[fpc-pascal] Generics vs templates

[Maciej Izak]

2018-01-08 21:30 GMT+01:00 Graeme Geldenhuys <mailinglists at geldenhuys.co.uk>
:

> Speaking of Generics and Benchmarks. Has anybody done some benchmarks on
> FPC's Generics vs "old-school" TList and TObjectList. Recently I did a very
> simple test with Delphi XE3 using TList<Integer> and a stock TList. Adding
> 50,000 and 200,000 integer values to each list, and timing the creation of
> the list and population of the list. Then I also timed the destruction of
> the list. I was horified to find out how much slower Delphi's Generics were
> compared to TList and TObjectList. Destruction was 250x slower in many
> cases. Creation and population of the list was 5x-10x slower.
>
> Lets hope FPC fares better. If nobody has done such tests, I can do it
> tomorrow at work with the same Delphi test code I created.
>

It depends on use case and on library design. For example in the FPC case,
generic TList<T> has better performance for larger lists (the capacity uses
golden ratio) than regular TList (for Integers). The performance difference
in daily usage is rather minor (if any).

We have available detailed tests for generic and non generics hash maps
thanks to Benito van der Zander (FPC has so many different maps! :) ):

http://www.benibela.de/fpc-map-benchmark_en.html

The results for rtl-generics should be better soon. I am working on new
version of rtl-generics library, so all should works faster (better hashing
function + optimizations for managed types - especially improved for
incoming smart pointers/objects).

-- 
Best regards,
Maciej Izak
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freepascal.org/pipermail/fpc-pascal/attachments/20180109/8518d9a2/attachment.html>