efriedma-quic wrote: This patch seems sort of contradictory to me. You're encouraging users to pass a non-constant slice_offset, presumably because we expect the backend to optimize it... but at the same time, you're special-casing constant slice_offsets because you don't expect the backend to optimize it. What do you actually expect the backend to do in practice?
I don't see any backend tests involving a non-zero slice_offset? We currently have some code in getMemVTFromNode() in AArch64ISelDAGToDAG.cpp that handles sme_ldr/sme_str. Is that code currently reachable? If it is reachable, does it need to be adjusted for this change? https://github.com/llvm/llvm-project/pull/68565 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
