[LV] Now for another bit of noise , ok any noise ...

[Maciej W. Rozycki]

On Tue, 3 Mar 2009, Jan-Benedict Glaw wrote:

> >  There is also the "long double" H-float type currently unsupported even 
> > by GCC.  It should be straightforward to add too -- no rocket science 
> > here.  It's optional for hardware, but at least the NVAX implements it in 
> > silicon which makes it attractive for me. ;)
> 
> Guesswork is to add an encoder/decoder function to gcc/real.c and to
> enable it. It's probably not all that clever to simply add it as
> TFmode to vax-modes.def, but to add a command line option instead,
> placing it into the override function (which already manages D vs. G
> float format)?

 I don't think there is a need for a flag here.  Whoever does computation 
on long doubles has to know what they are doing -- GCC and glibc already 
handle the diversity.

> Anyone to write an emulator for hardware missing H_float support? Then
> we'd unconditionally enable it for Linux :)

 There might be code available for some of the other platforms.  It would 
be more interesting to investigate whether to do this in the kernel (as 
the x86 and MIPS do for example) or in the C library (as suggested many 
times in discussions on LKML) in a signal handler.  I have a slight 
preference for the latter or at least I would like to try to see if that's 
feasible if we ever come to it.

 That's a task for a distant future though -- let's make the necessary 
bits first. ;)

  Maciej
_______________________________________________
Linux-Vax mailing list
Linux-Vax at pergamentum.com
http://www.pergamentum.com/mailman/listinfo/linux-vax