cluster_sparse_solver and distributed format

cluster_sparse_solver and distributed format

Hello,

I'm trying to use cluster_sparse_solver and solve a system in-place (iparm(6) = 1), with a distributed format (iparm(40) = 2). I adapted the example cl_solver_unsym_distr_c.c as you can see attached, and at runtime, on two MPI processes, I get the following output:

The solution out-of-place of the system is: 
 on zero process x [0] =  0.263109       rhs [0] =  1.000000
 on zero process x [1] =  0.305243       rhs [1] =  1.000000
 on zero process x [2] = -0.347378       rhs [2] =  0.250000

The solution out-of-place of the system is: 
 on first process x [0] = -0.347378      rhs [0] =  0.750000
 on first process x [1] =  0.205993      rhs [1] =  1.000000
 on first process x [2] =  0.288390      rhs [2] =  1.000000

Solving system in-place...
The solution in-place of the system is: 
 on zero process x [0] =  1.000000
 on zero process x [1] =  1.000000
 on zero process x [2] =  0.250000

The solution in-place of the system is: 
 on first process x [0] =  0.750000
 on first process x [1] =  1.000000
 on first process x [2] =  1.000000

Can you reproduce this behavior ? The solution in-place is obviously wrong. Do you see how to fix that ? Thank you in advance.

AttachmentSize
Download cl_solver_unsym_distr_c.c12.7 KB
7 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.

Hello,

Thanks for your report.  Our engineer owner could verify this is problem, and they are planning to fix it in our near future release.

Regards,
Chao

This problem is fixed and the fix would be available into the nearest update 1 -  MKL v.11.2.  We will make the announcement of this release on the Top of the forum. 

With the new update, I'm now getting errors that I can't reproduce on a simple program such as:

*** 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: 159 kbyte

The thing is that the matrix is really small (36 x 36 on 8 processes), and I'm turning iterative refinement off, so I don't understand why is PARDISO trying to allocate memory, and why it is failing (I have 6 GB of RAM on the machine I'm getting these errors on).

Here is the exact same test case as above, except this time, the RHS and the solution are centralized while the matrix is distributed. Clearly the problem is still here. Can you reproduce this problem on your side ?

Thank you.

$ icc -v
icc version 15.0.1 (gcc version 4.9.0 compatibility)
$ mpicc -cc=icc cl_solver_unsym_distr_c.c -lmkl_intel_thread -lmkl_core -lmkl_intel_lp64 -liomp5
$ mpirun -np 2 ./a.out
The solution out-of-place of the system is:
 on zero process x [0] =  0.149579   rhs [0] =  1.000000
 on zero process x [1] =  0.259831   rhs [1] =  1.000000
 on zero process x [2] = -0.370084   rhs [2] =  0.250000
 on zero process x [3] =  0.011236   rhs [3] =  1.000000
 on zero process x [4] =  0.415730   rhs [4] =  1.000000

Solving system in-place...
The solution in-place of the system is:
 on zero process x [0] =  0.149579
 on zero process x [1] =  0.259831
 on zero process x [2] = -0.370084
 on zero process x [3] =  1.000000
 on zero process x [4] =  1.000000

Attachments: 

AttachmentSize
Download cl_solver_unsym_distr_c.c12.09 KB

Could you confirm that you can reproduce this new issue please ? The original issue is fixed though. Thank you.

It has been three weeks since I posted this new issue, but I still don't have a clear answer, could you please let me know if the problem is on my side ?

Thank you.

Login to leave a comment.