打破循环for循环

Question

#include<iostream>
#include<fstream>
#include<time.h>
#include<omp.h>

using namespace std;
static long num_steps = 100;
#define NUM 8
double step;

void main()
{
    clock_t time =clock();
    ofstream result;
    result.open ("Result.txt");
    int a[100];
    double pi, sum=0.0; 
    step = 1.0/(double) num_steps;

    #pragma omp parallel num_threads(NUM)
    {           
        int i, ID;    
        double x, psum= 0.0; 
        int nthreads = omp_get_num_threads();
        ID = omp_get_thread_num();   
        for (i=ID;i<= num_steps; i+=nthreads)
        {
            x = (i+0.5)*step;
            psum += 4.0/(1.0+x*x);
        }
        #pragma omp critical
        sum += psum;
    }

    pi = step * sum; 
    for (int j=0;j<100;j++)
    result<<a[j]<<endl;

    time = clock() - time;

    result << "Time Elapsed: " << (((double)time)/CLOCKS_PER_SEC) << endl;

    result <<"======================================================================================="<<endl;
    result.close();
}

问题是:按以下顺序执行以下for (i=ID;i<= num_steps; i+=nthreads)循环: 01234567、01234567、01234567等等。分配的任务是更改for循环，使线程均匀分布，而不是以四舍五入的方式分配。先是零，然后是一，然后是二...那么我该如何改变forloop呢？

Answer 1

为此，您必须使用某种线程同步...

你给Visual studio加了标签，所以我假设Windows平台...

最近，这成了我的最爱：

// init
CRITICAL_SECTION hnd;
InitializeCriticalSectionAndSpinCount(&hnd,0x00000400);

// start lock
EnterCriticalSection(&hnd);
// stop lock
LeaveCriticalSection(&hnd);

// exit
DeleteCriticalSection(&hnd);

但是还有很多其他的方法。

• 你也可以尝试使你自己的锁或无锁线程
• ，但要知道，在像Windows7这样的较新的操作系统中，进程调度程序是不同的进程调度程序( process sheduler
• )，并且倾向于使用疯狂的进程调度程序(

)< code >H110，我的意思是100%工作的无锁代码在以前的OS-es上是断断续续或冻结的

• ，所以我更喜欢使用操作系统锁。

如果您错误地使用锁，则可能会失去多线程加速的任何好处。

如果你只是担心你的解决方案不能同时计算线程

在你的例子中不是并行的，而是串行的，而不是由以下原因引起的：

1. 处理时间granularity.

- any sheduled task is divided to chunks of time. 
- If your task is too short then it is done sooner then the other task even begin execution. 
- to test that try bigger payload (compute time > few seconds)
- enlarge number of cycles greatly
- add Sleep(time ms) to have longer computation time
- if the output will be mixed then it was it
- if not then you are still under granularity boundary
- or your multi-thread code is wrong

​

1. 错误的多线程代码

- are you shore your threads are created/running at the same time ?
- or do you synchronize to something wrong ? (like till the end of previous task)
- also some compilers do a big deal of volatile variables (add locks to it what sometimes do very weird things ... I stumped on it many times but mostly on MCU platforms and Eclipse)

​

1. 单核

- on some cases if you have just 1 CPU/Core/Computer for processing
- or just setted affinity mask to single CPU
- on some algorithms windows shedulers do not shedule the CPU time evenly
- even regardless the process/thread priority/class
- something similar appears sometimes on Windows 7 even for more CPUs ...
- especially with code mixed with Kernel mode code

​

要使用粒度，你可以使用他的：

// obtain OS time capabilities
TIMECAPS tim; 
timeGetDevCaps(&tim,sizeof(tim));

// set new granularity
if (timeBeginPeriod(time ms)!=TIMERR_NOERROR) log("time granularity out of range");

// return to previous hranularity
timeEndPeriod(time ms ... must be the same as beginperiod);out of range");

PS。关于这一点的非常好的东西在这里：

http://bitflipgames.com/2011/05/09/multithreaded-programming-part-1-the-critical-section-lock/ http://bitflipgames.com/2011/05/17/multithreaded-programming-part-2-multiple-readersingle-writer-lock/ http://bitflipgames.com/2011/05/20/multithreaded-programming-part-2-5-mrsw-lock-code/ http://bitflipgames.com/2011/05/25/multithreaded-programming-part-3-going-lockless/