There are several things that queued up: - scalar cleanup after final inlining should be done before computing aliasing (basically, re-run what we do for early optimization also after final inlining, but at least run copyprop, ccp, forwprop and dce)
- run an early complete loop unrolling pass unrolling _only_ innermost loops in loop nests (maybe only at -O3, setting up of loop info is expensive) - run DCE after vectorization, the IL is completely hosed for tree based costs otherwise (affects unrolling costs) -- Summary: Tree optimization pipeline needs re-tuning Product: gcc Version: 4.3.0 Status: UNCONFIRMED Keywords: missed-optimization Severity: normal Priority: P3 Component: tree-optimization AssignedTo: rguenth at gcc dot gnu dot org ReportedBy: rguenth at gcc dot gnu dot org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34416