等待已经完成的请求(Waiting requests which have been already completed)

 我正在尝试使用boost mpi库实现异步消息传递机制（检查点机制）。 在我的代码中，接收者通过irecv等待来自其他人的消息。 在irecv之后，他们不会调用等待函数。 相反，他们称之为测试功能。 如果测试成功，他们将处理收到的消息并启动新的irecv。 否则，他们将前进到检查站。 在检查点期间，他们使用等待功能等待所有传输消息完成。 换句话说，检查点是稳定的同步点。 
 
 我简化了下面的代码并希望它很清楚： 
 
 接收方： 
 
mpiReceiveRequest = RepastProcess::instance()->getCommunicator()->irecv(1, 100, receivedData);

for(int i=0; i<=20; i++){
    if(i%5 == 0){   // checkpoints
        while(true){
            mpiReceiveRequest.wait();
            if(receivedData > 0)
                mpiReceiveRequest = world.irecv(1, 100, receivedData);
            else
                break;
        }

    }

    else{
        if(mpiReceiveRequest.test()){
            if(receivedData > 0)
                mpiReceiveRequest = world.irecv(1, 100, receivedData);
        }
    }
}
 
 发件人方： 
 
for(int i=0; i<=20; i++){
    int randomNumber = // select a random number between 1-20
    if(i > randomNumber)
        world.isend(0, 100, i);
    else
        world.isend(0, 100, -1);
}    
 
 正如您在上面所看到的，我不得不为已完成的请求调用wait函数。 我的意思是，我为已经成功测试的消息调用了wait函数。 现在，由于这个调用，问题是一个分段错误。 并且，对于boost :: mpi api中的测试函数有一些解释。 它说 
 
 
 
 你应该注意，一旦@c test（）返回一个@c状态对象，请求就完成了，并且不应该调用@c wait（）。 
 
 
 实际上，我可以存储无法通过测试功能的请求。 然后，我只能为这些请求调用wait函数。 但是，我想知道如果可以有另一个更简单的解决方案来克服这个问题。 

I am trying to implement an asynchronous messaging mechanism (a checkpointing mechanism) using boost mpi library. In my code, receivers wait a message from the others via irecv. After the irecv, they do not call wait function. Instead, they call the test function. If test is successful, they will process the received message and start a new irecv. Else, they will advance until the checkpoint. During the checkpoint, they wait all of the in transit messages to be completed using wait function. In other words, checkpoints are stable synchronization points.
 
I simplified my code below and hope it is clear:
 
Receiver side:
 
mpiReceiveRequest = RepastProcess::instance()->getCommunicator()->irecv(1, 100, receivedData);

for(int i=0; i<=20; i++){
    if(i%5 == 0){   // checkpoints
        while(true){
            mpiReceiveRequest.wait();
            if(receivedData > 0)
                mpiReceiveRequest = world.irecv(1, 100, receivedData);
            else
                break;
        }

    }

    else{
        if(mpiReceiveRequest.test()){
            if(receivedData > 0)
                mpiReceiveRequest = world.irecv(1, 100, receivedData);
        }
    }
}
 
Sender side:
 
for(int i=0; i<=20; i++){
    int randomNumber = // select a random number between 1-20
    if(i > randomNumber)
        world.isend(0, 100, i);
    else
        world.isend(0, 100, -1);
}    
 
As you can see above, I had to call wait function for completed requests. I mean, I call the wait function for an already successfully tested message. Now, problem is a segmentation fault because of this call. And, there are some explanations about test function in boost::mpi api. It says that 
 
 
 
you should Note that once @c test() returns a @c status object, the request has completed and @c wait() should not be called.
 
 
Actually, I can store the requests which could not pass the test function. Then, I can call wait function for only these requests. However, I am wondering that if there can be another simpler solution for overcoming this issue.

原文：https://stackoverflow.com/questions/31301557