593 Commits over 426 Days - 0.06cph!
merge from indirect_instancing
Keep Indirect Instancing disabled by default (until rare wrong object bug is fixed)
* Improve job scheduling to make leaving the job phase more robust through loading screens
* Fix NRE when trying to remove instances before Indirect Instancing was initialised
* Made switching Indirect Instancing at runtime more robust
* Prevent some other possible exceptions
* Avoid some unnecessary calls to MotionList
* Renamed TransformMemory to MotionList (as it should be)
* Use managed HashSet instead of NativeHashSet because there seems to be no performance benefit to using native at the moment.
* Fixed exception caused by removing elements from the set while iterating
Use negative handles for fallback memory
Remove InstanceHandle and use a simple integer instead
* Added generic NativeMultiList to replace InstanceHandleList (WIP)
* Store Transform.InstanceIDs (and actual nullable Transform references) in PerInstanceMemory, indexed by seq_num
* Simplified MotionList add/removal procedures to HashSet operations
* Cleared up function naming in BaseEntity.Client.Motion.cs
* Consider all entities as a potential motion parent and utilise existing parent/child relations to call MotionStart / MotionStop
* Removed IMotionParent interface because it caused more confusion than anything else
* Correctly dispose of the hash set storing child instances when an entity is destroyed
(WIP) Large refactor: Make BaseEntity responsible for calling StartMotion / StopMotion with (potentially) a single array of numbers to mitigate spikes when large entities set into motion.
* Make instances identifiable by a single never changing integer
* Speed up adding/removing instances by using various look up tables to find instances instead of staged linear searching
Convert seq_num uint → int
Enable Indirect Instancing by default
No longer restrict Indirect Instancing (`indirect_instancing.enabled`) to developer accounts. I'm looking at you, staging players. 👀
Allow Indirect Instancing to be toggled in-game any time (Also correctly applies config value when in editor now)
* Remove the need for HasValidMesh() on LOD components
* Renamed a bunch of things for clarity
Disable Indirect Instancing by default to unblock editor issues (I'll be back)
Disable Indirect Instancing by default to unblock editor issues (I'll be back)
merge from indirect_instancing (for real this time)
merge from indirect_instancing
* Moved fallback bounds transformation to a job
* Removed old fallback culling code
Cleanup: Remove old MotionList code
* Fixed objects in fallback renderer not updating correctly when in motion.
* Added profiler markers for draw call submission of instanced and fallback renderers.
* Fixed streaming_use_mesh_uv_distribution_metrics not actually doing anything.
Fix 2 NREs with Indirect Instancing enabled
* Convert INDIRECT_INSTANCING_ON to a global keyword because it kind of is global
* Make sure INDIRECT_INSTANCING_ON is always defined in the GCINCLUDE section (Fixes shipping containers)
* Allow changing indirect_instancing.enabled at any time in the Editor (FIxes entering play mode with it enabled)
Use LOD quality setting "Object Quality" in Indirect Instancing to more closely resemble the LOD switching behavior of the old system.
Try skipping the extra pass to remove null transforms for an extra 0.2 ms (toggleable via ConVar)
Perform application of the motion list transforms in a job for better performance
Merge from indirect_instancing (also force-disables it for non developers)
Set new _INDIRECT_INSTANCING_ON keyword when Indirect Instancing is enabled
Mark all Indirect Instancing ConVars as DeveloperOnly.
Don't include DeveloperOnly commands in autocomplete.
Merge from main (to pull in mac_foliage_tearing_fix)
* Fix randomly disappearing in-motion objects since switching to the new TransformMemory implementation.
* Fix randomly disappearing stationary and in-motion objects in the deep sea.
* Fix concurrent access exception when in-motion objects are destroyed.
add some more debug info to TransformMemory
Improve performance of instances in motion by about 3x by mirroring all swap and copy operations of instance memory into a separate structure, and using that to apply new transforms of instances in motion.
Enable 222 new materials for GPU instancing (If you conflict with this changeset, simply override these changes and manualle re-enable GPU instancing on the affected materials afterwards)
Enable 222 new materials for GPU instancing (If you conflict with this changeset, simply override these changes and manualle re-enable GPU instancing on the affected materials afterwards) [Part 7 of 7, because Plastic is a broken mess]
Enable 222 new materials for GPU instancing (If you conflict with this changeset, simply override these changes and manualle re-enable GPU instancing on the affected materials afterwards) [Part 6 of X, because Plastic is a broken mess]
Enable 222 new materials for GPU instancing (If you conflict with this changeset, simply override these changes and manualle re-enable GPU instancing on the affected materials afterwards) [Part 4 of X, because Plastic is a broken mess]
Enable 222 new materials for GPU instancing (If you conflict with this changeset, simply override these changes and manualle re-enable GPU instancing on the affected materials afterwards) [Part 5 of X, because Plastic is a broken mess]
Enable 222 new materials for GPU instancing (If you conflict with this changeset, simply override these changes and manualle re-enable GPU instancing on the affected materials afterwards) [Part 3 of X, because Plastic is a broken mess]
