Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

These are definitely incredible.

Wonder how one can know to find and make these optimizations



Profiling. So much profiling.

Of course for that to be relevant you need to have tests that can actually hit the relevant code paths.

The fastest(hah!) way to get performance of something fixed is to provide a easy to run deterministic and vaguely realistic example where performance is significantly worse in one case vs another. Now that could be one JS engine vs another, but it could also be a piece of JS calling two functions doing the same thing, only one in JS and one in wasm, or say a switch vs a series of if/else


Most of these tests are attached to the pull requests with metrics on how the microbenchmarks improved :)


Most of these provide specializations for specific operations that “should” get optimized but currently aren’t. They were probably pulled from some other code that was slower than it should be, isolated to the specific thing that wasn’t optimized, and then the heuristics were written to match it.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: