Re: crypto benchmark results with 2.6.0-test6

[Fruhwirth Clemens <clemens-dated-1066221801.48ed@endorphin.org>]

[-- Attachment #1: Type: text/plain, Size: 1623 bytes --]

On Sun, Oct 05, 2003 at 02:18:00PM +0200, Christian Kujau wrote:
> Sorry, took me a while to retest.

Thanks :)

> Fruhwirth Clemens schrieb:
> >would you like to benchmark
> >http://clemens.endorphin.org/patches/aes-i586-asm-2.6.0-test5.diff
>
> yes, i did so, results on:
> http://www.nerdbynature.de/bench/prinz/

Looks promising. But it's not going to be merged because of objections by
the cryptoapi maintainer.

> >http://clemens.endorphin.org/twofish-i586/ (experimential) 
> 
> uh, i guess this masm/windoze/elf32 stuff is too much for me, i could 
> try, but don't have time to dig into this. but you could try my script, 
> and run benchmarks on your machine too.
> but: how is this twofish optimization supposed to go into mainline 
> anyway? one had to use a special compiling environment to compile a kernel?

It's not a patch, it's an add-on. I'm working on a gas version of the
assembler code so it can be merged, but it's far from being complete
(although it works).

> i set up these benchmarks for me too, because i needed to know what 
> cipher is fast enough for my own use. usually i sticked to serpent, now 
> aes-i586 looks quite good on this PC. i wonder if it will compile and so 
> something on ppc32 too, but probably not. which is a pity, because my 
> primary use of the cryptoloop is a ppc :-(

Well that's the cost of an assembler implementation ;) it's not portable. 
Probably you can one of IBM's compilers to compile the cipher of your
choice. http://www-3.ibm.com/software/awdtools/ccompilers/ ... Rumours say
they're fast than gcc.

Regards, Clemens

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]