Floating point overflow in ice_lim_flx

[ufla]

I got occasional floating point overflows in the second line (setting pevap_ice) of this loop in sbcice_lim.F90:

DO jl = 1, jpl
  pqns_ice (:,:,jl) = pqns_ice (:,:,jl) + pdqn_ice  (:,:,jl) * ( ptn_ice(:,:,jl) - ztem_m(:,:) )
  pevap_ice(:,:,jl) = pevap_ice(:,:,jl) + pdevap_ice(:,:,jl) * ( ptn_ice(:,:,jl) - ztem_m(:,:) )
  pqsr_ice (:,:,jl) = pqsr_ice (:,:,jl) * ( 1._wp - palb_ice(:,:,jl) ) / ( 1._wp - zalb_m(:,:) ) 
END DO

Testing further, it turns out that I can make this crash reproducible by initialising all allocatable fields in sbc_ice.F90 (function sbc_ice_alloc) to HUGE. Just setting evap_ice didn't reproduce the crash reliably.

I expect that some of the arrays from sbc_ice_alloc are used before initialisation, at least for k_limflx==2. I'm using revision 5456 of the trunk.

I'll do some further test to find out which array is causing the problem.

[ufla]

This is just to confirm: If I initialise all arrays in sbc_ice_alloc to HUGE, it crashes reliably. If I initialise them to ZERO, it runs reliably fine. Now we just need to find out which of the 21 arrays...

[ufla]

It's pdevap_ice that is not initialised (confirmed with TotalView? debugger).

[clem]

Thanks Uwe,
it should be fixed at revision 5471. I am waiting for your approval to close this ticket, and I'll have a look at your other ticket #1547 on the same topic.

[ufla]

Thanks for your response! The fix seems to do the trick, so you may close the ticket.