[foreign-memaccess] RFR: JDK-8242495: Restructure implementation of memory segments

[Jorn Vernee]

On Fri, 10 Apr 2020 16:51:08 GMT, Maurizio Cimadamore <mcimadamore at openjdk.org> wrote:

>> Interesting, I thought it might be the explicit casts introduced in the MH machinery of lambdas, and since everything
>> is final and stable (?) the casts usually help the JIT, if so then there is clearly more going on with the MH machinery.
>
> that's true, although I'd expect that with MH the Class constant is stored in some separate spinned class - e.g. it is
> a true constant, not a final field load?

Maybe the speedup comes from each lambda having a separate type (a class is spun for each), so C2 can apply type
profiling & guarded inlining, and within the inlined section a sharper array type can be assumed.

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

PR: https://git.openjdk.java.net/panama-foreign/pull/109