Rgb2Gray

• GPU上运行的函数又称为Kernel，用__global__修饰
• 调用Kernel函数时，用FunctionCall<<<block_shape, thread_shape, int>>>(args1, args2, ...)
• block_shape与thread shape都是一个三维的结构dim(x,y,z), 其中x为横向，y为纵向。变化最快的维度为x，其他为y，最后为z。所以维度为(i,j,k)的thread或block的Index实际上是：k * Dim.x * Dim.y + j * Dim.x +i
• 每个block中的线程数量有一定的限制，老的GPU通过为512个，较新的GPU则通常有1024个，所以每个block中的线程最好不要超过(32,32)
  
  

#include <iostream>
#include <cuda_runtime.h>
#include <opencv2/core/core.hpp>
#include <opencv2/highgui/highgui.hpp>

struct image_pixel {
    uchar r;
    uchar g;
    uchar b;
};

__global__ void Rgb2Gray(image_pixel *d_color_data, uchar *d_gray_data, int image_rows, int image_cols) {
    int i = blockIdx.x * blockDim.x + threadIdx.x;
    int j = blockIdx.y * blockDim.y + threadIdx.y;

    if (i < image_cols && j < image_rows) {
        int pixel_index = image_cols * j + i;
        image_pixel pixel = d_color_data[pixel_index];
        d_gray_data[pixel_index] = pixel.r * 0.299 + pixel.g * 0.587 + pixel.b * 0.114;
    }
}

int main(int argc, char *argv[]) {
    cv::Mat image = cv::imread("/home/yansheng/Pictures/fbb.jpg");
    if (!image.isContinuous()) {
        std::cout << "The image data is not continuous" << std::endl;
        return -1;
    }
    int image_rows = image.rows;
    int image_cols = image.cols;

    std::cout << "Image Width: " << image.cols << std::endl;
    std::cout << "Image Height: " << image.rows << std::endl;

    image_pixel *d_color_data;
    uchar *d_gray_data;
    cudaMalloc((void **)&d_color_data, image_rows * image_cols * sizeof(image_pixel));
    cudaMalloc((void **)&d_gray_data, image_rows * image_cols);

    cudaMemcpy(d_color_data, image.data, image_rows * image_cols * sizeof(image_pixel), cudaMemcpyHostToDevice);

    dim3 block_dim(16, 16);
    dim3 grad_dim(image_cols / block_dim.x + 1, image_rows / block_dim.y + 1);

    Rgb2Gray<<<grad_dim, block_dim>>>(d_color_data, d_gray_data, image_rows, image_cols);

    cv::Mat gray(image.size(), CV_8UC1);
    cudaMemcpy(gray.data, d_gray_data, image_rows * image_cols, cudaMemcpyDeviceToHost);

    cv::imshow("Gray", gray);
    cv::waitKey();

    return 0;
}