RFR: 8291065: Creating a VarHandle for a static field triggers class initialization [v14]

[Paul Sandoz]

On Tue, 27 Jun 2023 13:57:31 GMT, Chen Liang <liach at openjdk.org> wrote:

>> This patch implements lazy initialization for VarHandle working on static fields. It has a good initial call performance.
>> 
>> We introduce a new internal API, `target()` to unpack a lazy VarHandle in VH implementation methods. If called via MethodHandle, a barrier is added in the MethodHandle instead.
>> 
>> The new test ensures the correctness of Lazy VH for both direct and indirect invocation; the performance of MethodHandle version of lazy VH is not yet tested.
>> 
>> 
>> Benchmark                                            Mode  Cnt    Score    Error  Units
>> LazyStaticColdStart.methodHandleCreateEager            ss   10   41.490 ± 12.331  us/op
>> LazyStaticColdStart.methodHandleCreateLazy             ss   10   21.810 ± 16.964  us/op
>> LazyStaticColdStart.methodHandleInitializeCallEager    ss   10   57.860 ± 13.738  us/op
>> LazyStaticColdStart.methodHandleInitializeCallLazy     ss   10   93.300 ± 18.858  us/op
>> LazyStaticColdStart.varHandleCreateEager               ss   10   39.860 ±  9.362  us/op
>> LazyStaticColdStart.varHandleCreateLazy                ss   10   17.630 ±  1.111  us/op
>> LazyStaticColdStart.varHandleInitializeCallEager       ss   10  123.170 ± 62.468  us/op
>> LazyStaticColdStart.varHandleInitializeCallLazy        ss   10  105.390 ± 41.815  us/op
>
> Chen Liang has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Rollback VHG changes

Latest update looks very good, i like how it has got simpler and therefore easier to understand. I agree with a focus on the correctness for the MHs, as it is quite hard to reason about all this. The initializing MHs should optimize, it just takes more work to do so.

I shall provide some minor comments inline.

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

PR Comment: https://git.openjdk.org/jdk/pull/13821#issuecomment-1610190790