Date: Thu, 4 Oct 2012 18:41:59 -0300 From: Claudio André <claudioandre.br@...il.com> To: john-dev@...ts.openwall.com Subject: Re: Office2013 OpenCL 2012/10/4 magnum <john.magnum@...hmail.com>: > > I just committed an OpenCL Office 2013 format, using your SHA-512 (but optimised for loop use). It works fine on nvidias and CPU but fails self-test on HD7970. I think the code is fine so it may be some kind of driver bug that has to be worked around. If you have some spare time, please have a look. I will. Just for fun, let me tell you a story. I wasted a couple of day trying to figure out what i was doing wrong in the split task. After a lot of debugging i realized there is a bug in 12.6 AMD driver/compiler or whatever. Remenber i was running on my 6770 (not new) hardware, updated driver, etc. I saw you committed some "work around" for RAR on OSX. It is sad we are fighting so hard with AMD and NVIDIA (and APPLE?) software. > > BTW it does around 44 million SHA-2 per second on the 570 and over 75 million on HD7970. Are those speeds fair or lousy? There is a huge need for private memory so I have no idea how to get it faster. Well, i tried once to compare raw 512 and 512crypt. Then, i realized that on sha512crypt i measure rounds (not 'digest' operations, 'neigher' blocks operations). Not sure if i am clear here, but your number seems to be ok if i can compare your solution with mine. Anyway, i will try to think about it while looking at the code. And yes, the pressure for registers is amazing. I haven
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.