fubark
commented
6 months ago I think the end goal isn't really to compete with LuaJIT or even more advanced JIT tiers in JS engines. The demonstration showed that Cyber can take a different approach than a tracing JIT and the perf can still be in the same ballpark for select examples such as fibonacci. It remains to be seen how other algorithms compares. Cyber's advantage is that the compilation to native code is very fast so it can start executing faster code sooner, while other JITs need to warm up.
At this point, I'm more interested in showing that Cyber code can be compiled in one shot (tradeoff of much slower compile times) with perf comparable to C. Since Cyber supports static types, this makes the most sense going forward.
Your metrics show you outpacing LuaJIT, which is just insane. Kudos for getting that fast, that's really something.
That's the thing though: it's insane. It's hard to believe, for someone looking for the first time. They could of course run the benchmarks themself, but absent an automated harness that's tedious; I'd wager a lot of them will take the easier route of writing off the unknown project as a crank.
The solution, of course, is to have a reputable independent third party replicate the test, and get the same result. Off the top of my head I'm not certain if there's a formal process for that though. Professional consultation firms are probably expensive, but if you reached out to a tech journalist known for benchmarks and tell them "hey, we're faster than LuaJIT", you'd likely get their attention for free.
Of course, you'd then probably get a lot of eyes on the project, which might not be what you want at this stage.