Using li2018 auto scheduler: Error: Atomic updates are not supported inside Metal kernels

[vakokako]

I'm using li2018 autoscheduler for gpu target. I get the following exceptions on targets x86-64-osx-avx-f16c-metal-sse41 and x86-64-windows-avx-avx2-d3d12compute-f16c-fma-sse41:

Error: Atomic updates are not supported inside Metal kernels
Error: Atomics operations are not supported inside D3D12Compute kernel.

OpenCl and Cuda are working correctly without exception.

[alexreinking]

The Li2018 autoscheduler tries to generate schedules that use atomic, but that scheduling directive isn't implemented for d3d12 or metal. I'm not sure if anyone is working on that, but the error message does at least tell you that this scenario is not (yet) supported.