Performance of ranges verses direct

[John Colvin]

On Wednesday, 11 December 2013 at 07:29:39 UTC, Frustrated wrote:
>
> Has anyone done any work on comparing the performance of ranges 
> vs using direct straightforward code(optimized in the sense of 
> the way ranges work but by hand).
>
> e.g., How does filter compare to a simple loop over the 
> elements and comparing it. How does a a chain of UFCS compare 
> to doing the same in a simple loop.

In my very unscientific experiments:
range based, functional style D code runs about 75-100% speed of 
the equivalent explicit iterative version.

A lot of the performance loss is down to missed optimisations, in 
particular inlining. gdc/ldc are good at this, dmd not so good so 
you'll likely see bigger performance gaps with dmd than the other 
compilers.


General advice: use ranges, std.range and std.algorithm wherever 
possible. Profile the resulting code and write out the hotspots 
in an imperative style to see if you can squeeze out some extra 
speed.