Cluster Pardiso error for iparm(24)=1

[Serban_G_]

Hello all,

Last week I have been trying out Cluster Pardiso, testing various options and so on, and I found a problem. Cluster Pardiso fails with various error messages when I enable two-level factorization (that is iparm(24)=1).

The error can be easily reproduced in the following way:

1. Edit cl_solver_sym_sp_0_based_c.c example and add "iparm[23] = 0" at line 112 (or around there)

2. Recompile: mpiicc -mkl -mt_mpi cl_solver_sym_sp_0_based_c.c -o cl_solver_sym_sp_0_based_c

3. Run: export OMP_NUM_THREADS=2; mpirun -np 6 ./cl_solver_sym_sp_0_based_c

Here is the output:

*****************************************************************************************

=== CPARDISO: solving a symmetric indefinite system ===
0-based array is turned ON
CPARDISO single precision computation is turned ON
METIS algorithm at reorder step is turned ON
Matching is turned ON

Summary: ( reordering phase )
================

Times:
======
Time spent in calculations of symmetric matrix portrait (fulladj): 0.001182 s
Time spent in reordering of the initial matrix (reorder)         : 0.000001 s
Time spent in symbolic factorization (symbfct)                   : 0.002049 s
Time spent in data preparations for factorization (parlist)      : 0.000005 s
Time spent in allocation of internal data structures (malloc)    : 0.005455 s
Time spent in additional calculations                            : 0.000024 s
Total time spent                                                 : 0.008716 s

Statistics:
===========
Parallel Direct Factorization is running on 6 MPI and 2 OpenMP per MPI process

< Linear system Ax = b >
             number of equations:           8
             number of non-zeros in A:      18
             number of non-zeros in A (%): 28.125000

             number of right-hand sides:    1

< Factors L and U >
             number of columns for each panel: 64
             number of independent subgraphs:  0
< Preprocessing with state of the art partitioning metis>
             number of supernodes:                    4
             size of largest supernode:               4
             number of non-zeros in L:                31
             number of non-zeros in U:                1
             number of non-zeros in L+U:              32

Reordering completed ... *** Error in PARDISO  (     insufficient_memory) error_num= 10
*** Error in PARDISO memory allocation: SOLVING_ITERREF_WORK_DATA, allocation of 1 bytes failed
total memory wanted here: 144 kbyte

=== CPARDISO: solving a symmetric indefinite system ===
Two-level factorization algorithm is turned ON

Summary: ( factorization phase )
================

Times:
======
Time spent in copying matrix to internal data structure (A to LU): 0.000000 s
Time spent in factorization step (numfct)                        : 0.000000 s
Time spent in allocation of internal data structures (malloc)    : 0.000029 s
Time spent in additional calculations                            : 0.000109 s
Total time spent                                                 : 0.000138 s

Statistics:
===========
Parallel Direct Factorization is running on 6 MPI and 2 OpenMP per MPI process

< Linear system Ax = b >
             number of equations:           8
             number of non-zeros in A:      18
             number of non-zeros in A (%): 28.125000

             number of right-hand sides:    1

< Factors L and U >
             number of columns for each panel: 64
             number of independent subgraphs:  0
< Preprocessing with state of the art partitioning metis>
             number of supernodes:                    4
             size of largest supernode:               4
             number of non-zeros in L:                31
             number of non-zeros in U:                1
             number of non-zeros in L+U:              32
             gflop   for the numerical factorization: 0.000000

*****************************************************************************************

With the original example eveyrhing works out alright.
Also, on other matrices I got different types of errors - every time I enable the two-level factorization feature the solver fails.

Am I doing something wrong?

Any ideeas would be highly appreciated!

Thank you in advance,

Serban

[Gennady_F_Intel]

hello Serban,

please look at the documentation for cluster_sparse_solver iparm Parameter description - you will see the follow:

iparm(22) - iparm(26)

Reserved. Set to zero

--Gennady

[Serban_G_]

Hi Gennady,

Thanks a lot for the very quick reply.

In the documentation that I found here https://software.intel.com/en-us/node/521691 I did not found that information. It is written that iparm(26)  is reserved, but I got the impression that  iparm(24) can be used.

Am I looking in the wrong place?

Cheers,

Serban

[mecej4 · Accepted solution]

The link that Gennady gave is for the cluster parallel direct solver. The link in #3 is not. In #1 you said that you used the cluster version of the solver.

[Serban_G_]

Hi macej4,

Indeed, following your comment I've realized the I've been looking at the wrong documentation. Now all makes sense, that option was wrong.

Thank you and to Gennady for your help!

Serban