Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 11 Aug 2013 19:29:43 +0400
From: Solar Designer <solar@...nwall.com>
To: john-users@...ts.openwall.com
Subject: Re: Cuda issues on OSX

kzug -

On Sun, Aug 11, 2013 at 04:25:39PM +0200, magnum wrote:
> On 11 Aug, 2013, at 16:11 , kzug <kzug10@...il.com> wrote:
> > I have a bit of an issue compiling JtR Cuda and any help would be appreciated: 
> > When attempting to compile with  "make clean macosx-x86-64-cuda"
> > I got the following error: 
> > cd cuda; nvcc -c -Xptxas -v -arch sm_10 -m64 cuda_common.cu -o ../cuda_common.o
> > /bin/sh: nvcc: command not found
> 
> Add /usr/local/cuda/bin to your PATH.

Besides setting the proper PATH, you will likely want to use the -gpu or
-opencl make target to build JtR for, not -cuda.  Our OpenCL kernels are
more numerous and (maybe counter-intuitively) generally faster than CUDA
ones.  So you'll probably want to use the -opencl formats, not -cuda ones.

In part this is related to auto-tuning, which is available in many of
our OpenCL-enabled JtR formats, but not in any of the CUDA ones - so
to get good speeds out of CUDA you'd have to edit some .h files to fit
your specific GPU better (there is some info on that in doc/README-CUDA).
With most of our OpenCL-enabled formats, there's no such issue - so just
use those instead of CUDA ones.

Alexander

Powered by blists - more mailing lists

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