【NVIDIA讲座】CUDA Python

CUDA
compute Unified Device Architecture

CUDA C/C++
基于C/C++的编程方法
支持异构编程的扩展方法
简单明了的API

CUDA支持的编程语言：
C/C++/PYTHON/Fortran/Java/...

CUDA并行计算模式
并行计算是同时应用多个计算资源解决一个计算问题
（空间换时间）

异构计算
HOST CPU和内存（host memory）
DEVICE GPU和现存 （device memory）

32个CUDA核(一个warp)共享一个execution contexts

CUDA Python

host: The CPU
device: The GPU
host memory: The system main memory
device memory: Onboard memroy on a GPU card
kernals: a GPU function launched by the host and executed on the device
device function: a GPU function executed on the device which can only be called from the device (i.e. form a kernel or another device function)

定义Kernal函数：

@cuda.jit('void(int32[:]),int32[:]')
def foo(aryA,aryB):
      ...

调用Kernal函数：

griddim = 1,2
blockdim = 3,4
foo[griddim,blockdim](aryA,aryB)

查看CUDA 版本

nvcc -V

创建文件

touch 20200609-python-cuda-cv.py