undefined reference to 'd1mach_'

Question

I\'m trying to link a fortran subroutine with c++, but can\'t quite figure out what exactly is wrong here: The fortran subroutine calls some functions eg. d1mach or xermsg, whic

Answer 1

The reason why the problem persists might be simply because your lib3j6j9j.a does not include necessary files (such as d1mach). Actually, we can compile necessary files rather directly, so I will summarize the procedure below:

1) Download drc3jm.f (which calculates 3j-symbols) and dependencies from the Netlib/Slatec page (here or here). Unpack the archive file to get Fortran files (*.f).

tar xvf netlibfiles.tgz

2) Remove d1mach.f, i1mach.f, and r1mach.f (if any). Instead, download their alternative versions from Netlib/blas (*):

rm -f i1mach.f r1mach.f d1mach.f
wget http://www.netlib.org/blas/i1mach.f 
wget http://www.netlib.org/blas/r1mach.f
wget http://www.netlib.org/blas/d1mach.f 

3) Compile all *.f files

gfortran testf.f90 *.f

together with a main program testf.f90 (in free-format), e.g.,

program main
implicit none
integer, parameter :: N = 1000
double precision coef( N ), M2min, M2max, M2
integer ier
ier = 0 ; coef(:) = 0.0d0

call DRC3JM( 15.0d0, 30.0d0, 40.d0, 2.0d0,  M2min, M2max, coef, N, ier )
print *, "M2min, M2max, ier = ", M2min, M2max, ier

M2 = 2.0d0
print "(a, f20.15)", "coef = ", coef( nint(M2 - M2min+1) )  !! -0.019081579799192
end

Then running the executable gives the desired result.

---

3-a) We can also make these *.f as a library and link with C++ codes, e.g., as follows:

gfortran -c *.f
ar rv mylib.a *.o
g++ testc.cpp mylib.a -lgfortran

with a main program (testc.cpp)

#include <cstdio>
extern "C"
double drc3jm_ (double*, double*, double*, 
                double*, double*, double*, double*, int*, int*);

int main()
{
    double* coef;
    double L1, L2, L3, M1, M2min, M2max, M2;
    int ier, k, N = 1000;

    coef = new double [ N ];
    L1 = 15.0; L2 = 30.0; L3 = 40.0; M1 = 2.0;

    drc3jm_ ( &L1, &L2,    &L3,
              &M1, &M2min, &M2max, coef, &N, &ier );  
    printf( "M2min, M2max, ierr = %10.5f%10.5f%d\n", M2min, M2max, ier );

    M2 = 2.0;   
    k = (int)(M2 - M2min + 1.0e-3);
    printf( "coef = %20.15f\n", coef[ k ] );  // -0.019081579799192
    return 0;
}

We can see that the two programs give the same coefficient (-0.019081579799192) for

j1=15, j2=30, j3=40, m1=2, m2=2, m3=-4

You can also get the same result with an online tool, e.g., here.

---

But depending on cases, it may be simpler to use other libraries. One approach is to use the corresponding GSL routines (here) as

#include <cstdio>
extern "C"
double gsl_sf_coupling_3j (int two_ja, int two_jb, int two_jc,
                           int two_ma, int two_mb, int two_mc);  
int main()
{
    double coef;
    coef = gsl_sf_coupling_3j( 30, 60, 80, 4, 4, -8 );  // -0.019081579799205
    // NOTE: all j's and m's need to be doubled.

    printf( "coef = %20.15f\n", coef );
    return 0;
}

Here you need to link necessary GSL libraries (e.g., g++ test.cpp -lgsl or g++ test.cpp /usr/lib64/libgsl.so.0 /usr/lib64/libgslcblas.so.0 etc).

---

Yet another approach is to use a latest program WIGXJPF (the related paper is here). I tried this a bit and it seems extremely easy to install (only one make) and use. For example, enter the example/ directory and try gcc -I../inc csimple.c ../lib/libwigxjpf.a. According to the above paper, this program may offer some accuracy and performance advantage.

---

(*) For more details, please see the Netlib/FAQ page (thanks to @VladimirF in the comment). We could utilize the original d1mach.f etc in Slatec, but we need to modify them so as to obtain correct machine-dependent constants. The above BLAS versions of d1mach.f etc handle this automatically, so they are more convenient.