On 04/17/2012 04:29 AM, Richard Sandiford wrote:
Vladimir Makarov<vmaka...@redhat.com> writes:
On the other hand, I don't think that 1st insn scheduling will be ever
used for x86. And although the SPECFP2000 rate is the same on x86-64 I
saw that some SPECFP2000 tests benefit from your algorithm on x86-64
(one amazing difference is 70% improvement on swim on x86-64 although it
might be because of different reasons like alignment or cache
behaviour). So I think the algorithm might work better on processors
with more registers.
Notwithstanding that this is a goemean, I assume there were some bad
results to cancel out the gain?
Yes, mgrid had 4% degradation, mesa had 2%, facerec and ammp had 2.5%,
lucas had 15%. On the other hand, galgel had 2% improvement and equake
had 1%. All other differences are not considerable.
Oops, thanks :-)
Anyway, given those results and your mixed feelings, I think it would
be best to drop the patch. It's a lot of code to carry around, and its
ad-hoc nature would make it hard to change in future. There must be
better ways of achieving the same thing.
It is up to you, Richard. I don't mind if you commit it into the trunk.
There is something in your approach too. If it would be possible to get
the best of the two methods, we could see a big improvement.
