From: Stefan Keller <drezon_at_gmail.com>
Date: Fri, 12 Aug 2011 12:13:44 +0200

Hi Thomas,

On 08/11/2011 11:33 PM, Thomas Martitz wrote:
> Am 10.08.2011 13:11, schrieb Stefan Keller:
>> Reviews, comments, testing etc. welcome, of course.
>
> I'm not a guy that uses benefits from resampling as all my music is
> 44.1kHz, but I'd sure be willing to have better sound quality if possible.
>
> Did you check how much memory and cpu your resampler uses? Some targets
> might get into trouble for some codecs and perhaps better off with the
> simple linear one.

The memory footprint is roughly (15*256*2 + 90) * 4 Bytes. I.e. ~ 30KB.
That could be reduced to ~15KB by calculating the sinc using fp_sincos
in resampler_new_delta.
As for cpu, the code does 15 multiplications and 30 additions (when
downsampling) per output sample. Before implementing this, I asked on
IRC whether that would be doable and the response was that it's doable
although slowish on ARM.
Anyway, the patch makes the sinc resampling a compilation option. It is
only included if DSP_USE_SINC_RESAMPLING is defined. So for too slow /
too low on memory targets we could just not define it.

> /me wants to have a discussion as to whether we accept this and in which
> form :)

Yes, /me too. :)

Best regards,
Stefan

-- 
Stefan Keller

Received on 2011-08-12