According to http://www.suse.de/~gcctest/c++bench/polyhedron/, the run time of "NF" increased from 37.70s to 41.57s (+10%) between 071031 and 071101. This is an AMD Opteron system.
The same result I obtained at http://physik.fu-berlin.de/~tburnus/gcc-trunk/benchmark/ also on an AMD64. Interestingly, while x86-64 (-m64) gets slower, using -m32 I don't see any change (apart of noise) - neither for x87 nor for SSE(3). To boil it down to version numbers, I believe the change must have happened between the following versions: Fast: 4.3.0 20071031 (experimental) [trunk revision 129791] Slow: 4.3.0 20071031 (experimental) [trunk revision 129797] As culpits there are essentially only the following checkins possible: - PR tree-optimization/32377 (compute_overlap_steps_for_affine_univar) Make it work also for unknown number of iterations. [Sebastian Pop] - PR middle-end/33779 fold-const.c (extract_muldiv_1): Make sure to not introduce new undefined integer overflow. [Richard Guenther] - PR fortran/33897 decl.c (gfc_match_entry): Do not make ENTRY name global for contained procedures. parse.c (gfc_fixup_sibling_symbols): Fix code for determining whether a procedure is external. [Paul Thomas] The Polyhedron test case is available from: http://www.polyhedron.co.uk/MFL6VW74649 -- Summary: 10% performance regression since Nov 1 on Polyhedron's "NF" on AMD64 Product: gcc Version: 4.3.0 Status: UNCONFIRMED Severity: normal Priority: P3 Component: target AssignedTo: unassigned at gcc dot gnu dot org ReportedBy: burnus at gcc dot gnu dot org GCC target triplet: x86_64-unknown-linux-gnu http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34163
