Currently, a lot of the parallelization in this crate involves mindless rayon spam. However, there might be some significant performance improvements I can make by reducing false sharing via padding/alignment, especially in the atomic neuron value cache.
There may also be some similar improvements to be made with genetic-rs, but that's out of the scope of this issue.
Currently, a lot of the parallelization in this crate involves mindless
rayonspam. However, there might be some significant performance improvements I can make by reducing false sharing via padding/alignment, especially in the atomic neuron value cache.There may also be some similar improvements to be made with
genetic-rs, but that's out of the scope of this issue.