a78d1d9a8b
Currently, `ConvertVectorToLLVM` rejects strided memrefs when lowering `vector.gather` and `vector.scatter`. This PR adds tests to document that behavior. Supporting strided memrefs in the lowering is left as future work. However, it is still unclear whether gather/scatter on strided memrefs should be supported at all (see the Discourse discussion [1]). This PR also adds tests for `vector.load` and `vector.store` in `invalid.mlir` to document that these ops do not support strided memrefs. [1] https://discourse.llvm.org/t/rfc-semantics-of-vector-gather-indices-with-strided-memrefs