Andrei Alexandrescu needs to read this

welkam wwwelkam at
Wed Oct 23 23:20:07 UTC 2019

On Wednesday, 23 October 2019 at 22:03:29 UTC, Jonathan Marler 

> That's why Andrei says "always measure".
I see you didnt read the paper. The performance measurement that 
you talk about and Andrei does measures two things: 1. change due 
to code change and 2. change due to code layout changes.
When performance change is in order of magnitude then you can 
safely assume it was because of code change you made but when the 
difference is less than 10% it becomes unclear what actually is 
responsible for that difference. If you had read the paper you 
would find out that gcc's -O3 changes performance over -O2 from 
-8% to +12% on the same application.

Simple measurement is not sufficient to conclude that your change 
in algorithm is what is responsible for measured performance 

More information about the Digitalmars-d mailing list