Re: questionable use of qemu-m68k for building (was Re: ruby2.2 fuckup)

[Laurent Vivier <laurent@vivier.eu>]

Le 05/12/2015 20:59, Thorsten Glaser a écrit :
> Dixi quod…
>> Laurent Vivier dixit:
>>> Le 21/11/2015 13:07, Andreas Schwab a icrit :
>>
>>>> Does that include the FPU emulation?
>>>
>>> I don't know: FPU emulation is on 96 bits (in fact 80...) and can have
>>> some bugs (I didn't fix the NaN format for the moment).
>>
>> In this case, e.g. Python interpreters and related things (cffi/numpy?)
>> built on qemu-m68k may be broken.
> 
> It looks like Python running *on* qemu-m68k is, at the very least,
> broken (probably precisely because it does funny things to the FPU):

I think qemu-m68k is emulating precisely the FPU. I've checked the
result of testfloat on qemu-m68k, and moreover I've compared the result
of softloat on my Quadra 800 (68040 FPU) against the result of qemu-m68k.

As I said, NaN is not defined correctly but it should be hidden for all
applications that will not check the bit encoding.

http://www.jhauser.us/arithmetic/
https://github.com/vivier/m68k-testfloat
https://github.com/vivier/m68k-softfloat

Laurent