perf: faster relabeling for several variants including SAUF

[william-silversmith]

Hi YACCLAB maintainers!

I was running some experiments on YACCLAB and noticed in profiling that relabeling was contributing more time than I expected for variants that work one-pixel/voxel at a time. It's not a huge amount of time, but it is a constant tax against these variants that makes them look worse than they are and in-between them, softens the contribution of a faster first-pass.

I also fixed some compiler warnings.

I re-ran the 8-connected benchmark to demonstrate that there is a material speedup on these variants. The chip was a Macbook Pro M3. I did run them simultaneously, but I think that should be ok on this machine.

I've attached the results of my runs.

Here are some samples though:

Before

After

I hope you find this a useful contribution.

Sincerely,
Will Silversmith

faster_relabel.zip

master.zip

[prittt]

Dear @william-silversmith,

Thank you for your contribution! Actually, running the tests in parallel could skew the results (yours might not be the case). I will run some more tests on different machines/OS and let you know.

Best regards,
Federico.

[william-silversmith]

Thank you for taking a look Federico!

I did some short experiments that were single core and saw similar numbers which is why I felt more confident running the parallel version. Hoping that holds up in your experiments across architectures.