Floating divide by zero in cvmix intel compiler DEBUG build

[nichannah]

Link to failure in jenkins:

https://climate-cms.nci.org.au/jenkins/job/mom-ocean.org/job/MOM6_run/build=DEBUG,compiler=intel,experiment=ocean_only-single_column-BML,memory_type=dynamic/

Some error output:

forrtl: error (73): floating divide by zero
Image PC Routine Line Source
MOM6 000000000077D57D cvmix_math_mp_cvm 198 cvmix_math.F90
MOM6 0000000000A5E534 cvmix_kpp_mp_cvmi 1342 cvmix_kpp.F90
MOM6 000000000073FEEC mom_kpp_mp_kpp_ca 607 MOM_KPP.F90
MOM6 00000000031121C6 mom_diabatic_driv 570 MOM_diabatic_driver.F90
MOM6 0000000002D93105 mom_mp_step_mom_ 1010 MOM.F90
MOM6 0000000001BA84F9 MAIN__ 426 MOM_driver.F90
MOM6 000000000040C41C Unknown Unknown Unknown
libc.so.6 00002B43D815FD5D Unknown Unknown Unknown
MOM6 000000000040C319 Unknown Unknown Unknown

[adcroft]

It is true that the solver has an possible divide by zero at L198. I would suggest a ticket on CVMix's page but before that, we should establish what the inputs are from MOM_KPP.F90. It is entirely possible we are calling cvmis_kpp on a land point, or such.

That said, there are bracketing algorithms for root finders to avoid getting hung-up on a local inflexion.

[nichannah]

I ran this in the debugger and the inputs look normal. So yes, it's Newton's method getting stuck at the stationary point. The CVMix issue is here CVMix/CVMix-src#65

[adcroft]

@nicjhan I'm adding a work around at the MOM level to avoid this situation in the CVmix root finder. This is only a work around and CVmix still needs to be made robust to vanished layers. The work around adjusts the coordinates passed to CVmix by inflating vanished layers.