測試下預取的效果，利用CPU始終查看效果。根據實驗發現，預取地址在地址使用之前的十行左右代碼處效果比較好！

代碼如下:
#include <stdio.h>

#define MAX_LEN 1000000

static inline void prefetchnta(void *addr) //預取部分
{
    __asm__("movl %0, %%eax"::"a"(addr));
    __asm__(".byte 0x0f, 0x18, 0x00");
}

inline unsigned long long GetCPUTickCount()
{
    unsigned long high32 = 0;
    unsigned long low32 = 0;

    __asm__("RDTSC" : "=a"(low32), "=d"(high32));

    unsigned long long counter = high32;
    counter = (counter<<32) + low32;

    return counter;
}

int main(int argc, char* argv[])
{
    long long start, end;
    long long array[MAX_LEN];
    int i;

    for(i = 0; i < MAX_LEN； i++) //讓cache失效
        array[i]++;

    start = GetCPUTickCount();
    array[0]++;
    end = GetCPUTickCount();
    printf("don't use prefetch time:%ld/n", end - start);

    for(i = 0; i < MAX_LEN； i++)
        array[i]++;

    prefetchnta(array);
    start = GetCPUTickCount();
    array[0]++;
    end = GetCPUTickCount();
    printf("use prefetch time:%ld/n", end - start);

    return 0;
}