Open dotnetian opened 3 months ago
That sounds like quite the task! But would be awesome nonetheless.
NPUs are generally specialized for stuff like FMAs (like the Apple Neural Engine), so I don't think it's possible to use them to run Bend/HVM.
I would be happy if non-Nvidea GPUs (Apple silicone, namely, for me) were supported, nevermind the NPUs.
I don't think I see any branches working on that... any forks working on this that anyone knows about?
Is your feature request related to a problem? Please describe. NPUs, similar to GPUs can run a ton of tasks in parallel. And now with the release of a new gen of PCs using ARM-based CPUs also NPUs such as Qualcomm Hexagon are getting more popular and accessible for users. I'd suggest the Bend community consider supporting NPUs as another runtime environment for the language.