[OpenCL] Guard all half float usage based on cl_khr_fp16 extension

[jerryct]

Summary

Hi,

this patch guards all missed half float functions based on the availabiltiy of the OpenCL extension cl_khr_fp16.

Best regards,

Jerry

[yaxunl]

Only halfn type requires cl_khr_fp16. These functions do not use halfn type, therefore cl_khr_fp16 is not required.

[Anastasia]

In D46501#1089777, @yaxunl wrote:

Only halfn type requires cl_khr_fp16. These functions do not use halfn type, therefore cl_khr_fp16 is not required.

I think the extension is for all half https://www.khronos.org/registry/OpenCL/sdk/1.2/docs/man/xhtml/cl_khr_fp16.html

This extension adds support for half scalar and vector types as built-in types that can be used for arithmetic operations, conversions, etc. An application that wants to use half and halfn types will need to include the directive shown above.

[Anastasia]

Could you upload the full diff please, otherwise it's not easy to see all the functions guarded by the macro.

[bader]

lib/Headers/opencl-c.h

12097

These built-ins are part of the core specification, which doesn't allow using half data type directly, but user can declare a pointer to half data type and use built-ins like vload_half/vstore_half converting half data type to float data type.

cl_khr_fp16 enables regular uses of half data types as well as built-ins returning half data type instead of float - vload/vstore.