[Flow] Fold flow reshape with mismatching dyn dims

[IanWood1]

flow.tensor.reshape is foldable when there is a single mismatching dynamic dimension (unequal SSA values) because they must be the same.

Cleans up pattern found in migx_bert-large-uncased

[MaheshRavishankar]

Am I missing something here. The test added here does not seem to be an example of what is described in the PR description?

[IanWood1]

Are you looking at foldReshapeChain? The diff is a bit confusing. flattenReshapeChain was the old test to combine multiple reshapes, but because it only has 1 dynamic dim it got completely folded away. foldReshapeChain is the new test (which uses the old mlir input), but the difference is that it checks that all reshape ops get folded. But I could be wrong

[MaheshRavishankar]

Are you looking at foldReshapeChain? The diff is a bit confusing. flattenReshapeChain was the old test to combine multiple reshapes, but because it only has 1 dynamic dim it got completely folded away. foldReshapeChain is the new test (which uses the old mlir input), but the difference is that it checks that all reshape ops get folded. But I could be wrong

Ah interesting. I did miss that, I see now. I dont know how this change made flatterReshapeChain work.