Open stefan-k opened 8 months ago
I think this is also one of the sticking points for no_std
support, which would be an interesting feature (in particular, to use argmin inside GPU kernels).
I think the entire argmin_math
crate needs a refactoring with input from someone who has a better understanding of how to generalize the various math crates than I have. Performance and no_std
support should then be amongst the main concerns. Being able to use argmin on GPUs would be really neat.
Unfortunately I won't be able to work on this in the near future, but I am happy to support any development in that direction.
Due to the way the math traits are currently defined, no inplace operations are possible, often leading to unnecessary allocations. By adding inplace variants and changing solver code accordingly, we should be able to increase performance.