On Fri, Jun 10, 2005 at 10:50:01AM -0400, Jan-Willem Maessen wrote:
On Jun 10, 2005, at 6:33 AM, Keean Schupke wrote:
anyone know how to marshall data for Fortran?
Find something with a C interface and use that. Most LAPACK and BLAS libraries should have a C binding. My understanding is that Fortran dope vector formats are not standard, though the latest versions of the Fortran standard aim to fix that non-portability problem.
I've never heard of a problem with the Fortran matrix format. We call the fortran LAPACK/BLAS routines straight from C, using an autoconf macro to figure out how the fortran symbols are mangled. It's not so hard. Basically, a fortran matrix is just a C matrix (2D array) transposed. It's either M_ij = M[i+N*j] or M[j+N*i]. I can never remember which. Fortran also always passes by reference, so when calling from C you need to make every argument a pointer, but other than that you can pretty much read the LAPACK docs and code in C straight from that. Oddly enough, I think calling the fortran routines from C is more portable than trying to find a C binding. P.S. for examples of calling lapack from C/C++ (although it also does a lot more) you can run see the dft++ source code. darcs get http://dft.physics.cornell.edu/dft You can grep for F77_FUNC to find fortran-calling code. The fortran-calling and lapack-finding autoconf macros are cribbed from mpb, which is written by the Steven Johnson, who is one of the authors of fftw. So I suppose you could also look at the mpb source... -- David Roundy http://www.darcs.net