--- Comment #7 from rguenth at gcc dot gnu dot org 2009-12-01 16:46 ---
Just reverting rev. 154688 and using the training set gets us from
464.h264ref--228 -- S
to
464.h264ref--170 -- S
at -O3 -ffast-ma
--- Comment #8 from rguenth at gcc dot gnu dot org 2009-12-01 17:03 ---
The hot loop is mv-search.c:SetupFastFullPelSearch
for (pos = 0; pos < max_pos; pos++)
{
abs_y = offset_y + spiral_search_y[pos];
abs_x = offset_x + spiral_search_x[pos];
if (range_partly_outside)
--- Comment #9 from bernds_cb1 at t-online dot de 2009-12-01 17:25 ---
Created an attachment (id=19202)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19202&action=view)
Test patch to attempt to narrow down the problem
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42216
--- 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
t