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 "ir3/ir3_compiler.h"
57 #include "ir3/ir3_shader.h"
59 #include "adreno_common.xml.h"
60 #include "adreno_pm4.xml.h"
62 #include "fdl/freedreno_layout.h"
64 #include "tu_descriptor_set.h"
65 #include "tu_extensions.h"
68 /* Pre-declarations needed for WSI entrypoints */
71 typedef struct xcb_connection_t xcb_connection_t
;
72 typedef uint32_t xcb_visualid_t
;
73 typedef uint32_t xcb_window_t
;
75 #include <vulkan/vk_android_native_buffer.h>
76 #include <vulkan/vk_icd.h>
77 #include <vulkan/vulkan.h>
78 #include <vulkan/vulkan_intel.h>
80 #include "tu_entrypoints.h"
82 #include "vk_format.h"
85 #define MAX_VERTEX_ATTRIBS 32
87 #define MAX_VSC_PIPES 32
88 #define MAX_VIEWPORTS 1
89 #define MAX_SCISSORS 16
90 #define MAX_DISCARD_RECTANGLES 4
91 #define MAX_PUSH_CONSTANTS_SIZE 128
92 #define MAX_PUSH_DESCRIPTORS 32
93 #define MAX_DYNAMIC_UNIFORM_BUFFERS 16
94 #define MAX_DYNAMIC_STORAGE_BUFFERS 8
95 #define MAX_DYNAMIC_BUFFERS \
96 (MAX_DYNAMIC_UNIFORM_BUFFERS + MAX_DYNAMIC_STORAGE_BUFFERS)
97 #define TU_MAX_DRM_DEVICES 8
99 #define MAX_BIND_POINTS 2 /* compute + graphics */
100 /* The Qualcomm driver exposes 0x20000058 */
101 #define MAX_STORAGE_BUFFER_RANGE 0x20000000
102 /* We use ldc for uniform buffer loads, just like the Qualcomm driver, so
103 * expose the same maximum range.
104 * TODO: The SIZE bitfield is 15 bits, and in 4-dword units, so the actual
105 * range might be higher.
107 #define MAX_UNIFORM_BUFFER_RANGE 0x10000
109 #define A6XX_TEX_CONST_DWORDS 16
110 #define A6XX_TEX_SAMP_DWORDS 4
112 #define tu_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
114 static inline uint32_t
115 tu_minify(uint32_t n
, uint32_t levels
)
117 if (unlikely(n
== 0))
120 return MAX2(n
>> levels
, 1);
123 #define for_each_bit(b, dword) \
124 for (uint32_t __dword = (dword); \
125 (b) = __builtin_ffs(__dword) - 1, __dword; __dword &= ~(1 << (b)))
127 #define typed_memcpy(dest, src, count) \
129 STATIC_ASSERT(sizeof(*src) == sizeof(*dest)); \
130 memcpy((dest), (src), (count) * sizeof(*(src))); \
133 #define COND(bool, val) ((bool) ? (val) : 0)
134 #define BIT(bit) (1u << (bit))
136 /* Whenever we generate an error, pass it through this function. Useful for
137 * debugging, where we can break on it. Only call at error site, not when
138 * propagating errors. Might be useful to plug in a stack trace here.
144 __vk_errorf(struct tu_instance
*instance
,
151 #define vk_error(instance, error) \
152 __vk_errorf(instance, error, __FILE__, __LINE__, NULL);
153 #define vk_errorf(instance, error, format, ...) \
154 __vk_errorf(instance, error, __FILE__, __LINE__, format, ##__VA_ARGS__);
157 __tu_finishme(const char *file
, int line
, const char *format
, ...)
160 tu_loge(const char *format
, ...) tu_printflike(1, 2);
162 tu_logi(const char *format
, ...) tu_printflike(1, 2);
165 * Print a FINISHME message, including its source location.
167 #define tu_finishme(format, ...) \
169 static bool reported = false; \
171 __tu_finishme(__FILE__, __LINE__, format, ##__VA_ARGS__); \
178 tu_finishme("stub %s", __func__); \
182 tu_lookup_entrypoint_unchecked(const char *name
);
184 tu_lookup_entrypoint_checked(
186 uint32_t core_version
,
187 const struct tu_instance_extension_table
*instance
,
188 const struct tu_device_extension_table
*device
);
190 struct tu_physical_device
192 struct vk_object_base base
;
194 struct tu_instance
*instance
;
197 char name
[VK_MAX_PHYSICAL_DEVICE_NAME_SIZE
];
198 uint8_t driver_uuid
[VK_UUID_SIZE
];
199 uint8_t device_uuid
[VK_UUID_SIZE
];
200 uint8_t cache_uuid
[VK_UUID_SIZE
];
202 struct wsi_device wsi_device
;
210 uint32_t ccu_offset_gmem
;
211 uint32_t ccu_offset_bypass
;
212 /* alignment for size of tiles */
213 uint32_t tile_align_w
;
214 #define TILE_ALIGN_H 16
215 /* gmem store/load granularity */
216 #define GMEM_ALIGN_W 16
217 #define GMEM_ALIGN_H 4
220 uint32_t PC_UNKNOWN_9805
;
221 uint32_t SP_UNKNOWN_A0F8
;
224 int msm_major_version
;
225 int msm_minor_version
;
229 /* This is the drivers on-disk cache used as a fallback as opposed to
230 * the pipeline cache defined by apps.
232 struct disk_cache
*disk_cache
;
234 struct tu_device_extension_table supported_extensions
;
239 TU_DEBUG_STARTUP
= 1 << 0,
240 TU_DEBUG_NIR
= 1 << 1,
241 TU_DEBUG_IR3
= 1 << 2,
242 TU_DEBUG_NOBIN
= 1 << 3,
243 TU_DEBUG_SYSMEM
= 1 << 4,
244 TU_DEBUG_FORCEBIN
= 1 << 5,
245 TU_DEBUG_NOUBWC
= 1 << 6,
250 struct vk_object_base base
;
252 VkAllocationCallbacks alloc
;
254 uint32_t api_version
;
255 int physical_device_count
;
256 struct tu_physical_device physical_devices
[TU_MAX_DRM_DEVICES
];
258 enum tu_debug_flags debug_flags
;
260 struct vk_debug_report_instance debug_report_callbacks
;
262 struct tu_instance_extension_table enabled_extensions
;
266 tu_wsi_init(struct tu_physical_device
*physical_device
);
268 tu_wsi_finish(struct tu_physical_device
*physical_device
);
271 tu_instance_extension_supported(const char *name
);
273 tu_physical_device_api_version(struct tu_physical_device
*dev
);
275 tu_physical_device_extension_supported(struct tu_physical_device
*dev
,
280 struct tu_pipeline_cache
282 struct vk_object_base base
;
284 struct tu_device
*device
;
285 pthread_mutex_t mutex
;
289 uint32_t kernel_count
;
290 struct cache_entry
**hash_table
;
293 VkAllocationCallbacks alloc
;
296 struct tu_pipeline_key
302 #define TU_QUEUE_GENERAL 0
304 #define TU_MAX_QUEUE_FAMILIES 1
308 struct vk_object_base base
;
309 struct wsi_fence
*fence_wsi
;
315 tu_fence_init(struct tu_fence
*fence
, bool signaled
);
317 tu_fence_finish(struct tu_fence
*fence
);
319 tu_fence_update_fd(struct tu_fence
*fence
, int fd
);
321 tu_fence_copy(struct tu_fence
*fence
, const struct tu_fence
*src
);
323 tu_fence_signal(struct tu_fence
*fence
);
325 tu_fence_wait_idle(struct tu_fence
*fence
);
329 struct vk_object_base base
;
331 struct tu_device
*device
;
332 uint32_t queue_family_index
;
334 VkDeviceQueueCreateFlags flags
;
336 uint32_t msm_queue_id
;
337 struct tu_fence submit_fence
;
352 GLOBAL_SH_FS_CLEAR_MAX
= GLOBAL_SH_FS_CLEAR0
+ MAX_RTS
,
356 /* This struct defines the layout of the global_bo */
359 /* 6 bcolor_entry entries, one for each VK_BORDER_COLOR */
360 uint8_t border_color
[128 * 6];
362 /* clear/blit shaders, all <= 16 instrs (16 instr = 1 instrlen unit) */
363 instr_t shaders
[GLOBAL_SH_COUNT
][16];
365 uint32_t seqno_dummy
; /* dummy seqno for CP_EVENT_WRITE */
367 volatile uint32_t vsc_draw_overflow
;
369 volatile uint32_t vsc_prim_overflow
;
373 /* scratch space for VPC_SO[i].FLUSH_BASE_LO/HI, start on 32 byte boundary. */
379 #define gb_offset(member) offsetof(struct tu6_global, member)
380 #define global_iova(cmd, member) ((cmd)->device->global_bo.iova + gb_offset(member))
382 void tu_init_clear_blit_shaders(struct tu6_global
*global
);
384 /* extra space in vsc draw/prim streams */
390 struct tu_instance
*instance
;
392 struct tu_queue
*queues
[TU_MAX_QUEUE_FAMILIES
];
393 int queue_count
[TU_MAX_QUEUE_FAMILIES
];
395 struct tu_physical_device
*physical_device
;
398 struct ir3_compiler
*compiler
;
400 /* Backup in-memory cache to be used if the app doesn't provide one */
401 struct tu_pipeline_cache
*mem_cache
;
403 #define MIN_SCRATCH_BO_SIZE_LOG2 12 /* A page */
405 /* Currently the kernel driver uses a 32-bit GPU address space, but it
406 * should be impossible to go beyond 48 bits.
412 } scratch_bos
[48 - MIN_SCRATCH_BO_SIZE_LOG2
];
414 struct tu_bo global_bo
;
416 struct tu_device_extension_table enabled_extensions
;
418 uint32_t vsc_draw_strm_pitch
;
419 uint32_t vsc_prim_strm_pitch
;
423 VkResult
_tu_device_set_lost(struct tu_device
*device
,
424 const char *file
, int line
,
425 const char *msg
, ...) PRINTFLIKE(4, 5);
426 #define tu_device_set_lost(dev, ...) \
427 _tu_device_set_lost(dev, __FILE__, __LINE__, __VA_ARGS__)
430 tu_device_is_lost(struct tu_device
*device
)
432 return unlikely(p_atomic_read(&device
->_lost
));
436 tu_bo_init_new(struct tu_device
*dev
, struct tu_bo
*bo
, uint64_t size
);
438 tu_bo_init_dmabuf(struct tu_device
*dev
,
443 tu_bo_export_dmabuf(struct tu_device
*dev
, struct tu_bo
*bo
);
445 tu_bo_finish(struct tu_device
*dev
, struct tu_bo
*bo
);
447 tu_bo_map(struct tu_device
*dev
, struct tu_bo
*bo
);
449 /* Get a scratch bo for use inside a command buffer. This will always return
450 * the same bo given the same size or similar sizes, so only one scratch bo
451 * can be used at the same time. It's meant for short-lived things where we
452 * need to write to some piece of memory, read from it, and then immediately
456 tu_get_scratch_bo(struct tu_device
*dev
, uint64_t size
, struct tu_bo
**bo
);
461 const struct tu_bo
*bo
;
467 struct tu_cs_memory
{
472 struct tu_draw_state
{
477 enum tu_dynamic_state
479 /* re-use VK_DYNAMIC_STATE_ enums for non-extended dynamic states */
480 TU_DYNAMIC_STATE_SAMPLE_LOCATIONS
= VK_DYNAMIC_STATE_STENCIL_REFERENCE
+ 1,
481 TU_DYNAMIC_STATE_COUNT
,
484 enum tu_draw_state_group_id
486 TU_DRAW_STATE_PROGRAM
,
487 TU_DRAW_STATE_PROGRAM_BINNING
,
491 TU_DRAW_STATE_VI_BINNING
,
495 TU_DRAW_STATE_VS_CONST
,
496 TU_DRAW_STATE_HS_CONST
,
497 TU_DRAW_STATE_DS_CONST
,
498 TU_DRAW_STATE_GS_CONST
,
499 TU_DRAW_STATE_FS_CONST
,
500 TU_DRAW_STATE_DESC_SETS
,
501 TU_DRAW_STATE_DESC_SETS_LOAD
,
502 TU_DRAW_STATE_VS_PARAMS
,
503 TU_DRAW_STATE_INPUT_ATTACHMENTS_GMEM
,
504 TU_DRAW_STATE_INPUT_ATTACHMENTS_SYSMEM
,
506 /* dynamic state related draw states */
507 TU_DRAW_STATE_DYNAMIC
,
508 TU_DRAW_STATE_COUNT
= TU_DRAW_STATE_DYNAMIC
+ TU_DYNAMIC_STATE_COUNT
,
515 * A command stream in TU_CS_MODE_GROW mode grows automatically whenever it
516 * is full. tu_cs_begin must be called before command packet emission and
517 * tu_cs_end must be called after.
519 * This mode may create multiple entries internally. The entries must be
520 * submitted together.
525 * A command stream in TU_CS_MODE_EXTERNAL mode wraps an external,
526 * fixed-size buffer. tu_cs_begin and tu_cs_end are optional and have no
529 * This mode does not create any entry or any BO.
534 * A command stream in TU_CS_MODE_SUB_STREAM mode does not support direct
535 * command packet emission. tu_cs_begin_sub_stream must be called to get a
536 * sub-stream to emit comamnd packets to. When done with the sub-stream,
537 * tu_cs_end_sub_stream must be called.
539 * This mode does not create any entry internally.
541 TU_CS_MODE_SUB_STREAM
,
548 uint32_t *reserved_end
;
551 struct tu_device
*device
;
552 enum tu_cs_mode mode
;
553 uint32_t next_bo_size
;
555 struct tu_cs_entry
*entries
;
556 uint32_t entry_count
;
557 uint32_t entry_capacity
;
561 uint32_t bo_capacity
;
563 /* state for cond_exec_start/cond_exec_end */
565 uint32_t *cond_dwords
;
568 struct tu_device_memory
570 struct vk_object_base base
;
575 /* for dedicated allocations */
576 struct tu_image
*image
;
577 struct tu_buffer
*buffer
;
584 struct tu_descriptor_range
590 struct tu_descriptor_set
592 struct vk_object_base base
;
594 const struct tu_descriptor_set_layout
*layout
;
595 struct tu_descriptor_pool
*pool
;
599 uint32_t *mapped_ptr
;
601 uint32_t *dynamic_descriptors
;
603 struct tu_bo
*buffers
[0];
606 struct tu_push_descriptor_set
608 struct tu_descriptor_set set
;
612 struct tu_descriptor_pool_entry
616 struct tu_descriptor_set
*set
;
619 struct tu_descriptor_pool
621 struct vk_object_base base
;
624 uint64_t current_offset
;
627 uint8_t *host_memory_base
;
628 uint8_t *host_memory_ptr
;
629 uint8_t *host_memory_end
;
631 uint32_t entry_count
;
632 uint32_t max_entry_count
;
633 struct tu_descriptor_pool_entry entries
[0];
636 struct tu_descriptor_update_template_entry
638 VkDescriptorType descriptor_type
;
640 /* The number of descriptors to update */
641 uint32_t descriptor_count
;
643 /* Into mapped_ptr or dynamic_descriptors, in units of the respective array
647 /* In dwords. Not valid/used for dynamic descriptors */
650 uint32_t buffer_offset
;
652 /* Only valid for combined image samplers and samplers */
653 uint16_t has_sampler
;
659 /* For push descriptors */
660 const uint32_t *immutable_samplers
;
663 struct tu_descriptor_update_template
665 struct vk_object_base base
;
667 uint32_t entry_count
;
668 struct tu_descriptor_update_template_entry entry
[0];
673 struct vk_object_base base
;
677 VkBufferUsageFlags usage
;
678 VkBufferCreateFlags flags
;
681 VkDeviceSize bo_offset
;
684 static inline uint64_t
685 tu_buffer_iova(struct tu_buffer
*buffer
)
687 return buffer
->bo
->iova
+ buffer
->bo_offset
;
690 struct tu_vertex_binding
692 struct tu_buffer
*buffer
;
697 tu_get_debug_option_name(int id
);
700 tu_get_perftest_option_name(int id
);
702 struct tu_descriptor_state
704 struct tu_descriptor_set
*sets
[MAX_SETS
];
705 uint32_t dynamic_descriptors
[MAX_DYNAMIC_BUFFERS
* A6XX_TEX_CONST_DWORDS
];
708 enum tu_cmd_dirty_bits
710 TU_CMD_DIRTY_VERTEX_BUFFERS
= 1 << 2,
711 TU_CMD_DIRTY_DESC_SETS_LOAD
= 1 << 3,
712 TU_CMD_DIRTY_COMPUTE_DESC_SETS_LOAD
= 1 << 4,
713 TU_CMD_DIRTY_SHADER_CONSTS
= 1 << 5,
714 /* all draw states were disabled and need to be re-enabled: */
715 TU_CMD_DIRTY_DRAW_STATE
= 1 << 7,
718 /* There are only three cache domains we have to care about: the CCU, or
719 * color cache unit, which is used for color and depth/stencil attachments
720 * and copy/blit destinations, and is split conceptually into color and depth,
721 * and the universal cache or UCHE which is used for pretty much everything
722 * else, except for the CP (uncached) and host. We need to flush whenever data
723 * crosses these boundaries.
726 enum tu_cmd_access_mask
{
727 TU_ACCESS_UCHE_READ
= 1 << 0,
728 TU_ACCESS_UCHE_WRITE
= 1 << 1,
729 TU_ACCESS_CCU_COLOR_READ
= 1 << 2,
730 TU_ACCESS_CCU_COLOR_WRITE
= 1 << 3,
731 TU_ACCESS_CCU_DEPTH_READ
= 1 << 4,
732 TU_ACCESS_CCU_DEPTH_WRITE
= 1 << 5,
734 /* Experiments have shown that while it's safe to avoid flushing the CCU
735 * after each blit/renderpass, it's not safe to assume that subsequent
736 * lookups with a different attachment state will hit unflushed cache
737 * entries. That is, the CCU needs to be flushed and possibly invalidated
738 * when accessing memory with a different attachment state. Writing to an
739 * attachment under the following conditions after clearing using the
740 * normal 2d engine path is known to have issues:
742 * - It isn't the 0'th layer.
743 * - There are more than one attachment, and this isn't the 0'th attachment
744 * (this seems to also depend on the cpp of the attachments).
746 * Our best guess is that the layer/MRT state is used when computing
747 * the location of a cache entry in CCU, to avoid conflicts. We assume that
748 * any access in a renderpass after or before an access by a transfer needs
749 * a flush/invalidate, and use the _INCOHERENT variants to represent access
752 TU_ACCESS_CCU_COLOR_INCOHERENT_READ
= 1 << 6,
753 TU_ACCESS_CCU_COLOR_INCOHERENT_WRITE
= 1 << 7,
754 TU_ACCESS_CCU_DEPTH_INCOHERENT_READ
= 1 << 8,
755 TU_ACCESS_CCU_DEPTH_INCOHERENT_WRITE
= 1 << 9,
757 /* Accesses by the host */
758 TU_ACCESS_HOST_READ
= 1 << 10,
759 TU_ACCESS_HOST_WRITE
= 1 << 11,
761 /* Accesses by a GPU engine which bypasses any cache. e.g. writes via
762 * CP_EVENT_WRITE::BLIT and the CP are SYSMEM_WRITE.
764 TU_ACCESS_SYSMEM_READ
= 1 << 12,
765 TU_ACCESS_SYSMEM_WRITE
= 1 << 13,
767 /* Set if a WFI is required. This can be required for:
768 * - 2D engine which (on some models) doesn't wait for flushes to complete
770 * - CP draw indirect opcodes, where we need to wait for any flushes to
771 * complete but the CP implicitly waits for WFI's to complete and
772 * therefore we only need a WFI after the flushes.
774 TU_ACCESS_WFI_READ
= 1 << 14,
776 /* Set if a CP_WAIT_FOR_ME is required due to the data being read by the CP
777 * without it waiting for any WFI.
779 TU_ACCESS_WFM_READ
= 1 << 15,
781 /* Memory writes from the CP start in-order with draws and event writes,
782 * but execute asynchronously and hence need a CP_WAIT_MEM_WRITES if read.
784 TU_ACCESS_CP_WRITE
= 1 << 16,
787 TU_ACCESS_UCHE_READ
|
788 TU_ACCESS_CCU_COLOR_READ
|
789 TU_ACCESS_CCU_DEPTH_READ
|
790 TU_ACCESS_CCU_COLOR_INCOHERENT_READ
|
791 TU_ACCESS_CCU_DEPTH_INCOHERENT_READ
|
792 TU_ACCESS_HOST_READ
|
793 TU_ACCESS_SYSMEM_READ
|
798 TU_ACCESS_UCHE_WRITE
|
799 TU_ACCESS_CCU_COLOR_WRITE
|
800 TU_ACCESS_CCU_COLOR_INCOHERENT_WRITE
|
801 TU_ACCESS_CCU_DEPTH_WRITE
|
802 TU_ACCESS_CCU_DEPTH_INCOHERENT_WRITE
|
803 TU_ACCESS_HOST_WRITE
|
804 TU_ACCESS_SYSMEM_WRITE
|
812 enum tu_cmd_flush_bits
{
813 TU_CMD_FLAG_CCU_FLUSH_DEPTH
= 1 << 0,
814 TU_CMD_FLAG_CCU_FLUSH_COLOR
= 1 << 1,
815 TU_CMD_FLAG_CCU_INVALIDATE_DEPTH
= 1 << 2,
816 TU_CMD_FLAG_CCU_INVALIDATE_COLOR
= 1 << 3,
817 TU_CMD_FLAG_CACHE_FLUSH
= 1 << 4,
818 TU_CMD_FLAG_CACHE_INVALIDATE
= 1 << 5,
819 TU_CMD_FLAG_WAIT_MEM_WRITES
= 1 << 6,
820 TU_CMD_FLAG_WAIT_FOR_IDLE
= 1 << 7,
821 TU_CMD_FLAG_WAIT_FOR_ME
= 1 << 8,
823 TU_CMD_FLAG_ALL_FLUSH
=
824 TU_CMD_FLAG_CCU_FLUSH_DEPTH
|
825 TU_CMD_FLAG_CCU_FLUSH_COLOR
|
826 TU_CMD_FLAG_CACHE_FLUSH
|
827 /* Treat the CP as a sort of "cache" which may need to be "flushed" via
828 * waiting for writes to land with WAIT_FOR_MEM_WRITES.
830 TU_CMD_FLAG_WAIT_MEM_WRITES
,
832 TU_CMD_FLAG_GPU_INVALIDATE
=
833 TU_CMD_FLAG_CCU_INVALIDATE_DEPTH
|
834 TU_CMD_FLAG_CCU_INVALIDATE_COLOR
|
835 TU_CMD_FLAG_CACHE_INVALIDATE
,
837 TU_CMD_FLAG_ALL_INVALIDATE
=
838 TU_CMD_FLAG_GPU_INVALIDATE
|
839 /* Treat the CP as a sort of "cache" which may need to be "invalidated"
840 * via waiting for UCHE/CCU flushes to land with WFI/WFM.
842 TU_CMD_FLAG_WAIT_FOR_IDLE
|
843 TU_CMD_FLAG_WAIT_FOR_ME
,
846 /* Changing the CCU from sysmem mode to gmem mode or vice-versa is pretty
847 * heavy, involving a CCU cache flush/invalidate and a WFI in order to change
848 * which part of the gmem is used by the CCU. Here we keep track of what the
851 enum tu_cmd_ccu_state
{
857 struct tu_cache_state
{
858 /* Caches which must be made available (flushed) eventually if there are
859 * any users outside that cache domain, and caches which must be
860 * invalidated eventually if there are any reads.
862 enum tu_cmd_flush_bits pending_flush_bits
;
863 /* Pending flushes */
864 enum tu_cmd_flush_bits flush_bits
;
871 struct tu_pipeline
*pipeline
;
872 struct tu_pipeline
*compute_pipeline
;
877 struct tu_buffer
*buffers
[MAX_VBS
];
878 VkDeviceSize offsets
[MAX_VBS
];
881 /* for dynamic states that can't be emitted directly */
882 uint32_t dynamic_stencil_mask
;
883 uint32_t dynamic_stencil_wrmask
;
884 uint32_t dynamic_stencil_ref
;
885 uint32_t dynamic_gras_su_cntl
;
887 /* saved states to re-emit in TU_CMD_DIRTY_DRAW_STATE case */
888 struct tu_draw_state dynamic_state
[TU_DYNAMIC_STATE_COUNT
];
889 struct tu_draw_state vertex_buffers
;
890 struct tu_draw_state shader_const
[MESA_SHADER_STAGES
];
891 struct tu_draw_state desc_sets
;
893 struct tu_draw_state vs_params
;
897 uint32_t max_index_count
;
900 /* because streamout base has to be 32-byte aligned
901 * there is an extra offset to deal with when it is
904 uint8_t streamout_offset
[IR3_MAX_SO_BUFFERS
];
906 /* Renderpasses are tricky, because we may need to flush differently if
907 * using sysmem vs. gmem and therefore we have to delay any flushing that
908 * happens before a renderpass. So we have to have two copies of the flush
909 * state, one for intra-renderpass flushes (i.e. renderpass dependencies)
910 * and one for outside a renderpass.
912 struct tu_cache_state cache
;
913 struct tu_cache_state renderpass_cache
;
915 enum tu_cmd_ccu_state ccu_state
;
917 const struct tu_render_pass
*pass
;
918 const struct tu_subpass
*subpass
;
919 const struct tu_framebuffer
*framebuffer
;
920 VkRect2D render_area
;
922 struct tu_cs_entry tile_store_ib
;
926 bool has_subpass_predication
;
927 bool predication_active
;
932 struct vk_object_base base
;
934 VkAllocationCallbacks alloc
;
935 struct list_head cmd_buffers
;
936 struct list_head free_cmd_buffers
;
937 uint32_t queue_family_index
;
940 struct tu_cmd_buffer_upload
945 struct list_head list
;
948 enum tu_cmd_buffer_status
950 TU_CMD_BUFFER_STATUS_INVALID
,
951 TU_CMD_BUFFER_STATUS_INITIAL
,
952 TU_CMD_BUFFER_STATUS_RECORDING
,
953 TU_CMD_BUFFER_STATUS_EXECUTABLE
,
954 TU_CMD_BUFFER_STATUS_PENDING
,
957 #ifndef MSM_SUBMIT_BO_READ
958 #define MSM_SUBMIT_BO_READ 0x0001
959 #define MSM_SUBMIT_BO_WRITE 0x0002
960 #define MSM_SUBMIT_BO_DUMP 0x0004
962 struct drm_msm_gem_submit_bo
{
963 uint32_t flags
; /* in, mask of MSM_SUBMIT_BO_x */
964 uint32_t handle
; /* in, GEM handle */
965 uint64_t presumed
; /* in/out, presumed buffer address */
973 struct drm_msm_gem_submit_bo
*bo_infos
;
976 #define TU_BO_LIST_FAILED (~0)
979 tu_bo_list_init(struct tu_bo_list
*list
);
981 tu_bo_list_destroy(struct tu_bo_list
*list
);
983 tu_bo_list_reset(struct tu_bo_list
*list
);
985 tu_bo_list_add(struct tu_bo_list
*list
,
986 const struct tu_bo
*bo
,
989 tu_bo_list_merge(struct tu_bo_list
*list
, const struct tu_bo_list
*other
);
993 struct vk_object_base base
;
995 struct tu_device
*device
;
997 struct tu_cmd_pool
*pool
;
998 struct list_head pool_link
;
1000 VkCommandBufferUsageFlags usage_flags
;
1001 VkCommandBufferLevel level
;
1002 enum tu_cmd_buffer_status status
;
1004 struct tu_cmd_state state
;
1005 struct tu_vertex_binding vertex_bindings
[MAX_VBS
];
1006 uint32_t vertex_bindings_set
;
1007 uint32_t queue_family_index
;
1009 uint32_t push_constants
[MAX_PUSH_CONSTANTS_SIZE
/ 4];
1010 VkShaderStageFlags push_constant_stages
;
1011 struct tu_descriptor_set meta_push_descriptors
;
1013 struct tu_descriptor_state descriptors
[MAX_BIND_POINTS
];
1015 struct tu_cmd_buffer_upload upload
;
1017 VkResult record_result
;
1019 struct tu_bo_list bo_list
;
1021 struct tu_cs draw_cs
;
1022 struct tu_cs draw_epilogue_cs
;
1023 struct tu_cs sub_cs
;
1025 uint32_t vsc_draw_strm_pitch
;
1026 uint32_t vsc_prim_strm_pitch
;
1029 /* Temporary struct for tracking a register state to be written, used by
1030 * a6xx-pack.h and tu_cs_emit_regs()
1032 struct tu_reg_value
{
1043 void tu_emit_cache_flush_renderpass(struct tu_cmd_buffer
*cmd_buffer
,
1046 void tu_emit_cache_flush_ccu(struct tu_cmd_buffer
*cmd_buffer
,
1048 enum tu_cmd_ccu_state ccu_state
);
1051 tu6_emit_event_write(struct tu_cmd_buffer
*cmd
,
1053 enum vgt_event_type event
);
1055 static inline struct tu_descriptor_state
*
1056 tu_get_descriptors_state(struct tu_cmd_buffer
*cmd_buffer
,
1057 VkPipelineBindPoint bind_point
)
1059 return &cmd_buffer
->descriptors
[bind_point
];
1064 struct vk_object_base base
;
1068 struct tu_shader_module
1070 struct vk_object_base base
;
1072 unsigned char sha1
[20];
1075 const uint32_t *code
[0];
1078 struct tu_push_constant_range
1086 struct ir3_shader
*ir3_shader
;
1088 struct tu_push_constant_range push_consts
;
1089 uint8_t active_desc_sets
;
1093 tu_shader_create(struct tu_device
*dev
,
1094 gl_shader_stage stage
,
1095 const VkPipelineShaderStageCreateInfo
*stage_info
,
1096 struct tu_pipeline_layout
*layout
,
1097 const VkAllocationCallbacks
*alloc
);
1100 tu_shader_destroy(struct tu_device
*dev
,
1101 struct tu_shader
*shader
,
1102 const VkAllocationCallbacks
*alloc
);
1104 struct tu_program_descriptor_linkage
1106 struct ir3_const_state const_state
;
1110 struct tu_push_constant_range push_consts
;
1115 struct vk_object_base base
;
1119 struct tu_pipeline_layout
*layout
;
1121 bool need_indirect_descriptor_sets
;
1122 VkShaderStageFlags active_stages
;
1123 uint32_t active_desc_sets
;
1125 /* mask of enabled dynamic states
1126 * if BIT(i) is set, pipeline->dynamic_state[i] is *NOT* used
1128 uint32_t dynamic_state_mask
;
1129 struct tu_draw_state dynamic_state
[TU_DYNAMIC_STATE_COUNT
];
1131 /* gras_su_cntl without line width, used for dynamic line width state */
1132 uint32_t gras_su_cntl
;
1134 /* draw states for the pipeline */
1135 struct tu_draw_state load_state
, rast_state
, ds_state
, blend_state
;
1139 struct tu_draw_state state
;
1140 struct tu_draw_state binning_state
;
1142 struct tu_program_descriptor_linkage link
[MESA_SHADER_STAGES
];
1147 struct tu_draw_state state
;
1148 struct tu_draw_state binning_state
;
1149 uint32_t bindings_used
;
1154 enum pc_di_primtype primtype
;
1155 bool primitive_restart
;
1160 uint32_t patch_type
;
1161 uint32_t param_stride
;
1162 uint32_t hs_bo_regid
;
1163 uint32_t ds_bo_regid
;
1164 bool upper_left_domain_origin
;
1169 uint32_t local_size
[3];
1174 tu6_emit_viewport(struct tu_cs
*cs
, const VkViewport
*viewport
);
1177 tu6_emit_scissor(struct tu_cs
*cs
, const VkRect2D
*scissor
);
1180 tu6_emit_sample_locations(struct tu_cs
*cs
, const VkSampleLocationsInfoEXT
*samp_loc
);
1183 tu6_emit_depth_bias(struct tu_cs
*cs
,
1184 float constant_factor
,
1186 float slope_factor
);
1188 void tu6_emit_msaa(struct tu_cs
*cs
, VkSampleCountFlagBits samples
);
1190 void tu6_emit_window_scissor(struct tu_cs
*cs
, uint32_t x1
, uint32_t y1
, uint32_t x2
, uint32_t y2
);
1192 void tu6_emit_window_offset(struct tu_cs
*cs
, uint32_t x1
, uint32_t y1
);
1195 tu6_emit_xs_config(struct tu_cs
*cs
,
1196 gl_shader_stage stage
,
1197 const struct ir3_shader_variant
*xs
,
1198 uint64_t binary_iova
);
1201 tu6_emit_vpc(struct tu_cs
*cs
,
1202 const struct ir3_shader_variant
*vs
,
1203 const struct ir3_shader_variant
*hs
,
1204 const struct ir3_shader_variant
*ds
,
1205 const struct ir3_shader_variant
*gs
,
1206 const struct ir3_shader_variant
*fs
,
1207 uint32_t patch_control_points
,
1208 bool vshs_workgroup
);
1211 tu6_emit_fs_inputs(struct tu_cs
*cs
, const struct ir3_shader_variant
*fs
);
1213 struct tu_image_view
;
1216 tu_resolve_sysmem(struct tu_cmd_buffer
*cmd
,
1218 struct tu_image_view
*src
,
1219 struct tu_image_view
*dst
,
1221 const VkRect2D
*rect
);
1224 tu_clear_sysmem_attachment(struct tu_cmd_buffer
*cmd
,
1227 const VkRenderPassBeginInfo
*info
);
1230 tu_clear_gmem_attachment(struct tu_cmd_buffer
*cmd
,
1233 const VkRenderPassBeginInfo
*info
);
1236 tu_load_gmem_attachment(struct tu_cmd_buffer
*cmd
,
1241 /* expose this function to be able to emit load without checking LOAD_OP */
1243 tu_emit_load_gmem_attachment(struct tu_cmd_buffer
*cmd
, struct tu_cs
*cs
, uint32_t a
);
1245 /* note: gmem store can also resolve */
1247 tu_store_gmem_attachment(struct tu_cmd_buffer
*cmd
,
1252 enum tu_supported_formats
{
1258 struct tu_native_format
1260 enum a6xx_format fmt
: 8;
1261 enum a3xx_color_swap swap
: 8;
1262 enum a6xx_tile_mode tile_mode
: 8;
1263 enum tu_supported_formats supported
: 8;
1266 struct tu_native_format
tu6_format_vtx(VkFormat format
);
1267 struct tu_native_format
tu6_format_color(VkFormat format
, enum a6xx_tile_mode tile_mode
);
1268 struct tu_native_format
tu6_format_texture(VkFormat format
, enum a6xx_tile_mode tile_mode
);
1270 static inline enum a6xx_format
1271 tu6_base_format(VkFormat format
)
1273 /* note: tu6_format_color doesn't care about tiling for .fmt field */
1274 return tu6_format_color(format
, TILE6_LINEAR
).fmt
;
1279 struct vk_object_base base
;
1282 /* The original VkFormat provided by the client. This may not match any
1283 * of the actual surface formats.
1286 VkImageAspectFlags aspects
;
1287 VkImageUsageFlags usage
; /**< Superset of VkImageCreateInfo::usage. */
1288 VkImageTiling tiling
; /** VkImageCreateInfo::tiling */
1289 VkImageCreateFlags flags
; /** VkImageCreateInfo::flags */
1291 uint32_t level_count
;
1292 uint32_t layer_count
;
1293 VkSampleCountFlagBits samples
;
1295 struct fdl_layout layout
[3];
1296 uint32_t total_size
;
1298 unsigned queue_family_mask
;
1302 /* For VK_ANDROID_native_buffer, the WSI image owns the memory, */
1303 VkDeviceMemory owned_memory
;
1305 /* Set when bound */
1307 VkDeviceSize bo_offset
;
1310 static inline uint32_t
1311 tu_get_layerCount(const struct tu_image
*image
,
1312 const VkImageSubresourceRange
*range
)
1314 return range
->layerCount
== VK_REMAINING_ARRAY_LAYERS
1315 ? image
->layer_count
- range
->baseArrayLayer
1316 : range
->layerCount
;
1319 static inline uint32_t
1320 tu_get_levelCount(const struct tu_image
*image
,
1321 const VkImageSubresourceRange
*range
)
1323 return range
->levelCount
== VK_REMAINING_MIP_LEVELS
1324 ? image
->level_count
- range
->baseMipLevel
1325 : range
->levelCount
;
1328 struct tu_image_view
1330 struct vk_object_base base
;
1332 struct tu_image
*image
; /**< VkImageViewCreateInfo::image */
1336 uint32_t layer_size
;
1337 uint32_t ubwc_layer_size
;
1339 /* used to determine if fast gmem store path can be used */
1345 uint32_t descriptor
[A6XX_TEX_CONST_DWORDS
];
1347 /* Descriptor for use as a storage image as opposed to a sampled image.
1348 * This has a few differences for cube maps (e.g. type).
1350 uint32_t storage_descriptor
[A6XX_TEX_CONST_DWORDS
];
1352 /* pre-filled register values */
1354 uint32_t FLAG_BUFFER_PITCH
;
1356 uint32_t RB_MRT_BUF_INFO
;
1357 uint32_t SP_FS_MRT_REG
;
1359 uint32_t SP_PS_2D_SRC_INFO
;
1360 uint32_t SP_PS_2D_SRC_SIZE
;
1362 uint32_t RB_2D_DST_INFO
;
1364 uint32_t RB_BLIT_DST_INFO
;
1366 /* for d32s8 separate stencil */
1367 uint64_t stencil_base_addr
;
1368 uint32_t stencil_layer_size
;
1369 uint32_t stencil_PITCH
;
1372 struct tu_sampler_ycbcr_conversion
{
1373 struct vk_object_base base
;
1376 VkSamplerYcbcrModelConversion ycbcr_model
;
1377 VkSamplerYcbcrRange ycbcr_range
;
1378 VkComponentMapping components
;
1379 VkChromaLocation chroma_offsets
[2];
1380 VkFilter chroma_filter
;
1384 struct vk_object_base base
;
1386 uint32_t descriptor
[A6XX_TEX_SAMP_DWORDS
];
1387 struct tu_sampler_ycbcr_conversion
*ycbcr_sampler
;
1391 tu_cs_image_ref(struct tu_cs
*cs
, const struct tu_image_view
*iview
, uint32_t layer
);
1394 tu_cs_image_ref_2d(struct tu_cs
*cs
, const struct tu_image_view
*iview
, uint32_t layer
, bool src
);
1397 tu_cs_image_flag_ref(struct tu_cs
*cs
, const struct tu_image_view
*iview
, uint32_t layer
);
1400 tu_cs_image_stencil_ref(struct tu_cs
*cs
, const struct tu_image_view
*iview
, uint32_t layer
);
1402 #define tu_image_view_stencil(iview, x) \
1403 ((iview->x & ~A6XX_##x##_COLOR_FORMAT__MASK) | A6XX_##x##_COLOR_FORMAT(FMT6_8_UINT))
1406 tu_image_create(VkDevice _device
,
1407 const VkImageCreateInfo
*pCreateInfo
,
1408 const VkAllocationCallbacks
*alloc
,
1411 const VkSubresourceLayout
*plane_layouts
);
1414 tu_image_from_gralloc(VkDevice device_h
,
1415 const VkImageCreateInfo
*base_info
,
1416 const VkNativeBufferANDROID
*gralloc_info
,
1417 const VkAllocationCallbacks
*alloc
,
1418 VkImage
*out_image_h
);
1421 tu_image_view_init(struct tu_image_view
*iview
,
1422 const VkImageViewCreateInfo
*pCreateInfo
,
1423 bool limited_z24s8
);
1425 struct tu_buffer_view
1427 struct vk_object_base base
;
1429 uint32_t descriptor
[A6XX_TEX_CONST_DWORDS
];
1431 struct tu_buffer
*buffer
;
1434 tu_buffer_view_init(struct tu_buffer_view
*view
,
1435 struct tu_device
*device
,
1436 const VkBufferViewCreateInfo
*pCreateInfo
);
1438 struct tu_attachment_info
1440 struct tu_image_view
*attachment
;
1443 struct tu_framebuffer
1445 struct vk_object_base base
;
1451 /* size of the first tile */
1453 /* number of tiles */
1454 VkExtent2D tile_count
;
1456 /* size of the first VSC pipe */
1458 /* number of VSC pipes */
1459 VkExtent2D pipe_count
;
1461 /* pipe register values */
1462 uint32_t pipe_config
[MAX_VSC_PIPES
];
1463 uint32_t pipe_sizes
[MAX_VSC_PIPES
];
1465 uint32_t attachment_count
;
1466 struct tu_attachment_info attachments
[0];
1470 tu_framebuffer_tiling_config(struct tu_framebuffer
*fb
,
1471 const struct tu_device
*device
,
1472 const struct tu_render_pass
*pass
);
1474 struct tu_subpass_barrier
{
1475 VkPipelineStageFlags src_stage_mask
;
1476 VkAccessFlags src_access_mask
;
1477 VkAccessFlags dst_access_mask
;
1478 bool incoherent_ccu_color
, incoherent_ccu_depth
;
1481 struct tu_subpass_attachment
1483 uint32_t attachment
;
1488 uint32_t input_count
;
1489 uint32_t color_count
;
1490 struct tu_subpass_attachment
*input_attachments
;
1491 struct tu_subpass_attachment
*color_attachments
;
1492 struct tu_subpass_attachment
*resolve_attachments
;
1493 struct tu_subpass_attachment depth_stencil_attachment
;
1495 VkSampleCountFlagBits samples
;
1499 struct tu_subpass_barrier start_barrier
;
1502 struct tu_render_pass_attachment
1507 VkImageAspectFlags clear_mask
;
1510 int32_t gmem_offset
;
1511 /* for D32S8 separate stencil: */
1514 int32_t gmem_offset_stencil
;
1517 struct tu_render_pass
1519 struct vk_object_base base
;
1521 uint32_t attachment_count
;
1522 uint32_t subpass_count
;
1523 uint32_t gmem_pixels
;
1524 uint32_t tile_align_w
;
1525 struct tu_subpass_attachment
*subpass_attachments
;
1526 struct tu_render_pass_attachment
*attachments
;
1527 struct tu_subpass_barrier end_barrier
;
1528 struct tu_subpass subpasses
[0];
1531 struct tu_query_pool
1533 struct vk_object_base base
;
1538 uint32_t pipeline_statistics
;
1542 enum tu_semaphore_kind
1545 TU_SEMAPHORE_SYNCOBJ
,
1548 struct tu_semaphore_part
1550 enum tu_semaphore_kind kind
;
1558 struct vk_object_base base
;
1560 struct tu_semaphore_part permanent
;
1561 struct tu_semaphore_part temporary
;
1565 tu_set_descriptor_set(struct tu_cmd_buffer
*cmd_buffer
,
1566 VkPipelineBindPoint bind_point
,
1567 struct tu_descriptor_set
*set
,
1571 tu_update_descriptor_sets(struct tu_device
*device
,
1572 struct tu_cmd_buffer
*cmd_buffer
,
1573 VkDescriptorSet overrideSet
,
1574 uint32_t descriptorWriteCount
,
1575 const VkWriteDescriptorSet
*pDescriptorWrites
,
1576 uint32_t descriptorCopyCount
,
1577 const VkCopyDescriptorSet
*pDescriptorCopies
);
1580 tu_update_descriptor_set_with_template(
1581 struct tu_device
*device
,
1582 struct tu_cmd_buffer
*cmd_buffer
,
1583 struct tu_descriptor_set
*set
,
1584 VkDescriptorUpdateTemplate descriptorUpdateTemplate
,
1588 tu_physical_device_init(struct tu_physical_device
*device
,
1589 struct tu_instance
*instance
);
1591 tu_enumerate_devices(struct tu_instance
*instance
);
1594 tu_drm_submitqueue_new(const struct tu_device
*dev
,
1596 uint32_t *queue_id
);
1599 tu_drm_submitqueue_close(const struct tu_device
*dev
, uint32_t queue_id
);
1601 #define TU_DEFINE_HANDLE_CASTS(__tu_type, __VkType) \
1603 static inline struct __tu_type *__tu_type##_from_handle(__VkType _handle) \
1605 return (struct __tu_type *) _handle; \
1608 static inline __VkType __tu_type##_to_handle(struct __tu_type *_obj) \
1610 return (__VkType) _obj; \
1613 #define TU_DEFINE_NONDISP_HANDLE_CASTS(__tu_type, __VkType) \
1615 static inline struct __tu_type *__tu_type##_from_handle(__VkType _handle) \
1617 return (struct __tu_type *) (uintptr_t) _handle; \
1620 static inline __VkType __tu_type##_to_handle(struct __tu_type *_obj) \
1622 return (__VkType)(uintptr_t) _obj; \
1625 #define TU_FROM_HANDLE(__tu_type, __name, __handle) \
1626 struct __tu_type *__name = __tu_type##_from_handle(__handle)
1628 TU_DEFINE_HANDLE_CASTS(tu_cmd_buffer
, VkCommandBuffer
)
1629 TU_DEFINE_HANDLE_CASTS(tu_device
, VkDevice
)
1630 TU_DEFINE_HANDLE_CASTS(tu_instance
, VkInstance
)
1631 TU_DEFINE_HANDLE_CASTS(tu_physical_device
, VkPhysicalDevice
)
1632 TU_DEFINE_HANDLE_CASTS(tu_queue
, VkQueue
)
1634 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_cmd_pool
, VkCommandPool
)
1635 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_buffer
, VkBuffer
)
1636 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_buffer_view
, VkBufferView
)
1637 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_pool
, VkDescriptorPool
)
1638 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_set
, VkDescriptorSet
)
1639 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_set_layout
,
1640 VkDescriptorSetLayout
)
1641 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_update_template
,
1642 VkDescriptorUpdateTemplate
)
1643 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_device_memory
, VkDeviceMemory
)
1644 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_fence
, VkFence
)
1645 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_event
, VkEvent
)
1646 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_framebuffer
, VkFramebuffer
)
1647 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_image
, VkImage
)
1648 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_image_view
, VkImageView
);
1649 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_pipeline_cache
, VkPipelineCache
)
1650 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_pipeline
, VkPipeline
)
1651 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_pipeline_layout
, VkPipelineLayout
)
1652 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_query_pool
, VkQueryPool
)
1653 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_render_pass
, VkRenderPass
)
1654 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_sampler
, VkSampler
)
1655 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_sampler_ycbcr_conversion
, VkSamplerYcbcrConversion
)
1656 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_shader_module
, VkShaderModule
)
1657 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_semaphore
, VkSemaphore
)
1659 #endif /* TU_PRIVATE_H */