Add leapfrog pusher for plasma

[AlexanderSinn]

To use the new leapfrog pusher, compile with

-DHiPACE_PUSHER=LEAPFROG

Leapfrog: 40% faster plasma push

--------------------------------------------------------------------------------------------------
Name                                               NCalls  Excl. Min  Excl. Avg  Excl. Max   Max %
--------------------------------------------------------------------------------------------------
AnyDST::Execute()                                   18000      33.76      33.76      33.76  23.54%
DepositCurrent_PlasmaParticleContainer()             6000      26.52      26.52      26.52  18.49%
AdvancePlasmaParticles()                             6000      22.52      22.52      22.52  15.70%
ExplicitDeposition()                                 6000      22.36      22.36      22.36  15.59%
hpmg::MultiGrid::solve1()                            3000      21.49      21.49      21.49  14.98%

AB5:

--------------------------------------------------------------------------------------------------
Name                                               NCalls  Excl. Min  Excl. Avg  Excl. Max   Max %
--------------------------------------------------------------------------------------------------
AnyDST::Execute()                                   18000      33.68      33.68      33.68  21.85%
AdvancePlasmaParticles()                             6000      32.24      32.24      32.24  20.92%
DepositCurrent_PlasmaParticleContainer()             6000      26.65      26.65      26.65  17.29%
ExplicitDeposition()                                 6000      22.33      22.33      22.33  14.49%
hpmg::MultiGrid::solve1()                            3000      21.49      21.49      21.49  13.94%

Leapfrog: 2.5x less memory for plasma

-------------------------------------------------------------------
Name                                      Nalloc  Nfree  Max Memory
-------------------------------------------------------------------
Unprofiled                                     1      1      29 GiB
sortBeamParticlesByBox()                      44     44      11 GiB
BeamParticleContainer::InitParticles          10     10    7390 MiB
PlasmaParticleContainer::InitParticles        34     34    4737 MiB
DenseBins<T>::buildGPU                        36     36    1210 MiB

AB5:

-------------------------------------------------------------------
Name                                      Nalloc  Nfree  Max Memory
-------------------------------------------------------------------
Unprofiled                                     1      1      29 GiB
PlasmaParticleContainer::InitParticles        84     84      11 GiB
sortBeamParticlesByBox()                      44     44      11 GiB
BeamParticleContainer::InitParticles          10     10    7390 MiB
DenseBins<T>::buildGPU                        36     36    1210 MiB

Leapfrog: within ~10^-5 field results

AB5:

Delta:

• Small enough (< few 100s of lines), otherwise it should probably be split into smaller PRs
• Tested (describe the tests in the PR description)
• Runs on GPU (basic: the code compiles and run well with the new module)
• Contains an automated test (checksum and/or comparison with theory)
• Documented: all elements (classes and their members, functions, namespaces, etc.) are documented
• Constified (All that can be const is const)
• Code is clean (no unwanted comments, )
• Style and code conventions are respected at the bottom of https://github.com/Hi-PACE/hipace
• Proper label and GitHub project, if applicable

[AlexanderSinn]

The differences between the new and old AB5 pusher in CI can be explained by:

Ionization: for new electrons the old pusher would immediately push them with force terms equal to zero, while the new pusher would gather force terms first.

Collisions: Because it is more convenient with how temp slices work, with the new pusher collisions are done after the push while the old pusher pushed after collisions.

production.SI: This is a very tiny change in the checksum of fields like Bx, ExmBy, Bz etc. that don’t have a strong structure in on the xz slice, possibly caused by reordering some multiplications and divisions in the pusher.

So they are effectively equivalent. I will now change the default to leapfrog.

[SeverinDiederichs]

Suggestion I'd like: Since it is a compile time option, it would be great if it was printed in the output, similar to other compile time options like the precision. i.e.,

HiPACE++ (v22.12-33-gab7921595939) running in double precision
using CUDA version 11.7.64
using the leapfrog / Adams-Bashforth plasma particle pusher

[MaxThevenet]

Looks great, thanks for this PR! Indeed quite a bit of cleaning. See small comments below.

[MaxThevenet]

Nice, thanks for the comments!

[MaxThevenet]

For the record, could you update the description of your PR to reflect the new name of the compile-time option?

[ax3l]

@AlexanderSinn is it work making this a compile-time option?

I think pre-compiling both options and doing a runtime switch before the two specialized kernels would be more user friendly.

[ax3l]

Discussed: needs additional (25) force terms on the particle, so that's a bit of a heavy lift.

[AlexanderSinn]

As mentioned, it is a compile-time option not for performance of the push kernel, but rather to save on memory of the 25 Adams-Bashforth force terms (per plasma particle) if the leapfrog pusher is used. One would need to change these to runtime components to make this a runtime option.