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

["dominiq at lps dot ens.fr" <gcc-bugzilla at gcc dot gnu dot org>]

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

--- Comment #18 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-05-18 18:29:06 UTC ---
> Are conditional moves that bad on x86?  The change which uses them more for
> COND_EXPR was mine but really I think this was a latent bug or a way to say
> chose conditional move over jumps for some targets.

As said in comment #0 the first regression appeared  between revisions 162456
(2010-07-23) and 164728 (2010-09-29), so the problem is fairly old

[macbook] test/dbg_rnflow% /opt/gcc/gcc4.6p-162456/bin/gfortran -c -O3
cptrf2.f90
[macbook] test/dbg_rnflow% makeo ; time a.out > /dev/null
20.904u 0.345s 0:21.26 99.9%    0+0k 0+0io 0pf+0w
[macbook] test/dbg_rnflow% /opt/gcc/gcc4.6p-162456/bin/gfortran -c -O3
-fno-tree-loop-if-convert cptrf2.f90
[macbook] test/dbg_rnflow% makeo ; time a.out > /dev/null
20.898u 0.341s 0:21.24 99.9%    0+0k 0+0io 0pf+0w
[macbook] test/dbg_rnflow% /opt/gcc/gcc4.6p-164728/bin/gfortran -c -O3
cptrf2.f90
[macbook] test/dbg_rnflow% makeo ; time a.out > /dev/null
28.607u 0.346s 0:28.96 99.9%    0+0k 0+0io 0pf+0w
[macbook] test/dbg_rnflow% /opt/gcc/gcc4.6p-164728/bin/gfortran -c -O3
-fno-tree-loop-if-convert cptrf2.f90
[macbook] test/dbg_rnflow% makeo ; time a.out > /dev/null
21.153u 0.342s 0:21.50 99.9%    0+0k 0+0io 0pf+0w