Re: [hatari-devel] Possible bug in 060 exception stack frames

[ Thread Index | Date Index | More Archives ]

 > Enabled inexact...passed

I still get errors (see attached file). Also i still get an Address error when i allow it run the fmovem.l #xx,fpcr/fpsr test. Are there any differences in exception handling for Hatari/WINUAE? I noticed some ifdefs there, but i can't tell whether thats the reason.

It looks like FPU mode is not softfloat mode (or FPU config is not in "unimplemented" mode). Host FPU mode does not support FPU arithmetic exceptions and special case flags accurately. (NaN, operr etc..)

(I still haven't found any real world programs that use FPU arithmetic exceptions..)

I used your ftest/ files, I can't run FMOVEM test if it is disabled :)

btw, any idea if 68060 test would work with 68040 FPSP?

With the FPSP for 040? No, that is organized differently. Instead of the callouts, it is a library that is expected to be linked with certain user-supplied functions. Also i haven't found any binary, precompiled versions so far, so it would be difficult to rule out any assembler problems. And most of the tests are 060 specific anyway, the integer tests for example won't trigger an exception on 040.

I meant even if it reports errors (which it surely does if CPU is wrong model) but it still can be used to verify if emulation and real 6888x/68040 report exact same errors.

I'll do 68882 test later today.

What would be interesting to know though, whether the FPSP060 package can also be used for 040. So far i think that should be possible, but maybe i missed some tricks, especially when the fsave frame is manipulated.

I don't think it is easily possible. FSAVE frame is totally different between 040 and 060. 68040 frame is crazy, 68040 does some preprocessing internally before creating the stack frame and then it expects FPSP to do some more processing before returning with FRESTORE.

Mail converted by MHonArc 2.6.19+