Re: Long division in Mintlibs

[Chris Ridd <chris@imc.exec.nhs.uk>]

Eric wrote:
>Have you checked the code to make sure that the Lattice version of the
>division code is taking the parameters as Lattice passes them, and not
>as gcc does?

Lattice always calls the division routines with the args in d0 and d1, and
returns a result in d0, whatever argument passing model is used (stdargs or
regargs).

I did find one minor difference in the code (an add.w instead of an addq.w)
but the bug was rather more simple than that and hence took longer to
find :-(

The names of the unsigned and signed division routines in the Lattice port
are transposed, so that the unsigned routine was being called instead of
the signed one. The same goes for the multiplication routines as well.

Do I post patches to this list, or direct to nox and knarf?

Chris