Author: Remi Meier <[email protected]>
Branch: c8-faster-smallobj-sync
Changeset: r1930:fe7f0b6ebe7e
Date: 2015-08-11 13:01 +0200
http://bitbucket.org/pypy/stmgc/changeset/fe7f0b6ebe7e/
Log: think some more
diff --git a/c8/stm/smallmalloc.h b/c8/stm/smallmalloc.h
--- a/c8/stm/smallmalloc.h
+++ b/c8/stm/smallmalloc.h
@@ -52,15 +52,25 @@
/* lists the memory ranges of uncommitted/overflow objs that
need to be flushed to other segments on commit (like
- large_overflow_objects) */
+ large_overflow_objects). (unsorted, a range never overlaps
+ pages) */
struct list_s *uncommitted_ranges;
- /* List of smallobjs that were unreachable in a major GC. We
- really free them only on commit/abort because we would
- otherwise need to deal with holes generated in the
- uncommitted_ranges during major GCs (and I can't think of
- a non-quadratic algorithm to deal with this). */
- struct list_s *delayed_free_objs;
+ /* with "uncommitted_ranges", we do the following during major GCs:
+
+ for seg in segments:
+ for r in seg.uncommitted_ranges:
+ size = obj_size_in_seg(seg, (object_t*)r)
+ assert(r.size % size == 0)
+ for(obj = r; obj < r + r.size; r += size):
+ if obj_dies(obj):
+ adjust_range(r) // split/shrink
+
+ Optionally, we can also merge ranges, sort the list, etc.
+ But generally, I hope we don't have too many uncommitted
+ ranges; and I don't know, if there is another way to get
+ the size(-class) of the page/objs.
+ */
};
_______________________________________________
pypy-commit mailing list
[email protected]
https://mail.python.org/mailman/listinfo/pypy-commit
