[Mlir-commits] [mlir] [MLIR] Testing arith-to-emitc conversions using opaque types (PR #137936)

[Gil Rapaport]

aniragil wrote:

> > But the ArithToEmitC pass is essentially only a testing pass, so I think it shouldn't matter.
> 
> I am not sure this holds true. It might be the case for your specific use case but not in general and I sometimes wonder who uses EmitC for what purpose (if I ever hear of it). Thus if this is only for testing, I think it should be hidden behind a flag at least and furthermore needs some words written down on the motivation.

I agree with @marbre that the emitc passes are not testing passes, so hiding this behind a flag makes sense. An alternative might be to generalize this lowering into a option for handling all unsupported types under some `--unsupported-to-opaque` flag, which would consistently convert all types supported by arith but not by emitc, does that make sense?

https://github.com/llvm/llvm-project/pull/137936