Feature Request: OpenMP/TBB like Parallel For Loops

[steveklabnik]

Issue by siavashserver
Friday Feb 28, 2014 at 11:14 GMT

For earlier discussion, see rust-lang/rust#12619

This issue was labelled with: A-libs in the Rust repository

---

Is it possible to have some easy to use parallel for loops like OpenMP or Intel TBB in Rust? As far as I know their parallel for loops divide the range by CPU cores (or specified worker threads), then every worker thread will access a sequential stream of data in memory (better CPU cache utilization) and idle worker threads are also able to steal tasks from busy ones to balance the load (which is optional in OMP, and the default for the TBB).

Serial code:

void SerialApplyFoo( float b[], const float a[], size_t n ){
    for( size_t i=0; i<n; ++i )
        b[i] = Foo(a[i]);
}

Intel TBB example:

void ParallelApplyFoo (float b[], const float a[], size_t n){
    tbb::parallel_for (size_t(0), n, [=](size_t i) {
        b[i] = Foo(a[i]);
    });
}

OpenMP example:

void ParallelApplyFoo( float b[], const float a[], size_t n ){
    #pragma omp parallel for
    for( size_t i=0; i<n; ++i )
        b[i] = Foo(a[i]);
}

[Ericson2314]

New Send should make this easy to do out of tree.

[aturon]

cc @alexcrichton

[jnicholls]

This is possible with scoped threads, so probably already satisfied with simple_parallel + crossbeam no?

Would love to see more things enter into std.

[kamalmarhubi]

Also relevant is @nikomatsakis' rayon:

Rayon is a data-parallelism library for Rust. It is extremely lightweight and makes it easy to convert a sequential computation into a parallel one. It also guarantees data-race freedom.

Recent blog post: http://smallcultfollowing.com/babysteps/blog/2015/12/18/rayon-data-parallelism-in-rust/

[JayKickliter]

@steveklabnik can this be closed?

[steveklabnik]

yes