Ömercan Yazici

Yes I agree. We should also include the "fma" instruction to the math builtins (maybe with a fallback for non LLVM, e.g, OpenCL, etc). I think there are more general...

Yes. Unfortunately, there is no access to streams or other finer-grade barriers in the API. Having a common set between all the device types we support is quite difficult. Especially...

The hsa function requires signals (which might be useful for events [other PR]). What would be the best practice to provide them for each call without exposing it to the...

OpenCL support might be feasible with `clGetEventProfilingInfo` and other functions

For HSA (AMD) signals can be used. But we have bookkeep some lists to have the same approach as the CUDA events, most likely.

See issue #46 as well

The problem increased when introducing participating media as they should be encapsulated as well.

Improved usage by grouping similar lights together (currently only simple area lights)

Smart specialization and anonymization should make this a non-factor in the future

Should be exposed as another "texture". Can then be combined with other stuff.