[Bug tree-optimization/42216] [4.5 Regression] rev 154688 regress 464.h264ref peak 20%

[bernds_cb1 at t-online dot de]

------- Comment #10 from bernds_cb1 at t-online dot de  2009-12-01 17:27 -------
Yes, regrename should only affect the second scheduling pass.

I'm attaching a stab-in-the-dark patch.  It contains a fix for the ia64
regressions (testing in progress), it adds a few warnings which could be useful
to spot whether the testcase is producing the corner cases we refuse to handle,
and it makes some changes to the register selection.  The idea is to prefer
registers which aren't used in the basic block yet, which should give the best
freedom to the scheduler.

I'd be glad if you could play with this and let me know the results.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42216