811 Commits over 395 Days - 0.09cph!
clarified water trace setup
Gather trace commands in a job, requires copying some data into a native container first but it's still twice as fast overall
Test: projectile tests all still passing
clear benchmark timers after recording results, stops accumulating time between runs
jobified transform readback for effect updating, still slow to update scaled renderers though
pooling HitInfo for Projectile DoHit and don't string concat for every ricochet fx fire
moved most of HitTest generation to a transform job
further batched projectile improvemenets
- jobified position/rotation writeback, 10x faster
- adjusted HitTest to only populate some fields once for a projectile's lifetime (based on flag reset when it is pooled), some improvements
Tests: all projectile tests passing
reworked PlayerCache into a generic StableObjectCache to re-use it for batched Projectiles, making use of it for reading batches of projectile transform data which speeds it up by roughly 8x-10x compared to going through managed properties (cached or otherwise)
Tests: re-ran all projectile and PlayerCache tests, adjusted to new interface
added batched water tracing and various supporting features
- GamePhysics.UseBatchedWaterRays convar to disable for perf comparison, will remove eventually
- NativeArray/NativeList expansion can optionally use power of two expansion
- Added native-friendly query structures for TerrainHeightMap and ShoreVectors, can add more querying functionality as needed but keeping it lean for now
- Modified test cases to be more lenient on float comparisons as Burst compilation introduces variance of around 1e-8 in position results on waves which is more than acceptable
fixed potential break in TraceWaterRays/TraceWaterSpheres logic that would cause casts to miss water traces if an earlier one had hit the max number of colliders
sharing more data between batched projectile effect processing - only accessing transform pos/rot once per projectile and only checking retrieving local player entity data once per batch
added Projectile.UseBatched ClientVar to switch between serial and batched tick for projectiles
Jobified trace-testing in HitboxSystem, also replacing local NativeArray helper function with existing Expand static extension
corrected InitializeVelocity on projectile to keep Projectile performance test consistent between runs (increased run-count)
Optim: reduced unnecessary Transform reads in batched position/rotation update, also added AddRange(BufferList<>) function to BufferList to change to that for marginally faster loop speed
Test: consistency tests passing with serial using old method still
Added extra sanity checks to projectile consistency testing
added missing effects update on batched projectiles and retiring into the prefab pool
clear static test field in pet-test teardown
▋▄▍▄▉ ▆▉▉▌▆ █▋▇▆▇▋▄▇▅ ▌▅▌▌▌▊▄▇ ▅▋▄▄ ▆▋▍▇▌▊ ▋▌▅▍▍▋▍▅ ██ ▋▍▋▍▌ ▄▊▍, ▍▆ ▅██▌ ▇▋ ▌▅▆▍█-▍▉▅▆█▄▍ ▅▄▌▋▅▉▉▆▄ ▄▅▆ ▄▍ ▇▉ █▇▇ ▆▉ ▅▍▌▌█▄
- ▌█▉▊█▍▋▋▌█ █▆▍▆▅▋▋▄ ▍▌ █▉▍▊ ▍ ▄▋▇▄▅ (▄▌ ▋█▋▋ ██▋▆▇ ▆▉▅▄▄█▍ ▇█) ▅▋▍▉ ▊▇█ ▍▋▇▆█-▋▋█▊▉█▉▇ ▆▇▆▊▉▆▄ ▋▉▌ ▊▌ ▍▊▍▆▍ ▅▋▇▇▉▊▇ ▄▌ ▉▋▋ █▉▍▇▍▆ ▌▉▆▍█▇▊▇▌▋▇ █▄▋ ▅▊▇▊▆ ▄▊ ▊▌ ▊▊▋▆▄▅ ▇▊█▄█▄▋ ▊▆▇▄▊<->▅▋▊▊█ █▍▉▋▅▆▅▌▉
cached native allocations, batched projectiles now save between 30% - 50% of the runtime compared to serialized in performance tests (gains increase with projectile count)
multiple significant fixes to GameTrace, new Projectile and GameTrace tests are passing but seeing a slight differences in projectile counts with performance test that's indicating some untested edgecases
initial pass on batched projectiles, functional except for LOS checks but failing consistency tests which needs more diagnosing
- added batch support to multiple tests for comparison
reconfigured fluorescent ceiling light physics configuration
- using configurable joints for better motion, small linear limit for better motion and positional drive on the Y to stop it flying through the ceiling
- added toggle to reduce sleep threshold on CeilingLights
- needs a line renderer setup to replace mesh cables
improved topology-based max carve height to produce a better gradient
tapering MaxCarveDelta off using multiple topology queries of different radii
adjust sediment and waterheight values with carve limits so water and sediment are also corrected
fixing erosion issues waterside and stripe artefacts
- corrected slope angle calculation that was causing striped carving at low
- introduced procedural carving limit that eliminates waterside carving and just limits max-depth elsewhere
moved some TerrainTopologyMap job utility methods into own class and also promoted local function to a static method, burst seems to hate local functions in this case
merge from erosion (burst aot fixes)
removed uneeded burst compatibility test attributes from methods, caused the owning class to be tested for burst compatibility
fixed editor-only memory leak in shorevectors that would happen with domain reload
moved heightmap carving part of erosion significantly sooner in the worldgen, post canyons/lakes but pre-rivers
- simplest way to avoid affecting roads/rivers/etc too much was to just do it first and let them override the results
- splatting still happens after all this, uses the original heightDelta for strength but the current slope delta for tapering