RFR: Remove prefetch during mark

Per Liden per.liden at oracle.com
Wed Feb 14 09:07:00 UTC 2018


Hi Aleksey,

On 02/14/2018 09:48 AM, Aleksey Shipilev wrote:
> On 02/14/2018 09:43 AM, Per Liden wrote:
>> Small patch to remove the prefetch we do during marking. Charlie Hunt spent some time looking into
>> the effect of this, and in conclusion the prefetch doesn't help.
> 
> Our Shenandoah experience corroborates this: once you lay out the objects in traversal order (or
> some approximation of it), the mark prefetch becomes actively hostile: wastes cycles/instruction for
> the accesses that are already implicitly-prefetched well.

I think you're right. In our case, Charlie didn't see any major problems 
caused by the prefetch, but it also wasn't very helpful.

/Per


More information about the zgc-dev mailing list