search

PeterHjortLauritzen / opt-se-cslam

0 stars 0 forks source link

possible vertical remap bug #25

Closed PeterHjortLauritzen closed 6 years ago

PeterHjortLauritzen commented 6 years ago

When I was running CSLAM with DEBUG set, the bounds checker found bounds violation for the variable pio(kk) in line 536 of file vertremap_mod.F90. I 'fixed' this issue by replacing lines 527-529 with the following:

   do while ( pio(kk) <= pin(k+1) )
      kk = kk + 1
      if(kk==nlev+2) exit
    enddo
johnmauff commented 6 years ago

Totally forgot about this issue. Sorry that I did not commit it previously.

PeterHjortLauritzen commented 6 years ago

code has been checked in. It is unclear to me why "if(kk==nlev+2) exit" fixes the problem since pio(nlev+2) is guaranteed to be > pin(k+1) so we will always be done with the while-loop with kk<=nlev+2. I have checked that kk is in bounds after the while-loop and it always is. So why "if(kk==nlev+2) exit" is needed is a mystery to me ....

Crash does not happen on Hobart with NAG. Could it be a weird Intel compiler bug.