check for existing data before radiation_map_setup

[electronsandstuff]

This is related to the performance issue in #523. I think I was able to solve it by putting a check before where the radiation_map_setup subroutine is called.

Here is the performance on the example in the issue before the change.

Benchmark 1: tao -noplot -command exit
  Time (mean ± σ):      2.359 s ±  0.019 s    [User: 2.332 s, System: 0.024 s]
  Range (min … max):    2.336 s …  2.395 s    10 runs

Here is the performance with the checks.

Benchmark 1: tao -noplot -command exit
  Time (mean ± σ):      1.746 s ±  0.008 s    [User: 1.720 s, System: 0.023 s]
  Range (min … max):    1.736 s …  1.758 s    10 runs

It gives about a 25% reduction in running time for this example. Profiling it after the change confirms that the radiation_map_setup code takes an insignificant amount of time.

[ChristopherMayes]

track1_radiation is called by track1. With OpenMP, this is within a parallelized loop:

bmad-ecosystem/bmad/multiparticle/beam_utils.f90

Line 69 in 24282a4

!$OMP parallel do if (thread_safe)

@DavidSagan won't this cause a problem, modifying ele_struct by separate threads?

[DavidSagan]

@ChristopherMayes Yes it looks like there is an OpenMP problem here. I will take a look. Also the new test as constructed is invalid since Fortran does not have short-circuiting so testing associated(ele%rad_map) and ele%rad_map%stale in the same if statement is illegal.

[DavidSagan]

I modified the code to avoid the OpenMP problem and added code so that the radiation integral values would be reused if possible. This decreased the run time for the test program by about 15%. (Devel/step17 #534)

[DavidSagan]

With the Bmad code change mentioned above the difference with and without the PR code is about 5%. The next PR from me will contain this PR code modified to be valid.