Duplicate results

[BoySanic]

For a long time now, seeds that are output from this program always print out twice.

The most I could figure out is that the duplication must be happening in the GPU code, or maybe on the host thread running the GPU code. Because by the time it gets to the cpu filters where it will eventually print out the result, it's only going to print once per result ultimately, so it must have been given 2 of that result.

I have no basis for thinking this is true but I hope it's not the case that we're doing 2x the work all the time as evidenced by the duplicate results.

[mrzetti]

I also don't think this means we scan seed ranges twice. SeedIterator::next() uses an atomic fetch_add, so GPU workers claim disjoint ranges:
https://github.com/MinecraftAtHome/COMMISSION/blob/master/src/gpu.h#L5-L14

Each GPU worker loop claims one range here:
https://github.com/MinecraftAtHome/COMMISSION/blob/master/src/gpu.cu#L1763-L1765

The duplicate appears later in the candidate-position pipeline. Final GPU candidates are copied and all pushed into the CPU queue without a uniqueness check:
https://github.com/MinecraftAtHome/COMMISSION/blob/master/src/gpu.cu#L1839-L1853

The later filter stages emit candidate positions derived from earlier parent candidates:
https://github.com/MinecraftAtHome/COMMISSION/blob/master/src/gpu.cu#L1213-L1218

Because those parent candidate search windows can overlap, two GPU records can represent the same seed/area candidate! Then CPU verification canonicalizes the input position to the actual cubiomes island center/area:
https://github.com/MinecraftAtHome/COMMISSION/blob/master/src/cpu.cpp#L25-L32
https://github.com/MinecraftAtHome/COMMISSION/blob/master/src/cubiomes.c#L264-L280

So two different GPU candidate records can collapse to the same final printed {seed, x, z, area}.

Disclaimer, gpt-5.5 helped me find this, hope it helps, i am no cuda expert!

[Likqez]

always print out twice.

I just wanna add that i saw a 3x output for the same seed.