So I wanted to use a special maths function but couldn't find a C library that had it, however I found an old Fortran library slatec that had implemented it, so in order to use it, I adapted these instructions to my case :http://difdop.polytechnique.fr/wiki/index.php/How_to_Bessel_Functions_in_C
I'm using a makefile to compile, here's an extract of its contents in the GNU case
CFLAGS=-Wall -O2
CC=gcc
MPICC=mpicc -DDEBUG_INPUT
LIBS=-lm -lgfortran
THIS_DIR=$(shell pwd)
WEAKWEAK_OBJECTS=manager_weakweak.o worker_weakweak.o bunch_weak.o transform_weak.o cyclic_array.o
WEAKSTRONG_OBJECTS=manager_weakstrong.o worker_weakstrong.o bunch_strong.o fbi_weakstrong.o rw_weakstrong.o
../mbtrack-mpi : ./besselh/libamos.a mbtrack-mpi.o confmpi.o input.o test.o ${WEAKWEAK_OBJECTS} ${WEAKSTRONG_OBJECTS} bunch.o stats.o physic.o statistics.o tracking.o rw_table.o rw_new.o lagrange.o feedback.o nrutil.o
${MPICC} $(CFLAGS) $^ -o $@ $(LIBS) -L$(THIS_DIR)/besselh $(THIS_DIR)/besselh/libamos.a
Everything works marvelously with the GNU compilers, but I am currently trying to compile on a cluster that predominantly uses Intel Compilers, and -lgfortran
is not found.
I read up in Intel's documentation that using the -nofor-main
flag should allow the linking to occur, but the compiler just complains about a linking error:
cc -DDEBUG_INPUT -Wall -O2 -I/opt/local/include besselh/libamos.a mbtrack-mpi.o
confmpi.o input.o test.o manager_weakweak.o worker_weakweak.o bunch_weak.o
transform_weak.o cyclic_array.o manager_weakstrong.o worker_weakstrong.o
bunch_strong.o fbi_weakstrong.o rw_weakstrong.o bunch.o stats.o physic.o statistics.o
tracking.o rw_table.o rw_new.o lagrange.o feedback.o nrutil.o -o ../mbtrack-mpi -lm -
lifcore -lifcoremt -L/global/u1/j/jackb/work/new_rw_2/src/besselh
/global/u1/j/jackb/work/new_rw_2/src/besselh/libamos.a
/opt/intel/composer_xe_2013_sp1.2.144/compiler/lib/intel64/libifcore.a(for_main.o): dans la fonction « main »:
for_main.c:(.text+0x54): référence indéfinie vers « MAIN__ »
/usr/bin/ld: erreurs de lien trouvés, destruction de l'exécutable « ../mbtrack-mpi »
make: *** [../mbtrack-mpi] Erreur 1
I was wondering if there was anything as simple as the -lgfortran library flag that I used with the GNU compilers?
In an attempt to adapt the compilers on the cluster:
Here is the rule to compile the library:
./besselh/libamos.a : FORCE
cd besselh/; $(MAKE) libamos.a
And the contents of the Makefile in the besselh directory:
libamos.a: $(files_o)
# ftn -nofor-main -fdefault-real-8 -fdefault-double-8 -c $(files_f)
ar cru libamos.a $(files_o)
ranlib libamos.a
$(files_o): $(files_f)
ftn -nofor-main -c $(files_f)
clean:
rm -f $(files_o)
rm -f libamos.a
If you just want Bessel functions in C, aren't those in the C99 standard enough? http://www.gnu.org/software/libc/manual/html_node/Special-Functions.html
The correct option to disable the Fortran specific main
in Intel Fortran is -nofor-main
.
libgfortran
is the Fortran runtime library of gfortran. If you compiled the Fortran subroutines with gfortran and are just calling them from C source compiled with icc
it should work.
If you are using Intel Fortran, then it of course doesn't have it. It has its own runtime libraries libifcore
and libifcoremt
. You should link them via -lifcore -lifcoremt
or by linking using ifort
instead of icc
. But it is quite likely you will not need them.
Finally, it shouldn't be a problem to find a C library. Try the GNU Scientific Library. I think the Intel's MKL contains them too.