Problem with offload code ( dlopen() and undefined symbol: _ZTVN2cv11_InputArrayE )

Hello,

i tried to generete a code with intrinsics, but I having some problems with my code.

So, for publish in here, i generated only a simple code that causes the same problem.

I have three files:

main.cpp

#include <time.h>
#include <sys/time.h>
#include <opencv2/opencv.hpp>
#include <iostream>
#include "secondaryFunction.h"

static inline long long timestampInUS();

using namespace cv;
using namespace std;
int main(int argc, char **argv) {

        if (argc != 3) {
                cout << "Usage: ./processImage <im1> <im2>\n"<< endl;
                return 0;
        }

        Mat image1 = imread(argv[1], -1);
        Mat image2 = imread(argv[2], -1);

        Mat recon;
        uint64_t t1, t2;

        t1 = timestampInUS();
        recon = internalFunctionTwo<unsigned char>(image1, image2, 8);
        t2 = timestampInUS();
        cout << "Time: "<< t2 - t1 << " us"<< endl;
        imwrite("finalImage.jpg", recon);
        return 0;

}

static inline long long timestampInUS() {
        struct timeval ts;
        gettimeofday(&ts, NULL);
        return (ts.tv_sec * 1000000LL + (ts.tv_usec));
}

secondaryFunction.h

#ifndef SECONDARYFUNCTION_H_
#define SECONDARYFUNCTION_H_
#include "opencv/cv.hpp"

using namespace cv;
template <typename T>
cv::Mat internalFunctionTwo(const cv::Mat& image1, const cv::Mat& image2, int typeImage);

#endif

and secondaryFunction.cpp

#include <iostream>
#include <limits>
#include <opencv/cv.h>
#include "secondaryFunction.h"
#include <immintrin.h>

using namespace cv;
using namespace std;
template<typename T>
inline void internalFunctionOne(Mat& imageOne, Mat& imageTwo, int* out,
                int &iniO, int &endO, int &size, int typeImage) {

#pragma offload target(mic)
        {
                __m512i pixels, pval12, vectmp;
                __mmask16 p1mask, p2mask, finalMask;
//              p1mask = _mm512_int2mask(0xFF00);
//              p2mask = _mm512_int2mask(0x00FF);
// ...
        }

}

template<typename T>
Mat internalFunctionTwo(const Mat& image1, const Mat& image2, int typeImage) {

        Mat imageOut(image1.size() + Size(2, 2), imageOut.type());
        copyMakeBorder(image1, imageOut, 1, 1, 1, 1, BORDER_CONSTANT, 0);
        Mat imageIn(image2.size() + Size(2, 2), imageIn.type());
        copyMakeBorder(image2, imageIn, 1, 1, 1, 1, BORDER_CONSTANT, 0);

        int size = 10000;//test

        int* in = new int[size];
        int iniI = 0;
        int endI = 0;

        int* out = new int[size];
        int iniO = 0;
        int endO = 0;

        internalFunctionOne<T>(imageOut, imageIn, out, iniO, endO, size, typeImage);
        return imageOut;

}

template Mat internalFunctionTwo<unsigned char>(const Mat& image1,const Mat& image2, int typeImage);
template Mat internalFunctionTwo<float>(const Mat& image1, const Mat& image2,int typeImage);

The compilation is done with the following lines:

CXX     =       icpc
OPENCV_PATH     =       ${WORK}/opencv/opencv-2.4.5/install

AR      =       xiar
AR_FLAGS        =       -qoffload-build rcs

OPENCV_FLAGS    =       -I${OPENCV_PATH}/include/opencv -I${OPENCV_PATH}/include
OTHER_FLAGS     =       -fPIC -O2 -offload-attribute-target=mic -offload-option,mic,compiler,"-O2 -fma"

CXXFLAGS = ${OPENCV_FLAGS} ${OTHER_FLAGS}
CXXLIBS = ${OPENCV_PATH}/lib/libopencv_calib3d.so ${OPENCV_PATH}/lib/libopencv_contrib.so ${OPENCV_PATH}/lib/libopencv_core.so ${OPENCV_PATH}/lib/libopencv_features2d.so ${OPENCV_PATH}/lib/libopencv_flann.so ${OPENCV_PATH}/lib/libopencv_gpu.so ${OPENCV_PATH}/lib/libopencv_highgui.so ${OPENCV_PATH}/lib/libopencv_imgproc.so ${OPENCV_PATH}/lib/libopencv_legacy.so ${OPENCV_PATH}/lib/libopencv_ml.so ${OPENCV_PATH}/lib/libopencv_nonfree.so ${OPENCV_PATH}/lib/libopencv_objdetect.so ${OPENCV_PATH}/lib/libopencv_ocl.so ${OPENCV_PATH}/lib/libopencv_photo.so ${OPENCV_PATH}/lib/libopencv_stitching.so ${OPENCV_PATH}/lib/libopencv_superres.so ${OPENCV_PATH}/lib/libopencv_ts.so ${OPENCV_PATH}/lib/libopencv_video.so ${OPENCV_PATH}/lib/libopencv_videostab.so -lrt -lpthread -lm -ldl

TARGET = program

SOURCE = $(TARGET).cpp secondaryFunction.cpp

TARGET: libMO.a main.o
        ${CXX} ${CXXFLAGS} main.o libMO.a -o program ${CXXLIBS}

libMO.a: main.o secondaryFunction.o
        ${AR} ${AR_FLAGS} libMO.a secondaryFunction.o

main.o: main.cpp secondaryFunction.h
        ${CXX} ${CXXFLAGS} main.cpp -c

secondaryFunction.o: secondaryFunction.cpp
        ${CXX} ${CXXFLAGS} secondaryFunction.cpp -c

clean:
        rm -f *.o *.a *~ $(TARGET)

When I try to execute the program, I receive this below error:

host-1303$ ./program image01.jpg image02.jpg
On the remote process, dlopen() failed. The error message sent back from the sink is /tmp/coi_procs/1/3637/load_lib/icpcoutqtHLqK: undefined symbol: _ZTVN2cv11_InputArrayE
offload error: cannot load library to the device 0 (error code 20)
On the sink, dlopen() returned NULL. The result of dlerror() is "/tmp/coi_procs/1/3637/load_lib/icpcoutqtHLqK: undefined symbol: _ZTVN2cv11_InputArrayE"

This error occurs only when I put some offload code in secondaryFunction.cpp. I was thinking that this problem was in OpenCV libraries, but without her lines doesn't works too. Now I checking of xiar and compilation lines.