[Mlir-commits] [mlir] [mlir][vector] Clarify the semantics of gather/scatter indexing (nfc) (PR #181357)

Krzysztof Drewniak llvmlistbot at llvm.org
Mon Mar 16 09:00:14 PDT 2026


krzysz00 wrote:

1. The inconsistency has been fixed with your updates stating that the gather indices don't follow layout but continue in a dense, row-major way. I don't know if those _should_ be the semantics, given that `tensor.extract_strided_slice` / `memref.subview` exist - but that is what we're trying to document here.
2. I'd like to flag #184706 and @hanhanW because I think we'll be getting inconsistent semantics again from that change
3. I'm going to once again suggest that we should make gather take k <= rank -tuples of indices to fix this mess.

https://github.com/llvm/llvm-project/pull/181357


More information about the Mlir-commits mailing list