This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: CFG merge part 7 - superblock/trace scheduling
Jan Hubicka wrote:
>
> > Hi,
> > This patch adds option to enable superblock scheduling (aka EBB scheduler) for
> > non-IA-64 targets as well as "trace scheduling" (aka EBB scheduler preceeded by
> > the trace formation).
> > My experience with this feature are bit mixed - by default it causes 2%
> > regression on Athlon. By significantly improving MD description I was able to
>
> This regression seems to go away after changing the superblock choice
> algorithm I sent in merge patch 11.
Jan, I've just checked trace scheduling on ia64 (intel itanium 900MHz
server). I needed to make some modifications for the compiler to switch
it on for ia64. It still has some test degradation but I managed to
compile all SPECINT benchmarks except eon. Unfortunately, there is no
big improvement for code quality. Here the results in brief are:
o code quality improvement is 0.5% (Spec rate excluding eon is 406.38
vs 404.52).
o slower compiler speed about 2% (user time is 11m27s vs 11m14s).
o bigger code about 20% (overall size of text segments is 9486Kb vs
7662Kb).
As for code quality, there are improvements (the biggest is on gzip a
bit less 3%!) and degradation only on vpr parser gap and bzip2. So I
think the trace scheduling should be not used without profile
information.
Vlad