On Thu, 24 Oct 2024 02:13:20 GMT, Jasmine Karthikeyan <jkarthike...@openjdk.org> wrote:
> I think we shouldn't run `Ideal` on the graph, because there is a chance that > it could undo the lowering changes that we just did. This gives lowering more > freedom to change the graph in different ways that would otherwise be undone > by Ideal. We run `Value` on all of the transformed nodes mainly so the types > table is accurate, so we can call upon the type of any node during lowering. We need to preserve lowering canonicalizations, but lowered graph is susceptible to further idealizations, else we could do lowering during final graph reshaping just before matching. ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/21599#discussion_r1814176800
