http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14741
Joost VandeVondele <Joost.VandeVondele at mat dot ethz.ch> changed:
What |Removed |Added
----------------------------------------------------------------------------
Last reconfirmed|2006-04-23 17:57:20 |2013-03-29
CC| |Joost.VandeVondele at mat
| |dot ethz.ch
Blocks| |51119
Summary|missing transformations |graphite with loop blocking
|lead to poorly optimized |and interchanging doesn't
|code |optimize a matrix
| |multiplication loop
Known to fail|tree-ssa |4.9.0
--- Comment #17 from Joost VandeVondele <Joost.VandeVondele at mat dot ethz.ch>
2013-03-29 11:03:40 UTC ---
With 4.9 trunk using
gfortran -ffast-math -O3 -march=native -fgraphite -floop-interchange
-floop-block PR14741.f90
The testcase still runs about 8x slower than what ifort produces at -O3 -xHost.
This is too bad, and in my opinion blocks as simple solution for PR51119
