NdArrayCore.hpp master - Githubissues

dpilger26 / NumCpp

C++ implementation of the Python Numpy library

https://dpilger26.github.io/NumCpp

MIT License

3.56k stars 555 forks source link

NdArrayCore.hpp master #119

Closed mu578 closed 2 years ago

mu578 commented 2 years ago

memory and performance killer

In argsort Axis::ROW/COL you are acquiring a new vector of indexes at each iteration; please move it out of the loop construct.
It would be wise to think about writing a simple stable_sort column-wise instead of transposing in Axis::ROW.

-- Cheers!

dpilger26 commented 2 years ago

Item 1 was fixed in Release Version 2.7.0. Due to the cache unfriendliness of column wise operations it is unlikely that item 2 would result in any performance gains.

mu578 commented 2 years ago

item 2 ; is a broad cowboy junkie statement; not really. denied, air ball.