OpenMP and tensor_eval()

[andrea-afk]

I have problems parallelizing some part of code with OpenMP. This is an example:

  #include <iostream>
  #include <omp.h>
  #include <adolc/adolc.h> 
  #include <adolc/adolc_openmp.h>

int main(){
	int max_t = omp_get_num_threads();
double x[max_t];
double y[max_t];
#pragma omp parallel for ADOLC_OPENMP_NC 
    for(int i = 0; i < 10000; ++i)
{
	int numt = omp_get_thread_num();

	double** tensor2 = myalloc(1,3);
	double** S = myalloc(1,1);
	S[0][0] = 1.0;

   		adouble a_y, a_x;

	trace_on(numt);
	a_x <<= x[numt];
	a_y = pow(2*a_x,3);
	a_y >>= y[numt];
	trace_off();

	double o = 1.0*i;
	

//		function(numt,1,1,&o,&y[numt]);
//		zos_forward(numt,1,1,1,&o,&y[numt]);
//		fos_reverse(numt,1,1,&o,&y[numt]);
	
	tensor_eval(numt,1,1,2,1,&o,tensor2,S);
	std::cout  << tensor2[0][2] << std::endl;
//	std::cout  << y[numt] << std::endl;
}
return 0;
}

When executing with #pragma omp parallel for ADOLC_OPENMP_NC it gives me Segmentation fault (core dumped) , no problem with function() or zos_forward() and fos_reverse() . Thanks for your patience.

OS: Ubuntu 19.10 Adol-C version: tested with v2.7.2

[awalther1]

Hi,

thanks for your message. I will have a look at this as soon as possible.

Best

Andrea