It can be compiled by adding
#pragma OPENCL EXTENSION cl_khr_fp16 : enable
as a half-precision floating point support option (pragma) in the kernel code statement.
Please refer to the following document for details.
(Reference) The OpenCL Specification Version: 1.0 Document Revision: 48
https://www.khronos.org/registry/OpenCL/specs/opencl-1.0.pdf (9. 10 Half (Item of Floating-Point)
--------------------
Category: OpenCL™
Tools: FPGA SDK for OpenCL™
Devices: -