GST_DEBUG_CATEGORY_STATIC (gst_voaacenc_debug);
#define GST_CAT_DEFAULT gst_voaacenc_debug
-static void gst_voaacenc_finalize (GObject * object);
-
-static GstFlowReturn gst_voaacenc_chain (GstPad * pad, GstBuffer * buffer);
-static gboolean gst_voaacenc_setcaps (GstPad * pad, GstCaps * caps);
-static GstStateChangeReturn gst_voaacenc_state_change (GstElement * element,
- GstStateChange transition);
static gboolean voaacenc_core_init (GstVoAacEnc * voaacenc);
static gboolean voaacenc_core_set_parameter (GstVoAacEnc * voaacenc);
static void voaacenc_core_uninit (GstVoAacEnc * voaacenc);
-static GstCaps *gst_voaacenc_getcaps (GstPad * pad);
static GstCaps *gst_voaacenc_create_source_pad_caps (GstVoAacEnc * voaacenc);
static gint voaacenc_get_rate_index (gint rate);
+static gboolean gst_voaacenc_start (GstAudioEncoder * enc);
+static gboolean gst_voaacenc_stop (GstAudioEncoder * enc);
+static gboolean gst_voaacenc_set_format (GstAudioEncoder * enc,
+ GstAudioInfo * info);
+static GstFlowReturn gst_voaacenc_handle_frame (GstAudioEncoder * enc,
+ GstBuffer * in_buf);
+static GstCaps *gst_voaacenc_getcaps (GstAudioEncoder * enc);
+
static gpointer
gst_voaacenc_generate_sink_caps (gpointer data)
{
g_once (&g_once, gst_voaacenc_generate_sink_caps, NULL);
caps = g_once.retval;
- return gst_caps_ref (caps);
-}
-
-static void
-_do_init (GType object_type)
-{
- const GInterfaceInfo preset_interface_info = {
- NULL, /* interface init */
- NULL, /* interface finalize */
- NULL /* interface_data */
- };
-
- g_type_add_interface_static (object_type, GST_TYPE_PRESET,
- &preset_interface_info);
-
- GST_DEBUG_CATEGORY_INIT (gst_voaacenc_debug, "voaacenc", 0,
- "AAC audio encoder");
+ return caps;
}
-GST_BOILERPLATE_FULL (GstVoAacEnc, gst_voaacenc, GstElement, GST_TYPE_ELEMENT,
- _do_init);
+GST_BOILERPLATE (GstVoAacEnc, gst_voaacenc, GstAudioEncoder,
+ GST_TYPE_AUDIO_ENCODER);
static void
gst_voaacenc_set_property (GObject * object, guint prop_id,
gst_voaacenc_class_init (GstVoAacEncClass * klass)
{
GObjectClass *object_class = G_OBJECT_CLASS (klass);
- GstElementClass *element_class = GST_ELEMENT_CLASS (klass);
+ GstAudioEncoderClass *base_class = GST_AUDIO_ENCODER_CLASS (klass);
object_class->set_property = GST_DEBUG_FUNCPTR (gst_voaacenc_set_property);
object_class->get_property = GST_DEBUG_FUNCPTR (gst_voaacenc_get_property);
- object_class->finalize = GST_DEBUG_FUNCPTR (gst_voaacenc_finalize);
+
+ base_class->start = GST_DEBUG_FUNCPTR (gst_voaacenc_start);
+ base_class->stop = GST_DEBUG_FUNCPTR (gst_voaacenc_stop);
+ base_class->set_format = GST_DEBUG_FUNCPTR (gst_voaacenc_set_format);
+ base_class->handle_frame = GST_DEBUG_FUNCPTR (gst_voaacenc_handle_frame);
+ base_class->getcaps = GST_DEBUG_FUNCPTR (gst_voaacenc_getcaps);
g_object_class_install_property (object_class, PROP_BITRATE,
g_param_spec_int ("bitrate",
"Target Audio Bitrate",
0, G_MAXINT, VOAAC_ENC_DEFAULT_BITRATE,
G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
-
- element_class->change_state = GST_DEBUG_FUNCPTR (gst_voaacenc_state_change);
}
static void
gst_voaacenc_init (GstVoAacEnc * voaacenc, GstVoAacEncClass * klass)
{
- /* create the sink pad */
- voaacenc->sinkpad = gst_pad_new_from_static_template (&sink_template, "sink");
- gst_pad_set_setcaps_function (voaacenc->sinkpad,
- GST_DEBUG_FUNCPTR (gst_voaacenc_setcaps));
- gst_pad_set_getcaps_function (voaacenc->sinkpad,
- GST_DEBUG_FUNCPTR (gst_voaacenc_getcaps));
- gst_pad_set_chain_function (voaacenc->sinkpad,
- GST_DEBUG_FUNCPTR (gst_voaacenc_chain));
- gst_element_add_pad (GST_ELEMENT (voaacenc), voaacenc->sinkpad);
-
- /* create the src pad */
- voaacenc->srcpad = gst_pad_new_from_static_template (&src_template, "src");
- gst_pad_use_fixed_caps (voaacenc->srcpad);
- gst_element_add_pad (GST_ELEMENT (voaacenc), voaacenc->srcpad);
-
- voaacenc->adapter = gst_adapter_new ();
-
voaacenc->bitrate = VOAAC_ENC_DEFAULT_BITRATE;
voaacenc->output_format = VOAAC_ENC_DEFAULT_OUTPUTFORMAT;
voaacenc->handle = NULL;
}
-static void
-gst_voaacenc_finalize (GObject * object)
+static gboolean
+gst_voaacenc_start (GstAudioEncoder * enc)
{
- GstVoAacEnc *voaacenc;
+ GstVoAacEnc *voaacenc = GST_VOAACENC (enc);
+
+ GST_DEBUG_OBJECT (enc, "start");
- voaacenc = GST_VOAACENC (object);
+ if (voaacenc_core_init (voaacenc) == FALSE)
+ return FALSE;
- g_object_unref (G_OBJECT (voaacenc->adapter));
- voaacenc->adapter = NULL;
+ voaacenc->rate = 0;
+ voaacenc->channels = 0;
- G_OBJECT_CLASS (parent_class)->finalize (object);
+ return TRUE;
+}
+
+static gboolean
+gst_voaacenc_stop (GstAudioEncoder * enc)
+{
+ GstVoAacEnc *voaacenc = GST_VOAACENC (enc);
+
+ GST_DEBUG_OBJECT (enc, "stop");
+ voaacenc_core_uninit (voaacenc);
+
+ return TRUE;
}
/* check downstream caps to configure format */
{
GstCaps *caps;
- caps = gst_pad_get_allowed_caps (voaacenc->srcpad);
+ caps = gst_pad_get_allowed_caps (GST_AUDIO_ENCODER_SRC_PAD (voaacenc));
GST_DEBUG_OBJECT (voaacenc, "allowed caps: %" GST_PTR_FORMAT, caps);
}
static GstCaps *
-gst_voaacenc_getcaps (GstPad * pad)
+gst_voaacenc_getcaps (GstAudioEncoder * benc)
{
- return gst_voaacenc_get_sink_caps ();
+ return gst_audio_encoder_proxy_getcaps (benc, gst_voaacenc_get_sink_caps ());
}
-
static gboolean
-gst_voaacenc_setcaps (GstPad * pad, GstCaps * caps)
+gst_voaacenc_set_format (GstAudioEncoder * benc, GstAudioInfo * info)
{
gboolean ret = FALSE;
- GstStructure *structure;
GstVoAacEnc *voaacenc;
GstCaps *src_caps;
- voaacenc = GST_VOAACENC (GST_PAD_PARENT (pad));
-
- structure = gst_caps_get_structure (caps, 0);
+ voaacenc = GST_VOAACENC (benc);
/* get channel count */
- gst_structure_get_int (structure, "channels", &voaacenc->channels);
- gst_structure_get_int (structure, "rate", &voaacenc->rate);
+ voaacenc->channels = GST_AUDIO_INFO_CHANNELS (info);
+ voaacenc->rate = GST_AUDIO_INFO_RATE (info);
- /* precalc duration as it's constant now */
- voaacenc->duration =
- gst_util_uint64_scale_int (1024, GST_SECOND, voaacenc->rate);
+ /* precalc buffer size as it's constant now */
voaacenc->inbuf_size = voaacenc->channels * 2 * 1024;
gst_voaacenc_negotiate (voaacenc);
src_caps = gst_voaacenc_create_source_pad_caps (voaacenc);
if (src_caps) {
- gst_pad_set_caps (voaacenc->srcpad, src_caps);
+ gst_pad_set_caps (GST_AUDIO_ENCODER_SRC_PAD (voaacenc), src_caps);
gst_caps_unref (src_caps);
ret = voaacenc_core_set_parameter (voaacenc);
}
+
+ /* report needs to base class */
+ gst_audio_encoder_set_frame_samples_min (benc, 1024);
+ gst_audio_encoder_set_frame_samples_max (benc, 1024);
+ gst_audio_encoder_set_frame_max (benc, 1);
+
return ret;
}
static GstFlowReturn
-gst_voaacenc_chain (GstPad * pad, GstBuffer * buffer)
+gst_voaacenc_handle_frame (GstAudioEncoder * benc, GstBuffer * buf)
{
GstVoAacEnc *voaacenc;
- GstFlowReturn ret;
- guint64 timestamp, distance = 0;
+ GstFlowReturn ret = GST_FLOW_OK;
+ GstBuffer *out;
+ VO_AUDIO_OUTPUTINFO output_info = { {0} };
+ VO_CODECBUFFER input = { 0 };
+ VO_CODECBUFFER output = { 0 };
- voaacenc = GST_VOAACENC (GST_PAD_PARENT (pad));
+ voaacenc = GST_VOAACENC (benc);
- g_return_val_if_fail (voaacenc->handle, GST_FLOW_WRONG_STATE);
+ g_return_val_if_fail (voaacenc->handle, GST_FLOW_NOT_NEGOTIATED);
if (voaacenc->rate == 0 || voaacenc->channels == 0)
goto not_negotiated;
- /* discontinuity clears adapter, FIXME, maybe we can set some
- * encoder flag to mask the discont. */
- if (GST_BUFFER_FLAG_IS_SET (buffer, GST_BUFFER_FLAG_DISCONT)) {
- gst_adapter_clear (voaacenc->adapter);
- voaacenc->discont = TRUE;
+ /* we don't deal with squeezing remnants, so simply discard those */
+ if (G_UNLIKELY (buf == NULL)) {
+ GST_DEBUG_OBJECT (benc, "no data");
+ goto exit;
}
- ret = GST_FLOW_OK;
- gst_adapter_push (voaacenc->adapter, buffer);
-
- /* Collect samples until we have enough for an output frame */
- while (gst_adapter_available (voaacenc->adapter) >= voaacenc->inbuf_size) {
- GstBuffer *out;
- guint8 *data;
- VO_CODECBUFFER input = { 0 }
- , output = {
- 0};
- VO_AUDIO_OUTPUTINFO output_info = { {0}
- };
-
-
- /* max size */
- if ((ret =
- gst_pad_alloc_buffer_and_set_caps (voaacenc->srcpad, 0,
- voaacenc->inbuf_size, GST_PAD_CAPS (voaacenc->srcpad),
- &out)) != GST_FLOW_OK) {
- return ret;
- }
-
- output.Buffer = GST_BUFFER_DATA (out);
- output.Length = voaacenc->inbuf_size;
-
- if (voaacenc->discont) {
- GST_BUFFER_FLAG_SET (out, GST_BUFFER_FLAG_DISCONT);
- voaacenc->discont = FALSE;
- }
+ if (G_UNLIKELY (GST_BUFFER_SIZE (buf) < voaacenc->inbuf_size)) {
+ GST_DEBUG_OBJECT (voaacenc, "discarding trailing data %d",
+ buf ? GST_BUFFER_SIZE (buf) : 0);
+ ret = gst_audio_encoder_finish_frame (benc, NULL, -1);
+ goto exit;
+ }
- data =
- (guint8 *) gst_adapter_peek (voaacenc->adapter, voaacenc->inbuf_size);
- input.Buffer = data;
- input.Length = voaacenc->inbuf_size;
- voaacenc->codec_api.SetInputData (voaacenc->handle, &input);
-
- /* encode */
- if (voaacenc->codec_api.GetOutputData (voaacenc->handle, &output,
- &output_info) != VO_ERR_NONE) {
- gst_buffer_unref (out);
- return GST_FLOW_ERROR;
- }
+ /* max size */
+ if ((ret =
+ gst_pad_alloc_buffer_and_set_caps (GST_AUDIO_ENCODER_SRC_PAD
+ (voaacenc), 0, voaacenc->inbuf_size,
+ GST_PAD_CAPS (GST_AUDIO_ENCODER_SRC_PAD (voaacenc)),
+ &out)) != GST_FLOW_OK) {
+ goto exit;
+ }
- /* get timestamp from adapter */
- timestamp = gst_adapter_prev_timestamp (voaacenc->adapter, &distance);
+ output.Buffer = GST_BUFFER_DATA (out);
+ output.Length = voaacenc->inbuf_size;
- if (G_LIKELY (GST_CLOCK_TIME_IS_VALID (timestamp))) {
- GST_BUFFER_TIMESTAMP (out) =
- timestamp +
- GST_FRAMES_TO_CLOCK_TIME (distance / voaacenc->channels /
- VOAAC_ENC_BITS_PER_SAMPLE, voaacenc->rate);
- }
+ g_assert (GST_BUFFER_SIZE (buf) == voaacenc->inbuf_size);
+ input.Buffer = GST_BUFFER_DATA (buf);
+ input.Length = voaacenc->inbuf_size;
+ voaacenc->codec_api.SetInputData (voaacenc->handle, &input);
- GST_BUFFER_DURATION (out) =
- GST_FRAMES_TO_CLOCK_TIME (voaacenc->inbuf_size / voaacenc->channels /
- VOAAC_ENC_BITS_PER_SAMPLE, voaacenc->rate);
+ /* encode */
+ if (voaacenc->codec_api.GetOutputData (voaacenc->handle, &output,
+ &output_info) != VO_ERR_NONE) {
+ gst_buffer_unref (out);
+ goto encode_failed;
+ }
- GST_LOG_OBJECT (voaacenc, "Pushing out buffer time: %" GST_TIME_FORMAT
- " duration: %" GST_TIME_FORMAT,
- GST_TIME_ARGS (GST_BUFFER_TIMESTAMP (out)),
- GST_TIME_ARGS (GST_BUFFER_DURATION (out)));
+ GST_LOG_OBJECT (voaacenc, "encoded to %d bytes", output.Length);
+ GST_BUFFER_SIZE (out) = output.Length;
- GST_BUFFER_SIZE (out) = output.Length;
+ GST_LOG_OBJECT (voaacenc, "Pushing out buffer time: %" GST_TIME_FORMAT
+ " duration: %" GST_TIME_FORMAT,
+ GST_TIME_ARGS (GST_BUFFER_TIMESTAMP (out)),
+ GST_TIME_ARGS (GST_BUFFER_DURATION (out)));
- /* flush the among of data we have peek */
- gst_adapter_flush (voaacenc->adapter, voaacenc->inbuf_size);
+ ret = gst_audio_encoder_finish_frame (benc, out, 1024);
- /* play */
- if ((ret = gst_pad_push (voaacenc->srcpad, out)) != GST_FLOW_OK)
- break;
- }
+exit:
return ret;
/* ERRORS */
{
GST_ELEMENT_ERROR (voaacenc, STREAM, TYPE_NOT_FOUND,
(NULL), ("unknown type"));
- return GST_FLOW_NOT_NEGOTIATED;
- }
-}
-
-static GstStateChangeReturn
-gst_voaacenc_state_change (GstElement * element, GstStateChange transition)
-{
- GstVoAacEnc *voaacenc;
- GstStateChangeReturn ret;
-
- voaacenc = GST_VOAACENC (element);
-
- switch (transition) {
- case GST_STATE_CHANGE_NULL_TO_READY:
- if (voaacenc_core_init (voaacenc) == FALSE)
- return GST_STATE_CHANGE_FAILURE;
- break;
- case GST_STATE_CHANGE_READY_TO_PAUSED:
- voaacenc->rate = 0;
- voaacenc->channels = 0;
- voaacenc->discont = FALSE;
- gst_adapter_clear (voaacenc->adapter);
- break;
- default:
- break;
+ ret = GST_FLOW_NOT_NEGOTIATED;
+ goto exit;
}
-
- ret = GST_ELEMENT_CLASS (parent_class)->change_state (element, transition);
-
- switch (transition) {
- case GST_STATE_CHANGE_READY_TO_NULL:
- voaacenc_core_uninit (voaacenc);
- gst_adapter_clear (voaacenc->adapter);
- break;
- default:
- break;
+encode_failed:
+ {
+ GST_ELEMENT_ERROR (voaacenc, STREAM, ENCODE, (NULL), ("encode failed"));
+ ret = GST_FLOW_ERROR;
+ goto exit;
}
- return ret;
}
+
static GstCaps *
gst_voaacenc_create_source_pad_caps (GstVoAacEnc * voaacenc)
{