[Mlir-commits] [mlir] [mlir][vector][NFC] Make function name more meaningful in lit tests. (PR #94538)

[Andrzej Warzyński]

================
@@ -102,6 +102,25 @@ func.func @transfer_read_dims_mismatch_non_zero_indices(
 
 // -----
 
+func.func @transfer_read_dims_mismatch_contiguous_non_zero_idx(
----------------
banach-space wrote:

> %subview : memref<1x3x3x2xf32, strided<[40, 10, 2, 1], offset: ?>>

Note the strides - the memref is not contiguous when considering all dims. However, your point is correct:

> I think it is writing vector<2x2xf32> to a contiguous memory?

When only considering the **trailing 2 dims**, memref is indeed contiguous. IIRC, the point of this test wast to make the pattern work even if:
* overall, memref is non-contiguous, however
* the inner dims _are_ contiguous and that's sufficient.

Does it make sense? If yes, then we probably need a better name or a comment to explain 😅 

https://github.com/llvm/llvm-project/pull/94538