RFR: 8140393: Move WorkerDataArray to its own file
Tom Benson
tom.benson at oracle.com
Mon Oct 26 15:44:20 UTC 2015
Hi Eric,
Looks good to me. One very minor comment - In set_thread_work_item, the
assertion comment "No sub count" (which you didn't change) seems a
little off. I agree removing the caching code is good, though I was
wary at first. Not because of the expense of walking the array, but
the potential expense of dfloat ops (as on a previous platform I worked
on). But in reality I don't think the value will be recomputed much, if
ever.
Tom
On 10/26/2015 8:23 AM, Erik Helin wrote:
> Hi all,
>
> this patch moves the class WorkerDataArray into its own file. While
> moving to the code, I also refurbished it a bit. WorkerDataArray used
> to cache the results for calls to max(), min(), sum() etc because it
> might be expensive too loop over the array. However, the array we are
> iterating over will at most be as big as the number of GC threads (it is
> very rare to use more than 1000 GC threads). Performance runs shows no
> regressions at all, so I prefer to go with the simpler code.
>
> Enhancement:
> https://bugs.openjdk.java.net/browse/JDK-8140393
>
> Webrev:
> http://cr.openjdk.java.net/~ehelin/8140393/webrev.00/
>
> Testing:
> - JPRT
> - Perf Aurora for performance regressions
>
> Thanks,
> Erik
More information about the hotspot-gc-dev
mailing list