OpenACC Directive Optimization

OpenACC is a parallel computing programming standard co-developed by Cray, CAPS, NVIDIA, and PGI. Its primary goal is to simplify parallel programming for heterogeneous computing (CPU/GPU) systems.

Similar to OpenMP, OpenACC allows programmers to annotate specific code snippets in C, C++, and Fortran source files to indicate the targets to be accelerated using compiler directives or other functions. Like OpenMP 4.0 and later versions, OpenACC allows code to be executed on both CPUs and GPUs.

An OpenACC instruction consists of a directive and a clause.

#pragma acc loop independent

In the preceding command, #pragma acc loop is the directive, and independent is the clause. The directive is used to notify the compiler of the parallel code to be converted (function to be implemented) in the subsequent code. The clause is used to help the compiler modify the code more accurately. You can understand the specific functions of the clause in practical use.

int main()
{
    int N = 1024;
    int input_1[N][N], input_2[N][N], out[N][N];
    for (int i =  0; i < N; i++)
    {
          for (int j =  0; i < N; i++)
           {
                     input_1[i][j] = rand();
                     input_2[i][j] = rand();
           }
    }
    
    #pragma acc kernels
    for (int i =  0; i < N; i++)
    {
          for (int j =  0; i < N; i++)
           {
                    out[i][j] = input_1[i][j] + input_2[i][j];

           }
    }
}

Similar to OpenMP, OpenACC uses the #pragma acc kernels directive to enable parallel operations. Unlike OpenMP, OpenACC is designed for offloading parallel tasks to GPUs, which does not require manual data migration.