ANanonymoushn·Jun 20, 2022
> Nvm. Those methods use SIMD tricks, which while neat are hard to generalize.
I've been writing this sort of parser lately and I don't think people's fear of this stuff is at all proportional to the actual level of difficulty.
Looks like you can use this sort of technique on ~every CPU someone might plausibly be using as a laptop to write code on[0]. If you didn't use an off-the-shelf implementation, you would need 2 implementations of your inner loops to you target ~all x86 machines + M1 macs or 3 if want to go slightly faster on the vast majority of x86 machines that have AVX2 without dropping support for those x86 machines that have a bunch of SSE extensions but not AVX2.
> Also not sure what JSON-RPC is referring to? LSP? Are you arguing that JSON is suboptimal? Like maybe, but it's close enough to rival binary formats. Plus you can
Read the rest
parse GB/s of JSON.
I'm arguing that IPC is suboptimal for use cases that do not need IPC.
> Even text editors that forgo LSP have speed issues, unrelated to JSON.
I'm arguing that this is probably not due to the essential complexity of text editors.
As an unrelated toy example of the sort of thing that I'm talking about, recently I'm playing a game in which people implement somewhat fast wordcounts[1]. The reference solution uses std::iostream to put space-delimited strings into a std::unordered_map, then gets them all out into a std::vector and std::sorts them. This implementation works at about 33MB/s for large inputs. It's generally a lot more efficient than the sort of code you'd commonly find in a larger code base just because libstdc++ isn't as slow lots of other things and it uses vaguely correct data structures and algorithms. It seems like someone who never looks at any assembly listings and never uses a profiler but sort of has some vague idea of what data oriented design is could easily come up with a solution that processes large inputs at 200MB/s even if they are paying some minor overhead for the JVM, the CLR, the Go runtime, or something like that.
[0]: https://github.com/cyb70289/utf8
[1]: https://easyperf.net/blog/2022/05/28/Performance-analysis-an...