Re: [med-svn] r2775 - trunk/packages/velvet/trunk/debian
Le Sat, Dec 13, 2008 at 09:59:20PM -0600, Steve M. Robbins a écrit :
>
> As you say, it could be worthwhile to benchmark to see whether the
> extra optimization is useful. But benchmarking is a tricky business:
> it can be hard to benchmark all use cases of the software (I have no
> idea what "velvet" is, so I'm only speaking in generalities). So one
> might be inclined rather to follow upstream here. What would be the
> harm?
It would harm my lazyness :) For the moment, debian/rules is only four lines,
and automagically takes care of the "noopt" option of DEB_BUILD_OPTIONS. This
option goes through CFLAGS, which come before OPT in Upstream's makefile, so
leaving Upstrem's default breaks DEB_BUILD_OPTIONS. I of course can rewrite
debian/rules from scratch using Debhelper.
I admit that it I am not comptent enought to pick the best optimisation level.
In a very interseting thread on -devel it was even noticed that -Os could be
the fastest when the difference in size was just enough for the program to fit
on the processor's cache. The reverse is said elsewhere to be true for -O3…
I will do what Upstream thinks and let you know.
Have a nice Sunday,
--
Charles Plessy
Debian Med packaging team,
http://www.debian.org/devel/debian-med
Tsurumi, Kanagawa, Japan
Reply to: