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 05:24:36

On 20/03/14 21:57, Naveen wrote:
> Hi,
>
> Based on understanding the proposal, I'm just curious to know how we are
> gonna use the compiler optimizations when creating the parallel_xxx
> functions. Say any normal loop would be optimized by the compiler using any
> of the loop transformations like loop interchange, loop unrolling, loop
> fusion etc.,
>
> let us consider a sample FORTRAN code to :
>
> 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
>
>
> which may or may not be optimized by the compiler into something like this
>
> DO K = 1, 100
>    DO I = 1, 100
>      DO J = 1, 100
>         C(I,J) = C(I,J) + A(I,K) * B(K,J)
>      END DO
>    END DO
> END DO

That's unlikely to happen in C++.

> so, when we try to introduce our parallel algorithm in to the loop
> containing J, it may be interchanged depending upon the compiler
> properties. So, should we plan for such scenarios.
>
> When we try to introduce our parallel functions the only way possible is to
> make the compiler optimizations temporarily inactive on those tasks. But,
> by doing so the performance of the application will be affected.

There is no need. Compiler optimizations are not allowed to change 
program behaviour, regardless of what you do.

Introducing parallelism on any of the loops would definitely prevent 
re-ordering.