| commit | 0e3af05102caa6ddaaa7bfa16d2fce584b5c66b9 | |
| author | Rodrigo Kumpera <kumpera@gmail.com> | |
| Wed, 18 Apr 2012 21:07:22 +0000 (18 18:07 -0300) | ||
| committer | Rodrigo Kumpera <kumpera@gmail.com> | |
| Thu, 19 Apr 2012 21:35:10 +0000 (19 18:35 -0300) | ||
| tree | f23f403d85cd67f395ff7556647d5da32f09e493 | treesnapshot (tar.gz zip) |
| parent | 5a5128b2d8ccbca5c70a9bd240d58017d19dc95f | commitdiff |
Implement aging in the split nursery.
* sgen-split-nursery.c: Implement aging. Age information is
stored in 512 byte buckets in parallel to the nursery.
This increases our bandwidth requirements by a small bit
but has the advantage not requiring us to mess with the
status word of objects.
Reducing the number of available bits for hashing sounded
like a specially bad idea since ages above 1 are not very
usefull unless you're a long living object.
The object promotion code was split into a fast/slow path
since we now have per-age bump allocators. The fast path
is extremely efficient and does only 3 comparisons and a
bunch of cache hot lookups.
* sgen-split-nursery.c: Implement aging. Age information is
stored in 512 byte buckets in parallel to the nursery.
This increases our bandwidth requirements by a small bit
but has the advantage not requiring us to mess with the
status word of objects.
Reducing the number of available bits for hashing sounded
like a specially bad idea since ages above 1 are not very
usefull unless you're a long living object.
The object promotion code was split into a fast/slow path
since we now have per-age bump allocators. The fast path
is extremely efficient and does only 3 comparisons and a
bunch of cache hot lookups.
| mono/metadata/sgen-split-nursery.c | diffblobblamehistory |