8,084 Commits over 1,004 Days - 0.34cph!
Reduced the amount of pooled ballista projectiles 100 -> 3
Reviewed the amount of bullet impacts and decals we prewarm
- Rock, gravel, stones, metal ore and tundra impacts were only warmed with 1 copy, now 25 like the others.
- Glass impact 25 -> 32, it stays alive for 4s+
- Bullet decals 25 -> 30, forest was set to 1, cactus lowered to 5
Prewarm 10 copies of each sound templates, they were all set to 1
small-sound gets 50
Trying to get to a point where I can empty an AK magazine without instantiating any prefab
Fixed all the references to the default unity Quad in particles not actually using it
Any of these particle being set active in builds could spike up to 7ms, caused by ParticleSystemRenderer::UpdateCachedMesh walking the Quad whole user list to check whether its already registered
It was referenced by thousands of renderers in monument prefabs, this is why the AK muzzle was slow to spawn in
167737
Stop HideIfAiming deactivating the muzzle flash once its particles finish. Every reused attack effect had to reactivate it
Was spiking up to ~5-7ms in a build and only 0.2ms in editor somehow, now 0
It points at ParticleSystemRenderer::UpdateCachedMesh, checking that next
Keep sound template AudioSources enabled while pooled, disabling them on every recycle flushed unity's audio DSP, stalling up to ~1ms when a sound got cut mid play
RecycleSound in editor:
~0.45ms -> ~0.25ms
The occasional 1ms "flush dsp connections" is gone
Fixed quick craft spending ~1.6ms rebuilding after almost every shot (inventory update)
Only rebuilds when the inventory is open now
Pool 3 copies of thrown melee weapon and tool projectiles instead of 100. No point in filling our pool with 3900 copies you'll rarely see more than 3 of at the same time
Saves around 50MB of memory and gives me some leeway to eat that with gun effects next
Prewarm 5 copies of each ricochet effect instead of 1, first magazine dumped after connecting created up to 2 more per prefab at ~0.5ms each
Create the pools LaunchProjectileClientside uses at bootstrap instead of on the first shot, avoids a ~365KB alloc when the first projectile of the session is fired
Compile fix, missed a file..
Fixed
150531 not working in builds (Mega hack to bypass the PlayerUpdateCanvases call when unity renders the realtime cubemap, saving some time wasted on UI for nothing)
merge from cubemap_ui_optims2
merge from soundocclusion_optims
Revert ProjectSettings.asset changes
Fixed sounds started by destroyed monument client entities when loading not being cleared correctly
Sound occlusion optims:
- stop checking every wall between a sound and the camera once one wall muffles it
- looping sounds occlusion rays are batched
- virtualised sounds (sounds muted by unity) check occlusion every 1s instead of 0.25s
Standing outside an average base with 40 looping sounds in range:
~0.65ms -> ~0.05 ms per frame
merge from railcolliders_optim
Rail segments now cook their collider mesh in a job instead of on the main thread
5.2ms -> 0.03ms on the main thread per segment
merge from buildingprivilege_cache
Fixed regression with player boat privs when using the cache
Cleaned up building priv cache vs vis tests
merge from buildingprivilege_cache
Reordered the check so bases with no TC are cheaper
Convared the vehicle and entity privilege changes under buildingmanager.privilegecache
Legacy shelters, vehicles and boat building stations no longer use Vis queries
Same base same spot:
GetVehicleBuildingPrivilege 0.34ms -> 0.01ms
GetEntityBuildingPrivilege 0.84ms -> ~0
Sort the cells oldest first
Store each blocks bound centre and net id in the grid so:
- blocks out of the query radius are skipped without a dictionary lookup
- the age check doesnt read every block from memory
Only run HasColliderWithinSphere when the block bounds aren't entirely inside the search sphere
Around 0.09ms for GetBuildingPrivilege in that same base, same spot
- Skip the bounds check for blocks that aren't older than the current best block
- Visit the cell you're standing in first
Skips a lot of work when standing in the core or middle of a base
Attempt at global building privilege cache, using a simple grid holding all the building blocks + a collider test on the winning block, because triangles and roofs have bounds larger than their colliders
Thats enough to remove the Vis Entities call and keep the same precision. Works on the client and the server
In a big base, 575 blocks around me:
4.60ms -> ~0.50ms
Added alien costume to the skin viewer
merge from Halloween2026DLC
▉▍▍█▍▍ ▊▉▊▍▄▋ ▋▍▌▍▉▌ ▄▍ ▉▄▄ ▅▅▌▉ ▌▉▉▋█▊
▇▉▇▋▇▄ ▄█▋▍ ▆█▆▋▆▍▅▌▌▌█▅▋▇▇▊▊▌
Building privilege debug draw convar
▅▉▌▉▄▌ ▌▍▊▉█▉█ ▅█▊ ▍▄▄█▇▊▍▅▊▆▅ ▅▌▅▅▄▅▌▋ ▌▋▆▋▇▊▊█▆ ▊▇▇▌▍ ▊▋▉▋▇▊ █▉ ▋▍ ▋▋▋▄█▇█▇▉█▊▌▊▆▊▋ ▋▆ ▋▉▆▌ ▆▅▅ ▇▉▆ ▋▆▆▇▄▇▍ ▊▍▉▍ ▌▅▍▋▇▇▌▅ ▆▅▊▍▆ ▋▆▊▌█▉▍ ▄ ▌▆▉▇▆▇ ▋▇▇▄▌▅▆ ▌ ▇▇▄▍▋▇▆▆▄ ▅▆▍▉ ▌▊ ▄▆▇ ▊▍▄▍ ██ ▅▄▄ ▋▍▅▌▌▄▌▌▇▋▉▋ ▉▋▉▅▇▅▆ ▋▍ ▊▆▌██▆▌ █▌ ▊▋ ▋▆▉▇▍▇█▌▊▍ ▇▆▊▉▅▅█▋▄ ▆▌▆▄▄▋▆▇ ▇▌▄█ █▅ ▋▄▆▆ ▆▅▄▋█▅ ▄▇▍ ▅▍▇▅█▋ ▌█▍█▆█ ▉▊▄ ▄▄▉█▌▅▍█▌▌▆▉▍ ▉█▄▄▊▋▅▇ ▄▊ ▊▋▊▊▌▊▄ ▋▊ █▄▊ ▇▉▉█▋▅█ █▆▅▅▌▌█▇▆▋
▅█▅▄▅ ▊▉▅▅ ▊▌▍ ▉█▉ ▉▍▊▌▊▉▅█▇▆▆▅▌█ ▋▅▊▄ ▍▌▄▋▆ ▌▆▅▄▊█▌▄▍ ▇▊▍▉▌▊ █▆ ▍▄▌▌ ▌▉ ▊▊█ █▅▌▊▊▉ ▌▊▌▉██▉ ▍█ ▅▇▋▆▉ ▅▇▍▊ ▋▍ ▄▌ █▍▋█▇▅ ▄▋▅▇▋
▄▌▋▊▆▍▊▅ ▍▌ ▇▇▅█▅ ▇▅█ ▋▌▆▊▅▊▄▄█▊▍▅█▅▆▌▅▍▄▍▉▅▊██▆▌▆ ▆▇█▊▇▍▆▄ ▌█▋▅▊▆█▄ ▆▌█ ▆▅▆ ▆▆▆ ▆▍▊▆▋ ▅▌▋▋ ▋▄ ▅▆▌ ▉▌▆▄ ▋▅▍▇ ▆▊▉▆ ▊▋ ▉▅▄▄▆ █▅▍▉▋▄▆ ▌▇▌▉█ ▌▍▄▍ ▇▌▆█▄▇▌▊▄▊ ▉▉ ▄ ▅▉▊▄▍▌▊▉ ▆█ ▍▉▇▉█▋ ▆▋ ▇▊▌▆▆▋▉▆ ▄▆▆▉▄▉▋▌▅▆▌█ ▍ ▄▇▇ ▄▇▆▍▋ ▋ █▍▌▍▅▌ ▆▍▇▍█▋▋ █ ▆▋██▋ ██▅▌▌▌ ▆▆█ ▆▌▉▊▉▇▌▇▋ ▅▇▌ ▉▆▍▍▍▆ ▇▌▅▇ ▍▌▆ ▆▇▆▄▄▅▍▊▋▋█▇▋▅█▊▍▍▆▄ ▍▍▆▆▍ ▌▊▌█▇ ▍▌▋█▇ ▄▉▆ ▋▄█▆▌ ▉█▉▆ ▄▍▆ ▆▇▋▋ ▆▇▅▄▅ ▆█ ▄▅█▌
merge from automated_testing
Building privilege notices client tests
merge from buildingpriv_vitals_optims
Renamed the invoke UpdateLighting -> UpdateRgbLighting, was confusing when looking at the profiler
RgbEffects no longer looks up building privilege when no SteelSeries or Razer device is connected
Its now also using the ignorePosition cache lookup, so it no longer fires 10 times a second while moving
Fixed vitals calling GetBuildingPrivilege ~24 times a second when walking around in a base
Added an ignorePosition param for Privilege.IsCacheValid, now used by the client vitals
Was taking up the 6.5ms a frame in a big base, now back to one call per second
Stop LOD components registering with occlusion culling while its off
Occlusion culling never runs under RRP, but moving LODs still pushed their bounds every frame into a list that never got cleared
Was ~0.5ms a frame on dynamic occludee for nothing too on average
