[llvm] [InstCombine] Scalarize `(vec_ops (insert ?, X, Idx))` when only one element is demanded (PR #84645)

[Yingwei Zheng via llvm-commits]

dtcxzyw wrote:

> Are you saying that SLP is creating these vectorizations but shouldn't? Isnt that a cost issue with the alt opcode cases?

See the test `replace_through_casts` in `llvm/test/Transforms/InstCombine/shufflevector-inselt.ll`. The second element of `ui_v` and the first element of `si_v` are unused. Is this behavior expected in SLP?


https://github.com/llvm/llvm-project/pull/84645