During tests for AMM with MERGE2013 rev 4521, it appears that ifort compiler options as defined in ARCH/arch-x3750_ADA.fcm :

%FCFLAGS             -DCPP_PARA -i4 -r8 -O3 -xAVX -fp-model precise

are invalid, and produce strange results, including violent crash of NEMO. This has been confirmed on another test using ORCA1 configuration. Must be further explored carefully. Meanwhile, switching to

%FCFLAGS             -DCPP_PARA -i4 -r8 -O2 -fp-model precise

makes NEMO work fine again.

These results have been found on ada.idris.fr (IBM cluster-Intel ES 4650 processors) with ifort version 13.0.1. Other tests using ifort will be usefull:

Out-of-bounds errors occur in subroutines 'sol_sor_tan' and 'sol_sor_adj' (file NEMOGCM/NEMO/OPATAM_SRC/SOL/solsor_tam.f90) when array 'nitsor' is accessed during time steps nitend-1 and nitend in configurations with 'key_dynspg_flt' and 'nn_solv' set to 2 (as in the template configuration 'ORCA2_TAM'). The size of the array 'nitsor' (set on line 63 of NEMOGCM/NEMO/OPATAM_SRC/SOL/sol_oce_tam.f90) can be increased by 2 to prevent these out-of-bounds errors.

The test of the tangent-linear and adjoint versions of the tracer surface boundary condition subroutine (test 'tra_sbc_adj') reports a failure or a warning with a large relative error. In the absence of parallelisation, the reported relative error is somewhat smaller than with parallelisation (32 CPU cores), but it is still large engough to be reported as a warning.

Examination of the code in NEMOGCM/NEMO/OPATAM_SRC/TRA/trasbc_tam.F90 revealed a bug in the first of the two calls to 'lbc_lnk' in subroutine 'tra_sbc_tan': the first argument should be 'tsa_tl(:,:,:,jp_tem)', and not 'tsa(:,:,:,jp_tem)'.