Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Tue, 07 Jan 2014 06:44:05 +0100
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Re: CUDA multi-device support

On 2014-01-07 04:11, Muhammad Junaid Muzammil wrote:
> I am still not sure what was the issue with pull request as I think I
> sent a pull request for bleeding-jumbo branch. Will take a look at this.

It was from your bleeding-jumbo - but you accidentally targeted my 
unstable-jumbo branch (for various reasons that's unfortunately the 
default). But since it was correct in your fork, I could just reject the 
PR and cherry-pick them instead without trouble.

Always review your commits (use git gui and stage a hunk at a time as 
opposed to a blind "git add -a ." and always review things like PRs, 
traditional patches etcetera. If you look at 
https://github.com/magnumripper/JohnTheRipper/pull/469 you can 
immediately see that something is wrong. Example (top row): "junmuz 
wants to merge 2,815 commits into magnumripper:unstable-jumbo from 
junmuz:bleeding-jumbo" ;-)

Anyway, the patches themselves was good. We'll tweak some edge cases 
(eg. warn when MAX_CUDA_DEVICES get in the way, not silently skip some 
devices). Also, I think for CUDA we shold default to *all* devices as 
opposed to device 0. Anyone disagree? Doing the same to OpenCL is more 
questionable - unless we limit it to GPUs and accelerators (with 
fallback to CPUs if that's all there is).

BTW do you actually have any/some/unlimited access to a machine with 
more than one [logical] CUDA device? I think fixing pwsafe requires 
that. Unfortunately Solar's test machines can't provide this yet.

magnum

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.