Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 13 Aug 2012 08:33:45 +0530
From: Sayantan Datta <std2048@...il.com>
To: john-dev@...ts.openwall.com
Subject: Re: bf-opencl fails self-test on CPU

On Mon, Aug 13, 2012 at 5:52 AM, magnum <john.magnum@...hmail.com> wrote:

> On 2012-08-12 20:33, Solar Designer wrote:
> > On Sun, Aug 12, 2012 at 11:28:05PM +0530, Sayantan Datta wrote:
> >> With Intel SDK bf-opencl passed the self test on i5 2500k with 1450c/s .
> >> However it fails with AMD SDK on same machine. Probably its not a very
> good
> >> idea but still I was wondering if we could we use the Intel SDK for AMD
> >> cpu's too.
> >
> > We can, but my point was that there could be a bug somewhere in our
> > bf-opencl code that only shows up on some builds - if so, we'd want to
> > find and fix the bug even if it's of little practical relevance
> > currently.  Indeed, it is also possible that the bug is actually in
> > AMD's SDK.  We just don't know.
>
> I believe this patch (just committed to git) strongly indicates that
> it's a bug in the SDK - it works around the problem with no ill side
> effects.
>
> magnum@...l:src [1.7.9-jumbo-6-fixes]$ ../run/john -t -fo:bf-opencl
> -plat=1 -dev=1
> OpenCL platform 1: AMD Accelerated Parallel Processing, 2 device(s).
> Using device 1: AMD FX(tm)-8120 Eight-Core Processor
> ****Please see 'opencl_bf_std.h' for device specific optimizations****
> Benchmarking: OpenBSD Blowfish (x32) [OpenCL]... DONE
> Raw:    3056 c/s real, 386 c/s virtual
>
> magnum
>

Patch works great for both intel and amd SDK. However AMD SDK has better
performance even for intel CPU.

OpenCL platform 0: Intel(R) OpenCL, 1 device(s).
Using device 0:         Intel(R) Core(TM) i5-2500 CPU @ 3.30GHz
Compilation log: :116:59: warning: unknown attribute 'max_constant_size'
ignored
:117:46: warning: unknown attribute 'max_constant_size' ignored
:122:60: warning: unknown attribute 'max_constant_size' ignored
Build started
Kernel <blowfish> was not vectorized
Done.
****Please see 'opencl_bf_std.h' for device specific optimizations****
Benchmarking: OpenBSD Blowfish (x32) [OpenCL]... DONE
Raw:    1449 c/s real, 486 c/s virtual

OpenCL platform 1: AMD Accelerated Parallel Processing, 2 device(s).
Using device 1: Intel(R) Core(TM) i5-2500 CPU @ 3.30GHz
****Please see 'opencl_bf_std.h' for device specific optimizations****
Benchmarking: OpenBSD Blowfish (x32) [OpenCL]... DONE
Raw:    1873 c/s real, 473 c/s virtual

Now I really doubt there is anything equivalent to LDS on a CPU. Maybe the
intel SDK already used a bypass similar to the magnum's patch as there's no
impact on performance on Intel SDK due to this patch.

Regards,
Sayantan

[ CONTENT OF TYPE text/html SKIPPED ]

Powered by blists - more mailing lists

Your e-mail address:

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