Created attachment 460348 [details, diff] Patch to switch "--enable-64-bit-words" on Flac-1.3.2 has a configure switch to enable 8 bytes words on 64bit systems, but it's disable by default. See this git commit: https://git.xiph.org/?p=flac.git;a=commit;h=6cc1cbad2c40a4406fd911445b86b3df2291dd79 Working on 8 bytes at times give some performance (5%-10% IIRC) improvement, so it would be nice to have it enabled in the ebuild. I've added a check for CHOST=x86_64 to the flac-1.3.2-r1 ebuild, and it works on my system, but i'm no expert on ebuild writing.
This can be easily implemeted by adding $([[ ${CHOST} == *x86_64* ]] && echo "--enable-64-bit-words") to myeconfargs. Do we want this?
(In reply to Filip Kobierski from comment #1) > This can be easily implemeted by adding > > $([[ ${CHOST} == *x86_64* ]] && echo "--enable-64-bit-words") (Well, yeah, that's what the attached patch already did.)
> the attached patch already I did not notice it. I have done a simple (but long) "benchmark" of 1.5.0 by doing 'ebuild compile; time ebuild test': # :: gentoo 1. real 12m5.876s; user 9m38.188s; sys 5m35.370s 2. real 11m48.366s; user 9m25.594s; sys 5m23.649s 3. real 12m9.248s; user 9m39.520s; sys 5m40.103s # 64bit words 1. real 11m45.112s; user 9m23.116s; sys 5m22.070s 2. real 12m3.832s; user 9m36.254s; sys 5m33.531s 3. real 12m11.313s; user 10m15.681s; sys 5m23.508s I see no significant differences and think a proper benchmark is needed to judge the changes. Shark, how did you observe this 5-10% performance improvement? If this were to be added, it would make sense to test not only for amd64, but for a 64 bit arch in general, right?
