There has been a performance drop of approx. 8% in facerec benchmark with at -O2 optimizatio level, since 2005-10-17 on ppc.
After doing little regression hunt, found following patch causing the drop, http://gcc.gnu.org/ml/gcc-cvs/2005-10/msg00586.html As this patch has two parts, one fixes bug #23714 and other removes the call to mark_array_ref_addressable() and related definitions. As you can tell these two set of changes does two different things. First one fixes pr #23714. I have verified that, the second set of changes causing the performance drop. After reverting only the second part from the above patch and running facerec benchmark at -O2, I could recover the old better numbers (~+8%). Any thoughts on this? I could revert the change (only second part) with the new patch unless there is a reason (which for the moment I'm not able to understand) for removing the call to mark_array_ref_addressable(). -- Summary: facerec performance regression Product: gcc Version: unknown Status: UNCONFIRMED Severity: normal Priority: P2 Component: regression AssignedTo: unassigned at gcc dot gnu dot org ReportedBy: uttamp at us dot ibm dot com GCC build triplet: powerpc64-linux GCC host triplet: powerpc64-linux GCC target triplet: powerpc64-linux http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24565
