[Bug tree-optimization/53346] [4.6/4.7/4.8 Regression] Bad vectorization in the proc cptrf2 of rnflow.f90

ubizjak at gmail dot com gcc-bugzilla@gcc.gnu.org
Fri May 18 17:32:00 GMT 2012


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53346

--- Comment #11 from Uros Bizjak <ubizjak at gmail dot com> 2012-05-18 16:04:46 UTC ---
(In reply to comment #9)

> Uneducated guess: is it possible that failed attempts to vectorize may mess up
> further optimizations?

You are right. -ftree-vectorize implies -ftree-loop-if-convert and this option
makes all the difference!

-O2 -ftree-vectorize:

real    0m24.061s
user    0m23.789s
sys     0m0.225s

-O2 -ftree-vectorize -fno-tree-loop-if-convert

real    0m18.029s
user    0m17.761s
sys     0m0.220s

We were barking up to the wrong tree. ;)



More information about the Gcc-bugs mailing list