Using OpenCV 3.0 UMat on Odroid XU3

edit

Hi all, I am trying to use OpenCV3.0 OpenCL API on my Odroid XU3. Unfortunately, I am encountering some real weird problems.

The following is my devices information which is derived by OpenCL 1.1:

PlatformCount: 1
1. Device: Mali-T628
 1.1 Hardware version: OpenCL 1.1
 1.2 Software version: 1.1
 1.3 OpenCL C version: OpenCL C 1.1
 1.4 Parallel compute units: 4
2. Device: Mali-T628
 2.1 Hardware version: OpenCL 1.1
 2.2 Software version: 1.1
 2.3 OpenCL C version: OpenCL C 1.1
 2.4 Parallel compute units: 2

I tried to using do some performance test on Odroid by OpenCV3.0 OpenCL API, so I wrote following code:

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

using namespace std;

int main(){
     if (!cv::ocl::haveOpenCL())
        {
            cout << "OpenCL is not avaiable..." << endl;
            return 0;
        }
        cv::ocl::Context context;
        if (!context.create(cv::ocl::Device::TYPE_GPU))
        {
            cout << "Failed creating the context..." << endl;
            return 0;
        }

    // 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;
    break;
        }
    double tic, toc;

    cv::Mat mat_src = cv::imread("img_1.png", 0);
    cv::Mat mat_dst;

    cv::UMat umat_src = mat_src.getUMat(cv::ACCESS_READ, cv::USAGE_ALLOCATE_DEVICE_MEMORY);
    cv::UMat umat_dst;

    tic = (double)cv::getTickCount();
    for(int i = 0 ; i<1000; i++){
        cv::Canny(umat_src, umat_dst, 0, 50);
    }
    toc = (double)cv::getTickCount();
    cout<<"UMat X Canny Time: "<<(double)((toc-tic)/cv::getTickFrequency())<<endl;

    tic = (double)cv::getTickCount();
        for(int i = 0 ; i<1000; i++){
                cv::Canny(mat_src, mat_dst, 0, 50);
        }
        toc = (double)cv::getTickCount();
        cout<<"Mat X Canny Time: "<<(double)((toc-tic)/cv::getTickFrequency())<<endl;

    return 0;


}

and I got following results on my Odroid:

1 GPU devices are detected.
name                 : Mali-T628
available            : 1
imageSupport         : 1
OpenCL_C_Version     : OpenCL C 1.1
UMat X Canny Time: 44.6943
Mat X Canny Time: 44.0222

As you can see, the results are merely no differences whether use UMat or not. To ensure UMat really did some effect to performance, I try the exactly same code on my PC, and I got following results:

1 GPU devices are detected.
name                 : Intel(R) HD Graphics 4600
available            : 1
imageSupport         : 1
OpenCL_C_Version     : OpenCL C 1.2

UMat X Canny Time: 4.8207
Mat X Canny Time: 11.6929

The result showed that performance was dramatically improved by OpenCL.

So, why OpenCL API not work on Odroid? Is it related to the version of OpenCL?