Fof

Made a merge request out of this to make commenting easier.

@jwillis, I did a first round of comments, the code looks good!

A few higher-level comments:

• Please run format.sh, it makes it easier to read the code on my small laptop screen :)
• The current implementation does not support symmetry, i.e. each particle pair is inspected twice. This could be fixed by not looking at pairs of root-level cells if cid < cjd.
• Instead of implementing this as tasks (eventually), it may be fast enough to implement it as a mapper.

Cheers, Pedro

added 1 commit

• 51c499e4 - Formatting.

Compare with previous version

added 1 commit

• c41249dd - Removed unused variables and added comments.

Compare with previous version

added 1 commit

• c1dfa0d9 - Calculate the number of particles in each group at the end of the union-find…

Compare with previous version

added 4 commits

• b0179c60 - Split tree search into a self recursion and pair recursion. Fixed a bug where the incorrect
• cd46b57b - Fixed compilation error. Renamed variables and initialised group_sizes with bzero.
• 8c97d3ac - Removed old functions.
• a34ef085 - Renamed functions. Skip recursion of empty cells.

Compare with previous version

added 355 commits

• a34ef085...63ada53a - 353 commits from branch master
• 0e835ccb - Merge branch 'master' into FOF
• 4e7b4d49 - Use size_t in loop.

Compare with previous version

Just gone through some of the updates since last week. Looks good!

You added an escape path for empty cells to make them stop the recursion. That's good. Is it worth doing at other levels in the tree as well?

Thanks! I'm working on another optimisation so that the root of pi is found once in the outer loop and updated on the fly if it changes within the inner loop.

Yeah, I could check the number of particles at every level of recursion, it should improve performance.

Have you tried it on a different box? I am interested in having an idea of the time this takes to run.

Also, have you checked whether you get consistent results if you vary the number of top-level cells?

I tried the exit cell condition on leafs, but it takes longer 100ms longer.

I haven't tried this on any other EAGLE boxes yet, but I also haven't parallelised it yet. Do you want a serial run of a EAGLE_12/25?

I just tried EAGLE_6 with 15x15x15 top-level cells and it produced the same answer and ran faster.