[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 27 Aug 2018 19:36:25 +0200
From: magnum <john.magnum@...hmail.com>
To: john-users@...ts.openwall.com
Subject: Re: Beignet: unable to find good values for
local_work_size
On 2018-08-27 11:54, Matus UHLAR - fantomas wrote:
> for a few weeks, the beignet OpenCL driver reports this message:
>
> Loaded 1 password hash (mysql-sha1-opencl, MySQL 4.1+ [SHA1 OpenCL])
> Beignet: "unable to find good values for local_work_size[i], please
> provide\n" " local_work_size[] explicitly, you can find good values with\n"
> " trial-and-error method."
>
> I understand it as the driver does not provide any useful value and the
> john
> is just guesssing.
>
> How can I test for the proper value? can I do it at runtime?
That message comes from Beignet, not john. What is triggering it is
likely that we deliberately use a local work size of NULL for incomplete
batches (eg. the very last batch in a session). There's not really any
way you can provide it... short of reverting to some commit before
https://github.com/magnumripper/JohnTheRipper/issues/3223 where we had
another scheme (which had other problems).
I would report it to them as a bug, because according to the OpenCL
spec. it should handle it. I believe what most drivers/runtimes do is
find greatest divisor of GWS that is less than or equal to highest
possible LWS, and use that.
Does it crash/abort or is this just a warning message? It could be that
it simply warns and then use an LWS of 1 which is suboptimal but works
fine in this case.
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.
