Date: Sun, 11 Dec 2011 19:02:15 +0100 From: magnum <john.magnum@...hmail.com> To: john-dev@...ts.openwall.com Subject: SSE_PARA configuration I've been experimenting with different values for the *_SSE_PARA defines in arch.h and benchmarking raw-md4, raw-md5thick, md5 (crypt) and raw-sha1. It turns out the optimum values are not only dependant on compiler brand (which is what the current tests look at) but also exact compiler version, whether we're building 32-bit or 64-bit and what exact CPU we are building for - at least we need to distinguish intel from AMD. We could add #if clauses and make targets until our fingers bleed but I figure the real solution is to implement some sort of build-time tests similar to what happens in "make generic", but that finds the best *_SSE_PARA values for the specific platform. Some platforms even have one best MD5_SSE_PARA setting for raw-md5 and another for crypt-md5 and we need to pick one of them (probably the latter). I'm not sure where to start so I'll do nothing for now. Anyone have ideas? 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.