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"
50 #include "util/u_atomic.h"
52 #include "vk_object.h"
53 #include "vk_debug_report.h"
54 #include "wsi_common.h"
56 #include "drm-uapi/msm_drm.h"
57 #include "ir3/ir3_compiler.h"
58 #include "ir3/ir3_shader.h"
60 #include "adreno_common.xml.h"
61 #include "adreno_pm4.xml.h"
63 #include "fdl/freedreno_layout.h"
65 #include "tu_descriptor_set.h"
66 #include "tu_extensions.h"
69 /* Pre-declarations needed for WSI entrypoints */
72 typedef struct xcb_connection_t xcb_connection_t
;
73 typedef uint32_t xcb_visualid_t
;
74 typedef uint32_t xcb_window_t
;
76 #include <vulkan/vk_android_native_buffer.h>
77 #include <vulkan/vk_icd.h>
78 #include <vulkan/vulkan.h>
79 #include <vulkan/vulkan_intel.h>
81 #include "tu_entrypoints.h"
83 #include "vk_format.h"
86 #define MAX_VERTEX_ATTRIBS 32
88 #define MAX_VSC_PIPES 32
89 #define MAX_VIEWPORTS 1
90 #define MAX_SCISSORS 16
91 #define MAX_DISCARD_RECTANGLES 4
92 #define MAX_PUSH_CONSTANTS_SIZE 128
93 #define MAX_PUSH_DESCRIPTORS 32
94 #define MAX_DYNAMIC_UNIFORM_BUFFERS 16
95 #define MAX_DYNAMIC_STORAGE_BUFFERS 8
96 #define MAX_DYNAMIC_BUFFERS \
97 (MAX_DYNAMIC_UNIFORM_BUFFERS + MAX_DYNAMIC_STORAGE_BUFFERS)
98 #define TU_MAX_DRM_DEVICES 8
100 #define MAX_BIND_POINTS 2 /* compute + graphics */
101 /* The Qualcomm driver exposes 0x20000058 */
102 #define MAX_STORAGE_BUFFER_RANGE 0x20000000
103 /* We use ldc for uniform buffer loads, just like the Qualcomm driver, so
104 * expose the same maximum range.
105 * TODO: The SIZE bitfield is 15 bits, and in 4-dword units, so the actual
106 * range might be higher.
108 #define MAX_UNIFORM_BUFFER_RANGE 0x10000
110 #define A6XX_TEX_CONST_DWORDS 16
111 #define A6XX_TEX_SAMP_DWORDS 4
113 #define tu_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
115 static inline uint32_t
116 tu_minify(uint32_t n
, uint32_t levels
)
118 if (unlikely(n
== 0))
121 return MAX2(n
>> levels
, 1);
124 #define for_each_bit(b, dword) \
125 for (uint32_t __dword = (dword); \
126 (b) = __builtin_ffs(__dword) - 1, __dword; __dword &= ~(1 << (b)))
128 #define typed_memcpy(dest, src, count) \
130 STATIC_ASSERT(sizeof(*src) == sizeof(*dest)); \
131 memcpy((dest), (src), (count) * sizeof(*(src))); \
134 #define COND(bool, val) ((bool) ? (val) : 0)
135 #define BIT(bit) (1u << (bit))
137 /* Whenever we generate an error, pass it through this function. Useful for
138 * debugging, where we can break on it. Only call at error site, not when
139 * propagating errors. Might be useful to plug in a stack trace here.
145 __vk_errorf(struct tu_instance
*instance
,
152 #define vk_error(instance, error) \
153 __vk_errorf(instance, error, __FILE__, __LINE__, NULL);
154 #define vk_errorf(instance, error, format, ...) \
155 __vk_errorf(instance, error, __FILE__, __LINE__, format, ##__VA_ARGS__);
158 __tu_finishme(const char *file
, int line
, const char *format
, ...)
161 tu_loge(const char *format
, ...) tu_printflike(1, 2);
163 tu_logi(const char *format
, ...) tu_printflike(1, 2);
166 * Print a FINISHME message, including its source location.
168 #define tu_finishme(format, ...) \
170 static bool reported = false; \
172 __tu_finishme(__FILE__, __LINE__, format, ##__VA_ARGS__); \
179 tu_finishme("stub %s", __func__); \
183 tu_lookup_entrypoint_unchecked(const char *name
);
185 tu_lookup_entrypoint_checked(
187 uint32_t core_version
,
188 const struct tu_instance_extension_table
*instance
,
189 const struct tu_device_extension_table
*device
);
191 struct tu_physical_device
193 struct vk_object_base base
;
195 struct tu_instance
*instance
;
198 char name
[VK_MAX_PHYSICAL_DEVICE_NAME_SIZE
];
199 uint8_t driver_uuid
[VK_UUID_SIZE
];
200 uint8_t device_uuid
[VK_UUID_SIZE
];
201 uint8_t cache_uuid
[VK_UUID_SIZE
];
203 struct wsi_device wsi_device
;
211 uint32_t ccu_offset_gmem
;
212 uint32_t ccu_offset_bypass
;
213 /* alignment for size of tiles */
214 uint32_t tile_align_w
;
215 #define TILE_ALIGN_H 16
216 /* gmem store/load granularity */
217 #define GMEM_ALIGN_W 16
218 #define GMEM_ALIGN_H 4
221 uint32_t PC_UNKNOWN_9805
;
222 uint32_t SP_UNKNOWN_A0F8
;
225 int msm_major_version
;
226 int msm_minor_version
;
228 /* This is the drivers on-disk cache used as a fallback as opposed to
229 * the pipeline cache defined by apps.
231 struct disk_cache
*disk_cache
;
233 struct tu_device_extension_table supported_extensions
;
238 TU_DEBUG_STARTUP
= 1 << 0,
239 TU_DEBUG_NIR
= 1 << 1,
240 TU_DEBUG_IR3
= 1 << 2,
241 TU_DEBUG_NOBIN
= 1 << 3,
242 TU_DEBUG_SYSMEM
= 1 << 4,
243 TU_DEBUG_FORCEBIN
= 1 << 5,
244 TU_DEBUG_NOUBWC
= 1 << 6,
249 struct vk_object_base base
;
251 VkAllocationCallbacks alloc
;
253 uint32_t api_version
;
254 int physical_device_count
;
255 struct tu_physical_device physical_devices
[TU_MAX_DRM_DEVICES
];
257 enum tu_debug_flags debug_flags
;
259 struct vk_debug_report_instance debug_report_callbacks
;
261 struct tu_instance_extension_table enabled_extensions
;
265 tu_wsi_init(struct tu_physical_device
*physical_device
);
267 tu_wsi_finish(struct tu_physical_device
*physical_device
);
270 tu_instance_extension_supported(const char *name
);
272 tu_physical_device_api_version(struct tu_physical_device
*dev
);
274 tu_physical_device_extension_supported(struct tu_physical_device
*dev
,
279 struct tu_pipeline_cache
281 struct vk_object_base base
;
283 struct tu_device
*device
;
284 pthread_mutex_t mutex
;
288 uint32_t kernel_count
;
289 struct cache_entry
**hash_table
;
292 VkAllocationCallbacks alloc
;
295 struct tu_pipeline_key
301 #define TU_QUEUE_GENERAL 0
303 #define TU_MAX_QUEUE_FAMILIES 1
307 struct vk_object_base base
;
308 struct wsi_fence
*fence_wsi
;
314 tu_fence_init(struct tu_fence
*fence
, bool signaled
);
316 tu_fence_finish(struct tu_fence
*fence
);
318 tu_fence_update_fd(struct tu_fence
*fence
, int fd
);
320 tu_fence_copy(struct tu_fence
*fence
, const struct tu_fence
*src
);
322 tu_fence_signal(struct tu_fence
*fence
);
324 tu_fence_wait_idle(struct tu_fence
*fence
);
328 struct vk_object_base base
;
330 struct tu_device
*device
;
331 uint32_t queue_family_index
;
333 VkDeviceQueueCreateFlags flags
;
335 uint32_t msm_queue_id
;
336 struct tu_fence submit_fence
;
351 GLOBAL_SH_FS_CLEAR_MAX
= GLOBAL_SH_FS_CLEAR0
+ MAX_RTS
,
355 /* This struct defines the layout of the global_bo */
358 /* 6 bcolor_entry entries, one for each VK_BORDER_COLOR */
359 uint8_t border_color
[128 * 6];
361 /* clear/blit shaders, all <= 16 instrs (16 instr = 1 instrlen unit) */
362 instr_t shaders
[GLOBAL_SH_COUNT
][16];
364 uint32_t seqno_dummy
; /* dummy seqno for CP_EVENT_WRITE */
366 volatile uint32_t vsc_draw_overflow
;
368 volatile uint32_t vsc_prim_overflow
;
371 /* scratch space for VPC_SO[i].FLUSH_BASE_LO/HI, start on 32 byte boundary. */
377 #define gb_offset(member) offsetof(struct tu6_global, member)
378 #define global_iova(cmd, member) ((cmd)->device->global_bo.iova + gb_offset(member))
380 void tu_init_clear_blit_shaders(struct tu6_global
*global
);
382 /* extra space in vsc draw/prim streams */
388 struct tu_instance
*instance
;
390 struct tu_queue
*queues
[TU_MAX_QUEUE_FAMILIES
];
391 int queue_count
[TU_MAX_QUEUE_FAMILIES
];
393 struct tu_physical_device
*physical_device
;
396 struct ir3_compiler
*compiler
;
398 /* Backup in-memory cache to be used if the app doesn't provide one */
399 struct tu_pipeline_cache
*mem_cache
;
401 #define MIN_SCRATCH_BO_SIZE_LOG2 12 /* A page */
403 /* Currently the kernel driver uses a 32-bit GPU address space, but it
404 * should be impossible to go beyond 48 bits.
410 } scratch_bos
[48 - MIN_SCRATCH_BO_SIZE_LOG2
];
412 struct tu_bo global_bo
;
414 struct tu_device_extension_table enabled_extensions
;
416 uint32_t vsc_draw_strm_pitch
;
417 uint32_t vsc_prim_strm_pitch
;
421 VkResult
_tu_device_set_lost(struct tu_device
*device
,
422 const char *file
, int line
,
423 const char *msg
, ...) PRINTFLIKE(4, 5);
424 #define tu_device_set_lost(dev, ...) \
425 _tu_device_set_lost(dev, __FILE__, __LINE__, __VA_ARGS__)
428 tu_device_is_lost(struct tu_device
*device
)
430 return unlikely(p_atomic_read(&device
->_lost
));
434 tu_bo_init_new(struct tu_device
*dev
, struct tu_bo
*bo
, uint64_t size
);
436 tu_bo_init_dmabuf(struct tu_device
*dev
,
441 tu_bo_export_dmabuf(struct tu_device
*dev
, struct tu_bo
*bo
);
443 tu_bo_finish(struct tu_device
*dev
, struct tu_bo
*bo
);
445 tu_bo_map(struct tu_device
*dev
, struct tu_bo
*bo
);
447 /* Get a scratch bo for use inside a command buffer. This will always return
448 * the same bo given the same size or similar sizes, so only one scratch bo
449 * can be used at the same time. It's meant for short-lived things where we
450 * need to write to some piece of memory, read from it, and then immediately
454 tu_get_scratch_bo(struct tu_device
*dev
, uint64_t size
, struct tu_bo
**bo
);
459 const struct tu_bo
*bo
;
465 struct tu_cs_memory
{
470 struct tu_draw_state
{
475 enum tu_dynamic_state
477 /* re-use VK_DYNAMIC_STATE_ enums for non-extended dynamic states */
478 TU_DYNAMIC_STATE_SAMPLE_LOCATIONS
= VK_DYNAMIC_STATE_STENCIL_REFERENCE
+ 1,
479 TU_DYNAMIC_STATE_COUNT
,
482 enum tu_draw_state_group_id
484 TU_DRAW_STATE_PROGRAM
,
485 TU_DRAW_STATE_PROGRAM_BINNING
,
489 TU_DRAW_STATE_VI_BINNING
,
493 TU_DRAW_STATE_VS_CONST
,
494 TU_DRAW_STATE_HS_CONST
,
495 TU_DRAW_STATE_DS_CONST
,
496 TU_DRAW_STATE_GS_CONST
,
497 TU_DRAW_STATE_FS_CONST
,
498 TU_DRAW_STATE_DESC_SETS
,
499 TU_DRAW_STATE_DESC_SETS_LOAD
,
500 TU_DRAW_STATE_VS_PARAMS
,
501 TU_DRAW_STATE_INPUT_ATTACHMENTS_GMEM
,
502 TU_DRAW_STATE_INPUT_ATTACHMENTS_SYSMEM
,
504 /* dynamic state related draw states */
505 TU_DRAW_STATE_DYNAMIC
,
506 TU_DRAW_STATE_COUNT
= TU_DRAW_STATE_DYNAMIC
+ TU_DYNAMIC_STATE_COUNT
,
513 * A command stream in TU_CS_MODE_GROW mode grows automatically whenever it
514 * is full. tu_cs_begin must be called before command packet emission and
515 * tu_cs_end must be called after.
517 * This mode may create multiple entries internally. The entries must be
518 * submitted together.
523 * A command stream in TU_CS_MODE_EXTERNAL mode wraps an external,
524 * fixed-size buffer. tu_cs_begin and tu_cs_end are optional and have no
527 * This mode does not create any entry or any BO.
532 * A command stream in TU_CS_MODE_SUB_STREAM mode does not support direct
533 * command packet emission. tu_cs_begin_sub_stream must be called to get a
534 * sub-stream to emit comamnd packets to. When done with the sub-stream,
535 * tu_cs_end_sub_stream must be called.
537 * This mode does not create any entry internally.
539 TU_CS_MODE_SUB_STREAM
,
546 uint32_t *reserved_end
;
549 struct tu_device
*device
;
550 enum tu_cs_mode mode
;
551 uint32_t next_bo_size
;
553 struct tu_cs_entry
*entries
;
554 uint32_t entry_count
;
555 uint32_t entry_capacity
;
559 uint32_t bo_capacity
;
561 /* state for cond_exec_start/cond_exec_end */
563 uint32_t *cond_dwords
;
566 struct tu_device_memory
568 struct vk_object_base base
;
573 /* for dedicated allocations */
574 struct tu_image
*image
;
575 struct tu_buffer
*buffer
;
582 struct tu_descriptor_range
588 struct tu_descriptor_set
590 struct vk_object_base base
;
592 const struct tu_descriptor_set_layout
*layout
;
593 struct tu_descriptor_pool
*pool
;
597 uint32_t *mapped_ptr
;
599 uint32_t *dynamic_descriptors
;
601 struct tu_bo
*buffers
[0];
604 struct tu_push_descriptor_set
606 struct tu_descriptor_set set
;
610 struct tu_descriptor_pool_entry
614 struct tu_descriptor_set
*set
;
617 struct tu_descriptor_pool
619 struct vk_object_base base
;
622 uint64_t current_offset
;
625 uint8_t *host_memory_base
;
626 uint8_t *host_memory_ptr
;
627 uint8_t *host_memory_end
;
629 uint32_t entry_count
;
630 uint32_t max_entry_count
;
631 struct tu_descriptor_pool_entry entries
[0];
634 struct tu_descriptor_update_template_entry
636 VkDescriptorType descriptor_type
;
638 /* The number of descriptors to update */
639 uint32_t descriptor_count
;
641 /* Into mapped_ptr or dynamic_descriptors, in units of the respective array
645 /* In dwords. Not valid/used for dynamic descriptors */
648 uint32_t buffer_offset
;
650 /* Only valid for combined image samplers and samplers */
651 uint16_t has_sampler
;
657 /* For push descriptors */
658 const uint32_t *immutable_samplers
;
661 struct tu_descriptor_update_template
663 struct vk_object_base base
;
665 uint32_t entry_count
;
666 struct tu_descriptor_update_template_entry entry
[0];
671 struct vk_object_base base
;
675 VkBufferUsageFlags usage
;
676 VkBufferCreateFlags flags
;
679 VkDeviceSize bo_offset
;
682 static inline uint64_t
683 tu_buffer_iova(struct tu_buffer
*buffer
)
685 return buffer
->bo
->iova
+ buffer
->bo_offset
;
688 struct tu_vertex_binding
690 struct tu_buffer
*buffer
;
695 tu_get_debug_option_name(int id
);
698 tu_get_perftest_option_name(int id
);
700 struct tu_descriptor_state
702 struct tu_descriptor_set
*sets
[MAX_SETS
];
703 uint32_t dynamic_descriptors
[MAX_DYNAMIC_BUFFERS
* A6XX_TEX_CONST_DWORDS
];
706 enum tu_cmd_dirty_bits
708 TU_CMD_DIRTY_COMPUTE_PIPELINE
= 1 << 1,
709 TU_CMD_DIRTY_VERTEX_BUFFERS
= 1 << 2,
710 TU_CMD_DIRTY_DESC_SETS_LOAD
= 1 << 3,
711 TU_CMD_DIRTY_COMPUTE_DESCRIPTOR_SETS
= 1 << 4,
712 TU_CMD_DIRTY_SHADER_CONSTS
= 1 << 5,
713 /* all draw states were disabled and need to be re-enabled: */
714 TU_CMD_DIRTY_DRAW_STATE
= 1 << 7,
717 /* There are only three cache domains we have to care about: the CCU, or
718 * color cache unit, which is used for color and depth/stencil attachments
719 * and copy/blit destinations, and is split conceptually into color and depth,
720 * and the universal cache or UCHE which is used for pretty much everything
721 * else, except for the CP (uncached) and host. We need to flush whenever data
722 * crosses these boundaries.
725 enum tu_cmd_access_mask
{
726 TU_ACCESS_UCHE_READ
= 1 << 0,
727 TU_ACCESS_UCHE_WRITE
= 1 << 1,
728 TU_ACCESS_CCU_COLOR_READ
= 1 << 2,
729 TU_ACCESS_CCU_COLOR_WRITE
= 1 << 3,
730 TU_ACCESS_CCU_DEPTH_READ
= 1 << 4,
731 TU_ACCESS_CCU_DEPTH_WRITE
= 1 << 5,
733 /* Experiments have shown that while it's safe to avoid flushing the CCU
734 * after each blit/renderpass, it's not safe to assume that subsequent
735 * lookups with a different attachment state will hit unflushed cache
736 * entries. That is, the CCU needs to be flushed and possibly invalidated
737 * when accessing memory with a different attachment state. Writing to an
738 * attachment under the following conditions after clearing using the
739 * normal 2d engine path is known to have issues:
741 * - It isn't the 0'th layer.
742 * - There are more than one attachment, and this isn't the 0'th attachment
743 * (this seems to also depend on the cpp of the attachments).
745 * Our best guess is that the layer/MRT state is used when computing
746 * the location of a cache entry in CCU, to avoid conflicts. We assume that
747 * any access in a renderpass after or before an access by a transfer needs
748 * a flush/invalidate, and use the _INCOHERENT variants to represent access
751 TU_ACCESS_CCU_COLOR_INCOHERENT_READ
= 1 << 6,
752 TU_ACCESS_CCU_COLOR_INCOHERENT_WRITE
= 1 << 7,
753 TU_ACCESS_CCU_DEPTH_INCOHERENT_READ
= 1 << 8,
754 TU_ACCESS_CCU_DEPTH_INCOHERENT_WRITE
= 1 << 9,
756 TU_ACCESS_SYSMEM_READ
= 1 << 10,
757 TU_ACCESS_SYSMEM_WRITE
= 1 << 11,
759 /* Set if a WFI is required due to data being read by the CP or the 2D
762 TU_ACCESS_WFI_READ
= 1 << 12,
765 TU_ACCESS_UCHE_READ
|
766 TU_ACCESS_CCU_COLOR_READ
|
767 TU_ACCESS_CCU_DEPTH_READ
|
768 TU_ACCESS_CCU_COLOR_INCOHERENT_READ
|
769 TU_ACCESS_CCU_DEPTH_INCOHERENT_READ
|
770 TU_ACCESS_SYSMEM_READ
,
773 TU_ACCESS_UCHE_WRITE
|
774 TU_ACCESS_CCU_COLOR_WRITE
|
775 TU_ACCESS_CCU_COLOR_INCOHERENT_WRITE
|
776 TU_ACCESS_CCU_DEPTH_WRITE
|
777 TU_ACCESS_CCU_DEPTH_INCOHERENT_WRITE
|
778 TU_ACCESS_SYSMEM_WRITE
,
785 enum tu_cmd_flush_bits
{
786 TU_CMD_FLAG_CCU_FLUSH_DEPTH
= 1 << 0,
787 TU_CMD_FLAG_CCU_FLUSH_COLOR
= 1 << 1,
788 TU_CMD_FLAG_CCU_INVALIDATE_DEPTH
= 1 << 2,
789 TU_CMD_FLAG_CCU_INVALIDATE_COLOR
= 1 << 3,
790 TU_CMD_FLAG_CACHE_FLUSH
= 1 << 4,
791 TU_CMD_FLAG_CACHE_INVALIDATE
= 1 << 5,
793 TU_CMD_FLAG_ALL_FLUSH
=
794 TU_CMD_FLAG_CCU_FLUSH_DEPTH
|
795 TU_CMD_FLAG_CCU_FLUSH_COLOR
|
796 TU_CMD_FLAG_CACHE_FLUSH
,
798 TU_CMD_FLAG_ALL_INVALIDATE
=
799 TU_CMD_FLAG_CCU_INVALIDATE_DEPTH
|
800 TU_CMD_FLAG_CCU_INVALIDATE_COLOR
|
801 TU_CMD_FLAG_CACHE_INVALIDATE
,
803 TU_CMD_FLAG_WFI
= 1 << 6,
806 /* Changing the CCU from sysmem mode to gmem mode or vice-versa is pretty
807 * heavy, involving a CCU cache flush/invalidate and a WFI in order to change
808 * which part of the gmem is used by the CCU. Here we keep track of what the
811 enum tu_cmd_ccu_state
{
817 struct tu_cache_state
{
818 /* Caches which must be made available (flushed) eventually if there are
819 * any users outside that cache domain, and caches which must be
820 * invalidated eventually if there are any reads.
822 enum tu_cmd_flush_bits pending_flush_bits
;
823 /* Pending flushes */
824 enum tu_cmd_flush_bits flush_bits
;
831 struct tu_pipeline
*pipeline
;
832 struct tu_pipeline
*compute_pipeline
;
837 struct tu_buffer
*buffers
[MAX_VBS
];
838 VkDeviceSize offsets
[MAX_VBS
];
841 /* for dynamic states that can't be emitted directly */
842 uint32_t dynamic_stencil_mask
;
843 uint32_t dynamic_stencil_wrmask
;
844 uint32_t dynamic_stencil_ref
;
845 uint32_t dynamic_gras_su_cntl
;
847 /* saved states to re-emit in TU_CMD_DIRTY_DRAW_STATE case */
848 struct tu_draw_state dynamic_state
[TU_DYNAMIC_STATE_COUNT
];
849 struct tu_cs_entry vertex_buffers_ib
;
850 struct tu_cs_entry shader_const_ib
[MESA_SHADER_STAGES
];
851 struct tu_cs_entry desc_sets_ib
;
852 struct tu_cs_entry ia_gmem_ib
, ia_sysmem_ib
;
854 struct tu_draw_state vs_params
;
858 uint32_t max_index_count
;
861 /* because streamout base has to be 32-byte aligned
862 * there is an extra offset to deal with when it is
865 uint8_t streamout_offset
[IR3_MAX_SO_BUFFERS
];
867 /* Renderpasses are tricky, because we may need to flush differently if
868 * using sysmem vs. gmem and therefore we have to delay any flushing that
869 * happens before a renderpass. So we have to have two copies of the flush
870 * state, one for intra-renderpass flushes (i.e. renderpass dependencies)
871 * and one for outside a renderpass.
873 struct tu_cache_state cache
;
874 struct tu_cache_state renderpass_cache
;
876 enum tu_cmd_ccu_state ccu_state
;
878 const struct tu_render_pass
*pass
;
879 const struct tu_subpass
*subpass
;
880 const struct tu_framebuffer
*framebuffer
;
881 VkRect2D render_area
;
883 struct tu_cs_entry tile_store_ib
;
890 struct vk_object_base base
;
892 VkAllocationCallbacks alloc
;
893 struct list_head cmd_buffers
;
894 struct list_head free_cmd_buffers
;
895 uint32_t queue_family_index
;
898 struct tu_cmd_buffer_upload
903 struct list_head list
;
906 enum tu_cmd_buffer_status
908 TU_CMD_BUFFER_STATUS_INVALID
,
909 TU_CMD_BUFFER_STATUS_INITIAL
,
910 TU_CMD_BUFFER_STATUS_RECORDING
,
911 TU_CMD_BUFFER_STATUS_EXECUTABLE
,
912 TU_CMD_BUFFER_STATUS_PENDING
,
919 struct drm_msm_gem_submit_bo
*bo_infos
;
922 #define TU_BO_LIST_FAILED (~0)
925 tu_bo_list_init(struct tu_bo_list
*list
);
927 tu_bo_list_destroy(struct tu_bo_list
*list
);
929 tu_bo_list_reset(struct tu_bo_list
*list
);
931 tu_bo_list_add(struct tu_bo_list
*list
,
932 const struct tu_bo
*bo
,
935 tu_bo_list_merge(struct tu_bo_list
*list
, const struct tu_bo_list
*other
);
939 struct vk_object_base base
;
941 struct tu_device
*device
;
943 struct tu_cmd_pool
*pool
;
944 struct list_head pool_link
;
946 VkCommandBufferUsageFlags usage_flags
;
947 VkCommandBufferLevel level
;
948 enum tu_cmd_buffer_status status
;
950 struct tu_cmd_state state
;
951 struct tu_vertex_binding vertex_bindings
[MAX_VBS
];
952 uint32_t vertex_bindings_set
;
953 uint32_t queue_family_index
;
955 uint32_t push_constants
[MAX_PUSH_CONSTANTS_SIZE
/ 4];
956 VkShaderStageFlags push_constant_stages
;
957 struct tu_descriptor_set meta_push_descriptors
;
959 struct tu_descriptor_state descriptors
[MAX_BIND_POINTS
];
961 struct tu_cmd_buffer_upload upload
;
963 VkResult record_result
;
965 struct tu_bo_list bo_list
;
967 struct tu_cs draw_cs
;
968 struct tu_cs draw_epilogue_cs
;
973 uint32_t vsc_draw_strm_pitch
;
974 uint32_t vsc_prim_strm_pitch
;
977 /* Temporary struct for tracking a register state to be written, used by
978 * a6xx-pack.h and tu_cs_emit_regs()
980 struct tu_reg_value
{
991 void tu_emit_cache_flush_renderpass(struct tu_cmd_buffer
*cmd_buffer
,
994 void tu_emit_cache_flush_ccu(struct tu_cmd_buffer
*cmd_buffer
,
996 enum tu_cmd_ccu_state ccu_state
);
999 tu6_emit_event_write(struct tu_cmd_buffer
*cmd
,
1001 enum vgt_event_type event
);
1003 static inline struct tu_descriptor_state
*
1004 tu_get_descriptors_state(struct tu_cmd_buffer
*cmd_buffer
,
1005 VkPipelineBindPoint bind_point
)
1007 return &cmd_buffer
->descriptors
[bind_point
];
1012 struct vk_object_base base
;
1016 struct tu_shader_module
1018 struct vk_object_base base
;
1020 unsigned char sha1
[20];
1023 const uint32_t *code
[0];
1026 struct tu_push_constant_range
1034 struct ir3_shader
*ir3_shader
;
1036 struct tu_push_constant_range push_consts
;
1037 uint8_t active_desc_sets
;
1041 tu_shader_create(struct tu_device
*dev
,
1042 gl_shader_stage stage
,
1043 const VkPipelineShaderStageCreateInfo
*stage_info
,
1044 struct tu_pipeline_layout
*layout
,
1045 const VkAllocationCallbacks
*alloc
);
1048 tu_shader_destroy(struct tu_device
*dev
,
1049 struct tu_shader
*shader
,
1050 const VkAllocationCallbacks
*alloc
);
1052 struct tu_program_descriptor_linkage
1054 struct ir3_const_state const_state
;
1058 struct tu_push_constant_range push_consts
;
1063 struct vk_object_base base
;
1067 struct tu_pipeline_layout
*layout
;
1069 bool need_indirect_descriptor_sets
;
1070 VkShaderStageFlags active_stages
;
1071 uint32_t active_desc_sets
;
1073 /* mask of enabled dynamic states
1074 * if BIT(i) is set, pipeline->dynamic_state[i] is *NOT* used
1076 uint32_t dynamic_state_mask
;
1077 struct tu_draw_state dynamic_state
[TU_DYNAMIC_STATE_COUNT
];
1079 /* gras_su_cntl without line width, used for dynamic line width state */
1080 uint32_t gras_su_cntl
;
1084 struct tu_cs_entry state_ib
;
1085 struct tu_cs_entry binning_state_ib
;
1087 struct tu_program_descriptor_linkage link
[MESA_SHADER_STAGES
];
1092 struct tu_cs_entry state_ib
;
1097 struct tu_cs_entry state_ib
;
1098 struct tu_cs_entry binning_state_ib
;
1099 uint32_t bindings_used
;
1104 enum pc_di_primtype primtype
;
1105 bool primitive_restart
;
1110 uint32_t patch_type
;
1111 uint32_t param_stride
;
1112 uint32_t hs_bo_regid
;
1113 uint32_t ds_bo_regid
;
1114 bool upper_left_domain_origin
;
1119 struct tu_cs_entry state_ib
;
1124 struct tu_cs_entry state_ib
;
1129 struct tu_cs_entry state_ib
;
1134 uint32_t local_size
[3];
1139 tu6_emit_viewport(struct tu_cs
*cs
, const VkViewport
*viewport
);
1142 tu6_emit_scissor(struct tu_cs
*cs
, const VkRect2D
*scissor
);
1145 tu6_emit_sample_locations(struct tu_cs
*cs
, const VkSampleLocationsInfoEXT
*samp_loc
);
1148 tu6_emit_depth_bias(struct tu_cs
*cs
,
1149 float constant_factor
,
1151 float slope_factor
);
1153 void tu6_emit_msaa(struct tu_cs
*cs
, VkSampleCountFlagBits samples
);
1155 void tu6_emit_window_scissor(struct tu_cs
*cs
, uint32_t x1
, uint32_t y1
, uint32_t x2
, uint32_t y2
);
1157 void tu6_emit_window_offset(struct tu_cs
*cs
, uint32_t x1
, uint32_t y1
);
1160 tu6_emit_xs_config(struct tu_cs
*cs
,
1161 gl_shader_stage stage
,
1162 const struct ir3_shader_variant
*xs
,
1163 uint64_t binary_iova
);
1166 tu6_emit_vpc(struct tu_cs
*cs
,
1167 const struct ir3_shader_variant
*vs
,
1168 const struct ir3_shader_variant
*hs
,
1169 const struct ir3_shader_variant
*ds
,
1170 const struct ir3_shader_variant
*gs
,
1171 const struct ir3_shader_variant
*fs
);
1174 tu6_emit_fs_inputs(struct tu_cs
*cs
, const struct ir3_shader_variant
*fs
);
1176 struct tu_image_view
;
1179 tu_resolve_sysmem(struct tu_cmd_buffer
*cmd
,
1181 struct tu_image_view
*src
,
1182 struct tu_image_view
*dst
,
1184 const VkRect2D
*rect
);
1187 tu_clear_sysmem_attachment(struct tu_cmd_buffer
*cmd
,
1190 const VkRenderPassBeginInfo
*info
);
1193 tu_clear_gmem_attachment(struct tu_cmd_buffer
*cmd
,
1196 const VkRenderPassBeginInfo
*info
);
1199 tu_load_gmem_attachment(struct tu_cmd_buffer
*cmd
,
1204 /* expose this function to be able to emit load without checking LOAD_OP */
1206 tu_emit_load_gmem_attachment(struct tu_cmd_buffer
*cmd
, struct tu_cs
*cs
, uint32_t a
);
1208 /* note: gmem store can also resolve */
1210 tu_store_gmem_attachment(struct tu_cmd_buffer
*cmd
,
1215 enum tu_supported_formats
{
1221 struct tu_native_format
1223 enum a6xx_format fmt
: 8;
1224 enum a3xx_color_swap swap
: 8;
1225 enum a6xx_tile_mode tile_mode
: 8;
1226 enum tu_supported_formats supported
: 8;
1229 struct tu_native_format
tu6_format_vtx(VkFormat format
);
1230 struct tu_native_format
tu6_format_color(VkFormat format
, enum a6xx_tile_mode tile_mode
);
1231 struct tu_native_format
tu6_format_texture(VkFormat format
, enum a6xx_tile_mode tile_mode
);
1233 static inline enum a6xx_format
1234 tu6_base_format(VkFormat format
)
1236 /* note: tu6_format_color doesn't care about tiling for .fmt field */
1237 return tu6_format_color(format
, TILE6_LINEAR
).fmt
;
1242 struct vk_object_base base
;
1245 /* The original VkFormat provided by the client. This may not match any
1246 * of the actual surface formats.
1249 VkImageAspectFlags aspects
;
1250 VkImageUsageFlags usage
; /**< Superset of VkImageCreateInfo::usage. */
1251 VkImageTiling tiling
; /** VkImageCreateInfo::tiling */
1252 VkImageCreateFlags flags
; /** VkImageCreateInfo::flags */
1254 uint32_t level_count
;
1255 uint32_t layer_count
;
1256 VkSampleCountFlagBits samples
;
1258 struct fdl_layout layout
[3];
1259 uint32_t total_size
;
1261 unsigned queue_family_mask
;
1265 /* For VK_ANDROID_native_buffer, the WSI image owns the memory, */
1266 VkDeviceMemory owned_memory
;
1268 /* Set when bound */
1270 VkDeviceSize bo_offset
;
1273 static inline uint32_t
1274 tu_get_layerCount(const struct tu_image
*image
,
1275 const VkImageSubresourceRange
*range
)
1277 return range
->layerCount
== VK_REMAINING_ARRAY_LAYERS
1278 ? image
->layer_count
- range
->baseArrayLayer
1279 : range
->layerCount
;
1282 static inline uint32_t
1283 tu_get_levelCount(const struct tu_image
*image
,
1284 const VkImageSubresourceRange
*range
)
1286 return range
->levelCount
== VK_REMAINING_MIP_LEVELS
1287 ? image
->level_count
- range
->baseMipLevel
1288 : range
->levelCount
;
1291 struct tu_image_view
1293 struct vk_object_base base
;
1295 struct tu_image
*image
; /**< VkImageViewCreateInfo::image */
1299 uint32_t layer_size
;
1300 uint32_t ubwc_layer_size
;
1302 /* used to determine if fast gmem store path can be used */
1308 uint32_t descriptor
[A6XX_TEX_CONST_DWORDS
];
1310 /* Descriptor for use as a storage image as opposed to a sampled image.
1311 * This has a few differences for cube maps (e.g. type).
1313 uint32_t storage_descriptor
[A6XX_TEX_CONST_DWORDS
];
1315 /* pre-filled register values */
1317 uint32_t FLAG_BUFFER_PITCH
;
1319 uint32_t RB_MRT_BUF_INFO
;
1320 uint32_t SP_FS_MRT_REG
;
1322 uint32_t SP_PS_2D_SRC_INFO
;
1323 uint32_t SP_PS_2D_SRC_SIZE
;
1325 uint32_t RB_2D_DST_INFO
;
1327 uint32_t RB_BLIT_DST_INFO
;
1330 struct tu_sampler_ycbcr_conversion
{
1331 struct vk_object_base base
;
1334 VkSamplerYcbcrModelConversion ycbcr_model
;
1335 VkSamplerYcbcrRange ycbcr_range
;
1336 VkComponentMapping components
;
1337 VkChromaLocation chroma_offsets
[2];
1338 VkFilter chroma_filter
;
1342 struct vk_object_base base
;
1344 uint32_t descriptor
[A6XX_TEX_SAMP_DWORDS
];
1345 struct tu_sampler_ycbcr_conversion
*ycbcr_sampler
;
1349 tu_cs_image_ref(struct tu_cs
*cs
, const struct tu_image_view
*iview
, uint32_t layer
);
1352 tu_cs_image_ref_2d(struct tu_cs
*cs
, const struct tu_image_view
*iview
, uint32_t layer
, bool src
);
1355 tu_cs_image_flag_ref(struct tu_cs
*cs
, const struct tu_image_view
*iview
, uint32_t layer
);
1358 tu_image_create(VkDevice _device
,
1359 const VkImageCreateInfo
*pCreateInfo
,
1360 const VkAllocationCallbacks
*alloc
,
1363 const VkSubresourceLayout
*plane_layouts
);
1366 tu_image_from_gralloc(VkDevice device_h
,
1367 const VkImageCreateInfo
*base_info
,
1368 const VkNativeBufferANDROID
*gralloc_info
,
1369 const VkAllocationCallbacks
*alloc
,
1370 VkImage
*out_image_h
);
1373 tu_image_view_init(struct tu_image_view
*view
,
1374 const VkImageViewCreateInfo
*pCreateInfo
);
1376 struct tu_buffer_view
1378 struct vk_object_base base
;
1380 uint32_t descriptor
[A6XX_TEX_CONST_DWORDS
];
1382 struct tu_buffer
*buffer
;
1385 tu_buffer_view_init(struct tu_buffer_view
*view
,
1386 struct tu_device
*device
,
1387 const VkBufferViewCreateInfo
*pCreateInfo
);
1389 struct tu_attachment_info
1391 struct tu_image_view
*attachment
;
1394 struct tu_framebuffer
1396 struct vk_object_base base
;
1402 /* size of the first tile */
1404 /* number of tiles */
1405 VkExtent2D tile_count
;
1407 /* size of the first VSC pipe */
1409 /* number of VSC pipes */
1410 VkExtent2D pipe_count
;
1412 /* pipe register values */
1413 uint32_t pipe_config
[MAX_VSC_PIPES
];
1414 uint32_t pipe_sizes
[MAX_VSC_PIPES
];
1416 uint32_t attachment_count
;
1417 struct tu_attachment_info attachments
[0];
1421 tu_framebuffer_tiling_config(struct tu_framebuffer
*fb
,
1422 const struct tu_device
*device
,
1423 const struct tu_render_pass
*pass
);
1425 struct tu_subpass_barrier
{
1426 VkPipelineStageFlags src_stage_mask
;
1427 VkAccessFlags src_access_mask
;
1428 VkAccessFlags dst_access_mask
;
1429 bool incoherent_ccu_color
, incoherent_ccu_depth
;
1432 struct tu_subpass_attachment
1434 uint32_t attachment
;
1439 uint32_t input_count
;
1440 uint32_t color_count
;
1441 struct tu_subpass_attachment
*input_attachments
;
1442 struct tu_subpass_attachment
*color_attachments
;
1443 struct tu_subpass_attachment
*resolve_attachments
;
1444 struct tu_subpass_attachment depth_stencil_attachment
;
1446 VkSampleCountFlagBits samples
;
1450 struct tu_subpass_barrier start_barrier
;
1453 struct tu_render_pass_attachment
1458 VkImageAspectFlags clear_mask
;
1461 int32_t gmem_offset
;
1464 struct tu_render_pass
1466 struct vk_object_base base
;
1468 uint32_t attachment_count
;
1469 uint32_t subpass_count
;
1470 uint32_t gmem_pixels
;
1471 uint32_t tile_align_w
;
1472 struct tu_subpass_attachment
*subpass_attachments
;
1473 struct tu_render_pass_attachment
*attachments
;
1474 struct tu_subpass_barrier end_barrier
;
1475 struct tu_subpass subpasses
[0];
1478 struct tu_query_pool
1480 struct vk_object_base base
;
1485 uint32_t pipeline_statistics
;
1489 enum tu_semaphore_kind
1492 TU_SEMAPHORE_SYNCOBJ
,
1495 struct tu_semaphore_part
1497 enum tu_semaphore_kind kind
;
1505 struct vk_object_base base
;
1507 struct tu_semaphore_part permanent
;
1508 struct tu_semaphore_part temporary
;
1512 tu_set_descriptor_set(struct tu_cmd_buffer
*cmd_buffer
,
1513 VkPipelineBindPoint bind_point
,
1514 struct tu_descriptor_set
*set
,
1518 tu_update_descriptor_sets(struct tu_device
*device
,
1519 struct tu_cmd_buffer
*cmd_buffer
,
1520 VkDescriptorSet overrideSet
,
1521 uint32_t descriptorWriteCount
,
1522 const VkWriteDescriptorSet
*pDescriptorWrites
,
1523 uint32_t descriptorCopyCount
,
1524 const VkCopyDescriptorSet
*pDescriptorCopies
);
1527 tu_update_descriptor_set_with_template(
1528 struct tu_device
*device
,
1529 struct tu_cmd_buffer
*cmd_buffer
,
1530 struct tu_descriptor_set
*set
,
1531 VkDescriptorUpdateTemplate descriptorUpdateTemplate
,
1535 tu_drm_get_gpu_id(const struct tu_physical_device
*dev
, uint32_t *id
);
1538 tu_drm_get_gmem_size(const struct tu_physical_device
*dev
, uint32_t *size
);
1541 tu_drm_get_gmem_base(const struct tu_physical_device
*dev
, uint64_t *base
);
1544 tu_drm_submitqueue_new(const struct tu_device
*dev
,
1546 uint32_t *queue_id
);
1549 tu_drm_submitqueue_close(const struct tu_device
*dev
, uint32_t queue_id
);
1552 tu_gem_new(const struct tu_device
*dev
, uint64_t size
, uint32_t flags
);
1554 tu_gem_import_dmabuf(const struct tu_device
*dev
,
1558 tu_gem_export_dmabuf(const struct tu_device
*dev
, uint32_t gem_handle
);
1560 tu_gem_close(const struct tu_device
*dev
, uint32_t gem_handle
);
1562 tu_gem_info_offset(const struct tu_device
*dev
, uint32_t gem_handle
);
1564 tu_gem_info_iova(const struct tu_device
*dev
, uint32_t gem_handle
);
1566 #define TU_DEFINE_HANDLE_CASTS(__tu_type, __VkType) \
1568 static inline struct __tu_type *__tu_type##_from_handle(__VkType _handle) \
1570 return (struct __tu_type *) _handle; \
1573 static inline __VkType __tu_type##_to_handle(struct __tu_type *_obj) \
1575 return (__VkType) _obj; \
1578 #define TU_DEFINE_NONDISP_HANDLE_CASTS(__tu_type, __VkType) \
1580 static inline struct __tu_type *__tu_type##_from_handle(__VkType _handle) \
1582 return (struct __tu_type *) (uintptr_t) _handle; \
1585 static inline __VkType __tu_type##_to_handle(struct __tu_type *_obj) \
1587 return (__VkType)(uintptr_t) _obj; \
1590 #define TU_FROM_HANDLE(__tu_type, __name, __handle) \
1591 struct __tu_type *__name = __tu_type##_from_handle(__handle)
1593 TU_DEFINE_HANDLE_CASTS(tu_cmd_buffer
, VkCommandBuffer
)
1594 TU_DEFINE_HANDLE_CASTS(tu_device
, VkDevice
)
1595 TU_DEFINE_HANDLE_CASTS(tu_instance
, VkInstance
)
1596 TU_DEFINE_HANDLE_CASTS(tu_physical_device
, VkPhysicalDevice
)
1597 TU_DEFINE_HANDLE_CASTS(tu_queue
, VkQueue
)
1599 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_cmd_pool
, VkCommandPool
)
1600 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_buffer
, VkBuffer
)
1601 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_buffer_view
, VkBufferView
)
1602 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_pool
, VkDescriptorPool
)
1603 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_set
, VkDescriptorSet
)
1604 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_set_layout
,
1605 VkDescriptorSetLayout
)
1606 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_update_template
,
1607 VkDescriptorUpdateTemplate
)
1608 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_device_memory
, VkDeviceMemory
)
1609 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_fence
, VkFence
)
1610 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_event
, VkEvent
)
1611 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_framebuffer
, VkFramebuffer
)
1612 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_image
, VkImage
)
1613 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_image_view
, VkImageView
);
1614 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_pipeline_cache
, VkPipelineCache
)
1615 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_pipeline
, VkPipeline
)
1616 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_pipeline_layout
, VkPipelineLayout
)
1617 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_query_pool
, VkQueryPool
)
1618 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_render_pass
, VkRenderPass
)
1619 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_sampler
, VkSampler
)
1620 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_sampler_ycbcr_conversion
, VkSamplerYcbcrConversion
)
1621 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_shader_module
, VkShaderModule
)
1622 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_semaphore
, VkSemaphore
)
1624 #endif /* TU_PRIVATE_H */