Activity, fixes.

[anvilvapre]

Activity, fixes.

Do not call SkipDoneActivities method recursively via the
NextActivity property. Rather use the nextActivity member.
Avoiding additional functional calls and a recursively
growing stack.

Do not call ChildActivity and NextActivity properties
twice in a row. Once to test for null and after to access
it's value. It will cause the complete list of activities
to be traversed twice looking for non done activities.

Replace Queue method with a version that does not the
NextActivity property causing an extra call to
SkipDoneActivities. Avoid calling Queue recursively.

Similar replace QueueChild with a version that does
not call additional methods.

Note that ActivitiesImplementing returns only non
done activities. The method name does not suggest this.

Please consider making NextActivity a method to cleary indicate it
involves the logic of skipping Done activities. To let
the called know it is 'expensive'.

Please consider renaming the protected property ChildActivity to
FirstChildActivityNotDone to avoid it being used as being
used as childActivity.

Please consider maintaining a pointer to the first
non done activity. This avoids the need the each time find it.

[dnqbob]

After last engine upgrade the SP seems not good on performance. Will this PR be the cure?

[anvilvapre]

See also #18797

[teinarss]

Are we sure that this is note done in the optimization? Ie have you checked the IL?

[pchote]

I'm far from an expert at understanding IL, but this does appear to be evaluating ChildActivity twice on bleed:

// if ChildActivity != null
	IL_00c2:  ldarg.0 
	IL_00c3:  call instance class OpenRA.Activities.Activity class OpenRA.Activities.Activity::get_ChildActivity()
	IL_00c8:  brfalse.s IL_0101

// if ChildActivity.State == ActivityState.Queued
	IL_00ca:  ldarg.0 
	IL_00cb:  call instance class OpenRA.Activities.Activity class OpenRA.Activities.Activity::get_ChildActivity()
	IL_00d0:  callvirt instance valuetype OpenRA.Activities.ActivityState class OpenRA.Activities.Activity::get_State()
	IL_00d5:  brtrue.s IL_0101

// if ChildHasPriority
	IL_00d7:  ldarg.0 
	IL_00d8:  call instance bool class OpenRA.Activities.Activity::get_ChildHasPriority()
	IL_00dd:  brfalse.s IL_00f9

[anvilvapre]

I have not checked. But to me it seems impossible that IL optimization can optimize out SkipDoneActivities.
The difference between ChildActivity and childActivity. The last probably could have been optimized out.
(Unless properties work different from calling functions.)

[pchote]

https://github.com/OpenRA/OpenRA/wiki/Changelog-(bleed)/_compare/d423a15528b82797d1e1bfa1c4e7bfe4a1ab4107

[dnqbob]

I am going to have some tests on how much this branch can optimize to OpenRA later.

[dnqbob]

Feels better, at least 3 FPS improve, and

20 ms [10048] Activity: Move
20 ms [10048] Activity: AttackMoveActivity

12 ms [14665] Activity: Move
12 ms [14665] Activity: MoveWithinRange
12 ms [14665] Activity: Attack

which may still be some spike lag, but better than before.

shellmap of SP has noticeable performance improvement.