Performance regression on mutator time for semispace in PR#555 #610
Description
@wenyuzhao reported performance regression on semispace between 519e8f3(20211206) and cd6d898(20220322). The most noticeable issue is that the mutator time for lusearch is slowed down by 60-70%. The regression is only measurable with many mutator threads.
8 cores: http://squirrel.anu.edu.au/plotty-public/wenyuz/v8/p/xDFxUB Mar-2022 build is 7% faster on time, 3% slower on time.other
16 cores: http://squirrel.anu.edu.au/plotty-public/wenyuz/v8/p/CsGQdJ Mar-2022 build is 3% faster on time, 8% slower on time.other
32 cores: http://squirrel.anu.edu.au/plotty-public/wenyuz/v8/p/jCt85v Mar-2022 build is 16% slower on time, 77% slower on time.other
The complete discussion is here: https://mmtk.zulipchat.com/#narrow/stream/262673-mmtk-core/topic/Performance.20regression
After some measurement, the problematic commit is this one: #555, which added a lock in the allocation slowpath. That is possibly the issue. I will further look into it.
