Use linear interpolation in DragForcing

[marchdf]

Summary

This is much more efficient than the current implementation since it does a bisection search. Also instead of a nearest neighbor it uses a linear interpolation. One can recover the previous implementation by just using the new nearest_search.

Pull request type

Please check the type of change introduced:

• Bugfix
• Feature
• Code style update (formatting, renaming)
• Refactoring (no functional changes, no api changes)
• Build related changes
• Documentation content changes
• Other (please describe):

Checklist

The following is included:

• new unit-test(s)
• new regression test(s)
• documentation for new capability

This PR was tested by running:

• the unit tests
  • on GPU
  • on CPU
• the regression tests
  • on GPU
  • on CPU

[hgopalan]

Here are some timing tests for a case with ~ 13 million cells on 208 cores

Older nearest search: Time spent in Evolve(): 2914.012737
New linear interpolation: Time spent in Evolve(): 2579.18432

[marchdf]

@hgopalan Thanks for doing that comparison. I thought of a better/less complicated way. Would you mind checking the timing with the latest?

[hgopalan]

@hgopalan Thanks for doing that comparison. I thought of a better/less complicated way. Would you mind checking the timing with the latest?

Here are some timing tests for a case with ~ 13 million cells on 208 cores

Older nearest search: Time spent in Evolve(): 2914.012737
New Old linear interpolation: Time spent in Evolve(): 2579.18432
New linear interpolation: Time spent in Evolve(): 2343.574082

[marchdf]

1.25X speedup. I will take it ;)