SIMD based hull collision inspired by example code from @cairnc in #54
<img width="1676" height="1332" alt="image"
src="https://github.com/user-attachments/assets/58b137bd-bf7b-4c99-863c-64341b973a3d"
/>
- Fixed#78
- Fixed#85 and fixed CI to error on UB.
- Updated benchmarks
- Added AMD scalar benchmarks
- Added Neon benchmarks
- More determinism unit tests
- Sample leak detection
- Added b3Body_AllowFastRotation
- Added b3Body_IsFastRotationAllowed
- Bumped b3HullData version
- Bumped b3CompoundData version
Removed B3_BODY_NAME_LENGTH and B3_SHAPE_NAME_LENGTH.
Now using a name cache instead to save memory and make recordings more
universal.
This is a recording version bump.
Added a compile time option to use double precision body positions. Most
math remains single precision with minimal performance loss.
`BOX3D_DOUBLE_PRECISION`
Improved accuracy of body inertia computation when shapes are far from
the body origin.
WARNING: several APIs changed but migration should be straight forward
if doubles are not enabled.
Save memory when there are many duplicate hulls. Can improve performance
in some cases.
For consistency with b3MeshData, b3Hull is renamed to b3HullData.
- Fixed mesh point culling for vertically stacked points
- Fixed two point capsule vs capsule normal
- Fixed bug where body flags can go out of sync
- Added tests
- Updated benchmarks
- API to query compounds directly.
- Fix some uninitialized stuff (thanks MSAN).
- Fix for store impulses on overflow.
- AABB margin reduced for small shapes
- body contact recycling option
- angular threshold for contact recycling
- reduce AABB margin on smaller shapes
- fix for hull vs tri when axis aligned
- fully locked rotation uses zero inverse inertia