llvm-project

783bf90e - [SLP]Vectorize operand chains of non-vectorizable instructions

[SLP]Vectorize operand chains of non-vectorizable instructions Extend the post-process operand-chain seeding (previously only cmps) to non-vectorizable calls, invokes, callbrs, non-trivially-vectorizable intrinsics, atomicrmw, cmpxchg, returns, and stores. Stores are processed after every other vectorization attempt in the basic block. Reviewers: bababuck, hiraditya, RKSimon Pull Request: https://github.com/llvm/llvm-project/pull/194248