[Mlir-commits] [mlir] [mlir] [linalg] fix failure on specializing matmul with permuted loops (PR #184294)

[Renato Golin]

rengolin wrote:

> From my understanding a core property of linalg is that we do not embed any scheduling information into the operation itself. Which does mean that i should be always free to reorder the iterator_types in any way i want because the order carries no meaning for how i would materialize the actual loop nest.

Indeed. Even the contract operation renormalizes them by order of dimensions in the affine maps. I guess this should be fine, here, too.

> I can add more tests that cover different loop ordering + transpositions.

Thanks! Do we have existing tests for broadcasts? The number of dims won't change (and thus iterator_types), but the affine map of the inputs will be different. Just making sure we can cope with whatever comes.

https://github.com/llvm/llvm-project/pull/184294