Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20140911083912.GA8745@openwall.com>
Date: Thu, 11 Sep 2014 12:39:12 +0400
From: Solar Designer <solar@...nwall.com>
To: john-users@...ts.openwall.com
Subject: Re: nVidia Maxwell support (especially descrypt)?

On Thu, Sep 11, 2014 at 09:54:39AM +0200, magnum wrote:
> BTW I just dropped the DEScrypt-opencl format to the broken/ 
> subdirectory because it keeps giving me problems.

Ouch.  It worked for me before.

> For example, it fails 
> building under later AMD drivers due to excessive use of 'goto'. Some 
> other drivers commit suic^Wsegfault while trying to sort it out. I 
> believe using goto in an OpenCL kernel is really begging for problems.
> 
> To use it, simply move the files back from src/broken to src, 
> reconfigure and re-build. I hope Sayantan will fix it (as well as the 
> memory leak discussed before) at some point in time.

I'd say it's a compiler problem, but yes we may have to workaround it by
stopping to use goto, even though this kernel's uses of goto are
reasonable.  Code size might increase if we express the same things with
structured programming constructs, and thus speed might be lower.

Those goto's are not data-dependent (they depend on the current
iteration number only, which is the same for all work-items), so they are
OK for GPUs' hardware.

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.