Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 24 May 2015 05:57:17 +0300
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: PHC: Parallel in OpenCL

Agnieszka,

I changed the message Subject for this sub-thread.

On Sat, May 23, 2015 at 01:50:57PM +0200, Agnieszka Bielec wrote:
> I was trying to optimize sha512block in parallel but when i was
> copying this functions from various Claudio Andre's source codes the
> result wasn't better, was one case when the speed was greater on super
> for --dev=5 from 35k to 40k and on my laptop from 26k to 28k but on
> --dev=1 the speed decreased from 40k to 20k or 10k, also I was
> modifying slightly two versions of Claudio Andre's sha512 but the
> speed didn't increase

Please note that Claudio has a separate OpenCL kernel for GCN, such as
--dev=1 on super:

[solar@...er opencl]$ ls -1 cryptsha512_kernel_*
cryptsha512_kernel_DEFAULT.cl
cryptsha512_kernel_GCN.cl
cryptsha512_kernel_GPU.cl

I guess you took code from cryptsha512_kernel_GPU.cl?  It isn't meant to
be the best on --dev=1.  I guess you should use code from
cryptsha512_kernel_GCN.cl for it, with runtime detection of GCN arch
like in opencl_cryptsha512_fmt_plug.c: init().  It may be better to pass
some -D and use #ifdef than to use entirely separate kernel source
files, though.

Alexander

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.