RFR: 8317677: Specialize Vtablestubs::entry_for() for VtableBlob

Thomas Schatzl tschatzl at openjdk.org
Fri Oct 13 07:42:26 UTC 2023


On Thu, 12 Oct 2023 05:26:53 GMT, Tobias Hartmann <thartmann at openjdk.org> wrote:

>> Hi all,
>> 
>>   please review this change that specializes the call to `Vtablestubs::entry_for` for vtable blob.
>> 
>> This kind of code blobs must always be in the `Vtablestubs` table (they are only created by the `Vtablestubs` class), so trying to find that stub there using it is unnecessary. For verification I added an assert that verifies it's found.
>> 
>> Given that `Vtablestubs::entry_point()` would take a lock this removal of the lock results in a large speedup (https://bugs.openjdk.org/browse/JDK-8317527) in MT situations and is an even better change than that mentioned CR - and also worth the extra code (I did not retest performance since the `Vtablestubs` table lookup should be fairly fast).
>> 
>> Fwiw, the first commit contains a different variant of this change, keeping the decision out of `Vtablestubs`, but I liked the current one better for it encapsulating that decision more. I am fine with reverting the second commit.
>> 
>> Testing: tier1-7, gha, stress testing
>> 
>> Thanks,
>>   Thomas
>
> Nice improvement, looks good to me.

Thanks @TobiHartmann @vnkozlov for your reviews

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

PR Comment: https://git.openjdk.org/jdk/pull/16099#issuecomment-1761052291


More information about the hotspot-compiler-dev mailing list