gstreamer/subprojects/gst-plugins-bad/gst/videoparsers/gstav1parse.c

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

2324 lines
70 KiB
C
Raw Normal View History

/* GStreamer
* Copyright (C) 2020 He Junyan <junyan.he@intel.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.
*/
/*
* SECTION:element-av1parse
* @title: av1parse
* @short_description: An AV1 stream parse.
*
* The minimal unit should be the BYTE.
* There are four types of AV1 alignment in the AV1 stream.
*
* alignment: byte, obu, frame, tu
*
* 1. Aligned to byte. The basic and default one for input.
* 2. Aligned to obu(Open Bitstream Units).
* 3. Aligned to frame. The default one for output. This ensures that
* each buffer contains only one frame or frame header with the
* show_existing flag for the base or sub layer. It is useful for
* the decoder.
* 4. Aligned to tu(Temporal Unit). A temporal unit consists of all the
* OBUs that are associated with a specific, distinct time instant.
* When scalability is disabled, it contains just exact one showing
* frame(may contain several unshowing frames). When scalability is
* enabled, it contains frames depending on the layer number. It should
* begin with a temporal delimiter obu. It may be useful for mux/demux
* to index the data of some timestamp.
*
* The annex B define a special format for the temporal unit. The size of
* each temporal unit is extract out to the header of the buffer, and no
* size field inside the each obu. There are two stream formats:
*
* stream-format: obu-stream, annexb
*
* 1. obu-stream. The basic and default one.
* 2. annexb. A special stream of temporal unit. It also implies that the
* alignment should be TU.
*
* This AV1 parse implements the conversion between the alignments and the
* stream-formats. If the input and output have the same alignment and the
* same stream-format, it will check and bypass the data.
*
* ## Example launch line to generate annex B format AV1 stream:
* ```
* gst-launch-1.0 filesrc location=sample.av1 ! ivfparse ! av1parse ! \
* video/x-av1,alignment=\(string\)tu,stream-format=\(string\)annexb ! \
* filesink location=matroskamux ! filesink location=trans.mkv
* ```
*
* Since: 1.20
*/
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif
#include <gst/base/gstbitreader.h>
#include <gst/base/gstbitwriter.h>
#include <gst/codecparsers/gstav1parser.h>
#include <gst/video/video.h>
#include "gstvideoparserselements.h"
#include "gstav1parse.h"
#include <string.h>
#define GST_AV1_MAX_LEB_128_SIZE 8
GST_DEBUG_CATEGORY (av1_parse_debug);
#define GST_CAT_DEFAULT av1_parse_debug
/* We combine the stream format and the alignment
together. When stream format is annexb, the
alignment must be TU. */
typedef enum
{
GST_AV1_PARSE_ALIGN_ERROR = -1,
GST_AV1_PARSE_ALIGN_NONE = 0,
GST_AV1_PARSE_ALIGN_BYTE,
GST_AV1_PARSE_ALIGN_OBU,
GST_AV1_PARSE_ALIGN_FRAME,
GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT,
GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B,
} GstAV1ParseAligment;
struct _GstAV1Parse
{
GstBaseParse parent;
gint width;
gint height;
gint subsampling_x;
gint subsampling_y;
gboolean mono_chrome;
guint8 seq_level_idx;
guint8 seq_tier;
guint8 max_seq_level_idx;
guint8 max_seq_tier;
guint8 bit_depth;
gchar *colorimetry;
GstAV1Profile profile;
gint fps_n;
gint fps_d;
/* incaps framerate overwrites the AV1 time info */
gboolean has_input_fps;
GstAV1ParseAligment in_align;
gboolean detect_annex_b;
GstAV1ParseAligment align;
GstAV1Parser *parser;
GstAdapter *cache_out;
guint last_parsed_offset;
GstAdapter *frame_cache;
guint highest_spatial_id;
gint last_shown_frame_temporal_id;
gint last_shown_frame_spatial_id;
gboolean within_one_frame;
gboolean update_caps;
gboolean discont;
gboolean header;
gboolean keyframe;
gboolean show_frame;
av1parse: Correct the pts for frames and OBUs inside a TU. When the output alignment is smaller than the input alignment, for example, When the output alignment is "FRAME" and the parse is likely connecting to a decoder, the current PTS setting for AV1 frames inside a TU is not very correct. For example, a TU may begin with non-displayed frames and end with a displayed frame. The current way will assign the PTS to the first non-displayed frame, which is a decode-only frame and the PTS will be discarded in the video decoder. While the last displayed frame has invalid PTS, and so the video decoder needs to guess its PTS based on the frame rate and previous frame's PTS. This is not a decent and robust way. And more important, when the previous frames provide DTS, the video decoder will also guess the PTS based on the previous frames' DTS and trigger the warning like: gstvideodecoder.c:3147:gst_video_decoder_prepare_finish_frame: \ <vavp9dec0> decreasing timestame It sets the reordered_output and makes the decoder in free run mode. We should correct the PTS for a TU, let the non-displayed frames have no PTS while set the correct PTS to the displayed one. Also, when the AV1 stream has multi spatial layers, there are more than one displayed frames inside one TU with the same PTS. Note: If the input alignment is not TU aligned, we can not know the exact PTS of this TU, and so we just clear the PTS of the decode only frame and leave others unchanged. We also correct all the PTS if the output is OBU aligned. All their PTS and DTS are set to the input buffer's PTS. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3182>
2022-10-14 08:05:28 +00:00
GstClockTime buffer_pts;
GstClockTime buffer_dts;
GstClockTime buffer_duration;
};
static GstStaticPadTemplate sinktemplate = GST_STATIC_PAD_TEMPLATE ("sink",
GST_PAD_SINK,
GST_PAD_ALWAYS,
GST_STATIC_CAPS ("video/x-av1"));
static GstStaticPadTemplate srctemplate = GST_STATIC_PAD_TEMPLATE ("src",
GST_PAD_SRC,
GST_PAD_ALWAYS,
GST_STATIC_CAPS ("video/x-av1, parsed = (boolean) true, "
"stream-format=(string) { obu-stream, annexb }, "
"alignment=(string) { obu, tu, frame }"));
#define parent_class gst_av1_parse_parent_class
G_DEFINE_TYPE (GstAV1Parse, gst_av1_parse, GST_TYPE_BASE_PARSE);
GST_ELEMENT_REGISTER_DEFINE_WITH_CODE (av1parse, "av1parse", GST_RANK_SECONDARY,
GST_TYPE_AV1_PARSE, videoparsers_element_init (plugin));
static void
remove_fields (GstCaps * caps, gboolean all)
{
guint i, n;
n = gst_caps_get_size (caps);
for (i = 0; i < n; i++) {
GstStructure *s = gst_caps_get_structure (caps, i);
if (all) {
gst_structure_remove_field (s, "alignment");
gst_structure_remove_field (s, "stream-format");
}
gst_structure_remove_field (s, "parsed");
}
}
static const gchar *
_obu_name (GstAV1OBUType type)
{
switch (type) {
case GST_AV1_OBU_SEQUENCE_HEADER:
return "sequence header";
case GST_AV1_OBU_TEMPORAL_DELIMITER:
return "temporal delimiter";
case GST_AV1_OBU_FRAME_HEADER:
return "frame header";
case GST_AV1_OBU_TILE_GROUP:
return "tile group";
case GST_AV1_OBU_METADATA:
return "metadata";
case GST_AV1_OBU_FRAME:
return "frame";
case GST_AV1_OBU_REDUNDANT_FRAME_HEADER:
return "redundant frame header";
case GST_AV1_OBU_TILE_LIST:
return "tile list";
case GST_AV1_OBU_PADDING:
return "padding";
default:
return "unknown";
}
return NULL;
}
static guint32
_read_leb128 (guint8 * data, GstAV1ParserResult * retval, guint32 * comsumed)
{
guint8 leb128_byte = 0;
guint64 value = 0;
gint i;
gboolean result;
GstBitReader br;
guint32 cur_pos;
gst_bit_reader_init (&br, data, 8);
cur_pos = gst_bit_reader_get_pos (&br);
for (i = 0; i < 8; i++) {
leb128_byte = 0;
result = gst_bit_reader_get_bits_uint8 (&br, &leb128_byte, 8);
if (result == FALSE) {
*retval = GST_AV1_PARSER_BITSTREAM_ERROR;
return 0;
}
value |= (((gint) leb128_byte & 0x7f) << (i * 7));
if (!(leb128_byte & 0x80))
break;
}
*comsumed = (gst_bit_reader_get_pos (&br) - cur_pos) / 8;
/* check for bitstream conformance see chapter4.10.5 */
if (value < G_MAXUINT32) {
*retval = GST_AV1_PARSER_OK;
return (guint32) value;
} else {
GST_WARNING ("invalid leb128");
*retval = GST_AV1_PARSER_BITSTREAM_ERROR;
return 0;
}
}
static gsize
_leb_size_in_bytes (guint64 value)
{
gsize size = 0;
do {
++size;
} while ((value >>= 7) != 0);
return size;
}
static gboolean
_write_leb128 (guint8 * data, guint * len, guint64 value)
{
guint leb_size = _leb_size_in_bytes (value);
guint i;
if (value > G_MAXUINT32 || leb_size > GST_AV1_MAX_LEB_128_SIZE)
return FALSE;
for (i = 0; i < leb_size; ++i) {
guint8 byte = value & 0x7f;
value >>= 7;
/* Signal that more bytes follow. */
if (value != 0)
byte |= 0x80;
*(data + i) = byte;
}
*len = leb_size;
return TRUE;
}
static gboolean gst_av1_parse_start (GstBaseParse * parse);
static gboolean gst_av1_parse_stop (GstBaseParse * parse);
static GstFlowReturn gst_av1_parse_handle_frame (GstBaseParse * parse,
GstBaseParseFrame * frame, gint * skipsize);
static gboolean gst_av1_parse_set_sink_caps (GstBaseParse * parse,
GstCaps * caps);
static GstCaps *gst_av1_parse_get_sink_caps (GstBaseParse * parse,
GstCaps * filter);
av1parse: Correct the pts for frames and OBUs inside a TU. When the output alignment is smaller than the input alignment, for example, When the output alignment is "FRAME" and the parse is likely connecting to a decoder, the current PTS setting for AV1 frames inside a TU is not very correct. For example, a TU may begin with non-displayed frames and end with a displayed frame. The current way will assign the PTS to the first non-displayed frame, which is a decode-only frame and the PTS will be discarded in the video decoder. While the last displayed frame has invalid PTS, and so the video decoder needs to guess its PTS based on the frame rate and previous frame's PTS. This is not a decent and robust way. And more important, when the previous frames provide DTS, the video decoder will also guess the PTS based on the previous frames' DTS and trigger the warning like: gstvideodecoder.c:3147:gst_video_decoder_prepare_finish_frame: \ <vavp9dec0> decreasing timestame It sets the reordered_output and makes the decoder in free run mode. We should correct the PTS for a TU, let the non-displayed frames have no PTS while set the correct PTS to the displayed one. Also, when the AV1 stream has multi spatial layers, there are more than one displayed frames inside one TU with the same PTS. Note: If the input alignment is not TU aligned, we can not know the exact PTS of this TU, and so we just clear the PTS of the decode only frame and leave others unchanged. We also correct all the PTS if the output is OBU aligned. All their PTS and DTS are set to the input buffer's PTS. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3182>
2022-10-14 08:05:28 +00:00
static GstFlowReturn gst_av1_parse_pre_push_frame (GstBaseParse * parse,
GstBaseParseFrame * frame);
/* Clear the parse state related to data kind OBUs. */
static void
gst_av1_parse_reset_obu_data_state (GstAV1Parse * self)
{
self->last_shown_frame_temporal_id = -1;
self->last_shown_frame_spatial_id = -1;
self->within_one_frame = FALSE;
}
av1parse: Correct the pts for frames and OBUs inside a TU. When the output alignment is smaller than the input alignment, for example, When the output alignment is "FRAME" and the parse is likely connecting to a decoder, the current PTS setting for AV1 frames inside a TU is not very correct. For example, a TU may begin with non-displayed frames and end with a displayed frame. The current way will assign the PTS to the first non-displayed frame, which is a decode-only frame and the PTS will be discarded in the video decoder. While the last displayed frame has invalid PTS, and so the video decoder needs to guess its PTS based on the frame rate and previous frame's PTS. This is not a decent and robust way. And more important, when the previous frames provide DTS, the video decoder will also guess the PTS based on the previous frames' DTS and trigger the warning like: gstvideodecoder.c:3147:gst_video_decoder_prepare_finish_frame: \ <vavp9dec0> decreasing timestame It sets the reordered_output and makes the decoder in free run mode. We should correct the PTS for a TU, let the non-displayed frames have no PTS while set the correct PTS to the displayed one. Also, when the AV1 stream has multi spatial layers, there are more than one displayed frames inside one TU with the same PTS. Note: If the input alignment is not TU aligned, we can not know the exact PTS of this TU, and so we just clear the PTS of the decode only frame and leave others unchanged. We also correct all the PTS if the output is OBU aligned. All their PTS and DTS are set to the input buffer's PTS. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3182>
2022-10-14 08:05:28 +00:00
static void
gst_av1_parse_reset_tu_timestamp (GstAV1Parse * self)
{
self->buffer_pts = GST_CLOCK_TIME_NONE;
self->buffer_dts = GST_CLOCK_TIME_NONE;
self->buffer_duration = GST_CLOCK_TIME_NONE;
}
static void
gst_av1_parse_reset (GstAV1Parse * self)
{
self->width = 0;
self->height = 0;
self->subsampling_x = -1;
self->subsampling_y = -1;
self->mono_chrome = FALSE;
self->profile = GST_AV1_PROFILE_UNDEFINED;
self->seq_level_idx = GST_AV1_SEQ_LEVEL_MAX;
self->seq_tier = 0;
self->max_seq_level_idx = GST_AV1_SEQ_LEVEL_MAX;
self->max_seq_tier = 0;
self->bit_depth = 0;
self->align = GST_AV1_PARSE_ALIGN_NONE;
self->in_align = GST_AV1_PARSE_ALIGN_NONE;
self->detect_annex_b = FALSE;
self->discont = TRUE;
self->header = FALSE;
self->keyframe = FALSE;
self->show_frame = FALSE;
self->last_parsed_offset = 0;
self->highest_spatial_id = 0;
gst_av1_parse_reset_obu_data_state (self);
g_clear_pointer (&self->colorimetry, g_free);
g_clear_pointer (&self->parser, gst_av1_parser_free);
gst_adapter_clear (self->cache_out);
gst_adapter_clear (self->frame_cache);
av1parse: Correct the pts for frames and OBUs inside a TU. When the output alignment is smaller than the input alignment, for example, When the output alignment is "FRAME" and the parse is likely connecting to a decoder, the current PTS setting for AV1 frames inside a TU is not very correct. For example, a TU may begin with non-displayed frames and end with a displayed frame. The current way will assign the PTS to the first non-displayed frame, which is a decode-only frame and the PTS will be discarded in the video decoder. While the last displayed frame has invalid PTS, and so the video decoder needs to guess its PTS based on the frame rate and previous frame's PTS. This is not a decent and robust way. And more important, when the previous frames provide DTS, the video decoder will also guess the PTS based on the previous frames' DTS and trigger the warning like: gstvideodecoder.c:3147:gst_video_decoder_prepare_finish_frame: \ <vavp9dec0> decreasing timestame It sets the reordered_output and makes the decoder in free run mode. We should correct the PTS for a TU, let the non-displayed frames have no PTS while set the correct PTS to the displayed one. Also, when the AV1 stream has multi spatial layers, there are more than one displayed frames inside one TU with the same PTS. Note: If the input alignment is not TU aligned, we can not know the exact PTS of this TU, and so we just clear the PTS of the decode only frame and leave others unchanged. We also correct all the PTS if the output is OBU aligned. All their PTS and DTS are set to the input buffer's PTS. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3182>
2022-10-14 08:05:28 +00:00
gst_av1_parse_reset_tu_timestamp (self);
}
static void
gst_av1_parse_init (GstAV1Parse * self)
{
gst_base_parse_set_pts_interpolation (GST_BASE_PARSE (self), FALSE);
gst_base_parse_set_infer_ts (GST_BASE_PARSE (self), FALSE);
GST_PAD_SET_ACCEPT_INTERSECT (GST_BASE_PARSE_SINK_PAD (self));
GST_PAD_SET_ACCEPT_TEMPLATE (GST_BASE_PARSE_SINK_PAD (self));
self->cache_out = gst_adapter_new ();
self->frame_cache = gst_adapter_new ();
}
static void
gst_av1_parse_finalize (GObject * object)
{
GstAV1Parse *self = GST_AV1_PARSE (object);
gst_av1_parse_reset (self);
g_object_unref (self->cache_out);
g_object_unref (self->frame_cache);
G_OBJECT_CLASS (parent_class)->finalize (object);
}
static void
gst_av1_parse_class_init (GstAV1ParseClass * klass)
{
GObjectClass *gobject_class = (GObjectClass *) klass;
GstBaseParseClass *parse_class = GST_BASE_PARSE_CLASS (klass);
GstElementClass *element_class = GST_ELEMENT_CLASS (klass);
gobject_class->finalize = gst_av1_parse_finalize;
parse_class->start = GST_DEBUG_FUNCPTR (gst_av1_parse_start);
parse_class->stop = GST_DEBUG_FUNCPTR (gst_av1_parse_stop);
parse_class->handle_frame = GST_DEBUG_FUNCPTR (gst_av1_parse_handle_frame);
av1parse: Correct the pts for frames and OBUs inside a TU. When the output alignment is smaller than the input alignment, for example, When the output alignment is "FRAME" and the parse is likely connecting to a decoder, the current PTS setting for AV1 frames inside a TU is not very correct. For example, a TU may begin with non-displayed frames and end with a displayed frame. The current way will assign the PTS to the first non-displayed frame, which is a decode-only frame and the PTS will be discarded in the video decoder. While the last displayed frame has invalid PTS, and so the video decoder needs to guess its PTS based on the frame rate and previous frame's PTS. This is not a decent and robust way. And more important, when the previous frames provide DTS, the video decoder will also guess the PTS based on the previous frames' DTS and trigger the warning like: gstvideodecoder.c:3147:gst_video_decoder_prepare_finish_frame: \ <vavp9dec0> decreasing timestame It sets the reordered_output and makes the decoder in free run mode. We should correct the PTS for a TU, let the non-displayed frames have no PTS while set the correct PTS to the displayed one. Also, when the AV1 stream has multi spatial layers, there are more than one displayed frames inside one TU with the same PTS. Note: If the input alignment is not TU aligned, we can not know the exact PTS of this TU, and so we just clear the PTS of the decode only frame and leave others unchanged. We also correct all the PTS if the output is OBU aligned. All their PTS and DTS are set to the input buffer's PTS. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3182>
2022-10-14 08:05:28 +00:00
parse_class->pre_push_frame =
GST_DEBUG_FUNCPTR (gst_av1_parse_pre_push_frame);
parse_class->set_sink_caps = GST_DEBUG_FUNCPTR (gst_av1_parse_set_sink_caps);
parse_class->get_sink_caps = GST_DEBUG_FUNCPTR (gst_av1_parse_get_sink_caps);
gst_element_class_add_static_pad_template (element_class, &srctemplate);
gst_element_class_add_static_pad_template (element_class, &sinktemplate);
gst_element_class_set_static_metadata (element_class, "AV1 parser",
"Codec/Parser/Converter/Video",
"Parses AV1 streams", "He Junyan <junyan.he@intel.com>");
GST_DEBUG_CATEGORY_INIT (av1_parse_debug, "av1parse", 0, "av1 parser");
}
static gboolean
gst_av1_parse_start (GstBaseParse * parse)
{
GstAV1Parse *self = GST_AV1_PARSE (parse);
GST_DEBUG_OBJECT (self, "start");
gst_av1_parse_reset (self);
self->parser = gst_av1_parser_new ();
/* At least the OBU header. */
gst_base_parse_set_min_frame_size (parse, 1);
return TRUE;
}
static gboolean
gst_av1_parse_stop (GstBaseParse * parse)
{
GstAV1Parse *self = GST_AV1_PARSE (parse);
GST_DEBUG_OBJECT (self, "stop");
g_clear_pointer (&self->parser, gst_av1_parser_free);
return TRUE;
}
static const gchar *
gst_av1_parse_profile_to_string (GstAV1Profile profile)
{
switch (profile) {
case GST_AV1_PROFILE_0:
return "main";
case GST_AV1_PROFILE_1:
return "high";
case GST_AV1_PROFILE_2:
return "professional";
default:
break;
}
return NULL;
}
static const gchar *
gst_av1_parse_seq_level_idx_to_string (GstAV1SeqLevels seq_level_idx)
{
switch (seq_level_idx) {
case GST_AV1_SEQ_LEVEL_2_0:
return "2.0";
case GST_AV1_SEQ_LEVEL_2_1:
return "2.1";
case GST_AV1_SEQ_LEVEL_2_2:
return "2.2";
case GST_AV1_SEQ_LEVEL_2_3:
return "2.3";
case GST_AV1_SEQ_LEVEL_3_0:
return "3.0";
case GST_AV1_SEQ_LEVEL_3_1:
return "3.1";
case GST_AV1_SEQ_LEVEL_3_2:
return "3.2";
case GST_AV1_SEQ_LEVEL_3_3:
return "3.3";
case GST_AV1_SEQ_LEVEL_4_0:
return "4.0";
case GST_AV1_SEQ_LEVEL_4_1:
return "4.1";
case GST_AV1_SEQ_LEVEL_4_2:
return "4.2";
case GST_AV1_SEQ_LEVEL_4_3:
return "4.3";
case GST_AV1_SEQ_LEVEL_5_0:
return "5.0";
case GST_AV1_SEQ_LEVEL_5_1:
return "5.1";
case GST_AV1_SEQ_LEVEL_5_2:
return "5.2";
case GST_AV1_SEQ_LEVEL_5_3:
return "5.3";
case GST_AV1_SEQ_LEVEL_6_0:
return "6.0";
case GST_AV1_SEQ_LEVEL_6_1:
return "6.1";
case GST_AV1_SEQ_LEVEL_6_2:
return "6.2";
case GST_AV1_SEQ_LEVEL_6_3:
return "6.3";
case GST_AV1_SEQ_LEVEL_7_0:
return "7.0";
case GST_AV1_SEQ_LEVEL_7_1:
return "7.1";
case GST_AV1_SEQ_LEVEL_7_2:
return "7.2";
case GST_AV1_SEQ_LEVEL_7_3:
return "7.3";
case GST_AV1_SEQ_LEVEL_MAX:
case GST_AV1_SEQ_LEVELS:
/* not valid */
return NULL;
}
return NULL;
}
static const gchar *
gst_av1_parse_tier_to_string (guint8 seq_tier)
{
switch (seq_tier) {
case 0:
return "main";
case 1:
return "high";
}
return NULL;
}
static GstAV1Profile
gst_av1_parse_profile_from_string (const gchar * profile)
{
if (!profile)
return GST_AV1_PROFILE_UNDEFINED;
if (g_strcmp0 (profile, "main") == 0)
return GST_AV1_PROFILE_0;
else if (g_strcmp0 (profile, "high") == 0)
return GST_AV1_PROFILE_1;
else if (g_strcmp0 (profile, "professional") == 0)
return GST_AV1_PROFILE_2;
return GST_AV1_PROFILE_UNDEFINED;
}
static const gchar *
gst_av1_parse_alignment_to_steam_format_string (GstAV1ParseAligment align)
{
switch (align) {
case GST_AV1_PARSE_ALIGN_BYTE:
return "obu-stream";
case GST_AV1_PARSE_ALIGN_OBU:
case GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT:
case GST_AV1_PARSE_ALIGN_FRAME:
return "obu-stream";
case GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B:
return "annexb";
default:
GST_WARNING ("Unrecognized steam format");
break;
}
return NULL;
}
static const gchar *
gst_av1_parse_alignment_to_string (GstAV1ParseAligment align)
{
switch (align) {
case GST_AV1_PARSE_ALIGN_BYTE:
return "byte";
case GST_AV1_PARSE_ALIGN_OBU:
return "obu";
case GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT:
case GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B:
return "tu";
case GST_AV1_PARSE_ALIGN_FRAME:
return "frame";
default:
GST_WARNING ("Unrecognized alignment");
break;
}
return NULL;
}
static GstAV1ParseAligment
gst_av1_parse_alignment_from_string (const gchar * align,
const gchar * stream_format)
{
if (!align && !stream_format)
return GST_AV1_PARSE_ALIGN_NONE;
if (stream_format) {
if (g_strcmp0 (stream_format, "annexb") == 0) {
if (align && g_strcmp0 (align, "tu") != 0) {
/* annex b stream must align to TU. */
return GST_AV1_PARSE_ALIGN_ERROR;
} else {
return GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B;
}
} else if (g_strcmp0 (stream_format, "obu-stream") != 0) {
/* unrecognized */
return GST_AV1_PARSE_ALIGN_NONE;
}
/* stream-format is obu-stream, depends on align */
}
if (align) {
if (g_strcmp0 (align, "byte") == 0) {
return GST_AV1_PARSE_ALIGN_BYTE;
} else if (g_strcmp0 (align, "obu") == 0) {
return GST_AV1_PARSE_ALIGN_OBU;
} else if (g_strcmp0 (align, "tu") == 0) {
return GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT;
} else if (g_strcmp0 (align, "frame") == 0) {
return GST_AV1_PARSE_ALIGN_FRAME;
} else {
/* unrecognized */
return GST_AV1_PARSE_ALIGN_NONE;
}
}
return GST_AV1_PARSE_ALIGN_NONE;
}
static gboolean
gst_av1_parse_caps_has_alignment (GstCaps * caps, GstAV1ParseAligment alignment)
{
guint i, j, caps_size;
const gchar *cmp_align_str = NULL;
const gchar *cmp_stream_str = NULL;
GST_DEBUG ("Try to find alignment %d in caps: %" GST_PTR_FORMAT,
alignment, caps);
caps_size = gst_caps_get_size (caps);
if (caps_size == 0)
return FALSE;
switch (alignment) {
case GST_AV1_PARSE_ALIGN_BYTE:
cmp_align_str = "byte";
cmp_stream_str = "obu-stream";
break;
case GST_AV1_PARSE_ALIGN_OBU:
cmp_align_str = "obu";
cmp_stream_str = "obu-stream";
break;
case GST_AV1_PARSE_ALIGN_FRAME:
cmp_align_str = "frame";
cmp_stream_str = "obu-stream";
break;
case GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT:
cmp_align_str = "tu";
cmp_stream_str = "obu-stream";
break;
case GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B:
cmp_align_str = "tu";
cmp_stream_str = "annexb";
break;
default:
return FALSE;
}
for (i = 0; i < caps_size; i++) {
GstStructure *s = gst_caps_get_structure (caps, i);
const GValue *alignment_value = gst_structure_get_value (s, "alignment");
const GValue *stream_value = gst_structure_get_value (s, "stream-format");
if (!alignment_value || !stream_value)
continue;
if (G_VALUE_HOLDS_STRING (alignment_value)) {
const gchar *align_str = g_value_get_string (alignment_value);
if (g_strcmp0 (align_str, cmp_align_str) != 0)
continue;
} else if (GST_VALUE_HOLDS_LIST (alignment_value)) {
guint num_values = gst_value_list_get_size (alignment_value);
for (j = 0; j < num_values; j++) {
const GValue *v = gst_value_list_get_value (alignment_value, j);
const gchar *align_str = g_value_get_string (v);
if (g_strcmp0 (align_str, cmp_align_str) == 0)
break;
}
if (j == num_values)
continue;
}
if (G_VALUE_HOLDS_STRING (stream_value)) {
const gchar *stream_str = g_value_get_string (stream_value);
if (g_strcmp0 (stream_str, cmp_stream_str) != 0)
continue;
} else if (GST_VALUE_HOLDS_LIST (stream_value)) {
guint num_values = gst_value_list_get_size (stream_value);
for (j = 0; j < num_values; j++) {
const GValue *v = gst_value_list_get_value (stream_value, j);
const gchar *stream_str = g_value_get_string (v);
if (g_strcmp0 (stream_str, cmp_stream_str) == 0)
break;
}
if (j == num_values)
continue;
}
return TRUE;
}
return FALSE;
}
static GstAV1ParseAligment
gst_av1_parse_alignment_from_caps (GstCaps * caps)
{
GstAV1ParseAligment align;
align = GST_AV1_PARSE_ALIGN_NONE;
GST_DEBUG ("parsing caps: %" GST_PTR_FORMAT, caps);
if (caps && gst_caps_get_size (caps) > 0) {
GstStructure *s = gst_caps_get_structure (caps, 0);
const gchar *str_align = NULL;
const gchar *str_stream = NULL;
str_align = gst_structure_get_string (s, "alignment");
str_stream = gst_structure_get_string (s, "stream-format");
align = gst_av1_parse_alignment_from_string (str_align, str_stream);
}
return align;
}
static void
gst_av1_parse_update_src_caps (GstAV1Parse * self, GstCaps * caps)
{
GstCaps *sink_caps, *src_caps;
GstCaps *final_caps = NULL;
GstStructure *s = NULL;
gint width, height;
gint par_n = 0, par_d = 0;
const gchar *profile = NULL;
const gchar *level = NULL;
const gchar *tier = NULL;
if (G_UNLIKELY (!gst_pad_has_current_caps (GST_BASE_PARSE_SRC_PAD (self))))
self->update_caps = TRUE;
if (!self->update_caps)
return;
/* if this is being called from the first _setcaps call, caps on the sinkpad
* aren't set yet and so they need to be passed as an argument */
if (caps)
sink_caps = gst_caps_ref (caps);
else
sink_caps = gst_pad_get_current_caps (GST_BASE_PARSE_SINK_PAD (self));
/* carry over input caps as much as possible; override with our own stuff */
if (!sink_caps)
sink_caps = gst_caps_new_empty_simple ("video/x-av1");
else
s = gst_caps_get_structure (sink_caps, 0);
final_caps = gst_caps_copy (sink_caps);
if (s && gst_structure_has_field (s, "width") &&
gst_structure_has_field (s, "height")) {
gst_structure_get_int (s, "width", &width);
gst_structure_get_int (s, "height", &height);
} else {
width = self->width;
height = self->height;
}
if (width > 0 && height > 0)
gst_caps_set_simple (final_caps, "width", G_TYPE_INT, width,
"height", G_TYPE_INT, height, NULL);
if (s && gst_structure_get_fraction (s, "pixel-aspect-ratio", &par_n, &par_d)) {
if (par_n != 0 && par_d != 0) {
gst_caps_set_simple (final_caps, "pixel-aspect-ratio",
GST_TYPE_FRACTION, par_n, par_d, NULL);
}
}
if (self->fps_n > 0 && self->fps_d > 0) {
gst_caps_set_simple (final_caps, "framerate",
GST_TYPE_FRACTION, self->fps_n, self->fps_d, NULL);
gst_base_parse_set_frame_rate (GST_BASE_PARSE (self),
self->fps_n, self->fps_d, 0, 0);
}
/* When not RGB, the chroma format is needed. */
if (self->colorimetry == NULL ||
(g_strcmp0 (self->colorimetry, GST_VIDEO_COLORIMETRY_SRGB) != 0)) {
const gchar *chroma_format = NULL;
if (self->subsampling_x == 1 && self->subsampling_y == 1) {
if (!self->mono_chrome) {
chroma_format = "4:2:0";
} else {
chroma_format = "4:0:0";
}
} else if (self->subsampling_x == 1 && self->subsampling_y == 0) {
chroma_format = "4:2:2";
} else if (self->subsampling_x == 0 && self->subsampling_y == 0) {
chroma_format = "4:4:4";
}
if (chroma_format)
gst_caps_set_simple (final_caps,
"chroma-format", G_TYPE_STRING, chroma_format, NULL);
}
if (self->bit_depth)
gst_caps_set_simple (final_caps,
"bit-depth-luma", G_TYPE_UINT, self->bit_depth,
"bit-depth-chroma", G_TYPE_UINT, self->bit_depth, NULL);
if (self->colorimetry && (!s || !gst_structure_has_field (s, "colorimetry")))
gst_caps_set_simple (final_caps,
"colorimetry", G_TYPE_STRING, self->colorimetry, NULL);
g_assert (self->align > GST_AV1_PARSE_ALIGN_NONE);
gst_caps_set_simple (final_caps, "parsed", G_TYPE_BOOLEAN, TRUE,
"stream-format", G_TYPE_STRING,
gst_av1_parse_alignment_to_steam_format_string (self->align),
"alignment", G_TYPE_STRING,
gst_av1_parse_alignment_to_string (self->align), NULL);
profile = gst_av1_parse_profile_to_string (self->profile);
if (profile)
gst_caps_set_simple (final_caps, "profile", G_TYPE_STRING, profile, NULL);
level = gst_av1_parse_seq_level_idx_to_string (self->seq_level_idx);
if (level)
gst_caps_set_simple (final_caps, "level", G_TYPE_STRING, level, NULL);
tier = gst_av1_parse_tier_to_string (self->seq_tier);
if (tier)
gst_caps_set_simple (final_caps, "tier", G_TYPE_STRING, tier, NULL);
if ((self->max_seq_tier != self->seq_tier)
|| (self->max_seq_level_idx != self->seq_level_idx)) {
tier = gst_av1_parse_tier_to_string (self->max_seq_tier);
if (tier) {
gst_caps_set_simple (final_caps, "max-tier", G_TYPE_STRING, tier, NULL);
} else {
GST_WARNING_OBJECT (self, "Invalid max seq tier %d", self->max_seq_tier);
}
level = gst_av1_parse_seq_level_idx_to_string (self->max_seq_level_idx);
if (level) {
gst_caps_set_simple (final_caps, "max-level", G_TYPE_STRING, level, NULL);
} else {
GST_WARNING_OBJECT (self, "Invalid max level idx %d",
self->max_seq_level_idx);
}
}
src_caps = gst_pad_get_current_caps (GST_BASE_PARSE_SRC_PAD (self));
if (!(src_caps && gst_caps_is_strictly_equal (src_caps, final_caps))) {
GST_DEBUG_OBJECT (self, "Update src caps %" GST_PTR_FORMAT, final_caps);
gst_pad_set_caps (GST_BASE_PARSE_SRC_PAD (self), final_caps);
}
gst_clear_caps (&src_caps);
gst_caps_unref (final_caps);
gst_caps_unref (sink_caps);
self->update_caps = FALSE;
}
/* check downstream caps to configure format and alignment */
static void
gst_av1_parse_negotiate (GstAV1Parse * self, GstCaps * in_caps)
{
GstCaps *caps;
GstAV1ParseAligment align;
caps = gst_pad_get_allowed_caps (GST_BASE_PARSE_SRC_PAD (self));
GST_DEBUG_OBJECT (self, "allowed caps: %" GST_PTR_FORMAT, caps);
/* concentrate on leading structure, since decodebin parser
* capsfilter always includes parser template caps */
if (caps) {
caps = gst_caps_truncate (caps);
GST_DEBUG_OBJECT (self, "negotiating with caps: %" GST_PTR_FORMAT, caps);
}
/* prefer TU as default */
if (gst_av1_parse_caps_has_alignment (caps,
GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT)) {
self->align = GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT;
goto done;
}
/* Both upsteam and downstream support, best */
if (in_caps && caps) {
if (gst_caps_can_intersect (in_caps, caps)) {
GstCaps *common_caps = NULL;
common_caps = gst_caps_intersect (in_caps, caps);
align = gst_av1_parse_alignment_from_caps (common_caps);
gst_clear_caps (&common_caps);
if (align != GST_AV1_PARSE_ALIGN_NONE
&& align != GST_AV1_PARSE_ALIGN_ERROR) {
self->align = align;
goto done;
}
}
}
/* Select first one of downstream support */
if (caps && !gst_caps_is_empty (caps)) {
/* fixate to avoid ambiguity with lists when parsing */
caps = gst_caps_fixate (caps);
align = gst_av1_parse_alignment_from_caps (caps);
if (align != GST_AV1_PARSE_ALIGN_NONE && align != GST_AV1_PARSE_ALIGN_ERROR) {
self->align = align;
goto done;
}
}
/* default */
self->align = GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT;
done:
GST_INFO_OBJECT (self, "selected alignment %s",
gst_av1_parse_alignment_to_string (self->align));
gst_clear_caps (&caps);
}
static GstCaps *
gst_av1_parse_get_sink_caps (GstBaseParse * parse, GstCaps * filter)
{
GstCaps *peercaps, *templ;
GstCaps *res, *tmp, *pcopy;
templ = gst_pad_get_pad_template_caps (GST_BASE_PARSE_SINK_PAD (parse));
if (filter) {
GstCaps *fcopy = gst_caps_copy (filter);
/* Remove the fields we convert */
remove_fields (fcopy, TRUE);
peercaps = gst_pad_peer_query_caps (GST_BASE_PARSE_SRC_PAD (parse), fcopy);
gst_caps_unref (fcopy);
} else {
peercaps = gst_pad_peer_query_caps (GST_BASE_PARSE_SRC_PAD (parse), NULL);
}
pcopy = gst_caps_copy (peercaps);
remove_fields (pcopy, TRUE);
res = gst_caps_intersect_full (pcopy, templ, GST_CAPS_INTERSECT_FIRST);
gst_caps_unref (pcopy);
gst_caps_unref (templ);
if (filter) {
GstCaps *tmp = gst_caps_intersect_full (res, filter,
GST_CAPS_INTERSECT_FIRST);
gst_caps_unref (res);
res = tmp;
}
/* Try if we can put the downstream caps first */
pcopy = gst_caps_copy (peercaps);
remove_fields (pcopy, FALSE);
tmp = gst_caps_intersect_full (pcopy, res, GST_CAPS_INTERSECT_FIRST);
gst_caps_unref (pcopy);
if (!gst_caps_is_empty (tmp))
res = gst_caps_merge (tmp, res);
else
gst_caps_unref (tmp);
gst_caps_unref (peercaps);
return res;
}
static gboolean
gst_av1_parse_set_sink_caps (GstBaseParse * parse, GstCaps * caps)
{
GstAV1Parse *self = GST_AV1_PARSE (parse);
GstStructure *str;
GstAV1ParseAligment align;
GstCaps *in_caps = NULL;
const gchar *profile;
str = gst_caps_get_structure (caps, 0);
/* accept upstream info if provided */
gst_structure_get_int (str, "width", &self->width);
gst_structure_get_int (str, "height", &self->height);
profile = gst_structure_get_string (str, "profile");
if (profile)
self->profile = gst_av1_parse_profile_from_string (profile);
if (gst_structure_has_field (str, "framerate")) {
gst_structure_get_fraction (str, "framerate", &self->fps_n, &self->fps_d);
self->has_input_fps = TRUE;
} else {
self->fps_n = 0;
self->fps_d = 1;
self->has_input_fps = FALSE;
}
/* get upstream align from caps */
align = gst_av1_parse_alignment_from_caps (caps);
if (align == GST_AV1_PARSE_ALIGN_ERROR) {
GST_ERROR_OBJECT (self, "Sink caps %" GST_PTR_FORMAT " set stream-format"
" and alignment conflict.", caps);
return FALSE;
}
in_caps = gst_caps_copy (caps);
/* default */
if (align == GST_AV1_PARSE_ALIGN_NONE) {
align = GST_AV1_PARSE_ALIGN_BYTE;
gst_caps_set_simple (in_caps, "alignment", G_TYPE_STRING,
gst_av1_parse_alignment_to_string (align),
"stream-format", G_TYPE_STRING, "obu-stream", NULL);
}
/* negotiate with downstream, set output align */
gst_av1_parse_negotiate (self, in_caps);
self->update_caps = TRUE;
/* if all of decoder's capability related values are provided
* by upstream, update src caps now */
if (self->width > 0 && self->height > 0 && profile)
gst_av1_parse_update_src_caps (self, in_caps);
gst_caps_unref (in_caps);
self->in_align = align;
if (self->in_align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT)
self->detect_annex_b = TRUE;
if (self->in_align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B) {
gst_av1_parser_reset (self->parser, TRUE);
} else {
gst_av1_parser_reset (self->parser, FALSE);
}
return TRUE;
}
static GstFlowReturn
gst_av1_parse_push_data (GstAV1Parse * self, GstBaseParseFrame * frame,
guint32 finish_sz, gboolean frame_finished)
{
gsize sz;
GstBuffer *buf, *header_buf;
GstBuffer *buffer = frame->buffer;
GstFlowReturn ret = GST_FLOW_OK;
/* Need to generate the final TU annex-b format */
if (self->align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B) {
guint8 size_data[GST_AV1_MAX_LEB_128_SIZE];
guint size_len = 0;
guint len;
/* When push a TU, it must also be a frame end. */
g_assert (frame_finished);
/* Still some left in the frame cache */
len = gst_adapter_available (self->frame_cache);
if (len) {
buf = gst_adapter_take_buffer (self->frame_cache, len);
/* frame_unit_size */
_write_leb128 (size_data, &size_len, len);
header_buf = gst_buffer_new_memdup (size_data, size_len);
GST_BUFFER_PTS (header_buf) = GST_BUFFER_PTS (buf);
GST_BUFFER_DTS (header_buf) = GST_BUFFER_DTS (buf);
GST_BUFFER_DURATION (header_buf) = GST_BUFFER_DURATION (buf);
gst_adapter_push (self->cache_out, header_buf);
gst_adapter_push (self->cache_out, buf);
}
len = gst_adapter_available (self->cache_out);
if (len) {
buf = gst_adapter_take_buffer (self->cache_out, len);
/* temporal_unit_size */
_write_leb128 (size_data, &size_len, len);
header_buf = gst_buffer_new_memdup (size_data, size_len);
GST_BUFFER_PTS (header_buf) = GST_BUFFER_PTS (buf);
GST_BUFFER_DTS (header_buf) = GST_BUFFER_DTS (buf);
GST_BUFFER_DURATION (header_buf) = GST_BUFFER_DURATION (buf);
gst_adapter_push (self->cache_out, header_buf);
gst_adapter_push (self->cache_out, buf);
}
}
sz = gst_adapter_available (self->cache_out);
if (sz) {
buf = gst_adapter_take_buffer (self->cache_out, sz);
gst_buffer_copy_into (buf, buffer, GST_BUFFER_COPY_METADATA, 0, -1);
if (self->discont) {
GST_BUFFER_FLAG_SET (buf, GST_BUFFER_FLAG_DISCONT);
self->discont = FALSE;
} else {
GST_BUFFER_FLAG_UNSET (buf, GST_BUFFER_FLAG_DISCONT);
}
if (self->header) {
GST_BUFFER_FLAG_SET (buf, GST_BUFFER_FLAG_HEADER);
self->header = FALSE;
} else {
GST_BUFFER_FLAG_UNSET (buf, GST_BUFFER_FLAG_HEADER);
}
if (self->keyframe) {
GST_BUFFER_FLAG_UNSET (buf, GST_BUFFER_FLAG_DELTA_UNIT);
self->keyframe = FALSE;
} else {
GST_BUFFER_FLAG_SET (buf, GST_BUFFER_FLAG_DELTA_UNIT);
}
if (self->align == GST_AV1_PARSE_ALIGN_FRAME
|| self->align == GST_AV1_PARSE_ALIGN_OBU) {
if (!self->show_frame) {
GST_BUFFER_FLAG_SET (buf, GST_BUFFER_FLAG_DECODE_ONLY);
} else {
GST_BUFFER_FLAG_UNSET (buf, GST_BUFFER_FLAG_DECODE_ONLY);
if (frame_finished) {
GST_BUFFER_FLAG_SET (buf, GST_BUFFER_FLAG_MARKER);
} else {
GST_BUFFER_FLAG_UNSET (buf, GST_BUFFER_FLAG_MARKER);
}
}
} else {
/* Aligning to TUs which must contain a display frame. */
GST_BUFFER_FLAG_UNSET (buf, GST_BUFFER_FLAG_DECODE_ONLY);
if (frame_finished) {
GST_BUFFER_FLAG_SET (buf, GST_BUFFER_FLAG_MARKER);
} else {
GST_BUFFER_FLAG_UNSET (buf, GST_BUFFER_FLAG_MARKER);
}
}
gst_buffer_replace (&frame->out_buffer, buf);
gst_buffer_unref (buf);
gst_av1_parse_update_src_caps (self, NULL);
GST_LOG_OBJECT (self, "comsumed %d, output one buffer with size %"
G_GSSIZE_FORMAT, finish_sz, sz);
ret = gst_base_parse_finish_frame (GST_BASE_PARSE (self), frame, finish_sz);
}
return ret;
}
static void
gst_av1_parse_convert_to_annexb (GstAV1Parse * self, GstBuffer * buffer,
GstAV1OBU * obu, gboolean frame_complete)
{
guint8 size_data[GST_AV1_MAX_LEB_128_SIZE];
guint size_len = 0;
GstBitWriter bs;
GstBuffer *buf, *buf2;
guint8 *data;
guint len, len2, offset;
/* obu_length */
_write_leb128 (size_data, &size_len,
obu->obu_size + 1 + obu->header.obu_extention_flag);
gst_bit_writer_init_with_size (&bs, 128, FALSE);
/* obu_forbidden_bit */
gst_bit_writer_put_bits_uint8 (&bs, 0, 1);
/* obu_type */
gst_bit_writer_put_bits_uint8 (&bs, obu->obu_type, 4);
/* obu_extension_flag */
gst_bit_writer_put_bits_uint8 (&bs, obu->header.obu_extention_flag, 1);
/* obu_has_size_field */
gst_bit_writer_put_bits_uint8 (&bs, 0, 1);
/* obu_reserved_1bit */
gst_bit_writer_put_bits_uint8 (&bs, 0, 1);
if (obu->header.obu_extention_flag) {
/* temporal_id */
gst_bit_writer_put_bits_uint8 (&bs, obu->header.obu_temporal_id, 3);
/* spatial_id */
gst_bit_writer_put_bits_uint8 (&bs, obu->header.obu_spatial_id, 2);
/* extension_header_reserved_3bits */
gst_bit_writer_put_bits_uint8 (&bs, 0, 3);
}
g_assert (GST_BIT_WRITER_BIT_SIZE (&bs) % 8 == 0);
len = size_len;
len += GST_BIT_WRITER_BIT_SIZE (&bs) / 8;
len += obu->obu_size;
data = g_malloc (len);
offset = 0;
memcpy (data + offset, size_data, size_len);
offset += size_len;
memcpy (data + offset, GST_BIT_WRITER_DATA (&bs),
GST_BIT_WRITER_BIT_SIZE (&bs) / 8);
offset += GST_BIT_WRITER_BIT_SIZE (&bs) / 8;
memcpy (data + offset, obu->data, obu->obu_size);
/* The buf of this OBU */
buf = gst_buffer_new_wrapped (data, len);
GST_BUFFER_PTS (buf) = GST_BUFFER_PTS (buffer);
GST_BUFFER_DTS (buf) = GST_BUFFER_DTS (buffer);
GST_BUFFER_DURATION (buf) = GST_BUFFER_DURATION (buffer);
gst_adapter_push (self->frame_cache, buf);
if (frame_complete) {
len2 = gst_adapter_available (self->frame_cache);
buf2 = gst_adapter_take_buffer (self->frame_cache, len2);
/* frame_unit_size */
_write_leb128 (size_data, &size_len, len2);
buf = gst_buffer_new_memdup (size_data, size_len);
GST_BUFFER_PTS (buf) = GST_BUFFER_PTS (buf2);
GST_BUFFER_DTS (buf) = GST_BUFFER_DTS (buf2);
GST_BUFFER_DURATION (buf) = GST_BUFFER_DURATION (buf2);
gst_adapter_push (self->cache_out, buf);
gst_adapter_push (self->cache_out, buf2);
}
gst_bit_writer_reset (&bs);
}
static void
gst_av1_parse_convert_from_annexb (GstAV1Parse * self, GstBuffer * buffer,
GstAV1OBU * obu)
{
guint8 size_data[GST_AV1_MAX_LEB_128_SIZE];
guint size_len = 0;
GstBuffer *buf;
guint len, offset;
guint8 *data;
GstBitWriter bs;
_write_leb128 (size_data, &size_len, obu->obu_size);
/* obu_header */
len = 1;
if (obu->header.obu_extention_flag)
len += 1;
len += size_len;
len += obu->obu_size;
gst_bit_writer_init_with_size (&bs, 128, FALSE);
/* obu_forbidden_bit */
gst_bit_writer_put_bits_uint8 (&bs, 0, 1);
/* obu_type */
gst_bit_writer_put_bits_uint8 (&bs, obu->obu_type, 4);
/* obu_extension_flag */
gst_bit_writer_put_bits_uint8 (&bs, obu->header.obu_extention_flag, 1);
/* obu_has_size_field */
gst_bit_writer_put_bits_uint8 (&bs, 1, 1);
/* obu_reserved_1bit */
gst_bit_writer_put_bits_uint8 (&bs, 0, 1);
if (obu->header.obu_extention_flag) {
/* temporal_id */
gst_bit_writer_put_bits_uint8 (&bs, obu->header.obu_temporal_id, 3);
/* spatial_id */
gst_bit_writer_put_bits_uint8 (&bs, obu->header.obu_spatial_id, 2);
/* extension_header_reserved_3bits */
gst_bit_writer_put_bits_uint8 (&bs, 0, 3);
}
g_assert (GST_BIT_WRITER_BIT_SIZE (&bs) % 8 == 0);
data = g_malloc (len);
offset = 0;
memcpy (data + offset, GST_BIT_WRITER_DATA (&bs),
GST_BIT_WRITER_BIT_SIZE (&bs) / 8);
offset += GST_BIT_WRITER_BIT_SIZE (&bs) / 8;
memcpy (data + offset, size_data, size_len);
offset += size_len;
memcpy (data + offset, obu->data, obu->obu_size);
buf = gst_buffer_new_wrapped (data, len);
GST_BUFFER_PTS (buf) = GST_BUFFER_PTS (buffer);
GST_BUFFER_DTS (buf) = GST_BUFFER_DTS (buffer);
GST_BUFFER_DURATION (buf) = GST_BUFFER_DURATION (buffer);
gst_adapter_push (self->cache_out, buf);
gst_bit_writer_reset (&bs);
}
static void
gst_av1_parse_cache_one_obu (GstAV1Parse * self, GstBuffer * buffer,
GstAV1OBU * obu, guint8 * data, guint32 size, gboolean frame_complete)
{
gboolean need_convert = FALSE;
GstBuffer *buf;
if (self->in_align != self->align
&& (self->in_align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B
|| self->align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B))
need_convert = TRUE;
if (need_convert) {
if (self->in_align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B) {
gst_av1_parse_convert_from_annexb (self, buffer, obu);
} else {
gst_av1_parse_convert_to_annexb (self, buffer, obu, frame_complete);
}
} else if (self->align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B) {
g_assert (self->in_align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B);
gst_av1_parse_convert_to_annexb (self, buffer, obu, frame_complete);
} else {
buf = gst_buffer_new_memdup (data, size);
GST_BUFFER_PTS (buf) = GST_BUFFER_PTS (buffer);
GST_BUFFER_DTS (buf) = GST_BUFFER_DTS (buffer);
GST_BUFFER_DURATION (buf) = GST_BUFFER_DURATION (buffer);
gst_adapter_push (self->cache_out, buf);
}
}
static void
gst_av1_parse_calculate_framerate (GstAV1TimingInfo * ti,
gint * fps_n, gint * fps_d)
{
/* To calculate framerate, we use this formula:
*
* time_scale 1
* fps = ------------------------- x ---------------------------------
* num_units_in_display_tick num_ticks_per_picture_minus_1 + 1
*/
if (ti->equal_picture_interval) {
gint gcd;
*fps_n = ti->time_scale;
*fps_d = ti->num_units_in_display_tick *
(ti->num_ticks_per_picture_minus_1 + 1);
gcd = gst_util_greatest_common_divisor (*fps_n, *fps_d);
if (gcd) {
*fps_n /= gcd;
*fps_d /= gcd;
}
} else {
*fps_n = 0;
*fps_d = 1;
}
}
static GstAV1ParserResult
gst_av1_parse_handle_sequence_obu (GstAV1Parse * self, GstAV1OBU * obu)
{
GstAV1SequenceHeaderOBU seq_header;
GstAV1ParserResult res;
guint i;
guint val;
res = gst_av1_parser_parse_sequence_header_obu (self->parser,
obu, &seq_header);
if (res != GST_AV1_PARSER_OK)
return res;
if (self->width != seq_header.max_frame_width_minus_1 + 1) {
self->width = seq_header.max_frame_width_minus_1 + 1;
self->update_caps = TRUE;
}
if (self->height != seq_header.max_frame_height_minus_1 + 1) {
self->height = seq_header.max_frame_height_minus_1 + 1;
self->update_caps = TRUE;
}
if (seq_header.color_config.color_description_present_flag) {
GstVideoColorimetry cinfo;
gchar *colorimetry = NULL;
if (seq_header.color_config.color_range)
cinfo.range = GST_VIDEO_COLOR_RANGE_0_255;
else
cinfo.range = GST_VIDEO_COLOR_RANGE_16_235;
cinfo.matrix = gst_video_color_matrix_from_iso
(seq_header.color_config.matrix_coefficients);
cinfo.transfer = gst_video_transfer_function_from_iso
(seq_header.color_config.transfer_characteristics);
cinfo.primaries = gst_video_color_primaries_from_iso
(seq_header.color_config.color_primaries);
colorimetry = gst_video_colorimetry_to_string (&cinfo);
if (g_strcmp0 (colorimetry, self->colorimetry) != 0) {
g_free (self->colorimetry);
self->colorimetry = colorimetry;
colorimetry = NULL;
self->update_caps = TRUE;
}
g_clear_pointer (&colorimetry, g_free);
}
if (self->subsampling_x != seq_header.color_config.subsampling_x) {
self->subsampling_x = seq_header.color_config.subsampling_x;
self->update_caps = TRUE;
}
if (self->subsampling_y != seq_header.color_config.subsampling_y) {
self->subsampling_y = seq_header.color_config.subsampling_y;
self->update_caps = TRUE;
}
if (self->mono_chrome != seq_header.color_config.mono_chrome) {
self->mono_chrome = seq_header.color_config.mono_chrome;
self->update_caps = TRUE;
}
if (seq_header.operating_points_cnt_minus_1 + 1) {
/* Search for the actual max level and tier */
self->seq_level_idx = self->max_seq_level_idx =
seq_header.operating_points[0].seq_level_idx;
self->seq_tier = self->max_seq_tier =
seq_header.operating_points[0].seq_tier;
for (i = 1; i < seq_header.operating_points_cnt_minus_1 + 1; i++) {
if (self->max_seq_level_idx <
seq_header.operating_points[i].seq_level_idx) {
self->max_seq_level_idx = seq_header.operating_points[i].seq_level_idx;
self->max_seq_tier = seq_header.operating_points[i].seq_tier;
}
}
}
if (self->bit_depth != seq_header.bit_depth) {
self->bit_depth = seq_header.bit_depth;
self->update_caps = TRUE;
}
if (self->profile != seq_header.seq_profile) {
self->profile = seq_header.seq_profile;
self->update_caps = TRUE;
}
if (!self->has_input_fps) {
gint fps_n, fps_d;
gst_av1_parse_calculate_framerate (&seq_header.timing_info, &fps_n, &fps_d);
if (self->fps_n != fps_n || self->fps_d != fps_d) {
self->fps_n = fps_n;
self->fps_d = fps_d;
self->update_caps = TRUE;
}
}
val = (self->parser->state.operating_point_idc >> 8) & 0x0f;
for (i = 0; i < GST_AV1_MAX_NUM_SPATIAL_LAYERS; i++) {
if (val & (1 << i))
self->highest_spatial_id = i;
}
return GST_AV1_PARSER_OK;
}
/* Check whether the frame start a new TU.
The obu here should be a shown frame/frame header. */
static gboolean
gst_av1_parse_frame_start_new_temporal_unit (GstAV1Parse * self,
GstAV1OBU * obu)
{
gboolean ret = FALSE;
g_assert (obu->obu_type == GST_AV1_OBU_FRAME_HEADER
|| obu->obu_type == GST_AV1_OBU_FRAME);
/* 7.5.Ordering of OBUs: The value of temporal_id must be the same in all
OBU extension headers that are contained in the same temporal unit. */
if (self->last_shown_frame_temporal_id >= 0 &&
obu->header.obu_temporal_id != self->last_shown_frame_temporal_id) {
ret = TRUE;
goto new_tu;
}
/* If scalability is not being used, only one shown frame for each
temporal unit. So the new frame belongs to a new temporal unit. */
if (!self->within_one_frame && self->last_shown_frame_temporal_id >= 0 &&
self->parser->state.operating_point_idc == 0) {
ret = TRUE;
goto new_tu;
}
/* The new frame has the same layer IDs with the last shown frame,
it should belong to a new temporal unit. */
if (!self->within_one_frame &&
obu->header.obu_temporal_id == self->last_shown_frame_temporal_id &&
obu->header.obu_spatial_id == self->last_shown_frame_spatial_id) {
ret = TRUE;
goto new_tu;
}
new_tu:
if (ret) {
if (self->within_one_frame)
GST_WARNING_OBJECT (self,
"Start a new temporal unit with incompleted frame.");
gst_av1_parse_reset_obu_data_state (self);
gst_av1_parse_reset_tu_timestamp (self);
}
return ret;
}
/* frame_complete will be set true if it is the frame edge. */
static GstAV1ParserResult
gst_av1_parse_handle_one_obu (GstAV1Parse * self, GstAV1OBU * obu,
gboolean * frame_complete, gboolean * check_new_tu)
{
GstAV1ParserResult res = GST_AV1_PARSER_OK;
GstAV1MetadataOBU metadata;
GstAV1FrameHeaderOBU frame_header;
GstAV1TileListOBU tile_list;
GstAV1TileGroupOBU tile_group;
GstAV1FrameOBU frame;
*frame_complete = FALSE;
switch (obu->obu_type) {
case GST_AV1_OBU_TEMPORAL_DELIMITER:
res = gst_av1_parser_parse_temporal_delimiter_obu (self->parser, obu);
break;
case GST_AV1_OBU_SEQUENCE_HEADER:
res = gst_av1_parse_handle_sequence_obu (self, obu);
break;
case GST_AV1_OBU_REDUNDANT_FRAME_HEADER:
res = gst_av1_parser_parse_frame_header_obu (self->parser, obu,
&frame_header);
break;
case GST_AV1_OBU_FRAME_HEADER:
res = gst_av1_parser_parse_frame_header_obu (self->parser, obu,
&frame_header);
break;
case GST_AV1_OBU_FRAME:
res = gst_av1_parser_parse_frame_obu (self->parser, obu, &frame);
break;
case GST_AV1_OBU_METADATA:
res = gst_av1_parser_parse_metadata_obu (self->parser, obu, &metadata);
break;
case GST_AV1_OBU_TILE_GROUP:
res =
gst_av1_parser_parse_tile_group_obu (self->parser, obu, &tile_group);
break;
case GST_AV1_OBU_TILE_LIST:
res = gst_av1_parser_parse_tile_list_obu (self->parser, obu, &tile_list);
break;
case GST_AV1_OBU_PADDING:
break;
default:
GST_WARNING_OBJECT (self, "an unrecognized obu type %d", obu->obu_type);
res = GST_AV1_PARSER_BITSTREAM_ERROR;
break;
}
GST_LOG_OBJECT (self, "parsing the obu %s, result is %d",
_obu_name (obu->obu_type), res);
if (res != GST_AV1_PARSER_OK)
goto out;
/* 7.5:
All OBU extension headers that are contained in the same temporal
unit and have the same spatial_id value must have the same temporal_id
value.
And
OBUs with spatial level IDs (spatial_id) greater than 0 must
appear within a temporal unit in increasing order of the spatial
level ID values. */
if (obu->header.obu_spatial_id > self->highest_spatial_id) {
GST_WARNING_OBJECT (self,
"spatial_id %d is bigger than highest_spatial_id %d",
obu->header.obu_spatial_id, self->highest_spatial_id);
res = GST_AV1_PARSER_BITSTREAM_ERROR;
goto out;
}
/* If to check a new temporal starts, return early.
In 7.5.Ordering of OBUs: Sequence header OBUs may appear in any order
within a coded video sequence. So it is allowed to repeat the sequence
header within one temporal unit, and sequence header does not definitely
start a TU. We only check TD here. */
if (obu->obu_type == GST_AV1_OBU_TEMPORAL_DELIMITER) {
gst_av1_parse_reset_obu_data_state (self);
gst_av1_parse_reset_tu_timestamp (self);
if (check_new_tu) {
*check_new_tu = TRUE;
res = GST_AV1_PARSER_OK;
goto out;
}
}
if (obu->obu_type == GST_AV1_OBU_SEQUENCE_HEADER)
self->header = TRUE;
if (obu->obu_type == GST_AV1_OBU_FRAME_HEADER
|| obu->obu_type == GST_AV1_OBU_FRAME
|| obu->obu_type == GST_AV1_OBU_REDUNDANT_FRAME_HEADER) {
GstAV1FrameHeaderOBU *fh = &frame_header;
if (obu->obu_type == GST_AV1_OBU_FRAME)
fh = &frame.frame_header;
self->show_frame = fh->show_frame || fh->show_existing_frame;
if (self->show_frame) {
/* Check whether a new temporal starts, and return early. */
if (obu->obu_type != GST_AV1_OBU_REDUNDANT_FRAME_HEADER
&& gst_av1_parse_frame_start_new_temporal_unit (self, obu)) {
if (check_new_tu) {
*check_new_tu = TRUE;
res = GST_AV1_PARSER_OK;
goto out;
}
}
self->last_shown_frame_temporal_id = obu->header.obu_temporal_id;
self->last_shown_frame_spatial_id = obu->header.obu_spatial_id;
}
self->within_one_frame = TRUE;
/* if a show_existing_frame case, only update key frame.
otherwise, update all type of frame. */
if (!fh->show_existing_frame || fh->frame_type == GST_AV1_KEY_FRAME)
res = gst_av1_parser_reference_frame_update (self->parser, fh);
if (res != GST_AV1_PARSER_OK)
GST_WARNING_OBJECT (self, "update frame get result %d", res);
if (fh->show_existing_frame) {
*frame_complete = TRUE;
self->within_one_frame = FALSE;
}
if (fh->frame_type == GST_AV1_KEY_FRAME)
self->keyframe = TRUE;
}
if (obu->obu_type == GST_AV1_OBU_TILE_GROUP
|| obu->obu_type == GST_AV1_OBU_FRAME) {
GstAV1TileGroupOBU *tg = &tile_group;
self->within_one_frame = TRUE;
if (obu->obu_type == GST_AV1_OBU_FRAME)
tg = &frame.tile_group;
if (tg->tg_end == tg->num_tiles - 1) {
*frame_complete = TRUE;
self->within_one_frame = FALSE;
}
}
out:
if (res != GST_AV1_PARSER_OK) {
/* Some verbose OBU can be skip */
if (obu->obu_type == GST_AV1_OBU_REDUNDANT_FRAME_HEADER) {
GST_WARNING_OBJECT (self, "Ignore a verbose %s OBU parsing error",
_obu_name (obu->obu_type));
gst_av1_parse_reset_obu_data_state (self);
res = GST_AV1_PARSER_OK;
}
}
return res;
}
static GstFlowReturn
gst_av1_parse_handle_obu_to_obu (GstBaseParse * parse,
GstBaseParseFrame * frame, gint * skipsize)
{
GstAV1Parse *self = GST_AV1_PARSE (parse);
GstMapInfo map_info;
GstAV1OBU obu;
GstFlowReturn ret = GST_FLOW_OK;
GstAV1ParserResult res;
GstBuffer *buffer = gst_buffer_ref (frame->buffer);
guint32 consumed;
gboolean frame_complete;
if (!gst_buffer_map (buffer, &map_info, GST_MAP_READ)) {
*skipsize = 0;
GST_ERROR_OBJECT (parse, "Couldn't map incoming buffer");
return GST_FLOW_ERROR;
}
consumed = 0;
frame_complete = FALSE;
res = gst_av1_parser_identify_one_obu (self->parser, map_info.data,
map_info.size, &obu, &consumed);
if (res == GST_AV1_PARSER_OK)
res = gst_av1_parse_handle_one_obu (self, &obu, &frame_complete, NULL);
g_assert (consumed <= map_info.size);
if (res == GST_AV1_PARSER_BITSTREAM_ERROR ||
res == GST_AV1_PARSER_MISSING_OBU_REFERENCE) {
if (consumed) {
*skipsize = consumed;
} else {
*skipsize = map_info.size;
}
GST_WARNING_OBJECT (parse, "Parse obu error, discard %d.", *skipsize);
gst_av1_parse_reset_obu_data_state (self);
ret = GST_FLOW_OK;
goto out;
} else if (res == GST_AV1_PARSER_NO_MORE_DATA) {
*skipsize = 0;
if (self->in_align == GST_AV1_PARSE_ALIGN_OBU) {
/* The buffer is already aligned to OBU, should not happen. */
if (consumed) {
*skipsize = consumed;
} else {
*skipsize = map_info.size;
}
GST_WARNING_OBJECT (parse, "Parse obu need more data, discard %d.",
*skipsize);
gst_av1_parse_reset_obu_data_state (self);
}
ret = GST_FLOW_OK;
goto out;
} else if (res == GST_AV1_PARSER_DROP) {
GST_DEBUG_OBJECT (parse, "Drop %d data", consumed);
*skipsize = consumed;
gst_av1_parse_reset_obu_data_state (self);
ret = GST_FLOW_OK;
goto out;
} else if (res != GST_AV1_PARSER_OK) {
GST_ERROR_OBJECT (parse, "Parse obu get unexpect error %d", res);
*skipsize = 0;
ret = GST_FLOW_ERROR;
goto out;
}
g_assert (consumed);
gst_av1_parse_update_src_caps (self, NULL);
if (self->discont) {
GST_BUFFER_FLAG_SET (buffer, GST_BUFFER_FLAG_DISCONT);
self->discont = FALSE;
}
if (self->header) {
GST_BUFFER_FLAG_SET (buffer, GST_BUFFER_FLAG_HEADER);
self->header = FALSE;
}
/* happen to be a frame boundary */
if (frame_complete)
GST_BUFFER_FLAG_SET (buffer, GST_BUFFER_FLAG_MARKER);
GST_LOG_OBJECT (self, "Output one buffer with size %d", consumed);
ret = gst_base_parse_finish_frame (parse, frame, consumed);
*skipsize = 0;
out:
gst_buffer_unmap (buffer, &map_info);
gst_buffer_unref (buffer);
return ret;
}
static void
gst_av1_parse_create_subframe (GstBaseParseFrame * frame,
GstBaseParseFrame * subframe, GstBuffer * buffer)
{
gst_base_parse_frame_init (subframe);
subframe->flags |= frame->flags;
subframe->offset = frame->offset;
subframe->overhead = frame->overhead;
/* Just ref the input buffer. The base parse will check that
pointer, and it will be replaced by its out_buffer later. */
subframe->buffer = gst_buffer_ref (buffer);
}
static GstFlowReturn
gst_av1_parse_handle_to_small_and_equal_align (GstBaseParse * parse,
GstBaseParseFrame * frame, gint * skipsize)
{
GstAV1Parse *self = GST_AV1_PARSE (parse);
GstMapInfo map_info;
GstAV1OBU obu;
GstFlowReturn ret = GST_FLOW_OK;
GstAV1ParserResult res = GST_AV1_PARSER_INVALID_OPERATION;
GstBuffer *buffer = gst_buffer_ref (frame->buffer);
guint32 offset, consumed_before_push, consumed;
gboolean frame_complete;
GstBaseParseFrame subframe;
if (!gst_buffer_map (buffer, &map_info, GST_MAP_READ)) {
GST_ERROR_OBJECT (parse, "Couldn't map incoming buffer");
return GST_FLOW_ERROR;
}
av1parse: Correct the pts for frames and OBUs inside a TU. When the output alignment is smaller than the input alignment, for example, When the output alignment is "FRAME" and the parse is likely connecting to a decoder, the current PTS setting for AV1 frames inside a TU is not very correct. For example, a TU may begin with non-displayed frames and end with a displayed frame. The current way will assign the PTS to the first non-displayed frame, which is a decode-only frame and the PTS will be discarded in the video decoder. While the last displayed frame has invalid PTS, and so the video decoder needs to guess its PTS based on the frame rate and previous frame's PTS. This is not a decent and robust way. And more important, when the previous frames provide DTS, the video decoder will also guess the PTS based on the previous frames' DTS and trigger the warning like: gstvideodecoder.c:3147:gst_video_decoder_prepare_finish_frame: \ <vavp9dec0> decreasing timestame It sets the reordered_output and makes the decoder in free run mode. We should correct the PTS for a TU, let the non-displayed frames have no PTS while set the correct PTS to the displayed one. Also, when the AV1 stream has multi spatial layers, there are more than one displayed frames inside one TU with the same PTS. Note: If the input alignment is not TU aligned, we can not know the exact PTS of this TU, and so we just clear the PTS of the decode only frame and leave others unchanged. We also correct all the PTS if the output is OBU aligned. All their PTS and DTS are set to the input buffer's PTS. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3182>
2022-10-14 08:05:28 +00:00
self->buffer_pts = GST_BUFFER_PTS (buffer);
self->buffer_dts = GST_BUFFER_DTS (buffer);
self->buffer_duration = GST_BUFFER_DURATION (buffer);
consumed_before_push = 0;
offset = 0;
frame_complete = FALSE;
again:
while (offset < map_info.size) {
av1parse: Correct the pts for frames and OBUs inside a TU. When the output alignment is smaller than the input alignment, for example, When the output alignment is "FRAME" and the parse is likely connecting to a decoder, the current PTS setting for AV1 frames inside a TU is not very correct. For example, a TU may begin with non-displayed frames and end with a displayed frame. The current way will assign the PTS to the first non-displayed frame, which is a decode-only frame and the PTS will be discarded in the video decoder. While the last displayed frame has invalid PTS, and so the video decoder needs to guess its PTS based on the frame rate and previous frame's PTS. This is not a decent and robust way. And more important, when the previous frames provide DTS, the video decoder will also guess the PTS based on the previous frames' DTS and trigger the warning like: gstvideodecoder.c:3147:gst_video_decoder_prepare_finish_frame: \ <vavp9dec0> decreasing timestame It sets the reordered_output and makes the decoder in free run mode. We should correct the PTS for a TU, let the non-displayed frames have no PTS while set the correct PTS to the displayed one. Also, when the AV1 stream has multi spatial layers, there are more than one displayed frames inside one TU with the same PTS. Note: If the input alignment is not TU aligned, we can not know the exact PTS of this TU, and so we just clear the PTS of the decode only frame and leave others unchanged. We also correct all the PTS if the output is OBU aligned. All their PTS and DTS are set to the input buffer's PTS. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3182>
2022-10-14 08:05:28 +00:00
GST_BUFFER_OFFSET (buffer) = offset;
res = gst_av1_parser_identify_one_obu (self->parser,
map_info.data + offset, map_info.size - offset, &obu, &consumed);
if (res == GST_AV1_PARSER_OK)
res = gst_av1_parse_handle_one_obu (self, &obu, &frame_complete, NULL);
if (res != GST_AV1_PARSER_OK)
break;
if (obu.obu_type == GST_AV1_OBU_TEMPORAL_DELIMITER
&& consumed_before_push > 0) {
GST_DEBUG_OBJECT (self, "Encounter TD inside one %s aligned"
" buffer, should not happen normally.",
gst_av1_parse_alignment_to_string (self->in_align));
if (self->in_align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B)
gst_av1_parser_reset_annex_b (self->parser);
/* Not include this TD obu, it should belong to the next TU or frame,
we push all the data we already got. */
gst_av1_parse_create_subframe (frame, &subframe, buffer);
ret = gst_av1_parse_push_data (self, &subframe,
consumed_before_push, TRUE);
if (ret != GST_FLOW_OK)
goto out;
/* Begin to find the next. */
frame_complete = FALSE;
consumed_before_push = 0;
continue;
}
gst_av1_parse_cache_one_obu (self, buffer, &obu,
map_info.data + offset, consumed, frame_complete);
offset += consumed;
consumed_before_push += consumed;
if ((self->align == GST_AV1_PARSE_ALIGN_OBU) ||
(self->align == GST_AV1_PARSE_ALIGN_FRAME && frame_complete)) {
gst_av1_parse_create_subframe (frame, &subframe, buffer);
ret = gst_av1_parse_push_data (self, &subframe,
consumed_before_push, frame_complete);
if (ret != GST_FLOW_OK)
goto out;
/* Begin to find the next. */
frame_complete = FALSE;
consumed_before_push = 0;
continue;
}
}
if (res == GST_AV1_PARSER_BITSTREAM_ERROR ||
res == GST_AV1_PARSER_MISSING_OBU_REFERENCE) {
/* Discard the whole frame */
*skipsize = map_info.size;
GST_WARNING_OBJECT (parse, "Parse obu error, discard %d", *skipsize);
if (self->in_align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B)
gst_av1_parser_reset_annex_b (self->parser);
gst_av1_parse_reset_obu_data_state (self);
ret = GST_FLOW_OK;
goto out;
} else if (res == GST_AV1_PARSER_NO_MORE_DATA) {
/* Discard the whole buffer */
*skipsize = map_info.size;
GST_WARNING_OBJECT (parse, "Parse obu need more data, discard %d.",
*skipsize);
if (self->in_align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B)
gst_av1_parser_reset_annex_b (self->parser);
gst_av1_parse_reset_obu_data_state (self);
ret = GST_FLOW_OK;
goto out;
} else if (res == GST_AV1_PARSER_DROP) {
GST_DEBUG_OBJECT (parse, "Drop %d data", consumed);
offset += consumed;
gst_av1_parse_reset_obu_data_state (self);
res = GST_AV1_PARSER_OK;
goto again;
} else if (res != GST_AV1_PARSER_OK) {
GST_ERROR_OBJECT (parse, "Parse obu get unexpect error %d", res);
*skipsize = 0;
ret = GST_FLOW_ERROR;
goto out;
}
/* If the total buffer exhausted but frame is not complete, we just
push the left data and consider it as a frame. */
if (consumed_before_push > 0 && !frame_complete
&& self->align == GST_AV1_PARSE_ALIGN_FRAME) {
g_assert (offset >= map_info.size);
/* Warning and still consider the frame is complete */
GST_WARNING_OBJECT (self, "Exhaust the buffer but still incomplete frame,"
" should not happend in %s alignment",
gst_av1_parse_alignment_to_string (self->in_align));
}
ret = gst_av1_parse_push_data (self, frame, consumed_before_push, TRUE);
out:
gst_buffer_unmap (buffer, &map_info);
gst_buffer_unref (buffer);
av1parse: Correct the pts for frames and OBUs inside a TU. When the output alignment is smaller than the input alignment, for example, When the output alignment is "FRAME" and the parse is likely connecting to a decoder, the current PTS setting for AV1 frames inside a TU is not very correct. For example, a TU may begin with non-displayed frames and end with a displayed frame. The current way will assign the PTS to the first non-displayed frame, which is a decode-only frame and the PTS will be discarded in the video decoder. While the last displayed frame has invalid PTS, and so the video decoder needs to guess its PTS based on the frame rate and previous frame's PTS. This is not a decent and robust way. And more important, when the previous frames provide DTS, the video decoder will also guess the PTS based on the previous frames' DTS and trigger the warning like: gstvideodecoder.c:3147:gst_video_decoder_prepare_finish_frame: \ <vavp9dec0> decreasing timestame It sets the reordered_output and makes the decoder in free run mode. We should correct the PTS for a TU, let the non-displayed frames have no PTS while set the correct PTS to the displayed one. Also, when the AV1 stream has multi spatial layers, there are more than one displayed frames inside one TU with the same PTS. Note: If the input alignment is not TU aligned, we can not know the exact PTS of this TU, and so we just clear the PTS of the decode only frame and leave others unchanged. We also correct all the PTS if the output is OBU aligned. All their PTS and DTS are set to the input buffer's PTS. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3182>
2022-10-14 08:05:28 +00:00
gst_av1_parse_reset_tu_timestamp (self);
return ret;
}
static GstFlowReturn
gst_av1_parse_handle_to_big_align (GstBaseParse * parse,
GstBaseParseFrame * frame, gint * skipsize)
{
GstAV1Parse *self = GST_AV1_PARSE (parse);
GstMapInfo map_info;
GstAV1OBU obu;
GstFlowReturn ret = GST_FLOW_OK;
GstAV1ParserResult res = GST_AV1_PARSER_OK;
GstBuffer *buffer = gst_buffer_ref (frame->buffer);
guint32 consumed;
gboolean frame_complete;
gboolean check_new_tu;
gboolean complete;
g_assert (self->in_align <= GST_AV1_PARSE_ALIGN_FRAME);
if (!gst_buffer_map (buffer, &map_info, GST_MAP_READ)) {
*skipsize = 0;
GST_ERROR_OBJECT (parse, "Couldn't map incoming buffer");
return GST_FLOW_ERROR;
}
complete = FALSE;
again:
while (self->last_parsed_offset < map_info.size) {
res = gst_av1_parser_identify_one_obu (self->parser,
map_info.data + self->last_parsed_offset,
map_info.size - self->last_parsed_offset, &obu, &consumed);
if (res != GST_AV1_PARSER_OK)
break;
/* Take the DTS from the first OBU of the TU */
if (!GST_CLOCK_TIME_IS_VALID (self->buffer_dts))
self->buffer_dts = GST_BUFFER_DTS (buffer);
check_new_tu = FALSE;
if (self->align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT
|| self->align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B) {
res = gst_av1_parse_handle_one_obu (self, &obu, &frame_complete,
&check_new_tu);
} else {
res = gst_av1_parse_handle_one_obu (self, &obu, &frame_complete, NULL);
}
if (res != GST_AV1_PARSER_OK)
break;
if (check_new_tu && (gst_adapter_available (self->cache_out) ||
gst_adapter_available (self->frame_cache))) {
complete = TRUE;
break;
}
if (self->align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT ||
self->align == GST_AV1_PARSE_ALIGN_FRAME) {
GstBuffer *buf = gst_buffer_copy_region (buffer, GST_BUFFER_COPY_ALL,
self->last_parsed_offset, consumed);
gst_adapter_push (self->cache_out, buf);
} else if (self->align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B) {
gst_av1_parse_convert_to_annexb (self, buffer, &obu, frame_complete);
} else {
g_assert_not_reached ();
}
self->last_parsed_offset += consumed;
if (self->align == GST_AV1_PARSE_ALIGN_FRAME && frame_complete)
complete = TRUE;
if (complete)
break;
}
/* Finish a complete frame anyway */
if (complete || GST_BASE_PARSE_DRAINING (parse)) {
*skipsize = 0;
/* Save the oldest valid PTS as the TU PTS */
if (GST_CLOCK_TIME_IS_VALID (GST_BUFFER_PTS (buffer)))
self->buffer_pts = GST_BUFFER_PTS (buffer);
/* push the left anyway if no error */
if (res == GST_AV1_PARSER_OK)
ret = gst_av1_parse_push_data (self, frame,
self->last_parsed_offset, TRUE);
self->last_parsed_offset = 0;
goto out;
}
if (res == GST_AV1_PARSER_BITSTREAM_ERROR ||
res == GST_AV1_PARSER_MISSING_OBU_REFERENCE) {
*skipsize = map_info.size;
GST_WARNING_OBJECT (parse, "Parse obu error, discard whole buffer %d.",
*skipsize);
/* The adapter will be cleared in next loop because of
GST_BASE_PARSE_FRAME_FLAG_NEW_FRAME flag */
gst_av1_parse_reset_obu_data_state (self);
ret = GST_FLOW_OK;
} else if (res == GST_AV1_PARSER_NO_MORE_DATA) {
*skipsize = 0;
if (self->in_align >= GST_AV1_PARSE_ALIGN_OBU) {
/* The buffer is already aligned to OBU, should not happen.
The adapter will be cleared in next loop because of
GST_BASE_PARSE_FRAME_FLAG_NEW_FRAME flag */
*skipsize = map_info.size;
gst_av1_parse_reset_obu_data_state (self);
GST_WARNING_OBJECT (parse,
"Parse obu need more data, discard whole buffer %d.", *skipsize);
}
ret = GST_FLOW_OK;
} else if (res == GST_AV1_PARSER_DROP) {
GST_DEBUG_OBJECT (parse, "Drop %d data", consumed);
self->last_parsed_offset += consumed;
gst_av1_parse_reset_obu_data_state (self);
res = GST_AV1_PARSER_OK;
goto again;
} else if (res == GST_AV1_PARSER_OK) {
/* Everything is correct but still not get a frame or tu,
need more data */
GST_DEBUG_OBJECT (parse, "Need more data");
*skipsize = 0;
ret = GST_FLOW_OK;
} else {
GST_ERROR_OBJECT (parse, "Parse obu get unexpect error %d", res);
*skipsize = 0;
ret = GST_FLOW_ERROR;
}
out:
gst_buffer_unmap (buffer, &map_info);
gst_buffer_unref (buffer);
return ret;
}
/* Try to recognize whether the input is annex-b format.
return TRUE if we decide, FALSE if we can not decide or
encounter some error. */
static gboolean
gst_av1_parse_detect_stream_format (GstBaseParse * parse,
GstBaseParseFrame * frame)
{
GstAV1Parse *self = GST_AV1_PARSE (parse);
GstMapInfo map_info;
GstAV1OBU obu;
GstAV1ParserResult res = GST_AV1_PARSER_INVALID_OPERATION;
GstBuffer *buffer = gst_buffer_ref (frame->buffer);
gboolean got_seq, got_frame;
gboolean frame_complete;
guint32 consumed;
guint32 total_consumed;
guint32 tu_sz;
gboolean ret = FALSE;
g_assert (self->in_align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT);
g_assert (self->detect_annex_b == TRUE);
if (!gst_buffer_map (buffer, &map_info, GST_MAP_READ)) {
GST_ERROR_OBJECT (parse, "Couldn't map incoming buffer");
return FALSE;
}
gst_av1_parser_reset (self->parser, FALSE);
got_seq = FALSE;
got_frame = FALSE;
total_consumed = 0;
again:
while (total_consumed < map_info.size) {
res = gst_av1_parser_identify_one_obu (self->parser,
map_info.data + total_consumed, map_info.size - total_consumed,
&obu, &consumed);
if (res == GST_AV1_PARSER_OK) {
total_consumed += consumed;
res = gst_av1_parse_handle_one_obu (self, &obu, &frame_complete, NULL);
}
if (res != GST_AV1_PARSER_OK)
break;
if (obu.obu_type == GST_AV1_OBU_SEQUENCE_HEADER)
got_seq = TRUE;
if (obu.obu_type == GST_AV1_OBU_REDUNDANT_FRAME_HEADER ||
obu.obu_type == GST_AV1_OBU_FRAME ||
obu.obu_type == GST_AV1_OBU_FRAME_HEADER)
got_frame = TRUE;
if (got_seq || got_frame)
break;
}
gst_av1_parser_reset (self->parser, FALSE);
/* If succeed recognize seq or frame, it's done.
otherwise, just need to get more data. */
if (got_seq || got_frame) {
ret = TRUE;
self->detect_annex_b = FALSE;
goto out;
}
if (res == GST_AV1_PARSER_DROP) {
total_consumed += consumed;
res = GST_AV1_PARSER_OK;
gst_av1_parse_reset_obu_data_state (self);
goto again;
}
/* Try the annex b format. The buffer should contain the whole TU,
and the buffer start with the TU size in leb128() format. */
if (map_info.size < 8) {
/* Too small. */
goto out;
}
tu_sz = _read_leb128 (map_info.data, &res, &consumed);
if (tu_sz == 0 || res != GST_AV1_PARSER_OK) {
/* error to get the TU size, should not be annex b. */
goto out;
}
if (tu_sz + consumed != map_info.size) {
GST_DEBUG_OBJECT (self, "Buffer size %" G_GSSIZE_FORMAT ", TU size %d,"
" do not match.", map_info.size, tu_sz);
goto out;
}
GST_INFO_OBJECT (self, "Detect the annex-b format");
self->in_align = GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B;
self->detect_annex_b = FALSE;
gst_av1_parser_reset (self->parser, TRUE);
ret = TRUE;
out:
gst_av1_parse_reset_obu_data_state (self);
gst_buffer_unmap (buffer, &map_info);
gst_buffer_unref (buffer);
return ret;
}
static GstFlowReturn
gst_av1_parse_handle_frame (GstBaseParse * parse,
GstBaseParseFrame * frame, gint * skipsize)
{
GstAV1Parse *self = GST_AV1_PARSE (parse);
GstFlowReturn ret = GST_FLOW_OK;
guint in_level, out_level;
if (GST_BUFFER_FLAG_IS_SET (frame->buffer, GST_BUFFER_FLAG_DISCONT)) {
self->discont = TRUE;
if (frame->flags & GST_BASE_PARSE_FRAME_FLAG_NEW_FRAME) {
gst_av1_parse_reset_obu_data_state (self);
gst_av1_parse_reset_tu_timestamp (self);
}
} else {
self->discont = FALSE;
}
GST_LOG_OBJECT (self, "Input frame size %" G_GSSIZE_FORMAT,
gst_buffer_get_size (frame->buffer));
/* avoid stale cached parsing state */
if (frame->flags & GST_BASE_PARSE_FRAME_FLAG_NEW_FRAME) {
GST_LOG_OBJECT (self, "parsing new frame");
gst_adapter_clear (self->cache_out);
gst_adapter_clear (self->frame_cache);
self->last_parsed_offset = 0;
self->header = FALSE;
self->keyframe = FALSE;
self->show_frame = FALSE;
} else {
GST_LOG_OBJECT (self, "resuming frame parsing");
}
/* When in pull mode, the sink pad has no caps, we may get the
caps by query the upstream element */
if (self->in_align == GST_AV1_PARSE_ALIGN_NONE) {
GstCaps *upstream_caps;
upstream_caps =
gst_pad_peer_query_caps (GST_BASE_PARSE_SINK_PAD (self), NULL);
if (upstream_caps) {
if (!gst_caps_is_empty (upstream_caps)
&& !gst_caps_is_any (upstream_caps)) {
GstAV1ParseAligment align;
GST_LOG_OBJECT (self, "upstream caps: %" GST_PTR_FORMAT, upstream_caps);
/* fixate to avoid ambiguity with lists when parsing */
upstream_caps = gst_caps_fixate (upstream_caps);
align = gst_av1_parse_alignment_from_caps (upstream_caps);
if (align == GST_AV1_PARSE_ALIGN_ERROR) {
GST_ERROR_OBJECT (self, "upstream caps %" GST_PTR_FORMAT
" set stream-format and alignment conflict.", upstream_caps);
gst_caps_unref (upstream_caps);
return GST_FLOW_ERROR;
}
self->in_align = align;
}
gst_caps_unref (upstream_caps);
gst_av1_parser_reset (self->parser,
self->in_align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B);
}
if (self->in_align != GST_AV1_PARSE_ALIGN_NONE) {
GST_LOG_OBJECT (self, "Query the upstream get the alignment %s",
gst_av1_parse_alignment_to_string (self->in_align));
} else {
self->in_align = GST_AV1_PARSE_ALIGN_BYTE;
GST_DEBUG_OBJECT (self, "alignment set to default %s",
gst_av1_parse_alignment_to_string (GST_AV1_PARSE_ALIGN_BYTE));
}
}
if (self->in_align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT
&& self->detect_annex_b) {
/* Only happend at the first time of handle_frame, try to
recognize the annex b stream format. */
if (gst_av1_parse_detect_stream_format (parse, frame)) {
GST_INFO_OBJECT (self, "Input alignment %s",
gst_av1_parse_alignment_to_string (self->in_align));
} else {
/* Because the input is already TU aligned, we should skip
the whole problematic TU and check the next one. */
*skipsize = gst_buffer_get_size (frame->buffer);
GST_WARNING_OBJECT (self, "Fail to detect the stream format for TU,"
" skip the whole TU %d", *skipsize);
return GST_FLOW_OK;
}
}
/* We may in pull mode and no caps is set */
if (self->align == GST_AV1_PARSE_ALIGN_NONE)
gst_av1_parse_negotiate (self, NULL);
in_level = self->in_align;
if (self->in_align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B)
in_level = GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT;
out_level = self->align;
if (self->align == GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT_ANNEX_B)
out_level = GST_AV1_PARSE_ALIGN_TEMPORAL_UNIT;
if (self->in_align <= GST_AV1_PARSE_ALIGN_OBU
&& self->align == GST_AV1_PARSE_ALIGN_OBU) {
ret = gst_av1_parse_handle_obu_to_obu (parse, frame, skipsize);
} else if (in_level < out_level) {
ret = gst_av1_parse_handle_to_big_align (parse, frame, skipsize);
} else {
ret = gst_av1_parse_handle_to_small_and_equal_align (parse,
frame, skipsize);
}
return ret;
}
av1parse: Correct the pts for frames and OBUs inside a TU. When the output alignment is smaller than the input alignment, for example, When the output alignment is "FRAME" and the parse is likely connecting to a decoder, the current PTS setting for AV1 frames inside a TU is not very correct. For example, a TU may begin with non-displayed frames and end with a displayed frame. The current way will assign the PTS to the first non-displayed frame, which is a decode-only frame and the PTS will be discarded in the video decoder. While the last displayed frame has invalid PTS, and so the video decoder needs to guess its PTS based on the frame rate and previous frame's PTS. This is not a decent and robust way. And more important, when the previous frames provide DTS, the video decoder will also guess the PTS based on the previous frames' DTS and trigger the warning like: gstvideodecoder.c:3147:gst_video_decoder_prepare_finish_frame: \ <vavp9dec0> decreasing timestame It sets the reordered_output and makes the decoder in free run mode. We should correct the PTS for a TU, let the non-displayed frames have no PTS while set the correct PTS to the displayed one. Also, when the AV1 stream has multi spatial layers, there are more than one displayed frames inside one TU with the same PTS. Note: If the input alignment is not TU aligned, we can not know the exact PTS of this TU, and so we just clear the PTS of the decode only frame and leave others unchanged. We also correct all the PTS if the output is OBU aligned. All their PTS and DTS are set to the input buffer's PTS. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3182>
2022-10-14 08:05:28 +00:00
static GstFlowReturn
gst_av1_parse_pre_push_frame (GstBaseParse * parse, GstBaseParseFrame * frame)
{
GstAV1Parse *self = GST_AV1_PARSE (parse);
frame->flags |= GST_BASE_PARSE_FRAME_FLAG_CLIP;
if (!frame->buffer)
return GST_FLOW_OK;
if (self->align == GST_AV1_PARSE_ALIGN_FRAME) {
/* Input buffers may may contain more than one frames inside its buffer.
When splitting a TU into frames, the base parse class only assign the
PTS to the first frame and leave the others PTS invalid. But in fact,
all decode only frames should have invalid PTS while showable frames
should have correct PTS setting. */
if (GST_BUFFER_FLAG_IS_SET (frame->buffer, GST_BUFFER_FLAG_DECODE_ONLY)) {
GST_BUFFER_PTS (frame->buffer) = GST_CLOCK_TIME_NONE;
GST_BUFFER_DURATION (frame->buffer) = GST_CLOCK_TIME_NONE;
av1parse: Correct the pts for frames and OBUs inside a TU. When the output alignment is smaller than the input alignment, for example, When the output alignment is "FRAME" and the parse is likely connecting to a decoder, the current PTS setting for AV1 frames inside a TU is not very correct. For example, a TU may begin with non-displayed frames and end with a displayed frame. The current way will assign the PTS to the first non-displayed frame, which is a decode-only frame and the PTS will be discarded in the video decoder. While the last displayed frame has invalid PTS, and so the video decoder needs to guess its PTS based on the frame rate and previous frame's PTS. This is not a decent and robust way. And more important, when the previous frames provide DTS, the video decoder will also guess the PTS based on the previous frames' DTS and trigger the warning like: gstvideodecoder.c:3147:gst_video_decoder_prepare_finish_frame: \ <vavp9dec0> decreasing timestame It sets the reordered_output and makes the decoder in free run mode. We should correct the PTS for a TU, let the non-displayed frames have no PTS while set the correct PTS to the displayed one. Also, when the AV1 stream has multi spatial layers, there are more than one displayed frames inside one TU with the same PTS. Note: If the input alignment is not TU aligned, we can not know the exact PTS of this TU, and so we just clear the PTS of the decode only frame and leave others unchanged. We also correct all the PTS if the output is OBU aligned. All their PTS and DTS are set to the input buffer's PTS. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3182>
2022-10-14 08:05:28 +00:00
} else {
GST_BUFFER_PTS (frame->buffer) = self->buffer_pts;
GST_BUFFER_DURATION (frame->buffer) = self->buffer_duration;
av1parse: Correct the pts for frames and OBUs inside a TU. When the output alignment is smaller than the input alignment, for example, When the output alignment is "FRAME" and the parse is likely connecting to a decoder, the current PTS setting for AV1 frames inside a TU is not very correct. For example, a TU may begin with non-displayed frames and end with a displayed frame. The current way will assign the PTS to the first non-displayed frame, which is a decode-only frame and the PTS will be discarded in the video decoder. While the last displayed frame has invalid PTS, and so the video decoder needs to guess its PTS based on the frame rate and previous frame's PTS. This is not a decent and robust way. And more important, when the previous frames provide DTS, the video decoder will also guess the PTS based on the previous frames' DTS and trigger the warning like: gstvideodecoder.c:3147:gst_video_decoder_prepare_finish_frame: \ <vavp9dec0> decreasing timestame It sets the reordered_output and makes the decoder in free run mode. We should correct the PTS for a TU, let the non-displayed frames have no PTS while set the correct PTS to the displayed one. Also, when the AV1 stream has multi spatial layers, there are more than one displayed frames inside one TU with the same PTS. Note: If the input alignment is not TU aligned, we can not know the exact PTS of this TU, and so we just clear the PTS of the decode only frame and leave others unchanged. We also correct all the PTS if the output is OBU aligned. All their PTS and DTS are set to the input buffer's PTS. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3182>
2022-10-14 08:05:28 +00:00
}
GST_BUFFER_DTS (frame->buffer) = self->buffer_dts;
av1parse: Correct the pts for frames and OBUs inside a TU. When the output alignment is smaller than the input alignment, for example, When the output alignment is "FRAME" and the parse is likely connecting to a decoder, the current PTS setting for AV1 frames inside a TU is not very correct. For example, a TU may begin with non-displayed frames and end with a displayed frame. The current way will assign the PTS to the first non-displayed frame, which is a decode-only frame and the PTS will be discarded in the video decoder. While the last displayed frame has invalid PTS, and so the video decoder needs to guess its PTS based on the frame rate and previous frame's PTS. This is not a decent and robust way. And more important, when the previous frames provide DTS, the video decoder will also guess the PTS based on the previous frames' DTS and trigger the warning like: gstvideodecoder.c:3147:gst_video_decoder_prepare_finish_frame: \ <vavp9dec0> decreasing timestame It sets the reordered_output and makes the decoder in free run mode. We should correct the PTS for a TU, let the non-displayed frames have no PTS while set the correct PTS to the displayed one. Also, when the AV1 stream has multi spatial layers, there are more than one displayed frames inside one TU with the same PTS. Note: If the input alignment is not TU aligned, we can not know the exact PTS of this TU, and so we just clear the PTS of the decode only frame and leave others unchanged. We also correct all the PTS if the output is OBU aligned. All their PTS and DTS are set to the input buffer's PTS. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3182>
2022-10-14 08:05:28 +00:00
} else if (self->align == GST_AV1_PARSE_ALIGN_OBU) {
/* When we split a big frame or TU into OBUs, all OBUs should have the
same PTS and DTS of the input buffer, and should not have duration. */
if (self->in_align >= GST_AV1_PARSE_ALIGN_FRAME) {
GST_BUFFER_PTS (frame->buffer) = self->buffer_pts;
GST_BUFFER_DTS (frame->buffer) = self->buffer_dts;
GST_BUFFER_DURATION (frame->buffer) = GST_CLOCK_TIME_NONE;
}
}
GST_LOG_OBJECT (parse, "Adjust the frame buffer PTS/DTS/duration."
" The buffer of size %" G_GSIZE_FORMAT " now with dts %"
GST_TIME_FORMAT ", pts %" GST_TIME_FORMAT ", duration %"
GST_TIME_FORMAT, gst_buffer_get_size (frame->buffer),
GST_TIME_ARGS (GST_BUFFER_DTS (frame->buffer)),
GST_TIME_ARGS (GST_BUFFER_PTS (frame->buffer)),
GST_TIME_ARGS (GST_BUFFER_DURATION (frame->buffer)));
return GST_FLOW_OK;
}