branchrust_reboot/main/erosioncancel
33 Commits over 31 Days - 0.04cph!
adjusted splat ignoremask to include ocean and lake (using pre-existing Sand topology mask)
re-balanced erosion splat values to be less intense, included oceanside and lakeside to ignore mask to avoid wiping out beaches with stones
switched splat writing to iterate heightmap delta and write splat with normalized coords
- much better results, but needs jobifying as it's much slower
erosion texture fixes
- use proper splat types
- rebalanced splat values to use exponents
- don't splat monuments
added splatmap writing to the end of the grid-based erosion, re-used approach from old particle-based solution but with re-tuned values
- also rebalanced values again
remove unused watermap array from job
switched initial map prepare job to IJobParallelForBatch to reduce locking with parallel writer (temp alloc list of indices and add to write as batch per-job)
- also other various micro-optim
erosion tuning and fixes
- tweaked values again, bit more prominent
- splitting slope calc into its own job with tiling for better caching
- corrected error with flux calc where water was tending to flow in one direction
erosion adjustments
- fixed evaporation rate being stupidly high
- cache ocean indices on the start and refill the ocean each step, stops it seeping out the world border
- smoothstep sediment adjustment with water depth and threshold to stop erosion from high velocity deep water
- properly adjust water height by sediment adjustments
adjusted debug_erosion data dump to have resolution size in file name and also reference the data vis repo
more tuning and correcting velocity adjustment
removed unnecessary NoAlias attribute
fixes and tuning improvements
- corrected indexing mistakes that caused water to flow generally eastwards and slope angle to just be wrong
- water height threshold to stop ocean eroding seafloor
- sediment depositing uses timestep to help tuning be more consistent
- cleanup
some aggressive optimizations to vectorize better, mostly switching out repeated inline coordinate->index operations with single int2x4 operation - makes a difference with the loop count we're looking at
grid-based erosion improvements
- much smaller timestep, stops patterns from shoddy integration forming
- working with height in world-space and copying back
- better job dependency handling
- some disabled debug data dumping to help debug the process with python scripts
initial implementation of a grid-based hydraulic erosion system in the jobsystem
- super fast
- super wrong
- needs a lot more parameter/constant tuning to get good results, but it's functioning
small improvements/fixes to particle erosion setup
- starting to look like a bad approach for us though
pretty dirty WIP approach to erosion in a job, needs cleanup and splat writing fixing
terrain topology radius samples run through burst job, even when called from other threads
- 5000 size map without monuments, trains, roads takes 96s rather than 510s to generate, significant reduction in river layout and shore vector generation
Update: added GetTopologies jobs for TopologyMap
- Also covered with tests
- Sprinkled [WriteOnly]/[ReadOnly] attributes into jobs
- Fixed constant string allocation in tests
Tests: ran unit tests
Update: adding WaterMap height sampling burst jobs
- Also covered them with tests to validate output against non-job calls
Tests: ran unit tests
Merge: from terrainmap_nativearray2
- Fixed a bug of invalid reinterpret
Bringing across my tests since we're both working on the same thing.
Tests: ran the new tests
TerrainMap uses NativeArray instead of managed
