[Mlir-commits] [mlir] [mlir][Linalg] Allow more control in drop unit dims (PR #170104)

[Renato Golin]

rengolin wrote:

> So, the default behavior has always been to preserve encoding and will continue to do so, but it doesn't collapse operands with encoding. The new controls enable to do that as one possible use case.

If the end result still returns the same shape (`<1, 1, 42>`) via collapse and expand shapes, then what's the benefit of this pass?

If you want to return some other shape, then can't we just collapse the resulting shape?

The example doesn't seem to convey the real motivation behind this extra control (and interface complexity).

https://github.com/llvm/llvm-project/pull/170104