I bench-marked using Intel vTune.
for loop is interesting. It's why Tensorflow Core written in C++ instead Java.
I don't know any complex C++ program that employ their own GCs when C++ has RAII which is superior to GC.
Just give a try for C++11/14/17 and you will see which one is more maintainable and expressive.
Look at Chromium codebase. It's the most beautiful codebase I've ever been to.
I've done a lot of CRUD web apps in C++ using expresscpp [1] and I would say it's easy as ABC.
A lot of Java folks haven't tried C++11/14/17 (Modern C++).
C++ is Zen of OOP.