OpenMP: predetermined 'shared' for 'shared'?

Question

See this function (matrix-vector product):

std::vector<double> times(std::vector<std::vector<double> > const& A, std::vector<double> const& b, int m, int n) {

    std::vector<double> c;
    c.resize(n);

    int i, j;
    double sum;

    #pragma omp parallel for default(none) private(i, j, sum) shared(m, n, A, b, c)
    for (i = 0; i < m; ++i) {
        sum = 0.0;
        for (j = 0; j < n; j++) {
            sum += A[i][j] * b[j];
        }
        c[i] = sum;
    }

    return c;
}

When trying to compile this with OpenMP, the compiler fails with:

Invoking: GCC C++ Compiler
g++ -O0 -g3 -Wall -c -fmessage-length=0 -fopenmp -MMD -MP -MF"src/OpemMPTutorial.d" -MT"src/OpemMPTutorial.d" -o "src/OpemMPTutorial.o" "../src/OpemMPTutorial.cpp"
../src/OpemMPTutorial.cpp:127: warning: ignoring #pragma omp end
../src/OpemMPTutorial.cpp: In function 'std::vector<double, std::allocator<double> > times(std::vector<std::vector<double, std::allocator<double> >, std::allocator<std::vector<double, std::allocator<double> > > >&, std::vector<double, std::allocator<double> >&, int, int)':
../src/OpemMPTutorial.cpp:200: error: 'b' is predetermined 'shared' for 'shared'
../src/OpemMPTutorial.cpp:200: error: 'A' is predetermined 'shared' for 'shared'
make: *** [src/OpemMPTutorial.o] Error 1

What's wrong here?

(Note that simply removing the const results in the same error.)

Answer 1

I had a very similar problem and experienced that such a program can be compiled using Apple's GCC 4.2 after I removed the shared const variables from the shared section of OpenMP directive. They are predetermined as shared since they are constant and there is no need to make a copy for each thread. And the compiler seems to just not accept telling it explicitely when it knows already…

I would also remove the default(none) specification (but see the comment below ). OpenMP is intended to reduce explicit specifications, so let it do its job.

Answer 2

This is caused by insufficient OpenMP support in gcc-4.2. The code fragment compiles without problems using gcc-4.7.