some ideas

[mattsignorelli]

This is a running list of ideas I have we could discuss in the future:

• Maybe this exists already, but a generic interface for element attributes - for example, if someone wants to add attribute to the dict, one step should be to define a function att_update_rule that tells the bookkeeper when to recalculate that quantity and also how to calculate that quantity. Then all other attributes should satisfy this interface
• multiparticle spin tracking should only track the quaternion for each particle instead of Sx Sy Sz
• low-level track functions should be in-place (e.g track_quadrupole!(z0,z, ...)), allow aliasing with minimum temporaries produced inside the function (the track package API could be very similar to that of the MAD GTPSA C library), and fully compileable
• tracking functions need to be compatible with CuArray type or easily translatable to separate kernels for GPU porting with CUDA. Particle coordinates stored in 1D CuArray
• use PrecompileTools for tracking and acceleratorlattice to kill the JIT time for users, precompile all the tracking routines with Float64, and duals and TPSs
• how to convert the lattice to a new structure for maximum speed in tracking (can it fit on a single register in GPU?)