On Tue, Nov 19, 2013 at 1:00 PM, Andi Kleen <a...@firstfloor.org> wrote: > Teresa Johnson <tejohn...@google.com> writes: > >> This patch enables -freorder-blocks-and-partition by default for x86 >> at -O2 and up. It is showing some modest gains in cpu2006 performance >> with profile feedback and -O2 on an Intel Westmere system. Specifically, >> I am seeing consistent improvements in 401.bzip2 (1.5-3%), 483.xalancbmk >> (1.5-3%), and 453.povray (2.5-3%), and no apparent regressions. > > One thing that worries me is what this will do to profilers. > > I had to hack some assembler code using out of line sections > to able to handle the libunwind based perf dwarf unwinder. > > My understanding is that these out of line sections can be > only described properly in dwarf3, and there's still some > dwarf2 based unwinder code around. > > So this may cause problems with profiling and debugging. > > It's probably still a good idea, just may need some extra > care. > > -Andi
Sri has approval for a patch that should address this by giving the split cold sections a label. It should go in today as well: http://gcc.gnu.org/ml/gcc-patches/2013-11/msg02143.html Thanks, Teresa > > -- > a...@linux.intel.com -- Speaking for myself only -- Teresa Johnson | Software Engineer | tejohn...@google.com | 408-460-2413