Very impressive!

But I can't help but think if something like this can be done to this extend, I wonder what went wrong/why it's a struggle for OpenCL to unify the two fragmentized communities. While this is very practical and has a significant impact for people who develop GPGPU/AI applications, for the heterogeneous computing community as a whole, relying on/promoting a proprietary interface/API/language to become THE interface to work with different GPUs sounds like bad news.

Can someone educate me on why OpenCL seems to be out of scene in the comments/any of the recent discussions related to this topic?