RFR: 8233494: Avoid calling MallocTracker::record_malloc and record_free when NMT is off

Claes Redestad claes.redestad at oracle.com
Mon Nov 4 14:38:29 UTC 2019


Martin, Zhengyu,

thank you for reviewing!

/Claes

On 2019-11-04 15:18, Zhengyu Gu wrote:
> Looks good to me too.
> 
> Thanks,
> 
> -Zhengyu
> 
> On 11/4/19 9:14 AM, Doerr, Martin wrote:
>> Hi Claes,
>>
>> this makes sense. Change looks good to me.
>>
>> Best regards,
>> Martin
>>
>>
>>> -----Original Message-----
>>> From: hotspot-runtime-dev <hotspot-runtime-dev-
>>> bounces at openjdk.java.net> On Behalf Of Claes Redestad
>>> Sent: Montag, 4. November 2019 14:54
>>> To: Hotspot dev runtime <hotspot-runtime-dev at openjdk.java.net>
>>> Subject: RFR: 8233494: Avoid calling MallocTracker::record_malloc and
>>> record_free when NMT is off
>>>
>>> Hi,
>>>
>>> this patch removes some small but measurable NMT-related overheads
>>> when
>>> when NMT is disabled, by moving NMT_off checks out into MemTracker
>>> where
>>> they can be more aggressively inlined.
>>>
>>> Bug:    https://bugs.openjdk.java.net/browse/JDK-8233494
>>> Webrev: http://cr.openjdk.java.net/~redestad/8233494/open.00/
>>>
>>> Motivation:
>>>
>>> Overhead of calling MallocTracker methods account for ~15-25% of
>>> instructions retired by os::malloc/realloc/free. On a "Hello World" on
>>> my laptop (no large pages) we already do roughly 9k os::malloc calls, so
>>> this improvement means a reduction in instructions retired by ~250k, or
>>> ~0.2% of the total. There is no discernible difference in behavior or
>>> overhead for the case when NMT is enabled.
>>>
>>> Testing: tier1-2
>>>
>>> Thanks!
>>>
>>> /Claes
> 


More information about the hotspot-runtime-dev mailing list