Jeremy Kun

The form I gave above is the "generic" form of the IR, and in that case I think you would need to annotate the `dense` with the type (`dense :...

Oh, I see my original comment also had a syntax error. I think it should be ``` %poly1 = "poly.constant"() : !poly.poly : () -> !poly.poly ```

I was missing a closing `}` in the original comment, for one. But moreover there's the type annotation inside the attribute itself to say what integer type the content of...

@ljhsiun2 could you put what you have in a draft PR and I can clone it and take a look?

You may want to take a look at https://jeremykun.com/2023/09/11/mlir-folders/ for a tutorial on constant-like ops and constant materialization

Polynomial dialect is currently being upstreamed, and I have a PR open there that will resolve this https://github.com/llvm/llvm-project/pull/91655

The first step of this is done: optimizing the inner-most body of a loop-nest.

Last step: add a helper script that runs the loop unrolling with lots of different unroll factors, and prints the relative statistics.

I've been told that affine-scalrep can do this. Is the IR small enough to run affine-scalrep on?

Part of https://github.com/google/heir/issues/475