One small part of curl is now 36% faster

Notice the drop at R198? Actually, I caused it.

Have you ever used curl? Well, now one small part of it is 36% faster.

I submitted a PR that optimized the Base64 decode lookup table and rewrote the loop. Bagder, the maintainer, reviewed it and committed it himself. He even changed his mind about the approach and said he "now thinks differently" about using the full lookup table.

Here's what actually happened, and the results below were run by Bagder himself: before my change, 353.1 nanoseconds per decode. After my change, 226.43 nanoseconds per decode. 10 million operations to measure that. That's roughly 36% faster for this Base64 decoding benchmark.

Not all of curl got faster. Just this tiny function.

And now Bagder's performance infrastructure is recording it as part of curl's performance history.

So what changes for users? Not that their downloads suddenly become 36% faster. It means curl spends less CPU time on this particular piece of work whenever Base64 decoding is involved. Most people will probably never notice it.

GitHub: curl/curl #22903. Commit: e7ea5d1. Performance history: curl.se/perf (look at R198).