Ask Your Question
0

using GPU module with own code

asked 2014-01-03 03:27:55 -0500

Farhad gravatar image

updated 2014-01-03 16:00:19 -0500

berak gravatar image

I am new to CUDA programming, and am trying to use OpenCV's GPU module with my own CUDA code, but am having problem getting it to work so was wondering if anyone here can point out what I am doing wrong.

I have made a very simple toy problem.

----------------------------------
// main.cpp 

#include <iostream>
#include <fstream>


#include "opencv2/highgui/highgui.hpp"
#include "opencv2/gpu/gpu.hpp"

#include "test.h"

using namespace std;
using namespace cv;
using namespace cv::gpu;

int main()
{
    setDevice(0);

    Mat image = imread("./testset/image_0001.png");
    Mat bw_image(image.size(), CV_32FC1); 
    cvtColor(image, bw_image, CV_RGB2GRAY);
    GpuMat  d_image(bw_image);

    GpuMat  d_image_result(d_image.size(), d_image.type() );


    test_func(d_image, d_image_result);

    Mat mmm;
    d_image_result.download(mmm);

  return 0;
}


-----------------------------------------------------------
// test.h

#ifndef __TEST__
#define __TEST__

#ifndef SKIP_INCLUDES
#include <vector>
#include <memory>
#include <iosfwd>
#endif

#include "opencv2/core/gpumat.hpp"

using namespace std;
using namespace cv;
using namespace cv::gpu;

void do_test(PtrStepSzb src,PtrStepSzb dst);

CV_EXPORTS void test_func(const GpuMat& src, GpuMat& dst)
{
  do_test(src, dst);
}

#endif /* __TEST__ */

-----------------------------------------------------------------------
// test.cu

using namespace std;
using namespace cv;
using namespace cv::gpu;

__global__ void do_test_kernel(PtrStepSz<float> src, PtrStepSz<float> dst)
{

    int x = threadIdx.x + blockIdx.x * blockDim.x;
    int y = threadIdx.y + blockIdx.y * blockDim.y;

    if(x<dst.cols && y < dst.rows)
        {dst.ptr(y)[x] = src.ptr(y)[x];}
}

void    do_test(const PtrStepSz<float>& src, PtrStepSz<float>& dst)
{
    dim3 block(32,8);
    dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y));

    do_test_kernel<<<grid,block>>>(src,dst);
    cudaSafeCall( cudaGetLastError() );

    cudaSafeCall( cudaDeviceSynchronize() );
};

-------------------------------------------------------------

Now running the code explained above, results in this runtime error:

OpenCV Error: Gpu API call (unspecified launch failure) in caller, file /.../opencv-2.4.7/modules/gpu/src/cuda/matrix_reductions.cu, line 437 terminate called after throwing an instance of 'cv::Exception' what(): /.../opencv-2.4.7/modules/gpu/src/cuda/matrix_reductions.cu:437: error: (-217) unspecified launch failure in function caller

Can anyone please tell me what is the problem with my code?

edit retag flag offensive close merge delete

1 answer

Sort by ยป oldest newest most voted
2

answered 2014-01-03 10:39:59 -0500

Vladislav Vinogradov gravatar image

Your images have CV_8UC1 type. Even if you declare bw_image as CV_32FC1 it will have CV_8UC1 type, because image has CV_8UC3 type and cvtColor doesn't change depth. Thus cvtColor recreate bw_image with CV_8UC1 type. You mast use unsigned char in your kernel instead of float or explisitly convert d_image to CV_32FC type by convertTo method.

edit flag offensive delete link more

Comments

Thank you, that solved my problem.

Farhad gravatar imageFarhad ( 2014-01-03 11:46:50 -0500 )edit

I am unable to compile this program. I get the following error ->undefined reference to `do_test(cv::gpu::PtrStepSz<unsigned char>, cv::gpu::PtrStepSz<unsigned char>)' in the header file

zpot gravatar imagezpot ( 2014-04-06 17:25:32 -0500 )edit

Question Tools

1 follower

Stats

Asked: 2014-01-03 03:27:55 -0500

Seen: 2,311 times

Last updated: Jan 03 '14