1 #ifdef PETSC_RCS_HEADER 2 static char vcid[] = "$Id: baijfact.c,v 1.68 1999/02/03 17:22:53 balay Exp bsmith $"; 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 PetscMemcpy(ajtmp,ajnew,ainew[i]*sizeof(int)); 101 PetscFree(ajnew); 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 PetscFree(fill); 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 PetscFree(b->imax); 138 b->singlemalloc = 0; 139 /* the next line frees the default space generated by the Create() */ 140 PetscFree(b->a); PetscFree(b->ilen); 141 b->a = (MatScalar *) PetscMalloc((ainew[n]+1)*sizeof(MatScalar)*bs2);CHKPTRQ(b->a); 142 b->j = ajnew; 143 b->i = ainew; 144 b->diag = idnew; 145 b->ilen = 0; 146 b->imax = 0; 147 b->row = isrow; 148 b->col = iscol; 149 b->icol = isicol; 150 b->solve_work = (Scalar *) PetscMalloc( (bs*n+bs)*sizeof(Scalar));CHKPTRQ(b->solve_work); 151 /* In b structure: Free imax, ilen, old a, old j. 152 Allocate idnew, solve_work, new a, new j */ 153 PLogObjectMemory(*B,(ainew[n]-n)*(sizeof(int)+sizeof(MatScalar))); 154 b->maxnz = b->nz = ainew[n]; 155 156 (*B)->factor = FACTOR_LU; 157 (*B)->info.factor_mallocs = realloc; 158 (*B)->info.fill_ratio_given = f; 159 if (ai[n] != 0) { 160 (*B)->info.fill_ratio_needed = ((double)ainew[n])/((double)ai[n]); 161 } else { 162 (*B)->info.fill_ratio_needed = 0.0; 163 } 164 165 166 PetscFunctionReturn(0); 167 } 168 169 /* ----------------------------------------------------------- */ 170 #undef __FUNC__ 171 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_N" 172 int MatLUFactorNumeric_SeqBAIJ_N(Mat A,Mat *B) 173 { 174 Mat C = *B; 175 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 176 IS isrow = b->row, isicol = b->icol; 177 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 178 int *ajtmpold, *ajtmp, nz, row, bslog,*ai=a->i,*aj=a->j,k,flg; 179 int *diag_offset=b->diag,diag,bs=a->bs,bs2 = a->bs2,*v_pivots; 180 register int *pj; 181 register MatScalar *pv,*v,*rtmp,*multiplier,*v_work,*pc,*w; 182 MatScalar *ba = b->a,*aa = a->a; 183 184 PetscFunctionBegin; 185 ierr = ISGetIndices(isrow,&r); CHKERRQ(ierr); 186 ierr = ISGetIndices(isicol,&ic); CHKERRQ(ierr); 187 rtmp = (MatScalar *) PetscMalloc(bs2*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 188 PetscMemzero(rtmp,bs2*(n+1)*sizeof(MatScalar)); 189 /* generate work space needed by dense LU factorization */ 190 v_work = (MatScalar *) PetscMalloc(bs*sizeof(int) + (bs+bs2)*sizeof(MatScalar));CHKPTRQ(v_work); 191 multiplier = v_work + bs; 192 v_pivots = (int *) (multiplier + bs2); 193 194 /* flops in while loop */ 195 bslog = 2*bs*bs2; 196 197 for ( i=0; i<n; i++ ) { 198 nz = bi[i+1] - bi[i]; 199 ajtmp = bj + bi[i]; 200 for ( j=0; j<nz; j++ ) { 201 PetscMemzero(rtmp+bs2*ajtmp[j],bs2*sizeof(MatScalar)); 202 } 203 /* load in initial (unfactored row) */ 204 nz = ai[r[i]+1] - ai[r[i]]; 205 ajtmpold = aj + ai[r[i]]; 206 v = aa + bs2*ai[r[i]]; 207 for ( j=0; j<nz; j++ ) { 208 PetscMemcpy(rtmp+bs2*ic[ajtmpold[j]],v+bs2*j,bs2*sizeof(MatScalar)); 209 } 210 row = *ajtmp++; 211 while (row < i) { 212 pc = rtmp + bs2*row; 213 /* if (*pc) { */ 214 for ( flg=0,k=0; k<bs2; k++ ) { if (pc[k]!=0.0) { flg =1; break; }} 215 if (flg) { 216 pv = ba + bs2*diag_offset[row]; 217 pj = bj + diag_offset[row] + 1; 218 Kernel_A_gets_A_times_B(bs,pc,pv,multiplier); 219 nz = bi[row+1] - diag_offset[row] - 1; 220 pv += bs2; 221 for (j=0; j<nz; j++) { 222 Kernel_A_gets_A_minus_B_times_C(bs,rtmp+bs2*pj[j],pc,pv+bs2*j); 223 } 224 PLogFlops(bslog*(nz+1)-bs); 225 } 226 row = *ajtmp++; 227 } 228 /* finished row so stick it into b->a */ 229 pv = ba + bs2*bi[i]; 230 pj = bj + bi[i]; 231 nz = bi[i+1] - bi[i]; 232 for ( j=0; j<nz; j++ ) { 233 PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar)); 234 } 235 diag = diag_offset[i] - bi[i]; 236 /* invert diagonal block */ 237 w = pv + bs2*diag; 238 Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work); 239 } 240 241 PetscFree(rtmp); PetscFree(v_work); 242 ierr = ISRestoreIndices(isicol,&ic); CHKERRQ(ierr); 243 ierr = ISRestoreIndices(isrow,&r); CHKERRQ(ierr); 244 C->factor = FACTOR_LU; 245 C->assembled = PETSC_TRUE; 246 PLogFlops(1.3333*bs*bs2*b->mbs); /* from inverting diagonal blocks */ 247 PetscFunctionReturn(0); 248 } 249 /* ------------------------------------------------------------*/ 250 /* 251 Version for when blocks are 7 by 7 252 */ 253 #undef __FUNC__ 254 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_7" 255 int MatLUFactorNumeric_SeqBAIJ_7(Mat A,Mat *B) 256 { 257 Mat C = *B; 258 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 259 IS isrow = b->row, isicol = b->icol; 260 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 261 int *ajtmpold, *ajtmp, nz, row; 262 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 263 register int *pj; 264 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 265 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 266 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 267 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 268 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 269 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 270 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 271 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 272 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 273 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 274 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 275 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 276 MatScalar *ba = b->a,*aa = a->a; 277 278 PetscFunctionBegin; 279 ierr = ISGetIndices(isrow,&r); CHKERRQ(ierr); 280 ierr = ISGetIndices(isicol,&ic); CHKERRQ(ierr); 281 rtmp = (MatScalar *) PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 282 283 for ( i=0; i<n; i++ ) { 284 nz = bi[i+1] - bi[i]; 285 ajtmp = bj + bi[i]; 286 for ( j=0; j<nz; j++ ) { 287 x = rtmp+49*ajtmp[j]; 288 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 289 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 290 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 291 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 292 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 293 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = x[49] = 0.0 ; 294 } 295 /* load in initial (unfactored row) */ 296 idx = r[i]; 297 nz = ai[idx+1] - ai[idx]; 298 ajtmpold = aj + ai[idx]; 299 v = aa + 49*ai[idx]; 300 for ( j=0; j<nz; j++ ) { 301 x = rtmp+49*ic[ajtmpold[j]]; 302 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 303 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 304 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 305 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 306 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 307 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 308 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 309 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 310 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 311 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 312 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 313 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 314 x[48] = v[48]; 315 v += 49; 316 } 317 row = *ajtmp++; 318 while (row < i) { 319 pc = rtmp + 49*row; 320 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 321 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 322 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 323 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 324 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 325 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 326 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 327 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 328 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 329 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 330 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 331 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 332 p49 = pc[48]; 333 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 334 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 335 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 336 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 337 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 338 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 339 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 340 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 341 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 342 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 343 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 344 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 345 p49 != 0.0) { 346 pv = ba + 49*diag_offset[row]; 347 pj = bj + diag_offset[row] + 1; 348 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 349 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 350 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 351 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 352 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 353 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 354 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 355 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 356 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 357 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 358 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 359 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 360 x49 = pv[48]; 361 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 362 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 363 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 364 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 365 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 366 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 367 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 368 369 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 370 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 371 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 372 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 373 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 374 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 375 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 376 377 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 378 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 379 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 380 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 381 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 382 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 383 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 384 385 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 386 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 387 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 388 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 389 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 390 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 391 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 392 393 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 394 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 395 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 396 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 397 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 398 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 399 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 400 401 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 402 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 403 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 404 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 405 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 406 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 407 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 408 409 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 410 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 411 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 412 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 413 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 414 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 415 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 416 417 nz = bi[row+1] - diag_offset[row] - 1; 418 pv += 49; 419 for (j=0; j<nz; j++) { 420 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 421 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 422 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 423 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 424 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 425 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 426 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 427 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 428 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 429 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 430 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 431 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 432 x49 = pv[48]; 433 x = rtmp + 49*pj[j]; 434 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 435 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 436 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 437 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 438 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 439 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 440 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 441 442 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 443 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 444 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 445 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 446 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 447 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 448 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 449 450 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 451 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 452 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 453 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 454 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 455 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 456 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 457 458 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 459 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 460 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 461 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 462 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 463 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 464 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 465 466 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 467 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 468 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 469 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 470 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 471 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 472 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 473 474 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 475 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 476 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 477 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 478 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 479 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 480 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 481 482 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 483 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 484 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 485 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 486 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 487 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 488 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 489 pv += 49; 490 } 491 PLogFlops(686*nz+637); 492 } 493 row = *ajtmp++; 494 } 495 /* finished row so stick it into b->a */ 496 pv = ba + 49*bi[i]; 497 pj = bj + bi[i]; 498 nz = bi[i+1] - bi[i]; 499 for ( j=0; j<nz; j++ ) { 500 x = rtmp+49*pj[j]; 501 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 502 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 503 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 504 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 505 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 506 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 507 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 508 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 509 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 510 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 511 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 512 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 513 pv[48] = v[48]; 514 pv += 49; 515 } 516 /* invert diagonal block */ 517 w = ba + 49*diag_offset[i]; 518 ierr = Kernel_A_gets_inverse_A_7(w); CHKERRQ(ierr); 519 } 520 521 PetscFree(rtmp); 522 ierr = ISRestoreIndices(isicol,&ic); CHKERRQ(ierr); 523 ierr = ISRestoreIndices(isrow,&r); CHKERRQ(ierr); 524 C->factor = FACTOR_LU; 525 C->assembled = PETSC_TRUE; 526 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 527 PetscFunctionReturn(0); 528 } 529 /* 530 Version for when blocks are 7 by 7 Using natural ordering 531 */ 532 #undef __FUNC__ 533 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_7_NaturalOrdering" 534 int MatLUFactorNumeric_SeqBAIJ_7_NaturalOrdering(Mat A,Mat *B) 535 { 536 Mat C = *B; 537 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 538 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 539 int *ajtmpold, *ajtmp, nz, row; 540 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 541 register int *pj; 542 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 543 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 544 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 545 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 546 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 547 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 548 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 549 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 550 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 551 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 552 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 553 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 554 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 555 MatScalar *ba = b->a,*aa = a->a; 556 557 PetscFunctionBegin; 558 rtmp = (MatScalar *) PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 559 for ( i=0; i<n; i++ ) { 560 nz = bi[i+1] - bi[i]; 561 ajtmp = bj + bi[i]; 562 for ( j=0; j<nz; j++ ) { 563 x = rtmp+49*ajtmp[j]; 564 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 565 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 566 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 567 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 568 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 569 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = x[49] = 0.0 ; 570 } 571 /* load in initial (unfactored row) */ 572 nz = ai[i+1] - ai[i]; 573 ajtmpold = aj + ai[i]; 574 v = aa + 49*ai[i]; 575 for ( j=0; j<nz; j++ ) { 576 x = rtmp+49*ajtmpold[j]; 577 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 578 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 579 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 580 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 581 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 582 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 583 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 584 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 585 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 586 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 587 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 588 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 589 x[48] = v[48]; 590 v += 49; 591 } 592 row = *ajtmp++; 593 while (row < i) { 594 pc = rtmp + 49*row; 595 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 596 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 597 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 598 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 599 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 600 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 601 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 602 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 603 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 604 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 605 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 606 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 607 p49 = pc[48]; 608 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 609 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 610 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 611 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 612 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 613 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 614 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 615 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 616 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 617 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 618 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 619 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 620 p49 != 0.0) { 621 pv = ba + 36*diag_offset[row]; 622 pj = bj + diag_offset[row] + 1; 623 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 624 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 625 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 626 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 627 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 628 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 629 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 630 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 631 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 632 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 633 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 634 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 635 x49 = pv[48]; 636 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 637 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 638 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 639 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 640 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 641 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 642 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 643 644 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 645 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 646 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 647 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 648 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 649 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 650 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 651 652 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 653 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 654 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 655 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 656 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 657 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 658 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 659 660 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 661 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 662 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 663 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 664 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 665 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 666 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 667 668 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 669 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 670 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 671 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 672 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 673 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 674 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 675 676 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 677 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 678 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 679 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 680 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 681 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 682 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 683 684 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 685 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 686 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 687 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 688 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 689 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 690 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 691 692 nz = bi[row+1] - diag_offset[row] - 1; 693 pv += 49; 694 for (j=0; j<nz; j++) { 695 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 696 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 697 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 698 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 699 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 700 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 701 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 702 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 703 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 704 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 705 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 706 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 707 x49 = pv[48]; 708 x = rtmp + 49*pj[j]; 709 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 710 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 711 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 712 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 713 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 714 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 715 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 716 717 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 718 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 719 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 720 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 721 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 722 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 723 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 724 725 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 726 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 727 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 728 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 729 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 730 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 731 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 732 733 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 734 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 735 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 736 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 737 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 738 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 739 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 740 741 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 742 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 743 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 744 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 745 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 746 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 747 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 748 749 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 750 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 751 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 752 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 753 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 754 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 755 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 756 757 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 758 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 759 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 760 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 761 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 762 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 763 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 764 pv += 49; 765 } 766 PLogFlops(686*nz+637); 767 } 768 row = *ajtmp++; 769 } 770 /* finished row so stick it into b->a */ 771 pv = ba + 49*bi[i]; 772 pj = bj + bi[i]; 773 nz = bi[i+1] - bi[i]; 774 for ( j=0; j<nz; j++ ) { 775 x = rtmp+49*pj[j]; 776 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 777 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 778 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 779 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 780 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 781 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 782 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 783 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 784 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 785 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 786 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 787 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 788 pv[48] = v[48]; 789 pv += 49; 790 } 791 /* invert diagonal block */ 792 w = ba + 49*diag_offset[i]; 793 ierr = Kernel_A_gets_inverse_A_7(w); CHKERRQ(ierr); 794 } 795 796 PetscFree(rtmp); 797 C->factor = FACTOR_LU; 798 C->assembled = PETSC_TRUE; 799 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 800 PetscFunctionReturn(0); 801 } 802 803 /* ------------------------------------------------------------*/ 804 /* 805 Version for when blocks are 6 by 6 806 */ 807 #undef __FUNC__ 808 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_6" 809 int MatLUFactorNumeric_SeqBAIJ_6(Mat A,Mat *B) 810 { 811 Mat C = *B; 812 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 813 IS isrow = b->row, isicol = b->icol; 814 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 815 int *ajtmpold, *ajtmp, nz, row; 816 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 817 register int *pj; 818 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 819 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 820 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 821 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 822 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 823 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 824 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 825 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 826 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 827 MatScalar *ba = b->a,*aa = a->a; 828 829 PetscFunctionBegin; 830 ierr = ISGetIndices(isrow,&r); CHKERRQ(ierr); 831 ierr = ISGetIndices(isicol,&ic); CHKERRQ(ierr); 832 rtmp = (MatScalar *) PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 833 834 for ( i=0; i<n; i++ ) { 835 nz = bi[i+1] - bi[i]; 836 ajtmp = bj + bi[i]; 837 for ( j=0; j<nz; j++ ) { 838 x = rtmp+36*ajtmp[j]; 839 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 840 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 841 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 842 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 843 x[34] = x[35] = 0.0 ; 844 } 845 /* load in initial (unfactored row) */ 846 idx = r[i]; 847 nz = ai[idx+1] - ai[idx]; 848 ajtmpold = aj + ai[idx]; 849 v = aa + 36*ai[idx]; 850 for ( j=0; j<nz; j++ ) { 851 x = rtmp+36*ic[ajtmpold[j]]; 852 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 853 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 854 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 855 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 856 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 857 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 858 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 859 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 860 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 861 v += 36; 862 } 863 row = *ajtmp++; 864 while (row < i) { 865 pc = rtmp + 36*row; 866 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 867 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 868 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 869 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 870 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 871 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 872 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 873 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 874 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 875 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 876 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 877 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 878 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 879 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 880 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 881 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 882 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 883 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 884 pv = ba + 36*diag_offset[row]; 885 pj = bj + diag_offset[row] + 1; 886 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 887 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 888 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 889 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 890 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 891 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 892 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 893 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 894 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 895 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 896 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 897 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 898 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 899 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 900 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 901 902 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 903 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 904 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 905 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 906 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 907 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 908 909 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 910 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 911 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 912 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 913 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 914 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 915 916 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 917 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 918 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 919 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 920 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 921 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 922 923 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 924 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 925 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 926 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 927 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 928 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 929 930 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 931 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 932 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 933 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 934 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 935 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 936 937 nz = bi[row+1] - diag_offset[row] - 1; 938 pv += 36; 939 for (j=0; j<nz; j++) { 940 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 941 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 942 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 943 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 944 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 945 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 946 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 947 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 948 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 949 x = rtmp + 36*pj[j]; 950 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 951 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 952 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 953 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 954 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 955 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 956 957 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 958 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 959 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 960 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 961 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 962 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 963 964 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 965 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 966 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 967 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 968 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 969 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 970 971 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 972 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 973 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 974 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 975 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 976 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 977 978 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 979 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 980 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 981 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 982 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 983 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 984 985 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 986 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 987 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 988 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 989 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 990 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 991 992 pv += 36; 993 } 994 PLogFlops(432*nz+396); 995 } 996 row = *ajtmp++; 997 } 998 /* finished row so stick it into b->a */ 999 pv = ba + 36*bi[i]; 1000 pj = bj + bi[i]; 1001 nz = bi[i+1] - bi[i]; 1002 for ( j=0; j<nz; j++ ) { 1003 x = rtmp+36*pj[j]; 1004 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1005 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1006 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1007 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1008 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1009 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1010 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1011 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1012 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1013 pv += 36; 1014 } 1015 /* invert diagonal block */ 1016 w = ba + 36*diag_offset[i]; 1017 ierr = Kernel_A_gets_inverse_A_6(w); CHKERRQ(ierr); 1018 } 1019 1020 PetscFree(rtmp); 1021 ierr = ISRestoreIndices(isicol,&ic); CHKERRQ(ierr); 1022 ierr = ISRestoreIndices(isrow,&r); CHKERRQ(ierr); 1023 C->factor = FACTOR_LU; 1024 C->assembled = PETSC_TRUE; 1025 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1026 PetscFunctionReturn(0); 1027 } 1028 /* 1029 Version for when blocks are 6 by 6 Using natural ordering 1030 */ 1031 #undef __FUNC__ 1032 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_6_NaturalOrdering" 1033 int MatLUFactorNumeric_SeqBAIJ_6_NaturalOrdering(Mat A,Mat *B) 1034 { 1035 Mat C = *B; 1036 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1037 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1038 int *ajtmpold, *ajtmp, nz, row; 1039 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 1040 register int *pj; 1041 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1042 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1043 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1044 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1045 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1046 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1047 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1048 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 1049 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 1050 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 1051 MatScalar *ba = b->a,*aa = a->a; 1052 1053 PetscFunctionBegin; 1054 rtmp = (MatScalar *) PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1055 for ( i=0; i<n; i++ ) { 1056 nz = bi[i+1] - bi[i]; 1057 ajtmp = bj + bi[i]; 1058 for ( j=0; j<nz; j++ ) { 1059 x = rtmp+36*ajtmp[j]; 1060 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1061 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1062 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 1063 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 1064 x[34] = x[35] = 0.0 ; 1065 } 1066 /* load in initial (unfactored row) */ 1067 nz = ai[i+1] - ai[i]; 1068 ajtmpold = aj + ai[i]; 1069 v = aa + 36*ai[i]; 1070 for ( j=0; j<nz; j++ ) { 1071 x = rtmp+36*ajtmpold[j]; 1072 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1073 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 1074 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 1075 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 1076 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 1077 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1078 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 1079 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 1080 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 1081 v += 36; 1082 } 1083 row = *ajtmp++; 1084 while (row < i) { 1085 pc = rtmp + 36*row; 1086 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1087 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 1088 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 1089 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 1090 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 1091 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1092 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 1093 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 1094 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 1095 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 1096 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 1097 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 1098 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 1099 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 1100 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 1101 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 1102 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 1103 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 1104 pv = ba + 36*diag_offset[row]; 1105 pj = bj + diag_offset[row] + 1; 1106 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1107 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1108 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1109 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1110 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1111 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1112 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1113 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1114 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1115 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 1116 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 1117 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 1118 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 1119 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 1120 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 1121 1122 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 1123 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 1124 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 1125 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 1126 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 1127 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 1128 1129 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 1130 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 1131 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 1132 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 1133 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 1134 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 1135 1136 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 1137 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 1138 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 1139 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 1140 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 1141 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 1142 1143 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 1144 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 1145 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 1146 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 1147 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 1148 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 1149 1150 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 1151 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 1152 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 1153 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 1154 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 1155 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 1156 1157 nz = bi[row+1] - diag_offset[row] - 1; 1158 pv += 36; 1159 for (j=0; j<nz; j++) { 1160 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1161 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1162 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1163 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1164 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1165 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1166 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1167 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1168 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1169 x = rtmp + 36*pj[j]; 1170 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 1171 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 1172 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 1173 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 1174 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 1175 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 1176 1177 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 1178 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 1179 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 1180 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 1181 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 1182 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 1183 1184 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 1185 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 1186 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 1187 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 1188 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 1189 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 1190 1191 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 1192 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 1193 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 1194 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 1195 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 1196 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 1197 1198 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 1199 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 1200 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 1201 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 1202 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 1203 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 1204 1205 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 1206 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 1207 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 1208 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 1209 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 1210 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 1211 1212 pv += 36; 1213 } 1214 PLogFlops(432*nz+396); 1215 } 1216 row = *ajtmp++; 1217 } 1218 /* finished row so stick it into b->a */ 1219 pv = ba + 36*bi[i]; 1220 pj = bj + bi[i]; 1221 nz = bi[i+1] - bi[i]; 1222 for ( j=0; j<nz; j++ ) { 1223 x = rtmp+36*pj[j]; 1224 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1225 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1226 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1227 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1228 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1229 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1230 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1231 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1232 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1233 pv += 36; 1234 } 1235 /* invert diagonal block */ 1236 w = ba + 36*diag_offset[i]; 1237 ierr = Kernel_A_gets_inverse_A_6(w); CHKERRQ(ierr); 1238 } 1239 1240 PetscFree(rtmp); 1241 C->factor = FACTOR_LU; 1242 C->assembled = PETSC_TRUE; 1243 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1244 PetscFunctionReturn(0); 1245 } 1246 1247 /* ------------------------------------------------------------*/ 1248 /* 1249 Version for when blocks are 5 by 5 1250 */ 1251 #undef __FUNC__ 1252 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_5" 1253 int MatLUFactorNumeric_SeqBAIJ_5(Mat A,Mat *B) 1254 { 1255 Mat C = *B; 1256 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1257 IS isrow = b->row, isicol = b->icol; 1258 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1259 int *ajtmpold, *ajtmp, nz, row; 1260 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 1261 register int *pj; 1262 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1263 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1264 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1265 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 1266 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 1267 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1268 MatScalar *ba = b->a,*aa = a->a; 1269 1270 PetscFunctionBegin; 1271 ierr = ISGetIndices(isrow,&r); CHKERRQ(ierr); 1272 ierr = ISGetIndices(isicol,&ic); CHKERRQ(ierr); 1273 rtmp = (MatScalar *) PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1274 1275 for ( i=0; i<n; i++ ) { 1276 nz = bi[i+1] - bi[i]; 1277 ajtmp = bj + bi[i]; 1278 for ( j=0; j<nz; j++ ) { 1279 x = rtmp+25*ajtmp[j]; 1280 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1281 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1282 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1283 } 1284 /* load in initial (unfactored row) */ 1285 idx = r[i]; 1286 nz = ai[idx+1] - ai[idx]; 1287 ajtmpold = aj + ai[idx]; 1288 v = aa + 25*ai[idx]; 1289 for ( j=0; j<nz; j++ ) { 1290 x = rtmp+25*ic[ajtmpold[j]]; 1291 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1292 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1293 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1294 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; 1295 x[18] = v[18]; x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; 1296 x[22] = v[22]; x[23] = v[23]; x[24] = v[24]; 1297 v += 25; 1298 } 1299 row = *ajtmp++; 1300 while (row < i) { 1301 pc = rtmp + 25*row; 1302 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1303 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1304 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1305 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; 1306 p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1307 p25 = pc[24]; 1308 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1309 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1310 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1311 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || 1312 p20 != 0.0 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || 1313 p24 != 0.0 || p25 != 0.0) { 1314 pv = ba + 25*diag_offset[row]; 1315 pj = bj + diag_offset[row] + 1; 1316 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1317 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1318 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1319 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1320 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; 1321 x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1322 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1323 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1324 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1325 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1326 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1327 1328 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1329 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1330 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1331 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1332 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1333 1334 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1335 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1336 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1337 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1338 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1339 1340 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1341 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1342 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1343 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1344 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1345 1346 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1347 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1348 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1349 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1350 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1351 1352 nz = bi[row+1] - diag_offset[row] - 1; 1353 pv += 25; 1354 for (j=0; j<nz; j++) { 1355 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1356 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1357 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1358 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; 1359 x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; 1360 x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1361 x = rtmp + 25*pj[j]; 1362 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1363 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1364 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1365 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1366 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1367 1368 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1369 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1370 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1371 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1372 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1373 1374 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1375 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1376 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1377 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1378 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1379 1380 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1381 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1382 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1383 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1384 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1385 1386 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1387 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1388 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1389 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1390 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1391 1392 pv += 25; 1393 } 1394 PLogFlops(250*nz+225); 1395 } 1396 row = *ajtmp++; 1397 } 1398 /* finished row so stick it into b->a */ 1399 pv = ba + 25*bi[i]; 1400 pj = bj + bi[i]; 1401 nz = bi[i+1] - bi[i]; 1402 for ( j=0; j<nz; j++ ) { 1403 x = rtmp+25*pj[j]; 1404 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1405 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1406 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1407 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; 1408 pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; 1409 pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; pv[24] = x[24]; 1410 pv += 25; 1411 } 1412 /* invert diagonal block */ 1413 w = ba + 25*diag_offset[i]; 1414 ierr = Kernel_A_gets_inverse_A_5(w); CHKERRQ(ierr); 1415 } 1416 1417 PetscFree(rtmp); 1418 ierr = ISRestoreIndices(isicol,&ic); CHKERRQ(ierr); 1419 ierr = ISRestoreIndices(isrow,&r); CHKERRQ(ierr); 1420 C->factor = FACTOR_LU; 1421 C->assembled = PETSC_TRUE; 1422 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1423 PetscFunctionReturn(0); 1424 } 1425 /* 1426 Version for when blocks are 5 by 5 Using natural ordering 1427 */ 1428 #undef __FUNC__ 1429 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_5_NaturalOrdering" 1430 int MatLUFactorNumeric_SeqBAIJ_5_NaturalOrdering(Mat A,Mat *B) 1431 { 1432 Mat C = *B; 1433 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1434 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1435 int *ajtmpold, *ajtmp, nz, row; 1436 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 1437 register int *pj; 1438 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1439 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1440 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1441 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1442 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1443 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1444 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1445 MatScalar *ba = b->a,*aa = a->a; 1446 1447 PetscFunctionBegin; 1448 rtmp = (MatScalar *) PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1449 for ( i=0; i<n; i++ ) { 1450 nz = bi[i+1] - bi[i]; 1451 ajtmp = bj + bi[i]; 1452 for ( j=0; j<nz; j++ ) { 1453 x = rtmp+25*ajtmp[j]; 1454 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1455 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1456 x[16] = x[17] = x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1457 } 1458 /* load in initial (unfactored row) */ 1459 nz = ai[i+1] - ai[i]; 1460 ajtmpold = aj + ai[i]; 1461 v = aa + 25*ai[i]; 1462 for ( j=0; j<nz; j++ ) { 1463 x = rtmp+25*ajtmpold[j]; 1464 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1465 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1466 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1467 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; 1468 x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1469 x[24] = v[24]; 1470 v += 25; 1471 } 1472 row = *ajtmp++; 1473 while (row < i) { 1474 pc = rtmp + 25*row; 1475 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1476 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1477 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1478 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; 1479 p19 = pc[18]; p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; 1480 p24 = pc[23]; p25 = pc[24]; 1481 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1482 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1483 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1484 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 1485 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || p25 != 0.0) { 1486 pv = ba + 25*diag_offset[row]; 1487 pj = bj + diag_offset[row] + 1; 1488 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1489 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1490 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1491 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; 1492 x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1493 x25 = pv[24]; 1494 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1495 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1496 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1497 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1498 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1499 1500 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1501 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1502 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1503 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1504 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1505 1506 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1507 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1508 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1509 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1510 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1511 1512 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1513 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1514 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1515 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1516 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1517 1518 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1519 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1520 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1521 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1522 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1523 1524 nz = bi[row+1] - diag_offset[row] - 1; 1525 pv += 25; 1526 for (j=0; j<nz; j++) { 1527 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1528 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1529 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1530 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1531 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; 1532 x24 = pv[23]; x25 = pv[24]; 1533 x = rtmp + 25*pj[j]; 1534 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1535 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1536 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1537 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1538 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1539 1540 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1541 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1542 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1543 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1544 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1545 1546 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1547 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1548 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1549 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1550 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1551 1552 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1553 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1554 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1555 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1556 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1557 1558 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1559 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1560 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1561 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1562 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1563 pv += 25; 1564 } 1565 PLogFlops(250*nz+225); 1566 } 1567 row = *ajtmp++; 1568 } 1569 /* finished row so stick it into b->a */ 1570 pv = ba + 25*bi[i]; 1571 pj = bj + bi[i]; 1572 nz = bi[i+1] - bi[i]; 1573 for ( j=0; j<nz; j++ ) { 1574 x = rtmp+25*pj[j]; 1575 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1576 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1577 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1578 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; pv[17] = x[17]; 1579 pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; 1580 pv[23] = x[23]; pv[24] = x[24]; 1581 pv += 25; 1582 } 1583 /* invert diagonal block */ 1584 w = ba + 25*diag_offset[i]; 1585 ierr = Kernel_A_gets_inverse_A_5(w); CHKERRQ(ierr); 1586 } 1587 1588 PetscFree(rtmp); 1589 C->factor = FACTOR_LU; 1590 C->assembled = PETSC_TRUE; 1591 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1592 PetscFunctionReturn(0); 1593 } 1594 1595 /* ------------------------------------------------------------*/ 1596 /* 1597 Version for when blocks are 4 by 4 1598 */ 1599 #undef __FUNC__ 1600 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_4" 1601 int MatLUFactorNumeric_SeqBAIJ_4(Mat A,Mat *B) 1602 { 1603 Mat C = *B; 1604 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1605 IS isrow = b->row, isicol = b->icol; 1606 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1607 int *ajtmpold, *ajtmp, nz, row; 1608 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 1609 register int *pj; 1610 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1611 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1612 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1613 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1614 MatScalar m13,m14,m15,m16; 1615 MatScalar *ba = b->a,*aa = a->a; 1616 1617 PetscFunctionBegin; 1618 ierr = ISGetIndices(isrow,&r); CHKERRQ(ierr); 1619 ierr = ISGetIndices(isicol,&ic); CHKERRQ(ierr); 1620 rtmp = (MatScalar *) PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1621 1622 for ( i=0; i<n; i++ ) { 1623 nz = bi[i+1] - bi[i]; 1624 ajtmp = bj + bi[i]; 1625 for ( j=0; j<nz; j++ ) { 1626 x = rtmp+16*ajtmp[j]; 1627 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1628 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1629 } 1630 /* load in initial (unfactored row) */ 1631 idx = r[i]; 1632 nz = ai[idx+1] - ai[idx]; 1633 ajtmpold = aj + ai[idx]; 1634 v = aa + 16*ai[idx]; 1635 for ( j=0; j<nz; j++ ) { 1636 x = rtmp+16*ic[ajtmpold[j]]; 1637 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1638 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1639 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1640 x[14] = v[14]; x[15] = v[15]; 1641 v += 16; 1642 } 1643 row = *ajtmp++; 1644 while (row < i) { 1645 pc = rtmp + 16*row; 1646 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1647 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1648 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1649 p15 = pc[14]; p16 = pc[15]; 1650 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1651 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1652 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1653 || p16 != 0.0) { 1654 pv = ba + 16*diag_offset[row]; 1655 pj = bj + diag_offset[row] + 1; 1656 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1657 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1658 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1659 x15 = pv[14]; x16 = pv[15]; 1660 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1661 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1662 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1663 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1664 1665 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1666 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1667 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1668 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1669 1670 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1671 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1672 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1673 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1674 1675 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1676 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1677 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1678 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1679 1680 nz = bi[row+1] - diag_offset[row] - 1; 1681 pv += 16; 1682 for (j=0; j<nz; j++) { 1683 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1684 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1685 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1686 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1687 x = rtmp + 16*pj[j]; 1688 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1689 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1690 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1691 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1692 1693 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1694 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1695 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1696 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1697 1698 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1699 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1700 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1701 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1702 1703 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1704 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1705 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1706 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1707 1708 pv += 16; 1709 } 1710 PLogFlops(128*nz+112); 1711 } 1712 row = *ajtmp++; 1713 } 1714 /* finished row so stick it into b->a */ 1715 pv = ba + 16*bi[i]; 1716 pj = bj + bi[i]; 1717 nz = bi[i+1] - bi[i]; 1718 for ( j=0; j<nz; j++ ) { 1719 x = rtmp+16*pj[j]; 1720 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1721 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1722 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1723 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1724 pv += 16; 1725 } 1726 /* invert diagonal block */ 1727 w = ba + 16*diag_offset[i]; 1728 ierr = Kernel_A_gets_inverse_A_4(w); CHKERRQ(ierr); 1729 } 1730 1731 PetscFree(rtmp); 1732 ierr = ISRestoreIndices(isicol,&ic); CHKERRQ(ierr); 1733 ierr = ISRestoreIndices(isrow,&r); CHKERRQ(ierr); 1734 C->factor = FACTOR_LU; 1735 C->assembled = PETSC_TRUE; 1736 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1737 PetscFunctionReturn(0); 1738 } 1739 /* 1740 Version for when blocks are 4 by 4 Using natural ordering 1741 */ 1742 #undef __FUNC__ 1743 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_4_NaturalOrdering" 1744 int MatLUFactorNumeric_SeqBAIJ_4_NaturalOrdering(Mat A,Mat *B) 1745 { 1746 Mat C = *B; 1747 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1748 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1749 int *ajtmpold, *ajtmp, nz, row; 1750 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 1751 register int *pj; 1752 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1753 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1754 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1755 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1756 MatScalar m13,m14,m15,m16; 1757 MatScalar *ba = b->a,*aa = a->a; 1758 1759 PetscFunctionBegin; 1760 rtmp = (MatScalar *) PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1761 1762 for ( i=0; i<n; i++ ) { 1763 nz = bi[i+1] - bi[i]; 1764 ajtmp = bj + bi[i]; 1765 for ( j=0; j<nz; j++ ) { 1766 x = rtmp+16*ajtmp[j]; 1767 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1768 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1769 } 1770 /* load in initial (unfactored row) */ 1771 nz = ai[i+1] - ai[i]; 1772 ajtmpold = aj + ai[i]; 1773 v = aa + 16*ai[i]; 1774 for ( j=0; j<nz; j++ ) { 1775 x = rtmp+16*ajtmpold[j]; 1776 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1777 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1778 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1779 x[14] = v[14]; x[15] = v[15]; 1780 v += 16; 1781 } 1782 row = *ajtmp++; 1783 while (row < i) { 1784 pc = rtmp + 16*row; 1785 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1786 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1787 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1788 p15 = pc[14]; p16 = pc[15]; 1789 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1790 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1791 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1792 || p16 != 0.0) { 1793 pv = ba + 16*diag_offset[row]; 1794 pj = bj + diag_offset[row] + 1; 1795 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1796 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1797 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1798 x15 = pv[14]; x16 = pv[15]; 1799 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1800 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1801 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1802 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1803 1804 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1805 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1806 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1807 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1808 1809 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1810 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1811 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1812 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1813 1814 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1815 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1816 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1817 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1818 1819 nz = bi[row+1] - diag_offset[row] - 1; 1820 pv += 16; 1821 for (j=0; j<nz; j++) { 1822 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1823 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1824 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1825 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1826 x = rtmp + 16*pj[j]; 1827 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1828 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1829 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1830 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1831 1832 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1833 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1834 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1835 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1836 1837 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1838 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1839 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1840 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1841 1842 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1843 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1844 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1845 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1846 1847 pv += 16; 1848 } 1849 PLogFlops(128*nz+112); 1850 } 1851 row = *ajtmp++; 1852 } 1853 /* finished row so stick it into b->a */ 1854 pv = ba + 16*bi[i]; 1855 pj = bj + bi[i]; 1856 nz = bi[i+1] - bi[i]; 1857 for ( j=0; j<nz; j++ ) { 1858 x = rtmp+16*pj[j]; 1859 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1860 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1861 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1862 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1863 pv += 16; 1864 } 1865 /* invert diagonal block */ 1866 w = ba + 16*diag_offset[i]; 1867 ierr = Kernel_A_gets_inverse_A_4(w); CHKERRQ(ierr); 1868 } 1869 1870 PetscFree(rtmp); 1871 C->factor = FACTOR_LU; 1872 C->assembled = PETSC_TRUE; 1873 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1874 PetscFunctionReturn(0); 1875 } 1876 1877 1878 /* ------------------------------------------------------------*/ 1879 /* 1880 Version for when blocks are 3 by 3 1881 */ 1882 #undef __FUNC__ 1883 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_3" 1884 int MatLUFactorNumeric_SeqBAIJ_3(Mat A,Mat *B) 1885 { 1886 Mat C = *B; 1887 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1888 IS isrow = b->row, isicol = b->icol; 1889 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1890 int *ajtmpold, *ajtmp, nz, row, *ai=a->i,*aj=a->j; 1891 int *diag_offset = b->diag,idx; 1892 register int *pj; 1893 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1894 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1895 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 1896 MatScalar *ba = b->a,*aa = a->a; 1897 1898 PetscFunctionBegin; 1899 ierr = ISGetIndices(isrow,&r); CHKERRQ(ierr); 1900 ierr = ISGetIndices(isicol,&ic); CHKERRQ(ierr); 1901 rtmp = (MatScalar *) PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1902 1903 for ( i=0; i<n; i++ ) { 1904 nz = bi[i+1] - bi[i]; 1905 ajtmp = bj + bi[i]; 1906 for ( j=0; j<nz; j++ ) { 1907 x = rtmp + 9*ajtmp[j]; 1908 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 1909 } 1910 /* load in initial (unfactored row) */ 1911 idx = r[i]; 1912 nz = ai[idx+1] - ai[idx]; 1913 ajtmpold = aj + ai[idx]; 1914 v = aa + 9*ai[idx]; 1915 for ( j=0; j<nz; j++ ) { 1916 x = rtmp + 9*ic[ajtmpold[j]]; 1917 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1918 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1919 v += 9; 1920 } 1921 row = *ajtmp++; 1922 while (row < i) { 1923 pc = rtmp + 9*row; 1924 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1925 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1926 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1927 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 1928 pv = ba + 9*diag_offset[row]; 1929 pj = bj + diag_offset[row] + 1; 1930 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1931 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1932 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 1933 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 1934 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 1935 1936 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 1937 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 1938 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 1939 1940 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 1941 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 1942 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 1943 nz = bi[row+1] - diag_offset[row] - 1; 1944 pv += 9; 1945 for (j=0; j<nz; j++) { 1946 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1947 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1948 x = rtmp + 9*pj[j]; 1949 x[0] -= m1*x1 + m4*x2 + m7*x3; 1950 x[1] -= m2*x1 + m5*x2 + m8*x3; 1951 x[2] -= m3*x1 + m6*x2 + m9*x3; 1952 1953 x[3] -= m1*x4 + m4*x5 + m7*x6; 1954 x[4] -= m2*x4 + m5*x5 + m8*x6; 1955 x[5] -= m3*x4 + m6*x5 + m9*x6; 1956 1957 x[6] -= m1*x7 + m4*x8 + m7*x9; 1958 x[7] -= m2*x7 + m5*x8 + m8*x9; 1959 x[8] -= m3*x7 + m6*x8 + m9*x9; 1960 pv += 9; 1961 } 1962 PLogFlops(54*nz+36); 1963 } 1964 row = *ajtmp++; 1965 } 1966 /* finished row so stick it into b->a */ 1967 pv = ba + 9*bi[i]; 1968 pj = bj + bi[i]; 1969 nz = bi[i+1] - bi[i]; 1970 for ( j=0; j<nz; j++ ) { 1971 x = rtmp + 9*pj[j]; 1972 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1973 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1974 pv += 9; 1975 } 1976 /* invert diagonal block */ 1977 w = ba + 9*diag_offset[i]; 1978 ierr = Kernel_A_gets_inverse_A_3(w); CHKERRQ(ierr); 1979 } 1980 1981 PetscFree(rtmp); 1982 ierr = ISRestoreIndices(isicol,&ic); CHKERRQ(ierr); 1983 ierr = ISRestoreIndices(isrow,&r); CHKERRQ(ierr); 1984 C->factor = FACTOR_LU; 1985 C->assembled = PETSC_TRUE; 1986 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 1987 PetscFunctionReturn(0); 1988 } 1989 /* 1990 Version for when blocks are 3 by 3 Using natural ordering 1991 */ 1992 #undef __FUNC__ 1993 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_3_NaturalOrdering" 1994 int MatLUFactorNumeric_SeqBAIJ_3_NaturalOrdering(Mat A,Mat *B) 1995 { 1996 Mat C = *B; 1997 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1998 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1999 int *ajtmpold, *ajtmp, nz, row; 2000 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 2001 register int *pj; 2002 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2003 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 2004 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 2005 MatScalar *ba = b->a,*aa = a->a; 2006 2007 PetscFunctionBegin; 2008 rtmp = (MatScalar *) PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2009 2010 for ( i=0; i<n; i++ ) { 2011 nz = bi[i+1] - bi[i]; 2012 ajtmp = bj + bi[i]; 2013 for ( j=0; j<nz; j++ ) { 2014 x = rtmp+9*ajtmp[j]; 2015 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 2016 } 2017 /* load in initial (unfactored row) */ 2018 nz = ai[i+1] - ai[i]; 2019 ajtmpold = aj + ai[i]; 2020 v = aa + 9*ai[i]; 2021 for ( j=0; j<nz; j++ ) { 2022 x = rtmp+9*ajtmpold[j]; 2023 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2024 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 2025 v += 9; 2026 } 2027 row = *ajtmp++; 2028 while (row < i) { 2029 pc = rtmp + 9*row; 2030 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2031 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 2032 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 2033 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 2034 pv = ba + 9*diag_offset[row]; 2035 pj = bj + diag_offset[row] + 1; 2036 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2037 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2038 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 2039 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 2040 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 2041 2042 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 2043 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 2044 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 2045 2046 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 2047 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 2048 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 2049 2050 nz = bi[row+1] - diag_offset[row] - 1; 2051 pv += 9; 2052 for (j=0; j<nz; j++) { 2053 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2054 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2055 x = rtmp + 9*pj[j]; 2056 x[0] -= m1*x1 + m4*x2 + m7*x3; 2057 x[1] -= m2*x1 + m5*x2 + m8*x3; 2058 x[2] -= m3*x1 + m6*x2 + m9*x3; 2059 2060 x[3] -= m1*x4 + m4*x5 + m7*x6; 2061 x[4] -= m2*x4 + m5*x5 + m8*x6; 2062 x[5] -= m3*x4 + m6*x5 + m9*x6; 2063 2064 x[6] -= m1*x7 + m4*x8 + m7*x9; 2065 x[7] -= m2*x7 + m5*x8 + m8*x9; 2066 x[8] -= m3*x7 + m6*x8 + m9*x9; 2067 pv += 9; 2068 } 2069 PLogFlops(54*nz+36); 2070 } 2071 row = *ajtmp++; 2072 } 2073 /* finished row so stick it into b->a */ 2074 pv = ba + 9*bi[i]; 2075 pj = bj + bi[i]; 2076 nz = bi[i+1] - bi[i]; 2077 for ( j=0; j<nz; j++ ) { 2078 x = rtmp+9*pj[j]; 2079 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2080 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 2081 pv += 9; 2082 } 2083 /* invert diagonal block */ 2084 w = ba + 9*diag_offset[i]; 2085 ierr = Kernel_A_gets_inverse_A_3(w); CHKERRQ(ierr); 2086 } 2087 2088 PetscFree(rtmp); 2089 C->factor = FACTOR_LU; 2090 C->assembled = PETSC_TRUE; 2091 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 2092 PetscFunctionReturn(0); 2093 } 2094 2095 /* ------------------------------------------------------------*/ 2096 /* 2097 Version for when blocks are 2 by 2 2098 */ 2099 #undef __FUNC__ 2100 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_2" 2101 int MatLUFactorNumeric_SeqBAIJ_2(Mat A,Mat *B) 2102 { 2103 Mat C = *B; 2104 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 2105 IS isrow = b->row, isicol = b->icol; 2106 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 2107 int *ajtmpold, *ajtmp, nz, row, v_pivots[2]; 2108 int *diag_offset=b->diag,bs = 2,idx,*ai=a->i,*aj=a->j; 2109 register int *pj; 2110 register MatScalar *pv,*v,*rtmp,m1,m2,m3,m4,*pc,*w,*x,x1,x2,x3,x4; 2111 MatScalar p1,p2,p3,p4,v_work[2]; 2112 MatScalar *ba = b->a,*aa = a->a; 2113 2114 PetscFunctionBegin; 2115 ierr = ISGetIndices(isrow,&r); CHKERRQ(ierr); 2116 ierr = ISGetIndices(isicol,&ic); CHKERRQ(ierr); 2117 rtmp = (MatScalar *) PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2118 2119 for ( i=0; i<n; i++ ) { 2120 nz = bi[i+1] - bi[i]; 2121 ajtmp = bj + bi[i]; 2122 for ( j=0; j<nz; j++ ) { 2123 x = rtmp+4*ajtmp[j]; x[0] = x[1] = x[2] = x[3] = 0.0; 2124 } 2125 /* load in initial (unfactored row) */ 2126 idx = r[i]; 2127 nz = ai[idx+1] - ai[idx]; 2128 ajtmpold = aj + ai[idx]; 2129 v = aa + 4*ai[idx]; 2130 for ( j=0; j<nz; j++ ) { 2131 x = rtmp+4*ic[ajtmpold[j]]; 2132 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2133 v += 4; 2134 } 2135 row = *ajtmp++; 2136 while (row < i) { 2137 pc = rtmp + 4*row; 2138 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2139 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2140 pv = ba + 4*diag_offset[row]; 2141 pj = bj + diag_offset[row] + 1; 2142 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2143 pc[0] = m1 = p1*x1 + p3*x2; 2144 pc[1] = m2 = p2*x1 + p4*x2; 2145 pc[2] = m3 = p1*x3 + p3*x4; 2146 pc[3] = m4 = p2*x3 + p4*x4; 2147 nz = bi[row+1] - diag_offset[row] - 1; 2148 pv += 4; 2149 for (j=0; j<nz; j++) { 2150 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2151 x = rtmp + 4*pj[j]; 2152 x[0] -= m1*x1 + m3*x2; 2153 x[1] -= m2*x1 + m4*x2; 2154 x[2] -= m1*x3 + m3*x4; 2155 x[3] -= m2*x3 + m4*x4; 2156 pv += 4; 2157 } 2158 PLogFlops(16*nz+12); 2159 } 2160 row = *ajtmp++; 2161 } 2162 /* finished row so stick it into b->a */ 2163 pv = ba + 4*bi[i]; 2164 pj = bj + bi[i]; 2165 nz = bi[i+1] - bi[i]; 2166 for ( j=0; j<nz; j++ ) { 2167 x = rtmp+4*pj[j]; 2168 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2169 pv += 4; 2170 } 2171 /* invert diagonal block */ 2172 w = ba + 4*diag_offset[i]; 2173 ierr = Kernel_A_gets_inverse_A_2(w); CHKERRQ(ierr); 2174 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2175 } 2176 2177 PetscFree(rtmp); 2178 ierr = ISRestoreIndices(isicol,&ic); CHKERRQ(ierr); 2179 ierr = ISRestoreIndices(isrow,&r); CHKERRQ(ierr); 2180 C->factor = FACTOR_LU; 2181 C->assembled = PETSC_TRUE; 2182 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2183 PetscFunctionReturn(0); 2184 } 2185 /* 2186 Version for when blocks are 2 by 2 Using natural ordering 2187 */ 2188 #undef __FUNC__ 2189 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering" 2190 int MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering(Mat A,Mat *B) 2191 { 2192 Mat C = *B; 2193 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 2194 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 2195 int *ajtmpold, *ajtmp, nz, row, v_pivots[2]; 2196 int *diag_offset = b->diag, bs = 2,*ai=a->i,*aj=a->j; 2197 register int *pj; 2198 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2199 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,x1,x2,x3,x4,v_work[2]; 2200 MatScalar *ba = b->a,*aa = a->a; 2201 2202 PetscFunctionBegin; 2203 rtmp = (MatScalar *) PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2204 2205 for ( i=0; i<n; i++ ) { 2206 nz = bi[i+1] - bi[i]; 2207 ajtmp = bj + bi[i]; 2208 for ( j=0; j<nz; j++ ) { 2209 x = rtmp+4*ajtmp[j]; 2210 x[0] = x[1] = x[2] = x[3] = 0.0; 2211 } 2212 /* load in initial (unfactored row) */ 2213 nz = ai[i+1] - ai[i]; 2214 ajtmpold = aj + ai[i]; 2215 v = aa + 4*ai[i]; 2216 for ( j=0; j<nz; j++ ) { 2217 x = rtmp+4*ajtmpold[j]; 2218 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2219 v += 4; 2220 } 2221 row = *ajtmp++; 2222 while (row < i) { 2223 pc = rtmp + 4*row; 2224 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2225 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2226 pv = ba + 4*diag_offset[row]; 2227 pj = bj + diag_offset[row] + 1; 2228 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2229 pc[0] = m1 = p1*x1 + p3*x2; 2230 pc[1] = m2 = p2*x1 + p4*x2; 2231 pc[2] = m3 = p1*x3 + p3*x4; 2232 pc[3] = m4 = p2*x3 + p4*x4; 2233 nz = bi[row+1] - diag_offset[row] - 1; 2234 pv += 4; 2235 for (j=0; j<nz; j++) { 2236 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2237 x = rtmp + 4*pj[j]; 2238 x[0] -= m1*x1 + m3*x2; 2239 x[1] -= m2*x1 + m4*x2; 2240 x[2] -= m1*x3 + m3*x4; 2241 x[3] -= m2*x3 + m4*x4; 2242 pv += 4; 2243 } 2244 PLogFlops(16*nz+12); 2245 } 2246 row = *ajtmp++; 2247 } 2248 /* finished row so stick it into b->a */ 2249 pv = ba + 4*bi[i]; 2250 pj = bj + bi[i]; 2251 nz = bi[i+1] - bi[i]; 2252 for ( j=0; j<nz; j++ ) { 2253 x = rtmp+4*pj[j]; 2254 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2255 pv += 4; 2256 } 2257 /* invert diagonal block */ 2258 w = ba + 4*diag_offset[i]; 2259 ierr = Kernel_A_gets_inverse_A_2(w); CHKERRQ(ierr); 2260 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2261 } 2262 2263 PetscFree(rtmp); 2264 C->factor = FACTOR_LU; 2265 C->assembled = PETSC_TRUE; 2266 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2267 PetscFunctionReturn(0); 2268 } 2269 2270 /* ----------------------------------------------------------- */ 2271 /* 2272 Version for when blocks are 1 by 1. 2273 */ 2274 #undef __FUNC__ 2275 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_1" 2276 int MatLUFactorNumeric_SeqBAIJ_1(Mat A,Mat *B) 2277 { 2278 Mat C = *B; 2279 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data, *b = (Mat_SeqBAIJ *)C->data; 2280 IS isrow = b->row, isicol = b->icol; 2281 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 2282 int *ajtmpold, *ajtmp, nz, row,*ai = a->i,*aj = a->j; 2283 int *diag_offset = b->diag,diag; 2284 register int *pj; 2285 register MatScalar *pv,*v,*rtmp,multiplier,*pc; 2286 MatScalar *ba = b->a,*aa = a->a; 2287 2288 PetscFunctionBegin; 2289 ierr = ISGetIndices(isrow,&r); CHKERRQ(ierr); 2290 ierr = ISGetIndices(isicol,&ic); CHKERRQ(ierr); 2291 rtmp = (MatScalar *) PetscMalloc((n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2292 2293 for ( i=0; i<n; i++ ) { 2294 nz = bi[i+1] - bi[i]; 2295 ajtmp = bj + bi[i]; 2296 for ( j=0; j<nz; j++ ) rtmp[ajtmp[j]] = 0.0; 2297 2298 /* load in initial (unfactored row) */ 2299 nz = ai[r[i]+1] - ai[r[i]]; 2300 ajtmpold = aj + ai[r[i]]; 2301 v = aa + ai[r[i]]; 2302 for ( j=0; j<nz; j++ ) rtmp[ic[ajtmpold[j]]] = v[j]; 2303 2304 row = *ajtmp++; 2305 while (row < i) { 2306 pc = rtmp + row; 2307 if (*pc != 0.0) { 2308 pv = ba + diag_offset[row]; 2309 pj = bj + diag_offset[row] + 1; 2310 multiplier = *pc * *pv++; 2311 *pc = multiplier; 2312 nz = bi[row+1] - diag_offset[row] - 1; 2313 for (j=0; j<nz; j++) rtmp[pj[j]] -= multiplier * pv[j]; 2314 PLogFlops(1+2*nz); 2315 } 2316 row = *ajtmp++; 2317 } 2318 /* finished row so stick it into b->a */ 2319 pv = ba + bi[i]; 2320 pj = bj + bi[i]; 2321 nz = bi[i+1] - bi[i]; 2322 for ( j=0; j<nz; j++ ) {pv[j] = rtmp[pj[j]];} 2323 diag = diag_offset[i] - bi[i]; 2324 /* check pivot entry for current row */ 2325 if (pv[diag] == 0.0) { 2326 SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,0,"Zero pivot"); 2327 } 2328 pv[diag] = 1.0/pv[diag]; 2329 } 2330 2331 PetscFree(rtmp); 2332 ierr = ISRestoreIndices(isicol,&ic); CHKERRQ(ierr); 2333 ierr = ISRestoreIndices(isrow,&r); CHKERRQ(ierr); 2334 C->factor = FACTOR_LU; 2335 C->assembled = PETSC_TRUE; 2336 PLogFlops(b->n); 2337 PetscFunctionReturn(0); 2338 } 2339 2340 /* ----------------------------------------------------------- */ 2341 #undef __FUNC__ 2342 #define __FUNC__ "MatLUFactor_SeqBAIJ" 2343 int MatLUFactor_SeqBAIJ(Mat A,IS row,IS col,double f) 2344 { 2345 Mat_SeqBAIJ *mat = (Mat_SeqBAIJ *) A->data; 2346 int ierr; 2347 Mat C; 2348 PetscOps *Abops; 2349 MatOps Aops; 2350 2351 PetscFunctionBegin; 2352 ierr = MatLUFactorSymbolic(A,row,col,f,&C); CHKERRQ(ierr); 2353 ierr = MatLUFactorNumeric(A,&C); CHKERRQ(ierr); 2354 2355 /* free all the data structures from mat */ 2356 PetscFree(mat->a); 2357 if (!mat->singlemalloc) {PetscFree(mat->i); PetscFree(mat->j);} 2358 if (mat->diag) PetscFree(mat->diag); 2359 if (mat->ilen) PetscFree(mat->ilen); 2360 if (mat->imax) PetscFree(mat->imax); 2361 if (mat->solve_work) PetscFree(mat->solve_work); 2362 if (mat->mult_work) PetscFree(mat->mult_work); 2363 if (mat->icol) {ierr = ISDestroy(mat->icol);CHKERRQ(ierr);} 2364 PetscFree(mat); 2365 2366 ierr = MapDestroy(A->rmap);CHKERRQ(ierr); 2367 ierr = MapDestroy(A->cmap);CHKERRQ(ierr); 2368 2369 /* 2370 This is horrible, horrible code. We need to keep the 2371 A pointers for the bops and ops but copy everything 2372 else from C. 2373 */ 2374 Abops = A->bops; 2375 Aops = A->ops; 2376 PetscMemcpy(A,C,sizeof(struct _p_Mat)); 2377 mat = (Mat_SeqBAIJ *) A->data; 2378 PLogObjectParent(A,mat->icol); 2379 2380 A->bops = Abops; 2381 A->ops = Aops; 2382 A->qlist = 0; 2383 2384 PetscHeaderDestroy(C); 2385 PetscFunctionReturn(0); 2386 } 2387 2388 2389 2390