KABOOM! in 142 lines of Parallel ML

A port (MPL) of a port (Futhark) of the KABOOM (C++) tiny graphics program.

Here's my current result at 24 fps with MPL. (Also see boom-60fps.gif, but high framerate GIFs don't seem to play very well in the browser for some reason...)

Multicore performance

Machine: 4x Intel 18-core E7-8867 v4 Xeons (144 hyperthreads).

Time to generate one frame:

MPL	Futhark (CPU backend)
0.094	0.212

Specifics for running with MPL:

$ make main.mpl
$ ./main.mpl @mpl procs 144 set-affinity -- -frames 100
...
average time per frame: 0.0943s

And, running @athas's implementation in Futhark:

$ futhark bench --backend=multicore --runs=100 tinykaboom.fut
Compiling tinykaboom.fut...
Reporting average runtime of 100 runs for each dataset.

Results for tinykaboom.fut:
#0 ("640i64 480i64 0.1f32 6.28f32"):     212341μs (RSD: 0.108; min: -13%; max: +88%)