RFR (S) 8150463: StringConcat MH_INLINE_SIZED_EXACT should skip storage initialization
Chris Hegarty
chris.hegarty at oracle.com
Thu Mar 24 09:19:45 UTC 2016
On 24 Mar 2016, at 09:15, Peter Levart <peter.levart at gmail.com> wrote:
> Hi Aleksey,
>
> Looks nice to me, but I'm not a Reviewer. Quite an improvement for longer strings.
+1
-Chris.
> Regards, Peter
>
> On 03/24/2016 08:31 AM, Aleksey Shipilev wrote:
>> On 03/16/2016 11:13 AM, Aleksey Shipilev wrote:
>>> I'd like to improve our advanced String concat strategy,
>>> MH_INLINE_SIZED_EXACT, with skipping the storage initialization:
>>> https://bugs.openjdk.java.net/browse/JDK-8150463
>>>
>>> Webrev:
>>> http://cr.openjdk.java.net/~shade/8150463/webrev.01/
>>>
>>> Performance had improved, as planned:
>>> http://cr.openjdk.java.net/~shade/8150463/perf.txt
>>>
>>> The rest of the difference against BC_SB strategy is code-generation
>>> quirks, to be addressed separately. We pay some price for a runtime
>>> final index check to strengthen against inadvertent bugs in
>>> concatenation code.
>>>
>>> This patch does not affect the String concatenation at this point,
>>> because we are still running with BC_SB strategy. I ran through
>>> java/lang/String jtregs to see if MH_INLINE_SIZED_EXACT works as
>>> expected. Therefore, I don't think this needs to wait for Jigsaw
>>> integration. But, it can also wait, no problem. I'll let Jigsaw folks to
>>> decide on integration timing.
>> Anyone? I have re-based and re-tested the patch after Jigsaw
>> integration, and it still works fine (the webrev identical):
>> http://cr.openjdk.java.net/~shade/8150463/webrev.01/
>>
>> Thanks,
>> -Aleksey
>>
>>
>>
>
More information about the core-libs-dev
mailing list