[OpenCV] GpuMat and Mat, compare cvtColor perforemence

Introduction

I am going to measure the performence of my two GT650M and compare GPU with CPU version.

code segments

#include <iostream>
#include <omp.h>
using namespace std;

#include <cuda_runtime.h>

#include <opencv2/opencv.hpp>
#include <opencv2/gpu/gpu.hpp>
using namespace cv;

#define TB__(A) int64 A; A = cv::getTickCount()
#define TE__(A) cout << #A << " : " << 1.E3 * double(cv::getTickCount() - A)/double(cv::getTickFrequency()) << "ms" << endl

Vec3b max(cv::Mat &A, cv::Mat &B){
    int width = A.cols;
    int height = A.rows;
    Vec3b me(0,0,0);
    for (int y = 0; y < height; y++){
        for (int x = 0; x < width; x++){
            Vec3b p0 = A.at<Vec3b>(y,x);
            Vec3b p1 = B.at<Vec3b>(y,x);
            Vec3b pixel;
            pixel[0] = p1[0]>p0[0]? p1[0]-p0[0]:p0[0]-p1[0];
            pixel[1] = p1[1]>p0[1]? p1[1]-p0[1]:p0[1]-p1[1];
            pixel[2] = p1[2]>p0[2]? p1[2]-p0[2]:p0[2]-p1[2];
            if (me[0] < pixel[0])
                me[0] = pixel[0];
            if (me[1] < pixel[1])
                me[1] = pixel[1];
            if (me[2] < pixel[2])
                me[2] = pixel[2];
        }
    }
    return me;
}

#define REPEATES 10000

int main(){
  // please ensure that you have at least two Nvidia GPUs on your mathorboad.
    cv::Mat cpu_src = imread("86.jpg");
    cv::Mat cpu_dst;

    cv::Mat gpu_dst[2];

    TB__(cpu_cvt);
#pragma omp parallel for num_threads(4)
    for (int k = 0; k < REPEATES; k++)
        cv::cvtColor(cpu_src, cpu_dst, CV_BGR2Lab);
    TE__(cpu_cvt);

    TB__(gpu_cvt);
#pragma omp parallel num_threads(2)
    {
        int id = omp_get_thread_num();
        cv::gpu::setDevice(id);

        cv::gpu::GpuMat dev_src;
        cv::gpu::GpuMat dev_dst;

        dev_src.upload(cpu_src);

        //#pragma omp master
        //TB__(gpu_omp_dev);

        #pragma omp for
        for (int k = 0; k < REPEATES; k++){
            dev_src.upload(cpu_src);
            cv::gpu::cvtColor(dev_src, dev_dst, CV_BGR2Lab);
            dev_dst.download(gpu_dst[id]);
        }
        cudaDeviceSynchronize();
    }
    TE__(gpu_cvt);

  cout << cv::sum(cpu_dst - gpu_dst[0]) << endl;
  cout << cv::sum(cpu_dst - gpu_dst[1]) << endl;
  cout << cv::sum(gpu_dst[0] - gpu_dst[1]) << endl;

    cout << endl;

  cout << max(cpu_dst, gpu_dst[0]) << endl;
  cout << max(cpu_dst, gpu_dst[1]) << endl;
    return 0;
}