Improve run time of identifySubdomainMesh
The run time was improved due to:
- selection of better (smaller) ratio of number of mesh nodes per grid cell - this leads to lower number of distance computations / comparisons
- pre-computing the connected elements to nodes and use it multiple times
- use unordered_map instead of map
- make OpenMP parallelization available
The times given in the table are without reading / writing the meshes. The first number in the sum is the time for the MeshNodeSearcher construction, the second number is the time for the identification of nodes/elements.
OMP_NUM_THREADS | master | MR |
---|---|---|
1 | 0.50 + 536.76 | 0.78 + 142.58 |
2 | 0.75 + 80.23 | |
4 | 0.74 + 55.36 | |
8 | 0.76 + 43.05 |
-
Feature description was added to the changelog -
Tests covering your feature were added? - identifySubdomains tests passes in CI. -
Any new feature or behaviour change was documented?