OpenMP比单线程慢，即使令人尴尬的可并行化

Question

#include <stdio.h>
#include <stdlib.h>
#include <omp.h>
#include <math.h>
#include <time.h>


void func1(float *x, int n) {
    for (int ind=0; ind<n; ind++) {
        x[ind] = fabsf(x[ind]);
    }
}

void func2(float *x, int n) {
    #pragma omp parallel for
    for (int ind=0; ind<n; ind++) {
        x[ind] = fabsf(x[ind]);
    }
}


int main(void) {
    int TIMES = 10;
    int N = 1880000*128;
    float *arr = (float*)malloc(N*sizeof(float));

    float start_time1 = (float)clock() / CLOCKS_PER_SEC;
    for (int t=0; t<TIMES; t++) {
        func1(arr, N);
    }
    float duration1 = (float)clock() / CLOCKS_PER_SEC - start_time1;

    float start_time2 = (float)clock() / CLOCKS_PER_SEC;
    for (int t=0; t<TIMES; t++) {
        func2(arr, N);
    }
    float duration2 = (float)clock() / CLOCKS_PER_SEC - start_time2;

    free(arr);
    printf("duration1: %.4f\n", duration1 / TIMES);
    printf("duration2: %.4f\n", duration2 / TIMES);
    return 0;
}

我使用Ubuntu 14.04附带的GCC（4.8.4）编译：

> gcc -O3 -std=c99 -fopenmp -o test test.c -lm
> ./test
duration1: 0.1090
duration2: 0.3777

为什么func2会变慢？我在AWS中使用c4.2xlarge实例（8CPU，16GB内存）。