CUDA programming

Nvidia Blog, Even easier introduction CUDA

Nvidia Blog CUDA refressher

• Add Specifier __global__ to functions, which tells the CUDA C++ compiler that this is a function that runs on GPU and call be called from CPU, those __global__ functions are know as kernels.

• CUDA kernel launches are specified using the triple angle bracket syntax <<<numBlocks, blockSize>>>, execution configuration

• CUDA kernel launches don’t block the calling CPU thread, if need the CPU to wait until the kernel is done before it accesses the results, add cudaDeviceSynchronize()

• nvcc -x cu option allows us to write CUDA code directly in cpp file, Jupyter supports .cpp syntax hightlighting

!nvcc -x cu -arch=sm_70 -o monte_carlo_mgpu_cuda_peer exercises/monte_carlo_mgpu_cuda_peer.cpp
%%time
!./monte_carlo_mgpu_cuda_peer