Common used OpenMP situations / 常见OpenMP处理场景

[crazyzlj]

Classified by the characteristics of calculation

1. Independent calculation / 计算独立型

Simply add a #pragma omp parallel for before the for-loop.

#pragma omp parallel for
for (int i = 0; i < m_nCells; i++) {
    DoSomething(i)
}

2. Dependent calculation / 计算依赖型

The calculation of current computation unit is dependent on other units, e.g., the upstream units.

for (int iLayer = 0; iLayer < m_nRoutingLayers; iLayer++) {
    // There are not any flow relationship within each routing layer.
    // So parallelization can be done here.
    int nCells = (int) m_routingLayers[iLayer][0];
#pragma omp parallel for
    for (int iCell = 1; iCell <= nCells; ++iCell) {
        int id = (int) m_routingLayers[iLayer][iCell]; // cell index
        DoSomething(id);
    }
}

3. Reduce operation / 规约型计算

Using openmp for reduction an array should be paid much more attention, e.g., when summarizing raster data according to subbasin ID.

Here is a solution. https://stackoverflow.com/questions/20413995/reducing-on-array-in-openmp
It is worth to notice that #pragma omp parallel for reduction(+:myArray[:6]) is supported with OpenMP 4.5. However, as far as I know, currently, MSVC 2010-2015 are using OpenMP 2.0.

Below is a code snippet to solve this issue in current SEIMS.

#pragma omp parallel
    {
        float *tmp_qiSubbsn = new(nothrow) float[m_nSubbasin + 1];
        for (int i = 0; i <= m_nSubbasin; i++) {
            tmp_qiSubbsn[i] = 0.f;
        }
#pragma omp for
        for (int i = 0; i < m_nCells; i++) {
            tmp_qiSubbsn[int(m_subbasin[i])] = SummarizeSomething();
        }
#pragma omp critical
        {
            for (int i = 1; i <= m_nSubbasin; i++) {
                m_qiSubbasin[i] += tmp_qiSubbsn[i];
            }
        }
        delete[] tmp_qiSubbsn;
        tmp_qiSubbsn = nullptr;
    } /* END of #pragma omp parallel */

Q&A

1. Can I use OpenMP nest parallel for loops?

If your compiler supports OpenMP 3.0, you can use the collapse clause:

#pragma omp parallel for schedule(dynamic,1) collapse(2)
for (int x = 0; x < x_max; x++) {
    for (int y = 0; y < y_max; y++) { 
    // parallelize this code here
    }
// IMPORTANT: no code in here
}

If it doesn't (e.g. only OpenMP 2.5 is supported), there is a simple workaround:

#pragma omp parallel for schedule(dynamic,1)
for (int xy = 0; xy < x_max*y_max; xy++) {
    int x = xy / y_max;
    int y = xy % y_max;
    //parallelize this code here
}

You can enable nested parallelism with omp_set_nested(1); and your nested omp parallel for code will work but that might not be the best idea.

So, when the omp-embeded functions (e.g., Initialize1DArray and Initialize2DArray) are invoked in your code, please put outside the omp parallel for loop, or use new and delete[] instead.

• References:
  • https://msdn.microsoft.com/en-us/library/sk3zt8e1.aspx
  • https://stackoverflow.com/questions/10540760/openmp-nested-parallel-for-loops-vs-inner-parallel-for