Escali 4.4 3.10 Error and warning messages

Section: 3.10 Error and warning messages

Scali MPI Connect Release 4.4 Users Guide 33

Because of different buffering behavior, some programs may run with MPICH, but not with

SMC. Unsafe MPI programs may require resources that are not always guaranteed by SMC, and

deadlock might occur (since SMC uses spin locks, these may appear to be live locks). If you

want to know more about how to write portable MPI programs, see for example MPI: The

complete reference: vol. 1, the MPI core [2].

A typical example that will not work with SMC (for long messages):

while (...) {

MPI_Send(buf, cnt, dtype, partner, tag, comm);

MPI_Recv(buf, cnt, dtype, MPI_ANY_SOURCE, MPI_ANY_TAG, comm, sts);

doStuff();

}

Thsi code tries to se the same buffer for both sending and receiving. Such logic can be found,

e.g., where processes from a ring where they communicate with their neigbours. Unfortunately

writing the code this way leads to deadlock, and to make it work the MPI_Send() must be

replaced with MPI_Isend() and MPI_Wait(), or the whole construction should be replaced with

MPI_Sendrecv() or MPI_Sendrecv_replace().

The SMC library is written in C and all of its C names are prefixed with scampi_. Depending

on the compiler used, the user may run into problems if he/she has C code using the same

scampi_ prefix. In addition, there are a few global variables that may cause problems. All of

these functions and variables are listed in the include files mpi.h and mpif.h. Normally, these

files are installed in /opt/scali/include.

Given that SMC has not fixed its OS routines to specific libraries, it is good programming

practice to avoid using OS functions or standard C-lib functions as application function names.

Naming routines or global variables as send, recv, open, close, yield, internal_error,

failure, service or other OS reserved names may result in an unpredictable and undesirable

behavior.

3.10 Error and warning messages

User interface errors usually result from problems where the setup of the environment causes

difficulties for mpimon when starting an MPI program. mpimon will not start before the

environment is properly defined. These problems are usually easy to fix, by giving mpimon

the correct location of the necessary executable. The error message provides a straight forward

indication of what to do. Thus, only particularly troublesome user interface errors will be listed

here.

Using the -verbose option enables mpimon to print more detailed warnings.

When a fatal error occurs, SMC prints an error message before calling MPI_Abort() to shut

down all MPI-processes.