Add multithreading via OpenMP to Linac RF gaps, 3D FFT SC, Twiss calculation and Teapot transforms

[danielnoll]

No description provided.

[woodtp]

Hi @danielnoll, thanks for tackling this. Would you happen to have any performance benchmarks for your additions?

[danielnoll]

Hi @woodtp,

I did a test with our Linac model (European Spallation Source) on a Macbook Pro with an Apple M2 Pro (6 performance cores). The runtime of the model is dominated by the 3d FFT SC (64^3 mesh, 1M particles), which for the FFT is not parallel when running with MPI, and so the usage of the parallel FFTW library improves speed considerably.

Without SC, things look like this (1M particles). My guess at the reason behind the slightly worse scaling of the OpenMP version is that apertures/particle loss is not running in parallel, but I haven't looked into that further.