Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Thu, 10 Sep 2015 19:26:47 +0300
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: SHA-1 H()

Lei,

On Fri, Sep 11, 2015 at 12:04:21AM +0800, Lei Zhang wrote:
> The format prefix 'pbkdf2-hmac-' is omitted below.
> 
> On ARM (2xOMP):
> 
> [before]
> MD4:	2576 c/s real, 1288 c/s virtual
> MD5:	1651 c/s real, 825 c/s virtual
> SHA1:	578 c/s real, 290 c/s virtual
> 
> [after]
> MD4:	2608 c/s real, 1304 c/s virtual
> MD5:	1600 c/s real, 803 c/s virtual
> SHA1:	501 c/s real, 250 c/s virtual
> 
> MD4 becomes a little bit faster; MD5 & SHA1 become slower.

For MD5, you can (and should) repair this with OR-NOT.

For SHA-1, please investigate: take a look at the generated code in
both cases.

> On Power (8xOMP, bound to a single core)
> 
> [before]
> MD4:	28248 c/s real, 3531 c/s virtual
> MD5:	19980 c/s real, 2497 c/s virtual
> SHA1:	10593 c/s real, 1322 c/s virtual
> 
> [after]
> MD4:	31207 c/s real, 3882 c/s virtual
> MD5:	19980 c/s real, 2489 c/s virtual
> SHA1:	11273 c/s real, 1409 c/s virtual
> 
> On Power (1xOMP)
> 
> [before]
> MD4:	13398 c/s real, 13398 c/s virtual
> MD5:	10626 c/s real, 10626 c/s virtual
> SHA1:	8533 c/s real, 8533 c/s virtual
> 
> [after]
> MD4:	14628 c/s real, 14628 c/s virtual
> MD5:	10935 c/s real, 10935 c/s virtual
> SHA1:	8947 c/s real, 8947 c/s virtual
> 
> At least there's no performance drop on Power. BTW, It looks Power's SMT performance is not very impressive.

Why, a 2x+ speedup compared to 1 thread is very good.  It's similar to
or even better than what we're seeing with interleaving on x86.

Thanks,

Alexander

Powered by blists - more mailing lists

Your e-mail address:

Powered by Openwall GNU/*/Linux - Powered by OpenVZ