在 openMP 程序中嵌套 for 循环耗时太长

Question

I'm having a problem with parallelizing my program with openMP. The first for loop takes about 10 milliseconds, but the second takes about 45 seconds. I'm not sure if I'm just doing something wrong in the loop that is wasting time.

float A[M][M];
float B[M][M];
float C[M][M];

main(int argc, char** argv) {
float temp;
float real;
float error = 0;
int i,j,k;
double time_start;
double time_end;
double time_mid;
int n  = 12;

omp_set_num_threads(n);
time_start = omp_get_wtime();


#pragma omp parallel default(shared) private(i,j,k,temp,real) reduction(+:error)
#pragma omp for
for (i=0; i<M; i++) {
        for (j=0; j<M; j++) {
                A[i][j] = ((i+1)*(j+1))/(float)M;
                B[i][j] = (j+1)/(float)(i+1);
        }
}

time_mid = omp_get_wtime();
#pragma omp for
for (i=0; i<M; i++) {
        for (j=0; j<M; j++) {
                temp = 0;
                for (k=0; k<M; k++) {
                        temp += A[i][k]*B[k][j];
                }
            C[i][j] = temp;
            real =(float) (i+1)*(j+1);
            error = error + (float) fabs(temp-real)/real;

}
}


time_end = omp_get_wtime();
error = (100/(float)(M*M))*error;

printf("Percent error for C[][] is: %f\n", error);
printf("Time is: %f\n%f\n%f\n%f\n", time_end-time_start, time_start, time_mid, time_end);

return 0;
}

Answer 1

From OpenMP specifications (page 35, 2.1 Directive Format C/C++)
https://www.openmp.org/wp-content/uploads/openmp-4.5.pdf

An OpenMP executable directive applies to at most one succeeding statement, which must be a structured block.

The definition of a block in C++ is stmt.block

Therefore #pragma omp parallel default(shared) private(i,j,k,temp,real) reduction(+:error) will only apply to the first block (your first for loop)

The other loops are not in a ' #pragma omp parallel ' context.

Use #pragma omp parallel{} to enclose the second loop.