[Bug tree-optimization/54776] New: [4.8 Regression] tramp3d-v4: 20% performance regression using -O3

["markus at trippelsdorf dot de" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54776

             Bug #: 54776
           Summary: [4.8 Regression] tramp3d-v4: 20% performance
                    regression using -O3
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: markus@trippelsdorf.de


With gcc-4.8 (--enable-checking=release):

markus@x4 ~ % time c++ -w -O3 tramp3d-v4.cpp
c++ -w -O3 tramp3d-v4.cpp  24.87s user 0.34s system 99% cpu 25.293 total
markus@x4 ~ % ./a.out --cartvis 1.0 0.0 --rhomin 1e-8 -n 20
...
Time spent in iteration: 7.35642

With gcc-4.7.2:

markus@x4 ~ % time c++ -w -O3 tramp3d-v4.cpp
c++ -w -O3 tramp3d-v4.cpp  25.15s user 0.33s system 99% cpu 25.568 total
markus@x4 ~ % ./a.out --cartvis 1.0 0.0 --rhomin 1e-8 -n 20
...
Time spent in iteration: 5.81199

LTO doesn't help much (gcc-4.8):

markus@x4 ~ % time c++ -w -O3 -flto  tramp3d-v4.cpp
c++ -w -O3 -flto tramp3d-v4.cpp  45.78s user 0.95s system 99% cpu 47.012 total
markus@x4 ~ % ./a.out --cartvis 1.0 0.0 --rhomin 1e-8 -n 20
...
Time spent in iteration: 7.2111


(For comparison here are some clang results:

markus@x4 ~ % time clang++ -w -O3 tramp3d-v4.cpp
clang++ -w -O3 tramp3d-v4.cpp  14.67s user 0.12s system 99% cpu 14.874 total
markus@x4 ~ % ./a.out --cartvis 1.0 0.0 --rhomin 1e-8 -n 20
...
Time spent in iteration: 6.1923

markus@x4 ~ % time clang++ -w -O3 -flto tramp3d-v4.cpp
clang++ -w -O3 -flto tramp3d-v4.cpp  20.28s user 0.16s system 99% cpu 20.535
total
markus@x4 ~ % ./a.out --cartvis 1.0 0.0 --rhomin 1e-8 -n 20
...
Time spent in iteration: 4.47936

That's an almost 28% improvement due to -flto)