6_emit_scissor(struct tu_cs
*cs
, const VkRect2D
*scissor
);
1263 tu6_emit_gras_su_cntl(struct tu_cs
*cs
,
1264 uint32_t gras_su_cntl
,
1268 tu6_emit_depth_bias(struct tu_cs
*cs
,
1269 float constant_factor
,
1271 float slope_factor
);
1274 tu6_emit_stencil_compare_mask(struct tu_cs
*cs
,
1279 tu6_emit_stencil_write_mask(struct tu_cs
*cs
, uint32_t front
, uint32_t back
);
1282 tu6_emit_stencil_reference(struct tu_cs
*cs
, uint32_t front
, uint32_t back
);
1285 tu6_emit_blend_constants(struct tu_cs
*cs
, const float constants
[4]);
1287 struct tu_userdata_info
*
1288 tu_lookup_user_sgpr(struct tu_pipeline
*pipeline
,
1289 gl_shader_stage stage
,
1292 struct tu_shader_variant
*
1293 tu_get_shader(struct tu_pipeline
*pipeline
, gl_shader_stage stage
);
1295 struct tu_graphics_pipeline_create_info
1298 bool db_depth_clear
;
1299 bool db_stencil_clear
;
1300 bool db_depth_disable_expclear
;
1301 bool db_stencil_disable_expclear
;
1302 bool db_flush_depth_inplace
;
1303 bool db_flush_stencil_inplace
;
1304 bool db_resummarize
;
1305 uint32_t custom_blend_mode
;
1308 enum tu_supported_formats
{
1314 struct tu_native_format
1316 enum a6xx_format fmt
: 8;
1317 enum a3xx_color_swap swap
: 8;
1318 enum a6xx_tile_mode tile_mode
: 8;
1319 enum tu_supported_formats supported
: 8;
1322 struct tu_native_format
tu6_format_vtx(VkFormat format
);
1323 struct tu_native_format
tu6_format_color(VkFormat format
, enum a6xx_tile_mode tile_mode
);
1324 struct tu_native_format
tu6_format_texture(VkFormat format
, enum a6xx_tile_mode tile_mode
);
1326 static inline enum a6xx_format
1327 tu6_base_format(VkFormat format
)
1329 /* note: tu6_format_color doesn't care about tiling for .fmt field */
1330 return tu6_format_color(format
, TILE6_LINEAR
).fmt
;
1334 tu_pack_clear_value(const VkClearValue
*val
,
1339 tu_2d_clear_color(const VkClearColorValue
*val
, VkFormat format
, uint32_t buf
[4]);
1342 tu_2d_clear_zs(const VkClearDepthStencilValue
*val
, VkFormat format
, uint32_t buf
[4]);
1344 enum a6xx_2d_ifmt
tu6_fmt_to_ifmt(enum a6xx_format fmt
);
1345 enum a6xx_depth_format
tu6_pipe2depth(VkFormat format
);
1350 /* The original VkFormat provided by the client. This may not match any
1351 * of the actual surface formats.
1354 VkImageAspectFlags aspects
;
1355 VkImageUsageFlags usage
; /**< Superset of VkImageCreateInfo::usage. */
1356 VkImageTiling tiling
; /** VkImageCreateInfo::tiling */
1357 VkImageCreateFlags flags
; /** VkImageCreateInfo::flags */
1359 uint32_t level_count
;
1360 uint32_t layer_count
;
1361 VkSampleCountFlagBits samples
;
1363 struct fdl_layout layout
;
1365 unsigned queue_family_mask
;
1369 /* For VK_ANDROID_native_buffer, the WSI image owns the memory, */
1370 VkDeviceMemory owned_memory
;
1372 /* Set when bound */
1374 VkDeviceSize bo_offset
;
1378 tu_image_queue_family_mask(const struct tu_image
*image
,
1380 uint32_t queue_family
);
1382 static inline uint32_t
1383 tu_get_layerCount(const struct tu_image
*image
,
1384 const VkImageSubresourceRange
*range
)
1386 return range
->layerCount
== VK_REMAINING_ARRAY_LAYERS
1387 ? image
->layer_count
- range
->baseArrayLayer
1388 : range
->layerCount
;
1391 static inline uint32_t
1392 tu_get_levelCount(const struct tu_image
*image
,
1393 const VkImageSubresourceRange
*range
)
1395 return range
->levelCount
== VK_REMAINING_MIP_LEVELS
1396 ? image
->level_count
- range
->baseMipLevel
1397 : range
->levelCount
;
1400 static inline VkDeviceSize
1401 tu_layer_size(struct tu_image
*image
, int level
)
1403 return fdl_layer_stride(&image
->layout
, level
);
1406 static inline uint32_t
1407 tu_image_stride(struct tu_image
*image
, int level
)
1409 return image
->layout
.slices
[level
].pitch
* image
->layout
.cpp
;
1412 static inline uint64_t
1413 tu_image_base(struct tu_image
*image
, int level
, int layer
)
1415 return image
->bo
->iova
+ image
->bo_offset
+
1416 fdl_surface_offset(&image
->layout
, level
, layer
);
1419 #define tu_image_base_ref(image, level, layer) \
1421 .bo_offset = (image->bo_offset + fdl_surface_offset(&image->layout, \
1424 #define tu_image_view_base_ref(iview) \
1425 tu_image_base_ref(iview->image, iview->base_mip, iview->base_layer)
1427 static inline VkDeviceSize
1428 tu_image_ubwc_size(struct tu_image
*image
, int level
)
1430 return image
->layout
.ubwc_layer_size
;
1433 static inline uint32_t
1434 tu_image_ubwc_pitch(struct tu_image
*image
, int level
)
1436 return image
->layout
.ubwc_slices
[level
].pitch
;
1439 static inline uint64_t
1440 tu_image_ubwc_surface_offset(struct tu_image
*image
, int level
, int layer
)
1442 return image
->layout
.ubwc_slices
[level
].offset
+
1443 layer
* tu_image_ubwc_size(image
, level
);
1446 static inline uint64_t
1447 tu_image_ubwc_base(struct tu_image
*image
, int level
, int layer
)
1449 return image
->bo
->iova
+ image
->bo_offset
+
1450 tu_image_ubwc_surface_offset(image
, level
, layer
);
1453 #define tu_image_ubwc_base_ref(image, level, layer) \
1455 .bo_offset = (image->bo_offset + tu_image_ubwc_surface_offset(image, \
1458 #define tu_image_view_ubwc_base_ref(iview) \
1459 tu_image_ubwc_base_ref(iview->image, iview->base_mip, iview->base_layer)
1462 tu6_get_image_tile_mode(struct tu_image
*image
, int level
);
1463 enum a3xx_msaa_samples
1464 tu_msaa_samples(uint32_t samples
);
1466 static inline struct tu_native_format
1467 tu6_format_image(struct tu_image
*image
, VkFormat format
, uint32_t level
)
1469 struct tu_native_format fmt
=
1470 tu6_format_color(format
, image
->layout
.tile_mode
);
1471 fmt
.tile_mode
= tu6_get_image_tile_mode(image
, level
);
1475 static inline struct tu_native_format
1476 tu6_format_image_src(struct tu_image
*image
, VkFormat format
, uint32_t level
)
1478 struct tu_native_format fmt
=
1479 tu6_format_texture(format
, image
->layout
.tile_mode
);
1480 fmt
.tile_mode
= tu6_get_image_tile_mode(image
, level
);
1484 struct tu_image_view
1486 struct tu_image
*image
; /**< VkImageViewCreateInfo::image */
1488 VkImageViewType type
;
1489 VkImageAspectFlags aspect_mask
;
1491 uint32_t base_layer
;
1492 uint32_t layer_count
;
1494 uint32_t level_count
;
1495 VkExtent3D extent
; /**< Extent of VkImageViewCreateInfo::baseMipLevel. */
1497 uint32_t descriptor
[A6XX_TEX_CONST_DWORDS
];
1499 /* Descriptor for use as a storage image as opposed to a sampled image.
1500 * This has a few differences for cube maps (e.g. type).
1502 uint32_t storage_descriptor
[A6XX_TEX_CONST_DWORDS
];
1506 uint32_t descriptor
[A6XX_TEX_SAMP_DWORDS
];
1510 tu_image_create(VkDevice _device
,
1511 const VkImageCreateInfo
*pCreateInfo
,
1512 const VkAllocationCallbacks
*alloc
,
1517 tu_image_from_gralloc(VkDevice device_h
,
1518 const VkImageCreateInfo
*base_info
,
1519 const VkNativeBufferANDROID
*gralloc_info
,
1520 const VkAllocationCallbacks
*alloc
,
1521 VkImage
*out_image_h
);
1524 tu_image_view_init(struct tu_image_view
*view
,
1525 struct tu_device
*device
,
1526 const VkImageViewCreateInfo
*pCreateInfo
);
1528 struct tu_buffer_view
1530 uint32_t descriptor
[A6XX_TEX_CONST_DWORDS
];
1532 struct tu_buffer
*buffer
;
1535 tu_buffer_view_init(struct tu_buffer_view
*view
,
1536 struct tu_device
*device
,
1537 const VkBufferViewCreateInfo
*pCreateInfo
);
1539 static inline struct VkExtent3D
1540 tu_sanitize_image_extent(const VkImageType imageType
,
1541 const struct VkExtent3D imageExtent
)
1543 switch (imageType
) {
1544 case VK_IMAGE_TYPE_1D
:
1545 return (VkExtent3D
) { imageExtent
.width
, 1, 1 };
1546 case VK_IMAGE_TYPE_2D
:
1547 return (VkExtent3D
) { imageExtent
.width
, imageExtent
.height
, 1 };
1548 case VK_IMAGE_TYPE_3D
:
1551 unreachable("invalid image type");
1555 static inline struct VkOffset3D
1556 tu_sanitize_image_offset(const VkImageType imageType
,
1557 const struct VkOffset3D imageOffset
)
1559 switch (imageType
) {
1560 case VK_IMAGE_TYPE_1D
:
1561 return (VkOffset3D
) { imageOffset
.x
, 0, 0 };
1562 case VK_IMAGE_TYPE_2D
:
1563 return (VkOffset3D
) { imageOffset
.x
, imageOffset
.y
, 0 };
1564 case VK_IMAGE_TYPE_3D
:
1567 unreachable("invalid image type");
1571 struct tu_attachment_info
1573 struct tu_image_view
*attachment
;
1576 struct tu_framebuffer
1582 uint32_t attachment_count
;
1583 struct tu_attachment_info attachments
[0];
1586 struct tu_subpass_attachment
1588 uint32_t attachment
;
1593 uint32_t input_count
;
1594 uint32_t color_count
;
1595 struct tu_subpass_attachment
*input_attachments
;
1596 struct tu_subpass_attachment
*color_attachments
;
1597 struct tu_subpass_attachment
*resolve_attachments
;
1598 struct tu_subpass_attachment depth_stencil_attachment
;
1600 VkSampleCountFlagBits samples
;
1603 struct tu_render_pass_attachment
1608 VkAttachmentLoadOp load_op
;
1609 VkAttachmentLoadOp stencil_load_op
;
1610 VkAttachmentStoreOp store_op
;
1611 VkAttachmentStoreOp stencil_store_op
;
1612 int32_t gmem_offset
;
1615 struct tu_render_pass
1617 uint32_t attachment_count
;
1618 uint32_t subpass_count
;
1619 uint32_t gmem_pixels
;
1620 struct tu_subpass_attachment
*subpass_attachments
;
1621 struct tu_render_pass_attachment
*attachments
;
1622 struct tu_subpass subpasses
[0];
1626 tu_device_init_meta(struct tu_device
*device
);
1628 tu_device_finish_meta(struct tu_device
*device
);
1630 struct tu_query_pool
1635 uint32_t pipeline_statistics
;
1642 uint32_t temp_syncobj
;
1646 tu_set_descriptor_set(struct tu_cmd_buffer
*cmd_buffer
,
1647 VkPipelineBindPoint bind_point
,
1648 struct tu_descriptor_set
*set
,
1652 tu_update_descriptor_sets(struct tu_device
*device
,
1653 struct tu_cmd_buffer
*cmd_buffer
,
1654 VkDescriptorSet overrideSet
,
1655 uint32_t descriptorWriteCount
,
1656 const VkWriteDescriptorSet
*pDescriptorWrites
,
1657 uint32_t descriptorCopyCount
,
1658 const VkCopyDescriptorSet
*pDescriptorCopies
);
1661 tu_update_descriptor_set_with_template(
1662 struct tu_device
*device
,
1663 struct tu_cmd_buffer
*cmd_buffer
,
1664 struct tu_descriptor_set
*set
,
1665 VkDescriptorUpdateTemplate descriptorUpdateTemplate
,
1669 tu_meta_push_descriptor_set(struct tu_cmd_buffer
*cmd_buffer
,
1670 VkPipelineBindPoint pipelineBindPoint
,
1671 VkPipelineLayout _layout
,
1673 uint32_t descriptorWriteCount
,
1674 const VkWriteDescriptorSet
*pDescriptorWrites
);
1677 tu_drm_get_gpu_id(const struct tu_physical_device
*dev
, uint32_t *id
);
1680 tu_drm_get_gmem_size(const struct tu_physical_device
*dev
, uint32_t *size
);
1683 tu_drm_get_gmem_base(const struct tu_physical_device
*dev
, uint64_t *base
);
1686 tu_drm_submitqueue_new(const struct tu_device
*dev
,
1688 uint32_t *queue_id
);
1691 tu_drm_submitqueue_close(const struct tu_device
*dev
, uint32_t queue_id
);
1694 tu_gem_new(const struct tu_device
*dev
, uint64_t size
, uint32_t flags
);
1696 tu_gem_import_dmabuf(const struct tu_device
*dev
,
1700 tu_gem_export_dmabuf(const struct tu_device
*dev
, uint32_t gem_handle
);
1702 tu_gem_close(const struct tu_device
*dev
, uint32_t gem_handle
);
1704 tu_gem_info_offset(const struct tu_device
*dev
, uint32_t gem_handle
);
1706 tu_gem_info_iova(const struct tu_device
*dev
, uint32_t gem_handle
);
1710 tu_clear_sysmem_attachment(struct tu_cmd_buffer
*cmd
,
1712 uint32_t attachment
,
1713 const VkClearValue
*value
,
1714 const VkClearRect
*rect
);
1717 tu_clear_gmem_attachment(struct tu_cmd_buffer
*cmd
,
1719 uint32_t attachment
,
1720 uint8_t component_mask
,
1721 const VkClearValue
*value
);
1723 #define TU_DEFINE_HANDLE_CASTS(__tu_type, __VkType) \
1725 static inline struct __tu_type *__tu_type##_from_handle(__VkType _handle) \
1727 return (struct __tu_type *) _handle; \
1730 static inline __VkType __tu_type##_to_handle(struct __tu_type *_obj) \
1732 return (__VkType) _obj; \
1735 #define TU_DEFINE_NONDISP_HANDLE_CASTS(__tu_type, __VkType) \
1737 static inline struct __tu_type *__tu_type##_from_handle(__VkType _handle) \
1739 return (struct __tu_type *) (uintptr_t) _handle; \
1742 static inline __VkType __tu_type##_to_handle(struct __tu_type *_obj) \
1744 return (__VkType)(uintptr_t) _obj; \
1747 #define TU_FROM_HANDLE(__tu_type, __name, __handle) \
1748 struct __tu_type *__name = __tu_type##_from_handle(__handle)
1750 TU_DEFINE_HANDLE_CASTS(tu_cmd_buffer
, VkCommandBuffer
)
1751 TU_DEFINE_HANDLE_CASTS(tu_device
, VkDevice
)
1752 TU_DEFINE_HANDLE_CASTS(tu_instance
, VkInstance
)
1753 TU_DEFINE_HANDLE_CASTS(tu_physical_device
, VkPhysicalDevice
)
1754 TU_DEFINE_HANDLE_CASTS(tu_queue
, VkQueue
)
1756 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_cmd_pool
, VkCommandPool
)
1757 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_buffer
, VkBuffer
)
1758 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_buffer_view
, VkBufferView
)
1759 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_pool
, VkDescriptorPool
)
1760 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_set
, VkDescriptorSet
)
1761 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_set_layout
,
1762 VkDescriptorSetLayout
)
1763 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_update_template
,
1764 VkDescriptorUpdateTemplate
)
1765 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_device_memory
, VkDeviceMemory
)
1766 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_fence
, VkFence
)
1767 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_event
, VkEvent
)
1768 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_framebuffer
, VkFramebuffer
)
1769 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_image
, VkImage
)
1770 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_image_view
, VkImageView
);
1771 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_pipeline_cache
, VkPipelineCache
)
1772 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_pipeline
, VkPipeline
)
1773 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_pipeline_layout
, VkPipelineLayout
)
1774 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_query_pool
, VkQueryPool
)
1775 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_render_pass
, VkRenderPass
)
1776 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_sampler
, VkSampler
)
1777 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_shader_module
, VkShaderModule
)
1778 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_semaphore
, VkSemaphore
)
1780 #endif /* TU_PRIVATE_H */