Dennis Castleman (DennisCastleman@oaktech.com) writes:
> I'm trying to run soft-floating point functions on a r5000 core with
> a FPU. Without having to take the overhead of using a trap. Using
> the files fp-bit.c and dp-bit.c from the gcc source as a floating
> point lib. This implementation lack in accuracy in the least
> signeficant bit multiplication in division operations.
There's an IEEE-compatible set of software floating point routines in
the Linux kernel, invoked by the trap handler. The routines were
donated by Algorithmics (now part of MIPS Technologies).
If you wrapped them in the appropriate GCC skins, they should give you
a soft-float library which works better.