john-dev - Re: async key transfers to GPU

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

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

Message-ID: <CANJ2NMPpUavfAWrJL6m=z3pbXFqV3yT7ZouKHGD_6ER0xbK87Q@mail.gmail.com>
Date: Thu, 28 Jun 2012 00:33:07 +0800
From: myrice <qqlddg@...il.com>
To: john-dev@...ts.openwall.com
Subject: Re: async key transfers to GPU

Here are the result with 2 streams. Stream #1 memcpyH2D first half of
keys in set_key(). Stream #2 memcpyH2D second half of keys in
beginning of crypt_all(). Then I split xsha512 kernel into 2 with 2
streams.

Here is the profiler data: http://myrice.me/gsoc/async.png

There is only 4.2% overlap. memcpyH2D is 2 time slow on bull. And two
xsha512 kernels execution time are slight slow than one.

So avoid data copy and generate password on GPU would be very helpful.
It is time for me working it out quickly.

Thanks
myrice

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