Problems in testing


Up: Problems Next: General Previous: DEC ULTRIX

The MPICH test suite, in examples/test, performs a fairly complete test of an MPI implementation. If there is an error, it usually indicates a problem with the implementation of MPI; if you encounter such a problem, please report it to mpi-bugs@mcs.anl.gov. However, there are a few exceptions that are described here.


Up: Problems Next: General Previous: DEC ULTRIX

General


Up: Problems in testing Next: Acknowledgments Previous: Problems in testing

    1. Q: The test pt2pt/structf fails with 
    0 - Error in MPI_ADDRESS : Invalid argument: Address of location  
given to MPI_ADDRESS does not fit in Fortran integer 
[0] Aborting program!
    A: This is not an error; it is a gap in the MPI definition. This indicates that Fortran integers are not large enough to hold an address. This does indicate that MPI programs written in Fortran should not use the MPI_Address function on this system.


    2. Q: The test env/timers fails with 

    Timer around sleep(1) did not give 1 second; gave 0.399949
    A: The low-level software that MPICH uses probably makes use of the SIGALRM signal, thus denying it to the user's program. This is not an error (the standard permits systems to make use of any signals) though it is unfortunate.

    One system known to use SIGALAM is the IBM MPL/POE (device ch_mpl) software for using the High Performance Switch in the IBM SPx parallel computers.


Up: Problems in testing Next: Acknowledgments Previous: Problems in testing