mirror of
https://gitlab.freedesktop.org/gstreamer/gstreamer.git
synced 2025-02-22 05:56:31 +00:00
ffmpeg: rewrite audio encoders.
Rewrite the audio encoders to use the right API functions of ffmpeg. Also get rid of the handrolled cache and use adapter instead for formats that require fixed frame_size as input.
This commit is contained in:
Wim Taymans
Wim Taymans
parent
92b8a5adc9
commit
25b41829b9
2 changed files with 172 additions and 89 deletions
|
|
@ -255,6 +255,8 @@ gst_ffmpegenc_init (GstFFMpegEnc * ffmpegenc)
|
|||
|
||||
gst_element_add_pad (GST_ELEMENT (ffmpegenc), ffmpegenc->sinkpad);
|
||||
gst_element_add_pad (GST_ELEMENT (ffmpegenc), ffmpegenc->srcpad);
|
||||
|
||||
ffmpegenc->adapter = gst_adapter_new ();
|
||||
}
|
||||
|
||||
static void
|
||||
|
|
@ -277,6 +279,8 @@ gst_ffmpegenc_finalize (GObject * object)
|
|||
g_queue_free (ffmpegenc->delay);
|
||||
g_free (ffmpegenc->filename);
|
||||
|
||||
g_object_unref (ffmpegenc->adapter);
|
||||
|
||||
G_OBJECT_CLASS (parent_class)->finalize (object);
|
||||
}
|
||||
|
||||
|
|
@ -698,103 +702,179 @@ gst_ffmpegenc_chain_video (GstPad * pad, GstBuffer * inbuf)
|
|||
}
|
||||
|
||||
static GstFlowReturn
|
||||
gst_ffmpegenc_chain_audio (GstPad * pad, GstBuffer * inbuf)
|
||||
gst_ffmpegenc_encode_audio (GstFFMpegEnc * ffmpegenc, GstBuffer * inbuf,
|
||||
gint max_size, gboolean discont)
|
||||
{
|
||||
GstBuffer *outbuf = NULL, *subbuf;
|
||||
GstFFMpegEnc *ffmpegenc = (GstFFMpegEnc *) (GST_OBJECT_PARENT (pad));
|
||||
gint size, ret_size = 0, in_size, frame_size;
|
||||
GstBuffer *outbuf;
|
||||
AVCodecContext *ctx;
|
||||
guint8 *audio_out, *audio_in;
|
||||
GstClockTime timestamp, duration;
|
||||
gint res;
|
||||
GstFlowReturn ret;
|
||||
|
||||
size = GST_BUFFER_SIZE (inbuf);
|
||||
ctx = ffmpegenc->context;
|
||||
|
||||
/* FIXME: events (discont (flush!) and eos (close down) etc.) */
|
||||
/* caller should set timestamps on inbuf */
|
||||
timestamp = GST_BUFFER_TIMESTAMP (inbuf);
|
||||
duration = GST_BUFFER_DURATION (inbuf);
|
||||
audio_in = GST_BUFFER_DATA (inbuf);
|
||||
|
||||
frame_size = ffmpegenc->context->frame_size * 2 *
|
||||
ffmpegenc->context->channels;
|
||||
in_size = size;
|
||||
if (ffmpegenc->cache)
|
||||
in_size += GST_BUFFER_SIZE (ffmpegenc->cache);
|
||||
outbuf = gst_buffer_new_and_alloc (max_size);
|
||||
audio_out = GST_BUFFER_DATA (outbuf);
|
||||
|
||||
GST_DEBUG_OBJECT (ffmpegenc,
|
||||
"Received buffer of time %" GST_TIME_FORMAT " and size %d (cache: %d)",
|
||||
GST_TIME_ARGS (GST_BUFFER_TIMESTAMP (inbuf)), size, in_size - size);
|
||||
GST_LOG_OBJECT (ffmpegenc, "encoding buffer of max size %d", max_size);
|
||||
|
||||
while (1) {
|
||||
/* do we have enough data for one frame? */
|
||||
if (in_size / (2 * ffmpegenc->context->channels) <
|
||||
ffmpegenc->context->frame_size) {
|
||||
if (in_size > size) {
|
||||
/* this is panic! we got a buffer, but still don't have enough
|
||||
* data. Merge them and retry in the next cycle... */
|
||||
ffmpegenc->cache = gst_buffer_join (ffmpegenc->cache, inbuf);
|
||||
} else if (in_size == size) {
|
||||
/* exactly the same! how wonderful */
|
||||
ffmpegenc->cache = inbuf;
|
||||
} else if (in_size > 0) {
|
||||
ffmpegenc->cache = gst_buffer_create_sub (inbuf, size - in_size,
|
||||
in_size);
|
||||
GST_BUFFER_DURATION (ffmpegenc->cache) =
|
||||
GST_BUFFER_DURATION (inbuf) * GST_BUFFER_SIZE (ffmpegenc->cache) /
|
||||
size;
|
||||
GST_BUFFER_TIMESTAMP (ffmpegenc->cache) =
|
||||
GST_BUFFER_TIMESTAMP (inbuf) +
|
||||
(GST_BUFFER_DURATION (inbuf) * (size - in_size) / size);
|
||||
gst_buffer_unref (inbuf);
|
||||
} else {
|
||||
gst_buffer_unref (inbuf);
|
||||
}
|
||||
return GST_FLOW_OK;
|
||||
}
|
||||
res = avcodec_encode_audio (ctx, audio_out, max_size, (short *) audio_in);
|
||||
gst_buffer_unref (inbuf);
|
||||
|
||||
/* create the frame */
|
||||
if (in_size > size) {
|
||||
/* merge */
|
||||
subbuf = gst_buffer_create_sub (inbuf, 0, frame_size - (in_size - size));
|
||||
GST_BUFFER_DURATION (subbuf) =
|
||||
GST_BUFFER_DURATION (inbuf) * GST_BUFFER_SIZE (subbuf) / size;
|
||||
subbuf = gst_buffer_join (ffmpegenc->cache, subbuf);
|
||||
ffmpegenc->cache = NULL;
|
||||
} else {
|
||||
subbuf = gst_buffer_create_sub (inbuf, size - in_size, frame_size);
|
||||
GST_BUFFER_DURATION (subbuf) =
|
||||
GST_BUFFER_DURATION (inbuf) * GST_BUFFER_SIZE (subbuf) / size;
|
||||
GST_BUFFER_TIMESTAMP (subbuf) =
|
||||
GST_BUFFER_TIMESTAMP (inbuf) + (GST_BUFFER_DURATION (inbuf) *
|
||||
(size - in_size) / size);
|
||||
}
|
||||
|
||||
outbuf = gst_buffer_new_and_alloc (GST_BUFFER_SIZE (subbuf));
|
||||
ret_size = avcodec_encode_audio (ffmpegenc->context,
|
||||
GST_BUFFER_DATA (outbuf), GST_BUFFER_SIZE (outbuf), (const short int *)
|
||||
GST_BUFFER_DATA (subbuf));
|
||||
|
||||
if (ret_size < 0) {
|
||||
GST_ERROR_OBJECT (ffmpegenc, "Failed to encode buffer");
|
||||
gst_buffer_unref (inbuf);
|
||||
gst_buffer_unref (outbuf);
|
||||
gst_buffer_unref (subbuf);
|
||||
return GST_FLOW_OK;
|
||||
}
|
||||
|
||||
GST_BUFFER_SIZE (outbuf) = ret_size;
|
||||
GST_BUFFER_TIMESTAMP (outbuf) = GST_BUFFER_TIMESTAMP (subbuf);
|
||||
GST_BUFFER_DURATION (outbuf) = GST_BUFFER_DURATION (subbuf);
|
||||
gst_buffer_set_caps (outbuf, GST_PAD_CAPS (ffmpegenc->srcpad));
|
||||
gst_buffer_unref (subbuf);
|
||||
|
||||
ret = gst_pad_push (ffmpegenc->srcpad, outbuf);
|
||||
if (ret != GST_FLOW_OK) {
|
||||
/* cleanup and return */
|
||||
gst_buffer_unref (inbuf);
|
||||
return ret;
|
||||
}
|
||||
|
||||
in_size -= frame_size;
|
||||
if (res < 0) {
|
||||
GST_ERROR_OBJECT (ffmpegenc, "Failed to encode buffer: %d", res);
|
||||
gst_buffer_unref (outbuf);
|
||||
return GST_FLOW_OK;
|
||||
}
|
||||
GST_LOG_OBJECT (ffmpegenc, "got output size %d", res);
|
||||
|
||||
GST_BUFFER_SIZE (outbuf) = res;
|
||||
GST_BUFFER_TIMESTAMP (outbuf) = timestamp;
|
||||
GST_BUFFER_DURATION (outbuf) = duration;
|
||||
if (discont)
|
||||
GST_BUFFER_FLAG_SET (outbuf, GST_BUFFER_FLAG_DISCONT);
|
||||
gst_buffer_set_caps (outbuf, GST_PAD_CAPS (ffmpegenc->srcpad));
|
||||
|
||||
GST_LOG_OBJECT (ffmpegenc, "pushing size %d, timestamp %" GST_TIME_FORMAT,
|
||||
res, GST_TIME_ARGS (timestamp));
|
||||
|
||||
ret = gst_pad_push (ffmpegenc->srcpad, outbuf);
|
||||
|
||||
return ret;
|
||||
}
|
||||
|
||||
static GstFlowReturn
|
||||
gst_ffmpegenc_chain_audio (GstPad * pad, GstBuffer * inbuf)
|
||||
{
|
||||
GstFFMpegEnc *ffmpegenc;
|
||||
GstFFMpegEncClass *oclass;
|
||||
AVCodecContext *ctx;
|
||||
GstClockTime timestamp, duration;
|
||||
guint size, frame_size;
|
||||
gint osize;
|
||||
GstFlowReturn ret;
|
||||
gint out_size;
|
||||
gboolean discont;
|
||||
|
||||
ffmpegenc = (GstFFMpegEnc *) (GST_OBJECT_PARENT (pad));
|
||||
oclass = (GstFFMpegEncClass *) G_OBJECT_GET_CLASS (ffmpegenc);
|
||||
|
||||
ctx = ffmpegenc->context;
|
||||
|
||||
size = GST_BUFFER_SIZE (inbuf);
|
||||
timestamp = GST_BUFFER_TIMESTAMP (inbuf);
|
||||
duration = GST_BUFFER_DURATION (inbuf);
|
||||
discont = GST_BUFFER_IS_DISCONT (inbuf);
|
||||
|
||||
GST_DEBUG_OBJECT (ffmpegenc,
|
||||
"Received time %" GST_TIME_FORMAT ", duration %" GST_TIME_FORMAT
|
||||
", size %d", GST_TIME_ARGS (timestamp), GST_TIME_ARGS (duration), size);
|
||||
|
||||
frame_size = ctx->frame_size;
|
||||
osize = av_get_bits_per_sample_format (ctx->sample_fmt) / 8;
|
||||
|
||||
if (frame_size > 1) {
|
||||
/* we have a frame_size, feed the encoder multiples of this frame size */
|
||||
guint avail, frame_bytes;
|
||||
|
||||
if (discont) {
|
||||
GST_LOG_OBJECT (ffmpegenc, "DISCONT, clear adapter");
|
||||
gst_adapter_clear (ffmpegenc->adapter);
|
||||
ffmpegenc->discont = TRUE;
|
||||
}
|
||||
|
||||
if (gst_adapter_available (ffmpegenc->adapter) == 0) {
|
||||
/* lock on to new timestamp */
|
||||
GST_LOG_OBJECT (ffmpegenc, "taking buffer timestamp %" GST_TIME_FORMAT,
|
||||
GST_TIME_ARGS (timestamp));
|
||||
ffmpegenc->adapter_ts = timestamp;
|
||||
ffmpegenc->adapter_consumed = 0;
|
||||
} else {
|
||||
/* use timestamp at head of the adapter */
|
||||
GST_LOG_OBJECT (ffmpegenc, "taking adapter timestamp %" GST_TIME_FORMAT,
|
||||
GST_TIME_ARGS (ffmpegenc->adapter_ts));
|
||||
timestamp = ffmpegenc->adapter_ts;
|
||||
timestamp +=
|
||||
gst_util_uint64_scale (ffmpegenc->adapter_consumed, GST_SECOND,
|
||||
ctx->sample_rate);
|
||||
}
|
||||
|
||||
GST_LOG_OBJECT (ffmpegenc, "pushing buffer in adapter");
|
||||
gst_adapter_push (ffmpegenc->adapter, inbuf);
|
||||
|
||||
/* first see how many bytes we need to feed to the decoder. */
|
||||
frame_bytes = frame_size * osize * ctx->channels;
|
||||
avail = gst_adapter_available (ffmpegenc->adapter);
|
||||
|
||||
GST_LOG_OBJECT (ffmpegenc, "frame_bytes %u, avail %u", frame_bytes, avail);
|
||||
|
||||
/* while there is more than a frame size in the adapter, consume it */
|
||||
while (avail >= frame_bytes) {
|
||||
GST_LOG_OBJECT (ffmpegenc, "taking %u bytes from the adapter",
|
||||
frame_bytes);
|
||||
|
||||
/* take an audio buffer out of the adapter */
|
||||
inbuf = gst_adapter_take_buffer (ffmpegenc->adapter, frame_bytes);
|
||||
ffmpegenc->adapter_consumed += frame_size;
|
||||
|
||||
/* calculate timestamp and duration relative to start of adapter and to
|
||||
* the amount of samples we consumed */
|
||||
duration =
|
||||
gst_util_uint64_scale (ffmpegenc->adapter_consumed, GST_SECOND,
|
||||
ctx->sample_rate);
|
||||
duration -= (timestamp - ffmpegenc->adapter_ts);
|
||||
|
||||
GST_BUFFER_TIMESTAMP (inbuf) = timestamp;
|
||||
GST_BUFFER_DURATION (inbuf) = duration;
|
||||
|
||||
/* advance the adapter timestamp with the duration */
|
||||
timestamp += duration;
|
||||
|
||||
/* 4 times the input size should be big enough... */
|
||||
out_size = frame_bytes * 4;
|
||||
|
||||
ret = gst_ffmpegenc_encode_audio (ffmpegenc, inbuf, out_size,
|
||||
ffmpegenc->discont);
|
||||
if (ret != GST_FLOW_OK)
|
||||
goto push_failed;
|
||||
|
||||
ffmpegenc->discont = FALSE;
|
||||
avail = gst_adapter_available (ffmpegenc->adapter);
|
||||
}
|
||||
GST_LOG_OBJECT (ffmpegenc, "%u bytes left in the adapter", avail);
|
||||
} else {
|
||||
/* we have no frame_size, feed the encoder all the data and expect a fixed
|
||||
* output size */
|
||||
int coded_bps = av_get_bits_per_sample (oclass->in_plugin->id) / 8;
|
||||
|
||||
GST_LOG_OBJECT (ffmpegenc, "coded bps %d, osize %d", coded_bps, osize);
|
||||
|
||||
out_size = size / osize;
|
||||
if (coded_bps)
|
||||
out_size *= coded_bps;
|
||||
|
||||
ret = gst_ffmpegenc_encode_audio (ffmpegenc, inbuf, out_size, discont);
|
||||
if (ret != GST_FLOW_OK)
|
||||
goto push_failed;
|
||||
}
|
||||
|
||||
return GST_FLOW_OK;
|
||||
|
||||
/* ERRORS */
|
||||
push_failed:
|
||||
{
|
||||
GST_DEBUG_OBJECT (ffmpegenc, "Failed to push buffer %d (%s)", ret,
|
||||
gst_flow_get_name (ret));
|
||||
return ret;
|
||||
}
|
||||
}
|
||||
|
||||
static void
|
||||
gst_ffmpegenc_flush_buffers (GstFFMpegEnc * ffmpegenc, gboolean send)
|
||||
{
|
||||
|
|
@ -1000,10 +1080,8 @@ gst_ffmpegenc_change_state (GstElement * element, GstStateChange transition)
|
|||
gst_ffmpeg_avcodec_close (ffmpegenc->context);
|
||||
ffmpegenc->opened = FALSE;
|
||||
}
|
||||
if (ffmpegenc->cache) {
|
||||
gst_buffer_unref (ffmpegenc->cache);
|
||||
ffmpegenc->cache = NULL;
|
||||
}
|
||||
gst_adapter_clear (ffmpegenc->adapter);
|
||||
|
||||
if (ffmpegenc->file) {
|
||||
fclose (ffmpegenc->file);
|
||||
ffmpegenc->file = NULL;
|
||||
|
|
|
|||
|
|
@ -26,6 +26,8 @@
|
|||
|
||||
G_BEGIN_DECLS
|
||||
|
||||
#include <gst/base/gstadapter.h>
|
||||
|
||||
typedef struct _GstFFMpegEnc GstFFMpegEnc;
|
||||
|
||||
struct _GstFFMpegEnc
|
||||
|
|
@ -39,7 +41,10 @@ struct _GstFFMpegEnc
|
|||
AVCodecContext *context;
|
||||
AVFrame *picture;
|
||||
gboolean opened;
|
||||
GstBuffer *cache;
|
||||
GstClockTime adapter_ts;
|
||||
guint64 adapter_consumed;
|
||||
GstAdapter *adapter;
|
||||
gboolean discont;
|
||||
|
||||
/* cache */
|
||||
gulong bitrate;
|
||||
|
|
|
|||
Loading…
Reference in a new issue