mirrors/gstreamer
mirrors/gstreamer
1
1
Fork 0
mirror of https://gitlab.freedesktop.org/gstreamer/gstreamer.git synced 2024-11-28 04:31:06 +00:00
Code Issues 1 Projects Releases Wiki Activity

make v4l element do all calls in one thread, this makes performance much better (because of kernel locking, more thre...

Browse source 
Original commit message from CVS:
make v4l element do all calls in one thread, this makes performance much better (because of kernel locking, more threads doesn't make sense anyway)
This commit is contained in:
Ronald S. Bultje 2003-02-06 18:45:35 +00:00
parent 54341cbe3b
commit 39e02b7779
2 changed files with 83 additions and 27 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

1
sys/v4l/gstv4lsrc.h
View file

@ -65,6 +65,7 @@ struct _GstV4lSrc {
/* num of queued frames and some GThread stuff to wait if there's not enough */
guint16 num_queued_frames;
guint64 total_queued_frames;
GMutex * mutex_queued_frames;
GCond * cond_queued_frames;

109
sys/v4l/v4lsrc_calls.c
View file

@ -77,12 +77,9 @@ gst_v4lsrc_queue_frame (GstV4lSrc *v4lsrc,
v4lsrc->mmap.frame = num;
g_mutex_lock(v4lsrc->mutex_queued_frames);
if (v4lsrc->frame_queued[num] < 0)
if (v4lsrc->frame_queued[num] == -1)
{
//v4lsrc->frame_queued[num] = 0;
g_mutex_unlock(v4lsrc->mutex_queued_frames);
return TRUE;
}
@ -91,15 +88,12 @@ gst_v4lsrc_queue_frame (GstV4lSrc *v4lsrc,
gst_element_error(GST_ELEMENT(v4lsrc),
"Error queueing a buffer (%d): %s",
num, g_strerror(errno));
g_mutex_unlock(v4lsrc->mutex_queued_frames);
return FALSE;
}
v4lsrc->frame_queued[num] = 1;
v4lsrc->num_queued_frames++;
g_cond_broadcast(v4lsrc->cond_queued_frames);
g_mutex_unlock(v4lsrc->mutex_queued_frames);
v4lsrc->total_queued_frames++;
return TRUE;
}
@ -119,29 +113,86 @@ gst_v4lsrc_soft_sync_thread (void *arg)
DEBUG("starting software sync thread");
#if 0
/* Allow easy shutting down by other processes... */
pthread_setcancelstate( PTHREAD_CANCEL_ENABLE, NULL );
pthread_setcanceltype( PTHREAD_CANCEL_ASYNCHRONOUS, NULL );
#endif
while (1)
{
/* are there queued frames left? */
g_mutex_lock(v4lsrc->mutex_queued_frames);
if (v4lsrc->num_queued_frames < MIN_BUFFERS_QUEUED)
/* this cycle is non-onbligatory - we just queue frames
* as they become available, below, we'll wait for queues
* if we don't have enough of them */
while (1)
{
if (v4lsrc->frame_queued[frame] < 0) {
gint qframe = v4lsrc->total_queued_frames % v4lsrc->mbuf.frames;
g_mutex_lock(v4lsrc->mutex_queued_frames);
if (v4lsrc->frame_queued[qframe] == -2)
{
if (!gst_v4lsrc_queue_frame(v4lsrc, qframe))
{
g_mutex_unlock(v4lsrc->mutex_queued_frames);
g_mutex_lock(v4lsrc->mutex_soft_sync);
/* note that we use *frame* here, not *qframe* -
* reason is simple, the thread waiting for us needs
* to know that we stopped syncing on *this* frame,
* we didn't even start with qframe yet */
v4lsrc->isready_soft_sync[frame] = -1;
g_cond_broadcast(v4lsrc->cond_soft_sync[frame]);
g_mutex_unlock(v4lsrc->mutex_soft_sync);
goto end;
}
}
else {
g_mutex_unlock(v4lsrc->mutex_queued_frames);
break;
}
DEBUG("Waiting for new frames to be queued (%d < %d)",
v4lsrc->num_queued_frames, MIN_BUFFERS_QUEUED);
g_cond_wait(v4lsrc->cond_queued_frames, v4lsrc->mutex_queued_frames);
g_mutex_unlock(v4lsrc->mutex_queued_frames);
}
/* are there queued frames left? */
while (v4lsrc->num_queued_frames < MIN_BUFFERS_QUEUED)
{
gint qframe = v4lsrc->total_queued_frames % v4lsrc->mbuf.frames;
g_mutex_lock(v4lsrc->mutex_queued_frames);
if (v4lsrc->frame_queued[frame] == -1) {
g_mutex_unlock(v4lsrc->mutex_queued_frames);
break;
}
DEBUG("Waiting for new frames to be queued (%d < %d, frame=%d)",
v4lsrc->num_queued_frames, MIN_BUFFERS_QUEUED, qframe);
/* sleep for new buffers to be completed encoding. After that,
* requeue them so we have more than MIN_QUEUES_NEEDED buffers
* free */
while (v4lsrc->frame_queued[qframe] != -2)
{
g_cond_wait(v4lsrc->cond_queued_frames,
v4lsrc->mutex_queued_frames);
if (v4lsrc->frame_queued[qframe] == -1) {
g_mutex_unlock(v4lsrc->mutex_queued_frames);
goto end;
}
}
if (!gst_v4lsrc_queue_frame(v4lsrc, qframe))
{
g_mutex_unlock(v4lsrc->mutex_queued_frames);
g_mutex_lock(v4lsrc->mutex_soft_sync);
/* note that we use *frame* here, not *qframe* -
* reason is simple, the thread waiting for us needs
* to know that we stopped syncing on *this* frame,
* we didn't even start with qframe yet */
v4lsrc->isready_soft_sync[frame] = -1;
g_cond_broadcast(v4lsrc->cond_soft_sync[frame]);
g_mutex_unlock(v4lsrc->mutex_soft_sync);
goto end;
}
g_mutex_unlock(v4lsrc->mutex_queued_frames);
}
g_mutex_unlock(v4lsrc->mutex_queued_frames);
if (!v4lsrc->num_queued_frames)
{
@ -363,6 +414,7 @@ gst_v4lsrc_capture_start (GstV4lSrc *v4lsrc)
GST_V4L_CHECK_ACTIVE(GST_V4LELEMENT(v4lsrc));
v4lsrc->num_queued_frames = 0;
v4lsrc->total_queued_frames = 0;
/* queue all buffers, this starts streaming capture */
for (n=0;n<v4lsrc->mbuf.frames;n++)
@ -441,9 +493,11 @@ gst_v4lsrc_requeue_frame (GstV4lSrc *v4lsrc, gint num)
GST_V4L_CHECK_OPEN(GST_V4LELEMENT(v4lsrc));
GST_V4L_CHECK_ACTIVE(GST_V4LELEMENT(v4lsrc));
/* and let's queue the buffer */
if (!gst_v4lsrc_queue_frame(v4lsrc, num))
return FALSE;
/* mark frame as 'ready to requeue' */
g_mutex_lock(v4lsrc->mutex_queued_frames);
v4lsrc->frame_queued[num] = -2;
g_cond_broadcast(v4lsrc->cond_queued_frames);
g_mutex_unlock(v4lsrc->mutex_queued_frames);
return TRUE;
}
@ -464,7 +518,8 @@ gst_v4lsrc_capture_stop (GstV4lSrc *v4lsrc)
GST_V4L_CHECK_OPEN(GST_V4LELEMENT(v4lsrc));
GST_V4L_CHECK_ACTIVE(GST_V4LELEMENT(v4lsrc));
/* we actually need to sync on all queued buffers but not on the non-queued ones */
/* we actually need to sync on all queued buffers but
* not on the non-queued ones */
g_mutex_lock(v4lsrc->mutex_queued_frames);
for (n=0;n<v4lsrc->mbuf.frames;n++)
v4lsrc->frame_queued[n] = -1;