[pulseaudio-discuss] [PATCH v2 3/3] resampler: Add support for resamplers that consume less data than asked.

[Arun Raghavan]

On Thu, 2012-04-26 at 11:24 +0300, Tanu Kaskinen wrote:
> On Thu, 2012-04-26 at 08:20 +0200, David Henningsson wrote:
> > >   size_t pa_resampler_result(pa_resampler *r, size_t in_length) {
> > >       pa_assert(r);
> > >
> > > -    /* Let's round up here */
> > > -
> > > -    return (((((in_length + r->i_fz-1) / r->i_fz) * r->o_ss.rate) + r->i_ss.rate-1) / r->i_ss.rate) * r->o_fz;
> > > +    /* Let's round up here to ensure that the caller will always allocate big
> > > +     * enough output buffer. */
> > > +    return (((((in_length + r->i_fz-1) / r->i_fz + r->leftover_buf.length / r->o_fz) * r->o_ss.rate) + r->i_ss.rate-1) / r->i_ss.rate) * r->o_fz;
> > 
> > A temporary variable or two here would increase readability and ease 
> > debugging.
> > 
> > >   }
> > >
> > >   size_t pa_resampler_max_block_size(pa_resampler *r) {
> > > @@ -413,7 +428,7 @@ size_t pa_resampler_max_block_size(pa_resampler *r) {
> > >
> > >       fs = pa_frame_size(&ss);
> > >
> > > -    return (((block_size_max/fs - EXTRA_FRAMES)*r->i_ss.rate)/ss.rate)*r->i_fz;
> > > +    return (((block_size_max / fs - EXTRA_FRAMES - r->leftover_buf.length / r->o_fz) * r->i_ss.rate) / ss.rate) * r->i_fz;
> > 
> > A temporary variable or two here would increase readability and ease 
> > debugging.
> 
> Good suggestions. There was also wrong frame size used in these
> calculations: the leftover length needs to be divided by working frame
> size instead of output frame size.

Would be neat to have a test case for this -- something like a dummy
resampler that wraps some other resampler and intentionally consumes
less than the total.

-- Arun