- Add `qH` and `qHDiagInv` to `mjData` to save factorized modified inertia.
- Add `mj_EulerSkip`, `mj_implicitSkip`, to `engine_forward.c`.
- Using the above functions, implement `mj_stepSkip` in `engine_derivative.c`.
- Add `mjd_stepFD` and `mjd_transitionFD` to `engine_derivative.c` to compute `mj_step` Jacobians.
- Exploit "Skip" functionality for speed.
- Correctly handle quaternion derivatives.
- Handle warmstarts and control limits.
PiperOrigin-RevId: 456584811
Change-Id: Iee8541f11e7b66feb8f431cb102d9bbe65461f79
The introduction of additional mjData arrays for the implicit integrator caused in a shift in data alignment in memory. In certain combinations of machines and models, this change resulted in ~5% performance regression that appears to be attributable to L2 cache misses.
This change aligns all arrays in mjData to the boundary of a typical L2 cache line size. Benchmarking shows that this resolves the performance regression on a Broadwell machine.
PiperOrigin-RevId: 451652783
Change-Id: I4ed2c2d03453a83a0b4a9b8e224e2fd7c0dadea9
Yuval Tassa