Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 21 Nov 2011 09:36:00 +0400
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: OpenMP for MD5-crypt

magnum -

On Thu, Nov 10, 2011 at 02:11:30AM +0100, magnum wrote:
> 2011-11-09 20:45, Solar Designer wrote:
> > On Wed, Nov 09, 2011 at 08:29:06PM +0100, magnum wrote:
> >> 2011-11-09 13:39, Solar Designer wrote:
> >>> With these two changes, I think the check could be:
> >>>
> >>> #if !defined(MD5_in_sse_intrinsics) && defined(__GNUC__) && \
> >>>     (__GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ < 4)) && \
> >>>     !defined(_OPENMP) && !defined(USING_ICC_S_FILE)
> >>> #undef MD5_SSE_PARA
> >>> #endif
> >>
> >> Sounds reasonable. So if I understand correctly, the above check would
> >> replace the current one in MD5_std.h and there is no other things needed
> >> anywhere, right?
> > 
> > Yes, that's my understanding too.
> 
> This was incorporated in patch 0004.

I got OpenMP support for MD5-crypt into the main tree, without
intrinsics for now.  This means that the above check will need to be
revised for 1.7.9 again - remove the "!defined(_OPENMP) &&" portion from
it such that we don't force the use of intrinsics for OpenMP.

Oh, and I'd appreciate testing of code currently in CVS.  Things that
need testing are: MD5-crypt in various builds, --make-charset (and use
of the resulting .chr files), bitslice DES on more archs (turns out
there was a problem on big-endian in last week's revision of the code,
should be fixed now).

Thanks,

Alexander

Powered by blists - more mailing lists

Your e-mail address:

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