Closed smu160 closed 3 months ago
Yes, multiversion seems like the way to go. Do you want implement it?
@LaihoE
I'll start on a PR for this issue once the benchmarks PR is sorted out. That way, I can easily establish a baseline to test for regressions. Thank you!!
Sorry if you were working on this, It is just blocking some future ideas with SIMD_LEN, so I went ahead and added multiversioning.
@LaihoE All good! Apologies for the delays.
Hi @LaihoE,
I was wondering if you are open to uitilizing the multiversion crate to utilize dyanamic dispatch, rather than having users compile from source (to get the best performance).
I use multiversion for my own project and it's great! You can see example usage here.
I can run a before/after benchmark using your existing benchmarks to check for any regressions.
Thank you!