[Mlir-commits] [mlir] Remove Pure attribute from Linalg::IndexOp. (PR #68894)

[Mehdi Amini]

joker-eph wrote:

> what I was thinking was to say that there is an op that is valid only if it defined within a region. So it can be CSEed/moved around within the region, but cannot be "hoisted" out of the region for example. I agree this is a big change.

What I was describing was a change to the side effect mechanism instead to be able to model an effect that is not "escaping" from the region. Not something mechanical blocking CSE or anything (which would be a workaround for a transformation, not actually fixing the model).

> I think one solution is to introduce a single trailing block argument to Linalg operations. 

Seems to match my second bullet "There is a SSA value carrying this state...", so LGTM.


https://github.com/llvm/llvm-project/pull/68894