[Mlir-commits] [mlir] [mlir][vector] `vector.fma` is not `ElementwiseMappable` (PR #132611)

Andrzej WarzyƄski llvmlistbot at llvm.org
Mon Mar 24 11:57:03 PDT 2025


banach-space wrote:

> > And as a side question, why do we even need `vector.fma`? Looks like it's 1:1 equivalent of `math.fma` on vectors, except `math.fma` actually does support scalars and tensors.
> 
> +1. I'd be in favor of deprecating it and eventually dropping.

I don't have an opinion just yet. 

It would be good to identify the original rationale for introducing `vector.fma`  - why didn't people use `math.fma`? Also, removing `vector.fma` would mean introducing a dependency of `Vector` on `Math` - perhaps that's the reason `vector.fma` exists?

So, in principle, no objections (removing redundancy is good), but lets make sure it's an informed decision.

https://github.com/llvm/llvm-project/pull/132611


More information about the Mlir-commits mailing list