[Mlir-commits] [mlir] [MLIR][Linalg] Introduce broadcast/transpose semantic to 'linalg.batc… (PR #122275)

[Renato Golin]

rengolin wrote:

> Why not split this into two patches? Those two goals seem orthogonal to me and reviewing small patches is easier (at least on GitHub).

>From discussions on the previous PR (matmul), there should be no differences to what we did there. The ODS implementation is identical to the OpDSL one except the affine map. All previous tests should pass. So the extra logic that needs to be reviewed is actually a very small portion of the whole patch. If we split, the first patch would be almost as big as this one.

But getting the semantics correct is not trivial either. So splitting could lead to an interim state that is not quite here nor there, and we want to avoid that.

https://github.com/llvm/llvm-project/pull/122275