On Tue, Oct 30, 2012 at 12:25 PM, Steven Bosscher <stevenb....@gmail.com> wrote: > On Tue, Oct 30, 2012 at 5:59 PM, Teresa Johnson wrote: >> I will try testing your patch on top of mine with our fdo benchmarks. > > Thanks. But you should expect a lot of errors, hopefully you can make > something out of it for Bugzilla. > >> For the others on the cc list, you may need to include my patch as >> well for testing. Without it, -freorder-blocks-and-partition was DOA >> for me. For my patch, see >> http://gcc.gnu.org/ml/gcc-patches/2012-10/msg02692.html > > Ah, the fate of a pass that isn't enabled by default.
I see only 5 test cases under tree-prof for function splitting. More test cases are needed. David > From what I can > tell, looking at this code the past few hours, it's hopelessly broken > at the moment: > > * It doesn't work with cfglayout mode, even though it is in between > pass_into_cfg_layout_mode and pass_outof_cfg_layout_mode. Coming out > of cfglayout mode > > * Coming out of cfglayout mode, fixup_reorder_chain adds > REG_CROSSING_JUMP -- but so does force_nonfallthru_and_redirect called > just before it. So we end up with >1 such note, or with notes on jumps > that shouldn't have them. > > * The scheduler doesn't respect the partitioning at all. remove_notes > happily removes the section split note. > > * etc. > > > This pass may need some major work to be in good order again. > > Ciao! > Steven
