Re: [boost] [Parallel Algorithms] Queries regarding the Boost.Thread / Parallel algorithms

Subject: Re: [boost] [Parallel Algorithms] Queries regarding the Boost.Thread / Parallel algorithms
From: Mathias Gaunard (mathias.gaunard_at_[hidden])
Date: 2014-03-26 05:48:49

On 24/03/14 18:00, Naveen wrote:
> Hi,
> In my previous mail, I was speaking about the compiler optimization
> dependencies while creating a parallel version of the following code:
>
>   DO J = 1, 100
>      DO I = 1, 100
>        DO K = 1, 100
>           C(I,J) = C(I,J) + A(I,K) * B(K,J)
>        END DO
>      END DO
>   END DO
>
> As you know, the above code is the 2D-Matrix multiplication logic.

While this is a pretty bad implementation of matrix multiplication, it's 
interesting because it involves multiple loops.

A good implementation would parallelize regardless of the number of 
iterations in each loop, which are all set to 100 in your example but 
which might be smaller in other cases.

> The question now is, I have used the threads as shown in the pseudo code
> below:
>
> int main()
> {
>    create_pthreads(assign_thread_ID, call the function);
>
>    join_threads(thread_ID);
>
>    destroy_threads();
> }
>
> function_called_by_each_thread(thread_ID)
> {
>    all_computations;
> }
>
> All thread documentation reaches to some sort of options like this. Is this
> the correct way to approach the problem for creating parallel algorithms
> for boost threads.

That's a basic skeleton for parallel_transform on an SMP machine.

> Please clarify, whether there are any alternative approaches available to
> achieve parallelism using threads.

There are plenty. Just look at the literature on the subject.