Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Thu, 18 Dec 2014 16:45:35 +0300
From: Solar Designer <>
Subject: Re: Jumbo 1 release

On Tue, Dec 16, 2014 at 09:30:14AM +0100, magnum wrote:
> Current status: 

This currently lists four issues:

"Mention Markov mode and mask mode in doc/MODES?" - would be nice, but
if it's not done already, it might be too late for 1.8.0-jumbo-1, which
I might release _today_.

"DEScrypt-opencl disabled or not?" - it's a pity we "had to" disable
this format at all.  I doubt it's the only one producing errors with
_some_ Catalyst versions on _some_ cards.  Maybe we should have just
kept it enabled.  It's not like we have perfect quality code in other
formats anyway.  I am fine with whatever you guys decide, though, as
long as the format is easy to enable and use.

""linux-ppc64-altivec" target is broken in core" - this is tricky as I
don't have access to a system where I could reproduce the problem, and
the very same code built fine on another Linux/PPC64 system (just not on
Dhiru's).  This could as well be a distro/setup bug (which doesn't mean
we couldn't workaround it, though).  Not a release blocker anyway.

"release 1.8.0-jumbo1 (off bleeding-jumbo branch)" - sure.

> I believe we no longer have a single OpenCL format that doesn't 
> auto-tune! And all of them that desperately needed a "split kernel" has 
> got that too. A truckload of minor issues (including with performance) 
> were fixed the last couple of weeks and we're now just testing & fixing, 
> no new features will be added.


> Too bad we haven't heard from Sayantan for a while, I guess GPU-mask 
> will have to wait until J2. But the CPU-only mask mode has grown pretty 
> mature, with good speed, good node distribution and great support for 
> codepages if you wish (eg. ?L for lower-case 8-bit characters in the 
> currently selected encoding).

I think it's actually good we're not trying to get mask on GPU into this
same release.  There are way too many changes in it as it is.

> Incomplete list of tasks:
> - Fix the remaining issues.
> - Compile a human readable summary of changes. This is a huge job, there 
> are almost 6,000 commits. Perhaps the best way doing this is just 
> ignoring all separate commit descriptions and just review the format 
> list, relbench output, usage blob and new/changed doc files.
> - Review a single diff against core 1.8.
> - Review a single diff against J7.
> - Verify that the documentation is up to date.
> - Test on all odd systems we can think of (Sparc/BE tests were made 
> pretty recently).
> - More verifying of performance:
> -- J7 vs now.
> -- non-OMP vs OMP w/ just one thread.
> -- non-OMP vs OMP w/ threads.
> - More abort/resume tests, with and without nodes.
> - More ETA/progress tests, with and without nodes and before and after a 
> resume.

It feels unrealistic to "properly" take care of all of these, yet
release before the end of December.

I am tempted to go ahead and release what we've got today, with some
very minor changes of my own.

If I don't do it today, I might not do it for a while. ;-(


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.