Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Tue, 17 Apr 2012 00:59:52 +0200
From: magnum <john.magnum@...hmail.com>
To: "john-dev@...ts.openwall.com" <john-dev@...ts.openwall.com>
Subject: Re: Xsha512 OpenCL Code

I have seen this before. Just don't use constant and const at the same time (although you should be able to)

magnum



On 16 apr 2012, at 13:28, myrice <qqlddg@...il.com> wrote:

> 
> 
> On Sun, Apr 15, 2012 at 10:31 AM, Solar Designer <solar@...nwall.com> wrote:
> OK, I've just tried it out.  There's a performance regression for CUDA:
>  
> user@...l:~/john/myrice-JtR/src$ ../run/john -te -fo=xsha512-cuda
> Benchmarking: Mac OS X 10.7+ salted SHA-512 [CUDA]... DONE
> Many salts:     40348K c/s real, 40106K c/s virtual
> Only one salt:  11796K c/s real, 11796K c/s virtual
> 
> I think I haven't changed cuda format so much. I believe no performance difference.I still design how to make PLAINTEXT_LENGTH work. In myrice-JtR, the format mainly tested on my G9600M GS. So the THREADS, BLOCKS and ITERATIONS are not suitable for your GTX570 card. However, if I push it to magnum-jumbo, it should be fine.
> I just check my github, the PLAINTEXT_LENGTH is 107, BLOCK is 32 and ITERATION is 8. So the performance is slow.
> 
> Now the OpenCL stuff:
> 
> user@...l:~/john/myrice-JtR/src$ ../run/john -te -fo=xsha512-opencl
> OpenCL platform 0: NVIDIA CUDA, 1 device(s).
> Using device 0: GeForce GTX 570
> Benchmarking: Mac OS X 10.7+ salted SHA-512 [OpenCL]... DONE
> Many salts:     25204K c/s real, 22649K c/s virtual
> Only one salt:  17086K c/s real, 15649K c/s virtual
> 
> I will make it find best configuration in my next commit. 
>  
> user@...l:~/john/myrice-JtR/src$ ../run/john -te -fo=xsha512-opencl -pla=1
> OpenCL platform 1: AMD Accelerated Parallel Processing, 2 device(s).
> Using device 0: Tahiti
> Compilation log: /tmp/OCLbj5mOR.cl(121): warning: type qualifier specified more than once
>        __global uint64_t* hash, uint32_t offset, __constant const char* salt)
>                                                             ^
> 
> /tmp/OCLbj5mOR.cl(224): warning: type qualifier specified more than once
>        __constant const char *salt)
>                   ^
> Benchmarking: Mac OS X 10.7+ salted SHA-512 [OpenCL]... FAILED (get_hash[0](0))
> 
> It is strange. On my NVIDIA card, if I remove the __constant, there will be errors when compile it.
> 
> Thanks!
> Myrice

Content of type "text/html" skipped

Powered by blists - more mailing lists

Your e-mail address:

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