> The optimization might be useful for some other processors which
> have direct move insns for the two considered classes and when IRA for
> some reasons did not use the class union. At least I see
> that we could try this for ARM (spilling general regs into VF regs)
> and for extended powerpc architecture (spilling general regs into fp
> regs). What is only necessary is just to define two macros. I am
> going to do it for ARM and see is this optimization beneficial for
> OMAP4. Although I think it is not as fp units with VF regs in ARM
> implementations I know are too separate from integer units.
There is a cost associated with using the VFP register bank and on
older cores like the A8,
there is a penalty associated with moving values from the VFP register
bank to the integer register bank, so it needs to be carefully looked
at on a per core basis.
If you are benchmarking this on an A9 (which is an OMAP4),
I would suggest turning on Neon in your builds to see the full effect of this
rather than just defaulting to the standard vfpv3-d16 configuration just
because this then also brings in the SIMD unit into play.