Memory allocation failed in SuperLU_MT?

[SalmanMaths]

I am using the SuperLU_MT and the code from pdrepeat.c (https://github.com/xiaoyeli/superlu_mt/blob/master/EXAMPLE/pdrepeat.c). In this code the function dreadhb(&m, &n, &nnz, &a, &asub, &xa); is used. My matrix format is CRS and transform it into CCS by dCompRow_to_CompCol(m, n, nnz, const_cast<double>(_sk.data()), const_cast<int_t>(_ik.data()), const_cast<int_t*>(_id.data()), &a, &asub, &xa); And then use the dCreate_CompCol_Matrix(&A, m, n, nnz, a, asub, xa, SLU_NC, SLU_D, SLU_GE); to create the Supermatrix A.

When I run the code, it gives me the following error:

** On entry to sp_ienv, parameter number 1 had an illegal value Storage for L subscripts exceeded; Current column 0; Need at least 72; You may set it by the 8-th parameter in routine sp_ienv(). Memory allocation failed at line 222 in file pmemory.c

Any suggestions about this to handle the error?

[xiaoyeli]

It's difficult to tell what went wrong. You will need to do some debugging, for example, by printing out some info at various places of the source code.

[SalmanMaths]

The problem occurred to me at the position/function ( LU factorization of A): / ------------------------------------------------------------ Compute the LU factorization of A. The following routine will create nprocs threads. ------------------------------------------------------------/ pdgstrf(&superlumt_options, &AC, perm_r, &L, &U, &Gstat, &info);

[SalmanMaths]

It is must, to use the dreadhb(&m, &n, &nnz, &a, &asub, &xa)? instead of it, I use the following (it works in SuperLU):

// Initialize matrix A. m = n = f.size();//f.size is the number of rows of the matrix nnz = _sk.size(); if ( !(a = doubleMalloc(nnz)) ) SUPERLU_ABORT("Malloc fails for a[]."); if ( !(asub = intMalloc(nnz)) ) SUPERLU_ABORT("Malloc fails for asub[]."); if ( !(xa = intMalloc(n+1)) ) SUPERLU_ABORT("Malloc fails for xa[]."); dCompRow_to_CompCol(m, n, nnz, const_cast<double>(_sk.data()), const_cast<int_t>(_ik.data()), const_cast<int_t>(_id.data()), &a, &asub, &xa); / Set up the sparse matrix data structure for A. */ dCreate_CompCol_Matrix(&A, m, n, nnz, a, asub, xa, SLU_NC, SLU_D, SLU_GE);

[xiaoyeli]

I don't see anything wrong in your matrix setup routine. I suggest you go into the SuperLU source code, do some printing at various places, see whether your input matrix is correct. In SRC/pdutil.c, therre are a number of printing functions, such as: dPrint_CompCol_Matrix(), etc.

[SalmanMaths]

fp.txt I did this and printed the CCS matrix in the attached file.

[SalmanMaths]

Dear Sherry Li, Thank you so much for your support. I think my problem is from the not correct reading of matrix structure. Now I am going to save my matrix in a .hb file then I will call from dreadhb() function. So I hope it will work.

But I need your help to define the .hb file correctly. I defined it below but it is not working. Please help me in this (Particularly in the first two lines: fprintf(fp,"CompCol matrix %s\n", what); //fprintf(fp,"Stype %d, Dtype %d, Mtype %d\n", A->Stype,A->Dtype,A->Mtype); n = A->ncol; Astore = (NCformat ) A->Store; dp = (double ) Astore->nzval; fprintf(fp,"RUA"); fprintf(fp," %d %d %d ", A->nrow,A->ncol,Astore->nnz);//nrow, ncol, a, asub, xa, fprintf(fp,"0\n");):

`template void FEM_Matrix_2::Save_Matrix(std::vector const &f) const { SuperMatrix A1; SuperMatrix A; double a; int asub, xa; int m, n1, nnz; NCformat Astore; int i,n; double dp; char what; FILE fp; // // Initialize matrix A. m = n1 = f.size(); // number of rows of square matrix. nnz = _sk.size(); if ( !(a = doubleMalloc(nnz)) ) SUPERLU_ABORT("Malloc fails for a[]."); if ( !(asub = intMalloc(nnz)) ) SUPERLU_ABORT("Malloc fails for asub[]."); if ( !(xa = intMalloc(n1+1)) ) SUPERLU_ABORT("Malloc fails for xa[]."); dCompRow_to_CompCol(m, n1, nnz, const_cast<double>(_sk.data()), const_cast<int_t>(_ik.data()), const_cast<int_t*>(_id.data()), &a, &asub, &xa);

// Create matrix A in the format expected by SuperLU. 
dCreate_CompCol_Matrix(&A1, m, n1, nnz, a, asub, xa, SLU_NC, SLU_D, SLU_GE);
A = &A1;
fp = fopen("./SRC/fp.hb","w");
fprintf(fp,"CompCol matrix %s\n", what);
//fprintf(fp,"Stype %d, Dtype %d, Mtype %d\n", A->Stype,A->Dtype,A->Mtype);
n = A->ncol;
Astore = (NCformat *) A->Store;
dp = (double *) Astore->nzval;
fprintf(fp,"RUA");
fprintf(fp," %d %d  %d ", A->nrow,A->ncol,Astore->nnz);//nrow, ncol, a, asub, xa,
fprintf(fp,"0\n");
//fprintf(fp,"\nxa: ");//
for (i = 0; i <= n; ++i){
fprintf(fp,"%d  ", Astore->colptr[i]);
}
fprintf(fp,"\n");
//fprintf(fp,"\nasub: ");//
for (i = 0; i < Astore->colptr[n]; ++i){
fprintf(fp,"%d  ", Astore->rowind[i]);
}
fprintf(fp,"\n");
//fprintf(fp,"\na: ");//
for (i = 0; i < Astore->colptr[n]; ++i){ 
fprintf(fp,"%f  ", dp[i]);
}
fprintf(fp,"\n");
fclose(fp);
return; 

}`

[SalmanMaths]

fp.hb.zip Dear Sherry Li,

I am using the matrix in the attachment but give me the Segmentation fault (core dumped) at dreadhb(). What is wrong in the matrix or it is the code problem?

[SalmanMaths]

fp.zip

Dear Sherry Li,

I also try in this way but still segmentation fault.