[Vorbis-dev] [RFC PATCH v1 0/3] Introducing ARM SIMD Support

[Viswanath Puttagunta]

Hello All,

Please note that I haven't abandoned my patchset.. I am hoping the brains
in opus will review my code and accept it there. Just posted yesterday. You
can find it at [1] and [2].

Then, I can just redo my Vorbis neon patches accordingly and resubmit
here.. but expecting this will take some time.

[1]: http://lists.xiph.org/pipermail/opus/2014-November/002745.html
[2]: http://lists.xiph.org/pipermail/opus/2014-November/002744.html

Regards,
Vish (Viswanath Puttagunta)

On 2 October 2014 23:38, Chris Moore <moore at free.fr> wrote:

> Hi Vish,
>
> I am very interested in your Vorbis optimisations.
> Unfortunately I am not a NEON guru, nor even an ARM assembler one.
> I had a quick look at your code and the only remark I have concerns the
> pseudo-code in comments.
> In several places you write:
> a[i] = a[i]*b[n-i-1] + a[i]*c[i]
> but IMHO this should be:
> a[i] = a[i]*b[n-i-1] + b[i]*c[i]
>
> I have also done some work on Vorbis improvements, mainly to Tremor.
> I posted some information on them to this list early this year, mainly in
> March.
> Unfortunately nobody showed any interest in receiving patches :(
> I shall be interested to see if anybody integrates yours.
> Maybe we should be working on Opus ;-)
>
> I am afraid that I have not tested your modifications.
> For the moment I have been working on pure C code in the comfort of an
> x86_64 environment.
> But I have oldish ARM hardware in mind as Tremor is really for
> architectures without hardware floating point.
>
> Cheers,
> Chris
>
> Le 01/10/2014 20:33, Viswanath Puttagunta a écrit :
>
>>
>> On 1 October 2014 12:35, Ralph Giles <giles at thaumas.net <mailto:
>> giles at thaumas.net>> wrote:
>>
>>     On 2014-10-01 10:00 AM, Viswanath Puttagunta wrote:
>>
>>     > I would like to get feedback on "1" above from this list. I first
>>     want
>>     > to make sure the modifications to autoconf files is in line with the
>>     > project.
>>
>>     Not having looked at the patches, you should compare your work
>>     with the
>>     instruction set detection in the opus implementation. That has worked
>>     well so far, if you can say you did the same thing, it will be easier
>>     for the authors to review it.
>>
>>     https://git.xiph.org/?p=opus.git;a=summary
>>
>> >> Thanks, will do.
>>
>>
>>
>>     > For "2", which is ARM NEON specific reviews, if necessary, I can
>>     ping a
>>     > few of ARM's NEON gurus and I think they will be happy to join
>>     this list
>>     > and review the code.
>>
>>     Do you have output verification and benchmark results you can share?
>>
>> >> As mentioned in cover letter for patches, available here <
>> https://docs.google.com/a/vishndivi.com/document/d/
>> 1LuMORT2jIroiuDKAQaj7YpizCRQBpgruC8DKDlzc8ik/edit>.
>>
>>
>>      -r
>>
>>
>>
>> _______________________________________________
>> Vorbis-dev mailing list
>> Vorbis-dev at xiph.org
>> http://lists.xiph.org/mailman/listinfo/vorbis-dev
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.xiph.org/pipermail/vorbis-dev/attachments/20141110/2236c54a/attachment.htm