seems to cause a runtime failure of a double free or corruption:
*** Error in `/discover/nobackup/mathomp4/Experiments/scottbranch-2023Feb06-1day-c24-GNUGCMFixes-GNU-DEBUG/scratch/GEOSgcm.x': double free or corruption (!prev): 0x0000000026f69870 ***
and even this took some hunting. In the call above, vort and divg are optional, but so are UA,VA`, etc.
This PR has a workaround that is ugly, but seems to work. If building with GNU, it does not pass in vort and divg but rather uses the old external GetVorticity and GetDivergence codes that @wmputman moved into getAllWinds to be more efficient (less communications which is important at high-res).
I'll be looking at this with @tclune and @aoloso but as far as I can tell, the current Intel path is legal Fortran, so I don't know why GNU has an issue. Maybe there is bug elsewhere?
This PR has changes to allow the GNU compiler to run the
feature/sdrabenh/merge_v10.25.0_MSTRF
code from @sdrabenh and @wmputmanThe issue is that, for some reason, getting
vort
anddivg
from:seems to cause a runtime failure of a double free or corruption:
and even this took some hunting. In the call above,
vort
anddivg
are optional, but so areUA,
VA`, etc.This PR has a workaround that is ugly, but seems to work. If building with GNU, it does not pass in
vort
anddivg
but rather uses the old externalGetVorticity
andGetDivergence
codes that @wmputman moved intogetAllWinds
to be more efficient (less communications which is important at high-res).I'll be looking at this with @tclune and @aoloso but as far as I can tell, the current Intel path is legal Fortran, so I don't know why GNU has an issue. Maybe there is bug elsewhere?