[Mlir-commits] [mlir] [mlir] Let GPU ID bounds work on any FunctionOpInterfaces (PR #95166)

[Mehdi Amini]

joker-eph wrote:

> I remember there being a note on Discourse recently saying that non-kernel GPU functions could be func.funcs

Yes, but as far as I remember, the only reason for this is because we wanted to allow function that works on CPU and GPU: that is if a function does not make use of GPU specific features then it can be a func.func.

> Oh, and, 3, this'll make it work for spirv.func if that's a thing

But wouldn't a spirv.func be the result of this conversion? That is you perform dialect conversion from the GPU dialect to SPIRV and you get a spirv.func with the spirv operation matching this GPU operation?

https://github.com/llvm/llvm-project/pull/95166