1 #ifdef PETSC_RCS_HEADER 2 static char vcid[] = "$Id: baijfact.c,v 1.70 1999/03/17 23:23:13 bsmith Exp balay $"; 3 #endif 4 /* 5 Factorization code for BAIJ format. 6 */ 7 8 #include "src/mat/impls/baij/seq/baij.h" 9 #include "src/vec/vecimpl.h" 10 #include "src/inline/ilu.h" 11 12 13 /* 14 The symbolic factorization code is identical to that for AIJ format, 15 except for very small changes since this is now a SeqBAIJ datastructure. 16 NOT good code reuse. 17 */ 18 #undef __FUNC__ 19 #define __FUNC__ "MatLUFactorSymbolic_SeqBAIJ" 20 int MatLUFactorSymbolic_SeqBAIJ(Mat A,IS isrow,IS iscol,double f,Mat *B) 21 { 22 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data, *b; 23 IS isicol; 24 int *r,*ic, ierr, i, n = a->mbs, *ai = a->i, *aj = a->j; 25 int *ainew,*ajnew, jmax,*fill, *ajtmp, nz, bs = a->bs, bs2=a->bs2; 26 int *idnew, idx, row,m,fm, nnz, nzi,realloc = 0,nzbd,*im; 27 28 PetscFunctionBegin; 29 PetscValidHeaderSpecific(isrow,IS_COOKIE); 30 PetscValidHeaderSpecific(iscol,IS_COOKIE); 31 ierr = ISInvertPermutation(iscol,&isicol); CHKERRQ(ierr); 32 ISGetIndices(isrow,&r); ISGetIndices(isicol,&ic); 33 34 /* get new row pointers */ 35 ainew = (int *) PetscMalloc( (n+1)*sizeof(int) ); CHKPTRQ(ainew); 36 ainew[0] = 0; 37 /* don't know how many column pointers are needed so estimate */ 38 jmax = (int) (f*ai[n] + 1); 39 ajnew = (int *) PetscMalloc( (jmax)*sizeof(int) ); CHKPTRQ(ajnew); 40 /* fill is a linked list of nonzeros in active row */ 41 fill = (int *) PetscMalloc( (2*n+1)*sizeof(int)); CHKPTRQ(fill); 42 im = fill + n + 1; 43 /* idnew is location of diagonal in factor */ 44 idnew = (int *) PetscMalloc( (n+1)*sizeof(int)); CHKPTRQ(idnew); 45 idnew[0] = 0; 46 47 for ( i=0; i<n; i++ ) { 48 /* first copy previous fill into linked list */ 49 nnz = nz = ai[r[i]+1] - ai[r[i]]; 50 if (!nz) SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,1,"Empty row in matrix"); 51 ajtmp = aj + ai[r[i]]; 52 fill[n] = n; 53 while (nz--) { 54 fm = n; 55 idx = ic[*ajtmp++]; 56 do { 57 m = fm; 58 fm = fill[m]; 59 } while (fm < idx); 60 fill[m] = idx; 61 fill[idx] = fm; 62 } 63 row = fill[n]; 64 while ( row < i ) { 65 ajtmp = ajnew + idnew[row] + 1; 66 nzbd = 1 + idnew[row] - ainew[row]; 67 nz = im[row] - nzbd; 68 fm = row; 69 while (nz-- > 0) { 70 idx = *ajtmp++; 71 nzbd++; 72 if (idx == i) im[row] = nzbd; 73 do { 74 m = fm; 75 fm = fill[m]; 76 } while (fm < idx); 77 if (fm != idx) { 78 fill[m] = idx; 79 fill[idx] = fm; 80 fm = idx; 81 nnz++; 82 } 83 } 84 row = fill[row]; 85 } 86 /* copy new filled row into permanent storage */ 87 ainew[i+1] = ainew[i] + nnz; 88 if (ainew[i+1] > jmax) { 89 90 /* estimate how much additional space we will need */ 91 /* use the strategy suggested by David Hysom <hysom@perch-t.icase.edu> */ 92 /* just double the memory each time */ 93 int maxadd = jmax; 94 /* maxadd = (int) ((f*(ai[n]+1)*(n-i+5))/n); */ 95 if (maxadd < nnz) maxadd = (n-i)*(nnz+1); 96 jmax += maxadd; 97 98 /* allocate a longer ajnew */ 99 ajtmp = (int *) PetscMalloc( jmax*sizeof(int) );CHKPTRQ(ajtmp); 100 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] = 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] = x[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 /* 531 Version for when blocks are 7 by 7 Using natural ordering 532 */ 533 #undef __FUNC__ 534 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_7_NaturalOrdering" 535 int MatLUFactorNumeric_SeqBAIJ_7_NaturalOrdering(Mat A,Mat *B) 536 { 537 Mat C = *B; 538 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 539 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 540 int *ajtmpold, *ajtmp, nz, row; 541 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 542 register int *pj; 543 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 544 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 545 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 546 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 547 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 548 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 549 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 550 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 551 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 552 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 553 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 554 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 555 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 556 MatScalar *ba = b->a,*aa = a->a; 557 558 PetscFunctionBegin; 559 rtmp = (MatScalar *) PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 560 for ( i=0; i<n; i++ ) { 561 nz = bi[i+1] - bi[i]; 562 ajtmp = bj + bi[i]; 563 for ( j=0; j<nz; j++ ) { 564 x = rtmp+49*ajtmp[j]; 565 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 566 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 567 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 568 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 569 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 570 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 571 } 572 /* load in initial (unfactored row) */ 573 nz = ai[i+1] - ai[i]; 574 ajtmpold = aj + ai[i]; 575 v = aa + 49*ai[i]; 576 for ( j=0; j<nz; j++ ) { 577 x = rtmp+49*ajtmpold[j]; 578 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 579 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 580 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 581 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 582 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 583 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 584 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 585 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 586 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 587 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 588 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 589 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 590 x[48] = v[48]; 591 v += 49; 592 } 593 row = *ajtmp++; 594 while (row < i) { 595 pc = rtmp + 49*row; 596 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 597 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 598 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 599 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 600 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 601 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 602 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 603 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 604 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 605 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 606 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 607 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 608 p49 = pc[48]; 609 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 610 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 611 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 612 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 613 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 614 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 615 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 616 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 617 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 618 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 619 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 620 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 621 p49 != 0.0) { 622 pv = ba + 49*diag_offset[row]; 623 pj = bj + diag_offset[row] + 1; 624 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 625 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 626 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 627 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 628 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 629 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 630 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 631 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 632 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 633 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 634 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 635 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 636 x49 = pv[48]; 637 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 638 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 639 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 640 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 641 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 642 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 643 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 644 645 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 646 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 647 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 648 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 649 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 650 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 651 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 652 653 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 654 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 655 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 656 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 657 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 658 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 659 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 660 661 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 662 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 663 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 664 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 665 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 666 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 667 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 668 669 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 670 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 671 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 672 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 673 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 674 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 675 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 676 677 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 678 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 679 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 680 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 681 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 682 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 683 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 684 685 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 686 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 687 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 688 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 689 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 690 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 691 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 692 693 nz = bi[row+1] - diag_offset[row] - 1; 694 pv += 49; 695 for (j=0; j<nz; j++) { 696 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 697 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 698 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 699 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 700 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 701 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 702 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 703 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 704 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 705 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 706 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 707 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 708 x49 = pv[48]; 709 x = rtmp + 49*pj[j]; 710 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 711 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 712 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 713 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 714 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 715 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 716 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 717 718 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 719 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 720 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 721 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 722 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 723 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 724 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 725 726 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 727 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 728 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 729 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 730 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 731 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 732 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 733 734 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 735 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 736 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 737 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 738 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 739 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 740 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 741 742 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 743 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 744 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 745 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 746 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 747 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 748 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 749 750 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 751 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 752 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 753 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 754 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 755 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 756 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 757 758 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 759 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 760 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 761 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 762 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 763 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 764 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 765 pv += 49; 766 } 767 PLogFlops(686*nz+637); 768 } 769 row = *ajtmp++; 770 } 771 /* finished row so stick it into b->a */ 772 pv = ba + 49*bi[i]; 773 pj = bj + bi[i]; 774 nz = bi[i+1] - bi[i]; 775 for ( j=0; j<nz; j++ ) { 776 x = rtmp+49*pj[j]; 777 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 778 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 779 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 780 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 781 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 782 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 783 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 784 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 785 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 786 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 787 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 788 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 789 pv[48] = x[48]; 790 pv += 49; 791 } 792 /* invert diagonal block */ 793 w = ba + 49*diag_offset[i]; 794 ierr = Kernel_A_gets_inverse_A_7(w); CHKERRQ(ierr); 795 } 796 797 PetscFree(rtmp); 798 C->factor = FACTOR_LU; 799 C->assembled = PETSC_TRUE; 800 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 801 PetscFunctionReturn(0); 802 } 803 804 /* ------------------------------------------------------------*/ 805 /* 806 Version for when blocks are 6 by 6 807 */ 808 #undef __FUNC__ 809 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_6" 810 int MatLUFactorNumeric_SeqBAIJ_6(Mat A,Mat *B) 811 { 812 Mat C = *B; 813 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 814 IS isrow = b->row, isicol = b->icol; 815 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 816 int *ajtmpold, *ajtmp, nz, row; 817 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 818 register int *pj; 819 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 820 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 821 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 822 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 823 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 824 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 825 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 826 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 827 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 828 MatScalar *ba = b->a,*aa = a->a; 829 830 PetscFunctionBegin; 831 ierr = ISGetIndices(isrow,&r); CHKERRQ(ierr); 832 ierr = ISGetIndices(isicol,&ic); CHKERRQ(ierr); 833 rtmp = (MatScalar *) PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 834 835 for ( i=0; i<n; i++ ) { 836 nz = bi[i+1] - bi[i]; 837 ajtmp = bj + bi[i]; 838 for ( j=0; j<nz; j++ ) { 839 x = rtmp+36*ajtmp[j]; 840 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 841 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 842 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 843 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 844 x[34] = x[35] = 0.0 ; 845 } 846 /* load in initial (unfactored row) */ 847 idx = r[i]; 848 nz = ai[idx+1] - ai[idx]; 849 ajtmpold = aj + ai[idx]; 850 v = aa + 36*ai[idx]; 851 for ( j=0; j<nz; j++ ) { 852 x = rtmp+36*ic[ajtmpold[j]]; 853 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 854 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 855 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 856 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 857 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 858 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 859 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 860 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 861 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 862 v += 36; 863 } 864 row = *ajtmp++; 865 while (row < i) { 866 pc = rtmp + 36*row; 867 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 868 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 869 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 870 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 871 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 872 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 873 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 874 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 875 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 876 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 877 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 878 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 879 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 880 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 881 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 882 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 883 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 884 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 885 pv = ba + 36*diag_offset[row]; 886 pj = bj + diag_offset[row] + 1; 887 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 888 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 889 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 890 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 891 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 892 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 893 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 894 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 895 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 896 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 897 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 898 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 899 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 900 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 901 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 902 903 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 904 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 905 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 906 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 907 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 908 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 909 910 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 911 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 912 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 913 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 914 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 915 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 916 917 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 918 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 919 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 920 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 921 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 922 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 923 924 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 925 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 926 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 927 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 928 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 929 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 930 931 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 932 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 933 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 934 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 935 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 936 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 937 938 nz = bi[row+1] - diag_offset[row] - 1; 939 pv += 36; 940 for (j=0; j<nz; j++) { 941 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 942 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 943 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 944 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 945 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 946 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 947 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 948 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 949 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 950 x = rtmp + 36*pj[j]; 951 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 952 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 953 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 954 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 955 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 956 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 957 958 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 959 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 960 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 961 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 962 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 963 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 964 965 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 966 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 967 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 968 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 969 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 970 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 971 972 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 973 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 974 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 975 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 976 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 977 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 978 979 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 980 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 981 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 982 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 983 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 984 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 985 986 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 987 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 988 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 989 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 990 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 991 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 992 993 pv += 36; 994 } 995 PLogFlops(432*nz+396); 996 } 997 row = *ajtmp++; 998 } 999 /* finished row so stick it into b->a */ 1000 pv = ba + 36*bi[i]; 1001 pj = bj + bi[i]; 1002 nz = bi[i+1] - bi[i]; 1003 for ( j=0; j<nz; j++ ) { 1004 x = rtmp+36*pj[j]; 1005 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1006 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1007 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1008 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1009 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1010 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1011 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1012 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1013 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1014 pv += 36; 1015 } 1016 /* invert diagonal block */ 1017 w = ba + 36*diag_offset[i]; 1018 ierr = Kernel_A_gets_inverse_A_6(w); CHKERRQ(ierr); 1019 } 1020 1021 PetscFree(rtmp); 1022 ierr = ISRestoreIndices(isicol,&ic); CHKERRQ(ierr); 1023 ierr = ISRestoreIndices(isrow,&r); CHKERRQ(ierr); 1024 C->factor = FACTOR_LU; 1025 C->assembled = PETSC_TRUE; 1026 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1027 PetscFunctionReturn(0); 1028 } 1029 /* 1030 Version for when blocks are 6 by 6 Using natural ordering 1031 */ 1032 #undef __FUNC__ 1033 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_6_NaturalOrdering" 1034 int MatLUFactorNumeric_SeqBAIJ_6_NaturalOrdering(Mat A,Mat *B) 1035 { 1036 Mat C = *B; 1037 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1038 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1039 int *ajtmpold, *ajtmp, nz, row; 1040 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 1041 register int *pj; 1042 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1043 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1044 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1045 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1046 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1047 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1048 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1049 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 1050 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 1051 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 1052 MatScalar *ba = b->a,*aa = a->a; 1053 1054 PetscFunctionBegin; 1055 rtmp = (MatScalar *) PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1056 for ( i=0; i<n; i++ ) { 1057 nz = bi[i+1] - bi[i]; 1058 ajtmp = bj + bi[i]; 1059 for ( j=0; j<nz; j++ ) { 1060 x = rtmp+36*ajtmp[j]; 1061 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1062 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1063 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 1064 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 1065 x[34] = x[35] = 0.0 ; 1066 } 1067 /* load in initial (unfactored row) */ 1068 nz = ai[i+1] - ai[i]; 1069 ajtmpold = aj + ai[i]; 1070 v = aa + 36*ai[i]; 1071 for ( j=0; j<nz; j++ ) { 1072 x = rtmp+36*ajtmpold[j]; 1073 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1074 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 1075 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 1076 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 1077 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 1078 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1079 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 1080 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 1081 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 1082 v += 36; 1083 } 1084 row = *ajtmp++; 1085 while (row < i) { 1086 pc = rtmp + 36*row; 1087 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1088 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 1089 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 1090 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 1091 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 1092 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1093 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 1094 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 1095 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 1096 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 1097 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 1098 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 1099 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 1100 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 1101 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 1102 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 1103 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 1104 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 1105 pv = ba + 36*diag_offset[row]; 1106 pj = bj + diag_offset[row] + 1; 1107 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1108 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1109 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1110 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1111 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1112 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1113 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1114 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1115 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1116 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 1117 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 1118 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 1119 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 1120 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 1121 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 1122 1123 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 1124 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 1125 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 1126 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 1127 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 1128 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 1129 1130 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 1131 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 1132 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 1133 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 1134 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 1135 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 1136 1137 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 1138 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 1139 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 1140 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 1141 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 1142 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 1143 1144 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 1145 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 1146 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 1147 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 1148 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 1149 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 1150 1151 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 1152 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 1153 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 1154 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 1155 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 1156 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 1157 1158 nz = bi[row+1] - diag_offset[row] - 1; 1159 pv += 36; 1160 for (j=0; j<nz; j++) { 1161 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1162 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1163 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1164 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1165 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1166 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1167 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1168 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1169 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1170 x = rtmp + 36*pj[j]; 1171 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 1172 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 1173 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 1174 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 1175 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 1176 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 1177 1178 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 1179 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 1180 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 1181 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 1182 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 1183 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 1184 1185 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 1186 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 1187 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 1188 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 1189 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 1190 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 1191 1192 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 1193 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 1194 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 1195 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 1196 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 1197 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 1198 1199 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 1200 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 1201 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 1202 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 1203 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 1204 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 1205 1206 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 1207 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 1208 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 1209 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 1210 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 1211 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 1212 1213 pv += 36; 1214 } 1215 PLogFlops(432*nz+396); 1216 } 1217 row = *ajtmp++; 1218 } 1219 /* finished row so stick it into b->a */ 1220 pv = ba + 36*bi[i]; 1221 pj = bj + bi[i]; 1222 nz = bi[i+1] - bi[i]; 1223 for ( j=0; j<nz; j++ ) { 1224 x = rtmp+36*pj[j]; 1225 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1226 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1227 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1228 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1229 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1230 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1231 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1232 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1233 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1234 pv += 36; 1235 } 1236 /* invert diagonal block */ 1237 w = ba + 36*diag_offset[i]; 1238 ierr = Kernel_A_gets_inverse_A_6(w); CHKERRQ(ierr); 1239 } 1240 1241 PetscFree(rtmp); 1242 C->factor = FACTOR_LU; 1243 C->assembled = PETSC_TRUE; 1244 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1245 PetscFunctionReturn(0); 1246 } 1247 1248 /* ------------------------------------------------------------*/ 1249 /* 1250 Version for when blocks are 5 by 5 1251 */ 1252 #undef __FUNC__ 1253 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_5" 1254 int MatLUFactorNumeric_SeqBAIJ_5(Mat A,Mat *B) 1255 { 1256 Mat C = *B; 1257 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1258 IS isrow = b->row, isicol = b->icol; 1259 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1260 int *ajtmpold, *ajtmp, nz, row; 1261 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 1262 register int *pj; 1263 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1264 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1265 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1266 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 1267 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 1268 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1269 MatScalar *ba = b->a,*aa = a->a; 1270 1271 PetscFunctionBegin; 1272 ierr = ISGetIndices(isrow,&r); CHKERRQ(ierr); 1273 ierr = ISGetIndices(isicol,&ic); CHKERRQ(ierr); 1274 rtmp = (MatScalar *) PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1275 1276 for ( i=0; i<n; i++ ) { 1277 nz = bi[i+1] - bi[i]; 1278 ajtmp = bj + bi[i]; 1279 for ( j=0; j<nz; j++ ) { 1280 x = rtmp+25*ajtmp[j]; 1281 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1282 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1283 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1284 } 1285 /* load in initial (unfactored row) */ 1286 idx = r[i]; 1287 nz = ai[idx+1] - ai[idx]; 1288 ajtmpold = aj + ai[idx]; 1289 v = aa + 25*ai[idx]; 1290 for ( j=0; j<nz; j++ ) { 1291 x = rtmp+25*ic[ajtmpold[j]]; 1292 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1293 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1294 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1295 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; 1296 x[18] = v[18]; x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; 1297 x[22] = v[22]; x[23] = v[23]; x[24] = v[24]; 1298 v += 25; 1299 } 1300 row = *ajtmp++; 1301 while (row < i) { 1302 pc = rtmp + 25*row; 1303 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1304 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1305 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1306 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; 1307 p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1308 p25 = pc[24]; 1309 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1310 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1311 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1312 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || 1313 p20 != 0.0 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || 1314 p24 != 0.0 || p25 != 0.0) { 1315 pv = ba + 25*diag_offset[row]; 1316 pj = bj + diag_offset[row] + 1; 1317 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1318 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1319 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1320 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1321 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; 1322 x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1323 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1324 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1325 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1326 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1327 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1328 1329 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1330 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1331 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1332 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1333 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1334 1335 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1336 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1337 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1338 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1339 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1340 1341 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1342 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1343 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1344 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1345 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1346 1347 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1348 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1349 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1350 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1351 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1352 1353 nz = bi[row+1] - diag_offset[row] - 1; 1354 pv += 25; 1355 for (j=0; j<nz; j++) { 1356 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1357 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1358 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1359 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; 1360 x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; 1361 x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1362 x = rtmp + 25*pj[j]; 1363 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1364 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1365 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1366 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1367 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1368 1369 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1370 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1371 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1372 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1373 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1374 1375 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1376 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1377 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1378 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1379 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1380 1381 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1382 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1383 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1384 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1385 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1386 1387 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1388 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1389 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1390 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1391 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1392 1393 pv += 25; 1394 } 1395 PLogFlops(250*nz+225); 1396 } 1397 row = *ajtmp++; 1398 } 1399 /* finished row so stick it into b->a */ 1400 pv = ba + 25*bi[i]; 1401 pj = bj + bi[i]; 1402 nz = bi[i+1] - bi[i]; 1403 for ( j=0; j<nz; j++ ) { 1404 x = rtmp+25*pj[j]; 1405 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1406 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1407 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1408 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; 1409 pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; 1410 pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; pv[24] = x[24]; 1411 pv += 25; 1412 } 1413 /* invert diagonal block */ 1414 w = ba + 25*diag_offset[i]; 1415 ierr = Kernel_A_gets_inverse_A_5(w); CHKERRQ(ierr); 1416 } 1417 1418 PetscFree(rtmp); 1419 ierr = ISRestoreIndices(isicol,&ic); CHKERRQ(ierr); 1420 ierr = ISRestoreIndices(isrow,&r); CHKERRQ(ierr); 1421 C->factor = FACTOR_LU; 1422 C->assembled = PETSC_TRUE; 1423 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1424 PetscFunctionReturn(0); 1425 } 1426 /* 1427 Version for when blocks are 5 by 5 Using natural ordering 1428 */ 1429 #undef __FUNC__ 1430 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_5_NaturalOrdering" 1431 int MatLUFactorNumeric_SeqBAIJ_5_NaturalOrdering(Mat A,Mat *B) 1432 { 1433 Mat C = *B; 1434 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1435 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1436 int *ajtmpold, *ajtmp, nz, row; 1437 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 1438 register int *pj; 1439 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1440 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1441 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1442 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1443 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1444 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1445 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1446 MatScalar *ba = b->a,*aa = a->a; 1447 1448 PetscFunctionBegin; 1449 rtmp = (MatScalar *) PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1450 for ( i=0; i<n; i++ ) { 1451 nz = bi[i+1] - bi[i]; 1452 ajtmp = bj + bi[i]; 1453 for ( j=0; j<nz; j++ ) { 1454 x = rtmp+25*ajtmp[j]; 1455 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1456 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1457 x[16] = x[17] = x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1458 } 1459 /* load in initial (unfactored row) */ 1460 nz = ai[i+1] - ai[i]; 1461 ajtmpold = aj + ai[i]; 1462 v = aa + 25*ai[i]; 1463 for ( j=0; j<nz; j++ ) { 1464 x = rtmp+25*ajtmpold[j]; 1465 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1466 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1467 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1468 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; 1469 x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1470 x[24] = v[24]; 1471 v += 25; 1472 } 1473 row = *ajtmp++; 1474 while (row < i) { 1475 pc = rtmp + 25*row; 1476 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1477 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1478 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1479 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; 1480 p19 = pc[18]; p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; 1481 p24 = pc[23]; p25 = pc[24]; 1482 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1483 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1484 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1485 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 1486 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || p25 != 0.0) { 1487 pv = ba + 25*diag_offset[row]; 1488 pj = bj + diag_offset[row] + 1; 1489 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1490 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1491 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1492 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; 1493 x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1494 x25 = pv[24]; 1495 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1496 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1497 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1498 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1499 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1500 1501 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1502 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1503 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1504 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1505 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1506 1507 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1508 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1509 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1510 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1511 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1512 1513 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1514 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1515 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1516 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1517 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1518 1519 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1520 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1521 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1522 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1523 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1524 1525 nz = bi[row+1] - diag_offset[row] - 1; 1526 pv += 25; 1527 for (j=0; j<nz; j++) { 1528 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1529 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1530 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1531 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1532 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; 1533 x24 = pv[23]; x25 = pv[24]; 1534 x = rtmp + 25*pj[j]; 1535 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1536 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1537 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1538 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1539 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1540 1541 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1542 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1543 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1544 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1545 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1546 1547 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1548 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1549 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1550 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1551 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1552 1553 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1554 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1555 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1556 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1557 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1558 1559 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1560 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1561 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1562 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1563 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1564 pv += 25; 1565 } 1566 PLogFlops(250*nz+225); 1567 } 1568 row = *ajtmp++; 1569 } 1570 /* finished row so stick it into b->a */ 1571 pv = ba + 25*bi[i]; 1572 pj = bj + bi[i]; 1573 nz = bi[i+1] - bi[i]; 1574 for ( j=0; j<nz; j++ ) { 1575 x = rtmp+25*pj[j]; 1576 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1577 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1578 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1579 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; pv[17] = x[17]; 1580 pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; 1581 pv[23] = x[23]; pv[24] = x[24]; 1582 pv += 25; 1583 } 1584 /* invert diagonal block */ 1585 w = ba + 25*diag_offset[i]; 1586 ierr = Kernel_A_gets_inverse_A_5(w); CHKERRQ(ierr); 1587 } 1588 1589 PetscFree(rtmp); 1590 C->factor = FACTOR_LU; 1591 C->assembled = PETSC_TRUE; 1592 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1593 PetscFunctionReturn(0); 1594 } 1595 1596 /* ------------------------------------------------------------*/ 1597 /* 1598 Version for when blocks are 4 by 4 1599 */ 1600 #undef __FUNC__ 1601 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_4" 1602 int MatLUFactorNumeric_SeqBAIJ_4(Mat A,Mat *B) 1603 { 1604 Mat C = *B; 1605 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1606 IS isrow = b->row, isicol = b->icol; 1607 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1608 int *ajtmpold, *ajtmp, nz, row; 1609 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 1610 register int *pj; 1611 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1612 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1613 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1614 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1615 MatScalar m13,m14,m15,m16; 1616 MatScalar *ba = b->a,*aa = a->a; 1617 1618 PetscFunctionBegin; 1619 ierr = ISGetIndices(isrow,&r); CHKERRQ(ierr); 1620 ierr = ISGetIndices(isicol,&ic); CHKERRQ(ierr); 1621 rtmp = (MatScalar *) PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1622 1623 for ( i=0; i<n; i++ ) { 1624 nz = bi[i+1] - bi[i]; 1625 ajtmp = bj + bi[i]; 1626 for ( j=0; j<nz; j++ ) { 1627 x = rtmp+16*ajtmp[j]; 1628 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1629 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1630 } 1631 /* load in initial (unfactored row) */ 1632 idx = r[i]; 1633 nz = ai[idx+1] - ai[idx]; 1634 ajtmpold = aj + ai[idx]; 1635 v = aa + 16*ai[idx]; 1636 for ( j=0; j<nz; j++ ) { 1637 x = rtmp+16*ic[ajtmpold[j]]; 1638 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1639 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1640 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1641 x[14] = v[14]; x[15] = v[15]; 1642 v += 16; 1643 } 1644 row = *ajtmp++; 1645 while (row < i) { 1646 pc = rtmp + 16*row; 1647 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1648 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1649 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1650 p15 = pc[14]; p16 = pc[15]; 1651 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1652 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1653 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1654 || p16 != 0.0) { 1655 pv = ba + 16*diag_offset[row]; 1656 pj = bj + diag_offset[row] + 1; 1657 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1658 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1659 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1660 x15 = pv[14]; x16 = pv[15]; 1661 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1662 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1663 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1664 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1665 1666 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1667 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1668 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1669 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1670 1671 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1672 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1673 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1674 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1675 1676 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1677 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1678 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1679 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1680 1681 nz = bi[row+1] - diag_offset[row] - 1; 1682 pv += 16; 1683 for (j=0; j<nz; j++) { 1684 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1685 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1686 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1687 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1688 x = rtmp + 16*pj[j]; 1689 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1690 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1691 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1692 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1693 1694 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1695 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1696 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1697 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1698 1699 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1700 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1701 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1702 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1703 1704 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1705 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1706 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1707 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1708 1709 pv += 16; 1710 } 1711 PLogFlops(128*nz+112); 1712 } 1713 row = *ajtmp++; 1714 } 1715 /* finished row so stick it into b->a */ 1716 pv = ba + 16*bi[i]; 1717 pj = bj + bi[i]; 1718 nz = bi[i+1] - bi[i]; 1719 for ( j=0; j<nz; j++ ) { 1720 x = rtmp+16*pj[j]; 1721 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1722 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1723 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1724 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1725 pv += 16; 1726 } 1727 /* invert diagonal block */ 1728 w = ba + 16*diag_offset[i]; 1729 ierr = Kernel_A_gets_inverse_A_4(w); CHKERRQ(ierr); 1730 } 1731 1732 PetscFree(rtmp); 1733 ierr = ISRestoreIndices(isicol,&ic); CHKERRQ(ierr); 1734 ierr = ISRestoreIndices(isrow,&r); CHKERRQ(ierr); 1735 C->factor = FACTOR_LU; 1736 C->assembled = PETSC_TRUE; 1737 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1738 PetscFunctionReturn(0); 1739 } 1740 /* 1741 Version for when blocks are 4 by 4 Using natural ordering 1742 */ 1743 #undef __FUNC__ 1744 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_4_NaturalOrdering" 1745 int MatLUFactorNumeric_SeqBAIJ_4_NaturalOrdering(Mat A,Mat *B) 1746 { 1747 Mat C = *B; 1748 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1749 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1750 int *ajtmpold, *ajtmp, nz, row; 1751 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 1752 register int *pj; 1753 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1754 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1755 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1756 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1757 MatScalar m13,m14,m15,m16; 1758 MatScalar *ba = b->a,*aa = a->a; 1759 1760 PetscFunctionBegin; 1761 rtmp = (MatScalar *) PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1762 1763 for ( i=0; i<n; i++ ) { 1764 nz = bi[i+1] - bi[i]; 1765 ajtmp = bj + bi[i]; 1766 for ( j=0; j<nz; j++ ) { 1767 x = rtmp+16*ajtmp[j]; 1768 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1769 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1770 } 1771 /* load in initial (unfactored row) */ 1772 nz = ai[i+1] - ai[i]; 1773 ajtmpold = aj + ai[i]; 1774 v = aa + 16*ai[i]; 1775 for ( j=0; j<nz; j++ ) { 1776 x = rtmp+16*ajtmpold[j]; 1777 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1778 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1779 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1780 x[14] = v[14]; x[15] = v[15]; 1781 v += 16; 1782 } 1783 row = *ajtmp++; 1784 while (row < i) { 1785 pc = rtmp + 16*row; 1786 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1787 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1788 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1789 p15 = pc[14]; p16 = pc[15]; 1790 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1791 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1792 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1793 || p16 != 0.0) { 1794 pv = ba + 16*diag_offset[row]; 1795 pj = bj + diag_offset[row] + 1; 1796 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1797 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1798 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1799 x15 = pv[14]; x16 = pv[15]; 1800 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1801 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1802 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1803 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1804 1805 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1806 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1807 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1808 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1809 1810 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1811 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1812 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1813 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1814 1815 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1816 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1817 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1818 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1819 1820 nz = bi[row+1] - diag_offset[row] - 1; 1821 pv += 16; 1822 for (j=0; j<nz; j++) { 1823 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1824 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1825 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1826 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1827 x = rtmp + 16*pj[j]; 1828 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1829 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1830 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1831 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1832 1833 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1834 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1835 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1836 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1837 1838 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1839 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1840 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1841 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1842 1843 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1844 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1845 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1846 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1847 1848 pv += 16; 1849 } 1850 PLogFlops(128*nz+112); 1851 } 1852 row = *ajtmp++; 1853 } 1854 /* finished row so stick it into b->a */ 1855 pv = ba + 16*bi[i]; 1856 pj = bj + bi[i]; 1857 nz = bi[i+1] - bi[i]; 1858 for ( j=0; j<nz; j++ ) { 1859 x = rtmp+16*pj[j]; 1860 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1861 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1862 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1863 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1864 pv += 16; 1865 } 1866 /* invert diagonal block */ 1867 w = ba + 16*diag_offset[i]; 1868 ierr = Kernel_A_gets_inverse_A_4(w); CHKERRQ(ierr); 1869 } 1870 1871 PetscFree(rtmp); 1872 C->factor = FACTOR_LU; 1873 C->assembled = PETSC_TRUE; 1874 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1875 PetscFunctionReturn(0); 1876 } 1877 1878 1879 /* ------------------------------------------------------------*/ 1880 /* 1881 Version for when blocks are 3 by 3 1882 */ 1883 #undef __FUNC__ 1884 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_3" 1885 int MatLUFactorNumeric_SeqBAIJ_3(Mat A,Mat *B) 1886 { 1887 Mat C = *B; 1888 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1889 IS isrow = b->row, isicol = b->icol; 1890 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1891 int *ajtmpold, *ajtmp, nz, row, *ai=a->i,*aj=a->j; 1892 int *diag_offset = b->diag,idx; 1893 register int *pj; 1894 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1895 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1896 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 1897 MatScalar *ba = b->a,*aa = a->a; 1898 1899 PetscFunctionBegin; 1900 ierr = ISGetIndices(isrow,&r); CHKERRQ(ierr); 1901 ierr = ISGetIndices(isicol,&ic); CHKERRQ(ierr); 1902 rtmp = (MatScalar *) PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1903 1904 for ( i=0; i<n; i++ ) { 1905 nz = bi[i+1] - bi[i]; 1906 ajtmp = bj + bi[i]; 1907 for ( j=0; j<nz; j++ ) { 1908 x = rtmp + 9*ajtmp[j]; 1909 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 1910 } 1911 /* load in initial (unfactored row) */ 1912 idx = r[i]; 1913 nz = ai[idx+1] - ai[idx]; 1914 ajtmpold = aj + ai[idx]; 1915 v = aa + 9*ai[idx]; 1916 for ( j=0; j<nz; j++ ) { 1917 x = rtmp + 9*ic[ajtmpold[j]]; 1918 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1919 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1920 v += 9; 1921 } 1922 row = *ajtmp++; 1923 while (row < i) { 1924 pc = rtmp + 9*row; 1925 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1926 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1927 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1928 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 1929 pv = ba + 9*diag_offset[row]; 1930 pj = bj + diag_offset[row] + 1; 1931 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1932 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1933 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 1934 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 1935 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 1936 1937 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 1938 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 1939 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 1940 1941 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 1942 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 1943 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 1944 nz = bi[row+1] - diag_offset[row] - 1; 1945 pv += 9; 1946 for (j=0; j<nz; j++) { 1947 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1948 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1949 x = rtmp + 9*pj[j]; 1950 x[0] -= m1*x1 + m4*x2 + m7*x3; 1951 x[1] -= m2*x1 + m5*x2 + m8*x3; 1952 x[2] -= m3*x1 + m6*x2 + m9*x3; 1953 1954 x[3] -= m1*x4 + m4*x5 + m7*x6; 1955 x[4] -= m2*x4 + m5*x5 + m8*x6; 1956 x[5] -= m3*x4 + m6*x5 + m9*x6; 1957 1958 x[6] -= m1*x7 + m4*x8 + m7*x9; 1959 x[7] -= m2*x7 + m5*x8 + m8*x9; 1960 x[8] -= m3*x7 + m6*x8 + m9*x9; 1961 pv += 9; 1962 } 1963 PLogFlops(54*nz+36); 1964 } 1965 row = *ajtmp++; 1966 } 1967 /* finished row so stick it into b->a */ 1968 pv = ba + 9*bi[i]; 1969 pj = bj + bi[i]; 1970 nz = bi[i+1] - bi[i]; 1971 for ( j=0; j<nz; j++ ) { 1972 x = rtmp + 9*pj[j]; 1973 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1974 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1975 pv += 9; 1976 } 1977 /* invert diagonal block */ 1978 w = ba + 9*diag_offset[i]; 1979 ierr = Kernel_A_gets_inverse_A_3(w); CHKERRQ(ierr); 1980 } 1981 1982 PetscFree(rtmp); 1983 ierr = ISRestoreIndices(isicol,&ic); CHKERRQ(ierr); 1984 ierr = ISRestoreIndices(isrow,&r); CHKERRQ(ierr); 1985 C->factor = FACTOR_LU; 1986 C->assembled = PETSC_TRUE; 1987 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 1988 PetscFunctionReturn(0); 1989 } 1990 /* 1991 Version for when blocks are 3 by 3 Using natural ordering 1992 */ 1993 #undef __FUNC__ 1994 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_3_NaturalOrdering" 1995 int MatLUFactorNumeric_SeqBAIJ_3_NaturalOrdering(Mat A,Mat *B) 1996 { 1997 Mat C = *B; 1998 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1999 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 2000 int *ajtmpold, *ajtmp, nz, row; 2001 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 2002 register int *pj; 2003 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2004 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 2005 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 2006 MatScalar *ba = b->a,*aa = a->a; 2007 2008 PetscFunctionBegin; 2009 rtmp = (MatScalar *) PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2010 2011 for ( i=0; i<n; i++ ) { 2012 nz = bi[i+1] - bi[i]; 2013 ajtmp = bj + bi[i]; 2014 for ( j=0; j<nz; j++ ) { 2015 x = rtmp+9*ajtmp[j]; 2016 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 2017 } 2018 /* load in initial (unfactored row) */ 2019 nz = ai[i+1] - ai[i]; 2020 ajtmpold = aj + ai[i]; 2021 v = aa + 9*ai[i]; 2022 for ( j=0; j<nz; j++ ) { 2023 x = rtmp+9*ajtmpold[j]; 2024 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2025 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 2026 v += 9; 2027 } 2028 row = *ajtmp++; 2029 while (row < i) { 2030 pc = rtmp + 9*row; 2031 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2032 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 2033 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 2034 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 2035 pv = ba + 9*diag_offset[row]; 2036 pj = bj + diag_offset[row] + 1; 2037 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2038 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2039 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 2040 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 2041 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 2042 2043 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 2044 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 2045 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 2046 2047 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 2048 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 2049 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 2050 2051 nz = bi[row+1] - diag_offset[row] - 1; 2052 pv += 9; 2053 for (j=0; j<nz; j++) { 2054 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2055 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2056 x = rtmp + 9*pj[j]; 2057 x[0] -= m1*x1 + m4*x2 + m7*x3; 2058 x[1] -= m2*x1 + m5*x2 + m8*x3; 2059 x[2] -= m3*x1 + m6*x2 + m9*x3; 2060 2061 x[3] -= m1*x4 + m4*x5 + m7*x6; 2062 x[4] -= m2*x4 + m5*x5 + m8*x6; 2063 x[5] -= m3*x4 + m6*x5 + m9*x6; 2064 2065 x[6] -= m1*x7 + m4*x8 + m7*x9; 2066 x[7] -= m2*x7 + m5*x8 + m8*x9; 2067 x[8] -= m3*x7 + m6*x8 + m9*x9; 2068 pv += 9; 2069 } 2070 PLogFlops(54*nz+36); 2071 } 2072 row = *ajtmp++; 2073 } 2074 /* finished row so stick it into b->a */ 2075 pv = ba + 9*bi[i]; 2076 pj = bj + bi[i]; 2077 nz = bi[i+1] - bi[i]; 2078 for ( j=0; j<nz; j++ ) { 2079 x = rtmp+9*pj[j]; 2080 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2081 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 2082 pv += 9; 2083 } 2084 /* invert diagonal block */ 2085 w = ba + 9*diag_offset[i]; 2086 ierr = Kernel_A_gets_inverse_A_3(w); CHKERRQ(ierr); 2087 } 2088 2089 PetscFree(rtmp); 2090 C->factor = FACTOR_LU; 2091 C->assembled = PETSC_TRUE; 2092 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 2093 PetscFunctionReturn(0); 2094 } 2095 2096 /* ------------------------------------------------------------*/ 2097 /* 2098 Version for when blocks are 2 by 2 2099 */ 2100 #undef __FUNC__ 2101 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_2" 2102 int MatLUFactorNumeric_SeqBAIJ_2(Mat A,Mat *B) 2103 { 2104 Mat C = *B; 2105 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 2106 IS isrow = b->row, isicol = b->icol; 2107 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 2108 int *ajtmpold, *ajtmp, nz, row; 2109 int *diag_offset=b->diag,idx,*ai=a->i,*aj=a->j; 2110 register int *pj; 2111 register MatScalar *pv,*v,*rtmp,m1,m2,m3,m4,*pc,*w,*x,x1,x2,x3,x4; 2112 MatScalar p1,p2,p3,p4; 2113 MatScalar *ba = b->a,*aa = a->a; 2114 2115 PetscFunctionBegin; 2116 ierr = ISGetIndices(isrow,&r); CHKERRQ(ierr); 2117 ierr = ISGetIndices(isicol,&ic); CHKERRQ(ierr); 2118 rtmp = (MatScalar *) PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2119 2120 for ( i=0; i<n; i++ ) { 2121 nz = bi[i+1] - bi[i]; 2122 ajtmp = bj + bi[i]; 2123 for ( j=0; j<nz; j++ ) { 2124 x = rtmp+4*ajtmp[j]; x[0] = x[1] = x[2] = x[3] = 0.0; 2125 } 2126 /* load in initial (unfactored row) */ 2127 idx = r[i]; 2128 nz = ai[idx+1] - ai[idx]; 2129 ajtmpold = aj + ai[idx]; 2130 v = aa + 4*ai[idx]; 2131 for ( j=0; j<nz; j++ ) { 2132 x = rtmp+4*ic[ajtmpold[j]]; 2133 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2134 v += 4; 2135 } 2136 row = *ajtmp++; 2137 while (row < i) { 2138 pc = rtmp + 4*row; 2139 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2140 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2141 pv = ba + 4*diag_offset[row]; 2142 pj = bj + diag_offset[row] + 1; 2143 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2144 pc[0] = m1 = p1*x1 + p3*x2; 2145 pc[1] = m2 = p2*x1 + p4*x2; 2146 pc[2] = m3 = p1*x3 + p3*x4; 2147 pc[3] = m4 = p2*x3 + p4*x4; 2148 nz = bi[row+1] - diag_offset[row] - 1; 2149 pv += 4; 2150 for (j=0; j<nz; j++) { 2151 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2152 x = rtmp + 4*pj[j]; 2153 x[0] -= m1*x1 + m3*x2; 2154 x[1] -= m2*x1 + m4*x2; 2155 x[2] -= m1*x3 + m3*x4; 2156 x[3] -= m2*x3 + m4*x4; 2157 pv += 4; 2158 } 2159 PLogFlops(16*nz+12); 2160 } 2161 row = *ajtmp++; 2162 } 2163 /* finished row so stick it into b->a */ 2164 pv = ba + 4*bi[i]; 2165 pj = bj + bi[i]; 2166 nz = bi[i+1] - bi[i]; 2167 for ( j=0; j<nz; j++ ) { 2168 x = rtmp+4*pj[j]; 2169 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2170 pv += 4; 2171 } 2172 /* invert diagonal block */ 2173 w = ba + 4*diag_offset[i]; 2174 ierr = Kernel_A_gets_inverse_A_2(w); CHKERRQ(ierr); 2175 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2176 } 2177 2178 PetscFree(rtmp); 2179 ierr = ISRestoreIndices(isicol,&ic); CHKERRQ(ierr); 2180 ierr = ISRestoreIndices(isrow,&r); CHKERRQ(ierr); 2181 C->factor = FACTOR_LU; 2182 C->assembled = PETSC_TRUE; 2183 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2184 PetscFunctionReturn(0); 2185 } 2186 /* 2187 Version for when blocks are 2 by 2 Using natural ordering 2188 */ 2189 #undef __FUNC__ 2190 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering" 2191 int MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering(Mat A,Mat *B) 2192 { 2193 Mat C = *B; 2194 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 2195 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 2196 int *ajtmpold, *ajtmp, nz, row; 2197 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 2198 register int *pj; 2199 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2200 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,x1,x2,x3,x4; 2201 MatScalar *ba = b->a,*aa = a->a; 2202 2203 PetscFunctionBegin; 2204 rtmp = (MatScalar *) PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2205 2206 for ( i=0; i<n; i++ ) { 2207 nz = bi[i+1] - bi[i]; 2208 ajtmp = bj + bi[i]; 2209 for ( j=0; j<nz; j++ ) { 2210 x = rtmp+4*ajtmp[j]; 2211 x[0] = x[1] = x[2] = x[3] = 0.0; 2212 } 2213 /* load in initial (unfactored row) */ 2214 nz = ai[i+1] - ai[i]; 2215 ajtmpold = aj + ai[i]; 2216 v = aa + 4*ai[i]; 2217 for ( j=0; j<nz; j++ ) { 2218 x = rtmp+4*ajtmpold[j]; 2219 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2220 v += 4; 2221 } 2222 row = *ajtmp++; 2223 while (row < i) { 2224 pc = rtmp + 4*row; 2225 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2226 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2227 pv = ba + 4*diag_offset[row]; 2228 pj = bj + diag_offset[row] + 1; 2229 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2230 pc[0] = m1 = p1*x1 + p3*x2; 2231 pc[1] = m2 = p2*x1 + p4*x2; 2232 pc[2] = m3 = p1*x3 + p3*x4; 2233 pc[3] = m4 = p2*x3 + p4*x4; 2234 nz = bi[row+1] - diag_offset[row] - 1; 2235 pv += 4; 2236 for (j=0; j<nz; j++) { 2237 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2238 x = rtmp + 4*pj[j]; 2239 x[0] -= m1*x1 + m3*x2; 2240 x[1] -= m2*x1 + m4*x2; 2241 x[2] -= m1*x3 + m3*x4; 2242 x[3] -= m2*x3 + m4*x4; 2243 pv += 4; 2244 } 2245 PLogFlops(16*nz+12); 2246 } 2247 row = *ajtmp++; 2248 } 2249 /* finished row so stick it into b->a */ 2250 pv = ba + 4*bi[i]; 2251 pj = bj + bi[i]; 2252 nz = bi[i+1] - bi[i]; 2253 for ( j=0; j<nz; j++ ) { 2254 x = rtmp+4*pj[j]; 2255 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2256 pv += 4; 2257 } 2258 /* invert diagonal block */ 2259 w = ba + 4*diag_offset[i]; 2260 ierr = Kernel_A_gets_inverse_A_2(w); CHKERRQ(ierr); 2261 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2262 } 2263 2264 PetscFree(rtmp); 2265 C->factor = FACTOR_LU; 2266 C->assembled = PETSC_TRUE; 2267 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2268 PetscFunctionReturn(0); 2269 } 2270 2271 /* ----------------------------------------------------------- */ 2272 /* 2273 Version for when blocks are 1 by 1. 2274 */ 2275 #undef __FUNC__ 2276 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_1" 2277 int MatLUFactorNumeric_SeqBAIJ_1(Mat A,Mat *B) 2278 { 2279 Mat C = *B; 2280 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data, *b = (Mat_SeqBAIJ *)C->data; 2281 IS isrow = b->row, isicol = b->icol; 2282 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 2283 int *ajtmpold, *ajtmp, nz, row,*ai = a->i,*aj = a->j; 2284 int *diag_offset = b->diag,diag; 2285 register int *pj; 2286 register MatScalar *pv,*v,*rtmp,multiplier,*pc; 2287 MatScalar *ba = b->a,*aa = a->a; 2288 2289 PetscFunctionBegin; 2290 ierr = ISGetIndices(isrow,&r); CHKERRQ(ierr); 2291 ierr = ISGetIndices(isicol,&ic); CHKERRQ(ierr); 2292 rtmp = (MatScalar *) PetscMalloc((n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2293 2294 for ( i=0; i<n; i++ ) { 2295 nz = bi[i+1] - bi[i]; 2296 ajtmp = bj + bi[i]; 2297 for ( j=0; j<nz; j++ ) rtmp[ajtmp[j]] = 0.0; 2298 2299 /* load in initial (unfactored row) */ 2300 nz = ai[r[i]+1] - ai[r[i]]; 2301 ajtmpold = aj + ai[r[i]]; 2302 v = aa + ai[r[i]]; 2303 for ( j=0; j<nz; j++ ) rtmp[ic[ajtmpold[j]]] = v[j]; 2304 2305 row = *ajtmp++; 2306 while (row < i) { 2307 pc = rtmp + row; 2308 if (*pc != 0.0) { 2309 pv = ba + diag_offset[row]; 2310 pj = bj + diag_offset[row] + 1; 2311 multiplier = *pc * *pv++; 2312 *pc = multiplier; 2313 nz = bi[row+1] - diag_offset[row] - 1; 2314 for (j=0; j<nz; j++) rtmp[pj[j]] -= multiplier * pv[j]; 2315 PLogFlops(1+2*nz); 2316 } 2317 row = *ajtmp++; 2318 } 2319 /* finished row so stick it into b->a */ 2320 pv = ba + bi[i]; 2321 pj = bj + bi[i]; 2322 nz = bi[i+1] - bi[i]; 2323 for ( j=0; j<nz; j++ ) {pv[j] = rtmp[pj[j]];} 2324 diag = diag_offset[i] - bi[i]; 2325 /* check pivot entry for current row */ 2326 if (pv[diag] == 0.0) { 2327 SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,0,"Zero pivot"); 2328 } 2329 pv[diag] = 1.0/pv[diag]; 2330 } 2331 2332 PetscFree(rtmp); 2333 ierr = ISRestoreIndices(isicol,&ic); CHKERRQ(ierr); 2334 ierr = ISRestoreIndices(isrow,&r); CHKERRQ(ierr); 2335 C->factor = FACTOR_LU; 2336 C->assembled = PETSC_TRUE; 2337 PLogFlops(b->n); 2338 PetscFunctionReturn(0); 2339 } 2340 2341 /* ----------------------------------------------------------- */ 2342 #undef __FUNC__ 2343 #define __FUNC__ "MatLUFactor_SeqBAIJ" 2344 int MatLUFactor_SeqBAIJ(Mat A,IS row,IS col,double f) 2345 { 2346 Mat_SeqBAIJ *mat = (Mat_SeqBAIJ *) A->data; 2347 int ierr; 2348 Mat C; 2349 PetscOps *Abops; 2350 MatOps Aops; 2351 2352 PetscFunctionBegin; 2353 ierr = MatLUFactorSymbolic(A,row,col,f,&C); CHKERRQ(ierr); 2354 ierr = MatLUFactorNumeric(A,&C); CHKERRQ(ierr); 2355 2356 /* free all the data structures from mat */ 2357 PetscFree(mat->a); 2358 if (!mat->singlemalloc) {PetscFree(mat->i); PetscFree(mat->j);} 2359 if (mat->diag) PetscFree(mat->diag); 2360 if (mat->ilen) PetscFree(mat->ilen); 2361 if (mat->imax) PetscFree(mat->imax); 2362 if (mat->solve_work) PetscFree(mat->solve_work); 2363 if (mat->mult_work) PetscFree(mat->mult_work); 2364 if (mat->icol) {ierr = ISDestroy(mat->icol);CHKERRQ(ierr);} 2365 PetscFree(mat); 2366 2367 ierr = MapDestroy(A->rmap);CHKERRQ(ierr); 2368 ierr = MapDestroy(A->cmap);CHKERRQ(ierr); 2369 2370 /* 2371 This is horrible, horrible code. We need to keep the 2372 A pointers for the bops and ops but copy everything 2373 else from C. 2374 */ 2375 Abops = A->bops; 2376 Aops = A->ops; 2377 PetscMemcpy(A,C,sizeof(struct _p_Mat)); 2378 mat = (Mat_SeqBAIJ *) A->data; 2379 PLogObjectParent(A,mat->icol); 2380 2381 A->bops = Abops; 2382 A->ops = Aops; 2383 A->qlist = 0; 2384 2385 PetscHeaderDestroy(C); 2386 PetscFunctionReturn(0); 2387 } 2388 2389 2390 2391