[Mlir-commits] [mlir] [mlir][linalg] Introduce new `linalg.conv` op (PR #117688)

[Renato Golin]

rengolin wrote:

> but there is a reason for why discussion in that thread died down

It didn't die down, we're all taking hold of the discussed points and understanding how they affect our own work.

If you think a thread has died down, the easiest thing to do is just ping the thread asking if it did.

> The goal of this PR is to give the participants a working principle for this new Op and an opportunity to specify exactly how it would need to change for it to work for them.

That's a worthy goal, but without direction, it could make it harder for all parties to agree. All I'm asking is that you work with the affected people to put a proposal out. It has a much higher chance of avoiding endless discussions and does not detract the discussion from upstream.

> Like you said, the goal is to get buy-in of all parties on the right direction, not proposing the end-all-be-all solution for convolution in linalg.

That's not what I said. No one wants to go through this again. We want buy-in for the final solution. We already have "buy-in" for the wrong solution, and that's what we've been trying to correct.

Less haste, more speed.

https://github.com/llvm/llvm-project/pull/117688