http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51119
Tobias Burnus <burnus at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |burnus at gcc dot gnu.org --- Comment #2 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-11-14 13:08:49 UTC --- (In reply to comment #0) > Compared to ATLAS BLAS on an AMD 10h processor, MATMUL on square matrices with > n > 256 is around a factor of 8 slower. Side note: You can use -fexternal-blas -fblas-matmul-limit=<...> and link ATLAS BLAS. > Assigning to myself. > I have a cunning plan. I am looking forward to cunning ideas - at least if they are not too convoluted, work on all targets and are middle-end friendly.