john-dev - Re: [GSoC] John the Ripper support for PHC finalists

Follow @Openwall on Twitter for new release announcements and other news

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAKGDhHVKmjGoPWWEe1MAV3eVo806rrs3hwHL091isvV-2L1d5g@mail.gmail.com>
Date: Mon, 4 May 2015 01:18:46 +0200
From: Agnieszka Bielec <bielecagnieszka8@...il.com>
To: john-dev@...ts.openwall.com
Subject: Re: [GSoC] John the Ripper support for PHC finalists

2015-04-27 3:50 GMT+02:00 Solar Designer <solar@...nwall.com>:

> BTW, bumping into total GPU global memory size may be realistic with
> these memory-hard hashes.  Our TITAN's 6 GB was the performance
> limiting factor in some of the benchmarks here:
> http://www.openwall.com/lists/crypt-dev/2014/03/13/1

I use only 128MB
for the experimenst i removed almost everything from my code except
the biggest bottleneck:

#define F(i)  {                \
    i0 = ((i) - 0*4)  & mask1; \
    i1 = ((i) - 2*4)  & mask1; \
    i2 = ((i) - 3*4)  & mask1; \
    i3 = ((i) - 7*4)  & mask1; \
    i4 = ((i) - 13*4) & mask1; \
}

#define H(i, random_number)  {
             \
    for (j = 0; j < 128; j = j+4)
             \
    {
             \
        F(i+j);
             \
        index_local    = (((i + j) >> 2) - 0x1000 + (random_number &
0x1fff)) & mask;     \
        index_local    = index_local << 2;
                   \
\
        v1=vload4(0,S+sMAP(index_local));       \
        v= v+(v1<<1); \
        v1=v1+(v<<2);             \
        vstore4(v1,0,S+sMAP(index_local));               \
\
        random_number  = S[sMAP(i3)];              \
    }                                        \
}

and the gws number with the memory usage were the same, I can nothing
to do with this bottleneck

but If I remove everything from the code, GWS also doesn't differ

another thing

index_local    = (((i + j) >> 2) - 0x1000 + (random_number & 0x1fff))
& mask;     \

0x1fff*8 = 64KB, we could cache this segment but today graphic cards
rarely has 64KB of local memory, on super we have 48kB and 32KB
it is even worse because this is 64KB for the work-group

we don't know how much the __private memory we have , we can only see
if the kernel compilation failed or not, but I'm not sure of this

Im woried about using only 128 MB of memory

thanks

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.