opencl image2d_t不回写值

时间:2011-12-22 06:03:08

标签: opencl

Windows 7 AMD App SDK 2.6 Asic:Redwood

我正在尝试编写一个简单的pass-thru内核,看看问题是什么,我似乎无法找到错误。

void kernel_test(CLManager* clMgr, int W, int H)
{
    cl::ImageFormat format;
    format.image_channel_order = CL_RGBA;
    format.image_channel_data_type = CL_FLOAT;

    cl_float4* inp = new cl_float4[W * H];

    for (int i = 0; i < W * H; ++i)
    {
        inp[i].s[0] = 1.0f;
        inp[i].s[1] = 0.0f;
        inp[i].s[2] = 0.0f;
        inp[i].s[3] = 1.0f;
    }

    cl_float4* oup = new cl_float4[W * H];

    cl::Image2D clInputImage  = clMgr->createImage<cl::Image2D>(CL_MEM_READ_ONLY, format, W, H, 0, NULL);
    cl::Image2D clOutputImage = clMgr->createImage<cl::Image2D>(CL_MEM_WRITE_ONLY, format, W, H, 0, NULL);
    cl::Sampler sampler = clMgr->createSampler(CL_FALSE, CL_ADDRESS_CLAMP_TO_EDGE, CL_FILTER_NEAREST);

    cl::size_t<3> origin;
    origin[0] = 0; origin[1] = 0, origin[2] = 0;
    cl::size_t<3> region;
    region[0] = W; region[1] = H; region[2] = 1;

    unsigned int pgmID = clMgr->buildSource("CL/convolution.cl");

    cl::Kernel* kernel = clMgr->makeKernel(pgmID, "passthru");

    cl::CommandQueue* queue = clMgr->getCmdQueue();

    queue->enqueueWriteImage(clInputImage, CL_TRUE, origin, region, 0, 0, inp, 0, NULL);

    int status;

    status = kernel->setArg(0, clInputImage);
    status = kernel->setArg(1, clOutputImage);
    status = kernel->setArg(2, sampler);

    cl::NDRange globalSize(W, H);

    std::cout << "Starting Kernel: passthru" << std::endl;

    status = queue->enqueueNDRangeKernel(*kernel, 2, globalSize, cl::NullRange);

    std::cout << "Ending Kernel: passthru" << std::endl;

    status = queue->enqueueReadImage(clOutputImage, CL_TRUE, origin, region, 0, 0, oup);
}

内核看起来像这样

__kernel
void passthru(__read_only image2d_t sourceImage,
                 __write_only image2d_t outputImage,
                 sampler_t sampler)
{
    // Store each work-items unique row and column
    int2 coords = (int2){get_global_id(0), get_global_id(1)};
    float4 pixel = read_imagef(sourceImage, sampler, coords);
    write_imagef(outputImage, coords, pixel);
}

所以我将输入图像清除为RED,然后内核应该将RED颜色写回输出。但是,它只输出0.0f表示输出中的所有内容。我没有看到clOutputImage中的值的任何特殊原因?

1 个答案:

答案 0 :(得分:1)

问题在于这一行。
    status = queue->enqueueNDRangeKernel(*kernel, 2, globalSize, cl::NullRange);

我错误地认为是C参数列表并添加了2作为维度。删除2解决了问题。