jupyter-cpp-kernel

An kernel implementation for C++, it

treat each cell as a unique c++ main function.
support many magic templates.
support complex compile configuration, for example, you can use this for CUDA programming.

Each code cell will be compiled as a c++ file, so it is slow but stable, and this idea is different from the kernel for the well known cling project.

This project is based on the code of jupyter-c-kernel, great work!

install

the requirements are

to install as follows

There are some special tokens that helps to auto insert some code.

//% includes: full

will insert some high frequency header fills, such as

to include some specific header file, just includes its name

//% includes: algorithm

// is equal to #include <algorithm>

Kernal will automatically add main function wrapper, so just insert code block as follows will get

//% includes: full

std::cout << "hello world" << endl;

works.

But sometimes, we might need to define some data structure, so better to define a seperate main function.

//% includes: full
//% main: no

struct Node {
  int v;
  Node* left, *right;
};

int main() {
  Node root;
  
  return 0;
}

By setting cflags and ldflags, one can control the compile period, for example, to use Eigen

//% cflags: -I xxx/Eigen
//% includes: Eigen

// use eigen logic

//% compiler: nvcc
//% main: no

__global__
void VecAdd(float* A, float* B, float* C) {
  // ...
}