RE: [clisp-list] FFI with complex numbers?

[Goffioul M. <gof...@im...>]

> Michael Goffioul asks:
> >Is it possible to use complex numbers with FFI? What I want to=20
> >do is to pass=20
> >complex numbers from LISP to C code (whatever the format=20
> >received by the C function)
>=20
> Since there's no standard format for complex numbers in C,=20

"complex" type modifier is beginning to appear (I don't know if
it's a recent standard from *C99* or a gcc extensions), but I
don't to which memory model it corresponds. Anyway, I guess that
mapping a LISP complex object to double* is reasonable.

> the CLISP FFI doesn't provide standardized means to pass such objetcs.
> My guess is that a good enough approach is to pass real and=20
> imaginary part separately, each using a known format (e.g.=20
> double-float) and COERCE to this chosen format on the Lisp side.
>=20
> Is there anything else that would make you happy?

Well, yes: a MatLisp version that work with CLISP... :-)

>=20
> You could also define a C-STRUCT {double real, imag}, write=20
> two functions to convert Common Lisps COMPLEX numbers to that=20
> structure and vice-versa and pass along FFI pointers to that=20
> a structure.
> You should then investigate DEF-C-TYPE and DEF-C-STRUCT and=20
> the constructors automatically generated by the FFI. Probably=20
> this approach looks like a better design than the above.
>=20
> Don't the library you want to interface to provide some=20
> representation of complex numbers already?

The target math library is BLAS/LAPACK, which are FORTRAN libraries.
In FORTRAN, complex numbers are represented as real and imag part
next to each other: for example, a single complex number can be mapped
to a real array of 2 elements (this would corresponds the structure
you mentionned above).
Writing conversion routines is a possibility, but I fear that efficiency
would be low, especially if you want to convert large complex arrays
(if you need to do it in a loop). Is there a way to make C code access
LISP variables directly (without copying)?

Michael.