This covers the following ops:
`spirv.GroupNonUniform` x {`Bitwise`, `Logical`} x {`And`, `Or`, `Xor`}
We need these to efficiently lower from the `gpu.subgroup_reduce` op.
18 KiB
18 KiB