Basic GC.stat for MMTk

[chrisseaton]

Fix towards mmtk/mmtk-ruby#1.

before = GC.stat

pp before

1_0000_000.times do
  [1, 2, 3].map(&:succ)
end

after = GC.stat

pp after.map { |k, v| [k, v - (before[k] || 0)] }.to_h.select { |k, v| v > 0 }

{:total_allocated_objects=>57044,
 :malloc_increase_bytes=>4879488,
 :malloc_increase_bytes_limit=>16777216,
 :oldmalloc_increase_bytes=>4879488,
 :oldmalloc_increase_bytes_limit=>0,
 :mmtk_free_bytes=>995471360,
 :mmtk_total_bytes=>1000001536,
 :mmtk_used_bytes=>4530176,
 :mmtk_starting_heap_address=>2199023255552,
 :mmtk_last_heap_address=>35184372088832}
{:total_allocated_objects=>20003460, :malloc_increase_bytes=>227872, :oldmalloc_increase_bytes=>227872, :mmtk_used_bytes=>174497792}

Is mmtk_free_bytes working? Why doesn't it change with allocation and collection activity?

[wks]

It decreased, but you filtered out all fields with v > 0. Changing it to v != 0 and it shows the difference, for example:

{:total_allocated_objects=>9862, :malloc_increase_bytes=>2235696, :malloc_increase_bytes_limit=>16777216, :oldmalloc_increase_bytes=>2235696, :oldmalloc_increase_bytes_limit=>0, :mmtk_free_bytes=>523583488, :mmtk_total_bytes=>524288000, :mmtk_used_bytes=>704512, :mmtk_starting_heap_address=>2199023255552, :mmtk_last_heap_address=>35184372088832}
{:total_allocated_objects=>2000077, :malloc_increase_bytes=>14336, :oldmalloc_increase_bytes=>14336, :mmtk_free_bytes=>-117276672, :mmtk_used_bytes=>117276672}

I slightly changed the number of iterations to make it run faster. My program runs for one million iterations, so it effectively creates two million three-element arrays. Assume each three-element array fits in a 40-byte slot, and my mmtk-ruby binding adds one 8-byte hidden field to record the object size. It should be a total of 96MB of allocation, which is pretty close to -117276672, assuming that there are other objects created during the execution, or there is overhead for the underlying malloc implementation.

If I set MMTK_PLAN=Immix (as long as it doesn't trigger GC, it should be OK), the :mmtk_free_bytes will be -99483648. I think this measurement is more precise given that the Immix space is implemented inside mmtk-core.

[wks]

Since the code is actually working, I merged this PR.