int ff_audio_convert(AudioConvert *ac, AudioData *out, AudioData *in)
{
int use_generic = 1;
int len = in->nb_samples;
int p;
if (ac->dc) {
/* dithered conversion */
av_dlog(ac->avr, "%d samples - audio_convert: %s to %s (dithered)\n",
len, av_get_sample_fmt_name(ac->in_fmt),
av_get_sample_fmt_name(ac->out_fmt));
return ff_convert_dither(ac->dc, out, in);
}
/* determine whether to use the optimized function based on pointer and
samples alignment in both the input and output */
if (ac->has_optimized_func) {
int ptr_align = FFMIN(in->ptr_align, out->ptr_align);
int samples_align = FFMIN(in->samples_align, out->samples_align);
int aligned_len = FFALIGN(len, ac->samples_align);
if (!(ptr_align % ac->ptr_align) && samples_align >= aligned_len) {
len = aligned_len;
use_generic = 0;
}
}
av_dlog(ac->avr, "%d samples - audio_convert: %s to %s (%s)\n", len,
av_get_sample_fmt_name(ac->in_fmt),
av_get_sample_fmt_name(ac->out_fmt),
use_generic ? ac->func_descr_generic : ac->func_descr);
if (ac->apply_map) {
ChannelMapInfo *map = &ac->avr->ch_map_info;
if (!av_sample_fmt_is_planar(ac->out_fmt)) {
av_log(ac->avr, AV_LOG_ERROR, "cannot remap packed format during conversion\n");
return AVERROR(EINVAL);
}
if (map->do_remap) {
if (av_sample_fmt_is_planar(ac->in_fmt)) {
conv_func_flat *convert = use_generic ? ac->conv_flat_generic :
ac->conv_flat;
for (p = 0; p < ac->planes; p++)
if (map->channel_map[p] >= 0)
convert(out->data[p], in->data[map->channel_map[p]], len);
} else {
uint8_t *data[AVRESAMPLE_MAX_CHANNELS];
conv_func_deinterleave *convert = use_generic ?
ac->conv_deinterleave_generic :
ac->conv_deinterleave;
for (p = 0; p < ac->channels; p++)
data[map->input_map[p]] = out->data[p];
convert(data, in->data[0], len, ac->channels);
}
}
if (map->do_copy || map->do_zero) {
for (p = 0; p < ac->planes; p++) {
if (map->channel_copy[p])
memcpy(out->data[p], out->data[map->channel_copy[p]],
len * out->stride);
else if (map->channel_zero[p])
av_samples_set_silence(&out->data[p], 0, len, 1, ac->out_fmt);
}
}
} else {
switch (ac->func_type) {
case CONV_FUNC_TYPE_FLAT: {
if (!in->is_planar)
len *= in->channels;
if (use_generic) {
for (p = 0; p < ac->planes; p++)
ac->conv_flat_generic(out->data[p], in->data[p], len);
} else {
for (p = 0; p < ac->planes; p++)
ac->conv_flat(out->data[p], in->data[p], len);
}
break;
}
case CONV_FUNC_TYPE_INTERLEAVE:
if (use_generic)
ac->conv_interleave_generic(out->data[0], in->data, len,
ac->channels);
else
ac->conv_interleave(out->data[0], in->data, len, ac->channels);
break;
case CONV_FUNC_TYPE_DEINTERLEAVE:
if (use_generic)
ac->conv_deinterleave_generic(out->data, in->data[0], len,
ac->channels);
else
ac->conv_deinterleave(out->data, in->data[0], len,
ac->channels);
break;
}
}
out->nb_samples = in->nb_samples;
return 0;
}
int ff_audio_mix(AudioMix *am, AudioData *src)
{
int use_generic = 1;
int len = src->nb_samples;
int i, j;
/* determine whether to use the optimized function based on pointer and
samples alignment in both the input and output */
if (am->has_optimized_func) {
int aligned_len = FFALIGN(len, am->samples_align);
if (!(src->ptr_align % am->ptr_align) &&
src->samples_align >= aligned_len) {
len = aligned_len;
use_generic = 0;
}
}
av_dlog(am->avr, "audio_mix: %d samples - %d to %d channels (%s)\n",
src->nb_samples, am->in_channels, am->out_channels,
use_generic ? am->func_descr_generic : am->func_descr);
if (am->in_matrix_channels && am->out_matrix_channels) {
uint8_t **data;
uint8_t *data0[AVRESAMPLE_MAX_CHANNELS] = { NULL };
if (am->out_matrix_channels < am->out_channels ||
am->in_matrix_channels < am->in_channels) {
for (i = 0, j = 0; i < FFMAX(am->in_channels, am->out_channels); i++) {
if (am->input_skip[i] || am->output_skip[i] || am->output_zero[i])
continue;
data0[j++] = src->data[i];
}
data = data0;
} else {
data = src->data;
}
if (use_generic)
am->mix_generic(data, am->matrix, len, am->out_matrix_channels,
am->in_matrix_channels);
else
am->mix(data, am->matrix, len, am->out_matrix_channels,
am->in_matrix_channels);
}
if (am->out_matrix_channels < am->out_channels) {
for (i = 0; i < am->out_channels; i++)
if (am->output_zero[i])
av_samples_set_silence(&src->data[i], 0, len, 1, am->fmt);
}
ff_audio_data_set_channels(src, am->out_channels);
/* clip protection is only enabled when the internal sample format is fltp, so we don't need to check here */
if (am->clip_protection) {
for(i = 0; i < src->nb_samples; i++) {
for (j = 0; j < src->channels; j++) {
const float sample = fabs(((float **)src->data)[j][i]);
if (sample > am->clip_max) {
am->clip_max = sample;
av_log(am->avr, AV_LOG_INFO, "Clipping protection at %.3f\n", sample);
}
if (am->clip_max > 1.0f)
((float **)src->data)[j][i] /= am->clip_max;
}
}
}
return 0;
}
static int return_audio_frame(AVFilterContext *ctx)
{
AVFilterLink *link = ctx->outputs[0];
FifoContext *s = ctx->priv;
AVFilterBufferRef *head = s->root.next->buf;
AVFilterBufferRef *buf_out;
int ret;
if (!s->buf_out &&
head->audio->nb_samples >= link->request_samples &&
calc_ptr_alignment(head) >= 32) {
if (head->audio->nb_samples == link->request_samples) {
buf_out = head;
queue_pop(s);
} else {
buf_out = avfilter_ref_buffer(head, AV_PERM_READ);
if (!buf_out)
return AVERROR(ENOMEM);
buf_out->audio->nb_samples = link->request_samples;
buffer_offset(link, head, link->request_samples);
}
} else {
int nb_channels = av_get_channel_layout_nb_channels(link->channel_layout);
if (!s->buf_out) {
s->buf_out = ff_get_audio_buffer(link, AV_PERM_WRITE,
link->request_samples);
if (!s->buf_out)
return AVERROR(ENOMEM);
s->buf_out->audio->nb_samples = 0;
s->buf_out->pts = head->pts;
s->allocated_samples = link->request_samples;
} else if (link->request_samples != s->allocated_samples) {
av_log(ctx, AV_LOG_ERROR, "request_samples changed before the "
"buffer was returned.\n");
return AVERROR(EINVAL);
}
while (s->buf_out->audio->nb_samples < s->allocated_samples) {
int len = FFMIN(s->allocated_samples - s->buf_out->audio->nb_samples,
head->audio->nb_samples);
av_samples_copy(s->buf_out->extended_data, head->extended_data,
s->buf_out->audio->nb_samples, 0, len, nb_channels,
link->format);
s->buf_out->audio->nb_samples += len;
if (len == head->audio->nb_samples) {
avfilter_unref_buffer(head);
queue_pop(s);
if (!s->root.next &&
(ret = ff_request_frame(ctx->inputs[0])) < 0) {
if (ret == AVERROR_EOF) {
av_samples_set_silence(s->buf_out->extended_data,
s->buf_out->audio->nb_samples,
s->allocated_samples -
s->buf_out->audio->nb_samples,
nb_channels, link->format);
s->buf_out->audio->nb_samples = s->allocated_samples;
break;
}
return ret;
}
head = s->root.next->buf;
} else {
buffer_offset(link, head, len);
}
}
buf_out = s->buf_out;
s->buf_out = NULL;
}
return ff_filter_samples(link, buf_out);
}
static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
{
AVFilterContext *ctx = inlink->dst;
ASyncContext *s = ctx->priv;
AVFilterLink *outlink = ctx->outputs[0];
int nb_channels = av_get_channel_layout_nb_channels(buf->channel_layout);
int64_t pts = (buf->pts == AV_NOPTS_VALUE) ? buf->pts :
av_rescale_q(buf->pts, inlink->time_base, outlink->time_base);
int out_size, ret;
int64_t delta;
int64_t new_pts;
/* buffer data until we get the next timestamp */
if (s->pts == AV_NOPTS_VALUE || pts == AV_NOPTS_VALUE) {
if (pts != AV_NOPTS_VALUE) {
s->pts = pts - get_delay(s);
}
return write_to_fifo(s, buf);
}
if (s->first_pts != AV_NOPTS_VALUE) {
handle_trimming(ctx);
if (!avresample_available(s->avr))
return write_to_fifo(s, buf);
}
/* when we have two timestamps, compute how many samples would we have
* to add/remove to get proper sync between data and timestamps */
delta = pts - s->pts - get_delay(s);
out_size = avresample_available(s->avr);
if (labs(delta) > s->min_delta ||
(s->first_frame && delta && s->first_pts != AV_NOPTS_VALUE)) {
av_log(ctx, AV_LOG_VERBOSE, "Discontinuity - %"PRId64" samples.\n", delta);
out_size = av_clipl_int32((int64_t)out_size + delta);
} else {
if (s->resample) {
// adjust the compensation if delta is non-zero
int delay = get_delay(s);
int comp = s->comp + av_clip(delta * inlink->sample_rate / delay,
-s->max_comp, s->max_comp);
if (comp != s->comp) {
av_log(ctx, AV_LOG_VERBOSE, "Compensating %d samples per second.\n", comp);
if (avresample_set_compensation(s->avr, comp, inlink->sample_rate) == 0) {
s->comp = comp;
}
}
}
// adjust PTS to avoid monotonicity errors with input PTS jitter
pts -= delta;
delta = 0;
}
if (out_size > 0) {
AVFrame *buf_out = ff_get_audio_buffer(outlink, out_size);
if (!buf_out) {
ret = AVERROR(ENOMEM);
goto fail;
}
if (s->first_frame && delta > 0) {
int ch;
av_samples_set_silence(buf_out->extended_data, 0, delta,
nb_channels, buf->format);
for (ch = 0; ch < nb_channels; ch++)
buf_out->extended_data[ch] += delta;
avresample_read(s->avr, buf_out->extended_data, out_size);
for (ch = 0; ch < nb_channels; ch++)
buf_out->extended_data[ch] -= delta;
} else {
avresample_read(s->avr, buf_out->extended_data, out_size);
if (delta > 0) {
av_samples_set_silence(buf_out->extended_data, out_size - delta,
delta, nb_channels, buf->format);
}
}
buf_out->pts = s->pts;
ret = ff_filter_frame(outlink, buf_out);
if (ret < 0)
goto fail;
s->got_output = 1;
} else if (avresample_available(s->avr)) {
av_log(ctx, AV_LOG_WARNING, "Non-monotonous timestamps, dropping "
"whole buffer.\n");
}
/* drain any remaining buffered data */
avresample_read(s->avr, NULL, avresample_available(s->avr));
new_pts = pts - avresample_get_delay(s->avr);
/* check for s->pts monotonicity */
if (new_pts > s->pts) {
s->pts = new_pts;
ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
buf->linesize[0], buf->nb_samples);
} else {
av_log(ctx, AV_LOG_WARNING, "Non-monotonous timestamps, dropping "
"whole buffer.\n");
ret = 0;
}
s->first_frame = 0;
fail:
av_frame_free(&buf);
return ret;
}
static int return_audio_frame(AVFilterContext *ctx)
{
AVFilterLink *link = ctx->outputs[0];
FifoContext *s = ctx->priv;
AVFrame *head = s->root.next ? s->root.next->frame : NULL;
AVFrame *out;
int ret;
/* if head is NULL then we're flushing the remaining samples in out */
if (!head && !s->out)
return AVERROR_EOF;
if (!s->out &&
head->nb_samples >= link->request_samples &&
calc_ptr_alignment(head) >= 32) {
if (head->nb_samples == link->request_samples) {
out = head;
queue_pop(s);
} else {
out = av_frame_clone(head);
if (!out)
return AVERROR(ENOMEM);
out->nb_samples = link->request_samples;
buffer_offset(link, head, link->request_samples);
}
} else {
int nb_channels = av_get_channel_layout_nb_channels(link->channel_layout);
if (!s->out) {
s->out = ff_get_audio_buffer(link, link->request_samples);
if (!s->out)
return AVERROR(ENOMEM);
s->out->nb_samples = 0;
s->out->pts = head->pts;
s->allocated_samples = link->request_samples;
} else if (link->request_samples != s->allocated_samples) {
av_log(ctx, AV_LOG_ERROR, "request_samples changed before the "
"buffer was returned.\n");
return AVERROR(EINVAL);
}
while (s->out->nb_samples < s->allocated_samples) {
int len;
if (!s->root.next) {
ret = ff_request_frame(ctx->inputs[0]);
if (ret == AVERROR_EOF) {
av_samples_set_silence(s->out->extended_data,
s->out->nb_samples,
s->allocated_samples -
s->out->nb_samples,
nb_channels, link->format);
s->out->nb_samples = s->allocated_samples;
break;
} else if (ret < 0)
return ret;
av_assert0(s->root.next); // If ff_request_frame() succeeded then we should have a frame
}
head = s->root.next->frame;
len = FFMIN(s->allocated_samples - s->out->nb_samples,
head->nb_samples);
av_samples_copy(s->out->extended_data, head->extended_data,
s->out->nb_samples, 0, len, nb_channels,
link->format);
s->out->nb_samples += len;
if (len == head->nb_samples) {
av_frame_free(&head);
queue_pop(s);
} else {
buffer_offset(link, head, len);
}
}
out = s->out;
s->out = NULL;
}
return ff_filter_frame(link, out);
}
