The last time I ran the experiment of having a single jump, it was slower than jump-per-opcode-handler. It's true that predictors are able to see through multi

The last time I ran the experiment of having a single jump, it was slower than jump-per-opcode-handler. It's true that predictors are able to see through multiple levels, but a threaded interpreter gives them plus one level, and that ends up mattering as much as it always did.