Milestone Performance Improvements

Completed 4 weeks ago (Jun 20, 2018, 8:48:11 AM)

There are several simple ways to improve the performance of the more computationally intensive functions in LSL. Some of the ideas are:

  1. Reduce the number of array copies in functions like lsl.correlator.fx.FXMaster()
  2. Change the OpenMP loop structure in lsl.correlator._core so that it performs better with DRX data
  3. Unify the structure of lsl.correlator._core with lsl.correlator._spec

Other things that might change in LSL are:

  1. Switch from the FFTW fftwf_complex type over to the native C type. If this happens it will also hit the pulsar extension.
  2. Decide between using direct memory access for NumPy? data and the various access macros.

Update: Most of this will appear in version 1.2.1 of LSL.

Note: See TracRoadmap for help on using the roadmap.