RFR(S): 8239072: subtype check macro node causes node budget to be exhausted
Roland Westrelin
rwestrel at redhat.com
Tue Mar 31 08:52:28 UTC 2020
> I think we should consider rework how we calculate expansion size (number of ideal nodes) for macro nodes.
> One suggestion would be to follow what we do for Mach nodes - have method expansion_node_count() in macro nodes which
> give worst case number. Yes, it will complicate calculation code but at least we don't need to assume that all macro
> nodes have huge expansion and guess numbers based only on how many macro nodes we have.
I don't think that would be sufficient. Macro node expansion runs with
PhaseIterGVN::_delay_transform set to true. So there can be some
duplication of nodes that only fold once IGVN runs.
Actually, I like Vladimir I's suggestion so here is a new patch based on
it:
http://cr.openjdk.java.net/~roland/8239072/webrev.01/
Roland.
More information about the hotspot-compiler-dev
mailing list