turnip: Use the common base object type and struct.
[mesa.git] / src / freedreno / vulkan / tu_private.h
1 /*
2 * Copyright © 2016 Red Hat.
3 * Copyright © 2016 Bas Nieuwenhuizen
4 *
5 * based in part on anv driver which is:
6 * Copyright © 2015 Intel Corporation
7 *
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:
14 *
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
17 * Software.
18 *
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.
26 */
27
28 #ifndef TU_PRIVATE_H
29 #define TU_PRIVATE_H
30
31 #include <assert.h>
32 #include <pthread.h>
33 #include <stdbool.h>
34 #include <stdint.h>
35 #include <stdio.h>
36 #include <stdlib.h>
37 #include <string.h>
38 #ifdef HAVE_VALGRIND
39 #include <memcheck.h>
40 #include <valgrind.h>
41 #define VG(x) x
42 #else
43 #define VG(x) ((void)0)
44 #endif
45
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"
51 #include "vk_alloc.h"
52 #include "vk_object.h"
53 #include "vk_debug_report.h"
54 #include "wsi_common.h"
55
56 #include "drm-uapi/msm_drm.h"
57 #include "ir3/ir3_compiler.h"
58 #include "ir3/ir3_shader.h"
59
60 #include "adreno_common.xml.h"
61 #include "adreno_pm4.xml.h"
62 #include "a6xx.xml.h"
63 #include "fdl/freedreno_layout.h"
64
65 #include "tu_descriptor_set.h"
66 #include "tu_extensions.h"
67 #include "tu_util.h"
68
69 /* Pre-declarations needed for WSI entrypoints */
70 struct wl_surface;
71 struct wl_display;
72 typedef struct xcb_connection_t xcb_connection_t;
73 typedef uint32_t xcb_visualid_t;
74 typedef uint32_t xcb_window_t;
75
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>
80
81 #include "tu_entrypoints.h"
82
83 #include "vk_format.h"
84
85 #define MAX_VBS 32
86 #define MAX_VERTEX_ATTRIBS 32
87 #define MAX_RTS 8
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
99 #define MAX_VIEWS 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.
107 */
108 #define MAX_UNIFORM_BUFFER_RANGE 0x10000
109
110 #define A6XX_TEX_CONST_DWORDS 16
111 #define A6XX_TEX_SAMP_DWORDS 4
112
113 #define tu_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
114
115 static inline uint32_t
116 tu_minify(uint32_t n, uint32_t levels)
117 {
118 if (unlikely(n == 0))
119 return 0;
120 else
121 return MAX2(n >> levels, 1);
122 }
123
124 #define for_each_bit(b, dword) \
125 for (uint32_t __dword = (dword); \
126 (b) = __builtin_ffs(__dword) - 1, __dword; __dword &= ~(1 << (b)))
127
128 #define typed_memcpy(dest, src, count) \
129 ({ \
130 STATIC_ASSERT(sizeof(*src) == sizeof(*dest)); \
131 memcpy((dest), (src), (count) * sizeof(*(src))); \
132 })
133
134 #define COND(bool, val) ((bool) ? (val) : 0)
135 #define BIT(bit) (1u << (bit))
136
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.
140 */
141
142 struct tu_instance;
143
144 VkResult
145 __vk_errorf(struct tu_instance *instance,
146 VkResult error,
147 const char *file,
148 int line,
149 const char *format,
150 ...);
151
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__);
156
157 void
158 __tu_finishme(const char *file, int line, const char *format, ...)
159 tu_printflike(3, 4);
160 void
161 tu_loge(const char *format, ...) tu_printflike(1, 2);
162 void
163 tu_logi(const char *format, ...) tu_printflike(1, 2);
164
165 /**
166 * Print a FINISHME message, including its source location.
167 */
168 #define tu_finishme(format, ...) \
169 do { \
170 static bool reported = false; \
171 if (!reported) { \
172 __tu_finishme(__FILE__, __LINE__, format, ##__VA_ARGS__); \
173 reported = true; \
174 } \
175 } while (0)
176
177 #define tu_stub() \
178 do { \
179 tu_finishme("stub %s", __func__); \
180 } while (0)
181
182 void *
183 tu_lookup_entrypoint_unchecked(const char *name);
184 void *
185 tu_lookup_entrypoint_checked(
186 const char *name,
187 uint32_t core_version,
188 const struct tu_instance_extension_table *instance,
189 const struct tu_device_extension_table *device);
190
191 struct tu_physical_device
192 {
193 struct vk_object_base base;
194
195 struct tu_instance *instance;
196
197 char path[20];
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];
202
203 struct wsi_device wsi_device;
204
205 int local_fd;
206 int master_fd;
207
208 unsigned gpu_id;
209 uint32_t gmem_size;
210 uint64_t gmem_base;
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
219
220 struct {
221 uint32_t PC_UNKNOWN_9805;
222 uint32_t SP_UNKNOWN_A0F8;
223 } magic;
224
225 int msm_major_version;
226 int msm_minor_version;
227
228 /* This is the drivers on-disk cache used as a fallback as opposed to
229 * the pipeline cache defined by apps.
230 */
231 struct disk_cache *disk_cache;
232
233 struct tu_device_extension_table supported_extensions;
234 };
235
236 enum tu_debug_flags
237 {
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,
245 };
246
247 struct tu_instance
248 {
249 struct vk_object_base base;
250
251 VkAllocationCallbacks alloc;
252
253 uint32_t api_version;
254 int physical_device_count;
255 struct tu_physical_device physical_devices[TU_MAX_DRM_DEVICES];
256
257 enum tu_debug_flags debug_flags;
258
259 struct vk_debug_report_instance debug_report_callbacks;
260
261 struct tu_instance_extension_table enabled_extensions;
262 };
263
264 VkResult
265 tu_wsi_init(struct tu_physical_device *physical_device);
266 void
267 tu_wsi_finish(struct tu_physical_device *physical_device);
268
269 bool
270 tu_instance_extension_supported(const char *name);
271 uint32_t
272 tu_physical_device_api_version(struct tu_physical_device *dev);
273 bool
274 tu_physical_device_extension_supported(struct tu_physical_device *dev,
275 const char *name);
276
277 struct cache_entry;
278
279 struct tu_pipeline_cache
280 {
281 struct vk_object_base base;
282
283 struct tu_device *device;
284 pthread_mutex_t mutex;
285
286 uint32_t total_size;
287 uint32_t table_size;
288 uint32_t kernel_count;
289 struct cache_entry **hash_table;
290 bool modified;
291
292 VkAllocationCallbacks alloc;
293 };
294
295 struct tu_pipeline_key
296 {
297 };
298
299
300 /* queue types */
301 #define TU_QUEUE_GENERAL 0
302
303 #define TU_MAX_QUEUE_FAMILIES 1
304
305 struct tu_fence
306 {
307 struct vk_object_base base;
308 struct wsi_fence *fence_wsi;
309 bool signaled;
310 int fd;
311 };
312
313 void
314 tu_fence_init(struct tu_fence *fence, bool signaled);
315 void
316 tu_fence_finish(struct tu_fence *fence);
317 void
318 tu_fence_update_fd(struct tu_fence *fence, int fd);
319 void
320 tu_fence_copy(struct tu_fence *fence, const struct tu_fence *src);
321 void
322 tu_fence_signal(struct tu_fence *fence);
323 void
324 tu_fence_wait_idle(struct tu_fence *fence);
325
326 struct tu_queue
327 {
328 struct vk_object_base base;
329
330 struct tu_device *device;
331 uint32_t queue_family_index;
332 int queue_idx;
333 VkDeviceQueueCreateFlags flags;
334
335 uint32_t msm_queue_id;
336 struct tu_fence submit_fence;
337 };
338
339 struct tu_bo
340 {
341 uint32_t gem_handle;
342 uint64_t size;
343 uint64_t iova;
344 void *map;
345 };
346
347 enum global_shader {
348 GLOBAL_SH_VS,
349 GLOBAL_SH_VS_LAYER,
350 GLOBAL_SH_GS_LAYER,
351 GLOBAL_SH_FS_BLIT,
352 GLOBAL_SH_FS_CLEAR0,
353 GLOBAL_SH_FS_CLEAR_MAX = GLOBAL_SH_FS_CLEAR0 + MAX_RTS,
354 GLOBAL_SH_COUNT,
355 };
356
357 /* This struct defines the layout of the global_bo */
358 struct tu6_global
359 {
360 /* 6 bcolor_entry entries, one for each VK_BORDER_COLOR */
361 uint8_t border_color[128 * 6];
362
363 /* clear/blit shaders, all <= 16 instrs (16 instr = 1 instrlen unit) */
364 instr_t shaders[GLOBAL_SH_COUNT][16];
365
366 uint32_t seqno_dummy; /* dummy seqno for CP_EVENT_WRITE */
367 uint32_t _pad0;
368 volatile uint32_t vsc_draw_overflow;
369 uint32_t _pad1;
370 volatile uint32_t vsc_prim_overflow;
371 uint32_t _pad2[3];
372
373 /* scratch space for VPC_SO[i].FLUSH_BASE_LO/HI, start on 32 byte boundary. */
374 struct {
375 uint32_t offset;
376 uint32_t pad[7];
377 } flush_base[4];
378 };
379 #define gb_offset(member) offsetof(struct tu6_global, member)
380 #define global_iova(cmd, member) ((cmd)->device->global_bo.iova + gb_offset(member))
381
382 void tu_init_clear_blit_shaders(struct tu6_global *global);
383
384 /* extra space in vsc draw/prim streams */
385 #define VSC_PAD 0x40
386
387 struct tu_device
388 {
389 struct vk_device vk;
390 struct tu_instance *instance;
391
392 struct tu_queue *queues[TU_MAX_QUEUE_FAMILIES];
393 int queue_count[TU_MAX_QUEUE_FAMILIES];
394
395 struct tu_physical_device *physical_device;
396 int _lost;
397
398 struct ir3_compiler *compiler;
399
400 /* Backup in-memory cache to be used if the app doesn't provide one */
401 struct tu_pipeline_cache *mem_cache;
402
403 #define MIN_SCRATCH_BO_SIZE_LOG2 12 /* A page */
404
405 /* Currently the kernel driver uses a 32-bit GPU address space, but it
406 * should be impossible to go beyond 48 bits.
407 */
408 struct {
409 struct tu_bo bo;
410 mtx_t construct_mtx;
411 bool initialized;
412 } scratch_bos[48 - MIN_SCRATCH_BO_SIZE_LOG2];
413
414 struct tu_bo global_bo;
415
416 struct tu_device_extension_table enabled_extensions;
417
418 uint32_t vsc_draw_strm_pitch;
419 uint32_t vsc_prim_strm_pitch;
420 mtx_t vsc_pitch_mtx;
421 };
422
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__)
428
429 static inline bool
430 tu_device_is_lost(struct tu_device *device)
431 {
432 return unlikely(p_atomic_read(&device->_lost));
433 }
434
435 VkResult
436 tu_bo_init_new(struct tu_device *dev, struct tu_bo *bo, uint64_t size);
437 VkResult
438 tu_bo_init_dmabuf(struct tu_device *dev,
439 struct tu_bo *bo,
440 uint64_t size,
441 int fd);
442 int
443 tu_bo_export_dmabuf(struct tu_device *dev, struct tu_bo *bo);
444 void
445 tu_bo_finish(struct tu_device *dev, struct tu_bo *bo);
446 VkResult
447 tu_bo_map(struct tu_device *dev, struct tu_bo *bo);
448
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
453 * discard it.
454 */
455 VkResult
456 tu_get_scratch_bo(struct tu_device *dev, uint64_t size, struct tu_bo **bo);
457
458 struct tu_cs_entry
459 {
460 /* No ownership */
461 const struct tu_bo *bo;
462
463 uint32_t size;
464 uint32_t offset;
465 };
466
467 struct tu_cs_memory {
468 uint32_t *map;
469 uint64_t iova;
470 };
471
472 struct tu_draw_state {
473 uint64_t iova : 48;
474 uint32_t size : 16;
475 };
476
477 enum tu_dynamic_state
478 {
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,
482 };
483
484 enum tu_draw_state_group_id
485 {
486 TU_DRAW_STATE_PROGRAM,
487 TU_DRAW_STATE_PROGRAM_BINNING,
488 TU_DRAW_STATE_TESS,
489 TU_DRAW_STATE_VB,
490 TU_DRAW_STATE_VI,
491 TU_DRAW_STATE_VI_BINNING,
492 TU_DRAW_STATE_RAST,
493 TU_DRAW_STATE_DS,
494 TU_DRAW_STATE_BLEND,
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,
505
506 /* dynamic state related draw states */
507 TU_DRAW_STATE_DYNAMIC,
508 TU_DRAW_STATE_COUNT = TU_DRAW_STATE_DYNAMIC + TU_DYNAMIC_STATE_COUNT,
509 };
510
511 enum tu_cs_mode
512 {
513
514 /*
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.
518 *
519 * This mode may create multiple entries internally. The entries must be
520 * submitted together.
521 */
522 TU_CS_MODE_GROW,
523
524 /*
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
527 * effect on it.
528 *
529 * This mode does not create any entry or any BO.
530 */
531 TU_CS_MODE_EXTERNAL,
532
533 /*
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.
538 *
539 * This mode does not create any entry internally.
540 */
541 TU_CS_MODE_SUB_STREAM,
542 };
543
544 struct tu_cs
545 {
546 uint32_t *start;
547 uint32_t *cur;
548 uint32_t *reserved_end;
549 uint32_t *end;
550
551 struct tu_device *device;
552 enum tu_cs_mode mode;
553 uint32_t next_bo_size;
554
555 struct tu_cs_entry *entries;
556 uint32_t entry_count;
557 uint32_t entry_capacity;
558
559 struct tu_bo **bos;
560 uint32_t bo_count;
561 uint32_t bo_capacity;
562
563 /* state for cond_exec_start/cond_exec_end */
564 uint32_t cond_flags;
565 uint32_t *cond_dwords;
566 };
567
568 struct tu_device_memory
569 {
570 struct vk_object_base base;
571
572 struct tu_bo bo;
573 VkDeviceSize size;
574
575 /* for dedicated allocations */
576 struct tu_image *image;
577 struct tu_buffer *buffer;
578
579 uint32_t type_index;
580 void *map;
581 void *user_ptr;
582 };
583
584 struct tu_descriptor_range
585 {
586 uint64_t va;
587 uint32_t size;
588 };
589
590 struct tu_descriptor_set
591 {
592 struct vk_object_base base;
593
594 const struct tu_descriptor_set_layout *layout;
595 struct tu_descriptor_pool *pool;
596 uint32_t size;
597
598 uint64_t va;
599 uint32_t *mapped_ptr;
600
601 uint32_t *dynamic_descriptors;
602
603 struct tu_bo *buffers[0];
604 };
605
606 struct tu_push_descriptor_set
607 {
608 struct tu_descriptor_set set;
609 uint32_t capacity;
610 };
611
612 struct tu_descriptor_pool_entry
613 {
614 uint32_t offset;
615 uint32_t size;
616 struct tu_descriptor_set *set;
617 };
618
619 struct tu_descriptor_pool
620 {
621 struct vk_object_base base;
622
623 struct tu_bo bo;
624 uint64_t current_offset;
625 uint64_t size;
626
627 uint8_t *host_memory_base;
628 uint8_t *host_memory_ptr;
629 uint8_t *host_memory_end;
630
631 uint32_t entry_count;
632 uint32_t max_entry_count;
633 struct tu_descriptor_pool_entry entries[0];
634 };
635
636 struct tu_descriptor_update_template_entry
637 {
638 VkDescriptorType descriptor_type;
639
640 /* The number of descriptors to update */
641 uint32_t descriptor_count;
642
643 /* Into mapped_ptr or dynamic_descriptors, in units of the respective array
644 */
645 uint32_t dst_offset;
646
647 /* In dwords. Not valid/used for dynamic descriptors */
648 uint32_t dst_stride;
649
650 uint32_t buffer_offset;
651
652 /* Only valid for combined image samplers and samplers */
653 uint16_t has_sampler;
654
655 /* In bytes */
656 size_t src_offset;
657 size_t src_stride;
658
659 /* For push descriptors */
660 const uint32_t *immutable_samplers;
661 };
662
663 struct tu_descriptor_update_template
664 {
665 struct vk_object_base base;
666
667 uint32_t entry_count;
668 struct tu_descriptor_update_template_entry entry[0];
669 };
670
671 struct tu_buffer
672 {
673 struct vk_object_base base;
674
675 VkDeviceSize size;
676
677 VkBufferUsageFlags usage;
678 VkBufferCreateFlags flags;
679
680 struct tu_bo *bo;
681 VkDeviceSize bo_offset;
682 };
683
684 static inline uint64_t
685 tu_buffer_iova(struct tu_buffer *buffer)
686 {
687 return buffer->bo->iova + buffer->bo_offset;
688 }
689
690 struct tu_vertex_binding
691 {
692 struct tu_buffer *buffer;
693 VkDeviceSize offset;
694 };
695
696 const char *
697 tu_get_debug_option_name(int id);
698
699 const char *
700 tu_get_perftest_option_name(int id);
701
702 struct tu_descriptor_state
703 {
704 struct tu_descriptor_set *sets[MAX_SETS];
705 uint32_t dynamic_descriptors[MAX_DYNAMIC_BUFFERS * A6XX_TEX_CONST_DWORDS];
706 };
707
708 enum tu_cmd_dirty_bits
709 {
710 TU_CMD_DIRTY_COMPUTE_PIPELINE = 1 << 1,
711 TU_CMD_DIRTY_VERTEX_BUFFERS = 1 << 2,
712 TU_CMD_DIRTY_DESCRIPTOR_SETS = 1 << 3,
713 TU_CMD_DIRTY_COMPUTE_DESCRIPTOR_SETS = 1 << 4,
714 TU_CMD_DIRTY_SHADER_CONSTS = 1 << 5,
715 /* all draw states were disabled and need to be re-enabled: */
716 TU_CMD_DIRTY_DRAW_STATE = 1 << 7,
717 };
718
719 /* There are only three cache domains we have to care about: the CCU, or
720 * color cache unit, which is used for color and depth/stencil attachments
721 * and copy/blit destinations, and is split conceptually into color and depth,
722 * and the universal cache or UCHE which is used for pretty much everything
723 * else, except for the CP (uncached) and host. We need to flush whenever data
724 * crosses these boundaries.
725 */
726
727 enum tu_cmd_access_mask {
728 TU_ACCESS_UCHE_READ = 1 << 0,
729 TU_ACCESS_UCHE_WRITE = 1 << 1,
730 TU_ACCESS_CCU_COLOR_READ = 1 << 2,
731 TU_ACCESS_CCU_COLOR_WRITE = 1 << 3,
732 TU_ACCESS_CCU_DEPTH_READ = 1 << 4,
733 TU_ACCESS_CCU_DEPTH_WRITE = 1 << 5,
734
735 /* Experiments have shown that while it's safe to avoid flushing the CCU
736 * after each blit/renderpass, it's not safe to assume that subsequent
737 * lookups with a different attachment state will hit unflushed cache
738 * entries. That is, the CCU needs to be flushed and possibly invalidated
739 * when accessing memory with a different attachment state. Writing to an
740 * attachment under the following conditions after clearing using the
741 * normal 2d engine path is known to have issues:
742 *
743 * - It isn't the 0'th layer.
744 * - There are more than one attachment, and this isn't the 0'th attachment
745 * (this seems to also depend on the cpp of the attachments).
746 *
747 * Our best guess is that the layer/MRT state is used when computing
748 * the location of a cache entry in CCU, to avoid conflicts. We assume that
749 * any access in a renderpass after or before an access by a transfer needs
750 * a flush/invalidate, and use the _INCOHERENT variants to represent access
751 * by a transfer.
752 */
753 TU_ACCESS_CCU_COLOR_INCOHERENT_READ = 1 << 6,
754 TU_ACCESS_CCU_COLOR_INCOHERENT_WRITE = 1 << 7,
755 TU_ACCESS_CCU_DEPTH_INCOHERENT_READ = 1 << 8,
756 TU_ACCESS_CCU_DEPTH_INCOHERENT_WRITE = 1 << 9,
757
758 TU_ACCESS_SYSMEM_READ = 1 << 10,
759 TU_ACCESS_SYSMEM_WRITE = 1 << 11,
760
761 /* Set if a WFI is required due to data being read by the CP or the 2D
762 * engine.
763 */
764 TU_ACCESS_WFI_READ = 1 << 12,
765
766 TU_ACCESS_READ =
767 TU_ACCESS_UCHE_READ |
768 TU_ACCESS_CCU_COLOR_READ |
769 TU_ACCESS_CCU_DEPTH_READ |
770 TU_ACCESS_CCU_COLOR_INCOHERENT_READ |
771 TU_ACCESS_CCU_DEPTH_INCOHERENT_READ |
772 TU_ACCESS_SYSMEM_READ,
773
774 TU_ACCESS_WRITE =
775 TU_ACCESS_UCHE_WRITE |
776 TU_ACCESS_CCU_COLOR_WRITE |
777 TU_ACCESS_CCU_COLOR_INCOHERENT_WRITE |
778 TU_ACCESS_CCU_DEPTH_WRITE |
779 TU_ACCESS_CCU_DEPTH_INCOHERENT_WRITE |
780 TU_ACCESS_SYSMEM_WRITE,
781
782 TU_ACCESS_ALL =
783 TU_ACCESS_READ |
784 TU_ACCESS_WRITE,
785 };
786
787 enum tu_cmd_flush_bits {
788 TU_CMD_FLAG_CCU_FLUSH_DEPTH = 1 << 0,
789 TU_CMD_FLAG_CCU_FLUSH_COLOR = 1 << 1,
790 TU_CMD_FLAG_CCU_INVALIDATE_DEPTH = 1 << 2,
791 TU_CMD_FLAG_CCU_INVALIDATE_COLOR = 1 << 3,
792 TU_CMD_FLAG_CACHE_FLUSH = 1 << 4,
793 TU_CMD_FLAG_CACHE_INVALIDATE = 1 << 5,
794
795 TU_CMD_FLAG_ALL_FLUSH =
796 TU_CMD_FLAG_CCU_FLUSH_DEPTH |
797 TU_CMD_FLAG_CCU_FLUSH_COLOR |
798 TU_CMD_FLAG_CACHE_FLUSH,
799
800 TU_CMD_FLAG_ALL_INVALIDATE =
801 TU_CMD_FLAG_CCU_INVALIDATE_DEPTH |
802 TU_CMD_FLAG_CCU_INVALIDATE_COLOR |
803 TU_CMD_FLAG_CACHE_INVALIDATE,
804
805 TU_CMD_FLAG_WFI = 1 << 6,
806 };
807
808 /* Changing the CCU from sysmem mode to gmem mode or vice-versa is pretty
809 * heavy, involving a CCU cache flush/invalidate and a WFI in order to change
810 * which part of the gmem is used by the CCU. Here we keep track of what the
811 * state of the CCU.
812 */
813 enum tu_cmd_ccu_state {
814 TU_CMD_CCU_SYSMEM,
815 TU_CMD_CCU_GMEM,
816 TU_CMD_CCU_UNKNOWN,
817 };
818
819 struct tu_cache_state {
820 /* Caches which must be made available (flushed) eventually if there are
821 * any users outside that cache domain, and caches which must be
822 * invalidated eventually if there are any reads.
823 */
824 enum tu_cmd_flush_bits pending_flush_bits;
825 /* Pending flushes */
826 enum tu_cmd_flush_bits flush_bits;
827 };
828
829 struct tu_cmd_state
830 {
831 uint32_t dirty;
832
833 struct tu_pipeline *pipeline;
834 struct tu_pipeline *compute_pipeline;
835
836 /* Vertex buffers */
837 struct
838 {
839 struct tu_buffer *buffers[MAX_VBS];
840 VkDeviceSize offsets[MAX_VBS];
841 } vb;
842
843 /* for dynamic states that can't be emitted directly */
844 uint32_t dynamic_stencil_mask;
845 uint32_t dynamic_stencil_wrmask;
846 uint32_t dynamic_stencil_ref;
847 uint32_t dynamic_gras_su_cntl;
848
849 /* saved states to re-emit in TU_CMD_DIRTY_DRAW_STATE case */
850 struct tu_draw_state dynamic_state[TU_DYNAMIC_STATE_COUNT];
851 struct tu_cs_entry vertex_buffers_ib;
852 struct tu_cs_entry shader_const_ib[MESA_SHADER_STAGES];
853 struct tu_cs_entry desc_sets_ib, desc_sets_load_ib;
854 struct tu_cs_entry ia_gmem_ib, ia_sysmem_ib;
855
856 struct tu_draw_state vs_params;
857
858 /* Index buffer */
859 uint64_t index_va;
860 uint32_t max_index_count;
861 uint8_t index_size;
862
863 /* because streamout base has to be 32-byte aligned
864 * there is an extra offset to deal with when it is
865 * unaligned
866 */
867 uint8_t streamout_offset[IR3_MAX_SO_BUFFERS];
868
869 /* Renderpasses are tricky, because we may need to flush differently if
870 * using sysmem vs. gmem and therefore we have to delay any flushing that
871 * happens before a renderpass. So we have to have two copies of the flush
872 * state, one for intra-renderpass flushes (i.e. renderpass dependencies)
873 * and one for outside a renderpass.
874 */
875 struct tu_cache_state cache;
876 struct tu_cache_state renderpass_cache;
877
878 enum tu_cmd_ccu_state ccu_state;
879
880 const struct tu_render_pass *pass;
881 const struct tu_subpass *subpass;
882 const struct tu_framebuffer *framebuffer;
883 VkRect2D render_area;
884
885 struct tu_cs_entry tile_store_ib;
886
887 bool xfb_used;
888 };
889
890 struct tu_cmd_pool
891 {
892 struct vk_object_base base;
893
894 VkAllocationCallbacks alloc;
895 struct list_head cmd_buffers;
896 struct list_head free_cmd_buffers;
897 uint32_t queue_family_index;
898 };
899
900 struct tu_cmd_buffer_upload
901 {
902 uint8_t *map;
903 unsigned offset;
904 uint64_t size;
905 struct list_head list;
906 };
907
908 enum tu_cmd_buffer_status
909 {
910 TU_CMD_BUFFER_STATUS_INVALID,
911 TU_CMD_BUFFER_STATUS_INITIAL,
912 TU_CMD_BUFFER_STATUS_RECORDING,
913 TU_CMD_BUFFER_STATUS_EXECUTABLE,
914 TU_CMD_BUFFER_STATUS_PENDING,
915 };
916
917 struct tu_bo_list
918 {
919 uint32_t count;
920 uint32_t capacity;
921 struct drm_msm_gem_submit_bo *bo_infos;
922 };
923
924 #define TU_BO_LIST_FAILED (~0)
925
926 void
927 tu_bo_list_init(struct tu_bo_list *list);
928 void
929 tu_bo_list_destroy(struct tu_bo_list *list);
930 void
931 tu_bo_list_reset(struct tu_bo_list *list);
932 uint32_t
933 tu_bo_list_add(struct tu_bo_list *list,
934 const struct tu_bo *bo,
935 uint32_t flags);
936 VkResult
937 tu_bo_list_merge(struct tu_bo_list *list, const struct tu_bo_list *other);
938
939 struct tu_cmd_buffer
940 {
941 struct vk_object_base base;
942
943 struct tu_device *device;
944
945 struct tu_cmd_pool *pool;
946 struct list_head pool_link;
947
948 VkCommandBufferUsageFlags usage_flags;
949 VkCommandBufferLevel level;
950 enum tu_cmd_buffer_status status;
951
952 struct tu_cmd_state state;
953 struct tu_vertex_binding vertex_bindings[MAX_VBS];
954 uint32_t vertex_bindings_set;
955 uint32_t queue_family_index;
956
957 uint32_t push_constants[MAX_PUSH_CONSTANTS_SIZE / 4];
958 VkShaderStageFlags push_constant_stages;
959 struct tu_descriptor_set meta_push_descriptors;
960
961 struct tu_descriptor_state descriptors[MAX_BIND_POINTS];
962
963 struct tu_cmd_buffer_upload upload;
964
965 VkResult record_result;
966
967 struct tu_bo_list bo_list;
968 struct tu_cs cs;
969 struct tu_cs draw_cs;
970 struct tu_cs draw_epilogue_cs;
971 struct tu_cs sub_cs;
972
973 bool has_tess;
974
975 uint32_t vsc_draw_strm_pitch;
976 uint32_t vsc_prim_strm_pitch;
977 };
978
979 /* Temporary struct for tracking a register state to be written, used by
980 * a6xx-pack.h and tu_cs_emit_regs()
981 */
982 struct tu_reg_value {
983 uint32_t reg;
984 uint64_t value;
985 bool is_address;
986 struct tu_bo *bo;
987 bool bo_write;
988 uint32_t bo_offset;
989 uint32_t bo_shift;
990 };
991
992
993 void tu_emit_cache_flush_renderpass(struct tu_cmd_buffer *cmd_buffer,
994 struct tu_cs *cs);
995
996 void tu_emit_cache_flush_ccu(struct tu_cmd_buffer *cmd_buffer,
997 struct tu_cs *cs,
998 enum tu_cmd_ccu_state ccu_state);
999
1000 void
1001 tu6_emit_event_write(struct tu_cmd_buffer *cmd,
1002 struct tu_cs *cs,
1003 enum vgt_event_type event);
1004
1005 static inline struct tu_descriptor_state *
1006 tu_get_descriptors_state(struct tu_cmd_buffer *cmd_buffer,
1007 VkPipelineBindPoint bind_point)
1008 {
1009 return &cmd_buffer->descriptors[bind_point];
1010 }
1011
1012 struct tu_event
1013 {
1014 struct vk_object_base base;
1015 struct tu_bo bo;
1016 };
1017
1018 struct tu_shader_module
1019 {
1020 struct vk_object_base base;
1021
1022 unsigned char sha1[20];
1023
1024 uint32_t code_size;
1025 const uint32_t *code[0];
1026 };
1027
1028 struct tu_push_constant_range
1029 {
1030 uint32_t lo;
1031 uint32_t count;
1032 };
1033
1034 struct tu_shader
1035 {
1036 struct ir3_shader *ir3_shader;
1037
1038 struct tu_push_constant_range push_consts;
1039 uint8_t active_desc_sets;
1040 };
1041
1042 struct tu_shader *
1043 tu_shader_create(struct tu_device *dev,
1044 gl_shader_stage stage,
1045 const VkPipelineShaderStageCreateInfo *stage_info,
1046 struct tu_pipeline_layout *layout,
1047 const VkAllocationCallbacks *alloc);
1048
1049 void
1050 tu_shader_destroy(struct tu_device *dev,
1051 struct tu_shader *shader,
1052 const VkAllocationCallbacks *alloc);
1053
1054 struct tu_program_descriptor_linkage
1055 {
1056 struct ir3_const_state const_state;
1057
1058 uint32_t constlen;
1059
1060 struct tu_push_constant_range push_consts;
1061 };
1062
1063 struct tu_pipeline
1064 {
1065 struct vk_object_base base;
1066
1067 struct tu_cs cs;
1068
1069 struct tu_pipeline_layout *layout;
1070
1071 bool need_indirect_descriptor_sets;
1072 VkShaderStageFlags active_stages;
1073 uint32_t active_desc_sets;
1074
1075 /* mask of enabled dynamic states
1076 * if BIT(i) is set, pipeline->dynamic_state[i] is *NOT* used
1077 */
1078 uint32_t dynamic_state_mask;
1079 struct tu_draw_state dynamic_state[TU_DYNAMIC_STATE_COUNT];
1080
1081 /* gras_su_cntl without line width, used for dynamic line width state */
1082 uint32_t gras_su_cntl;
1083
1084 struct
1085 {
1086 struct tu_cs_entry state_ib;
1087 struct tu_cs_entry binning_state_ib;
1088
1089 struct tu_program_descriptor_linkage link[MESA_SHADER_STAGES];
1090 } program;
1091
1092 struct
1093 {
1094 struct tu_cs_entry state_ib;
1095 } load_state;
1096
1097 struct
1098 {
1099 struct tu_cs_entry state_ib;
1100 struct tu_cs_entry binning_state_ib;
1101 uint32_t bindings_used;
1102 } vi;
1103
1104 struct
1105 {
1106 enum pc_di_primtype primtype;
1107 bool primitive_restart;
1108 } ia;
1109
1110 struct
1111 {
1112 uint32_t patch_type;
1113 uint32_t param_stride;
1114 uint32_t hs_bo_regid;
1115 uint32_t ds_bo_regid;
1116 bool upper_left_domain_origin;
1117 } tess;
1118
1119 struct
1120 {
1121 struct tu_cs_entry state_ib;
1122 } rast;
1123
1124 struct
1125 {
1126 struct tu_cs_entry state_ib;
1127 } ds;
1128
1129 struct
1130 {
1131 struct tu_cs_entry state_ib;
1132 } blend;
1133
1134 struct
1135 {
1136 uint32_t local_size[3];
1137 } compute;
1138 };
1139
1140 void
1141 tu6_emit_viewport(struct tu_cs *cs, const VkViewport *viewport);
1142
1143 void
1144 tu6_emit_scissor(struct tu_cs *cs, const VkRect2D *scissor);
1145
1146 void
1147 tu6_emit_sample_locations(struct tu_cs *cs, const VkSampleLocationsInfoEXT *samp_loc);
1148
1149 void
1150 tu6_emit_depth_bias(struct tu_cs *cs,
1151 float constant_factor,
1152 float clamp,
1153 float slope_factor);
1154
1155 void tu6_emit_msaa(struct tu_cs *cs, VkSampleCountFlagBits samples);
1156
1157 void tu6_emit_window_scissor(struct tu_cs *cs, uint32_t x1, uint32_t y1, uint32_t x2, uint32_t y2);
1158
1159 void tu6_emit_window_offset(struct tu_cs *cs, uint32_t x1, uint32_t y1);
1160
1161 void
1162 tu6_emit_xs_config(struct tu_cs *cs,
1163 gl_shader_stage stage,
1164 const struct ir3_shader_variant *xs,
1165 uint64_t binary_iova);
1166
1167 void
1168 tu6_emit_vpc(struct tu_cs *cs,
1169 const struct ir3_shader_variant *vs,
1170 const struct ir3_shader_variant *hs,
1171 const struct ir3_shader_variant *ds,
1172 const struct ir3_shader_variant *gs,
1173 const struct ir3_shader_variant *fs);
1174
1175 void
1176 tu6_emit_fs_inputs(struct tu_cs *cs, const struct ir3_shader_variant *fs);
1177
1178 struct tu_image_view;
1179
1180 void
1181 tu_resolve_sysmem(struct tu_cmd_buffer *cmd,
1182 struct tu_cs *cs,
1183 struct tu_image_view *src,
1184 struct tu_image_view *dst,
1185 uint32_t layers,
1186 const VkRect2D *rect);
1187
1188 void
1189 tu_clear_sysmem_attachment(struct tu_cmd_buffer *cmd,
1190 struct tu_cs *cs,
1191 uint32_t a,
1192 const VkRenderPassBeginInfo *info);
1193
1194 void
1195 tu_clear_gmem_attachment(struct tu_cmd_buffer *cmd,
1196 struct tu_cs *cs,
1197 uint32_t a,
1198 const VkRenderPassBeginInfo *info);
1199
1200 void
1201 tu_load_gmem_attachment(struct tu_cmd_buffer *cmd,
1202 struct tu_cs *cs,
1203 uint32_t a,
1204 bool force_load);
1205
1206 /* expose this function to be able to emit load without checking LOAD_OP */
1207 void
1208 tu_emit_load_gmem_attachment(struct tu_cmd_buffer *cmd, struct tu_cs *cs, uint32_t a);
1209
1210 /* note: gmem store can also resolve */
1211 void
1212 tu_store_gmem_attachment(struct tu_cmd_buffer *cmd,
1213 struct tu_cs *cs,
1214 uint32_t a,
1215 uint32_t gmem_a);
1216
1217 enum tu_supported_formats {
1218 FMT_VERTEX = 1,
1219 FMT_TEXTURE = 2,
1220 FMT_COLOR = 4,
1221 };
1222
1223 struct tu_native_format
1224 {
1225 enum a6xx_format fmt : 8;
1226 enum a3xx_color_swap swap : 8;
1227 enum a6xx_tile_mode tile_mode : 8;
1228 enum tu_supported_formats supported : 8;
1229 };
1230
1231 struct tu_native_format tu6_format_vtx(VkFormat format);
1232 struct tu_native_format tu6_format_color(VkFormat format, enum a6xx_tile_mode tile_mode);
1233 struct tu_native_format tu6_format_texture(VkFormat format, enum a6xx_tile_mode tile_mode);
1234
1235 static inline enum a6xx_format
1236 tu6_base_format(VkFormat format)
1237 {
1238 /* note: tu6_format_color doesn't care about tiling for .fmt field */
1239 return tu6_format_color(format, TILE6_LINEAR).fmt;
1240 }
1241
1242 struct tu_image
1243 {
1244 struct vk_object_base base;
1245
1246 VkImageType type;
1247 /* The original VkFormat provided by the client. This may not match any
1248 * of the actual surface formats.
1249 */
1250 VkFormat vk_format;
1251 VkImageAspectFlags aspects;
1252 VkImageUsageFlags usage; /**< Superset of VkImageCreateInfo::usage. */
1253 VkImageTiling tiling; /** VkImageCreateInfo::tiling */
1254 VkImageCreateFlags flags; /** VkImageCreateInfo::flags */
1255 VkExtent3D extent;
1256 uint32_t level_count;
1257 uint32_t layer_count;
1258 VkSampleCountFlagBits samples;
1259
1260 struct fdl_layout layout[3];
1261 uint32_t total_size;
1262
1263 unsigned queue_family_mask;
1264 bool exclusive;
1265 bool shareable;
1266
1267 /* For VK_ANDROID_native_buffer, the WSI image owns the memory, */
1268 VkDeviceMemory owned_memory;
1269
1270 /* Set when bound */
1271 struct tu_bo *bo;
1272 VkDeviceSize bo_offset;
1273 };
1274
1275 static inline uint32_t
1276 tu_get_layerCount(const struct tu_image *image,
1277 const VkImageSubresourceRange *range)
1278 {
1279 return range->layerCount == VK_REMAINING_ARRAY_LAYERS
1280 ? image->layer_count - range->baseArrayLayer
1281 : range->layerCount;
1282 }
1283
1284 static inline uint32_t
1285 tu_get_levelCount(const struct tu_image *image,
1286 const VkImageSubresourceRange *range)
1287 {
1288 return range->levelCount == VK_REMAINING_MIP_LEVELS
1289 ? image->level_count - range->baseMipLevel
1290 : range->levelCount;
1291 }
1292
1293 struct tu_image_view
1294 {
1295 struct vk_object_base base;
1296
1297 struct tu_image *image; /**< VkImageViewCreateInfo::image */
1298
1299 uint64_t base_addr;
1300 uint64_t ubwc_addr;
1301 uint32_t layer_size;
1302 uint32_t ubwc_layer_size;
1303
1304 /* used to determine if fast gmem store path can be used */
1305 VkExtent2D extent;
1306 bool need_y2_align;
1307
1308 bool ubwc_enabled;
1309
1310 uint32_t descriptor[A6XX_TEX_CONST_DWORDS];
1311
1312 /* Descriptor for use as a storage image as opposed to a sampled image.
1313 * This has a few differences for cube maps (e.g. type).
1314 */
1315 uint32_t storage_descriptor[A6XX_TEX_CONST_DWORDS];
1316
1317 /* pre-filled register values */
1318 uint32_t PITCH;
1319 uint32_t FLAG_BUFFER_PITCH;
1320
1321 uint32_t RB_MRT_BUF_INFO;
1322 uint32_t SP_FS_MRT_REG;
1323
1324 uint32_t SP_PS_2D_SRC_INFO;
1325 uint32_t SP_PS_2D_SRC_SIZE;
1326
1327 uint32_t RB_2D_DST_INFO;
1328
1329 uint32_t RB_BLIT_DST_INFO;
1330 };
1331
1332 struct tu_sampler_ycbcr_conversion {
1333 struct vk_object_base base;
1334
1335 VkFormat format;
1336 VkSamplerYcbcrModelConversion ycbcr_model;
1337 VkSamplerYcbcrRange ycbcr_range;
1338 VkComponentMapping components;
1339 VkChromaLocation chroma_offsets[2];
1340 VkFilter chroma_filter;
1341 };
1342
1343 struct tu_sampler {
1344 struct vk_object_base base;
1345
1346 uint32_t descriptor[A6XX_TEX_SAMP_DWORDS];
1347 struct tu_sampler_ycbcr_conversion *ycbcr_sampler;
1348 };
1349
1350 void
1351 tu_cs_image_ref(struct tu_cs *cs, const struct tu_image_view *iview, uint32_t layer);
1352
1353 void
1354 tu_cs_image_ref_2d(struct tu_cs *cs, const struct tu_image_view *iview, uint32_t layer, bool src);
1355
1356 void
1357 tu_cs_image_flag_ref(struct tu_cs *cs, const struct tu_image_view *iview, uint32_t layer);
1358
1359 VkResult
1360 tu_image_create(VkDevice _device,
1361 const VkImageCreateInfo *pCreateInfo,
1362 const VkAllocationCallbacks *alloc,
1363 VkImage *pImage,
1364 uint64_t modifier,
1365 const VkSubresourceLayout *plane_layouts);
1366
1367 VkResult
1368 tu_image_from_gralloc(VkDevice device_h,
1369 const VkImageCreateInfo *base_info,
1370 const VkNativeBufferANDROID *gralloc_info,
1371 const VkAllocationCallbacks *alloc,
1372 VkImage *out_image_h);
1373
1374 void
1375 tu_image_view_init(struct tu_image_view *view,
1376 const VkImageViewCreateInfo *pCreateInfo);
1377
1378 struct tu_buffer_view
1379 {
1380 struct vk_object_base base;
1381
1382 uint32_t descriptor[A6XX_TEX_CONST_DWORDS];
1383
1384 struct tu_buffer *buffer;
1385 };
1386 void
1387 tu_buffer_view_init(struct tu_buffer_view *view,
1388 struct tu_device *device,
1389 const VkBufferViewCreateInfo *pCreateInfo);
1390
1391 struct tu_attachment_info
1392 {
1393 struct tu_image_view *attachment;
1394 };
1395
1396 struct tu_framebuffer
1397 {
1398 struct vk_object_base base;
1399
1400 uint32_t width;
1401 uint32_t height;
1402 uint32_t layers;
1403
1404 /* size of the first tile */
1405 VkExtent2D tile0;
1406 /* number of tiles */
1407 VkExtent2D tile_count;
1408
1409 /* size of the first VSC pipe */
1410 VkExtent2D pipe0;
1411 /* number of VSC pipes */
1412 VkExtent2D pipe_count;
1413
1414 /* pipe register values */
1415 uint32_t pipe_config[MAX_VSC_PIPES];
1416 uint32_t pipe_sizes[MAX_VSC_PIPES];
1417
1418 uint32_t attachment_count;
1419 struct tu_attachment_info attachments[0];
1420 };
1421
1422 void
1423 tu_framebuffer_tiling_config(struct tu_framebuffer *fb,
1424 const struct tu_device *device,
1425 const struct tu_render_pass *pass);
1426
1427 struct tu_subpass_barrier {
1428 VkPipelineStageFlags src_stage_mask;
1429 VkAccessFlags src_access_mask;
1430 VkAccessFlags dst_access_mask;
1431 bool incoherent_ccu_color, incoherent_ccu_depth;
1432 };
1433
1434 struct tu_subpass_attachment
1435 {
1436 uint32_t attachment;
1437 };
1438
1439 struct tu_subpass
1440 {
1441 uint32_t input_count;
1442 uint32_t color_count;
1443 struct tu_subpass_attachment *input_attachments;
1444 struct tu_subpass_attachment *color_attachments;
1445 struct tu_subpass_attachment *resolve_attachments;
1446 struct tu_subpass_attachment depth_stencil_attachment;
1447
1448 VkSampleCountFlagBits samples;
1449
1450 uint32_t srgb_cntl;
1451
1452 struct tu_subpass_barrier start_barrier;
1453 };
1454
1455 struct tu_render_pass_attachment
1456 {
1457 VkFormat format;
1458 uint32_t samples;
1459 uint32_t cpp;
1460 VkImageAspectFlags clear_mask;
1461 bool load;
1462 bool store;
1463 int32_t gmem_offset;
1464 };
1465
1466 struct tu_render_pass
1467 {
1468 struct vk_object_base base;
1469
1470 uint32_t attachment_count;
1471 uint32_t subpass_count;
1472 uint32_t gmem_pixels;
1473 uint32_t tile_align_w;
1474 struct tu_subpass_attachment *subpass_attachments;
1475 struct tu_render_pass_attachment *attachments;
1476 struct tu_subpass_barrier end_barrier;
1477 struct tu_subpass subpasses[0];
1478 };
1479
1480 struct tu_query_pool
1481 {
1482 struct vk_object_base base;
1483
1484 VkQueryType type;
1485 uint32_t stride;
1486 uint64_t size;
1487 uint32_t pipeline_statistics;
1488 struct tu_bo bo;
1489 };
1490
1491 enum tu_semaphore_kind
1492 {
1493 TU_SEMAPHORE_NONE,
1494 TU_SEMAPHORE_SYNCOBJ,
1495 };
1496
1497 struct tu_semaphore_part
1498 {
1499 enum tu_semaphore_kind kind;
1500 union {
1501 uint32_t syncobj;
1502 };
1503 };
1504
1505 struct tu_semaphore
1506 {
1507 struct vk_object_base base;
1508
1509 struct tu_semaphore_part permanent;
1510 struct tu_semaphore_part temporary;
1511 };
1512
1513 void
1514 tu_set_descriptor_set(struct tu_cmd_buffer *cmd_buffer,
1515 VkPipelineBindPoint bind_point,
1516 struct tu_descriptor_set *set,
1517 unsigned idx);
1518
1519 void
1520 tu_update_descriptor_sets(struct tu_device *device,
1521 struct tu_cmd_buffer *cmd_buffer,
1522 VkDescriptorSet overrideSet,
1523 uint32_t descriptorWriteCount,
1524 const VkWriteDescriptorSet *pDescriptorWrites,
1525 uint32_t descriptorCopyCount,
1526 const VkCopyDescriptorSet *pDescriptorCopies);
1527
1528 void
1529 tu_update_descriptor_set_with_template(
1530 struct tu_device *device,
1531 struct tu_cmd_buffer *cmd_buffer,
1532 struct tu_descriptor_set *set,
1533 VkDescriptorUpdateTemplate descriptorUpdateTemplate,
1534 const void *pData);
1535
1536 int
1537 tu_drm_get_gpu_id(const struct tu_physical_device *dev, uint32_t *id);
1538
1539 int
1540 tu_drm_get_gmem_size(const struct tu_physical_device *dev, uint32_t *size);
1541
1542 int
1543 tu_drm_get_gmem_base(const struct tu_physical_device *dev, uint64_t *base);
1544
1545 int
1546 tu_drm_submitqueue_new(const struct tu_device *dev,
1547 int priority,
1548 uint32_t *queue_id);
1549
1550 void
1551 tu_drm_submitqueue_close(const struct tu_device *dev, uint32_t queue_id);
1552
1553 uint32_t
1554 tu_gem_new(const struct tu_device *dev, uint64_t size, uint32_t flags);
1555 uint32_t
1556 tu_gem_import_dmabuf(const struct tu_device *dev,
1557 int prime_fd,
1558 uint64_t size);
1559 int
1560 tu_gem_export_dmabuf(const struct tu_device *dev, uint32_t gem_handle);
1561 void
1562 tu_gem_close(const struct tu_device *dev, uint32_t gem_handle);
1563 uint64_t
1564 tu_gem_info_offset(const struct tu_device *dev, uint32_t gem_handle);
1565 uint64_t
1566 tu_gem_info_iova(const struct tu_device *dev, uint32_t gem_handle);
1567
1568 #define TU_DEFINE_HANDLE_CASTS(__tu_type, __VkType) \
1569 \
1570 static inline struct __tu_type *__tu_type##_from_handle(__VkType _handle) \
1571 { \
1572 return (struct __tu_type *) _handle; \
1573 } \
1574 \
1575 static inline __VkType __tu_type##_to_handle(struct __tu_type *_obj) \
1576 { \
1577 return (__VkType) _obj; \
1578 }
1579
1580 #define TU_DEFINE_NONDISP_HANDLE_CASTS(__tu_type, __VkType) \
1581 \
1582 static inline struct __tu_type *__tu_type##_from_handle(__VkType _handle) \
1583 { \
1584 return (struct __tu_type *) (uintptr_t) _handle; \
1585 } \
1586 \
1587 static inline __VkType __tu_type##_to_handle(struct __tu_type *_obj) \
1588 { \
1589 return (__VkType)(uintptr_t) _obj; \
1590 }
1591
1592 #define TU_FROM_HANDLE(__tu_type, __name, __handle) \
1593 struct __tu_type *__name = __tu_type##_from_handle(__handle)
1594
1595 TU_DEFINE_HANDLE_CASTS(tu_cmd_buffer, VkCommandBuffer)
1596 TU_DEFINE_HANDLE_CASTS(tu_device, VkDevice)
1597 TU_DEFINE_HANDLE_CASTS(tu_instance, VkInstance)
1598 TU_DEFINE_HANDLE_CASTS(tu_physical_device, VkPhysicalDevice)
1599 TU_DEFINE_HANDLE_CASTS(tu_queue, VkQueue)
1600
1601 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_cmd_pool, VkCommandPool)
1602 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_buffer, VkBuffer)
1603 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_buffer_view, VkBufferView)
1604 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_pool, VkDescriptorPool)
1605 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_set, VkDescriptorSet)
1606 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_set_layout,
1607 VkDescriptorSetLayout)
1608 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_update_template,
1609 VkDescriptorUpdateTemplate)
1610 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_device_memory, VkDeviceMemory)
1611 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_fence, VkFence)
1612 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_event, VkEvent)
1613 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_framebuffer, VkFramebuffer)
1614 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_image, VkImage)
1615 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_image_view, VkImageView);
1616 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_pipeline_cache, VkPipelineCache)
1617 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_pipeline, VkPipeline)
1618 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_pipeline_layout, VkPipelineLayout)
1619 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_query_pool, VkQueryPool)
1620 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_render_pass, VkRenderPass)
1621 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_sampler, VkSampler)
1622 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_sampler_ycbcr_conversion, VkSamplerYcbcrConversion)
1623 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_shader_module, VkShaderModule)
1624 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_semaphore, VkSemaphore)
1625
1626 #endif /* TU_PRIVATE_H */