Double allocation error of FIELD variable [w3wavemd] #14
Conversation
Also removed redundant OMPX switches.
|
A selection of regtests show no unexpected differences (ww3_tp2.18 difference in ww3_prtide_current.out is an existing issue). |
|
The simplest way to solve this conflict is to comment off the following line in w3wavemd.ftn:
!/OMPX FLOMP = .TRUE.
If OMPX option is no longer required, tiding up the code by removing all OMPX related lines will also correct the bug.
One question is that why the OMPX option is still tested if it is no longer used?
|
|
Thanks @ukmo-jianguo-li. In response:
I've done more that that - I've removed it and all other lines relating to OMPX and FLOMP. I've also removed the second ALLOCATE statement in the OMP block.
Yup - that's what I've done. :)
It's true that the w3wavemd module was the only module using the OMPX switch. I have not removed the switch from the list of valid switches as it is likely that future development might make use of it. Are you happy with the changes I have made? |
|
The changes are ok as long as no one complains about removing the OMPX option. As most of the OMPX lines have been commented off in the old code, it seems fine to go ahead with your changes. |
* Fb rtd2 (#7) FB_rtd2: Output boundary conditions to rotated grids * This feature allows the user to specify the pole of each nested grid output of boundary conditions (nest) file. - There is no effect unless the switch RTD is set for compilation. - The contents of b.c. output files nestI.ww3 are not affected. - A model compiled with combined switches SMC and RTD is not affected. * Updated manual Sec. "3.4.9 Rotated grids" * Added regtests for testing input/output BCs in rotated pole context. * Updated revision of ww3_bound.ftn to 7.11 * Added new ww3_tr1 regtest (rotated pole) to matrix.base. Also, split the SMC/RTD regtests into their own separate switches. Co-authored-by: ukmo-chris.bunney <[email protected]> * Double allocation error of FIELD variable [w3wavemd] (#14) * Fixed allocation of FIELD variable. * Removed defunct OMPX switches and FLOMP variable * Change coupling condition so that it works with all compilers (#12) Co-authored-by: Carsten Hansen <[email protected]> Co-authored-by: Juan Manuel Castillo Sanchez <[email protected]>
Relates to issue NOAA-EMC#252
When running on an
SMCgrid usingSHRDandOMPG/OMPXswitches,ww3_shelcrashes in thew3wavemdmodule due to an array being allocated twice.The issue arises due to some incorrect logic relating to the allocation of the
FIELDvariable. There is a defunctOMP PARALLELsection starting at line 1731 that was disabled some time ago in PR NOAA-EMC#160 due to it causing bit-for-bit differences. However, the logic for allocating the FILED array as a private variable inside the parallel section (rather than a shared variable outside) was left in - and this logic is incorrect. When the model is run with a combination ofSMCandOMPXswitches, the array is erroneously allocated twice.The simplest solution is to keep only the initial allocation at line 580 - 583 and remove the allocation in the parallel section. Furthermore, as the
OMPXstatements are now defunct, we can remove them and the associatedFLOMPlogical. This also requires only a singleDEALLOCATEstatement.NOTE: If it is considered that the OMP section starting at line 1731 might be re-instated in the future, I can keep the second allocation and fix the incorrect logic.