2018-11-05 14:24:58 +00:00
|
|
|
/*
|
|
|
|
* GStreamer
|
|
|
|
* Copyright (C) 2018 Sebastian Dröge <sebastian@centricular.com>
|
|
|
|
*
|
|
|
|
* This library is free software; you can redistribute it and/or
|
|
|
|
* modify it under the terms of the GNU Library General Public
|
|
|
|
* License as published by the Free Software Foundation; either
|
|
|
|
* version 2 of the License, or (at your option) any later version.
|
|
|
|
*
|
|
|
|
* This library is distributed in the hope that it will be useful,
|
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
|
* Library General Public License for more details.
|
|
|
|
*
|
|
|
|
* 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., 51 Franklin St, Fifth Floor,
|
|
|
|
* Boston, MA 02110-1301, USA.
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
|
|
#ifdef HAVE_CONFIG_H
|
|
|
|
# include <config.h>
|
|
|
|
#endif
|
|
|
|
|
|
|
|
#include <gst/gst.h>
|
|
|
|
#include <gst/base/base.h>
|
|
|
|
#include <gst/video/video.h>
|
|
|
|
#include <string.h>
|
|
|
|
|
|
|
|
#include "gstcccombiner.h"
|
|
|
|
|
|
|
|
GST_DEBUG_CATEGORY_STATIC (gst_cc_combiner_debug);
|
|
|
|
#define GST_CAT_DEFAULT gst_cc_combiner_debug
|
|
|
|
|
|
|
|
static GstStaticPadTemplate sinktemplate = GST_STATIC_PAD_TEMPLATE ("sink",
|
|
|
|
GST_PAD_SINK,
|
|
|
|
GST_PAD_ALWAYS,
|
|
|
|
GST_STATIC_CAPS_ANY);
|
|
|
|
|
|
|
|
static GstStaticPadTemplate srctemplate = GST_STATIC_PAD_TEMPLATE ("src",
|
|
|
|
GST_PAD_SRC,
|
|
|
|
GST_PAD_ALWAYS,
|
|
|
|
GST_STATIC_CAPS_ANY);
|
|
|
|
|
|
|
|
static GstStaticPadTemplate captiontemplate =
|
|
|
|
GST_STATIC_PAD_TEMPLATE ("caption",
|
|
|
|
GST_PAD_SINK,
|
|
|
|
GST_PAD_REQUEST,
|
|
|
|
GST_STATIC_CAPS
|
2018-12-10 13:54:49 +00:00
|
|
|
("closedcaption/x-cea-608,format={ (string) raw, (string) s334-1a}; "
|
2018-11-05 14:24:58 +00:00
|
|
|
"closedcaption/x-cea-708,format={ (string) cc_data, (string) cdp }"));
|
|
|
|
|
|
|
|
#define parent_class gst_cc_combiner_parent_class
|
2021-02-17 10:31:35 +00:00
|
|
|
G_DEFINE_TYPE (GstCCCombiner, gst_cc_combiner, GST_TYPE_AGGREGATOR);
|
|
|
|
GST_ELEMENT_REGISTER_DEFINE (cccombiner, "cccombiner",
|
|
|
|
GST_RANK_NONE, GST_TYPE_CCCOMBINER);
|
2018-11-05 14:24:58 +00:00
|
|
|
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
enum
|
|
|
|
{
|
|
|
|
PROP_0,
|
|
|
|
PROP_SCHEDULE,
|
2022-02-01 22:46:43 +00:00
|
|
|
PROP_OUTPUT_PADDING,
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
PROP_MAX_SCHEDULED,
|
|
|
|
};
|
|
|
|
|
|
|
|
#define DEFAULT_MAX_SCHEDULED 30
|
|
|
|
#define DEFAULT_SCHEDULE TRUE
|
2022-02-01 22:46:43 +00:00
|
|
|
#define DEFAULT_OUTPUT_PADDING TRUE
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
|
2018-11-05 14:24:58 +00:00
|
|
|
typedef struct
|
|
|
|
{
|
|
|
|
GstVideoCaptionType caption_type;
|
|
|
|
GstBuffer *buffer;
|
|
|
|
} CaptionData;
|
|
|
|
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
typedef struct
|
|
|
|
{
|
|
|
|
GstBuffer *buffer;
|
|
|
|
GstClockTime running_time;
|
|
|
|
GstClockTime stream_time;
|
|
|
|
} CaptionQueueItem;
|
|
|
|
|
2018-11-05 14:24:58 +00:00
|
|
|
static void
|
|
|
|
caption_data_clear (CaptionData * data)
|
|
|
|
{
|
|
|
|
gst_buffer_unref (data->buffer);
|
|
|
|
}
|
|
|
|
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
static void
|
|
|
|
clear_scheduled (CaptionQueueItem * item)
|
|
|
|
{
|
|
|
|
gst_buffer_unref (item->buffer);
|
|
|
|
}
|
|
|
|
|
2018-11-05 14:24:58 +00:00
|
|
|
static void
|
|
|
|
gst_cc_combiner_finalize (GObject * object)
|
|
|
|
{
|
|
|
|
GstCCCombiner *self = GST_CCCOMBINER (object);
|
|
|
|
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
gst_queue_array_free (self->scheduled[0]);
|
|
|
|
gst_queue_array_free (self->scheduled[1]);
|
2018-11-05 14:24:58 +00:00
|
|
|
g_array_unref (self->current_frame_captions);
|
|
|
|
self->current_frame_captions = NULL;
|
|
|
|
|
|
|
|
G_OBJECT_CLASS (parent_class)->finalize (object);
|
|
|
|
}
|
|
|
|
|
|
|
|
#define GST_FLOW_NEED_DATA GST_FLOW_CUSTOM_SUCCESS
|
|
|
|
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
static const guint8 *
|
|
|
|
extract_cdp (const guint8 * cdp, guint cdp_len, guint * cc_data_len)
|
|
|
|
{
|
|
|
|
GstByteReader br;
|
|
|
|
guint16 u16;
|
|
|
|
guint8 u8;
|
|
|
|
guint8 flags;
|
|
|
|
guint len = 0;
|
|
|
|
const guint8 *cc_data = NULL;
|
|
|
|
|
|
|
|
*cc_data_len = 0;
|
|
|
|
|
|
|
|
/* Header + footer length */
|
|
|
|
if (cdp_len < 11) {
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
|
|
|
|
gst_byte_reader_init (&br, cdp, cdp_len);
|
|
|
|
u16 = gst_byte_reader_get_uint16_be_unchecked (&br);
|
|
|
|
if (u16 != 0x9669) {
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
|
|
|
|
u8 = gst_byte_reader_get_uint8_unchecked (&br);
|
|
|
|
if (u8 != cdp_len) {
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
|
|
|
|
gst_byte_reader_skip_unchecked (&br, 1);
|
|
|
|
|
|
|
|
flags = gst_byte_reader_get_uint8_unchecked (&br);
|
|
|
|
|
|
|
|
/* No cc_data? */
|
|
|
|
if ((flags & 0x40) == 0) {
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* cdp_hdr_sequence_cntr */
|
|
|
|
gst_byte_reader_skip_unchecked (&br, 2);
|
|
|
|
|
|
|
|
/* time_code_present */
|
|
|
|
if (flags & 0x80) {
|
|
|
|
if (gst_byte_reader_get_remaining (&br) < 5) {
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
gst_byte_reader_skip_unchecked (&br, 5);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* ccdata_present */
|
|
|
|
if (flags & 0x40) {
|
|
|
|
guint8 cc_count;
|
|
|
|
|
|
|
|
if (gst_byte_reader_get_remaining (&br) < 2) {
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
u8 = gst_byte_reader_get_uint8_unchecked (&br);
|
|
|
|
if (u8 != 0x72) {
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
|
|
|
|
cc_count = gst_byte_reader_get_uint8_unchecked (&br);
|
|
|
|
if ((cc_count & 0xe0) != 0xe0) {
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
cc_count &= 0x1f;
|
|
|
|
|
|
|
|
if (cc_count == 0)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
len = 3 * cc_count;
|
|
|
|
if (gst_byte_reader_get_remaining (&br) < len)
|
|
|
|
goto done;
|
|
|
|
|
|
|
|
cc_data = gst_byte_reader_get_data_unchecked (&br, len);
|
|
|
|
*cc_data_len = len;
|
|
|
|
}
|
|
|
|
|
|
|
|
done:
|
|
|
|
return cc_data;
|
|
|
|
}
|
|
|
|
|
|
|
|
#define MAX_CDP_PACKET_LEN 256
|
|
|
|
#define MAX_CEA608_LEN 32
|
|
|
|
|
|
|
|
static const struct cdp_fps_entry cdp_fps_table[] = {
|
|
|
|
{0x1f, 24000, 1001, 25, 22, 3 /* FIXME: alternating max cea608 count! */ },
|
|
|
|
{0x2f, 24, 1, 25, 22, 2},
|
|
|
|
{0x3f, 25, 1, 24, 22, 2},
|
|
|
|
{0x4f, 30000, 1001, 20, 18, 2},
|
|
|
|
{0x5f, 30, 1, 20, 18, 2},
|
|
|
|
{0x6f, 50, 1, 12, 11, 1},
|
|
|
|
{0x7f, 60000, 1001, 10, 9, 1},
|
|
|
|
{0x8f, 60, 1, 10, 9, 1},
|
|
|
|
};
|
|
|
|
static const struct cdp_fps_entry null_fps_entry = { 0, 0, 0, 0 };
|
|
|
|
|
|
|
|
static const struct cdp_fps_entry *
|
|
|
|
cdp_fps_entry_from_fps (guint fps_n, guint fps_d)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
for (i = 0; i < G_N_ELEMENTS (cdp_fps_table); i++) {
|
|
|
|
if (cdp_fps_table[i].fps_n == fps_n && cdp_fps_table[i].fps_d == fps_d)
|
|
|
|
return &cdp_fps_table[i];
|
|
|
|
}
|
|
|
|
return &null_fps_entry;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static GstBuffer *
|
|
|
|
make_cdp (GstCCCombiner * self, const guint8 * cc_data, guint cc_data_len,
|
|
|
|
const struct cdp_fps_entry *fps_entry, const GstVideoTimeCode * tc)
|
|
|
|
{
|
|
|
|
GstByteWriter bw;
|
|
|
|
guint8 flags, checksum;
|
|
|
|
guint i, len;
|
|
|
|
GstBuffer *ret = gst_buffer_new_allocate (NULL, MAX_CDP_PACKET_LEN, NULL);
|
|
|
|
GstMapInfo map;
|
|
|
|
|
|
|
|
gst_buffer_map (ret, &map, GST_MAP_WRITE);
|
|
|
|
|
|
|
|
gst_byte_writer_init_with_data (&bw, map.data, MAX_CDP_PACKET_LEN, FALSE);
|
|
|
|
gst_byte_writer_put_uint16_be_unchecked (&bw, 0x9669);
|
|
|
|
/* Write a length of 0 for now */
|
|
|
|
gst_byte_writer_put_uint8_unchecked (&bw, 0);
|
|
|
|
|
|
|
|
gst_byte_writer_put_uint8_unchecked (&bw, fps_entry->fps_idx);
|
|
|
|
|
|
|
|
/* caption_service_active */
|
|
|
|
flags = 0x02;
|
|
|
|
|
|
|
|
/* ccdata_present */
|
|
|
|
flags |= 0x40;
|
|
|
|
|
|
|
|
if (tc && tc->config.fps_n > 0)
|
|
|
|
flags |= 0x80;
|
|
|
|
|
|
|
|
/* reserved */
|
|
|
|
flags |= 0x01;
|
|
|
|
|
|
|
|
gst_byte_writer_put_uint8_unchecked (&bw, flags);
|
|
|
|
|
|
|
|
gst_byte_writer_put_uint16_be_unchecked (&bw, self->cdp_hdr_sequence_cntr);
|
|
|
|
|
|
|
|
if (tc && tc->config.fps_n > 0) {
|
|
|
|
guint8 u8;
|
|
|
|
|
|
|
|
gst_byte_writer_put_uint8_unchecked (&bw, 0x71);
|
|
|
|
/* reserved 11 - 2 bits */
|
|
|
|
u8 = 0xc0;
|
|
|
|
/* tens of hours - 2 bits */
|
|
|
|
u8 |= ((tc->hours / 10) & 0x3) << 4;
|
|
|
|
/* units of hours - 4 bits */
|
|
|
|
u8 |= (tc->hours % 10) & 0xf;
|
|
|
|
gst_byte_writer_put_uint8_unchecked (&bw, u8);
|
|
|
|
|
|
|
|
/* reserved 1 - 1 bit */
|
|
|
|
u8 = 0x80;
|
|
|
|
/* tens of minutes - 3 bits */
|
|
|
|
u8 |= ((tc->minutes / 10) & 0x7) << 4;
|
|
|
|
/* units of minutes - 4 bits */
|
|
|
|
u8 |= (tc->minutes % 10) & 0xf;
|
|
|
|
gst_byte_writer_put_uint8_unchecked (&bw, u8);
|
|
|
|
|
|
|
|
/* field flag - 1 bit */
|
|
|
|
u8 = tc->field_count < 2 ? 0x00 : 0x80;
|
|
|
|
/* tens of seconds - 3 bits */
|
|
|
|
u8 |= ((tc->seconds / 10) & 0x7) << 4;
|
|
|
|
/* units of seconds - 4 bits */
|
|
|
|
u8 |= (tc->seconds % 10) & 0xf;
|
|
|
|
gst_byte_writer_put_uint8_unchecked (&bw, u8);
|
|
|
|
|
|
|
|
/* drop frame flag - 1 bit */
|
|
|
|
u8 = (tc->config.flags & GST_VIDEO_TIME_CODE_FLAGS_DROP_FRAME) ? 0x80 :
|
|
|
|
0x00;
|
|
|
|
/* reserved0 - 1 bit */
|
|
|
|
/* tens of frames - 2 bits */
|
|
|
|
u8 |= ((tc->frames / 10) & 0x3) << 4;
|
|
|
|
/* units of frames 4 bits */
|
|
|
|
u8 |= (tc->frames % 10) & 0xf;
|
|
|
|
gst_byte_writer_put_uint8_unchecked (&bw, u8);
|
|
|
|
}
|
|
|
|
|
|
|
|
gst_byte_writer_put_uint8_unchecked (&bw, 0x72);
|
|
|
|
gst_byte_writer_put_uint8_unchecked (&bw, 0xe0 | fps_entry->max_cc_count);
|
|
|
|
gst_byte_writer_put_data_unchecked (&bw, cc_data, cc_data_len);
|
|
|
|
while (fps_entry->max_cc_count > cc_data_len / 3) {
|
|
|
|
gst_byte_writer_put_uint8_unchecked (&bw, 0xfa);
|
|
|
|
gst_byte_writer_put_uint8_unchecked (&bw, 0x00);
|
|
|
|
gst_byte_writer_put_uint8_unchecked (&bw, 0x00);
|
|
|
|
cc_data_len += 3;
|
|
|
|
}
|
|
|
|
|
|
|
|
gst_byte_writer_put_uint8_unchecked (&bw, 0x74);
|
|
|
|
gst_byte_writer_put_uint16_be_unchecked (&bw, self->cdp_hdr_sequence_cntr);
|
|
|
|
self->cdp_hdr_sequence_cntr++;
|
|
|
|
/* We calculate the checksum afterwards */
|
|
|
|
gst_byte_writer_put_uint8_unchecked (&bw, 0);
|
|
|
|
|
|
|
|
len = gst_byte_writer_get_pos (&bw);
|
|
|
|
gst_byte_writer_set_pos (&bw, 2);
|
|
|
|
gst_byte_writer_put_uint8_unchecked (&bw, len);
|
|
|
|
|
|
|
|
checksum = 0;
|
|
|
|
for (i = 0; i < len; i++) {
|
|
|
|
checksum += map.data[i];
|
|
|
|
}
|
|
|
|
checksum &= 0xff;
|
|
|
|
checksum = 256 - checksum;
|
|
|
|
map.data[len - 1] = checksum;
|
|
|
|
|
|
|
|
gst_buffer_unmap (ret, &map);
|
|
|
|
|
|
|
|
gst_buffer_set_size (ret, len);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static GstBuffer *
|
|
|
|
make_padding (GstCCCombiner * self, const GstVideoTimeCode * tc, guint field)
|
|
|
|
{
|
|
|
|
GstBuffer *ret = NULL;
|
|
|
|
|
|
|
|
switch (self->caption_type) {
|
|
|
|
case GST_VIDEO_CAPTION_TYPE_CEA708_CDP:
|
|
|
|
{
|
2021-07-08 00:24:18 +00:00
|
|
|
const guint8 cc_data[6] = { 0xfc, 0x80, 0x80, 0xf9, 0x80, 0x80 };
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
|
|
|
|
ret = make_cdp (self, cc_data, 6, self->cdp_fps_entry, tc);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case GST_VIDEO_CAPTION_TYPE_CEA708_RAW:
|
|
|
|
{
|
|
|
|
GstMapInfo map;
|
|
|
|
|
|
|
|
ret = gst_buffer_new_allocate (NULL, 3, NULL);
|
|
|
|
|
|
|
|
gst_buffer_map (ret, &map, GST_MAP_WRITE);
|
|
|
|
|
|
|
|
map.data[0] = 0xfc | (field & 0x01);
|
|
|
|
map.data[1] = 0x80;
|
|
|
|
map.data[2] = 0x80;
|
|
|
|
|
|
|
|
gst_buffer_unmap (ret, &map);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case GST_VIDEO_CAPTION_TYPE_CEA608_S334_1A:
|
|
|
|
{
|
|
|
|
GstMapInfo map;
|
|
|
|
|
|
|
|
ret = gst_buffer_new_allocate (NULL, 3, NULL);
|
|
|
|
|
|
|
|
gst_buffer_map (ret, &map, GST_MAP_WRITE);
|
|
|
|
|
2022-01-07 23:16:29 +00:00
|
|
|
map.data[0] = field == 0 ? 0x80 : 0x00;
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
map.data[1] = 0x80;
|
|
|
|
map.data[2] = 0x80;
|
|
|
|
|
|
|
|
gst_buffer_unmap (ret, &map);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case GST_VIDEO_CAPTION_TYPE_CEA608_RAW:
|
|
|
|
{
|
|
|
|
GstMapInfo map;
|
|
|
|
|
|
|
|
ret = gst_buffer_new_allocate (NULL, 2, NULL);
|
|
|
|
|
|
|
|
gst_buffer_map (ret, &map, GST_MAP_WRITE);
|
|
|
|
|
|
|
|
map.data[0] = 0x80;
|
|
|
|
map.data[1] = 0x80;
|
|
|
|
|
|
|
|
gst_buffer_unmap (ret, &map);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
queue_caption (GstCCCombiner * self, GstBuffer * scheduled, guint field)
|
|
|
|
{
|
|
|
|
GstAggregatorPad *caption_pad;
|
|
|
|
CaptionQueueItem item;
|
|
|
|
|
|
|
|
if (self->progressive && field == 1) {
|
|
|
|
gst_buffer_unref (scheduled);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
caption_pad =
|
|
|
|
GST_AGGREGATOR_PAD_CAST (gst_element_get_static_pad (GST_ELEMENT_CAST
|
|
|
|
(self), "caption"));
|
|
|
|
|
|
|
|
g_assert (gst_queue_array_get_length (self->scheduled[field]) <=
|
|
|
|
self->max_scheduled);
|
|
|
|
|
|
|
|
if (gst_queue_array_get_length (self->scheduled[field]) ==
|
|
|
|
self->max_scheduled) {
|
|
|
|
CaptionQueueItem *dropped =
|
|
|
|
gst_queue_array_pop_tail_struct (self->scheduled[field]);
|
|
|
|
|
|
|
|
GST_WARNING_OBJECT (self,
|
|
|
|
"scheduled queue runs too long, dropping %" GST_PTR_FORMAT, dropped);
|
|
|
|
|
|
|
|
gst_element_post_message (GST_ELEMENT_CAST (self),
|
|
|
|
gst_message_new_qos (GST_OBJECT_CAST (self), FALSE,
|
|
|
|
dropped->running_time, dropped->stream_time,
|
|
|
|
GST_BUFFER_PTS (dropped->buffer), GST_BUFFER_DURATION (dropped)));
|
|
|
|
|
|
|
|
gst_buffer_unref (dropped->buffer);
|
|
|
|
}
|
|
|
|
|
|
|
|
gst_object_unref (caption_pad);
|
|
|
|
|
|
|
|
item.buffer = scheduled;
|
|
|
|
item.running_time =
|
|
|
|
gst_segment_to_running_time (&caption_pad->segment, GST_FORMAT_TIME,
|
|
|
|
GST_BUFFER_PTS (scheduled));
|
|
|
|
item.stream_time =
|
|
|
|
gst_segment_to_stream_time (&caption_pad->segment, GST_FORMAT_TIME,
|
|
|
|
GST_BUFFER_PTS (scheduled));
|
|
|
|
|
|
|
|
gst_queue_array_push_tail_struct (self->scheduled[field], &item);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
schedule_cdp (GstCCCombiner * self, const GstVideoTimeCode * tc,
|
|
|
|
const guint8 * data, guint len, GstClockTime pts, GstClockTime duration)
|
|
|
|
{
|
|
|
|
const guint8 *cc_data;
|
|
|
|
guint cc_data_len;
|
|
|
|
gboolean inject = FALSE;
|
|
|
|
|
|
|
|
if ((cc_data = extract_cdp (data, len, &cc_data_len))) {
|
|
|
|
guint8 i;
|
|
|
|
|
|
|
|
for (i = 0; i < cc_data_len / 3; i++) {
|
|
|
|
gboolean cc_valid = (cc_data[i * 3] & 0x04) == 0x04;
|
|
|
|
guint8 cc_type = cc_data[i * 3] & 0x03;
|
|
|
|
|
|
|
|
if (!cc_valid)
|
2021-07-27 21:53:06 +00:00
|
|
|
continue;
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
|
|
|
|
if (cc_type == 0x00 || cc_type == 0x01) {
|
|
|
|
if (cc_data[i * 3 + 1] != 0x80 || cc_data[i * 3 + 2] != 0x80) {
|
|
|
|
inject = TRUE;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
continue;
|
|
|
|
} else {
|
|
|
|
inject = TRUE;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (inject) {
|
|
|
|
GstBuffer *buf =
|
|
|
|
make_cdp (self, cc_data, cc_data_len, self->cdp_fps_entry, tc);
|
|
|
|
|
|
|
|
/* We only set those for QoS reporting purposes */
|
|
|
|
GST_BUFFER_PTS (buf) = pts;
|
|
|
|
GST_BUFFER_DURATION (buf) = duration;
|
|
|
|
|
|
|
|
queue_caption (self, buf, 0);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
schedule_cea608_s334_1a (GstCCCombiner * self, guint8 * data, guint len,
|
|
|
|
GstClockTime pts, GstClockTime duration)
|
|
|
|
{
|
|
|
|
guint8 field0_data[3], field1_data[3];
|
|
|
|
guint field0_len = 0, field1_len = 0;
|
|
|
|
guint i;
|
|
|
|
gboolean field0_608 = FALSE, field1_608 = FALSE;
|
|
|
|
|
|
|
|
if (len % 3 != 0) {
|
|
|
|
GST_WARNING ("Invalid cc_data buffer size %u. Truncating to a multiple "
|
|
|
|
"of 3", len);
|
|
|
|
len = len - (len % 3);
|
|
|
|
}
|
|
|
|
|
|
|
|
for (i = 0; i < len / 3; i++) {
|
2022-01-07 23:16:29 +00:00
|
|
|
if (data[i * 3] & 0x80) {
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
if (field0_608)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
field0_608 = TRUE;
|
|
|
|
|
|
|
|
if (data[i * 3 + 1] == 0x80 && data[i * 3 + 2] == 0x80)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
field0_data[field0_len++] = data[i * 3];
|
|
|
|
field0_data[field0_len++] = data[i * 3 + 1];
|
|
|
|
field0_data[field0_len++] = data[i * 3 + 2];
|
2022-01-07 23:16:29 +00:00
|
|
|
} else {
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
if (field1_608)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
field1_608 = TRUE;
|
|
|
|
|
|
|
|
if (data[i * 3 + 1] == 0x80 && data[i * 3 + 2] == 0x80)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
field1_data[field1_len++] = data[i * 3];
|
|
|
|
field1_data[field1_len++] = data[i * 3 + 1];
|
|
|
|
field1_data[field1_len++] = data[i * 3 + 2];
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (field0_len > 0) {
|
|
|
|
GstBuffer *buf = gst_buffer_new_allocate (NULL, field0_len, NULL);
|
|
|
|
|
|
|
|
gst_buffer_fill (buf, 0, field0_data, field0_len);
|
|
|
|
GST_BUFFER_PTS (buf) = pts;
|
|
|
|
GST_BUFFER_DURATION (buf) = duration;
|
|
|
|
|
|
|
|
queue_caption (self, buf, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (field1_len > 0) {
|
|
|
|
GstBuffer *buf = gst_buffer_new_allocate (NULL, field1_len, NULL);
|
|
|
|
|
|
|
|
gst_buffer_fill (buf, 0, field1_data, field1_len);
|
|
|
|
GST_BUFFER_PTS (buf) = pts;
|
|
|
|
GST_BUFFER_DURATION (buf) = duration;
|
|
|
|
|
|
|
|
queue_caption (self, buf, 1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
schedule_cea708_raw (GstCCCombiner * self, guint8 * data, guint len,
|
|
|
|
GstClockTime pts, GstClockTime duration)
|
|
|
|
{
|
|
|
|
guint8 field0_data[MAX_CDP_PACKET_LEN], field1_data[3];
|
|
|
|
guint field0_len = 0, field1_len = 0;
|
|
|
|
guint i;
|
|
|
|
gboolean field0_608 = FALSE, field1_608 = FALSE;
|
|
|
|
gboolean started_ccp = FALSE;
|
|
|
|
|
|
|
|
if (len % 3 != 0) {
|
|
|
|
GST_WARNING ("Invalid cc_data buffer size %u. Truncating to a multiple "
|
|
|
|
"of 3", len);
|
|
|
|
len = len - (len % 3);
|
|
|
|
}
|
|
|
|
|
|
|
|
for (i = 0; i < len / 3; i++) {
|
|
|
|
gboolean cc_valid = (data[i * 3] & 0x04) == 0x04;
|
|
|
|
guint8 cc_type = data[i * 3] & 0x03;
|
|
|
|
|
|
|
|
if (!started_ccp) {
|
|
|
|
if (cc_type == 0x00) {
|
|
|
|
if (!cc_valid)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (field0_608)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
field0_608 = TRUE;
|
|
|
|
|
|
|
|
if (data[i * 3 + 1] == 0x80 && data[i * 3 + 2] == 0x80)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
field0_data[field0_len++] = data[i * 3];
|
|
|
|
field0_data[field0_len++] = data[i * 3 + 1];
|
|
|
|
field0_data[field0_len++] = data[i * 3 + 2];
|
|
|
|
} else if (cc_type == 0x01) {
|
|
|
|
if (!cc_valid)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (field1_608)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
field1_608 = TRUE;
|
|
|
|
|
|
|
|
if (data[i * 3 + 1] == 0x80 && data[i * 3 + 2] == 0x80)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
field1_data[field1_len++] = data[i * 3];
|
|
|
|
field1_data[field1_len++] = data[i * 3 + 1];
|
|
|
|
field1_data[field1_len++] = data[i * 3 + 2];
|
|
|
|
}
|
|
|
|
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (cc_type & 0x10)
|
|
|
|
started_ccp = TRUE;
|
|
|
|
|
|
|
|
if (!cc_valid)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (cc_type == 0x00 || cc_type == 0x01)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
field0_data[field0_len++] = data[i * 3];
|
|
|
|
field0_data[field0_len++] = data[i * 3 + 1];
|
|
|
|
field0_data[field0_len++] = data[i * 3 + 2];
|
|
|
|
}
|
|
|
|
|
|
|
|
if (field0_len > 0) {
|
|
|
|
GstBuffer *buf = gst_buffer_new_allocate (NULL, field0_len, NULL);
|
|
|
|
|
|
|
|
gst_buffer_fill (buf, 0, field0_data, field0_len);
|
|
|
|
GST_BUFFER_PTS (buf) = pts;
|
|
|
|
GST_BUFFER_DURATION (buf) = duration;
|
|
|
|
|
|
|
|
queue_caption (self, buf, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (field1_len > 0) {
|
|
|
|
GstBuffer *buf = gst_buffer_new_allocate (NULL, field1_len, NULL);
|
|
|
|
|
|
|
|
gst_buffer_fill (buf, 0, field1_data, field1_len);
|
|
|
|
GST_BUFFER_PTS (buf) = pts;
|
|
|
|
GST_BUFFER_DURATION (buf) = duration;
|
|
|
|
|
|
|
|
queue_caption (self, buf, 1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
schedule_cea608_raw (GstCCCombiner * self, guint8 * data, guint len,
|
|
|
|
GstBuffer * buffer)
|
|
|
|
{
|
|
|
|
if (len < 2) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (data[0] != 0x80 || data[1] != 0x80) {
|
|
|
|
queue_caption (self, gst_buffer_ref (buffer), 0);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void
|
|
|
|
schedule_caption (GstCCCombiner * self, GstBuffer * caption_buf,
|
|
|
|
const GstVideoTimeCode * tc)
|
|
|
|
{
|
|
|
|
GstMapInfo map;
|
|
|
|
GstClockTime pts, duration;
|
|
|
|
|
|
|
|
pts = GST_BUFFER_PTS (caption_buf);
|
|
|
|
duration = GST_BUFFER_DURATION (caption_buf);
|
|
|
|
|
|
|
|
gst_buffer_map (caption_buf, &map, GST_MAP_READ);
|
|
|
|
|
|
|
|
switch (self->caption_type) {
|
|
|
|
case GST_VIDEO_CAPTION_TYPE_CEA708_CDP:
|
|
|
|
schedule_cdp (self, tc, map.data, map.size, pts, duration);
|
|
|
|
break;
|
|
|
|
case GST_VIDEO_CAPTION_TYPE_CEA708_RAW:
|
|
|
|
schedule_cea708_raw (self, map.data, map.size, pts, duration);
|
|
|
|
break;
|
|
|
|
case GST_VIDEO_CAPTION_TYPE_CEA608_S334_1A:
|
|
|
|
schedule_cea608_s334_1a (self, map.data, map.size, pts, duration);
|
|
|
|
break;
|
|
|
|
case GST_VIDEO_CAPTION_TYPE_CEA608_RAW:
|
|
|
|
schedule_cea608_raw (self, map.data, map.size, caption_buf);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
gst_buffer_unmap (caption_buf, &map);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
2022-01-05 21:48:31 +00:00
|
|
|
dequeue_caption_one_field (GstCCCombiner * self, const GstVideoTimeCode * tc,
|
|
|
|
guint field, gboolean drain)
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
{
|
|
|
|
CaptionQueueItem *scheduled;
|
|
|
|
CaptionData caption_data;
|
|
|
|
|
|
|
|
if ((scheduled = gst_queue_array_pop_head_struct (self->scheduled[field]))) {
|
|
|
|
caption_data.buffer = scheduled->buffer;
|
|
|
|
caption_data.caption_type = self->caption_type;
|
|
|
|
g_array_append_val (self->current_frame_captions, caption_data);
|
2022-02-01 22:46:43 +00:00
|
|
|
} else if (!drain && self->output_padding) {
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
caption_data.caption_type = self->caption_type;
|
|
|
|
caption_data.buffer = make_padding (self, tc, field);
|
|
|
|
g_array_append_val (self->current_frame_captions, caption_data);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2022-01-05 21:48:31 +00:00
|
|
|
static void
|
|
|
|
dequeue_caption_both_fields (GstCCCombiner * self, const GstVideoTimeCode * tc,
|
|
|
|
gboolean drain)
|
|
|
|
{
|
|
|
|
CaptionQueueItem *field0_scheduled, *field1_scheduled;
|
2022-02-01 22:46:43 +00:00
|
|
|
GstBuffer *field0_buffer = NULL, *field1_buffer = NULL;
|
2022-01-05 21:48:31 +00:00
|
|
|
CaptionData caption_data;
|
|
|
|
|
|
|
|
field0_scheduled = gst_queue_array_pop_head_struct (self->scheduled[0]);
|
|
|
|
field1_scheduled = gst_queue_array_pop_head_struct (self->scheduled[1]);
|
|
|
|
|
|
|
|
if (drain && !field0_scheduled && !field1_scheduled) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (field0_scheduled) {
|
|
|
|
field0_buffer = field0_scheduled->buffer;
|
2022-02-01 22:46:43 +00:00
|
|
|
} else if (self->output_padding) {
|
2022-01-05 21:48:31 +00:00
|
|
|
field0_buffer = make_padding (self, tc, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (field1_scheduled) {
|
|
|
|
field1_buffer = field1_scheduled->buffer;
|
2022-02-01 22:46:43 +00:00
|
|
|
} else if (self->output_padding) {
|
2022-01-05 21:48:31 +00:00
|
|
|
field1_buffer = make_padding (self, tc, 1);
|
|
|
|
}
|
|
|
|
|
2022-02-01 22:46:43 +00:00
|
|
|
if (field0_buffer || field1_buffer) {
|
|
|
|
if (field0_buffer && field1_buffer) {
|
|
|
|
caption_data.buffer = gst_buffer_append (field0_buffer, field1_buffer);
|
|
|
|
} else if (field0_buffer) {
|
|
|
|
caption_data.buffer = field0_buffer;
|
|
|
|
} else if (field1_buffer) {
|
|
|
|
caption_data.buffer = field1_buffer;
|
|
|
|
} else {
|
|
|
|
g_assert_not_reached ();
|
|
|
|
}
|
|
|
|
|
|
|
|
caption_data.caption_type = self->caption_type;
|
2022-01-05 21:48:31 +00:00
|
|
|
|
2022-02-01 22:46:43 +00:00
|
|
|
g_array_append_val (self->current_frame_captions, caption_data);
|
|
|
|
}
|
2022-01-05 21:48:31 +00:00
|
|
|
}
|
|
|
|
|
2018-11-05 14:24:58 +00:00
|
|
|
static GstFlowReturn
|
2019-03-13 12:43:54 +00:00
|
|
|
gst_cc_combiner_collect_captions (GstCCCombiner * self, gboolean timeout)
|
2018-11-05 14:24:58 +00:00
|
|
|
{
|
2019-03-13 12:43:54 +00:00
|
|
|
GstAggregatorPad *src_pad =
|
|
|
|
GST_AGGREGATOR_PAD (GST_AGGREGATOR_SRC_PAD (self));
|
2018-11-05 14:24:58 +00:00
|
|
|
GstAggregatorPad *caption_pad;
|
|
|
|
GstBuffer *video_buf;
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
GstVideoTimeCodeMeta *tc_meta;
|
|
|
|
GstVideoTimeCode *tc = NULL;
|
2021-10-25 23:09:58 +00:00
|
|
|
gboolean caption_pad_is_eos = FALSE;
|
2018-11-05 14:24:58 +00:00
|
|
|
|
|
|
|
g_assert (self->current_video_buffer != NULL);
|
|
|
|
|
|
|
|
caption_pad =
|
|
|
|
GST_AGGREGATOR_PAD_CAST (gst_element_get_static_pad (GST_ELEMENT_CAST
|
|
|
|
(self), "caption"));
|
|
|
|
/* No caption pad, forward buffer directly */
|
|
|
|
if (!caption_pad) {
|
|
|
|
GST_LOG_OBJECT (self, "No caption pad, passing through video");
|
|
|
|
video_buf = self->current_video_buffer;
|
2020-08-05 14:57:15 +00:00
|
|
|
gst_aggregator_selected_samples (GST_AGGREGATOR_CAST (self),
|
|
|
|
GST_BUFFER_PTS (video_buf), GST_BUFFER_DTS (video_buf),
|
2020-08-07 06:37:20 +00:00
|
|
|
GST_BUFFER_DURATION (video_buf), NULL);
|
2021-10-26 13:58:26 +00:00
|
|
|
self->current_video_buffer = NULL;
|
2019-03-13 12:43:54 +00:00
|
|
|
goto done;
|
2018-11-05 14:24:58 +00:00
|
|
|
}
|
|
|
|
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
tc_meta = gst_buffer_get_video_time_code_meta (self->current_video_buffer);
|
|
|
|
|
|
|
|
if (tc_meta) {
|
|
|
|
tc = &tc_meta->tc;
|
|
|
|
}
|
|
|
|
|
2018-11-05 14:24:58 +00:00
|
|
|
GST_LOG_OBJECT (self, "Trying to collect captions for queued video buffer");
|
|
|
|
do {
|
|
|
|
GstBuffer *caption_buf;
|
|
|
|
GstClockTime caption_time;
|
|
|
|
CaptionData caption_data;
|
|
|
|
|
|
|
|
caption_buf = gst_aggregator_pad_peek_buffer (caption_pad);
|
|
|
|
if (!caption_buf) {
|
|
|
|
if (gst_aggregator_pad_is_eos (caption_pad)) {
|
|
|
|
GST_DEBUG_OBJECT (self, "Caption pad is EOS, we're done");
|
2021-10-25 23:09:58 +00:00
|
|
|
|
|
|
|
caption_pad_is_eos = TRUE;
|
2018-11-05 14:24:58 +00:00
|
|
|
break;
|
2019-03-13 12:43:54 +00:00
|
|
|
} else if (!timeout) {
|
2018-11-05 14:24:58 +00:00
|
|
|
GST_DEBUG_OBJECT (self, "Need more caption data");
|
|
|
|
gst_object_unref (caption_pad);
|
|
|
|
return GST_FLOW_NEED_DATA;
|
2019-03-13 12:43:54 +00:00
|
|
|
} else {
|
|
|
|
GST_DEBUG_OBJECT (self, "No caption data on timeout");
|
|
|
|
break;
|
2018-11-05 14:24:58 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
caption_time = GST_BUFFER_PTS (caption_buf);
|
|
|
|
if (!GST_CLOCK_TIME_IS_VALID (caption_time)) {
|
|
|
|
GST_ERROR_OBJECT (self, "Caption buffer without PTS");
|
|
|
|
|
|
|
|
gst_buffer_unref (caption_buf);
|
|
|
|
gst_object_unref (caption_pad);
|
|
|
|
|
|
|
|
return GST_FLOW_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
caption_time =
|
|
|
|
gst_segment_to_running_time (&caption_pad->segment, GST_FORMAT_TIME,
|
|
|
|
caption_time);
|
|
|
|
|
|
|
|
if (!GST_CLOCK_TIME_IS_VALID (caption_time)) {
|
|
|
|
GST_DEBUG_OBJECT (self, "Caption buffer outside segment, dropping");
|
|
|
|
|
|
|
|
gst_aggregator_pad_drop_buffer (caption_pad);
|
|
|
|
gst_buffer_unref (caption_buf);
|
|
|
|
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
2020-04-22 14:51:08 +00:00
|
|
|
if (gst_buffer_get_size (caption_buf) == 0 &&
|
|
|
|
GST_BUFFER_FLAG_IS_SET (caption_buf, GST_BUFFER_FLAG_GAP)) {
|
|
|
|
/* This is a gap, we can go ahead. We only consume it once its end point
|
|
|
|
* is behind the current video running time. Important to note that
|
|
|
|
* we can't deal with gaps with no duration (-1)
|
|
|
|
*/
|
|
|
|
if (!GST_CLOCK_TIME_IS_VALID (GST_BUFFER_DURATION (caption_buf))) {
|
|
|
|
GST_ERROR_OBJECT (self, "GAP buffer without a duration");
|
|
|
|
|
|
|
|
gst_buffer_unref (caption_buf);
|
|
|
|
gst_object_unref (caption_pad);
|
|
|
|
|
|
|
|
return GST_FLOW_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
gst_buffer_unref (caption_buf);
|
|
|
|
|
|
|
|
if (caption_time + GST_BUFFER_DURATION (caption_buf) <
|
|
|
|
self->current_video_running_time_end) {
|
|
|
|
gst_aggregator_pad_drop_buffer (caption_pad);
|
|
|
|
continue;
|
|
|
|
} else {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-11-05 14:24:58 +00:00
|
|
|
/* Collected all caption buffers for this video buffer */
|
|
|
|
if (caption_time >= self->current_video_running_time_end) {
|
|
|
|
gst_buffer_unref (caption_buf);
|
|
|
|
break;
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
} else if (!self->schedule) {
|
|
|
|
if (GST_CLOCK_TIME_IS_VALID (self->previous_video_running_time_end)) {
|
|
|
|
if (caption_time < self->previous_video_running_time_end) {
|
|
|
|
GST_WARNING_OBJECT (self,
|
|
|
|
"Caption buffer before end of last video frame, dropping");
|
|
|
|
|
|
|
|
gst_aggregator_pad_drop_buffer (caption_pad);
|
|
|
|
gst_buffer_unref (caption_buf);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
} else if (caption_time < self->current_video_running_time) {
|
2020-04-22 14:53:00 +00:00
|
|
|
GST_WARNING_OBJECT (self,
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
"Caption buffer before current video frame, dropping");
|
2020-04-22 14:53:00 +00:00
|
|
|
|
|
|
|
gst_aggregator_pad_drop_buffer (caption_pad);
|
|
|
|
gst_buffer_unref (caption_buf);
|
|
|
|
continue;
|
|
|
|
}
|
2018-11-05 14:24:58 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* This caption buffer has to be collected */
|
|
|
|
GST_LOG_OBJECT (self,
|
|
|
|
"Collecting caption buffer %p %" GST_TIME_FORMAT " for video buffer %p",
|
|
|
|
caption_buf, GST_TIME_ARGS (caption_time), self->current_video_buffer);
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
|
|
|
|
caption_data.caption_type = self->caption_type;
|
|
|
|
|
2018-11-05 14:24:58 +00:00
|
|
|
gst_aggregator_pad_drop_buffer (caption_pad);
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
|
|
|
|
if (!self->schedule) {
|
|
|
|
caption_data.buffer = caption_buf;
|
|
|
|
g_array_append_val (self->current_frame_captions, caption_data);
|
|
|
|
} else {
|
|
|
|
schedule_caption (self, caption_buf, tc);
|
|
|
|
gst_buffer_unref (caption_buf);
|
|
|
|
}
|
2018-11-05 14:24:58 +00:00
|
|
|
} while (TRUE);
|
|
|
|
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
/* FIXME pad correctly according to fps */
|
|
|
|
if (self->schedule) {
|
|
|
|
g_assert (self->current_frame_captions->len == 0);
|
|
|
|
|
|
|
|
switch (self->caption_type) {
|
|
|
|
case GST_VIDEO_CAPTION_TYPE_CEA708_CDP:
|
|
|
|
{
|
|
|
|
/* Only relevant in alternate and mixed mode, no need to look at the caps */
|
|
|
|
if (GST_BUFFER_FLAG_IS_SET (self->current_video_buffer,
|
|
|
|
GST_VIDEO_BUFFER_FLAG_INTERLACED)) {
|
2021-07-15 19:10:14 +00:00
|
|
|
if (!GST_VIDEO_BUFFER_IS_BOTTOM_FIELD (self->current_video_buffer)) {
|
2022-01-05 21:48:31 +00:00
|
|
|
dequeue_caption_one_field (self, tc, 0, caption_pad_is_eos);
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
}
|
|
|
|
} else {
|
2022-01-05 21:48:31 +00:00
|
|
|
dequeue_caption_one_field (self, tc, 0, caption_pad_is_eos);
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case GST_VIDEO_CAPTION_TYPE_CEA708_RAW:
|
|
|
|
case GST_VIDEO_CAPTION_TYPE_CEA608_S334_1A:
|
|
|
|
{
|
|
|
|
if (self->progressive) {
|
2022-01-05 21:48:31 +00:00
|
|
|
dequeue_caption_one_field (self, tc, 0, caption_pad_is_eos);
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
} else if (GST_BUFFER_FLAG_IS_SET (self->current_video_buffer,
|
2021-07-15 19:10:14 +00:00
|
|
|
GST_VIDEO_BUFFER_FLAG_INTERLACED) &&
|
|
|
|
GST_BUFFER_FLAG_IS_SET (self->current_video_buffer,
|
|
|
|
GST_VIDEO_BUFFER_FLAG_ONEFIELD)) {
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
if (GST_VIDEO_BUFFER_IS_TOP_FIELD (self->current_video_buffer)) {
|
2022-01-05 21:48:31 +00:00
|
|
|
dequeue_caption_one_field (self, tc, 0, caption_pad_is_eos);
|
2021-07-15 19:10:14 +00:00
|
|
|
} else {
|
2022-01-05 21:48:31 +00:00
|
|
|
dequeue_caption_one_field (self, tc, 1, caption_pad_is_eos);
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
}
|
|
|
|
} else {
|
2022-01-05 21:48:31 +00:00
|
|
|
dequeue_caption_both_fields (self, tc, caption_pad_is_eos);
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case GST_VIDEO_CAPTION_TYPE_CEA608_RAW:
|
|
|
|
{
|
|
|
|
if (self->progressive) {
|
2022-01-05 21:48:31 +00:00
|
|
|
dequeue_caption_one_field (self, tc, 0, caption_pad_is_eos);
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
} else if (GST_BUFFER_FLAG_IS_SET (self->current_video_buffer,
|
|
|
|
GST_VIDEO_BUFFER_FLAG_INTERLACED)) {
|
2021-07-15 19:10:14 +00:00
|
|
|
if (!GST_VIDEO_BUFFER_IS_BOTTOM_FIELD (self->current_video_buffer)) {
|
2022-01-05 21:48:31 +00:00
|
|
|
dequeue_caption_one_field (self, tc, 0, caption_pad_is_eos);
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
}
|
|
|
|
} else {
|
2022-01-05 21:48:31 +00:00
|
|
|
dequeue_caption_one_field (self, tc, 0, caption_pad_is_eos);
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2020-08-05 14:57:15 +00:00
|
|
|
gst_aggregator_selected_samples (GST_AGGREGATOR_CAST (self),
|
|
|
|
GST_BUFFER_PTS (self->current_video_buffer),
|
|
|
|
GST_BUFFER_DTS (self->current_video_buffer),
|
2020-08-07 06:37:20 +00:00
|
|
|
GST_BUFFER_DURATION (self->current_video_buffer), NULL);
|
2020-07-01 01:59:56 +00:00
|
|
|
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
GST_LOG_OBJECT (self, "Attaching %u captions to buffer %p",
|
|
|
|
self->current_frame_captions->len, self->current_video_buffer);
|
|
|
|
|
2018-11-05 14:24:58 +00:00
|
|
|
if (self->current_frame_captions->len > 0) {
|
|
|
|
guint i;
|
|
|
|
|
|
|
|
video_buf = gst_buffer_make_writable (self->current_video_buffer);
|
|
|
|
self->current_video_buffer = NULL;
|
|
|
|
|
|
|
|
for (i = 0; i < self->current_frame_captions->len; i++) {
|
|
|
|
CaptionData *caption_data =
|
|
|
|
&g_array_index (self->current_frame_captions, CaptionData, i);
|
|
|
|
GstMapInfo map;
|
|
|
|
|
|
|
|
gst_buffer_map (caption_data->buffer, &map, GST_MAP_READ);
|
|
|
|
gst_buffer_add_video_caption_meta (video_buf, caption_data->caption_type,
|
|
|
|
map.data, map.size);
|
|
|
|
gst_buffer_unmap (caption_data->buffer, &map);
|
|
|
|
}
|
|
|
|
|
|
|
|
g_array_set_size (self->current_frame_captions, 0);
|
|
|
|
} else {
|
|
|
|
GST_LOG_OBJECT (self, "No captions for buffer %p",
|
|
|
|
self->current_video_buffer);
|
|
|
|
video_buf = self->current_video_buffer;
|
|
|
|
self->current_video_buffer = NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
gst_object_unref (caption_pad);
|
|
|
|
|
2019-03-13 12:43:54 +00:00
|
|
|
done:
|
|
|
|
src_pad->segment.position =
|
|
|
|
GST_BUFFER_PTS (video_buf) + GST_BUFFER_DURATION (video_buf);
|
|
|
|
|
2018-11-05 14:24:58 +00:00
|
|
|
return gst_aggregator_finish_buffer (GST_AGGREGATOR_CAST (self), video_buf);
|
|
|
|
}
|
|
|
|
|
|
|
|
static GstFlowReturn
|
|
|
|
gst_cc_combiner_aggregate (GstAggregator * aggregator, gboolean timeout)
|
|
|
|
{
|
|
|
|
GstCCCombiner *self = GST_CCCOMBINER (aggregator);
|
|
|
|
GstFlowReturn flow_ret = GST_FLOW_OK;
|
|
|
|
|
|
|
|
/* If we have no current video buffer, queue one. If we have one but
|
|
|
|
* its end running time is not known yet, try to determine it from the
|
|
|
|
* next video buffer */
|
|
|
|
if (!self->current_video_buffer
|
|
|
|
|| !GST_CLOCK_TIME_IS_VALID (self->current_video_running_time_end)) {
|
|
|
|
GstAggregatorPad *video_pad;
|
|
|
|
GstClockTime video_start;
|
|
|
|
GstBuffer *video_buf;
|
|
|
|
|
|
|
|
video_pad =
|
|
|
|
GST_AGGREGATOR_PAD_CAST (gst_element_get_static_pad (GST_ELEMENT_CAST
|
|
|
|
(aggregator), "sink"));
|
|
|
|
video_buf = gst_aggregator_pad_peek_buffer (video_pad);
|
|
|
|
if (!video_buf) {
|
|
|
|
if (gst_aggregator_pad_is_eos (video_pad)) {
|
|
|
|
GST_DEBUG_OBJECT (aggregator, "Video pad is EOS, we're done");
|
|
|
|
|
|
|
|
/* Assume that this buffer ends where it started +50ms (25fps) and handle it */
|
|
|
|
if (self->current_video_buffer) {
|
|
|
|
self->current_video_running_time_end =
|
|
|
|
self->current_video_running_time + 50 * GST_MSECOND;
|
2019-03-13 12:43:54 +00:00
|
|
|
flow_ret = gst_cc_combiner_collect_captions (self, timeout);
|
2018-11-05 14:24:58 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* If we collected all captions for the remaining video frame we're
|
|
|
|
* done, otherwise get called another time and go directly into the
|
|
|
|
* outer branch for finishing the current video frame */
|
|
|
|
if (flow_ret == GST_FLOW_NEED_DATA)
|
|
|
|
flow_ret = GST_FLOW_OK;
|
|
|
|
else
|
|
|
|
flow_ret = GST_FLOW_EOS;
|
|
|
|
} else {
|
|
|
|
flow_ret = GST_FLOW_OK;
|
|
|
|
}
|
|
|
|
|
|
|
|
gst_object_unref (video_pad);
|
|
|
|
return flow_ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
video_start = GST_BUFFER_PTS (video_buf);
|
|
|
|
if (!GST_CLOCK_TIME_IS_VALID (video_start)) {
|
|
|
|
gst_buffer_unref (video_buf);
|
|
|
|
gst_object_unref (video_pad);
|
|
|
|
|
|
|
|
GST_ERROR_OBJECT (aggregator, "Video buffer without PTS");
|
|
|
|
|
|
|
|
return GST_FLOW_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
video_start =
|
|
|
|
gst_segment_to_running_time (&video_pad->segment, GST_FORMAT_TIME,
|
|
|
|
video_start);
|
|
|
|
if (!GST_CLOCK_TIME_IS_VALID (video_start)) {
|
|
|
|
GST_DEBUG_OBJECT (aggregator, "Buffer outside segment, dropping");
|
|
|
|
gst_aggregator_pad_drop_buffer (video_pad);
|
|
|
|
gst_buffer_unref (video_buf);
|
|
|
|
gst_object_unref (video_pad);
|
|
|
|
return GST_FLOW_OK;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (self->current_video_buffer) {
|
|
|
|
/* If we already have a video buffer just update the current end running
|
|
|
|
* time accordingly. That's what was missing and why we got here */
|
|
|
|
self->current_video_running_time_end = video_start;
|
|
|
|
gst_buffer_unref (video_buf);
|
|
|
|
GST_LOG_OBJECT (self,
|
|
|
|
"Determined end timestamp for video buffer: %p %" GST_TIME_FORMAT
|
|
|
|
" - %" GST_TIME_FORMAT, self->current_video_buffer,
|
|
|
|
GST_TIME_ARGS (self->current_video_running_time),
|
|
|
|
GST_TIME_ARGS (self->current_video_running_time_end));
|
|
|
|
} else {
|
|
|
|
/* Otherwise we had no buffer queued currently. Let's do that now
|
|
|
|
* so that we can collect captions for it */
|
|
|
|
gst_buffer_replace (&self->current_video_buffer, video_buf);
|
|
|
|
self->current_video_running_time = video_start;
|
|
|
|
gst_aggregator_pad_drop_buffer (video_pad);
|
|
|
|
gst_buffer_unref (video_buf);
|
|
|
|
|
|
|
|
if (GST_BUFFER_DURATION_IS_VALID (video_buf)) {
|
|
|
|
GstClockTime end_time =
|
|
|
|
GST_BUFFER_PTS (video_buf) + GST_BUFFER_DURATION (video_buf);
|
|
|
|
if (video_pad->segment.stop != -1 && end_time > video_pad->segment.stop)
|
|
|
|
end_time = video_pad->segment.stop;
|
|
|
|
self->current_video_running_time_end =
|
|
|
|
gst_segment_to_running_time (&video_pad->segment, GST_FORMAT_TIME,
|
|
|
|
end_time);
|
|
|
|
} else if (self->video_fps_n != 0 && self->video_fps_d != 0) {
|
|
|
|
GstClockTime end_time =
|
|
|
|
GST_BUFFER_PTS (video_buf) + gst_util_uint64_scale_int (GST_SECOND,
|
|
|
|
self->video_fps_d, self->video_fps_n);
|
|
|
|
if (video_pad->segment.stop != -1 && end_time > video_pad->segment.stop)
|
|
|
|
end_time = video_pad->segment.stop;
|
|
|
|
self->current_video_running_time_end =
|
|
|
|
gst_segment_to_running_time (&video_pad->segment, GST_FORMAT_TIME,
|
|
|
|
end_time);
|
|
|
|
} else {
|
|
|
|
self->current_video_running_time_end = GST_CLOCK_TIME_NONE;
|
|
|
|
}
|
|
|
|
|
|
|
|
GST_LOG_OBJECT (self,
|
|
|
|
"Queued new video buffer: %p %" GST_TIME_FORMAT " - %"
|
|
|
|
GST_TIME_FORMAT, self->current_video_buffer,
|
|
|
|
GST_TIME_ARGS (self->current_video_running_time),
|
|
|
|
GST_TIME_ARGS (self->current_video_running_time_end));
|
|
|
|
}
|
|
|
|
|
|
|
|
gst_object_unref (video_pad);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* At this point we have a video buffer queued and can start collecting
|
|
|
|
* caption buffers for it */
|
|
|
|
g_assert (self->current_video_buffer != NULL);
|
|
|
|
g_assert (GST_CLOCK_TIME_IS_VALID (self->current_video_running_time));
|
|
|
|
g_assert (GST_CLOCK_TIME_IS_VALID (self->current_video_running_time_end));
|
|
|
|
|
2019-03-13 12:43:54 +00:00
|
|
|
flow_ret = gst_cc_combiner_collect_captions (self, timeout);
|
2018-11-05 14:24:58 +00:00
|
|
|
|
|
|
|
/* Only if we collected all captions we replace the current video buffer
|
|
|
|
* with NULL and continue with the next one on the next call */
|
|
|
|
if (flow_ret == GST_FLOW_NEED_DATA) {
|
|
|
|
flow_ret = GST_FLOW_OK;
|
|
|
|
} else {
|
|
|
|
gst_buffer_replace (&self->current_video_buffer, NULL);
|
2020-04-22 14:53:00 +00:00
|
|
|
self->previous_video_running_time_end =
|
|
|
|
self->current_video_running_time_end;
|
2018-11-05 14:24:58 +00:00
|
|
|
self->current_video_running_time = self->current_video_running_time_end =
|
|
|
|
GST_CLOCK_TIME_NONE;
|
|
|
|
}
|
|
|
|
|
|
|
|
return flow_ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static gboolean
|
|
|
|
gst_cc_combiner_sink_event (GstAggregator * aggregator,
|
|
|
|
GstAggregatorPad * agg_pad, GstEvent * event)
|
|
|
|
{
|
|
|
|
GstCCCombiner *self = GST_CCCOMBINER (aggregator);
|
|
|
|
|
|
|
|
switch (GST_EVENT_TYPE (event)) {
|
|
|
|
case GST_EVENT_CAPS:{
|
|
|
|
GstCaps *caps;
|
|
|
|
GstStructure *s;
|
|
|
|
|
|
|
|
gst_event_parse_caps (event, &caps);
|
|
|
|
s = gst_caps_get_structure (caps, 0);
|
|
|
|
|
|
|
|
if (strcmp (GST_OBJECT_NAME (agg_pad), "caption") == 0) {
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
GstVideoCaptionType caption_type =
|
|
|
|
gst_video_caption_type_from_caps (caps);
|
|
|
|
|
|
|
|
if (self->caption_type != GST_VIDEO_CAPTION_TYPE_UNKNOWN &&
|
|
|
|
caption_type != self->caption_type) {
|
|
|
|
GST_ERROR_OBJECT (self, "Changing caption type is not allowed");
|
|
|
|
|
|
|
|
GST_ELEMENT_ERROR (self, CORE, NEGOTIATION, (NULL),
|
|
|
|
("Changing caption type is not allowed"));
|
|
|
|
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
self->caption_type = caption_type;
|
2018-11-05 14:24:58 +00:00
|
|
|
} else {
|
2019-03-13 12:43:54 +00:00
|
|
|
gint fps_n, fps_d;
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
const gchar *interlace_mode;
|
2019-03-13 12:43:54 +00:00
|
|
|
|
|
|
|
fps_n = fps_d = 0;
|
|
|
|
|
|
|
|
gst_structure_get_fraction (s, "framerate", &fps_n, &fps_d);
|
|
|
|
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
interlace_mode = gst_structure_get_string (s, "interlace-mode");
|
|
|
|
|
|
|
|
self->progressive = !interlace_mode
|
|
|
|
|| !g_strcmp0 (interlace_mode, "progressive");
|
|
|
|
|
2019-03-13 12:43:54 +00:00
|
|
|
if (fps_n != self->video_fps_n || fps_d != self->video_fps_d) {
|
|
|
|
GstClockTime latency;
|
|
|
|
|
|
|
|
latency = gst_util_uint64_scale (GST_SECOND, fps_d, fps_n);
|
|
|
|
gst_aggregator_set_latency (aggregator, latency, latency);
|
|
|
|
}
|
|
|
|
|
|
|
|
self->video_fps_n = fps_n;
|
|
|
|
self->video_fps_d = fps_d;
|
2018-11-05 14:24:58 +00:00
|
|
|
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
self->cdp_fps_entry = cdp_fps_entry_from_fps (fps_n, fps_d);
|
|
|
|
|
2019-08-14 07:13:52 +00:00
|
|
|
gst_aggregator_set_src_caps (aggregator, caps);
|
2018-11-05 14:24:58 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
break;
|
2020-07-08 15:24:06 +00:00
|
|
|
}
|
|
|
|
case GST_EVENT_SEGMENT:{
|
|
|
|
if (strcmp (GST_OBJECT_NAME (agg_pad), "sink") == 0) {
|
|
|
|
const GstSegment *segment;
|
|
|
|
|
|
|
|
gst_event_parse_segment (event, &segment);
|
|
|
|
gst_aggregator_update_segment (aggregator, segment);
|
|
|
|
}
|
|
|
|
break;
|
2018-11-05 14:24:58 +00:00
|
|
|
}
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
return GST_AGGREGATOR_CLASS (parent_class)->sink_event (aggregator, agg_pad,
|
|
|
|
event);
|
|
|
|
}
|
|
|
|
|
|
|
|
static gboolean
|
|
|
|
gst_cc_combiner_stop (GstAggregator * aggregator)
|
|
|
|
{
|
|
|
|
GstCCCombiner *self = GST_CCCOMBINER (aggregator);
|
|
|
|
|
|
|
|
self->video_fps_n = self->video_fps_d = 0;
|
|
|
|
self->current_video_running_time = self->current_video_running_time_end =
|
2020-04-22 14:53:00 +00:00
|
|
|
self->previous_video_running_time_end = GST_CLOCK_TIME_NONE;
|
2018-11-05 14:24:58 +00:00
|
|
|
gst_buffer_replace (&self->current_video_buffer, NULL);
|
|
|
|
|
|
|
|
g_array_set_size (self->current_frame_captions, 0);
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
self->caption_type = GST_VIDEO_CAPTION_TYPE_UNKNOWN;
|
|
|
|
|
|
|
|
gst_queue_array_clear (self->scheduled[0]);
|
|
|
|
gst_queue_array_clear (self->scheduled[1]);
|
|
|
|
self->cdp_fps_entry = &null_fps_entry;
|
2018-11-05 14:24:58 +00:00
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
static GstFlowReturn
|
|
|
|
gst_cc_combiner_flush (GstAggregator * aggregator)
|
|
|
|
{
|
|
|
|
GstCCCombiner *self = GST_CCCOMBINER (aggregator);
|
2019-03-13 12:43:54 +00:00
|
|
|
GstAggregatorPad *src_pad =
|
|
|
|
GST_AGGREGATOR_PAD (GST_AGGREGATOR_SRC_PAD (aggregator));
|
2018-11-05 14:24:58 +00:00
|
|
|
|
|
|
|
self->current_video_running_time = self->current_video_running_time_end =
|
2020-04-22 14:53:00 +00:00
|
|
|
self->previous_video_running_time_end = GST_CLOCK_TIME_NONE;
|
2018-11-05 14:24:58 +00:00
|
|
|
gst_buffer_replace (&self->current_video_buffer, NULL);
|
|
|
|
|
|
|
|
g_array_set_size (self->current_frame_captions, 0);
|
|
|
|
|
2019-03-13 12:43:54 +00:00
|
|
|
src_pad->segment.position = GST_CLOCK_TIME_NONE;
|
|
|
|
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
self->cdp_hdr_sequence_cntr = 0;
|
|
|
|
gst_queue_array_clear (self->scheduled[0]);
|
|
|
|
gst_queue_array_clear (self->scheduled[1]);
|
|
|
|
|
2018-11-05 14:24:58 +00:00
|
|
|
return GST_FLOW_OK;
|
|
|
|
}
|
|
|
|
|
|
|
|
static GstAggregatorPad *
|
|
|
|
gst_cc_combiner_create_new_pad (GstAggregator * aggregator,
|
|
|
|
GstPadTemplate * templ, const gchar * req_name, const GstCaps * caps)
|
|
|
|
{
|
|
|
|
GstCCCombiner *self = GST_CCCOMBINER (aggregator);
|
|
|
|
GstAggregatorPad *agg_pad;
|
|
|
|
|
|
|
|
if (templ->direction != GST_PAD_SINK)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
if (templ->presence != GST_PAD_REQUEST)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
if (strcmp (templ->name_template, "caption") != 0)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
GST_OBJECT_LOCK (self);
|
|
|
|
agg_pad = g_object_new (GST_TYPE_AGGREGATOR_PAD,
|
|
|
|
"name", "caption", "direction", GST_PAD_SINK, "template", templ, NULL);
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
self->caption_type = GST_VIDEO_CAPTION_TYPE_UNKNOWN;
|
2018-11-05 14:24:58 +00:00
|
|
|
GST_OBJECT_UNLOCK (self);
|
|
|
|
|
|
|
|
return agg_pad;
|
|
|
|
}
|
|
|
|
|
2019-07-09 09:43:53 +00:00
|
|
|
static gboolean
|
|
|
|
gst_cc_combiner_src_query (GstAggregator * aggregator, GstQuery * query)
|
|
|
|
{
|
|
|
|
GstPad *video_sinkpad =
|
|
|
|
gst_element_get_static_pad (GST_ELEMENT_CAST (aggregator), "sink");
|
|
|
|
gboolean ret;
|
|
|
|
|
|
|
|
switch (GST_QUERY_TYPE (query)) {
|
|
|
|
case GST_QUERY_POSITION:
|
|
|
|
case GST_QUERY_DURATION:
|
|
|
|
case GST_QUERY_URI:
|
|
|
|
case GST_QUERY_CAPS:
|
|
|
|
case GST_QUERY_ALLOCATION:
|
|
|
|
ret = gst_pad_peer_query (video_sinkpad, query);
|
|
|
|
break;
|
|
|
|
case GST_QUERY_ACCEPT_CAPS:{
|
|
|
|
GstCaps *caps;
|
|
|
|
GstCaps *templ = gst_static_pad_template_get_caps (&srctemplate);
|
|
|
|
|
|
|
|
gst_query_parse_accept_caps (query, &caps);
|
|
|
|
gst_query_set_accept_caps_result (query, gst_caps_is_subset (caps,
|
|
|
|
templ));
|
2020-08-13 11:19:21 +00:00
|
|
|
gst_caps_unref (templ);
|
2019-07-09 09:43:53 +00:00
|
|
|
ret = TRUE;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
default:
|
|
|
|
ret = GST_AGGREGATOR_CLASS (parent_class)->src_query (aggregator, query);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
gst_object_unref (video_sinkpad);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static gboolean
|
|
|
|
gst_cc_combiner_sink_query (GstAggregator * aggregator,
|
|
|
|
GstAggregatorPad * aggpad, GstQuery * query)
|
|
|
|
{
|
|
|
|
GstPad *video_sinkpad =
|
|
|
|
gst_element_get_static_pad (GST_ELEMENT_CAST (aggregator), "sink");
|
|
|
|
GstPad *srcpad = GST_AGGREGATOR_SRC_PAD (aggregator);
|
|
|
|
|
|
|
|
gboolean ret;
|
|
|
|
|
|
|
|
switch (GST_QUERY_TYPE (query)) {
|
|
|
|
case GST_QUERY_POSITION:
|
|
|
|
case GST_QUERY_DURATION:
|
|
|
|
case GST_QUERY_URI:
|
|
|
|
case GST_QUERY_ALLOCATION:
|
|
|
|
if (GST_PAD_CAST (aggpad) == video_sinkpad) {
|
|
|
|
ret = gst_pad_peer_query (srcpad, query);
|
|
|
|
} else {
|
|
|
|
ret =
|
2020-08-13 11:19:21 +00:00
|
|
|
GST_AGGREGATOR_CLASS (parent_class)->sink_query (aggregator,
|
|
|
|
aggpad, query);
|
2019-07-09 09:43:53 +00:00
|
|
|
}
|
|
|
|
break;
|
|
|
|
case GST_QUERY_CAPS:
|
|
|
|
if (GST_PAD_CAST (aggpad) == video_sinkpad) {
|
|
|
|
ret = gst_pad_peer_query (srcpad, query);
|
|
|
|
} else {
|
|
|
|
GstCaps *filter;
|
|
|
|
GstCaps *templ = gst_static_pad_template_get_caps (&captiontemplate);
|
|
|
|
|
|
|
|
gst_query_parse_caps (query, &filter);
|
|
|
|
|
|
|
|
if (filter) {
|
|
|
|
GstCaps *caps =
|
|
|
|
gst_caps_intersect_full (filter, templ, GST_CAPS_INTERSECT_FIRST);
|
|
|
|
gst_query_set_caps_result (query, caps);
|
|
|
|
gst_caps_unref (caps);
|
|
|
|
} else {
|
|
|
|
gst_query_set_caps_result (query, templ);
|
|
|
|
}
|
|
|
|
gst_caps_unref (templ);
|
|
|
|
ret = TRUE;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case GST_QUERY_ACCEPT_CAPS:
|
|
|
|
if (GST_PAD_CAST (aggpad) == video_sinkpad) {
|
|
|
|
ret = gst_pad_peer_query (srcpad, query);
|
|
|
|
} else {
|
|
|
|
GstCaps *caps;
|
|
|
|
GstCaps *templ = gst_static_pad_template_get_caps (&captiontemplate);
|
|
|
|
|
|
|
|
gst_query_parse_accept_caps (query, &caps);
|
|
|
|
gst_query_set_accept_caps_result (query, gst_caps_is_subset (caps,
|
|
|
|
templ));
|
2020-08-13 11:19:21 +00:00
|
|
|
gst_caps_unref (templ);
|
2019-07-09 09:43:53 +00:00
|
|
|
ret = TRUE;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
default:
|
2020-08-13 11:19:21 +00:00
|
|
|
ret = GST_AGGREGATOR_CLASS (parent_class)->sink_query (aggregator,
|
|
|
|
aggpad, query);
|
2019-07-09 09:43:53 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
gst_object_unref (video_sinkpad);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2020-07-01 01:59:56 +00:00
|
|
|
static GstSample *
|
|
|
|
gst_cc_combiner_peek_next_sample (GstAggregator * agg,
|
|
|
|
GstAggregatorPad * aggpad)
|
|
|
|
{
|
|
|
|
GstAggregatorPad *caption_pad, *video_pad;
|
|
|
|
GstCCCombiner *self = GST_CCCOMBINER (agg);
|
|
|
|
GstSample *res = NULL;
|
|
|
|
|
|
|
|
caption_pad =
|
|
|
|
GST_AGGREGATOR_PAD_CAST (gst_element_get_static_pad (GST_ELEMENT_CAST
|
|
|
|
(self), "caption"));
|
|
|
|
video_pad =
|
|
|
|
GST_AGGREGATOR_PAD_CAST (gst_element_get_static_pad (GST_ELEMENT_CAST
|
|
|
|
(self), "sink"));
|
|
|
|
|
|
|
|
if (aggpad == caption_pad) {
|
|
|
|
if (self->current_frame_captions->len > 0) {
|
|
|
|
GstCaps *caps = gst_pad_get_current_caps (GST_PAD (aggpad));
|
|
|
|
GstBufferList *buflist = gst_buffer_list_new ();
|
|
|
|
guint i;
|
|
|
|
|
|
|
|
for (i = 0; i < self->current_frame_captions->len; i++) {
|
|
|
|
CaptionData *caption_data =
|
|
|
|
&g_array_index (self->current_frame_captions, CaptionData, i);
|
|
|
|
gst_buffer_list_add (buflist, gst_buffer_ref (caption_data->buffer));
|
|
|
|
}
|
|
|
|
|
|
|
|
res = gst_sample_new (NULL, caps, &aggpad->segment, NULL);
|
|
|
|
gst_caps_unref (caps);
|
|
|
|
|
|
|
|
gst_sample_set_buffer_list (res, buflist);
|
|
|
|
gst_buffer_list_unref (buflist);
|
|
|
|
}
|
|
|
|
} else if (aggpad == video_pad) {
|
|
|
|
if (self->current_video_buffer) {
|
|
|
|
GstCaps *caps = gst_pad_get_current_caps (GST_PAD (aggpad));
|
|
|
|
res = gst_sample_new (self->current_video_buffer,
|
|
|
|
caps, &aggpad->segment, NULL);
|
|
|
|
gst_caps_unref (caps);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (caption_pad)
|
|
|
|
gst_object_unref (caption_pad);
|
|
|
|
|
|
|
|
if (video_pad)
|
|
|
|
gst_object_unref (video_pad);
|
|
|
|
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
static GstStateChangeReturn
|
|
|
|
gst_cc_combiner_change_state (GstElement * element, GstStateChange transition)
|
|
|
|
{
|
|
|
|
GstCCCombiner *self = GST_CCCOMBINER (element);
|
|
|
|
|
|
|
|
switch (transition) {
|
|
|
|
case GST_STATE_CHANGE_READY_TO_PAUSED:
|
|
|
|
self->schedule = self->prop_schedule;
|
|
|
|
self->max_scheduled = self->prop_max_scheduled;
|
2022-02-01 22:46:43 +00:00
|
|
|
self->output_padding = self->prop_output_padding;
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
return GST_ELEMENT_CLASS (parent_class)->change_state (element, transition);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
gst_cc_combiner_set_property (GObject * object, guint prop_id,
|
|
|
|
const GValue * value, GParamSpec * pspec)
|
|
|
|
{
|
|
|
|
GstCCCombiner *self = GST_CCCOMBINER (object);
|
|
|
|
|
|
|
|
switch (prop_id) {
|
|
|
|
case PROP_SCHEDULE:
|
|
|
|
self->prop_schedule = g_value_get_boolean (value);
|
|
|
|
break;
|
|
|
|
case PROP_MAX_SCHEDULED:
|
|
|
|
self->prop_max_scheduled = g_value_get_uint (value);
|
|
|
|
break;
|
2022-02-01 22:46:43 +00:00
|
|
|
case PROP_OUTPUT_PADDING:
|
|
|
|
self->prop_output_padding = g_value_get_boolean (value);
|
|
|
|
break;
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
default:
|
|
|
|
G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
gst_cc_combiner_get_property (GObject * object, guint prop_id, GValue * value,
|
|
|
|
GParamSpec * pspec)
|
|
|
|
{
|
|
|
|
GstCCCombiner *self = GST_CCCOMBINER (object);
|
|
|
|
|
|
|
|
switch (prop_id) {
|
|
|
|
case PROP_SCHEDULE:
|
|
|
|
g_value_set_boolean (value, self->prop_schedule);
|
|
|
|
break;
|
|
|
|
case PROP_MAX_SCHEDULED:
|
|
|
|
g_value_set_uint (value, self->prop_max_scheduled);
|
|
|
|
break;
|
2022-02-01 22:46:43 +00:00
|
|
|
case PROP_OUTPUT_PADDING:
|
|
|
|
g_value_set_boolean (value, self->prop_output_padding);
|
|
|
|
break;
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
default:
|
|
|
|
G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-11-05 14:24:58 +00:00
|
|
|
static void
|
|
|
|
gst_cc_combiner_class_init (GstCCCombinerClass * klass)
|
|
|
|
{
|
|
|
|
GObjectClass *gobject_class;
|
|
|
|
GstElementClass *gstelement_class;
|
|
|
|
GstAggregatorClass *aggregator_class;
|
|
|
|
|
|
|
|
gobject_class = (GObjectClass *) klass;
|
|
|
|
gstelement_class = (GstElementClass *) klass;
|
|
|
|
aggregator_class = (GstAggregatorClass *) klass;
|
|
|
|
|
|
|
|
gobject_class->finalize = gst_cc_combiner_finalize;
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
gobject_class->set_property = gst_cc_combiner_set_property;
|
|
|
|
gobject_class->get_property = gst_cc_combiner_get_property;
|
2018-11-05 14:24:58 +00:00
|
|
|
|
|
|
|
gst_element_class_set_static_metadata (gstelement_class,
|
|
|
|
"Closed Caption Combiner",
|
|
|
|
"Filter",
|
|
|
|
"Combines GstVideoCaptionMeta with video input stream",
|
|
|
|
"Sebastian Dröge <sebastian@centricular.com>");
|
|
|
|
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
/**
|
|
|
|
* GstCCCombiner:schedule:
|
|
|
|
*
|
|
|
|
* Controls whether caption buffers should be smoothly scheduled
|
|
|
|
* in order to have exactly one per output video buffer.
|
|
|
|
*
|
|
|
|
* This can involve rewriting input captions, for example when the
|
|
|
|
* input is CDP sequence counters are rewritten, time codes are dropped
|
|
|
|
* and potentially re-injected if the input video frame had a time code
|
|
|
|
* meta.
|
|
|
|
*
|
|
|
|
* Caption buffers may also get split up in order to assign captions to
|
|
|
|
* the correct field when the input is interlaced.
|
|
|
|
*
|
|
|
|
* This can also imply that the input will drift from synchronization,
|
|
|
|
* when there isn't enough padding in the input stream to catch up. In
|
|
|
|
* that case the element will start dropping old caption buffers once
|
|
|
|
* the number of buffers in its internal queue reaches
|
|
|
|
* #GstCCCombiner:max-scheduled.
|
|
|
|
*
|
|
|
|
* When this is set to %FALSE, the behaviour of this element is essentially
|
|
|
|
* that of a funnel.
|
|
|
|
*
|
|
|
|
* Since: 1.20
|
|
|
|
*/
|
|
|
|
g_object_class_install_property (G_OBJECT_CLASS (klass),
|
|
|
|
PROP_SCHEDULE, g_param_spec_boolean ("schedule",
|
|
|
|
"Schedule",
|
|
|
|
"Schedule caption buffers so that exactly one is output per video frame",
|
2021-10-26 14:00:36 +00:00
|
|
|
DEFAULT_SCHEDULE,
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS |
|
|
|
|
GST_PARAM_MUTABLE_READY));
|
|
|
|
|
|
|
|
/**
|
|
|
|
* GstCCCombiner:max-scheduled:
|
|
|
|
*
|
|
|
|
* Controls the number of scheduled buffers after which the element
|
|
|
|
* will start dropping old buffers from its internal queues. See
|
|
|
|
* #GstCCCombiner:schedule.
|
|
|
|
*
|
|
|
|
* Since: 1.20
|
|
|
|
*/
|
|
|
|
g_object_class_install_property (G_OBJECT_CLASS (klass),
|
2021-04-19 10:06:23 +00:00
|
|
|
PROP_MAX_SCHEDULED, g_param_spec_uint ("max-scheduled",
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
"Max Scheduled",
|
|
|
|
"Maximum number of buffers to queue for scheduling", 0, G_MAXUINT,
|
|
|
|
DEFAULT_MAX_SCHEDULED,
|
|
|
|
G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS |
|
|
|
|
GST_PARAM_MUTABLE_READY));
|
|
|
|
|
2022-02-01 22:46:43 +00:00
|
|
|
/**
|
|
|
|
* GstCCCombiner:output-padding:
|
|
|
|
*
|
|
|
|
* When #GstCCCombiner:schedule is %TRUE, this property controls
|
|
|
|
* whether the output closed caption meta stream will be padded.
|
|
|
|
*
|
|
|
|
* Since: 1.22
|
|
|
|
*/
|
|
|
|
g_object_class_install_property (G_OBJECT_CLASS (klass),
|
|
|
|
PROP_OUTPUT_PADDING, g_param_spec_boolean ("output-padding",
|
|
|
|
"Output padding",
|
|
|
|
"Whether to output padding packets when schedule=true",
|
|
|
|
DEFAULT_OUTPUT_PADDING,
|
|
|
|
G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS |
|
|
|
|
GST_PARAM_MUTABLE_READY));
|
|
|
|
|
|
|
|
|
2018-11-05 14:24:58 +00:00
|
|
|
gst_element_class_add_static_pad_template_with_gtype (gstelement_class,
|
|
|
|
&sinktemplate, GST_TYPE_AGGREGATOR_PAD);
|
|
|
|
gst_element_class_add_static_pad_template_with_gtype (gstelement_class,
|
|
|
|
&srctemplate, GST_TYPE_AGGREGATOR_PAD);
|
|
|
|
gst_element_class_add_static_pad_template_with_gtype (gstelement_class,
|
|
|
|
&captiontemplate, GST_TYPE_AGGREGATOR_PAD);
|
|
|
|
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
gstelement_class->change_state =
|
|
|
|
GST_DEBUG_FUNCPTR (gst_cc_combiner_change_state);
|
|
|
|
|
2018-11-05 14:24:58 +00:00
|
|
|
aggregator_class->aggregate = gst_cc_combiner_aggregate;
|
|
|
|
aggregator_class->stop = gst_cc_combiner_stop;
|
|
|
|
aggregator_class->flush = gst_cc_combiner_flush;
|
|
|
|
aggregator_class->create_new_pad = gst_cc_combiner_create_new_pad;
|
|
|
|
aggregator_class->sink_event = gst_cc_combiner_sink_event;
|
2019-08-14 07:13:52 +00:00
|
|
|
aggregator_class->negotiate = NULL;
|
2019-03-13 12:43:54 +00:00
|
|
|
aggregator_class->get_next_time = gst_aggregator_simple_get_next_time;
|
2019-07-09 09:43:53 +00:00
|
|
|
aggregator_class->src_query = gst_cc_combiner_src_query;
|
|
|
|
aggregator_class->sink_query = gst_cc_combiner_sink_query;
|
2020-07-01 01:59:56 +00:00
|
|
|
aggregator_class->peek_next_sample = gst_cc_combiner_peek_next_sample;
|
2018-11-05 14:24:58 +00:00
|
|
|
|
|
|
|
GST_DEBUG_CATEGORY_INIT (gst_cc_combiner_debug, "cccombiner",
|
|
|
|
0, "Closed Caption combiner");
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
gst_cc_combiner_init (GstCCCombiner * self)
|
|
|
|
{
|
|
|
|
GstPadTemplate *templ;
|
|
|
|
GstAggregatorPad *agg_pad;
|
|
|
|
|
|
|
|
templ = gst_static_pad_template_get (&sinktemplate);
|
|
|
|
agg_pad = g_object_new (GST_TYPE_AGGREGATOR_PAD,
|
|
|
|
"name", "sink", "direction", GST_PAD_SINK, "template", templ, NULL);
|
|
|
|
gst_object_unref (templ);
|
|
|
|
gst_element_add_pad (GST_ELEMENT_CAST (self), GST_PAD_CAST (agg_pad));
|
|
|
|
|
|
|
|
self->current_frame_captions =
|
|
|
|
g_array_new (FALSE, FALSE, sizeof (CaptionData));
|
|
|
|
g_array_set_clear_func (self->current_frame_captions,
|
|
|
|
(GDestroyNotify) caption_data_clear);
|
|
|
|
|
|
|
|
self->current_video_running_time = self->current_video_running_time_end =
|
2020-04-22 14:53:00 +00:00
|
|
|
self->previous_video_running_time_end = GST_CLOCK_TIME_NONE;
|
2018-11-05 14:24:58 +00:00
|
|
|
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
self->caption_type = GST_VIDEO_CAPTION_TYPE_UNKNOWN;
|
|
|
|
|
|
|
|
self->prop_schedule = DEFAULT_SCHEDULE;
|
|
|
|
self->prop_max_scheduled = DEFAULT_MAX_SCHEDULED;
|
2022-02-01 22:46:43 +00:00
|
|
|
self->prop_output_padding = DEFAULT_OUTPUT_PADDING;
|
cccombiner: implement scheduling
Prior to that, cccombiner's behaviour was essentially that of
a funnel: it strictly looked at input timestamps to associate
together video and caption buffers.
This patch instead exposes a "schedule" property, with a default
of TRUE, to control whether caption buffers should be smoothly
scheduled, in order to have exactly one per output video buffer.
This can involve rewriting input captions, for example when the
input is CDP sequence counters are rewritten, time codes are dropped
and potentially re-injected if the input video frame had a time code
meta.
Caption buffers may also get split up in order to assign captions to
the correct field when the input is interlaced.
This can also imply that the input will drift from synchronization,
when there isn't enough padding in the input stream to catch up. In
that case the element will start dropping old caption buffers once
the number of buffers in its internal queue reaches a certain limit
(configurable).
The property is exposed so that existing users of cccombiner can
revert back to the original behaviour, but should eventually be
removed, as that behaviour was simply inadequate.
This commit also disallows changing the input caption type, as
this would needlessly complicate implementation, and removes
the corresponding test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2076>
2021-03-09 12:22:10 +00:00
|
|
|
self->scheduled[0] =
|
|
|
|
gst_queue_array_new_for_struct (sizeof (CaptionQueueItem), 0);
|
|
|
|
self->scheduled[1] =
|
|
|
|
gst_queue_array_new_for_struct (sizeof (CaptionQueueItem), 0);
|
|
|
|
gst_queue_array_set_clear_func (self->scheduled[0],
|
|
|
|
(GDestroyNotify) clear_scheduled);
|
|
|
|
gst_queue_array_set_clear_func (self->scheduled[1],
|
|
|
|
(GDestroyNotify) clear_scheduled);
|
|
|
|
self->cdp_hdr_sequence_cntr = 0;
|
|
|
|
self->cdp_fps_entry = &null_fps_entry;
|
2018-11-05 14:24:58 +00:00
|
|
|
}
|