README.CRAY-X1

# 2 May 2005
# pnetcdf on the Cray X1

I performed the following steps to get Parallel-NetCDF to build on the Cray X1
at Oak Ridge (phoenix.ccs.ornl.gov).   Note that out-of-tree (or VPATH) builds
do not work for the Fortran interface as of 1.0.0-pre2, but we will try to
address this issue in a future release.  

prompt:$ module load mpt
prompt:$ export CC=cc
prompt:$ export FC=ftn
prompt:$ export MPIF77=$FC
prompt:$ export MPICC=$CC
prompt:$ export FFLAGS="-eh"
prompt:$ ./configure --prefix=/path/to/install
# note: configure takes a fairly long time. 
prompt:$ make

The "nc_test" test will exhaust the available MPI datatypes on the X1.  Your
application might see this error:

   MPI has run out of internal datatype entries.                  
   Please set the environment variable MPI_TYPE_MAX for additional space.
   The current value of MPI_TYPE_MAX is 2098                      

I did as asked and nc_test completed with MPI_TYPE_MAX set to 4096

If you run on the login node, expect to see a lot of these messages:

  Process [nc_test] 89345 generated trap, but has signal 8 held or ignored
        epc 0x1219bb4 ra 0x1219b94 badvaddr 0x40004f0004000020 

The messages don't *appear* to impact the program results, and additionally do
not show up if you submit the job to PBS.

Fortran codes should use '-eh' so that the Cray ftn compiler will use 1 byte
for int*1 and 2 bytes for int*2.  Otherwise, our Fortran bindings will pass
incorrect values to the C routines.