OpenCV support with OpenCL on imx8mp

[meetgandhi-dev]

Hi,

I'm trying to enable OpenCL support in OpenCV. I'm on hardknott branch.

With reference to this discussion #320 , I have added below parameters in my local.conf file:

PACKAGECONFIG_append_pn-opencv = " opencl"
PREFERRED_PROVIDER_opencl-headers_imxgpu     = "imx-gpu-viv"

I have build the opencv package. As per the yocto's log.do_configure the support for OpenCL is enabled:

--   OpenCL:                        YES (no extra features)
--     Include path:                /home/test/imx8mp-sr/imx-yocto/build-xwayland-imx8mpsolidrun/tmp/work/cortexa53-crypto-mx8mp-poky-linux/opencv/4.5.2.imx-r0/git/3rdparty/include/opencl/1.2
--     Link libraries:              Dynamic load
--

I have verified the same from the opencv2/cvconfig.h file as well:

/* OpenCL Support */
#define HAVE_OPENCL
/* #undef HAVE_OPENCL_STATIC */
/* #undef HAVE_OPENCL_SVM */

But when I try to check the OpenCV in the generated image, it says OpenCL is disabled. Below is the output of opencv_version --opencl command:

4.5.2
OpenCL is disabled

I have also tried below code snippet:

#include <iostream>
#include <fstream>
#include <string>
#include <iterator>
#include <opencv2/cvconfig.h>
#include <opencv2/opencv.hpp>
#include <opencv2/core/ocl.hpp>

using namespace std;

int main()
{
    cv::ocl::setUseOpenCL(true);
    if (!cv::ocl::haveOpenCL())
    {
        cout << "OpenCL is not avaiable..." << endl;
    }
    cv::ocl::Context context;
    if (!context.create(cv::ocl::Device::TYPE_ALL))
    {
        cout << "Failed creating the context..." << endl;
    }

    // In OpenCV 3.0.0 beta, only a single device is detected.
    cout << context.ndevices() << " GPU devices are detected." << endl;
    for (int i = 0; i < context.ndevices(); i++)
    {
        cv::ocl::Device device = context.device(i);
        cout << "name                 : " << device.name() << endl;
        cout << "available            : " << device.available() << endl;
        cout << "imageSupport         : " << device.imageSupport() << endl;
        cout << "OpenCL_C_Version     : " << device.OpenCL_C_Version() << endl;
        cout << endl;
    }
    return 0;
}

I got below output:

OpenCL is not avaiable...
Failed creating the context...
0 GPU devices are detected.

I tried using the OpenCL SDK directly. Using that my test application was able to detect the GPUs:

#include <iostream>
#include <string>
#include <vector>

#include <CL/opencl.hpp>

static cl_int PrintPlatformInfoSummary(cl::Platform platform)
{
    std::cout << "\tName:           " << platform.getInfo<CL_PLATFORM_NAME>()
              << "\n";
    std::cout << "\tVendor:         " << platform.getInfo<CL_PLATFORM_VENDOR>()
              << "\n";
    std::cout << "\tDriver Version: " << platform.getInfo<CL_PLATFORM_VERSION>()
              << "\n";

    return CL_SUCCESS;
}

static void PrintDeviceType(const std::string& label, cl_device_type type)
{
    std::cout << label << ((type & CL_DEVICE_TYPE_DEFAULT) ? "DEFAULT " : "")
              << ((type & CL_DEVICE_TYPE_CPU) ? "CPU " : "")
              << ((type & CL_DEVICE_TYPE_GPU) ? "GPU " : "")
              << ((type & CL_DEVICE_TYPE_ACCELERATOR) ? "ACCELERATOR " : "")
              << ((type & CL_DEVICE_TYPE_CUSTOM) ? "CUSTOM " : "") << "\n";
}

static cl_int PrintDeviceInfoSummary(const std::vector<cl::Device> devices)
{
    for (size_t i = 0; i < devices.size(); i++)
    {
        std::cout << "Device[" << i << "]:\n";

        cl_device_type deviceType = devices[i].getInfo<CL_DEVICE_TYPE>();
        PrintDeviceType("\tType:           ", deviceType);

        std::cout << "\tName:           "
                  << devices[i].getInfo<CL_DEVICE_NAME>() << "\n";
        std::cout << "\tVendor:         "
                  << devices[i].getInfo<CL_DEVICE_VENDOR>() << "\n";
        std::cout << "\tDevice Version: "
                  << devices[i].getInfo<CL_DEVICE_VERSION>() << "\n";
        std::cout << "\tDevice Profile: "
                  << devices[i].getInfo<CL_DEVICE_PROFILE>() << "\n";
        std::cout << "\tDriver Version: "
                  << devices[i].getInfo<CL_DRIVER_VERSION>() << "\n";
    }

    return CL_SUCCESS;
}

int main(int, char**)
{
    std::vector<cl::Platform> platforms;
    cl::Platform::get(&platforms);
    std::cout << "Enumerated " << platforms.size() << " platforms.\n\n";

    for (size_t i = 0; i < platforms.size(); i++)
    {
        std::cout << "Platform[" << i << "]:\n";
        PrintPlatformInfoSummary(platforms[i]);

        std::vector<cl::Device> devices;
        platforms[i].getDevices(CL_DEVICE_TYPE_ALL, &devices);

        PrintDeviceInfoSummary(devices);
        std::cout << "\n";
    }

    std::cout << "Done.\n";

    return 0;
}

The output is shown below:

Enumerated 1 platforms.

Platform[0]:
    Name:           Vivante OpenCL Platform
    Vendor:         Vivante Corporation
    Driver Version: OpenCL 3.0 V6.4.3.p2.336687
Device[0]:
    Type:           GPU 
    Name:           Vivante OpenCL Device GC7000UL.6204.0000
    Vendor:         Vivante Corporation
    Device Version: OpenCL 3.0 
    Device Profile: FULL_PROFILE
    Driver Version: OpenCL 3.0 V6.4.3.p2.336687
Device[1]:
    Type:           GPU 
    Name:           Vivante OpenCL Device VIP8000Nano-S+I.8002.0000
    Vendor:         Vivante Corporation
    Device Version: OpenCL 3.0 
    Device Profile: FULL_PROFILE
    Driver Version: OpenCL 3.0 V6.4.3.p2.336687
Done

Am I missing something here, when building the OpenCV with OpenCL? Let me know if you need any additional information.

[fabioestevam]

Hardknott is no longer supported. Please try scarthgap instead.

[meetgandhi-dev]

Upgrading the entire Yocto version isn’t feasible at the moment. Could you suggest some alternative solutions or workarounds?

[otavio]

As Fabio pointed out, it is very difficult for us to assist you with such an old version. Unfortunately, this is not maintained anymore, as it has reached the end of life. So, upgrading the release is likely to become imperative very soon.

To work on that kind of situation, we usually make some specific backports or minor modifications to allow the customers to keep using the old releases. However, this ends up becoming more expensive and, in the long term, unmaintainable.

We can help you commercially if you want, but our advice is to try to Upgrade the version as fast as possible.