Does NumCpp generates vectorized instructions (AVX512)?

[IshaanKarnik]

Does NumCpp generates vectorized instructions (e.g. AVX512) after compiling with proper flags?

NumPy is able to leverage performance benefits from SIMD (vectorized) instruction set as per this link.

[SnowWalkerJ]

I'd like to know the answer too.

I'm considering putting NumCpp into production. And I'm facing a situation where performance really matters.

And also, does this project welcome PRs? What instructions should I follow if I want to contribute?

[dpilger26]

The short answer is that the NumCpp library does not do anything special to assure vectorized instructions are generated. The element-wise operations are for the most part built on top of the standard <algorithm> algorithms. If you compile with the NUMCPP_USE_MULTITHREAD option then they will default to using the std::par_unseq execution policy. So depending on your compiler and standard library implementation it can/may generate vectorized instructions.