PARDISOMEX_dke_yp

Link with the parallel matrix solver package PARDISO written in C/C++
for MEX file (single processor).  

Solve the linear system of equation AX=B

 INPUTS:

    - A: Matrix [n,n]
   - B: Vector for solving the linear system of equations AX=B [n,1]  
   - job: PARDISO job type (1,2,3,5,6,-2) see MUMPS documentation
  
 OUTPUTS:

   - X: Solution of the linear system of equation [n,1]
   - flag: execution flag (see the PARDISO documentation) [1,1]

By Yves Peysson (CEA/DSM/IRFM, yves.peysson@cea.fr) and Joan Decker (CEA/DSM/IRFM, joan.decker@cea.fr) and

0001 %
0002 %Link with the parallel matrix solver package PARDISO written in C/C++
0003 %for MEX file (single processor).
0004 %
0005 %Solve the linear system of equation AX=B
0006 %
0007 % INPUTS:
0008 %
0009 %    - A: Matrix [n,n]
0010 %   - B: Vector for solving the linear system of equations AX=B [n,1]
0011 %   - job: PARDISO job type (1,2,3,5,6,-2) see MUMPS documentation
0012 %
0013 % OUTPUTS:
0014 %
0015 %   - X: Solution of the linear system of equation [n,1]
0016 %   - flag: execution flag (see the PARDISO documentation) [1,1]
0017 %
0018 %By Yves Peysson (CEA/DSM/IRFM, yves.peysson@cea.fr) and Joan Decker (CEA/DSM/IRFM, joan.decker@cea.fr) and
0019 %
0020 verbose_PARDISO = false;
0021 %
0022 if job_PARDISOMEX == 1,
0023     % Initialize the PARDISO internal data structures. We've told PARDISO to
0024     % handle real non-symmetric matrices using the sparse direct solver.
0025     %
0026     info_PARDISO = pardisoinit(11,0);
0027 %    info_PARDISO.iparm(3) = 6;
0028 %    info_PARDISO.iparm(4) = 61;
0029 %    info_PARDISO.iparm(8) = 10;
0030     %
0031 elseif job_PARDISOMEX == 2,
0032     %
0033     A_PARDISOMEX = MMXPC_f_t;
0034     %
0035     % Analyze the matrix and compute a symbolic factorization.
0036     %
0037     info_PARDISO = pardisoreorder(A_PARDISOMEX,info_PARDISO,verbose_PARDISO);
0038     %fprintf('The factors have %d nonzero entries.\n',info_PARDISO.iparm(18));
0039     %
0040     % Compute the numeric factorization.
0041     %
0042     info_PARDISO = pardisofactor(A_PARDISOMEX,info_PARDISO,verbose_PARDISO);
0043     %
0044 elseif job_PARDISOMEX == 3,
0045     %
0046     B_PARDISOMEX = MMXR_t;
0047     %
0048     % Compute the solutions X using the symbolic factorization.
0049     %
0050     [X_PARDISOMEX,info_PARDISO] = pardisosolve(A_PARDISOMEX,B_PARDISOMEX,info_PARDISO,verbose_PARDISO);
0051     %
0052     fprintf('PARDISO performed %d iterative refinement steps.\n',info_PARDISO.iparm(7));        
0053     %
0054     % Compute the residuals.
0055     R_PARDISOMEX = max(abs(A_PARDISOMEX*X_PARDISOMEX - B_PARDISOMEX));
0056     fprintf('The maximum residual for the solution X is %0.3g.\n',max(R_PARDISOMEX(:)));
0057     flag_PARDISOMEX = 0;
0058 elseif job_PARDISOMEX == -2,
0059     % Free the PARDISO data structures.
0060     pardisofree(info_PARDISO);
0061     % clear info
0062 end 
0063     
0064     
0065     
0066     
0067     
0068     
0069     
0070