CUDA编程主函数模板及实例

//CUDA编程主函数模板


int main(){
//Allocate memory on GPU(分配GPU内存）
float *Md;
cudaMalloc((void**)&Md, size);
//Copy data from CPU to GPU（从CPU内存中拷贝数据到GPU中）
cud M (Md M i d M H tT D i ) aMemcpy(Md, M, size, cudaMemcpyHostToDevice);
//Call GPU kernel function（调用CUDA内核函数完成程序指定的运算）
kernel<<<dimGrid, dimBlock>>> (arguments);
//Copy data from GPU back to CPU（将数据从GPU拷贝回CPU内存中）
CopyFromDeviceMatrix(M, Md);
//Free device matrices（释放GPU内存空间）
FreeDeviceMatrix(Md);
}
应用实例：基于CUDA的两幅数字图像相加
部分主函数代码如下：
int main() {
	// source images, gray  
	Mat img1_host = imread("D:\\pictures\\3.jpg", IMREAD_GRAYSCALE);
	Mat img2_host = imread("D:\\pictures\\4.jpg", IMREAD_GRAYSCALE);

	int row = img1_host.rows;
	int col = img1_host.cols;
	int length = row * col;

	// memory size  
	int memSize = length * sizeof(uchar);

	// device memory  
	uchar* img1_device;
	uchar* img2_device;
	uchar* imgres_device;

	cudaMalloc((void**)&img1_device, memSize);
	cudaMalloc((void**)&img2_device, memSize);
	cudaMalloc((void**)&imgres_device, memSize);

	// copy host to device  
	cudaMemcpy(img1_device, img1_host.data, memSize, cudaMemcpyHostToDevice);
	cudaMemcpy(img2_device, img2_host.data, memSize, cudaMemcpyHostToDevice);

	// setting parameters and run the kernel function  
	dim3 grid(1 + (length / (32 * 32 + 1)), 1, 1);      // grid  
	dim3 block(32, 32, 1);                              // block  
	imageAdd << <grid, block >> >(img1_device, img2_device, imgres_device, length);

	// copy device to host  
	Mat imgres_host = Mat::zeros(row, col, CV_8UC1);
	cudaMemcpy(imgres_host.data, imgres_device, memSize, cudaMemcpyDeviceToHost);

	// show source and result images  
	imshow("img1", img1_host);
	imshow("img2", img2_host);
	imshow("imgres", imgres_host);
	waitKey(0);
	cudaFree(img1_device);
	cudaFree(img2_device);
	cudaFree(imgres_device);
	system("pause");
	return 0;
}
主函数中包括了上述编程的模板。