在python jupyter下运行cuda c++程序

Pre-Install

• For Windows Users
  Install N-GPU drivers
• Ubuntu22.04 WSL Users
  WSL CUDA Drivers Install

Install

run this on jupyter(*.ipynb) files

!pip3 install nvcc4jupyter

Usage

load the extension to enable the magic commands:

%load_ext nvcc4jupyter

Run cuda test

%%cuda
#include <stdio.h>

__global__ void hello(){
    printf("Hello from block: %u, thread: %u\n", blockIdx.x, threadIdx.x);
}

int main(){
    hello<<<6, 12>>>();
    cudaDeviceSynchronize();
}

output
Hello from block: 0, thread: 0
Hello from block: 0, thread: 1
Hello from block: 0, thread: 2
Hello from block: 0, thread: 3
Hello from block: 0, thread: 4
Hello from block: 0, thread: 5
Hello from block: 0, thread: 6
Hello from block: 0, thread: 7
Hello from block: 0, thread: 8
Hello from block: 0, thread: 9
Hello from block: 0, thread: 10
Hello from block: 0, thread: 11
Hello from block: 2, thread: 0
Hello from block: 2, thread: 1
Hello from block: 2, thread: 2
Hello from block: 2, thread: 3
Hello from block: 2, thread: 4
Hello from block: 2, thread: 5
Hello from block: 2, thread: 6
Hello from block: 2, thread: 7
Hello from block: 2, thread: 8
Hello from block: 2, thread: 9
Hello from block: 2, thread: 10
Hello from block: 2, thread: 11
Hello from block: 1, thread: 0
...
Hello from block: 5, thread: 9
Hello from block: 5, thread: 10
Hello from block: 5, thread: 11

Output is truncated.

Other Problem

ModuleNotFoundError: No module named 'nvcc_plugin'

try this

!pip3 install nvcc_plugin

or

!pip3 install nvcc4jupyter==1.0.0