topic Eigenvalue Solver Error (dfeast_scsrgv ) in Intel® oneAPI Math Kernel Library

Eigenvalue Solver Error (dfeast_scsrgv )

vahid_s_ — Mon, 09 Jun 2014 00:28:19 GMT

I'm trying to solve eigenvalue problem for large sparse matrices using dfeast_scsrgv function. The function works fine for small problems (ex: 8*8 sparse matrix) but it gives System.StackOverflowException error for larger problems( ex: 200*200 sparse matrix) . I'm using Visual Studio 2008 and MKL version 11 with most recent updates installed. My system is windows 64 bit and the programming language is C++. Following I provided the eigenvalue solver code that I'm using. In debug mode when I reach dfeast_scsrgv line it gives me Stack Overflow error. I do not think I am using any infinite loop or unnecessary large arrays. I would appreciate if someone can help me to fix the problem. Thanks!

[cpp]   
//Convert stiffness and mass matrix to CSR format - Seldon library
int NumStiff = M_GStiff.GetDataSize();   
Vector<double> V_GStiffVal   (NumStiff);    
Vector<int>    V_GStiffColInd(NumStiff);
Vector<int>    V_GStiffRowPtr(PrbDim+1);
ConvertToCSR(M_GStiff, prop, V_GStiffRowPtr, V_GStiffColInd, V_GStiffVal);  
 
int NumMass = M_GMass.GetDataSize();
Vector<double> V_GMassVal   (NumMass);
Vector<int>    V_GMassColInd(NumMass);
Vector<int>    V_GMassRowPtr(PrbDim+1);
ConvertToCSR(M_GMass, prop, V_GMassRowPtr, V_GMassColInd, V_GMassVal);   
 
//Release memory
M_GStiff.Clear();
M_GMass.Clear();
 
//Convert Seldon format to typical C array
double* a = V_GStiffVal.GetData();
int*   ia = V_GStiffRowPtr.GetData();
int*   ja = V_GStiffColInd.GetData();   
 
double* b = V_GMassVal.GetData();
int*   ib = V_GMassRowPtr.GetData();
int*   jb = V_GMassColInd.GetData();
 
// Convert matrix from 0-based C-notation to Fortran 1-based
int nnz = ia[PrbDim]; 
for (int i = 0; i < PrbDim+1; i++)        ia += 1;    
for (int i = 0; i < nnz; i++)                 ja += 1;  
 
for (int i = 0; i < PrbDim+1; i++)      ib += 1;
for (int i = 0; i < nnz; i++)              jb += 1;
 
// Initialize variables for the solver
double Error    = 0;
int    Loop     = 0;
int    NumMode  = 10;
double Emin     = 0;
double Emax     = pow(10.0,10.0);    
int    Flag     = 0;
char   MTyp     = 'U';
int    NumEigen = NumMode;
 
vector<int>    V_FPM (128,0);
vector<double> V_Eigen(NumMode,0);
vector<double> V_Res (NumMode,0);    
 
V_FPM[0]  = 1;
V_FPM[1]  = 8;
V_FPM[2]  = 12;
V_FPM[3]  = 20;
V_FPM[4]  = 0;
V_FPM[5]  = 0;
V_FPM[6]  = 5;
V_FPM[13] = 0;
V_FPM[63] = 0;
 
int*    P_FPM          = &V_FPM[0];
double* P_Eigen        = &V_Eigen[0];  
double* P_Res          = &V_Res[0];    
double dDum;
 
// Call Eigenvalue Solver
dfeast_scsrgv (&MTyp, &PrbDim, a, ia, ja, b, ib, jb,
                P_FPM, &Error, &Loop, &Emin, &Emax, &NumMode, P_Eigen, &dDum, &NumEigen, P_Res, &Flag);

[/cpp]

Can you give the reproducer

Gennady_F_Intel — Mon, 09 Jun 2014 03:16:33 GMT

Can you give the reproducer which we can compile and debug the problem on our side?

Attached is my C++ code,

vahid_s_ — Mon, 09 Jun 2014 03:40:18 GMT

Attached is my C++ code, sparse matrices example (binary file) and seldon library.

I see the failure on my side

Gennady_F_Intel — Mon, 09 Jun 2014 06:10:39 GMT

I see the failure on my side with ml v.11.1. We will debug the code and will back in the case any update.

Thanks Gennady. Would you

vahid_s_ — Mon, 09 Jun 2014 17:01:51 GMT

Thanks Gennady. Would you please let me know approximately how long it will take to fix the problem and make the new update? Thank you again for your help and time.

please see into attachment

Gennady_F_Intel — Wed, 11 Jun 2014 06:15:52 GMT

please see into attachment your code which was slightly modified. the modified part marked by --GF--

it would also make sense to

Gennady_F_Intel — Wed, 11 Jun 2014 06:17:03 GMT

it would also make sense to notice the output results I have on my side:

Intel MKL Extended Eigensolvers: double precision driver
Intel MKL Extended Eigensolvers: List of input parameters fpm(1:64)-- if different from default
Intel MKL Extended Eigensolvers: fpm(1)=1
Intel MKL Extended Eigensolvers: fpm(11)=0
Search interval [0.000000000000000e+000;1.000000000000000e+003]
Intel MKL Extended Eigensolvers: Size subspace 3
#Loop | #Eig | Trace | Error-Trace | Max-Residual
Intel MKL Extended Eigensolvers: Resize subspace 1
0,1,1.749966841512209e+000,1.000000000000000e+000,6.859928113611897e-015
Intel MKL Extended Eigensolvers have successfully converged (to desired tolerance).
dfeast_scsrgv PASSED
1.74997
END!