RFR: 8318446: C2: optimize stores into primitive arrays by combining values into larger store [v11]

[Vladimir Kozlov]

On Fri, 23 Feb 2024 15:21:53 GMT, Emanuel Peter <epeter at openjdk.org> wrote:

>> src/hotspot/share/opto/memnode.cpp line 2860:
>> 
>>> 2858:   Node* new_mem  = first->in(MemNode::Memory);
>>> 2859:   Node* new_adr  = first->in(MemNode::Address);
>>> 2860:   const TypePtr* atp = TypeRawPtr::BOTTOM;
>> 
>> What does C2 prevent from moving a load from one of the array elements to above the raw store, assuming they are independent?
>
> I don't know. I have added a test now for that, but probably need more tests.
> @vnkozlov @rwestrel  do you think I need to do something about this?

Yes, this could be the issue if we don't check that memory accesses offset overlaps.

There was issue we recently fixed [#16015](https://github.com/openjdk/jdk/pull/16015) where we did not take into account wider memory access to array.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/16245#discussion_r1501017756