[llvm] [NFC][LoopVectorize] Make replaceVPBBWithIRVPBB more efficient (PR #111514)

[Florian Hahn via llvm-commits]

https://github.com/fhahn commented:

Thanks for putting up the patch! 

The order of successors and predecessors is meaningful in VPlan, e.g. to determine the order of operands in a branch instruction for VPBlocks with multiple successors. So it should definitely be preserved here, it wasn't an issue so far as the middle block always had a single predecessor



https://github.com/llvm/llvm-project/pull/111514