Date: Sat, 9 May 2015 02:53:02 +0300 From: Solar Designer <solar@...nwall.com> To: john-dev@...ts.openwall.com Subject: Re: [john dev] [PHC support] parallel Agnieszka, On Sat, May 09, 2015 at 01:25:22AM +0200, Agnieszka Bielec wrote: > I've implemented the basic version of parallel (PHS) (SHA-512). I took a look at commit 5746012c73e10f29d662c02f4421859d15589d91. Here are some assorted issues I noticed, in arbitrary order: You edited parallel.h compared to Steve's original, yet your edited file still says it's by Steve, and does not mention you. This is wrong. Steve shouldn't be responsible for your bugs, if any, and it should be clear who the copyright holders of each file are and what it's license is. Steve's parallel.h doesn't mention a license, but perhaps it was elsewhere in the Parallel package? In fact, it's unclear which Parallel .h file your parallel.h is based on. Can you clarify? "#ifdef ARCH_LITTLE_ENDIAN" is wrong. We use "#if ARCH_LITTLE_ENDIAN" elsewhere, since we define this to either 0 or 1. Among the test vectors, please include some with non-zero sequential cost. Notice how Parallel uses its t_cost: parallelLoops = 3 * 5 * 128 * calcLoopCount(t_cost & 0xffff); sequentialLoops = calcLoopCount(t_cost >> 16); What do you mean by "//to do: change char to short integers" for get_salt()? cmp_all() and cmp_one() include some magic for length 256, I guess inherited from POMELO and irrelevant for Parallel. The "struct fmt_main fmt_parallel" initializer is badly mis-formatted, just like POMELO's. You need to format them properly. Please see dummy.c for a cleaner format definition. Overall, this stuff looks extremely dirty. You need to clean it up before you proceed to work on it further. After the cleanups, your next priority may be OpenCL implementation, and after that (or while waiting for feedback from me) a SIMD implementation. Thanks, 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.