mirror of
https://gitlab.freedesktop.org/gstreamer/gstreamer.git
synced 2024-11-18 07:47:17 +00:00
d36a22553b
If width or height are zero ... there's no video :)
1015 lines
31 KiB
C
1015 lines
31 KiB
C
/* GStreamer
|
|
* Copyright (C) 2004 Benjamin Otte <in7y118@public.uni-hamburg.de>
|
|
* Copyright (c) 2012 Collabora Ltd.
|
|
* Author : Edward Hervey <edward@collabora.com>
|
|
* Author : Mark Nauwelaerts <mark.nauwelaerts@collabora.co.uk>
|
|
*
|
|
* 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-theoradec
|
|
* @title: theoradec
|
|
* @see_also: theoraenc, oggdemux
|
|
*
|
|
* This element decodes theora streams into raw video
|
|
* <ulink url="http://www.theora.org/">Theora</ulink> is a royalty-free
|
|
* video codec maintained by the <ulink url="http://www.xiph.org/">Xiph.org
|
|
* Foundation</ulink>, based on the VP3 codec.
|
|
*
|
|
* ## Example pipeline
|
|
* |[
|
|
* gst-launch-1.0 -v filesrc location=videotestsrc.ogg ! oggdemux ! theoradec ! videoconvert ! videoscale ! autovideosink
|
|
* ]|
|
|
* This example pipeline will decode an ogg stream and decodes the theora video in it.
|
|
* Refer to the theoraenc example to create the ogg file.
|
|
*
|
|
*/
|
|
|
|
#ifdef HAVE_CONFIG_H
|
|
# include "config.h"
|
|
#endif
|
|
|
|
#include "gsttheoradec.h"
|
|
#include <gst/tag/tag.h>
|
|
#include <gst/video/video.h>
|
|
#include <gst/video/gstvideometa.h>
|
|
#include <gst/video/gstvideopool.h>
|
|
|
|
#define GST_CAT_DEFAULT theoradec_debug
|
|
GST_DEBUG_CATEGORY_STATIC (theoradec_debug);
|
|
GST_DEBUG_CATEGORY_STATIC (CAT_PERFORMANCE);
|
|
|
|
#define THEORA_DEF_TELEMETRY_MV 0
|
|
#define THEORA_DEF_TELEMETRY_MBMODE 0
|
|
#define THEORA_DEF_TELEMETRY_QI 0
|
|
#define THEORA_DEF_TELEMETRY_BITS 0
|
|
|
|
/* This was removed from the base class, this is used as a
|
|
temporary return to signal the need to call _drop_frame,
|
|
and does not leave theoraenc. */
|
|
#define GST_CUSTOM_FLOW_DROP GST_FLOW_CUSTOM_SUCCESS_1
|
|
|
|
enum
|
|
{
|
|
PROP_0,
|
|
PROP_TELEMETRY_MV,
|
|
PROP_TELEMETRY_MBMODE,
|
|
PROP_TELEMETRY_QI,
|
|
PROP_TELEMETRY_BITS
|
|
};
|
|
|
|
static GstStaticPadTemplate theora_dec_src_factory =
|
|
GST_STATIC_PAD_TEMPLATE ("src",
|
|
GST_PAD_SRC,
|
|
GST_PAD_ALWAYS,
|
|
GST_STATIC_CAPS ("video/x-raw, "
|
|
"format = (string) { I420, Y42B, Y444 }, "
|
|
"framerate = (fraction) [0/1, MAX], "
|
|
"width = (int) [ 1, MAX ], " "height = (int) [ 1, MAX ]")
|
|
);
|
|
|
|
static GstStaticPadTemplate theora_dec_sink_factory =
|
|
GST_STATIC_PAD_TEMPLATE ("sink",
|
|
GST_PAD_SINK,
|
|
GST_PAD_ALWAYS,
|
|
GST_STATIC_CAPS ("video/x-theora")
|
|
);
|
|
|
|
#define gst_theora_dec_parent_class parent_class
|
|
G_DEFINE_TYPE (GstTheoraDec, gst_theora_dec, GST_TYPE_VIDEO_DECODER);
|
|
|
|
static void theora_dec_get_property (GObject * object, guint prop_id,
|
|
GValue * value, GParamSpec * pspec);
|
|
static void theora_dec_set_property (GObject * object, guint prop_id,
|
|
const GValue * value, GParamSpec * pspec);
|
|
|
|
static gboolean theora_dec_start (GstVideoDecoder * decoder);
|
|
static gboolean theora_dec_stop (GstVideoDecoder * decoder);
|
|
static gboolean theora_dec_set_format (GstVideoDecoder * decoder,
|
|
GstVideoCodecState * state);
|
|
static gboolean theora_dec_flush (GstVideoDecoder * decoder);
|
|
static GstFlowReturn theora_dec_parse (GstVideoDecoder * decoder,
|
|
GstVideoCodecFrame * frame, GstAdapter * adapter, gboolean at_eos);
|
|
static GstFlowReturn theora_dec_handle_frame (GstVideoDecoder * decoder,
|
|
GstVideoCodecFrame * frame);
|
|
static gboolean theora_dec_decide_allocation (GstVideoDecoder * decoder,
|
|
GstQuery * query);
|
|
|
|
static GstFlowReturn theora_dec_decode_buffer (GstTheoraDec * dec,
|
|
GstBuffer * buf, GstVideoCodecFrame * frame);
|
|
|
|
static gboolean
|
|
gst_theora_dec_ctl_is_supported (int req)
|
|
{
|
|
/* should return TH_EFAULT or TH_EINVAL if supported, and TH_EIMPL if not */
|
|
return (th_decode_ctl (NULL, req, NULL, 0) != TH_EIMPL);
|
|
}
|
|
|
|
static void
|
|
gst_theora_dec_class_init (GstTheoraDecClass * klass)
|
|
{
|
|
GObjectClass *gobject_class = G_OBJECT_CLASS (klass);
|
|
GstElementClass *element_class = GST_ELEMENT_CLASS (klass);
|
|
GstVideoDecoderClass *video_decoder_class = GST_VIDEO_DECODER_CLASS (klass);
|
|
|
|
gobject_class->set_property = theora_dec_set_property;
|
|
gobject_class->get_property = theora_dec_get_property;
|
|
|
|
if (gst_theora_dec_ctl_is_supported (TH_DECCTL_SET_TELEMETRY_MV)) {
|
|
g_object_class_install_property (gobject_class, PROP_TELEMETRY_MV,
|
|
g_param_spec_int ("visualize-motion-vectors",
|
|
"Visualize motion vectors",
|
|
"Show motion vector selection overlaid on image. "
|
|
"Value gives a mask for motion vector (MV) modes to show",
|
|
0, 0xffff, THEORA_DEF_TELEMETRY_MV,
|
|
G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
|
|
}
|
|
|
|
if (gst_theora_dec_ctl_is_supported (TH_DECCTL_SET_TELEMETRY_MBMODE)) {
|
|
g_object_class_install_property (gobject_class, PROP_TELEMETRY_MBMODE,
|
|
g_param_spec_int ("visualize-macroblock-modes",
|
|
"Visualize macroblock modes",
|
|
"Show macroblock mode selection overlaid on image. "
|
|
"Value gives a mask for macroblock (MB) modes to show",
|
|
0, 0xffff, THEORA_DEF_TELEMETRY_MBMODE,
|
|
G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
|
|
}
|
|
|
|
if (gst_theora_dec_ctl_is_supported (TH_DECCTL_SET_TELEMETRY_QI)) {
|
|
g_object_class_install_property (gobject_class, PROP_TELEMETRY_QI,
|
|
g_param_spec_int ("visualize-quantization-modes",
|
|
"Visualize adaptive quantization modes",
|
|
"Show adaptive quantization mode selection overlaid on image. "
|
|
"Value gives a mask for quantization (QI) modes to show",
|
|
0, 0xffff, THEORA_DEF_TELEMETRY_QI,
|
|
G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
|
|
}
|
|
|
|
if (gst_theora_dec_ctl_is_supported (TH_DECCTL_SET_TELEMETRY_BITS)) {
|
|
/* FIXME: make this a boolean instead? The value scales the bars so
|
|
* they're less wide. Default is to use full width, and anything else
|
|
* doesn't seem particularly useful, since the smaller bars just disappear
|
|
* then (they almost disappear for a value of 2 already). */
|
|
g_object_class_install_property (gobject_class, PROP_TELEMETRY_BITS,
|
|
g_param_spec_int ("visualize-bit-usage",
|
|
"Visualize bitstream usage breakdown",
|
|
"Sets the bitstream breakdown visualization mode. "
|
|
"Values influence the width of the bit usage bars to show",
|
|
0, 0xff, THEORA_DEF_TELEMETRY_BITS,
|
|
G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
|
|
}
|
|
|
|
gst_element_class_add_static_pad_template (element_class,
|
|
&theora_dec_src_factory);
|
|
gst_element_class_add_static_pad_template (element_class,
|
|
&theora_dec_sink_factory);
|
|
gst_element_class_set_static_metadata (element_class, "Theora video decoder",
|
|
"Codec/Decoder/Video", "decode raw theora streams to raw YUV video",
|
|
"Benjamin Otte <otte@gnome.org>, Wim Taymans <wim@fluendo.com>");
|
|
|
|
video_decoder_class->start = GST_DEBUG_FUNCPTR (theora_dec_start);
|
|
video_decoder_class->stop = GST_DEBUG_FUNCPTR (theora_dec_stop);
|
|
video_decoder_class->flush = GST_DEBUG_FUNCPTR (theora_dec_flush);
|
|
video_decoder_class->set_format = GST_DEBUG_FUNCPTR (theora_dec_set_format);
|
|
video_decoder_class->parse = GST_DEBUG_FUNCPTR (theora_dec_parse);
|
|
video_decoder_class->handle_frame =
|
|
GST_DEBUG_FUNCPTR (theora_dec_handle_frame);
|
|
video_decoder_class->decide_allocation =
|
|
GST_DEBUG_FUNCPTR (theora_dec_decide_allocation);
|
|
|
|
GST_DEBUG_CATEGORY_INIT (theoradec_debug, "theoradec", 0, "Theora decoder");
|
|
GST_DEBUG_CATEGORY_GET (CAT_PERFORMANCE, "GST_PERFORMANCE");
|
|
}
|
|
|
|
static void
|
|
gst_theora_dec_init (GstTheoraDec * dec)
|
|
{
|
|
dec->telemetry_mv = THEORA_DEF_TELEMETRY_MV;
|
|
dec->telemetry_mbmode = THEORA_DEF_TELEMETRY_MBMODE;
|
|
dec->telemetry_qi = THEORA_DEF_TELEMETRY_QI;
|
|
dec->telemetry_bits = THEORA_DEF_TELEMETRY_BITS;
|
|
|
|
/* input is packetized,
|
|
* but is not marked that way so data gets parsed and keyframes marked */
|
|
gst_video_decoder_set_packetized (GST_VIDEO_DECODER (dec), FALSE);
|
|
gst_video_decoder_set_needs_format (GST_VIDEO_DECODER (dec), TRUE);
|
|
|
|
gst_video_decoder_set_use_default_pad_acceptcaps (GST_VIDEO_DECODER_CAST
|
|
(dec), TRUE);
|
|
GST_PAD_SET_ACCEPT_TEMPLATE (GST_VIDEO_DECODER_SINK_PAD (dec));
|
|
}
|
|
|
|
static gboolean
|
|
theora_dec_start (GstVideoDecoder * decoder)
|
|
{
|
|
GstTheoraDec *dec = GST_THEORA_DEC (decoder);
|
|
|
|
GST_DEBUG_OBJECT (dec, "start");
|
|
GST_DEBUG_OBJECT (dec, "Setting have_header to FALSE");
|
|
dec->have_header = FALSE;
|
|
dec->can_crop = FALSE;
|
|
|
|
return TRUE;
|
|
}
|
|
|
|
static gboolean
|
|
theora_dec_stop (GstVideoDecoder * decoder)
|
|
{
|
|
GstTheoraDec *dec = GST_THEORA_DEC (decoder);
|
|
|
|
GST_DEBUG_OBJECT (dec, "stop");
|
|
|
|
th_info_clear (&dec->info);
|
|
th_comment_clear (&dec->comment);
|
|
if (dec->setup) {
|
|
th_setup_free (dec->setup);
|
|
dec->setup = NULL;
|
|
}
|
|
if (dec->decoder) {
|
|
th_decode_free (dec->decoder);
|
|
dec->decoder = NULL;
|
|
}
|
|
|
|
if (dec->input_state) {
|
|
gst_video_codec_state_unref (dec->input_state);
|
|
dec->input_state = NULL;
|
|
}
|
|
if (dec->output_state) {
|
|
gst_video_codec_state_unref (dec->output_state);
|
|
dec->output_state = NULL;
|
|
}
|
|
dec->can_crop = FALSE;
|
|
|
|
return TRUE;
|
|
}
|
|
|
|
static gboolean
|
|
theora_dec_flush (GstVideoDecoder * decoder)
|
|
{
|
|
GstTheoraDec *dec = GST_THEORA_DEC (decoder);
|
|
|
|
dec->need_keyframe = TRUE;
|
|
|
|
return TRUE;
|
|
}
|
|
|
|
static GstFlowReturn
|
|
theora_dec_parse (GstVideoDecoder * decoder,
|
|
GstVideoCodecFrame * frame, GstAdapter * adapter, gboolean at_eos)
|
|
{
|
|
gint av;
|
|
const guint8 *data;
|
|
|
|
av = gst_adapter_available (adapter);
|
|
|
|
if (av > 0) {
|
|
data = gst_adapter_map (adapter, 1);
|
|
/* check for keyframe; must not be header packet (0x80 | 0x40) */
|
|
if (!(data[0] & 0xc0)) {
|
|
GST_VIDEO_CODEC_FRAME_SET_SYNC_POINT (frame);
|
|
GST_LOG_OBJECT (decoder, "Found keyframe");
|
|
}
|
|
gst_adapter_unmap (adapter);
|
|
}
|
|
|
|
/* and pass along all */
|
|
gst_video_decoder_add_to_frame (decoder, av);
|
|
return gst_video_decoder_have_frame (decoder);
|
|
}
|
|
|
|
|
|
static gboolean
|
|
theora_dec_set_format (GstVideoDecoder * bdec, GstVideoCodecState * state)
|
|
{
|
|
GstTheoraDec *dec;
|
|
|
|
dec = GST_THEORA_DEC (bdec);
|
|
|
|
/* Keep a copy of the input state */
|
|
if (dec->input_state)
|
|
gst_video_codec_state_unref (dec->input_state);
|
|
dec->input_state = gst_video_codec_state_ref (state);
|
|
|
|
/* FIXME : Interesting, we always accept any kind of caps ? */
|
|
if (state->codec_data) {
|
|
GstBuffer *buffer;
|
|
GstMapInfo minfo;
|
|
guint8 *data;
|
|
guint size;
|
|
guint offset;
|
|
|
|
buffer = state->codec_data;
|
|
gst_buffer_map (buffer, &minfo, GST_MAP_READ);
|
|
|
|
offset = 0;
|
|
size = minfo.size;
|
|
data = (guint8 *) minfo.data;
|
|
|
|
while (size > 2) {
|
|
guint psize;
|
|
GstBuffer *buf;
|
|
|
|
psize = (data[0] << 8) | data[1];
|
|
/* skip header */
|
|
data += 2;
|
|
size -= 2;
|
|
offset += 2;
|
|
|
|
/* make sure we don't read too much */
|
|
psize = MIN (psize, size);
|
|
|
|
buf = gst_buffer_copy_region (buffer, GST_BUFFER_COPY_ALL, offset, psize);
|
|
|
|
/* first buffer is a discont buffer */
|
|
if (offset == 2)
|
|
GST_BUFFER_FLAG_SET (buf, GST_BUFFER_FLAG_DISCONT);
|
|
|
|
/* now feed it to the decoder we can ignore the error */
|
|
theora_dec_decode_buffer (dec, buf, NULL);
|
|
gst_buffer_unref (buf);
|
|
|
|
/* skip the data */
|
|
size -= psize;
|
|
data += psize;
|
|
offset += psize;
|
|
}
|
|
|
|
gst_buffer_unmap (buffer, &minfo);
|
|
}
|
|
|
|
GST_DEBUG_OBJECT (dec, "Done");
|
|
|
|
return TRUE;
|
|
}
|
|
|
|
static GstFlowReturn
|
|
theora_handle_comment_packet (GstTheoraDec * dec, ogg_packet * packet)
|
|
{
|
|
gchar *encoder = NULL;
|
|
GstTagList *list;
|
|
|
|
GST_DEBUG_OBJECT (dec, "parsing comment packet");
|
|
|
|
list =
|
|
gst_tag_list_from_vorbiscomment (packet->packet, packet->bytes,
|
|
(guint8 *) "\201theora", 7, &encoder);
|
|
|
|
if (!list) {
|
|
GST_ERROR_OBJECT (dec, "couldn't decode comments");
|
|
list = gst_tag_list_new_empty ();
|
|
}
|
|
if (encoder) {
|
|
gst_tag_list_add (list, GST_TAG_MERGE_REPLACE,
|
|
GST_TAG_ENCODER, encoder, NULL);
|
|
g_free (encoder);
|
|
}
|
|
gst_tag_list_add (list, GST_TAG_MERGE_REPLACE,
|
|
GST_TAG_ENCODER_VERSION, dec->info.version_major,
|
|
GST_TAG_VIDEO_CODEC, "Theora", NULL);
|
|
|
|
if (dec->info.target_bitrate > 0) {
|
|
gst_tag_list_add (list, GST_TAG_MERGE_REPLACE,
|
|
GST_TAG_BITRATE, dec->info.target_bitrate,
|
|
GST_TAG_NOMINAL_BITRATE, dec->info.target_bitrate, NULL);
|
|
}
|
|
|
|
gst_video_decoder_merge_tags (GST_VIDEO_DECODER (dec),
|
|
list, GST_TAG_MERGE_REPLACE);
|
|
|
|
gst_tag_list_unref (list);
|
|
|
|
return GST_FLOW_OK;
|
|
}
|
|
|
|
static GstFlowReturn
|
|
theora_handle_type_packet (GstTheoraDec * dec)
|
|
{
|
|
gint par_num, par_den;
|
|
GstFlowReturn ret = GST_FLOW_OK;
|
|
GstVideoCodecState *state;
|
|
GstVideoFormat fmt;
|
|
GstVideoInfo *info;
|
|
|
|
info = &dec->input_state->info;
|
|
|
|
GST_DEBUG_OBJECT (dec, "fps %d/%d, PAR %d/%d",
|
|
dec->info.fps_numerator, dec->info.fps_denominator,
|
|
dec->info.aspect_numerator, dec->info.aspect_denominator);
|
|
|
|
/* calculate par
|
|
* the info.aspect_* values reflect PAR;
|
|
* 0:x and x:0 are allowed and can be interpreted as 1:1.
|
|
*/
|
|
par_num = GST_VIDEO_INFO_PAR_N (info);
|
|
par_den = GST_VIDEO_INFO_PAR_D (info);
|
|
|
|
/* If we have a default PAR, see if the decoder specified a different one */
|
|
if (par_num == 1 && par_den == 1 &&
|
|
(dec->info.aspect_numerator != 0 && dec->info.aspect_denominator != 0)) {
|
|
par_num = dec->info.aspect_numerator;
|
|
par_den = dec->info.aspect_denominator;
|
|
}
|
|
/* theora has:
|
|
*
|
|
* width/height : dimension of the encoded frame
|
|
* pic_width/pic_height : dimension of the visible part
|
|
* pic_x/pic_y : offset in encoded frame where visible part starts
|
|
*/
|
|
GST_DEBUG_OBJECT (dec, "dimension %dx%d, PAR %d/%d", dec->info.pic_width,
|
|
dec->info.pic_height, par_num, par_den);
|
|
GST_DEBUG_OBJECT (dec, "frame dimension %dx%d, offset %d:%d",
|
|
dec->info.pic_width, dec->info.pic_height,
|
|
dec->info.pic_x, dec->info.pic_y);
|
|
|
|
switch (dec->info.pixel_fmt) {
|
|
case TH_PF_420:
|
|
fmt = GST_VIDEO_FORMAT_I420;
|
|
break;
|
|
case TH_PF_422:
|
|
fmt = GST_VIDEO_FORMAT_Y42B;
|
|
break;
|
|
case TH_PF_444:
|
|
fmt = GST_VIDEO_FORMAT_Y444;
|
|
break;
|
|
default:
|
|
goto unsupported_format;
|
|
}
|
|
|
|
GST_VIDEO_INFO_WIDTH (info) = dec->info.pic_width;
|
|
GST_VIDEO_INFO_HEIGHT (info) = dec->info.pic_height;
|
|
|
|
/* Ensure correct offsets in chroma for formats that need it
|
|
* by rounding the offset. libtheora will add proper pixels,
|
|
* so no need to handle them ourselves. */
|
|
if (dec->info.pic_x & 1 && dec->info.pixel_fmt != TH_PF_444) {
|
|
GST_VIDEO_INFO_WIDTH (info)++;
|
|
}
|
|
if (dec->info.pic_y & 1 && dec->info.pixel_fmt == TH_PF_420) {
|
|
GST_VIDEO_INFO_HEIGHT (info)++;
|
|
}
|
|
|
|
GST_DEBUG_OBJECT (dec, "after fixup frame dimension %dx%d, offset %d:%d",
|
|
info->width, info->height, dec->info.pic_x, dec->info.pic_y);
|
|
|
|
if (info->width == 0 || info->height == 0)
|
|
goto invalid_dimensions;
|
|
|
|
/* done */
|
|
dec->decoder = th_decode_alloc (&dec->info, dec->setup);
|
|
|
|
if (th_decode_ctl (dec->decoder, TH_DECCTL_SET_TELEMETRY_MV,
|
|
&dec->telemetry_mv, sizeof (dec->telemetry_mv)) != TH_EIMPL) {
|
|
GST_WARNING_OBJECT (dec, "Could not enable MV visualisation");
|
|
}
|
|
if (th_decode_ctl (dec->decoder, TH_DECCTL_SET_TELEMETRY_MBMODE,
|
|
&dec->telemetry_mbmode, sizeof (dec->telemetry_mbmode)) != TH_EIMPL) {
|
|
GST_WARNING_OBJECT (dec, "Could not enable MB mode visualisation");
|
|
}
|
|
if (th_decode_ctl (dec->decoder, TH_DECCTL_SET_TELEMETRY_QI,
|
|
&dec->telemetry_qi, sizeof (dec->telemetry_qi)) != TH_EIMPL) {
|
|
GST_WARNING_OBJECT (dec, "Could not enable QI mode visualisation");
|
|
}
|
|
if (th_decode_ctl (dec->decoder, TH_DECCTL_SET_TELEMETRY_BITS,
|
|
&dec->telemetry_bits, sizeof (dec->telemetry_bits)) != TH_EIMPL) {
|
|
GST_WARNING_OBJECT (dec, "Could not enable BITS mode visualisation");
|
|
}
|
|
|
|
/* Create the output state */
|
|
dec->output_state = state =
|
|
gst_video_decoder_set_output_state (GST_VIDEO_DECODER (dec), fmt,
|
|
info->width, info->height, dec->input_state);
|
|
|
|
/* FIXME : Do we still need to set fps/par now that we pass the reference input stream ? */
|
|
state->info.fps_n = dec->info.fps_numerator;
|
|
state->info.fps_d = dec->info.fps_denominator;
|
|
state->info.par_n = par_num;
|
|
state->info.par_d = par_den;
|
|
|
|
/* these values are for all versions of the colorspace specified in the
|
|
* theora info */
|
|
state->info.chroma_site = GST_VIDEO_CHROMA_SITE_JPEG;
|
|
state->info.colorimetry.range = GST_VIDEO_COLOR_RANGE_16_235;
|
|
state->info.colorimetry.matrix = GST_VIDEO_COLOR_MATRIX_BT601;
|
|
state->info.colorimetry.transfer = GST_VIDEO_TRANSFER_BT709;
|
|
switch (dec->info.colorspace) {
|
|
case TH_CS_ITU_REC_470M:
|
|
state->info.colorimetry.primaries = GST_VIDEO_COLOR_PRIMARIES_BT470M;
|
|
break;
|
|
case TH_CS_ITU_REC_470BG:
|
|
state->info.colorimetry.primaries = GST_VIDEO_COLOR_PRIMARIES_BT470BG;
|
|
break;
|
|
default:
|
|
state->info.colorimetry.primaries = GST_VIDEO_COLOR_PRIMARIES_UNKNOWN;
|
|
break;
|
|
}
|
|
|
|
dec->uncropped_info = state->info;
|
|
|
|
if (!gst_video_decoder_negotiate (GST_VIDEO_DECODER (dec)))
|
|
goto not_negotiated;
|
|
|
|
dec->have_header = TRUE;
|
|
|
|
return ret;
|
|
|
|
/* ERRORS */
|
|
unsupported_format:
|
|
{
|
|
GST_ERROR_OBJECT (dec, "Invalid pixel format %d", dec->info.pixel_fmt);
|
|
return GST_FLOW_ERROR;
|
|
}
|
|
|
|
not_negotiated:
|
|
{
|
|
GST_ERROR_OBJECT (dec, "Failed to negotiate");
|
|
return GST_FLOW_NOT_NEGOTIATED;
|
|
}
|
|
|
|
invalid_dimensions:
|
|
{
|
|
GST_ERROR_OBJECT (dec, "Invalid dimensions (width:%d, height:%d)",
|
|
info->width, info->height);
|
|
return GST_FLOW_ERROR;
|
|
}
|
|
}
|
|
|
|
static GstFlowReturn
|
|
theora_handle_header_packet (GstTheoraDec * dec, ogg_packet * packet)
|
|
{
|
|
GstFlowReturn res;
|
|
int ret;
|
|
|
|
GST_DEBUG_OBJECT (dec, "parsing header packet");
|
|
|
|
ret = th_decode_headerin (&dec->info, &dec->comment, &dec->setup, packet);
|
|
if (ret < 0)
|
|
goto header_read_error;
|
|
|
|
switch (packet->packet[0]) {
|
|
case 0x81:
|
|
res = theora_handle_comment_packet (dec, packet);
|
|
break;
|
|
case 0x82:
|
|
res = theora_handle_type_packet (dec);
|
|
break;
|
|
default:
|
|
/* ignore */
|
|
g_warning ("unknown theora header packet found");
|
|
case 0x80:
|
|
/* nothing special, this is the identification header */
|
|
res = GST_FLOW_OK;
|
|
break;
|
|
}
|
|
return res;
|
|
|
|
/* ERRORS */
|
|
header_read_error:
|
|
{
|
|
GST_ELEMENT_ERROR (GST_ELEMENT (dec), STREAM, DECODE,
|
|
(NULL), ("couldn't read header packet"));
|
|
return GST_FLOW_ERROR;
|
|
}
|
|
}
|
|
|
|
#define MIN_NUM_HEADERS 3
|
|
static GstFlowReturn
|
|
theoradec_handle_header_caps (GstTheoraDec * dec)
|
|
{
|
|
GstFlowReturn result = GST_CUSTOM_FLOW_DROP;
|
|
GstCaps *caps;
|
|
GstStructure *s = NULL;
|
|
const GValue *array = NULL;
|
|
|
|
GST_DEBUG_OBJECT (dec, "Looking for Theora headers in caps");
|
|
caps = gst_pad_get_current_caps (GST_VIDEO_DECODER_SINK_PAD (dec));
|
|
if (caps)
|
|
s = gst_caps_get_structure (caps, 0);
|
|
if (s)
|
|
array = gst_structure_get_value (s, "streamheader");
|
|
|
|
if (caps)
|
|
gst_caps_unref (caps);
|
|
|
|
if (array && (gst_value_array_get_size (array) >= MIN_NUM_HEADERS)) {
|
|
const GValue *value = NULL;
|
|
GstBuffer *buf = NULL;
|
|
gint i = 0;
|
|
|
|
while (result == GST_CUSTOM_FLOW_DROP
|
|
&& i < gst_value_array_get_size (array)) {
|
|
value = gst_value_array_get_value (array, i);
|
|
buf = gst_value_get_buffer (value);
|
|
if (!buf)
|
|
goto null_buffer;
|
|
GST_LOG_OBJECT (dec, "Submitting header packet");
|
|
result = theora_dec_decode_buffer (dec, buf, NULL);
|
|
i++;
|
|
}
|
|
} else
|
|
goto array_error;
|
|
|
|
done:
|
|
return (result !=
|
|
GST_CUSTOM_FLOW_DROP ? GST_FLOW_NOT_NEGOTIATED : GST_FLOW_OK);
|
|
|
|
/* ERRORS */
|
|
array_error:
|
|
{
|
|
GST_WARNING_OBJECT (dec, "streamheader array not found");
|
|
result = GST_FLOW_ERROR;
|
|
goto done;
|
|
}
|
|
null_buffer:
|
|
{
|
|
GST_WARNING_OBJECT (dec, "streamheader with null buffer received");
|
|
result = GST_FLOW_ERROR;
|
|
goto done;
|
|
}
|
|
}
|
|
|
|
/* Allocate buffer and copy image data into Y444 format */
|
|
static GstFlowReturn
|
|
theora_handle_image (GstTheoraDec * dec, th_ycbcr_buffer buf,
|
|
GstVideoCodecFrame * frame)
|
|
{
|
|
GstVideoDecoder *decoder = GST_VIDEO_DECODER (dec);
|
|
gint width, height, stride;
|
|
GstFlowReturn result;
|
|
gint i, comp;
|
|
guint8 *dest, *src;
|
|
GstVideoFrame vframe;
|
|
gint pic_width, pic_height;
|
|
gint offset_x, offset_y;
|
|
|
|
result = gst_video_decoder_allocate_output_frame (decoder, frame);
|
|
|
|
if (G_UNLIKELY (result != GST_FLOW_OK)) {
|
|
GST_DEBUG_OBJECT (dec, "could not get buffer, reason: %s",
|
|
gst_flow_get_name (result));
|
|
return result;
|
|
}
|
|
|
|
if (!dec->can_crop) {
|
|
/* we need to crop the hard way */
|
|
offset_x = dec->info.pic_x;
|
|
offset_y = dec->info.pic_y;
|
|
pic_width = dec->info.pic_width;
|
|
pic_height = dec->info.pic_height;
|
|
/* Ensure correct offsets in chroma for formats that need it
|
|
* by rounding the offset. libtheora will add proper pixels,
|
|
* so no need to handle them ourselves. */
|
|
if (offset_x & 1 && dec->info.pixel_fmt != TH_PF_444)
|
|
offset_x--;
|
|
if (offset_y & 1 && dec->info.pixel_fmt == TH_PF_420)
|
|
offset_y--;
|
|
} else {
|
|
/* copy the whole frame */
|
|
offset_x = 0;
|
|
offset_y = 0;
|
|
pic_width = dec->info.frame_width;
|
|
pic_height = dec->info.frame_height;
|
|
|
|
if (dec->info.pic_width != dec->info.frame_width ||
|
|
dec->info.pic_height != dec->info.frame_height ||
|
|
dec->info.pic_x != 0 || dec->info.pic_y != 0) {
|
|
GstVideoMeta *vmeta;
|
|
GstVideoCropMeta *cmeta;
|
|
|
|
vmeta = gst_buffer_get_video_meta (frame->output_buffer);
|
|
/* If the buffer pool didn't add the meta already
|
|
* we add it ourselves here */
|
|
if (!vmeta)
|
|
vmeta = gst_buffer_add_video_meta (frame->output_buffer,
|
|
GST_VIDEO_FRAME_FLAG_NONE,
|
|
dec->output_state->info.finfo->format,
|
|
dec->info.frame_width, dec->info.frame_height);
|
|
|
|
/* Just to be sure that the buffer pool doesn't do something
|
|
* completely weird and we would crash later
|
|
*/
|
|
g_assert (vmeta->format == dec->output_state->info.finfo->format);
|
|
g_assert (vmeta->width == dec->info.frame_width);
|
|
g_assert (vmeta->height == dec->info.frame_height);
|
|
|
|
cmeta = gst_buffer_add_video_crop_meta (frame->output_buffer);
|
|
|
|
/* we can do things slightly more efficient when we know that
|
|
* downstream understands clipping */
|
|
cmeta->x = dec->info.pic_x;
|
|
cmeta->y = dec->info.pic_y;
|
|
cmeta->width = dec->info.pic_width;
|
|
cmeta->height = dec->info.pic_height;
|
|
}
|
|
}
|
|
|
|
/* if only libtheora would allow us to give it a destination frame */
|
|
GST_CAT_TRACE_OBJECT (CAT_PERFORMANCE, dec,
|
|
"doing unavoidable video frame copy");
|
|
|
|
if (G_UNLIKELY (!gst_video_frame_map (&vframe, &dec->uncropped_info,
|
|
frame->output_buffer, GST_MAP_WRITE)))
|
|
goto invalid_frame;
|
|
|
|
for (comp = 0; comp < 3; comp++) {
|
|
width =
|
|
GST_VIDEO_FORMAT_INFO_SCALE_WIDTH (vframe.info.finfo, comp, pic_width);
|
|
height =
|
|
GST_VIDEO_FORMAT_INFO_SCALE_HEIGHT (vframe.info.finfo, comp,
|
|
pic_height);
|
|
stride = GST_VIDEO_FRAME_COMP_STRIDE (&vframe, comp);
|
|
dest = GST_VIDEO_FRAME_COMP_DATA (&vframe, comp);
|
|
|
|
src = buf[comp].data;
|
|
src += ((height == pic_height) ? offset_y : offset_y / 2)
|
|
* buf[comp].stride;
|
|
src += (width == pic_width) ? offset_x : offset_x / 2;
|
|
|
|
for (i = 0; i < height; i++) {
|
|
memcpy (dest, src, width);
|
|
|
|
dest += stride;
|
|
src += buf[comp].stride;
|
|
}
|
|
}
|
|
gst_video_frame_unmap (&vframe);
|
|
|
|
return GST_FLOW_OK;
|
|
invalid_frame:
|
|
{
|
|
GST_DEBUG_OBJECT (dec, "could not map video frame");
|
|
return GST_FLOW_ERROR;
|
|
}
|
|
}
|
|
|
|
static GstFlowReturn
|
|
theora_handle_data_packet (GstTheoraDec * dec, ogg_packet * packet,
|
|
GstVideoCodecFrame * frame)
|
|
{
|
|
/* normal data packet */
|
|
th_ycbcr_buffer buf;
|
|
gboolean keyframe;
|
|
GstFlowReturn result;
|
|
ogg_int64_t gp;
|
|
|
|
if (G_UNLIKELY (!dec->have_header)) {
|
|
result = theoradec_handle_header_caps (dec);
|
|
if (result != GST_FLOW_OK)
|
|
goto not_initialized;
|
|
}
|
|
|
|
/* the second most significant bit of the first data byte is cleared
|
|
* for keyframes. We can only check it if it's not a zero-length packet. */
|
|
keyframe = packet->bytes && ((packet->packet[0] & 0x40) == 0);
|
|
if (G_UNLIKELY (keyframe)) {
|
|
GST_DEBUG_OBJECT (dec, "we have a keyframe");
|
|
dec->need_keyframe = FALSE;
|
|
} else if (G_UNLIKELY (dec->need_keyframe)) {
|
|
goto dropping;
|
|
}
|
|
|
|
GST_DEBUG_OBJECT (dec, "parsing data packet");
|
|
|
|
/* this does the decoding */
|
|
if (G_UNLIKELY (th_decode_packetin (dec->decoder, packet, &gp) < 0))
|
|
goto decode_error;
|
|
|
|
if (frame &&
|
|
(gst_video_decoder_get_max_decode_time (GST_VIDEO_DECODER (dec),
|
|
frame) < 0))
|
|
goto dropping_qos;
|
|
|
|
/* this does postprocessing and set up the decoded frame
|
|
* pointers in our yuv variable */
|
|
if (G_UNLIKELY (th_decode_ycbcr_out (dec->decoder, buf) < 0))
|
|
goto no_yuv;
|
|
|
|
if (G_UNLIKELY ((buf[0].width != dec->info.frame_width)
|
|
|| (buf[0].height != dec->info.frame_height)))
|
|
goto wrong_dimensions;
|
|
|
|
result = theora_handle_image (dec, buf, frame);
|
|
|
|
return result;
|
|
|
|
/* ERRORS */
|
|
not_initialized:
|
|
{
|
|
GST_ELEMENT_ERROR (GST_ELEMENT (dec), STREAM, DECODE,
|
|
(NULL), ("no header sent yet"));
|
|
return GST_FLOW_ERROR;
|
|
}
|
|
dropping:
|
|
{
|
|
GST_WARNING_OBJECT (dec, "dropping frame because we need a keyframe");
|
|
return GST_CUSTOM_FLOW_DROP;
|
|
}
|
|
dropping_qos:
|
|
{
|
|
GST_WARNING_OBJECT (dec, "dropping frame because of QoS");
|
|
return GST_CUSTOM_FLOW_DROP;
|
|
}
|
|
decode_error:
|
|
{
|
|
GST_ELEMENT_ERROR (GST_ELEMENT (dec), STREAM, DECODE,
|
|
(NULL), ("theora decoder did not decode data packet"));
|
|
return GST_FLOW_ERROR;
|
|
}
|
|
no_yuv:
|
|
{
|
|
GST_ELEMENT_ERROR (GST_ELEMENT (dec), STREAM, DECODE,
|
|
(NULL), ("couldn't read out YUV image"));
|
|
return GST_FLOW_ERROR;
|
|
}
|
|
wrong_dimensions:
|
|
{
|
|
GST_ELEMENT_ERROR (GST_ELEMENT (dec), STREAM, FORMAT,
|
|
(NULL), ("dimensions of image do not match header"));
|
|
return GST_FLOW_ERROR;
|
|
}
|
|
}
|
|
|
|
static GstFlowReturn
|
|
theora_dec_decode_buffer (GstTheoraDec * dec, GstBuffer * buf,
|
|
GstVideoCodecFrame * frame)
|
|
{
|
|
ogg_packet packet;
|
|
GstFlowReturn result = GST_FLOW_OK;
|
|
GstMapInfo minfo;
|
|
|
|
/* make ogg_packet out of the buffer */
|
|
gst_buffer_map (buf, &minfo, GST_MAP_READ);
|
|
packet.packet = minfo.data;
|
|
packet.bytes = minfo.size;
|
|
packet.granulepos = -1;
|
|
packet.packetno = 0; /* we don't really care */
|
|
packet.b_o_s = dec->have_header ? 0 : 1;
|
|
/* EOS does not matter for the decoder */
|
|
packet.e_o_s = 0;
|
|
|
|
GST_LOG_OBJECT (dec, "decode buffer of size %ld", packet.bytes);
|
|
|
|
GST_DEBUG_OBJECT (dec, "header=%02x", packet.bytes ? packet.packet[0] : -1);
|
|
|
|
/* switch depending on packet type. A zero byte packet is always a data
|
|
* packet; we don't dereference it in that case. */
|
|
if (packet.bytes && packet.packet[0] & 0x80) {
|
|
/* header packets are not meant to be displayed - return FLOW_DROP */
|
|
if (dec->have_header) {
|
|
GST_WARNING_OBJECT (GST_OBJECT (dec), "Ignoring header");
|
|
result = GST_CUSTOM_FLOW_DROP;
|
|
goto done;
|
|
}
|
|
if ((result = theora_handle_header_packet (dec, &packet)) != GST_FLOW_OK)
|
|
goto done;
|
|
result = GST_CUSTOM_FLOW_DROP;
|
|
} else {
|
|
result = theora_handle_data_packet (dec, &packet, frame);
|
|
}
|
|
|
|
done:
|
|
gst_buffer_unmap (buf, &minfo);
|
|
|
|
return result;
|
|
}
|
|
|
|
static GstFlowReturn
|
|
theora_dec_handle_frame (GstVideoDecoder * bdec, GstVideoCodecFrame * frame)
|
|
{
|
|
GstTheoraDec *dec;
|
|
GstFlowReturn res;
|
|
|
|
dec = GST_THEORA_DEC (bdec);
|
|
|
|
res = theora_dec_decode_buffer (dec, frame->input_buffer, frame);
|
|
switch (res) {
|
|
case GST_FLOW_OK:
|
|
res = gst_video_decoder_finish_frame (bdec, frame);
|
|
break;
|
|
case GST_CUSTOM_FLOW_DROP:
|
|
res = gst_video_decoder_drop_frame (bdec, frame);
|
|
break;
|
|
default:
|
|
gst_video_codec_frame_unref (frame);
|
|
break;
|
|
}
|
|
|
|
return res;
|
|
}
|
|
|
|
static gboolean
|
|
theora_dec_decide_allocation (GstVideoDecoder * decoder, GstQuery * query)
|
|
{
|
|
GstTheoraDec *dec = GST_THEORA_DEC (decoder);
|
|
GstVideoCodecState *state;
|
|
GstBufferPool *pool;
|
|
guint size, min, max;
|
|
GstStructure *config;
|
|
|
|
if (!GST_VIDEO_DECODER_CLASS (parent_class)->decide_allocation (decoder,
|
|
query))
|
|
return FALSE;
|
|
|
|
state = gst_video_decoder_get_output_state (decoder);
|
|
|
|
gst_query_parse_nth_allocation_pool (query, 0, &pool, &size, &min, &max);
|
|
|
|
dec->can_crop = FALSE;
|
|
config = gst_buffer_pool_get_config (pool);
|
|
if (gst_query_find_allocation_meta (query, GST_VIDEO_META_API_TYPE, NULL)) {
|
|
gst_buffer_pool_config_add_option (config,
|
|
GST_BUFFER_POOL_OPTION_VIDEO_META);
|
|
dec->can_crop =
|
|
gst_query_find_allocation_meta (query, GST_VIDEO_CROP_META_API_TYPE,
|
|
NULL);
|
|
}
|
|
|
|
if (dec->can_crop) {
|
|
GstVideoInfo *info = &dec->uncropped_info;
|
|
GstCaps *caps;
|
|
|
|
GST_LOG_OBJECT (decoder, "Using GstVideoCropMeta, uncropped wxh = %dx%d",
|
|
info->width, info->height);
|
|
|
|
gst_video_info_set_format (info, info->finfo->format, dec->info.frame_width,
|
|
dec->info.frame_height);
|
|
|
|
/* Calculate uncropped size */
|
|
size = MAX (size, info->size);
|
|
caps = gst_video_info_to_caps (info);
|
|
gst_buffer_pool_config_set_params (config, caps, size, min, max);
|
|
gst_caps_unref (caps);
|
|
}
|
|
|
|
gst_buffer_pool_set_config (pool, config);
|
|
|
|
gst_query_set_nth_allocation_pool (query, 0, pool, size, min, max);
|
|
|
|
gst_object_unref (pool);
|
|
gst_video_codec_state_unref (state);
|
|
|
|
return TRUE;
|
|
}
|
|
|
|
static void
|
|
theora_dec_set_property (GObject * object, guint prop_id,
|
|
const GValue * value, GParamSpec * pspec)
|
|
{
|
|
GstTheoraDec *dec = GST_THEORA_DEC (object);
|
|
|
|
switch (prop_id) {
|
|
case PROP_TELEMETRY_MV:
|
|
dec->telemetry_mv = g_value_get_int (value);
|
|
break;
|
|
case PROP_TELEMETRY_MBMODE:
|
|
dec->telemetry_mbmode = g_value_get_int (value);
|
|
break;
|
|
case PROP_TELEMETRY_QI:
|
|
dec->telemetry_qi = g_value_get_int (value);
|
|
break;
|
|
case PROP_TELEMETRY_BITS:
|
|
dec->telemetry_bits = g_value_get_int (value);
|
|
break;
|
|
default:
|
|
G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
|
|
break;
|
|
}
|
|
}
|
|
|
|
static void
|
|
theora_dec_get_property (GObject * object, guint prop_id,
|
|
GValue * value, GParamSpec * pspec)
|
|
{
|
|
GstTheoraDec *dec = GST_THEORA_DEC (object);
|
|
|
|
switch (prop_id) {
|
|
case PROP_TELEMETRY_MV:
|
|
g_value_set_int (value, dec->telemetry_mv);
|
|
break;
|
|
case PROP_TELEMETRY_MBMODE:
|
|
g_value_set_int (value, dec->telemetry_mbmode);
|
|
break;
|
|
case PROP_TELEMETRY_QI:
|
|
g_value_set_int (value, dec->telemetry_qi);
|
|
break;
|
|
case PROP_TELEMETRY_BITS:
|
|
g_value_set_int (value, dec->telemetry_bits);
|
|
break;
|
|
default:
|
|
G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
|
|
break;
|
|
}
|
|
}
|