program crashed during calling Pardiso() with intel fortran 11.1.035 on win32 with VS2005

[joey_hylton]

1. !C -Prep. for PARDISO
2. PT=0
3. MAXFCT=1
4. MNUM=1
5. MMTYPE=13
6. IPARM(1)=0
7. IPARM(3)=omp_get_max_threads()
8. IPARM(3)=1
9. MSGLVL=0
11. !C -Compute the PML parameters and build the PML matrix
12. CALL BUILD_PML(N1,N2,DN1,DN2)
14. C -Build the source term
15. Q_VEC=0
16. CALL SOURCE_VEC(F,DN1,DN2,Q_VEC)
18. CALL STIFF_MAT(F,DN1,DN2,NOE,NNZ,IA,JA,A_PARD)
20. !C -Try PARDISO
21. PHASE=12
22. CALL PARDISO(PT,MAXFCT,MNUM,MMTYPE,PHASE,NOE,A_PARD,IA,JA,
23. & PERM,NRHS,IPARM,MSGLVL,DDUM,DDUM,ERROR)
25. !C -Solve
26. DO SRC_INDEX=1,LS
27. Q_VEC1=Q_VEC(:,SRC_INDEX)
29. PHASE=33
30. NRHS=1
31. CALL PARDISO(PT,MAXFCT,MNUM,MMTYPE,PHASE,NOE,A_PARD,IA,JA,
32. & PERM,NRHS,IPARM,MSGLVL,Q_VEC1,S_VEC1,ERROR)
33. S_VEC(:,SRC_INDEX)=S_VEC1
34. ENDDO
36. WRITE(*,*) '*********************************'
37. WRITE(14,*) '*********************************'
38. WRITE(17,*) LS

I have a fortran code goes like above for calling pardiso() in with intel fortran 11.1.035

In x64 machine with 8GB memory, it goes ok. But for a 32bit machine with 2GB memory, it always has problems.

PROBLEM 1:

I have case which has LS=30 (line 26). When I debug it, the SRC_INDEX can go up to 13, then crashed by give me some error information:

*** Error in PARDISO ( sequence_ido,parameters) error_num= 2

*** Input check: max_fac_store_new 2327064 _old 1 are incompatible

*** Input parameters: inconsistent error= 2 max_fac_store_in: 2327064

matrix_number_in : 811400488 matrix_type_in : 11161568

ido_in : 33 neqns_in : 12444

ia(neqns_in+1)-1 : 60444 nb_in : 1

PT value:

*******************************************
1 2384432 0 825434024 0 0
0 843382848 0 0 0 832962624
825434096 825533704 0 0 2385440 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0

PROBLEM 2:

I have a case with LS=1, then the program crashed at line 36 by giving me these information:

forrtl: severe (8): internal consistency check failure, file for_put.c, line 898
Image PC Routine Line Source
program.exe 0066EE0A Unknown Unknown Unknown
program.exe 0066AB08 Unknown Unknown Unknown
program.exe 005C15C0 Unknown Unknown Unknown
program.exe 005D6E9B Unknown Unknown Unknown
program.exe 005BA1DB Unknown Unknown Unknown
program.exe 005B8A39 Unknown Unknown Unknown
program.exe 00429A52 _MAIN__ 36 mycode.for
program.exe 00A1E233 Unknown Unknown Unknown
program.exe 0060DE03 Unknown Unknown Unknown
program.exe 0060DBCD Unknown Unknown Unknown
kernel32.dll 7C817067 Unknown Unknown Unknown

PT value:

*******************************************
1 2289104 0 818806664 0 0
0 831324224 0 0 0 820904000
2293656 818806800 0 0 818806736 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0

In release mode, case 2 can finish without error. But I am not sure the results is correct or not.

Libraries I used:

libguide.lib
mkl_core.lib
mkl_intel_s.lib
mkl_intel_thread.lib
mkl_solver.lib

[Vladimir_Lunev]

Hello joey_hilton,

I see that you supposes using CVF calling convention (mkl_intel_s.lib). What compiler options were used to compile your fortran code?

Thanks,
-Vladimir

[Gennady_F_Intel]

Joey,
Do you have any updates on the issue you encountered?
Can you please send us a reproducible test case?
This tread is the private, so you don't care about privacy
--Gennady

[joey_hylton]

Quoting - Gennady Fedorov (Intel)

Joey,
Do you have any updates on the issue you encountered?
Can you please send us a reproducible test case?
This tread is the private, so you don't care about privacy
--Gennady

Gennady and Vladimir,

Sorry for the late reply. I solved this problem by changing the link libraries as #4 in Gennady's suggestions, and following this in the userguide:

10.(For 32-bit)Static linking of myprog.f, sequential version of an iterative sparse solver, and
sequential Intel MKL supporting cdecl interface:
ifort myprog.f mkl_solver_sequential.lib mkl_intel_c.libmkl_sequential.lib mkl_core.lib