[Doubt] How are various dtypes handled in NumCpp

[faze-geek]

Hello everyone, This is more of a doubt than an issue, I could not find a discussion forum or mailing list so I'm asking it here itself. I am new to this interesting idea of implementing Numpy through CPP and had a very basic doubt.

How are the various dtypes of Numpy ndarray like int8, int16, int32, float32 handled in c++ ? Can someone point me to the implementation and briefly explain the thought behind it ?

[dpilger26]

Do you have a more specific question, I'm not certain what you are really asking?

The dtype is handled throughout the library as a template parameter, for example:

auto data = nc::NdArray<double>{5, 6};

This will create an NdArray of dtype double and dimensions 5x6.

[faze-geek]

Hi @dpilger26 . There are various dtypes we in numpy - https://numpy.org/doc/stable/reference/arrays.dtypes.html I meant how do you smartly handle the dtype, ensure an annotation of right size and memory is used. Like If I want to pick int32 (integer of 32 bits) or let's say something more advanced like complex128 (floating point complex number of 128 bits), what will be the cpp equivalent used.

[faze-geek]

If you can point me out to the files where the dtypes for NumCPP are designed it will be helpful as well.

[dpilger26]

I don't think you are understanding... it sounds like maybe you come from a Python background and haven't done a lot of C++? The dtypes aren't "designed" like you say. The dtype is a template parameter into NdArray<dtype>, and can be any C++ arithmetic type, as well as std::complex<T>, and can also act as an array for structured data types while disabling the arithmetic operations. The data is stored under the hood as a contiguous, row major block of memory of the chosen dtype. The NdArrayCore.hpp file is where this is all defined:

https://github.com/dpilger26/NumCpp/blob/master/include/NumCpp/NdArray/NdArrayCore.hpp

[faze-geek]

I got what I wanted. Thanks for answering my silly doubts :)

[faze-geek]

@dpilger26

The dtype is a template parameter into NdArray<dtype>, and can be any C++ arithmetic type, as well as std::complex<T>

Can the dtype be precision types like int8, int16, int32, int64 as well or all integers are just int?

[faze-geek]

I think for precision types C++ libraries like cstdint provide int8_t, int16_t and so on.

[faze-geek]

I see that is handled here. Thanks ! https://github.com/dpilger26/NumCpp/blob/5e40aab74d14e257d65d3dc385c9ff9e2120c60e/include/NumCpp/Core/Types.hpp#L30-L43