Skip to content

Benchmark updates

Frederik Hennig requested to merge da15siwa/walberla:benchmark-updates into master

Introduces various fixes and updates to our lbm-generated benchmark apps:

UniformGridCPU benchmark

  • Fix lid driven cavity setup in uniform benchmark
  • Change method identifiers to be compliant with recently published lbmpy paper
  • Uniformly switched to compressible methods

NonUniformGridCPU benchmark

  • Separated grid generation into separate application
  • Added Python callbacks and database output

Separating grid generation from the actual benchmark run turned out to be necessary to avoid the SetupBlockForest to be created on each process. This lead to out-of-memory errors at >384 nodes both on SuperMUC NG and Fugaku. Also, it makes no sense to create the SetupBlockForest redundantly thousands of times.

Merge request reports