Re: ia64 XFmode patch rides again (need ia64-linux testing)

[Jan Hubicka <hubicka at ucw dot cz>]

> Jan Hubicka <hubicka@ucw.cz> writes:
> 
> >> I am confused.  I thought __float128 and "long double" were both going
> >> to be 80-bit IEEE extended in an 128-bit field on x86-64.  What you
> >> are saying now makes it sound like __float128 is a different type
> >> altogether (IEEE quad?).  If this is so, you'll need to mention TFmode
> >
> > __float128 is supposed to be IEEE quad.  long double is supposed to be
> > 80-bit IEEE in 128-bit.
> 
> Ah, okay, then you will need both XFmode and TFmode in i386-modes.def.
> You can probably copy much of ia64-modes.def; just get rid of the HPUX
> specific gunk.

Well, that is precisely what I did.
> 
> >> Do you have the most recent version of genmodes.c?
> >
> > So the sollution really is easy :)  I didn't updated since yesterday.
> 
> Odd, that code went in on Friday with the ia64 backend changes.  Check
> for sticky tags.

Actually I was traveling for few days so it is quite posible that my
tree was older.  I however had the ia-64 changes in.  Weird..
I guess my tree got just partly updated for some reason, so no surprise
that things got messed up in my tests.  Sorry for the confusion.
> 
> > So the right sollution would be to provide both *xf *tf with the same
> > semnatics + TFmode libcalls under new names...
> 
> Yes, I think so.  And, as Richard points out, the *xf symbols should
> be versioned GCC_3.4; come to think of it, the same is true on ia64,
> but I don't know how to do that.

Well, me neither.  For now I will copy what you do have in IA-64 so we
are in sync.  There is not much need to make these different and lets
work this out afterwards.
> 
> >> You'll have to rename all the TFmode libcalls to some other
> >> convention.  For IA64, it seems sensible to make the HP-UX 
> >> convention be used everywhere.  Does AMD maybe have a document
> >> defining software float library routines?
> >
> > No, but we can write one.  Any particular prefferences in
> > convention?  Or taking the HP-UX names you do use would be OK?
> 
> I have no preference at present.  You can look in
> ia64.c:ia64_hpux_init_libfuncs (IIRC) for the HP-UX names.  Note that

I copied these already...

> GCC's floating point emulation library does not support IEEE quad, so
> the question is moot until that gets resolved.  (Or does x86-64
> support this in hardware?)

No, it doesn't.  I tought that there is at least one implementation in
glibc used by SPARC sofware emulation available but I never really
checked it out.  Not sure whether it is C or assembly or whether it
exists at all..

Honza
> 
> zw