Date: Sun, 26 Apr 2015 00:31:49 +0300 From: Solar Designer <solar@...nwall.com> To: john-dev@...ts.openwall.com Subject: Re: [GSoC] John the Ripper support for PHC finalists On Sat, Apr 25, 2015 at 10:28:31PM +0200, Agnieszka Bielec wrote: > 2015-04-25 21:20 GMT+02:00 Solar Designer <solar@...nwall.com>: > > "4 GPU units" sounded weird and confusing to me. > > If you put a hash of a known password in a text file, and put the > > password somewhere down a wordlist, and run "john" with pomelo-opencl > > against that, will it crack the hash? > > It works Can you explain how/why it works? I am probably missing something. I would expect it to fail most of the time. > > You don't actually have any test vectors in > > opencl_pomelo_fmt_plug.c, so when you --test you don't actually test, > > right? :-( > > no, now I keep in fmt_tests many hashes with various costs > and it's 100 lines of code. Before I was generating hashes to structure > fmt_test dynamically because it was convenient for me. > this version of code was rather intended for experiments Oh, I looked at the gpu4_cache branch. Now I see you do have test vectors in the main branch. That's good. > > Also, I think you're over-using "unsigned long", such as for loop > > variables. This may have performance impact on GPUs. You should prefer > > "uint" except where a wider type is actually needed. > > unsigned long is necessary where m_cost is equal to 21 or bigger > I can test if it will increase the speed but I'm sceptic about that We don't need m_cost >= 21 in the OpenCL-enabled format. OpenCL typically means running lots of concurrent instances, and m_cost = 21 is way too high for that on current devices. That's 16 GB per instance! Please check for this in the OpenCL-enabled format's valid(), print a warning (at most once), and have valid() return 0 to skip such hashes. 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.