avconv: split audio transcoding out of output_packet().

This commit is contained in:
Anton Khirnov 2011-11-21 14:39:22 +01:00
parent 78162b4ea2
commit ded28ba35b
1 changed files with 114 additions and 95 deletions

209
avconv.c
View File

@ -1615,6 +1615,109 @@ static void rate_emu_sleep(InputStream *ist)
}
}
static int transcode_audio(InputStream *ist, AVPacket *pkt, int *got_output)
{
static unsigned int samples_size = 0;
int bps = av_get_bytes_per_sample(ist->st->codec->sample_fmt);
uint8_t *decoded_data_buf = NULL;
int decoded_data_size = 0;
int i, ret;
if (pkt && samples_size < FFMAX(pkt->size * bps, AVCODEC_MAX_AUDIO_FRAME_SIZE)) {
av_free(samples);
samples_size = FFMAX(pkt->size * bps, AVCODEC_MAX_AUDIO_FRAME_SIZE);
samples = av_malloc(samples_size);
}
decoded_data_size = samples_size;
ret = avcodec_decode_audio3(ist->st->codec, samples, &decoded_data_size,
pkt);
if (ret < 0)
return ret;
pkt->data += ret;
pkt->size -= ret;
*got_output = decoded_data_size > 0;
/* Some bug in mpeg audio decoder gives */
/* decoded_data_size < 0, it seems they are overflows */
if (!*got_output) {
/* no audio frame */
return 0;
}
decoded_data_buf = (uint8_t *)samples;
ist->next_pts += ((int64_t)AV_TIME_BASE/bps * decoded_data_size) /
(ist->st->codec->sample_rate * ist->st->codec->channels);
// preprocess audio (volume)
if (audio_volume != 256) {
switch (ist->st->codec->sample_fmt) {
case AV_SAMPLE_FMT_U8:
{
uint8_t *volp = samples;
for (i = 0; i < (decoded_data_size / sizeof(*volp)); i++) {
int v = (((*volp - 128) * audio_volume + 128) >> 8) + 128;
*volp++ = av_clip_uint8(v);
}
break;
}
case AV_SAMPLE_FMT_S16:
{
int16_t *volp = samples;
for (i = 0; i < (decoded_data_size / sizeof(*volp)); i++) {
int v = ((*volp) * audio_volume + 128) >> 8;
*volp++ = av_clip_int16(v);
}
break;
}
case AV_SAMPLE_FMT_S32:
{
int32_t *volp = samples;
for (i = 0; i < (decoded_data_size / sizeof(*volp)); i++) {
int64_t v = (((int64_t)*volp * audio_volume + 128) >> 8);
*volp++ = av_clipl_int32(v);
}
break;
}
case AV_SAMPLE_FMT_FLT:
{
float *volp = samples;
float scale = audio_volume / 256.f;
for (i = 0; i < (decoded_data_size / sizeof(*volp)); i++) {
*volp++ *= scale;
}
break;
}
case AV_SAMPLE_FMT_DBL:
{
double *volp = samples;
double scale = audio_volume / 256.;
for (i = 0; i < (decoded_data_size / sizeof(*volp)); i++) {
*volp++ *= scale;
}
break;
}
default:
av_log(NULL, AV_LOG_FATAL,
"Audio volume adjustment on sample format %s is not supported.\n",
av_get_sample_fmt_name(ist->st->codec->sample_fmt));
exit_program(1);
}
}
rate_emu_sleep(ist);
for (i = 0; i < nb_output_streams; i++) {
OutputStream *ost = &output_streams[i];
if (!check_output_constraints(ist, ost) || !ost->encoding_needed)
continue;
do_audio_out(output_files[ost->file_index].ctx, ost, ist,
decoded_data_buf, decoded_data_size);
}
return 0;
}
/* pkt = NULL means EOF (needed to flush decoder buffers) */
static int output_packet(InputStream *ist, int ist_index,
OutputStream *ost_table, int nb_ostreams,
@ -1625,7 +1728,6 @@ static int output_packet(InputStream *ist, int ist_index,
int ret = 0, i;
int got_output;
void *buffer_to_free = NULL;
static unsigned int samples_size= 0;
AVSubtitle subtitle, *subtitle_to_free;
int64_t pkt_pts = AV_NOPTS_VALUE;
#if CONFIG_AVFILTER
@ -1634,7 +1736,6 @@ static int output_packet(InputStream *ist, int ist_index,
float quality;
AVPacket avpkt;
int bps = av_get_bytes_per_sample(ist->st->codec->sample_fmt);
if(ist->next_pts == AV_NOPTS_VALUE)
ist->next_pts= ist->pts;
@ -1656,8 +1757,6 @@ static int output_packet(InputStream *ist, int ist_index,
//while we have more to decode or while the decoder did output something on EOF
while (ist->decoding_needed && (avpkt.size > 0 || (!pkt && got_output))) {
uint8_t *decoded_data_buf;
int decoded_data_size;
AVFrame *decoded_frame, *filtered_frame;
handle_eof:
ist->pts= ist->next_pts;
@ -1667,38 +1766,19 @@ static int output_packet(InputStream *ist, int ist_index,
"Multiple frames in a packet from stream %d\n", pkt->stream_index);
ist->showed_multi_packet_warning=1;
/* decode the packet if needed */
decoded_frame = filtered_frame = NULL;
decoded_data_buf = NULL; /* fail safe */
decoded_data_size= 0;
subtitle_to_free = NULL;
switch(ist->st->codec->codec_type) {
case AVMEDIA_TYPE_AUDIO:{
if(pkt && samples_size < FFMAX(pkt->size * bps, AVCODEC_MAX_AUDIO_FRAME_SIZE)) {
samples_size = FFMAX(pkt->size * bps, AVCODEC_MAX_AUDIO_FRAME_SIZE);
av_free(samples);
samples= av_malloc(samples_size);
}
decoded_data_size= samples_size;
/* XXX: could avoid copy if PCM 16 bits with same
endianness as CPU */
ret = avcodec_decode_audio3(ist->st->codec, samples, &decoded_data_size,
&avpkt);
// XXX temporary hack, will be turned to a switch() once all codec
// types are split out
if (ist->st->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
ret = transcode_audio(ist, &avpkt, &got_output);
if (ret < 0)
return ret;
avpkt.data += ret;
avpkt.size -= ret;
got_output = decoded_data_size > 0;
/* Some bug in mpeg audio decoder gives */
/* decoded_data_size < 0, it seems they are overflows */
if (!got_output) {
/* no audio frame */
continue;
}
decoded_data_buf = (uint8_t *)samples;
ist->next_pts += ((int64_t)AV_TIME_BASE/bps * decoded_data_size) /
(ist->st->codec->sample_rate * ist->st->codec->channels);
break;}
continue;
}
/* decode the packet if needed */
decoded_frame = filtered_frame = NULL;
subtitle_to_free = NULL;
switch(ist->st->codec->codec_type) {
case AVMEDIA_TYPE_VIDEO:
if (!(decoded_frame = avcodec_alloc_frame()))
return AVERROR(ENOMEM);
@ -1743,64 +1823,6 @@ static int output_packet(InputStream *ist, int ist_index,
return -1;
}
// preprocess audio (volume)
if (ist->st->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
if (audio_volume != 256) {
switch (ist->st->codec->sample_fmt) {
case AV_SAMPLE_FMT_U8:
{
uint8_t *volp = samples;
for (i = 0; i < (decoded_data_size / sizeof(*volp)); i++) {
int v = (((*volp - 128) * audio_volume + 128) >> 8) + 128;
*volp++ = av_clip_uint8(v);
}
break;
}
case AV_SAMPLE_FMT_S16:
{
int16_t *volp = samples;
for (i = 0; i < (decoded_data_size / sizeof(*volp)); i++) {
int v = ((*volp) * audio_volume + 128) >> 8;
*volp++ = av_clip_int16(v);
}
break;
}
case AV_SAMPLE_FMT_S32:
{
int32_t *volp = samples;
for (i = 0; i < (decoded_data_size / sizeof(*volp)); i++) {
int64_t v = (((int64_t)*volp * audio_volume + 128) >> 8);
*volp++ = av_clipl_int32(v);
}
break;
}
case AV_SAMPLE_FMT_FLT:
{
float *volp = samples;
float scale = audio_volume / 256.f;
for (i = 0; i < (decoded_data_size / sizeof(*volp)); i++) {
*volp++ *= scale;
}
break;
}
case AV_SAMPLE_FMT_DBL:
{
double *volp = samples;
double scale = audio_volume / 256.;
for (i = 0; i < (decoded_data_size / sizeof(*volp)); i++) {
*volp++ *= scale;
}
break;
}
default:
av_log(NULL, AV_LOG_FATAL,
"Audio volume adjustment on sample format %s is not supported.\n",
av_get_sample_fmt_name(ist->st->codec->sample_fmt));
exit_program(1);
}
}
}
/* frame rate emulation */
rate_emu_sleep(ist);
@ -1846,9 +1868,6 @@ static int output_packet(InputStream *ist, int ist_index,
av_assert0(ist->decoding_needed);
switch(ost->st->codec->codec_type) {
case AVMEDIA_TYPE_AUDIO:
do_audio_out(os, ost, ist, decoded_data_buf, decoded_data_size);
break;
case AVMEDIA_TYPE_VIDEO:
#if CONFIG_AVFILTER
if (ost->picref->video && !ost->frame_aspect_ratio)