No floating Point coprocessor
cubranic at cs.ubc.ca
Mon Jun 22 16:16:02 PDT 1998
Paul Caamano <pcaamano at cse.ucsc.edu> has said:
> Our group is working on a port of Kaffe to a National Semiconductor
> NS486SXF board. So far we have the interpreter working and are proceeding
> with the JIT version. However, we're running into some problems since the
> JIT seems to depend on the presence of a FPU (our board doesn't have one).
> Does anyone know of any simple way to bypass these instructions?
How about lifting math coprocessor emulation from Linux or any other open
source OS and hooking it into whatever kernel you're running on your box?
Admittedly, this is not an ideal solution because it adds the overhead of
catching the exception for every floating-point operation, but it might be the
quickest way to get you up and running (if you can alter the OS on your board,
that is). Alternatively, you might be able to replace floating-point operation
codes that JIT would generate with calls to the routines in the said emulation
More information about the kaffe