Multithreading, tbb, local variable problem

idleman · 12-05-2009, 08:53 AM

Hi all,
I Just have just getting started to use intels Threading Building Blocks, but it just took some minutes before I got some problem. If I am running the code below will m_type.size() always return 0 even if i am pushing new data in a second thread:

Code:

#include <string>
#include <iostream>
#include <boost/thread.hpp>
#include "tbb/concurrent_vector.h"


class Test
{
public:
	Test(tbb::concurrent_vector<std::string> & t)
		//: m_type(t) 
	{
	}
	void worker()
	{
		std::cout << "Producing now..";
		while(true)
		{	
			if(m_type.size() < 1000)
			{
				m_type.push_back(std::string("Mohaha"));
				std::cout << "worker: size = " << m_type.size() << std::endl;
			}
		}
	}

	void consumer()
	{
		std::cout << "Consuming now..";
		while(true)
		{
			std::cout << "Consumer: size = " << m_type.size() << std::endl;
			if(m_type.size())
			{
				std::string message = m_type.back();
				std::cout << "Message: " << message << std::endl;
				if(m_type.size() > 100)
				{
					m_type.clear();
				}
			}
		}
	}
private:
	tbb::concurrent_vector<std::string> m_type;
};


int main(int argc, char** argv)
{
	tbb::concurrent_vector<std::string> vec;
	Test test(vec);


	boost::thread thread(boost::bind(&Test::worker, test));
	boost::thread thread2(boost::bind(&Test::consumer, test));
	thread2.join();
	thread.join();
}

But if I am making m_type to a reference (&) will it work without problems.

I know that local variables can be read into a processor cache and that is what I think happens, but how do I prevent it? And if it isn't the problem, what I am doing wrong then?

I second question:
Do any of you have any good multi threading links to share?

Thanks in advance

C_ntua · 12-05-2009, 11:21 AM

Can you post the exact output of the program?

idleman · 12-05-2009, 12:00 PM

Exactly? No, it is different every time I run it of course.

The program acts like it got 2 tbb::concurrent_vector<std::string> separated for each thread, even if it just got one. Even if the worker method pushes new data to the vector does the consumer never see it. It is like it is 2 different variables separated from each other. Worker thread says m_type.size() have size (x > 0) and consumer always 0. If I am in another way make m_type to a reference will it work, probably because the compiler cannot make some optimizes and if I am trying to make the variable of type volatile I just got compile errors.

I have search little and I probably need a memory barrier that force the processor to clean its L1/L2 cache and register every time, but I always have used boost::mutex before this so I don't really know how I fix it. I could add a mutex (mutex force memory barrier as I know), but what should be the point to use a concurrent container then?

abachler · 12-05-2009, 12:07 PM

You need to use critical sections to prevent simultaneous access to a modifiable object.

idleman · 12-05-2009, 12:16 PM

What do you mean? It is a concurrent container, it is thread safe to access from multiple threads, both read and write it says in the documentation.

**CornedBee** · 12-06-2009, 03:08 PM

Give Test a private copy constructor, and you'll see the problem (or not, given bind's error messages): Boost.Bind binds its arguments as copies. In other words, each of these threads has its private Test object, with its own private vector. You need to bind by reference, which you can achieve by using Boost's reference_wrapper:

Code:

	boost::thread thread(boost::bind(&Test::worker, boost::ref(test)));
	boost::thread thread2(boost::bind(&Test::consumer, boost::ref(test)));

idleman · 12-06-2009, 03:30 PM

Thanks CornedBee, I never thought about it

12-05-2009, 11:21 AM	#2
C_ntua Registered User Join Date: Jun 2008 Posts: 1,526	Can you post the exact output of the program?
C_ntua is offline

12-05-2009, 12:00 PM	#3
idleman Registered User Join Date: Feb 2009 Posts: 40	Exactly? No, it is different every time I run it of course. The program acts like it got 2 tbb::concurrent_vector<std::string> separated for each thread, even if it just got one. Even if the worker method pushes new data to the vector does the consumer never see it. It is like it is 2 different variables separated from each other. Worker thread says m_type.size() have size (x > 0) and consumer always 0. If I am in another way make m_type to a reference will it work, probably because the compiler cannot make some optimizes and if I am trying to make the variable of type volatile I just got compile errors. I have search little and I probably need a memory barrier that force the processor to clean its L1/L2 cache and register every time, but I always have used boost::mutex before this so I don't really know how I fix it. I could add a mutex (mutex force memory barrier as I know), but what should be the point to use a concurrent container then?
idleman is offline

12-05-2009, 12:07 PM	#4
abachler Malum in se Join Date: Apr 2007 Posts: 3,188	You need to use critical sections to prevent simultaneous access to a modifiable object. __________________ Until you can build a working general purpose reprogrammable computer out of basic components from radio shack, you are not fit to call yourself a programmer in my presence. This is cwhizard, signing off.
abachler is offline

12-05-2009, 12:16 PM	#5
idleman Registered User Join Date: Feb 2009 Posts: 40	What do you mean? It is a concurrent container, it is thread safe to access from multiple threads, both read and write it says in the documentation.
idleman is offline

12-06-2009, 03:08 PM	#6
CornedBee Cat without Hat Join Date: Apr 2003 Posts: 8,579	Give Test a private copy constructor, and you'll see the problem (or not, given bind's error messages): Boost.Bind binds its arguments as copies. In other words, each of these threads has its private Test object, with its own private vector. You need to bind by reference, which you can achieve by using Boost's reference_wrapper: Code: boost::thread thread(boost::bind(&Test::worker, boost::ref(test))); boost::thread thread2(boost::bind(&Test::consumer, boost::ref(test))); __________________ All the buzzt! CornedBee "There is not now, nor has there ever been, nor will there ever be, any programming language in which it is the least bit difficult to write bad code." - Flon's Law
CornedBee is offline

12-06-2009, 03:30 PM	#7
idleman Registered User Join Date: Feb 2009 Posts: 40	Thanks CornedBee, I never thought about it
idleman is offline

Thread Tools
Show Printable Version Email this Page
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Variable Allocation Problem	bavetta	C Programming	3	11-14-2009 12:52 AM
Problem when trying to return union type variable from function	limp	C Programming	3	10-13-2009 08:49 AM
VERY strange problem. My program hangs unless I add a test variable and never use it.	rasheemo	C Programming	2	10-10-2009 10:38 PM
Warning : returns address of a local variable YET correct output	tapti	C Programming	5	08-14-2009 03:15 AM
Need some help...	darkconvoy	C Programming	32	04-29-2008 03:33 PM