多线程伪共享(false sharing)问题分析

#include<stdio.h> 
#include<omp.h> 
#include<string.h> 
#include<time.h>
#include<stdlib.h>
#include<iostream>
using namespace std;



#define NMAX 4096 * 10 
#define NUM_CORE 4 

int g_iBuff[NMAX];


int main()
{
    int alignPos = 0;
    for (int i = 0; i < 65; ++i)
    {
        int * addr = &g_iBuff[i];
        char buf[32];
        int int_addr = atoi(buf);
        if ( int_addr % 64 == 0)
        {
            alignPos = i;
            cout<<"alignPos:"<<int_addr<<endl;
            break;
        }
    }

    for (int pos = 1; pos <= 4096; pos <<= 1)
    {
        memset(g_iBuff, 0 , sizeof(g_iBuff));
        time_t beg = time(0);
#pragma omp parallel for num_threads(NUM_CORE) 
        for (int i = 0; i < NUM_CORE; ++i)
        {
            int tp = i * pos + alignPos;
            for (int j = 0; j < 999999999; ++j)
            {
                g_iBuff[tp]++;
            }
        }
        time_t end = time(0);

        cout << "false shareing: step = "<<pos<<" time = "<< end - beg << "s" <<endl;
    }

    return 0;
}

# g++ XX.cpp -o xx -fopenmp

# ./xx 

我们首先来看一下这段代码的输出结果：

false shareing: step = 1 time = 12s
false shareing: step = 2 time = 13s
false shareing: step = 4 time = 13s
false shareing: step = 8 time = 6s
false shareing: step = 16 time = 3s
false shareing: step = 32 time = 3s
false shareing: step = 64 time = 3s
false shareing: step = 128 time = 3s
false shareing: step = 256 time = 3s
false shareing: step = 512 time = 3s
false shareing: step = 1024 time = 3s
false shareing: step = 2048 time = 3s
false shareing: step = 4096 time = 3s

接下来我们来分析一个程序：