Actual source code: baijfact11.c

  1: /*$Id: baijfact11.c,v 1.4 2001/03/23 23:22:07 balay Exp $*/
  2: /*
  3:     Factorization code for BAIJ format. 
  4: */
 5:  #include src/mat/impls/baij/seq/baij.h
 6:  #include src/vec/vecimpl.h
 7:  #include src/inline/ilu.h

  9: /* ------------------------------------------------------------*/
 10: /*
 11:       Version for when blocks are 4 by 4
 12: */
 13: #undef __FUNCT__  
 15: int MatLUFactorNumeric_SeqBAIJ_4(Mat A,Mat *B)
 16: {
 17:   Mat         C = *B;
 18:   Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data;
 19:   IS          isrow = b->row,isicol = b->icol;
 20:   int         *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j;
 21:   int         *ajtmpold,*ajtmp,nz,row;
 22:   int         *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj;
 23:   MatScalar   *pv,*v,*rtmp,*pc,*w,*x;
 24:   MatScalar   p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4;
 25:   MatScalar   p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16;
 26:   MatScalar   p10,p11,p12,p13,p14,p15,p16,m10,m11,m12;
 27:   MatScalar   m13,m14,m15,m16;
 28:   MatScalar   *ba = b->a,*aa = a->a;
 29:   PetscTruth  pivotinblocks = b->pivotinblocks;

 32:   ISGetIndices(isrow,&r);
 33:   ISGetIndices(isicol,&ic);
 34:   PetscMalloc(16*(n+1)*sizeof(MatScalar),&rtmp);

 36:   for (i=0; i<n; i++) {
 37:     nz    = bi[i+1] - bi[i];
 38:     ajtmp = bj + bi[i];
 39:     for  (j=0; j<nz; j++) {
 40:       x = rtmp+16*ajtmp[j];
 41:       x[0]  = x[1]  = x[2]  = x[3]  = x[4]  = x[5]  = x[6] = x[7] = x[8] = x[9] = 0.0;
 42:       x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0;
 43:     }
 44:     /* load in initial (unfactored row) */
 45:     idx      = r[i];
 46:     nz       = ai[idx+1] - ai[idx];
 47:     ajtmpold = aj + ai[idx];
 48:     v        = aa + 16*ai[idx];
 49:     for (j=0; j<nz; j++) {
 50:       x    = rtmp+16*ic[ajtmpold[j]];
 51:       x[0]  = v[0];  x[1]  = v[1];  x[2]  = v[2];  x[3]  = v[3];
 52:       x[4]  = v[4];  x[5]  = v[5];  x[6]  = v[6];  x[7]  = v[7];  x[8]  = v[8];
 53:       x[9]  = v[9];  x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13];
 54:       x[14] = v[14]; x[15] = v[15];
 55:       v    += 16;
 56:     }
 57:     row = *ajtmp++;
 58:     while (row < i) {
 59:       pc  = rtmp + 16*row;
 60:       p1  = pc[0];  p2  = pc[1];  p3  = pc[2];  p4  = pc[3];
 61:       p5  = pc[4];  p6  = pc[5];  p7  = pc[6];  p8  = pc[7];  p9  = pc[8];
 62:       p10 = pc[9];  p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13];
 63:       p15 = pc[14]; p16 = pc[15];
 64:       if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 ||
 65:           p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 ||
 66:           p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0
 67:           || p16 != 0.0) {
 68:         pv = ba + 16*diag_offset[row];
 69:         pj = bj + diag_offset[row] + 1;
 70:         x1  = pv[0];  x2  = pv[1];  x3  = pv[2];  x4  = pv[3];
 71:         x5  = pv[4];  x6  = pv[5];  x7  = pv[6];  x8  = pv[7];  x9  = pv[8];
 72:         x10 = pv[9];  x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13];
 73:         x15 = pv[14]; x16 = pv[15];
 74:         pc[0] = m1 = p1*x1 + p5*x2  + p9*x3  + p13*x4;
 75:         pc[1] = m2 = p2*x1 + p6*x2  + p10*x3 + p14*x4;
 76:         pc[2] = m3 = p3*x1 + p7*x2  + p11*x3 + p15*x4;
 77:         pc[3] = m4 = p4*x1 + p8*x2  + p12*x3 + p16*x4;

 79:         pc[4] = m5 = p1*x5 + p5*x6  + p9*x7  + p13*x8;
 80:         pc[5] = m6 = p2*x5 + p6*x6  + p10*x7 + p14*x8;
 81:         pc[6] = m7 = p3*x5 + p7*x6  + p11*x7 + p15*x8;
 82:         pc[7] = m8 = p4*x5 + p8*x6  + p12*x7 + p16*x8;

 84:         pc[8]  = m9  = p1*x9 + p5*x10  + p9*x11  + p13*x12;
 85:         pc[9]  = m10 = p2*x9 + p6*x10  + p10*x11 + p14*x12;
 86:         pc[10] = m11 = p3*x9 + p7*x10  + p11*x11 + p15*x12;
 87:         pc[11] = m12 = p4*x9 + p8*x10  + p12*x11 + p16*x12;

 89:         pc[12] = m13 = p1*x13 + p5*x14  + p9*x15  + p13*x16;
 90:         pc[13] = m14 = p2*x13 + p6*x14  + p10*x15 + p14*x16;
 91:         pc[14] = m15 = p3*x13 + p7*x14  + p11*x15 + p15*x16;
 92:         pc[15] = m16 = p4*x13 + p8*x14  + p12*x15 + p16*x16;

 94:         nz = bi[row+1] - diag_offset[row] - 1;
 95:         pv += 16;
 96:         for (j=0; j<nz; j++) {
 97:           x1   = pv[0];  x2  = pv[1];   x3 = pv[2];  x4  = pv[3];
 98:           x5   = pv[4];  x6  = pv[5];   x7 = pv[6];  x8  = pv[7]; x9 = pv[8];
 99:           x10  = pv[9];  x11 = pv[10]; x12 = pv[11]; x13 = pv[12];
100:           x14  = pv[13]; x15 = pv[14]; x16 = pv[15];
101:           x    = rtmp + 16*pj[j];
102:           x[0] -= m1*x1 + m5*x2  + m9*x3  + m13*x4;
103:           x[1] -= m2*x1 + m6*x2  + m10*x3 + m14*x4;
104:           x[2] -= m3*x1 + m7*x2  + m11*x3 + m15*x4;
105:           x[3] -= m4*x1 + m8*x2  + m12*x3 + m16*x4;

107:           x[4] -= m1*x5 + m5*x6  + m9*x7  + m13*x8;
108:           x[5] -= m2*x5 + m6*x6  + m10*x7 + m14*x8;
109:           x[6] -= m3*x5 + m7*x6  + m11*x7 + m15*x8;
110:           x[7] -= m4*x5 + m8*x6  + m12*x7 + m16*x8;

112:           x[8]  -= m1*x9 + m5*x10 + m9*x11  + m13*x12;
113:           x[9]  -= m2*x9 + m6*x10 + m10*x11 + m14*x12;
114:           x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12;
115:           x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12;

117:           x[12] -= m1*x13 + m5*x14  + m9*x15  + m13*x16;
118:           x[13] -= m2*x13 + m6*x14  + m10*x15 + m14*x16;
119:           x[14] -= m3*x13 + m7*x14  + m11*x15 + m15*x16;
120:           x[15] -= m4*x13 + m8*x14  + m12*x15 + m16*x16;

122:           pv   += 16;
123:         }
124:         PetscLogFlops(128*nz+112);
125:       }
126:       row = *ajtmp++;
127:     }
128:     /* finished row so stick it into b->a */
129:     pv = ba + 16*bi[i];
130:     pj = bj + bi[i];
131:     nz = bi[i+1] - bi[i];
132:     for (j=0; j<nz; j++) {
133:       x      = rtmp+16*pj[j];
134:       pv[0]  = x[0];  pv[1]  = x[1];  pv[2]  = x[2];  pv[3]  = x[3];
135:       pv[4]  = x[4];  pv[5]  = x[5];  pv[6]  = x[6];  pv[7]  = x[7]; pv[8] = x[8];
136:       pv[9]  = x[9];  pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12];
137:       pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15];
138:       pv   += 16;
139:     }
140:     /* invert diagonal block */
141:     w    = ba + 16*diag_offset[i];
142:     if (pivotinblocks) {
143:       Kernel_A_gets_inverse_A_4(w);
144:     } else {
145:       Kernel_A_gets_inverse_A_4_nopivot(w);
146:     }
147:   }

149:   PetscFree(rtmp);
150:   ISRestoreIndices(isicol,&ic);
151:   ISRestoreIndices(isrow,&r);
152:   C->factor = FACTOR_LU;
153:   C->assembled = PETSC_TRUE;
154:   PetscLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */
155:   return(0);
156: }