Re: crypto_scrypt-sse.c speedup

[Colin Percival <cperciva@tarsnap.com>]

On 11/15/12 02:52, Solar Designer wrote:
> The attached patch for crypto_scrypt-sse.c (against its revision in
> scrypt-1.1.6.tgz) speeds it up by 30% on AMD Bulldozer (tested on
> FX-8120) in -march=native or -mxop builds, and by 5% to 10% on Intel
> CPUs (tested on Xeon E5649), when run on the official test vectors.

Interesting!  What compiler is this?  I'm surprised that the "inline"
does anything, and it seems odd that the loop unrolling wouldn't happen
automatically too.

> Please let me know if I should add a copyright statement, although maybe
> these changes are too minor to be subject to copyright.

Up to you -- either declare your changes to be public domain (and don't add
a copyright line in your name) or declare them to be 2-clause BSD licensed
(and add a copyright line).  Either is fine with me, but I need you to pick
one. :-)

-- 
Colin Percival
Security Officer Emeritus, FreeBSD | The power to serve
Founder, Tarsnap | www.tarsnap.com | Online backups for the truly paranoid