1 /**************************************************************************
3 * Copyright 2017 Advanced Micro Devices, Inc.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * on the rights to use, copy, modify, merge, publish, distribute, sub
10 * license, and/or sell copies of the Software, and to permit persons to whom
11 * the Software is furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice (including the next
14 * paragraph) shall be included in all copies or substantial portions of the
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
21 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
22 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
23 * USE OR OTHER DEALINGS IN THE SOFTWARE.
25 **************************************************************************/
27 #include "util/u_threaded_context.h"
28 #include "util/u_cpu_detect.h"
29 #include "util/u_format.h"
30 #include "util/u_inlines.h"
31 #include "util/u_memory.h"
32 #include "util/u_upload_mgr.h"
34 /* 0 = disabled, 1 = assertions, 2 = printfs */
38 #define tc_assert assert
44 #define tc_printf printf
45 #define tc_asprintf asprintf
46 #define tc_strcmp strcmp
48 #define tc_printf(...)
49 #define tc_asprintf(...) 0
50 #define tc_strcmp(...) 0
53 #define TC_SENTINEL 0x5ca1ab1e
56 #define CALL(name) TC_CALL_##name,
57 #include "u_threaded_context_calls.h"
62 typedef void (*tc_execute
)(struct pipe_context
*pipe
, union tc_payload
*payload
);
64 static const tc_execute execute_func
[TC_NUM_CALLS
];
67 tc_batch_check(struct tc_batch
*batch
)
69 tc_assert(batch
->sentinel
== TC_SENTINEL
);
70 tc_assert(batch
->num_total_call_slots
<= TC_CALLS_PER_BATCH
);
74 tc_debug_check(struct threaded_context
*tc
)
76 for (unsigned i
= 0; i
< TC_MAX_BATCHES
; i
++) {
77 tc_batch_check(&tc
->batch_slots
[i
]);
78 tc_assert(tc
->batch_slots
[i
].pipe
== tc
->pipe
);
83 tc_batch_execute(void *job
, int thread_index
)
85 struct tc_batch
*batch
= job
;
86 struct pipe_context
*pipe
= batch
->pipe
;
87 struct tc_call
*last
= &batch
->call
[batch
->num_total_call_slots
];
89 tc_batch_check(batch
);
91 for (struct tc_call
*iter
= batch
->call
; iter
!= last
;
92 iter
+= iter
->num_call_slots
) {
93 tc_assert(iter
->sentinel
== TC_SENTINEL
);
94 execute_func
[iter
->call_id
](pipe
, &iter
->payload
);
97 tc_batch_check(batch
);
98 batch
->num_total_call_slots
= 0;
102 tc_batch_flush(struct threaded_context
*tc
)
104 struct tc_batch
*next
= &tc
->batch_slots
[tc
->next
];
106 tc_assert(next
->num_total_call_slots
!= 0);
107 tc_batch_check(next
);
109 p_atomic_add(&tc
->num_offloaded_slots
, next
->num_total_call_slots
);
111 util_queue_add_job(&tc
->queue
, next
, &next
->fence
, tc_batch_execute
,
114 tc
->next
= (tc
->next
+ 1) % TC_MAX_BATCHES
;
117 /* This is the function that adds variable-sized calls into the current
118 * batch. It also flushes the batch if there is not enough space there.
119 * All other higher-level "add" functions use it.
121 static union tc_payload
*
122 tc_add_sized_call(struct threaded_context
*tc
, enum tc_call_id id
,
123 unsigned payload_size
)
125 struct tc_batch
*next
= &tc
->batch_slots
[tc
->next
];
126 unsigned total_size
= offsetof(struct tc_call
, payload
) + payload_size
;
127 unsigned num_call_slots
= DIV_ROUND_UP(total_size
, sizeof(struct tc_call
));
131 if (unlikely(next
->num_total_call_slots
+ num_call_slots
> TC_CALLS_PER_BATCH
)) {
133 next
= &tc
->batch_slots
[tc
->next
];
134 tc_assert(next
->num_total_call_slots
== 0);
137 tc_assert(util_queue_fence_is_signalled(&next
->fence
));
139 struct tc_call
*call
= &next
->call
[next
->num_total_call_slots
];
140 next
->num_total_call_slots
+= num_call_slots
;
142 call
->sentinel
= TC_SENTINEL
;
144 call
->num_call_slots
= num_call_slots
;
147 return &call
->payload
;
150 #define tc_add_struct_typed_call(tc, execute, type) \
151 ((struct type*)tc_add_sized_call(tc, execute, sizeof(struct type)))
153 #define tc_add_slot_based_call(tc, execute, type, num_slots) \
154 ((struct type*)tc_add_sized_call(tc, execute, \
155 sizeof(struct type) + \
156 sizeof(((struct type*)NULL)->slot[0]) * \
159 static union tc_payload
*
160 tc_add_small_call(struct threaded_context
*tc
, enum tc_call_id id
)
162 return tc_add_sized_call(tc
, id
, 0);
166 _tc_sync(struct threaded_context
*tc
, const char *info
, const char *func
)
168 struct tc_batch
*last
= &tc
->batch_slots
[tc
->last
];
169 struct tc_batch
*next
= &tc
->batch_slots
[tc
->next
];
174 /* Only wait for queued calls... */
175 if (!util_queue_fence_is_signalled(&last
->fence
)) {
176 util_queue_fence_wait(&last
->fence
);
182 /* .. and execute unflushed calls directly. */
183 if (next
->num_total_call_slots
) {
184 p_atomic_add(&tc
->num_direct_slots
, next
->num_total_call_slots
);
185 tc_batch_execute(next
, 0);
190 p_atomic_inc(&tc
->num_syncs
);
192 if (tc_strcmp(func
, "tc_destroy") != 0)
193 tc_printf("sync %s %s\n", func
, info
);
199 #define tc_sync(tc) _tc_sync(tc, "", __func__)
200 #define tc_sync_msg(tc, info) _tc_sync(tc, info, __func__)
203 tc_set_resource_reference(struct pipe_resource
**dst
, struct pipe_resource
*src
)
206 pipe_resource_reference(dst
, src
);
210 threaded_resource_init(struct pipe_resource
*res
)
212 struct threaded_resource
*tres
= threaded_resource(res
);
214 tres
->latest
= &tres
->b
;
215 util_range_init(&tres
->valid_buffer_range
);
216 tres
->base_valid_buffer_range
= &tres
->valid_buffer_range
;
217 tres
->is_shared
= false;
218 tres
->is_user_ptr
= false;
222 threaded_resource_deinit(struct pipe_resource
*res
)
224 struct threaded_resource
*tres
= threaded_resource(res
);
226 if (tres
->latest
!= &tres
->b
)
227 pipe_resource_reference(&tres
->latest
, NULL
);
228 util_range_destroy(&tres
->valid_buffer_range
);
231 struct pipe_context
*
232 threaded_context_unwrap_sync(struct pipe_context
*pipe
)
234 if (!pipe
|| !pipe
->priv
)
237 tc_sync(threaded_context(pipe
));
238 return (struct pipe_context
*)pipe
->priv
;
242 /********************************************************************
246 #define TC_FUNC1(func, m_payload, qualifier, type, deref, deref2) \
248 tc_call_##func(struct pipe_context *pipe, union tc_payload *payload) \
250 pipe->func(pipe, deref2((type*)payload)); \
254 tc_##func(struct pipe_context *_pipe, qualifier type deref param) \
256 struct threaded_context *tc = threaded_context(_pipe); \
257 type *p = (type*)tc_add_sized_call(tc, TC_CALL_##func, sizeof(type)); \
261 TC_FUNC1(set_active_query_state
, flags
, , boolean
, , *)
263 TC_FUNC1(set_blend_color
, blend_color
, const, struct pipe_blend_color
, *, )
264 TC_FUNC1(set_stencil_ref
, stencil_ref
, const, struct pipe_stencil_ref
, *, )
265 TC_FUNC1(set_clip_state
, clip_state
, const, struct pipe_clip_state
, *, )
266 TC_FUNC1(set_sample_mask
, sample_mask
, , unsigned, , *)
267 TC_FUNC1(set_min_samples
, min_samples
, , unsigned, , *)
268 TC_FUNC1(set_polygon_stipple
, polygon_stipple
, const, struct pipe_poly_stipple
, *, )
270 TC_FUNC1(texture_barrier
, flags
, , unsigned, , *)
271 TC_FUNC1(memory_barrier
, flags
, , unsigned, , *)
274 /********************************************************************
278 static struct pipe_query
*
279 tc_create_query(struct pipe_context
*_pipe
, unsigned query_type
,
282 struct threaded_context
*tc
= threaded_context(_pipe
);
283 struct pipe_context
*pipe
= tc
->pipe
;
285 return pipe
->create_query(pipe
, query_type
, index
);
288 static struct pipe_query
*
289 tc_create_batch_query(struct pipe_context
*_pipe
, unsigned num_queries
,
290 unsigned *query_types
)
292 struct threaded_context
*tc
= threaded_context(_pipe
);
293 struct pipe_context
*pipe
= tc
->pipe
;
295 return pipe
->create_batch_query(pipe
, num_queries
, query_types
);
299 tc_call_destroy_query(struct pipe_context
*pipe
, union tc_payload
*payload
)
301 pipe
->destroy_query(pipe
, payload
->query
);
305 tc_destroy_query(struct pipe_context
*_pipe
, struct pipe_query
*query
)
307 struct threaded_context
*tc
= threaded_context(_pipe
);
308 struct threaded_query
*tq
= threaded_query(query
);
310 if (tq
->head_unflushed
.next
)
311 LIST_DEL(&tq
->head_unflushed
);
313 tc_add_small_call(tc
, TC_CALL_destroy_query
)->query
= query
;
317 tc_call_begin_query(struct pipe_context
*pipe
, union tc_payload
*payload
)
319 pipe
->begin_query(pipe
, payload
->query
);
323 tc_begin_query(struct pipe_context
*_pipe
, struct pipe_query
*query
)
325 struct threaded_context
*tc
= threaded_context(_pipe
);
326 union tc_payload
*payload
= tc_add_small_call(tc
, TC_CALL_begin_query
);
328 payload
->query
= query
;
329 return true; /* we don't care about the return value for this call */
333 tc_call_end_query(struct pipe_context
*pipe
, union tc_payload
*payload
)
335 pipe
->end_query(pipe
, payload
->query
);
339 tc_end_query(struct pipe_context
*_pipe
, struct pipe_query
*query
)
341 struct threaded_context
*tc
= threaded_context(_pipe
);
342 struct threaded_query
*tq
= threaded_query(query
);
343 union tc_payload
*payload
= tc_add_small_call(tc
, TC_CALL_end_query
);
345 payload
->query
= query
;
348 if (!tq
->head_unflushed
.next
)
349 LIST_ADD(&tq
->head_unflushed
, &tc
->unflushed_queries
);
351 return true; /* we don't care about the return value for this call */
355 tc_get_query_result(struct pipe_context
*_pipe
,
356 struct pipe_query
*query
, boolean wait
,
357 union pipe_query_result
*result
)
359 struct threaded_context
*tc
= threaded_context(_pipe
);
360 struct threaded_query
*tq
= threaded_query(query
);
361 struct pipe_context
*pipe
= tc
->pipe
;
364 tc_sync_msg(tc
, wait
? "wait" : "nowait");
366 bool success
= pipe
->get_query_result(pipe
, query
, wait
, result
);
370 if (tq
->head_unflushed
.next
)
371 LIST_DEL(&tq
->head_unflushed
);
376 struct tc_query_result_resource
{
377 struct pipe_query
*query
;
379 enum pipe_query_value_type result_type
;
381 struct pipe_resource
*resource
;
386 tc_call_get_query_result_resource(struct pipe_context
*pipe
,
387 union tc_payload
*payload
)
389 struct tc_query_result_resource
*p
= (struct tc_query_result_resource
*)payload
;
391 pipe
->get_query_result_resource(pipe
, p
->query
, p
->wait
, p
->result_type
,
392 p
->index
, p
->resource
, p
->offset
);
393 pipe_resource_reference(&p
->resource
, NULL
);
397 tc_get_query_result_resource(struct pipe_context
*_pipe
,
398 struct pipe_query
*query
, boolean wait
,
399 enum pipe_query_value_type result_type
, int index
,
400 struct pipe_resource
*resource
, unsigned offset
)
402 struct threaded_context
*tc
= threaded_context(_pipe
);
403 struct tc_query_result_resource
*p
=
404 tc_add_struct_typed_call(tc
, TC_CALL_get_query_result_resource
,
405 tc_query_result_resource
);
409 p
->result_type
= result_type
;
411 tc_set_resource_reference(&p
->resource
, resource
);
415 struct tc_render_condition
{
416 struct pipe_query
*query
;
422 tc_call_render_condition(struct pipe_context
*pipe
, union tc_payload
*payload
)
424 struct tc_render_condition
*p
= (struct tc_render_condition
*)payload
;
425 pipe
->render_condition(pipe
, p
->query
, p
->condition
, p
->mode
);
429 tc_render_condition(struct pipe_context
*_pipe
,
430 struct pipe_query
*query
, boolean condition
,
431 enum pipe_render_cond_flag mode
)
433 struct threaded_context
*tc
= threaded_context(_pipe
);
434 struct tc_render_condition
*p
=
435 tc_add_struct_typed_call(tc
, TC_CALL_render_condition
, tc_render_condition
);
438 p
->condition
= condition
;
443 /********************************************************************
444 * constant (immutable) states
447 #define TC_CSO_CREATE(name, sname) \
449 tc_create_##name##_state(struct pipe_context *_pipe, \
450 const struct pipe_##sname##_state *state) \
452 struct pipe_context *pipe = threaded_context(_pipe)->pipe; \
453 return pipe->create_##name##_state(pipe, state); \
456 #define TC_CSO_BIND(name) TC_FUNC1(bind_##name##_state, cso, , void *, , *)
457 #define TC_CSO_DELETE(name) TC_FUNC1(delete_##name##_state, cso, , void *, , *)
459 #define TC_CSO_WHOLE2(name, sname) \
460 TC_CSO_CREATE(name, sname) \
464 #define TC_CSO_WHOLE(name) TC_CSO_WHOLE2(name, name)
467 TC_CSO_WHOLE(rasterizer
)
468 TC_CSO_WHOLE(depth_stencil_alpha
)
469 TC_CSO_WHOLE(compute
)
470 TC_CSO_WHOLE2(fs
, shader
)
471 TC_CSO_WHOLE2(vs
, shader
)
472 TC_CSO_WHOLE2(gs
, shader
)
473 TC_CSO_WHOLE2(tcs
, shader
)
474 TC_CSO_WHOLE2(tes
, shader
)
475 TC_CSO_CREATE(sampler
, sampler
)
476 TC_CSO_DELETE(sampler
)
477 TC_CSO_BIND(vertex_elements
)
478 TC_CSO_DELETE(vertex_elements
)
481 tc_create_vertex_elements_state(struct pipe_context
*_pipe
, unsigned count
,
482 const struct pipe_vertex_element
*elems
)
484 struct pipe_context
*pipe
= threaded_context(_pipe
)->pipe
;
486 return pipe
->create_vertex_elements_state(pipe
, count
, elems
);
489 struct tc_sampler_states
{
490 ubyte shader
, start
, count
;
491 void *slot
[0]; /* more will be allocated if needed */
495 tc_call_bind_sampler_states(struct pipe_context
*pipe
, union tc_payload
*payload
)
497 struct tc_sampler_states
*p
= (struct tc_sampler_states
*)payload
;
498 pipe
->bind_sampler_states(pipe
, p
->shader
, p
->start
, p
->count
, p
->slot
);
502 tc_bind_sampler_states(struct pipe_context
*_pipe
,
503 enum pipe_shader_type shader
,
504 unsigned start
, unsigned count
, void **states
)
509 struct threaded_context
*tc
= threaded_context(_pipe
);
510 struct tc_sampler_states
*p
=
511 tc_add_slot_based_call(tc
, TC_CALL_bind_sampler_states
, tc_sampler_states
, count
);
516 memcpy(p
->slot
, states
, count
* sizeof(states
[0]));
520 /********************************************************************
525 tc_call_set_framebuffer_state(struct pipe_context
*pipe
, union tc_payload
*payload
)
527 struct pipe_framebuffer_state
*p
= (struct pipe_framebuffer_state
*)payload
;
529 pipe
->set_framebuffer_state(pipe
, p
);
531 unsigned nr_cbufs
= p
->nr_cbufs
;
532 for (unsigned i
= 0; i
< nr_cbufs
; i
++)
533 pipe_surface_reference(&p
->cbufs
[i
], NULL
);
534 pipe_surface_reference(&p
->zsbuf
, NULL
);
538 tc_set_framebuffer_state(struct pipe_context
*_pipe
,
539 const struct pipe_framebuffer_state
*fb
)
541 struct threaded_context
*tc
= threaded_context(_pipe
);
542 struct pipe_framebuffer_state
*p
=
543 tc_add_struct_typed_call(tc
, TC_CALL_set_framebuffer_state
,
544 pipe_framebuffer_state
);
545 unsigned nr_cbufs
= fb
->nr_cbufs
;
547 p
->width
= fb
->width
;
548 p
->height
= fb
->height
;
549 p
->samples
= fb
->samples
;
550 p
->layers
= fb
->layers
;
551 p
->nr_cbufs
= nr_cbufs
;
553 for (unsigned i
= 0; i
< nr_cbufs
; i
++) {
555 pipe_surface_reference(&p
->cbufs
[i
], fb
->cbufs
[i
]);
558 pipe_surface_reference(&p
->zsbuf
, fb
->zsbuf
);
562 tc_call_set_tess_state(struct pipe_context
*pipe
, union tc_payload
*payload
)
564 float *p
= (float*)payload
;
565 pipe
->set_tess_state(pipe
, p
, p
+ 4);
569 tc_set_tess_state(struct pipe_context
*_pipe
,
570 const float default_outer_level
[4],
571 const float default_inner_level
[2])
573 struct threaded_context
*tc
= threaded_context(_pipe
);
574 float *p
= (float*)tc_add_sized_call(tc
, TC_CALL_set_tess_state
,
577 memcpy(p
, default_outer_level
, 4 * sizeof(float));
578 memcpy(p
+ 4, default_inner_level
, 2 * sizeof(float));
581 struct tc_constant_buffer
{
583 struct pipe_constant_buffer cb
;
587 tc_call_set_constant_buffer(struct pipe_context
*pipe
, union tc_payload
*payload
)
589 struct tc_constant_buffer
*p
= (struct tc_constant_buffer
*)payload
;
591 pipe
->set_constant_buffer(pipe
,
595 pipe_resource_reference(&p
->cb
.buffer
, NULL
);
599 tc_set_constant_buffer(struct pipe_context
*_pipe
,
600 enum pipe_shader_type shader
, uint index
,
601 const struct pipe_constant_buffer
*cb
)
603 struct threaded_context
*tc
= threaded_context(_pipe
);
604 struct pipe_resource
*buffer
= NULL
;
607 /* This must be done before adding set_constant_buffer, because it could
608 * generate e.g. transfer_unmap and flush partially-uninitialized
609 * set_constant_buffer to the driver if it was done afterwards.
611 if (cb
&& cb
->user_buffer
) {
612 u_upload_data(tc
->base
.const_uploader
, 0, cb
->buffer_size
, 64,
613 cb
->user_buffer
, &offset
, &buffer
);
616 struct tc_constant_buffer
*p
=
617 tc_add_struct_typed_call(tc
, TC_CALL_set_constant_buffer
,
623 if (cb
->user_buffer
) {
624 p
->cb
.buffer_size
= cb
->buffer_size
;
625 p
->cb
.user_buffer
= NULL
;
626 p
->cb
.buffer_offset
= offset
;
627 p
->cb
.buffer
= buffer
;
629 tc_set_resource_reference(&p
->cb
.buffer
,
631 memcpy(&p
->cb
, cb
, sizeof(*cb
));
634 memset(&p
->cb
, 0, sizeof(*cb
));
640 struct pipe_scissor_state slot
[0]; /* more will be allocated if needed */
644 tc_call_set_scissor_states(struct pipe_context
*pipe
, union tc_payload
*payload
)
646 struct tc_scissors
*p
= (struct tc_scissors
*)payload
;
647 pipe
->set_scissor_states(pipe
, p
->start
, p
->count
, p
->slot
);
651 tc_set_scissor_states(struct pipe_context
*_pipe
,
652 unsigned start
, unsigned count
,
653 const struct pipe_scissor_state
*states
)
655 struct threaded_context
*tc
= threaded_context(_pipe
);
656 struct tc_scissors
*p
=
657 tc_add_slot_based_call(tc
, TC_CALL_set_scissor_states
, tc_scissors
, count
);
661 memcpy(&p
->slot
, states
, count
* sizeof(states
[0]));
664 struct tc_viewports
{
666 struct pipe_viewport_state slot
[0]; /* more will be allocated if needed */
670 tc_call_set_viewport_states(struct pipe_context
*pipe
, union tc_payload
*payload
)
672 struct tc_viewports
*p
= (struct tc_viewports
*)payload
;
673 pipe
->set_viewport_states(pipe
, p
->start
, p
->count
, p
->slot
);
677 tc_set_viewport_states(struct pipe_context
*_pipe
,
678 unsigned start
, unsigned count
,
679 const struct pipe_viewport_state
*states
)
684 struct threaded_context
*tc
= threaded_context(_pipe
);
685 struct tc_viewports
*p
=
686 tc_add_slot_based_call(tc
, TC_CALL_set_viewport_states
, tc_viewports
, count
);
690 memcpy(&p
->slot
, states
, count
* sizeof(states
[0]));
693 struct tc_window_rects
{
696 struct pipe_scissor_state slot
[0]; /* more will be allocated if needed */
700 tc_call_set_window_rectangles(struct pipe_context
*pipe
,
701 union tc_payload
*payload
)
703 struct tc_window_rects
*p
= (struct tc_window_rects
*)payload
;
704 pipe
->set_window_rectangles(pipe
, p
->include
, p
->count
, p
->slot
);
708 tc_set_window_rectangles(struct pipe_context
*_pipe
, boolean include
,
710 const struct pipe_scissor_state
*rects
)
712 struct threaded_context
*tc
= threaded_context(_pipe
);
713 struct tc_window_rects
*p
=
714 tc_add_slot_based_call(tc
, TC_CALL_set_window_rectangles
, tc_window_rects
, count
);
716 p
->include
= include
;
718 memcpy(p
->slot
, rects
, count
* sizeof(rects
[0]));
721 struct tc_sampler_views
{
722 ubyte shader
, start
, count
;
723 struct pipe_sampler_view
*slot
[0]; /* more will be allocated if needed */
727 tc_call_set_sampler_views(struct pipe_context
*pipe
, union tc_payload
*payload
)
729 struct tc_sampler_views
*p
= (struct tc_sampler_views
*)payload
;
730 unsigned count
= p
->count
;
732 pipe
->set_sampler_views(pipe
, p
->shader
, p
->start
, p
->count
, p
->slot
);
733 for (unsigned i
= 0; i
< count
; i
++)
734 pipe_sampler_view_reference(&p
->slot
[i
], NULL
);
738 tc_set_sampler_views(struct pipe_context
*_pipe
,
739 enum pipe_shader_type shader
,
740 unsigned start
, unsigned count
,
741 struct pipe_sampler_view
**views
)
746 struct threaded_context
*tc
= threaded_context(_pipe
);
747 struct tc_sampler_views
*p
=
748 tc_add_slot_based_call(tc
, TC_CALL_set_sampler_views
, tc_sampler_views
, count
);
755 for (unsigned i
= 0; i
< count
; i
++) {
757 pipe_sampler_view_reference(&p
->slot
[i
], views
[i
]);
760 memset(p
->slot
, 0, count
* sizeof(views
[0]));
764 struct tc_shader_images
{
765 ubyte shader
, start
, count
;
767 struct pipe_image_view slot
[0]; /* more will be allocated if needed */
771 tc_call_set_shader_images(struct pipe_context
*pipe
, union tc_payload
*payload
)
773 struct tc_shader_images
*p
= (struct tc_shader_images
*)payload
;
774 unsigned count
= p
->count
;
777 pipe
->set_shader_images(pipe
, p
->shader
, p
->start
, p
->count
, NULL
);
781 pipe
->set_shader_images(pipe
, p
->shader
, p
->start
, p
->count
, p
->slot
);
783 for (unsigned i
= 0; i
< count
; i
++)
784 pipe_resource_reference(&p
->slot
[i
].resource
, NULL
);
788 tc_set_shader_images(struct pipe_context
*_pipe
,
789 enum pipe_shader_type shader
,
790 unsigned start
, unsigned count
,
791 const struct pipe_image_view
*images
)
796 struct threaded_context
*tc
= threaded_context(_pipe
);
797 struct tc_shader_images
*p
=
798 tc_add_slot_based_call(tc
, TC_CALL_set_shader_images
, tc_shader_images
,
804 p
->unbind
= images
== NULL
;
807 for (unsigned i
= 0; i
< count
; i
++) {
808 tc_set_resource_reference(&p
->slot
[i
].resource
, images
[i
].resource
);
810 if (images
[i
].access
& PIPE_IMAGE_ACCESS_WRITE
&&
811 images
[i
].resource
&&
812 images
[i
].resource
->target
== PIPE_BUFFER
) {
813 struct threaded_resource
*tres
=
814 threaded_resource(images
[i
].resource
);
816 util_range_add(&tres
->valid_buffer_range
, images
[i
].u
.buf
.offset
,
817 images
[i
].u
.buf
.offset
+ images
[i
].u
.buf
.size
);
820 memcpy(p
->slot
, images
, count
* sizeof(images
[0]));
824 struct tc_shader_buffers
{
825 ubyte shader
, start
, count
;
827 struct pipe_shader_buffer slot
[0]; /* more will be allocated if needed */
831 tc_call_set_shader_buffers(struct pipe_context
*pipe
, union tc_payload
*payload
)
833 struct tc_shader_buffers
*p
= (struct tc_shader_buffers
*)payload
;
834 unsigned count
= p
->count
;
837 pipe
->set_shader_buffers(pipe
, p
->shader
, p
->start
, p
->count
, NULL
);
841 pipe
->set_shader_buffers(pipe
, p
->shader
, p
->start
, p
->count
, p
->slot
);
843 for (unsigned i
= 0; i
< count
; i
++)
844 pipe_resource_reference(&p
->slot
[i
].buffer
, NULL
);
848 tc_set_shader_buffers(struct pipe_context
*_pipe
,
849 enum pipe_shader_type shader
,
850 unsigned start
, unsigned count
,
851 const struct pipe_shader_buffer
*buffers
)
856 struct threaded_context
*tc
= threaded_context(_pipe
);
857 struct tc_shader_buffers
*p
=
858 tc_add_slot_based_call(tc
, TC_CALL_set_shader_buffers
, tc_shader_buffers
,
859 buffers
? count
: 0);
864 p
->unbind
= buffers
== NULL
;
867 for (unsigned i
= 0; i
< count
; i
++) {
868 struct pipe_shader_buffer
*dst
= &p
->slot
[i
];
869 const struct pipe_shader_buffer
*src
= buffers
+ i
;
871 tc_set_resource_reference(&dst
->buffer
, src
->buffer
);
872 dst
->buffer_offset
= src
->buffer_offset
;
873 dst
->buffer_size
= src
->buffer_size
;
876 struct threaded_resource
*tres
= threaded_resource(src
->buffer
);
878 util_range_add(&tres
->valid_buffer_range
, src
->buffer_offset
,
879 src
->buffer_offset
+ src
->buffer_size
);
885 struct tc_vertex_buffers
{
888 struct pipe_vertex_buffer slot
[0]; /* more will be allocated if needed */
892 tc_call_set_vertex_buffers(struct pipe_context
*pipe
, union tc_payload
*payload
)
894 struct tc_vertex_buffers
*p
= (struct tc_vertex_buffers
*)payload
;
895 unsigned count
= p
->count
;
898 pipe
->set_vertex_buffers(pipe
, p
->start
, count
, NULL
);
902 for (unsigned i
= 0; i
< count
; i
++)
903 tc_assert(!p
->slot
[i
].is_user_buffer
);
905 pipe
->set_vertex_buffers(pipe
, p
->start
, count
, p
->slot
);
906 for (unsigned i
= 0; i
< count
; i
++)
907 pipe_resource_reference(&p
->slot
[i
].buffer
.resource
, NULL
);
911 tc_set_vertex_buffers(struct pipe_context
*_pipe
,
912 unsigned start
, unsigned count
,
913 const struct pipe_vertex_buffer
*buffers
)
915 struct threaded_context
*tc
= threaded_context(_pipe
);
921 struct tc_vertex_buffers
*p
=
922 tc_add_slot_based_call(tc
, TC_CALL_set_vertex_buffers
, tc_vertex_buffers
, count
);
927 for (unsigned i
= 0; i
< count
; i
++) {
928 struct pipe_vertex_buffer
*dst
= &p
->slot
[i
];
929 const struct pipe_vertex_buffer
*src
= buffers
+ i
;
931 tc_assert(!src
->is_user_buffer
);
932 dst
->stride
= src
->stride
;
933 dst
->is_user_buffer
= false;
934 tc_set_resource_reference(&dst
->buffer
.resource
,
935 src
->buffer
.resource
);
936 dst
->buffer_offset
= src
->buffer_offset
;
939 struct tc_vertex_buffers
*p
=
940 tc_add_slot_based_call(tc
, TC_CALL_set_vertex_buffers
, tc_vertex_buffers
, 0);
947 struct tc_stream_outputs
{
949 struct pipe_stream_output_target
*targets
[PIPE_MAX_SO_BUFFERS
];
950 unsigned offsets
[PIPE_MAX_SO_BUFFERS
];
954 tc_call_set_stream_output_targets(struct pipe_context
*pipe
, union tc_payload
*payload
)
956 struct tc_stream_outputs
*p
= (struct tc_stream_outputs
*)payload
;
957 unsigned count
= p
->count
;
959 pipe
->set_stream_output_targets(pipe
, count
, p
->targets
, p
->offsets
);
960 for (unsigned i
= 0; i
< count
; i
++)
961 pipe_so_target_reference(&p
->targets
[i
], NULL
);
965 tc_set_stream_output_targets(struct pipe_context
*_pipe
,
967 struct pipe_stream_output_target
**tgs
,
968 const unsigned *offsets
)
970 struct threaded_context
*tc
= threaded_context(_pipe
);
971 struct tc_stream_outputs
*p
=
972 tc_add_struct_typed_call(tc
, TC_CALL_set_stream_output_targets
,
975 for (unsigned i
= 0; i
< count
; i
++) {
976 p
->targets
[i
] = NULL
;
977 pipe_so_target_reference(&p
->targets
[i
], tgs
[i
]);
980 memcpy(p
->offsets
, offsets
, count
* sizeof(unsigned));
984 tc_set_compute_resources(struct pipe_context
*_pipe
, unsigned start
,
985 unsigned count
, struct pipe_surface
**resources
)
987 struct threaded_context
*tc
= threaded_context(_pipe
);
988 struct pipe_context
*pipe
= tc
->pipe
;
991 pipe
->set_compute_resources(pipe
, start
, count
, resources
);
995 tc_set_global_binding(struct pipe_context
*_pipe
, unsigned first
,
996 unsigned count
, struct pipe_resource
**resources
,
999 struct threaded_context
*tc
= threaded_context(_pipe
);
1000 struct pipe_context
*pipe
= tc
->pipe
;
1003 pipe
->set_global_binding(pipe
, first
, count
, resources
, handles
);
1007 /********************************************************************
1011 static struct pipe_surface
*
1012 tc_create_surface(struct pipe_context
*_pipe
,
1013 struct pipe_resource
*resource
,
1014 const struct pipe_surface
*surf_tmpl
)
1016 struct pipe_context
*pipe
= threaded_context(_pipe
)->pipe
;
1017 struct pipe_surface
*view
=
1018 pipe
->create_surface(pipe
, resource
, surf_tmpl
);
1021 view
->context
= _pipe
;
1026 tc_surface_destroy(struct pipe_context
*_pipe
,
1027 struct pipe_surface
*surf
)
1029 struct pipe_context
*pipe
= threaded_context(_pipe
)->pipe
;
1031 pipe
->surface_destroy(pipe
, surf
);
1034 static struct pipe_sampler_view
*
1035 tc_create_sampler_view(struct pipe_context
*_pipe
,
1036 struct pipe_resource
*resource
,
1037 const struct pipe_sampler_view
*templ
)
1039 struct pipe_context
*pipe
= threaded_context(_pipe
)->pipe
;
1040 struct pipe_sampler_view
*view
=
1041 pipe
->create_sampler_view(pipe
, resource
, templ
);
1044 view
->context
= _pipe
;
1049 tc_sampler_view_destroy(struct pipe_context
*_pipe
,
1050 struct pipe_sampler_view
*view
)
1052 struct pipe_context
*pipe
= threaded_context(_pipe
)->pipe
;
1054 pipe
->sampler_view_destroy(pipe
, view
);
1057 static struct pipe_stream_output_target
*
1058 tc_create_stream_output_target(struct pipe_context
*_pipe
,
1059 struct pipe_resource
*res
,
1060 unsigned buffer_offset
,
1061 unsigned buffer_size
)
1063 struct pipe_context
*pipe
= threaded_context(_pipe
)->pipe
;
1064 struct threaded_resource
*tres
= threaded_resource(res
);
1065 struct pipe_stream_output_target
*view
;
1067 tc_sync(threaded_context(_pipe
));
1068 util_range_add(&tres
->valid_buffer_range
, buffer_offset
,
1069 buffer_offset
+ buffer_size
);
1071 view
= pipe
->create_stream_output_target(pipe
, res
, buffer_offset
,
1074 view
->context
= _pipe
;
1079 tc_stream_output_target_destroy(struct pipe_context
*_pipe
,
1080 struct pipe_stream_output_target
*target
)
1082 struct pipe_context
*pipe
= threaded_context(_pipe
)->pipe
;
1084 pipe
->stream_output_target_destroy(pipe
, target
);
1088 /********************************************************************
1092 struct tc_replace_buffer_storage
{
1093 struct pipe_resource
*dst
;
1094 struct pipe_resource
*src
;
1095 tc_replace_buffer_storage_func func
;
1099 tc_call_replace_buffer_storage(struct pipe_context
*pipe
,
1100 union tc_payload
*payload
)
1102 struct tc_replace_buffer_storage
*p
=
1103 (struct tc_replace_buffer_storage
*)payload
;
1105 p
->func(pipe
, p
->dst
, p
->src
);
1106 pipe_resource_reference(&p
->dst
, NULL
);
1107 pipe_resource_reference(&p
->src
, NULL
);
1111 tc_invalidate_buffer(struct threaded_context
*tc
,
1112 struct threaded_resource
*tbuf
)
1114 /* We can't check if the buffer is idle, so we invalidate it
1115 * unconditionally. */
1116 struct pipe_screen
*screen
= tc
->base
.screen
;
1117 struct pipe_resource
*new_buf
;
1119 /* Shared, pinned, and sparse buffers can't be reallocated. */
1120 if (tbuf
->is_shared
||
1121 tbuf
->is_user_ptr
||
1122 tbuf
->b
.flags
& PIPE_RESOURCE_FLAG_SPARSE
)
1125 /* Allocate a new one. */
1126 new_buf
= screen
->resource_create(screen
, &tbuf
->b
);
1130 /* Replace the "latest" pointer. */
1131 if (tbuf
->latest
!= &tbuf
->b
)
1132 pipe_resource_reference(&tbuf
->latest
, NULL
);
1134 tbuf
->latest
= new_buf
;
1135 util_range_set_empty(&tbuf
->valid_buffer_range
);
1137 /* The valid range should point to the original buffer. */
1138 threaded_resource(new_buf
)->base_valid_buffer_range
=
1139 &tbuf
->valid_buffer_range
;
1141 /* Enqueue storage replacement of the original buffer. */
1142 struct tc_replace_buffer_storage
*p
=
1143 tc_add_struct_typed_call(tc
, TC_CALL_replace_buffer_storage
,
1144 tc_replace_buffer_storage
);
1146 p
->func
= tc
->replace_buffer_storage
;
1147 tc_set_resource_reference(&p
->dst
, &tbuf
->b
);
1148 tc_set_resource_reference(&p
->src
, new_buf
);
1153 tc_improve_map_buffer_flags(struct threaded_context
*tc
,
1154 struct threaded_resource
*tres
, unsigned usage
,
1155 unsigned offset
, unsigned size
)
1157 /* Sparse buffers can't be mapped directly and can't be reallocated
1158 * (fully invalidated). That may just be a radeonsi limitation, but
1159 * the threaded context must obey it with radeonsi.
1161 if (tres
->b
.flags
& PIPE_RESOURCE_FLAG_SPARSE
) {
1162 /* We can use DISCARD_RANGE instead of full discard. This is the only
1163 * fast path for sparse buffers that doesn't need thread synchronization.
1165 if (usage
& PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
)
1166 usage
|= PIPE_TRANSFER_DISCARD_RANGE
;
1168 /* Allow DISCARD_WHOLE_RESOURCE and infering UNSYNCHRONIZED in drivers.
1169 * The threaded context doesn't do unsychronized mappings and invalida-
1170 * tions of sparse buffers, therefore a correct driver behavior won't
1171 * result in an incorrect behavior with the threaded context.
1176 /* Handle CPU reads trivially. */
1177 if (usage
& PIPE_TRANSFER_READ
) {
1178 /* Driver aren't allowed to do buffer invalidations. */
1179 return (usage
& ~PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
) |
1180 TC_TRANSFER_MAP_NO_INVALIDATE
|
1181 TC_TRANSFER_MAP_IGNORE_VALID_RANGE
;
1184 /* See if the buffer range being mapped has never been initialized,
1185 * in which case it can be mapped unsynchronized. */
1186 if (!(usage
& PIPE_TRANSFER_UNSYNCHRONIZED
) &&
1188 !util_ranges_intersect(&tres
->valid_buffer_range
, offset
, offset
+ size
))
1189 usage
|= PIPE_TRANSFER_UNSYNCHRONIZED
;
1191 if (!(usage
& PIPE_TRANSFER_UNSYNCHRONIZED
)) {
1192 /* If discarding the entire range, discard the whole resource instead. */
1193 if (usage
& PIPE_TRANSFER_DISCARD_RANGE
&&
1194 offset
== 0 && size
== tres
->b
.width0
)
1195 usage
|= PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
;
1197 /* Discard the whole resource if needed. */
1198 if (usage
& PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
) {
1199 if (tc_invalidate_buffer(tc
, tres
))
1200 usage
|= PIPE_TRANSFER_UNSYNCHRONIZED
;
1202 usage
|= PIPE_TRANSFER_DISCARD_RANGE
; /* fallback */
1206 /* We won't need this flag anymore. */
1207 /* TODO: We might not need TC_TRANSFER_MAP_NO_INVALIDATE with this. */
1208 usage
&= ~PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
;
1210 /* GL_AMD_pinned_memory and persistent mappings can't use staging
1212 if (usage
& (PIPE_TRANSFER_UNSYNCHRONIZED
|
1213 PIPE_TRANSFER_PERSISTENT
) ||
1215 usage
&= ~PIPE_TRANSFER_DISCARD_RANGE
;
1217 /* Unsychronized buffer mappings don't have to synchronize the thread. */
1218 if (usage
& PIPE_TRANSFER_UNSYNCHRONIZED
)
1219 usage
|= TC_TRANSFER_MAP_THREADED_UNSYNC
; /* notify the driver */
1221 /* Never invalidate inside the driver and never infer "unsynchronized". */
1223 TC_TRANSFER_MAP_NO_INVALIDATE
|
1224 TC_TRANSFER_MAP_IGNORE_VALID_RANGE
;
1228 tc_transfer_map(struct pipe_context
*_pipe
,
1229 struct pipe_resource
*resource
, unsigned level
,
1230 unsigned usage
, const struct pipe_box
*box
,
1231 struct pipe_transfer
**transfer
)
1233 struct threaded_context
*tc
= threaded_context(_pipe
);
1234 struct threaded_resource
*tres
= threaded_resource(resource
);
1235 struct pipe_context
*pipe
= tc
->pipe
;
1237 if (resource
->target
== PIPE_BUFFER
) {
1238 usage
= tc_improve_map_buffer_flags(tc
, tres
, usage
, box
->x
, box
->width
);
1240 /* Do a staging transfer within the threaded context. The driver should
1241 * only get resource_copy_region.
1243 if (usage
& PIPE_TRANSFER_DISCARD_RANGE
) {
1244 struct threaded_transfer
*ttrans
= slab_alloc(&tc
->pool_transfers
);
1247 ttrans
->staging
= NULL
;
1249 u_upload_alloc(tc
->base
.stream_uploader
, 0,
1250 box
->width
+ (box
->x
% tc
->map_buffer_alignment
),
1251 64, &ttrans
->offset
, &ttrans
->staging
, (void**)&map
);
1253 slab_free(&tc
->pool_transfers
, ttrans
);
1257 tc_set_resource_reference(&ttrans
->b
.resource
, resource
);
1258 ttrans
->b
.level
= 0;
1259 ttrans
->b
.usage
= usage
;
1260 ttrans
->b
.box
= *box
;
1261 ttrans
->b
.stride
= 0;
1262 ttrans
->b
.layer_stride
= 0;
1263 *transfer
= &ttrans
->b
;
1264 return map
+ (box
->x
% tc
->map_buffer_alignment
);
1268 /* Unsychronized buffer mappings don't have to synchronize the thread. */
1269 if (!(usage
& TC_TRANSFER_MAP_THREADED_UNSYNC
))
1270 tc_sync_msg(tc
, resource
->target
!= PIPE_BUFFER
? " texture" :
1271 usage
& PIPE_TRANSFER_DISCARD_RANGE
? " discard_range" :
1272 usage
& PIPE_TRANSFER_READ
? " read" : " ??");
1274 return pipe
->transfer_map(pipe
, tres
->latest
? tres
->latest
: resource
,
1275 level
, usage
, box
, transfer
);
1278 struct tc_transfer_flush_region
{
1279 struct pipe_transfer
*transfer
;
1280 struct pipe_box box
;
1284 tc_call_transfer_flush_region(struct pipe_context
*pipe
,
1285 union tc_payload
*payload
)
1287 struct tc_transfer_flush_region
*p
=
1288 (struct tc_transfer_flush_region
*)payload
;
1290 pipe
->transfer_flush_region(pipe
, p
->transfer
, &p
->box
);
1293 struct tc_resource_copy_region
{
1294 struct pipe_resource
*dst
;
1296 unsigned dstx
, dsty
, dstz
;
1297 struct pipe_resource
*src
;
1299 struct pipe_box src_box
;
1303 tc_resource_copy_region(struct pipe_context
*_pipe
,
1304 struct pipe_resource
*dst
, unsigned dst_level
,
1305 unsigned dstx
, unsigned dsty
, unsigned dstz
,
1306 struct pipe_resource
*src
, unsigned src_level
,
1307 const struct pipe_box
*src_box
);
1310 tc_buffer_do_flush_region(struct threaded_context
*tc
,
1311 struct threaded_transfer
*ttrans
,
1312 const struct pipe_box
*box
)
1314 struct threaded_resource
*tres
= threaded_resource(ttrans
->b
.resource
);
1316 if (ttrans
->staging
) {
1317 struct pipe_box src_box
;
1319 u_box_1d(ttrans
->offset
+ box
->x
% tc
->map_buffer_alignment
,
1320 box
->width
, &src_box
);
1322 /* Copy the staging buffer into the original one. */
1323 tc_resource_copy_region(&tc
->base
, ttrans
->b
.resource
, 0, box
->x
, 0, 0,
1324 ttrans
->staging
, 0, &src_box
);
1327 util_range_add(tres
->base_valid_buffer_range
, box
->x
, box
->x
+ box
->width
);
1331 tc_transfer_flush_region(struct pipe_context
*_pipe
,
1332 struct pipe_transfer
*transfer
,
1333 const struct pipe_box
*rel_box
)
1335 struct threaded_context
*tc
= threaded_context(_pipe
);
1336 struct threaded_transfer
*ttrans
= threaded_transfer(transfer
);
1337 struct threaded_resource
*tres
= threaded_resource(transfer
->resource
);
1338 unsigned required_usage
= PIPE_TRANSFER_WRITE
|
1339 PIPE_TRANSFER_FLUSH_EXPLICIT
;
1341 if (tres
->b
.target
== PIPE_BUFFER
) {
1342 if ((transfer
->usage
& required_usage
) == required_usage
) {
1343 struct pipe_box box
;
1345 u_box_1d(transfer
->box
.x
+ rel_box
->x
, rel_box
->width
, &box
);
1346 tc_buffer_do_flush_region(tc
, ttrans
, &box
);
1349 /* Staging transfers don't send the call to the driver. */
1350 if (ttrans
->staging
)
1354 struct tc_transfer_flush_region
*p
=
1355 tc_add_struct_typed_call(tc
, TC_CALL_transfer_flush_region
,
1356 tc_transfer_flush_region
);
1357 p
->transfer
= transfer
;
1362 tc_call_transfer_unmap(struct pipe_context
*pipe
, union tc_payload
*payload
)
1364 pipe
->transfer_unmap(pipe
, payload
->transfer
);
1368 tc_transfer_unmap(struct pipe_context
*_pipe
, struct pipe_transfer
*transfer
)
1370 struct threaded_context
*tc
= threaded_context(_pipe
);
1371 struct threaded_transfer
*ttrans
= threaded_transfer(transfer
);
1372 struct threaded_resource
*tres
= threaded_resource(transfer
->resource
);
1374 if (tres
->b
.target
== PIPE_BUFFER
) {
1375 if (transfer
->usage
& PIPE_TRANSFER_WRITE
&&
1376 !(transfer
->usage
& PIPE_TRANSFER_FLUSH_EXPLICIT
))
1377 tc_buffer_do_flush_region(tc
, ttrans
, &transfer
->box
);
1379 /* Staging transfers don't send the call to the driver. */
1380 if (ttrans
->staging
) {
1381 pipe_resource_reference(&ttrans
->staging
, NULL
);
1382 pipe_resource_reference(&ttrans
->b
.resource
, NULL
);
1383 slab_free(&tc
->pool_transfers
, ttrans
);
1388 tc_add_small_call(tc
, TC_CALL_transfer_unmap
)->transfer
= transfer
;
1391 struct tc_buffer_subdata
{
1392 struct pipe_resource
*resource
;
1393 unsigned usage
, offset
, size
;
1394 char slot
[0]; /* more will be allocated if needed */
1398 tc_call_buffer_subdata(struct pipe_context
*pipe
, union tc_payload
*payload
)
1400 struct tc_buffer_subdata
*p
= (struct tc_buffer_subdata
*)payload
;
1402 pipe
->buffer_subdata(pipe
, p
->resource
, p
->usage
, p
->offset
, p
->size
,
1404 pipe_resource_reference(&p
->resource
, NULL
);
1408 tc_buffer_subdata(struct pipe_context
*_pipe
,
1409 struct pipe_resource
*resource
,
1410 unsigned usage
, unsigned offset
,
1411 unsigned size
, const void *data
)
1413 struct threaded_context
*tc
= threaded_context(_pipe
);
1414 struct threaded_resource
*tres
= threaded_resource(resource
);
1419 usage
|= PIPE_TRANSFER_WRITE
|
1420 PIPE_TRANSFER_DISCARD_RANGE
;
1422 usage
= tc_improve_map_buffer_flags(tc
, tres
, usage
, offset
, size
);
1424 /* Unsychronized and big transfers should use transfer_map. Also handle
1425 * full invalidations, because drivers aren't allowed to do them.
1427 if (usage
& (PIPE_TRANSFER_UNSYNCHRONIZED
|
1428 PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
) ||
1429 size
> TC_MAX_SUBDATA_BYTES
) {
1430 struct pipe_transfer
*transfer
;
1431 struct pipe_box box
;
1432 uint8_t *map
= NULL
;
1434 u_box_1d(offset
, size
, &box
);
1436 map
= tc_transfer_map(_pipe
, resource
, 0, usage
, &box
, &transfer
);
1438 memcpy(map
, data
, size
);
1439 tc_transfer_unmap(_pipe
, transfer
);
1444 util_range_add(&tres
->valid_buffer_range
, offset
, offset
+ size
);
1446 /* The upload is small. Enqueue it. */
1447 struct tc_buffer_subdata
*p
=
1448 tc_add_slot_based_call(tc
, TC_CALL_buffer_subdata
, tc_buffer_subdata
, size
);
1450 tc_set_resource_reference(&p
->resource
, resource
);
1454 memcpy(p
->slot
, data
, size
);
1457 struct tc_texture_subdata
{
1458 struct pipe_resource
*resource
;
1459 unsigned level
, usage
, stride
, layer_stride
;
1460 struct pipe_box box
;
1461 char slot
[0]; /* more will be allocated if needed */
1465 tc_call_texture_subdata(struct pipe_context
*pipe
, union tc_payload
*payload
)
1467 struct tc_texture_subdata
*p
= (struct tc_texture_subdata
*)payload
;
1469 pipe
->texture_subdata(pipe
, p
->resource
, p
->level
, p
->usage
, &p
->box
,
1470 p
->slot
, p
->stride
, p
->layer_stride
);
1471 pipe_resource_reference(&p
->resource
, NULL
);
1475 tc_texture_subdata(struct pipe_context
*_pipe
,
1476 struct pipe_resource
*resource
,
1477 unsigned level
, unsigned usage
,
1478 const struct pipe_box
*box
,
1479 const void *data
, unsigned stride
,
1480 unsigned layer_stride
)
1482 struct threaded_context
*tc
= threaded_context(_pipe
);
1485 assert(box
->height
>= 1);
1486 assert(box
->depth
>= 1);
1488 size
= (box
->depth
- 1) * layer_stride
+
1489 (box
->height
- 1) * stride
+
1490 box
->width
* util_format_get_blocksize(resource
->format
);
1494 /* Small uploads can be enqueued, big uploads must sync. */
1495 if (size
<= TC_MAX_SUBDATA_BYTES
) {
1496 struct tc_texture_subdata
*p
=
1497 tc_add_slot_based_call(tc
, TC_CALL_texture_subdata
, tc_texture_subdata
, size
);
1499 tc_set_resource_reference(&p
->resource
, resource
);
1504 p
->layer_stride
= layer_stride
;
1505 memcpy(p
->slot
, data
, size
);
1507 struct pipe_context
*pipe
= tc
->pipe
;
1510 pipe
->texture_subdata(pipe
, resource
, level
, usage
, box
, data
,
1511 stride
, layer_stride
);
1516 /********************************************************************
1520 #define TC_FUNC_SYNC_RET0(ret_type, func) \
1522 tc_##func(struct pipe_context *_pipe) \
1524 struct threaded_context *tc = threaded_context(_pipe); \
1525 struct pipe_context *pipe = tc->pipe; \
1527 return pipe->func(pipe); \
1530 TC_FUNC_SYNC_RET0(enum pipe_reset_status
, get_device_reset_status
)
1531 TC_FUNC_SYNC_RET0(uint64_t, get_timestamp
)
1534 tc_get_sample_position(struct pipe_context
*_pipe
,
1535 unsigned sample_count
, unsigned sample_index
,
1538 struct threaded_context
*tc
= threaded_context(_pipe
);
1539 struct pipe_context
*pipe
= tc
->pipe
;
1542 pipe
->get_sample_position(pipe
, sample_count
, sample_index
,
1547 tc_set_device_reset_callback(struct pipe_context
*_pipe
,
1548 const struct pipe_device_reset_callback
*cb
)
1550 struct threaded_context
*tc
= threaded_context(_pipe
);
1551 struct pipe_context
*pipe
= tc
->pipe
;
1554 pipe
->set_device_reset_callback(pipe
, cb
);
1557 struct tc_string_marker
{
1559 char slot
[0]; /* more will be allocated if needed */
1563 tc_call_emit_string_marker(struct pipe_context
*pipe
, union tc_payload
*payload
)
1565 struct tc_string_marker
*p
= (struct tc_string_marker
*)payload
;
1566 pipe
->emit_string_marker(pipe
, p
->slot
, p
->len
);
1570 tc_emit_string_marker(struct pipe_context
*_pipe
,
1571 const char *string
, int len
)
1573 struct threaded_context
*tc
= threaded_context(_pipe
);
1575 if (len
<= TC_MAX_STRING_MARKER_BYTES
) {
1576 struct tc_string_marker
*p
=
1577 tc_add_slot_based_call(tc
, TC_CALL_emit_string_marker
, tc_string_marker
, len
);
1579 memcpy(p
->slot
, string
, len
);
1582 struct pipe_context
*pipe
= tc
->pipe
;
1585 pipe
->emit_string_marker(pipe
, string
, len
);
1590 tc_dump_debug_state(struct pipe_context
*_pipe
, FILE *stream
,
1593 struct threaded_context
*tc
= threaded_context(_pipe
);
1594 struct pipe_context
*pipe
= tc
->pipe
;
1597 pipe
->dump_debug_state(pipe
, stream
, flags
);
1601 tc_set_debug_callback(struct pipe_context
*_pipe
,
1602 const struct pipe_debug_callback
*cb
)
1604 struct threaded_context
*tc
= threaded_context(_pipe
);
1605 struct pipe_context
*pipe
= tc
->pipe
;
1607 /* Drop all synchronous debug callbacks. Drivers are expected to be OK
1608 * with this. shader-db will use an environment variable to disable
1609 * the threaded context.
1611 if (cb
&& cb
->debug_message
&& !cb
->async
)
1615 pipe
->set_debug_callback(pipe
, cb
);
1619 tc_create_fence_fd(struct pipe_context
*_pipe
,
1620 struct pipe_fence_handle
**fence
, int fd
)
1622 struct threaded_context
*tc
= threaded_context(_pipe
);
1623 struct pipe_context
*pipe
= tc
->pipe
;
1626 pipe
->create_fence_fd(pipe
, fence
, fd
);
1630 tc_fence_server_sync(struct pipe_context
*_pipe
,
1631 struct pipe_fence_handle
*fence
)
1633 struct threaded_context
*tc
= threaded_context(_pipe
);
1634 struct pipe_context
*pipe
= tc
->pipe
;
1637 pipe
->fence_server_sync(pipe
, fence
);
1640 static struct pipe_video_codec
*
1641 tc_create_video_codec(struct pipe_context
*_pipe
,
1642 const struct pipe_video_codec
*templ
)
1644 unreachable("Threaded context should not be enabled for video APIs");
1648 static struct pipe_video_buffer
*
1649 tc_create_video_buffer(struct pipe_context
*_pipe
,
1650 const struct pipe_video_buffer
*templ
)
1652 unreachable("Threaded context should not be enabled for video APIs");
1657 /********************************************************************
1658 * draw, launch, clear, blit, copy, flush
1662 tc_flush(struct pipe_context
*_pipe
, struct pipe_fence_handle
**fence
,
1665 struct threaded_context
*tc
= threaded_context(_pipe
);
1666 struct pipe_context
*pipe
= tc
->pipe
;
1667 struct threaded_query
*tq
, *tmp
;
1669 LIST_FOR_EACH_ENTRY_SAFE(tq
, tmp
, &tc
->unflushed_queries
, head_unflushed
) {
1671 LIST_DEL(&tq
->head_unflushed
);
1674 /* TODO: deferred flushes? */
1675 tc_sync_msg(tc
, flags
& PIPE_FLUSH_END_OF_FRAME
? "end of frame" :
1676 flags
& PIPE_FLUSH_DEFERRED
? "deferred fence" : "normal");
1677 pipe
->flush(pipe
, fence
, flags
);
1680 /* This is actually variable-sized, because indirect isn't allocated if it's
1682 struct tc_full_draw_info
{
1683 struct pipe_draw_info draw
;
1684 struct pipe_draw_indirect_info indirect
;
1688 tc_call_draw_vbo(struct pipe_context
*pipe
, union tc_payload
*payload
)
1690 struct tc_full_draw_info
*info
= (struct tc_full_draw_info
*)payload
;
1692 pipe
->draw_vbo(pipe
, &info
->draw
);
1693 pipe_so_target_reference(&info
->draw
.count_from_stream_output
, NULL
);
1694 if (info
->draw
.index_size
)
1695 pipe_resource_reference(&info
->draw
.index
.resource
, NULL
);
1696 if (info
->draw
.indirect
) {
1697 pipe_resource_reference(&info
->indirect
.buffer
, NULL
);
1698 pipe_resource_reference(&info
->indirect
.indirect_draw_count
, NULL
);
1702 static struct tc_full_draw_info
*
1703 tc_add_draw_vbo(struct pipe_context
*_pipe
, bool indirect
)
1705 return (struct tc_full_draw_info
*)
1706 tc_add_sized_call(threaded_context(_pipe
), TC_CALL_draw_vbo
,
1707 indirect
? sizeof(struct tc_full_draw_info
) :
1708 sizeof(struct pipe_draw_info
));
1712 tc_draw_vbo(struct pipe_context
*_pipe
, const struct pipe_draw_info
*info
)
1714 struct threaded_context
*tc
= threaded_context(_pipe
);
1715 struct pipe_draw_indirect_info
*indirect
= info
->indirect
;
1716 unsigned index_size
= info
->index_size
;
1717 bool has_user_indices
= info
->has_user_indices
;
1719 if (index_size
&& has_user_indices
) {
1720 unsigned size
= info
->count
* index_size
;
1721 struct pipe_resource
*buffer
= NULL
;
1724 tc_assert(!indirect
);
1726 /* This must be done before adding draw_vbo, because it could generate
1727 * e.g. transfer_unmap and flush partially-uninitialized draw_vbo
1728 * to the driver if it was done afterwards.
1730 u_upload_data(tc
->base
.stream_uploader
, 0, size
, 4, info
->index
.user
,
1732 if (unlikely(!buffer
))
1735 struct tc_full_draw_info
*p
= tc_add_draw_vbo(_pipe
, false);
1736 p
->draw
.count_from_stream_output
= NULL
;
1737 pipe_so_target_reference(&p
->draw
.count_from_stream_output
,
1738 info
->count_from_stream_output
);
1739 memcpy(&p
->draw
, info
, sizeof(*info
));
1740 p
->draw
.has_user_indices
= false;
1741 p
->draw
.index
.resource
= buffer
;
1742 p
->draw
.start
= offset
/ index_size
;
1744 /* Non-indexed call or indexed with a real index buffer. */
1745 struct tc_full_draw_info
*p
= tc_add_draw_vbo(_pipe
, indirect
!= NULL
);
1746 p
->draw
.count_from_stream_output
= NULL
;
1747 pipe_so_target_reference(&p
->draw
.count_from_stream_output
,
1748 info
->count_from_stream_output
);
1750 tc_set_resource_reference(&p
->draw
.index
.resource
,
1751 info
->index
.resource
);
1753 memcpy(&p
->draw
, info
, sizeof(*info
));
1756 tc_set_resource_reference(&p
->draw
.indirect
->buffer
, indirect
->buffer
);
1757 tc_set_resource_reference(&p
->indirect
.indirect_draw_count
,
1758 indirect
->indirect_draw_count
);
1759 memcpy(&p
->indirect
, indirect
, sizeof(*indirect
));
1760 p
->draw
.indirect
= &p
->indirect
;
1766 tc_call_launch_grid(struct pipe_context
*pipe
, union tc_payload
*payload
)
1768 struct pipe_grid_info
*p
= (struct pipe_grid_info
*)payload
;
1770 pipe
->launch_grid(pipe
, p
);
1771 pipe_resource_reference(&p
->indirect
, NULL
);
1775 tc_launch_grid(struct pipe_context
*_pipe
,
1776 const struct pipe_grid_info
*info
)
1778 struct threaded_context
*tc
= threaded_context(_pipe
);
1779 struct pipe_grid_info
*p
= tc_add_struct_typed_call(tc
, TC_CALL_launch_grid
,
1781 assert(info
->input
== NULL
);
1783 tc_set_resource_reference(&p
->indirect
, info
->indirect
);
1784 memcpy(p
, info
, sizeof(*info
));
1788 tc_call_resource_copy_region(struct pipe_context
*pipe
, union tc_payload
*payload
)
1790 struct tc_resource_copy_region
*p
= (struct tc_resource_copy_region
*)payload
;
1792 pipe
->resource_copy_region(pipe
, p
->dst
, p
->dst_level
, p
->dstx
, p
->dsty
,
1793 p
->dstz
, p
->src
, p
->src_level
, &p
->src_box
);
1794 pipe_resource_reference(&p
->dst
, NULL
);
1795 pipe_resource_reference(&p
->src
, NULL
);
1799 tc_resource_copy_region(struct pipe_context
*_pipe
,
1800 struct pipe_resource
*dst
, unsigned dst_level
,
1801 unsigned dstx
, unsigned dsty
, unsigned dstz
,
1802 struct pipe_resource
*src
, unsigned src_level
,
1803 const struct pipe_box
*src_box
)
1805 struct threaded_context
*tc
= threaded_context(_pipe
);
1806 struct threaded_resource
*tdst
= threaded_resource(dst
);
1807 struct tc_resource_copy_region
*p
=
1808 tc_add_struct_typed_call(tc
, TC_CALL_resource_copy_region
,
1809 tc_resource_copy_region
);
1811 tc_set_resource_reference(&p
->dst
, dst
);
1812 p
->dst_level
= dst_level
;
1816 tc_set_resource_reference(&p
->src
, src
);
1817 p
->src_level
= src_level
;
1818 p
->src_box
= *src_box
;
1820 if (dst
->target
== PIPE_BUFFER
)
1821 util_range_add(&tdst
->valid_buffer_range
, dstx
, dstx
+ src_box
->width
);
1825 tc_call_blit(struct pipe_context
*pipe
, union tc_payload
*payload
)
1827 struct pipe_blit_info
*blit
= (struct pipe_blit_info
*)payload
;
1829 pipe
->blit(pipe
, blit
);
1830 pipe_resource_reference(&blit
->dst
.resource
, NULL
);
1831 pipe_resource_reference(&blit
->src
.resource
, NULL
);
1835 tc_blit(struct pipe_context
*_pipe
, const struct pipe_blit_info
*info
)
1837 struct threaded_context
*tc
= threaded_context(_pipe
);
1838 struct pipe_blit_info
*blit
=
1839 tc_add_struct_typed_call(tc
, TC_CALL_blit
, pipe_blit_info
);
1841 tc_set_resource_reference(&blit
->dst
.resource
, info
->dst
.resource
);
1842 tc_set_resource_reference(&blit
->src
.resource
, info
->src
.resource
);
1843 memcpy(blit
, info
, sizeof(*info
));
1846 struct tc_generate_mipmap
{
1847 struct pipe_resource
*res
;
1848 enum pipe_format format
;
1849 unsigned base_level
;
1850 unsigned last_level
;
1851 unsigned first_layer
;
1852 unsigned last_layer
;
1856 tc_call_generate_mipmap(struct pipe_context
*pipe
, union tc_payload
*payload
)
1858 struct tc_generate_mipmap
*p
= (struct tc_generate_mipmap
*)payload
;
1859 bool result
= pipe
->generate_mipmap(pipe
, p
->res
, p
->format
, p
->base_level
,
1860 p
->last_level
, p
->first_layer
,
1863 pipe_resource_reference(&p
->res
, NULL
);
1867 tc_generate_mipmap(struct pipe_context
*_pipe
,
1868 struct pipe_resource
*res
,
1869 enum pipe_format format
,
1870 unsigned base_level
,
1871 unsigned last_level
,
1872 unsigned first_layer
,
1873 unsigned last_layer
)
1875 struct threaded_context
*tc
= threaded_context(_pipe
);
1876 struct pipe_context
*pipe
= tc
->pipe
;
1877 struct pipe_screen
*screen
= pipe
->screen
;
1878 unsigned bind
= PIPE_BIND_SAMPLER_VIEW
;
1880 if (util_format_is_depth_or_stencil(format
))
1881 bind
= PIPE_BIND_DEPTH_STENCIL
;
1883 bind
= PIPE_BIND_RENDER_TARGET
;
1885 if (!screen
->is_format_supported(screen
, format
, res
->target
,
1886 res
->nr_samples
, bind
))
1889 struct tc_generate_mipmap
*p
=
1890 tc_add_struct_typed_call(tc
, TC_CALL_generate_mipmap
, tc_generate_mipmap
);
1892 tc_set_resource_reference(&p
->res
, res
);
1894 p
->base_level
= base_level
;
1895 p
->last_level
= last_level
;
1896 p
->first_layer
= first_layer
;
1897 p
->last_layer
= last_layer
;
1902 tc_call_flush_resource(struct pipe_context
*pipe
, union tc_payload
*payload
)
1904 pipe
->flush_resource(pipe
, payload
->resource
);
1905 pipe_resource_reference(&payload
->resource
, NULL
);
1909 tc_flush_resource(struct pipe_context
*_pipe
,
1910 struct pipe_resource
*resource
)
1912 struct threaded_context
*tc
= threaded_context(_pipe
);
1913 union tc_payload
*payload
= tc_add_small_call(tc
, TC_CALL_flush_resource
);
1915 tc_set_resource_reference(&payload
->resource
, resource
);
1919 tc_call_invalidate_resource(struct pipe_context
*pipe
, union tc_payload
*payload
)
1921 pipe
->invalidate_resource(pipe
, payload
->resource
);
1922 pipe_resource_reference(&payload
->resource
, NULL
);
1926 tc_invalidate_resource(struct pipe_context
*_pipe
,
1927 struct pipe_resource
*resource
)
1929 struct threaded_context
*tc
= threaded_context(_pipe
);
1931 if (resource
->target
== PIPE_BUFFER
) {
1932 tc_invalidate_buffer(tc
, threaded_resource(resource
));
1936 union tc_payload
*payload
= tc_add_small_call(tc
, TC_CALL_invalidate_resource
);
1937 tc_set_resource_reference(&payload
->resource
, resource
);
1942 union pipe_color_union color
;
1948 tc_call_clear(struct pipe_context
*pipe
, union tc_payload
*payload
)
1950 struct tc_clear
*p
= (struct tc_clear
*)payload
;
1951 pipe
->clear(pipe
, p
->buffers
, &p
->color
, p
->depth
, p
->stencil
);
1955 tc_clear(struct pipe_context
*_pipe
, unsigned buffers
,
1956 const union pipe_color_union
*color
, double depth
,
1959 struct threaded_context
*tc
= threaded_context(_pipe
);
1960 struct tc_clear
*p
= tc_add_struct_typed_call(tc
, TC_CALL_clear
, tc_clear
);
1962 p
->buffers
= buffers
;
1965 p
->stencil
= stencil
;
1969 tc_clear_render_target(struct pipe_context
*_pipe
,
1970 struct pipe_surface
*dst
,
1971 const union pipe_color_union
*color
,
1972 unsigned dstx
, unsigned dsty
,
1973 unsigned width
, unsigned height
,
1974 bool render_condition_enabled
)
1976 struct threaded_context
*tc
= threaded_context(_pipe
);
1977 struct pipe_context
*pipe
= tc
->pipe
;
1980 pipe
->clear_render_target(pipe
, dst
, color
, dstx
, dsty
, width
, height
,
1981 render_condition_enabled
);
1985 tc_clear_depth_stencil(struct pipe_context
*_pipe
,
1986 struct pipe_surface
*dst
, unsigned clear_flags
,
1987 double depth
, unsigned stencil
, unsigned dstx
,
1988 unsigned dsty
, unsigned width
, unsigned height
,
1989 bool render_condition_enabled
)
1991 struct threaded_context
*tc
= threaded_context(_pipe
);
1992 struct pipe_context
*pipe
= tc
->pipe
;
1995 pipe
->clear_depth_stencil(pipe
, dst
, clear_flags
, depth
, stencil
,
1996 dstx
, dsty
, width
, height
,
1997 render_condition_enabled
);
2000 struct tc_clear_buffer
{
2001 struct pipe_resource
*res
;
2004 char clear_value
[16];
2005 int clear_value_size
;
2009 tc_call_clear_buffer(struct pipe_context
*pipe
, union tc_payload
*payload
)
2011 struct tc_clear_buffer
*p
= (struct tc_clear_buffer
*)payload
;
2013 pipe
->clear_buffer(pipe
, p
->res
, p
->offset
, p
->size
, p
->clear_value
,
2014 p
->clear_value_size
);
2015 pipe_resource_reference(&p
->res
, NULL
);
2019 tc_clear_buffer(struct pipe_context
*_pipe
, struct pipe_resource
*res
,
2020 unsigned offset
, unsigned size
,
2021 const void *clear_value
, int clear_value_size
)
2023 struct threaded_context
*tc
= threaded_context(_pipe
);
2024 struct threaded_resource
*tres
= threaded_resource(res
);
2025 struct tc_clear_buffer
*p
=
2026 tc_add_struct_typed_call(tc
, TC_CALL_clear_buffer
, tc_clear_buffer
);
2028 tc_set_resource_reference(&p
->res
, res
);
2031 memcpy(p
->clear_value
, clear_value
, clear_value_size
);
2032 p
->clear_value_size
= clear_value_size
;
2034 util_range_add(&tres
->valid_buffer_range
, offset
, offset
+ size
);
2037 struct tc_clear_texture
{
2038 struct pipe_resource
*res
;
2040 struct pipe_box box
;
2045 tc_call_clear_texture(struct pipe_context
*pipe
, union tc_payload
*payload
)
2047 struct tc_clear_texture
*p
= (struct tc_clear_texture
*)payload
;
2049 pipe
->clear_texture(pipe
, p
->res
, p
->level
, &p
->box
, p
->data
);
2050 pipe_resource_reference(&p
->res
, NULL
);
2054 tc_clear_texture(struct pipe_context
*_pipe
, struct pipe_resource
*res
,
2055 unsigned level
, const struct pipe_box
*box
, const void *data
)
2057 struct threaded_context
*tc
= threaded_context(_pipe
);
2058 struct tc_clear_texture
*p
=
2059 tc_add_struct_typed_call(tc
, TC_CALL_clear_texture
, tc_clear_texture
);
2061 tc_set_resource_reference(&p
->res
, res
);
2064 memcpy(p
->data
, data
,
2065 util_format_get_blocksize(res
->format
));
2068 struct tc_resource_commit
{
2069 struct pipe_resource
*res
;
2071 struct pipe_box box
;
2076 tc_call_resource_commit(struct pipe_context
*pipe
, union tc_payload
*payload
)
2078 struct tc_resource_commit
*p
= (struct tc_resource_commit
*)payload
;
2080 pipe
->resource_commit(pipe
, p
->res
, p
->level
, &p
->box
, p
->commit
);
2081 pipe_resource_reference(&p
->res
, NULL
);
2085 tc_resource_commit(struct pipe_context
*_pipe
, struct pipe_resource
*res
,
2086 unsigned level
, struct pipe_box
*box
, bool commit
)
2088 struct threaded_context
*tc
= threaded_context(_pipe
);
2089 struct tc_resource_commit
*p
=
2090 tc_add_struct_typed_call(tc
, TC_CALL_resource_commit
, tc_resource_commit
);
2092 tc_set_resource_reference(&p
->res
, res
);
2096 return true; /* we don't care about the return value for this call */
2100 /********************************************************************
2105 tc_destroy(struct pipe_context
*_pipe
)
2107 struct threaded_context
*tc
= threaded_context(_pipe
);
2108 struct pipe_context
*pipe
= tc
->pipe
;
2112 if (util_queue_is_initialized(&tc
->queue
)) {
2113 util_queue_destroy(&tc
->queue
);
2115 for (unsigned i
= 0; i
< TC_MAX_BATCHES
; i
++)
2116 util_queue_fence_destroy(&tc
->batch_slots
[i
].fence
);
2119 if (tc
->base
.const_uploader
&&
2120 tc
->base
.stream_uploader
!= tc
->base
.const_uploader
)
2121 u_upload_destroy(tc
->base
.const_uploader
);
2123 if (tc
->base
.stream_uploader
)
2124 u_upload_destroy(tc
->base
.stream_uploader
);
2126 slab_destroy_child(&tc
->pool_transfers
);
2127 pipe
->destroy(pipe
);
2128 os_free_aligned(tc
);
2131 static const tc_execute execute_func
[TC_NUM_CALLS
] = {
2132 #define CALL(name) tc_call_##name,
2133 #include "u_threaded_context_calls.h"
2138 * Wrap an existing pipe_context into a threaded_context.
2140 * \param pipe pipe_context to wrap
2141 * \param parent_transfer_pool parent slab pool set up for creating pipe_-
2142 * transfer objects; the driver should have one
2144 * \param replace_buffer callback for replacing a pipe_resource's storage
2145 * with another pipe_resource's storage.
2146 * \param out if successful, the threaded_context will be returned here in
2147 * addition to the return value if "out" != NULL
2149 struct pipe_context
*
2150 threaded_context_create(struct pipe_context
*pipe
,
2151 struct slab_parent_pool
*parent_transfer_pool
,
2152 tc_replace_buffer_storage_func replace_buffer
,
2153 struct threaded_context
**out
)
2155 struct threaded_context
*tc
;
2157 STATIC_ASSERT(sizeof(union tc_payload
) <= 8);
2158 STATIC_ASSERT(sizeof(struct tc_call
) <= 16);
2165 if (!debug_get_bool_option("GALLIUM_THREAD", util_cpu_caps
.nr_cpus
> 1))
2168 tc
= os_malloc_aligned(sizeof(struct threaded_context
), 16);
2170 pipe
->destroy(pipe
);
2173 memset(tc
, 0, sizeof(*tc
));
2175 assert((uintptr_t)tc
% 16 == 0);
2176 /* These should be static asserts, but they don't work with MSVC */
2177 assert(offsetof(struct threaded_context
, batch_slots
) % 16 == 0);
2178 assert(offsetof(struct threaded_context
, batch_slots
[0].call
) % 16 == 0);
2179 assert(offsetof(struct threaded_context
, batch_slots
[0].call
[1]) % 16 == 0);
2180 assert(offsetof(struct threaded_context
, batch_slots
[1].call
) % 16 == 0);
2182 /* The driver context isn't wrapped, so set its "priv" to NULL. */
2186 tc
->replace_buffer_storage
= replace_buffer
;
2187 tc
->map_buffer_alignment
=
2188 pipe
->screen
->get_param(pipe
->screen
, PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT
);
2189 tc
->base
.priv
= pipe
; /* priv points to the wrapped driver context */
2190 tc
->base
.screen
= pipe
->screen
;
2191 tc
->base
.destroy
= tc_destroy
;
2193 tc
->base
.stream_uploader
= u_upload_clone(&tc
->base
, pipe
->stream_uploader
);
2194 if (pipe
->stream_uploader
== pipe
->const_uploader
)
2195 tc
->base
.const_uploader
= tc
->base
.stream_uploader
;
2197 tc
->base
.const_uploader
= u_upload_clone(&tc
->base
, pipe
->const_uploader
);
2199 if (!tc
->base
.stream_uploader
|| !tc
->base
.const_uploader
)
2202 /* The queue size is the number of batches "waiting". Batches are removed
2203 * from the queue before being executed, so keep one tc_batch slot for that
2204 * execution. Also, keep one unused slot for an unflushed batch.
2206 if (!util_queue_init(&tc
->queue
, "gallium_drv", TC_MAX_BATCHES
- 2, 1))
2209 for (unsigned i
= 0; i
< TC_MAX_BATCHES
; i
++) {
2210 tc
->batch_slots
[i
].sentinel
= TC_SENTINEL
;
2211 tc
->batch_slots
[i
].pipe
= pipe
;
2212 util_queue_fence_init(&tc
->batch_slots
[i
].fence
);
2215 LIST_INITHEAD(&tc
->unflushed_queries
);
2217 slab_create_child(&tc
->pool_transfers
, parent_transfer_pool
);
2219 #define CTX_INIT(_member) \
2220 tc->base._member = tc->pipe->_member ? tc_##_member : NULL
2224 CTX_INIT(launch_grid
);
2225 CTX_INIT(resource_copy_region
);
2228 CTX_INIT(clear_render_target
);
2229 CTX_INIT(clear_depth_stencil
);
2230 CTX_INIT(clear_buffer
);
2231 CTX_INIT(clear_texture
);
2232 CTX_INIT(flush_resource
);
2233 CTX_INIT(generate_mipmap
);
2234 CTX_INIT(render_condition
);
2235 CTX_INIT(create_query
);
2236 CTX_INIT(create_batch_query
);
2237 CTX_INIT(destroy_query
);
2238 CTX_INIT(begin_query
);
2239 CTX_INIT(end_query
);
2240 CTX_INIT(get_query_result
);
2241 CTX_INIT(get_query_result_resource
);
2242 CTX_INIT(set_active_query_state
);
2243 CTX_INIT(create_blend_state
);
2244 CTX_INIT(bind_blend_state
);
2245 CTX_INIT(delete_blend_state
);
2246 CTX_INIT(create_sampler_state
);
2247 CTX_INIT(bind_sampler_states
);
2248 CTX_INIT(delete_sampler_state
);
2249 CTX_INIT(create_rasterizer_state
);
2250 CTX_INIT(bind_rasterizer_state
);
2251 CTX_INIT(delete_rasterizer_state
);
2252 CTX_INIT(create_depth_stencil_alpha_state
);
2253 CTX_INIT(bind_depth_stencil_alpha_state
);
2254 CTX_INIT(delete_depth_stencil_alpha_state
);
2255 CTX_INIT(create_fs_state
);
2256 CTX_INIT(bind_fs_state
);
2257 CTX_INIT(delete_fs_state
);
2258 CTX_INIT(create_vs_state
);
2259 CTX_INIT(bind_vs_state
);
2260 CTX_INIT(delete_vs_state
);
2261 CTX_INIT(create_gs_state
);
2262 CTX_INIT(bind_gs_state
);
2263 CTX_INIT(delete_gs_state
);
2264 CTX_INIT(create_tcs_state
);
2265 CTX_INIT(bind_tcs_state
);
2266 CTX_INIT(delete_tcs_state
);
2267 CTX_INIT(create_tes_state
);
2268 CTX_INIT(bind_tes_state
);
2269 CTX_INIT(delete_tes_state
);
2270 CTX_INIT(create_compute_state
);
2271 CTX_INIT(bind_compute_state
);
2272 CTX_INIT(delete_compute_state
);
2273 CTX_INIT(create_vertex_elements_state
);
2274 CTX_INIT(bind_vertex_elements_state
);
2275 CTX_INIT(delete_vertex_elements_state
);
2276 CTX_INIT(set_blend_color
);
2277 CTX_INIT(set_stencil_ref
);
2278 CTX_INIT(set_sample_mask
);
2279 CTX_INIT(set_min_samples
);
2280 CTX_INIT(set_clip_state
);
2281 CTX_INIT(set_constant_buffer
);
2282 CTX_INIT(set_framebuffer_state
);
2283 CTX_INIT(set_polygon_stipple
);
2284 CTX_INIT(set_scissor_states
);
2285 CTX_INIT(set_viewport_states
);
2286 CTX_INIT(set_window_rectangles
);
2287 CTX_INIT(set_sampler_views
);
2288 CTX_INIT(set_tess_state
);
2289 CTX_INIT(set_shader_buffers
);
2290 CTX_INIT(set_shader_images
);
2291 CTX_INIT(set_vertex_buffers
);
2292 CTX_INIT(create_stream_output_target
);
2293 CTX_INIT(stream_output_target_destroy
);
2294 CTX_INIT(set_stream_output_targets
);
2295 CTX_INIT(create_sampler_view
);
2296 CTX_INIT(sampler_view_destroy
);
2297 CTX_INIT(create_surface
);
2298 CTX_INIT(surface_destroy
);
2299 CTX_INIT(transfer_map
);
2300 CTX_INIT(transfer_flush_region
);
2301 CTX_INIT(transfer_unmap
);
2302 CTX_INIT(buffer_subdata
);
2303 CTX_INIT(texture_subdata
);
2304 CTX_INIT(texture_barrier
);
2305 CTX_INIT(memory_barrier
);
2306 CTX_INIT(resource_commit
);
2307 CTX_INIT(create_video_codec
);
2308 CTX_INIT(create_video_buffer
);
2309 CTX_INIT(set_compute_resources
);
2310 CTX_INIT(set_global_binding
);
2311 CTX_INIT(get_sample_position
);
2312 CTX_INIT(invalidate_resource
);
2313 CTX_INIT(get_device_reset_status
);
2314 CTX_INIT(set_device_reset_callback
);
2315 CTX_INIT(dump_debug_state
);
2316 CTX_INIT(emit_string_marker
);
2317 CTX_INIT(set_debug_callback
);
2318 CTX_INIT(create_fence_fd
);
2319 CTX_INIT(fence_server_sync
);
2320 CTX_INIT(get_timestamp
);
2329 tc_destroy(&tc
->base
);