CUDA编程模型系列四(卷积 or sobel边缘检测)

CUDA编程模型系列四(卷积 or sobel边缘检测)

本系列教程将介绍具体的CUDA编程代码的细节

CUDA编程模型系列四(卷积 or sobel边缘检测)

#include <opencv2/opencv.hpp>
#include <iostream>
#include "error.cuh"

using namespace std;
using namespace cv;

// Sobel:
// kernel x00 x01 x02        m00 m01 m02 m03 m04    r00 r01 r02 r03 r04
//        x10 x11 x12        m10 m11 m12 m13 m14    r10 r11 r12 r13 r14
//        x20 x21 x22        m20 m21 m22 m23 m24    r20 r21 r22 r23 r24
//                           m30 m31 m32 m33 m34    r30 r31 r32 r33 r34
//                           m40 m41 m42 m43 m44    r40 r41 r42 r43 r44
//
// r11 = m00*x00 + m01*x01 + x02*m02 + x10 * m10 +x11*m11 + x12*m12 + x20*m20 +x21*m21 + x22*m22
// Gx 1 0 -1    Gy  1  2  1
//    2 0 -2        0  0  0
//    1 0 -1       -1 -2 -1
// Gx_r11 = m00 + 2*m10 + m20 - m02 - 2*m12 -m22
// Gy_r11 = m00 + 2*m01 + m02 - m20 - 2*m21 -m22
// r11 = (Gx + Gy)/2

__global__ void sobel_gpu(unsigned char* in, unsigned char* out, const int Height, const int Width)
{
    
    
    int x = blockDim.x * blockIdx.x + threadIdx.x;
    int y = blockDim.y * blockIdx.y + threadIdx.y;

    int index = y * Width + x;

    int Gx =0;
    int Gy =0;
    unsigned char x0, x1, x2, x3, x5, x6, x7, x8;
    //SM register
    if( x > 0 && x < (Width -1) && y>0 && y <(Height-1) )
    {
    
    
        x0 = in[(y-1)*Width + (x-1)];
        x1 = in[(y-1)*Width + (x)];
        x2 = in[(y-1)*Width + (x+1)];
        x3 = in[(y)*Width + (x-1)];
        //x4 = in[y * Width +  x];
        x5 = in[(y)*Width + (x+1)];
        x6 = in[(y+1)*Width + (x-1)];
        x7 = in[(y+1)*Width + (x)];
        x8 = in[(y+1)*Width + (x+1)];

        Gx = (x0 + 2*x3 +x6) - (x2 + 2*x5 + x8);
        Gy = (x0 + 2*x1 +x2) - (x6 + 2*x7 + x8);

        out[index] = (abs(Gx) + abs(Gy))/2;
    }

}

int main()
{
    
    
    Mat img = imread("1.jpg", 0);
    int height = img.rows;
    int width  = img.cols;

    Mat gaussImg;
    GaussianBlur(img, gaussImg, Size(3,3), 0, 0, BORDER_DEFAULT);

    Mat dst_gpu(height, width, CV_8UC1, Scalar(0));

    int memsize = height * width *sizeof(unsigned char);

    unsigned char* in_gpu;
    unsigned char* out_gpu;

    cudaMalloc((void**)&in_gpu, memsize);
    cudaMalloc((void**)&out_gpu, memsize);

    dim3 threadsPerBlock(32, 32);
    dim3 blocksPerGrid((width + threadsPerBlock.x -1)/threadsPerBlock.x, (height + threadsPerBlock.y -1)/threadsPerBlock.y);
    
    cudaMemcpy(in_gpu, gaussImg.data, memsize, cudaMemcpyHostToDevice);

    sobel_gpu<<<blocksPerGrid, threadsPerBlock>>>(in_gpu, out_gpu, height, width);
    cudaError_t error_code;
    error_code = cudaGetLastError();

    if(error_code != cudaSuccess)
    {
    
    
        printf("error_code: %s\n", cudaGetErrorString(error_code));
    }

    cudaMemcpy(dst_gpu.data, out_gpu, memsize, cudaMemcpyDeviceToHost);

    imwrite("save.png", dst_gpu);
    cudaFree(in_gpu);
    cudaFree(out_gpu);

    return 0;
}