[llvm] [VectorCombine] New folding pattern for extract/binop/shuffle chains (PR #145232)

[Rajveer Singh Bharadwaj via llvm-commits]

Rajveer100 wrote:

Do you mean just by logic perspective or making them into smaller lambda functions? 

Also, I think `SelectionDAG::matchBinOpReduction` handles only for powers of two. Another difference is, here we can have either an intrinsic instruction / bin op.

In terms of performance, although I haven't tested, my version may be faster although more complex.

https://github.com/llvm/llvm-project/pull/145232