Question

I’m a newbie using OpenMP with C++. I was doing a simple function, loading two arrays using a for loop. These arrays are defined as complex.

#include <omp.h>
#include <iostream>
#include <stdlib.h>
#include <complex>
#define CHUNKSIZE   10
#define N       100

using namespace std;

int main (int argc, char *argv[]) 
{
int nthreads, tid, i, chunk;
complex<double> a[N], b[N], c[N];

/* Some initializations */
for (i=0; i < N; i++)
    a[i].real() = b[i].real() = i * 1.0;
chunk = CHUNKSIZE;

#pragma omp parallel shared(a,b,c,nthreads,chunk) private(i,tid)
{
    tid = omp_get_thread_num();
    if (tid == 0)
    {
        nthreads = omp_get_num_threads();
        printf("Number of threads = %d\n", nthreads);
    }
    printf("Thread %d starting...\n",tid);

#pragma omp for schedule(dynamic,chunk)
    for (i=0; i<N; i++)
    {
        c[i] = a[i] + b[i];
        printf("Thread %d: c[%d]= %e\n",tid,i,c[i]);
    }

}  /* end of parallel section */

}

When I compile, I get this warning:

omp_complex.cpp:43: warning: cannot pass objects of non-POD type ‘struct std::complex’ through ‘…’; call will abort at runtime

and If I run a.out, I get an “Illegal instruction” message on the screen. I was trying to find out what’s going on, but I didn’t find any good reference. Do anyone know if complex types are allowed in OpenMP directives?

Answer 1

The error is here:

printf("Thread %d: c[%d]= %e\n",tid,i,c[i]);

printf doesn’t know (and has no way of knowing) how to handle a std::complex. Use C++ streaming operations to output complex types.

Furthermore, in order to avoid concurrency issues, you need to stream into a thread-local buffer before you can write to stdout, otherwise the C++ streaming syntax creates race conditions.

I usually use a macro for that (but C++11 makes this easier with variadic templates):

#define THREAD_SAFE_OUT(out, message) \
    do { \
        std::ostringstream buffer; \
        buffer << message; \
        out << buffer.str(); \
    while (false)

…

THREAD_SAFE_OUT(std::cout, "Thread " << tid << ": c[" << i << "] = " << c[i]);

---

A word on style when using OpenMP:

Don’t use the private directive. This is just a workaround for languages who require all variables to be declared at the beginning of the method. Since C++ doesn’t mandate this, it’s better (always) to declare the variables on first use, i.e. inside the parallel section. That way, they are also thread-private.