[Mlir-commits] [mlir] [mlir][TOSA] Fix shape inference when operand was inferred (PR #66906)

[Eric Kunze]

eric-k256 wrote:

I think @maxbartel is right based on what we have today. It does point out a problem that there is no way to reshape down to a rank0 tensor with the current `new_shape` argument. The spec says rank1 to MAX_RANK today, but I think it needs to support rank 0 also (and plan to change the spec to allow it). Maybe a new_shape of array<i64: 0> as a special case for rank 0, but there are other options.

https://github.com/llvm/llvm-project/pull/66906