[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20121012223115.GD30294@openwall.com>
Date: Sat, 13 Oct 2012 02:31:15 +0400
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: bitslice DES on GPU
Hi Sayantan,
I'm sorry I have no time to seriously look into this sub-project lately.
On Fri, Oct 12, 2012 at 09:04:32AM +0530, Sayantan Datta wrote:
> I am now using bitselect() and AMD sboxes for AMD gpus.
Great. Does this no longer cause a slowdown?
> So my kernel is
> mostly at its final stages requiring the tweaks for latency and registers
> in the sboxes. Can you provide some idea regarding this.
I think it's premature. In fact, I think that several of your recent
optimizations were premature (I'm sorry I did not have time to comment).
What use is speeding it up by a few percent when it's like 10x slower
than expected? Well, you may do that to make it a bit more useful
temporarily, until we have time to seriously look into the bottlenecks
and hopefully resolve them.
> Also I'm thinking
> to port cmp_all and cmp_one to opencl. Also is it possible/viable to port
> get_hash() to opencl?
Of these, under the current formats interface, only cmp_all() is likely
reasonable to have on GPU. Further speedup in this area is possible
once we integrate the formats interface changes that I implemented for
myrice's work this summer (they're not committed yet, except to myrice's
experimental trees).
Thanks,
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.
