turnip: Support tess for draws
[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 "vk_alloc.h"
51 #include "vk_debug_report.h"
52 #include "wsi_common.h"
53
54 #include "drm-uapi/msm_drm.h"
55 #include "ir3/ir3_compiler.h"
56 #include "ir3/ir3_shader.h"
57
58 #include "adreno_common.xml.h"
59 #include "adreno_pm4.xml.h"
60 #include "a6xx.xml.h"
61 #include "fdl/freedreno_layout.h"
62
63 #include "tu_descriptor_set.h"
64 #include "tu_extensions.h"
65 #include "tu_util.h"
66
67 /* Pre-declarations needed for WSI entrypoints */
68 struct wl_surface;
69 struct wl_display;
70 typedef struct xcb_connection_t xcb_connection_t;
71 typedef uint32_t xcb_visualid_t;
72 typedef uint32_t xcb_window_t;
73
74 #include <vulkan/vk_android_native_buffer.h>
75 #include <vulkan/vk_icd.h>
76 #include <vulkan/vulkan.h>
77 #include <vulkan/vulkan_intel.h>
78
79 #include "tu_entrypoints.h"
80
81 #include "vk_format.h"
82
83 #define MAX_VBS 32
84 #define MAX_VERTEX_ATTRIBS 32
85 #define MAX_RTS 8
86 #define MAX_VSC_PIPES 32
87 #define MAX_VIEWPORTS 1
88 #define MAX_SCISSORS 16
89 #define MAX_DISCARD_RECTANGLES 4
90 #define MAX_PUSH_CONSTANTS_SIZE 128
91 #define MAX_PUSH_DESCRIPTORS 32
92 #define MAX_DYNAMIC_UNIFORM_BUFFERS 16
93 #define MAX_DYNAMIC_STORAGE_BUFFERS 8
94 #define MAX_DYNAMIC_BUFFERS \
95 (MAX_DYNAMIC_UNIFORM_BUFFERS + MAX_DYNAMIC_STORAGE_BUFFERS)
96 #define TU_MAX_DRM_DEVICES 8
97 #define MAX_VIEWS 8
98 #define MAX_BIND_POINTS 2 /* compute + graphics */
99 /* The Qualcomm driver exposes 0x20000058 */
100 #define MAX_STORAGE_BUFFER_RANGE 0x20000000
101 /* We use ldc for uniform buffer loads, just like the Qualcomm driver, so
102 * expose the same maximum range.
103 * TODO: The SIZE bitfield is 15 bits, and in 4-dword units, so the actual
104 * range might be higher.
105 */
106 #define MAX_UNIFORM_BUFFER_RANGE 0x10000
107
108 #define A6XX_TEX_CONST_DWORDS 16
109 #define A6XX_TEX_SAMP_DWORDS 4
110
111 #define tu_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
112
113 static inline uint32_t
114 tu_minify(uint32_t n, uint32_t levels)
115 {
116 if (unlikely(n == 0))
117 return 0;
118 else
119 return MAX2(n >> levels, 1);
120 }
121
122 #define for_each_bit(b, dword) \
123 for (uint32_t __dword = (dword); \
124 (b) = __builtin_ffs(__dword) - 1, __dword; __dword &= ~(1 << (b)))
125
126 #define typed_memcpy(dest, src, count) \
127 ({ \
128 STATIC_ASSERT(sizeof(*src) == sizeof(*dest)); \
129 memcpy((dest), (src), (count) * sizeof(*(src))); \
130 })
131
132 #define COND(bool, val) ((bool) ? (val) : 0)
133 #define BIT(bit) (1u << (bit))
134
135 /* Whenever we generate an error, pass it through this function. Useful for
136 * debugging, where we can break on it. Only call at error site, not when
137 * propagating errors. Might be useful to plug in a stack trace here.
138 */
139
140 struct tu_instance;
141
142 VkResult
143 __vk_errorf(struct tu_instance *instance,
144 VkResult error,
145 const char *file,
146 int line,
147 const char *format,
148 ...);
149
150 #define vk_error(instance, error) \
151 __vk_errorf(instance, error, __FILE__, __LINE__, NULL);
152 #define vk_errorf(instance, error, format, ...) \
153 __vk_errorf(instance, error, __FILE__, __LINE__, format, ##__VA_ARGS__);
154
155 void
156 __tu_finishme(const char *file, int line, const char *format, ...)
157 tu_printflike(3, 4);
158 void
159 tu_loge(const char *format, ...) tu_printflike(1, 2);
160 void
161 tu_logi(const char *format, ...) tu_printflike(1, 2);
162
163 /**
164 * Print a FINISHME message, including its source location.
165 */
166 #define tu_finishme(format, ...) \
167 do { \
168 static bool reported = false; \
169 if (!reported) { \
170 __tu_finishme(__FILE__, __LINE__, format, ##__VA_ARGS__); \
171 reported = true; \
172 } \
173 } while (0)
174
175 #define tu_stub() \
176 do { \
177 tu_finishme("stub %s", __func__); \
178 } while (0)
179
180 void *
181 tu_lookup_entrypoint_unchecked(const char *name);
182 void *
183 tu_lookup_entrypoint_checked(
184 const char *name,
185 uint32_t core_version,
186 const struct tu_instance_extension_table *instance,
187 const struct tu_device_extension_table *device);
188
189 struct tu_physical_device
190 {
191 VK_LOADER_DATA _loader_data;
192
193 struct tu_instance *instance;
194
195 char path[20];
196 char name[VK_MAX_PHYSICAL_DEVICE_NAME_SIZE];
197 uint8_t driver_uuid[VK_UUID_SIZE];
198 uint8_t device_uuid[VK_UUID_SIZE];
199 uint8_t cache_uuid[VK_UUID_SIZE];
200
201 struct wsi_device wsi_device;
202
203 int local_fd;
204 int master_fd;
205
206 unsigned gpu_id;
207 uint32_t gmem_size;
208 uint64_t gmem_base;
209 uint32_t ccu_offset_gmem;
210 uint32_t ccu_offset_bypass;
211 /* alignment for size of tiles */
212 uint32_t tile_align_w;
213 #define TILE_ALIGN_H 16
214 /* gmem store/load granularity */
215 #define GMEM_ALIGN_W 16
216 #define GMEM_ALIGN_H 4
217
218 struct {
219 uint32_t PC_UNKNOWN_9805;
220 uint32_t SP_UNKNOWN_A0F8;
221 } magic;
222
223 /* This is the drivers on-disk cache used as a fallback as opposed to
224 * the pipeline cache defined by apps.
225 */
226 struct disk_cache *disk_cache;
227
228 struct tu_device_extension_table supported_extensions;
229 };
230
231 enum tu_debug_flags
232 {
233 TU_DEBUG_STARTUP = 1 << 0,
234 TU_DEBUG_NIR = 1 << 1,
235 TU_DEBUG_IR3 = 1 << 2,
236 TU_DEBUG_NOBIN = 1 << 3,
237 TU_DEBUG_SYSMEM = 1 << 4,
238 TU_DEBUG_FORCEBIN = 1 << 5,
239 TU_DEBUG_NOUBWC = 1 << 6,
240 };
241
242 struct tu_instance
243 {
244 VK_LOADER_DATA _loader_data;
245
246 VkAllocationCallbacks alloc;
247
248 uint32_t api_version;
249 int physical_device_count;
250 struct tu_physical_device physical_devices[TU_MAX_DRM_DEVICES];
251
252 enum tu_debug_flags debug_flags;
253
254 struct vk_debug_report_instance debug_report_callbacks;
255
256 struct tu_instance_extension_table enabled_extensions;
257 };
258
259 VkResult
260 tu_wsi_init(struct tu_physical_device *physical_device);
261 void
262 tu_wsi_finish(struct tu_physical_device *physical_device);
263
264 bool
265 tu_instance_extension_supported(const char *name);
266 uint32_t
267 tu_physical_device_api_version(struct tu_physical_device *dev);
268 bool
269 tu_physical_device_extension_supported(struct tu_physical_device *dev,
270 const char *name);
271
272 struct cache_entry;
273
274 struct tu_pipeline_cache
275 {
276 struct tu_device *device;
277 pthread_mutex_t mutex;
278
279 uint32_t total_size;
280 uint32_t table_size;
281 uint32_t kernel_count;
282 struct cache_entry **hash_table;
283 bool modified;
284
285 VkAllocationCallbacks alloc;
286 };
287
288 struct tu_pipeline_key
289 {
290 };
291
292
293 /* queue types */
294 #define TU_QUEUE_GENERAL 0
295
296 #define TU_MAX_QUEUE_FAMILIES 1
297
298 struct tu_fence
299 {
300 struct wsi_fence *fence_wsi;
301 bool signaled;
302 int fd;
303 };
304
305 void
306 tu_fence_init(struct tu_fence *fence, bool signaled);
307 void
308 tu_fence_finish(struct tu_fence *fence);
309 void
310 tu_fence_update_fd(struct tu_fence *fence, int fd);
311 void
312 tu_fence_copy(struct tu_fence *fence, const struct tu_fence *src);
313 void
314 tu_fence_signal(struct tu_fence *fence);
315 void
316 tu_fence_wait_idle(struct tu_fence *fence);
317
318 struct tu_queue
319 {
320 VK_LOADER_DATA _loader_data;
321 struct tu_device *device;
322 uint32_t queue_family_index;
323 int queue_idx;
324 VkDeviceQueueCreateFlags flags;
325
326 uint32_t msm_queue_id;
327 struct tu_fence submit_fence;
328 };
329
330 struct tu_bo
331 {
332 uint32_t gem_handle;
333 uint64_t size;
334 uint64_t iova;
335 void *map;
336 };
337
338 struct tu_device
339 {
340 VK_LOADER_DATA _loader_data;
341
342 VkAllocationCallbacks alloc;
343
344 struct tu_instance *instance;
345
346 struct tu_queue *queues[TU_MAX_QUEUE_FAMILIES];
347 int queue_count[TU_MAX_QUEUE_FAMILIES];
348
349 struct tu_physical_device *physical_device;
350
351 struct ir3_compiler *compiler;
352
353 /* Backup in-memory cache to be used if the app doesn't provide one */
354 struct tu_pipeline_cache *mem_cache;
355
356 struct tu_bo vsc_draw_strm;
357 struct tu_bo vsc_prim_strm;
358 uint32_t vsc_draw_strm_pitch;
359 uint32_t vsc_prim_strm_pitch;
360
361 #define MIN_SCRATCH_BO_SIZE_LOG2 12 /* A page */
362
363 /* Currently the kernel driver uses a 32-bit GPU address space, but it
364 * should be impossible to go beyond 48 bits.
365 */
366 struct {
367 struct tu_bo bo;
368 mtx_t construct_mtx;
369 bool initialized;
370 } scratch_bos[48 - MIN_SCRATCH_BO_SIZE_LOG2];
371
372 struct tu_bo border_color;
373
374 struct tu_device_extension_table enabled_extensions;
375 };
376
377 VkResult
378 tu_bo_init_new(struct tu_device *dev, struct tu_bo *bo, uint64_t size);
379 VkResult
380 tu_bo_init_dmabuf(struct tu_device *dev,
381 struct tu_bo *bo,
382 uint64_t size,
383 int fd);
384 int
385 tu_bo_export_dmabuf(struct tu_device *dev, struct tu_bo *bo);
386 void
387 tu_bo_finish(struct tu_device *dev, struct tu_bo *bo);
388 VkResult
389 tu_bo_map(struct tu_device *dev, struct tu_bo *bo);
390
391 /* Get a scratch bo for use inside a command buffer. This will always return
392 * the same bo given the same size or similar sizes, so only one scratch bo
393 * can be used at the same time. It's meant for short-lived things where we
394 * need to write to some piece of memory, read from it, and then immediately
395 * discard it.
396 */
397 VkResult
398 tu_get_scratch_bo(struct tu_device *dev, uint64_t size, struct tu_bo **bo);
399
400 struct tu_cs_entry
401 {
402 /* No ownership */
403 const struct tu_bo *bo;
404
405 uint32_t size;
406 uint32_t offset;
407 };
408
409 struct ts_cs_memory {
410 uint32_t *map;
411 uint64_t iova;
412 };
413
414 struct tu_draw_state {
415 uint64_t iova : 48;
416 uint32_t size : 16;
417 };
418
419 enum tu_dynamic_state
420 {
421 /* re-use VK_DYNAMIC_STATE_ enums for non-extended dynamic states */
422 TU_DYNAMIC_STATE_SAMPLE_LOCATIONS = VK_DYNAMIC_STATE_STENCIL_REFERENCE + 1,
423 TU_DYNAMIC_STATE_COUNT,
424 };
425
426 enum tu_draw_state_group_id
427 {
428 TU_DRAW_STATE_PROGRAM,
429 TU_DRAW_STATE_PROGRAM_BINNING,
430 TU_DRAW_STATE_TESS,
431 TU_DRAW_STATE_VB,
432 TU_DRAW_STATE_VI,
433 TU_DRAW_STATE_VI_BINNING,
434 TU_DRAW_STATE_RAST,
435 TU_DRAW_STATE_DS,
436 TU_DRAW_STATE_BLEND,
437 TU_DRAW_STATE_VS_CONST,
438 TU_DRAW_STATE_HS_CONST,
439 TU_DRAW_STATE_DS_CONST,
440 TU_DRAW_STATE_GS_CONST,
441 TU_DRAW_STATE_FS_CONST,
442 TU_DRAW_STATE_DESC_SETS,
443 TU_DRAW_STATE_DESC_SETS_LOAD,
444 TU_DRAW_STATE_VS_PARAMS,
445 TU_DRAW_STATE_INPUT_ATTACHMENTS_GMEM,
446 TU_DRAW_STATE_INPUT_ATTACHMENTS_SYSMEM,
447
448 /* dynamic state related draw states */
449 TU_DRAW_STATE_DYNAMIC,
450 TU_DRAW_STATE_COUNT = TU_DRAW_STATE_DYNAMIC + TU_DYNAMIC_STATE_COUNT,
451 };
452
453 enum tu_cs_mode
454 {
455
456 /*
457 * A command stream in TU_CS_MODE_GROW mode grows automatically whenever it
458 * is full. tu_cs_begin must be called before command packet emission and
459 * tu_cs_end must be called after.
460 *
461 * This mode may create multiple entries internally. The entries must be
462 * submitted together.
463 */
464 TU_CS_MODE_GROW,
465
466 /*
467 * A command stream in TU_CS_MODE_EXTERNAL mode wraps an external,
468 * fixed-size buffer. tu_cs_begin and tu_cs_end are optional and have no
469 * effect on it.
470 *
471 * This mode does not create any entry or any BO.
472 */
473 TU_CS_MODE_EXTERNAL,
474
475 /*
476 * A command stream in TU_CS_MODE_SUB_STREAM mode does not support direct
477 * command packet emission. tu_cs_begin_sub_stream must be called to get a
478 * sub-stream to emit comamnd packets to. When done with the sub-stream,
479 * tu_cs_end_sub_stream must be called.
480 *
481 * This mode does not create any entry internally.
482 */
483 TU_CS_MODE_SUB_STREAM,
484 };
485
486 struct tu_cs
487 {
488 uint32_t *start;
489 uint32_t *cur;
490 uint32_t *reserved_end;
491 uint32_t *end;
492
493 struct tu_device *device;
494 enum tu_cs_mode mode;
495 uint32_t next_bo_size;
496
497 struct tu_cs_entry *entries;
498 uint32_t entry_count;
499 uint32_t entry_capacity;
500
501 struct tu_bo **bos;
502 uint32_t bo_count;
503 uint32_t bo_capacity;
504
505 /* state for cond_exec_start/cond_exec_end */
506 uint32_t cond_flags;
507 uint32_t *cond_dwords;
508 };
509
510 struct tu_device_memory
511 {
512 struct tu_bo bo;
513 VkDeviceSize size;
514
515 /* for dedicated allocations */
516 struct tu_image *image;
517 struct tu_buffer *buffer;
518
519 uint32_t type_index;
520 void *map;
521 void *user_ptr;
522 };
523
524 struct tu_descriptor_range
525 {
526 uint64_t va;
527 uint32_t size;
528 };
529
530 struct tu_descriptor_set
531 {
532 const struct tu_descriptor_set_layout *layout;
533 struct tu_descriptor_pool *pool;
534 uint32_t size;
535
536 uint64_t va;
537 uint32_t *mapped_ptr;
538
539 uint32_t *dynamic_descriptors;
540
541 struct tu_bo *buffers[0];
542 };
543
544 struct tu_push_descriptor_set
545 {
546 struct tu_descriptor_set set;
547 uint32_t capacity;
548 };
549
550 struct tu_descriptor_pool_entry
551 {
552 uint32_t offset;
553 uint32_t size;
554 struct tu_descriptor_set *set;
555 };
556
557 struct tu_descriptor_pool
558 {
559 struct tu_bo bo;
560 uint64_t current_offset;
561 uint64_t size;
562
563 uint8_t *host_memory_base;
564 uint8_t *host_memory_ptr;
565 uint8_t *host_memory_end;
566
567 uint32_t entry_count;
568 uint32_t max_entry_count;
569 struct tu_descriptor_pool_entry entries[0];
570 };
571
572 struct tu_descriptor_update_template_entry
573 {
574 VkDescriptorType descriptor_type;
575
576 /* The number of descriptors to update */
577 uint32_t descriptor_count;
578
579 /* Into mapped_ptr or dynamic_descriptors, in units of the respective array
580 */
581 uint32_t dst_offset;
582
583 /* In dwords. Not valid/used for dynamic descriptors */
584 uint32_t dst_stride;
585
586 uint32_t buffer_offset;
587
588 /* Only valid for combined image samplers and samplers */
589 uint16_t has_sampler;
590
591 /* In bytes */
592 size_t src_offset;
593 size_t src_stride;
594
595 /* For push descriptors */
596 const uint32_t *immutable_samplers;
597 };
598
599 struct tu_descriptor_update_template
600 {
601 uint32_t entry_count;
602 struct tu_descriptor_update_template_entry entry[0];
603 };
604
605 struct tu_buffer
606 {
607 VkDeviceSize size;
608
609 VkBufferUsageFlags usage;
610 VkBufferCreateFlags flags;
611
612 struct tu_bo *bo;
613 VkDeviceSize bo_offset;
614 };
615
616 static inline uint64_t
617 tu_buffer_iova(struct tu_buffer *buffer)
618 {
619 return buffer->bo->iova + buffer->bo_offset;
620 }
621
622 struct tu_vertex_binding
623 {
624 struct tu_buffer *buffer;
625 VkDeviceSize offset;
626 };
627
628 const char *
629 tu_get_debug_option_name(int id);
630
631 const char *
632 tu_get_perftest_option_name(int id);
633
634 struct tu_descriptor_state
635 {
636 struct tu_descriptor_set *sets[MAX_SETS];
637 uint32_t dynamic_descriptors[MAX_DYNAMIC_BUFFERS * A6XX_TEX_CONST_DWORDS];
638 };
639
640 struct tu_tile
641 {
642 uint8_t pipe;
643 uint8_t slot;
644 VkOffset2D begin;
645 VkOffset2D end;
646 };
647
648 struct tu_tiling_config
649 {
650 VkRect2D render_area;
651
652 /* position and size of the first tile */
653 VkRect2D tile0;
654 /* number of tiles */
655 VkExtent2D tile_count;
656
657 /* size of the first VSC pipe */
658 VkExtent2D pipe0;
659 /* number of VSC pipes */
660 VkExtent2D pipe_count;
661
662 /* pipe register values */
663 uint32_t pipe_config[MAX_VSC_PIPES];
664 uint32_t pipe_sizes[MAX_VSC_PIPES];
665
666 /* Whether sysmem rendering must be used */
667 bool force_sysmem;
668 };
669
670 enum tu_cmd_dirty_bits
671 {
672 TU_CMD_DIRTY_COMPUTE_PIPELINE = 1 << 1,
673 TU_CMD_DIRTY_VERTEX_BUFFERS = 1 << 2,
674 TU_CMD_DIRTY_DESCRIPTOR_SETS = 1 << 3,
675 TU_CMD_DIRTY_COMPUTE_DESCRIPTOR_SETS = 1 << 4,
676 TU_CMD_DIRTY_SHADER_CONSTS = 1 << 5,
677 TU_CMD_DIRTY_STREAMOUT_BUFFERS = 1 << 6,
678 /* all draw states were disabled and need to be re-enabled: */
679 TU_CMD_DIRTY_DRAW_STATE = 1 << 7,
680 };
681
682 struct tu_streamout_state {
683 uint16_t stride[IR3_MAX_SO_BUFFERS];
684 uint32_t ncomp[IR3_MAX_SO_BUFFERS];
685 uint32_t prog[IR3_MAX_SO_OUTPUTS * 2];
686 uint32_t prog_count;
687 uint32_t vpc_so_buf_cntl;
688 };
689
690 /* There are only three cache domains we have to care about: the CCU, or
691 * color cache unit, which is used for color and depth/stencil attachments
692 * and copy/blit destinations, and is split conceptually into color and depth,
693 * and the universal cache or UCHE which is used for pretty much everything
694 * else, except for the CP (uncached) and host. We need to flush whenever data
695 * crosses these boundaries.
696 */
697
698 enum tu_cmd_access_mask {
699 TU_ACCESS_UCHE_READ = 1 << 0,
700 TU_ACCESS_UCHE_WRITE = 1 << 1,
701 TU_ACCESS_CCU_COLOR_READ = 1 << 2,
702 TU_ACCESS_CCU_COLOR_WRITE = 1 << 3,
703 TU_ACCESS_CCU_DEPTH_READ = 1 << 4,
704 TU_ACCESS_CCU_DEPTH_WRITE = 1 << 5,
705
706 /* Experiments have shown that while it's safe to avoid flushing the CCU
707 * after each blit/renderpass, it's not safe to assume that subsequent
708 * lookups with a different attachment state will hit unflushed cache
709 * entries. That is, the CCU needs to be flushed and possibly invalidated
710 * when accessing memory with a different attachment state. Writing to an
711 * attachment under the following conditions after clearing using the
712 * normal 2d engine path is known to have issues:
713 *
714 * - It isn't the 0'th layer.
715 * - There are more than one attachment, and this isn't the 0'th attachment
716 * (this seems to also depend on the cpp of the attachments).
717 *
718 * Our best guess is that the layer/MRT state is used when computing
719 * the location of a cache entry in CCU, to avoid conflicts. We assume that
720 * any access in a renderpass after or before an access by a transfer needs
721 * a flush/invalidate, and use the _INCOHERENT variants to represent access
722 * by a transfer.
723 */
724 TU_ACCESS_CCU_COLOR_INCOHERENT_READ = 1 << 6,
725 TU_ACCESS_CCU_COLOR_INCOHERENT_WRITE = 1 << 7,
726 TU_ACCESS_CCU_DEPTH_INCOHERENT_READ = 1 << 8,
727 TU_ACCESS_CCU_DEPTH_INCOHERENT_WRITE = 1 << 9,
728
729 TU_ACCESS_SYSMEM_READ = 1 << 10,
730 TU_ACCESS_SYSMEM_WRITE = 1 << 11,
731
732 /* Set if a WFI is required due to data being read by the CP or the 2D
733 * engine.
734 */
735 TU_ACCESS_WFI_READ = 1 << 12,
736
737 TU_ACCESS_READ =
738 TU_ACCESS_UCHE_READ |
739 TU_ACCESS_CCU_COLOR_READ |
740 TU_ACCESS_CCU_DEPTH_READ |
741 TU_ACCESS_CCU_COLOR_INCOHERENT_READ |
742 TU_ACCESS_CCU_DEPTH_INCOHERENT_READ |
743 TU_ACCESS_SYSMEM_READ,
744
745 TU_ACCESS_WRITE =
746 TU_ACCESS_UCHE_WRITE |
747 TU_ACCESS_CCU_COLOR_WRITE |
748 TU_ACCESS_CCU_COLOR_INCOHERENT_WRITE |
749 TU_ACCESS_CCU_DEPTH_WRITE |
750 TU_ACCESS_CCU_DEPTH_INCOHERENT_WRITE |
751 TU_ACCESS_SYSMEM_WRITE,
752
753 TU_ACCESS_ALL =
754 TU_ACCESS_READ |
755 TU_ACCESS_WRITE,
756 };
757
758 enum tu_cmd_flush_bits {
759 TU_CMD_FLAG_CCU_FLUSH_DEPTH = 1 << 0,
760 TU_CMD_FLAG_CCU_FLUSH_COLOR = 1 << 1,
761 TU_CMD_FLAG_CCU_INVALIDATE_DEPTH = 1 << 2,
762 TU_CMD_FLAG_CCU_INVALIDATE_COLOR = 1 << 3,
763 TU_CMD_FLAG_CACHE_FLUSH = 1 << 4,
764 TU_CMD_FLAG_CACHE_INVALIDATE = 1 << 5,
765
766 TU_CMD_FLAG_ALL_FLUSH =
767 TU_CMD_FLAG_CCU_FLUSH_DEPTH |
768 TU_CMD_FLAG_CCU_FLUSH_COLOR |
769 TU_CMD_FLAG_CACHE_FLUSH,
770
771 TU_CMD_FLAG_ALL_INVALIDATE =
772 TU_CMD_FLAG_CCU_INVALIDATE_DEPTH |
773 TU_CMD_FLAG_CCU_INVALIDATE_COLOR |
774 TU_CMD_FLAG_CACHE_INVALIDATE,
775
776 TU_CMD_FLAG_WFI = 1 << 6,
777 };
778
779 /* Changing the CCU from sysmem mode to gmem mode or vice-versa is pretty
780 * heavy, involving a CCU cache flush/invalidate and a WFI in order to change
781 * which part of the gmem is used by the CCU. Here we keep track of what the
782 * state of the CCU.
783 */
784 enum tu_cmd_ccu_state {
785 TU_CMD_CCU_SYSMEM,
786 TU_CMD_CCU_GMEM,
787 TU_CMD_CCU_UNKNOWN,
788 };
789
790 struct tu_cache_state {
791 /* Caches which must be made available (flushed) eventually if there are
792 * any users outside that cache domain, and caches which must be
793 * invalidated eventually if there are any reads.
794 */
795 enum tu_cmd_flush_bits pending_flush_bits;
796 /* Pending flushes */
797 enum tu_cmd_flush_bits flush_bits;
798 };
799
800 struct tu_cmd_state
801 {
802 uint32_t dirty;
803
804 struct tu_pipeline *pipeline;
805 struct tu_pipeline *compute_pipeline;
806
807 /* Vertex buffers */
808 struct
809 {
810 struct tu_buffer *buffers[MAX_VBS];
811 VkDeviceSize offsets[MAX_VBS];
812 } vb;
813
814 /* for dynamic states that can't be emitted directly */
815 uint32_t dynamic_stencil_mask;
816 uint32_t dynamic_stencil_wrmask;
817 uint32_t dynamic_stencil_ref;
818 uint32_t dynamic_gras_su_cntl;
819
820 /* saved states to re-emit in TU_CMD_DIRTY_DRAW_STATE case */
821 struct tu_draw_state dynamic_state[TU_DYNAMIC_STATE_COUNT];
822 struct tu_cs_entry vertex_buffers_ib;
823 struct tu_cs_entry shader_const_ib[MESA_SHADER_STAGES];
824 struct tu_cs_entry desc_sets_ib, desc_sets_load_ib;
825 struct tu_cs_entry ia_gmem_ib, ia_sysmem_ib;
826
827 /* Stream output buffers */
828 struct
829 {
830 struct tu_buffer *buffers[IR3_MAX_SO_BUFFERS];
831 VkDeviceSize offsets[IR3_MAX_SO_BUFFERS];
832 VkDeviceSize sizes[IR3_MAX_SO_BUFFERS];
833 } streamout_buf;
834
835 uint8_t streamout_reset;
836 uint8_t streamout_enabled;
837
838 /* Index buffer */
839 struct tu_buffer *index_buffer;
840 uint64_t index_offset;
841 uint32_t index_type;
842 uint32_t max_index_count;
843 uint64_t index_va;
844
845 /* Renderpasses are tricky, because we may need to flush differently if
846 * using sysmem vs. gmem and therefore we have to delay any flushing that
847 * happens before a renderpass. So we have to have two copies of the flush
848 * state, one for intra-renderpass flushes (i.e. renderpass dependencies)
849 * and one for outside a renderpass.
850 */
851 struct tu_cache_state cache;
852 struct tu_cache_state renderpass_cache;
853
854 enum tu_cmd_ccu_state ccu_state;
855
856 const struct tu_render_pass *pass;
857 const struct tu_subpass *subpass;
858 const struct tu_framebuffer *framebuffer;
859
860 struct tu_tiling_config tiling_config;
861
862 struct tu_cs_entry tile_store_ib;
863 };
864
865 struct tu_cmd_pool
866 {
867 VkAllocationCallbacks alloc;
868 struct list_head cmd_buffers;
869 struct list_head free_cmd_buffers;
870 uint32_t queue_family_index;
871 };
872
873 struct tu_cmd_buffer_upload
874 {
875 uint8_t *map;
876 unsigned offset;
877 uint64_t size;
878 struct list_head list;
879 };
880
881 enum tu_cmd_buffer_status
882 {
883 TU_CMD_BUFFER_STATUS_INVALID,
884 TU_CMD_BUFFER_STATUS_INITIAL,
885 TU_CMD_BUFFER_STATUS_RECORDING,
886 TU_CMD_BUFFER_STATUS_EXECUTABLE,
887 TU_CMD_BUFFER_STATUS_PENDING,
888 };
889
890 struct tu_bo_list
891 {
892 uint32_t count;
893 uint32_t capacity;
894 struct drm_msm_gem_submit_bo *bo_infos;
895 };
896
897 #define TU_BO_LIST_FAILED (~0)
898
899 void
900 tu_bo_list_init(struct tu_bo_list *list);
901 void
902 tu_bo_list_destroy(struct tu_bo_list *list);
903 void
904 tu_bo_list_reset(struct tu_bo_list *list);
905 uint32_t
906 tu_bo_list_add(struct tu_bo_list *list,
907 const struct tu_bo *bo,
908 uint32_t flags);
909 VkResult
910 tu_bo_list_merge(struct tu_bo_list *list, const struct tu_bo_list *other);
911
912 /* This struct defines the layout of the scratch_bo */
913 struct tu6_control
914 {
915 uint32_t seqno_dummy; /* dummy seqno for CP_EVENT_WRITE */
916 uint32_t _pad0;
917 volatile uint32_t vsc_overflow;
918 uint32_t _pad1;
919 /* flag set from cmdstream when VSC overflow detected: */
920 uint32_t vsc_scratch;
921 uint32_t _pad2;
922 uint32_t _pad3;
923 uint32_t _pad4;
924
925 /* scratch space for VPC_SO[i].FLUSH_BASE_LO/HI, start on 32 byte boundary. */
926 struct {
927 uint32_t offset;
928 uint32_t pad[7];
929 } flush_base[4];
930 };
931
932 #define ctrl_offset(member) offsetof(struct tu6_control, member)
933
934 struct tu_cmd_buffer
935 {
936 VK_LOADER_DATA _loader_data;
937
938 struct tu_device *device;
939
940 struct tu_cmd_pool *pool;
941 struct list_head pool_link;
942
943 VkCommandBufferUsageFlags usage_flags;
944 VkCommandBufferLevel level;
945 enum tu_cmd_buffer_status status;
946
947 struct tu_cmd_state state;
948 struct tu_vertex_binding vertex_bindings[MAX_VBS];
949 uint32_t vertex_bindings_set;
950 uint32_t queue_family_index;
951
952 uint32_t push_constants[MAX_PUSH_CONSTANTS_SIZE / 4];
953 VkShaderStageFlags push_constant_stages;
954 struct tu_descriptor_set meta_push_descriptors;
955
956 struct tu_descriptor_state descriptors[MAX_BIND_POINTS];
957
958 struct tu_cmd_buffer_upload upload;
959
960 VkResult record_result;
961
962 struct tu_bo_list bo_list;
963 struct tu_cs cs;
964 struct tu_cs draw_cs;
965 struct tu_cs draw_epilogue_cs;
966 struct tu_cs sub_cs;
967
968 struct tu_bo scratch_bo;
969
970 struct tu_bo vsc_draw_strm;
971 struct tu_bo vsc_prim_strm;
972 uint32_t vsc_draw_strm_pitch;
973 uint32_t vsc_prim_strm_pitch;
974 bool use_vsc_data;
975 };
976
977 /* Temporary struct for tracking a register state to be written, used by
978 * a6xx-pack.h and tu_cs_emit_regs()
979 */
980 struct tu_reg_value {
981 uint32_t reg;
982 uint64_t value;
983 bool is_address;
984 struct tu_bo *bo;
985 bool bo_write;
986 uint32_t bo_offset;
987 uint32_t bo_shift;
988 };
989
990
991 void tu_emit_cache_flush_renderpass(struct tu_cmd_buffer *cmd_buffer,
992 struct tu_cs *cs);
993
994 void tu_emit_cache_flush_ccu(struct tu_cmd_buffer *cmd_buffer,
995 struct tu_cs *cs,
996 enum tu_cmd_ccu_state ccu_state);
997
998 void
999 tu6_emit_event_write(struct tu_cmd_buffer *cmd,
1000 struct tu_cs *cs,
1001 enum vgt_event_type event);
1002
1003 static inline struct tu_descriptor_state *
1004 tu_get_descriptors_state(struct tu_cmd_buffer *cmd_buffer,
1005 VkPipelineBindPoint bind_point)
1006 {
1007 return &cmd_buffer->descriptors[bind_point];
1008 }
1009
1010 struct tu_event
1011 {
1012 struct tu_bo bo;
1013 };
1014
1015 struct tu_shader_module
1016 {
1017 unsigned char sha1[20];
1018
1019 uint32_t code_size;
1020 const uint32_t *code[0];
1021 };
1022
1023 struct tu_push_constant_range
1024 {
1025 uint32_t lo;
1026 uint32_t count;
1027 };
1028
1029 struct tu_shader
1030 {
1031 struct ir3_shader *ir3_shader;
1032
1033 struct tu_push_constant_range push_consts;
1034 uint8_t active_desc_sets;
1035 };
1036
1037 struct tu_shader *
1038 tu_shader_create(struct tu_device *dev,
1039 gl_shader_stage stage,
1040 const VkPipelineShaderStageCreateInfo *stage_info,
1041 struct tu_pipeline_layout *layout,
1042 const VkAllocationCallbacks *alloc);
1043
1044 void
1045 tu_shader_destroy(struct tu_device *dev,
1046 struct tu_shader *shader,
1047 const VkAllocationCallbacks *alloc);
1048
1049 struct tu_program_descriptor_linkage
1050 {
1051 struct ir3_const_state const_state;
1052
1053 uint32_t constlen;
1054
1055 struct tu_push_constant_range push_consts;
1056 };
1057
1058 struct tu_pipeline
1059 {
1060 struct tu_cs cs;
1061
1062 struct tu_pipeline_layout *layout;
1063
1064 bool need_indirect_descriptor_sets;
1065 VkShaderStageFlags active_stages;
1066 uint32_t active_desc_sets;
1067
1068 struct tu_streamout_state streamout;
1069
1070 /* mask of enabled dynamic states
1071 * if BIT(i) is set, pipeline->dynamic_state[i] is *NOT* used
1072 */
1073 uint32_t dynamic_state_mask;
1074 struct tu_draw_state dynamic_state[TU_DYNAMIC_STATE_COUNT];
1075
1076 /* gras_su_cntl without line width, used for dynamic line width state */
1077 uint32_t gras_su_cntl;
1078
1079 struct
1080 {
1081 struct tu_bo binary_bo;
1082 struct tu_cs_entry state_ib;
1083 struct tu_cs_entry binning_state_ib;
1084
1085 struct tu_program_descriptor_linkage link[MESA_SHADER_STAGES];
1086 } program;
1087
1088 struct
1089 {
1090 struct tu_cs_entry state_ib;
1091 } load_state;
1092
1093 struct
1094 {
1095 struct tu_cs_entry state_ib;
1096 struct tu_cs_entry binning_state_ib;
1097 uint32_t bindings_used;
1098 } vi;
1099
1100 struct
1101 {
1102 enum pc_di_primtype primtype;
1103 bool primitive_restart;
1104 } ia;
1105
1106 struct
1107 {
1108 uint32_t patch_type;
1109 uint32_t per_vertex_output_size;
1110 uint32_t per_patch_output_size;
1111 uint32_t hs_bo_regid;
1112 uint32_t ds_bo_regid;
1113 bool upper_left_domain_origin;
1114 } tess;
1115
1116 struct
1117 {
1118 struct tu_cs_entry state_ib;
1119 } rast;
1120
1121 struct
1122 {
1123 struct tu_cs_entry state_ib;
1124 } ds;
1125
1126 struct
1127 {
1128 struct tu_cs_entry state_ib;
1129 } blend;
1130
1131 struct
1132 {
1133 uint32_t local_size[3];
1134 } compute;
1135 };
1136
1137 void
1138 tu6_emit_viewport(struct tu_cs *cs, const VkViewport *viewport);
1139
1140 void
1141 tu6_emit_scissor(struct tu_cs *cs, const VkRect2D *scissor);
1142
1143 void
1144 tu6_emit_sample_locations(struct tu_cs *cs, const VkSampleLocationsInfoEXT *samp_loc);
1145
1146 void
1147 tu6_emit_depth_bias(struct tu_cs *cs,
1148 float constant_factor,
1149 float clamp,
1150 float slope_factor);
1151
1152 void tu6_emit_msaa(struct tu_cs *cs, VkSampleCountFlagBits samples);
1153
1154 void tu6_emit_window_scissor(struct tu_cs *cs, uint32_t x1, uint32_t y1, uint32_t x2, uint32_t y2);
1155
1156 void tu6_emit_window_offset(struct tu_cs *cs, uint32_t x1, uint32_t y1);
1157
1158 void
1159 tu6_emit_xs_config(struct tu_cs *cs,
1160 gl_shader_stage stage,
1161 const struct ir3_shader_variant *xs,
1162 uint64_t binary_iova);
1163
1164 void
1165 tu6_emit_vpc(struct tu_cs *cs,
1166 const struct ir3_shader_variant *vs,
1167 const struct ir3_shader_variant *hs,
1168 const struct ir3_shader_variant *ds,
1169 const struct ir3_shader_variant *gs,
1170 const struct ir3_shader_variant *fs,
1171 struct tu_streamout_state *tf);
1172
1173 void
1174 tu6_emit_fs_inputs(struct tu_cs *cs, const struct ir3_shader_variant *fs);
1175
1176 struct tu_image_view;
1177
1178 void
1179 tu_resolve_sysmem(struct tu_cmd_buffer *cmd,
1180 struct tu_cs *cs,
1181 struct tu_image_view *src,
1182 struct tu_image_view *dst,
1183 uint32_t layers,
1184 const VkRect2D *rect);
1185
1186 void
1187 tu_clear_sysmem_attachment(struct tu_cmd_buffer *cmd,
1188 struct tu_cs *cs,
1189 uint32_t a,
1190 const VkRenderPassBeginInfo *info);
1191
1192 void
1193 tu_clear_gmem_attachment(struct tu_cmd_buffer *cmd,
1194 struct tu_cs *cs,
1195 uint32_t a,
1196 const VkRenderPassBeginInfo *info);
1197
1198 void
1199 tu_load_gmem_attachment(struct tu_cmd_buffer *cmd,
1200 struct tu_cs *cs,
1201 uint32_t a,
1202 bool force_load);
1203
1204 /* expose this function to be able to emit load without checking LOAD_OP */
1205 void
1206 tu_emit_load_gmem_attachment(struct tu_cmd_buffer *cmd, struct tu_cs *cs, uint32_t a);
1207
1208 /* note: gmem store can also resolve */
1209 void
1210 tu_store_gmem_attachment(struct tu_cmd_buffer *cmd,
1211 struct tu_cs *cs,
1212 uint32_t a,
1213 uint32_t gmem_a);
1214
1215 enum tu_supported_formats {
1216 FMT_VERTEX = 1,
1217 FMT_TEXTURE = 2,
1218 FMT_COLOR = 4,
1219 };
1220
1221 struct tu_native_format
1222 {
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;
1227 };
1228
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);
1232
1233 static inline enum a6xx_format
1234 tu6_base_format(VkFormat format)
1235 {
1236 /* note: tu6_format_color doesn't care about tiling for .fmt field */
1237 return tu6_format_color(format, TILE6_LINEAR).fmt;
1238 }
1239
1240 struct tu_image
1241 {
1242 VkImageType type;
1243 /* The original VkFormat provided by the client. This may not match any
1244 * of the actual surface formats.
1245 */
1246 VkFormat vk_format;
1247 VkImageAspectFlags aspects;
1248 VkImageUsageFlags usage; /**< Superset of VkImageCreateInfo::usage. */
1249 VkImageTiling tiling; /** VkImageCreateInfo::tiling */
1250 VkImageCreateFlags flags; /** VkImageCreateInfo::flags */
1251 VkExtent3D extent;
1252 uint32_t level_count;
1253 uint32_t layer_count;
1254 VkSampleCountFlagBits samples;
1255
1256 struct fdl_layout layout;
1257
1258 unsigned queue_family_mask;
1259 bool exclusive;
1260 bool shareable;
1261
1262 /* For VK_ANDROID_native_buffer, the WSI image owns the memory, */
1263 VkDeviceMemory owned_memory;
1264
1265 /* Set when bound */
1266 struct tu_bo *bo;
1267 VkDeviceSize bo_offset;
1268 };
1269
1270 static inline uint32_t
1271 tu_get_layerCount(const struct tu_image *image,
1272 const VkImageSubresourceRange *range)
1273 {
1274 return range->layerCount == VK_REMAINING_ARRAY_LAYERS
1275 ? image->layer_count - range->baseArrayLayer
1276 : range->layerCount;
1277 }
1278
1279 static inline uint32_t
1280 tu_get_levelCount(const struct tu_image *image,
1281 const VkImageSubresourceRange *range)
1282 {
1283 return range->levelCount == VK_REMAINING_MIP_LEVELS
1284 ? image->level_count - range->baseMipLevel
1285 : range->levelCount;
1286 }
1287
1288 struct tu_image_view
1289 {
1290 struct tu_image *image; /**< VkImageViewCreateInfo::image */
1291
1292 uint64_t base_addr;
1293 uint64_t ubwc_addr;
1294 uint32_t layer_size;
1295 uint32_t ubwc_layer_size;
1296
1297 /* used to determine if fast gmem store path can be used */
1298 VkExtent2D extent;
1299 bool need_y2_align;
1300
1301 bool ubwc_enabled;
1302
1303 uint32_t descriptor[A6XX_TEX_CONST_DWORDS];
1304
1305 /* Descriptor for use as a storage image as opposed to a sampled image.
1306 * This has a few differences for cube maps (e.g. type).
1307 */
1308 uint32_t storage_descriptor[A6XX_TEX_CONST_DWORDS];
1309
1310 /* pre-filled register values */
1311 uint32_t PITCH;
1312 uint32_t FLAG_BUFFER_PITCH;
1313
1314 uint32_t RB_MRT_BUF_INFO;
1315 uint32_t SP_FS_MRT_REG;
1316
1317 uint32_t SP_PS_2D_SRC_INFO;
1318 uint32_t SP_PS_2D_SRC_SIZE;
1319
1320 uint32_t RB_2D_DST_INFO;
1321
1322 uint32_t RB_BLIT_DST_INFO;
1323 };
1324
1325 struct tu_sampler_ycbcr_conversion {
1326 VkFormat format;
1327 VkSamplerYcbcrModelConversion ycbcr_model;
1328 VkSamplerYcbcrRange ycbcr_range;
1329 VkComponentMapping components;
1330 VkChromaLocation chroma_offsets[2];
1331 VkFilter chroma_filter;
1332 };
1333
1334 struct tu_sampler {
1335 uint32_t descriptor[A6XX_TEX_SAMP_DWORDS];
1336 struct tu_sampler_ycbcr_conversion *ycbcr_sampler;
1337 };
1338
1339 void
1340 tu_cs_image_ref(struct tu_cs *cs, const struct tu_image_view *iview, uint32_t layer);
1341
1342 void
1343 tu_cs_image_ref_2d(struct tu_cs *cs, const struct tu_image_view *iview, uint32_t layer, bool src);
1344
1345 void
1346 tu_cs_image_flag_ref(struct tu_cs *cs, const struct tu_image_view *iview, uint32_t layer);
1347
1348 VkResult
1349 tu_image_create(VkDevice _device,
1350 const VkImageCreateInfo *pCreateInfo,
1351 const VkAllocationCallbacks *alloc,
1352 VkImage *pImage,
1353 uint64_t modifier,
1354 const VkSubresourceLayout *plane_layouts);
1355
1356 VkResult
1357 tu_image_from_gralloc(VkDevice device_h,
1358 const VkImageCreateInfo *base_info,
1359 const VkNativeBufferANDROID *gralloc_info,
1360 const VkAllocationCallbacks *alloc,
1361 VkImage *out_image_h);
1362
1363 void
1364 tu_image_view_init(struct tu_image_view *view,
1365 const VkImageViewCreateInfo *pCreateInfo);
1366
1367 struct tu_buffer_view
1368 {
1369 uint32_t descriptor[A6XX_TEX_CONST_DWORDS];
1370
1371 struct tu_buffer *buffer;
1372 };
1373 void
1374 tu_buffer_view_init(struct tu_buffer_view *view,
1375 struct tu_device *device,
1376 const VkBufferViewCreateInfo *pCreateInfo);
1377
1378 struct tu_attachment_info
1379 {
1380 struct tu_image_view *attachment;
1381 };
1382
1383 struct tu_framebuffer
1384 {
1385 uint32_t width;
1386 uint32_t height;
1387 uint32_t layers;
1388
1389 uint32_t attachment_count;
1390 struct tu_attachment_info attachments[0];
1391 };
1392
1393 struct tu_subpass_barrier {
1394 VkPipelineStageFlags src_stage_mask;
1395 VkAccessFlags src_access_mask;
1396 VkAccessFlags dst_access_mask;
1397 bool incoherent_ccu_color, incoherent_ccu_depth;
1398 };
1399
1400 struct tu_subpass_attachment
1401 {
1402 uint32_t attachment;
1403 VkImageLayout layout;
1404 };
1405
1406 struct tu_subpass
1407 {
1408 uint32_t input_count;
1409 uint32_t color_count;
1410 struct tu_subpass_attachment *input_attachments;
1411 struct tu_subpass_attachment *color_attachments;
1412 struct tu_subpass_attachment *resolve_attachments;
1413 struct tu_subpass_attachment depth_stencil_attachment;
1414
1415 VkSampleCountFlagBits samples;
1416 bool has_external_src, has_external_dst;
1417
1418 uint32_t srgb_cntl;
1419
1420 struct tu_subpass_barrier start_barrier;
1421 };
1422
1423 struct tu_render_pass_attachment
1424 {
1425 VkFormat format;
1426 uint32_t samples;
1427 uint32_t cpp;
1428 VkImageAspectFlags clear_mask;
1429 bool load;
1430 bool store;
1431 VkImageLayout initial_layout, final_layout;
1432 int32_t gmem_offset;
1433 };
1434
1435 struct tu_render_pass
1436 {
1437 uint32_t attachment_count;
1438 uint32_t subpass_count;
1439 uint32_t gmem_pixels;
1440 uint32_t tile_align_w;
1441 struct tu_subpass_attachment *subpass_attachments;
1442 struct tu_render_pass_attachment *attachments;
1443 struct tu_subpass_barrier end_barrier;
1444 struct tu_subpass subpasses[0];
1445 };
1446
1447 struct tu_query_pool
1448 {
1449 VkQueryType type;
1450 uint32_t stride;
1451 uint64_t size;
1452 uint32_t pipeline_statistics;
1453 struct tu_bo bo;
1454 };
1455
1456 struct tu_semaphore
1457 {
1458 uint32_t syncobj;
1459 uint32_t temp_syncobj;
1460 };
1461
1462 void
1463 tu_set_descriptor_set(struct tu_cmd_buffer *cmd_buffer,
1464 VkPipelineBindPoint bind_point,
1465 struct tu_descriptor_set *set,
1466 unsigned idx);
1467
1468 void
1469 tu_update_descriptor_sets(struct tu_device *device,
1470 struct tu_cmd_buffer *cmd_buffer,
1471 VkDescriptorSet overrideSet,
1472 uint32_t descriptorWriteCount,
1473 const VkWriteDescriptorSet *pDescriptorWrites,
1474 uint32_t descriptorCopyCount,
1475 const VkCopyDescriptorSet *pDescriptorCopies);
1476
1477 void
1478 tu_update_descriptor_set_with_template(
1479 struct tu_device *device,
1480 struct tu_cmd_buffer *cmd_buffer,
1481 struct tu_descriptor_set *set,
1482 VkDescriptorUpdateTemplate descriptorUpdateTemplate,
1483 const void *pData);
1484
1485 int
1486 tu_drm_get_gpu_id(const struct tu_physical_device *dev, uint32_t *id);
1487
1488 int
1489 tu_drm_get_gmem_size(const struct tu_physical_device *dev, uint32_t *size);
1490
1491 int
1492 tu_drm_get_gmem_base(const struct tu_physical_device *dev, uint64_t *base);
1493
1494 int
1495 tu_drm_submitqueue_new(const struct tu_device *dev,
1496 int priority,
1497 uint32_t *queue_id);
1498
1499 void
1500 tu_drm_submitqueue_close(const struct tu_device *dev, uint32_t queue_id);
1501
1502 uint32_t
1503 tu_gem_new(const struct tu_device *dev, uint64_t size, uint32_t flags);
1504 uint32_t
1505 tu_gem_import_dmabuf(const struct tu_device *dev,
1506 int prime_fd,
1507 uint64_t size);
1508 int
1509 tu_gem_export_dmabuf(const struct tu_device *dev, uint32_t gem_handle);
1510 void
1511 tu_gem_close(const struct tu_device *dev, uint32_t gem_handle);
1512 uint64_t
1513 tu_gem_info_offset(const struct tu_device *dev, uint32_t gem_handle);
1514 uint64_t
1515 tu_gem_info_iova(const struct tu_device *dev, uint32_t gem_handle);
1516
1517 #define TU_DEFINE_HANDLE_CASTS(__tu_type, __VkType) \
1518 \
1519 static inline struct __tu_type *__tu_type##_from_handle(__VkType _handle) \
1520 { \
1521 return (struct __tu_type *) _handle; \
1522 } \
1523 \
1524 static inline __VkType __tu_type##_to_handle(struct __tu_type *_obj) \
1525 { \
1526 return (__VkType) _obj; \
1527 }
1528
1529 #define TU_DEFINE_NONDISP_HANDLE_CASTS(__tu_type, __VkType) \
1530 \
1531 static inline struct __tu_type *__tu_type##_from_handle(__VkType _handle) \
1532 { \
1533 return (struct __tu_type *) (uintptr_t) _handle; \
1534 } \
1535 \
1536 static inline __VkType __tu_type##_to_handle(struct __tu_type *_obj) \
1537 { \
1538 return (__VkType)(uintptr_t) _obj; \
1539 }
1540
1541 #define TU_FROM_HANDLE(__tu_type, __name, __handle) \
1542 struct __tu_type *__name = __tu_type##_from_handle(__handle)
1543
1544 TU_DEFINE_HANDLE_CASTS(tu_cmd_buffer, VkCommandBuffer)
1545 TU_DEFINE_HANDLE_CASTS(tu_device, VkDevice)
1546 TU_DEFINE_HANDLE_CASTS(tu_instance, VkInstance)
1547 TU_DEFINE_HANDLE_CASTS(tu_physical_device, VkPhysicalDevice)
1548 TU_DEFINE_HANDLE_CASTS(tu_queue, VkQueue)
1549
1550 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_cmd_pool, VkCommandPool)
1551 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_buffer, VkBuffer)
1552 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_buffer_view, VkBufferView)
1553 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_pool, VkDescriptorPool)
1554 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_set, VkDescriptorSet)
1555 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_set_layout,
1556 VkDescriptorSetLayout)
1557 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_update_template,
1558 VkDescriptorUpdateTemplate)
1559 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_device_memory, VkDeviceMemory)
1560 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_fence, VkFence)
1561 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_event, VkEvent)
1562 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_framebuffer, VkFramebuffer)
1563 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_image, VkImage)
1564 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_image_view, VkImageView);
1565 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_pipeline_cache, VkPipelineCache)
1566 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_pipeline, VkPipeline)
1567 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_pipeline_layout, VkPipelineLayout)
1568 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_query_pool, VkQueryPool)
1569 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_render_pass, VkRenderPass)
1570 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_sampler, VkSampler)
1571 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_sampler_ycbcr_conversion, VkSamplerYcbcrConversion)
1572 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_shader_module, VkShaderModule)
1573 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_semaphore, VkSemaphore)
1574
1575 #endif /* TU_PRIVATE_H */