The primary improvements in GTSAM 3.0 result from a complete rewrite of the core linear and symbolic optimization pipeline. Multithreading using TBB is now mostly ubiquitous, and redesigning algorithms eliminated much bookkeeping, both resulting in significant performance improvements and data structures that are easier to deal with in user code.

Details:

• Complete rewrite of the linear and symbolic solver with the following improvements:
  • Multithreaded from the ground up using Intel TBB.
  • Numerous performance improvements by redesign that eliminates much bookkeeping.
  • Simplification and elimination of bookkeeping makes manipulating and interpreting linearized graphs as easy as nonlinear graphs from user code.
• Computationally-intensive code now parallelized with TBB (in addition to the core solver).
• MKL automatically detected and used for further performance improvements.
• Performance and functionality improvements in ISAM2.