Date: Mon, 9 Jul 2012 01:24:14 +0530 From: Sayantan Datta <std2048@...il.com> To: john-dev@...ts.openwall.com Subject: Re: Rotate and bitselect investigation On Sun, Jul 8, 2012 at 2:48 PM, Solar Designer <solar@...nwall.com> wrote: > On Sun, Jul 08, 2012 at 02:41:08PM +0530, Sayantan Datta wrote: > > So the results are little slow on 570 and also on 4890. So if you want we > > can have two binary kernel specifiaclly tailored for these two and one > > general kernel for all other devices. Or we may go the opposite way too > i.e > > using binary for 7970. > > No, let's not bother with multiple kernel versions just because of this. > > Rather, if you have time, you may try optimizing SHA1_digest() further > (based on the known constants for some W elements), such that it > relies on the optimizer less. I suspect that it's this reliance on the > optimizer that results in bitselect() hurting performance on some cards. > This suggests that the overall speedup we're seeing on 7970 is less than > full as well - in other words, that there's some slowdown (more limited > propagation of the known constants in the optimizer) and a greater > speedup from bitselect() itself at once. The slowdown portion may be > avoided by better written source code. > > > Do you want me to commit it and push to git repo? > > Yes, please. Thanks. > > Alexander > I was able to squeeze in another bitselect optimization in the kernel resulting in 2% performance jump. std2048@...l:~/bin/run$ ./john -te -fo=mscash2-opencl -pla=1 OpenCL platform 1: AMD Accelerated Parallel Processing, 2 device(s). Using device 0: Tahiti Optimal Work Group Size:128 Kernel Execution Speed (Higher is better):1.581469 Benchmarking: M$ Cache Hash 2 (DCC2) PBKDF2-HMAC-SHA-1 [OpenCL]... DONE Raw: 103819 c/s real, 103819 c/s virtual Regards, Sayantan 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.