Skip to content

TVF - Process chunks in a more convenient way to avoid clone checking.

Daniel Campora Perez requested to merge kalman_optimizedChunks into 2017-patches

Gain in [1-5]% speed for TVF depending on dataset.

  • Compared to TMF (approx values): magup2016: -18%, upgrade-baseline: -2%.
  • to_process is populated with chunks of size BatchSize.
  • BatchSize is defaulted to 50 after parameter exploration.
  • Population of nodes in TVF casts smooth state vector only once (very small gain).

Merge request reports