[petsc-users] Intel Internal compiler error: segmentation violation, signal raised

[Åsmund Ervik]

> From: TAY wee-beng <zonexo at gmail.com> To: Barry Smith
> <bsmith at mcs.anl.gov>, Jed Brown <jed at jedbrown.org> Cc: PETSc list
> <petsc-users at mcs.anl.gov> Subject: Re: [petsc-users] Intel Internal
> compiler error: segmentation violation signal raised
> 
> Hi Barry, ...
> 
> Also, can anyone recommend options to get optimized results in
> gfortran?

Hi all,

Just my 2 cents: we're using

-O3 -march=native -fno-protect-parens -fstack-arrays

This is supposed to mimick the behaviour of ifort with "-O3 -xHost".
Be aware that using "-ffast-math" can be a little dangerous in some
situations, so unless this option gives your code a large speedup I
would personally remove it.

For gfortran version 4.8.1 you also have to add
-fno-predictive-commoning
which is due to a bug that is fixed in 4.8.2


I would also add that actively using more than one compiler (we
currently use pgf90, ifort, gfortran) together with nightly regression
testing (using Jenkins) has helped us in discovering bugs much faster
than if we were just using one compiler. We've seen code that will
segfault when compiled with ifort and run just fine with gfortran, and
vice versa. Same goes for debug vs. optim builds, we do a full set of
both for our nightly testing.


Regards,
Åsmund


> 
> I'm using :
> 
> -fno-signed-zeros  -fno-trapping-math -ffast-math -march=native 
> -funroll-loops -ffree-line-length-none -O3
> 
> Thank you.
> 
> Yours sincerely,
> 
> TAY wee-beng
> 
>> 
>> 
> 
> 
> ------------------------------
> 
> _______________________________________________ petsc-users mailing
> list petsc-users at mcs.anl.gov 
> https://lists.mcs.anl.gov/mailman/listinfo/petsc-users
> 
> 
> End of petsc-users Digest, Vol 64, Issue 21 
> *******************************************
>