Add benchmarks

[efaulhaber]

Based on #15.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 89.86%. Comparing base (f4109dd) to head (eb64d9c).

Additional details and impacted files

@@           Coverage Diff           @@
##             main      #18   +/-   ##
=======================================
  Coverage   89.86%   89.86%           
=======================================
  Files          13       13           
  Lines         306      306           
=======================================
  Hits          275      275           
  Misses         31       31           

Flag	Coverage Δ
unit	89.86% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[svchb]

What is the intention here for actually showing the benchmark results?
Do you want to add them to a gh page later?

The current output:

TrivialNeighborhoodSearch
with 50 = 50 particles finished in 2.235 μs

GridNeighborhoodSearch
with 50 = 50 particles finished in 1.041 μs

ProcomputedNeighborhoodSearch
with 50 = 50 particles finished in 105.863 ns
... 

needs to be more structured its unclear what the differences are between for example the second output:

TrivialNeighborhoodSearch
with 50 = 50 particles finished in 3.202 μs

It would also be nice to have a benchmark just for the initialization of the methods to cover PrecomputedNeighborhoodSearch.

[efaulhaber]

What is the intention here for actually showing the benchmark results?
Do you want to add them to a gh page later?

Yes, the plots are supposed to go in the docs and/or README. Especially with #8, we can also very nicely show the difference to other Julia packages. And as opposed to other packages, I wanted to make our benchmark code easily available, so that one only has to run a single command to produce these plots.

The current output:

TrivialNeighborhoodSearch
with 50 = 50 particles finished in 2.235 μs

GridNeighborhoodSearch
with 50 = 50 particles finished in 1.041 μs

ProcomputedNeighborhoodSearch
with 50 = 50 particles finished in 105.863 ns
...

needs to be more structured its unclear what the differences are between for example the second output:

TrivialNeighborhoodSearch
with 50 = 50 particles finished in 3.202 μs

I don't understand what you mean. The output very clearly shows the timings of the different implementations at different problem sizes:

julia> plot_benchmarks(benchmark_count_neighbors, (100, 100), 2)
TrivialNeighborhoodSearch
with 100x100 = 10000 particles finished in 8.618 ms

GridNeighborhoodSearch
with 100x100 = 10000 particles finished in 769.375 μs

PrecomputedNeighborhoodSearch
with 100x100 = 10000 particles finished in 53.708 μs

TrivialNeighborhoodSearch
with 200x200 = 40000 particles finished in 136.701 ms

GridNeighborhoodSearch
with 200x200 = 40000 particles finished in 3.472 ms

PrecomputedNeighborhoodSearch
with 200x200 = 40000 particles finished in 215.458 μs

It would also be nice to have a benchmark just for the initialization of the methods to cover PrecomputedNeighborhoodSearch.

My future plans for the benchmarks are:

1. Add real-life SPH benchmarks calling interact! for WCSPH and TLSPH (Add real-life SPH benchmarks #29).
2. Add an artificial benchmark for update! by updating the same neighborhood search, alternating between two slightly different coordinate arrays (point_cloud with different seeds).
3. Add a real-life benchmark of a full WCSPH simulation (combining update and query performance).