Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 9 Mar 2012 00:23:17 +0200
From: Milen Rangelov <>
Subject: Re: Re: [JtR PATCH] Support rar's -p mode by spawning
 external unrar process.

Not a good idea to decrypt on GPU with -hp, not at least on AMD hardware
with OpenCL (yes, I tried it ). This increases the GPR pressure and that
impacts the whole kernel performance, it's not worth it. What I did not try
is separate AES kernel post the setcryptkeys one, but I suspect kernel
launch latency and memory transfers would make the whole excercise

I also had a look at crark's CUDA code and I don't think it's quite
optimal.  Not quite glad with the speeds on my AMD hardware too. Porting
that to OpenCL would be a disaster, at least on AMD hardware, it's like a
quick port from CPU to GPU :) Too much loops, too much branches, byte
addressable stores, the SHA1 implementation is like taken from OpenSSL too

Anyway I am working on 7z now, things are even more fun here.

On Thu, Mar 8, 2012 at 10:31 PM, magnum <> wrote:

> On 02/29/2012 03:28 PM, Samuele Giovanni Tonon wrote:
> > btw
> > for those interested there's the cuda source available
> > for crark .
> >
> >
> I wonder if that code is what is used today (timestamps say 2009). The
> code seems to be almost pristine OpenSSL C code. Nothing special except
> he skips update() and inlines that functionality in the loop. Looks like
> a good idea. Our current sha1 kernel could easily be modified to keep a
> context.
> He decrypts in GPU too. For -hp mode that might be a good idea but for
> -p mode I suspect AES-NI is they key to success if your CPU supports it.
> I have tried crark but not igrargpu, are they about the same in speed on
> a specific card?
> magnum

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.