Optimize computation in iaf_psc_alpha_ps model

[heshpdx]

Replace FP divides with FP multiplies and pre-compute loop invariants. Gives about +10% performance. Similar to #3304.

[heplesser]

@heshpdx Thank you for this improvement, a great gain for little work. Your PR also made me realize that I really need to post documentation on NEST model structure. Briefly P_ contains parameters that are user-visible and kept constant during simulation. S_ are state variables, user-visible and evolving during simulation, B_ (buffers) and V_ (variables) are implementation details. The difference between B_ and V_ is that B_ is initialized upon the first Simulate() call (specifically, by the Prepare() step of Simulate()) and persists on subsequent Simulate() calls, while V_ is re-initialized on every Simulate()/Prepare() call. B_ contains typically buffers for input spikes and ODE solver objects, while V_ contains pre-computed quantities. Initialization functions are init_buffers_() for B_ and pre_run_hook() for V_.

So to make a long story short, the new inv_* members you introduced should go into V_ instead of P_ and the code for computing them into pre_run_hook().

[heplesser]

@heshpdx Thanks, all good now! One CI runner failed because a dependency could not be downloaded (SIONLIB). I have restarted the runner now.

[heshpdx]

@clinssen anything left for me to do here?

[clinssen]

Hi, thanks for the updates, this looks great and I'm giving it my thumbs up!