[opus] opus Digest, Vol 72, Issue 17

Phil Wang phil.wang at arm.com
Tue Feb 3 22:20:33 PST 2015


Hi Timothy and Viswanath,

Thanks for your quick response. The remaining problem in Ne10 could be
fixing our build script for Linux. I will take care of it.
As for scaled forward power-of-2 FFT, it will take a longer time. If current
non-power-of-2 FFT does cover more than 95% of user cases, I don't think we
should give high priority to it.
Also, similar modification (scaled forward FFT) will be made to Ne10 for
fixed-point FFT. It would not take long, but it has lower priority until
Viswanath's patch is accepted. I take his patch as a milestone which means
we have already finished optimizing floating-point non-power-of-2 FFT. If
fixed-point FFT should have had higher priority, please let me know.

Best Regards,
Phil Wang

> -----Original Message-----
> From: Timothy B. Terriberry [mailto:tterribe at xiph.org]
> Sent: Wednesday, February 04, 2015 12:18 PM
> To: Viswanath Puttagunta; Phil Wang
> Cc: Jey Patel; Steve Bannister; Zhou (Joe) Yu; Yang Zhang; Steve at xiph.org;
> Zhongwei Yao; opus at xiph.org
> Subject: Re: [opus] opus Digest, Vol 72, Issue 17
> 
> Viswanath Puttagunta wrote:
> > What should we do for power-of-2? I really want to avoid putting
> > runtime checks if nfft is power of 2 in opus_fft_float_neon.
> 
> Given the tests that had to be disabled for NE10, I suspect we will not
really be
> able to use it for CUSTOM_MODES, which should be the only time nfft is a
> power of 2. So I'd suggest just disabling the support when CUSTOM_MODES is
> enabled and putting in an assert.
> 
> >> It would not be too ugly if we put them together somewhere and
> document them well enough. It is a tiny cost to integrate a 3rd party
library.
> >
> > Timothy/All,
> > Can we please get some guidance/consensus on this so we can move
> forward?
> 
> I made this comment mostly for educational purposes. If it's too late for
NE10
> to make these kind of sweeping changes to its API, I'd just not worry
about it in
> libopus.






More information about the opus mailing list