Fix tile iteration bug and speed up min(dij)

[graeme-a-stewart]

Fix a bug in the tile iterator, so that if a tile has no jets (i.e., the first "jet" is not valid) then the iterator returns nothing

Speed-up the search for the best dij value by using the LoopVectorisation package

Some small fixes for comments and additions to .gitignore

[graeme-a-stewart]

Hi @grasph - I tracked down the bug that was causing your implementation to deviate from the FastJet results. If a tile had no valid jets it still returned one value in the iterator of noTiledJet, which sometimes caused a jet with no neighbours to merge with the beam prematurely. This fix costs about 4% runtime.

I have also applied the nice optimisation to the search for the lowest dij value at each iteration, using LoopParallelisation. This saves a lot and the final code is now about x1.45 faster than the FastJet N2Tiled algorithm. (251us/event vs. 173us/event on my MacBook). I was a little concerned that the final dij picked here might have a race condition, but I ran it hundreds of times and it aways gets the correct value.

[graeme-a-stewart]

Just doing some more benchmarks on x86 boxes, the speed up is not quite as high there (tested on one Intel and one AMD), with a 16% gain. Still allows us to beat FastJet, but only by a small margin, x1.06.

[grasph]

Excellent!