Date: Tue, 30 Jun 2015 13:51:17 -0400 From: Rich Felker <dalias@...c.org> To: musl@...ts.openwall.com Subject: Re: [PATCH v2 5/6] dynlink.c: compute modulus via magic multiplication On Sun, Jun 28, 2015 at 02:48:34AM +0300, Alexander Monakov wrote: > Based on http://ridiculousfish.com/blog/posts/labor-of-division-episode-iii.html > > Do a little hand-holding for the compiler and fold magic post-shift into > 32-bit high word -> low word shift on 64-bit platforms. > --- On 32-bit Atom, I was able to get an improvement from 116ms to 113ms running an empty program linked with a .so produced using the attached script. This is probably roughly best-case improvement, so the umod stuff is probably not likely to be worthwhile unless/until we can improve the areas that are currently dominating the runtime. That's rather disappointing, but maybe we can improve enough other stuff for it to make sense to do the umod optimization. Rich Download attachment "gen.sh" of type "application/x-sh" (104 bytes)
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.