Occupancy 计算器

1// Target: blockDim = 256, at least 8 resident blocks per SM2__global__ void __launch_bounds__(256, 8)3myKernel(const float* __restrict__ in, float* out, int n) {4    __shared__ float tile[2048];   // 8 KB5    ...6}7 8// Confirm real register count at compile time:9//   nvcc -O3 -arch=sm_80 -Xptxas -v -c kernel.cu10// Let CUDA recommend blockDim at runtime:11//   cudaOccupancyMaxPotentialBlockSize(&minGrid, &blockSize, myKernel, 0, 0);