[Bug 1177260] Switching python-numpy to plain BLAS causes significant performance regression
https://bugzilla.suse.com/show_bug.cgi?id=1177260 https://bugzilla.suse.com/show_bug.cgi?id=1177260#c51 --- Comment #51 from Egbert Eich <eich@suse.com> --- (In reply to Richard Biener from comment #50)
Maybe you can just override LD? OTOH to get libgfortran linked the link probably uses FC, not CC, or does it explicitely add -lgfortran? To me it looks like an issue in the openblas makefiles.
It uses CC for linking - which points to gcc11 (on Leap/SLE) while FC which points to the stock gfortran is used to build the fortran sources. The latter may not be all that problematic, however, a binary may use other fortran libraries built with the stock gfortran compiler or be built/linked itself using this compiler. And, yes, the final link (done with $CC) adds -lgfortran. The problem is that we need to link C and Fortran objects together and make sure, the right libgfortran is used. I will try what happens if I naively set -lgfortran4. Likely, gcc11 will not find this library. -- You are receiving this mail because: You are on the CC list for the bug.
participants (1)
-
bugzilla_noreply@suse.com