Vectorization Calculations for Fast CFD Computation Times

I developed an innovative vectorization algorithm for my CFD-FSI simulations to speed up intra-particle interaction computations. This vectorization algorithm was benchmarked against the following (see graphic):

1. Simulations without intra-particle interaction computations (with time-steps of \(\Delta t = 10^{-3}\)); listed as “Local:Optimized”

2. Simulations without intra-particle interaction computations (with time-steps of \(\Delta t = 10^{-4}\)); listed as “Local:Unoptimized”

3. Simulations with intra-particle interaction computations (with time-steps of \(\Delta t = 10^{-4}\)) using nested for-loops; listed as “Non-Local:Non-Vectorized”

4. Simulations with intra-particle interaction computations (with time-steps of \(\Delta t = 10^{-4}\)) using nested back-end C/C++ compiler for the nested for-loops; listed as “Non-Local:Numba Optimized”

5. Simulations with intra-particle interaction computations (with time-steps of \(\Delta t = 10^{-4}\)) using full vectorization of these computations; listed as “Non-Local:Vectorized”

Notes

The benchmarks displayed here are for single-particle simulations. For multi-particle calculations, the vectorization boost in decreased computation times is expected to be approximately the same amount for single-particle calculations or even greater.

Link to GitHub Repo will be created soon!