RFR: 8305236: The LoadLoad barrier has been useless after JDK-8220051

Tue Apr 4 02:26:05 UTC 2023

On Thu, 30 Mar 2023 12:49:09 GMT, SUN Guoyun <duke at openjdk.org> wrote:

> After JDK-8220051, Interpreter::notice_safepoints() only be executed at a safe point, so LoadLoad barrier is useless.
> Barrier directives are generally time-consuming, so this patch removes LoadLoad be used for aarch64 and riscv.
> 
> Jtreg tier1-2 testing for aarch64 has been done.
> 
> Please help review it.
> 
> Thanks.

This really needs approval by someone involved in the Aarch64 and RiscV ports, but given `notice_safepoints` is now only used by JVMTI and called from a VM operation at a safepoint, I concur that these `loadload` barriers have no role to play in "seeing" the updated dispatch table.

Thanks.

May I suggest changing the PR and JBS issue title to:

8305236: Some LoadLoad barriers in the interpreter are unnecessary after JDK-8220051

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

Marked as reviewed by dholmes (Reviewer).

PR Review: https://git.openjdk.org/jdk/pull/13244#pullrequestreview-1370079110
PR Comment: https://git.openjdk.org/jdk/pull/13244#issuecomment-1495248434