This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

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

> Jan Hubicka <> writes:
> >> 
> >> Having now, I hope, removed all the obstructions, this patch once
> >> again attempts to support simultaneous use of 80-bit and 128-bit
> >> floating point in the ia64 back end.  It's much the same as it was the
> >> last time -- replace TFmode with XFmode in all hardware insns
> >> referring to the extended floating point type; use TFmode exclusively
> >> for software emulated IEEE quad.  However, instead of using
> >> ROUND_TYPE_ALIGN as before, I am using ADJUST_BYTESIZE and
> >> ADJUST_ALIGNMENT declarations in ia64-modes.def, which *should* mean
> >> that everything obeys the requirements.
> >
> > This looks cool.  If I understand it right, to get the same done on
> > x86-64, where we do have 128-bit long doubles in XFmode format and
> > 128-bit __float128, all I need is to define
> > ADJUST_BYTESIZE/ADJUST_ALIGNMENT and kill the TFmode duplicates in
> >, right?
> I think so.  You will also need to define ADJUST_FLOAT_FORMAT; the
> selected floating point format has to match the byte size.

It would be nice thing to do.  At first it would kill many duplicated
patterns in that will likely speed up the compiler and at second
we do have __float128 specified by ABI (even tought it is optional).  I
am not sure about the timming.  Do you think something like this would
be safe and applicable for mainline at this stage? (I can do that at
monday or tuesday then)  If not,  I guess I should wait for
post-tree-ssa merge to avoid headaches with maintaining the patch given
that there is no BIB....

> zw

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]