hoknamahn

twod
We try to remain as device-agnostic as possible by using OpenCL. Currently the denoiser is locked to Nvidia GPUs (as it's their denoiser), but we have plans to support other denoisers as well. As RobW mentioned, you should be able to run karma on both a GPU-less farm machine via CPU CL and a local workstation with GPU CL, without any user intervention. And if you're lucky enough to have farm machines with GPUs, you can use the GPU on both.
When the Karma GPU variant is released, of course

As far as I'm aware AMD has dropped the support of CPU CL. Does it mean that Karma would not work on AMD CPUs? Or maybe the third party OCL implementations like POCL would be a viable solution?

edward

hoknamahn
I totally agree. But this is the way wranglers work.

Ok, so I tried this and looked at the VEX code inside an Attribute Wrangler. So if the wranglers work this way, then I guess functions work within functions right now.

They do. But only if a value we return is not of array type. It's the case in both normal VEX and a wranglified one. So it's not a problem of wrangler's preprocessor as I initially thought. Looks like vcc needs more love

edward
I don't know. Personally, I hate functions within functions from my days of modifying badly written Pascal code.

I totally agree. But this is the way wranglers work. Plus their behaviour is not consistent nor intuitive (otherwise we would not see such questions).
On a contrary, we have a workaround