f2c (+gcc)

The f2c program comes standard with all GNU/Linux distributions with which I am familiar. The home site of f2c is at AT&T's netlib software repository. f2c converts standard Fortran-77 code to standard C code. On GNU/linux, this is further compiled into assembly by gcc. Three immediate advantages of f2c are:

it is free,
it is extremely well-debugged,
and it produces pretty fast code when coupled with gcc.

A further advantage, if you are a nuclear or particle physicist, is that

it has been used to compile all of the CERN libraries, so using it you can get your favorite GEANT code or data analyzer running.

Disadvantages of f2c+gcc are that

it really accepts standard Fortran-77 code, and little else. Most Fortran programs out there in the physics world (in my experience) use many extensions, so you might have to do some editing. Trivial examples I've seen are the "q" format specifier, and the TYPE * statement which are both VAXisms. These two are fairly easy to fix. A nontrivial example is the STRUCTURE data type. If your code makes heavy use of it, you're looking at a serious rewrite if you want to use f2c+gcc. Look at Wolfgang Wander's f77 Problem Page for more information about this.

Debugging f2c+gcc-compiled code --- There are two ways to do this:

Use f77/reorder. This allows your code to be debugged under gdb.
Install the f2c_stabs package (available from standard linux software archives.) This package only runs under the Emacs editor.

Getting correct results --- schuetze@math.tu-dresden.de (T.Schuetze) pointed out that if one turns on optimization, the "paranoia" suite fails. This can be cured by passing the option -ffloat-store to the gcc phase of the compilation. If you use fort77 as a compiler driver, you can simply specify the option as printed above; some other compiler drivers may require you to pass it to gcc using some special option prefix. -ffloat-store forces all real variables to have their stated (either real*4 or real*8) precision. If you don't specify it, certain variables which only live in registers may have the precision of the Intel FP registers, which is different than real*4 or real*8.

Jeff Templon

Last modified: Wed Apr 11 14:54:45 EDT 2001