Re: [Boost-users] Beginning with thread

Subject: Re: [Boost-users] Beginning with thread
From: Philipp Kraus (philipp.kraus_at_[hidden])
Date: 2010-08-13 09:43:58

On 2010-08-13 15:21:29 +0200, Benjamin Sobotta said:

> On Friday 13 August 2010 13:46, Kraus Philipp wrote:
>> Hi,
>> 
>> I would like to use Boost threads, but I don't work with them. I know
>> only Java Threads.
>> 
>> The problem, in which should use threads:
>> I have some stochastic and numerical optimization, which all can
>> calculate independent (no mutex, synchronizsation or something else).
>> 
>> I've found boost::thread and boost::bind for creating threads (call a
>> method of an object).
>> 
>> In Java I would write this code structure:
>> 
>> myThread first = new myThread();
>> first.setParameter1(....)
>> first.setParameter1(....)
>> 
>> Thread th = new Thread(first);
>> th.start();
>> 
>> first.getData()
>> 
>> 
>> I would like to have an object of calculation, which receives only
>> after creating different parameters and then calculates the data on
>> more than one thread.
>> After completion of the calculation, I can let me deliver the data
>> back from any object, like:
>> 
>> std::vector<myWorker> x;
>> for(int i=0; i < numThreads; ++i) {
>> x.push_back( myWorker() );
>> x[i].setParameter(....);
>> }
>> 
>> createThreadGroup (x);
>> wait until ThreadGroup is finish
>> 
>> for(int i=0; i < numThreads; ++i)
>> x[i].getData()
>> 
>> How can do this with Boost Thread? Is this the correct way or can I do
>> this more efficient?
>> 
>> Thanks
>> 
>> Phil
> 
> Hi!
> 
> Ok if I understand this right you want one thread per worker object?! I would
> do something like

Yes. Each Object (Worker) = one Thread

> 
> std::vector<myWorker> x;
> for(int i=0; i < numThreads; ++i) {
>        x.push_back( myWorker() );
>        x[i].setParameter(....);
> }
> 
> boost::thread_group threads;
> 
> for(int i=0; i < numThreads; ++i)
> {
> 	threads.create_thread(boost::bind(&myWorker::compute_stuff), x[i]);
> }
> 
> threads.join_all();
> 
> // retrieve information from the workers
> 
> That would send a thread into each myWorker with entry point compute_stuff().
> At least in theory. Right now I've no way / time of testing a setup like
> this.

I will test it and report

Thanks

Phil