Benchmarks on modern CPUs

[dumblob]

I wonder whether the efficiency "cross point" is still around 200 bytes even on modern CPUs.

Does anyone have any recent measurements on modern CPUs (e.g. some 2021 Ryzens)?

(I myself do not have access to any such new CPU, so I can't run test/tests.c myself 😢 )

[josephg]

At this point all my development has moved to rust. I've rewritten this library in rust and gotten a pretty big performance gain in the process. I'm actually not sure why the rust version is so much faster than the C code - not all of it is attributable to the changes I've made.

Not sure when the "cross point" is, but the repository there has some much more recent performance numbers on a ryzen 5800x. In a real-world editing scenario, jumprope does 35M edits per second - compared to librope's 15M per second.

[dumblob]

Thanks for the info and the link (I totally missed you're working on a Rust-based successor).

The speed is really almost unbeliavable - so basically jumprope can become a substitute for nearly any string as it'll outperform contiguous string already at maybe 50-100 bytes of length?

Btw. have you thought of an optimization for small-length strings by first using a contiguous block and then after reaching some guesstimated number of bytes "upgrade" to jumprope? It seems it could be doable with gap buffer though only measurements will show whether it further increases overall efficiency for small length strings...

[josephg]

Interesting idea. Could be worthwhile but not super relevant for my use case. You wanna throw something together? I’d love to see how it performs.

[dumblob]

Interesting idea. Could be worthwhile but not super relevant for my use case. You wanna throw something together?

Not now (I'm too busy unfortunately) - it was just a flash idea when seeing the sheer speed of jumprope-rs (assuming the speed is not a side effect of some bug).

I’d love to see how it performs.

Me too 😉.