As a followup to the issue that was opened in: KhronosGroup/OpenCL-Docs#1266

It was clarified regarding floating point negation that it should behave like every other floating point operation (i.e. add, sub, mul, div, fabs) in the sense that:

• Denorms are preserved on devices that support denorms
• Denorms may be flushed to zero on devices that don't support them.

A further conclusion of the discussion was that CTS coverage for the operation is encouraged to emphasize this behavior.