This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [graphite] Remove checks for flag_loop_strip_mine, flag_loop_block, flag_loop_interchange in graphite_trans_bb_block.


On Sat, Aug 16, 2008 at 12:11:34AM -0300, Tobias Grosser wrote:
> 
> I saw a SEGFAULT, but as I forgot to remove the line in my patch, we
> wrote into some unknown memory location. So about every error message
> was possible.
> 
> At the moment the graphite branch works for me (Testcases and polyhedron
> with "-O2 -floop-strip-mine -floop-interchange -floop-block
> -fgraphite").
> 
> It seems to be hard to move a patch from git to svn. Sorry for the
> disruption. The next time I will compile the patch again after applying
> to my svn checkout. 
> 
> Tobias

Tobias,
   The compilation failure in aermod.f90 seems to be triggered by...

gfortran -ffast-math -O2 -floop-interchange -fgraphite aermod.f90 -o aermod
gfortran -ffast-math -O2 -floop-strip-mine -fgraphite aermod.f90 -o aermod
gfortran -ffast-math -O2 -floop-block -fgraphite aermod.f90 -o aermod

which all produce the error...

aermod.f90: In function âvrtcblâ:
aermod.f90:14521: internal compiler error: in instantiate_scev_1, at tree-scalar-evolution.c:2220
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

So we seem to have some bad interactions between -ffast-math and the new loop optimizations.

gfortran -O2 -floop-block -floop-interchange -floop-strip-mine -fgraphite aermod.f90 aermod

compiles aermod fine on the other hand.
                   Jack


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]