RFR: 8213352: Separate BufferNode allocation from PtrQueueSet
thomas.schatzl at oracle.com
Wed Nov 7 23:22:42 UTC 2018
On Sun, 2018-11-04 at 23:51 -0500, Kim Barrett wrote:
> Please review this change to PtrQueue buffer allocation, moving the
> implementation of free-list based allocation from PtrQueueSet to a
> separate BufferNode::Allocator object. The old code supported
> of free-lists by having one PtrQueueSet delegate to another; now
> free-list sharing is provided by using the same allocator object.
> mach5 tier1-5
> Added gtest for new allocator class.
- copyright for the gtest should be 2018 only
- out of curiosity, not an issue in the code, but why the need for
Atomic::load() in BufferNode::Allocator::free_count? It's okay, but I
do not see what Atomic::load provides you in this case compared to a
normal load. Is this the new preferred style to load from a volatile
variable of primitive type?
Don't need to see the new webrev for the copyright date change.
More information about the hotspot-gc-dev