You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In pio_internal.h we have a workaround for mpi libraries that do not (IMHO) correctly interpret the MPI standard with respect to the MPI_DATATYPE_NULL variable and generate errors when this datatype is passed to the library even when the associated data count is 0. If this is not the intent of MPI_DATATYPE_NULL then I have no idea what is. The code in pio_internal.h works around the problem for openmpi where we lost an argument here. And the HPE/SGI mpt implementation (pointless to argue, no issue was filed).
This recently came up again as an ESMF issue. when MPICH 4.0.0 is used. An issue in the mpich repo indicates that the mpich developers fixed this issue in 4.0.1. Do we add another exception in pio_internal.h or just use our workaround all the time. @edwardhartnett?
The text was updated successfully, but these errors were encountered:
In pio_internal.h we have a workaround for mpi libraries that do not (IMHO) correctly interpret the MPI standard with respect to the MPI_DATATYPE_NULL variable and generate errors when this datatype is passed to the library even when the associated data count is 0. If this is not the intent of MPI_DATATYPE_NULL then I have no idea what is. The code in pio_internal.h works around the problem for openmpi where we lost an argument here. And the HPE/SGI mpt implementation (pointless to argue, no issue was filed).
This recently came up again as an ESMF issue. when MPICH 4.0.0 is used.
An issue in the mpich repo indicates that the mpich developers fixed this issue in 4.0.1. Do we add another exception in pio_internal.h or just use our workaround all the time. @edwardhartnett?
The text was updated successfully, but these errors were encountered: