问题描述
我正在阅读的教科书中发现了这个问题.解决方案也在下面给出.我无法理解最小值可能是2的原因.为什么一个线程不能读取0，所有其他线程都执行并且写入1?而无论是1还是2，最后写入的线程仍必须完成自己的循环吗?
I found this question in a textbook I am reading. The solution is given below it as well. I'm having trouble understanding how the minimum could be 2. Why couldn't a thread read 0, all other threads execute and it writes 1? And whether it is 1 or 2, the thread writing last must still complete its own loop?
int n = 0;
int main(int argc, char **argv) {
for (i = 0; i < 5; i++) {
int tmp = n;
tmp = tmp + 1;
n = tmp;
}
return 0;
}
解决方案:
推荐答案
假设每个线程都有一个本地i(即，每个线程无论如何都将运行5次迭代)，让我们尝试获得1作为结果.这意味着 last 线程要写入值，必须在第5 次迭代中为n读取0.发生这种情况的唯一方法是，如果该线程的第5次迭代开始时尚未将任何线程写入n，但要使该线程处于其第5次迭代，该线程本身必须已写入n，因此它是不可能.
Assuming every thread has a local i (i.e., every thread will run for 5 iterations no matter what), let's try to get 1 as the result. This would mean the last thread to write a value would have to read 0 for n on its 5th iteration. The only way this could happen is if no thread has yet written to n at the start of that thread's 5th iteration, yet for that thread to be on its 5th iteration that thread itself must have written to n, hence it is not possible.
因此，可能的最小结果是2，它可能会发生，例如，如下所示:最后一个写入n的线程已完成4次迭代，然后另一个线程写入了1，最后一个线程在其开始处读取了1.第5次迭代，所有其他线程在最后一个线程之前完成所有迭代，最后最后一个线程完成第5次迭代，写入2.
Thus the smallest possible result is 2, which can occur, e.g., as follows: the last thread to write n has completed 4 iterations, then another thread writes 1, the last thread reads the 1 at the start of its 5th iteration, all other threads complete all their iterations before the last thread, and finally the last thread completes its 5th iteration writing the 2.
免责声明:我正在回答关于多线程的概念问题-正如其他人所指出的那样，如果没有C语言，缺乏原子性可能会导致不确定的行为和任意结果提出的按原样使用.基于问题的不言而喻"最大案例，我猜测教科书的作者可能没有意识到这一点，或者正在使用类似C的伪代码来说明这一概念.如果是前者，那么正确的答案将是该书是错误的，但我认为后者的答案也是有教育意义的.
Disclaimer: I am answering the conceptual question about multithreading – as others have pointed out, the lack of atomicity might lead to undefined behaviour and arbitrary results if the C code presented were used as is. Based on the question's "self-evident" largest number case I'm guessing the textbook's author either doesn't realise this, or is using a C-like pseudo code to illustrate the concept. If the former, then the correct answer would be that the book is wrong, but I think the answer in the latter case is also educational.
这篇关于多线程访问一个变量的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持！