PATCH: i386 target support for decimal float moves

[Menezes, Evandro]

> > The outcome of the discussion for the x86-64 ABI is to pass 
> decimal32 
> > like float, and decimal64 like double and decimal128 like 
> __float128 - 
> > therefore passing all of them in the XMM registers.
> 
> Are you sure?  This will require unnecessary overhead moving 
> them to and from integer registers in all the libcalls.

The consensus was that this overhead is small and that if and ever it's implemented in HW, the likely place would be the FPU, when the overhead or passing them in GPR would be much larger.


-- 
_______________________________________________________
Evandro Menezes            AMD               Austin, TX