Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Fri, 14 Jun 2019 13:55:00 +0200
From: Solar Designer <solar@...nwall.com>
To: john-users@...ts.openwall.com
Subject: Re: Unable to use OpenCL on arm64v8/debian stretch

On Thu, Jun 13, 2019 at 10:30:54PM -0400, billwhittaker@...e-mail.net wrote:
>       ./configure --enable-mpi

BTW, why do you care to build with MPI support?  How many of these
boards do you intend to use?

>       make -s clean && make -sj6
>       ../run/john --test=0 --format=cpu
> Testing: crypt, generic crypt(3) [?/64]... (6xOMP) PASS
> All 404 formats passed self-tests!

This looks right.  We have 407 CPU formats in 1.9.0-jumbo-1 by default,
but a few get disabled at compile-time on non-x86 systems.

> Incidentally, I am working on a Rock Pi. When I run the binary compiled on a Rock64, I see,
> All 705 formats passed self-tests!

Perhaps you don't have all of the .conf files in place when you run the
other binary.  We have dynamic_disabled.conf, which disables plenty of
dynamic formats by default.

Looking at the specs for these boards, the Rock Pi has two kinds of CPU
cores: two faster cores and four slower cores.  This means OpenMP might
not use them optimally by default.  You'd need to split them into two
sets (using GOMP_CPU_AFFINITY and OMP_NUM_THREADS) and use separate
instances of JtR (with "--session"), or try OpenMP dynamic scheduling
("OMP_SCHEDULE=dynamic") along with our auto-tuning ("--tune=auto"), or
give up on OpenMP and use "--fork" (will result in best performance, but
has other drawbacks).  This is in case you use CPU formats, not OpenCL.

I say "might not" rather than "will not" "use them optimally" because
we don't know whether the CPU would possibly throttle when you attempt
to use it fully.  So it is possible that under-using the two faster
cores is actually optimal.

Of course, you'd only try this for fun.  These boards won't deliver
great performance no matter what you do, unless you have a lot of them.

For comparison, here are some other benchmarks of password cracking on
mobile CPU and GPU vs. desktop CPU with its embedded GPU:

https://hashsuite.openwall.net/android

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.