Created by: kexinzhao

Currently, adding float16 header file in data_type.h will cause the failure of the tensor_util_test. (Weird segfault on GPU test)

Need to fix this and add isnan, isinf check for float16.

This issue has been solved in a WIP PR.