Date: Sun, 7 Oct 2012 20:28:13 +0200 From: magnum <john.magnum@...hmail.com> To: john-dev@...ts.openwall.com Subject: Re: Office2013 OpenCL On 6 Oct, 2012, at 18:47 , Claudio André <claudioandre.br@...il.com> wrote: > Em 06-10-2012 09:34, magnum escreveu: >> On 5 Oct, 2012, at 16:31 , Claudio André <claudioandre.br@...il.com> wrote: >> >>> Hi, your code works fine here (as is). Full output below. >>> >>> In such conditions, find the error on 7970 could take some time. >> Yes. Dhiru reports it is working fine on 6970 too. >> >> Thanks, >> magnum > > I have a candidate. Do you remember we are using this not (only) because of performance, but because do the 'shifts' was not working? > #define SWAP64(n) (as_ulong(as_uchar8(n).s76543210)) > --- > > I think we should investigate the 'shift's you are using inside your code. Seems to be a good start, and you are using a few of them. The same SHA-512 function (and shift stuff etc.) works fine in your's and Myrice's formats, I just changed it to work directly on the input buffer, and skip using a context struct. I have tried changing the swaps and bitselect stuff to the generic code the nvidias use, but it doesn't change a thing. Also, I change the hash loops to use uint throughout (getting rid of lots of shifting) but it neither affected speed nor outcome - it still fails (and still passes on other GPUs). Given it works fine on all other systems I know of (well except Apple but that is clearly a driver problem too), I regard this a driver bug and I will just ignore it for now. Thanks, magnum
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.