Why is v0.7 faster?



As we know, there are so many versions of Julia available. I downloaded three of them, and compared the speed of my code using them. I have used the “@profile” macro to find the hotspots, it showed that most of the cpu time was spend on operations like

@. a = b * c
plan_fft! * a
plan_fft! \ a

The result showed that the speed of the three Julia version is about
(0.6.1 stable) : (juliapro : (0.7-DEV.2158) = 1 : 1: 1.3

That is amazing progress if my bechmark is meaningful.
My question is, what makes the difference? Are some new features introduced in v0.7? Is the difference in performance just because the stable version like v0.6 just throw away some optimization strategies for stability?


This thread Fantastic progress in master branch! discusses the advances in v0.7 in depth.


NEWS.md should probably mention the highlights.


Practically, there is just a stable version (v0.6.1), and a version in development that will become v0.7 at some point.