[Mlir-commits] [mlir] [mlir][vector] Preserve dynamic mask bounds in shape cast folding (PR #229699)

[purnima shrivastava]

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

----------------
purnima-nlp wrote:

You're right. I was keeping this case to isolate the `all_of` → `any_of` part, since with four operands and two dropped dims the old slice-length calculation happens to produce the correct length.

But looking at it again, the simpler `no_fold_shape_cast_with_mask_leading_unit_dynamic_unknown` case already catches both issues independently: it fails if the slice length is wrong, and it also fails if the correct slice is used with the old `all_of` condition.

So this additional case is redundant. I'll remove it. Thanks for pushing on this!

https://github.com/llvm/llvm-project/pull/229699