Happy path

Roman Kennke rkennke at redhat.com
Wed Feb 14 10:39:30 UTC 2018


Am 14.02.2018 um 11:03 schrieb Roland Westrelin:
>
> http://cr.openjdk.java.net/~roland/shenandoah/happypath/webrev.00/
>
> This changes the implementation of the write barrier and pre barrier so
> they first check whether the gc state is zero. Back to back tests are
> merged, loops are unswitched around a gc state test. There's more work
> to do to merge 2 gc state ifs if there's some control flow between,
> push read barriers in gc state ifs, remove read barriers on the gc state
> zero branch. For instance:
>
>      static void m(A a, Object o) {
>          a.o = o;
>      }
>
> still has 2 checks for gc state is zero because there's some control
> flow between the 2.
>
> Roland.
>

I can only trust you with the actual patch. :-)

Have you run hotspot_gc_shenandoah tests and maybe one or the other
application with it?

Is there any chance to keep Shenandoah paths more contained?

Thanks,
Roman


More information about the shenandoah-dev mailing list