Are the Reserved Types in OpenCL reserved in the sense that the OpenCL people might want to enable them in future implementations, or are they actually available for use in the 1.0 Specification?
Specifically, are complex types available in OpenCL?
Cuda C++ support pretty smoothly gave me complex types for those kernels, but especially now that AMD has OpenCL for their x86s, I’m wanting to try out our kernels in OpenCL.
Also, are there limits to the number of arguments in an OpenCL kernel?
It’s so early that I think it’d be implementation dependent whether complex support is provided, I’d recommend trying it (let us know).
The maximum number of kernel arguments is not defined in the specification, so unless it’s written in some nvidia document, we don’t really know.
I dropped complex variable declarations in both Nvidia and AMD SDK samples, and both complained at JIT compile. I guess ‘reserved type’ means ‘not now, maybe later’.
Hopefully later won’t be too far out :). Hybrid with a single code path would be great, and programming data parallel on CPUs would be nice too.