|
|
@ -993,8 +993,6 @@ static av_cold void uninit(AVFilterContext *ctx) |
|
|
|
yae_release_buffers(atempo); |
|
|
|
yae_release_buffers(atempo); |
|
|
|
} |
|
|
|
} |
|
|
|
|
|
|
|
|
|
|
|
static int query_formats(AVFilterContext *ctx) |
|
|
|
|
|
|
|
{ |
|
|
|
|
|
|
|
// WSOLA necessitates an internal sliding window ring buffer
|
|
|
|
// WSOLA necessitates an internal sliding window ring buffer
|
|
|
|
// for incoming audio stream.
|
|
|
|
// for incoming audio stream.
|
|
|
|
//
|
|
|
|
//
|
|
|
@ -1009,16 +1007,6 @@ static int query_formats(AVFilterContext *ctx) |
|
|
|
AV_SAMPLE_FMT_DBL, |
|
|
|
AV_SAMPLE_FMT_DBL, |
|
|
|
AV_SAMPLE_FMT_NONE |
|
|
|
AV_SAMPLE_FMT_NONE |
|
|
|
}; |
|
|
|
}; |
|
|
|
int ret = ff_set_common_all_channel_counts(ctx); |
|
|
|
|
|
|
|
if (ret < 0) |
|
|
|
|
|
|
|
return ret; |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ret = ff_set_common_formats_from_list(ctx, sample_fmts); |
|
|
|
|
|
|
|
if (ret < 0) |
|
|
|
|
|
|
|
return ret; |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
return ff_set_common_all_samplerates(ctx); |
|
|
|
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
static int config_props(AVFilterLink *inlink) |
|
|
|
static int config_props(AVFilterLink *inlink) |
|
|
|
{ |
|
|
|
{ |
|
|
@ -1193,5 +1181,5 @@ const AVFilter ff_af_atempo = { |
|
|
|
.priv_class = &atempo_class, |
|
|
|
.priv_class = &atempo_class, |
|
|
|
FILTER_INPUTS(atempo_inputs), |
|
|
|
FILTER_INPUTS(atempo_inputs), |
|
|
|
FILTER_OUTPUTS(atempo_outputs), |
|
|
|
FILTER_OUTPUTS(atempo_outputs), |
|
|
|
FILTER_QUERY_FUNC(query_formats), |
|
|
|
FILTER_SAMPLEFMTS_ARRAY(sample_fmts), |
|
|
|
}; |
|
|
|
}; |
|
|
|