> It may just be projection on my part, but I feel like there's enough interest 
> for an async API to be validated fairly quickly

I agree with this and I'd be happy to participate in the validation in
C, C++, and/or Rust :)

> it's trivially easy to use a CPU-memory ArrowArray as an ArrowDeviceArray

Being trivially easy depends on the ArrowDeviceArray existing, which
it doesn't in some language bindings (e.g., arrow-rs). We could
declare the portion of the ArrowDeviceArray that is actually used
(CUDA, CUDA_HOST, CPU) to be stable and spend some time implementing
it everywhere, but we'd have to do that knowing the current design
probably won't work for Apple GPUs. Having a CPU version would
decouple those conversations which may be positive for both (since the
intersection of those already limited pools is small or may never
exist).

Cheers,

-dewey

Reply via email to