Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sat, 21 Sep 2013 02:36:38 +0400
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: ZedBoard / Parallella: bcrypt

Hi Katja,

Thank you for the status updates!

On Fri, Sep 20, 2013 at 11:04:07PM +0200, Katja Malvoni wrote:
> With true dual port BRAM used to store Sbox, performance is 25.6 c/s with
> 100 MHz clock.
> With 150 MHz clock, performance is 50.4 c/s but some timing constraints
> aren't met and parallella-bcrypt format doesn't work. However, I tried
> connecting bcrypt IP to 200 MHz clock and performance is again 50.4 c/s and
> timing constraints aren't met. This time parallella-bcrypt format works. My
> guess is that tool did clock conversion to known working value or something
> like this because otherwise performance numbers don't make sense.

Yes, something weird is going on.  Can you calculate the expected c/s
rate for your current bcrypt IP core e.g. at 100 MHz, then figure out
the actual clock rates from that?  What's the maximum clock rate for
which timing constraints are met?

> This is performance with 2 cycle latency inferred block RAM. My next step
> is to use 1 cycle latency BRAM (non-registered outputs) and do more
> optimizations.

Sounds good, and it's OK if you look into the clock rate issues after
making those optimizations if you prefer.

Alexander

Powered by blists - more mailing lists

Your e-mail address:

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