Articles were making bold claims, but it's just optimizing libc++ and abseil's implementations. There are already known implementations faster than either of th

Articles were making bold claims, but it's just optimizing libc++ and abseil's implementations. There are already known implementations faster than either of these in the right scenarios. As always, speed depends on the use case. What's optimal for a sorting algorithm depends on the size of the data, its distribution, and also how it fits into the rest of the program and how that program fits into what else the machine is doing. For example, you can make things faster by using more hardware resources, but that would penalize the rest of your program that doesn't have access to them anymore. I've frequently seen cases where a function was made faster in a microbenchmark by using all of the registers available, but then real programs ended up slower because the rest of the code had to spill to call it.