When writing mathematical software to run on a FPU-less machine, using fixed point rather than floating point for non-integer calculations usually results in significant speed improvement. This approach is well supported in SANE, but comes with some loss of precision.

Ken