Collect live bytes per space, and report by space

[qinsoon]

The current count_live_bytes_in_gc feature adds the size of all live objects and compare with the used pages reported by the plan. There are two issues with the feature:

1. VM space is not included in the used pages reported by the plan, but the live objects include objects in the VM space. So the reported fragmentation/utilization is wrong when the VM space is in use.
2. Spaces/policies have very different fragmentation ratio. Reporting the fragmentation for the entire heap is not useful.

This PR refactors the current count_live_bytes_in_gc feature so we collect live bytes per space, and report by space.

[wks]

Using space name as the key may be inefficient. Can we use SpaceDescriptor::get_index() to get an integer ordinal for the space, and use it to index a Vec<LiveBytesStats> (or [LiveBytesStats; MAX_SPACES]) instead?

In MMTK::aggregate_live_bytes_in_last_gc, we can get the SpaceDescriptor using space.common().descriptor.

In increase_live_bytes, we can use VMMap::get_descriptor_for_address to get the SpaceDescriptor. The SFT should work, too, but we need to add a new method for getting the descriptor, or getting a reference to CommonSpace from SFT.

[qinsoon]

Using space name as the key may be inefficient. Can we use SpaceDescriptor::get_index() to get an integer ordinal for the space, and use it to index a Vec<LiveBytesStats> (or [LiveBytesStats; MAX_SPACES]) instead?

Sounds good. I have made the change.

We might be able to use a runtime option for collecting live bytes data, and might be able to leave it enabled by default. I will run benchmarks first to know the overhead. Probably let's postpone the review for this PR until I get the benchmark results.

[wks]

AtomicRefCell.borrow_mut() has a cost. If this ends up being the bottleneck, we can do the borrowing in ScanObjectWork::do_work_common just before the for loop, and pass the &mut [...] to increase_live_bytes.

[qinsoon]

The following graph shows 3 runs: 1. count_live_bytes_in_gc as a feature and the feature is disabled, 2. count_live_bytes_in_gc as a runtime option and it is turned off, and 3. count_live_bytes_in_gc as a runtime option and it is turned on. The overhead from runtime option is little (0.7% GC time when it is turned off, and 1% GC time when it is turned on). So I made it as a runtime option.

plotty

[wks]

LGTM