Optimize region count #56471
Optimize region count #56471
Conversation
…ating over all the regions that we have reserved address space for, iterate just over the regions actually in use.
|
Tagging subscribers to this area: @dotnet/gc Issue DetailsWe have a couple of arrays with an entry per region - instead of iterating over all the regions that we have reserved address space for, iterate just over the regions actually in use. Arrays affected by this are mark_list_piece_start, mark_list_piece_end, survived_per_region and old_card_survived per_region. To avoid having inconsistent counts, we get the count of regions actually in use at the start of mark_phase. Any regions added during GC should not matter, because they shouldn't contain any marked objects.
|
src/coreclr/gc/gcpriv.h
Outdated
| return ((region_map_right_start == region_map_right_end) | ||
| ? (region_map_left_end - region_map_left_start) | ||
| : (region_map_right_end - region_map_left_start)); |
There was a problem hiding this comment.
actually I do think we don't need to care about the region_map_right case... since it's not used at all and the way we were thinking of using it will make this calculation incorrect anyway. so for now we can assume we don't need to worry about it at all.
There was a problem hiding this comment.
Ok, perhaps put an assert instead of the ?:
…xecuted) by an assert.
|
LGTM! |
ghost
locked as resolved and limited conversation to collaborators
We have a couple of arrays with an entry per region - instead of iterating over all the regions that we have reserved address space for, iterate just over the regions actually in use.
Arrays affected by this are mark_list_piece_start, mark_list_piece_end, survived_per_region and old_card_survived per_region.
To avoid having inconsistent counts, we get the count of regions actually in use at the start of mark_phase. Any regions added during GC should not matter, because they shouldn't contain any marked objects.