This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [tree-profiling] Reimplement profile driven heuristics


On Tue, 7 Dec 2004, Jan Hubicka wrote:
> I tested the patch on SPEC compiled with IMA and -O2. It improve peroformance
> by roughly 1% (on K8, PPC testing is in progress) both with and without
> profiling relative to the original heuristics not taking profiling into
> account.  With profiling the code size savings are roughly 40% (I hope
> to get close to -O2 numbers if heuristic is tuned curefully enought).

Not bad, contratulations!

> For gerald application the results are as follows:
> [...]
> So there is one regression but otherwise the results are fine.

I love them. :-)

> In some sense this is the end of original plan I had concerning profile
> driven optimizations in GCC so I feel bit sentimental and I would like
> thank everyone who helped with this task!

Once you have done those final adjustments/cleanup you mentioned, perhaps
an announcement on the GCC mainpage might be a good idea?

Gerald


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]