topic Meaning of PARDISO error flags ? in Intel® oneAPI Math Kernel Library

Meaning of PARDISO error flags ?

michel_lestrade — Wed, 30 Dec 2009 20:48:38 GMT

Hi,

I read the sticky on PARDISO and I need more assistance: I'm looking to find more information about the error flags I'm getting. PARDISO seemed to work OK on the first few small problems I gave it but now that I'm moving on to a larger problem (and 64-bit) I'm getting some issues.

Our code does phase=11 followed by phase=23: the matrix does not change shape after the initial setup so we can skip 11 in subsequent runs. However, this failure occurs on the very 1st run-through (during phase 23 I think):

============================================

*** Error in PARDISO ( sequence_ido,parameters) error_num= 100000
*** Error in PARDISO: clean impossible (ido?)

================ PARDISO: solving a real nonsymmetric system ================

Summary PARDISO: ( factorize to solve )
================

Times:
======
Time total : 0.002070 s total - sum: 0.002070 s

Statistics:
===========
< Parallel Direct Factorization with #processors: > 4
< Hybrid Solver PARDISO with CGS/CG Iteration >

< Linear system Ax = b>
#equations: 543447
#non-zeros in A: 5997873
non-zeros in A (): 0.002031

#right-hand sides: 1

< Factors L and U >
#columns for each panel: 96
#independent subgraphs: 0
< Preprocessing with state of the art partitioning metis>
#supernodes: 0
size of largest supernode: 0
number of nonzeros in L 0
number of nonzeros in U 0
number of nonzeros in L+U 0
gflop for the numerical factorization: 0.000000
The following ERROR was detected in PARDISO: -1
Input inconsistent
======================================

At first, I thought there was a problem with the storage order but I have iparm(27)=1 turned on and the message doesn't seem to indicate that is the case. I printed the data and everything seems to be in the correct order.

I also read elsewhere that PARDISO requires the diagonal to be stored explicitly so I checked: I always have a non-zero element in the diagonal.

At this point, I'm not too sure what the problem actually is. To reiterate a comment on an earlier thread, the documentation could use a more thorough description of the error messages and their common causes.

Here is a list of relevant parameters I am using:

iparm(1:64)= 0
iparm(1) = 1
iparm(2) = 2
iparm(3) = mkl_get_max_threads()
iparm(4) = 61 ! this is zero during phase=11
iparm(8) = 2
iparm(10) = 13
iparm(11) = 1
iparm(13) = 1
iparm(27) = 1
mtype=11

I am using the MKL that comes with version 11.1.051 of the Fortran compiler for Windows: not sure which version that is.

Regards,

Michel Lestrade
Crosslight Software

Re: Meaning of PARDISO error flags ?

Gennady_F_Intel — Thu, 31 Dec 2009 10:03:02 GMT

- yes, the error's flag information is not competed and required a lot of improvements.
we are working on it.. ( in the form of KB article )
- *** Error in PARDISO ( sequence_ido,parameters) error_num= 100000
this error returned when the handle doesn't contains the info from the previous phase..
Can you please check it?
- "I am using the MKL that comes with version 11.1.051 of the Fortran compiler for Windows: not sure which version that is."
it's pretty easy to check the version of MKL follow the article.
--Gennady

Re: Meaning of PARDISO error flags ?

michel_lestrade — Thu, 31 Dec 2009 19:13:34 GMT

Quoting - Gennady Fedorov (Intel)

Thank you, that was helpful. You are right and phase 11 was not being done. I am tracking down the exact reason why but it seems like the flag in my code that was supposed to set this is not being passed correctly when I compile in 64 bits.

For the MKL, it looks like I'm on 10.2 Update 2. I'll check later if there's anything in Update 3 that requires an immediate update of our tools.