turnip: Update VFD_CONTROL with tess system values
[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_GS_CONST,
439 TU_DRAW_STATE_FS_CONST,
440 TU_DRAW_STATE_DESC_SETS,
441 TU_DRAW_STATE_DESC_SETS_LOAD,
442 TU_DRAW_STATE_VS_PARAMS,
443 TU_DRAW_STATE_INPUT_ATTACHMENTS_GMEM,
444 TU_DRAW_STATE_INPUT_ATTACHMENTS_SYSMEM,
445
446 /* dynamic state related draw states */
447 TU_DRAW_STATE_DYNAMIC,
448 TU_DRAW_STATE_COUNT = TU_DRAW_STATE_DYNAMIC + TU_DYNAMIC_STATE_COUNT,
449 };
450
451 enum tu_cs_mode
452 {
453
454 /*
455 * A command stream in TU_CS_MODE_GROW mode grows automatically whenever it
456 * is full. tu_cs_begin must be called before command packet emission and
457 * tu_cs_end must be called after.
458 *
459 * This mode may create multiple entries internally. The entries must be
460 * submitted together.
461 */
462 TU_CS_MODE_GROW,
463
464 /*
465 * A command stream in TU_CS_MODE_EXTERNAL mode wraps an external,
466 * fixed-size buffer. tu_cs_begin and tu_cs_end are optional and have no
467 * effect on it.
468 *
469 * This mode does not create any entry or any BO.
470 */
471 TU_CS_MODE_EXTERNAL,
472
473 /*
474 * A command stream in TU_CS_MODE_SUB_STREAM mode does not support direct
475 * command packet emission. tu_cs_begin_sub_stream must be called to get a
476 * sub-stream to emit comamnd packets to. When done with the sub-stream,
477 * tu_cs_end_sub_stream must be called.
478 *
479 * This mode does not create any entry internally.
480 */
481 TU_CS_MODE_SUB_STREAM,
482 };
483
484 struct tu_cs
485 {
486 uint32_t *start;
487 uint32_t *cur;
488 uint32_t *reserved_end;
489 uint32_t *end;
490
491 struct tu_device *device;
492 enum tu_cs_mode mode;
493 uint32_t next_bo_size;
494
495 struct tu_cs_entry *entries;
496 uint32_t entry_count;
497 uint32_t entry_capacity;
498
499 struct tu_bo **bos;
500 uint32_t bo_count;
501 uint32_t bo_capacity;
502
503 /* state for cond_exec_start/cond_exec_end */
504 uint32_t cond_flags;
505 uint32_t *cond_dwords;
506 };
507
508 struct tu_device_memory
509 {
510 struct tu_bo bo;
511 VkDeviceSize size;
512
513 /* for dedicated allocations */
514 struct tu_image *image;
515 struct tu_buffer *buffer;
516
517 uint32_t type_index;
518 void *map;
519 void *user_ptr;
520 };
521
522 struct tu_descriptor_range
523 {
524 uint64_t va;
525 uint32_t size;
526 };
527
528 struct tu_descriptor_set
529 {
530 const struct tu_descriptor_set_layout *layout;
531 struct tu_descriptor_pool *pool;
532 uint32_t size;
533
534 uint64_t va;
535 uint32_t *mapped_ptr;
536
537 uint32_t *dynamic_descriptors;
538
539 struct tu_bo *buffers[0];
540 };
541
542 struct tu_push_descriptor_set
543 {
544 struct tu_descriptor_set set;
545 uint32_t capacity;
546 };
547
548 struct tu_descriptor_pool_entry
549 {
550 uint32_t offset;
551 uint32_t size;
552 struct tu_descriptor_set *set;
553 };
554
555 struct tu_descriptor_pool
556 {
557 struct tu_bo bo;
558 uint64_t current_offset;
559 uint64_t size;
560
561 uint8_t *host_memory_base;
562 uint8_t *host_memory_ptr;
563 uint8_t *host_memory_end;
564
565 uint32_t entry_count;
566 uint32_t max_entry_count;
567 struct tu_descriptor_pool_entry entries[0];
568 };
569
570 struct tu_descriptor_update_template_entry
571 {
572 VkDescriptorType descriptor_type;
573
574 /* The number of descriptors to update */
575 uint32_t descriptor_count;
576
577 /* Into mapped_ptr or dynamic_descriptors, in units of the respective array
578 */
579 uint32_t dst_offset;
580
581 /* In dwords. Not valid/used for dynamic descriptors */
582 uint32_t dst_stride;
583
584 uint32_t buffer_offset;
585
586 /* Only valid for combined image samplers and samplers */
587 uint16_t has_sampler;
588
589 /* In bytes */
590 size_t src_offset;
591 size_t src_stride;
592
593 /* For push descriptors */
594 const uint32_t *immutable_samplers;
595 };
596
597 struct tu_descriptor_update_template
598 {
599 uint32_t entry_count;
600 struct tu_descriptor_update_template_entry entry[0];
601 };
602
603 struct tu_buffer
604 {
605 VkDeviceSize size;
606
607 VkBufferUsageFlags usage;
608 VkBufferCreateFlags flags;
609
610 struct tu_bo *bo;
611 VkDeviceSize bo_offset;
612 };
613
614 static inline uint64_t
615 tu_buffer_iova(struct tu_buffer *buffer)
616 {
617 return buffer->bo->iova + buffer->bo_offset;
618 }
619
620 struct tu_vertex_binding
621 {
622 struct tu_buffer *buffer;
623 VkDeviceSize offset;
624 };
625
626 const char *
627 tu_get_debug_option_name(int id);
628
629 const char *
630 tu_get_perftest_option_name(int id);
631
632 struct tu_descriptor_state
633 {
634 struct tu_descriptor_set *sets[MAX_SETS];
635 uint32_t dynamic_descriptors[MAX_DYNAMIC_BUFFERS * A6XX_TEX_CONST_DWORDS];
636 };
637
638 struct tu_tile
639 {
640 uint8_t pipe;
641 uint8_t slot;
642 VkOffset2D begin;
643 VkOffset2D end;
644 };
645
646 struct tu_tiling_config
647 {
648 VkRect2D render_area;
649
650 /* position and size of the first tile */
651 VkRect2D tile0;
652 /* number of tiles */
653 VkExtent2D tile_count;
654
655 /* size of the first VSC pipe */
656 VkExtent2D pipe0;
657 /* number of VSC pipes */
658 VkExtent2D pipe_count;
659
660 /* pipe register values */
661 uint32_t pipe_config[MAX_VSC_PIPES];
662 uint32_t pipe_sizes[MAX_VSC_PIPES];
663
664 /* Whether sysmem rendering must be used */
665 bool force_sysmem;
666 };
667
668 enum tu_cmd_dirty_bits
669 {
670 TU_CMD_DIRTY_COMPUTE_PIPELINE = 1 << 1,
671 TU_CMD_DIRTY_VERTEX_BUFFERS = 1 << 2,
672 TU_CMD_DIRTY_DESCRIPTOR_SETS = 1 << 3,
673 TU_CMD_DIRTY_COMPUTE_DESCRIPTOR_SETS = 1 << 4,
674 TU_CMD_DIRTY_SHADER_CONSTS = 1 << 5,
675 TU_CMD_DIRTY_STREAMOUT_BUFFERS = 1 << 6,
676 /* all draw states were disabled and need to be re-enabled: */
677 TU_CMD_DIRTY_DRAW_STATE = 1 << 7,
678 };
679
680 struct tu_streamout_state {
681 uint16_t stride[IR3_MAX_SO_BUFFERS];
682 uint32_t ncomp[IR3_MAX_SO_BUFFERS];
683 uint32_t prog[IR3_MAX_SO_OUTPUTS * 2];
684 uint32_t prog_count;
685 uint32_t vpc_so_buf_cntl;
686 };
687
688 /* There are only three cache domains we have to care about: the CCU, or
689 * color cache unit, which is used for color and depth/stencil attachments
690 * and copy/blit destinations, and is split conceptually into color and depth,
691 * and the universal cache or UCHE which is used for pretty much everything
692 * else, except for the CP (uncached) and host. We need to flush whenever data
693 * crosses these boundaries.
694 */
695
696 enum tu_cmd_access_mask {
697 TU_ACCESS_UCHE_READ = 1 << 0,
698 TU_ACCESS_UCHE_WRITE = 1 << 1,
699 TU_ACCESS_CCU_COLOR_READ = 1 << 2,
700 TU_ACCESS_CCU_COLOR_WRITE = 1 << 3,
701 TU_ACCESS_CCU_DEPTH_READ = 1 << 4,
702 TU_ACCESS_CCU_DEPTH_WRITE = 1 << 5,
703
704 /* Experiments have shown that while it's safe to avoid flushing the CCU
705 * after each blit/renderpass, it's not safe to assume that subsequent
706 * lookups with a different attachment state will hit unflushed cache
707 * entries. That is, the CCU needs to be flushed and possibly invalidated
708 * when accessing memory with a different attachment state. Writing to an
709 * attachment under the following conditions after clearing using the
710 * normal 2d engine path is known to have issues:
711 *
712 * - It isn't the 0'th layer.
713 * - There are more than one attachment, and this isn't the 0'th attachment
714 * (this seems to also depend on the cpp of the attachments).
715 *
716 * Our best guess is that the layer/MRT state is used when computing
717 * the location of a cache entry in CCU, to avoid conflicts. We assume that
718 * any access in a renderpass after or before an access by a transfer needs
719 * a flush/invalidate, and use the _INCOHERENT variants to represent access
720 * by a transfer.
721 */
722 TU_ACCESS_CCU_COLOR_INCOHERENT_READ = 1 << 6,
723 TU_ACCESS_CCU_COLOR_INCOHERENT_WRITE = 1 << 7,
724 TU_ACCESS_CCU_DEPTH_INCOHERENT_READ = 1 << 8,
725 TU_ACCESS_CCU_DEPTH_INCOHERENT_WRITE = 1 << 9,
726
727 TU_ACCESS_SYSMEM_READ = 1 << 10,
728 TU_ACCESS_SYSMEM_WRITE = 1 << 11,
729
730 /* Set if a WFI is required due to data being read by the CP or the 2D
731 * engine.
732 */
733 TU_ACCESS_WFI_READ = 1 << 12,
734
735 TU_ACCESS_READ =
736 TU_ACCESS_UCHE_READ |
737 TU_ACCESS_CCU_COLOR_READ |
738 TU_ACCESS_CCU_DEPTH_READ |
739 TU_ACCESS_CCU_COLOR_INCOHERENT_READ |
740 TU_ACCESS_CCU_DEPTH_INCOHERENT_READ |
741 TU_ACCESS_SYSMEM_READ,
742
743 TU_ACCESS_WRITE =
744 TU_ACCESS_UCHE_WRITE |
745 TU_ACCESS_CCU_COLOR_WRITE |
746 TU_ACCESS_CCU_COLOR_INCOHERENT_WRITE |
747 TU_ACCESS_CCU_DEPTH_WRITE |
748 TU_ACCESS_CCU_DEPTH_INCOHERENT_WRITE |
749 TU_ACCESS_SYSMEM_WRITE,
750
751 TU_ACCESS_ALL =
752 TU_ACCESS_READ |
753 TU_ACCESS_WRITE,
754 };
755
756 enum tu_cmd_flush_bits {
757 TU_CMD_FLAG_CCU_FLUSH_DEPTH = 1 << 0,
758 TU_CMD_FLAG_CCU_FLUSH_COLOR = 1 << 1,
759 TU_CMD_FLAG_CCU_INVALIDATE_DEPTH = 1 << 2,
760 TU_CMD_FLAG_CCU_INVALIDATE_COLOR = 1 << 3,
761 TU_CMD_FLAG_CACHE_FLUSH = 1 << 4,
762 TU_CMD_FLAG_CACHE_INVALIDATE = 1 << 5,
763
764 TU_CMD_FLAG_ALL_FLUSH =
765 TU_CMD_FLAG_CCU_FLUSH_DEPTH |
766 TU_CMD_FLAG_CCU_FLUSH_COLOR |
767 TU_CMD_FLAG_CACHE_FLUSH,
768
769 TU_CMD_FLAG_ALL_INVALIDATE =
770 TU_CMD_FLAG_CCU_INVALIDATE_DEPTH |
771 TU_CMD_FLAG_CCU_INVALIDATE_COLOR |
772 TU_CMD_FLAG_CACHE_INVALIDATE,
773
774 TU_CMD_FLAG_WFI = 1 << 6,
775 };
776
777 /* Changing the CCU from sysmem mode to gmem mode or vice-versa is pretty
778 * heavy, involving a CCU cache flush/invalidate and a WFI in order to change
779 * which part of the gmem is used by the CCU. Here we keep track of what the
780 * state of the CCU.
781 */
782 enum tu_cmd_ccu_state {
783 TU_CMD_CCU_SYSMEM,
784 TU_CMD_CCU_GMEM,
785 TU_CMD_CCU_UNKNOWN,
786 };
787
788 struct tu_cache_state {
789 /* Caches which must be made available (flushed) eventually if there are
790 * any users outside that cache domain, and caches which must be
791 * invalidated eventually if there are any reads.
792 */
793 enum tu_cmd_flush_bits pending_flush_bits;
794 /* Pending flushes */
795 enum tu_cmd_flush_bits flush_bits;
796 };
797
798 struct tu_cmd_state
799 {
800 uint32_t dirty;
801
802 struct tu_pipeline *pipeline;
803 struct tu_pipeline *compute_pipeline;
804
805 /* Vertex buffers */
806 struct
807 {
808 struct tu_buffer *buffers[MAX_VBS];
809 VkDeviceSize offsets[MAX_VBS];
810 } vb;
811
812 /* for dynamic states that can't be emitted directly */
813 uint32_t dynamic_stencil_mask;
814 uint32_t dynamic_stencil_wrmask;
815 uint32_t dynamic_stencil_ref;
816 uint32_t dynamic_gras_su_cntl;
817
818 /* saved states to re-emit in TU_CMD_DIRTY_DRAW_STATE case */
819 struct tu_draw_state dynamic_state[TU_DYNAMIC_STATE_COUNT];
820 struct tu_cs_entry vertex_buffers_ib;
821 struct tu_cs_entry shader_const_ib[MESA_SHADER_STAGES];
822 struct tu_cs_entry desc_sets_ib, desc_sets_load_ib;
823 struct tu_cs_entry ia_gmem_ib, ia_sysmem_ib;
824
825 /* Stream output buffers */
826 struct
827 {
828 struct tu_buffer *buffers[IR3_MAX_SO_BUFFERS];
829 VkDeviceSize offsets[IR3_MAX_SO_BUFFERS];
830 VkDeviceSize sizes[IR3_MAX_SO_BUFFERS];
831 } streamout_buf;
832
833 uint8_t streamout_reset;
834 uint8_t streamout_enabled;
835
836 /* Index buffer */
837 struct tu_buffer *index_buffer;
838 uint64_t index_offset;
839 uint32_t index_type;
840 uint32_t max_index_count;
841 uint64_t index_va;
842
843 /* Renderpasses are tricky, because we may need to flush differently if
844 * using sysmem vs. gmem and therefore we have to delay any flushing that
845 * happens before a renderpass. So we have to have two copies of the flush
846 * state, one for intra-renderpass flushes (i.e. renderpass dependencies)
847 * and one for outside a renderpass.
848 */
849 struct tu_cache_state cache;
850 struct tu_cache_state renderpass_cache;
851
852 enum tu_cmd_ccu_state ccu_state;
853
854 const struct tu_render_pass *pass;
855 const struct tu_subpass *subpass;
856 const struct tu_framebuffer *framebuffer;
857
858 struct tu_tiling_config tiling_config;
859
860 struct tu_cs_entry tile_store_ib;
861 };
862
863 struct tu_cmd_pool
864 {
865 VkAllocationCallbacks alloc;
866 struct list_head cmd_buffers;
867 struct list_head free_cmd_buffers;
868 uint32_t queue_family_index;
869 };
870
871 struct tu_cmd_buffer_upload
872 {
873 uint8_t *map;
874 unsigned offset;
875 uint64_t size;
876 struct list_head list;
877 };
878
879 enum tu_cmd_buffer_status
880 {
881 TU_CMD_BUFFER_STATUS_INVALID,
882 TU_CMD_BUFFER_STATUS_INITIAL,
883 TU_CMD_BUFFER_STATUS_RECORDING,
884 TU_CMD_BUFFER_STATUS_EXECUTABLE,
885 TU_CMD_BUFFER_STATUS_PENDING,
886 };
887
888 struct tu_bo_list
889 {
890 uint32_t count;
891 uint32_t capacity;
892 struct drm_msm_gem_submit_bo *bo_infos;
893 };
894
895 #define TU_BO_LIST_FAILED (~0)
896
897 void
898 tu_bo_list_init(struct tu_bo_list *list);
899 void
900 tu_bo_list_destroy(struct tu_bo_list *list);
901 void
902 tu_bo_list_reset(struct tu_bo_list *list);
903 uint32_t
904 tu_bo_list_add(struct tu_bo_list *list,
905 const struct tu_bo *bo,
906 uint32_t flags);
907 VkResult
908 tu_bo_list_merge(struct tu_bo_list *list, const struct tu_bo_list *other);
909
910 /* This struct defines the layout of the scratch_bo */
911 struct tu6_control
912 {
913 uint32_t seqno_dummy; /* dummy seqno for CP_EVENT_WRITE */
914 uint32_t _pad0;
915 volatile uint32_t vsc_overflow;
916 uint32_t _pad1;
917 /* flag set from cmdstream when VSC overflow detected: */
918 uint32_t vsc_scratch;
919 uint32_t _pad2;
920 uint32_t _pad3;
921 uint32_t _pad4;
922
923 /* scratch space for VPC_SO[i].FLUSH_BASE_LO/HI, start on 32 byte boundary. */
924 struct {
925 uint32_t offset;
926 uint32_t pad[7];
927 } flush_base[4];
928 };
929
930 #define ctrl_offset(member) offsetof(struct tu6_control, member)
931
932 struct tu_cmd_buffer
933 {
934 VK_LOADER_DATA _loader_data;
935
936 struct tu_device *device;
937
938 struct tu_cmd_pool *pool;
939 struct list_head pool_link;
940
941 VkCommandBufferUsageFlags usage_flags;
942 VkCommandBufferLevel level;
943 enum tu_cmd_buffer_status status;
944
945 struct tu_cmd_state state;
946 struct tu_vertex_binding vertex_bindings[MAX_VBS];
947 uint32_t vertex_bindings_set;
948 uint32_t queue_family_index;
949
950 uint32_t push_constants[MAX_PUSH_CONSTANTS_SIZE / 4];
951 VkShaderStageFlags push_constant_stages;
952 struct tu_descriptor_set meta_push_descriptors;
953
954 struct tu_descriptor_state descriptors[MAX_BIND_POINTS];
955
956 struct tu_cmd_buffer_upload upload;
957
958 VkResult record_result;
959
960 struct tu_bo_list bo_list;
961 struct tu_cs cs;
962 struct tu_cs draw_cs;
963 struct tu_cs draw_epilogue_cs;
964 struct tu_cs sub_cs;
965
966 struct tu_bo scratch_bo;
967
968 struct tu_bo vsc_draw_strm;
969 struct tu_bo vsc_prim_strm;
970 uint32_t vsc_draw_strm_pitch;
971 uint32_t vsc_prim_strm_pitch;
972 bool use_vsc_data;
973 };
974
975 /* Temporary struct for tracking a register state to be written, used by
976 * a6xx-pack.h and tu_cs_emit_regs()
977 */
978 struct tu_reg_value {
979 uint32_t reg;
980 uint64_t value;
981 bool is_address;
982 struct tu_bo *bo;
983 bool bo_write;
984 uint32_t bo_offset;
985 uint32_t bo_shift;
986 };
987
988
989 void tu_emit_cache_flush_renderpass(struct tu_cmd_buffer *cmd_buffer,
990 struct tu_cs *cs);
991
992 void tu_emit_cache_flush_ccu(struct tu_cmd_buffer *cmd_buffer,
993 struct tu_cs *cs,
994 enum tu_cmd_ccu_state ccu_state);
995
996 void
997 tu6_emit_event_write(struct tu_cmd_buffer *cmd,
998 struct tu_cs *cs,
999 enum vgt_event_type event);
1000
1001 static inline struct tu_descriptor_state *
1002 tu_get_descriptors_state(struct tu_cmd_buffer *cmd_buffer,
1003 VkPipelineBindPoint bind_point)
1004 {
1005 return &cmd_buffer->descriptors[bind_point];
1006 }
1007
1008 struct tu_event
1009 {
1010 struct tu_bo bo;
1011 };
1012
1013 struct tu_shader_module
1014 {
1015 unsigned char sha1[20];
1016
1017 uint32_t code_size;
1018 const uint32_t *code[0];
1019 };
1020
1021 struct tu_push_constant_range
1022 {
1023 uint32_t lo;
1024 uint32_t count;
1025 };
1026
1027 struct tu_shader
1028 {
1029 struct ir3_shader *ir3_shader;
1030
1031 struct tu_push_constant_range push_consts;
1032 uint8_t active_desc_sets;
1033 };
1034
1035 struct tu_shader *
1036 tu_shader_create(struct tu_device *dev,
1037 gl_shader_stage stage,
1038 const VkPipelineShaderStageCreateInfo *stage_info,
1039 struct tu_pipeline_layout *layout,
1040 const VkAllocationCallbacks *alloc);
1041
1042 void
1043 tu_shader_destroy(struct tu_device *dev,
1044 struct tu_shader *shader,
1045 const VkAllocationCallbacks *alloc);
1046
1047 struct tu_program_descriptor_linkage
1048 {
1049 struct ir3_const_state const_state;
1050
1051 uint32_t constlen;
1052
1053 struct tu_push_constant_range push_consts;
1054 };
1055
1056 struct tu_pipeline
1057 {
1058 struct tu_cs cs;
1059
1060 struct tu_pipeline_layout *layout;
1061
1062 bool need_indirect_descriptor_sets;
1063 VkShaderStageFlags active_stages;
1064 uint32_t active_desc_sets;
1065
1066 struct tu_streamout_state streamout;
1067
1068 /* mask of enabled dynamic states
1069 * if BIT(i) is set, pipeline->dynamic_state[i] is *NOT* used
1070 */
1071 uint32_t dynamic_state_mask;
1072 struct tu_draw_state dynamic_state[TU_DYNAMIC_STATE_COUNT];
1073
1074 /* gras_su_cntl without line width, used for dynamic line width state */
1075 uint32_t gras_su_cntl;
1076
1077 struct
1078 {
1079 struct tu_bo binary_bo;
1080 struct tu_cs_entry state_ib;
1081 struct tu_cs_entry binning_state_ib;
1082
1083 struct tu_program_descriptor_linkage link[MESA_SHADER_STAGES];
1084 } program;
1085
1086 struct
1087 {
1088 struct tu_cs_entry state_ib;
1089 } load_state;
1090
1091 struct
1092 {
1093 struct tu_cs_entry state_ib;
1094 struct tu_cs_entry binning_state_ib;
1095 uint32_t bindings_used;
1096 } vi;
1097
1098 struct
1099 {
1100 enum pc_di_primtype primtype;
1101 bool primitive_restart;
1102 } ia;
1103
1104 struct
1105 {
1106 uint32_t patch_type;
1107 uint32_t per_vertex_output_size;
1108 uint32_t per_patch_output_size;
1109 uint32_t hs_bo_regid;
1110 uint32_t ds_bo_regid;
1111 } tess;
1112
1113 struct
1114 {
1115 struct tu_cs_entry state_ib;
1116 } rast;
1117
1118 struct
1119 {
1120 struct tu_cs_entry state_ib;
1121 } ds;
1122
1123 struct
1124 {
1125 struct tu_cs_entry state_ib;
1126 } blend;
1127
1128 struct
1129 {
1130 uint32_t local_size[3];
1131 } compute;
1132 };
1133
1134 void
1135 tu6_emit_viewport(struct tu_cs *cs, const VkViewport *viewport);
1136
1137 void
1138 tu6_emit_scissor(struct tu_cs *cs, const VkRect2D *scissor);
1139
1140 void
1141 tu6_emit_sample_locations(struct tu_cs *cs, const VkSampleLocationsInfoEXT *samp_loc);
1142
1143 void
1144 tu6_emit_depth_bias(struct tu_cs *cs,
1145 float constant_factor,
1146 float clamp,
1147 float slope_factor);
1148
1149 void tu6_emit_msaa(struct tu_cs *cs, VkSampleCountFlagBits samples);
1150
1151 void tu6_emit_window_scissor(struct tu_cs *cs, uint32_t x1, uint32_t y1, uint32_t x2, uint32_t y2);
1152
1153 void tu6_emit_window_offset(struct tu_cs *cs, uint32_t x1, uint32_t y1);
1154
1155 void
1156 tu6_emit_xs_config(struct tu_cs *cs,
1157 gl_shader_stage stage,
1158 const struct ir3_shader_variant *xs,
1159 uint64_t binary_iova);
1160
1161 void
1162 tu6_emit_vpc(struct tu_cs *cs,
1163 const struct ir3_shader_variant *vs,
1164 const struct ir3_shader_variant *hs,
1165 const struct ir3_shader_variant *ds,
1166 const struct ir3_shader_variant *gs,
1167 const struct ir3_shader_variant *fs,
1168 struct tu_streamout_state *tf);
1169
1170 void
1171 tu6_emit_fs_inputs(struct tu_cs *cs, const struct ir3_shader_variant *fs);
1172
1173 struct tu_image_view;
1174
1175 void
1176 tu_resolve_sysmem(struct tu_cmd_buffer *cmd,
1177 struct tu_cs *cs,
1178 struct tu_image_view *src,
1179 struct tu_image_view *dst,
1180 uint32_t layers,
1181 const VkRect2D *rect);
1182
1183 void
1184 tu_clear_sysmem_attachment(struct tu_cmd_buffer *cmd,
1185 struct tu_cs *cs,
1186 uint32_t a,
1187 const VkRenderPassBeginInfo *info);
1188
1189 void
1190 tu_clear_gmem_attachment(struct tu_cmd_buffer *cmd,
1191 struct tu_cs *cs,
1192 uint32_t a,
1193 const VkRenderPassBeginInfo *info);
1194
1195 void
1196 tu_load_gmem_attachment(struct tu_cmd_buffer *cmd,
1197 struct tu_cs *cs,
1198 uint32_t a,
1199 bool force_load);
1200
1201 /* expose this function to be able to emit load without checking LOAD_OP */
1202 void
1203 tu_emit_load_gmem_attachment(struct tu_cmd_buffer *cmd, struct tu_cs *cs, uint32_t a);
1204
1205 /* note: gmem store can also resolve */
1206 void
1207 tu_store_gmem_attachment(struct tu_cmd_buffer *cmd,
1208 struct tu_cs *cs,
1209 uint32_t a,
1210 uint32_t gmem_a);
1211
1212 enum tu_supported_formats {
1213 FMT_VERTEX = 1,
1214 FMT_TEXTURE = 2,
1215 FMT_COLOR = 4,
1216 };
1217
1218 struct tu_native_format
1219 {
1220 enum a6xx_format fmt : 8;
1221 enum a3xx_color_swap swap : 8;
1222 enum a6xx_tile_mode tile_mode : 8;
1223 enum tu_supported_formats supported : 8;
1224 };
1225
1226 struct tu_native_format tu6_format_vtx(VkFormat format);
1227 struct tu_native_format tu6_format_color(VkFormat format, enum a6xx_tile_mode tile_mode);
1228 struct tu_native_format tu6_format_texture(VkFormat format, enum a6xx_tile_mode tile_mode);
1229
1230 static inline enum a6xx_format
1231 tu6_base_format(VkFormat format)
1232 {
1233 /* note: tu6_format_color doesn't care about tiling for .fmt field */
1234 return tu6_format_color(format, TILE6_LINEAR).fmt;
1235 }
1236
1237 struct tu_image
1238 {
1239 VkImageType type;
1240 /* The original VkFormat provided by the client. This may not match any
1241 * of the actual surface formats.
1242 */
1243 VkFormat vk_format;
1244 VkImageAspectFlags aspects;
1245 VkImageUsageFlags usage; /**< Superset of VkImageCreateInfo::usage. */
1246 VkImageTiling tiling; /** VkImageCreateInfo::tiling */
1247 VkImageCreateFlags flags; /** VkImageCreateInfo::flags */
1248 VkExtent3D extent;
1249 uint32_t level_count;
1250 uint32_t layer_count;
1251 VkSampleCountFlagBits samples;
1252
1253 struct fdl_layout layout;
1254
1255 unsigned queue_family_mask;
1256 bool exclusive;
1257 bool shareable;
1258
1259 /* For VK_ANDROID_native_buffer, the WSI image owns the memory, */
1260 VkDeviceMemory owned_memory;
1261
1262 /* Set when bound */
1263 struct tu_bo *bo;
1264 VkDeviceSize bo_offset;
1265 };
1266
1267 static inline uint32_t
1268 tu_get_layerCount(const struct tu_image *image,
1269 const VkImageSubresourceRange *range)
1270 {
1271 return range->layerCount == VK_REMAINING_ARRAY_LAYERS
1272 ? image->layer_count - range->baseArrayLayer
1273 : range->layerCount;
1274 }
1275
1276 static inline uint32_t
1277 tu_get_levelCount(const struct tu_image *image,
1278 const VkImageSubresourceRange *range)
1279 {
1280 return range->levelCount == VK_REMAINING_MIP_LEVELS
1281 ? image->level_count - range->baseMipLevel
1282 : range->levelCount;
1283 }
1284
1285 struct tu_image_view
1286 {
1287 struct tu_image *image; /**< VkImageViewCreateInfo::image */
1288
1289 uint64_t base_addr;
1290 uint64_t ubwc_addr;
1291 uint32_t layer_size;
1292 uint32_t ubwc_layer_size;
1293
1294 /* used to determine if fast gmem store path can be used */
1295 VkExtent2D extent;
1296 bool need_y2_align;
1297
1298 bool ubwc_enabled;
1299
1300 uint32_t descriptor[A6XX_TEX_CONST_DWORDS];
1301
1302 /* Descriptor for use as a storage image as opposed to a sampled image.
1303 * This has a few differences for cube maps (e.g. type).
1304 */
1305 uint32_t storage_descriptor[A6XX_TEX_CONST_DWORDS];
1306
1307 /* pre-filled register values */
1308 uint32_t PITCH;
1309 uint32_t FLAG_BUFFER_PITCH;
1310
1311 uint32_t RB_MRT_BUF_INFO;
1312 uint32_t SP_FS_MRT_REG;
1313
1314 uint32_t SP_PS_2D_SRC_INFO;
1315 uint32_t SP_PS_2D_SRC_SIZE;
1316
1317 uint32_t RB_2D_DST_INFO;
1318
1319 uint32_t RB_BLIT_DST_INFO;
1320 };
1321
1322 struct tu_sampler_ycbcr_conversion {
1323 VkFormat format;
1324 VkSamplerYcbcrModelConversion ycbcr_model;
1325 VkSamplerYcbcrRange ycbcr_range;
1326 VkComponentMapping components;
1327 VkChromaLocation chroma_offsets[2];
1328 VkFilter chroma_filter;
1329 };
1330
1331 struct tu_sampler {
1332 uint32_t descriptor[A6XX_TEX_SAMP_DWORDS];
1333 struct tu_sampler_ycbcr_conversion *ycbcr_sampler;
1334 };
1335
1336 void
1337 tu_cs_image_ref(struct tu_cs *cs, const struct tu_image_view *iview, uint32_t layer);
1338
1339 void
1340 tu_cs_image_ref_2d(struct tu_cs *cs, const struct tu_image_view *iview, uint32_t layer, bool src);
1341
1342 void
1343 tu_cs_image_flag_ref(struct tu_cs *cs, const struct tu_image_view *iview, uint32_t layer);
1344
1345 VkResult
1346 tu_image_create(VkDevice _device,
1347 const VkImageCreateInfo *pCreateInfo,
1348 const VkAllocationCallbacks *alloc,
1349 VkImage *pImage,
1350 uint64_t modifier,
1351 const VkSubresourceLayout *plane_layouts);
1352
1353 VkResult
1354 tu_image_from_gralloc(VkDevice device_h,
1355 const VkImageCreateInfo *base_info,
1356 const VkNativeBufferANDROID *gralloc_info,
1357 const VkAllocationCallbacks *alloc,
1358 VkImage *out_image_h);
1359
1360 void
1361 tu_image_view_init(struct tu_image_view *view,
1362 const VkImageViewCreateInfo *pCreateInfo);
1363
1364 struct tu_buffer_view
1365 {
1366 uint32_t descriptor[A6XX_TEX_CONST_DWORDS];
1367
1368 struct tu_buffer *buffer;
1369 };
1370 void
1371 tu_buffer_view_init(struct tu_buffer_view *view,
1372 struct tu_device *device,
1373 const VkBufferViewCreateInfo *pCreateInfo);
1374
1375 struct tu_attachment_info
1376 {
1377 struct tu_image_view *attachment;
1378 };
1379
1380 struct tu_framebuffer
1381 {
1382 uint32_t width;
1383 uint32_t height;
1384 uint32_t layers;
1385
1386 uint32_t attachment_count;
1387 struct tu_attachment_info attachments[0];
1388 };
1389
1390 struct tu_subpass_barrier {
1391 VkPipelineStageFlags src_stage_mask;
1392 VkAccessFlags src_access_mask;
1393 VkAccessFlags dst_access_mask;
1394 bool incoherent_ccu_color, incoherent_ccu_depth;
1395 };
1396
1397 struct tu_subpass_attachment
1398 {
1399 uint32_t attachment;
1400 VkImageLayout layout;
1401 };
1402
1403 struct tu_subpass
1404 {
1405 uint32_t input_count;
1406 uint32_t color_count;
1407 struct tu_subpass_attachment *input_attachments;
1408 struct tu_subpass_attachment *color_attachments;
1409 struct tu_subpass_attachment *resolve_attachments;
1410 struct tu_subpass_attachment depth_stencil_attachment;
1411
1412 VkSampleCountFlagBits samples;
1413 bool has_external_src, has_external_dst;
1414
1415 uint32_t srgb_cntl;
1416
1417 struct tu_subpass_barrier start_barrier;
1418 };
1419
1420 struct tu_render_pass_attachment
1421 {
1422 VkFormat format;
1423 uint32_t samples;
1424 uint32_t cpp;
1425 VkImageAspectFlags clear_mask;
1426 bool load;
1427 bool store;
1428 VkImageLayout initial_layout, final_layout;
1429 int32_t gmem_offset;
1430 };
1431
1432 struct tu_render_pass
1433 {
1434 uint32_t attachment_count;
1435 uint32_t subpass_count;
1436 uint32_t gmem_pixels;
1437 uint32_t tile_align_w;
1438 struct tu_subpass_attachment *subpass_attachments;
1439 struct tu_render_pass_attachment *attachments;
1440 struct tu_subpass_barrier end_barrier;
1441 struct tu_subpass subpasses[0];
1442 };
1443
1444 struct tu_query_pool
1445 {
1446 VkQueryType type;
1447 uint32_t stride;
1448 uint64_t size;
1449 uint32_t pipeline_statistics;
1450 struct tu_bo bo;
1451 };
1452
1453 struct tu_semaphore
1454 {
1455 uint32_t syncobj;
1456 uint32_t temp_syncobj;
1457 };
1458
1459 void
1460 tu_set_descriptor_set(struct tu_cmd_buffer *cmd_buffer,
1461 VkPipelineBindPoint bind_point,
1462 struct tu_descriptor_set *set,
1463 unsigned idx);
1464
1465 void
1466 tu_update_descriptor_sets(struct tu_device *device,
1467 struct tu_cmd_buffer *cmd_buffer,
1468 VkDescriptorSet overrideSet,
1469 uint32_t descriptorWriteCount,
1470 const VkWriteDescriptorSet *pDescriptorWrites,
1471 uint32_t descriptorCopyCount,
1472 const VkCopyDescriptorSet *pDescriptorCopies);
1473
1474 void
1475 tu_update_descriptor_set_with_template(
1476 struct tu_device *device,
1477 struct tu_cmd_buffer *cmd_buffer,
1478 struct tu_descriptor_set *set,
1479 VkDescriptorUpdateTemplate descriptorUpdateTemplate,
1480 const void *pData);
1481
1482 int
1483 tu_drm_get_gpu_id(const struct tu_physical_device *dev, uint32_t *id);
1484
1485 int
1486 tu_drm_get_gmem_size(const struct tu_physical_device *dev, uint32_t *size);
1487
1488 int
1489 tu_drm_get_gmem_base(const struct tu_physical_device *dev, uint64_t *base);
1490
1491 int
1492 tu_drm_submitqueue_new(const struct tu_device *dev,
1493 int priority,
1494 uint32_t *queue_id);
1495
1496 void
1497 tu_drm_submitqueue_close(const struct tu_device *dev, uint32_t queue_id);
1498
1499 uint32_t
1500 tu_gem_new(const struct tu_device *dev, uint64_t size, uint32_t flags);
1501 uint32_t
1502 tu_gem_import_dmabuf(const struct tu_device *dev,
1503 int prime_fd,
1504 uint64_t size);
1505 int
1506 tu_gem_export_dmabuf(const struct tu_device *dev, uint32_t gem_handle);
1507 void
1508 tu_gem_close(const struct tu_device *dev, uint32_t gem_handle);
1509 uint64_t
1510 tu_gem_info_offset(const struct tu_device *dev, uint32_t gem_handle);
1511 uint64_t
1512 tu_gem_info_iova(const struct tu_device *dev, uint32_t gem_handle);
1513
1514 #define TU_DEFINE_HANDLE_CASTS(__tu_type, __VkType) \
1515 \
1516 static inline struct __tu_type *__tu_type##_from_handle(__VkType _handle) \
1517 { \
1518 return (struct __tu_type *) _handle; \
1519 } \
1520 \
1521 static inline __VkType __tu_type##_to_handle(struct __tu_type *_obj) \
1522 { \
1523 return (__VkType) _obj; \
1524 }
1525
1526 #define TU_DEFINE_NONDISP_HANDLE_CASTS(__tu_type, __VkType) \
1527 \
1528 static inline struct __tu_type *__tu_type##_from_handle(__VkType _handle) \
1529 { \
1530 return (struct __tu_type *) (uintptr_t) _handle; \
1531 } \
1532 \
1533 static inline __VkType __tu_type##_to_handle(struct __tu_type *_obj) \
1534 { \
1535 return (__VkType)(uintptr_t) _obj; \
1536 }
1537
1538 #define TU_FROM_HANDLE(__tu_type, __name, __handle) \
1539 struct __tu_type *__name = __tu_type##_from_handle(__handle)
1540
1541 TU_DEFINE_HANDLE_CASTS(tu_cmd_buffer, VkCommandBuffer)
1542 TU_DEFINE_HANDLE_CASTS(tu_device, VkDevice)
1543 TU_DEFINE_HANDLE_CASTS(tu_instance, VkInstance)
1544 TU_DEFINE_HANDLE_CASTS(tu_physical_device, VkPhysicalDevice)
1545 TU_DEFINE_HANDLE_CASTS(tu_queue, VkQueue)
1546
1547 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_cmd_pool, VkCommandPool)
1548 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_buffer, VkBuffer)
1549 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_buffer_view, VkBufferView)
1550 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_pool, VkDescriptorPool)
1551 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_set, VkDescriptorSet)
1552 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_set_layout,
1553 VkDescriptorSetLayout)
1554 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_descriptor_update_template,
1555 VkDescriptorUpdateTemplate)
1556 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_device_memory, VkDeviceMemory)
1557 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_fence, VkFence)
1558 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_event, VkEvent)
1559 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_framebuffer, VkFramebuffer)
1560 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_image, VkImage)
1561 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_image_view, VkImageView);
1562 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_pipeline_cache, VkPipelineCache)
1563 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_pipeline, VkPipeline)
1564 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_pipeline_layout, VkPipelineLayout)
1565 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_query_pool, VkQueryPool)
1566 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_render_pass, VkRenderPass)
1567 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_sampler, VkSampler)
1568 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_sampler_ycbcr_conversion, VkSamplerYcbcrConversion)
1569 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_shader_module, VkShaderModule)
1570 TU_DEFINE_NONDISP_HANDLE_CASTS(tu_semaphore, VkSemaphore)
1571
1572 #endif /* TU_PRIVATE_H */