[Mlir-commits] [mlir] [mlir][Linalg] Refine how broadcast dims are treated (PR #99015)

[llvmlistbot at llvm.org]

MaheshRavishankar wrote:

I have not looked into details of this PR, and there is value in having the lower levels of the stack being robust, but 
```
 #map1 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, 0)>
```
is not the canonical representation of broadcast IMO. This is.

```
 #map1 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2)>
```

So just signalling that here.

https://github.com/llvm/llvm-project/pull/99015