[PATCH] Resolve issues with fmuladd intrinsic handling across multiple backends

[Tim Northover]

> (For the AArch64 backend in particular, I made the assumption, for
> now, that FMAs should not be formed on fp128, but this can be changed
> with a one-line fix.)

I think that's reasonable, at least for now (I don't actually know how libgcc implements the fp128 arithmetic); the AArch64 changes look sensible to me.

Tim.