Date: Tue, 21 Aug 2012 22:45:43 +0400 From: Solar Designer <solar@...nwall.com> To: john-dev@...ts.openwall.com Subject: Re: OpenCL optimization trick for md* Hi Alain, On Tue, Aug 21, 2012 at 09:51:52AM -0700, Alain Espinosa wrote: > Hi. Share some small OpenCL optimization i will use in next Hash Suite. Thank you! > Work in nt_kernel.cl, md4_kernel.cl, md5_kernel.cl, sha1_kernel.cl and > possible others. These are deliberately not optimized at this time - no point in having their crypto code run faster when it's bottlenecked by password generation on CPU anyway. In myrice's branch, md5_kernel.cl includes this kind of optimizations, though, because that branch is about password generation on GPU experiments. > In nt_kernel.cl change all round 1 steps: > a += (d ^ (b & (c ^ d))) + nt_buffer ; a = rotate(a , 3u ); > by > a += bitselect(d, c, b) + nt_buffer ; a = rotate(a , 3u ); Yeah. > Work in nt_kernel.cl and md4_kernel.cl but not sure if it is an > improvement. In nt_kernel.cl change all round 2 steps: > a += ((b & (c | d)) | (c & d)) + nt_buffer + SQRT_2; a = rotate(a , 3u ); > by > a += bitselect(bitselect(b,c,d), bitselect(d,b,c), b) + nt_buffer + > SQRT_2; a = rotate(a , 3u ); Now this is curious. It's similar to what Sayantan did for SHA-1's round 3. We should try it for XOP and see if it helps there. Thanks again, 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.