2 * Copyright © 2012 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
26 * Support functions for the glthread feature of Mesa.
28 * In multicore systems, many applications end up CPU-bound with about half
29 * their time spent inside their rendering thread and half inside Mesa. To
30 * alleviate this, we put a shim layer in Mesa at the GL dispatch level that
31 * quickly logs the GL commands to a buffer to be processed by a worker
35 #include "main/mtypes.h"
36 #include "main/glthread.h"
37 #include "main/marshal.h"
38 #include "util/u_atomic.h"
39 #include "util/u_thread.h"
43 glthread_unmarshal_batch(void *job
, int thread_index
)
45 struct glthread_batch
*batch
= (struct glthread_batch
*)job
;
46 struct gl_context
*ctx
= batch
->ctx
;
48 int used
= batch
->used
;
49 uint8_t *buffer
= batch
->buffer
;
51 _glapi_set_dispatch(ctx
->CurrentServerDispatch
);
54 const struct marshal_cmd_base
*cmd
=
55 (const struct marshal_cmd_base
*)&buffer
[pos
];
57 _mesa_unmarshal_dispatch
[cmd
->cmd_id
](ctx
, cmd
);
66 glthread_thread_initialization(void *job
, int thread_index
)
68 struct gl_context
*ctx
= (struct gl_context
*)job
;
70 ctx
->Driver
.SetBackgroundContext(ctx
, &ctx
->GLThread
->stats
);
71 _glapi_set_context(ctx
);
75 _mesa_glthread_init(struct gl_context
*ctx
)
77 struct glthread_state
*glthread
= calloc(1, sizeof(*glthread
));
82 if (!util_queue_init(&glthread
->queue
, "gl", MARSHAL_MAX_BATCHES
- 2,
88 ctx
->MarshalExec
= _mesa_create_marshal_table(ctx
);
89 if (!ctx
->MarshalExec
) {
90 util_queue_destroy(&glthread
->queue
);
95 for (unsigned i
= 0; i
< MARSHAL_MAX_BATCHES
; i
++) {
96 glthread
->batches
[i
].ctx
= ctx
;
97 util_queue_fence_init(&glthread
->batches
[i
].fence
);
100 glthread
->stats
.queue
= &glthread
->queue
;
101 ctx
->CurrentClientDispatch
= ctx
->MarshalExec
;
102 ctx
->GLThread
= glthread
;
104 /* Execute the thread initialization function in the thread. */
105 struct util_queue_fence fence
;
106 util_queue_fence_init(&fence
);
107 util_queue_add_job(&glthread
->queue
, ctx
, &fence
,
108 glthread_thread_initialization
, NULL
, 0);
109 util_queue_fence_wait(&fence
);
110 util_queue_fence_destroy(&fence
);
114 _mesa_glthread_destroy(struct gl_context
*ctx
)
116 struct glthread_state
*glthread
= ctx
->GLThread
;
121 _mesa_glthread_finish(ctx
);
122 util_queue_destroy(&glthread
->queue
);
124 for (unsigned i
= 0; i
< MARSHAL_MAX_BATCHES
; i
++)
125 util_queue_fence_destroy(&glthread
->batches
[i
].fence
);
128 ctx
->GLThread
= NULL
;
130 _mesa_glthread_restore_dispatch(ctx
, "destroy");
134 _mesa_glthread_restore_dispatch(struct gl_context
*ctx
, const char *func
)
136 /* Remove ourselves from the dispatch table except if another ctx/thread
137 * already installed a new dispatch table.
139 * Typically glxMakeCurrent will bind a new context (install new table) then
140 * old context might be deleted.
142 if (_glapi_get_dispatch() == ctx
->MarshalExec
) {
143 ctx
->CurrentClientDispatch
= ctx
->CurrentServerDispatch
;
144 _glapi_set_dispatch(ctx
->CurrentClientDispatch
);
146 printf("glthread disabled: %s\n", func
);
152 _mesa_glthread_flush_batch(struct gl_context
*ctx
)
154 struct glthread_state
*glthread
= ctx
->GLThread
;
158 struct glthread_batch
*next
= &glthread
->batches
[glthread
->next
];
162 /* Debug: execute the batch immediately from this thread.
164 * Note that glthread_unmarshal_batch() changes the dispatch table so we'll
165 * need to restore it when it returns.
168 glthread_unmarshal_batch(next
, 0);
169 _glapi_set_dispatch(ctx
->CurrentClientDispatch
);
173 p_atomic_add(&glthread
->stats
.num_offloaded_items
, next
->used
);
175 util_queue_add_job(&glthread
->queue
, next
, &next
->fence
,
176 glthread_unmarshal_batch
, NULL
, 0);
177 glthread
->last
= glthread
->next
;
178 glthread
->next
= (glthread
->next
+ 1) % MARSHAL_MAX_BATCHES
;
182 * Waits for all pending batches have been unmarshaled.
184 * This can be used by the main thread to synchronize access to the context,
185 * since the worker thread will be idle after this.
188 _mesa_glthread_finish(struct gl_context
*ctx
)
190 struct glthread_state
*glthread
= ctx
->GLThread
;
194 /* If this is called from the worker thread, then we've hit a path that
195 * might be called from either the main thread or the worker (such as some
196 * dri interface entrypoints), in which case we don't need to actually
197 * synchronize against ourself.
199 if (u_thread_is_self(glthread
->queue
.threads
[0]))
202 struct glthread_batch
*last
= &glthread
->batches
[glthread
->last
];
203 struct glthread_batch
*next
= &glthread
->batches
[glthread
->next
];
206 if (!util_queue_fence_is_signalled(&last
->fence
)) {
207 util_queue_fence_wait(&last
->fence
);
212 p_atomic_add(&glthread
->stats
.num_direct_items
, next
->used
);
214 /* Since glthread_unmarshal_batch changes the dispatch to direct,
215 * restore it after it's done.
217 struct _glapi_table
*dispatch
= _glapi_get_dispatch();
218 glthread_unmarshal_batch(next
, 0);
219 _glapi_set_dispatch(dispatch
);
221 /* It's not a sync because we don't enqueue partial batches, but
222 * it would be a sync if we did. So count it anyway.
228 p_atomic_inc(&glthread
->stats
.num_syncs
);