*
* You should have received a copy of the GNU Library General Public
* License along with this library; if not, write to the
- * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
- * Boston, MA 02111-1307, USA.
+ * Free Software Foundation, Inc., 51 Franklin St, Fifth Floor,
+ * Boston, MA 02110-1301, USA.
*/
/**
* <itemizedlist><title>Multiple streamhandling</title>
* <listitem><para>
* The element handles queueing data on more than one stream at once. To
- * achieve such a feature it has request sink pads (sink%d) and
- * 'sometimes' src pads (src%d).
+ * achieve such a feature it has request sink pads (sink%u) and
+ * 'sometimes' src pads (src%u).
* </para><para>
- * When requesting a given sinkpad with gst_element_get_request_pad(),
+ * When requesting a given sinkpad with gst_element_request_pad(),
* the associated srcpad for that stream will be created.
* Example: requesting sink1 will generate src1.
* </para></listitem>
* queues is filled.
* Both signals are emitted from the context of the streaming thread.
* </para>
+ * <para>
+ * When using #GstMultiQueue:sync-by-running-time the unlinked streams will
+ * be throttled by the highest running-time of linked streams. This allows
+ * further relinking of those unlinked streams without them being in the
+ * future (i.e. to achieve gapless playback).
+ * When dealing with streams which have got different consumption requirements
+ * downstream (ex: video decoders which will consume more buffer (in time) than
+ * audio decoders), it is recommended to group streams of the same type
+ * by using the pad "group-id" property. This will further throttle streams
+ * in time within that group.
+ * </para>
* </refsect2>
- *
- * Last reviewed on 2008-01-25 (0.10.17)
*/
#ifdef HAVE_CONFIG_H
{
/* unique identifier of the queue */
guint id;
+ /* group of streams to which this queue belongs to */
+ guint groupid;
+ GstClockTimeDiff group_high_time;
GstMultiQueue *mqueue;
/* segments */
GstSegment sink_segment;
GstSegment src_segment;
+ gboolean has_src_segment; /* preferred over initializing the src_segment to
+ * UNDEFINED as this doesn't requires adding ifs
+ * in every segment usage */
/* position of src/sink */
- GstClockTime sinktime, srctime;
+ GstClockTimeDiff sinktime, srctime;
+ /* cached input value, used for interleave */
+ GstClockTimeDiff cached_sinktime;
/* TRUE if either position needs to be recalculated */
gboolean sink_tainted, src_tainted;
GstDataQueueSize max_size, extra_size;
GstClockTime cur_time;
gboolean is_eos;
+ gboolean is_sparse;
gboolean flushing;
+ gboolean active;
/* Protected by global lock */
guint32 nextid; /* ID of the next object waiting to be pushed */
guint32 oldid; /* ID of the last object pushed (last in a series) */
guint32 last_oldid; /* Previously observed old_id, reset to MAXUINT32 on flush */
- GstClockTime next_time; /* End running time of next buffer to be pushed */
- GstClockTime last_time; /* Start running time of last pushed buffer */
+ GstClockTimeDiff next_time; /* End running time of next buffer to be pushed */
+ GstClockTimeDiff last_time; /* Start running time of last pushed buffer */
GCond turn; /* SingleQueue turn waiting conditional */
/* for serialized queries */
GCond query_handled;
gboolean last_query;
+ GstQuery *last_handled_query;
+
+ /* For interleave calculation */
+ GThread *thread;
};
GDestroyNotify destroy;
guint32 posid;
+
+ gboolean is_query;
};
static GstSingleQueue *gst_single_queue_new (GstMultiQueue * mqueue, guint id);
static void wake_up_next_non_linked (GstMultiQueue * mq);
static void compute_high_id (GstMultiQueue * mq);
-static void compute_high_time (GstMultiQueue * mq);
+static void compute_high_time (GstMultiQueue * mq, guint groupid);
static void single_queue_overrun_cb (GstDataQueue * dq, GstSingleQueue * sq);
static void single_queue_underrun_cb (GstDataQueue * dq, GstSingleQueue * sq);
+static void update_buffering (GstMultiQueue * mq, GstSingleQueue * sq);
+static void gst_multi_queue_post_buffering (GstMultiQueue * mq);
+static void recheck_buffering_status (GstMultiQueue * mq);
+
+static void gst_single_queue_flush_queue (GstSingleQueue * sq, gboolean full);
+
static GstStaticPadTemplate sinktemplate = GST_STATIC_PAD_TEMPLATE ("sink_%u",
GST_PAD_SINK,
GST_PAD_REQUEST,
#define DEFAULT_EXTRA_SIZE_TIME 3 * GST_SECOND
#define DEFAULT_USE_BUFFERING FALSE
-#define DEFAULT_LOW_PERCENT 10
-#define DEFAULT_HIGH_PERCENT 99
+#define DEFAULT_LOW_WATERMARK 0.01
+#define DEFAULT_HIGH_WATERMARK 0.99
#define DEFAULT_SYNC_BY_RUNNING_TIME FALSE
+#define DEFAULT_USE_INTERLEAVE FALSE
+#define DEFAULT_UNLINKED_CACHE_TIME 250 * GST_MSECOND
enum
{
PROP_USE_BUFFERING,
PROP_LOW_PERCENT,
PROP_HIGH_PERCENT,
+ PROP_LOW_WATERMARK,
+ PROP_HIGH_WATERMARK,
PROP_SYNC_BY_RUNNING_TIME,
+ PROP_USE_INTERLEAVE,
+ PROP_UNLINKED_CACHE_TIME,
PROP_LAST
};
+/* Explanation for buffer levels and percentages:
+ *
+ * The buffering_level functions here return a value in a normalized range
+ * that specifies the current fill level of a queue. The range goes from 0 to
+ * MAX_BUFFERING_LEVEL. The low/high watermarks also use this same range.
+ *
+ * This is not to be confused with the buffering_percent value, which is
+ * a *relative* quantity - relative to the low/high watermarks.
+ * buffering_percent = 0% means overall buffering_level is at the low watermark.
+ * buffering_percent = 100% means overall buffering_level is at the high watermark.
+ * buffering_percent is used for determining if the fill level has reached
+ * the high watermark, and for producing BUFFERING messages. This value
+ * always uses a 0..100 range (since it is a percentage).
+ *
+ * To avoid future confusions, whenever "buffering level" is mentioned, it
+ * refers to the absolute level which is in the 0..MAX_BUFFERING_LEVEL
+ * range. Whenever "buffering_percent" is mentioned, it refers to the
+ * percentage value that is relative to the low/high watermark. */
+
+/* Using a buffering level range of 0..1000000 to allow for a
+ * resolution in ppm (1 ppm = 0.0001%) */
+#define MAX_BUFFERING_LEVEL 1000000
+
+/* How much 1% makes up in the buffer level range */
+#define BUF_LEVEL_PERCENT_FACTOR ((MAX_BUFFERING_LEVEL) / 100)
+
+/* GstMultiQueuePad */
+
+#define DEFAULT_PAD_GROUP_ID 0
+
+enum
+{
+ PROP_PAD_0,
+ PROP_PAD_GROUP_ID,
+};
+
+#define GST_TYPE_MULTIQUEUE_PAD (gst_multiqueue_pad_get_type())
+#define GST_MULTIQUEUE_PAD(obj) (G_TYPE_CHECK_INSTANCE_CAST((obj),GST_TYPE_MULTIQUEUE_PAD,GstMultiQueuePad))
+#define GST_IS_MULTIQUEUE_PAD(obj) (G_TYPE_CHECK_INSTANCE_TYPE((obj),GST_TYPE_MULTIQUEUE_PAD))
+#define GST_MULTIQUEUE_PAD_CLASS(klass) (G_TYPE_CHECK_CLASS_CAST((klass) ,GST_TYPE_MULTIQUEUE_PAD,GstMultiQueuePadClass))
+#define GST_IS_MULTIQUEUE_PAD_CLASS(klass) (G_TYPE_CHECK_CLASS_TYPE((klass) ,GST_TYPE_MULTIQUEUE_PAD))
+#define GST_MULTIQUEUE_PAD_GET_CLASS(obj) (G_TYPE_INSTANCE_GET_CLASS((obj) ,GST_TYPE_MULTIQUEUE_PAD,GstMultiQueuePadClass))
+
+struct _GstMultiQueuePad
+{
+ GstPad parent;
+
+ GstSingleQueue *sq;
+};
+
+struct _GstMultiQueuePadClass
+{
+ GstPadClass parent_class;
+};
+
+GType gst_multiqueue_pad_get_type (void);
+
+G_DEFINE_TYPE (GstMultiQueuePad, gst_multiqueue_pad, GST_TYPE_PAD);
+static void
+gst_multiqueue_pad_get_property (GObject * object, guint prop_id,
+ GValue * value, GParamSpec * pspec)
+{
+ GstMultiQueuePad *pad = GST_MULTIQUEUE_PAD (object);
+
+ switch (prop_id) {
+ case PROP_PAD_GROUP_ID:
+ if (pad->sq)
+ g_value_set_uint (value, pad->sq->groupid);
+ break;
+ default:
+ G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
+ break;
+ }
+}
+
+static void
+gst_multiqueue_pad_set_property (GObject * object, guint prop_id,
+ const GValue * value, GParamSpec * pspec)
+{
+ GstMultiQueuePad *pad = GST_MULTIQUEUE_PAD (object);
+
+ switch (prop_id) {
+ case PROP_PAD_GROUP_ID:
+ GST_OBJECT_LOCK (pad);
+ if (pad->sq)
+ pad->sq->groupid = g_value_get_uint (value);
+ GST_OBJECT_UNLOCK (pad);
+ break;
+ default:
+ G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
+ break;
+ }
+}
+
+static void
+gst_multiqueue_pad_class_init (GstMultiQueuePadClass * klass)
+{
+ GObjectClass *gobject_class = (GObjectClass *) klass;
+
+ gobject_class->set_property = gst_multiqueue_pad_set_property;
+ gobject_class->get_property = gst_multiqueue_pad_get_property;
+
+ /**
+ * GstMultiQueuePad:group-id:
+ *
+ * Group to which this pad belongs.
+ *
+ * Since: 1.10
+ */
+ g_object_class_install_property (gobject_class, PROP_PAD_GROUP_ID,
+ g_param_spec_uint ("group-id", "Group ID",
+ "Group to which this pad belongs", 0, G_MAXUINT32,
+ DEFAULT_PAD_GROUP_ID, G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
+}
+
+static void
+gst_multiqueue_pad_init (GstMultiQueuePad * pad)
+{
+
+}
+
+
#define GST_MULTI_QUEUE_MUTEX_LOCK(q) G_STMT_START { \
g_mutex_lock (&q->qlock); \
} G_STMT_END
g_mutex_unlock (&q->qlock); \
} G_STMT_END
+#define SET_PERCENT(mq, perc) G_STMT_START { \
+ if (perc != mq->buffering_percent) { \
+ mq->buffering_percent = perc; \
+ mq->buffering_percent_changed = TRUE; \
+ GST_DEBUG_OBJECT (mq, "buffering %d percent", perc); \
+ } \
+} G_STMT_END
+
+/* Convenience function */
+static inline GstClockTimeDiff
+my_segment_to_running_time (GstSegment * segment, GstClockTime val)
+{
+ GstClockTimeDiff res = GST_CLOCK_STIME_NONE;
+
+ if (GST_CLOCK_TIME_IS_VALID (val)) {
+ gboolean sign =
+ gst_segment_to_running_time_full (segment, GST_FORMAT_TIME, val, &val);
+ if (sign > 0)
+ res = val;
+ else if (sign < 0)
+ res = -val;
+ }
+ return res;
+}
+
static void gst_multi_queue_finalize (GObject * object);
static void gst_multi_queue_set_property (GObject * object,
guint prop_id, const GValue * value, GParamSpec * pspec);
/**
* GstMultiQueue::underrun:
- * @multiqueue: the multqueue instance
+ * @multiqueue: the multiqueue instance
*
* This signal is emitted from the streaming thread when there is
* no data in any of the queues inside the multiqueue instance (underrun).
g_param_spec_uint ("max-size-bytes", "Max. size (kB)",
"Max. amount of data in the queue (bytes, 0=disable)",
0, G_MAXUINT, DEFAULT_MAX_SIZE_BYTES,
- G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
+ G_PARAM_READWRITE | GST_PARAM_MUTABLE_PLAYING |
+ G_PARAM_STATIC_STRINGS));
g_object_class_install_property (gobject_class, PROP_MAX_SIZE_BUFFERS,
g_param_spec_uint ("max-size-buffers", "Max. size (buffers)",
"Max. number of buffers in the queue (0=disable)", 0, G_MAXUINT,
DEFAULT_MAX_SIZE_BUFFERS,
- G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
+ G_PARAM_READWRITE | GST_PARAM_MUTABLE_PLAYING |
+ G_PARAM_STATIC_STRINGS));
g_object_class_install_property (gobject_class, PROP_MAX_SIZE_TIME,
g_param_spec_uint64 ("max-size-time", "Max. size (ns)",
"Max. amount of data in the queue (in ns, 0=disable)", 0, G_MAXUINT64,
- DEFAULT_MAX_SIZE_TIME, G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
+ DEFAULT_MAX_SIZE_TIME, G_PARAM_READWRITE | GST_PARAM_MUTABLE_PLAYING |
+ G_PARAM_STATIC_STRINGS));
g_object_class_install_property (gobject_class, PROP_EXTRA_SIZE_BYTES,
g_param_spec_uint ("extra-size-bytes", "Extra Size (kB)",
* GstMultiQueue:use-buffering
*
* Enable the buffering option in multiqueue so that BUFFERING messages are
- * emited based on low-/high-percent thresholds.
- *
- * Since: 0.10.26
+ * emitted based on low-/high-percent thresholds.
*/
g_object_class_install_property (gobject_class, PROP_USE_BUFFERING,
g_param_spec_boolean ("use-buffering", "Use buffering",
"Emit GST_MESSAGE_BUFFERING based on low-/high-percent thresholds",
- DEFAULT_USE_BUFFERING, G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
+ DEFAULT_USE_BUFFERING, G_PARAM_READWRITE | GST_PARAM_MUTABLE_PLAYING |
+ G_PARAM_STATIC_STRINGS));
/**
* GstMultiQueue:low-percent
*
* Low threshold percent for buffering to start.
- *
- * Since: 0.10.26
*/
g_object_class_install_property (gobject_class, PROP_LOW_PERCENT,
g_param_spec_int ("low-percent", "Low percent",
- "Low threshold for buffering to start", 0, 100,
- DEFAULT_LOW_PERCENT, G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
+ "Low threshold for buffering to start. Only used if use-buffering is True "
+ "(Deprecated: use low-watermark instead)",
+ 0, 100, DEFAULT_LOW_WATERMARK * 100,
+ G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
/**
* GstMultiQueue:high-percent
*
* High threshold percent for buffering to finish.
- *
- * Since: 0.10.26
*/
g_object_class_install_property (gobject_class, PROP_HIGH_PERCENT,
g_param_spec_int ("high-percent", "High percent",
- "High threshold for buffering to finish", 0, 100,
- DEFAULT_HIGH_PERCENT, G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
+ "High threshold for buffering to finish. Only used if use-buffering is True "
+ "(Deprecated: use high-watermark instead)",
+ 0, 100, DEFAULT_HIGH_WATERMARK * 100,
+ G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
+ /**
+ * GstMultiQueue:low-watermark
+ *
+ * Low threshold watermark for buffering to start.
+ *
+ * Since: 1.10
+ */
+ g_object_class_install_property (gobject_class, PROP_LOW_WATERMARK,
+ g_param_spec_double ("low-watermark", "Low watermark",
+ "Low threshold for buffering to start. Only used if use-buffering is True",
+ 0.0, 1.0, DEFAULT_LOW_WATERMARK,
+ G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
+ /**
+ * GstMultiQueue:high-watermark
+ *
+ * High threshold watermark for buffering to finish.
+ *
+ * Since: 1.10
+ */
+ g_object_class_install_property (gobject_class, PROP_HIGH_WATERMARK,
+ g_param_spec_double ("high-watermark", "High watermark",
+ "High threshold for buffering to finish. Only used if use-buffering is True",
+ 0.0, 1.0, DEFAULT_HIGH_WATERMARK,
+ G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
/**
* GstMultiQueue:sync-by-running-time
* Otherwise multiqueue will synchronize the deactivated or not-linked
* streams by keeping the order in which buffers and events arrived compared
* to active and linked streams.
- *
- * Since: 0.10.36
*/
g_object_class_install_property (gobject_class, PROP_SYNC_BY_RUNNING_TIME,
g_param_spec_boolean ("sync-by-running-time", "Sync By Running Time",
DEFAULT_SYNC_BY_RUNNING_TIME,
G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
+ g_object_class_install_property (gobject_class, PROP_USE_INTERLEAVE,
+ g_param_spec_boolean ("use-interleave", "Use interleave",
+ "Adjust time limits based on input interleave",
+ DEFAULT_USE_INTERLEAVE, G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
+
+ g_object_class_install_property (gobject_class, PROP_UNLINKED_CACHE_TIME,
+ g_param_spec_uint64 ("unlinked-cache-time", "Unlinked cache time (ns)",
+ "Extra buffering in time for unlinked streams (if 'sync-by-running-time')",
+ 0, G_MAXUINT64, DEFAULT_UNLINKED_CACHE_TIME,
+ G_PARAM_READWRITE | GST_PARAM_MUTABLE_PLAYING |
+ G_PARAM_STATIC_STRINGS));
+
+
gobject_class->finalize = gst_multi_queue_finalize;
gst_element_class_set_static_metadata (gstelement_class,
"MultiQueue",
"Generic", "Multiple data queue", "Edward Hervey <edward@fluendo.com>");
- gst_element_class_add_pad_template (gstelement_class,
- gst_static_pad_template_get (&sinktemplate));
- gst_element_class_add_pad_template (gstelement_class,
- gst_static_pad_template_get (&srctemplate));
+ gst_element_class_add_static_pad_template (gstelement_class, &sinktemplate);
+ gst_element_class_add_static_pad_template (gstelement_class, &srctemplate);
gstelement_class->request_new_pad =
GST_DEBUG_FUNCPTR (gst_multi_queue_request_new_pad);
mqueue->extra_size.time = DEFAULT_EXTRA_SIZE_TIME;
mqueue->use_buffering = DEFAULT_USE_BUFFERING;
- mqueue->low_percent = DEFAULT_LOW_PERCENT;
- mqueue->high_percent = DEFAULT_HIGH_PERCENT;
+ mqueue->low_watermark = DEFAULT_LOW_WATERMARK * MAX_BUFFERING_LEVEL;
+ mqueue->high_watermark = DEFAULT_HIGH_WATERMARK * MAX_BUFFERING_LEVEL;
mqueue->sync_by_running_time = DEFAULT_SYNC_BY_RUNNING_TIME;
+ mqueue->use_interleave = DEFAULT_USE_INTERLEAVE;
+ mqueue->unlinked_cache_time = DEFAULT_UNLINKED_CACHE_TIME;
mqueue->counter = 1;
mqueue->highid = -1;
- mqueue->high_time = GST_CLOCK_TIME_NONE;
+ mqueue->high_time = GST_CLOCK_STIME_NONE;
g_mutex_init (&mqueue->qlock);
+ g_mutex_init (&mqueue->buffering_post_lock);
}
static void
/* free/unref instance data */
g_mutex_clear (&mqueue->qlock);
+ g_mutex_clear (&mqueue->buffering_post_lock);
G_OBJECT_CLASS (parent_class)->finalize (object);
}
while (tmp) { \
GstSingleQueue *q = (GstSingleQueue*)tmp->data; \
q->max_size.format = mq->max_size.format; \
+ update_buffering (mq, q); \
+ gst_data_queue_limits_changed (q->queue); \
tmp = g_list_next(tmp); \
}; \
} G_STMT_END
mq->max_size.bytes = g_value_get_uint (value);
SET_CHILD_PROPERTY (mq, bytes);
GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ gst_multi_queue_post_buffering (mq);
break;
case PROP_MAX_SIZE_BUFFERS:
+ {
+ GList *tmp;
+ gint new_size = g_value_get_uint (value);
+
GST_MULTI_QUEUE_MUTEX_LOCK (mq);
- mq->max_size.visible = g_value_get_uint (value);
- SET_CHILD_PROPERTY (mq, visible);
+
+ mq->max_size.visible = new_size;
+
+ tmp = mq->queues;
+ while (tmp) {
+ GstDataQueueSize size;
+ GstSingleQueue *q = (GstSingleQueue *) tmp->data;
+ gst_data_queue_get_level (q->queue, &size);
+
+ GST_DEBUG_OBJECT (mq, "Queue %d: Requested buffers size: %d,"
+ " current: %d, current max %d", q->id, new_size, size.visible,
+ q->max_size.visible);
+
+ /* do not reduce max size below current level if the single queue
+ * has grown because of empty queue */
+ if (new_size == 0) {
+ q->max_size.visible = new_size;
+ } else if (q->max_size.visible == 0) {
+ q->max_size.visible = MAX (new_size, size.visible);
+ } else if (new_size > size.visible) {
+ q->max_size.visible = new_size;
+ }
+ update_buffering (mq, q);
+ gst_data_queue_limits_changed (q->queue);
+ tmp = g_list_next (tmp);
+ }
+
GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ gst_multi_queue_post_buffering (mq);
+
break;
+ }
case PROP_MAX_SIZE_TIME:
GST_MULTI_QUEUE_MUTEX_LOCK (mq);
mq->max_size.time = g_value_get_uint64 (value);
SET_CHILD_PROPERTY (mq, time);
GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ gst_multi_queue_post_buffering (mq);
break;
case PROP_EXTRA_SIZE_BYTES:
mq->extra_size.bytes = g_value_get_uint (value);
break;
case PROP_USE_BUFFERING:
mq->use_buffering = g_value_get_boolean (value);
+ recheck_buffering_status (mq);
break;
case PROP_LOW_PERCENT:
- mq->low_percent = g_value_get_int (value);
+ mq->low_watermark = g_value_get_int (value) * BUF_LEVEL_PERCENT_FACTOR;
+ /* Recheck buffering status - the new low_watermark value might
+ * be above the current fill level. If the old low_watermark one
+ * was below the current level, this means that mq->buffering is
+ * disabled and needs to be re-enabled. */
+ recheck_buffering_status (mq);
break;
case PROP_HIGH_PERCENT:
- mq->high_percent = g_value_get_int (value);
+ mq->high_watermark = g_value_get_int (value) * BUF_LEVEL_PERCENT_FACTOR;
+ recheck_buffering_status (mq);
+ break;
+ case PROP_LOW_WATERMARK:
+ mq->low_watermark = g_value_get_double (value) * MAX_BUFFERING_LEVEL;
+ recheck_buffering_status (mq);
+ break;
+ case PROP_HIGH_WATERMARK:
+ mq->high_watermark = g_value_get_double (value) * MAX_BUFFERING_LEVEL;
+ recheck_buffering_status (mq);
break;
case PROP_SYNC_BY_RUNNING_TIME:
mq->sync_by_running_time = g_value_get_boolean (value);
break;
+ case PROP_USE_INTERLEAVE:
+ mq->use_interleave = g_value_get_boolean (value);
+ break;
+ case PROP_UNLINKED_CACHE_TIME:
+ GST_MULTI_QUEUE_MUTEX_LOCK (mq);
+ mq->unlinked_cache_time = g_value_get_uint64 (value);
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ gst_multi_queue_post_buffering (mq);
+ break;
default:
G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
break;
g_value_set_boolean (value, mq->use_buffering);
break;
case PROP_LOW_PERCENT:
- g_value_set_int (value, mq->low_percent);
+ g_value_set_int (value, mq->low_watermark / BUF_LEVEL_PERCENT_FACTOR);
break;
case PROP_HIGH_PERCENT:
- g_value_set_int (value, mq->high_percent);
+ g_value_set_int (value, mq->high_watermark / BUF_LEVEL_PERCENT_FACTOR);
+ break;
+ case PROP_LOW_WATERMARK:
+ g_value_set_double (value, mq->low_watermark /
+ (gdouble) MAX_BUFFERING_LEVEL);
+ break;
+ case PROP_HIGH_WATERMARK:
+ g_value_set_double (value, mq->high_watermark /
+ (gdouble) MAX_BUFFERING_LEVEL);
break;
case PROP_SYNC_BY_RUNNING_TIME:
g_value_set_boolean (value, mq->sync_by_running_time);
break;
+ case PROP_USE_INTERLEAVE:
+ g_value_set_boolean (value, mq->use_interleave);
+ break;
+ case PROP_UNLINKED_CACHE_TIME:
+ g_value_set_uint64 (value, mq->unlinked_cache_time);
+ break;
default:
G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
break;
{
GstMultiQueue *mqueue = GST_MULTI_QUEUE (element);
GstSingleQueue *squeue;
+ GstPad *new_pad;
guint temp_id = -1;
if (name) {
/* Create a new single queue, add the sink and source pad and return the sink pad */
squeue = gst_single_queue_new (mqueue, temp_id);
- GST_DEBUG_OBJECT (mqueue, "Returning pad %s:%s",
- GST_DEBUG_PAD_NAME (squeue->sinkpad));
+ new_pad = squeue ? squeue->sinkpad : NULL;
+
+ GST_DEBUG_OBJECT (mqueue, "Returning pad %" GST_PTR_FORMAT, new_pad);
- return squeue ? squeue->sinkpad : NULL;
+ return new_pad;
}
static void
sq = (GstSingleQueue *) tmp->data;
sq->flushing = FALSE;
}
+
+ /* the visible limit might not have been set on single queues that have grown because of other queueus were empty */
+ SET_CHILD_PROPERTY (mqueue, visible);
+
GST_MULTI_QUEUE_MUTEX_UNLOCK (mqueue);
+ gst_multi_queue_post_buffering (mqueue);
+
break;
}
case GST_STATE_CHANGE_PAUSED_TO_READY:{
sq = (GstSingleQueue *) tmp->data;
sq->flushing = TRUE;
g_cond_signal (&sq->turn);
+
+ sq->last_query = FALSE;
+ g_cond_signal (&sq->query_handled);
}
GST_MULTI_QUEUE_MUTEX_UNLOCK (mqueue);
break;
}
return result;
-
-
-
}
static gboolean
-gst_single_queue_flush (GstMultiQueue * mq, GstSingleQueue * sq, gboolean flush)
+gst_single_queue_flush (GstMultiQueue * mq, GstSingleQueue * sq, gboolean flush,
+ gboolean full)
{
gboolean result;
gst_data_queue_set_flushing (sq->queue, TRUE);
sq->flushing = TRUE;
- GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
/* wake up non-linked task */
GST_LOG_OBJECT (mq, "SingleQueue %d : waking up eventually waiting task",
result = gst_pad_pause_task (sq->srcpad);
sq->sink_tainted = sq->src_tainted = TRUE;
} else {
+ gst_single_queue_flush_queue (sq, full);
+
GST_MULTI_QUEUE_MUTEX_LOCK (mq);
- gst_data_queue_flush (sq->queue);
gst_segment_init (&sq->sink_segment, GST_FORMAT_TIME);
gst_segment_init (&sq->src_segment, GST_FORMAT_TIME);
+ sq->has_src_segment = FALSE;
/* All pads start off not-linked for a smooth kick-off */
sq->srcresult = GST_FLOW_OK;
sq->pushed = FALSE;
sq->nextid = 0;
sq->oldid = 0;
sq->last_oldid = G_MAXUINT32;
- sq->next_time = GST_CLOCK_TIME_NONE;
- sq->last_time = GST_CLOCK_TIME_NONE;
+ sq->next_time = GST_CLOCK_STIME_NONE;
+ sq->last_time = GST_CLOCK_STIME_NONE;
+ sq->cached_sinktime = GST_CLOCK_STIME_NONE;
+ sq->group_high_time = GST_CLOCK_STIME_NONE;
gst_data_queue_set_flushing (sq->queue, FALSE);
+ /* We will become active again on the next buffer/gap */
+ sq->active = FALSE;
+
/* Reset high time to be recomputed next */
- mq->high_time = GST_CLOCK_TIME_NONE;
+ mq->high_time = GST_CLOCK_STIME_NONE;
sq->flushing = FALSE;
GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
GST_LOG_OBJECT (mq, "SingleQueue %d : starting task", sq->id);
result =
gst_pad_start_task (sq->srcpad, (GstTaskFunction) gst_multi_queue_loop,
- sq->srcpad);
+ sq->srcpad, NULL);
}
return result;
}
-static void
-update_buffering (GstMultiQueue * mq, GstSingleQueue * sq)
+/* WITH LOCK TAKEN */
+static gint
+get_buffering_level (GstSingleQueue * sq)
{
GstDataQueueSize size;
- gint percent, tmp;
- gboolean post = FALSE;
-
- /* nothing to dowhen we are not in buffering mode */
- if (!mq->use_buffering)
- return;
+ gint buffering_level, tmp;
gst_data_queue_get_level (sq->queue, &size);
- GST_DEBUG_OBJECT (mq,
+ GST_DEBUG_OBJECT (sq->mqueue,
"queue %d: visible %u/%u, bytes %u/%u, time %" G_GUINT64_FORMAT "/%"
G_GUINT64_FORMAT, sq->id, size.visible, sq->max_size.visible,
size.bytes, sq->max_size.bytes, sq->cur_time, sq->max_size.time);
- /* get bytes and time percentages and take the max */
- if (sq->is_eos) {
- percent = 100;
+ /* get bytes and time buffer levels and take the max */
+ if (sq->is_eos || sq->srcresult == GST_FLOW_NOT_LINKED || sq->is_sparse) {
+ buffering_level = MAX_BUFFERING_LEVEL;
} else {
- percent = 0;
+ buffering_level = 0;
if (sq->max_size.time > 0) {
- tmp = (sq->cur_time * 100) / sq->max_size.time;
- percent = MAX (percent, tmp);
+ tmp =
+ gst_util_uint64_scale (sq->cur_time,
+ MAX_BUFFERING_LEVEL, sq->max_size.time);
+ buffering_level = MAX (buffering_level, tmp);
}
if (sq->max_size.bytes > 0) {
- tmp = (size.bytes * 100) / sq->max_size.bytes;
- percent = MAX (percent, tmp);
+ tmp =
+ gst_util_uint64_scale_int (size.bytes,
+ MAX_BUFFERING_LEVEL, sq->max_size.bytes);
+ buffering_level = MAX (buffering_level, tmp);
}
}
+ return buffering_level;
+}
+
+/* WITH LOCK TAKEN */
+static void
+update_buffering (GstMultiQueue * mq, GstSingleQueue * sq)
+{
+ gint buffering_level, percent;
+
+ /* nothing to dowhen we are not in buffering mode */
+ if (!mq->use_buffering)
+ return;
+
+ buffering_level = get_buffering_level (sq);
+
+ /* scale so that if buffering_level equals the high watermark,
+ * the percentage is 100% */
+ percent = gst_util_uint64_scale (buffering_level, 100, mq->high_watermark);
+ /* clip */
+ if (percent > 100)
+ percent = 100;
+
if (mq->buffering) {
- post = TRUE;
- if (percent >= mq->high_percent) {
+ if (buffering_level >= mq->high_watermark) {
mq->buffering = FALSE;
}
/* make sure it increases */
- percent = MAX (mq->percent, percent);
+ percent = MAX (mq->buffering_percent, percent);
- if (percent == mq->percent)
- /* don't post if nothing changed */
- post = FALSE;
- else
- /* else keep last value we posted */
- mq->percent = percent;
+ SET_PERCENT (mq, percent);
} else {
- if (percent < mq->low_percent) {
+ GList *iter;
+ gboolean is_buffering = TRUE;
+
+ for (iter = mq->queues; iter; iter = g_list_next (iter)) {
+ GstSingleQueue *oq = (GstSingleQueue *) iter->data;
+
+ if (get_buffering_level (oq) >= mq->high_watermark) {
+ is_buffering = FALSE;
+
+ break;
+ }
+ }
+
+ if (is_buffering && buffering_level < mq->low_watermark) {
mq->buffering = TRUE;
- mq->percent = percent;
- post = TRUE;
+ SET_PERCENT (mq, percent);
+ }
+ }
+}
+
+static void
+gst_multi_queue_post_buffering (GstMultiQueue * mq)
+{
+ GstMessage *msg = NULL;
+
+ g_mutex_lock (&mq->buffering_post_lock);
+ GST_MULTI_QUEUE_MUTEX_LOCK (mq);
+ if (mq->buffering_percent_changed) {
+ gint percent = mq->buffering_percent;
+
+ mq->buffering_percent_changed = FALSE;
+
+ GST_DEBUG_OBJECT (mq, "Going to post buffering: %d%%", percent);
+ msg = gst_message_new_buffering (GST_OBJECT_CAST (mq), percent);
+ }
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+
+ if (msg != NULL)
+ gst_element_post_message (GST_ELEMENT_CAST (mq), msg);
+
+ g_mutex_unlock (&mq->buffering_post_lock);
+}
+
+static void
+recheck_buffering_status (GstMultiQueue * mq)
+{
+ if (!mq->use_buffering && mq->buffering) {
+ GST_MULTI_QUEUE_MUTEX_LOCK (mq);
+ mq->buffering = FALSE;
+ GST_DEBUG_OBJECT (mq,
+ "Buffering property disabled, but queue was still buffering; "
+ "setting buffering percentage to 100%%");
+ SET_PERCENT (mq, 100);
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ }
+
+ if (mq->use_buffering) {
+ GList *tmp;
+ gint old_perc;
+
+ GST_MULTI_QUEUE_MUTEX_LOCK (mq);
+
+ /* force buffering percentage to be recalculated */
+ old_perc = mq->buffering_percent;
+ mq->buffering_percent = 0;
+
+ tmp = mq->queues;
+ while (tmp) {
+ GstSingleQueue *q = (GstSingleQueue *) tmp->data;
+ update_buffering (mq, q);
+ gst_data_queue_limits_changed (q->queue);
+ tmp = g_list_next (tmp);
}
+
+ GST_DEBUG_OBJECT (mq,
+ "Recalculated buffering percentage: old: %d%% new: %d%%",
+ old_perc, mq->buffering_percent);
+
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
}
- if (post) {
- GstMessage *message;
- /* scale to high percent so that it becomes the 100% mark */
- percent = percent * 100 / mq->high_percent;
- /* clip */
- if (percent > 100)
- percent = 100;
+ gst_multi_queue_post_buffering (mq);
+}
- GST_DEBUG_OBJECT (mq, "buffering %d percent", percent);
- message = gst_message_new_buffering (GST_OBJECT_CAST (mq), percent);
+static void
+calculate_interleave (GstMultiQueue * mq)
+{
+ GstClockTimeDiff low, high;
+ GstClockTime interleave;
+ GList *tmp;
- gst_element_post_message (GST_ELEMENT_CAST (mq), message);
- } else {
- GST_DEBUG_OBJECT (mq, "filled %d percent", percent);
+ low = high = GST_CLOCK_STIME_NONE;
+ interleave = mq->interleave;
+ /* Go over all single queues and calculate lowest/highest value */
+ for (tmp = mq->queues; tmp; tmp = tmp->next) {
+ GstSingleQueue *sq = (GstSingleQueue *) tmp->data;
+ /* Ignore sparse streams for interleave calculation */
+ if (sq->is_sparse)
+ continue;
+ /* If a stream is not active yet (hasn't received any buffers), set
+ * a maximum interleave to allow it to receive more data */
+ if (!sq->active) {
+ GST_LOG_OBJECT (mq,
+ "queue %d is not active yet, forcing interleave to 5s", sq->id);
+ mq->interleave = 5 * GST_SECOND;
+ /* Update max-size time */
+ mq->max_size.time = mq->interleave;
+ SET_CHILD_PROPERTY (mq, time);
+ goto beach;
+ }
+ if (GST_CLOCK_STIME_IS_VALID (sq->cached_sinktime)) {
+ if (low == GST_CLOCK_STIME_NONE || sq->cached_sinktime < low)
+ low = sq->cached_sinktime;
+ if (high == GST_CLOCK_STIME_NONE || sq->cached_sinktime > high)
+ high = sq->cached_sinktime;
+ }
+ GST_LOG_OBJECT (mq,
+ "queue %d , sinktime:%" GST_STIME_FORMAT " low:%" GST_STIME_FORMAT
+ " high:%" GST_STIME_FORMAT, sq->id,
+ GST_STIME_ARGS (sq->cached_sinktime), GST_STIME_ARGS (low),
+ GST_STIME_ARGS (high));
}
+
+ if (GST_CLOCK_STIME_IS_VALID (low) && GST_CLOCK_STIME_IS_VALID (high)) {
+ interleave = high - low;
+ /* Padding of interleave and minimum value */
+ /* FIXME : Make the minimum time interleave a property */
+ interleave = (150 * interleave / 100) + 250 * GST_MSECOND;
+
+ /* Update the stored interleave if:
+ * * No data has arrived yet (high == low)
+ * * Or it went higher
+ * * Or it went lower and we've gone past the previous interleave needed */
+ if (high == low || interleave > mq->interleave ||
+ ((mq->last_interleave_update + (2 * MIN (GST_SECOND,
+ mq->interleave)) < low)
+ && interleave < (mq->interleave * 3 / 4))) {
+ /* Update the interleave */
+ mq->interleave = interleave;
+ mq->last_interleave_update = high;
+ /* Update max-size time */
+ mq->max_size.time = mq->interleave;
+ SET_CHILD_PROPERTY (mq, time);
+ }
+ }
+
+beach:
+ GST_DEBUG_OBJECT (mq,
+ "low:%" GST_STIME_FORMAT " high:%" GST_STIME_FORMAT " interleave:%"
+ GST_TIME_FORMAT " mq->interleave:%" GST_TIME_FORMAT
+ " last_interleave_update:%" GST_STIME_FORMAT, GST_STIME_ARGS (low),
+ GST_STIME_ARGS (high), GST_TIME_ARGS (interleave),
+ GST_TIME_ARGS (mq->interleave),
+ GST_STIME_ARGS (mq->last_interleave_update));
}
+
/* calculate the diff between running time on the sink and src of the queue.
* This is the total amount of time in the queue.
* WITH LOCK TAKEN */
static void
update_time_level (GstMultiQueue * mq, GstSingleQueue * sq)
{
- gint64 sink_time, src_time;
+ GstClockTimeDiff sink_time, src_time;
if (sq->sink_tainted) {
- sink_time = sq->sinktime =
- gst_segment_to_running_time (&sq->sink_segment, GST_FORMAT_TIME,
+ sink_time = sq->sinktime = my_segment_to_running_time (&sq->sink_segment,
sq->sink_segment.position);
- if (G_UNLIKELY (sink_time != GST_CLOCK_TIME_NONE))
+ GST_DEBUG_OBJECT (mq,
+ "queue %d sink_segment.position:%" GST_TIME_FORMAT ", sink_time:%"
+ GST_STIME_FORMAT, sq->id, GST_TIME_ARGS (sq->sink_segment.position),
+ GST_STIME_ARGS (sink_time));
+
+ if (G_UNLIKELY (sq->last_time == GST_CLOCK_STIME_NONE)) {
+ /* If the single queue still doesn't have a last_time set, this means
+ * that nothing has been pushed out yet.
+ * In order for the high_time computation to be as efficient as possible,
+ * we set the last_time */
+ sq->last_time = sink_time;
+ }
+ if (G_UNLIKELY (sink_time != GST_CLOCK_STIME_NONE)) {
/* if we have a time, we become untainted and use the time */
sq->sink_tainted = FALSE;
+ if (mq->use_interleave) {
+ sq->cached_sinktime = sink_time;
+ calculate_interleave (mq);
+ }
+ }
} else
sink_time = sq->sinktime;
if (sq->src_tainted) {
- src_time = sq->srctime =
- gst_segment_to_running_time (&sq->src_segment, GST_FORMAT_TIME,
- sq->src_segment.position);
+ GstSegment *segment;
+ gint64 position;
+
+ if (sq->has_src_segment) {
+ segment = &sq->src_segment;
+ position = sq->src_segment.position;
+ } else {
+ /*
+ * If the src pad had no segment yet, use the sink segment
+ * to avoid signalling overrun if the received sink segment has a
+ * a position > max-size-time while the src pad time would be the default=0
+ *
+ * This can happen when switching pads on chained/adaptive streams and the
+ * new chain has a segment with a much larger position
+ */
+ segment = &sq->sink_segment;
+ position = sq->sink_segment.position;
+ }
+
+ src_time = sq->srctime = my_segment_to_running_time (segment, position);
/* if we have a time, we become untainted and use the time */
- if (G_UNLIKELY (src_time != GST_CLOCK_TIME_NONE))
+ if (G_UNLIKELY (src_time != GST_CLOCK_STIME_NONE)) {
sq->src_tainted = FALSE;
+ }
} else
src_time = sq->srctime;
GST_DEBUG_OBJECT (mq,
- "queue %d, sink %" GST_TIME_FORMAT ", src %" GST_TIME_FORMAT, sq->id,
- GST_TIME_ARGS (sink_time), GST_TIME_ARGS (src_time));
+ "queue %d, sink %" GST_STIME_FORMAT ", src %" GST_STIME_FORMAT, sq->id,
+ GST_STIME_ARGS (sink_time), GST_STIME_ARGS (src_time));
/* This allows for streams with out of order timestamping - sometimes the
* emerging timestamp is later than the arriving one(s) */
- if (G_LIKELY (sink_time != -1 && src_time != -1 && sink_time > src_time))
+ if (G_LIKELY (GST_CLOCK_STIME_IS_VALID (sink_time) &&
+ GST_CLOCK_STIME_IS_VALID (src_time) && sink_time > src_time))
sq->cur_time = sink_time - src_time;
else
sq->cur_time = 0;
}
GST_MULTI_QUEUE_MUTEX_LOCK (mq);
+ /* Make sure we have a valid initial segment position (and not garbage
+ * from upstream) */
+ if (segment->rate > 0.0)
+ segment->position = segment->start;
+ else
+ segment->position = segment->stop;
if (segment == &sq->sink_segment)
sq->sink_tainted = TRUE;
- else
+ else {
+ sq->has_src_segment = TRUE;
sq->src_tainted = TRUE;
+ }
GST_DEBUG_OBJECT (mq,
"queue %d, configured SEGMENT %" GST_SEGMENT_FORMAT, sq->id, segment);
update_time_level (mq, sq);
GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ gst_multi_queue_post_buffering (mq);
}
/* take a buffer and update segment, updating the time level of the queue. */
if (duration != GST_CLOCK_TIME_NONE)
timestamp += duration;
- GST_DEBUG_OBJECT (mq, "queue %d, position updated to %" GST_TIME_FORMAT,
- sq->id, GST_TIME_ARGS (timestamp));
+ GST_DEBUG_OBJECT (mq, "queue %d, %s position updated to %" GST_TIME_FORMAT,
+ sq->id, segment == &sq->sink_segment ? "sink" : "src",
+ GST_TIME_ARGS (timestamp));
segment->position = timestamp;
/* calc diff with other end */
update_time_level (mq, sq);
GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ gst_multi_queue_post_buffering (mq);
}
-static GstClockTime
+static void
+apply_gap (GstMultiQueue * mq, GstSingleQueue * sq, GstEvent * event,
+ GstSegment * segment)
+{
+ GstClockTime timestamp;
+ GstClockTime duration;
+
+ GST_MULTI_QUEUE_MUTEX_LOCK (mq);
+
+ gst_event_parse_gap (event, ×tamp, &duration);
+
+ if (GST_CLOCK_TIME_IS_VALID (timestamp)) {
+
+ if (GST_CLOCK_TIME_IS_VALID (duration)) {
+ timestamp += duration;
+ }
+
+ segment->position = timestamp;
+
+ if (segment == &sq->sink_segment)
+ sq->sink_tainted = TRUE;
+ else
+ sq->src_tainted = TRUE;
+
+ /* calc diff with other end */
+ update_time_level (mq, sq);
+ }
+
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ gst_multi_queue_post_buffering (mq);
+}
+
+static GstClockTimeDiff
get_running_time (GstSegment * segment, GstMiniObject * object, gboolean end)
{
- GstClockTime time = GST_CLOCK_TIME_NONE;
+ GstClockTimeDiff time = GST_CLOCK_STIME_NONE;
if (GST_IS_BUFFER (object)) {
GstBuffer *buf = GST_BUFFER_CAST (object);
+ GstClockTime btime = GST_BUFFER_DTS_OR_PTS (buf);
- if (GST_BUFFER_TIMESTAMP_IS_VALID (buf)) {
- time = GST_BUFFER_TIMESTAMP (buf);
+ if (GST_CLOCK_TIME_IS_VALID (btime)) {
if (end && GST_BUFFER_DURATION_IS_VALID (buf))
- time += GST_BUFFER_DURATION (buf);
- if (time > segment->stop)
- time = segment->stop;
- time = gst_segment_to_running_time (segment, GST_FORMAT_TIME, time);
+ btime += GST_BUFFER_DURATION (buf);
+ if (btime > segment->stop)
+ btime = segment->stop;
+ time = my_segment_to_running_time (segment, btime);
}
} else if (GST_IS_BUFFER_LIST (object)) {
GstBufferList *list = GST_BUFFER_LIST_CAST (object);
n = gst_buffer_list_length (list);
for (i = 0; i < n; i++) {
+ GstClockTime btime;
buf = gst_buffer_list_get (list, i);
- if (GST_BUFFER_TIMESTAMP_IS_VALID (buf)) {
- time = GST_BUFFER_TIMESTAMP (buf);
+ btime = GST_BUFFER_DTS_OR_PTS (buf);
+ if (GST_CLOCK_TIME_IS_VALID (btime)) {
if (end && GST_BUFFER_DURATION_IS_VALID (buf))
- time += GST_BUFFER_DURATION (buf);
- if (time > segment->stop)
- time = segment->stop;
- time = gst_segment_to_running_time (segment, GST_FORMAT_TIME, time);
+ btime += GST_BUFFER_DURATION (buf);
+ if (btime > segment->stop)
+ btime = segment->stop;
+ time = my_segment_to_running_time (segment, btime);
if (!end)
goto done;
} else if (!end) {
gst_event_parse_segment (event, &new_segment);
if (new_segment->format == GST_FORMAT_TIME) {
time =
- gst_segment_to_running_time (new_segment, GST_FORMAT_TIME,
+ my_segment_to_running_time ((GstSegment *) new_segment,
new_segment->start);
}
}
static GstFlowReturn
gst_single_queue_push_one (GstMultiQueue * mq, GstSingleQueue * sq,
- GstMiniObject * object)
+ GstMiniObject * object, gboolean * allow_drop)
{
- GstFlowReturn result = GST_FLOW_OK;
+ GstFlowReturn result = sq->srcresult;
if (GST_IS_BUFFER (object)) {
GstBuffer *buffer;
GstClockTime timestamp, duration;
buffer = GST_BUFFER_CAST (object);
- timestamp = GST_BUFFER_TIMESTAMP (buffer);
+ timestamp = GST_BUFFER_DTS_OR_PTS (buffer);
duration = GST_BUFFER_DURATION (buffer);
apply_buffer (mq, sq, timestamp, duration, &sq->src_segment);
/* Applying the buffer may have made the queue non-full again, unblock it if needed */
gst_data_queue_limits_changed (sq->queue);
- GST_DEBUG_OBJECT (mq,
- "SingleQueue %d : Pushing buffer %p with ts %" GST_TIME_FORMAT,
- sq->id, buffer, GST_TIME_ARGS (timestamp));
-
- result = gst_pad_push (sq->srcpad, buffer);
+ if (G_UNLIKELY (*allow_drop)) {
+ GST_DEBUG_OBJECT (mq,
+ "SingleQueue %d : Dropping EOS buffer %p with ts %" GST_TIME_FORMAT,
+ sq->id, buffer, GST_TIME_ARGS (timestamp));
+ gst_buffer_unref (buffer);
+ } else {
+ GST_DEBUG_OBJECT (mq,
+ "SingleQueue %d : Pushing buffer %p with ts %" GST_TIME_FORMAT,
+ sq->id, buffer, GST_TIME_ARGS (timestamp));
+ result = gst_pad_push (sq->srcpad, buffer);
+ }
} else if (GST_IS_EVENT (object)) {
GstEvent *event;
switch (GST_EVENT_TYPE (event)) {
case GST_EVENT_EOS:
result = GST_FLOW_EOS;
+ if (G_UNLIKELY (*allow_drop))
+ *allow_drop = FALSE;
break;
case GST_EVENT_SEGMENT:
apply_segment (mq, sq, event, &sq->src_segment);
/* Applying the segment may have made the queue non-full again, unblock it if needed */
gst_data_queue_limits_changed (sq->queue);
+ if (G_UNLIKELY (*allow_drop)) {
+ result = GST_FLOW_OK;
+ *allow_drop = FALSE;
+ }
+ break;
+ case GST_EVENT_GAP:
+ apply_gap (mq, sq, event, &sq->src_segment);
+ /* Applying the gap may have made the queue non-full again, unblock it if needed */
+ gst_data_queue_limits_changed (sq->queue);
break;
default:
break;
}
- GST_DEBUG_OBJECT (mq,
- "SingleQueue %d : Pushing event %p of type %s",
- sq->id, event, GST_EVENT_TYPE_NAME (event));
+ if (G_UNLIKELY (*allow_drop)) {
+ GST_DEBUG_OBJECT (mq,
+ "SingleQueue %d : Dropping EOS event %p of type %s",
+ sq->id, event, GST_EVENT_TYPE_NAME (event));
+ gst_event_unref (event);
+ } else {
+ GST_DEBUG_OBJECT (mq,
+ "SingleQueue %d : Pushing event %p of type %s",
+ sq->id, event, GST_EVENT_TYPE_NAME (event));
- gst_pad_push_event (sq->srcpad, event);
+ gst_pad_push_event (sq->srcpad, event);
+ }
} else if (GST_IS_QUERY (object)) {
GstQuery *query;
gboolean res;
query = GST_QUERY_CAST (object);
- res = gst_pad_peer_query (sq->srcpad, query);
+ if (G_UNLIKELY (*allow_drop)) {
+ GST_DEBUG_OBJECT (mq,
+ "SingleQueue %d : Dropping EOS query %p", sq->id, query);
+ gst_query_unref (query);
+ res = FALSE;
+ } else {
+ res = gst_pad_peer_query (sq->srcpad, query);
+ }
GST_MULTI_QUEUE_MUTEX_LOCK (mq);
sq->last_query = res;
+ sq->last_handled_query = query;
g_cond_signal (&sq->query_handled);
GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
} else {
- g_warning ("Unexpected object in singlequeue %d (refcounting problem?)",
+ g_warning ("Unexpected object in singlequeue %u (refcounting problem?)",
sq->id);
}
return result;
static void
gst_multi_queue_item_destroy (GstMultiQueueItem * item)
{
- if (item->object)
+ if (!item->is_query && item->object)
gst_mini_object_unref (item->object);
g_slice_free (GstMultiQueueItem, item);
}
item->object = object;
item->destroy = (GDestroyNotify) gst_multi_queue_item_destroy;
item->posid = curid;
+ item->is_query = GST_IS_QUERY (object);
item->size = gst_buffer_get_size (GST_BUFFER_CAST (object));
item->duration = GST_BUFFER_DURATION (object);
item->object = object;
item->destroy = (GDestroyNotify) gst_multi_queue_item_destroy;
item->posid = curid;
+ item->is_query = GST_IS_QUERY (object);
item->size = 0;
item->duration = 0;
GstMiniObject *object = NULL;
guint32 newid;
GstFlowReturn result;
- GstClockTime next_time;
+ GstClockTimeDiff next_time;
+ gboolean is_buffer;
+ gboolean do_update_buffering = FALSE;
+ gboolean dropping = FALSE;
sq = (GstSingleQueue *) gst_pad_get_element_private (pad);
mq = sq->mqueue;
+next:
GST_DEBUG_OBJECT (mq, "SingleQueue %d : trying to pop an object", sq->id);
if (sq->flushing)
object = gst_multi_queue_item_steal_object (item);
gst_multi_queue_item_destroy (item);
- /* Get running time of the item. Events will have GST_CLOCK_TIME_NONE */
- next_time = get_running_time (&sq->src_segment, object, TRUE);
+ is_buffer = GST_IS_BUFFER (object);
+
+ /* Get running time of the item. Events will have GST_CLOCK_STIME_NONE */
+ next_time = get_running_time (&sq->src_segment, object, FALSE);
GST_LOG_OBJECT (mq, "SingleQueue %d : newid:%d , oldid:%d",
sq->id, newid, sq->last_oldid);
/* If we're not-linked, we do some extra work because we might need to
* wait before pushing. If we're linked but there's a gap in the IDs,
- * or it's the first loop, or we just passed the previous highid,
- * we might need to wake some sleeping pad up, so there's extra work
+ * or it's the first loop, or we just passed the previous highid,
+ * we might need to wake some sleeping pad up, so there's extra work
* there too */
GST_MULTI_QUEUE_MUTEX_LOCK (mq);
if (sq->srcresult == GST_FLOW_NOT_LINKED
/* Update the nextid so other threads know when to wake us up */
sq->nextid = newid;
+ /* Take into account the extra cache time since we're unlinked */
+ if (GST_CLOCK_STIME_IS_VALID (next_time))
+ next_time += mq->unlinked_cache_time;
sq->next_time = next_time;
/* Update the oldid (the last ID we output) for highid tracking */
sq->oldid = sq->last_oldid;
if (sq->srcresult == GST_FLOW_NOT_LINKED) {
+ gboolean should_wait;
/* Go to sleep until it's time to push this buffer */
/* Recompute the highid */
compute_high_id (mq);
/* Recompute the high time */
- compute_high_time (mq);
+ compute_high_time (mq, sq->groupid);
+
+ GST_DEBUG_OBJECT (mq,
+ "groupid %d high_time %" GST_STIME_FORMAT " next_time %"
+ GST_STIME_FORMAT, sq->groupid, GST_STIME_ARGS (sq->group_high_time),
+ GST_STIME_ARGS (next_time));
+
+ if (mq->sync_by_running_time) {
+ if (sq->group_high_time == GST_CLOCK_STIME_NONE) {
+ should_wait = GST_CLOCK_STIME_IS_VALID (next_time) &&
+ (mq->high_time == GST_CLOCK_STIME_NONE
+ || next_time > mq->high_time);
+ } else {
+ should_wait = GST_CLOCK_STIME_IS_VALID (next_time) &&
+ next_time > sq->group_high_time;
+ }
+ } else
+ should_wait = newid > mq->highid;
- while (((mq->sync_by_running_time && next_time != GST_CLOCK_TIME_NONE &&
- (mq->high_time == GST_CLOCK_TIME_NONE
- || next_time >= mq->high_time))
- || (!mq->sync_by_running_time && newid > mq->highid))
- && sq->srcresult == GST_FLOW_NOT_LINKED) {
+ while (should_wait && sq->srcresult == GST_FLOW_NOT_LINKED) {
GST_DEBUG_OBJECT (mq,
"queue %d sleeping for not-linked wakeup with "
- "newid %u, highid %u, next_time %" GST_TIME_FORMAT
- ", high_time %" GST_TIME_FORMAT, sq->id, newid, mq->highid,
- GST_TIME_ARGS (next_time), GST_TIME_ARGS (mq->high_time));
+ "newid %u, highid %u, next_time %" GST_STIME_FORMAT
+ ", high_time %" GST_STIME_FORMAT, sq->id, newid, mq->highid,
+ GST_STIME_ARGS (next_time), GST_STIME_ARGS (sq->group_high_time));
/* Wake up all non-linked pads before we sleep */
wake_up_next_non_linked (mq);
goto out_flushing;
}
- /* Recompute the high time */
- compute_high_time (mq);
+ /* Recompute the high time and ID */
+ compute_high_time (mq, sq->groupid);
+ compute_high_id (mq);
GST_DEBUG_OBJECT (mq, "queue %d woken from sleeping for not-linked "
- "wakeup with newid %u, highid %u, next_time %" GST_TIME_FORMAT
- ", high_time %" GST_TIME_FORMAT, sq->id, newid, mq->highid,
- GST_TIME_ARGS (next_time), GST_TIME_ARGS (mq->high_time));
+ "wakeup with newid %u, highid %u, next_time %" GST_STIME_FORMAT
+ ", high_time %" GST_STIME_FORMAT " mq high_time %" GST_STIME_FORMAT,
+ sq->id, newid, mq->highid,
+ GST_STIME_ARGS (next_time), GST_STIME_ARGS (sq->group_high_time),
+ GST_STIME_ARGS (mq->high_time));
+
+ if (mq->sync_by_running_time) {
+ if (sq->group_high_time == GST_CLOCK_STIME_NONE) {
+ should_wait = GST_CLOCK_STIME_IS_VALID (next_time) &&
+ (mq->high_time == GST_CLOCK_STIME_NONE
+ || next_time > mq->high_time);
+ } else {
+ should_wait = GST_CLOCK_STIME_IS_VALID (next_time) &&
+ next_time > sq->group_high_time;
+ }
+ } else
+ should_wait = newid > mq->highid;
}
/* Re-compute the high_id in case someone else pushed */
compute_high_id (mq);
+ compute_high_time (mq, sq->groupid);
} else {
compute_high_id (mq);
+ compute_high_time (mq, sq->groupid);
/* Wake up all non-linked pads */
wake_up_next_non_linked (mq);
}
/* We're done waiting, we can clear the nextid and nexttime */
sq->nextid = 0;
- sq->next_time = GST_CLOCK_TIME_NONE;
+ sq->next_time = GST_CLOCK_STIME_NONE;
}
GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
if (sq->flushing)
goto out_flushing;
- GST_LOG_OBJECT (mq, "BEFORE PUSHING sq->srcresult: %s",
+ GST_LOG_OBJECT (mq, "sq:%d BEFORE PUSHING sq->srcresult: %s", sq->id,
gst_flow_get_name (sq->srcresult));
/* Update time stats */
GST_MULTI_QUEUE_MUTEX_LOCK (mq);
- next_time = get_running_time (&sq->src_segment, object, FALSE);
- if (next_time != GST_CLOCK_TIME_NONE) {
- if (sq->last_time == GST_CLOCK_TIME_NONE || sq->last_time < next_time)
+ next_time = get_running_time (&sq->src_segment, object, TRUE);
+ if (GST_CLOCK_STIME_IS_VALID (next_time)) {
+ if (sq->last_time == GST_CLOCK_STIME_NONE || sq->last_time < next_time)
sq->last_time = next_time;
- if (mq->high_time == GST_CLOCK_TIME_NONE || mq->high_time <= next_time) {
+ if (mq->high_time == GST_CLOCK_STIME_NONE || mq->high_time <= next_time) {
/* Wake up all non-linked pads now that we advanced the high time */
mq->high_time = next_time;
wake_up_next_non_linked (mq);
GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
/* Try to push out the new object */
- result = gst_single_queue_push_one (mq, sq, object);
+ result = gst_single_queue_push_one (mq, sq, object, &dropping);
+ object = NULL;
/* Check if we pushed something already and if this is
* now a switch from an active to a non-active stream.
sq->id);
compute_high_id (mq);
+ compute_high_time (mq, sq->groupid);
+ do_update_buffering = TRUE;
/* maybe no-one is waiting */
if (mq->numwaiting > 0) {
}
}
- if (GST_IS_BUFFER (object))
+ if (is_buffer)
sq->pushed = TRUE;
+
+ /* now hold on a bit;
+ * can not simply throw this result to upstream, because
+ * that might already be onto another segment, so we have to make
+ * sure we are relaying the correct info wrt proper segment */
+ if (result == GST_FLOW_EOS && !dropping &&
+ sq->srcresult != GST_FLOW_NOT_LINKED) {
+ GST_DEBUG_OBJECT (mq, "starting EOS drop on sq %d", sq->id);
+ dropping = TRUE;
+ /* pretend we have not seen EOS yet for upstream's sake */
+ result = sq->srcresult;
+ } else if (dropping && gst_data_queue_is_empty (sq->queue)) {
+ /* queue empty, so stop dropping
+ * we can commit the result we have now,
+ * which is either OK after a segment, or EOS */
+ GST_DEBUG_OBJECT (mq, "committed EOS drop on sq %d", sq->id);
+ dropping = FALSE;
+ result = GST_FLOW_EOS;
+ }
sq->srcresult = result;
sq->last_oldid = newid;
+
+ if (do_update_buffering)
+ update_buffering (mq, sq);
+
GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ gst_multi_queue_post_buffering (mq);
- object = NULL;
+ GST_LOG_OBJECT (mq, "sq:%d AFTER PUSHING sq->srcresult: %s (is_eos:%d)",
+ sq->id, gst_flow_get_name (sq->srcresult), GST_PAD_IS_EOS (sq->srcpad));
+
+ /* Need to make sure wake up any sleeping pads when we exit */
+ GST_MULTI_QUEUE_MUTEX_LOCK (mq);
+ if (mq->numwaiting > 0 && (GST_PAD_IS_EOS (sq->srcpad)
+ || sq->srcresult == GST_FLOW_EOS)) {
+ compute_high_time (mq, sq->groupid);
+ compute_high_id (mq);
+ wake_up_next_non_linked (mq);
+ }
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+
+ if (dropping)
+ goto next;
if (result != GST_FLOW_OK && result != GST_FLOW_NOT_LINKED
&& result != GST_FLOW_EOS)
goto out_flushing;
- GST_LOG_OBJECT (mq, "AFTER PUSHING sq->srcresult: %s",
- gst_flow_get_name (sq->srcresult));
-
return;
out_flushing:
if (object)
gst_mini_object_unref (object);
- /* Need to make sure wake up any sleeping pads when we exit */
GST_MULTI_QUEUE_MUTEX_LOCK (mq);
- compute_high_time (mq);
- compute_high_id (mq);
- wake_up_next_non_linked (mq);
- GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ sq->last_query = FALSE;
+ g_cond_signal (&sq->query_handled);
+
+ /* Post an error message if we got EOS while downstream
+ * has returned an error flow return. After EOS there
+ * will be no further buffer which could propagate the
+ * error upstream */
+ if (sq->is_eos && sq->srcresult < GST_FLOW_EOS) {
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ GST_ELEMENT_FLOW_ERROR (mq, sq->srcresult);
+ } else {
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ }
/* upstream needs to see fatal result ASAP to shut things down,
* but might be stuck in one of our other full queues;
* so empty this one and trigger dynamic queue growth. At
* this point the srcresult is not OK, NOT_LINKED
* or EOS, i.e. a real failure */
- gst_data_queue_flush (sq->queue);
+ gst_single_queue_flush_queue (sq, FALSE);
single_queue_underrun_cb (sq->queue, sq);
gst_data_queue_set_flushing (sq->queue, TRUE);
gst_pad_pause_task (sq->srcpad);
if (sq->is_eos)
goto was_eos;
+ sq->active = TRUE;
+
/* Get a unique incrementing id */
curid = g_atomic_int_add ((gint *) & mq->counter, 1);
- GST_LOG_OBJECT (mq, "SingleQueue %d : about to enqueue buffer %p with id %d",
- sq->id, buffer, curid);
+ timestamp = GST_BUFFER_DTS_OR_PTS (buffer);
+ duration = GST_BUFFER_DURATION (buffer);
+
+ GST_LOG_OBJECT (mq,
+ "SingleQueue %d : about to enqueue buffer %p with id %d (pts:%"
+ GST_TIME_FORMAT " dts:%" GST_TIME_FORMAT " dur:%" GST_TIME_FORMAT ")",
+ sq->id, buffer, curid, GST_TIME_ARGS (GST_BUFFER_PTS (buffer)),
+ GST_TIME_ARGS (GST_BUFFER_DTS (buffer)), GST_TIME_ARGS (duration));
item = gst_multi_queue_buffer_item_new (GST_MINI_OBJECT_CAST (buffer), curid);
- timestamp = GST_BUFFER_TIMESTAMP (buffer);
- duration = GST_BUFFER_DURATION (buffer);
+ /* Update interleave before pushing data into queue */
+ if (mq->use_interleave) {
+ GstClockTime val = timestamp;
+ GstClockTimeDiff dval;
+
+ GST_MULTI_QUEUE_MUTEX_LOCK (mq);
+ if (val == GST_CLOCK_TIME_NONE)
+ val = sq->sink_segment.position;
+ if (duration != GST_CLOCK_TIME_NONE)
+ val += duration;
+
+ dval = my_segment_to_running_time (&sq->sink_segment, val);
+ if (GST_CLOCK_STIME_IS_VALID (dval)) {
+ sq->cached_sinktime = dval;
+ GST_DEBUG_OBJECT (mq,
+ "Queue %d cached sink time now %" G_GINT64_FORMAT " %"
+ GST_STIME_FORMAT, sq->id, sq->cached_sinktime,
+ GST_STIME_ARGS (sq->cached_sinktime));
+ calculate_interleave (mq);
+ }
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ }
if (!(gst_data_queue_push (sq->queue, (GstDataQueueItem *) item)))
goto flushing;
/* All pads start off linked until they push one buffer */
sq->srcresult = GST_FLOW_OK;
sq->pushed = FALSE;
+ gst_data_queue_set_flushing (sq->queue, FALSE);
} else {
sq->srcresult = GST_FLOW_FLUSHING;
+ sq->last_query = FALSE;
+ g_cond_signal (&sq->query_handled);
+ gst_data_queue_set_flushing (sq->queue, TRUE);
+
+ /* Wait until streaming thread has finished */
+ if (mq)
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ GST_PAD_STREAM_LOCK (pad);
+ if (mq)
+ GST_MULTI_QUEUE_MUTEX_LOCK (mq);
gst_data_queue_flush (sq->queue);
+ if (mq)
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ GST_PAD_STREAM_UNLOCK (pad);
+ if (mq)
+ GST_MULTI_QUEUE_MUTEX_LOCK (mq);
}
res = TRUE;
break;
return res;
}
-static gboolean
+static GstFlowReturn
gst_multi_queue_sink_event (GstPad * pad, GstObject * parent, GstEvent * event)
{
GstSingleQueue *sq;
GstMultiQueue *mq;
guint32 curid;
GstMultiQueueItem *item;
- gboolean res;
+ gboolean res = TRUE;
+ GstFlowReturn flowret = GST_FLOW_OK;
GstEventType type;
GstEvent *sref = NULL;
type = GST_EVENT_TYPE (event);
switch (type) {
+ case GST_EVENT_STREAM_START:
+ {
+ if (mq->sync_by_running_time) {
+ GstStreamFlags stream_flags;
+ gst_event_parse_stream_flags (event, &stream_flags);
+ if ((stream_flags & GST_STREAM_FLAG_SPARSE)) {
+ GST_INFO_OBJECT (mq, "SingleQueue %d is a sparse stream", sq->id);
+ sq->is_sparse = TRUE;
+ }
+ }
+
+ sq->thread = g_thread_self ();
+
+ /* Remove EOS flag */
+ sq->is_eos = FALSE;
+ break;
+ }
case GST_EVENT_FLUSH_START:
GST_DEBUG_OBJECT (mq, "SingleQueue %d : received flush start event",
sq->id);
res = gst_pad_push_event (sq->srcpad, event);
- gst_single_queue_flush (mq, sq, TRUE);
+ gst_single_queue_flush (mq, sq, TRUE, FALSE);
goto done;
case GST_EVENT_FLUSH_STOP:
res = gst_pad_push_event (sq->srcpad, event);
- gst_single_queue_flush (mq, sq, FALSE);
+ gst_single_queue_flush (mq, sq, FALSE, FALSE);
goto done;
+
case GST_EVENT_SEGMENT:
+ sref = gst_event_ref (event);
+ break;
+ case GST_EVENT_GAP:
/* take ref because the queue will take ownership and we need the event
* afterwards to update the segment */
sref = gst_event_ref (event);
+ if (mq->use_interleave) {
+ GstClockTime val, dur;
+ GstClockTime stime;
+ gst_event_parse_gap (event, &val, &dur);
+ if (GST_CLOCK_TIME_IS_VALID (val)) {
+ GST_MULTI_QUEUE_MUTEX_LOCK (mq);
+ if (GST_CLOCK_TIME_IS_VALID (dur))
+ val += dur;
+ stime = my_segment_to_running_time (&sq->sink_segment, val);
+ if (GST_CLOCK_STIME_IS_VALID (stime)) {
+ sq->cached_sinktime = stime;
+ calculate_interleave (mq);
+ }
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ }
+ }
break;
default:
"SingleQueue %d : Enqueuing event %p of type %s with id %d",
sq->id, event, GST_EVENT_TYPE_NAME (event), curid);
- if (!(res = gst_data_queue_push (sq->queue, (GstDataQueueItem *) item)))
+ if (!gst_data_queue_push (sq->queue, (GstDataQueueItem *) item))
goto flushing;
/* mark EOS when we received one, we must do that after putting the
- * buffer in the queue because EOS marks the buffer as filled. No need to take
- * a lock, the _check_full happens from this thread only, right before pushing
- * into dataqueue. */
+ * buffer in the queue because EOS marks the buffer as filled. */
switch (type) {
case GST_EVENT_EOS:
+ GST_MULTI_QUEUE_MUTEX_LOCK (mq);
sq->is_eos = TRUE;
+
+ /* Post an error message if we got EOS while downstream
+ * has returned an error flow return. After EOS there
+ * will be no further buffer which could propagate the
+ * error upstream */
+ if (sq->srcresult < GST_FLOW_EOS) {
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ GST_ELEMENT_FLOW_ERROR (mq, sq->srcresult);
+ } else {
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ }
+
/* EOS affects the buffering state */
+ GST_MULTI_QUEUE_MUTEX_LOCK (mq);
update_buffering (mq, sq);
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
single_queue_overrun_cb (sq->queue, sq);
+ gst_multi_queue_post_buffering (mq);
break;
case GST_EVENT_SEGMENT:
apply_segment (mq, sq, sref, &sq->sink_segment);
gst_event_unref (sref);
+ /* a new segment allows us to accept more buffers if we got EOS
+ * from downstream */
+ GST_MULTI_QUEUE_MUTEX_LOCK (mq);
+ if (sq->srcresult == GST_FLOW_EOS)
+ sq->srcresult = GST_FLOW_OK;
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
break;
+ case GST_EVENT_GAP:
+ sq->active = TRUE;
+ apply_gap (mq, sq, sref, &sq->sink_segment);
+ gst_event_unref (sref);
default:
break;
}
+
done:
- return res;
+ if (res == FALSE)
+ flowret = GST_FLOW_ERROR;
+ GST_DEBUG_OBJECT (mq, "SingleQueue %d : returning %s", sq->id,
+ gst_flow_get_name (flowret));
+ return flowret;
flushing:
{
if (sref)
gst_event_unref (sref);
gst_multi_queue_item_destroy (item);
- goto done;
+ return sq->srcresult;
}
was_eos:
{
- GST_DEBUG_OBJECT (mq, "we are EOS, dropping event, return FALSE");
+ GST_DEBUG_OBJECT (mq, "we are EOS, dropping event, return GST_FLOW_EOS");
gst_event_unref (event);
- res = FALSE;
- goto done;
+ return GST_FLOW_EOS;
}
}
guint32 curid;
GstMultiQueueItem *item;
- /* Get an unique incrementing id. */
- curid = g_atomic_int_add ((gint *) & mq->counter, 1);
-
- item = gst_multi_queue_mo_item_new ((GstMiniObject *) query, curid);
-
- GST_DEBUG_OBJECT (mq,
- "SingleQueue %d : Enqueuing query %p of type %s with id %d",
- sq->id, query, GST_QUERY_TYPE_NAME (query), curid);
-
GST_MULTI_QUEUE_MUTEX_LOCK (mq);
- res = gst_data_queue_push (sq->queue, (GstDataQueueItem *) item);
- g_cond_wait (&sq->query_handled, &mq->qlock);
- res = sq->last_query;
+ if (sq->srcresult != GST_FLOW_OK)
+ goto out_flushing;
+
+ /* serialized events go in the queue. We need to be certain that we
+ * don't cause deadlocks waiting for the query return value. We check if
+ * the queue is empty (nothing is blocking downstream and the query can
+ * be pushed for sure) or we are not buffering. If we are buffering,
+ * the pipeline waits to unblock downstream until our queue fills up
+ * completely, which can not happen if we block on the query..
+ * Therefore we only potentially block when we are not buffering. */
+ if (!mq->use_buffering || gst_data_queue_is_empty (sq->queue)) {
+ /* Get an unique incrementing id. */
+ curid = g_atomic_int_add ((gint *) & mq->counter, 1);
+
+ item = gst_multi_queue_mo_item_new ((GstMiniObject *) query, curid);
+
+ GST_DEBUG_OBJECT (mq,
+ "SingleQueue %d : Enqueuing query %p of type %s with id %d",
+ sq->id, query, GST_QUERY_TYPE_NAME (query), curid);
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ res = gst_data_queue_push (sq->queue, (GstDataQueueItem *) item);
+ GST_MULTI_QUEUE_MUTEX_LOCK (mq);
+ if (!res || sq->flushing)
+ goto out_flushing;
+ /* it might be that the query has been taken out of the queue
+ * while we were unlocked. So, we need to check if the last
+ * handled query is the same one than the one we just
+ * pushed. If it is, we don't need to wait for the condition
+ * variable, otherwise we wait for the condition variable to
+ * be signaled. */
+ if (sq->last_handled_query != query)
+ g_cond_wait (&sq->query_handled, &mq->qlock);
+ res = sq->last_query;
+ sq->last_handled_query = NULL;
+ } else {
+ GST_DEBUG_OBJECT (mq, "refusing query, we are buffering and the "
+ "queue is not empty");
+ res = FALSE;
+ }
GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
} else {
/* default handling */
break;
}
return res;
+
+out_flushing:
+ {
+ GST_DEBUG_OBJECT (mq, "Flushing");
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+ return FALSE;
+ }
}
static gboolean
switch (mode) {
case GST_PAD_MODE_PUSH:
if (active) {
- result = gst_single_queue_flush (mq, sq, FALSE);
+ result = gst_single_queue_flush (mq, sq, FALSE, TRUE);
} else {
- result = gst_single_queue_flush (mq, sq, TRUE);
+ result = gst_single_queue_flush (mq, sq, TRUE, TRUE);
/* make sure streaming finishes */
result |= gst_pad_stop_task (pad);
}
gst_multi_queue_src_event (GstPad * pad, GstObject * parent, GstEvent * event)
{
GstSingleQueue *sq = gst_pad_get_element_private (pad);
+ GstMultiQueue *mq = sq->mqueue;
+ gboolean ret;
- return gst_pad_push_event (sq->sinkpad, event);
+ switch (GST_EVENT_TYPE (event)) {
+ case GST_EVENT_RECONFIGURE:
+ GST_MULTI_QUEUE_MUTEX_LOCK (mq);
+ if (sq->srcresult == GST_FLOW_NOT_LINKED) {
+ sq->srcresult = GST_FLOW_OK;
+ g_cond_signal (&sq->turn);
+ }
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
+
+ ret = gst_pad_push_event (sq->sinkpad, event);
+ break;
+ default:
+ ret = gst_pad_push_event (sq->sinkpad, event);
+ break;
+ }
+
+ return ret;
}
static gboolean
if (mq->numwaiting < 1)
return;
- /* Else figure out which singlequeue(s) need waking up */
- for (tmp = mq->queues; tmp; tmp = g_list_next (tmp)) {
- GstSingleQueue *sq = (GstSingleQueue *) tmp->data;
-
- if (sq->srcresult == GST_FLOW_NOT_LINKED) {
- if ((mq->sync_by_running_time && mq->high_time != GST_CLOCK_TIME_NONE
- && sq->next_time != GST_CLOCK_TIME_NONE
- && sq->next_time >= mq->high_time)
- || (sq->nextid != 0 && sq->nextid <= mq->highid)) {
+ if (mq->sync_by_running_time && GST_CLOCK_STIME_IS_VALID (mq->high_time)) {
+ /* Else figure out which singlequeue(s) need waking up */
+ for (tmp = mq->queues; tmp; tmp = tmp->next) {
+ GstSingleQueue *sq = (GstSingleQueue *) tmp->data;
+ if (sq->srcresult == GST_FLOW_NOT_LINKED) {
+ GstClockTimeDiff high_time;
+
+ if (GST_CLOCK_STIME_IS_VALID (sq->group_high_time))
+ high_time = sq->group_high_time;
+ else
+ high_time = mq->high_time;
+
+ if (GST_CLOCK_STIME_IS_VALID (sq->next_time) &&
+ GST_CLOCK_STIME_IS_VALID (high_time)
+ && sq->next_time <= high_time) {
+ GST_LOG_OBJECT (mq, "Waking up singlequeue %d", sq->id);
+ g_cond_signal (&sq->turn);
+ }
+ }
+ }
+ } else {
+ /* Else figure out which singlequeue(s) need waking up */
+ for (tmp = mq->queues; tmp; tmp = tmp->next) {
+ GstSingleQueue *sq = (GstSingleQueue *) tmp->data;
+ if (sq->srcresult == GST_FLOW_NOT_LINKED &&
+ sq->nextid != 0 && sq->nextid <= mq->highid) {
GST_LOG_OBJECT (mq, "Waking up singlequeue %d", sq->id);
g_cond_signal (&sq->turn);
}
if (sq->nextid < lowest)
lowest = sq->nextid;
- } else if (sq->srcresult != GST_FLOW_EOS) {
+ } else if (!GST_PAD_IS_EOS (sq->srcpad) && sq->srcresult != GST_FLOW_EOS) {
/* If we don't have a global highid, or the global highid is lower than
* this single queue's last outputted id, store the queue's one,
- * unless the singlequeue is at EOS (srcresult = EOS) */
+ * unless the singlequeue output is at EOS */
if ((highid == G_MAXUINT32) || (sq->oldid > highid))
highid = sq->oldid;
}
/* WITH LOCK TAKEN */
static void
-compute_high_time (GstMultiQueue * mq)
+compute_high_time (GstMultiQueue * mq, guint groupid)
{
- /* The high-id is either the highest id among the linked pads, or if all
- * pads are not-linked, it's the lowest not-linked pad */
+ /* The high-time is either the highest last time among the linked
+ * pads, or if all pads are not-linked, it's the lowest nex time of
+ * not-linked pad */
GList *tmp;
- GstClockTime highest = GST_CLOCK_TIME_NONE;
- GstClockTime lowest = GST_CLOCK_TIME_NONE;
+ GstClockTimeDiff highest = GST_CLOCK_STIME_NONE;
+ GstClockTimeDiff lowest = GST_CLOCK_STIME_NONE;
+ GstClockTimeDiff group_high = GST_CLOCK_STIME_NONE;
+ GstClockTimeDiff group_low = GST_CLOCK_STIME_NONE;
+ GstClockTimeDiff res;
+ /* Number of streams which belong to groupid */
+ guint group_count = 0;
+
+ if (!mq->sync_by_running_time)
+ /* return GST_CLOCK_STIME_NONE; */
+ return;
- for (tmp = mq->queues; tmp; tmp = g_list_next (tmp)) {
+ for (tmp = mq->queues; tmp; tmp = tmp->next) {
GstSingleQueue *sq = (GstSingleQueue *) tmp->data;
GST_LOG_OBJECT (mq,
- "inspecting sq:%d , next_time:%" GST_TIME_FORMAT ", last_time:%"
- GST_TIME_FORMAT ", srcresult:%s", sq->id, GST_TIME_ARGS (sq->next_time),
- GST_TIME_ARGS (sq->last_time), gst_flow_get_name (sq->srcresult));
+ "inspecting sq:%d (group:%d) , next_time:%" GST_STIME_FORMAT
+ ", last_time:%" GST_STIME_FORMAT ", srcresult:%s", sq->id, sq->groupid,
+ GST_STIME_ARGS (sq->next_time), GST_STIME_ARGS (sq->last_time),
+ gst_flow_get_name (sq->srcresult));
+
+ if (sq->groupid == groupid)
+ group_count++;
if (sq->srcresult == GST_FLOW_NOT_LINKED) {
/* No need to consider queues which are not waiting */
- if (sq->next_time == GST_CLOCK_TIME_NONE) {
+ if (!GST_CLOCK_STIME_IS_VALID (sq->next_time)) {
GST_LOG_OBJECT (mq, "sq:%d is not waiting - ignoring", sq->id);
continue;
}
- if (lowest == GST_CLOCK_TIME_NONE || sq->next_time < lowest)
+ if (lowest == GST_CLOCK_STIME_NONE || sq->next_time < lowest)
lowest = sq->next_time;
- } else if (sq->srcresult != GST_FLOW_EOS) {
- /* If we don't have a global highid, or the global highid is lower than
- * this single queue's last outputted id, store the queue's one,
- * unless the singlequeue is at EOS (srcresult = EOS) */
- if (highest == GST_CLOCK_TIME_NONE || sq->last_time > highest)
+ if (sq->groupid == groupid && (group_low == GST_CLOCK_STIME_NONE
+ || sq->next_time < group_low))
+ group_low = sq->next_time;
+ } else if (!GST_PAD_IS_EOS (sq->srcpad) && sq->srcresult != GST_FLOW_EOS) {
+ /* If we don't have a global high time, or the global high time
+ * is lower than this single queue's last outputted time, store
+ * the queue's one, unless the singlequeue output is at EOS. */
+ if (highest == GST_CLOCK_STIME_NONE
+ || (sq->last_time != GST_CLOCK_STIME_NONE && sq->last_time > highest))
highest = sq->last_time;
+ if (sq->groupid == groupid && (group_high == GST_CLOCK_STIME_NONE
+ || (sq->last_time != GST_CLOCK_STIME_NONE
+ && sq->last_time > group_high)))
+ group_high = sq->last_time;
}
+ GST_LOG_OBJECT (mq,
+ "highest now %" GST_STIME_FORMAT " lowest %" GST_STIME_FORMAT,
+ GST_STIME_ARGS (highest), GST_STIME_ARGS (lowest));
+ if (sq->groupid == groupid)
+ GST_LOG_OBJECT (mq,
+ "grouphigh %" GST_STIME_FORMAT " grouplow %" GST_STIME_FORMAT,
+ GST_STIME_ARGS (group_high), GST_STIME_ARGS (group_low));
}
- mq->high_time = highest;
+ if (highest == GST_CLOCK_STIME_NONE)
+ mq->high_time = lowest;
+ else
+ mq->high_time = highest;
+
+ /* If there's only one stream of a given type, use the global high */
+ if (group_count < 2)
+ res = GST_CLOCK_STIME_NONE;
+ else if (group_high == GST_CLOCK_STIME_NONE)
+ res = group_low;
+ else
+ res = group_high;
+ GST_LOG_OBJECT (mq, "group count %d for groupid %u", group_count, groupid);
GST_LOG_OBJECT (mq,
- "High time is now : %" GST_TIME_FORMAT ", lowest non-linked %"
- GST_TIME_FORMAT, GST_TIME_ARGS (mq->high_time), GST_TIME_ARGS (lowest));
+ "MQ High time is now : %" GST_STIME_FORMAT ", group %d high time %"
+ GST_STIME_FORMAT ", lowest non-linked %" GST_STIME_FORMAT,
+ GST_STIME_ARGS (mq->high_time), groupid, GST_STIME_ARGS (mq->high_time),
+ GST_STIME_ARGS (lowest));
+
+ for (tmp = mq->queues; tmp; tmp = tmp->next) {
+ GstSingleQueue *sq = (GstSingleQueue *) tmp->data;
+ if (groupid == sq->groupid)
+ sq->group_high_time = res;
+ }
}
#define IS_FILLED(q, format, value) (((q)->max_size.format) != 0 && \
GstMultiQueue *mq = sq->mqueue;
GList *tmp;
GstDataQueueSize size;
- gboolean filled = FALSE;
+ gboolean filled = TRUE;
+ gboolean empty_found = FALSE;
gst_data_queue_get_level (sq->queue, &size);
- GST_LOG_OBJECT (mq, "Single Queue %d is full", sq->id);
+ GST_LOG_OBJECT (mq,
+ "Single Queue %d: EOS %d, visible %u/%u, bytes %u/%u, time %"
+ G_GUINT64_FORMAT "/%" G_GUINT64_FORMAT, sq->id, sq->is_eos, size.visible,
+ sq->max_size.visible, size.bytes, sq->max_size.bytes, sq->cur_time,
+ sq->max_size.time);
GST_MULTI_QUEUE_MUTEX_LOCK (mq);
+
+ /* check if we reached the hard time/bytes limits;
+ time limit is only taken into account for non-sparse streams */
+ if (sq->is_eos || IS_FILLED (sq, bytes, size.bytes) ||
+ (!sq->is_sparse && IS_FILLED (sq, time, sq->cur_time))) {
+ goto done;
+ }
+
+ /* Search for empty queues */
for (tmp = mq->queues; tmp; tmp = g_list_next (tmp)) {
GstSingleQueue *oq = (GstSingleQueue *) tmp->data;
- GstDataQueueSize ssize;
- GST_LOG_OBJECT (mq, "Checking Queue %d", oq->id);
+ if (oq == sq)
+ continue;
- if (gst_data_queue_is_empty (oq->queue)) {
+ if (oq->srcresult == GST_FLOW_NOT_LINKED) {
+ GST_LOG_OBJECT (mq, "Queue %d is not-linked", oq->id);
+ continue;
+ }
+
+ GST_LOG_OBJECT (mq, "Checking Queue %d", oq->id);
+ if (gst_data_queue_is_empty (oq->queue) && !oq->is_sparse) {
GST_LOG_OBJECT (mq, "Queue %d is empty", oq->id);
- if (IS_FILLED (sq, visible, size.visible)) {
- sq->max_size.visible = size.visible + 1;
- GST_DEBUG_OBJECT (mq,
- "Another queue is empty, bumping single queue %d max visible to %d",
- sq->id, sq->max_size.visible);
- }
+ empty_found = TRUE;
+ break;
}
- /* check if we reached the hard time/bytes limits */
- gst_data_queue_get_level (oq->queue, &ssize);
+ }
- GST_DEBUG_OBJECT (mq,
- "queue %d: visible %u/%u, bytes %u/%u, time %" G_GUINT64_FORMAT "/%"
- G_GUINT64_FORMAT, oq->id, ssize.visible, oq->max_size.visible,
- ssize.bytes, oq->max_size.bytes, oq->cur_time, oq->max_size.time);
-
- /* if this queue is filled completely we must signal overrun.
- * FIXME, this seems wrong in many ways
- * - we're comparing the filled level of this queue against the
- * values of the other one
- * - we should only do this after we found no empty queues, ie, move
- * this check outside of the loop
- * - the debug statement talks about a different queue than the one
- * we are checking here.
- */
- if (sq->is_eos || IS_FILLED (sq, bytes, ssize.bytes) ||
- IS_FILLED (sq, time, sq->cur_time)) {
- GST_LOG_OBJECT (mq, "Queue %d is filled", oq->id);
- filled = TRUE;
+ /* if hard limits are not reached then we allow one more buffer in the full
+ * queue, but only if any of the other singelqueues are empty */
+ if (empty_found) {
+ if (IS_FILLED (sq, visible, size.visible)) {
+ sq->max_size.visible = size.visible + 1;
+ GST_DEBUG_OBJECT (mq,
+ "Bumping single queue %d max visible to %d",
+ sq->id, sq->max_size.visible);
+ filled = FALSE;
}
}
- /* no queues were empty */
+
+done:
GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
/* Overrun is always forwarded, since this is blocking the upstream element */
if (filled) {
- GST_DEBUG_OBJECT (mq, "A queue is filled, signalling overrun");
+ GST_DEBUG_OBJECT (mq, "Queue %d is filled, signalling overrun", sq->id);
g_signal_emit (mq, gst_multi_queue_signals[SIGNAL_OVERRUN], 0);
}
-
- return;
}
static void
GstMultiQueue *mq = sq->mqueue;
GList *tmp;
- GST_LOG_OBJECT (mq,
- "Single Queue %d is empty, Checking other single queues", sq->id);
+ if (sq->srcresult == GST_FLOW_NOT_LINKED) {
+ GST_LOG_OBJECT (mq, "Single Queue %d is empty but not-linked", sq->id);
+ return;
+ } else {
+ GST_LOG_OBJECT (mq,
+ "Single Queue %d is empty, Checking other single queues", sq->id);
+ }
GST_MULTI_QUEUE_MUTEX_LOCK (mq);
for (tmp = mq->queues; tmp; tmp = g_list_next (tmp)) {
gst_data_queue_limits_changed (oq->queue);
}
}
- if (!gst_data_queue_is_empty (oq->queue))
+ if (!gst_data_queue_is_empty (oq->queue) || oq->is_sparse)
empty = FALSE;
}
GST_MULTI_QUEUE_MUTEX_UNLOCK (mq);
return TRUE;
/* check time or bytes */
- res = IS_FILLED (sq, time, sq->cur_time) || IS_FILLED (sq, bytes, bytes);
+ res = IS_FILLED (sq, bytes, bytes);
+ /* We only care about limits in time if we're not a sparse stream or
+ * we're not syncing by running time */
+ if (!sq->is_sparse || !mq->sync_by_running_time) {
+ /* If unlinked, take into account the extra unlinked cache time */
+ if (mq->sync_by_running_time && sq->srcresult == GST_FLOW_NOT_LINKED) {
+ if (sq->cur_time > mq->unlinked_cache_time)
+ res |= IS_FILLED (sq, time, sq->cur_time - mq->unlinked_cache_time);
+ else
+ res = FALSE;
+ } else
+ res |= IS_FILLED (sq, time, sq->cur_time);
+ }
return res;
}
static void
+gst_single_queue_flush_queue (GstSingleQueue * sq, gboolean full)
+{
+ GstDataQueueItem *sitem;
+ GstMultiQueueItem *mitem;
+ gboolean was_flushing = FALSE;
+
+ while (!gst_data_queue_is_empty (sq->queue)) {
+ GstMiniObject *data;
+
+ /* FIXME: If this fails here although the queue is not empty,
+ * we're flushing... but we want to rescue all sticky
+ * events nonetheless.
+ */
+ if (!gst_data_queue_pop (sq->queue, &sitem)) {
+ was_flushing = TRUE;
+ gst_data_queue_set_flushing (sq->queue, FALSE);
+ continue;
+ }
+
+ mitem = (GstMultiQueueItem *) sitem;
+
+ data = sitem->object;
+
+ if (!full && !mitem->is_query && GST_IS_EVENT (data)
+ && GST_EVENT_IS_STICKY (data)
+ && GST_EVENT_TYPE (data) != GST_EVENT_SEGMENT
+ && GST_EVENT_TYPE (data) != GST_EVENT_EOS) {
+ gst_pad_store_sticky_event (sq->srcpad, GST_EVENT_CAST (data));
+ }
+
+ sitem->destroy (sitem);
+ }
+
+ gst_data_queue_flush (sq->queue);
+ if (was_flushing)
+ gst_data_queue_set_flushing (sq->queue, TRUE);
+
+ GST_MULTI_QUEUE_MUTEX_LOCK (sq->mqueue);
+ update_buffering (sq->mqueue, sq);
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (sq->mqueue);
+ gst_multi_queue_post_buffering (sq->mqueue);
+}
+
+static void
gst_single_queue_free (GstSingleQueue * sq)
{
/* DRAIN QUEUE */
gst_single_queue_new (GstMultiQueue * mqueue, guint id)
{
GstSingleQueue *sq;
+ GstMultiQueuePad *mqpad;
+ GstPadTemplate *templ;
gchar *name;
GList *tmp;
guint temp_id = (id == -1) ? 0 : id;
if (sq2->id == temp_id) {
/* If this ID was requested by the caller return NULL,
* otherwise just get us the next one */
- if (id == -1)
+ if (id == -1) {
temp_id = sq2->id + 1;
- else
+ } else {
+ GST_MULTI_QUEUE_MUTEX_UNLOCK (mqueue);
return NULL;
+ }
} else if (sq2->id > temp_id) {
break;
}
sq = g_new0 (GstSingleQueue, 1);
mqueue->nbqueues++;
sq->id = temp_id;
+ sq->groupid = DEFAULT_PAD_GROUP_ID;
+ sq->group_high_time = GST_CLOCK_STIME_NONE;
mqueue->queues = g_list_insert_before (mqueue->queues, tmp, sq);
mqueue->queues_cookie++;
sq->mqueue = mqueue;
sq->srcresult = GST_FLOW_FLUSHING;
sq->pushed = FALSE;
- sq->queue = gst_data_queue_new_full ((GstDataQueueCheckFullFunction)
+ sq->queue = gst_data_queue_new ((GstDataQueueCheckFullFunction)
single_queue_check_full,
(GstDataQueueFullCallback) single_queue_overrun_cb,
(GstDataQueueEmptyCallback) single_queue_underrun_cb, sq);
sq->is_eos = FALSE;
+ sq->is_sparse = FALSE;
sq->flushing = FALSE;
+ sq->active = FALSE;
gst_segment_init (&sq->sink_segment, GST_FORMAT_TIME);
gst_segment_init (&sq->src_segment, GST_FORMAT_TIME);
sq->nextid = 0;
sq->oldid = 0;
- sq->next_time = GST_CLOCK_TIME_NONE;
- sq->last_time = GST_CLOCK_TIME_NONE;
+ sq->next_time = GST_CLOCK_STIME_NONE;
+ sq->last_time = GST_CLOCK_STIME_NONE;
g_cond_init (&sq->turn);
g_cond_init (&sq->query_handled);
- sq->sinktime = GST_CLOCK_TIME_NONE;
- sq->srctime = GST_CLOCK_TIME_NONE;
+ sq->sinktime = GST_CLOCK_STIME_NONE;
+ sq->srctime = GST_CLOCK_STIME_NONE;
sq->sink_tainted = TRUE;
sq->src_tainted = TRUE;
name = g_strdup_printf ("sink_%u", sq->id);
- sq->sinkpad = gst_pad_new_from_static_template (&sinktemplate, name);
+ templ = gst_static_pad_template_get (&sinktemplate);
+ sq->sinkpad = g_object_new (GST_TYPE_MULTIQUEUE_PAD, "name", name,
+ "direction", templ->direction, "template", templ, NULL);
+ gst_object_unref (templ);
g_free (name);
+ mqpad = (GstMultiQueuePad *) sq->sinkpad;
+ mqpad->sq = sq;
+
gst_pad_set_chain_function (sq->sinkpad,
GST_DEBUG_FUNCPTR (gst_multi_queue_chain));
gst_pad_set_activatemode_function (sq->sinkpad,
GST_DEBUG_FUNCPTR (gst_multi_queue_sink_activate_mode));
- gst_pad_set_event_function (sq->sinkpad,
+ gst_pad_set_event_full_function (sq->sinkpad,
GST_DEBUG_FUNCPTR (gst_multi_queue_sink_event));
gst_pad_set_query_function (sq->sinkpad,
GST_DEBUG_FUNCPTR (gst_multi_queue_sink_query));