Ongoing Optimization: Relaxed Semantic with CppMem

With the relaxed semantic, we have no synchronisations and ordering constraints on atomic operations.

Relaxed semantic

With the relaxed semantic, there is only the atomicity of the operations on atomics left.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25

// ongoingOptimizationRelaxedSemantic.cpp

#include <atomic>
#include <iostream>
#include <thread>

std::atomic<int> x{0};
std::atomic<int> y{0};

void writing(){  
  x.store(2000,std::memory_order_relaxed);  
  y.store(11,std::memory_order_relaxed);
}

void reading(){  
  std::cout << y.load(std::memory_order_relaxed) << " ";  
  std::cout << x.load(std::memory_order_relaxed) << std::endl;
}

int main(){
  std::thread thread1(writing);
  std::thread thread2(reading);
  thread1.join();
  thread2.join();
};

Now, the questions are very easy to answer. Does the program have well defined behaviour? Which values for x and y are possible? At the one hand, all operations on x and y are atomic. So the program is well defined. On the other hand, there are no restrictions on the interleavings of the threads. In the end, thread 2 can see the operations on thread 1 in different order. So this is the first time in our process of ongoing optimizations, that thread 2 can display x == 0 and y == 1. All combinations of x and y are possible.

 I'm curious, how the graph of CppMem will look like for x == 0 and y == 1?

CppMem

int main(){
  atomic_int x= 0;
  atomic_int y= 0;
  {{{ { 
      x.store(2000, memory_order_relaxed);
      y.store(11,memory_order_relaxed);
      }
  ||| {
      y.load(memory_order_relaxed);
      x.load(memory_order_relaxed);
      }
  }}}
}

That was the CppMem program. Now to the graph.

Execution for (y=0,x=2000)

The graph shows crystal clear the unintuitive behaviour.

x reads the value 2000 from the writing thread, but y reads the value 0 from the main thread. That happens, although the reading of y is sequenced before the reading of x. Sequenced before exactly means, that the operation e:Rrix sb  is sequenced-before the operation f:Rrix.

What's next?

This was the last post in my mini series about ongoing optimization. So, what's next? There are a lot of issues with the singleton pattern. I'm totally aware of that. But the singleton pattern is an ideal use case for a variable, which has to be initialized in a thread safe way. From that point on, you can use it without synchronization.
So in the next post, I discuss different ways to initialize a singleton in a multithreading envrironment. You get the performance numbers and can reason about your uses cases for the thread safe initialization of a variable.