Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 23 Mar 2012 00:59:09 +0100
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Re: -DDEVICE_IS_CPU for OpenCL builds (was: #Include in
 opencl files)

On 03/06/2012 09:03 PM, Claudio André wrote:
> I noticed you guys did not use #include directive in .cl files. I used
> it, so i'm asking if:
> - it is acceptable? Necessary diff file attached.
> 
> Copy (and duplicate) source code to the .cl file is not good neither.
> 
> We only have to copy the .h included to the run folder (as done with .cl
> files).

I think we should move these OpenCL include files as well as CUDA and
OpenCL kernels to a subdirectory to $JOHN before releasing next Jumbo.
Like $JOHN/gpu. Would that pose any problems?

Anyway, the real reason for this post is I just added (to Claudio's
include_source() function, cause it was easiest) another thing: When
building for CPU, the option -DDEVICE_IS_CPU is added to the build
options. I use this in the upcoming rar-opencl format for self-adjusting
the kernel to device type. We could add some more similar macros as needed.

Speaking of RAR I got 750 c/s (for an eight character test vector) on
first try on a GTX280. Nowhere good enough, but not too bad either: It's
ten times faster than four CPU cores. I'm hoping I can get it a lot
faster once I get a working profiler. I'll switch to a shorter test
vector for better comparisons with cRARk and the likes.

magnum



Powered by blists - more mailing lists

Your e-mail address:

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