Skip to content

RAP-100 Reduce memory usage scalar solver

Mark de Wever requested to merge RAP-100-reduce-memory-usage-scalar-solver into master

The Scalar solver needs temporary matrixes. It always allocates these matrixes for all channel blocks to process. If less threads are available then channel blocks only n threads of these matrixes are used.

This change does the following:

  • Avoid creating more threads than channel blocks to process.
  • Limit the number of matrixes allocated to the number of threads used.
  • Some minor cleanups.

More information regarding the problem is documented in confluence https://support.astron.nl/confluence/display/SDCP/Increased+memory+usage+in+DP3%27s+DDECal

Based on @offringa's insight this makes similar changes to the DiagonalSolver and FullJonesSolver.

Closes RAP-100

Edited by Mark de Wever

Merge request reports