This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Inlining and estimate_num_insns
On Sun, 27 Feb 2005 02:04:19 +0100, Richard Guenther
<rguenth@tat.physik.uni-tuebingen.de> wrote:
> Steven Bosscher wrote:
> > On Saturday 26 February 2005 23:03, Jan Hubicka wrote:
> >
> >>Mark? I would say that there is little risk in this patch corectness
> >>wise, might have negative effect on compilation times since we re-start
> >>inlining more like we did in old days.
> >
> >
> > Can we see some timings with and without this patch?
>
> tramp3d-v3 -O3 compile- and run-times without checking are
>
> compile size run
> 3.4 1m7s 4421288 51s
> 4.0 1m2s 4531197 1m57s
> 4.0 patched 2m39s 5391554 27s
> 3.4 leafify 1m36s 4539503 34s
> 4.0 leafify 1m52s 4844784 15s
Further testing shows that the tramp3d testcase benefits most by the
removed abstraction penalty. With --param max-inline-insns-auto=80
--param max-inline-insns-single=80 we get
4.0 0m55s 4663068 45s
4.0 patched 1m32s 5069229 26s
uh oh? I don't understand the unpatched number - that looks like we
do not "sort" our inlining decisions to first. 500 as default for
max-inline-insns-single
looks very large anyway.
--param max-inline-insns-auto=50 --param max-inline-insns-single=50
4.0 0m49s 4749444 1m2s
4.0 patched 0m58s 4952254 40s