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(MAYBE_UNUSED
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
, UNUSED
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 assert(!batch
->token
);
93 for (struct tc_call
*iter
= batch
->call
; iter
!= last
;
94 iter
+= iter
->num_call_slots
) {
95 tc_assert(iter
->sentinel
== TC_SENTINEL
);
96 execute_func
[iter
->call_id
](pipe
, &iter
->payload
);
99 tc_batch_check(batch
);
100 batch
->num_total_call_slots
= 0;
104 tc_batch_flush(struct threaded_context
*tc
)
106 struct tc_batch
*next
= &tc
->batch_slots
[tc
->next
];
108 tc_assert(next
->num_total_call_slots
!= 0);
109 tc_batch_check(next
);
111 p_atomic_add(&tc
->num_offloaded_slots
, next
->num_total_call_slots
);
114 next
->token
->tc
= NULL
;
115 tc_unflushed_batch_token_reference(&next
->token
, NULL
);
118 util_queue_add_job(&tc
->queue
, next
, &next
->fence
, tc_batch_execute
,
121 tc
->next
= (tc
->next
+ 1) % TC_MAX_BATCHES
;
124 /* This is the function that adds variable-sized calls into the current
125 * batch. It also flushes the batch if there is not enough space there.
126 * All other higher-level "add" functions use it.
128 static union tc_payload
*
129 tc_add_sized_call(struct threaded_context
*tc
, enum tc_call_id id
,
130 unsigned payload_size
)
132 struct tc_batch
*next
= &tc
->batch_slots
[tc
->next
];
133 unsigned total_size
= offsetof(struct tc_call
, payload
) + payload_size
;
134 unsigned num_call_slots
= DIV_ROUND_UP(total_size
, sizeof(struct tc_call
));
138 if (unlikely(next
->num_total_call_slots
+ num_call_slots
> TC_CALLS_PER_BATCH
)) {
140 next
= &tc
->batch_slots
[tc
->next
];
141 tc_assert(next
->num_total_call_slots
== 0);
144 tc_assert(util_queue_fence_is_signalled(&next
->fence
));
146 struct tc_call
*call
= &next
->call
[next
->num_total_call_slots
];
147 next
->num_total_call_slots
+= num_call_slots
;
149 call
->sentinel
= TC_SENTINEL
;
151 call
->num_call_slots
= num_call_slots
;
154 return &call
->payload
;
157 #define tc_add_struct_typed_call(tc, execute, type) \
158 ((struct type*)tc_add_sized_call(tc, execute, sizeof(struct type)))
160 #define tc_add_slot_based_call(tc, execute, type, num_slots) \
161 ((struct type*)tc_add_sized_call(tc, execute, \
162 sizeof(struct type) + \
163 sizeof(((struct type*)NULL)->slot[0]) * \
166 static union tc_payload
*
167 tc_add_small_call(struct threaded_context
*tc
, enum tc_call_id id
)
169 return tc_add_sized_call(tc
, id
, 0);
173 tc_is_sync(struct threaded_context
*tc
)
175 struct tc_batch
*last
= &tc
->batch_slots
[tc
->last
];
176 struct tc_batch
*next
= &tc
->batch_slots
[tc
->next
];
178 return util_queue_fence_is_signalled(&last
->fence
) &&
179 !next
->num_total_call_slots
;
183 _tc_sync(struct threaded_context
*tc
, MAYBE_UNUSED
const char *info
, MAYBE_UNUSED
const char *func
)
185 struct tc_batch
*last
= &tc
->batch_slots
[tc
->last
];
186 struct tc_batch
*next
= &tc
->batch_slots
[tc
->next
];
191 /* Only wait for queued calls... */
192 if (!util_queue_fence_is_signalled(&last
->fence
)) {
193 util_queue_fence_wait(&last
->fence
);
200 next
->token
->tc
= NULL
;
201 tc_unflushed_batch_token_reference(&next
->token
, NULL
);
204 /* .. and execute unflushed calls directly. */
205 if (next
->num_total_call_slots
) {
206 p_atomic_add(&tc
->num_direct_slots
, next
->num_total_call_slots
);
207 tc_batch_execute(next
, 0);
212 p_atomic_inc(&tc
->num_syncs
);
214 if (tc_strcmp(func
, "tc_destroy") != 0) {
215 tc_printf("sync %s %s\n", func
, info
);
222 #define tc_sync(tc) _tc_sync(tc, "", __func__)
223 #define tc_sync_msg(tc, info) _tc_sync(tc, info, __func__)
226 * Call this from fence_finish for same-context fence waits of deferred fences
227 * that haven't been flushed yet.
229 * The passed pipe_context must be the one passed to pipe_screen::fence_finish,
230 * i.e., the wrapped one.
233 threaded_context_flush(struct pipe_context
*_pipe
,
234 struct tc_unflushed_batch_token
*token
,
237 struct threaded_context
*tc
= threaded_context(_pipe
);
239 /* This is called from the state-tracker / application thread. */
240 if (token
->tc
&& token
->tc
== tc
) {
241 struct tc_batch
*last
= &tc
->batch_slots
[tc
->last
];
243 /* Prefer to do the flush in the driver thread if it is already
244 * running. That should be better for cache locality.
246 if (prefer_async
|| !util_queue_fence_is_signalled(&last
->fence
))
254 tc_set_resource_reference(struct pipe_resource
**dst
, struct pipe_resource
*src
)
257 pipe_resource_reference(dst
, src
);
261 threaded_resource_init(struct pipe_resource
*res
)
263 struct threaded_resource
*tres
= threaded_resource(res
);
265 tres
->latest
= &tres
->b
;
266 util_range_init(&tres
->valid_buffer_range
);
267 tres
->base_valid_buffer_range
= &tres
->valid_buffer_range
;
268 tres
->is_shared
= false;
269 tres
->is_user_ptr
= false;
273 threaded_resource_deinit(struct pipe_resource
*res
)
275 struct threaded_resource
*tres
= threaded_resource(res
);
277 if (tres
->latest
!= &tres
->b
)
278 pipe_resource_reference(&tres
->latest
, NULL
);
279 util_range_destroy(&tres
->valid_buffer_range
);
282 struct pipe_context
*
283 threaded_context_unwrap_sync(struct pipe_context
*pipe
)
285 if (!pipe
|| !pipe
->priv
)
288 tc_sync(threaded_context(pipe
));
289 return (struct pipe_context
*)pipe
->priv
;
293 /********************************************************************
297 #define TC_FUNC1(func, m_payload, qualifier, type, deref, deref2) \
299 tc_call_##func(struct pipe_context *pipe, union tc_payload *payload) \
301 pipe->func(pipe, deref2((type*)payload)); \
305 tc_##func(struct pipe_context *_pipe, qualifier type deref param) \
307 struct threaded_context *tc = threaded_context(_pipe); \
308 type *p = (type*)tc_add_sized_call(tc, TC_CALL_##func, sizeof(type)); \
312 TC_FUNC1(set_active_query_state
, flags
, , bool, , *)
314 TC_FUNC1(set_blend_color
, blend_color
, const, struct pipe_blend_color
, *, )
315 TC_FUNC1(set_stencil_ref
, stencil_ref
, const, struct pipe_stencil_ref
, *, )
316 TC_FUNC1(set_clip_state
, clip_state
, const, struct pipe_clip_state
, *, )
317 TC_FUNC1(set_sample_mask
, sample_mask
, , unsigned, , *)
318 TC_FUNC1(set_min_samples
, min_samples
, , unsigned, , *)
319 TC_FUNC1(set_polygon_stipple
, polygon_stipple
, const, struct pipe_poly_stipple
, *, )
321 TC_FUNC1(texture_barrier
, flags
, , unsigned, , *)
322 TC_FUNC1(memory_barrier
, flags
, , unsigned, , *)
325 /********************************************************************
329 static struct pipe_query
*
330 tc_create_query(struct pipe_context
*_pipe
, unsigned query_type
,
333 struct threaded_context
*tc
= threaded_context(_pipe
);
334 struct pipe_context
*pipe
= tc
->pipe
;
336 return pipe
->create_query(pipe
, query_type
, index
);
339 static struct pipe_query
*
340 tc_create_batch_query(struct pipe_context
*_pipe
, unsigned num_queries
,
341 unsigned *query_types
)
343 struct threaded_context
*tc
= threaded_context(_pipe
);
344 struct pipe_context
*pipe
= tc
->pipe
;
346 return pipe
->create_batch_query(pipe
, num_queries
, query_types
);
350 tc_call_destroy_query(struct pipe_context
*pipe
, union tc_payload
*payload
)
352 struct threaded_query
*tq
= threaded_query(payload
->query
);
354 if (tq
->head_unflushed
.next
)
355 LIST_DEL(&tq
->head_unflushed
);
357 pipe
->destroy_query(pipe
, payload
->query
);
361 tc_destroy_query(struct pipe_context
*_pipe
, struct pipe_query
*query
)
363 struct threaded_context
*tc
= threaded_context(_pipe
);
365 tc_add_small_call(tc
, TC_CALL_destroy_query
)->query
= query
;
369 tc_call_begin_query(struct pipe_context
*pipe
, union tc_payload
*payload
)
371 pipe
->begin_query(pipe
, payload
->query
);
375 tc_begin_query(struct pipe_context
*_pipe
, struct pipe_query
*query
)
377 struct threaded_context
*tc
= threaded_context(_pipe
);
378 union tc_payload
*payload
= tc_add_small_call(tc
, TC_CALL_begin_query
);
380 payload
->query
= query
;
381 return true; /* we don't care about the return value for this call */
384 struct tc_end_query_payload
{
385 struct threaded_context
*tc
;
386 struct pipe_query
*query
;
390 tc_call_end_query(struct pipe_context
*pipe
, union tc_payload
*payload
)
392 struct tc_end_query_payload
*p
= (struct tc_end_query_payload
*)payload
;
393 struct threaded_query
*tq
= threaded_query(p
->query
);
395 if (!tq
->head_unflushed
.next
)
396 LIST_ADD(&tq
->head_unflushed
, &p
->tc
->unflushed_queries
);
398 pipe
->end_query(pipe
, p
->query
);
402 tc_end_query(struct pipe_context
*_pipe
, struct pipe_query
*query
)
404 struct threaded_context
*tc
= threaded_context(_pipe
);
405 struct threaded_query
*tq
= threaded_query(query
);
406 struct tc_end_query_payload
*payload
=
407 tc_add_struct_typed_call(tc
, TC_CALL_end_query
, tc_end_query_payload
);
410 payload
->query
= query
;
414 return true; /* we don't care about the return value for this call */
418 tc_get_query_result(struct pipe_context
*_pipe
,
419 struct pipe_query
*query
, bool wait
,
420 union pipe_query_result
*result
)
422 struct threaded_context
*tc
= threaded_context(_pipe
);
423 struct threaded_query
*tq
= threaded_query(query
);
424 struct pipe_context
*pipe
= tc
->pipe
;
427 tc_sync_msg(tc
, wait
? "wait" : "nowait");
429 bool success
= pipe
->get_query_result(pipe
, query
, wait
, result
);
433 if (tq
->head_unflushed
.next
) {
434 /* This is safe because it can only happen after we sync'd. */
435 LIST_DEL(&tq
->head_unflushed
);
441 struct tc_query_result_resource
{
442 struct pipe_query
*query
;
444 enum pipe_query_value_type result_type
;
446 struct pipe_resource
*resource
;
451 tc_call_get_query_result_resource(struct pipe_context
*pipe
,
452 union tc_payload
*payload
)
454 struct tc_query_result_resource
*p
= (struct tc_query_result_resource
*)payload
;
456 pipe
->get_query_result_resource(pipe
, p
->query
, p
->wait
, p
->result_type
,
457 p
->index
, p
->resource
, p
->offset
);
458 pipe_resource_reference(&p
->resource
, NULL
);
462 tc_get_query_result_resource(struct pipe_context
*_pipe
,
463 struct pipe_query
*query
, bool wait
,
464 enum pipe_query_value_type result_type
, int index
,
465 struct pipe_resource
*resource
, unsigned offset
)
467 struct threaded_context
*tc
= threaded_context(_pipe
);
468 struct tc_query_result_resource
*p
=
469 tc_add_struct_typed_call(tc
, TC_CALL_get_query_result_resource
,
470 tc_query_result_resource
);
474 p
->result_type
= result_type
;
476 tc_set_resource_reference(&p
->resource
, resource
);
480 struct tc_render_condition
{
481 struct pipe_query
*query
;
487 tc_call_render_condition(struct pipe_context
*pipe
, union tc_payload
*payload
)
489 struct tc_render_condition
*p
= (struct tc_render_condition
*)payload
;
490 pipe
->render_condition(pipe
, p
->query
, p
->condition
, p
->mode
);
494 tc_render_condition(struct pipe_context
*_pipe
,
495 struct pipe_query
*query
, bool condition
,
496 enum pipe_render_cond_flag mode
)
498 struct threaded_context
*tc
= threaded_context(_pipe
);
499 struct tc_render_condition
*p
=
500 tc_add_struct_typed_call(tc
, TC_CALL_render_condition
, tc_render_condition
);
503 p
->condition
= condition
;
508 /********************************************************************
509 * constant (immutable) states
512 #define TC_CSO_CREATE(name, sname) \
514 tc_create_##name##_state(struct pipe_context *_pipe, \
515 const struct pipe_##sname##_state *state) \
517 struct pipe_context *pipe = threaded_context(_pipe)->pipe; \
518 return pipe->create_##name##_state(pipe, state); \
521 #define TC_CSO_BIND(name) TC_FUNC1(bind_##name##_state, cso, , void *, , *)
522 #define TC_CSO_DELETE(name) TC_FUNC1(delete_##name##_state, cso, , void *, , *)
524 #define TC_CSO_WHOLE2(name, sname) \
525 TC_CSO_CREATE(name, sname) \
529 #define TC_CSO_WHOLE(name) TC_CSO_WHOLE2(name, name)
532 TC_CSO_WHOLE(rasterizer
)
533 TC_CSO_WHOLE(depth_stencil_alpha
)
534 TC_CSO_WHOLE(compute
)
535 TC_CSO_WHOLE2(fs
, shader
)
536 TC_CSO_WHOLE2(vs
, shader
)
537 TC_CSO_WHOLE2(gs
, shader
)
538 TC_CSO_WHOLE2(tcs
, shader
)
539 TC_CSO_WHOLE2(tes
, shader
)
540 TC_CSO_CREATE(sampler
, sampler
)
541 TC_CSO_DELETE(sampler
)
542 TC_CSO_BIND(vertex_elements
)
543 TC_CSO_DELETE(vertex_elements
)
546 tc_create_vertex_elements_state(struct pipe_context
*_pipe
, unsigned count
,
547 const struct pipe_vertex_element
*elems
)
549 struct pipe_context
*pipe
= threaded_context(_pipe
)->pipe
;
551 return pipe
->create_vertex_elements_state(pipe
, count
, elems
);
554 struct tc_sampler_states
{
555 ubyte shader
, start
, count
;
556 void *slot
[0]; /* more will be allocated if needed */
560 tc_call_bind_sampler_states(struct pipe_context
*pipe
, union tc_payload
*payload
)
562 struct tc_sampler_states
*p
= (struct tc_sampler_states
*)payload
;
563 pipe
->bind_sampler_states(pipe
, p
->shader
, p
->start
, p
->count
, p
->slot
);
567 tc_bind_sampler_states(struct pipe_context
*_pipe
,
568 enum pipe_shader_type shader
,
569 unsigned start
, unsigned count
, void **states
)
574 struct threaded_context
*tc
= threaded_context(_pipe
);
575 struct tc_sampler_states
*p
=
576 tc_add_slot_based_call(tc
, TC_CALL_bind_sampler_states
, tc_sampler_states
, count
);
581 memcpy(p
->slot
, states
, count
* sizeof(states
[0]));
585 /********************************************************************
590 tc_call_set_framebuffer_state(struct pipe_context
*pipe
, union tc_payload
*payload
)
592 struct pipe_framebuffer_state
*p
= (struct pipe_framebuffer_state
*)payload
;
594 pipe
->set_framebuffer_state(pipe
, p
);
596 unsigned nr_cbufs
= p
->nr_cbufs
;
597 for (unsigned i
= 0; i
< nr_cbufs
; i
++)
598 pipe_surface_reference(&p
->cbufs
[i
], NULL
);
599 pipe_surface_reference(&p
->zsbuf
, NULL
);
603 tc_set_framebuffer_state(struct pipe_context
*_pipe
,
604 const struct pipe_framebuffer_state
*fb
)
606 struct threaded_context
*tc
= threaded_context(_pipe
);
607 struct pipe_framebuffer_state
*p
=
608 tc_add_struct_typed_call(tc
, TC_CALL_set_framebuffer_state
,
609 pipe_framebuffer_state
);
610 unsigned nr_cbufs
= fb
->nr_cbufs
;
612 p
->width
= fb
->width
;
613 p
->height
= fb
->height
;
614 p
->samples
= fb
->samples
;
615 p
->layers
= fb
->layers
;
616 p
->nr_cbufs
= nr_cbufs
;
618 for (unsigned i
= 0; i
< nr_cbufs
; i
++) {
620 pipe_surface_reference(&p
->cbufs
[i
], fb
->cbufs
[i
]);
623 pipe_surface_reference(&p
->zsbuf
, fb
->zsbuf
);
627 tc_call_set_tess_state(struct pipe_context
*pipe
, union tc_payload
*payload
)
629 float *p
= (float*)payload
;
630 pipe
->set_tess_state(pipe
, p
, p
+ 4);
634 tc_set_tess_state(struct pipe_context
*_pipe
,
635 const float default_outer_level
[4],
636 const float default_inner_level
[2])
638 struct threaded_context
*tc
= threaded_context(_pipe
);
639 float *p
= (float*)tc_add_sized_call(tc
, TC_CALL_set_tess_state
,
642 memcpy(p
, default_outer_level
, 4 * sizeof(float));
643 memcpy(p
+ 4, default_inner_level
, 2 * sizeof(float));
646 struct tc_constant_buffer
{
648 struct pipe_constant_buffer cb
;
652 tc_call_set_constant_buffer(struct pipe_context
*pipe
, union tc_payload
*payload
)
654 struct tc_constant_buffer
*p
= (struct tc_constant_buffer
*)payload
;
656 pipe
->set_constant_buffer(pipe
,
660 pipe_resource_reference(&p
->cb
.buffer
, NULL
);
664 tc_set_constant_buffer(struct pipe_context
*_pipe
,
665 enum pipe_shader_type shader
, uint index
,
666 const struct pipe_constant_buffer
*cb
)
668 struct threaded_context
*tc
= threaded_context(_pipe
);
669 struct pipe_resource
*buffer
= NULL
;
672 /* This must be done before adding set_constant_buffer, because it could
673 * generate e.g. transfer_unmap and flush partially-uninitialized
674 * set_constant_buffer to the driver if it was done afterwards.
676 if (cb
&& cb
->user_buffer
) {
677 u_upload_data(tc
->base
.const_uploader
, 0, cb
->buffer_size
, 64,
678 cb
->user_buffer
, &offset
, &buffer
);
679 u_upload_unmap(tc
->base
.const_uploader
);
682 struct tc_constant_buffer
*p
=
683 tc_add_struct_typed_call(tc
, TC_CALL_set_constant_buffer
,
689 if (cb
->user_buffer
) {
690 p
->cb
.buffer_size
= cb
->buffer_size
;
691 p
->cb
.user_buffer
= NULL
;
692 p
->cb
.buffer_offset
= offset
;
693 p
->cb
.buffer
= buffer
;
695 tc_set_resource_reference(&p
->cb
.buffer
,
697 memcpy(&p
->cb
, cb
, sizeof(*cb
));
700 memset(&p
->cb
, 0, sizeof(*cb
));
706 struct pipe_scissor_state slot
[0]; /* more will be allocated if needed */
710 tc_call_set_scissor_states(struct pipe_context
*pipe
, union tc_payload
*payload
)
712 struct tc_scissors
*p
= (struct tc_scissors
*)payload
;
713 pipe
->set_scissor_states(pipe
, p
->start
, p
->count
, p
->slot
);
717 tc_set_scissor_states(struct pipe_context
*_pipe
,
718 unsigned start
, unsigned count
,
719 const struct pipe_scissor_state
*states
)
721 struct threaded_context
*tc
= threaded_context(_pipe
);
722 struct tc_scissors
*p
=
723 tc_add_slot_based_call(tc
, TC_CALL_set_scissor_states
, tc_scissors
, count
);
727 memcpy(&p
->slot
, states
, count
* sizeof(states
[0]));
730 struct tc_viewports
{
732 struct pipe_viewport_state slot
[0]; /* more will be allocated if needed */
736 tc_call_set_viewport_states(struct pipe_context
*pipe
, union tc_payload
*payload
)
738 struct tc_viewports
*p
= (struct tc_viewports
*)payload
;
739 pipe
->set_viewport_states(pipe
, p
->start
, p
->count
, p
->slot
);
743 tc_set_viewport_states(struct pipe_context
*_pipe
,
744 unsigned start
, unsigned count
,
745 const struct pipe_viewport_state
*states
)
750 struct threaded_context
*tc
= threaded_context(_pipe
);
751 struct tc_viewports
*p
=
752 tc_add_slot_based_call(tc
, TC_CALL_set_viewport_states
, tc_viewports
, count
);
756 memcpy(&p
->slot
, states
, count
* sizeof(states
[0]));
759 struct tc_window_rects
{
762 struct pipe_scissor_state slot
[0]; /* more will be allocated if needed */
766 tc_call_set_window_rectangles(struct pipe_context
*pipe
,
767 union tc_payload
*payload
)
769 struct tc_window_rects
*p
= (struct tc_window_rects
*)payload
;
770 pipe
->set_window_rectangles(pipe
, p
->include
, p
->count
, p
->slot
);
774 tc_set_window_rectangles(struct pipe_context
*_pipe
, bool include
,
776 const struct pipe_scissor_state
*rects
)
778 struct threaded_context
*tc
= threaded_context(_pipe
);
779 struct tc_window_rects
*p
=
780 tc_add_slot_based_call(tc
, TC_CALL_set_window_rectangles
, tc_window_rects
, count
);
782 p
->include
= include
;
784 memcpy(p
->slot
, rects
, count
* sizeof(rects
[0]));
787 struct tc_sampler_views
{
788 ubyte shader
, start
, count
;
789 struct pipe_sampler_view
*slot
[0]; /* more will be allocated if needed */
793 tc_call_set_sampler_views(struct pipe_context
*pipe
, union tc_payload
*payload
)
795 struct tc_sampler_views
*p
= (struct tc_sampler_views
*)payload
;
796 unsigned count
= p
->count
;
798 pipe
->set_sampler_views(pipe
, p
->shader
, p
->start
, p
->count
, p
->slot
);
799 for (unsigned i
= 0; i
< count
; i
++)
800 pipe_sampler_view_reference(&p
->slot
[i
], NULL
);
804 tc_set_sampler_views(struct pipe_context
*_pipe
,
805 enum pipe_shader_type shader
,
806 unsigned start
, unsigned count
,
807 struct pipe_sampler_view
**views
)
812 struct threaded_context
*tc
= threaded_context(_pipe
);
813 struct tc_sampler_views
*p
=
814 tc_add_slot_based_call(tc
, TC_CALL_set_sampler_views
, tc_sampler_views
, count
);
821 for (unsigned i
= 0; i
< count
; i
++) {
823 pipe_sampler_view_reference(&p
->slot
[i
], views
[i
]);
826 memset(p
->slot
, 0, count
* sizeof(views
[0]));
830 struct tc_shader_images
{
831 ubyte shader
, start
, count
;
833 struct pipe_image_view slot
[0]; /* more will be allocated if needed */
837 tc_call_set_shader_images(struct pipe_context
*pipe
, union tc_payload
*payload
)
839 struct tc_shader_images
*p
= (struct tc_shader_images
*)payload
;
840 unsigned count
= p
->count
;
843 pipe
->set_shader_images(pipe
, p
->shader
, p
->start
, p
->count
, NULL
);
847 pipe
->set_shader_images(pipe
, p
->shader
, p
->start
, p
->count
, p
->slot
);
849 for (unsigned i
= 0; i
< count
; i
++)
850 pipe_resource_reference(&p
->slot
[i
].resource
, NULL
);
854 tc_set_shader_images(struct pipe_context
*_pipe
,
855 enum pipe_shader_type shader
,
856 unsigned start
, unsigned count
,
857 const struct pipe_image_view
*images
)
862 struct threaded_context
*tc
= threaded_context(_pipe
);
863 struct tc_shader_images
*p
=
864 tc_add_slot_based_call(tc
, TC_CALL_set_shader_images
, tc_shader_images
,
870 p
->unbind
= images
== NULL
;
873 for (unsigned i
= 0; i
< count
; i
++) {
874 tc_set_resource_reference(&p
->slot
[i
].resource
, images
[i
].resource
);
876 if (images
[i
].access
& PIPE_IMAGE_ACCESS_WRITE
&&
877 images
[i
].resource
&&
878 images
[i
].resource
->target
== PIPE_BUFFER
) {
879 struct threaded_resource
*tres
=
880 threaded_resource(images
[i
].resource
);
882 util_range_add(&tres
->valid_buffer_range
, images
[i
].u
.buf
.offset
,
883 images
[i
].u
.buf
.offset
+ images
[i
].u
.buf
.size
);
886 memcpy(p
->slot
, images
, count
* sizeof(images
[0]));
890 struct tc_shader_buffers
{
891 ubyte shader
, start
, count
;
893 unsigned writable_bitmask
;
894 struct pipe_shader_buffer slot
[0]; /* more will be allocated if needed */
898 tc_call_set_shader_buffers(struct pipe_context
*pipe
, union tc_payload
*payload
)
900 struct tc_shader_buffers
*p
= (struct tc_shader_buffers
*)payload
;
901 unsigned count
= p
->count
;
904 pipe
->set_shader_buffers(pipe
, p
->shader
, p
->start
, p
->count
, NULL
, 0);
908 pipe
->set_shader_buffers(pipe
, p
->shader
, p
->start
, p
->count
, p
->slot
,
909 p
->writable_bitmask
);
911 for (unsigned i
= 0; i
< count
; i
++)
912 pipe_resource_reference(&p
->slot
[i
].buffer
, NULL
);
916 tc_set_shader_buffers(struct pipe_context
*_pipe
,
917 enum pipe_shader_type shader
,
918 unsigned start
, unsigned count
,
919 const struct pipe_shader_buffer
*buffers
,
920 unsigned writable_bitmask
)
925 struct threaded_context
*tc
= threaded_context(_pipe
);
926 struct tc_shader_buffers
*p
=
927 tc_add_slot_based_call(tc
, TC_CALL_set_shader_buffers
, tc_shader_buffers
,
928 buffers
? count
: 0);
933 p
->unbind
= buffers
== NULL
;
934 p
->writable_bitmask
= writable_bitmask
;
937 for (unsigned i
= 0; i
< count
; i
++) {
938 struct pipe_shader_buffer
*dst
= &p
->slot
[i
];
939 const struct pipe_shader_buffer
*src
= buffers
+ i
;
941 tc_set_resource_reference(&dst
->buffer
, src
->buffer
);
942 dst
->buffer_offset
= src
->buffer_offset
;
943 dst
->buffer_size
= src
->buffer_size
;
946 struct threaded_resource
*tres
= threaded_resource(src
->buffer
);
948 util_range_add(&tres
->valid_buffer_range
, src
->buffer_offset
,
949 src
->buffer_offset
+ src
->buffer_size
);
955 struct tc_vertex_buffers
{
958 struct pipe_vertex_buffer slot
[0]; /* more will be allocated if needed */
962 tc_call_set_vertex_buffers(struct pipe_context
*pipe
, union tc_payload
*payload
)
964 struct tc_vertex_buffers
*p
= (struct tc_vertex_buffers
*)payload
;
965 unsigned count
= p
->count
;
968 pipe
->set_vertex_buffers(pipe
, p
->start
, count
, NULL
);
972 for (unsigned i
= 0; i
< count
; i
++)
973 tc_assert(!p
->slot
[i
].is_user_buffer
);
975 pipe
->set_vertex_buffers(pipe
, p
->start
, count
, p
->slot
);
976 for (unsigned i
= 0; i
< count
; i
++)
977 pipe_resource_reference(&p
->slot
[i
].buffer
.resource
, NULL
);
981 tc_set_vertex_buffers(struct pipe_context
*_pipe
,
982 unsigned start
, unsigned count
,
983 const struct pipe_vertex_buffer
*buffers
)
985 struct threaded_context
*tc
= threaded_context(_pipe
);
991 struct tc_vertex_buffers
*p
=
992 tc_add_slot_based_call(tc
, TC_CALL_set_vertex_buffers
, tc_vertex_buffers
, count
);
997 for (unsigned i
= 0; i
< count
; i
++) {
998 struct pipe_vertex_buffer
*dst
= &p
->slot
[i
];
999 const struct pipe_vertex_buffer
*src
= buffers
+ i
;
1001 tc_assert(!src
->is_user_buffer
);
1002 dst
->stride
= src
->stride
;
1003 dst
->is_user_buffer
= false;
1004 tc_set_resource_reference(&dst
->buffer
.resource
,
1005 src
->buffer
.resource
);
1006 dst
->buffer_offset
= src
->buffer_offset
;
1009 struct tc_vertex_buffers
*p
=
1010 tc_add_slot_based_call(tc
, TC_CALL_set_vertex_buffers
, tc_vertex_buffers
, 0);
1017 struct tc_stream_outputs
{
1019 struct pipe_stream_output_target
*targets
[PIPE_MAX_SO_BUFFERS
];
1020 unsigned offsets
[PIPE_MAX_SO_BUFFERS
];
1024 tc_call_set_stream_output_targets(struct pipe_context
*pipe
, union tc_payload
*payload
)
1026 struct tc_stream_outputs
*p
= (struct tc_stream_outputs
*)payload
;
1027 unsigned count
= p
->count
;
1029 pipe
->set_stream_output_targets(pipe
, count
, p
->targets
, p
->offsets
);
1030 for (unsigned i
= 0; i
< count
; i
++)
1031 pipe_so_target_reference(&p
->targets
[i
], NULL
);
1035 tc_set_stream_output_targets(struct pipe_context
*_pipe
,
1037 struct pipe_stream_output_target
**tgs
,
1038 const unsigned *offsets
)
1040 struct threaded_context
*tc
= threaded_context(_pipe
);
1041 struct tc_stream_outputs
*p
=
1042 tc_add_struct_typed_call(tc
, TC_CALL_set_stream_output_targets
,
1045 for (unsigned i
= 0; i
< count
; i
++) {
1046 p
->targets
[i
] = NULL
;
1047 pipe_so_target_reference(&p
->targets
[i
], tgs
[i
]);
1050 memcpy(p
->offsets
, offsets
, count
* sizeof(unsigned));
1054 tc_set_compute_resources(struct pipe_context
*_pipe
, unsigned start
,
1055 unsigned count
, struct pipe_surface
**resources
)
1057 struct threaded_context
*tc
= threaded_context(_pipe
);
1058 struct pipe_context
*pipe
= tc
->pipe
;
1061 pipe
->set_compute_resources(pipe
, start
, count
, resources
);
1065 tc_set_global_binding(struct pipe_context
*_pipe
, unsigned first
,
1066 unsigned count
, struct pipe_resource
**resources
,
1069 struct threaded_context
*tc
= threaded_context(_pipe
);
1070 struct pipe_context
*pipe
= tc
->pipe
;
1073 pipe
->set_global_binding(pipe
, first
, count
, resources
, handles
);
1077 /********************************************************************
1081 static struct pipe_surface
*
1082 tc_create_surface(struct pipe_context
*_pipe
,
1083 struct pipe_resource
*resource
,
1084 const struct pipe_surface
*surf_tmpl
)
1086 struct pipe_context
*pipe
= threaded_context(_pipe
)->pipe
;
1087 struct pipe_surface
*view
=
1088 pipe
->create_surface(pipe
, resource
, surf_tmpl
);
1091 view
->context
= _pipe
;
1096 tc_surface_destroy(struct pipe_context
*_pipe
,
1097 struct pipe_surface
*surf
)
1099 struct pipe_context
*pipe
= threaded_context(_pipe
)->pipe
;
1101 pipe
->surface_destroy(pipe
, surf
);
1104 static struct pipe_sampler_view
*
1105 tc_create_sampler_view(struct pipe_context
*_pipe
,
1106 struct pipe_resource
*resource
,
1107 const struct pipe_sampler_view
*templ
)
1109 struct pipe_context
*pipe
= threaded_context(_pipe
)->pipe
;
1110 struct pipe_sampler_view
*view
=
1111 pipe
->create_sampler_view(pipe
, resource
, templ
);
1114 view
->context
= _pipe
;
1119 tc_sampler_view_destroy(struct pipe_context
*_pipe
,
1120 struct pipe_sampler_view
*view
)
1122 struct pipe_context
*pipe
= threaded_context(_pipe
)->pipe
;
1124 pipe
->sampler_view_destroy(pipe
, view
);
1127 static struct pipe_stream_output_target
*
1128 tc_create_stream_output_target(struct pipe_context
*_pipe
,
1129 struct pipe_resource
*res
,
1130 unsigned buffer_offset
,
1131 unsigned buffer_size
)
1133 struct pipe_context
*pipe
= threaded_context(_pipe
)->pipe
;
1134 struct threaded_resource
*tres
= threaded_resource(res
);
1135 struct pipe_stream_output_target
*view
;
1137 tc_sync(threaded_context(_pipe
));
1138 util_range_add(&tres
->valid_buffer_range
, buffer_offset
,
1139 buffer_offset
+ buffer_size
);
1141 view
= pipe
->create_stream_output_target(pipe
, res
, buffer_offset
,
1144 view
->context
= _pipe
;
1149 tc_stream_output_target_destroy(struct pipe_context
*_pipe
,
1150 struct pipe_stream_output_target
*target
)
1152 struct pipe_context
*pipe
= threaded_context(_pipe
)->pipe
;
1154 pipe
->stream_output_target_destroy(pipe
, target
);
1158 /********************************************************************
1163 tc_create_texture_handle(struct pipe_context
*_pipe
,
1164 struct pipe_sampler_view
*view
,
1165 const struct pipe_sampler_state
*state
)
1167 struct threaded_context
*tc
= threaded_context(_pipe
);
1168 struct pipe_context
*pipe
= tc
->pipe
;
1171 return pipe
->create_texture_handle(pipe
, view
, state
);
1175 tc_call_delete_texture_handle(struct pipe_context
*pipe
,
1176 union tc_payload
*payload
)
1178 pipe
->delete_texture_handle(pipe
, payload
->handle
);
1182 tc_delete_texture_handle(struct pipe_context
*_pipe
, uint64_t handle
)
1184 struct threaded_context
*tc
= threaded_context(_pipe
);
1185 union tc_payload
*payload
=
1186 tc_add_small_call(tc
, TC_CALL_delete_texture_handle
);
1188 payload
->handle
= handle
;
1191 struct tc_make_texture_handle_resident
1198 tc_call_make_texture_handle_resident(struct pipe_context
*pipe
,
1199 union tc_payload
*payload
)
1201 struct tc_make_texture_handle_resident
*p
=
1202 (struct tc_make_texture_handle_resident
*)payload
;
1204 pipe
->make_texture_handle_resident(pipe
, p
->handle
, p
->resident
);
1208 tc_make_texture_handle_resident(struct pipe_context
*_pipe
, uint64_t handle
,
1211 struct threaded_context
*tc
= threaded_context(_pipe
);
1212 struct tc_make_texture_handle_resident
*p
=
1213 tc_add_struct_typed_call(tc
, TC_CALL_make_texture_handle_resident
,
1214 tc_make_texture_handle_resident
);
1217 p
->resident
= resident
;
1221 tc_create_image_handle(struct pipe_context
*_pipe
,
1222 const struct pipe_image_view
*image
)
1224 struct threaded_context
*tc
= threaded_context(_pipe
);
1225 struct pipe_context
*pipe
= tc
->pipe
;
1228 return pipe
->create_image_handle(pipe
, image
);
1232 tc_call_delete_image_handle(struct pipe_context
*pipe
,
1233 union tc_payload
*payload
)
1235 pipe
->delete_image_handle(pipe
, payload
->handle
);
1239 tc_delete_image_handle(struct pipe_context
*_pipe
, uint64_t handle
)
1241 struct threaded_context
*tc
= threaded_context(_pipe
);
1242 union tc_payload
*payload
=
1243 tc_add_small_call(tc
, TC_CALL_delete_image_handle
);
1245 payload
->handle
= handle
;
1248 struct tc_make_image_handle_resident
1256 tc_call_make_image_handle_resident(struct pipe_context
*pipe
,
1257 union tc_payload
*payload
)
1259 struct tc_make_image_handle_resident
*p
=
1260 (struct tc_make_image_handle_resident
*)payload
;
1262 pipe
->make_image_handle_resident(pipe
, p
->handle
, p
->access
, p
->resident
);
1266 tc_make_image_handle_resident(struct pipe_context
*_pipe
, uint64_t handle
,
1267 unsigned access
, bool resident
)
1269 struct threaded_context
*tc
= threaded_context(_pipe
);
1270 struct tc_make_image_handle_resident
*p
=
1271 tc_add_struct_typed_call(tc
, TC_CALL_make_image_handle_resident
,
1272 tc_make_image_handle_resident
);
1276 p
->resident
= resident
;
1280 /********************************************************************
1284 struct tc_replace_buffer_storage
{
1285 struct pipe_resource
*dst
;
1286 struct pipe_resource
*src
;
1287 tc_replace_buffer_storage_func func
;
1291 tc_call_replace_buffer_storage(struct pipe_context
*pipe
,
1292 union tc_payload
*payload
)
1294 struct tc_replace_buffer_storage
*p
=
1295 (struct tc_replace_buffer_storage
*)payload
;
1297 p
->func(pipe
, p
->dst
, p
->src
);
1298 pipe_resource_reference(&p
->dst
, NULL
);
1299 pipe_resource_reference(&p
->src
, NULL
);
1303 tc_invalidate_buffer(struct threaded_context
*tc
,
1304 struct threaded_resource
*tbuf
)
1306 /* We can't check if the buffer is idle, so we invalidate it
1307 * unconditionally. */
1308 struct pipe_screen
*screen
= tc
->base
.screen
;
1309 struct pipe_resource
*new_buf
;
1311 /* Shared, pinned, and sparse buffers can't be reallocated. */
1312 if (tbuf
->is_shared
||
1313 tbuf
->is_user_ptr
||
1314 tbuf
->b
.flags
& PIPE_RESOURCE_FLAG_SPARSE
)
1317 /* Allocate a new one. */
1318 new_buf
= screen
->resource_create(screen
, &tbuf
->b
);
1322 /* Replace the "latest" pointer. */
1323 if (tbuf
->latest
!= &tbuf
->b
)
1324 pipe_resource_reference(&tbuf
->latest
, NULL
);
1326 tbuf
->latest
= new_buf
;
1327 util_range_set_empty(&tbuf
->valid_buffer_range
);
1329 /* The valid range should point to the original buffer. */
1330 threaded_resource(new_buf
)->base_valid_buffer_range
=
1331 &tbuf
->valid_buffer_range
;
1333 /* Enqueue storage replacement of the original buffer. */
1334 struct tc_replace_buffer_storage
*p
=
1335 tc_add_struct_typed_call(tc
, TC_CALL_replace_buffer_storage
,
1336 tc_replace_buffer_storage
);
1338 p
->func
= tc
->replace_buffer_storage
;
1339 tc_set_resource_reference(&p
->dst
, &tbuf
->b
);
1340 tc_set_resource_reference(&p
->src
, new_buf
);
1345 tc_improve_map_buffer_flags(struct threaded_context
*tc
,
1346 struct threaded_resource
*tres
, unsigned usage
,
1347 unsigned offset
, unsigned size
)
1349 /* Never invalidate inside the driver and never infer "unsynchronized". */
1350 unsigned tc_flags
= TC_TRANSFER_MAP_NO_INVALIDATE
|
1351 TC_TRANSFER_MAP_NO_INFER_UNSYNCHRONIZED
;
1353 /* Prevent a reentry. */
1354 if (usage
& tc_flags
)
1357 /* Use the staging upload if it's preferred. */
1358 if (usage
& (PIPE_TRANSFER_DISCARD_RANGE
|
1359 PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
) &&
1360 !(usage
& PIPE_TRANSFER_PERSISTENT
) &&
1361 /* Try not to decrement the counter if it's not positive. Still racy,
1362 * but it makes it harder to wrap the counter from INT_MIN to INT_MAX. */
1363 tres
->max_forced_staging_uploads
> 0 &&
1364 p_atomic_dec_return(&tres
->max_forced_staging_uploads
) >= 0) {
1365 usage
&= ~(PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
|
1366 PIPE_TRANSFER_UNSYNCHRONIZED
);
1368 return usage
| tc_flags
| PIPE_TRANSFER_DISCARD_RANGE
;
1371 /* Sparse buffers can't be mapped directly and can't be reallocated
1372 * (fully invalidated). That may just be a radeonsi limitation, but
1373 * the threaded context must obey it with radeonsi.
1375 if (tres
->b
.flags
& PIPE_RESOURCE_FLAG_SPARSE
) {
1376 /* We can use DISCARD_RANGE instead of full discard. This is the only
1377 * fast path for sparse buffers that doesn't need thread synchronization.
1379 if (usage
& PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
)
1380 usage
|= PIPE_TRANSFER_DISCARD_RANGE
;
1382 /* Allow DISCARD_WHOLE_RESOURCE and infering UNSYNCHRONIZED in drivers.
1383 * The threaded context doesn't do unsychronized mappings and invalida-
1384 * tions of sparse buffers, therefore a correct driver behavior won't
1385 * result in an incorrect behavior with the threaded context.
1392 /* Handle CPU reads trivially. */
1393 if (usage
& PIPE_TRANSFER_READ
) {
1394 /* Drivers aren't allowed to do buffer invalidations. */
1395 return usage
& ~PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
;
1398 /* See if the buffer range being mapped has never been initialized,
1399 * in which case it can be mapped unsynchronized. */
1400 if (!(usage
& PIPE_TRANSFER_UNSYNCHRONIZED
) &&
1402 !util_ranges_intersect(&tres
->valid_buffer_range
, offset
, offset
+ size
))
1403 usage
|= PIPE_TRANSFER_UNSYNCHRONIZED
;
1405 if (!(usage
& PIPE_TRANSFER_UNSYNCHRONIZED
)) {
1406 /* If discarding the entire range, discard the whole resource instead. */
1407 if (usage
& PIPE_TRANSFER_DISCARD_RANGE
&&
1408 offset
== 0 && size
== tres
->b
.width0
)
1409 usage
|= PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
;
1411 /* Discard the whole resource if needed. */
1412 if (usage
& PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
) {
1413 if (tc_invalidate_buffer(tc
, tres
))
1414 usage
|= PIPE_TRANSFER_UNSYNCHRONIZED
;
1416 usage
|= PIPE_TRANSFER_DISCARD_RANGE
; /* fallback */
1420 /* We won't need this flag anymore. */
1421 /* TODO: We might not need TC_TRANSFER_MAP_NO_INVALIDATE with this. */
1422 usage
&= ~PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
;
1424 /* GL_AMD_pinned_memory and persistent mappings can't use staging
1426 if (usage
& (PIPE_TRANSFER_UNSYNCHRONIZED
|
1427 PIPE_TRANSFER_PERSISTENT
) ||
1429 usage
&= ~PIPE_TRANSFER_DISCARD_RANGE
;
1431 /* Unsychronized buffer mappings don't have to synchronize the thread. */
1432 if (usage
& PIPE_TRANSFER_UNSYNCHRONIZED
) {
1433 usage
&= ~PIPE_TRANSFER_DISCARD_RANGE
;
1434 usage
|= TC_TRANSFER_MAP_THREADED_UNSYNC
; /* notify the driver */
1441 tc_transfer_map(struct pipe_context
*_pipe
,
1442 struct pipe_resource
*resource
, unsigned level
,
1443 unsigned usage
, const struct pipe_box
*box
,
1444 struct pipe_transfer
**transfer
)
1446 struct threaded_context
*tc
= threaded_context(_pipe
);
1447 struct threaded_resource
*tres
= threaded_resource(resource
);
1448 struct pipe_context
*pipe
= tc
->pipe
;
1450 if (resource
->target
== PIPE_BUFFER
) {
1451 usage
= tc_improve_map_buffer_flags(tc
, tres
, usage
, box
->x
, box
->width
);
1453 /* Do a staging transfer within the threaded context. The driver should
1454 * only get resource_copy_region.
1456 if (usage
& PIPE_TRANSFER_DISCARD_RANGE
) {
1457 struct threaded_transfer
*ttrans
= slab_alloc(&tc
->pool_transfers
);
1460 ttrans
->staging
= NULL
;
1462 u_upload_alloc(tc
->base
.stream_uploader
, 0,
1463 box
->width
+ (box
->x
% tc
->map_buffer_alignment
),
1464 64, &ttrans
->offset
, &ttrans
->staging
, (void**)&map
);
1466 slab_free(&tc
->pool_transfers
, ttrans
);
1470 tc_set_resource_reference(&ttrans
->b
.resource
, resource
);
1471 ttrans
->b
.level
= 0;
1472 ttrans
->b
.usage
= usage
;
1473 ttrans
->b
.box
= *box
;
1474 ttrans
->b
.stride
= 0;
1475 ttrans
->b
.layer_stride
= 0;
1476 *transfer
= &ttrans
->b
;
1477 return map
+ (box
->x
% tc
->map_buffer_alignment
);
1481 /* Unsychronized buffer mappings don't have to synchronize the thread. */
1482 if (!(usage
& TC_TRANSFER_MAP_THREADED_UNSYNC
))
1483 tc_sync_msg(tc
, resource
->target
!= PIPE_BUFFER
? " texture" :
1484 usage
& PIPE_TRANSFER_DISCARD_RANGE
? " discard_range" :
1485 usage
& PIPE_TRANSFER_READ
? " read" : " ??");
1487 return pipe
->transfer_map(pipe
, tres
->latest
? tres
->latest
: resource
,
1488 level
, usage
, box
, transfer
);
1491 struct tc_transfer_flush_region
{
1492 struct pipe_transfer
*transfer
;
1493 struct pipe_box box
;
1497 tc_call_transfer_flush_region(struct pipe_context
*pipe
,
1498 union tc_payload
*payload
)
1500 struct tc_transfer_flush_region
*p
=
1501 (struct tc_transfer_flush_region
*)payload
;
1503 pipe
->transfer_flush_region(pipe
, p
->transfer
, &p
->box
);
1506 struct tc_resource_copy_region
{
1507 struct pipe_resource
*dst
;
1509 unsigned dstx
, dsty
, dstz
;
1510 struct pipe_resource
*src
;
1512 struct pipe_box src_box
;
1516 tc_resource_copy_region(struct pipe_context
*_pipe
,
1517 struct pipe_resource
*dst
, unsigned dst_level
,
1518 unsigned dstx
, unsigned dsty
, unsigned dstz
,
1519 struct pipe_resource
*src
, unsigned src_level
,
1520 const struct pipe_box
*src_box
);
1523 tc_buffer_do_flush_region(struct threaded_context
*tc
,
1524 struct threaded_transfer
*ttrans
,
1525 const struct pipe_box
*box
)
1527 struct threaded_resource
*tres
= threaded_resource(ttrans
->b
.resource
);
1529 if (ttrans
->staging
) {
1530 struct pipe_box src_box
;
1532 u_box_1d(ttrans
->offset
+ ttrans
->b
.box
.x
% tc
->map_buffer_alignment
+
1533 (box
->x
- ttrans
->b
.box
.x
),
1534 box
->width
, &src_box
);
1536 /* Copy the staging buffer into the original one. */
1537 tc_resource_copy_region(&tc
->base
, ttrans
->b
.resource
, 0, box
->x
, 0, 0,
1538 ttrans
->staging
, 0, &src_box
);
1541 util_range_add(tres
->base_valid_buffer_range
, box
->x
, box
->x
+ box
->width
);
1545 tc_transfer_flush_region(struct pipe_context
*_pipe
,
1546 struct pipe_transfer
*transfer
,
1547 const struct pipe_box
*rel_box
)
1549 struct threaded_context
*tc
= threaded_context(_pipe
);
1550 struct threaded_transfer
*ttrans
= threaded_transfer(transfer
);
1551 struct threaded_resource
*tres
= threaded_resource(transfer
->resource
);
1552 unsigned required_usage
= PIPE_TRANSFER_WRITE
|
1553 PIPE_TRANSFER_FLUSH_EXPLICIT
;
1555 if (tres
->b
.target
== PIPE_BUFFER
) {
1556 if ((transfer
->usage
& required_usage
) == required_usage
) {
1557 struct pipe_box box
;
1559 u_box_1d(transfer
->box
.x
+ rel_box
->x
, rel_box
->width
, &box
);
1560 tc_buffer_do_flush_region(tc
, ttrans
, &box
);
1563 /* Staging transfers don't send the call to the driver. */
1564 if (ttrans
->staging
)
1568 struct tc_transfer_flush_region
*p
=
1569 tc_add_struct_typed_call(tc
, TC_CALL_transfer_flush_region
,
1570 tc_transfer_flush_region
);
1571 p
->transfer
= transfer
;
1576 tc_call_transfer_unmap(struct pipe_context
*pipe
, union tc_payload
*payload
)
1578 pipe
->transfer_unmap(pipe
, payload
->transfer
);
1582 tc_transfer_unmap(struct pipe_context
*_pipe
, struct pipe_transfer
*transfer
)
1584 struct threaded_context
*tc
= threaded_context(_pipe
);
1585 struct threaded_transfer
*ttrans
= threaded_transfer(transfer
);
1586 struct threaded_resource
*tres
= threaded_resource(transfer
->resource
);
1588 if (tres
->b
.target
== PIPE_BUFFER
) {
1589 if (transfer
->usage
& PIPE_TRANSFER_WRITE
&&
1590 !(transfer
->usage
& PIPE_TRANSFER_FLUSH_EXPLICIT
))
1591 tc_buffer_do_flush_region(tc
, ttrans
, &transfer
->box
);
1593 /* Staging transfers don't send the call to the driver. */
1594 if (ttrans
->staging
) {
1595 pipe_resource_reference(&ttrans
->staging
, NULL
);
1596 pipe_resource_reference(&ttrans
->b
.resource
, NULL
);
1597 slab_free(&tc
->pool_transfers
, ttrans
);
1602 tc_add_small_call(tc
, TC_CALL_transfer_unmap
)->transfer
= transfer
;
1605 struct tc_buffer_subdata
{
1606 struct pipe_resource
*resource
;
1607 unsigned usage
, offset
, size
;
1608 char slot
[0]; /* more will be allocated if needed */
1612 tc_call_buffer_subdata(struct pipe_context
*pipe
, union tc_payload
*payload
)
1614 struct tc_buffer_subdata
*p
= (struct tc_buffer_subdata
*)payload
;
1616 pipe
->buffer_subdata(pipe
, p
->resource
, p
->usage
, p
->offset
, p
->size
,
1618 pipe_resource_reference(&p
->resource
, NULL
);
1622 tc_buffer_subdata(struct pipe_context
*_pipe
,
1623 struct pipe_resource
*resource
,
1624 unsigned usage
, unsigned offset
,
1625 unsigned size
, const void *data
)
1627 struct threaded_context
*tc
= threaded_context(_pipe
);
1628 struct threaded_resource
*tres
= threaded_resource(resource
);
1633 usage
|= PIPE_TRANSFER_WRITE
;
1635 /* PIPE_TRANSFER_MAP_DIRECTLY supresses implicit DISCARD_RANGE. */
1636 if (!(usage
& PIPE_TRANSFER_MAP_DIRECTLY
))
1637 usage
|= PIPE_TRANSFER_DISCARD_RANGE
;
1639 usage
= tc_improve_map_buffer_flags(tc
, tres
, usage
, offset
, size
);
1641 /* Unsychronized and big transfers should use transfer_map. Also handle
1642 * full invalidations, because drivers aren't allowed to do them.
1644 if (usage
& (PIPE_TRANSFER_UNSYNCHRONIZED
|
1645 PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
) ||
1646 size
> TC_MAX_SUBDATA_BYTES
) {
1647 struct pipe_transfer
*transfer
;
1648 struct pipe_box box
;
1649 uint8_t *map
= NULL
;
1651 u_box_1d(offset
, size
, &box
);
1653 map
= tc_transfer_map(_pipe
, resource
, 0, usage
, &box
, &transfer
);
1655 memcpy(map
, data
, size
);
1656 tc_transfer_unmap(_pipe
, transfer
);
1661 util_range_add(&tres
->valid_buffer_range
, offset
, offset
+ size
);
1663 /* The upload is small. Enqueue it. */
1664 struct tc_buffer_subdata
*p
=
1665 tc_add_slot_based_call(tc
, TC_CALL_buffer_subdata
, tc_buffer_subdata
, size
);
1667 tc_set_resource_reference(&p
->resource
, resource
);
1671 memcpy(p
->slot
, data
, size
);
1674 struct tc_texture_subdata
{
1675 struct pipe_resource
*resource
;
1676 unsigned level
, usage
, stride
, layer_stride
;
1677 struct pipe_box box
;
1678 char slot
[0]; /* more will be allocated if needed */
1682 tc_call_texture_subdata(struct pipe_context
*pipe
, union tc_payload
*payload
)
1684 struct tc_texture_subdata
*p
= (struct tc_texture_subdata
*)payload
;
1686 pipe
->texture_subdata(pipe
, p
->resource
, p
->level
, p
->usage
, &p
->box
,
1687 p
->slot
, p
->stride
, p
->layer_stride
);
1688 pipe_resource_reference(&p
->resource
, NULL
);
1692 tc_texture_subdata(struct pipe_context
*_pipe
,
1693 struct pipe_resource
*resource
,
1694 unsigned level
, unsigned usage
,
1695 const struct pipe_box
*box
,
1696 const void *data
, unsigned stride
,
1697 unsigned layer_stride
)
1699 struct threaded_context
*tc
= threaded_context(_pipe
);
1702 assert(box
->height
>= 1);
1703 assert(box
->depth
>= 1);
1705 size
= (box
->depth
- 1) * layer_stride
+
1706 (box
->height
- 1) * stride
+
1707 box
->width
* util_format_get_blocksize(resource
->format
);
1711 /* Small uploads can be enqueued, big uploads must sync. */
1712 if (size
<= TC_MAX_SUBDATA_BYTES
) {
1713 struct tc_texture_subdata
*p
=
1714 tc_add_slot_based_call(tc
, TC_CALL_texture_subdata
, tc_texture_subdata
, size
);
1716 tc_set_resource_reference(&p
->resource
, resource
);
1721 p
->layer_stride
= layer_stride
;
1722 memcpy(p
->slot
, data
, size
);
1724 struct pipe_context
*pipe
= tc
->pipe
;
1727 pipe
->texture_subdata(pipe
, resource
, level
, usage
, box
, data
,
1728 stride
, layer_stride
);
1733 /********************************************************************
1737 #define TC_FUNC_SYNC_RET0(ret_type, func) \
1739 tc_##func(struct pipe_context *_pipe) \
1741 struct threaded_context *tc = threaded_context(_pipe); \
1742 struct pipe_context *pipe = tc->pipe; \
1744 return pipe->func(pipe); \
1747 TC_FUNC_SYNC_RET0(enum pipe_reset_status
, get_device_reset_status
)
1748 TC_FUNC_SYNC_RET0(uint64_t, get_timestamp
)
1751 tc_get_sample_position(struct pipe_context
*_pipe
,
1752 unsigned sample_count
, unsigned sample_index
,
1755 struct threaded_context
*tc
= threaded_context(_pipe
);
1756 struct pipe_context
*pipe
= tc
->pipe
;
1759 pipe
->get_sample_position(pipe
, sample_count
, sample_index
,
1764 tc_set_device_reset_callback(struct pipe_context
*_pipe
,
1765 const struct pipe_device_reset_callback
*cb
)
1767 struct threaded_context
*tc
= threaded_context(_pipe
);
1768 struct pipe_context
*pipe
= tc
->pipe
;
1771 pipe
->set_device_reset_callback(pipe
, cb
);
1774 struct tc_string_marker
{
1776 char slot
[0]; /* more will be allocated if needed */
1780 tc_call_emit_string_marker(struct pipe_context
*pipe
, union tc_payload
*payload
)
1782 struct tc_string_marker
*p
= (struct tc_string_marker
*)payload
;
1783 pipe
->emit_string_marker(pipe
, p
->slot
, p
->len
);
1787 tc_emit_string_marker(struct pipe_context
*_pipe
,
1788 const char *string
, int len
)
1790 struct threaded_context
*tc
= threaded_context(_pipe
);
1792 if (len
<= TC_MAX_STRING_MARKER_BYTES
) {
1793 struct tc_string_marker
*p
=
1794 tc_add_slot_based_call(tc
, TC_CALL_emit_string_marker
, tc_string_marker
, len
);
1796 memcpy(p
->slot
, string
, len
);
1799 struct pipe_context
*pipe
= tc
->pipe
;
1802 pipe
->emit_string_marker(pipe
, string
, len
);
1807 tc_dump_debug_state(struct pipe_context
*_pipe
, FILE *stream
,
1810 struct threaded_context
*tc
= threaded_context(_pipe
);
1811 struct pipe_context
*pipe
= tc
->pipe
;
1814 pipe
->dump_debug_state(pipe
, stream
, flags
);
1818 tc_set_debug_callback(struct pipe_context
*_pipe
,
1819 const struct pipe_debug_callback
*cb
)
1821 struct threaded_context
*tc
= threaded_context(_pipe
);
1822 struct pipe_context
*pipe
= tc
->pipe
;
1824 /* Drop all synchronous debug callbacks. Drivers are expected to be OK
1825 * with this. shader-db will use an environment variable to disable
1826 * the threaded context.
1828 if (cb
&& cb
->debug_message
&& !cb
->async
)
1832 pipe
->set_debug_callback(pipe
, cb
);
1836 tc_set_log_context(struct pipe_context
*_pipe
, struct u_log_context
*log
)
1838 struct threaded_context
*tc
= threaded_context(_pipe
);
1839 struct pipe_context
*pipe
= tc
->pipe
;
1842 pipe
->set_log_context(pipe
, log
);
1846 tc_create_fence_fd(struct pipe_context
*_pipe
,
1847 struct pipe_fence_handle
**fence
, int fd
,
1848 enum pipe_fd_type type
)
1850 struct threaded_context
*tc
= threaded_context(_pipe
);
1851 struct pipe_context
*pipe
= tc
->pipe
;
1854 pipe
->create_fence_fd(pipe
, fence
, fd
, type
);
1858 tc_call_fence_server_sync(struct pipe_context
*pipe
, union tc_payload
*payload
)
1860 pipe
->fence_server_sync(pipe
, payload
->fence
);
1861 pipe
->screen
->fence_reference(pipe
->screen
, &payload
->fence
, NULL
);
1865 tc_fence_server_sync(struct pipe_context
*_pipe
,
1866 struct pipe_fence_handle
*fence
)
1868 struct threaded_context
*tc
= threaded_context(_pipe
);
1869 struct pipe_screen
*screen
= tc
->pipe
->screen
;
1870 union tc_payload
*payload
= tc_add_small_call(tc
, TC_CALL_fence_server_sync
);
1872 payload
->fence
= NULL
;
1873 screen
->fence_reference(screen
, &payload
->fence
, fence
);
1877 tc_call_fence_server_signal(struct pipe_context
*pipe
, union tc_payload
*payload
)
1879 pipe
->fence_server_signal(pipe
, payload
->fence
);
1880 pipe
->screen
->fence_reference(pipe
->screen
, &payload
->fence
, NULL
);
1884 tc_fence_server_signal(struct pipe_context
*_pipe
,
1885 struct pipe_fence_handle
*fence
)
1887 struct threaded_context
*tc
= threaded_context(_pipe
);
1888 struct pipe_screen
*screen
= tc
->pipe
->screen
;
1889 union tc_payload
*payload
= tc_add_small_call(tc
, TC_CALL_fence_server_signal
);
1891 payload
->fence
= NULL
;
1892 screen
->fence_reference(screen
, &payload
->fence
, fence
);
1895 static struct pipe_video_codec
*
1896 tc_create_video_codec(UNUSED
struct pipe_context
*_pipe
,
1897 UNUSED
const struct pipe_video_codec
*templ
)
1899 unreachable("Threaded context should not be enabled for video APIs");
1903 static struct pipe_video_buffer
*
1904 tc_create_video_buffer(UNUSED
struct pipe_context
*_pipe
,
1905 UNUSED
const struct pipe_video_buffer
*templ
)
1907 unreachable("Threaded context should not be enabled for video APIs");
1911 struct tc_context_param
{
1912 enum pipe_context_param param
;
1917 tc_call_set_context_param(struct pipe_context
*pipe
,
1918 union tc_payload
*payload
)
1920 struct tc_context_param
*p
= (struct tc_context_param
*)payload
;
1922 if (pipe
->set_context_param
)
1923 pipe
->set_context_param(pipe
, p
->param
, p
->value
);
1927 tc_set_context_param(struct pipe_context
*_pipe
,
1928 enum pipe_context_param param
,
1931 struct threaded_context
*tc
= threaded_context(_pipe
);
1933 if (tc
->pipe
->set_context_param
) {
1934 struct tc_context_param
*payload
=
1935 tc_add_struct_typed_call(tc
, TC_CALL_set_context_param
,
1938 payload
->param
= param
;
1939 payload
->value
= value
;
1942 if (param
== PIPE_CONTEXT_PARAM_PIN_THREADS_TO_L3_CACHE
) {
1943 /* Pin the gallium thread as requested. */
1944 util_pin_thread_to_L3(tc
->queue
.threads
[0], value
,
1945 util_cpu_caps
.cores_per_L3
);
1950 /********************************************************************
1951 * draw, launch, clear, blit, copy, flush
1954 struct tc_flush_payload
{
1955 struct threaded_context
*tc
;
1956 struct pipe_fence_handle
*fence
;
1961 tc_flush_queries(struct threaded_context
*tc
)
1963 struct threaded_query
*tq
, *tmp
;
1964 LIST_FOR_EACH_ENTRY_SAFE(tq
, tmp
, &tc
->unflushed_queries
, head_unflushed
) {
1965 LIST_DEL(&tq
->head_unflushed
);
1967 /* Memory release semantics: due to a possible race with
1968 * tc_get_query_result, we must ensure that the linked list changes
1969 * are visible before setting tq->flushed.
1971 p_atomic_set(&tq
->flushed
, true);
1976 tc_call_flush(struct pipe_context
*pipe
, union tc_payload
*payload
)
1978 struct tc_flush_payload
*p
= (struct tc_flush_payload
*)payload
;
1979 struct pipe_screen
*screen
= pipe
->screen
;
1981 pipe
->flush(pipe
, p
->fence
? &p
->fence
: NULL
, p
->flags
);
1982 screen
->fence_reference(screen
, &p
->fence
, NULL
);
1984 if (!(p
->flags
& PIPE_FLUSH_DEFERRED
))
1985 tc_flush_queries(p
->tc
);
1989 tc_flush(struct pipe_context
*_pipe
, struct pipe_fence_handle
**fence
,
1992 struct threaded_context
*tc
= threaded_context(_pipe
);
1993 struct pipe_context
*pipe
= tc
->pipe
;
1994 struct pipe_screen
*screen
= pipe
->screen
;
1995 bool async
= flags
& PIPE_FLUSH_DEFERRED
;
1997 if (flags
& PIPE_FLUSH_ASYNC
) {
1998 struct tc_batch
*last
= &tc
->batch_slots
[tc
->last
];
2000 /* Prefer to do the flush in the driver thread, but avoid the inter-thread
2001 * communication overhead if the driver thread is currently idle and the
2002 * caller is going to wait for the fence immediately anyway.
2004 if (!(util_queue_fence_is_signalled(&last
->fence
) &&
2005 (flags
& PIPE_FLUSH_HINT_FINISH
)))
2009 if (async
&& tc
->create_fence
) {
2011 struct tc_batch
*next
= &tc
->batch_slots
[tc
->next
];
2014 next
->token
= malloc(sizeof(*next
->token
));
2018 pipe_reference_init(&next
->token
->ref
, 1);
2019 next
->token
->tc
= tc
;
2022 screen
->fence_reference(screen
, fence
, tc
->create_fence(pipe
, next
->token
));
2027 struct tc_flush_payload
*p
=
2028 tc_add_struct_typed_call(tc
, TC_CALL_flush
, tc_flush_payload
);
2030 p
->fence
= fence
? *fence
: NULL
;
2031 p
->flags
= flags
| TC_FLUSH_ASYNC
;
2033 if (!(flags
& PIPE_FLUSH_DEFERRED
))
2039 tc_sync_msg(tc
, flags
& PIPE_FLUSH_END_OF_FRAME
? "end of frame" :
2040 flags
& PIPE_FLUSH_DEFERRED
? "deferred fence" : "normal");
2042 if (!(flags
& PIPE_FLUSH_DEFERRED
))
2043 tc_flush_queries(tc
);
2044 pipe
->flush(pipe
, fence
, flags
);
2047 /* This is actually variable-sized, because indirect isn't allocated if it's
2049 struct tc_full_draw_info
{
2050 struct pipe_draw_info draw
;
2051 struct pipe_draw_indirect_info indirect
;
2055 tc_call_draw_vbo(struct pipe_context
*pipe
, union tc_payload
*payload
)
2057 struct tc_full_draw_info
*info
= (struct tc_full_draw_info
*)payload
;
2059 pipe
->draw_vbo(pipe
, &info
->draw
);
2060 pipe_so_target_reference(&info
->draw
.count_from_stream_output
, NULL
);
2061 if (info
->draw
.index_size
)
2062 pipe_resource_reference(&info
->draw
.index
.resource
, NULL
);
2063 if (info
->draw
.indirect
) {
2064 pipe_resource_reference(&info
->indirect
.buffer
, NULL
);
2065 pipe_resource_reference(&info
->indirect
.indirect_draw_count
, NULL
);
2069 static struct tc_full_draw_info
*
2070 tc_add_draw_vbo(struct pipe_context
*_pipe
, bool indirect
)
2072 return (struct tc_full_draw_info
*)
2073 tc_add_sized_call(threaded_context(_pipe
), TC_CALL_draw_vbo
,
2074 indirect
? sizeof(struct tc_full_draw_info
) :
2075 sizeof(struct pipe_draw_info
));
2079 tc_draw_vbo(struct pipe_context
*_pipe
, const struct pipe_draw_info
*info
)
2081 struct threaded_context
*tc
= threaded_context(_pipe
);
2082 struct pipe_draw_indirect_info
*indirect
= info
->indirect
;
2083 unsigned index_size
= info
->index_size
;
2084 bool has_user_indices
= info
->has_user_indices
;
2086 if (index_size
&& has_user_indices
) {
2087 unsigned size
= info
->count
* index_size
;
2088 struct pipe_resource
*buffer
= NULL
;
2091 tc_assert(!indirect
);
2093 /* This must be done before adding draw_vbo, because it could generate
2094 * e.g. transfer_unmap and flush partially-uninitialized draw_vbo
2095 * to the driver if it was done afterwards.
2097 u_upload_data(tc
->base
.stream_uploader
, 0, size
, 4, info
->index
.user
,
2099 if (unlikely(!buffer
))
2102 struct tc_full_draw_info
*p
= tc_add_draw_vbo(_pipe
, false);
2103 p
->draw
.count_from_stream_output
= NULL
;
2104 pipe_so_target_reference(&p
->draw
.count_from_stream_output
,
2105 info
->count_from_stream_output
);
2106 memcpy(&p
->draw
, info
, sizeof(*info
));
2107 p
->draw
.has_user_indices
= false;
2108 p
->draw
.index
.resource
= buffer
;
2109 p
->draw
.start
= offset
/ index_size
;
2111 /* Non-indexed call or indexed with a real index buffer. */
2112 struct tc_full_draw_info
*p
= tc_add_draw_vbo(_pipe
, indirect
!= NULL
);
2113 p
->draw
.count_from_stream_output
= NULL
;
2114 pipe_so_target_reference(&p
->draw
.count_from_stream_output
,
2115 info
->count_from_stream_output
);
2117 tc_set_resource_reference(&p
->draw
.index
.resource
,
2118 info
->index
.resource
);
2120 memcpy(&p
->draw
, info
, sizeof(*info
));
2123 tc_set_resource_reference(&p
->draw
.indirect
->buffer
, indirect
->buffer
);
2124 tc_set_resource_reference(&p
->indirect
.indirect_draw_count
,
2125 indirect
->indirect_draw_count
);
2126 memcpy(&p
->indirect
, indirect
, sizeof(*indirect
));
2127 p
->draw
.indirect
= &p
->indirect
;
2133 tc_call_launch_grid(struct pipe_context
*pipe
, union tc_payload
*payload
)
2135 struct pipe_grid_info
*p
= (struct pipe_grid_info
*)payload
;
2137 pipe
->launch_grid(pipe
, p
);
2138 pipe_resource_reference(&p
->indirect
, NULL
);
2142 tc_launch_grid(struct pipe_context
*_pipe
,
2143 const struct pipe_grid_info
*info
)
2145 struct threaded_context
*tc
= threaded_context(_pipe
);
2146 struct pipe_grid_info
*p
= tc_add_struct_typed_call(tc
, TC_CALL_launch_grid
,
2148 assert(info
->input
== NULL
);
2150 tc_set_resource_reference(&p
->indirect
, info
->indirect
);
2151 memcpy(p
, info
, sizeof(*info
));
2155 tc_call_resource_copy_region(struct pipe_context
*pipe
, union tc_payload
*payload
)
2157 struct tc_resource_copy_region
*p
= (struct tc_resource_copy_region
*)payload
;
2159 pipe
->resource_copy_region(pipe
, p
->dst
, p
->dst_level
, p
->dstx
, p
->dsty
,
2160 p
->dstz
, p
->src
, p
->src_level
, &p
->src_box
);
2161 pipe_resource_reference(&p
->dst
, NULL
);
2162 pipe_resource_reference(&p
->src
, NULL
);
2166 tc_resource_copy_region(struct pipe_context
*_pipe
,
2167 struct pipe_resource
*dst
, unsigned dst_level
,
2168 unsigned dstx
, unsigned dsty
, unsigned dstz
,
2169 struct pipe_resource
*src
, unsigned src_level
,
2170 const struct pipe_box
*src_box
)
2172 struct threaded_context
*tc
= threaded_context(_pipe
);
2173 struct threaded_resource
*tdst
= threaded_resource(dst
);
2174 struct tc_resource_copy_region
*p
=
2175 tc_add_struct_typed_call(tc
, TC_CALL_resource_copy_region
,
2176 tc_resource_copy_region
);
2178 tc_set_resource_reference(&p
->dst
, dst
);
2179 p
->dst_level
= dst_level
;
2183 tc_set_resource_reference(&p
->src
, src
);
2184 p
->src_level
= src_level
;
2185 p
->src_box
= *src_box
;
2187 if (dst
->target
== PIPE_BUFFER
)
2188 util_range_add(&tdst
->valid_buffer_range
, dstx
, dstx
+ src_box
->width
);
2192 tc_call_blit(struct pipe_context
*pipe
, union tc_payload
*payload
)
2194 struct pipe_blit_info
*blit
= (struct pipe_blit_info
*)payload
;
2196 pipe
->blit(pipe
, blit
);
2197 pipe_resource_reference(&blit
->dst
.resource
, NULL
);
2198 pipe_resource_reference(&blit
->src
.resource
, NULL
);
2202 tc_blit(struct pipe_context
*_pipe
, const struct pipe_blit_info
*info
)
2204 struct threaded_context
*tc
= threaded_context(_pipe
);
2205 struct pipe_blit_info
*blit
=
2206 tc_add_struct_typed_call(tc
, TC_CALL_blit
, pipe_blit_info
);
2208 tc_set_resource_reference(&blit
->dst
.resource
, info
->dst
.resource
);
2209 tc_set_resource_reference(&blit
->src
.resource
, info
->src
.resource
);
2210 memcpy(blit
, info
, sizeof(*info
));
2213 struct tc_generate_mipmap
{
2214 struct pipe_resource
*res
;
2215 enum pipe_format format
;
2216 unsigned base_level
;
2217 unsigned last_level
;
2218 unsigned first_layer
;
2219 unsigned last_layer
;
2223 tc_call_generate_mipmap(struct pipe_context
*pipe
, union tc_payload
*payload
)
2225 struct tc_generate_mipmap
*p
= (struct tc_generate_mipmap
*)payload
;
2226 MAYBE_UNUSED
bool result
= pipe
->generate_mipmap(pipe
, p
->res
, p
->format
,
2232 pipe_resource_reference(&p
->res
, NULL
);
2236 tc_generate_mipmap(struct pipe_context
*_pipe
,
2237 struct pipe_resource
*res
,
2238 enum pipe_format format
,
2239 unsigned base_level
,
2240 unsigned last_level
,
2241 unsigned first_layer
,
2242 unsigned last_layer
)
2244 struct threaded_context
*tc
= threaded_context(_pipe
);
2245 struct pipe_context
*pipe
= tc
->pipe
;
2246 struct pipe_screen
*screen
= pipe
->screen
;
2247 unsigned bind
= PIPE_BIND_SAMPLER_VIEW
;
2249 if (util_format_is_depth_or_stencil(format
))
2250 bind
= PIPE_BIND_DEPTH_STENCIL
;
2252 bind
= PIPE_BIND_RENDER_TARGET
;
2254 if (!screen
->is_format_supported(screen
, format
, res
->target
,
2255 res
->nr_samples
, res
->nr_storage_samples
,
2259 struct tc_generate_mipmap
*p
=
2260 tc_add_struct_typed_call(tc
, TC_CALL_generate_mipmap
, tc_generate_mipmap
);
2262 tc_set_resource_reference(&p
->res
, res
);
2264 p
->base_level
= base_level
;
2265 p
->last_level
= last_level
;
2266 p
->first_layer
= first_layer
;
2267 p
->last_layer
= last_layer
;
2272 tc_call_flush_resource(struct pipe_context
*pipe
, union tc_payload
*payload
)
2274 pipe
->flush_resource(pipe
, payload
->resource
);
2275 pipe_resource_reference(&payload
->resource
, NULL
);
2279 tc_flush_resource(struct pipe_context
*_pipe
,
2280 struct pipe_resource
*resource
)
2282 struct threaded_context
*tc
= threaded_context(_pipe
);
2283 union tc_payload
*payload
= tc_add_small_call(tc
, TC_CALL_flush_resource
);
2285 tc_set_resource_reference(&payload
->resource
, resource
);
2289 tc_call_invalidate_resource(struct pipe_context
*pipe
, union tc_payload
*payload
)
2291 pipe
->invalidate_resource(pipe
, payload
->resource
);
2292 pipe_resource_reference(&payload
->resource
, NULL
);
2296 tc_invalidate_resource(struct pipe_context
*_pipe
,
2297 struct pipe_resource
*resource
)
2299 struct threaded_context
*tc
= threaded_context(_pipe
);
2301 if (resource
->target
== PIPE_BUFFER
) {
2302 tc_invalidate_buffer(tc
, threaded_resource(resource
));
2306 union tc_payload
*payload
= tc_add_small_call(tc
, TC_CALL_invalidate_resource
);
2307 tc_set_resource_reference(&payload
->resource
, resource
);
2312 union pipe_color_union color
;
2318 tc_call_clear(struct pipe_context
*pipe
, union tc_payload
*payload
)
2320 struct tc_clear
*p
= (struct tc_clear
*)payload
;
2321 pipe
->clear(pipe
, p
->buffers
, &p
->color
, p
->depth
, p
->stencil
);
2325 tc_clear(struct pipe_context
*_pipe
, unsigned buffers
,
2326 const union pipe_color_union
*color
, double depth
,
2329 struct threaded_context
*tc
= threaded_context(_pipe
);
2330 struct tc_clear
*p
= tc_add_struct_typed_call(tc
, TC_CALL_clear
, tc_clear
);
2332 p
->buffers
= buffers
;
2335 p
->stencil
= stencil
;
2339 tc_clear_render_target(struct pipe_context
*_pipe
,
2340 struct pipe_surface
*dst
,
2341 const union pipe_color_union
*color
,
2342 unsigned dstx
, unsigned dsty
,
2343 unsigned width
, unsigned height
,
2344 bool render_condition_enabled
)
2346 struct threaded_context
*tc
= threaded_context(_pipe
);
2347 struct pipe_context
*pipe
= tc
->pipe
;
2350 pipe
->clear_render_target(pipe
, dst
, color
, dstx
, dsty
, width
, height
,
2351 render_condition_enabled
);
2355 tc_clear_depth_stencil(struct pipe_context
*_pipe
,
2356 struct pipe_surface
*dst
, unsigned clear_flags
,
2357 double depth
, unsigned stencil
, unsigned dstx
,
2358 unsigned dsty
, unsigned width
, unsigned height
,
2359 bool render_condition_enabled
)
2361 struct threaded_context
*tc
= threaded_context(_pipe
);
2362 struct pipe_context
*pipe
= tc
->pipe
;
2365 pipe
->clear_depth_stencil(pipe
, dst
, clear_flags
, depth
, stencil
,
2366 dstx
, dsty
, width
, height
,
2367 render_condition_enabled
);
2370 struct tc_clear_buffer
{
2371 struct pipe_resource
*res
;
2374 char clear_value
[16];
2375 int clear_value_size
;
2379 tc_call_clear_buffer(struct pipe_context
*pipe
, union tc_payload
*payload
)
2381 struct tc_clear_buffer
*p
= (struct tc_clear_buffer
*)payload
;
2383 pipe
->clear_buffer(pipe
, p
->res
, p
->offset
, p
->size
, p
->clear_value
,
2384 p
->clear_value_size
);
2385 pipe_resource_reference(&p
->res
, NULL
);
2389 tc_clear_buffer(struct pipe_context
*_pipe
, struct pipe_resource
*res
,
2390 unsigned offset
, unsigned size
,
2391 const void *clear_value
, int clear_value_size
)
2393 struct threaded_context
*tc
= threaded_context(_pipe
);
2394 struct threaded_resource
*tres
= threaded_resource(res
);
2395 struct tc_clear_buffer
*p
=
2396 tc_add_struct_typed_call(tc
, TC_CALL_clear_buffer
, tc_clear_buffer
);
2398 tc_set_resource_reference(&p
->res
, res
);
2401 memcpy(p
->clear_value
, clear_value
, clear_value_size
);
2402 p
->clear_value_size
= clear_value_size
;
2404 util_range_add(&tres
->valid_buffer_range
, offset
, offset
+ size
);
2407 struct tc_clear_texture
{
2408 struct pipe_resource
*res
;
2410 struct pipe_box box
;
2415 tc_call_clear_texture(struct pipe_context
*pipe
, union tc_payload
*payload
)
2417 struct tc_clear_texture
*p
= (struct tc_clear_texture
*)payload
;
2419 pipe
->clear_texture(pipe
, p
->res
, p
->level
, &p
->box
, p
->data
);
2420 pipe_resource_reference(&p
->res
, NULL
);
2424 tc_clear_texture(struct pipe_context
*_pipe
, struct pipe_resource
*res
,
2425 unsigned level
, const struct pipe_box
*box
, const void *data
)
2427 struct threaded_context
*tc
= threaded_context(_pipe
);
2428 struct tc_clear_texture
*p
=
2429 tc_add_struct_typed_call(tc
, TC_CALL_clear_texture
, tc_clear_texture
);
2431 tc_set_resource_reference(&p
->res
, res
);
2434 memcpy(p
->data
, data
,
2435 util_format_get_blocksize(res
->format
));
2438 struct tc_resource_commit
{
2439 struct pipe_resource
*res
;
2441 struct pipe_box box
;
2446 tc_call_resource_commit(struct pipe_context
*pipe
, union tc_payload
*payload
)
2448 struct tc_resource_commit
*p
= (struct tc_resource_commit
*)payload
;
2450 pipe
->resource_commit(pipe
, p
->res
, p
->level
, &p
->box
, p
->commit
);
2451 pipe_resource_reference(&p
->res
, NULL
);
2455 tc_resource_commit(struct pipe_context
*_pipe
, struct pipe_resource
*res
,
2456 unsigned level
, struct pipe_box
*box
, bool commit
)
2458 struct threaded_context
*tc
= threaded_context(_pipe
);
2459 struct tc_resource_commit
*p
=
2460 tc_add_struct_typed_call(tc
, TC_CALL_resource_commit
, tc_resource_commit
);
2462 tc_set_resource_reference(&p
->res
, res
);
2466 return true; /* we don't care about the return value for this call */
2470 /********************************************************************
2474 struct tc_callback_payload
{
2475 void (*fn
)(void *data
);
2480 tc_call_callback(UNUSED
struct pipe_context
*pipe
, union tc_payload
*payload
)
2482 struct tc_callback_payload
*p
= (struct tc_callback_payload
*)payload
;
2488 tc_callback(struct pipe_context
*_pipe
, void (*fn
)(void *), void *data
,
2491 struct threaded_context
*tc
= threaded_context(_pipe
);
2493 if (asap
&& tc_is_sync(tc
)) {
2498 struct tc_callback_payload
*p
=
2499 tc_add_struct_typed_call(tc
, TC_CALL_callback
, tc_callback_payload
);
2505 /********************************************************************
2510 tc_destroy(struct pipe_context
*_pipe
)
2512 struct threaded_context
*tc
= threaded_context(_pipe
);
2513 struct pipe_context
*pipe
= tc
->pipe
;
2515 if (tc
->base
.const_uploader
&&
2516 tc
->base
.stream_uploader
!= tc
->base
.const_uploader
)
2517 u_upload_destroy(tc
->base
.const_uploader
);
2519 if (tc
->base
.stream_uploader
)
2520 u_upload_destroy(tc
->base
.stream_uploader
);
2524 if (util_queue_is_initialized(&tc
->queue
)) {
2525 util_queue_destroy(&tc
->queue
);
2527 for (unsigned i
= 0; i
< TC_MAX_BATCHES
; i
++) {
2528 util_queue_fence_destroy(&tc
->batch_slots
[i
].fence
);
2529 assert(!tc
->batch_slots
[i
].token
);
2533 slab_destroy_child(&tc
->pool_transfers
);
2534 assert(tc
->batch_slots
[tc
->next
].num_total_call_slots
== 0);
2535 pipe
->destroy(pipe
);
2536 os_free_aligned(tc
);
2539 static const tc_execute execute_func
[TC_NUM_CALLS
] = {
2540 #define CALL(name) tc_call_##name,
2541 #include "u_threaded_context_calls.h"
2546 * Wrap an existing pipe_context into a threaded_context.
2548 * \param pipe pipe_context to wrap
2549 * \param parent_transfer_pool parent slab pool set up for creating pipe_-
2550 * transfer objects; the driver should have one
2552 * \param replace_buffer callback for replacing a pipe_resource's storage
2553 * with another pipe_resource's storage.
2554 * \param out if successful, the threaded_context will be returned here in
2555 * addition to the return value if "out" != NULL
2557 struct pipe_context
*
2558 threaded_context_create(struct pipe_context
*pipe
,
2559 struct slab_parent_pool
*parent_transfer_pool
,
2560 tc_replace_buffer_storage_func replace_buffer
,
2561 tc_create_fence_func create_fence
,
2562 struct threaded_context
**out
)
2564 struct threaded_context
*tc
;
2566 STATIC_ASSERT(sizeof(union tc_payload
) <= 8);
2567 STATIC_ASSERT(sizeof(struct tc_call
) <= 16);
2574 if (!debug_get_bool_option("GALLIUM_THREAD", util_cpu_caps
.nr_cpus
> 1))
2577 tc
= os_malloc_aligned(sizeof(struct threaded_context
), 16);
2579 pipe
->destroy(pipe
);
2582 memset(tc
, 0, sizeof(*tc
));
2584 assert((uintptr_t)tc
% 16 == 0);
2585 /* These should be static asserts, but they don't work with MSVC */
2586 assert(offsetof(struct threaded_context
, batch_slots
) % 16 == 0);
2587 assert(offsetof(struct threaded_context
, batch_slots
[0].call
) % 16 == 0);
2588 assert(offsetof(struct threaded_context
, batch_slots
[0].call
[1]) % 16 == 0);
2589 assert(offsetof(struct threaded_context
, batch_slots
[1].call
) % 16 == 0);
2591 /* The driver context isn't wrapped, so set its "priv" to NULL. */
2595 tc
->replace_buffer_storage
= replace_buffer
;
2596 tc
->create_fence
= create_fence
;
2597 tc
->map_buffer_alignment
=
2598 pipe
->screen
->get_param(pipe
->screen
, PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT
);
2599 tc
->base
.priv
= pipe
; /* priv points to the wrapped driver context */
2600 tc
->base
.screen
= pipe
->screen
;
2601 tc
->base
.destroy
= tc_destroy
;
2602 tc
->base
.callback
= tc_callback
;
2604 tc
->base
.stream_uploader
= u_upload_clone(&tc
->base
, pipe
->stream_uploader
);
2605 if (pipe
->stream_uploader
== pipe
->const_uploader
)
2606 tc
->base
.const_uploader
= tc
->base
.stream_uploader
;
2608 tc
->base
.const_uploader
= u_upload_clone(&tc
->base
, pipe
->const_uploader
);
2610 if (!tc
->base
.stream_uploader
|| !tc
->base
.const_uploader
)
2613 /* The queue size is the number of batches "waiting". Batches are removed
2614 * from the queue before being executed, so keep one tc_batch slot for that
2615 * execution. Also, keep one unused slot for an unflushed batch.
2617 if (!util_queue_init(&tc
->queue
, "gdrv", TC_MAX_BATCHES
- 2, 1, 0))
2620 for (unsigned i
= 0; i
< TC_MAX_BATCHES
; i
++) {
2621 tc
->batch_slots
[i
].sentinel
= TC_SENTINEL
;
2622 tc
->batch_slots
[i
].pipe
= pipe
;
2623 util_queue_fence_init(&tc
->batch_slots
[i
].fence
);
2626 LIST_INITHEAD(&tc
->unflushed_queries
);
2628 slab_create_child(&tc
->pool_transfers
, parent_transfer_pool
);
2630 tc
->base
.set_context_param
= tc_set_context_param
; /* always set this */
2632 #define CTX_INIT(_member) \
2633 tc->base._member = tc->pipe->_member ? tc_##_member : NULL
2637 CTX_INIT(launch_grid
);
2638 CTX_INIT(resource_copy_region
);
2641 CTX_INIT(clear_render_target
);
2642 CTX_INIT(clear_depth_stencil
);
2643 CTX_INIT(clear_buffer
);
2644 CTX_INIT(clear_texture
);
2645 CTX_INIT(flush_resource
);
2646 CTX_INIT(generate_mipmap
);
2647 CTX_INIT(render_condition
);
2648 CTX_INIT(create_query
);
2649 CTX_INIT(create_batch_query
);
2650 CTX_INIT(destroy_query
);
2651 CTX_INIT(begin_query
);
2652 CTX_INIT(end_query
);
2653 CTX_INIT(get_query_result
);
2654 CTX_INIT(get_query_result_resource
);
2655 CTX_INIT(set_active_query_state
);
2656 CTX_INIT(create_blend_state
);
2657 CTX_INIT(bind_blend_state
);
2658 CTX_INIT(delete_blend_state
);
2659 CTX_INIT(create_sampler_state
);
2660 CTX_INIT(bind_sampler_states
);
2661 CTX_INIT(delete_sampler_state
);
2662 CTX_INIT(create_rasterizer_state
);
2663 CTX_INIT(bind_rasterizer_state
);
2664 CTX_INIT(delete_rasterizer_state
);
2665 CTX_INIT(create_depth_stencil_alpha_state
);
2666 CTX_INIT(bind_depth_stencil_alpha_state
);
2667 CTX_INIT(delete_depth_stencil_alpha_state
);
2668 CTX_INIT(create_fs_state
);
2669 CTX_INIT(bind_fs_state
);
2670 CTX_INIT(delete_fs_state
);
2671 CTX_INIT(create_vs_state
);
2672 CTX_INIT(bind_vs_state
);
2673 CTX_INIT(delete_vs_state
);
2674 CTX_INIT(create_gs_state
);
2675 CTX_INIT(bind_gs_state
);
2676 CTX_INIT(delete_gs_state
);
2677 CTX_INIT(create_tcs_state
);
2678 CTX_INIT(bind_tcs_state
);
2679 CTX_INIT(delete_tcs_state
);
2680 CTX_INIT(create_tes_state
);
2681 CTX_INIT(bind_tes_state
);
2682 CTX_INIT(delete_tes_state
);
2683 CTX_INIT(create_compute_state
);
2684 CTX_INIT(bind_compute_state
);
2685 CTX_INIT(delete_compute_state
);
2686 CTX_INIT(create_vertex_elements_state
);
2687 CTX_INIT(bind_vertex_elements_state
);
2688 CTX_INIT(delete_vertex_elements_state
);
2689 CTX_INIT(set_blend_color
);
2690 CTX_INIT(set_stencil_ref
);
2691 CTX_INIT(set_sample_mask
);
2692 CTX_INIT(set_min_samples
);
2693 CTX_INIT(set_clip_state
);
2694 CTX_INIT(set_constant_buffer
);
2695 CTX_INIT(set_framebuffer_state
);
2696 CTX_INIT(set_polygon_stipple
);
2697 CTX_INIT(set_scissor_states
);
2698 CTX_INIT(set_viewport_states
);
2699 CTX_INIT(set_window_rectangles
);
2700 CTX_INIT(set_sampler_views
);
2701 CTX_INIT(set_tess_state
);
2702 CTX_INIT(set_shader_buffers
);
2703 CTX_INIT(set_shader_images
);
2704 CTX_INIT(set_vertex_buffers
);
2705 CTX_INIT(create_stream_output_target
);
2706 CTX_INIT(stream_output_target_destroy
);
2707 CTX_INIT(set_stream_output_targets
);
2708 CTX_INIT(create_sampler_view
);
2709 CTX_INIT(sampler_view_destroy
);
2710 CTX_INIT(create_surface
);
2711 CTX_INIT(surface_destroy
);
2712 CTX_INIT(transfer_map
);
2713 CTX_INIT(transfer_flush_region
);
2714 CTX_INIT(transfer_unmap
);
2715 CTX_INIT(buffer_subdata
);
2716 CTX_INIT(texture_subdata
);
2717 CTX_INIT(texture_barrier
);
2718 CTX_INIT(memory_barrier
);
2719 CTX_INIT(resource_commit
);
2720 CTX_INIT(create_video_codec
);
2721 CTX_INIT(create_video_buffer
);
2722 CTX_INIT(set_compute_resources
);
2723 CTX_INIT(set_global_binding
);
2724 CTX_INIT(get_sample_position
);
2725 CTX_INIT(invalidate_resource
);
2726 CTX_INIT(get_device_reset_status
);
2727 CTX_INIT(set_device_reset_callback
);
2728 CTX_INIT(dump_debug_state
);
2729 CTX_INIT(set_log_context
);
2730 CTX_INIT(emit_string_marker
);
2731 CTX_INIT(set_debug_callback
);
2732 CTX_INIT(create_fence_fd
);
2733 CTX_INIT(fence_server_sync
);
2734 CTX_INIT(fence_server_signal
);
2735 CTX_INIT(get_timestamp
);
2736 CTX_INIT(create_texture_handle
);
2737 CTX_INIT(delete_texture_handle
);
2738 CTX_INIT(make_texture_handle_resident
);
2739 CTX_INIT(create_image_handle
);
2740 CTX_INIT(delete_image_handle
);
2741 CTX_INIT(make_image_handle_resident
);
2750 tc_destroy(&tc
->base
);