> Well, at least with the current I/O setup it seems to be impossible > to assert anything useful for the optimizers just depending on > the function that is called. You could have two aliases with the same asm name and give them different attributes, or just make the alias in libgfortran. Paolo