OpenCL：内核仅读取第一个像素

Question

灰度内核仅读取第一个像素

以下是我的

grayscale.cl

内核实现。我面临的问题是内核似乎仅在第一个像素上执行灰度计算，并通过迭代将该值分配给所有其他像素。 我是否正确使用了

read_imagef

和

write_imagef

？

grayscale.cl

__kernel void grayscale(
    __read_only image2d_t inputImage,
    __write_only image2d_t outputImage,
    const int width,
    const int height)
{
    int x = get_global_id(0); // X-coordinate
    int y = get_global_id(1); // Y-coordinate

    if (x < get_image_width(inputImage) && y < get_image_height(inputImage)) {
        int2 coord = (int2)(x, y);
        float4 pixel = read_imagef(inputImage, CLK_NORMALIZED_COORDS_FALSE | CLK_ADDRESS_CLAMP | CLK_FILTER_NEAREST, coord);

        // Converts un-normalised values
        float f_x = pixel.x * 255.0f;
        float f_y = pixel.y * 255.0f;
        float f_z = pixel.z * 255.0f;

        printf("Processing pixel: (%d, %d) : %f, %f, %f\n", x, y, f_x, f_y, f_z);
        
        // Grayscale calculation
        float gray = 0.299f * f_x + 0.587f * f_y + 0.114f * f_z;

        // Write the grayscale value to the output image
        write_imagef(outputImage, coord, (float4)(gray, gray, gray, 1.0f));
    }
}

内核调试输出

Processing pixel: (292, 233) : 23.000000, 5.000000, 3.000000
Processing pixel: (293, 233) : 23.000000, 5.000000, 3.000000
Processing pixel: (294, 233) : 23.000000, 5.000000, 3.000000
Processing pixel: (295, 233) : 23.000000, 5.000000, 3.000000

以下是上下文的相应（汇总）主机代码。

使用 OpenCV 获取图像

// Load the input image using OpenCV
    cv::Mat image = cv::imread(image_path, cv::IMREAD_COLOR);
    if(image.empty()){
        logger.log("Failed to load image", Logger::LogLevel::ERROR);
    }

    // Display image (if necessary)
    if(DISPLAY_IMAGES){
        cv::imshow("Reference Image Window", image);
        cv::waitKey(0);
    }

    // Convert to RGBA and get image dimensions
    cv::cvtColor(image, image, cv::COLOR_BGR2RGBA);
    *width = image.cols;
    *height = image.rows;

    // Flatten image into uchar array
    std::vector<unsigned char> _input_data(image.data, image.data + image.total() * 4);

图像对象

// Define cl_image variables and format
        cl_image_format input_format;
        input_format.image_channel_order = CL_RGBA;     // RGB
        input_format.image_channel_data_type = CL_UNORM_INT8;

        cl_image_format output_format;
        output_format.image_channel_order = CL_R;       // Single channel (grayscale)
        output_format.image_channel_data_type = CL_FLOAT;

// Create memory objects
        cl_mem input_image = clCreateImage2D(*context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, &input_format, width, height, 0, input_data.data(), &err_num);
        if(err_num != CL_SUCCESS){
            logger.log("Failed to create cl_image input_image mem object", Logger::LogLevel::ERROR);
        }

        cl_mem output_image = clCreateImage2D(*context, CL_MEM_WRITE_ONLY, &output_format, width, height, 0, nullptr, &err_num);
        if(err_num != CL_SUCCESS){
            logger.log("Failed to create cl_image output_image mem object", Logger::LogLevel::ERROR);
        }

内核：写入、执行和读取

// Initialise input image
        size_t origin[3] = {0, 0, 0};
        size_t region[3] = {width, height, 1};

        // Set kernel arguments
        err_num = clSetKernelArg(*kernel, 0, sizeof(cl_mem), &input_image);
        err_num |= clSetKernelArg(*kernel, 1, sizeof(cl_mem), &output_image);
        err_num |= clSetKernelArg(*kernel, 2, sizeof(int), &width);
        err_num |= clSetKernelArg(*kernel, 3, sizeof(int), &height);
        if(err_num != CL_SUCCESS){
            logger.log("Failed to set kernel arguments", Logger::LogLevel::ERROR);
        }

        err_num = clEnqueueWriteImage(*command_queue, input_image, CL_TRUE, origin, region, 0, 0, input_data.data(), 0, nullptr, &write_event);
        if(err_num != CL_SUCCESS){
            logger.log("Failed to write cl_image", Logger::LogLevel::ERROR);
        }

        // Perform kernel
        err_num = clEnqueueNDRangeKernel(*command_queue, *kernel, 2, nullptr, global_work_size, nullptr, 0, nullptr, &kernel_event);
        if(err_num != CL_SUCCESS){
            logger.log("Failed when executing kernel", Logger::LogLevel::ERROR);
        }

        // Read back image data
        err_num = clEnqueueReadImage(*command_queue, output_image, CL_TRUE, origin, region, 0, 0, output_data.data(), 0, nullptr, &read_event);
        if(err_num != CL_SUCCESS){
            logger.log("Failed to read back image data from kernel", Logger::LogLevel::ERROR);
        }

我是 OpenCL 新手，因此我们将不胜感激。

提前致谢！