[Mlir-commits] [mlir] [MLIR][Vector] Generalize DropUnitDimFromElementwiseOps to non leading / trailing dimensions. (PR #92934)

[Han-Chung Wang]

hanhanW wrote:

The commit breaks downstream project. Here is the repro: `mlir-opt -test-vector-transfer-flatten-patterns ~/repro.mlir`

```mlir
func.func @unit_dim_folding(%arg0: vector<1x1xf32>) -> vector<1x1xf32> {
  %cst = arith.constant dense<0.000000e+00> : vector<1x1xf32>
  %0 = arith.mulf %arg0, %cst : vector<1x1xf32>
  return %0 : vector<1x1xf32>
}
```

Error:

```
repro.mlir:3:8: error: 'arith.mulf' op operand #0 must be floating-point-like, but got 'vector<f32>'
  %0 = arith.mulf %arg0, %cst : vector<1x1xf32>
```

The commit seems problematic for corner cases, can we revert it and re-land it with a fix?

https://llvm.org/docs/DeveloperPolicy.html#patch-reversion-policy

https://github.com/llvm/llvm-project/pull/92934