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

[Toni Wilen <twilen@xxxxxxxxxx>]

>  > FScc, FDBcc and FTRAPcc does not modify FPIAR
>
> Hm, are you sure about this? The manual states:
>
> "For the subset of the FPU instructions that generate exception traps, 
> the 32-bit FPIAR is loaded with the logical address of the instruction 
> before the processor executes it."
>
> And why are they zero now? Shouldn't that still be the address of the 
> previous FPU instruction, like it was before your recent changes? Guess 
> that has to verified on real hardware.

Yes, I tested it with real 68040, those 3 instructions won't modify 
FPIAR. FPIAR probably is zero because test loads all registers with zero 
(Using FMOVEM.L) before running the test.

I think this is yet another case of documentation writer not being clear 
enough ("looking through FPU colored glasses"). I am sure it should say 
"generate arithmetic exception traps", not any possible exception.

>  >Not sure why SNAN test does not work. Does it even cause an exception?
>
> I addes a printf in fpsr_check_arithmetic_exception (the value is the 
> FPSR). This is what i get with 060:

It looks like some early softload nan/invalid value detection functions 
won't check for signalling nan. I'll debug this later today.