timing arrayfire

add.cpp

#include <iostream>
#include <limits>
#include <arrayfire.h>

using namespace std;
using namespace af;

int comp(const void * elem1, const void * elem2) {
    int f = *((int*)elem1);
    int s = *((int*)elem2);
    if (f > s) return  1;
    if (f < s) return -1;
    return 0;
}

void addOnDevice(const af::array lhs,const  af::array rhs, int count) {
    double totalTime = 0;
    af::timer start = af::timer::start();
    for (int idx = 0; idx < count; idx++) {
        af::array result = lhs + rhs;
        result.eval();
    }
    af::sync();
    totalTime = af::timer::stop(start);
    printf("function addOnDevice took an average of  %g seconds to add %d elements after %d iterations\n", totalTime/count, lhs.elements(), count);
}

void addOnHost(int* lhs, int* rhs, int n, int count) {
    int* result = (int *)malloc(sizeof(int) * n);

    double totalTime = 0;

    af::timer start = af::timer::start();
    for (int idx = 0; idx < count; idx++) {
        for (int i = 0; i < n; i++) {
            result[i] = lhs[i] + rhs[i];
        }
    }
    totalTime += af::timer::stop(start);
    free(result);
    printf("function addOnHost took an average of %g seconds to add %d elements after %d iterations\n", totalTime/count, n, count);
}

int main()
{
    af::deviceset(0);
    af::info();

    int n = 50e6;
    int iterations = 10;

    // allocate and populate the host data
    int* hdata1 = (int *)malloc(sizeof(int) * n);
    for (int i = 0; i < n; i++) {
        hdata1[i] = int(rand() % 20);
    }

    int* hdata2 = (int *)malloc(sizeof(int) * n);
    for (int i = 0; i < n; i++) {
        hdata2[i] = int(rand() % 20);
    }
    // copy the same data to the devicw
    af::array ddata1(n, 1, hdata1, af::afHost);
    af::array ddata2(n, 1, hdata2, af::afHost);
    af::sync();

    addOnHost(hdata1, hdata2, n, iterations);
    addOnDevice(ddata1, ddata2, iterations);

    // manage memory
    free(hdata1);
    free(hdata2);
}