我正在尝试编写一个程序，将RGB图像转换为灰度图像。我从Udacity问题集中得到了灵感。问题是，当我在Udacity web环境中编写内核时，它会显示我的代码有效，但是当我在本地计算机上尝试执行它时，虽然没有出现错误，但是我的图像不是灰度图像，而是完全变成了灰色。它看起来像一张与我加载的图像相同尺寸的灰色方框。您能帮助我找到我的代码中的错误吗？我已经将其与Udacity版本进行了比较，但似乎找不到错误。
#include "cuda_runtime.h"
#include "device_launch_parameters.h"
#include <string>
#include <cuda.h>
#include <stdio.h>
#include <opencv\cv.h>
#include <opencv\highgui.h>
#include <iostream>
#define CUDA_ERROR_CHECK
#define CudaSafeCall( err ) __cudaSafeCall( err, __FILE__, __LINE__ )
#define CudaCheckError() __cudaCheckError( __FILE__, __LINE__ )
inline void __cudaSafeCall(cudaError err, const char *file, const int line)
{
#ifdef CUDA_ERROR_CHECK
if (cudaSuccess != err)
{
fprintf(stderr, "cudaSafeCall() failed at %s:%i : %s\n",
file, line, cudaGetErrorString(err));
exit(-1);
}
#endif
return;
}
inline void __cudaCheckError(const char *file, const int line)
{
#ifdef CUDA_ERROR_CHECK
cudaError err = cudaGetLastError();
if (cudaSuccess != err)
{
fprintf(stderr, "cudaCheckError() failed at %s:%i : %s\n",
file, line, cudaGetErrorString(err));
exit(-1);
}
err = cudaDeviceSynchronize();
if (cudaSuccess != err)
{
fprintf(stderr, "cudaCheckError() with sync failed at %s:%i : %s\n",
file, line, cudaGetErrorString(err));
exit(-1);
}
#endif
return;
}
__global__ void rgb_2_grey(uchar* const greyImage, const uchar4* const rgbImage, int rows, int columns)
{
int rgb_x = blockIdx.x * blockDim.x + threadIdx.x; //x coordinate of pixel
int rgb_y = blockIdx.y * blockDim.y + threadIdx.y; //y coordinate of pixel
if ((rgb_x >= columns) && (rgb_y >= rows)) {
return;
}
int rgb_ab = rgb_y*columns + rgb_x; //absolute pixel position
uchar4 rgb_Img = rgbImage[rgb_ab];
greyImage[rgb_ab] = uchar((float(rgb_Img.x))*0.299f + (float(rgb_Img.y))*0.587f + (float(rgb_Img.z))*0.114f);
}
using namespace cv;
using namespace std;
void Proc_Img(uchar4** h_RGBImage, uchar** h_greyImage, uchar4 **d_RGBImage, uchar** d_greyImage);
void RGB_2_Greyscale(uchar* const d_greyImage, uchar4* const d_RGBImage, size_t num_Rows, size_t num_Cols);
void Save_Img();
Mat img_RGB;
Mat img_Grey;
uchar4 *d_rgbImg;
uchar *d_greyImg;
int main()
{
uchar4* h_rgbImg;
//uchar4* d_rgbImge=0;
uchar* h_greyImg;
//uchar* d_greyImge=0;
Proc_Img(&h_rgbImg, &h_greyImg, &d_rgbImg, &d_greyImg);
RGB_2_Greyscale(d_greyImg, d_rgbImg, img_RGB.rows, img_RGB.cols);
Save_Img();
return 0;
}
void Proc_Img(uchar4** h_RGBImage, uchar** h_greyImage, uchar4 **d_RGBImage, uchar** d_greyImage){
cudaFree(0);
CudaCheckError();
//loads image into a matrix object along with the colors in BGR format (must convert to rgb).
Mat img = imread("C:\\Users\\Austin\\Pictures\\wallpapers\\IMG_3581.JPG", CV_LOAD_IMAGE_COLOR);
if (img.empty()){
cerr << "couldnt open file dumbas..." << "C:\\Users\\Austin\\Pictures\\wallpapers\\IMG_3581.JPG" << endl;
exit(1);
}
//converts color type from BGR to RGB
cvtColor(img, img_RGB, CV_BGR2RGBA);
//allocate memory for new greyscale image.
//img.rows returns the range of pixels in y, img.cols returns range of pixels in x
//CV_8UC1 means 8 bit unsigned(non-negative) single channel of color, aka greyscale.
//all three of the parameters allow the create function in the Mat class to determine how much memory to allocate
img_Grey.create(img.rows, img.cols, CV_8UC1);
//creates rgb and greyscale image arrays
*h_RGBImage = (uchar4*)img_RGB.ptr<uchar>(0); //.ptr is a method in the mat class that returns a pointer to the first element of the matrix.
*h_greyImage = (uchar*)img_Grey.ptr<uchar>(0); //this is just like a regular array/pointer mem address to first element of the array. This is templated
//in this case the compiler runs the function for returning pointer of type unsigned char. for rgb image it is
//cast to uchar4 struct to hold r,g, and b values.
const size_t num_pix = (img_RGB.rows) * (img_RGB.cols); //amount of pixels
//allocate memory on gpu
cudaMalloc(d_RGBImage, sizeof(uchar4) * num_pix); //bites of 1 uchar4 times # of pixels gives number of bites necessary for array
CudaCheckError();
cudaMalloc(d_greyImage, sizeof(uchar) * num_pix);//bites of uchar times # pixels gives number of bites necessary for array
CudaCheckError();
cudaMemset(*d_greyImage, 0, sizeof(uchar) * num_pix);
CudaCheckError();
//copy array into allocated space
cudaMemcpy(*d_RGBImage, *h_RGBImage, sizeof(uchar4)*num_pix, cudaMemcpyHostToDevice);
CudaCheckError();
d_rgbImg = *d_RGBImage;
d_greyImg = *d_greyImage;
}
void RGB_2_Greyscale(uchar* const d_greyImage, uchar4* const d_RGBImage, size_t num_Rows, size_t num_Cols){
const int BS = 16;
const dim3 blockSize(BS, BS);
const dim3 gridSize((num_Cols / BS) + 1, (num_Rows / BS) + 1);
rgb_2_grey <<<gridSize, blockSize>>>(d_greyImage, d_RGBImage, num_Rows, num_Cols);
cudaDeviceSynchronize(); CudaCheckError();
}
void Save_Img(){
const size_t num_pix = (img_RGB.rows) * (img_RGB.cols);
cudaMemcpy(img_Grey.ptr<uchar>(0), d_greyImg, sizeof(uchar)*num_pix, cudaMemcpyDeviceToHost);
CudaCheckError();
imwrite("C:\\Users\\Austin\\Pictures\\wallpapers\\IMG_3581GR.JPG", img_Grey);
cudaFree(d_rgbImg);
cudaFree(d_greyImg);
}
变量
d_rgbIme未初始化就被使用了
但是我已经在上面初始化过了。如果我把它们设置为0，则会出现CUDA错误提示：
遇到非法内存访问
我尝试运行cuda-memcheck，但却收到无法运行文件的错误提示...
if ((rgb_x < columns) && (rgb_y < rows)) {，然后再进行读写操作。另外，尝试使用cuda-memcheck运行你的代码，并为你的代码添加适当的CUDA错误检查。 - Robert Crovellafloat类型时（0.299f 是一个float常量），不需要显式地转换为float。 - Robert Crovellacuda-memcheck，它非常简单。这个陈述并没有告诉我太多信息：“我尝试运行cuda-memcheck，但是我得到了无法运行文件的错误...” 你确切地使用了什么命令行来运行cuda-memcheck？确切的输出是什么？如果你回答这些问题，你可能可以学会如何使用它，并且它将对调试CUDA程序非常有用。 - Robert Crovella