More precise writebarrier for regions

[PeterSolMS]

This introduces a lookup table for regions where we can find the current generation and the planned generation efficiently.

The table has byte-sized elements where the low nibble is the current generation and the high nibble is the planned generation.

The table is used in mark_through_cards_helper and in the write barriers (for now only the most frequently used ones, Array.Copy has its own way of setting cards that I haven't fixed).

I have changed the write barrier to only set single bits for the case where a pointer to younger generation is stored into an object in an older generation. This costs an interlocked operation in the case the bit is not already set. Hopefully though this will be more than compensated by lower cost in card marking.

I haven't implemented yet committing only the part of the lookup table that is needed.

[msftbot[bot]]

Tagging subscribers to this area: @dotnet/gc See info in area-owners.md if you want to be subscribed.

Issue Details

---

This introduces a lookup table for regions where we can find the current generation and the planned generation efficiently.

The table has byte-sized elements where the low nibble is the current generation and the high nibble is the planned generation.

The table is used in mark_through_cards_helper and in the write barriers (for now only the most frequently used ones, Array.Copy has its own way of setting cards that I haven't fixed).

I have changed the write barrier to only set single bits for the case where a pointer to younger generation is stored into an object in an older generation. This costs an interlocked operation in the case the bit is not already set. Hopefully though this will be more than compensated by lower cost in card marking.

I haven't implemented yet committing only the part of the lookup table that is needed.

Author:	PeterSolMS
Assignees:	PeterSolMS
Labels:	area-GC-coreclr
Milestone:	-

[Maoni0]

running on a 1st party prod workload -

index		Baseline	New	Diff	Diff %
3	Process Duration (Sec)	53,887.63	53,858.88	-28.751	-0.053
4	Total Allocated MB	16,252,981.58	15,819,053.88	-433,927.69	-2.67
5	Max Size Peak MB	26,878.44	27,017.04	138.607	0.516
6	GC Count	7,269.00	7,573.00	304	4.182
7	Heap Count	48	48	0	0
8	Gen0 Count	3,630.00	3,779.00	149	4.105
9	Gen1 Count	3,466.00	3,621.00	155	4.472
10	Ephemeral Count	7,096.00	7,400.00	304	4.284
11	Gen2 Blocking Count	4	4	0	0
12	BGC Count	169	169	0	0
13	Gen0 Total Pause Time MSec	269,123.47	231,960.87	-37,162.60	-13.81
14	Gen1 Total Pause Time MSec	355,468.13	337,579.93	-17,888.20	-5.032
15	Ephemeral Total Pause Time MSec	624,591.60	569,540.80	-55,050.80	-8.814
16	Blocking Gen2 Total Pause Time MSec	4,260.20	2,376.57	-1,883.63	-44.22
17	BGC Total Pause Time MSec	14,630.25	13,270.05	-1,360.20	-9.297
18	GC Pause Time %	1.194	1.087	-0.108	-9.011
19	Avg. Gen0 Pause Time (ms)	74.139	61.382	-12.757	-17.21
20	Avg. Gen1 Pause Time (ms)	102.559	93.228	-9.33	-9.097
21	Avg. Gen0 Promoted (mb)	170.597	166.073	-4.524	-2.652
22	Avg. Gen1 Promoted (mb)	343.586	332.5	-11.087	-3.227
23	Avg. Gen0 Speed (mb/ms)	2.301	2.706	0.405	17.58
24	Avg. Gen1 Speed (mb/ms)	3.35	3.567	0.216	6.458

looking at 500 GCs during steady state as an example -

[AndyAyersMS]

This improved crossgen2 throughput: