1 #ifdef PETSC_RCS_HEADER 2 static char vcid[] = "$Id: baijfact.c,v 1.72 1999/05/04 20:32:27 balay Exp balay $"; 3 #endif 4 /* 5 Factorization code for BAIJ format. 6 */ 7 8 #include "src/mat/impls/baij/seq/baij.h" 9 #include "src/vec/vecimpl.h" 10 #include "src/inline/ilu.h" 11 12 13 /* 14 The symbolic factorization code is identical to that for AIJ format, 15 except for very small changes since this is now a SeqBAIJ datastructure. 16 NOT good code reuse. 17 */ 18 #undef __FUNC__ 19 #define __FUNC__ "MatLUFactorSymbolic_SeqBAIJ" 20 int MatLUFactorSymbolic_SeqBAIJ(Mat A,IS isrow,IS iscol,double f,Mat *B) 21 { 22 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data, *b; 23 IS isicol; 24 int *r,*ic, ierr, i, n = a->mbs, *ai = a->i, *aj = a->j; 25 int *ainew,*ajnew, jmax,*fill, *ajtmp, nz, bs = a->bs, bs2=a->bs2; 26 int *idnew, idx, row,m,fm, nnz, nzi,realloc = 0,nzbd,*im; 27 28 PetscFunctionBegin; 29 PetscValidHeaderSpecific(isrow,IS_COOKIE); 30 PetscValidHeaderSpecific(iscol,IS_COOKIE); 31 ierr = ISInvertPermutation(iscol,&isicol);CHKERRQ(ierr); 32 ISGetIndices(isrow,&r); ISGetIndices(isicol,&ic); 33 34 /* get new row pointers */ 35 ainew = (int *) PetscMalloc( (n+1)*sizeof(int) );CHKPTRQ(ainew); 36 ainew[0] = 0; 37 /* don't know how many column pointers are needed so estimate */ 38 jmax = (int) (f*ai[n] + 1); 39 ajnew = (int *) PetscMalloc( (jmax)*sizeof(int) );CHKPTRQ(ajnew); 40 /* fill is a linked list of nonzeros in active row */ 41 fill = (int *) PetscMalloc( (2*n+1)*sizeof(int));CHKPTRQ(fill); 42 im = fill + n + 1; 43 /* idnew is location of diagonal in factor */ 44 idnew = (int *) PetscMalloc( (n+1)*sizeof(int));CHKPTRQ(idnew); 45 idnew[0] = 0; 46 47 for ( i=0; i<n; i++ ) { 48 /* first copy previous fill into linked list */ 49 nnz = nz = ai[r[i]+1] - ai[r[i]]; 50 if (!nz) SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,1,"Empty row in matrix"); 51 ajtmp = aj + ai[r[i]]; 52 fill[n] = n; 53 while (nz--) { 54 fm = n; 55 idx = ic[*ajtmp++]; 56 do { 57 m = fm; 58 fm = fill[m]; 59 } while (fm < idx); 60 fill[m] = idx; 61 fill[idx] = fm; 62 } 63 row = fill[n]; 64 while ( row < i ) { 65 ajtmp = ajnew + idnew[row] + 1; 66 nzbd = 1 + idnew[row] - ainew[row]; 67 nz = im[row] - nzbd; 68 fm = row; 69 while (nz-- > 0) { 70 idx = *ajtmp++; 71 nzbd++; 72 if (idx == i) im[row] = nzbd; 73 do { 74 m = fm; 75 fm = fill[m]; 76 } while (fm < idx); 77 if (fm != idx) { 78 fill[m] = idx; 79 fill[idx] = fm; 80 fm = idx; 81 nnz++; 82 } 83 } 84 row = fill[row]; 85 } 86 /* copy new filled row into permanent storage */ 87 ainew[i+1] = ainew[i] + nnz; 88 if (ainew[i+1] > jmax) { 89 90 /* estimate how much additional space we will need */ 91 /* use the strategy suggested by David Hysom <hysom@perch-t.icase.edu> */ 92 /* just double the memory each time */ 93 int maxadd = jmax; 94 /* maxadd = (int) ((f*(ai[n]+1)*(n-i+5))/n); */ 95 if (maxadd < nnz) maxadd = (n-i)*(nnz+1); 96 jmax += maxadd; 97 98 /* allocate a longer ajnew */ 99 ajtmp = (int *) PetscMalloc( jmax*sizeof(int) );CHKPTRQ(ajtmp); 100 ierr = PetscMemcpy(ajtmp,ajnew,ainew[i]*sizeof(int));CHKERRQ(ierr); 101 ierr = PetscFree(ajnew);CHKERRQ(ierr); 102 ajnew = ajtmp; 103 realloc++; /* count how many times we realloc */ 104 } 105 ajtmp = ajnew + ainew[i]; 106 fm = fill[n]; 107 nzi = 0; 108 im[i] = nnz; 109 while (nnz--) { 110 if (fm < i) nzi++; 111 *ajtmp++ = fm; 112 fm = fill[fm]; 113 } 114 idnew[i] = ainew[i] + nzi; 115 } 116 117 if (ai[n] != 0) { 118 double af = ((double)ainew[n])/((double)ai[n]); 119 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:Reallocs %d Fill ratio:given %g needed %g\n", 120 realloc,f,af); 121 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:Run with -pc_lu_fill %g or use \n",af); 122 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:PCLUSetFill(pc,%g);\n",af); 123 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:for best performance.\n"); 124 } else { 125 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:Empty matrix.\n"); 126 } 127 128 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 129 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 130 131 ierr = PetscFree(fill);CHKERRQ(ierr); 132 133 /* put together the new matrix */ 134 ierr = MatCreateSeqBAIJ(A->comm,bs,bs*n,bs*n,0,PETSC_NULL,B);CHKERRQ(ierr); 135 PLogObjectParent(*B,isicol); 136 b = (Mat_SeqBAIJ *) (*B)->data; 137 ierr = PetscFree(b->imax);CHKERRQ(ierr); 138 b->singlemalloc = 0; 139 /* the next line frees the default space generated by the Create() */ 140 ierr = PetscFree(b->a);CHKERRQ(ierr); 141 ierr = PetscFree(b->ilen);CHKERRQ(ierr); 142 b->a = (MatScalar *) PetscMalloc((ainew[n]+1)*sizeof(MatScalar)*bs2);CHKPTRQ(b->a); 143 b->j = ajnew; 144 b->i = ainew; 145 b->diag = idnew; 146 b->ilen = 0; 147 b->imax = 0; 148 b->row = isrow; 149 b->col = iscol; 150 b->icol = isicol; 151 b->solve_work = (Scalar *) PetscMalloc( (bs*n+bs)*sizeof(Scalar));CHKPTRQ(b->solve_work); 152 /* In b structure: Free imax, ilen, old a, old j. 153 Allocate idnew, solve_work, new a, new j */ 154 PLogObjectMemory(*B,(ainew[n]-n)*(sizeof(int)+sizeof(MatScalar))); 155 b->maxnz = b->nz = ainew[n]; 156 157 (*B)->factor = FACTOR_LU; 158 (*B)->info.factor_mallocs = realloc; 159 (*B)->info.fill_ratio_given = f; 160 if (ai[n] != 0) { 161 (*B)->info.fill_ratio_needed = ((double)ainew[n])/((double)ai[n]); 162 } else { 163 (*B)->info.fill_ratio_needed = 0.0; 164 } 165 166 167 PetscFunctionReturn(0); 168 } 169 170 /* ----------------------------------------------------------- */ 171 #undef __FUNC__ 172 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_N" 173 int MatLUFactorNumeric_SeqBAIJ_N(Mat A,Mat *B) 174 { 175 Mat C = *B; 176 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 177 IS isrow = b->row, isicol = b->icol; 178 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 179 int *ajtmpold, *ajtmp, nz, row, bslog,*ai=a->i,*aj=a->j,k,flg; 180 int *diag_offset=b->diag,diag,bs=a->bs,bs2 = a->bs2,*v_pivots; 181 register int *pj; 182 register MatScalar *pv,*v,*rtmp,*multiplier,*v_work,*pc,*w; 183 MatScalar *ba = b->a,*aa = a->a; 184 185 PetscFunctionBegin; 186 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 187 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 188 rtmp = (MatScalar *) PetscMalloc(bs2*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 189 ierr = PetscMemzero(rtmp,bs2*(n+1)*sizeof(MatScalar));CHKERRQ(ierr); 190 /* generate work space needed by dense LU factorization */ 191 v_work = (MatScalar *) PetscMalloc(bs*sizeof(int) + (bs+bs2)*sizeof(MatScalar));CHKPTRQ(v_work); 192 multiplier = v_work + bs; 193 v_pivots = (int *) (multiplier + bs2); 194 195 /* flops in while loop */ 196 bslog = 2*bs*bs2; 197 198 for ( i=0; i<n; i++ ) { 199 nz = bi[i+1] - bi[i]; 200 ajtmp = bj + bi[i]; 201 for ( j=0; j<nz; j++ ) { 202 ierr = PetscMemzero(rtmp+bs2*ajtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 203 } 204 /* load in initial (unfactored row) */ 205 nz = ai[r[i]+1] - ai[r[i]]; 206 ajtmpold = aj + ai[r[i]]; 207 v = aa + bs2*ai[r[i]]; 208 for ( j=0; j<nz; j++ ) { 209 ierr = PetscMemcpy(rtmp+bs2*ic[ajtmpold[j]],v+bs2*j,bs2*sizeof(MatScalar));CHKERRQ(ierr); 210 } 211 row = *ajtmp++; 212 while (row < i) { 213 pc = rtmp + bs2*row; 214 /* if (*pc) { */ 215 for ( flg=0,k=0; k<bs2; k++ ) { if (pc[k]!=0.0) { flg =1; break; }} 216 if (flg) { 217 pv = ba + bs2*diag_offset[row]; 218 pj = bj + diag_offset[row] + 1; 219 Kernel_A_gets_A_times_B(bs,pc,pv,multiplier); 220 nz = bi[row+1] - diag_offset[row] - 1; 221 pv += bs2; 222 for (j=0; j<nz; j++) { 223 Kernel_A_gets_A_minus_B_times_C(bs,rtmp+bs2*pj[j],pc,pv+bs2*j); 224 } 225 PLogFlops(bslog*(nz+1)-bs); 226 } 227 row = *ajtmp++; 228 } 229 /* finished row so stick it into b->a */ 230 pv = ba + bs2*bi[i]; 231 pj = bj + bi[i]; 232 nz = bi[i+1] - bi[i]; 233 for ( j=0; j<nz; j++ ) { 234 ierr = PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 235 } 236 diag = diag_offset[i] - bi[i]; 237 /* invert diagonal block */ 238 w = pv + bs2*diag; 239 Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work); 240 } 241 242 ierr = PetscFree(rtmp);CHKERRQ(ierr); 243 ierr = PetscFree(v_work);CHKERRQ(ierr); 244 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 245 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 246 C->factor = FACTOR_LU; 247 C->assembled = PETSC_TRUE; 248 PLogFlops(1.3333*bs*bs2*b->mbs); /* from inverting diagonal blocks */ 249 PetscFunctionReturn(0); 250 } 251 /* ------------------------------------------------------------*/ 252 /* 253 Version for when blocks are 7 by 7 254 */ 255 #undef __FUNC__ 256 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_7" 257 int MatLUFactorNumeric_SeqBAIJ_7(Mat A,Mat *B) 258 { 259 Mat C = *B; 260 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 261 IS isrow = b->row, isicol = b->icol; 262 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 263 int *ajtmpold, *ajtmp, nz, row; 264 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 265 register int *pj; 266 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 267 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 268 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 269 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 270 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 271 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 272 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 273 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 274 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 275 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 276 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 277 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 278 MatScalar *ba = b->a,*aa = a->a; 279 280 PetscFunctionBegin; 281 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 282 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 283 rtmp = (MatScalar *) PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 284 285 for ( i=0; i<n; i++ ) { 286 nz = bi[i+1] - bi[i]; 287 ajtmp = bj + bi[i]; 288 for ( j=0; j<nz; j++ ) { 289 x = rtmp+49*ajtmp[j]; 290 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 291 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 292 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 293 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 294 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 295 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 296 } 297 /* load in initial (unfactored row) */ 298 idx = r[i]; 299 nz = ai[idx+1] - ai[idx]; 300 ajtmpold = aj + ai[idx]; 301 v = aa + 49*ai[idx]; 302 for ( j=0; j<nz; j++ ) { 303 x = rtmp+49*ic[ajtmpold[j]]; 304 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 305 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 306 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 307 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 308 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 309 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 310 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 311 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 312 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 313 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 314 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 315 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 316 x[48] = v[48]; 317 v += 49; 318 } 319 row = *ajtmp++; 320 while (row < i) { 321 pc = rtmp + 49*row; 322 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 323 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 324 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 325 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 326 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 327 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 328 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 329 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 330 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 331 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 332 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 333 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 334 p49 = pc[48]; 335 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 336 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 337 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 338 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 339 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 340 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 341 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 342 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 343 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 344 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 345 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 346 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 347 p49 != 0.0) { 348 pv = ba + 49*diag_offset[row]; 349 pj = bj + diag_offset[row] + 1; 350 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 351 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 352 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 353 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 354 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 355 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 356 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 357 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 358 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 359 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 360 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 361 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 362 x49 = pv[48]; 363 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 364 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 365 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 366 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 367 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 368 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 369 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 370 371 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 372 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 373 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 374 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 375 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 376 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 377 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 378 379 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 380 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 381 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 382 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 383 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 384 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 385 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 386 387 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 388 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 389 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 390 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 391 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 392 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 393 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 394 395 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 396 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 397 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 398 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 399 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 400 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 401 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 402 403 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 404 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 405 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 406 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 407 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 408 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 409 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 410 411 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 412 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 413 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 414 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 415 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 416 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 417 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 418 419 nz = bi[row+1] - diag_offset[row] - 1; 420 pv += 49; 421 for (j=0; j<nz; j++) { 422 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 423 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 424 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 425 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 426 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 427 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 428 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 429 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 430 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 431 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 432 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 433 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 434 x49 = pv[48]; 435 x = rtmp + 49*pj[j]; 436 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 437 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 438 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 439 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 440 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 441 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 442 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 443 444 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 445 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 446 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 447 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 448 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 449 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 450 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 451 452 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 453 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 454 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 455 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 456 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 457 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 458 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 459 460 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 461 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 462 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 463 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 464 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 465 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 466 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 467 468 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 469 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 470 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 471 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 472 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 473 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 474 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 475 476 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 477 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 478 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 479 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 480 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 481 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 482 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 483 484 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 485 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 486 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 487 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 488 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 489 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 490 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 491 pv += 49; 492 } 493 PLogFlops(686*nz+637); 494 } 495 row = *ajtmp++; 496 } 497 /* finished row so stick it into b->a */ 498 pv = ba + 49*bi[i]; 499 pj = bj + bi[i]; 500 nz = bi[i+1] - bi[i]; 501 for ( j=0; j<nz; j++ ) { 502 x = rtmp+49*pj[j]; 503 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 504 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 505 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 506 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 507 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 508 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 509 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 510 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 511 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 512 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 513 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 514 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 515 pv[48] = x[48]; 516 pv += 49; 517 } 518 /* invert diagonal block */ 519 w = ba + 49*diag_offset[i]; 520 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 521 } 522 523 ierr = PetscFree(rtmp);CHKERRQ(ierr); 524 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 525 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 526 C->factor = FACTOR_LU; 527 C->assembled = PETSC_TRUE; 528 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 529 PetscFunctionReturn(0); 530 } 531 532 /* 533 Version for when blocks are 7 by 7 Using natural ordering 534 */ 535 #undef __FUNC__ 536 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_7_NaturalOrdering" 537 int MatLUFactorNumeric_SeqBAIJ_7_NaturalOrdering(Mat A,Mat *B) 538 { 539 Mat C = *B; 540 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 541 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 542 int *ajtmpold, *ajtmp, nz, row; 543 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 544 register int *pj; 545 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 546 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 547 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 548 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 549 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 550 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 551 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 552 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 553 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 554 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 555 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 556 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 557 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 558 MatScalar *ba = b->a,*aa = a->a; 559 560 PetscFunctionBegin; 561 rtmp = (MatScalar *) PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 562 for ( i=0; i<n; i++ ) { 563 nz = bi[i+1] - bi[i]; 564 ajtmp = bj + bi[i]; 565 for ( j=0; j<nz; j++ ) { 566 x = rtmp+49*ajtmp[j]; 567 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 568 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 569 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 570 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 571 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 572 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 573 } 574 /* load in initial (unfactored row) */ 575 nz = ai[i+1] - ai[i]; 576 ajtmpold = aj + ai[i]; 577 v = aa + 49*ai[i]; 578 for ( j=0; j<nz; j++ ) { 579 x = rtmp+49*ajtmpold[j]; 580 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 581 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 582 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 583 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 584 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 585 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 586 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 587 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 588 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 589 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 590 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 591 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 592 x[48] = v[48]; 593 v += 49; 594 } 595 row = *ajtmp++; 596 while (row < i) { 597 pc = rtmp + 49*row; 598 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 599 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 600 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 601 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 602 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 603 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 604 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 605 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 606 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 607 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 608 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 609 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 610 p49 = pc[48]; 611 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 612 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 613 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 614 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 615 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 616 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 617 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 618 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 619 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 620 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 621 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 622 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 623 p49 != 0.0) { 624 pv = ba + 49*diag_offset[row]; 625 pj = bj + diag_offset[row] + 1; 626 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 627 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 628 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 629 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 630 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 631 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 632 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 633 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 634 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 635 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 636 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 637 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 638 x49 = pv[48]; 639 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 640 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 641 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 642 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 643 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 644 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 645 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 646 647 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 648 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 649 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 650 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 651 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 652 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 653 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 654 655 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 656 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 657 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 658 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 659 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 660 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 661 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 662 663 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 664 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 665 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 666 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 667 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 668 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 669 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 670 671 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 672 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 673 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 674 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 675 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 676 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 677 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 678 679 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 680 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 681 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 682 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 683 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 684 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 685 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 686 687 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 688 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 689 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 690 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 691 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 692 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 693 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 694 695 nz = bi[row+1] - diag_offset[row] - 1; 696 pv += 49; 697 for (j=0; j<nz; j++) { 698 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 699 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 700 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 701 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 702 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 703 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 704 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 705 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 706 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 707 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 708 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 709 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 710 x49 = pv[48]; 711 x = rtmp + 49*pj[j]; 712 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 713 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 714 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 715 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 716 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 717 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 718 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 719 720 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 721 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 722 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 723 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 724 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 725 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 726 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 727 728 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 729 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 730 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 731 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 732 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 733 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 734 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 735 736 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 737 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 738 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 739 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 740 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 741 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 742 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 743 744 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 745 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 746 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 747 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 748 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 749 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 750 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 751 752 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 753 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 754 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 755 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 756 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 757 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 758 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 759 760 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 761 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 762 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 763 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 764 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 765 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 766 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 767 pv += 49; 768 } 769 PLogFlops(686*nz+637); 770 } 771 row = *ajtmp++; 772 } 773 /* finished row so stick it into b->a */ 774 pv = ba + 49*bi[i]; 775 pj = bj + bi[i]; 776 nz = bi[i+1] - bi[i]; 777 for ( j=0; j<nz; j++ ) { 778 x = rtmp+49*pj[j]; 779 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 780 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 781 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 782 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 783 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 784 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 785 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 786 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 787 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 788 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 789 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 790 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 791 pv[48] = x[48]; 792 pv += 49; 793 } 794 /* invert diagonal block */ 795 w = ba + 49*diag_offset[i]; 796 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 797 } 798 799 ierr = PetscFree(rtmp);CHKERRQ(ierr); 800 C->factor = FACTOR_LU; 801 C->assembled = PETSC_TRUE; 802 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 803 PetscFunctionReturn(0); 804 } 805 806 /* ------------------------------------------------------------*/ 807 /* 808 Version for when blocks are 6 by 6 809 */ 810 #undef __FUNC__ 811 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_6" 812 int MatLUFactorNumeric_SeqBAIJ_6(Mat A,Mat *B) 813 { 814 Mat C = *B; 815 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 816 IS isrow = b->row, isicol = b->icol; 817 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 818 int *ajtmpold, *ajtmp, nz, row; 819 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 820 register int *pj; 821 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 822 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 823 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 824 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 825 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 826 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 827 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 828 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 829 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 830 MatScalar *ba = b->a,*aa = a->a; 831 832 PetscFunctionBegin; 833 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 834 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 835 rtmp = (MatScalar *) PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 836 837 for ( i=0; i<n; i++ ) { 838 nz = bi[i+1] - bi[i]; 839 ajtmp = bj + bi[i]; 840 for ( j=0; j<nz; j++ ) { 841 x = rtmp+36*ajtmp[j]; 842 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 843 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 844 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 845 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 846 x[34] = x[35] = 0.0 ; 847 } 848 /* load in initial (unfactored row) */ 849 idx = r[i]; 850 nz = ai[idx+1] - ai[idx]; 851 ajtmpold = aj + ai[idx]; 852 v = aa + 36*ai[idx]; 853 for ( j=0; j<nz; j++ ) { 854 x = rtmp+36*ic[ajtmpold[j]]; 855 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 856 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 857 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 858 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 859 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 860 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 861 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 862 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 863 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 864 v += 36; 865 } 866 row = *ajtmp++; 867 while (row < i) { 868 pc = rtmp + 36*row; 869 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 870 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 871 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 872 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 873 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 874 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 875 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 876 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 877 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 878 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 879 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 880 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 881 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 882 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 883 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 884 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 885 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 886 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 887 pv = ba + 36*diag_offset[row]; 888 pj = bj + diag_offset[row] + 1; 889 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 890 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 891 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 892 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 893 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 894 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 895 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 896 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 897 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 898 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 899 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 900 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 901 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 902 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 903 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 904 905 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 906 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 907 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 908 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 909 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 910 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 911 912 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 913 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 914 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 915 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 916 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 917 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 918 919 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 920 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 921 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 922 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 923 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 924 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 925 926 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 927 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 928 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 929 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 930 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 931 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 932 933 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 934 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 935 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 936 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 937 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 938 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 939 940 nz = bi[row+1] - diag_offset[row] - 1; 941 pv += 36; 942 for (j=0; j<nz; j++) { 943 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 944 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 945 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 946 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 947 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 948 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 949 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 950 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 951 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 952 x = rtmp + 36*pj[j]; 953 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 954 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 955 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 956 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 957 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 958 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 959 960 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 961 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 962 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 963 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 964 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 965 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 966 967 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 968 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 969 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 970 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 971 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 972 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 973 974 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 975 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 976 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 977 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 978 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 979 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 980 981 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 982 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 983 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 984 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 985 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 986 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 987 988 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 989 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 990 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 991 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 992 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 993 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 994 995 pv += 36; 996 } 997 PLogFlops(432*nz+396); 998 } 999 row = *ajtmp++; 1000 } 1001 /* finished row so stick it into b->a */ 1002 pv = ba + 36*bi[i]; 1003 pj = bj + bi[i]; 1004 nz = bi[i+1] - bi[i]; 1005 for ( j=0; j<nz; j++ ) { 1006 x = rtmp+36*pj[j]; 1007 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1008 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1009 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1010 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1011 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1012 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1013 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1014 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1015 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1016 pv += 36; 1017 } 1018 /* invert diagonal block */ 1019 w = ba + 36*diag_offset[i]; 1020 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1021 } 1022 1023 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1024 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1025 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1026 C->factor = FACTOR_LU; 1027 C->assembled = PETSC_TRUE; 1028 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1029 PetscFunctionReturn(0); 1030 } 1031 /* 1032 Version for when blocks are 6 by 6 Using natural ordering 1033 */ 1034 #undef __FUNC__ 1035 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_6_NaturalOrdering" 1036 int MatLUFactorNumeric_SeqBAIJ_6_NaturalOrdering(Mat A,Mat *B) 1037 { 1038 Mat C = *B; 1039 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1040 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1041 int *ajtmpold, *ajtmp, nz, row; 1042 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 1043 register int *pj; 1044 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1045 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1046 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1047 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1048 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1049 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1050 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1051 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 1052 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 1053 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 1054 MatScalar *ba = b->a,*aa = a->a; 1055 1056 PetscFunctionBegin; 1057 rtmp = (MatScalar *) PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1058 for ( i=0; i<n; i++ ) { 1059 nz = bi[i+1] - bi[i]; 1060 ajtmp = bj + bi[i]; 1061 for ( j=0; j<nz; j++ ) { 1062 x = rtmp+36*ajtmp[j]; 1063 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1064 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1065 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 1066 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 1067 x[34] = x[35] = 0.0 ; 1068 } 1069 /* load in initial (unfactored row) */ 1070 nz = ai[i+1] - ai[i]; 1071 ajtmpold = aj + ai[i]; 1072 v = aa + 36*ai[i]; 1073 for ( j=0; j<nz; j++ ) { 1074 x = rtmp+36*ajtmpold[j]; 1075 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1076 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 1077 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 1078 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 1079 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 1080 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1081 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 1082 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 1083 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 1084 v += 36; 1085 } 1086 row = *ajtmp++; 1087 while (row < i) { 1088 pc = rtmp + 36*row; 1089 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1090 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 1091 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 1092 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 1093 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 1094 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1095 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 1096 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 1097 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 1098 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 1099 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 1100 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 1101 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 1102 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 1103 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 1104 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 1105 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 1106 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 1107 pv = ba + 36*diag_offset[row]; 1108 pj = bj + diag_offset[row] + 1; 1109 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1110 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1111 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1112 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1113 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1114 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1115 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1116 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1117 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1118 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 1119 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 1120 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 1121 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 1122 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 1123 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 1124 1125 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 1126 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 1127 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 1128 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 1129 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 1130 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 1131 1132 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 1133 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 1134 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 1135 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 1136 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 1137 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 1138 1139 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 1140 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 1141 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 1142 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 1143 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 1144 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 1145 1146 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 1147 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 1148 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 1149 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 1150 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 1151 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 1152 1153 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 1154 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 1155 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 1156 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 1157 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 1158 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 1159 1160 nz = bi[row+1] - diag_offset[row] - 1; 1161 pv += 36; 1162 for (j=0; j<nz; j++) { 1163 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1164 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1165 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1166 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1167 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1168 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1169 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1170 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1171 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1172 x = rtmp + 36*pj[j]; 1173 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 1174 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 1175 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 1176 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 1177 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 1178 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 1179 1180 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 1181 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 1182 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 1183 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 1184 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 1185 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 1186 1187 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 1188 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 1189 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 1190 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 1191 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 1192 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 1193 1194 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 1195 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 1196 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 1197 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 1198 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 1199 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 1200 1201 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 1202 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 1203 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 1204 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 1205 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 1206 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 1207 1208 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 1209 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 1210 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 1211 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 1212 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 1213 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 1214 1215 pv += 36; 1216 } 1217 PLogFlops(432*nz+396); 1218 } 1219 row = *ajtmp++; 1220 } 1221 /* finished row so stick it into b->a */ 1222 pv = ba + 36*bi[i]; 1223 pj = bj + bi[i]; 1224 nz = bi[i+1] - bi[i]; 1225 for ( j=0; j<nz; j++ ) { 1226 x = rtmp+36*pj[j]; 1227 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1228 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1229 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1230 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1231 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1232 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1233 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1234 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1235 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1236 pv += 36; 1237 } 1238 /* invert diagonal block */ 1239 w = ba + 36*diag_offset[i]; 1240 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1241 } 1242 1243 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1244 C->factor = FACTOR_LU; 1245 C->assembled = PETSC_TRUE; 1246 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1247 PetscFunctionReturn(0); 1248 } 1249 1250 /* ------------------------------------------------------------*/ 1251 /* 1252 Version for when blocks are 5 by 5 1253 */ 1254 #undef __FUNC__ 1255 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_5" 1256 int MatLUFactorNumeric_SeqBAIJ_5(Mat A,Mat *B) 1257 { 1258 Mat C = *B; 1259 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1260 IS isrow = b->row, isicol = b->icol; 1261 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1262 int *ajtmpold, *ajtmp, nz, row; 1263 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 1264 register int *pj; 1265 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1266 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1267 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1268 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 1269 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 1270 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1271 MatScalar *ba = b->a,*aa = a->a; 1272 1273 PetscFunctionBegin; 1274 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1275 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1276 rtmp = (MatScalar *) PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1277 1278 for ( i=0; i<n; i++ ) { 1279 nz = bi[i+1] - bi[i]; 1280 ajtmp = bj + bi[i]; 1281 for ( j=0; j<nz; j++ ) { 1282 x = rtmp+25*ajtmp[j]; 1283 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1284 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1285 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1286 } 1287 /* load in initial (unfactored row) */ 1288 idx = r[i]; 1289 nz = ai[idx+1] - ai[idx]; 1290 ajtmpold = aj + ai[idx]; 1291 v = aa + 25*ai[idx]; 1292 for ( j=0; j<nz; j++ ) { 1293 x = rtmp+25*ic[ajtmpold[j]]; 1294 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1295 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1296 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1297 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; 1298 x[18] = v[18]; x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; 1299 x[22] = v[22]; x[23] = v[23]; x[24] = v[24]; 1300 v += 25; 1301 } 1302 row = *ajtmp++; 1303 while (row < i) { 1304 pc = rtmp + 25*row; 1305 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1306 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1307 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1308 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; 1309 p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1310 p25 = pc[24]; 1311 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1312 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1313 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1314 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || 1315 p20 != 0.0 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || 1316 p24 != 0.0 || p25 != 0.0) { 1317 pv = ba + 25*diag_offset[row]; 1318 pj = bj + diag_offset[row] + 1; 1319 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1320 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1321 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1322 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1323 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; 1324 x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1325 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1326 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1327 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1328 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1329 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1330 1331 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1332 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1333 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1334 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1335 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1336 1337 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1338 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1339 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1340 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1341 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1342 1343 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1344 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1345 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1346 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1347 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1348 1349 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1350 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1351 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1352 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1353 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1354 1355 nz = bi[row+1] - diag_offset[row] - 1; 1356 pv += 25; 1357 for (j=0; j<nz; j++) { 1358 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1359 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1360 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1361 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; 1362 x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; 1363 x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1364 x = rtmp + 25*pj[j]; 1365 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1366 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1367 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1368 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1369 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1370 1371 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1372 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1373 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1374 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1375 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1376 1377 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1378 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1379 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1380 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1381 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1382 1383 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1384 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1385 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1386 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1387 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1388 1389 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1390 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1391 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1392 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1393 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1394 1395 pv += 25; 1396 } 1397 PLogFlops(250*nz+225); 1398 } 1399 row = *ajtmp++; 1400 } 1401 /* finished row so stick it into b->a */ 1402 pv = ba + 25*bi[i]; 1403 pj = bj + bi[i]; 1404 nz = bi[i+1] - bi[i]; 1405 for ( j=0; j<nz; j++ ) { 1406 x = rtmp+25*pj[j]; 1407 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1408 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1409 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1410 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; 1411 pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; 1412 pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; pv[24] = x[24]; 1413 pv += 25; 1414 } 1415 /* invert diagonal block */ 1416 w = ba + 25*diag_offset[i]; 1417 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1418 } 1419 1420 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1421 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1422 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1423 C->factor = FACTOR_LU; 1424 C->assembled = PETSC_TRUE; 1425 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1426 PetscFunctionReturn(0); 1427 } 1428 /* 1429 Version for when blocks are 5 by 5 Using natural ordering 1430 */ 1431 #undef __FUNC__ 1432 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_5_NaturalOrdering" 1433 int MatLUFactorNumeric_SeqBAIJ_5_NaturalOrdering(Mat A,Mat *B) 1434 { 1435 Mat C = *B; 1436 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1437 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1438 int *ajtmpold, *ajtmp, nz, row; 1439 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 1440 register int *pj; 1441 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1442 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1443 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1444 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1445 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1446 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1447 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1448 MatScalar *ba = b->a,*aa = a->a; 1449 1450 PetscFunctionBegin; 1451 rtmp = (MatScalar *) PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1452 for ( i=0; i<n; i++ ) { 1453 nz = bi[i+1] - bi[i]; 1454 ajtmp = bj + bi[i]; 1455 for ( j=0; j<nz; j++ ) { 1456 x = rtmp+25*ajtmp[j]; 1457 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1458 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1459 x[16] = x[17] = x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1460 } 1461 /* load in initial (unfactored row) */ 1462 nz = ai[i+1] - ai[i]; 1463 ajtmpold = aj + ai[i]; 1464 v = aa + 25*ai[i]; 1465 for ( j=0; j<nz; j++ ) { 1466 x = rtmp+25*ajtmpold[j]; 1467 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1468 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1469 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1470 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; 1471 x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1472 x[24] = v[24]; 1473 v += 25; 1474 } 1475 row = *ajtmp++; 1476 while (row < i) { 1477 pc = rtmp + 25*row; 1478 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1479 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1480 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1481 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; 1482 p19 = pc[18]; p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; 1483 p24 = pc[23]; p25 = pc[24]; 1484 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1485 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1486 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1487 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 1488 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || p25 != 0.0) { 1489 pv = ba + 25*diag_offset[row]; 1490 pj = bj + diag_offset[row] + 1; 1491 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1492 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1493 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1494 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; 1495 x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1496 x25 = pv[24]; 1497 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1498 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1499 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1500 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1501 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1502 1503 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1504 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1505 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1506 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1507 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1508 1509 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1510 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1511 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1512 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1513 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1514 1515 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1516 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1517 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1518 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1519 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1520 1521 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1522 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1523 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1524 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1525 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1526 1527 nz = bi[row+1] - diag_offset[row] - 1; 1528 pv += 25; 1529 for (j=0; j<nz; j++) { 1530 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1531 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1532 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1533 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1534 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; 1535 x24 = pv[23]; x25 = pv[24]; 1536 x = rtmp + 25*pj[j]; 1537 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1538 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1539 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1540 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1541 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1542 1543 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1544 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1545 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1546 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1547 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1548 1549 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1550 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1551 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1552 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1553 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1554 1555 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1556 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1557 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1558 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1559 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1560 1561 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1562 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1563 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1564 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1565 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1566 pv += 25; 1567 } 1568 PLogFlops(250*nz+225); 1569 } 1570 row = *ajtmp++; 1571 } 1572 /* finished row so stick it into b->a */ 1573 pv = ba + 25*bi[i]; 1574 pj = bj + bi[i]; 1575 nz = bi[i+1] - bi[i]; 1576 for ( j=0; j<nz; j++ ) { 1577 x = rtmp+25*pj[j]; 1578 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1579 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1580 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1581 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; pv[17] = x[17]; 1582 pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; 1583 pv[23] = x[23]; pv[24] = x[24]; 1584 pv += 25; 1585 } 1586 /* invert diagonal block */ 1587 w = ba + 25*diag_offset[i]; 1588 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1589 } 1590 1591 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1592 C->factor = FACTOR_LU; 1593 C->assembled = PETSC_TRUE; 1594 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1595 PetscFunctionReturn(0); 1596 } 1597 1598 /* ------------------------------------------------------------*/ 1599 /* 1600 Version for when blocks are 4 by 4 1601 */ 1602 #undef __FUNC__ 1603 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_4" 1604 int MatLUFactorNumeric_SeqBAIJ_4(Mat A,Mat *B) 1605 { 1606 Mat C = *B; 1607 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1608 IS isrow = b->row, isicol = b->icol; 1609 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1610 int *ajtmpold, *ajtmp, nz, row; 1611 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 1612 register int *pj; 1613 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1614 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1615 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1616 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1617 MatScalar m13,m14,m15,m16; 1618 MatScalar *ba = b->a,*aa = a->a; 1619 1620 PetscFunctionBegin; 1621 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1622 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1623 rtmp = (MatScalar *) PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1624 1625 for ( i=0; i<n; i++ ) { 1626 nz = bi[i+1] - bi[i]; 1627 ajtmp = bj + bi[i]; 1628 for ( j=0; j<nz; j++ ) { 1629 x = rtmp+16*ajtmp[j]; 1630 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1631 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1632 } 1633 /* load in initial (unfactored row) */ 1634 idx = r[i]; 1635 nz = ai[idx+1] - ai[idx]; 1636 ajtmpold = aj + ai[idx]; 1637 v = aa + 16*ai[idx]; 1638 for ( j=0; j<nz; j++ ) { 1639 x = rtmp+16*ic[ajtmpold[j]]; 1640 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1641 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1642 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1643 x[14] = v[14]; x[15] = v[15]; 1644 v += 16; 1645 } 1646 row = *ajtmp++; 1647 while (row < i) { 1648 pc = rtmp + 16*row; 1649 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1650 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1651 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1652 p15 = pc[14]; p16 = pc[15]; 1653 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1654 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1655 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1656 || p16 != 0.0) { 1657 pv = ba + 16*diag_offset[row]; 1658 pj = bj + diag_offset[row] + 1; 1659 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1660 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1661 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1662 x15 = pv[14]; x16 = pv[15]; 1663 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1664 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1665 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1666 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1667 1668 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1669 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1670 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1671 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1672 1673 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1674 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1675 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1676 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1677 1678 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1679 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1680 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1681 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1682 1683 nz = bi[row+1] - diag_offset[row] - 1; 1684 pv += 16; 1685 for (j=0; j<nz; j++) { 1686 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1687 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1688 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1689 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1690 x = rtmp + 16*pj[j]; 1691 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1692 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1693 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1694 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1695 1696 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1697 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1698 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1699 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1700 1701 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1702 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1703 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1704 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1705 1706 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1707 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1708 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1709 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1710 1711 pv += 16; 1712 } 1713 PLogFlops(128*nz+112); 1714 } 1715 row = *ajtmp++; 1716 } 1717 /* finished row so stick it into b->a */ 1718 pv = ba + 16*bi[i]; 1719 pj = bj + bi[i]; 1720 nz = bi[i+1] - bi[i]; 1721 for ( j=0; j<nz; j++ ) { 1722 x = rtmp+16*pj[j]; 1723 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1724 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1725 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1726 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1727 pv += 16; 1728 } 1729 /* invert diagonal block */ 1730 w = ba + 16*diag_offset[i]; 1731 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1732 } 1733 1734 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1735 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1736 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1737 C->factor = FACTOR_LU; 1738 C->assembled = PETSC_TRUE; 1739 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1740 PetscFunctionReturn(0); 1741 } 1742 /* 1743 Version for when blocks are 4 by 4 Using natural ordering 1744 */ 1745 #undef __FUNC__ 1746 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_4_NaturalOrdering" 1747 int MatLUFactorNumeric_SeqBAIJ_4_NaturalOrdering(Mat A,Mat *B) 1748 { 1749 Mat C = *B; 1750 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1751 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1752 int *ajtmpold, *ajtmp, nz, row; 1753 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 1754 register int *pj; 1755 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1756 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1757 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1758 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1759 MatScalar m13,m14,m15,m16; 1760 MatScalar *ba = b->a,*aa = a->a; 1761 1762 PetscFunctionBegin; 1763 rtmp = (MatScalar *) PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1764 1765 for ( i=0; i<n; i++ ) { 1766 nz = bi[i+1] - bi[i]; 1767 ajtmp = bj + bi[i]; 1768 for ( j=0; j<nz; j++ ) { 1769 x = rtmp+16*ajtmp[j]; 1770 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1771 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1772 } 1773 /* load in initial (unfactored row) */ 1774 nz = ai[i+1] - ai[i]; 1775 ajtmpold = aj + ai[i]; 1776 v = aa + 16*ai[i]; 1777 for ( j=0; j<nz; j++ ) { 1778 x = rtmp+16*ajtmpold[j]; 1779 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1780 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1781 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1782 x[14] = v[14]; x[15] = v[15]; 1783 v += 16; 1784 } 1785 row = *ajtmp++; 1786 while (row < i) { 1787 pc = rtmp + 16*row; 1788 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1789 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1790 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1791 p15 = pc[14]; p16 = pc[15]; 1792 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1793 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1794 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1795 || p16 != 0.0) { 1796 pv = ba + 16*diag_offset[row]; 1797 pj = bj + diag_offset[row] + 1; 1798 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1799 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1800 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1801 x15 = pv[14]; x16 = pv[15]; 1802 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1803 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1804 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1805 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1806 1807 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1808 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1809 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1810 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1811 1812 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1813 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1814 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1815 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1816 1817 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1818 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1819 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1820 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1821 1822 nz = bi[row+1] - diag_offset[row] - 1; 1823 pv += 16; 1824 for (j=0; j<nz; j++) { 1825 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1826 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1827 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1828 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1829 x = rtmp + 16*pj[j]; 1830 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1831 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1832 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1833 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1834 1835 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1836 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1837 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1838 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1839 1840 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1841 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1842 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1843 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1844 1845 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1846 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1847 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1848 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1849 1850 pv += 16; 1851 } 1852 PLogFlops(128*nz+112); 1853 } 1854 row = *ajtmp++; 1855 } 1856 /* finished row so stick it into b->a */ 1857 pv = ba + 16*bi[i]; 1858 pj = bj + bi[i]; 1859 nz = bi[i+1] - bi[i]; 1860 for ( j=0; j<nz; j++ ) { 1861 x = rtmp+16*pj[j]; 1862 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1863 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1864 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1865 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1866 pv += 16; 1867 } 1868 /* invert diagonal block */ 1869 w = ba + 16*diag_offset[i]; 1870 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1871 } 1872 1873 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1874 C->factor = FACTOR_LU; 1875 C->assembled = PETSC_TRUE; 1876 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1877 PetscFunctionReturn(0); 1878 } 1879 1880 1881 /* ------------------------------------------------------------*/ 1882 /* 1883 Version for when blocks are 3 by 3 1884 */ 1885 #undef __FUNC__ 1886 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_3" 1887 int MatLUFactorNumeric_SeqBAIJ_3(Mat A,Mat *B) 1888 { 1889 Mat C = *B; 1890 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1891 IS isrow = b->row, isicol = b->icol; 1892 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1893 int *ajtmpold, *ajtmp, nz, row, *ai=a->i,*aj=a->j; 1894 int *diag_offset = b->diag,idx; 1895 register int *pj; 1896 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1897 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1898 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 1899 MatScalar *ba = b->a,*aa = a->a; 1900 1901 PetscFunctionBegin; 1902 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1903 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1904 rtmp = (MatScalar *) PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1905 1906 for ( i=0; i<n; i++ ) { 1907 nz = bi[i+1] - bi[i]; 1908 ajtmp = bj + bi[i]; 1909 for ( j=0; j<nz; j++ ) { 1910 x = rtmp + 9*ajtmp[j]; 1911 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 1912 } 1913 /* load in initial (unfactored row) */ 1914 idx = r[i]; 1915 nz = ai[idx+1] - ai[idx]; 1916 ajtmpold = aj + ai[idx]; 1917 v = aa + 9*ai[idx]; 1918 for ( j=0; j<nz; j++ ) { 1919 x = rtmp + 9*ic[ajtmpold[j]]; 1920 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1921 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1922 v += 9; 1923 } 1924 row = *ajtmp++; 1925 while (row < i) { 1926 pc = rtmp + 9*row; 1927 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1928 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1929 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1930 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 1931 pv = ba + 9*diag_offset[row]; 1932 pj = bj + diag_offset[row] + 1; 1933 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1934 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1935 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 1936 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 1937 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 1938 1939 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 1940 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 1941 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 1942 1943 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 1944 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 1945 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 1946 nz = bi[row+1] - diag_offset[row] - 1; 1947 pv += 9; 1948 for (j=0; j<nz; j++) { 1949 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1950 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1951 x = rtmp + 9*pj[j]; 1952 x[0] -= m1*x1 + m4*x2 + m7*x3; 1953 x[1] -= m2*x1 + m5*x2 + m8*x3; 1954 x[2] -= m3*x1 + m6*x2 + m9*x3; 1955 1956 x[3] -= m1*x4 + m4*x5 + m7*x6; 1957 x[4] -= m2*x4 + m5*x5 + m8*x6; 1958 x[5] -= m3*x4 + m6*x5 + m9*x6; 1959 1960 x[6] -= m1*x7 + m4*x8 + m7*x9; 1961 x[7] -= m2*x7 + m5*x8 + m8*x9; 1962 x[8] -= m3*x7 + m6*x8 + m9*x9; 1963 pv += 9; 1964 } 1965 PLogFlops(54*nz+36); 1966 } 1967 row = *ajtmp++; 1968 } 1969 /* finished row so stick it into b->a */ 1970 pv = ba + 9*bi[i]; 1971 pj = bj + bi[i]; 1972 nz = bi[i+1] - bi[i]; 1973 for ( j=0; j<nz; j++ ) { 1974 x = rtmp + 9*pj[j]; 1975 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1976 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1977 pv += 9; 1978 } 1979 /* invert diagonal block */ 1980 w = ba + 9*diag_offset[i]; 1981 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 1982 } 1983 1984 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1985 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1986 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1987 C->factor = FACTOR_LU; 1988 C->assembled = PETSC_TRUE; 1989 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 1990 PetscFunctionReturn(0); 1991 } 1992 /* 1993 Version for when blocks are 3 by 3 Using natural ordering 1994 */ 1995 #undef __FUNC__ 1996 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_3_NaturalOrdering" 1997 int MatLUFactorNumeric_SeqBAIJ_3_NaturalOrdering(Mat A,Mat *B) 1998 { 1999 Mat C = *B; 2000 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 2001 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 2002 int *ajtmpold, *ajtmp, nz, row; 2003 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 2004 register int *pj; 2005 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2006 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 2007 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 2008 MatScalar *ba = b->a,*aa = a->a; 2009 2010 PetscFunctionBegin; 2011 rtmp = (MatScalar *) PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2012 2013 for ( i=0; i<n; i++ ) { 2014 nz = bi[i+1] - bi[i]; 2015 ajtmp = bj + bi[i]; 2016 for ( j=0; j<nz; j++ ) { 2017 x = rtmp+9*ajtmp[j]; 2018 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 2019 } 2020 /* load in initial (unfactored row) */ 2021 nz = ai[i+1] - ai[i]; 2022 ajtmpold = aj + ai[i]; 2023 v = aa + 9*ai[i]; 2024 for ( j=0; j<nz; j++ ) { 2025 x = rtmp+9*ajtmpold[j]; 2026 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2027 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 2028 v += 9; 2029 } 2030 row = *ajtmp++; 2031 while (row < i) { 2032 pc = rtmp + 9*row; 2033 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2034 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 2035 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 2036 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 2037 pv = ba + 9*diag_offset[row]; 2038 pj = bj + diag_offset[row] + 1; 2039 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2040 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2041 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 2042 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 2043 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 2044 2045 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 2046 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 2047 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 2048 2049 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 2050 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 2051 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 2052 2053 nz = bi[row+1] - diag_offset[row] - 1; 2054 pv += 9; 2055 for (j=0; j<nz; j++) { 2056 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2057 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2058 x = rtmp + 9*pj[j]; 2059 x[0] -= m1*x1 + m4*x2 + m7*x3; 2060 x[1] -= m2*x1 + m5*x2 + m8*x3; 2061 x[2] -= m3*x1 + m6*x2 + m9*x3; 2062 2063 x[3] -= m1*x4 + m4*x5 + m7*x6; 2064 x[4] -= m2*x4 + m5*x5 + m8*x6; 2065 x[5] -= m3*x4 + m6*x5 + m9*x6; 2066 2067 x[6] -= m1*x7 + m4*x8 + m7*x9; 2068 x[7] -= m2*x7 + m5*x8 + m8*x9; 2069 x[8] -= m3*x7 + m6*x8 + m9*x9; 2070 pv += 9; 2071 } 2072 PLogFlops(54*nz+36); 2073 } 2074 row = *ajtmp++; 2075 } 2076 /* finished row so stick it into b->a */ 2077 pv = ba + 9*bi[i]; 2078 pj = bj + bi[i]; 2079 nz = bi[i+1] - bi[i]; 2080 for ( j=0; j<nz; j++ ) { 2081 x = rtmp+9*pj[j]; 2082 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2083 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 2084 pv += 9; 2085 } 2086 /* invert diagonal block */ 2087 w = ba + 9*diag_offset[i]; 2088 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 2089 } 2090 2091 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2092 C->factor = FACTOR_LU; 2093 C->assembled = PETSC_TRUE; 2094 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 2095 PetscFunctionReturn(0); 2096 } 2097 2098 /* ------------------------------------------------------------*/ 2099 /* 2100 Version for when blocks are 2 by 2 2101 */ 2102 #undef __FUNC__ 2103 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_2" 2104 int MatLUFactorNumeric_SeqBAIJ_2(Mat A,Mat *B) 2105 { 2106 Mat C = *B; 2107 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 2108 IS isrow = b->row, isicol = b->icol; 2109 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 2110 int *ajtmpold, *ajtmp, nz, row; 2111 int *diag_offset=b->diag,idx,*ai=a->i,*aj=a->j; 2112 register int *pj; 2113 register MatScalar *pv,*v,*rtmp,m1,m2,m3,m4,*pc,*w,*x,x1,x2,x3,x4; 2114 MatScalar p1,p2,p3,p4; 2115 MatScalar *ba = b->a,*aa = a->a; 2116 2117 PetscFunctionBegin; 2118 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 2119 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 2120 rtmp = (MatScalar *) PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2121 2122 for ( i=0; i<n; i++ ) { 2123 nz = bi[i+1] - bi[i]; 2124 ajtmp = bj + bi[i]; 2125 for ( j=0; j<nz; j++ ) { 2126 x = rtmp+4*ajtmp[j]; x[0] = x[1] = x[2] = x[3] = 0.0; 2127 } 2128 /* load in initial (unfactored row) */ 2129 idx = r[i]; 2130 nz = ai[idx+1] - ai[idx]; 2131 ajtmpold = aj + ai[idx]; 2132 v = aa + 4*ai[idx]; 2133 for ( j=0; j<nz; j++ ) { 2134 x = rtmp+4*ic[ajtmpold[j]]; 2135 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2136 v += 4; 2137 } 2138 row = *ajtmp++; 2139 while (row < i) { 2140 pc = rtmp + 4*row; 2141 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2142 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2143 pv = ba + 4*diag_offset[row]; 2144 pj = bj + diag_offset[row] + 1; 2145 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2146 pc[0] = m1 = p1*x1 + p3*x2; 2147 pc[1] = m2 = p2*x1 + p4*x2; 2148 pc[2] = m3 = p1*x3 + p3*x4; 2149 pc[3] = m4 = p2*x3 + p4*x4; 2150 nz = bi[row+1] - diag_offset[row] - 1; 2151 pv += 4; 2152 for (j=0; j<nz; j++) { 2153 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2154 x = rtmp + 4*pj[j]; 2155 x[0] -= m1*x1 + m3*x2; 2156 x[1] -= m2*x1 + m4*x2; 2157 x[2] -= m1*x3 + m3*x4; 2158 x[3] -= m2*x3 + m4*x4; 2159 pv += 4; 2160 } 2161 PLogFlops(16*nz+12); 2162 } 2163 row = *ajtmp++; 2164 } 2165 /* finished row so stick it into b->a */ 2166 pv = ba + 4*bi[i]; 2167 pj = bj + bi[i]; 2168 nz = bi[i+1] - bi[i]; 2169 for ( j=0; j<nz; j++ ) { 2170 x = rtmp+4*pj[j]; 2171 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2172 pv += 4; 2173 } 2174 /* invert diagonal block */ 2175 w = ba + 4*diag_offset[i]; 2176 ierr = Kernel_A_gets_inverse_A_2(w);CHKERRQ(ierr); 2177 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2178 } 2179 2180 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2181 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 2182 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 2183 C->factor = FACTOR_LU; 2184 C->assembled = PETSC_TRUE; 2185 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2186 PetscFunctionReturn(0); 2187 } 2188 /* 2189 Version for when blocks are 2 by 2 Using natural ordering 2190 */ 2191 #undef __FUNC__ 2192 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering" 2193 int MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering(Mat A,Mat *B) 2194 { 2195 Mat C = *B; 2196 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 2197 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 2198 int *ajtmpold, *ajtmp, nz, row; 2199 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 2200 register int *pj; 2201 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2202 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,x1,x2,x3,x4; 2203 MatScalar *ba = b->a,*aa = a->a; 2204 2205 PetscFunctionBegin; 2206 rtmp = (MatScalar *) PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2207 2208 for ( i=0; i<n; i++ ) { 2209 nz = bi[i+1] - bi[i]; 2210 ajtmp = bj + bi[i]; 2211 for ( j=0; j<nz; j++ ) { 2212 x = rtmp+4*ajtmp[j]; 2213 x[0] = x[1] = x[2] = x[3] = 0.0; 2214 } 2215 /* load in initial (unfactored row) */ 2216 nz = ai[i+1] - ai[i]; 2217 ajtmpold = aj + ai[i]; 2218 v = aa + 4*ai[i]; 2219 for ( j=0; j<nz; j++ ) { 2220 x = rtmp+4*ajtmpold[j]; 2221 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2222 v += 4; 2223 } 2224 row = *ajtmp++; 2225 while (row < i) { 2226 pc = rtmp + 4*row; 2227 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2228 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2229 pv = ba + 4*diag_offset[row]; 2230 pj = bj + diag_offset[row] + 1; 2231 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2232 pc[0] = m1 = p1*x1 + p3*x2; 2233 pc[1] = m2 = p2*x1 + p4*x2; 2234 pc[2] = m3 = p1*x3 + p3*x4; 2235 pc[3] = m4 = p2*x3 + p4*x4; 2236 nz = bi[row+1] - diag_offset[row] - 1; 2237 pv += 4; 2238 for (j=0; j<nz; j++) { 2239 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2240 x = rtmp + 4*pj[j]; 2241 x[0] -= m1*x1 + m3*x2; 2242 x[1] -= m2*x1 + m4*x2; 2243 x[2] -= m1*x3 + m3*x4; 2244 x[3] -= m2*x3 + m4*x4; 2245 pv += 4; 2246 } 2247 PLogFlops(16*nz+12); 2248 } 2249 row = *ajtmp++; 2250 } 2251 /* finished row so stick it into b->a */ 2252 pv = ba + 4*bi[i]; 2253 pj = bj + bi[i]; 2254 nz = bi[i+1] - bi[i]; 2255 for ( j=0; j<nz; j++ ) { 2256 x = rtmp+4*pj[j]; 2257 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2258 pv += 4; 2259 } 2260 /* invert diagonal block */ 2261 w = ba + 4*diag_offset[i]; 2262 ierr = Kernel_A_gets_inverse_A_2(w);CHKERRQ(ierr); 2263 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2264 } 2265 2266 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2267 C->factor = FACTOR_LU; 2268 C->assembled = PETSC_TRUE; 2269 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2270 PetscFunctionReturn(0); 2271 } 2272 2273 /* ----------------------------------------------------------- */ 2274 /* 2275 Version for when blocks are 1 by 1. 2276 */ 2277 #undef __FUNC__ 2278 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_1" 2279 int MatLUFactorNumeric_SeqBAIJ_1(Mat A,Mat *B) 2280 { 2281 Mat C = *B; 2282 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data, *b = (Mat_SeqBAIJ *)C->data; 2283 IS isrow = b->row, isicol = b->icol; 2284 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 2285 int *ajtmpold, *ajtmp, nz, row,*ai = a->i,*aj = a->j; 2286 int *diag_offset = b->diag,diag; 2287 register int *pj; 2288 register MatScalar *pv,*v,*rtmp,multiplier,*pc; 2289 MatScalar *ba = b->a,*aa = a->a; 2290 2291 PetscFunctionBegin; 2292 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 2293 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 2294 rtmp = (MatScalar *) PetscMalloc((n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2295 2296 for ( i=0; i<n; i++ ) { 2297 nz = bi[i+1] - bi[i]; 2298 ajtmp = bj + bi[i]; 2299 for ( j=0; j<nz; j++ ) rtmp[ajtmp[j]] = 0.0; 2300 2301 /* load in initial (unfactored row) */ 2302 nz = ai[r[i]+1] - ai[r[i]]; 2303 ajtmpold = aj + ai[r[i]]; 2304 v = aa + ai[r[i]]; 2305 for ( j=0; j<nz; j++ ) rtmp[ic[ajtmpold[j]]] = v[j]; 2306 2307 row = *ajtmp++; 2308 while (row < i) { 2309 pc = rtmp + row; 2310 if (*pc != 0.0) { 2311 pv = ba + diag_offset[row]; 2312 pj = bj + diag_offset[row] + 1; 2313 multiplier = *pc * *pv++; 2314 *pc = multiplier; 2315 nz = bi[row+1] - diag_offset[row] - 1; 2316 for (j=0; j<nz; j++) rtmp[pj[j]] -= multiplier * pv[j]; 2317 PLogFlops(1+2*nz); 2318 } 2319 row = *ajtmp++; 2320 } 2321 /* finished row so stick it into b->a */ 2322 pv = ba + bi[i]; 2323 pj = bj + bi[i]; 2324 nz = bi[i+1] - bi[i]; 2325 for ( j=0; j<nz; j++ ) {pv[j] = rtmp[pj[j]];} 2326 diag = diag_offset[i] - bi[i]; 2327 /* check pivot entry for current row */ 2328 if (pv[diag] == 0.0) { 2329 SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,0,"Zero pivot"); 2330 } 2331 pv[diag] = 1.0/pv[diag]; 2332 } 2333 2334 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2335 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 2336 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 2337 C->factor = FACTOR_LU; 2338 C->assembled = PETSC_TRUE; 2339 PLogFlops(b->n); 2340 PetscFunctionReturn(0); 2341 } 2342 2343 /* ----------------------------------------------------------- */ 2344 #undef __FUNC__ 2345 #define __FUNC__ "MatLUFactor_SeqBAIJ" 2346 int MatLUFactor_SeqBAIJ(Mat A,IS row,IS col,double f) 2347 { 2348 Mat_SeqBAIJ *mat = (Mat_SeqBAIJ *) A->data; 2349 int ierr; 2350 Mat C; 2351 PetscOps *Abops; 2352 MatOps Aops; 2353 2354 PetscFunctionBegin; 2355 ierr = MatLUFactorSymbolic(A,row,col,f,&C);CHKERRQ(ierr); 2356 ierr = MatLUFactorNumeric(A,&C);CHKERRQ(ierr); 2357 2358 /* free all the data structures from mat */ 2359 ierr = PetscFree(mat->a);CHKERRQ(ierr); 2360 if (!mat->singlemalloc) { 2361 ierr = PetscFree(mat->i);CHKERRQ(ierr); 2362 ierr = PetscFree(mat->j);CHKERRQ(ierr); 2363 } 2364 if (mat->diag) {ierr = PetscFree(mat->diag);CHKERRQ(ierr);} 2365 if (mat->ilen) {ierr = PetscFree(mat->ilen);CHKERRQ(ierr);} 2366 if (mat->imax) {ierr = PetscFree(mat->imax);CHKERRQ(ierr);} 2367 if (mat->solve_work) {ierr = PetscFree(mat->solve_work);CHKERRQ(ierr);} 2368 if (mat->mult_work) {ierr = PetscFree(mat->mult_work);CHKERRQ(ierr);} 2369 if (mat->icol) {ierr = ISDestroy(mat->icol);CHKERRQ(ierr);} 2370 ierr = PetscFree(mat);CHKERRQ(ierr); 2371 2372 ierr = MapDestroy(A->rmap);CHKERRQ(ierr); 2373 ierr = MapDestroy(A->cmap);CHKERRQ(ierr); 2374 2375 /* 2376 This is horrible, horrible code. We need to keep the 2377 A pointers for the bops and ops but copy everything 2378 else from C. 2379 */ 2380 Abops = A->bops; 2381 Aops = A->ops; 2382 ierr = PetscMemcpy(A,C,sizeof(struct _p_Mat));CHKERRQ(ierr); 2383 mat = (Mat_SeqBAIJ *) A->data; 2384 PLogObjectParent(A,mat->icol); 2385 2386 A->bops = Abops; 2387 A->ops = Aops; 2388 A->qlist = 0; 2389 2390 PetscHeaderDestroy(C); 2391 PetscFunctionReturn(0); 2392 } 2393 2394 2395 2396