[Mlir-commits] [mlir] [mlir][vector] Test that transpose folds keep per-dim info (PR #219611)

[Andrzej Warzyński]

================

----------------
banach-space wrote:

> the test should attach the dim_tags at the first op,

I agree. @dcaballe , I think that you meant this instead:
```mlir
 %1 = vector.transpose %0, [2, 0, 1] {dim_tags = array<i64: 111, 222, 333>}
      : vector<4x1x1xf32> to vector<1x4x1xf32>
%0 = vector.shape_cast %arg0 : vector<4xf32> to vector<4x1x1xf32>
```

However, with an example like that, it's clear that the "user" does not care about tracking dimensions (otherwise they would not fold <4x1x1> to <4> via `vector.shape_cast`).

https://github.com/llvm/llvm-project/pull/219611