I want to generate a C-program that does the following. All the variables are real and the calculations are done as double precision, the reading and writing can be unformatted (READ( * , * ) and WRITE( * , * ) in FORTRAN):

1. Read some coefficients from an input file:

   a1=first read value, ..., a8=last read value.
   

2. Assign a set of known values (data) to some 'design' variables:

   x1=list1, ..., x12=list12
   

Note: The amount of 'columns' (lists of x's) is not related to the amount of coefficients, because this is a non-linear regression problem.

1. Calculate a function value for each 'row'. The function f is non-linear but relatively easy to code (powers, exponentials, etc.)

   yi=f(a1, a2, ...a8, x1i, x2i,...,x12i)
   

2. Print the yi's.

The program is easy to write in FORTRAN (the harder part is to automate the assignment of dataset values into the lists of x's, because the dataset is large). I'd rather not code in C if I can avoid it, so I was hoping Mathematica could do it for me.

The purpose is to couple the compiled C code to some external non linear regression solver and compare the results with the NonlinearModelFit in Mathematica. The reason for wanting something compiled is to avoid the overhead of launching the Mathematica kernel and evaluating not compiled code each time the external solver calls the code.

Any ideas will be appreciated.

OL.