[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <20150825060655.GB30942@openwall.com>
Date: Tue, 25 Aug 2015 09:06:55 +0300
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: LWS and GWS auto-tuning
magnum, all -
We ought to do something about the auto-tuning. Here are some ideas:
Maybe have a table of per card type likely optimal LWS (or multipliers
for powers of 2).
For both LWS and GWS, maybe after an optimal power of 2 is determined,
try intermediate values between it and two adjacent ones (previously
tested). We may apply existing numerical methods here, starting from
trivial bisection for LWS or GWS alone (with the other fixed), but not
necessarily limited to just that:
https://en.wikipedia.org/wiki/Mathematical_optimization#Computational_optimization_techniques
With a good algorithm choice, we could have it fairly quickly converge
to much more optimal LWS and GWS than the auto-tuning achieves now.
This sounds like an interesting project for someone curious in applying
numerical methods to a real-world task.
We could also add a long-running auto-tuning mode that would test more
combinations and populate a cache file with info on different GPUs and
formats (or one --format if specified). We could then distribute a
pre-populated cache file with JtR.
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.
