int minGridSize, blockSize;
cudaOccupancyMaxPotentialBlockSize(&minGridSize, &blockSize, myKernel, 0, 0);
printf("Suggested block size: %d\n", blockSize);

int numBlocks;
cudaOccupancyMaxActiveBlocksPerMultiprocessor(&numBlocks, myKernel, blockSize, 0);
int device;
cudaGetDevice(&device);
cudaDeviceProp prop;
cudaGetDeviceProperties(&prop, device);
float occupancy = (numBlocks * blockSize / 32.0f) / (prop.maxThreadsPerMultiProcessor / 32.0f);
printf("Achieved occupancy: %.1f%%\n", occupancy * 100);
