This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug tree-optimization/54776] [4.8 Regression] tramp3d-v4: 20% performance regression using -O3
- From: "hubicka at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: Sat, 06 Oct 2012 15:07:58 +0000
- Subject: [Bug tree-optimization/54776] [4.8 Regression] tramp3d-v4: 20% performance regression using -O3
- Auto-submitted: auto-generated
- References: <bug-54776-4@http.gcc.gnu.org/bugzilla/>
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54776
--- Comment #9 from Jan Hubicka <hubicka at gcc dot gnu.org> 2012-10-06 15:07:58 UTC ---
Mainline -O3
Time spent in iteration: 11.0926
LTO (with today fix for resolution info)
Time spent in iteration: 10.9666
LTO with V1 API hack disabled
Time spent in iteration: 9.58262
Whole program
Time spent in iteration: 9.59799
Mainline -O3 --param early-inlining-insns=13 (default is 11)
Time spent in iteration: 8.94558
So the LTO problem is solved. I will try to look at the functions with
estimated cost of 12 and 13 to see if I can reduce it. Otherwise I guess we
could simply bump up the early-inlining-insns. It was set lower based on
tramp3d and because of accounting inconsistency.