I believe we have a compiler problem with gcc42.
I had a problem with scilab under both -stable and -current.
I installed the ports with the default make files. scilab failed at
run time. ports/116378.
Math/scilab uses math/lapack and math/blas and builds everything with
gcc42 and gfortran42.
Charles Cox wrote:
I would like to add that some of us who do a lot of numerically intensive
programming, and that need to squeeze every last available cycle out of
our CPU's would really appreciate having -O2 available for userland
programs. To me, getting rid of the -O2+ switch would be
Actually I have several Alpha's, and one I have to reload due to this, so
I will give this a try on the other machine and will let you know if I have
good success with it. If this all works, then I agree with you that we
want to avoid, or at least note the problem with -O2 as I have been using
I'm sorry, at best here we're talking a few tenths of a
percentage point of increased speed (if that), and maybe a few
milliseconds saved -- right?
Actually the difference between -O and -O2 the times i've looked
at the generated code was substantial, but I guess that brings
along
