Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [day] [month] [year] [list]
Date: Tue, 24 Apr 2012 21:49:14 +0200
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Vectorizing OpenCL

I had a go at vectorizing raw-sha1-opencl just to see how it's done. It
was *extremely* easy (I had a sneak at Lukas' phpass for hints).

I expected it to boost AMD GPU's but it doesn't (not on Cedar at least,
12% drop). It had a devastating effect on 9600GT, about half the speed.
and GTX580 had a 20% drop.

For some reason, Intel's OpenCL-CPU compiler fails to vectorize it (ie.
make SSE2 out of it). That's strange. But the AMD CPU compiler seemed to
get the hint, it doesn't say anything in the build log but got like 3x
faster (though it was much slower than Intel before the change).

All in all, there seems to be no point in using this code. I'm posting
it here as an example. Nothing is optimised.

magnum

View attachment "0001-Raw-sha1-openssl-vectorized-just-for-getting-the-han.patch" of type "text/x-patch" (6566 bytes)

Powered by blists - more mailing lists

Your e-mail address:

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