This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Experimenting with tree inliner parameter settings
- From: Steven Bosscher <s dot bosscher at student dot tudelft dot nl>
- To: Richard Guenther <rguenth at tat dot physik dot uni-tuebingen dot de>
- Cc: gcc at gcc dot gnu dot org
- Date: 04 May 2003 22:55:22 +0200
- Subject: Re: Experimenting with tree inliner parameter settings
- References: <Pine.LNX.4.44.0305042240420.478-100000@goofy>
Op zo 04-05-2003, om 22:44 schreef Richard Guenther:
> [Summary / analysis of inline params snipped]
>
> The SPEC results show, that at least defaults for C and Fortran do not
> necessarily apply to C++ code (as the testcases from the PRs show).
Hmm, where do you see that indicated in the SPEC results? I have to
admit I did not look at that. I'm not sure how C++ and C would be
different -- what works for C++ should work for C as well, no?
For Fortran all this doesn't make a difference at all, simply because
g77 does not use the tree inliner.
> So may I suggest again splitting the defaults per language? Especially for
> templated C++ code with exceptions the instruction counting can be _very_
> wrong.
The instruction counting can indeed be very wrong, for C and C++. The
way to fix that may not be not to adjust the tree inliner parameters,
but instead to change how the instruction accounting is done in the
front end.
Greetz
Steven