[Mlir-commits] [mlir] [mlir][bufferization] Handle disjoint subsets in One-Shot Analysis (PR #224184)

[Matthias Springer]

https://github.com/matthias-springer commented:

I think this PR is functionally correct, but it could drastically increase compilation time on larger programs. The `findValueInReverseUseDefChain` call may traverse the full IR and is nested within multiple loops. In addition to that, `operatesOnDisjointSubset` etc. are potentially expensive because they build a constraint set based on yet another IR traversal.

Can you do some compilation time measurements before and after this PR? 

https://github.com/llvm/llvm-project/pull/224184