Re: [hatari-devel] FPU update

[Andreas Grabher <andreas.grabher@xxxxxxxxxxxx>]

Am 19.01.2017 um 20:59 schrieb Toni Wilen <twilen@xxxxxxxxxx>:

>>> I got these results:
>>> 
>>> 00000000 20000000 00000000 00000000 40000000 00000000 00000000
>>> 80000000 00000000  <--- If it was the x87 way, this would have
>>> jumped to 00010000 80000000 00000000 00010000 80000000 00000000 
>>> 00020000 80000000 00000000 00030000 80000000 00000000
>>> 
>> Thank you for the very quick response! So yes, that matches the data
>> sheet and means i will have to do some work on SoftFloat to correctly
>> emulate denormals. I also need to think about a solution for native
>> floats. But that will be non-trivial. Time is starting to get a
>> limiting factor now.
> 
> I don't think native floats need any updates (if it is even possible),
> no one has noticed anything until now anyway :)
> 
> btw, I started to "port" your changes back to QEMU softfloat-2a.

I think that the use of denormal extended floats is not very common. That might be the reason that nobody noticed the issue.

I now patched SoftFloat to use 68k format for denormals: https://sourceforge.net/p/previous/code/692/

I used pre-processor conditions to make my changes reversible and visible. That does not look very clean, but i want to keep it until i am confident, that the patch works under all conditions. I did some very basic testing and it seems to do the job. But i would be happy if someone could review or test this. I don't want to introduce new problems.