#include <iostream>
#include "cl_class.h"
#include <opencv2/opencv.hpp>
#include <opencv2/core.hpp>
#include <opencv2/core/mat.hpp>
// Maximum number of corners.
#define FAST_COUNT (1 << 18)

int main() {
    opencl gpu;
    /*
    gpu.make_kernel("int_add");
    int A[] = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 };
    int B[] = { 0, 1, 2, 0, 1, 2, 0, 1, 2, 0 };
    gpu.write_buf(0,0,10,A);
    gpu.write_buf(0,1,10,B);
    int C[10];
    gpu.run(0,10,1,1,2,10,C);
    std::cout << " result: \n";
    for (int i = 0; i < 10; i++){
        std::cout << C[i] << " ";
    }
    std::cout << std::endl;
    int A1[] = { 10, 1, 2, 3, 4, 5, 6, 7, 8, 9 };
    int B1[] = { 0, 1, 2, 0, 1, 2, 0, 1, 2, 10};
    gpu.reload_buf(0,0,A1);
    gpu.reload_buf(0,1,B1);
    int C1[10];
    gpu.run(0,10,1,1,2,10,C1);
    std::cout << " result: \n";
    for (int i = 0; i < 10; i++){
        std::cout << C1[i] << " ";
    }
    std::cout << std::endl;
    gpu.make_kernel("double_add");
    double A2[] = { 0.2, 1.3, 2.5, 3.6, 4.9, 5.12, 6.5, 7.6, 8.4, 9.1 };
    double B2[] = { 0.0, 1.0, 2.0, 0.0, 1.0, 2.0, 0.0, 1.0, 2.0, 0.0 };
    gpu.write_buf(1,0,10,A2);
    gpu.write_buf(1,1,10,B2);
    double C2[10];
    gpu.run(1,10,1,1,2,10,C2);
    std::cout << " result: \n";
    for (int i = 0; i < 10; i++){
        std::cout << C2[i] << " ";
    }
    std::cout << std::endl;
     */
    gpu.make_kernel("fast_gray");
    cv::Mat img = cv::imread("../frame0287.jpg", CV_8UC1);
    cv::Mat imageRGBA;
    cv::Mat downSampled;
    //cv::imshow("test2",img);
    //cv::waitKey();
    cv::resize(img, downSampled, cv::Size(752, 480));
    cv::cvtColor(downSampled, imageRGBA, CV_GRAY2RGBA);
    gpu.write_buf(0,0,downSampled);
    //cv::imshow("test2",downSampled);
    //cv::waitKey();
    cl_int2 corners_[740*450];
    gpu.write_buf(0,1,740*450,corners_);
    int icorner[1]={0};
    gpu.write_buf(0,2,1,icorner);
    int fast_count=FAST_COUNT;
    gpu.write_buf(0,3,fast_count);
    printf("\nOpenCL version\n");
    double time_accumulator = 0;
    for(int i=0;i<1000;++i){
        double t = (double)cv::getTickCount();
        gpu.reload_buf(0,0,downSampled);
        int icorner[1]={0};
        gpu.reload_buf(0,2,icorner);
        gpu.run(0,752,480,1);
       time_accumulator +=  ((cv::getTickCount() - t) / cv::getTickFrequency());
    }
    printf("OpenCL took %f ms (average over %d trials).\n", time_accumulator/((double)1000)*1000.0, 1000 );
    int icorner_[1]={0};
    gpu.read(0,2,icorner_);
    cl_int2 corners[740*450];
    gpu.read(0,1,corners);
    //double t = (double)cv::getTickCount();
    //cv::Mat condition=cv::Mat::ones(cv::Size(752,480),CV_8UC1);
    //gpu.run(2,752,480,1,6,360960,condition.data);

    int count=0;
    for(int i=0;i<icorner_[0];++i){
        if(corners[i].x<0 || corners[i].x>750 || corners[i].y>450 || corners[i].y<0)continue;
        cv::circle(imageRGBA, cv::Point(corners[i].x,corners[i].y ), 1, cv::Scalar(0, 0,255),1);
        ++count;
    }
    cv::imwrite("test3.png",imageRGBA);
    std::cerr<<count<<'\n';

    cv::waitKey();
    gpu.clear_buf();
    return 0;
}