[llvm] [LLVM][CodeGen][SVE] Use pairwise instruction for partial_reduce_[s/u]mla. (PR #212772)

[Gaëtan Bossu via llvm-commits]

https://github.com/gbossu approved this pull request.

Came across this PR so I thought I might as well look at it. This looks like a clear win to me. The ADALP family is perfect for 2x partial reductions. 

https://github.com/llvm/llvm-project/pull/212772