This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: [RFC/patch] Callgraph based inlining heuristics
- From: Gerald Pfeifer <pfeifer at dbai dot tuwien dot ac dot at>
- To: Jan Hubicka <jh at suse dot cz>
- Cc: gcc at gcc dot gnu dot org, Mark Mitchell <mark at codesourcery dot com>,rth at redhat dot com, aj at suse dot de
- Date: Mon, 23 Jun 2003 12:50:50 +0200 (CEST)
- Subject: Re: [RFC/patch] Callgraph based inlining heuristics
- References: <20030622232032.GC3266@kam.mff.cuni.cz>
On Mon, 23 Jun 2003, Jan Hubicka wrote:
> I tested libstdc++ compilation with -O2 and get 10% savings of the .so
> library size (with debug info in) and about 5% compiler speedup. I also
> tried PR/8362 testcase. With -O2 -fno-unit-at-a-time I get:
That's 8361, right?
> TOTAL : 134.67 2.86 137.79
>
> With -O2 -funit-at-a-time I get:
>
> TOTAL : 92.61 3.19 96.50
>
> And I get 24% fewer instructions produced.
Wow, that's impressive! How's the situation for -O3 (because only for
-O3 do we full inlining, as far as I understood)?
> I also tested it on some of C SPEC2000 sources and it does not seem to
> supress any important inlining. I hope to do some behchmarking tomorrow
> as I need some sleep now.
If you send me (privately) the full set of patches required for this, I'll
happily perform testsuite runs and benchmarks for the full application
which contains the code of PR/8361.
This looks very exciting!
Gerald