Wrong (?) optimisation with gfortran 4.8.1 svn

[Thomas Koenig]

Am 26.08.2013 17:51, schrieb Sylvestre Ledru:
> Indeed but that requires to rebuild Scilab ;)

In this case, please open a PR with a description of what
went wrong, even if it is only "make; make install; make check",
at least with enough information that somebody can try to reproduce
it, for example by going back through gcc revisions and finding
out when the problem started. Depending on the build time of
scilab, this may be quite possible.

This may be a wrong-code regression in gcc (from what you described,
maybe not the Fortran front end, but possibly the middle end
or the i686 back end). It may also be a bug in the code that has
finally surfaced due to some optimizations that the compiler previously
didn't do. Both things have happened in the past.

One final question: Did you try to run under "-flto" ?

Regards

Thomas