Would it be easy to modify ocamlopt so that it put "float array"s on
an 8-byte boundary?  If I want to use Fortran libraries like lapack
with ocamlopt on a Sparc I have to copy them before giving them to the
Fortran code, or else get a bus error.  I know the docs don't promise
that you can access doubles direct but it would be very nice.

[please don't read this bit ... :)]

Serait-il facile de modifier ocamlopt afin qu'il poserait les "float
array"s sur une adresse XXX0 ou XXX8?  Si je veux utilitiser les
libraries Fortran (par exemple, lapack) avec ocamlopt sur un CPU
sparc, je dois copier tous mes "float array"s avant de les donner à
lapack.  Sinon ... "bus error".