fullfp16

Enable full half-precision floating point.

CPUs: