[RegCNET] RegCM3 on SGI Altix with Intel Itanium CPUs
Peng Li
pli888 at gmail.com
Thu Jun 29 15:38:41 CEST 2006
Hello,
We were able to build the June 16 version of RegCM3 on X86 and X86_64
computers with both PGI and Intel compilers.
However, we had some problems on a SGI Altix system. The system is
equiped with Intel Itantium 2 (IA64) CPUs, SUSE Enterprise Linux
Server 9 and Intel compilers v.9. PGI compilers do not support IA64
for the moment. We built netcdf 3.6.1 and 3.6.2 in /opt/i/netcdf and
it passed the "make test" step without any errors.
We were able to compile terrain by running terrain.x without any errors.
However, the resulted DOMIAN.INFO was not correct. When plotting the
data with GrADS, it showed "Contouring: nan to nan interval inf".
Altix is a 64-bit little endian machine. So we kept "-convert
big_endian" in the Makefile and set ibyte=1 in domain.param. We also
tried removing "-convert big_endian" and set ibyte=4 but the results
were still not right. The datasets we used were GLCC10MIN_BATS.CDF and
GTOPO30_10MIN.CDF. I am attaching the makefile below.
SGI Altix is different from Origins which runs IRIX. We checked the
porting guide and did not see anything relevant in terrain.f. Here is
the link to the document.
http://techpubs.sgi.com/library/tpl/cgi-bin/browse.cgi?coll=linux&db=bks&cmd=toc&pth=/SGI_Developer/Porting_Guide
We would appreciate any hint. Does RegCM3 still depend on UDUNITS?
Thanks,
Peng
#Makefile for RegCM@ run on PC_LINUX (Intel 64bit CPU)
# by using Intel FORTRAN compiler version 8 or 9
FC = ifort -g -cm -w -w90 -w95 -convert big_endian
FFLAGS = -L/opt/i/netcdf/lib -lnetcdf -Vaxlib
LD = $(FC)
SRC = terrain.f
OBJ = terrain.o
all: terrain
.f.o:
$(FC) -c $(FFLAGS) $*.f
terrain: $(OBJ)
$(LD) -o $@ $(OBJ) $(FFLAGS)
clean:
rm -f *.o terrain
# --------------------------------------------------------------------
# DO NOT DELETE THIS LINE -- make depend depends on it.
terrain.o: domain.param terrain.f
More information about the RegCNET
mailing list