Added using new versions of libav audio and video decoding and encoding functions.

Added conversion/resampling functionality.
Added using of libav audio fifo buffer.

Works with latest git master of libav.

Signed-off-by: Michel Verbraak <info@1st-setup.nl>
This commit is contained in:
Michel Verbraak 2014-09-09 18:55:25 +02:00 committed by Jaroslav Kysela
parent 5f0cc2c2dd
commit 4e098bd877
2 changed files with 646 additions and 55 deletions

4
configure vendored
View file

@ -295,6 +295,10 @@ if enabled_or_auto libav; then
has_libav=false
fi
if $has_libav && ! check_pkg libavresample ">=2.1.0"; then
has_libav=false
fi
if $has_libav; then
enable libav
elif enabled libav; then

View file

@ -20,6 +20,14 @@
#include <libavformat/avformat.h>
#include <libavcodec/avcodec.h>
#include <libswscale/swscale.h>
#if LIBAVCODEC_VERSION_MAJOR > 54 || (LIBAVCODEC_VERSION_MAJOR == 54 && LIBAVCODEC_VERSION_MINOR >= 25)
#include <libavresample/avresample.h>
#include <libavutil/opt.h>
#include <libavutil/mathematics.h>
#include <libavutil/imgutils.h>
#include <libavutil/frame.h>
#include <libavutil/audio_fifo.h>
#endif
#include <libavutil/dict.h>
#include <libavutil/audioconvert.h>
@ -65,6 +73,12 @@ typedef struct audio_stream {
int8_t aud_channels;
int32_t aud_bitrate;
#if LIBAVCODEC_VERSION_MAJOR > 54 || (LIBAVCODEC_VERSION_MAJOR == 54 && LIBAVCODEC_VERSION_MINOR >= 25)
AVAudioResampleContext *resample_context;
AVAudioFifo *fifo;
int resample;
#endif
} audio_stream_t;
@ -110,7 +124,7 @@ typedef struct transcoder {
#define WORKING_ENCODER(x) (x == AV_CODEC_ID_H264 || x == AV_CODEC_ID_MPEG2VIDEO || \
x == AV_CODEC_ID_VP8 || x == AV_CODEC_ID_AAC || \
x == AV_CODEC_ID_MP2 || x == AV_CODEC_ID_VORBIS)
x == AV_CODEC_ID_MP2 || x == AV_CODEC_ID_VORBIS)
/**
@ -248,6 +262,7 @@ transcoder_stream_subtitle(transcoder_stream_t *ts, th_pkt_t *pkt)
}
#if LIBAVCODEC_VERSION_MAJOR < 54 || (LIBAVCODEC_VERSION_MAJOR == 54 && LIBAVCODEC_VERSION_MINOR < 25)
/**
*
*/
@ -383,6 +398,7 @@ transcoder_stream_audio(transcoder_stream_t *ts, th_pkt_t *pkt)
octx->flags |= CODEC_FLAG_GLOBAL_HEADER;
octx->channels = 2; // Only stereo suported by libavcodec
octx->global_quality = 4*FF_QP2LAMBDA;
break;
default:
@ -448,7 +464,589 @@ transcoder_stream_audio(transcoder_stream_t *ts, th_pkt_t *pkt)
av_free_packet(&packet);
pkt_ref_dec(pkt);
}
#else
static char *const get_error_text(const int error)
{
static char error_buffer[255];
av_strerror(error, error_buffer, sizeof(error_buffer));
return error_buffer;
}
/**
*
*/
static void
transcoder_stream_audio(transcoder_stream_t *ts, th_pkt_t *pkt)
{
AVCodec *icodec, *ocodec;
AVCodecContext *ictx, *octx;
AVPacket packet;
int length, len;
uint32_t frame_bytes;
streaming_message_t *sm;
th_pkt_t *n;
audio_stream_t *as = (audio_stream_t*)ts;
int got_frame, got_packet_ptr;
AVFrame *frame = av_frame_alloc();
uint8_t **output = NULL;
ictx = as->aud_ictx;
octx = as->aud_octx;
icodec = as->aud_icodec;
ocodec = as->aud_ocodec;
if (ictx->codec_id == AV_CODEC_ID_NONE) {
ictx->codec_id = icodec->id;
if (avcodec_open2(ictx, icodec, NULL) < 0) {
tvhlog(LOG_ERR, "transcode", "Unable to open %s decoder", icodec->name);
ts->ts_index = 0;
goto cleanup;
}
as->aud_dec_pts = pkt->pkt_pts;
}
if (pkt->pkt_pts > as->aud_dec_pts) {
tvhlog(LOG_WARNING, "transcode", "Detected framedrop in audio");
as->aud_enc_pts += (pkt->pkt_pts - as->aud_dec_pts);
as->aud_dec_pts += (pkt->pkt_pts - as->aud_dec_pts);
}
pkt = pkt_merge_header(pkt);
av_init_packet(&packet);
packet.data = pktbuf_ptr(pkt->pkt_payload);
packet.size = pktbuf_len(pkt->pkt_payload);
packet.pts = pkt->pkt_pts;
packet.dts = pkt->pkt_dts;
packet.duration = pkt->pkt_duration;
if ((len = as->aud_dec_size - as->aud_dec_offset) <= 0) {
tvhlog(LOG_ERR, "transcode", "Decoder buffer overflow");
ts->ts_index = 0;
goto cleanup;
}
length = avcodec_decode_audio4( ictx
, frame
, &got_frame
, &packet);
av_free_packet(&packet);
tvhlog(LOG_DEBUG, "transcode", "Decoded packet. length-consumed=%d from in-length=%zu, got_frame=%d.\n", length, pktbuf_len(pkt->pkt_payload),got_frame);
if (!got_frame) {
tvhlog(LOG_DEBUG, "transcode", "Did not have a full frame in the packet.\n");
}
if (length < 0) {
tvhlog(LOG_ERR, "transcode", "length < 0.\n");
ts->ts_index = 0;
goto cleanup;
}
if (length != pktbuf_len(pkt->pkt_payload)) {
tvhlog(LOG_DEBUG, "transcode", "Not all data from packet was decoded. length=%d from packet.size=%zu.\n", length, pktbuf_len(pkt->pkt_payload));
}
if (length == 0 || !got_frame) {
tvhlog(LOG_DEBUG, "transcode", "Not yet enough data for decoding.");
goto cleanup;
}
if (octx->codec_id == AV_CODEC_ID_NONE) {
octx->sample_rate = ictx->sample_rate;
octx->sample_fmt = ictx->sample_fmt;
if (ocodec->sample_fmts) {
// Find if we have a matching sample_fmt;
int acount = 0;
octx->sample_fmt = -1;
while ((octx->sample_fmt == -1) && (ocodec->sample_fmts[acount] > -1)) {
if (ocodec->sample_fmts[acount] == ictx->sample_fmt)
octx->sample_fmt = ictx->sample_fmt;
acount++;
}
if (octx->sample_fmt == -1) {
if (acount > 0) {
tvhlog(LOG_DEBUG, "transcode", "Did not find matching sample_fmt for encoder. Will use first supported: %d.", ocodec->sample_fmts[acount-1]);
octx->sample_fmt = ocodec->sample_fmts[acount-1];
}
else {
tvhlog(LOG_ERR, "transcode", "Encoder does not support a sample_fmt!!??.");
ts->ts_index = 0;
goto cleanup;
}
}
else {
tvhlog(LOG_DEBUG, "transcode", "Encoder supports same sample_fmt as decoder.");
}
}
octx->time_base.den = 90000;
octx->time_base.num = 1;
octx->channels = as->aud_channels ? as->aud_channels : ictx->channels;
octx->bit_rate = as->aud_bitrate ? as->aud_bitrate : ictx->bit_rate;
octx->channels = MIN(octx->channels, ictx->channels);
octx->bit_rate = MIN(octx->bit_rate, ictx->bit_rate);
switch (octx->channels) {
case 1:
octx->channel_layout = AV_CH_LAYOUT_MONO;
break;
case 2:
octx->channel_layout = AV_CH_LAYOUT_STEREO;
break;
case 3:
octx->channel_layout = AV_CH_LAYOUT_SURROUND;
break;
case 4:
octx->channel_layout = AV_CH_LAYOUT_QUAD;
break;
case 5:
octx->channel_layout = AV_CH_LAYOUT_5POINT0;
break;
case 6:
octx->channel_layout = AV_CH_LAYOUT_5POINT1;
break;
case 7:
octx->channel_layout = AV_CH_LAYOUT_6POINT1;
break;
case 8:
octx->channel_layout = AV_CH_LAYOUT_7POINT1;
break;
default:
break;
}
if (ocodec->channel_layouts) {
// Find if we have a matching channel_layout;
int acount = 0, maxchannels = 0, maxacount = 0;
octx->channel_layout = 0;
while ((octx->channel_layout == 0) && (ocodec->channel_layouts[acount] > 0)) {
if ((av_get_channel_layout_nb_channels(ocodec->channel_layouts[acount]) >= maxchannels) && (av_get_channel_layout_nb_channels(ocodec->channel_layouts[acount]) <= ictx->channels)) {
maxchannels = av_get_channel_layout_nb_channels(ocodec->channel_layouts[acount]);
maxacount = acount;
}
if (ocodec->channel_layouts[acount] == ictx->channel_layout)
octx->channel_layout = ictx->channel_layout;
acount++;
}
if (octx->channel_layout == 0) {
if (acount > 0) {
// find next which has same or less channels than decoder.
tvhlog(LOG_DEBUG, "transcode", "Did not find matching channel_layout for encoder. Will use last supported: %zu with %d channels.", ocodec->channel_layouts[maxacount], maxchannels);
octx->channel_layout = ocodec->channel_layouts[maxacount];
octx->channels = maxchannels;
}
else {
tvhlog(LOG_ERR, "transcode", "Encoder does not support a channel_layout!!??.");
ts->ts_index = 0;
goto cleanup;
}
}
else {
tvhlog(LOG_DEBUG, "transcode", "Encoder supports same channel_layout as decoder.");
}
}
else {
tvhlog(LOG_DEBUG, "transcode", "Encoder does not show which channel_layouts it supports.");
}
switch (ts->ts_type) {
case SCT_MPEG2AUDIO:
octx->bit_rate = 128000;
break;
case SCT_AAC:
octx->global_quality = 4*FF_QP2LAMBDA;
octx->flags |= CODEC_FLAG_QSCALE;
break;
case SCT_VORBIS:
octx->flags |= CODEC_FLAG_QSCALE;
octx->flags |= CODEC_FLAG_GLOBAL_HEADER;
octx->global_quality = 4*FF_QP2LAMBDA;
break;
default:
break;
}
as->resample = ((ictx->channels != octx->channels) ||
(ictx->channel_layout != octx->channel_layout) ||
(ictx->sample_fmt != octx->sample_fmt) ||
(ictx->sample_rate != octx->sample_rate));
octx->codec_id = ocodec->id;
if (avcodec_open2(octx, ocodec, NULL) < 0) {
tvhlog(LOG_ERR, "transcode", "Unable to open %s encoder", ocodec->name);
ts->ts_index = 0;
goto cleanup;
}
as->resample_context = NULL;
if (as->resample) {
if (!(as->resample_context = avresample_alloc_context())) {
tvhlog(LOG_ERR, "transcode", "Could not allocate resample context");
ts->ts_index = 0;
goto cleanup;
}
// Convert audio
tvhlog(LOG_DEBUG, "transcode", "converting audio");
tvhlog(LOG_DEBUG, "transcode", "ictx->channels:%d", ictx->channels);
tvhlog(LOG_DEBUG, "transcode", "ictx->channel_layout:%" PRIu64, ictx->channel_layout);
tvhlog(LOG_DEBUG, "transcode", "ictx->sample_rate:%d", ictx->sample_rate);
tvhlog(LOG_DEBUG, "transcode", "ictx->sample_fmt:%d", ictx->sample_fmt);
tvhlog(LOG_DEBUG, "transcode", "ictx->bit_rate:%d", ictx->bit_rate);
tvhlog(LOG_DEBUG, "transcode", "octx->channels:%d", octx->channels);
tvhlog(LOG_DEBUG, "transcode", "octx->channel_layout:%" PRIu64, octx->channel_layout);
tvhlog(LOG_DEBUG, "transcode", "octx->sample_rate:%d", octx->sample_rate);
tvhlog(LOG_DEBUG, "transcode", "octx->sample_fmt:%d", octx->sample_fmt);
tvhlog(LOG_DEBUG, "transcode", "octx->bit_rate:%d", octx->bit_rate);
int opt_error;
if ((opt_error = av_opt_set_int(as->resample_context, "in_channel_layout", ictx->channel_layout, 0)) != 0) {
tvhlog(LOG_ERR, "transcode", "Could not set option in_channel_layout (error '%s')",get_error_text(opt_error));
ts->ts_index = 0;
goto cleanup;
}
if ((opt_error = av_opt_set_int(as->resample_context, "out_channel_layout", octx->channel_layout, 0)) != 0) {
tvhlog(LOG_ERR, "transcode", "Could not set option out_channel_layout (error '%s')",get_error_text(opt_error));
ts->ts_index = 0;
goto cleanup;
}
if ((opt_error = av_opt_set_int(as->resample_context, "in_sample_rate", ictx->sample_rate, 0)) != 0) {
tvhlog(LOG_ERR, "transcode", "Could not set option in_sample_rate (error '%s')",get_error_text(opt_error));
ts->ts_index = 0;
goto cleanup;
}
if ((opt_error = av_opt_set_int(as->resample_context, "out_sample_rate", octx->sample_rate, 0)) != 0) {
tvhlog(LOG_ERR, "transcode", "Could not set option out_sample_rate (error '%s')",get_error_text(opt_error));
ts->ts_index = 0;
goto cleanup;
}
if ((opt_error = av_opt_set_int(as->resample_context, "in_sample_fmt", ictx->sample_fmt, 0)) != 0) {
tvhlog(LOG_ERR, "transcode", "Could not set option in_sample_fmt (error '%s')",get_error_text(opt_error));
ts->ts_index = 0;
goto cleanup;
}
if ((opt_error = av_opt_set_int(as->resample_context, "out_sample_fmt", octx->sample_fmt, 0)) != 0) {
tvhlog(LOG_ERR, "transcode", "Could not set option out_sample_fmt (error '%s')",get_error_text(opt_error));
ts->ts_index = 0;
goto cleanup;
}
if (avresample_open(as->resample_context) < 0) {
tvhlog(LOG_ERR, "transcode", "Error avresample_open.\n");
ts->ts_index = 0;
goto cleanup;
}
}
as->fifo = av_audio_fifo_alloc(octx->sample_fmt, octx->channels, 1);
if (!as->fifo) {
tvhlog(LOG_ERR, "transcode", "Could not allocate fifo");
ts->ts_index = 0;
goto cleanup;
}
}
if (as->resample) {
if (!(output = calloc(octx->channels, sizeof(output)))) {
tvhlog(LOG_ERR, "transcode", "Could not allocate converted input sample pointers");
ts->ts_index = 0;
goto cleanup;
}
if (av_samples_alloc(output, NULL, octx->channels, frame->nb_samples, octx->sample_fmt, 0) < 0) {
tvhlog(LOG_ERR, "transcode", "Could not allocate converted input sample");
ts->ts_index = 0;
goto cleanup;
}
if (!output) {
tvhlog(LOG_ERR, "transcode", "Could not allocate memory for converted input sample pointers");
ts->ts_index = 0;
goto cleanup;
}
int out_samples = avresample_convert(as->resample_context, NULL, 0, frame->nb_samples,
frame->extended_data, 0, frame->nb_samples);
tvhlog(LOG_DEBUG, "transcode", "after avresample_convert. out_samples=%d",out_samples);
while (avresample_available(as->resample_context) > 0) {
out_samples = avresample_read(as->resample_context, output, frame->nb_samples);
if (out_samples > 0) {
if (av_audio_fifo_realloc(as->fifo, av_audio_fifo_size(as->fifo) + out_samples) < 0) {
tvhlog(LOG_ERR, "transcode", "Could not reallocate FIFO.");
ts->ts_index = 0;
goto cleanup;
}
if (av_audio_fifo_write(as->fifo, (void **)output, out_samples) < out_samples) {
tvhlog(LOG_ERR, "transcode", "Could not write to FIFO.");
ts->ts_index = 0;
goto cleanup;
}
}
}
/* Need to find out where we are going to do this. Normally at the end.
int delay_samples = avresample_get_delay(as->resample_context);
if (delay_samples) {
tvhlog(LOG_DEBUG, "transcode", "%d samples in resamples delay buffer.", delay_samples);
av_freep(output);
goto cleanup;
}
*/
}
else {
tvhlog(LOG_DEBUG, "transcode", "No conversion needed");
if (av_audio_fifo_realloc(as->fifo, av_audio_fifo_size(as->fifo) + frame->nb_samples) < 0) {
tvhlog(LOG_ERR, "transcode", "Could not reallocate FIFO.");
ts->ts_index = 0;
goto cleanup;
}
if (av_audio_fifo_write(as->fifo, (void **)frame->extended_data, frame->nb_samples) < frame->nb_samples) {
tvhlog(LOG_ERR, "transcode", "Could not write to FIFO.");
ts->ts_index = 0;
goto cleanup;
}
}
as->aud_dec_pts += pkt->pkt_duration;
frame_bytes = av_get_bytes_per_sample(octx->sample_fmt) * octx->frame_size * octx->channels;
while (av_audio_fifo_size(as->fifo) >= octx->frame_size) {
tvhlog(LOG_DEBUG, "transcode", "start encoding loop: av_audio_fifo_size:%d, octx->frame_size=%d", av_audio_fifo_size(as->fifo), octx->frame_size);
av_frame_free(&frame);
frame = av_frame_alloc();
frame->nb_samples = octx->frame_size;
frame->format = octx->sample_fmt;
frame->channel_layout = octx->channel_layout;
frame->sample_rate = octx->sample_rate;
if (av_frame_get_buffer(frame, 0) < 0) {
tvhlog(LOG_ERR, "transcode", "Could not allocate output frame samples.");
ts->ts_index = 0;
goto cleanup;
}
int samples_read;
if ((samples_read = av_audio_fifo_read(as->fifo, (void **)frame->data, octx->frame_size)) < 0) {
tvhlog(LOG_ERR, "transcode", "Could not read data from FIFO.");
ts->ts_index = 0;
goto cleanup;
}
tvhlog(LOG_DEBUG, "transcode", "before encoding: frame->linesize[0]=%d, samples_read=%d", frame->linesize[0], samples_read);
av_init_packet(&packet);
packet.data = NULL;
packet.size = 0;
int ret = avcodec_encode_audio2(octx,
&packet,
frame,
&got_packet_ptr);
tvhlog(LOG_DEBUG, "transcode", "encoded: packet.size=%d, ret=%d, got_packet_ptr=%d", packet.size, ret, got_packet_ptr);
if ((ret < 0) || (got_packet_ptr < -1)) {
tvhlog(LOG_ERR, "transcode", "Unable to encode audio (%d:%d)", ret, got_packet_ptr);
ts->ts_index = 0;
goto cleanup;
} else if (got_packet_ptr) {
length = packet.size;
tvhlog(LOG_DEBUG, "transcode", "encoded: packet.pts=%" PRIu64 ", packet.dts=%" PRIu64 ", as->aud_enc_pts=%lu", packet.pts, packet.dts, as->aud_enc_pts);
n = pkt_alloc(packet.data, packet.size, as->aud_enc_pts, as->aud_enc_pts);
n->pkt_componentindex = ts->ts_index;
n->pkt_frametype = pkt->pkt_frametype;
n->pkt_channels = octx->channels;
n->pkt_sri = pkt->pkt_sri;
if (octx->coded_frame && octx->coded_frame->pts != AV_NOPTS_VALUE)
n->pkt_duration = octx->coded_frame->pts - as->aud_enc_pts;
else
n->pkt_duration = frame_bytes*90000 / (2 * octx->channels * octx->sample_rate);
as->aud_enc_pts += n->pkt_duration;
if (octx->extradata_size)
n->pkt_header = pktbuf_alloc(octx->extradata, octx->extradata_size);
sm = streaming_msg_create_pkt(n);
streaming_target_deliver2(ts->ts_target, sm);
pkt_ref_dec(n);
}
av_free_packet(&packet);
}
cleanup:
if ((output) && (output[0]))
av_freep(&(output)[0]);
if (output)
av_freep(output);
av_frame_free(&frame);
av_free_packet(&packet);
pkt_ref_dec(pkt);
}
#endif
/**
*
*/
static void send_video_packet(transcoder_stream_t *ts, th_pkt_t *pkt, uint8_t *out, int length, AVCodecContext *octx)
{
streaming_message_t *sm;
th_pkt_t *n;
if (length <= 0) {
if (length) {
tvhlog(LOG_ERR, "transcode", "Unable to encode video (%d)", length);
ts->ts_index = 0;
}
return;
}
if (!octx->coded_frame)
return;
n = pkt_alloc(out, length, octx->coded_frame->pkt_pts, octx->coded_frame->pkt_dts);
switch (octx->coded_frame->pict_type) {
case AV_PICTURE_TYPE_I:
n->pkt_frametype = PKT_I_FRAME;
break;
case AV_PICTURE_TYPE_P:
n->pkt_frametype = PKT_P_FRAME;
break;
case AV_PICTURE_TYPE_B:
n->pkt_frametype = PKT_B_FRAME;
break;
default:
break;
}
n->pkt_duration = pkt->pkt_duration;
n->pkt_commercial = pkt->pkt_commercial;
n->pkt_componentindex = pkt->pkt_componentindex;
n->pkt_field = pkt->pkt_field;
n->pkt_aspect_num = pkt->pkt_aspect_num;
n->pkt_aspect_den = pkt->pkt_aspect_den;
if(octx->coded_frame && octx->coded_frame->pts != AV_NOPTS_VALUE) {
if(n->pkt_dts != PTS_UNSET)
n->pkt_dts -= n->pkt_pts;
n->pkt_pts = octx->coded_frame->pts;
if(n->pkt_dts != PTS_UNSET)
n->pkt_dts += n->pkt_pts;
}
if (octx->extradata_size)
n->pkt_header = pktbuf_alloc(octx->extradata, octx->extradata_size);
else {
if (octx->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
uint32_t *mpeg2_header = (uint32_t *)out;
if (*mpeg2_header == 0xb3010000) { // SEQ_START_CODE
// Need to determine lentgh of header.
/*
From: http://en.wikipedia.org/wiki/Elementary_stream
Field Name # of bits Description
start code 32 0x000001B3
Horizontal Size 12
Vertical Size 12
Aspect ratio 4
Frame rate code 4
Bit rate 18 Actual bit rate = bit rate * 400, rounded upwards. Use 0x3FFFF for variable bit rate.
Marker bit 1 Always 1.
VBV buf size 10 Size of video buffer verifier = 16*1024*vbv buf size
constrained parameters flag 1
load intra quantizer matrix 1 If bit set then intra quantizer matrix follows, otherwise use default values.
intra quantizer matrix 0 or 64*8
load non intra quantizer matrix 1 If bit set then non intra quantizer matrix follows.
non intra quantizer matrix 0 or 64*8
Minimal of 12 bytes.
*/
int header_size = 12;
// load intra quantizer matrix
uint8_t matrix_enabled = (((uint8_t)*(out+(header_size-1)) & 0x02) == 0x02);
if (matrix_enabled)
header_size += 64;
//load non intra quantizer matrix
matrix_enabled = (((uint8_t)*(out+(header_size-1)) & 0x01) == 0x01);
if (matrix_enabled)
header_size += 64;
// See if we have the first EXT_START_CODE. Normally 10 bytes
// https://git.libav.org/?p=libav.git;a=blob;f=libavcodec/mpeg12enc.c;h=3376f1075f4b7582a8e4556e98deddab3e049dab;hb=HEAD#l272
mpeg2_header = (uint32_t *)(out+(header_size));
if (*mpeg2_header == 0xb5010000) { // EXT_START_CODE
header_size += 10;
// See if we have the second EXT_START_CODE. Normally 12 bytes
// https://git.libav.org/?p=libav.git;a=blob;f=libavcodec/mpeg12enc.c;h=3376f1075f4b7582a8e4556e98deddab3e049dab;hb=HEAD#l291
mpeg2_header = (uint32_t *)(out+(header_size));
if (*mpeg2_header == 0xb5010000) { // EXT_START_CODE
header_size += 12;
// See if we have the second GOP_START_CODE. Normally 31 bits == 4 bytes
// https://git.libav.org/?p=libav.git;a=blob;f=libavcodec/mpeg12enc.c;h=3376f1075f4b7582a8e4556e98deddab3e049dab;hb=HEAD#l304
mpeg2_header = (uint32_t *)(out+(header_size));
if (*mpeg2_header == 0xb8010000) // GOP_START_CODE
header_size += 4;
}
}
n->pkt_header = pktbuf_alloc(out, header_size);
}
}
}
sm = streaming_msg_create_pkt(n);
streaming_target_deliver2(ts->ts_target, sm);
pkt_ref_dec(n);
}
/**
*
@ -463,8 +1061,6 @@ transcoder_stream_video(transcoder_stream_t *ts, th_pkt_t *pkt)
AVPicture deint_pic;
uint8_t *buf, *out, *deint;
int length, len, got_picture;
streaming_message_t *sm;
th_pkt_t *n;
video_stream_t *vs = (video_stream_t*)ts;
ictx = vs->vid_ictx;
@ -655,10 +1251,6 @@ transcoder_stream_video(transcoder_stream_t *ts, th_pkt_t *pkt)
}
len = avpicture_get_size(octx->pix_fmt, ictx->width, ictx->height);
out = av_malloc(len + FF_INPUT_BUFFER_PADDING_SIZE);
memset(out, 0, len);
vs->vid_enc_frame->pkt_pts = vs->vid_dec_frame->pkt_pts;
vs->vid_enc_frame->pkt_dts = vs->vid_dec_frame->pkt_dts;
@ -667,62 +1259,38 @@ transcoder_stream_video(transcoder_stream_t *ts, th_pkt_t *pkt)
else if (ictx->coded_frame && ictx->coded_frame->pts != AV_NOPTS_VALUE)
vs->vid_enc_frame->pts = vs->vid_dec_frame->pts;
#if LIBAVCODEC_VERSION_MAJOR < 54 || (LIBAVCODEC_VERSION_MAJOR == 54 && LIBAVCODEC_VERSION_MINOR < 25)
len = avpicture_get_size(octx->pix_fmt, ictx->width, ictx->height);
out = av_malloc(len + FF_INPUT_BUFFER_PADDING_SIZE);
memset(out, 0, len);
length = avcodec_encode_video(octx, out, len, vs->vid_enc_frame);
if (length <= 0) {
if (length) {
tvhlog(LOG_ERR, "transcode", "Unable to encode video (%d)", length);
ts->ts_index = 0;
}
send_video_packet(ts, pkt, out, length, octx);
#else
AVPacket packet2;
int ret, got_output;
av_init_packet(&packet2);
packet2.data = NULL; // packet data will be allocated by the encoder
packet2.size = 0;
ret = avcodec_encode_video2(octx, &packet2, vs->vid_enc_frame, &got_output);
if (ret < 0) {
tvhlog(LOG_ERR, "transcode", "Error encoding frame. avcodec_encode_video2()");
av_free_packet(&packet2);
ts->ts_index = 0;
goto cleanup;
}
if (!octx->coded_frame)
goto cleanup;
n = pkt_alloc(out, length, octx->coded_frame->pkt_pts, octx->coded_frame->pkt_dts);
switch (octx->coded_frame->pict_type) {
case AV_PICTURE_TYPE_I:
n->pkt_frametype = PKT_I_FRAME;
break;
case AV_PICTURE_TYPE_P:
n->pkt_frametype = PKT_P_FRAME;
break;
case AV_PICTURE_TYPE_B:
n->pkt_frametype = PKT_B_FRAME;
break;
default:
break;
if (got_output) {
send_video_packet(ts, pkt, packet2.data, packet2.size, octx);
}
n->pkt_duration = pkt->pkt_duration;
n->pkt_commercial = pkt->pkt_commercial;
n->pkt_componentindex = pkt->pkt_componentindex;
n->pkt_field = pkt->pkt_field;
n->pkt_aspect_num = pkt->pkt_aspect_num;
n->pkt_aspect_den = pkt->pkt_aspect_den;
if(octx->coded_frame && octx->coded_frame->pts != AV_NOPTS_VALUE) {
if(n->pkt_dts != PTS_UNSET)
n->pkt_dts -= n->pkt_pts;
n->pkt_pts = octx->coded_frame->pts;
if(n->pkt_dts != PTS_UNSET)
n->pkt_dts += n->pkt_pts;
}
if (octx->extradata_size)
n->pkt_header = pktbuf_alloc(octx->extradata, octx->extradata_size);
sm = streaming_msg_create_pkt(n);
streaming_target_deliver2(ts->ts_target, sm);
pkt_ref_dec(n);
av_free_packet(&packet2);
#endif
cleanup:
av_free_packet(&packet);
@ -898,6 +1466,13 @@ transcoder_destroy_audio(transcoder_stream_t *ts)
if(as->aud_enc_sample)
av_free(as->aud_enc_sample);
#if LIBAVCODEC_VERSION_MAJOR > 54 || (LIBAVCODEC_VERSION_MAJOR == 54 && LIBAVCODEC_VERSION_MINOR >= 25)
if ((as->resample_context) && (avresample_is_open(as->resample_context)) )
avresample_close(as->resample_context);
av_audio_fifo_free(as->fifo);
#endif
free(ts);
}
@ -975,6 +1550,13 @@ transcoder_init_audio(transcoder_t *t, streaming_start_component_t *ssc)
as->aud_channels = 0;
as->aud_bitrate = as->aud_channels * 64000;
#if LIBAVCODEC_VERSION_MAJOR > 54 || (LIBAVCODEC_VERSION_MAJOR == 54 && LIBAVCODEC_VERSION_MINOR >= 25)
as->resample_context = NULL;
as->fifo = NULL;
as->resample = 0;
#endif
return 1;
}
@ -1102,6 +1684,8 @@ transcoder_calc_stream_count(transcoder_t *t, streaming_start_t *ss) {
int subtitle = 0;
streaming_start_component_t *ssc = NULL;
tvhlog(LOG_DEBUG, "transcoder", "transcoder_calc_stream_count: ss->ss_num_components=%d", ss->ss_num_components);
for (i = 0; i < ss->ss_num_components; i++) {
ssc = &ss->ss_components[i];
@ -1134,6 +1718,9 @@ transcoder_calc_stream_count(transcoder_t *t, streaming_start_t *ss) {
}
}
tvhlog(LOG_DEBUG, "transcoder", "transcoder_calc_stream_count=%d", (video + audio + subtitle));
return (video + audio + subtitle);
}