Andy Wingo
3ee2009de9
Move a lot of mark_space state to heap
2022-04-18 20:56:48 +02:00
Andy Wingo
119e273fa4
Rename mark-sweep "markers" to "tracers"
...
There could be other reasons than marking to trace the heap.
2022-04-18 15:19:55 +02:00
Andy Wingo
19f7f72b68
Rename mark-sweep "large" objects to "medium"
2022-04-18 10:00:44 +02:00
Andy Wingo
3f54fb3dbf
Fix semispace page stealing
...
Ensure number of stolen pages is even. Avoid madvising on every
collection. Cache the page size.
2022-04-17 21:51:20 +02:00
Andy Wingo
3315fc7477
Add large object space to semi-space collector
2022-04-14 22:20:27 +02:00
Andy Wingo
619a49ba41
Add large object space
...
Not wired up yet.
2022-04-13 21:43:18 +02:00
Andy Wingo
d425620d37
Add address map and set
2022-04-12 21:41:26 +02:00
Andy Wingo
b0b4c4d893
Remove unneeded files
2022-03-31 09:24:54 +02:00
Andy Wingo
54ce801c72
Update README now that we have parallel mutators
2022-03-30 23:21:45 +02:00
Andy Wingo
a1dbbfd6ae
Speed up sweeping for small objects
...
When sweeping for small objects of a known size, instead of fitting
swept regions into the largest available bucket size, eagerly break
the regions into the requested size. Throw away any fragmented space;
the next collection will get it.
When allocating small objects, just look in the size-segmented freelist;
don't grovel in other sizes on the global freelist. The thought is that
we only add to the global freelists when allocating large objects, and
in that case some fragmentation is OK. Perhaps this is the wrong
dynamic.
Reclaim 32 kB at a time instead of 1 kB. This helps remove scalability
bottlenecks.
2022-03-30 23:15:29 +02:00
Andy Wingo
6300203738
Add call_without_gc API
...
This lets us call pthread_join safely
2022-03-29 21:58:52 +02:00
Andy Wingo
680032fa89
Minor stop-the-world optimizations. There are still bugs
...
Probably should switch to using a semaphore; no need to reacquire the
lock on wakeup.
2022-03-29 15:47:19 +02:00
Andy Wingo
d879a01913
Remove gcbench in favor of mt-gcbench. Update quads
2022-03-29 15:12:56 +02:00
Andy Wingo
5522d827e3
mt-gcbench: write the "j" field in the binary tree nodes.
2022-03-29 15:12:56 +02:00
Andy Wingo
ac57e01e31
BDW doesn't have mutator-local freelists for pointerless objects
2022-03-29 15:12:56 +02:00
Andy Wingo
e837d51f53
mark-sweep collector allows parallel mutators
2022-03-29 15:12:56 +02:00
Andy Wingo
ded3b3c7a3
Update parallel marker API to use struct gcobj
2022-03-29 15:12:56 +02:00
Andy Wingo
14529f11e9
mark-sweep: add global small object freelist
...
This will be useful to collect data when sweeping, if a mutator doesn't
need those objects.
2022-03-29 15:11:56 +02:00
Andy Wingo
2d1e76eccc
mark-sweep: remote markers can send roots via mark buffers
...
When you have multiple mutators -- perhaps many more than marker threads
-- they can mark their roots in parallel but they can't enqueue them on
the same mark queue concurrently -- mark queues are single-producer,
multiple-consumer queues. Therefore, mutator threads will collect grey
roots from their own root sets, and then send them to the mutator that
is controlling GC, for it to add to the mark queue (somehow).
2022-03-29 15:07:59 +02:00
Andy Wingo
be90f7ba49
mark-sweep: Remove context, use mark space instead
...
This is the end of a series of refactors before adding thread-local
allocation.
2022-03-29 15:07:32 +02:00
Andy Wingo
9b0bc6e975
mark-sweep: Update markers to deal in heap and spaces
...
This will let us get rid of "struct context".
2022-03-29 15:06:28 +02:00
Andy Wingo
2401732e31
mark-sweep: mutator data structure separate from heap
...
This will allow thread-local allocation buffers.
2022-03-29 15:05:59 +02:00
Andy Wingo
61d38e4205
Refactor mark-sweep to send mutator to collect()
...
This will let the mutator hold a pointer to the heap.
2022-03-29 15:05:33 +02:00
Andy Wingo
edd46d8fe2
Start to adapt mark-sweep collector for separate heap/mutator
...
The current hack is that the mutator contains the heap. We'll relax
later on.
2022-03-29 15:04:53 +02:00
Andy Wingo
5a92b43e94
Change serial marker to deal in struct gcobj* instead of uintptr
...
"struct gcobj*" is how we denote live objects, and the marker will only
see live objects.
2022-03-29 15:04:20 +02:00
Andy Wingo
81037fd6d2
Convert semi-space collector to new API
2022-03-28 22:18:25 +02:00
Andy Wingo
06a213d1ed
Adapt GC API to have separate heap and mutator structs
...
Only BDW is adapted, so far.
2022-03-28 20:49:24 +02:00
Andy Wingo
883a761775
Stub out support for multiple mutator threads on semi, mark-sweep
...
For semi probably we never implement support for multiple mutator
threads. We will do local freelists for mark-sweep though.
2022-03-20 21:03:26 +01:00
Andy Wingo
a654a790b9
Add inline allocation for small objects for bdw-gc
2022-03-18 22:57:41 +01:00
Andy Wingo
d63288048c
Add mt-gcbench
2022-03-18 16:19:42 +01:00
Andy Wingo
e703568857
Remove heap-stretching phase
...
We should separate evaluation of the heap stretching heuristics from the
evaluation of the GC itself, otherwise our analysis of the GC itself
will be too sensitive to the details of the final heap size. Anyway
this doesn't affect results as we already specified the heap size
precisely.
2022-03-18 14:36:48 +01:00
Andy Wingo
4b7fb84ba0
gcbench takes heap multiplier on command line
2022-03-18 14:29:59 +01:00
Andy Wingo
887bdd5441
Clean up gcbench naming, to be consistent
2022-03-18 09:42:20 +01:00
Andy Wingo
7dda5b992d
Refactor pop_handle to not take the handle
2022-03-16 21:36:21 +01:00
Andy Wingo
32ddaa7624
Allocate GC context in GC-managed heap
2022-03-16 21:31:51 +01:00
Andy Wingo
f04b0bbd45
Simplify output of quads test
2022-03-16 14:28:49 +01:00
Andy Wingo
e7a3f83bcc
Add quads benchmark
...
Also expand GC interface with "allocate_pointerless". Limit lazy
sweeping to the allocation size that is causing the sweep, without
adding to fragmentation.
2022-03-16 14:16:22 +01:00
Andy Wingo
aac0faf4cf
Refactor type definitions
2022-03-16 09:05:31 +01:00
Andy Wingo
a1b4311cfc
Update status
2022-03-13 21:55:58 +01:00
Andy Wingo
a693c4ea8a
Bugfix to mark-sweep
...
Before this, the last sweep would cause premature gc
2022-03-13 21:45:20 +01:00
Andy Wingo
fddd4d9416
Hey parallel marking is finally an improvement??
2022-03-13 21:38:59 +01:00
Andy Wingo
4d7041bfa9
Another attempt at parallel marking, avoiding the channel
...
Not great though!
2022-03-13 13:54:58 +01:00
Andy Wingo
7ce07de670
First crack at parallel marking
2022-03-12 21:09:17 +01:00
Andy Wingo
9c89672c88
Put a local mark queue in front of the work-stealing queue
2022-03-11 11:57:14 +01:00
Andy Wingo
df9edfdff2
Remove tiny objects from mark-sweep
2022-03-11 11:48:26 +01:00
Andy Wingo
f57a1b8a55
Refactor to separate gcbench from gc
2022-03-11 11:48:26 +01:00
Andy Wingo
77ac530360
Add beginnings of parallel marker
2022-03-11 11:48:26 +01:00
Andy Wingo
01d3f9627e
Further accelerate sweeping
2022-03-11 11:48:17 +01:00
Andy Wingo
f6ac9d2571
Ability to set heap size on command line
2022-03-11 11:48:04 +01:00
Andy Wingo
5edc4fa81a
More efficient sweep
2022-03-11 11:44:11 +01:00
