Remove duplicated work in ILS/ICM encoding
una-dinosauria opened this issue · comments
This is somewhat easy when there is only one CPU doing the work. When there are multiple CPUs working, they all have to do all the ILS iterations separately, which makes it hard(er) to keep track of progress per ILS iteration. I think I'll focus on #3 -- LSQ on the GPU first.
If we use the c++ implementation, which is ~25% faster, we can simply use OMP. I'm going with this by default.