You are not logged in.
Pages: 1
I just grabbed the octave PKGBUILD from abs, and part of it looks like this:
# http://www.nabble.com/Random-rounding-errors-td16010966.html
FFLAGS="-O -ffloat-store" \
./configure --prefix=/usr --libexecdir=/usr/lib \
--enable-shared --disable-static
Why -O and not -O2 ? When I compile from this PKGBUILD the gfortran calls definitely have -O in them. I'd have though -O2 would be appropriate, especially for a numerical package.
Anyway, on my system I rebuilt with -O2 and it's working fine. I haven't done any tests to see if anything is actually faster, though.
Offline
According to the link in the comment in the PKGBUILD (http://www.nabble.com/Random-rounding-e … 10966.html), there are detectable and unpredictable rounding errors in Octave when not using -ffloat-store. -O2 may enable some math optimizations that make them show up, too. However, that thread is from 2008, and the issues may well be fixed by now -- someone should test it.
Offline
Pages: 1