// N is size of physical grid from which the sparse system is derived
// diag is the value of each diagonal element of the sparse matrix
// recipdiag is the reciprocal of each diagonal element
// odiag is the value of each off-diagonal non-zero element in matrix
// eps is the threshold for the convergence criterion
// nprfreq is the number of iterations per output of currwnt residual
// xnew and xold are used to hold the N*N guess solution vector components
// resid holds the current residual
// rhoinit, rhonew hold the initial and current residual error norms, resp.
int main (int argc, char * argv[])
{
double b[(N+2)*(N+2)];
double xold[(N+2)*(N+2)];
double xnew[(N+2)*(N+2)];
double resid[(N+2)*(N+2)];
double rhoinit,rhonew;
int i,j,iter,u;
omp_set_num_threads(atoi(argv[1]));
init(xold,xnew,b);
rhoinit = rhocalc(b);
clkbegin = rtclock();
for(iter=0;iter<maxiter;iter++){
update(xold,xnew,resid,b);
rhonew = rhocalc(resid);
if(rhonew<eps){
clkend = rtclock();
t = clkend-clkbegin;
printf("Solution converged in %d iterations\n",iter);
printf("Final residual norm = %f\n",rhonew);
printf("Solution at center and four corners of interior N/2 by N/2 grid : \n");
i=(N+2)/4; j=(N+2)/4; printf("xnew[%d][%d]=%f\n",i,j,xnew[i*(N+2)+j]);
i=(N+2)/4; j=3*(N+2)/4; printf("xnew[%d][%d]=%f\n",i,j,xnew[i*(N+2)+j]);
i=(N+1)/2; j=(N+1)/2; printf("xnew[%d][%d]=%f\n",i,j,xnew[i*(N+2)+j]);
i=3*(N+2)/4; j=(N+2)/4; printf("xnew[%d][%d]=%f\n",i,j,xnew[i*(N+2)+j]);
i=3*(N+2)/4; j=3*(N+2)/4; printf("xnew[%d][%d]=%f\n",i,j,xnew[i*(N+2)+j]);
printf("Sequential Jacobi: Matrix Size = %d; %.1f GFLOPS; Time = %.3f sec; \n",
N,13.0*1e-9*N*N*(iter+1)/t,t);
break;
}
copy(xold,xnew);
if((iter%nprfreq)==0)
printf("Iter = %d Resid Norm = %f\n",iter,rhonew);
}
}
int main (int argc, char * argv[])
{
double * b = malloc(sizeof(double)*(N+2)*(N+2));
double * xold = malloc(sizeof(double)*(N+2)*(N+2));
double * xnew = malloc(sizeof(double)*(N+2)*(N+2));
double * resid = malloc(sizeof(double)*(N+2)*(N+2));
double rhoinit,rhonew;
int i,j,iter,u;
init(xold,xnew,b);
rhoinit = rhocalc(b);
clkbegin = rtclock();
for(iter=0;iter<maxiter;iter++){
// This can of course not be parallelized. The number of iterations will be the same no matter how we divide the work up amongst the processors.
update(xold,xnew,resid,b);
rhonew = rhocalc(resid);
if(rhonew<eps){
clkend = rtclock();
t = clkend-clkbegin;
printf("Solution converged in %d iterations\n",iter);
printf("Final residual norm = %f\n",rhonew);
printf("Solution at center and four corners of interior N/2 by N/2 grid : \n");
i=(N+2)/4; j=(N+2)/4; printf("xnew[%d][%d]=%f\n",i,j,xnew[i*(N+2)+j]);
i=(N+2)/4; j=3*(N+2)/4; printf("xnew[%d][%d]=%f\n",i,j,xnew[i*(N+2)+j]);
i=(N+1)/2; j=(N+1)/2; printf("xnew[%d][%d]=%f\n",i,j,xnew[i*(N+2)+j]);
i=3*(N+2)/4; j=(N+2)/4; printf("xnew[%d][%d]=%f\n",i,j,xnew[i*(N+2)+j]);
i=3*(N+2)/4; j=3*(N+2)/4; printf("xnew[%d][%d]=%f\n",i,j,xnew[i*(N+2)+j]);
printf("Sequential Jacobi: Matrix Size = %d; %.1f GFLOPS; Time = %.3f sec; \n",
N,13.0*1e-9*N*N*(iter+1)/t,t);
break;
}
copy(xold,xnew);
if((iter%nprfreq)==0)
printf("Iter = %d Resid Norm = %f\n",iter,rhonew);
}
}
