[MPICH] mpich2-1.05 on the intel Mac OS X
Jiming Jin
JimingJin at lbl.gov
Tue Mar 6 17:37:57 CST 2007
Hi Everyone,
I just installed MPICH2-1.0.5p3 on my Pro Mac using the gfortran
compiler (Mac OS X with intel chips, 4 processors total). I did
'make testing', and I get a message saying "No Errors".
I assume the installation was successful. I started mpd on this
computer, did several other tests and everything looked well.
I start to run my model (it is a climate model (WRF)) with 'mpiexec -
n 4 wrf.exe and I only see one processor is running with more than
90% cpu, and other three processors are sleeping(with 0% cpu). When I
put the run to the background (Ctr-Z), I get the following error
messages:
pmpiexec_storm.lbl.gov: mpd_uncaught_except_tb handling:
exceptions.IOError: [Errno 9] Bad file descriptor
/usr/local/mpich/bin/mpiexec 1041 handle_stdin_input
sys.stdin.flush() # probably does nothing
/usr/local/mpich/bin/mpdlib.py 743 handle_active_streams
handler(stream,*args)
/usr/local/mpich/bin/mpiexec 500 mpiexec
rv = streamHandler.handle_active_streams(timeout=1.0)
/usr/local/mpich/bin/mpiexec 1399 ?
mpiexec()
Could someone give me suggestions on this problem? Thanks.
I compiled my model using the following options:
FC = mpif90 -f90=gfortran
LD = mpif90 -f90=gfortran
CC = mpicc -cc=/usr/local/gfortran/bin/gcc -
DMPI2_SUPPORT
SCC = /usr/local/gfortran/bin/gcc
SFC = gfortran
RWORDSIZE = $(NATIVE_RWORDSIZE)
# g95 does not like -r4. if you want promotion to -r8 add it to
PROMOTION below
#PROMOTION = -r$(RWORDSIZE) -i4
PROMOTION = -i4
CFLAGS = -DDM_PARALLEL \
-DMAXDOM_MAKE=$(MAX_DOMAINS) -DMAXPROC_MAKE=$
(MAX_PROC) -I../external/RSL_LITE
FCOPTIM = -O2 # -fast
FCDEBUG = #-g -O0
FCBASEOPTS = -fno-second-underscore -ffree-form -fmax-
stack-var-size=99999768 $(FCDEBUG) # -Mlfs
FCFLAGS = $(FCOPTIM) $(FCBASEOPTS)
ARCHFLAGS = -DGRIB1 -DINTIO -DDM_PARALLEL \
-DIWORDSIZE=4 -DDWORDSIZE=8 -DRWORDSIZE=$
(RWORDSIZE) -DLWORDSIZE=4 -DNETCDF \
-DLIMIT_ARGS -DNMM_NEST=$(WRF_NMM_NEST) -DG95
CPP = /usr/local/gfortran/bin/cpp -C -P -traditional
More information about the mpich-discuss
mailing list