This patch adds intrinsic support for conversion routines from FP16 to int/long and double and also adds c2 mid-end vectorization support for these operations and backend aarch64 where ever possible.