Created by: kexinzhao

fix #9222 (closed)

Added device function for multiplying two float16 numbers on GPU device, which is needed in the following code:

Y.device(place) = X * static_cast<T>(1.0f - dropout_prob);