Fast 6x12 Connected Components using bit-optimized BFS

emh
1,446 views

Open Source Your Knowledge, Become a Contributor

Technology knowledge has to be shared and made accessible for free. Join the movement.

Create Content

Fast 6x12 Connected Components using bit-optimized BFS

This is pretty standard BFS, just bit-optimized and with a lookup table for neighbours. Performance varies a lot depending on how densely populated the bitboards are, and that in turn depends a lot on the random seed they are initialized with. But you can expect 2-5 millions of iterations per 100ms, suitable for Smash the Code. Performance seems to be about same with and without global target set to sse+avx, but I do have an avx2 target locally around popcount code which improves performance by about 50%. For historical reasons I had some code that didn't optimize well with AVX on older gcc than 12, that's why I chose to enable it only for critical sections by default. Global default is without avx target. Enough talk, here's to the code:

Open Source Your Knowledge: become a Contributor and help others learn. Create New Content