[Mlir-commits] [mlir] [mlir][Tensor] Generalize the pattern to swap `tensor.collapse_shape` -> `tensor.expand_shape`. (PR #133819)

[llvmlistbot at llvm.org]

================
@@ -48,14 +48,67 @@ func.func @no_bubble_partial_intersecting_reshapes(%arg0: tensor<?x?x?x?xf32>, %
 
 // -----
 
-func.func @no_bubble_0d_tensor_reshapes(%arg0: tensor<?xf32>, %s0: index, %s1: index, %s2: index, %s3: index) -> tensor<?x?x?x?xf32> {
-  %collapse = tensor.collapse_shape %arg0 [] : tensor<?xf32> into tensor<f32>
+func.func @no_bubble_0d_tensor_reshapes(%arg0: tensor<1x1xf32>) -> tensor<1x1x1xf32> {
----------------
MaheshRavishankar wrote:

Yeah, actually the original reshape is not really well defined. The only valid value for `?` in the original example was 1. So its hard to assert in the verifier for the op, but at runtime, any value other than 1 would be invalid. So its kind of a corner case representation thing. So just fixed the test to avoid this case.

https://github.com/llvm/llvm-project/pull/133819