[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAKGDhHX0UWNMUjSuXayK=2YpW9UK69tCrWmHVWbKHAudtZepYg@mail.gmail.com>
Date: Fri, 3 Apr 2015 04:31:28 +0200
From: Agnieszka Bielec <bielecagnieszka8@...il.com>
To: john-dev@...ts.openwall.com
Subject: Re: [GSoC] John the Ripper support for PHC finalists
2015-04-02 10:33 GMT+02:00 Agnieszka Bielec <bielecagnieszka8@...il.com>:
> 2015-04-01 19:48 GMT+02:00 Frank Dittrich <frank.dittrich@...lbox.org>:
> >On 04/01/2015 06:56 PM, Agnieszka Bielec wrote:
> >> there is a way to pass options by comand line somethig alike:
-m_cost=x ?
> >> I need it for the auto_tune and compiling the kernel.
>
> >There is a --cost= command line option, see doc/OPTIONS
> >The --cost option can be used in regular runs and with --test, as long
> >as you have test vectors matching the requested cost settings.
>
> >To use thze --cost option for pomelo, you need to implement reporting
of
> >the tunable costs.
> >It is easy. See the django-scrypt format as an example.
> >(bleeding-jumbo)JtR $ grep -A 5 "#if FMT_MAIN_VERSION > 11"
> >src/django_scrypt_fmt_plug.c
> >#if FMT_MAIN_VERSION > 11
> >static unsigned int tunable_cost_N(void *salt)
> >{
> > static struct custom_salt *my_salt;
> >
> > my_salt = salt;
> >--
> >#if FMT_MAIN_VERSION > 11
> > {
> > "N",
> > "r",
> > "p"
> > },
> >--
> >#if FMT_MAIN_VERSION > 11
> > {
> > tunable_cost_N,
> > tunable_cost_r,
> > tunable_cost_p
> > },
> thanks but it seems that first is executed init function (where I need
access to cost) before functions tunable_cost_*
if this is impossible I must add my option to JohnTheRipper or make
another big modifications. Should I do it? I wanted to avoid this
and today I was fighting with the format and moved the init function
to be called inside set_salt
"
if(last_t_cost!=t_cost || last_m_cost!=m_cost)
{
if(last_t_cost !=0 && last_m_cost !=0)
done();
last_t_cost=t_cost;
last_m_cost=m_cost;
init(fmt_pomelo_tmp,t_cost,m_cost);
}
"
and many more tricks, now it works but I have some bugs and
also as before it is necessary to make modifications in files
like loader.c
I've added atomatic generating hashes for autotune
I know, there are some things I must to correct but I want to make
most important things before
now in my repository M_COST and T_COST are hardcoded
to value 2
tests for M_COST=2 and T_COST=2:
It's better than before but 2 is a very little value for POMELO
[a@...er run]$ ./john --format=pomelo-opencl -dev=2 --test
Device 2: Intel(R) Xeon(R) CPU E5-2670 0 @ 2.60GHz
Local worksize (LWS) 1, global worksize (GWS) 2048
Benchmarking: pomelo-opencl, POMELO [POMELO OpenCL (inefficient,
development use only)]... DONE
Raw: 200704 c/s real, 6432 c/s virtual
[a@...er run]$ ./john --format=pomelo-opencl -dev=3 --test
Device 3: Intel(R) Xeon(R) CPU E5-2670 0 @ 2.60GHz
Build log: Compilation started
Compilation done
Linking started
Linking done
Kernel <pomelo_crypt_kernel> was not vectorized
Done.
Local worksize (LWS) 8, global worksize (GWS) 65536
Benchmarking: pomelo-opencl, POMELO [POMELO OpenCL (inefficient,
development use only)]... DONE
Raw: 198593 c/s real, 6403 c/s virtual
not vectorized but fast, should I worry about this?
[a@...er run]$ ./john --format=pomelo-opencl -dev=4 --test
Device 4: Intel(R) Many Integrated Core Acceleration Card
Build log: Compilation started
Compilation done
Linking started
Linking done
Build started
Kernel <pomelo_crypt_kernel> was successfully vectorized
Done.
Local worksize (LWS) 64, global worksize (GWS) 64
Benchmarking: pomelo-opencl, POMELO [POMELO OpenCL (inefficient,
development use only)]... DONE
Raw: 34.2 c/s real, 6400 c/s virtual
[a@...er run]$ ./john --format=pomelo-opencl -dev=5 --test
Device 5: GeForce GTX TITAN
Local worksize (LWS) 64, global worksize (GWS) 8192
Benchmarking: pomelo-opencl, POMELO [POMELO OpenCL (inefficient,
development use only)]... DONE
Raw: 98304 c/s real, 98304 c/s virtual
but
./john --format=pomelo-opencl -dev=1 --test
Device 1: Tahiti [AMD Radeon HD 7900 Series]
OpenCL error (CL_OUT_OF_RESOURCES) in file (opencl_pomelo_fmt_plug.c) at
line (519) - (failed in clEnqueueNDRangeKernel)
it seems that I can allocate only wery small amounts of __private memory
even if I set MEM_SIZE to 3000 --dev=1 crashes
2015-03-30 11:51 GMT+02:00 magnum <john.magnum@...hmail.com>:
>On 2015-03-30 11:11, magnum wrote:
>> Also, your auto-tune settings are totally wrong, possibly ending up in
>> suboptimal work sizes. I will fix them and submit a patch.
>This patch fixes the auto-tune problems. BTW with my other suggestions
>it also runs fine on the GTX Titan, at about same ~10K c/s speed as on
>HD7970. And now it also runs on my Macbook GPU (1290 c/s).
thank you
Content of type "text/html" skipped
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.
