Provide conversion functions between cl_half and other types about opencl-headers HOT 6 CLOSED

We should probably also provide isnan, isnormal, isfinite and isinf.

from opencl-headers.

+1 to this, and I can do the work to get the initial conversion functions into the headers.

The functions in the CTS make extensive use of the MAKE_HEX_FLOAT macros. They also use fabsf. My current thinking is to rework them a little to avoid depending on these two things (easy enough to perform the comparisons as hex integers). Does this sound OK, or is there a desire to keep the functions exactly as they are in the CTS?

from opencl-headers.

That's great! Sounds good to me. I don't think we need/want to keep the functions exactly as they are in the CTS. I thought that we'd actually change the CTS to use the functions provided by the headers. The overall thinking was:

• starting from the functions in the CTS guarantees that the behaviour of the headers functions match what implementations do today
• using the functions provided in the headers in the CTS guarantees that applications use what implementations are verified with so at least the behaviour is consistent
• the last step would be to add unit tests for the headers functions vs. the spec and deal with the outcome of that.

from opencl-headers.

@jrprice #60 has done what I originally had in mind here. I've created KhronosGroup/OpenCL-CTS#870 so we don't forget to transition the CTS to these routines. Since you added tests as well under #60 I'm not sure there's anything left to do here other than creating new issues to cover making conversions automatic in C++ programs and consider adding all the is* routines. Did you have something else in mind?

from opencl-headers.

Did you have something else in mind?

No, the C++ sugar and is* were the only reasons I didn't close this issue.

from opencl-headers.

Thanks! I've created #102 and #103. Closing this issue.

from opencl-headers.