[master] RFR: Smaller class pointers

[Thomas Stuefe]

On Mon, 6 Dec 2021 13:19:31 GMT, Roman Kennke <rkennke at openjdk.org> wrote:

> Hi Thomas, thanks for this great work! 

Thanks, it has been interesting :)

> I could not find anything obviously wrong in the changeset. 22bits Klass* allows for another 2 bits for the hashcode, all in the upper 24 bits, and the remaining 8 bits for locking, GC age, and we even end up with some extra bits for future use. Together, it will allow 32bit headers, once we figure out how to restrict locking to the lower 32 (ideally lowest 3 or even 0) bits.
> 
> Go for it!

Should I integrate, or should I wait for more opinions?

@iklam already had an off-list look at the changes with his CDS eyes.

About the metaspace changes, I feel sufficiently confident myself.

The only thing I am hesitant about is the compiler changes, even if they are straightforward. Especially the aarch64 ones, since I cannot test them as thoroughly as x64.

Cheers, Thomas

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

PR: https://git.openjdk.java.net/lilliput/pull/13