I believe we should bite the bullet and delete NewGVN.
Various people have sunk a lot of time into it over the years, and we’re still at a point where the pass is nowhere close to production ready, both in terms of correctness and optimization parity. The algorithm behind NewGVN is nice in theory, but not so nice in practice.
Our efforts would be better spent improving the current GVN implementation, e.g. by migrating it to use MemorySSA.