RFR (S) 8231410: Shenandoah: clone barrier should use base pointer

[Aleksey Shipilev]

On 9/25/19 2:13 PM, Roman Kennke wrote:
> I guess it's more about not touching headers. Ultimately it's a little bit of overkill, but I am
> suspicious of clone() because I noticed some benchmarks make heavy use of it. The runtime is only
> touched on >32 or unknown numbers of fields though. Anyhow, I think it doesn't hurt to touch less
> memory while copying, especially because the compiled setup-code computes all those interior
> pointers and counts for us already anyway.
> 
> I removed printing, fixed the signatures and also added include for copy.hpp:
> 
> http://cr.openjdk.java.net/~rkennke/JDK-8231410/webrev.04/

I missed this the last time around: are we good with having jlong* in the signatures? That seems to
be the implementation detail for the actual copy. We are passing the addresses there, and even
though they should be aligned to 8, they are not necessarily jlong*-s. It seems the older way of
passing oopDesc* is just as good?

> Running full test suites. Ok, given passing tests?

Yeah.

-- 
Thanks,
-Aleksey