On Wed, 4 Mar 2026 10:16:31 GMT, Eric Fang <[email protected]> wrote:
> For a particular inlining order, VectorStoreMaskNode never gets added to the > IGVN worklist. That sounds like the real problem to me. We should make sure that they are always added, no? > At that point, in PhaseVector::do_cleanup() we can add all vector nodes to > the IGVN worklist, which helps trigger optimizations that might otherwise be > missed. I'm not yet sure that is the solution, but I'd have to dig a bit deeper myself. Why can't we just add all the nodes to the IGVN worklist when we do the inlining? How do we do this with "regular" inlining of scalar methods? We must put those nodes on the IGVN worklist as well, I'm thinking. ------------- PR Comment: https://git.openjdk.org/jdk/pull/28313#issuecomment-3996688231
