Oscar Smith

I've added some tests that use this. I'm not sure how much needs testing.

I believe everything should now work. The last issues were that tstep and invγdt were being computed correctly, but the incorrect values for them were being passed into the nlp_params,...

no. I haven't had much time to play with this. I still think that ~5-10x faster than the current algorithm should be achievable. but it is one where you need...

What computer are you on? 35ns for Float64 `exp` is way too slow. I get ~5ns on an 11th gen laptop on battery. ``` julia> @btime exp($(Ref(x)[])) 4.885 ns (0...

also, I think it would be nice if the benchmarks were line graphs rather than tables.

The reason splitting speeds up latency is that it means that packages that only rely on a subset of the functionality don't have to load all the code. This benefit...

The same thing applies for packages that only require some of the functions in terms of load speed. Loading more code takes more time and memory. It's a lot faster...

Closing as discussed in https://github.com/JuliaLang/julia/pull/54372#issuecomment-2102885169

IMO this is probably not worth it. For simple operations like this, a for loop is as fast as you can go because LLVM will be able to vectorize the...

Agreed, but a lot of these changes seem good (at least from a usability perspective)