OpenCV对OpenCL的支持

最近调查OpenCV能否使用SoC内置的GPU进行图像处理加速，所采用的SoC是NXP的i.MX6Q，GPU型号是Vivante GC2000。从SoC的年代来说，这款GPU的配置还是不错的，其中：

1. 1个GPGPU Core
2. 4个Shader Core

3. 32GFLOPS

   不过，GPU支持的OpenCL标准只到了OpenCL 1.1 EP。
   在OpenCV中，打开了OpenCL支持之后，运行出错，代码如下：

#include <opencv2/opencv.hpp>
#include <unistd.h>

using namespace cv;
using namespace std;

#define USE_T_API   1

/*
 * ===  FUNCTION  ======================================================================
 *         Name:  example
 *  Description:
 * =====================================================================================
 */
#ifndef USE_T_API
void example( Mat & img )
#else
void example( UMat & img )
{
#endif
#ifndef USE_T_API
    Mat _gray, _canny;
#else
    UMat _gray, _canny;
#endif
    double t = (double)getTickCount();

    cvtColor( img, _gray, COLOR_BGR2GRAY );
    GaussianBlur( _gray, _gray, Size(5, 5), 3, 3 );
    Canny( _gray, _canny, 10, 100, 3, true );

    t = ((double)getTickCount() - t)/getTickFrequency();
    cout << "Times passed in seconds: " << t << endl;

    namedWindow( "Gray", WINDOW_AUTOSIZE );
    namedWindow( "Canny", WINDOW_AUTOSIZE );

    imshow( "Gray", _gray );
    imshow( "Canny", _canny );
}       /* -----  end of function example  ----- */
/*
 * ===  FUNCTION  ======================================================================
 *         Name:  main
 *  Description:
 * =====================================================================================
 */
int main ( int argc, char *argv[] )
{
#ifndef USE_T_API
    Mat _img    =   imread( argv[1], -1 );
#else
    UMat _img;

    imread( argv[1], -1 ).copyTo( _img );
#endif

    if( _img.empty() ) return -1;

    example( _img );

    waitKey( 0 );

    return 0;
}               /* ----------  end of function main  ---------- */

调查原因，在OpenCV网站上，找到如下对OpenCL模块的说明：

Some tidbits:

 1. OpenCL version should be larger than 1.1 with FULL PROFILE.
 2. Currently there’s only one OpenCL context and command queue. We hope to implement multi device and multi queue support in the future.
 3. Many kernels use 256 as its workgroup size if possible, so the max work group size of the device must larger than 256. All GPU devices we are aware of indeed support 256 workitems in a workgroup, however non GPU devices may not. This will be improved in the future.
 4. ...

OpenCV对OpenCL要求的最低要求是v1.1以上的Full版本，所以针对i.MX6的OpenCV采用OpenCL加速的调查，可以暂时结束了。

另外，考虑到我是要做实时图像处理，即使支持OpenCL加速，那么OpenCV和OpenCL之间数据传递的延时，也可能不满足要求，这个之后条件成熟了再调查。

更新：
部分OpenCV接口在i.MX6上，可以基于OpenCL 1.1 EP执行，但性能与CPU执行相比更慢，原因不明。