你先读下我贴的那个link。说的相当全面了。而且有matlab的代码做对比。 How much of this increase is due to native performance adventures of Julia, and how much is simply due to the improvements in design that came from rebuilding this project from the ground up? It is of course difficult to say , and it is important to emphasize that one cannot be sure what portion of the performance increase can be attributed to inherent language features as opposed to design differences. Indeed, our MATLAB code suffers from many inefficiencies due to its long, cumulative development, and support for a plethora of models and features. Meanwhile, these design issues have been largely addressed in our Julia package. To best isolate differences in the languages themselves, we can look at our code to compute the model solution with gensys and apply the Kalman filter with kalman_filter. These two functions have relatively little redesign and optimization as compared to the MATLAB code and provide the most comparable, though still imperfect, measurements of performance. The reduction of 1/5th to 3/4th in computing time, therefore, could be taken as a first estimate of Julia's advantage in this single arena of computation.