[MITgcm-support] strange mpi results (using mpi?)
m. r. schaferkotter
schaferk at bellsouth.net
Thu Aug 12 17:19:33 EDT 2004
greetings;
after running a successful lock exchange problem in serial mode, i
decided to run as mpi
on lintel with pgi5.2, mpich-1.2.6 by modifying linux_ia32_pgf77+mpi,
which has no MPI defines in it, only
some paths to mpich includes and mpich libs.
after a successful build, the program runs for the prescribed number of
timesteps (10- 300 in testing).
i/ve compiled with 2,4, and 8 processors and a similar phenomenon
exists in all.
the program runs with mpirun, but i don/t see where the program been
told to use mpi. (below usingMPI is F).
(PID.TID 0000.0001) > nTx=1, ! nTx - No. threads per process in X
(PID.TID 0000.0001) > nTy=1 ! nTy - No. threads per process in Y
(PID.TID 0000.0001) nPx = 2 ; /* No. processes in X */
(PID.TID 0000.0001) nPy = 1 ; /* No. processes in Y */
(PID.TID 0000.0001) nSx = 1 ; /* No. tiles in X per process */
(PID.TID 0000.0001) nSy = 1 ; /* No. tiles in Y per process */
(PID.TID 0000.0001) sNx = 800 ; /* Tile size in X */
(PID.TID 0000.0001) sNy = 1 ; /* Tile size in Y */
(PID.TID 0000.0001) Nr = 51 ; /* No. levels in the vertical */
(PID.TID 0000.0001) nX = 1600 ; /* Total domain size in X ( =
nPx*nSx*sNx ) */
(PID.TID 0000.0001) nY = 1 ; /* Total domain size in Y ( =
nPy*nSy*sNy ) */
(PID.TID 0000.0001) nTiles = 1 ; /* Total no. tiles per process (
= nSx*nSy ) */
(PID.TID 0000.0001) nProcs = 2 ; /* Total no. processes ( =
nPx*nPy ) */
(PID.TID 0000.0001) nThreads = 1 ; /* Total no. threads per process
( = nTx*nTy ) */
(PID.TID 0000.0001) usingMPI = F ; /* Flag used to control whether
MPI is in use */
(PID.TID 0000.0001) /* note: To execute a program
with MPI calls */
(PID.TID 0000.0001) /* it must be launched
appropriately e.g */
(PID.TID 0000.0001) /* "mpirun -np 64 ......"
*/
when i view the output of the temperature (initial) i find that the
temperature looks ok,
but there is a large white blank square at the bottom of the domain, in
tile 2 of a 4 processor run
where does one set usingMPI? what about the eedata namelist
&EEPARMS
nTx=1, ! nTx - No. threads per process in X
nTy=1 ! nTy - No. threads per process in Y
&end
where are the instructions on getting mpi runs going?
m. r.
More information about the MITgcm-support
mailing list