1 /*$Id: baijfact.c,v 1.75 1999/11/05 14:45:32 bsmith Exp bsmith $*/ 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" 8 9 /* 10 The symbolic factorization code is identical to that for AIJ format, 11 except for very small changes since this is now a SeqBAIJ datastructure. 12 NOT good code reuse. 13 */ 14 #undef __FUNC__ 15 #define __FUNC__ "MatLUFactorSymbolic_SeqBAIJ" 16 int MatLUFactorSymbolic_SeqBAIJ(Mat A,IS isrow,IS iscol,double f,Mat *B) 17 { 18 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data, *b; 19 IS isicol; 20 int *r,*ic, ierr, i, n = a->mbs, *ai = a->i, *aj = a->j; 21 int *ainew,*ajnew, jmax,*fill, *ajtmp, nz, bs = a->bs, bs2=a->bs2; 22 int *idnew, idx, row,m,fm, nnz, nzi,realloc = 0,nzbd,*im; 23 24 PetscFunctionBegin; 25 PetscValidHeaderSpecific(isrow,IS_COOKIE); 26 PetscValidHeaderSpecific(iscol,IS_COOKIE); 27 if (A->M != A->N) SETERRQ(PETSC_ERR_ARG_WRONG,0,"matrix must be square"); 28 ierr = ISInvertPermutation(iscol,&isicol);CHKERRQ(ierr); 29 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 30 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 31 32 /* get new row pointers */ 33 ainew = (int *) PetscMalloc( (n+1)*sizeof(int) );CHKPTRQ(ainew); 34 ainew[0] = 0; 35 /* don't know how many column pointers are needed so estimate */ 36 jmax = (int) (f*ai[n] + 1); 37 ajnew = (int *) PetscMalloc( (jmax)*sizeof(int) );CHKPTRQ(ajnew); 38 /* fill is a linked list of nonzeros in active row */ 39 fill = (int *) PetscMalloc( (2*n+1)*sizeof(int));CHKPTRQ(fill); 40 im = fill + n + 1; 41 /* idnew is location of diagonal in factor */ 42 idnew = (int *) PetscMalloc( (n+1)*sizeof(int));CHKPTRQ(idnew); 43 idnew[0] = 0; 44 45 for ( i=0; i<n; i++ ) { 46 /* first copy previous fill into linked list */ 47 nnz = nz = ai[r[i]+1] - ai[r[i]]; 48 if (!nz) SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,1,"Empty row in matrix"); 49 ajtmp = aj + ai[r[i]]; 50 fill[n] = n; 51 while (nz--) { 52 fm = n; 53 idx = ic[*ajtmp++]; 54 do { 55 m = fm; 56 fm = fill[m]; 57 } while (fm < idx); 58 fill[m] = idx; 59 fill[idx] = fm; 60 } 61 row = fill[n]; 62 while ( row < i ) { 63 ajtmp = ajnew + idnew[row] + 1; 64 nzbd = 1 + idnew[row] - ainew[row]; 65 nz = im[row] - nzbd; 66 fm = row; 67 while (nz-- > 0) { 68 idx = *ajtmp++; 69 nzbd++; 70 if (idx == i) im[row] = nzbd; 71 do { 72 m = fm; 73 fm = fill[m]; 74 } while (fm < idx); 75 if (fm != idx) { 76 fill[m] = idx; 77 fill[idx] = fm; 78 fm = idx; 79 nnz++; 80 } 81 } 82 row = fill[row]; 83 } 84 /* copy new filled row into permanent storage */ 85 ainew[i+1] = ainew[i] + nnz; 86 if (ainew[i+1] > jmax) { 87 88 /* estimate how much additional space we will need */ 89 /* use the strategy suggested by David Hysom <hysom@perch-t.icase.edu> */ 90 /* just double the memory each time */ 91 int maxadd = jmax; 92 /* maxadd = (int) ((f*(ai[n]+1)*(n-i+5))/n); */ 93 if (maxadd < nnz) maxadd = (n-i)*(nnz+1); 94 jmax += maxadd; 95 96 /* allocate a longer ajnew */ 97 ajtmp = (int *) PetscMalloc( jmax*sizeof(int) );CHKPTRQ(ajtmp); 98 ierr = PetscMemcpy(ajtmp,ajnew,ainew[i]*sizeof(int));CHKERRQ(ierr); 99 ierr = PetscFree(ajnew);CHKERRQ(ierr); 100 ajnew = ajtmp; 101 realloc++; /* count how many times we realloc */ 102 } 103 ajtmp = ajnew + ainew[i]; 104 fm = fill[n]; 105 nzi = 0; 106 im[i] = nnz; 107 while (nnz--) { 108 if (fm < i) nzi++; 109 *ajtmp++ = fm; 110 fm = fill[fm]; 111 } 112 idnew[i] = ainew[i] + nzi; 113 } 114 115 if (ai[n] != 0) { 116 double af = ((double)ainew[n])/((double)ai[n]); 117 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:Reallocs %d Fill ratio:given %g needed %g\n",realloc,f,af); 118 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:Run with -pc_lu_fill %g or use \n",af); 119 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:PCLUSetFill(pc,%g);\n",af); 120 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:for best performance.\n"); 121 } else { 122 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:Empty matrix.\n"); 123 } 124 125 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 126 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 127 128 ierr = PetscFree(fill);CHKERRQ(ierr); 129 130 /* put together the new matrix */ 131 ierr = MatCreateSeqBAIJ(A->comm,bs,bs*n,bs*n,0,PETSC_NULL,B);CHKERRQ(ierr); 132 PLogObjectParent(*B,isicol); 133 b = (Mat_SeqBAIJ *) (*B)->data; 134 ierr = PetscFree(b->imax);CHKERRQ(ierr); 135 b->singlemalloc = 0; 136 /* the next line frees the default space generated by the Create() */ 137 ierr = PetscFree(b->a);CHKERRQ(ierr); 138 ierr = PetscFree(b->ilen);CHKERRQ(ierr); 139 b->a = (MatScalar *) PetscMalloc((ainew[n]+1)*sizeof(MatScalar)*bs2);CHKPTRQ(b->a); 140 b->j = ajnew; 141 b->i = ainew; 142 b->diag = idnew; 143 b->ilen = 0; 144 b->imax = 0; 145 b->row = isrow; 146 b->col = iscol; 147 ierr = PetscObjectReference((PetscObject)isrow);CHKERRQ(ierr); 148 ierr = PetscObjectReference((PetscObject)iscol);CHKERRQ(ierr); 149 b->icol = isicol; 150 b->solve_work = (Scalar *) PetscMalloc( (bs*n+bs)*sizeof(Scalar));CHKPTRQ(b->solve_work); 151 /* In b structure: Free imax, ilen, old a, old j. 152 Allocate idnew, solve_work, new a, new j */ 153 PLogObjectMemory(*B,(ainew[n]-n)*(sizeof(int)+sizeof(MatScalar))); 154 b->maxnz = b->nz = ainew[n]; 155 156 (*B)->factor = FACTOR_LU; 157 (*B)->info.factor_mallocs = realloc; 158 (*B)->info.fill_ratio_given = f; 159 if (ai[n] != 0) { 160 (*B)->info.fill_ratio_needed = ((double)ainew[n])/((double)ai[n]); 161 } else { 162 (*B)->info.fill_ratio_needed = 0.0; 163 } 164 165 166 PetscFunctionReturn(0); 167 } 168 169 /* ----------------------------------------------------------- */ 170 #undef __FUNC__ 171 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_N" 172 int MatLUFactorNumeric_SeqBAIJ_N(Mat A,Mat *B) 173 { 174 Mat C = *B; 175 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 176 IS isrow = b->row, isicol = b->icol; 177 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 178 int *ajtmpold, *ajtmp, nz, row, bslog,*ai=a->i,*aj=a->j,k,flg; 179 int *diag_offset=b->diag,diag,bs=a->bs,bs2 = a->bs2,*v_pivots; 180 register int *pj; 181 register MatScalar *pv,*v,*rtmp,*multiplier,*v_work,*pc,*w; 182 MatScalar *ba = b->a,*aa = a->a; 183 184 PetscFunctionBegin; 185 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 186 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 187 rtmp = (MatScalar *) PetscMalloc(bs2*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 188 ierr = PetscMemzero(rtmp,bs2*(n+1)*sizeof(MatScalar));CHKERRQ(ierr); 189 /* generate work space needed by dense LU factorization */ 190 v_work = (MatScalar *) PetscMalloc(bs*sizeof(int) + (bs+bs2)*sizeof(MatScalar));CHKPTRQ(v_work); 191 multiplier = v_work + bs; 192 v_pivots = (int *) (multiplier + bs2); 193 194 /* flops in while loop */ 195 bslog = 2*bs*bs2; 196 197 for ( i=0; i<n; i++ ) { 198 nz = bi[i+1] - bi[i]; 199 ajtmp = bj + bi[i]; 200 for ( j=0; j<nz; j++ ) { 201 ierr = PetscMemzero(rtmp+bs2*ajtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 202 } 203 /* load in initial (unfactored row) */ 204 nz = ai[r[i]+1] - ai[r[i]]; 205 ajtmpold = aj + ai[r[i]]; 206 v = aa + bs2*ai[r[i]]; 207 for ( j=0; j<nz; j++ ) { 208 ierr = PetscMemcpy(rtmp+bs2*ic[ajtmpold[j]],v+bs2*j,bs2*sizeof(MatScalar));CHKERRQ(ierr); 209 } 210 row = *ajtmp++; 211 while (row < i) { 212 pc = rtmp + bs2*row; 213 /* if (*pc) { */ 214 for ( flg=0,k=0; k<bs2; k++ ) { if (pc[k]!=0.0) { flg =1; break; }} 215 if (flg) { 216 pv = ba + bs2*diag_offset[row]; 217 pj = bj + diag_offset[row] + 1; 218 Kernel_A_gets_A_times_B(bs,pc,pv,multiplier); 219 nz = bi[row+1] - diag_offset[row] - 1; 220 pv += bs2; 221 for (j=0; j<nz; j++) { 222 Kernel_A_gets_A_minus_B_times_C(bs,rtmp+bs2*pj[j],pc,pv+bs2*j); 223 } 224 PLogFlops(bslog*(nz+1)-bs); 225 } 226 row = *ajtmp++; 227 } 228 /* finished row so stick it into b->a */ 229 pv = ba + bs2*bi[i]; 230 pj = bj + bi[i]; 231 nz = bi[i+1] - bi[i]; 232 for ( j=0; j<nz; j++ ) { 233 ierr = PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 234 } 235 diag = diag_offset[i] - bi[i]; 236 /* invert diagonal block */ 237 w = pv + bs2*diag; 238 Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work); 239 } 240 241 ierr = PetscFree(rtmp);CHKERRQ(ierr); 242 ierr = PetscFree(v_work);CHKERRQ(ierr); 243 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 244 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 245 C->factor = FACTOR_LU; 246 C->assembled = PETSC_TRUE; 247 PLogFlops(1.3333*bs*bs2*b->mbs); /* from inverting diagonal blocks */ 248 PetscFunctionReturn(0); 249 } 250 /* ------------------------------------------------------------*/ 251 /* 252 Version for when blocks are 7 by 7 253 */ 254 #undef __FUNC__ 255 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_7" 256 int MatLUFactorNumeric_SeqBAIJ_7(Mat A,Mat *B) 257 { 258 Mat C = *B; 259 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 260 IS isrow = b->row, isicol = b->icol; 261 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 262 int *ajtmpold, *ajtmp, nz, row; 263 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 264 register int *pj; 265 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 266 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 267 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 268 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 269 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 270 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 271 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 272 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 273 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 274 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 275 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 276 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 277 MatScalar *ba = b->a,*aa = a->a; 278 279 PetscFunctionBegin; 280 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 281 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 282 rtmp = (MatScalar *) PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 283 284 for ( i=0; i<n; i++ ) { 285 nz = bi[i+1] - bi[i]; 286 ajtmp = bj + bi[i]; 287 for ( j=0; j<nz; j++ ) { 288 x = rtmp+49*ajtmp[j]; 289 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 290 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 291 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 292 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 293 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 294 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 295 } 296 /* load in initial (unfactored row) */ 297 idx = r[i]; 298 nz = ai[idx+1] - ai[idx]; 299 ajtmpold = aj + ai[idx]; 300 v = aa + 49*ai[idx]; 301 for ( j=0; j<nz; j++ ) { 302 x = rtmp+49*ic[ajtmpold[j]]; 303 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 304 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 305 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 306 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 307 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 308 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 309 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 310 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 311 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 312 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 313 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 314 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 315 x[48] = v[48]; 316 v += 49; 317 } 318 row = *ajtmp++; 319 while (row < i) { 320 pc = rtmp + 49*row; 321 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 322 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 323 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 324 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 325 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 326 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 327 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 328 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 329 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 330 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 331 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 332 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 333 p49 = pc[48]; 334 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 335 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 336 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 337 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 338 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 339 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 340 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 341 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 342 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 343 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 344 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 345 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 346 p49 != 0.0) { 347 pv = ba + 49*diag_offset[row]; 348 pj = bj + diag_offset[row] + 1; 349 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 350 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 351 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 352 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 353 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 354 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 355 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 356 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 357 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 358 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 359 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 360 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 361 x49 = pv[48]; 362 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 363 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 364 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 365 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 366 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 367 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 368 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 369 370 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 371 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 372 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 373 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 374 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 375 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 376 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 377 378 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 379 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 380 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 381 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 382 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 383 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 384 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 385 386 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 387 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 388 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 389 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 390 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 391 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 392 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 393 394 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 395 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 396 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 397 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 398 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 399 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 400 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 401 402 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 403 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 404 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 405 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 406 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 407 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 408 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 409 410 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 411 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 412 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 413 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 414 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 415 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 416 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 417 418 nz = bi[row+1] - diag_offset[row] - 1; 419 pv += 49; 420 for (j=0; j<nz; j++) { 421 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 422 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 423 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 424 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 425 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 426 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 427 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 428 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 429 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 430 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 431 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 432 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 433 x49 = pv[48]; 434 x = rtmp + 49*pj[j]; 435 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 436 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 437 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 438 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 439 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 440 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 441 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 442 443 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 444 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 445 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 446 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 447 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 448 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 449 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 450 451 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 452 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 453 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 454 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 455 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 456 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 457 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 458 459 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 460 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 461 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 462 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 463 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 464 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 465 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 466 467 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 468 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 469 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 470 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 471 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 472 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 473 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 474 475 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 476 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 477 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 478 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 479 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 480 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 481 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 482 483 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 484 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 485 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 486 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 487 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 488 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 489 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 490 pv += 49; 491 } 492 PLogFlops(686*nz+637); 493 } 494 row = *ajtmp++; 495 } 496 /* finished row so stick it into b->a */ 497 pv = ba + 49*bi[i]; 498 pj = bj + bi[i]; 499 nz = bi[i+1] - bi[i]; 500 for ( j=0; j<nz; j++ ) { 501 x = rtmp+49*pj[j]; 502 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 503 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 504 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 505 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 506 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 507 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 508 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 509 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 510 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 511 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 512 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 513 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 514 pv[48] = x[48]; 515 pv += 49; 516 } 517 /* invert diagonal block */ 518 w = ba + 49*diag_offset[i]; 519 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 520 } 521 522 ierr = PetscFree(rtmp);CHKERRQ(ierr); 523 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 524 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 525 C->factor = FACTOR_LU; 526 C->assembled = PETSC_TRUE; 527 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 528 PetscFunctionReturn(0); 529 } 530 531 /* 532 Version for when blocks are 7 by 7 Using natural ordering 533 */ 534 #undef __FUNC__ 535 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_7_NaturalOrdering" 536 int MatLUFactorNumeric_SeqBAIJ_7_NaturalOrdering(Mat A,Mat *B) 537 { 538 Mat C = *B; 539 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 540 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 541 int *ajtmpold, *ajtmp, nz, row; 542 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 543 register int *pj; 544 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 545 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 546 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 547 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 548 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 549 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 550 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 551 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 552 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 553 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 554 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 555 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 556 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 557 MatScalar *ba = b->a,*aa = a->a; 558 559 PetscFunctionBegin; 560 rtmp = (MatScalar *) PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 561 for ( i=0; i<n; i++ ) { 562 nz = bi[i+1] - bi[i]; 563 ajtmp = bj + bi[i]; 564 for ( j=0; j<nz; j++ ) { 565 x = rtmp+49*ajtmp[j]; 566 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 567 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 568 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 569 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 570 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 571 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 572 } 573 /* load in initial (unfactored row) */ 574 nz = ai[i+1] - ai[i]; 575 ajtmpold = aj + ai[i]; 576 v = aa + 49*ai[i]; 577 for ( j=0; j<nz; j++ ) { 578 x = rtmp+49*ajtmpold[j]; 579 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 580 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 581 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 582 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 583 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 584 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 585 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 586 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 587 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 588 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 589 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 590 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 591 x[48] = v[48]; 592 v += 49; 593 } 594 row = *ajtmp++; 595 while (row < i) { 596 pc = rtmp + 49*row; 597 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 598 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 599 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 600 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 601 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 602 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 603 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 604 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 605 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 606 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 607 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 608 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 609 p49 = pc[48]; 610 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 611 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 612 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 613 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 614 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 615 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 616 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 617 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 618 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 619 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 620 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 621 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 622 p49 != 0.0) { 623 pv = ba + 49*diag_offset[row]; 624 pj = bj + diag_offset[row] + 1; 625 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 626 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 627 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 628 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 629 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 630 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 631 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 632 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 633 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 634 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 635 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 636 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 637 x49 = pv[48]; 638 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 639 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 640 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 641 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 642 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 643 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 644 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 645 646 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 647 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 648 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 649 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 650 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 651 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 652 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 653 654 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 655 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 656 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 657 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 658 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 659 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 660 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 661 662 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 663 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 664 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 665 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 666 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 667 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 668 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 669 670 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 671 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 672 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 673 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 674 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 675 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 676 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 677 678 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 679 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 680 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 681 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 682 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 683 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 684 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 685 686 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 687 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 688 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 689 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 690 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 691 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 692 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 693 694 nz = bi[row+1] - diag_offset[row] - 1; 695 pv += 49; 696 for (j=0; j<nz; j++) { 697 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 698 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 699 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 700 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 701 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 702 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 703 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 704 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 705 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 706 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 707 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 708 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 709 x49 = pv[48]; 710 x = rtmp + 49*pj[j]; 711 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 712 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 713 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 714 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 715 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 716 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 717 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 718 719 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 720 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 721 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 722 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 723 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 724 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 725 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 726 727 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 728 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 729 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 730 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 731 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 732 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 733 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 734 735 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 736 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 737 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 738 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 739 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 740 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 741 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 742 743 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 744 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 745 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 746 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 747 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 748 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 749 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 750 751 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 752 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 753 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 754 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 755 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 756 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 757 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 758 759 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 760 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 761 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 762 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 763 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 764 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 765 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 766 pv += 49; 767 } 768 PLogFlops(686*nz+637); 769 } 770 row = *ajtmp++; 771 } 772 /* finished row so stick it into b->a */ 773 pv = ba + 49*bi[i]; 774 pj = bj + bi[i]; 775 nz = bi[i+1] - bi[i]; 776 for ( j=0; j<nz; j++ ) { 777 x = rtmp+49*pj[j]; 778 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 779 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 780 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 781 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 782 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 783 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 784 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 785 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 786 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 787 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 788 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 789 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 790 pv[48] = x[48]; 791 pv += 49; 792 } 793 /* invert diagonal block */ 794 w = ba + 49*diag_offset[i]; 795 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 796 } 797 798 ierr = PetscFree(rtmp);CHKERRQ(ierr); 799 C->factor = FACTOR_LU; 800 C->assembled = PETSC_TRUE; 801 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 802 PetscFunctionReturn(0); 803 } 804 805 /* ------------------------------------------------------------*/ 806 /* 807 Version for when blocks are 6 by 6 808 */ 809 #undef __FUNC__ 810 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_6" 811 int MatLUFactorNumeric_SeqBAIJ_6(Mat A,Mat *B) 812 { 813 Mat C = *B; 814 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 815 IS isrow = b->row, isicol = b->icol; 816 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 817 int *ajtmpold, *ajtmp, nz, row; 818 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 819 register int *pj; 820 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 821 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 822 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 823 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 824 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 825 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 826 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 827 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 828 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 829 MatScalar *ba = b->a,*aa = a->a; 830 831 PetscFunctionBegin; 832 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 833 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 834 rtmp = (MatScalar *) PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 835 836 for ( i=0; i<n; i++ ) { 837 nz = bi[i+1] - bi[i]; 838 ajtmp = bj + bi[i]; 839 for ( j=0; j<nz; j++ ) { 840 x = rtmp+36*ajtmp[j]; 841 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 842 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 843 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 844 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 845 x[34] = x[35] = 0.0 ; 846 } 847 /* load in initial (unfactored row) */ 848 idx = r[i]; 849 nz = ai[idx+1] - ai[idx]; 850 ajtmpold = aj + ai[idx]; 851 v = aa + 36*ai[idx]; 852 for ( j=0; j<nz; j++ ) { 853 x = rtmp+36*ic[ajtmpold[j]]; 854 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 855 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 856 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 857 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 858 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 859 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 860 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 861 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 862 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 863 v += 36; 864 } 865 row = *ajtmp++; 866 while (row < i) { 867 pc = rtmp + 36*row; 868 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 869 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 870 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 871 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 872 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 873 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 874 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 875 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 876 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 877 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 878 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 879 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 880 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 881 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 882 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 883 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 884 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 885 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 886 pv = ba + 36*diag_offset[row]; 887 pj = bj + diag_offset[row] + 1; 888 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 889 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 890 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 891 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 892 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 893 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 894 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 895 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 896 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 897 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 898 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 899 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 900 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 901 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 902 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 903 904 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 905 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 906 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 907 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 908 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 909 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 910 911 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 912 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 913 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 914 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 915 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 916 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 917 918 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 919 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 920 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 921 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 922 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 923 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 924 925 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 926 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 927 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 928 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 929 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 930 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 931 932 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 933 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 934 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 935 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 936 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 937 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 938 939 nz = bi[row+1] - diag_offset[row] - 1; 940 pv += 36; 941 for (j=0; j<nz; j++) { 942 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 943 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 944 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 945 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 946 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 947 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 948 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 949 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 950 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 951 x = rtmp + 36*pj[j]; 952 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 953 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 954 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 955 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 956 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 957 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 958 959 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 960 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 961 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 962 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 963 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 964 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 965 966 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 967 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 968 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 969 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 970 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 971 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 972 973 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 974 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 975 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 976 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 977 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 978 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 979 980 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 981 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 982 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 983 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 984 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 985 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 986 987 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 988 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 989 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 990 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 991 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 992 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 993 994 pv += 36; 995 } 996 PLogFlops(432*nz+396); 997 } 998 row = *ajtmp++; 999 } 1000 /* finished row so stick it into b->a */ 1001 pv = ba + 36*bi[i]; 1002 pj = bj + bi[i]; 1003 nz = bi[i+1] - bi[i]; 1004 for ( j=0; j<nz; j++ ) { 1005 x = rtmp+36*pj[j]; 1006 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1007 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1008 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1009 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1010 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1011 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1012 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1013 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1014 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1015 pv += 36; 1016 } 1017 /* invert diagonal block */ 1018 w = ba + 36*diag_offset[i]; 1019 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1020 } 1021 1022 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1023 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1024 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1025 C->factor = FACTOR_LU; 1026 C->assembled = PETSC_TRUE; 1027 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1028 PetscFunctionReturn(0); 1029 } 1030 /* 1031 Version for when blocks are 6 by 6 Using natural ordering 1032 */ 1033 #undef __FUNC__ 1034 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_6_NaturalOrdering" 1035 int MatLUFactorNumeric_SeqBAIJ_6_NaturalOrdering(Mat A,Mat *B) 1036 { 1037 Mat C = *B; 1038 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1039 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1040 int *ajtmpold, *ajtmp, nz, row; 1041 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 1042 register int *pj; 1043 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1044 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1045 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1046 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1047 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1048 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1049 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1050 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 1051 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 1052 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 1053 MatScalar *ba = b->a,*aa = a->a; 1054 1055 PetscFunctionBegin; 1056 rtmp = (MatScalar *) PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1057 for ( i=0; i<n; i++ ) { 1058 nz = bi[i+1] - bi[i]; 1059 ajtmp = bj + bi[i]; 1060 for ( j=0; j<nz; j++ ) { 1061 x = rtmp+36*ajtmp[j]; 1062 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1063 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1064 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 1065 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 1066 x[34] = x[35] = 0.0 ; 1067 } 1068 /* load in initial (unfactored row) */ 1069 nz = ai[i+1] - ai[i]; 1070 ajtmpold = aj + ai[i]; 1071 v = aa + 36*ai[i]; 1072 for ( j=0; j<nz; j++ ) { 1073 x = rtmp+36*ajtmpold[j]; 1074 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1075 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 1076 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 1077 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 1078 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 1079 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1080 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 1081 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 1082 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 1083 v += 36; 1084 } 1085 row = *ajtmp++; 1086 while (row < i) { 1087 pc = rtmp + 36*row; 1088 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1089 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 1090 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 1091 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 1092 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 1093 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1094 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 1095 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 1096 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 1097 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 1098 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 1099 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 1100 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 1101 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 1102 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 1103 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 1104 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 1105 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 1106 pv = ba + 36*diag_offset[row]; 1107 pj = bj + diag_offset[row] + 1; 1108 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1109 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1110 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1111 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1112 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1113 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1114 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1115 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1116 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1117 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 1118 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 1119 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 1120 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 1121 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 1122 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 1123 1124 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 1125 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 1126 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 1127 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 1128 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 1129 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 1130 1131 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 1132 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 1133 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 1134 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 1135 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 1136 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 1137 1138 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 1139 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 1140 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 1141 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 1142 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 1143 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 1144 1145 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 1146 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 1147 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 1148 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 1149 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 1150 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 1151 1152 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 1153 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 1154 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 1155 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 1156 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 1157 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 1158 1159 nz = bi[row+1] - diag_offset[row] - 1; 1160 pv += 36; 1161 for (j=0; j<nz; j++) { 1162 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1163 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1164 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1165 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1166 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1167 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1168 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1169 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1170 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1171 x = rtmp + 36*pj[j]; 1172 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 1173 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 1174 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 1175 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 1176 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 1177 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 1178 1179 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 1180 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 1181 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 1182 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 1183 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 1184 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 1185 1186 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 1187 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 1188 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 1189 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 1190 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 1191 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 1192 1193 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 1194 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 1195 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 1196 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 1197 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 1198 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 1199 1200 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 1201 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 1202 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 1203 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 1204 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 1205 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 1206 1207 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 1208 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 1209 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 1210 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 1211 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 1212 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 1213 1214 pv += 36; 1215 } 1216 PLogFlops(432*nz+396); 1217 } 1218 row = *ajtmp++; 1219 } 1220 /* finished row so stick it into b->a */ 1221 pv = ba + 36*bi[i]; 1222 pj = bj + bi[i]; 1223 nz = bi[i+1] - bi[i]; 1224 for ( j=0; j<nz; j++ ) { 1225 x = rtmp+36*pj[j]; 1226 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1227 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1228 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1229 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1230 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1231 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1232 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1233 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1234 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1235 pv += 36; 1236 } 1237 /* invert diagonal block */ 1238 w = ba + 36*diag_offset[i]; 1239 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1240 } 1241 1242 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1243 C->factor = FACTOR_LU; 1244 C->assembled = PETSC_TRUE; 1245 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1246 PetscFunctionReturn(0); 1247 } 1248 1249 /* ------------------------------------------------------------*/ 1250 /* 1251 Version for when blocks are 5 by 5 1252 */ 1253 #undef __FUNC__ 1254 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_5" 1255 int MatLUFactorNumeric_SeqBAIJ_5(Mat A,Mat *B) 1256 { 1257 Mat C = *B; 1258 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1259 IS isrow = b->row, isicol = b->icol; 1260 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1261 int *ajtmpold, *ajtmp, nz, row; 1262 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 1263 register int *pj; 1264 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1265 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1266 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1267 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 1268 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 1269 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1270 MatScalar *ba = b->a,*aa = a->a; 1271 1272 PetscFunctionBegin; 1273 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1274 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1275 rtmp = (MatScalar *) PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1276 1277 for ( i=0; i<n; i++ ) { 1278 nz = bi[i+1] - bi[i]; 1279 ajtmp = bj + bi[i]; 1280 for ( j=0; j<nz; j++ ) { 1281 x = rtmp+25*ajtmp[j]; 1282 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1283 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1284 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1285 } 1286 /* load in initial (unfactored row) */ 1287 idx = r[i]; 1288 nz = ai[idx+1] - ai[idx]; 1289 ajtmpold = aj + ai[idx]; 1290 v = aa + 25*ai[idx]; 1291 for ( j=0; j<nz; j++ ) { 1292 x = rtmp+25*ic[ajtmpold[j]]; 1293 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1294 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1295 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1296 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; 1297 x[18] = v[18]; x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; 1298 x[22] = v[22]; x[23] = v[23]; x[24] = v[24]; 1299 v += 25; 1300 } 1301 row = *ajtmp++; 1302 while (row < i) { 1303 pc = rtmp + 25*row; 1304 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1305 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1306 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1307 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; 1308 p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1309 p25 = pc[24]; 1310 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1311 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1312 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1313 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || 1314 p20 != 0.0 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || 1315 p24 != 0.0 || p25 != 0.0) { 1316 pv = ba + 25*diag_offset[row]; 1317 pj = bj + diag_offset[row] + 1; 1318 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1319 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1320 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1321 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1322 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; 1323 x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1324 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1325 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1326 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1327 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1328 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1329 1330 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1331 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1332 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1333 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1334 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1335 1336 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1337 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1338 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1339 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1340 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1341 1342 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1343 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1344 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1345 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1346 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1347 1348 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1349 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1350 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1351 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1352 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1353 1354 nz = bi[row+1] - diag_offset[row] - 1; 1355 pv += 25; 1356 for (j=0; j<nz; j++) { 1357 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1358 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1359 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1360 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; 1361 x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; 1362 x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1363 x = rtmp + 25*pj[j]; 1364 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1365 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1366 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1367 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1368 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1369 1370 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1371 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1372 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1373 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1374 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1375 1376 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1377 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1378 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1379 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1380 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1381 1382 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1383 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1384 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1385 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1386 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1387 1388 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1389 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1390 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1391 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1392 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1393 1394 pv += 25; 1395 } 1396 PLogFlops(250*nz+225); 1397 } 1398 row = *ajtmp++; 1399 } 1400 /* finished row so stick it into b->a */ 1401 pv = ba + 25*bi[i]; 1402 pj = bj + bi[i]; 1403 nz = bi[i+1] - bi[i]; 1404 for ( j=0; j<nz; j++ ) { 1405 x = rtmp+25*pj[j]; 1406 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1407 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1408 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1409 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; 1410 pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; 1411 pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; pv[24] = x[24]; 1412 pv += 25; 1413 } 1414 /* invert diagonal block */ 1415 w = ba + 25*diag_offset[i]; 1416 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1417 } 1418 1419 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1420 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1421 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1422 C->factor = FACTOR_LU; 1423 C->assembled = PETSC_TRUE; 1424 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1425 PetscFunctionReturn(0); 1426 } 1427 /* 1428 Version for when blocks are 5 by 5 Using natural ordering 1429 */ 1430 #undef __FUNC__ 1431 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_5_NaturalOrdering" 1432 int MatLUFactorNumeric_SeqBAIJ_5_NaturalOrdering(Mat A,Mat *B) 1433 { 1434 Mat C = *B; 1435 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1436 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1437 int *ajtmpold, *ajtmp, nz, row; 1438 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 1439 register int *pj; 1440 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1441 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1442 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1443 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1444 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1445 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1446 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1447 MatScalar *ba = b->a,*aa = a->a; 1448 1449 PetscFunctionBegin; 1450 rtmp = (MatScalar *) PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1451 for ( i=0; i<n; i++ ) { 1452 nz = bi[i+1] - bi[i]; 1453 ajtmp = bj + bi[i]; 1454 for ( j=0; j<nz; j++ ) { 1455 x = rtmp+25*ajtmp[j]; 1456 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1457 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1458 x[16] = x[17] = x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1459 } 1460 /* load in initial (unfactored row) */ 1461 nz = ai[i+1] - ai[i]; 1462 ajtmpold = aj + ai[i]; 1463 v = aa + 25*ai[i]; 1464 for ( j=0; j<nz; j++ ) { 1465 x = rtmp+25*ajtmpold[j]; 1466 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1467 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1468 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1469 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; 1470 x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1471 x[24] = v[24]; 1472 v += 25; 1473 } 1474 row = *ajtmp++; 1475 while (row < i) { 1476 pc = rtmp + 25*row; 1477 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1478 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1479 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1480 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; 1481 p19 = pc[18]; p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; 1482 p24 = pc[23]; p25 = pc[24]; 1483 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1484 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1485 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1486 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 1487 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || p25 != 0.0) { 1488 pv = ba + 25*diag_offset[row]; 1489 pj = bj + diag_offset[row] + 1; 1490 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1491 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1492 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1493 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; 1494 x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1495 x25 = pv[24]; 1496 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1497 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1498 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1499 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1500 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1501 1502 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1503 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1504 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1505 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1506 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1507 1508 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1509 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1510 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1511 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1512 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1513 1514 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1515 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1516 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1517 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1518 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1519 1520 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1521 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1522 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1523 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1524 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1525 1526 nz = bi[row+1] - diag_offset[row] - 1; 1527 pv += 25; 1528 for (j=0; j<nz; j++) { 1529 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1530 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1531 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1532 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1533 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; 1534 x24 = pv[23]; x25 = pv[24]; 1535 x = rtmp + 25*pj[j]; 1536 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1537 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1538 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1539 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1540 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1541 1542 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1543 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1544 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1545 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1546 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1547 1548 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1549 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1550 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1551 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1552 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1553 1554 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1555 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1556 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1557 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1558 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1559 1560 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1561 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1562 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1563 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1564 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1565 pv += 25; 1566 } 1567 PLogFlops(250*nz+225); 1568 } 1569 row = *ajtmp++; 1570 } 1571 /* finished row so stick it into b->a */ 1572 pv = ba + 25*bi[i]; 1573 pj = bj + bi[i]; 1574 nz = bi[i+1] - bi[i]; 1575 for ( j=0; j<nz; j++ ) { 1576 x = rtmp+25*pj[j]; 1577 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1578 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1579 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1580 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; pv[17] = x[17]; 1581 pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; 1582 pv[23] = x[23]; pv[24] = x[24]; 1583 pv += 25; 1584 } 1585 /* invert diagonal block */ 1586 w = ba + 25*diag_offset[i]; 1587 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1588 } 1589 1590 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1591 C->factor = FACTOR_LU; 1592 C->assembled = PETSC_TRUE; 1593 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1594 PetscFunctionReturn(0); 1595 } 1596 1597 /* ------------------------------------------------------------*/ 1598 /* 1599 Version for when blocks are 4 by 4 1600 */ 1601 #undef __FUNC__ 1602 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_4" 1603 int MatLUFactorNumeric_SeqBAIJ_4(Mat A,Mat *B) 1604 { 1605 Mat C = *B; 1606 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1607 IS isrow = b->row, isicol = b->icol; 1608 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1609 int *ajtmpold, *ajtmp, nz, row; 1610 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 1611 register int *pj; 1612 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1613 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1614 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1615 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1616 MatScalar m13,m14,m15,m16; 1617 MatScalar *ba = b->a,*aa = a->a; 1618 1619 PetscFunctionBegin; 1620 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1621 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1622 rtmp = (MatScalar *) PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1623 1624 for ( i=0; i<n; i++ ) { 1625 nz = bi[i+1] - bi[i]; 1626 ajtmp = bj + bi[i]; 1627 for ( j=0; j<nz; j++ ) { 1628 x = rtmp+16*ajtmp[j]; 1629 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1630 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1631 } 1632 /* load in initial (unfactored row) */ 1633 idx = r[i]; 1634 nz = ai[idx+1] - ai[idx]; 1635 ajtmpold = aj + ai[idx]; 1636 v = aa + 16*ai[idx]; 1637 for ( j=0; j<nz; j++ ) { 1638 x = rtmp+16*ic[ajtmpold[j]]; 1639 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1640 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1641 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1642 x[14] = v[14]; x[15] = v[15]; 1643 v += 16; 1644 } 1645 row = *ajtmp++; 1646 while (row < i) { 1647 pc = rtmp + 16*row; 1648 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1649 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1650 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1651 p15 = pc[14]; p16 = pc[15]; 1652 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1653 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1654 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1655 || p16 != 0.0) { 1656 pv = ba + 16*diag_offset[row]; 1657 pj = bj + diag_offset[row] + 1; 1658 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1659 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1660 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1661 x15 = pv[14]; x16 = pv[15]; 1662 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1663 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1664 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1665 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1666 1667 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1668 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1669 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1670 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1671 1672 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1673 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1674 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1675 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1676 1677 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1678 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1679 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1680 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1681 1682 nz = bi[row+1] - diag_offset[row] - 1; 1683 pv += 16; 1684 for (j=0; j<nz; j++) { 1685 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1686 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1687 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1688 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1689 x = rtmp + 16*pj[j]; 1690 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1691 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1692 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1693 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1694 1695 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1696 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1697 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1698 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1699 1700 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1701 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1702 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1703 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1704 1705 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1706 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1707 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1708 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1709 1710 pv += 16; 1711 } 1712 PLogFlops(128*nz+112); 1713 } 1714 row = *ajtmp++; 1715 } 1716 /* finished row so stick it into b->a */ 1717 pv = ba + 16*bi[i]; 1718 pj = bj + bi[i]; 1719 nz = bi[i+1] - bi[i]; 1720 for ( j=0; j<nz; j++ ) { 1721 x = rtmp+16*pj[j]; 1722 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1723 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1724 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1725 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1726 pv += 16; 1727 } 1728 /* invert diagonal block */ 1729 w = ba + 16*diag_offset[i]; 1730 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1731 } 1732 1733 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1734 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1735 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1736 C->factor = FACTOR_LU; 1737 C->assembled = PETSC_TRUE; 1738 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1739 PetscFunctionReturn(0); 1740 } 1741 /* 1742 Version for when blocks are 4 by 4 Using natural ordering 1743 */ 1744 #undef __FUNC__ 1745 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_4_NaturalOrdering" 1746 int MatLUFactorNumeric_SeqBAIJ_4_NaturalOrdering(Mat A,Mat *B) 1747 { 1748 Mat C = *B; 1749 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1750 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1751 int *ajtmpold, *ajtmp, nz, row; 1752 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 1753 register int *pj; 1754 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1755 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1756 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1757 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1758 MatScalar m13,m14,m15,m16; 1759 MatScalar *ba = b->a,*aa = a->a; 1760 1761 PetscFunctionBegin; 1762 rtmp = (MatScalar *) PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1763 1764 for ( i=0; i<n; i++ ) { 1765 nz = bi[i+1] - bi[i]; 1766 ajtmp = bj + bi[i]; 1767 for ( j=0; j<nz; j++ ) { 1768 x = rtmp+16*ajtmp[j]; 1769 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1770 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1771 } 1772 /* load in initial (unfactored row) */ 1773 nz = ai[i+1] - ai[i]; 1774 ajtmpold = aj + ai[i]; 1775 v = aa + 16*ai[i]; 1776 for ( j=0; j<nz; j++ ) { 1777 x = rtmp+16*ajtmpold[j]; 1778 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1779 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1780 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1781 x[14] = v[14]; x[15] = v[15]; 1782 v += 16; 1783 } 1784 row = *ajtmp++; 1785 while (row < i) { 1786 pc = rtmp + 16*row; 1787 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1788 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1789 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1790 p15 = pc[14]; p16 = pc[15]; 1791 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1792 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1793 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1794 || p16 != 0.0) { 1795 pv = ba + 16*diag_offset[row]; 1796 pj = bj + diag_offset[row] + 1; 1797 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1798 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1799 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1800 x15 = pv[14]; x16 = pv[15]; 1801 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1802 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1803 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1804 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1805 1806 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1807 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1808 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1809 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1810 1811 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1812 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1813 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1814 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1815 1816 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1817 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1818 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1819 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1820 1821 nz = bi[row+1] - diag_offset[row] - 1; 1822 pv += 16; 1823 for (j=0; j<nz; j++) { 1824 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1825 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1826 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1827 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1828 x = rtmp + 16*pj[j]; 1829 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1830 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1831 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1832 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1833 1834 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1835 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1836 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1837 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1838 1839 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1840 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1841 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1842 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1843 1844 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1845 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1846 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1847 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1848 1849 pv += 16; 1850 } 1851 PLogFlops(128*nz+112); 1852 } 1853 row = *ajtmp++; 1854 } 1855 /* finished row so stick it into b->a */ 1856 pv = ba + 16*bi[i]; 1857 pj = bj + bi[i]; 1858 nz = bi[i+1] - bi[i]; 1859 for ( j=0; j<nz; j++ ) { 1860 x = rtmp+16*pj[j]; 1861 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1862 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1863 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1864 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1865 pv += 16; 1866 } 1867 /* invert diagonal block */ 1868 w = ba + 16*diag_offset[i]; 1869 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1870 } 1871 1872 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1873 C->factor = FACTOR_LU; 1874 C->assembled = PETSC_TRUE; 1875 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1876 PetscFunctionReturn(0); 1877 } 1878 1879 1880 /* ------------------------------------------------------------*/ 1881 /* 1882 Version for when blocks are 3 by 3 1883 */ 1884 #undef __FUNC__ 1885 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_3" 1886 int MatLUFactorNumeric_SeqBAIJ_3(Mat A,Mat *B) 1887 { 1888 Mat C = *B; 1889 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1890 IS isrow = b->row, isicol = b->icol; 1891 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1892 int *ajtmpold, *ajtmp, nz, row, *ai=a->i,*aj=a->j; 1893 int *diag_offset = b->diag,idx; 1894 register int *pj; 1895 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1896 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1897 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 1898 MatScalar *ba = b->a,*aa = a->a; 1899 1900 PetscFunctionBegin; 1901 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1902 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1903 rtmp = (MatScalar *) PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1904 1905 for ( i=0; i<n; i++ ) { 1906 nz = bi[i+1] - bi[i]; 1907 ajtmp = bj + bi[i]; 1908 for ( j=0; j<nz; j++ ) { 1909 x = rtmp + 9*ajtmp[j]; 1910 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 1911 } 1912 /* load in initial (unfactored row) */ 1913 idx = r[i]; 1914 nz = ai[idx+1] - ai[idx]; 1915 ajtmpold = aj + ai[idx]; 1916 v = aa + 9*ai[idx]; 1917 for ( j=0; j<nz; j++ ) { 1918 x = rtmp + 9*ic[ajtmpold[j]]; 1919 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1920 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1921 v += 9; 1922 } 1923 row = *ajtmp++; 1924 while (row < i) { 1925 pc = rtmp + 9*row; 1926 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1927 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1928 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1929 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 1930 pv = ba + 9*diag_offset[row]; 1931 pj = bj + diag_offset[row] + 1; 1932 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1933 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1934 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 1935 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 1936 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 1937 1938 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 1939 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 1940 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 1941 1942 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 1943 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 1944 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 1945 nz = bi[row+1] - diag_offset[row] - 1; 1946 pv += 9; 1947 for (j=0; j<nz; j++) { 1948 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1949 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1950 x = rtmp + 9*pj[j]; 1951 x[0] -= m1*x1 + m4*x2 + m7*x3; 1952 x[1] -= m2*x1 + m5*x2 + m8*x3; 1953 x[2] -= m3*x1 + m6*x2 + m9*x3; 1954 1955 x[3] -= m1*x4 + m4*x5 + m7*x6; 1956 x[4] -= m2*x4 + m5*x5 + m8*x6; 1957 x[5] -= m3*x4 + m6*x5 + m9*x6; 1958 1959 x[6] -= m1*x7 + m4*x8 + m7*x9; 1960 x[7] -= m2*x7 + m5*x8 + m8*x9; 1961 x[8] -= m3*x7 + m6*x8 + m9*x9; 1962 pv += 9; 1963 } 1964 PLogFlops(54*nz+36); 1965 } 1966 row = *ajtmp++; 1967 } 1968 /* finished row so stick it into b->a */ 1969 pv = ba + 9*bi[i]; 1970 pj = bj + bi[i]; 1971 nz = bi[i+1] - bi[i]; 1972 for ( j=0; j<nz; j++ ) { 1973 x = rtmp + 9*pj[j]; 1974 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1975 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1976 pv += 9; 1977 } 1978 /* invert diagonal block */ 1979 w = ba + 9*diag_offset[i]; 1980 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 1981 } 1982 1983 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1984 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1985 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1986 C->factor = FACTOR_LU; 1987 C->assembled = PETSC_TRUE; 1988 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 1989 PetscFunctionReturn(0); 1990 } 1991 /* 1992 Version for when blocks are 3 by 3 Using natural ordering 1993 */ 1994 #undef __FUNC__ 1995 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_3_NaturalOrdering" 1996 int MatLUFactorNumeric_SeqBAIJ_3_NaturalOrdering(Mat A,Mat *B) 1997 { 1998 Mat C = *B; 1999 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 2000 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 2001 int *ajtmpold, *ajtmp, nz, row; 2002 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 2003 register int *pj; 2004 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2005 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 2006 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 2007 MatScalar *ba = b->a,*aa = a->a; 2008 2009 PetscFunctionBegin; 2010 rtmp = (MatScalar *) PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2011 2012 for ( i=0; i<n; i++ ) { 2013 nz = bi[i+1] - bi[i]; 2014 ajtmp = bj + bi[i]; 2015 for ( j=0; j<nz; j++ ) { 2016 x = rtmp+9*ajtmp[j]; 2017 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 2018 } 2019 /* load in initial (unfactored row) */ 2020 nz = ai[i+1] - ai[i]; 2021 ajtmpold = aj + ai[i]; 2022 v = aa + 9*ai[i]; 2023 for ( j=0; j<nz; j++ ) { 2024 x = rtmp+9*ajtmpold[j]; 2025 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2026 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 2027 v += 9; 2028 } 2029 row = *ajtmp++; 2030 while (row < i) { 2031 pc = rtmp + 9*row; 2032 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2033 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 2034 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 2035 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 2036 pv = ba + 9*diag_offset[row]; 2037 pj = bj + diag_offset[row] + 1; 2038 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2039 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2040 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 2041 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 2042 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 2043 2044 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 2045 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 2046 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 2047 2048 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 2049 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 2050 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 2051 2052 nz = bi[row+1] - diag_offset[row] - 1; 2053 pv += 9; 2054 for (j=0; j<nz; j++) { 2055 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2056 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2057 x = rtmp + 9*pj[j]; 2058 x[0] -= m1*x1 + m4*x2 + m7*x3; 2059 x[1] -= m2*x1 + m5*x2 + m8*x3; 2060 x[2] -= m3*x1 + m6*x2 + m9*x3; 2061 2062 x[3] -= m1*x4 + m4*x5 + m7*x6; 2063 x[4] -= m2*x4 + m5*x5 + m8*x6; 2064 x[5] -= m3*x4 + m6*x5 + m9*x6; 2065 2066 x[6] -= m1*x7 + m4*x8 + m7*x9; 2067 x[7] -= m2*x7 + m5*x8 + m8*x9; 2068 x[8] -= m3*x7 + m6*x8 + m9*x9; 2069 pv += 9; 2070 } 2071 PLogFlops(54*nz+36); 2072 } 2073 row = *ajtmp++; 2074 } 2075 /* finished row so stick it into b->a */ 2076 pv = ba + 9*bi[i]; 2077 pj = bj + bi[i]; 2078 nz = bi[i+1] - bi[i]; 2079 for ( j=0; j<nz; j++ ) { 2080 x = rtmp+9*pj[j]; 2081 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2082 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 2083 pv += 9; 2084 } 2085 /* invert diagonal block */ 2086 w = ba + 9*diag_offset[i]; 2087 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 2088 } 2089 2090 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2091 C->factor = FACTOR_LU; 2092 C->assembled = PETSC_TRUE; 2093 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 2094 PetscFunctionReturn(0); 2095 } 2096 2097 /* ------------------------------------------------------------*/ 2098 /* 2099 Version for when blocks are 2 by 2 2100 */ 2101 #undef __FUNC__ 2102 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_2" 2103 int MatLUFactorNumeric_SeqBAIJ_2(Mat A,Mat *B) 2104 { 2105 Mat C = *B; 2106 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 2107 IS isrow = b->row, isicol = b->icol; 2108 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 2109 int *ajtmpold, *ajtmp, nz, row; 2110 int *diag_offset=b->diag,idx,*ai=a->i,*aj=a->j; 2111 register int *pj; 2112 register MatScalar *pv,*v,*rtmp,m1,m2,m3,m4,*pc,*w,*x,x1,x2,x3,x4; 2113 MatScalar p1,p2,p3,p4; 2114 MatScalar *ba = b->a,*aa = a->a; 2115 2116 PetscFunctionBegin; 2117 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 2118 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 2119 rtmp = (MatScalar *) PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2120 2121 for ( i=0; i<n; i++ ) { 2122 nz = bi[i+1] - bi[i]; 2123 ajtmp = bj + bi[i]; 2124 for ( j=0; j<nz; j++ ) { 2125 x = rtmp+4*ajtmp[j]; x[0] = x[1] = x[2] = x[3] = 0.0; 2126 } 2127 /* load in initial (unfactored row) */ 2128 idx = r[i]; 2129 nz = ai[idx+1] - ai[idx]; 2130 ajtmpold = aj + ai[idx]; 2131 v = aa + 4*ai[idx]; 2132 for ( j=0; j<nz; j++ ) { 2133 x = rtmp+4*ic[ajtmpold[j]]; 2134 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2135 v += 4; 2136 } 2137 row = *ajtmp++; 2138 while (row < i) { 2139 pc = rtmp + 4*row; 2140 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2141 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2142 pv = ba + 4*diag_offset[row]; 2143 pj = bj + diag_offset[row] + 1; 2144 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2145 pc[0] = m1 = p1*x1 + p3*x2; 2146 pc[1] = m2 = p2*x1 + p4*x2; 2147 pc[2] = m3 = p1*x3 + p3*x4; 2148 pc[3] = m4 = p2*x3 + p4*x4; 2149 nz = bi[row+1] - diag_offset[row] - 1; 2150 pv += 4; 2151 for (j=0; j<nz; j++) { 2152 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2153 x = rtmp + 4*pj[j]; 2154 x[0] -= m1*x1 + m3*x2; 2155 x[1] -= m2*x1 + m4*x2; 2156 x[2] -= m1*x3 + m3*x4; 2157 x[3] -= m2*x3 + m4*x4; 2158 pv += 4; 2159 } 2160 PLogFlops(16*nz+12); 2161 } 2162 row = *ajtmp++; 2163 } 2164 /* finished row so stick it into b->a */ 2165 pv = ba + 4*bi[i]; 2166 pj = bj + bi[i]; 2167 nz = bi[i+1] - bi[i]; 2168 for ( j=0; j<nz; j++ ) { 2169 x = rtmp+4*pj[j]; 2170 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2171 pv += 4; 2172 } 2173 /* invert diagonal block */ 2174 w = ba + 4*diag_offset[i]; 2175 ierr = Kernel_A_gets_inverse_A_2(w);CHKERRQ(ierr); 2176 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2177 } 2178 2179 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2180 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 2181 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 2182 C->factor = FACTOR_LU; 2183 C->assembled = PETSC_TRUE; 2184 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2185 PetscFunctionReturn(0); 2186 } 2187 /* 2188 Version for when blocks are 2 by 2 Using natural ordering 2189 */ 2190 #undef __FUNC__ 2191 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering" 2192 int MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering(Mat A,Mat *B) 2193 { 2194 Mat C = *B; 2195 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 2196 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 2197 int *ajtmpold, *ajtmp, nz, row; 2198 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 2199 register int *pj; 2200 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2201 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,x1,x2,x3,x4; 2202 MatScalar *ba = b->a,*aa = a->a; 2203 2204 PetscFunctionBegin; 2205 rtmp = (MatScalar *) PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2206 2207 for ( i=0; i<n; i++ ) { 2208 nz = bi[i+1] - bi[i]; 2209 ajtmp = bj + bi[i]; 2210 for ( j=0; j<nz; j++ ) { 2211 x = rtmp+4*ajtmp[j]; 2212 x[0] = x[1] = x[2] = x[3] = 0.0; 2213 } 2214 /* load in initial (unfactored row) */ 2215 nz = ai[i+1] - ai[i]; 2216 ajtmpold = aj + ai[i]; 2217 v = aa + 4*ai[i]; 2218 for ( j=0; j<nz; j++ ) { 2219 x = rtmp+4*ajtmpold[j]; 2220 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2221 v += 4; 2222 } 2223 row = *ajtmp++; 2224 while (row < i) { 2225 pc = rtmp + 4*row; 2226 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2227 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2228 pv = ba + 4*diag_offset[row]; 2229 pj = bj + diag_offset[row] + 1; 2230 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2231 pc[0] = m1 = p1*x1 + p3*x2; 2232 pc[1] = m2 = p2*x1 + p4*x2; 2233 pc[2] = m3 = p1*x3 + p3*x4; 2234 pc[3] = m4 = p2*x3 + p4*x4; 2235 nz = bi[row+1] - diag_offset[row] - 1; 2236 pv += 4; 2237 for (j=0; j<nz; j++) { 2238 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2239 x = rtmp + 4*pj[j]; 2240 x[0] -= m1*x1 + m3*x2; 2241 x[1] -= m2*x1 + m4*x2; 2242 x[2] -= m1*x3 + m3*x4; 2243 x[3] -= m2*x3 + m4*x4; 2244 pv += 4; 2245 } 2246 PLogFlops(16*nz+12); 2247 } 2248 row = *ajtmp++; 2249 } 2250 /* finished row so stick it into b->a */ 2251 pv = ba + 4*bi[i]; 2252 pj = bj + bi[i]; 2253 nz = bi[i+1] - bi[i]; 2254 for ( j=0; j<nz; j++ ) { 2255 x = rtmp+4*pj[j]; 2256 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2257 pv += 4; 2258 } 2259 /* invert diagonal block */ 2260 w = ba + 4*diag_offset[i]; 2261 ierr = Kernel_A_gets_inverse_A_2(w);CHKERRQ(ierr); 2262 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2263 } 2264 2265 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2266 C->factor = FACTOR_LU; 2267 C->assembled = PETSC_TRUE; 2268 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2269 PetscFunctionReturn(0); 2270 } 2271 2272 /* ----------------------------------------------------------- */ 2273 /* 2274 Version for when blocks are 1 by 1. 2275 */ 2276 #undef __FUNC__ 2277 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_1" 2278 int MatLUFactorNumeric_SeqBAIJ_1(Mat A,Mat *B) 2279 { 2280 Mat C = *B; 2281 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data, *b = (Mat_SeqBAIJ *)C->data; 2282 IS isrow = b->row, isicol = b->icol; 2283 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 2284 int *ajtmpold, *ajtmp, nz, row,*ai = a->i,*aj = a->j; 2285 int *diag_offset = b->diag,diag; 2286 register int *pj; 2287 register MatScalar *pv,*v,*rtmp,multiplier,*pc; 2288 MatScalar *ba = b->a,*aa = a->a; 2289 2290 PetscFunctionBegin; 2291 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 2292 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 2293 rtmp = (MatScalar *) PetscMalloc((n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2294 2295 for ( i=0; i<n; i++ ) { 2296 nz = bi[i+1] - bi[i]; 2297 ajtmp = bj + bi[i]; 2298 for ( j=0; j<nz; j++ ) rtmp[ajtmp[j]] = 0.0; 2299 2300 /* load in initial (unfactored row) */ 2301 nz = ai[r[i]+1] - ai[r[i]]; 2302 ajtmpold = aj + ai[r[i]]; 2303 v = aa + ai[r[i]]; 2304 for ( j=0; j<nz; j++ ) rtmp[ic[ajtmpold[j]]] = v[j]; 2305 2306 row = *ajtmp++; 2307 while (row < i) { 2308 pc = rtmp + row; 2309 if (*pc != 0.0) { 2310 pv = ba + diag_offset[row]; 2311 pj = bj + diag_offset[row] + 1; 2312 multiplier = *pc * *pv++; 2313 *pc = multiplier; 2314 nz = bi[row+1] - diag_offset[row] - 1; 2315 for (j=0; j<nz; j++) rtmp[pj[j]] -= multiplier * pv[j]; 2316 PLogFlops(1+2*nz); 2317 } 2318 row = *ajtmp++; 2319 } 2320 /* finished row so stick it into b->a */ 2321 pv = ba + bi[i]; 2322 pj = bj + bi[i]; 2323 nz = bi[i+1] - bi[i]; 2324 for ( j=0; j<nz; j++ ) {pv[j] = rtmp[pj[j]];} 2325 diag = diag_offset[i] - bi[i]; 2326 /* check pivot entry for current row */ 2327 if (pv[diag] == 0.0) { 2328 SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,0,"Zero pivot"); 2329 } 2330 pv[diag] = 1.0/pv[diag]; 2331 } 2332 2333 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2334 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 2335 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 2336 C->factor = FACTOR_LU; 2337 C->assembled = PETSC_TRUE; 2338 PLogFlops(b->n); 2339 PetscFunctionReturn(0); 2340 } 2341 2342 /* ----------------------------------------------------------- */ 2343 #undef __FUNC__ 2344 #define __FUNC__ "MatLUFactor_SeqBAIJ" 2345 int MatLUFactor_SeqBAIJ(Mat A,IS row,IS col,double f) 2346 { 2347 Mat_SeqBAIJ *mat = (Mat_SeqBAIJ *) A->data; 2348 int ierr; 2349 Mat C; 2350 PetscOps *Abops; 2351 MatOps Aops; 2352 2353 PetscFunctionBegin; 2354 ierr = MatLUFactorSymbolic(A,row,col,f,&C);CHKERRQ(ierr); 2355 ierr = MatLUFactorNumeric(A,&C);CHKERRQ(ierr); 2356 2357 /* free all the data structures from mat */ 2358 ierr = PetscFree(mat->a);CHKERRQ(ierr); 2359 if (!mat->singlemalloc) { 2360 ierr = PetscFree(mat->i);CHKERRQ(ierr); 2361 ierr = PetscFree(mat->j);CHKERRQ(ierr); 2362 } 2363 if (mat->diag) {ierr = PetscFree(mat->diag);CHKERRQ(ierr);} 2364 if (mat->ilen) {ierr = PetscFree(mat->ilen);CHKERRQ(ierr);} 2365 if (mat->imax) {ierr = PetscFree(mat->imax);CHKERRQ(ierr);} 2366 if (mat->solve_work) {ierr = PetscFree(mat->solve_work);CHKERRQ(ierr);} 2367 if (mat->mult_work) {ierr = PetscFree(mat->mult_work);CHKERRQ(ierr);} 2368 if (mat->icol) {ierr = ISDestroy(mat->icol);CHKERRQ(ierr);} 2369 ierr = PetscFree(mat);CHKERRQ(ierr); 2370 2371 ierr = MapDestroy(A->rmap);CHKERRQ(ierr); 2372 ierr = MapDestroy(A->cmap);CHKERRQ(ierr); 2373 2374 /* 2375 This is horrible, horrible code. We need to keep the 2376 A pointers for the bops and ops but copy everything 2377 else from C. 2378 */ 2379 Abops = A->bops; 2380 Aops = A->ops; 2381 ierr = PetscMemcpy(A,C,sizeof(struct _p_Mat));CHKERRQ(ierr); 2382 mat = (Mat_SeqBAIJ *) A->data; 2383 PLogObjectParent(A,mat->icol); 2384 2385 A->bops = Abops; 2386 A->ops = Aops; 2387 A->qlist = 0; 2388 /* copy over the type_name and name */ 2389 ierr = PetscStrallocpy(C->type_name,&A->type_name);CHKERRQ(ierr); 2390 ierr = PetscStrallocpy(C->name,&A->name);CHKERRQ(ierr); 2391 2392 PetscHeaderDestroy(C); 2393 PetscFunctionReturn(0); 2394 } 2395 2396 2397 2398