[Mlir-commits] [mlir] [MLIR][Linalg] Ternary Op & Linalg select (PR #91461)

[Matthias Springer]

https://github.com/matthias-springer approved this pull request.

I'm not too familiar with the Op DSL part, but this op makes sense to me.

There is actually a TODO in the `BufferizableOpInterface` implementation of `arith.select`: only `i1` conditions are supported. That is because there is no efficient way to bufferize `arith.select` with tensor conditions because the op is not in destination style. It is good to have a `linalg.select` destination-style op that can bufferize efficiently.

https://github.com/llvm/llvm-project/pull/91461