This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: GCC inline parameters (PR 10160 testcase)
Steven Bosscher <s.bosscher@student.tudelft.nl> writes:
> Op vr 02-05-2003, om 20:30 schreef Andreas Jaeger:
>> What kind of performance difference does this give us? I fear that
>> this change will speed up compilation but slow down execution and
>> would therefore like to see some performance data of compiled code
>> also.
>
> It is a real shame that this has to come up so late in the preparation
> for 3.3, because these issues have been known for almost a year now.
>
> Well, we'll never know until somebody benchmarks different settings.
>
> I've asked a couple of weeks ago if anybody had done some timings with
> different inline parameter settings, and there was _no_ reply at all.
> Ought so have seen at least a few for something so often debated in this
> mailing list as tree inlining issues...
>
> Apparently nobody has the numbers, and nobody can tell what TRT is until
> we _do_.
>
> What would be good testcases, i.e. some representative piece of code
> that is easy to measure runtime for? Maybe your SPEC tester could be
> tweaked to test a bunch of different inline parameter settings?
Good idea. If you give me a bunch of compiler flags, I'll do the
benchmarking and send you the results. Ok?
I'm just starting the following two SPECint runs:
-O2 -march=athlon
-O2 -march=athlon --param max-inline-insns-single=150 --param max-inline-insns-auto=150 --param max-inline-insns=300 --param min-inline-insns=100
But if you send more invocations, I can run them.
Andreas
--
Andreas Jaeger
SuSE Labs aj@suse.de
private aj@arthur.inka.de
http://www.suse.de/~aj