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 "main/marshal_generated.h"
43 glthread_allocate_batch(struct gl_context
*ctx
)
45 struct glthread_state
*glthread
= ctx
->GLThread
;
47 /* TODO: handle memory allocation failure. */
48 glthread
->batch
= calloc(1, sizeof(*glthread
->batch
));
51 glthread
->batch
->buffer
= malloc(MARSHAL_MAX_CMD_SIZE
);
55 glthread_unmarshal_batch(struct gl_context
*ctx
, struct glthread_batch
*batch
)
59 _glapi_set_dispatch(ctx
->CurrentServerDispatch
);
61 while (pos
< batch
->used
)
62 pos
+= _mesa_unmarshal_dispatch_cmd(ctx
, &batch
->buffer
[pos
]);
64 assert(pos
== batch
->used
);
71 glthread_worker(void *data
)
73 struct gl_context
*ctx
= data
;
74 struct glthread_state
*glthread
= ctx
->GLThread
;
76 ctx
->Driver
.SetBackgroundContext(ctx
);
77 _glapi_set_context(ctx
);
79 pthread_mutex_lock(&glthread
->mutex
);
82 struct glthread_batch
*batch
;
84 /* Block (dropping the lock) until new work arrives for us. */
85 while (!glthread
->batch_queue
&& !glthread
->shutdown
) {
86 pthread_cond_broadcast(&glthread
->work_done
);
87 pthread_cond_wait(&glthread
->new_work
, &glthread
->mutex
);
90 batch
= glthread
->batch_queue
;
92 if (glthread
->shutdown
&& !batch
) {
93 pthread_cond_broadcast(&glthread
->work_done
);
94 pthread_mutex_unlock(&glthread
->mutex
);
97 glthread
->batch_queue
= batch
->next
;
98 if (glthread
->batch_queue_tail
== &batch
->next
)
99 glthread
->batch_queue_tail
= &glthread
->batch_queue
;
101 glthread
->busy
= true;
102 pthread_mutex_unlock(&glthread
->mutex
);
104 glthread_unmarshal_batch(ctx
, batch
);
106 pthread_mutex_lock(&glthread
->mutex
);
107 glthread
->busy
= false;
115 _mesa_glthread_init(struct gl_context
*ctx
)
117 struct glthread_state
*glthread
= calloc(1, sizeof(*glthread
));
122 ctx
->MarshalExec
= _mesa_create_marshal_table(ctx
);
123 if (!ctx
->MarshalExec
) {
128 ctx
->CurrentClientDispatch
= ctx
->MarshalExec
;
130 pthread_mutex_init(&glthread
->mutex
, NULL
);
131 pthread_cond_init(&glthread
->new_work
, NULL
);
132 pthread_cond_init(&glthread
->work_done
, NULL
);
134 glthread
->batch_queue_tail
= &glthread
->batch_queue
;
135 ctx
->GLThread
= glthread
;
137 glthread_allocate_batch(ctx
);
139 pthread_create(&glthread
->thread
, NULL
, glthread_worker
, ctx
);
143 _mesa_glthread_destroy(struct gl_context
*ctx
)
145 struct glthread_state
*glthread
= ctx
->GLThread
;
150 _mesa_glthread_flush_batch(ctx
);
152 pthread_mutex_lock(&glthread
->mutex
);
153 glthread
->shutdown
= true;
154 pthread_cond_broadcast(&glthread
->new_work
);
155 pthread_mutex_unlock(&glthread
->mutex
);
157 /* Since this waits for the thread to exit, it means that all queued work
158 * will have been completed.
160 pthread_join(glthread
->thread
, NULL
);
162 pthread_cond_destroy(&glthread
->new_work
);
163 pthread_cond_destroy(&glthread
->work_done
);
164 pthread_mutex_destroy(&glthread
->mutex
);
166 /* Due to the join above, there should be one empty batch allocated at this
167 * point, and no batches queued.
169 assert(!glthread
->batch
->used
);
170 assert(!glthread
->batch
->next
);
171 free(glthread
->batch
);
172 assert(!glthread
->batch_queue
);
175 ctx
->GLThread
= NULL
;
177 /* Remove ourselves from the dispatch table. */
178 ctx
->CurrentClientDispatch
= ctx
->CurrentServerDispatch
;
179 _glapi_set_dispatch(ctx
->CurrentClientDispatch
);
183 _mesa_glthread_flush_batch(struct gl_context
*ctx
)
185 struct glthread_state
*glthread
= ctx
->GLThread
;
186 struct glthread_batch
*batch
;
191 batch
= glthread
->batch
;
195 /* Immediately reallocate a new batch, since the next marshalled call would
198 glthread_allocate_batch(ctx
);
200 /* Debug: execute the batch immediately from this thread.
202 * Note that glthread_unmarshal_batch() changes the dispatch table so we'll
203 * need to restore it when it returns.
206 glthread_unmarshal_batch(ctx
, batch
);
207 _glapi_set_dispatch(ctx
->CurrentClientDispatch
);
211 pthread_mutex_lock(&glthread
->mutex
);
212 *glthread
->batch_queue_tail
= batch
;
213 glthread
->batch_queue_tail
= &batch
->next
;
214 pthread_cond_broadcast(&glthread
->new_work
);
215 pthread_mutex_unlock(&glthread
->mutex
);
219 * Waits for all pending batches have been unmarshaled.
221 * This can be used by the main thread to synchronize access to the context,
222 * since the worker thread will be idle after this.
225 _mesa_glthread_finish(struct gl_context
*ctx
)
227 struct glthread_state
*glthread
= ctx
->GLThread
;
232 /* If this is called from the worker thread, then we've hit a path that
233 * might be called from either the main thread or the worker (such as some
234 * dri interface entrypoints), in which case we don't need to actually
235 * synchronize against ourself.
237 if (pthread_self() == glthread
->thread
)
240 _mesa_glthread_flush_batch(ctx
);
242 pthread_mutex_lock(&glthread
->mutex
);
244 while (glthread
->batch_queue
|| glthread
->busy
)
245 pthread_cond_wait(&glthread
->work_done
, &glthread
->mutex
);
247 pthread_mutex_unlock(&glthread
->mutex
);