This is the mail archive of the
mailing list for the GCC project.
Re: [RFA] Kill artificial inlining limit
- From: Michael Matz <matz at suse dot de>
- To: Steven Bosscher <s dot bosscher at student dot tudelft dot nl>
- Cc: Richard Guenther <rguenth at tat dot physik dot uni-tuebingen dot de>,<gcc-patches at gcc dot gnu dot org>
- Date: Tue, 13 May 2003 17:15:52 +0200 (CEST)
- Subject: Re: [RFA] Kill artificial inlining limit
On 13 May 2003, Steven Bosscher wrote:
> > > Seriously though, the SPEC tests that Andreas and I have done clearly
> > > show that the extreme inline limits your application needs are unusual.
> > I agree to most of what you wrote except that one. Richards and Geralds
> > code is heavily templated C++. SPEC has exactly one C++ test (eon), and
> > that one doesn't use exactly advanced methods. So deducing from SPEC to
> > unusualness of requirements is clearly wrong.
> Wouldn't that indicate a separate problem (or the same, the example I
Yes of course. I just reacted on your wrong conclusion, which followed
from non-bad effects in SPEC, that Richard has unusual inlining needs. In
fact I rather think, that SPEC has unusual ones.
The underlying reason is most probably what you speculated about, namely
the (sometimes) totally wrong heuristic of calculating the number of
instructions based on the tree statements.