Applied - I will look into including it in other compressors.

Thanks for the contribution!

Regards, Klaus Post

http://www.klauspost.com


On Thu, Jan 9, 2014 at 6:33 PM, Prakash Punnoor <[email protected]> wrote:

>  Hi again,
>
> sorry, the first patch contained some unnecessary diff. This one is better.
>
>
> On 09.01.2014 18:24, Prakash Punnoor wrote:
>
> Hi,
>
> following patch yields a speed-up of roughly 1.4 (~210ms to ~150ms) for a
> testimage from my Nikon D5100 using GCC 4.8.2 on Linux on my AMD Phenom II
> X4 840.
>
> Unfortunately line endings seem to be messed up, so sorry for that in
> advance  (dos2unix/unix2dos applied to the source file helps).
>
> Some comments:
> - BitPumpMsb doesn't need to be a field in Nikon decompressor, as it is
> only instantiated on Decompress.
> - Allocating BitPumpMsb on stack seems to speed up decoding. (I guess,
> because of memory locality and/or the compiler can do a better job).
> - No need for dtor in Nikon decompressor anymore.
> - BitPumpMsb: Allocating current_buffer inline also helps locality.
> - BitPumpMsb has no need for a (virtual) dtor anymore.
> - Maybe aligning current_buffer using some compiler attributes might
> further help.
> - Maybe other decompressors might profit from similar code changes.
>
> Cheers,
>
> Prakash
>
>
> _______________________________________________
> Rawstudio-dev mailing 
> [email protected]http://rawstudio.org/cgi-bin/mailman/listinfo/rawstudio-dev
>
>
>
> _______________________________________________
> Rawstudio-dev mailing list
> [email protected]
> http://rawstudio.org/cgi-bin/mailman/listinfo/rawstudio-dev
>
>
_______________________________________________
Rawstudio-dev mailing list
[email protected]
http://rawstudio.org/cgi-bin/mailman/listinfo/rawstudio-dev

Reply via email to