[master] RFR: JDK-8325104: Lilliput: Shrink Classpointers [v3]

[John R Rose]

On Tue, 26 Mar 2024 09:21:20 GMT, Thomas Stuefe <stuefe at openjdk.org> wrote:

> My short-term contingency plan had been to use some of the now free four bits to increase the Klass ID bits and thus bring alignment back to 64 bytes. That would work at least for 64-bit headers, as long as Valhalla does not need those bits too.

Thanks for the response, Thomas.  (It looks like you remembered our conversation better than I did.)  Taking back the four free bits is a good tactic for now.

It may possibly squeeze Valhalla, at which point we will have to evaluate our choices:  Should we encode the valhalla bits with a joint multi-bit encoding embedded in the klass-ID?  Should we look again at mod-1088 encodings discussed here?  Should we use a [joint bit encoding](https://cr.openjdk.org/~jrose/jvm/joint-bit-encodings.html) for all array klass IDs, since Valhalla needs special markings on arrays anyway?  Those are the sorts of questions that could arise.

-------------

PR Comment: https://git.openjdk.org/lilliput/pull/128#issuecomment-2023717893