1: /*
  2:     Factorization code for BAIJ format.
  3: */
  4: #include <../src/mat/impls/baij/seq/baij.h>
  5: #include <petsc/private/kernels/blockinvert.h>

  7: PetscErrorCode MatILUFactorNumeric_SeqBAIJ_N_inplace(Mat C, Mat A, const MatFactorInfo *info)
  8: {
  9:   Mat_SeqBAIJ    *a = (Mat_SeqBAIJ *)A->data, *b = (Mat_SeqBAIJ *)C->data;
 10:   IS              isrow = b->row, isicol = b->icol;
 11:   const PetscInt *r, *ic;
 12:   PetscInt        i, j, n = a->mbs, *bi = b->i, *bj = b->j;
 13:   PetscInt       *ajtmpold, *ajtmp, nz, row, *ai = a->i, *aj = a->j, k, flg;
 14:   const PetscInt *diag_offset;
 15:   PetscInt        diag, bs = A->rmap->bs, bs2 = a->bs2, *pj, *v_pivots;
 16:   MatScalar      *ba = b->a, *aa = a->a, *pv, *v, *rtmp, *multiplier, *v_work, *pc, *w;
 17:   PetscBool       allowzeropivot, zeropivotdetected;

 19:   PetscFunctionBegin;
 20:   /* Since A is C and C is labeled as a factored matrix we need to lie to MatGetDiagonalMarkers_SeqBAIJ() to get it to compute the diagonals */
 21:   A->factortype = MAT_FACTOR_NONE;
 22:   PetscCall(MatGetDiagonalMarkers_SeqBAIJ(A, &diag_offset, NULL));
 23:   A->factortype = MAT_FACTOR_ILU;
 24:   PetscCall(ISGetIndices(isrow, &r));
 25:   PetscCall(ISGetIndices(isicol, &ic));
 26:   allowzeropivot = PetscNot(A->erroriffailure);

 28:   PetscCall(PetscCalloc1(bs2 * (n + 1), &rtmp));
 29:   /* generate work space needed by dense LU factorization */
 30:   PetscCall(PetscMalloc3(bs, &v_work, bs2, &multiplier, bs, &v_pivots));

 32:   for (i = 0; i < n; i++) {
 33:     nz    = bi[i + 1] - bi[i];
 34:     ajtmp = bj + bi[i];
 35:     for (j = 0; j < nz; j++) PetscCall(PetscArrayzero(rtmp + bs2 * ajtmp[j], bs2));
 36:     /* load in initial (unfactored row) */
 37:     nz       = ai[r[i] + 1] - ai[r[i]];
 38:     ajtmpold = aj + ai[r[i]];
 39:     v        = aa + bs2 * ai[r[i]];
 40:     for (j = 0; j < nz; j++) PetscCall(PetscArraycpy(rtmp + bs2 * ic[ajtmpold[j]], v + bs2 * j, bs2));
 41:     row = *ajtmp++;
 42:     while (row < i) {
 43:       pc = rtmp + bs2 * row;
 44:       /*      if (*pc) { */
 45:       for (flg = 0, k = 0; k < bs2; k++) {
 46:         if (pc[k] != 0.0) {
 47:           flg = 1;
 48:           break;
 49:         }
 50:       }
 51:       if (flg) {
 52:         pv = ba + bs2 * diag_offset[row];
 53:         pj = bj + diag_offset[row] + 1;
 54:         PetscKernel_A_gets_A_times_B(bs, pc, pv, multiplier);
 55:         nz = bi[row + 1] - diag_offset[row] - 1;
 56:         pv += bs2;
 57:         for (j = 0; j < nz; j++) PetscKernel_A_gets_A_minus_B_times_C(bs, rtmp + bs2 * pj[j], pc, pv + bs2 * j);
 58:         PetscCall(PetscLogFlops(2.0 * bs * bs2 * (nz + 1.0) - bs));
 59:       }
 60:       row = *ajtmp++;
 61:     }
 62:     /* finished row so stick it into b->a */
 63:     pv = ba + bs2 * bi[i];
 64:     pj = bj + bi[i];
 65:     nz = bi[i + 1] - bi[i];
 66:     for (j = 0; j < nz; j++) PetscCall(PetscArraycpy(pv + bs2 * j, rtmp + bs2 * pj[j], bs2));
 67:     diag = diag_offset[i] - bi[i];
 68:     /* invert diagonal block */
 69:     w = pv + bs2 * diag;

 71:     PetscCall(PetscKernel_A_gets_inverse_A(bs, w, v_pivots, v_work, allowzeropivot, &zeropivotdetected));
 72:     if (zeropivotdetected) C->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
 73:   }

 75:   PetscCall(PetscFree(rtmp));
 76:   PetscCall(PetscFree3(v_work, multiplier, v_pivots));
 77:   PetscCall(ISRestoreIndices(isicol, &ic));
 78:   PetscCall(ISRestoreIndices(isrow, &r));

 80:   C->ops->solve          = MatSolve_SeqBAIJ_N_inplace;
 81:   C->ops->solvetranspose = MatSolveTranspose_SeqBAIJ_N_inplace;
 82:   C->assembled           = PETSC_TRUE;

 84:   PetscCall(PetscLogFlops(1.333333333333 * bs * bs2 * b->mbs)); /* from inverting diagonal blocks */
 85:   PetscFunctionReturn(PETSC_SUCCESS);
 86: }