[llvm] [GenericDomTreeConstruction] Speed up Semi-NCA construction (PR #207474)

[Alexis Engelke via llvm-commits]

https://github.com/aengelke commented:

LGTM otherwise. Nice improvement.

In the domtree, getNodeInfo tends to be expensive due to SmallVector<SmallVector<>> resize, which has to initialize all SmallVector pointer/capacities. But I suppose there's no way to change this without a new way of handling predecessors...

https://github.com/llvm/llvm-project/pull/207474