So they wanted to speed up numpy and machine learning algs and I have no idea how I was going to do it.
To cut a long story short, I compiled python 3 for servers with a customized makefile:
```
./configure --enable-optimizations --with-lto
```
and then, I replaced `-O3` in make file with `-Ofast`.
That was it. And of course, I wonder why these are not defaults.
--
