In what release to expect?
For SDK 2.2:
Execution capabilities:
Execute OpenCL kernels: Yes
Execute native function: No
Some times ago Macah wrote that availability of native functions will be adressed in upcoming release.
It was before SDK 2.2
To avoid excessive branch/comparison instruction usage I need, for example, fast max instruction. My dataset consists of "ordinary" numbers, no NaNs and other edge cases handling needed.
There is fmax() native function in OpenCL guide that should be faster than max(), right?
When it will be implemented?