[Mlir-commits] [mlir] [mlir][sparse] Reuse input sparsity for materialized outputs (PR #215579)

[Kohei Yamaguchi]

sott0n wrote:

@aartbik Thanks, I understand the concern and agree.

Although the support-containment analysis is format-independent, the current patch couples it to a CSR-specific materialization path without a concrete plan for supporting other sparse tensor formats. I agree that this is not the right shape for an upstream sparse compiler change.

I will pause this PR and prepare an RFC on MLIR Discourse before continuing the implementation. The RFC will cover the motivating use cases, the format-independent abstraction, where it should live in the compilation pipeline, how it should lower across sparse tensor encodings, and the intended support and testing plan.

We can then revisit the implementation after reaching agreement on the design.

https://github.com/llvm/llvm-project/pull/215579