1 /*$Id: baijfact.c,v 1.73 1999/06/30 23:51:46 balay Exp bsmith $*/ 2 /* 3 Factorization code for BAIJ format. 4 */ 5 #include "src/mat/impls/baij/seq/baij.h" 6 #include "src/vec/vecimpl.h" 7 #include "src/inline/ilu.h" 8 9 /* 10 The symbolic factorization code is identical to that for AIJ format, 11 except for very small changes since this is now a SeqBAIJ datastructure. 12 NOT good code reuse. 13 */ 14 #undef __FUNC__ 15 #define __FUNC__ "MatLUFactorSymbolic_SeqBAIJ" 16 int MatLUFactorSymbolic_SeqBAIJ(Mat A,IS isrow,IS iscol,double f,Mat *B) 17 { 18 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data, *b; 19 IS isicol; 20 int *r,*ic, ierr, i, n = a->mbs, *ai = a->i, *aj = a->j; 21 int *ainew,*ajnew, jmax,*fill, *ajtmp, nz, bs = a->bs, bs2=a->bs2; 22 int *idnew, idx, row,m,fm, nnz, nzi,realloc = 0,nzbd,*im; 23 24 PetscFunctionBegin; 25 PetscValidHeaderSpecific(isrow,IS_COOKIE); 26 PetscValidHeaderSpecific(iscol,IS_COOKIE); 27 ierr = ISInvertPermutation(iscol,&isicol);CHKERRQ(ierr); 28 ISGetIndices(isrow,&r); ISGetIndices(isicol,&ic); 29 30 /* get new row pointers */ 31 ainew = (int *) PetscMalloc( (n+1)*sizeof(int) );CHKPTRQ(ainew); 32 ainew[0] = 0; 33 /* don't know how many column pointers are needed so estimate */ 34 jmax = (int) (f*ai[n] + 1); 35 ajnew = (int *) PetscMalloc( (jmax)*sizeof(int) );CHKPTRQ(ajnew); 36 /* fill is a linked list of nonzeros in active row */ 37 fill = (int *) PetscMalloc( (2*n+1)*sizeof(int));CHKPTRQ(fill); 38 im = fill + n + 1; 39 /* idnew is location of diagonal in factor */ 40 idnew = (int *) PetscMalloc( (n+1)*sizeof(int));CHKPTRQ(idnew); 41 idnew[0] = 0; 42 43 for ( i=0; i<n; i++ ) { 44 /* first copy previous fill into linked list */ 45 nnz = nz = ai[r[i]+1] - ai[r[i]]; 46 if (!nz) SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,1,"Empty row in matrix"); 47 ajtmp = aj + ai[r[i]]; 48 fill[n] = n; 49 while (nz--) { 50 fm = n; 51 idx = ic[*ajtmp++]; 52 do { 53 m = fm; 54 fm = fill[m]; 55 } while (fm < idx); 56 fill[m] = idx; 57 fill[idx] = fm; 58 } 59 row = fill[n]; 60 while ( row < i ) { 61 ajtmp = ajnew + idnew[row] + 1; 62 nzbd = 1 + idnew[row] - ainew[row]; 63 nz = im[row] - nzbd; 64 fm = row; 65 while (nz-- > 0) { 66 idx = *ajtmp++; 67 nzbd++; 68 if (idx == i) im[row] = nzbd; 69 do { 70 m = fm; 71 fm = fill[m]; 72 } while (fm < idx); 73 if (fm != idx) { 74 fill[m] = idx; 75 fill[idx] = fm; 76 fm = idx; 77 nnz++; 78 } 79 } 80 row = fill[row]; 81 } 82 /* copy new filled row into permanent storage */ 83 ainew[i+1] = ainew[i] + nnz; 84 if (ainew[i+1] > jmax) { 85 86 /* estimate how much additional space we will need */ 87 /* use the strategy suggested by David Hysom <hysom@perch-t.icase.edu> */ 88 /* just double the memory each time */ 89 int maxadd = jmax; 90 /* maxadd = (int) ((f*(ai[n]+1)*(n-i+5))/n); */ 91 if (maxadd < nnz) maxadd = (n-i)*(nnz+1); 92 jmax += maxadd; 93 94 /* allocate a longer ajnew */ 95 ajtmp = (int *) PetscMalloc( jmax*sizeof(int) );CHKPTRQ(ajtmp); 96 ierr = PetscMemcpy(ajtmp,ajnew,ainew[i]*sizeof(int));CHKERRQ(ierr); 97 ierr = PetscFree(ajnew);CHKERRQ(ierr); 98 ajnew = ajtmp; 99 realloc++; /* count how many times we realloc */ 100 } 101 ajtmp = ajnew + ainew[i]; 102 fm = fill[n]; 103 nzi = 0; 104 im[i] = nnz; 105 while (nnz--) { 106 if (fm < i) nzi++; 107 *ajtmp++ = fm; 108 fm = fill[fm]; 109 } 110 idnew[i] = ainew[i] + nzi; 111 } 112 113 if (ai[n] != 0) { 114 double af = ((double)ainew[n])/((double)ai[n]); 115 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:Reallocs %d Fill ratio:given %g needed %g\n", 116 realloc,f,af); 117 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:Run with -pc_lu_fill %g or use \n",af); 118 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:PCLUSetFill(pc,%g);\n",af); 119 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:for best performance.\n"); 120 } else { 121 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:Empty matrix.\n"); 122 } 123 124 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 125 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 126 127 ierr = PetscFree(fill);CHKERRQ(ierr); 128 129 /* put together the new matrix */ 130 ierr = MatCreateSeqBAIJ(A->comm,bs,bs*n,bs*n,0,PETSC_NULL,B);CHKERRQ(ierr); 131 PLogObjectParent(*B,isicol); 132 b = (Mat_SeqBAIJ *) (*B)->data; 133 ierr = PetscFree(b->imax);CHKERRQ(ierr); 134 b->singlemalloc = 0; 135 /* the next line frees the default space generated by the Create() */ 136 ierr = PetscFree(b->a);CHKERRQ(ierr); 137 ierr = PetscFree(b->ilen);CHKERRQ(ierr); 138 b->a = (MatScalar *) PetscMalloc((ainew[n]+1)*sizeof(MatScalar)*bs2);CHKPTRQ(b->a); 139 b->j = ajnew; 140 b->i = ainew; 141 b->diag = idnew; 142 b->ilen = 0; 143 b->imax = 0; 144 b->row = isrow; 145 b->col = iscol; 146 b->icol = isicol; 147 b->solve_work = (Scalar *) PetscMalloc( (bs*n+bs)*sizeof(Scalar));CHKPTRQ(b->solve_work); 148 /* In b structure: Free imax, ilen, old a, old j. 149 Allocate idnew, solve_work, new a, new j */ 150 PLogObjectMemory(*B,(ainew[n]-n)*(sizeof(int)+sizeof(MatScalar))); 151 b->maxnz = b->nz = ainew[n]; 152 153 (*B)->factor = FACTOR_LU; 154 (*B)->info.factor_mallocs = realloc; 155 (*B)->info.fill_ratio_given = f; 156 if (ai[n] != 0) { 157 (*B)->info.fill_ratio_needed = ((double)ainew[n])/((double)ai[n]); 158 } else { 159 (*B)->info.fill_ratio_needed = 0.0; 160 } 161 162 163 PetscFunctionReturn(0); 164 } 165 166 /* ----------------------------------------------------------- */ 167 #undef __FUNC__ 168 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_N" 169 int MatLUFactorNumeric_SeqBAIJ_N(Mat A,Mat *B) 170 { 171 Mat C = *B; 172 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 173 IS isrow = b->row, isicol = b->icol; 174 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 175 int *ajtmpold, *ajtmp, nz, row, bslog,*ai=a->i,*aj=a->j,k,flg; 176 int *diag_offset=b->diag,diag,bs=a->bs,bs2 = a->bs2,*v_pivots; 177 register int *pj; 178 register MatScalar *pv,*v,*rtmp,*multiplier,*v_work,*pc,*w; 179 MatScalar *ba = b->a,*aa = a->a; 180 181 PetscFunctionBegin; 182 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 183 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 184 rtmp = (MatScalar *) PetscMalloc(bs2*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 185 ierr = PetscMemzero(rtmp,bs2*(n+1)*sizeof(MatScalar));CHKERRQ(ierr); 186 /* generate work space needed by dense LU factorization */ 187 v_work = (MatScalar *) PetscMalloc(bs*sizeof(int) + (bs+bs2)*sizeof(MatScalar));CHKPTRQ(v_work); 188 multiplier = v_work + bs; 189 v_pivots = (int *) (multiplier + bs2); 190 191 /* flops in while loop */ 192 bslog = 2*bs*bs2; 193 194 for ( i=0; i<n; i++ ) { 195 nz = bi[i+1] - bi[i]; 196 ajtmp = bj + bi[i]; 197 for ( j=0; j<nz; j++ ) { 198 ierr = PetscMemzero(rtmp+bs2*ajtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 199 } 200 /* load in initial (unfactored row) */ 201 nz = ai[r[i]+1] - ai[r[i]]; 202 ajtmpold = aj + ai[r[i]]; 203 v = aa + bs2*ai[r[i]]; 204 for ( j=0; j<nz; j++ ) { 205 ierr = PetscMemcpy(rtmp+bs2*ic[ajtmpold[j]],v+bs2*j,bs2*sizeof(MatScalar));CHKERRQ(ierr); 206 } 207 row = *ajtmp++; 208 while (row < i) { 209 pc = rtmp + bs2*row; 210 /* if (*pc) { */ 211 for ( flg=0,k=0; k<bs2; k++ ) { if (pc[k]!=0.0) { flg =1; break; }} 212 if (flg) { 213 pv = ba + bs2*diag_offset[row]; 214 pj = bj + diag_offset[row] + 1; 215 Kernel_A_gets_A_times_B(bs,pc,pv,multiplier); 216 nz = bi[row+1] - diag_offset[row] - 1; 217 pv += bs2; 218 for (j=0; j<nz; j++) { 219 Kernel_A_gets_A_minus_B_times_C(bs,rtmp+bs2*pj[j],pc,pv+bs2*j); 220 } 221 PLogFlops(bslog*(nz+1)-bs); 222 } 223 row = *ajtmp++; 224 } 225 /* finished row so stick it into b->a */ 226 pv = ba + bs2*bi[i]; 227 pj = bj + bi[i]; 228 nz = bi[i+1] - bi[i]; 229 for ( j=0; j<nz; j++ ) { 230 ierr = PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 231 } 232 diag = diag_offset[i] - bi[i]; 233 /* invert diagonal block */ 234 w = pv + bs2*diag; 235 Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work); 236 } 237 238 ierr = PetscFree(rtmp);CHKERRQ(ierr); 239 ierr = PetscFree(v_work);CHKERRQ(ierr); 240 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 241 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 242 C->factor = FACTOR_LU; 243 C->assembled = PETSC_TRUE; 244 PLogFlops(1.3333*bs*bs2*b->mbs); /* from inverting diagonal blocks */ 245 PetscFunctionReturn(0); 246 } 247 /* ------------------------------------------------------------*/ 248 /* 249 Version for when blocks are 7 by 7 250 */ 251 #undef __FUNC__ 252 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_7" 253 int MatLUFactorNumeric_SeqBAIJ_7(Mat A,Mat *B) 254 { 255 Mat C = *B; 256 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 257 IS isrow = b->row, isicol = b->icol; 258 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 259 int *ajtmpold, *ajtmp, nz, row; 260 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 261 register int *pj; 262 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 263 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 264 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 265 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 266 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 267 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 268 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 269 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 270 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 271 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 272 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 273 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 274 MatScalar *ba = b->a,*aa = a->a; 275 276 PetscFunctionBegin; 277 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 278 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 279 rtmp = (MatScalar *) PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 280 281 for ( i=0; i<n; i++ ) { 282 nz = bi[i+1] - bi[i]; 283 ajtmp = bj + bi[i]; 284 for ( j=0; j<nz; j++ ) { 285 x = rtmp+49*ajtmp[j]; 286 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 287 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 288 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 289 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 290 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 291 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 292 } 293 /* load in initial (unfactored row) */ 294 idx = r[i]; 295 nz = ai[idx+1] - ai[idx]; 296 ajtmpold = aj + ai[idx]; 297 v = aa + 49*ai[idx]; 298 for ( j=0; j<nz; j++ ) { 299 x = rtmp+49*ic[ajtmpold[j]]; 300 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 301 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 302 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 303 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 304 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 305 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 306 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 307 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 308 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 309 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 310 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 311 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 312 x[48] = v[48]; 313 v += 49; 314 } 315 row = *ajtmp++; 316 while (row < i) { 317 pc = rtmp + 49*row; 318 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 319 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 320 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 321 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 322 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 323 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 324 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 325 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 326 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 327 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 328 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 329 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 330 p49 = pc[48]; 331 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 332 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 333 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 334 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 335 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 336 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 337 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 338 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 339 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 340 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 341 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 342 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 343 p49 != 0.0) { 344 pv = ba + 49*diag_offset[row]; 345 pj = bj + diag_offset[row] + 1; 346 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 347 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 348 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 349 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 350 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 351 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 352 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 353 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 354 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 355 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 356 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 357 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 358 x49 = pv[48]; 359 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 360 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 361 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 362 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 363 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 364 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 365 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 366 367 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 368 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 369 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 370 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 371 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 372 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 373 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 374 375 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 376 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 377 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 378 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 379 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 380 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 381 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 382 383 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 384 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 385 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 386 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 387 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 388 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 389 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 390 391 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 392 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 393 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 394 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 395 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 396 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 397 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 398 399 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 400 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 401 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 402 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 403 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 404 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 405 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 406 407 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 408 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 409 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 410 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 411 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 412 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 413 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 414 415 nz = bi[row+1] - diag_offset[row] - 1; 416 pv += 49; 417 for (j=0; j<nz; j++) { 418 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 419 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 420 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 421 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 422 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 423 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 424 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 425 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 426 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 427 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 428 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 429 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 430 x49 = pv[48]; 431 x = rtmp + 49*pj[j]; 432 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 433 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 434 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 435 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 436 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 437 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 438 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 439 440 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 441 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 442 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 443 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 444 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 445 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 446 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 447 448 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 449 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 450 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 451 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 452 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 453 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 454 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 455 456 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 457 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 458 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 459 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 460 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 461 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 462 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 463 464 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 465 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 466 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 467 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 468 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 469 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 470 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 471 472 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 473 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 474 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 475 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 476 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 477 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 478 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 479 480 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 481 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 482 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 483 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 484 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 485 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 486 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 487 pv += 49; 488 } 489 PLogFlops(686*nz+637); 490 } 491 row = *ajtmp++; 492 } 493 /* finished row so stick it into b->a */ 494 pv = ba + 49*bi[i]; 495 pj = bj + bi[i]; 496 nz = bi[i+1] - bi[i]; 497 for ( j=0; j<nz; j++ ) { 498 x = rtmp+49*pj[j]; 499 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 500 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 501 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 502 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 503 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 504 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 505 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 506 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 507 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 508 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 509 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 510 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 511 pv[48] = x[48]; 512 pv += 49; 513 } 514 /* invert diagonal block */ 515 w = ba + 49*diag_offset[i]; 516 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 517 } 518 519 ierr = PetscFree(rtmp);CHKERRQ(ierr); 520 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 521 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 522 C->factor = FACTOR_LU; 523 C->assembled = PETSC_TRUE; 524 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 525 PetscFunctionReturn(0); 526 } 527 528 /* 529 Version for when blocks are 7 by 7 Using natural ordering 530 */ 531 #undef __FUNC__ 532 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_7_NaturalOrdering" 533 int MatLUFactorNumeric_SeqBAIJ_7_NaturalOrdering(Mat A,Mat *B) 534 { 535 Mat C = *B; 536 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 537 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 538 int *ajtmpold, *ajtmp, nz, row; 539 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 540 register int *pj; 541 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 542 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 543 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 544 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 545 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 546 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 547 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 548 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 549 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 550 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 551 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 552 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 553 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 554 MatScalar *ba = b->a,*aa = a->a; 555 556 PetscFunctionBegin; 557 rtmp = (MatScalar *) PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 558 for ( i=0; i<n; i++ ) { 559 nz = bi[i+1] - bi[i]; 560 ajtmp = bj + bi[i]; 561 for ( j=0; j<nz; j++ ) { 562 x = rtmp+49*ajtmp[j]; 563 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 564 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 565 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 566 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 567 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 568 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 569 } 570 /* load in initial (unfactored row) */ 571 nz = ai[i+1] - ai[i]; 572 ajtmpold = aj + ai[i]; 573 v = aa + 49*ai[i]; 574 for ( j=0; j<nz; j++ ) { 575 x = rtmp+49*ajtmpold[j]; 576 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 577 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 578 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 579 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 580 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 581 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 582 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 583 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 584 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 585 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 586 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 587 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 588 x[48] = v[48]; 589 v += 49; 590 } 591 row = *ajtmp++; 592 while (row < i) { 593 pc = rtmp + 49*row; 594 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 595 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 596 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 597 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 598 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 599 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 600 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 601 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 602 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 603 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 604 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 605 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 606 p49 = pc[48]; 607 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 608 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 609 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 610 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 611 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 612 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 613 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 614 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 615 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 616 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 617 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 618 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 619 p49 != 0.0) { 620 pv = ba + 49*diag_offset[row]; 621 pj = bj + diag_offset[row] + 1; 622 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 623 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 624 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 625 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 626 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 627 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 628 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 629 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 630 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 631 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 632 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 633 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 634 x49 = pv[48]; 635 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 636 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 637 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 638 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 639 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 640 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 641 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 642 643 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 644 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 645 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 646 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 647 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 648 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 649 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 650 651 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 652 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 653 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 654 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 655 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 656 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 657 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 658 659 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 660 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 661 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 662 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 663 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 664 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 665 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 666 667 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 668 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 669 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 670 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 671 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 672 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 673 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 674 675 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 676 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 677 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 678 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 679 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 680 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 681 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 682 683 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 684 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 685 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 686 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 687 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 688 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 689 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 690 691 nz = bi[row+1] - diag_offset[row] - 1; 692 pv += 49; 693 for (j=0; j<nz; j++) { 694 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 695 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 696 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 697 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 698 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 699 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 700 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 701 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 702 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 703 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 704 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 705 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 706 x49 = pv[48]; 707 x = rtmp + 49*pj[j]; 708 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 709 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 710 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 711 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 712 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 713 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 714 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 715 716 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 717 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 718 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 719 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 720 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 721 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 722 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 723 724 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 725 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 726 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 727 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 728 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 729 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 730 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 731 732 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 733 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 734 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 735 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 736 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 737 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 738 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 739 740 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 741 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 742 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 743 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 744 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 745 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 746 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 747 748 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 749 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 750 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 751 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 752 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 753 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 754 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 755 756 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 757 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 758 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 759 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 760 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 761 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 762 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 763 pv += 49; 764 } 765 PLogFlops(686*nz+637); 766 } 767 row = *ajtmp++; 768 } 769 /* finished row so stick it into b->a */ 770 pv = ba + 49*bi[i]; 771 pj = bj + bi[i]; 772 nz = bi[i+1] - bi[i]; 773 for ( j=0; j<nz; j++ ) { 774 x = rtmp+49*pj[j]; 775 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 776 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 777 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 778 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 779 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 780 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 781 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 782 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 783 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 784 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 785 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 786 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 787 pv[48] = x[48]; 788 pv += 49; 789 } 790 /* invert diagonal block */ 791 w = ba + 49*diag_offset[i]; 792 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 793 } 794 795 ierr = PetscFree(rtmp);CHKERRQ(ierr); 796 C->factor = FACTOR_LU; 797 C->assembled = PETSC_TRUE; 798 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 799 PetscFunctionReturn(0); 800 } 801 802 /* ------------------------------------------------------------*/ 803 /* 804 Version for when blocks are 6 by 6 805 */ 806 #undef __FUNC__ 807 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_6" 808 int MatLUFactorNumeric_SeqBAIJ_6(Mat A,Mat *B) 809 { 810 Mat C = *B; 811 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 812 IS isrow = b->row, isicol = b->icol; 813 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 814 int *ajtmpold, *ajtmp, nz, row; 815 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 816 register int *pj; 817 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 818 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 819 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 820 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 821 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 822 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 823 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 824 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 825 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 826 MatScalar *ba = b->a,*aa = a->a; 827 828 PetscFunctionBegin; 829 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 830 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 831 rtmp = (MatScalar *) PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 832 833 for ( i=0; i<n; i++ ) { 834 nz = bi[i+1] - bi[i]; 835 ajtmp = bj + bi[i]; 836 for ( j=0; j<nz; j++ ) { 837 x = rtmp+36*ajtmp[j]; 838 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 839 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 840 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 841 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 842 x[34] = x[35] = 0.0 ; 843 } 844 /* load in initial (unfactored row) */ 845 idx = r[i]; 846 nz = ai[idx+1] - ai[idx]; 847 ajtmpold = aj + ai[idx]; 848 v = aa + 36*ai[idx]; 849 for ( j=0; j<nz; j++ ) { 850 x = rtmp+36*ic[ajtmpold[j]]; 851 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 852 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 853 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 854 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 855 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 856 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 857 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 858 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 859 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 860 v += 36; 861 } 862 row = *ajtmp++; 863 while (row < i) { 864 pc = rtmp + 36*row; 865 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 866 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 867 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 868 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 869 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 870 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 871 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 872 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 873 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 874 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 875 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 876 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 877 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 878 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 879 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 880 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 881 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 882 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 883 pv = ba + 36*diag_offset[row]; 884 pj = bj + diag_offset[row] + 1; 885 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 886 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 887 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 888 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 889 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 890 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 891 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 892 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 893 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 894 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 895 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 896 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 897 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 898 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 899 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 900 901 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 902 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 903 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 904 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 905 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 906 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 907 908 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 909 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 910 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 911 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 912 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 913 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 914 915 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 916 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 917 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 918 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 919 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 920 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 921 922 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 923 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 924 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 925 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 926 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 927 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 928 929 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 930 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 931 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 932 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 933 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 934 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 935 936 nz = bi[row+1] - diag_offset[row] - 1; 937 pv += 36; 938 for (j=0; j<nz; j++) { 939 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 940 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 941 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 942 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 943 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 944 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 945 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 946 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 947 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 948 x = rtmp + 36*pj[j]; 949 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 950 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 951 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 952 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 953 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 954 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 955 956 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 957 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 958 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 959 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 960 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 961 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 962 963 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 964 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 965 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 966 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 967 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 968 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 969 970 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 971 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 972 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 973 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 974 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 975 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 976 977 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 978 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 979 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 980 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 981 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 982 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 983 984 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 985 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 986 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 987 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 988 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 989 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 990 991 pv += 36; 992 } 993 PLogFlops(432*nz+396); 994 } 995 row = *ajtmp++; 996 } 997 /* finished row so stick it into b->a */ 998 pv = ba + 36*bi[i]; 999 pj = bj + bi[i]; 1000 nz = bi[i+1] - bi[i]; 1001 for ( j=0; j<nz; j++ ) { 1002 x = rtmp+36*pj[j]; 1003 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1004 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1005 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1006 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1007 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1008 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1009 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1010 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1011 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1012 pv += 36; 1013 } 1014 /* invert diagonal block */ 1015 w = ba + 36*diag_offset[i]; 1016 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1017 } 1018 1019 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1020 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1021 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1022 C->factor = FACTOR_LU; 1023 C->assembled = PETSC_TRUE; 1024 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1025 PetscFunctionReturn(0); 1026 } 1027 /* 1028 Version for when blocks are 6 by 6 Using natural ordering 1029 */ 1030 #undef __FUNC__ 1031 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_6_NaturalOrdering" 1032 int MatLUFactorNumeric_SeqBAIJ_6_NaturalOrdering(Mat A,Mat *B) 1033 { 1034 Mat C = *B; 1035 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1036 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1037 int *ajtmpold, *ajtmp, nz, row; 1038 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 1039 register int *pj; 1040 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1041 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1042 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1043 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1044 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1045 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1046 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1047 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 1048 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 1049 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 1050 MatScalar *ba = b->a,*aa = a->a; 1051 1052 PetscFunctionBegin; 1053 rtmp = (MatScalar *) PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1054 for ( i=0; i<n; i++ ) { 1055 nz = bi[i+1] - bi[i]; 1056 ajtmp = bj + bi[i]; 1057 for ( j=0; j<nz; j++ ) { 1058 x = rtmp+36*ajtmp[j]; 1059 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1060 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1061 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 1062 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 1063 x[34] = x[35] = 0.0 ; 1064 } 1065 /* load in initial (unfactored row) */ 1066 nz = ai[i+1] - ai[i]; 1067 ajtmpold = aj + ai[i]; 1068 v = aa + 36*ai[i]; 1069 for ( j=0; j<nz; j++ ) { 1070 x = rtmp+36*ajtmpold[j]; 1071 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1072 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 1073 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 1074 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 1075 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 1076 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1077 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 1078 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 1079 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 1080 v += 36; 1081 } 1082 row = *ajtmp++; 1083 while (row < i) { 1084 pc = rtmp + 36*row; 1085 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1086 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 1087 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 1088 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 1089 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 1090 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1091 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 1092 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 1093 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 1094 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 1095 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 1096 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 1097 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 1098 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 1099 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 1100 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 1101 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 1102 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 1103 pv = ba + 36*diag_offset[row]; 1104 pj = bj + diag_offset[row] + 1; 1105 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1106 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1107 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1108 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1109 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1110 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1111 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1112 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1113 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1114 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 1115 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 1116 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 1117 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 1118 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 1119 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 1120 1121 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 1122 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 1123 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 1124 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 1125 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 1126 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 1127 1128 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 1129 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 1130 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 1131 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 1132 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 1133 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 1134 1135 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 1136 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 1137 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 1138 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 1139 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 1140 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 1141 1142 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 1143 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 1144 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 1145 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 1146 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 1147 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 1148 1149 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 1150 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 1151 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 1152 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 1153 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 1154 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 1155 1156 nz = bi[row+1] - diag_offset[row] - 1; 1157 pv += 36; 1158 for (j=0; j<nz; j++) { 1159 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1160 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1161 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1162 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1163 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1164 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1165 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1166 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1167 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1168 x = rtmp + 36*pj[j]; 1169 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 1170 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 1171 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 1172 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 1173 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 1174 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 1175 1176 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 1177 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 1178 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 1179 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 1180 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 1181 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 1182 1183 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 1184 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 1185 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 1186 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 1187 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 1188 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 1189 1190 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 1191 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 1192 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 1193 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 1194 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 1195 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 1196 1197 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 1198 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 1199 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 1200 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 1201 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 1202 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 1203 1204 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 1205 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 1206 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 1207 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 1208 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 1209 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 1210 1211 pv += 36; 1212 } 1213 PLogFlops(432*nz+396); 1214 } 1215 row = *ajtmp++; 1216 } 1217 /* finished row so stick it into b->a */ 1218 pv = ba + 36*bi[i]; 1219 pj = bj + bi[i]; 1220 nz = bi[i+1] - bi[i]; 1221 for ( j=0; j<nz; j++ ) { 1222 x = rtmp+36*pj[j]; 1223 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1224 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1225 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1226 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1227 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1228 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1229 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1230 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1231 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1232 pv += 36; 1233 } 1234 /* invert diagonal block */ 1235 w = ba + 36*diag_offset[i]; 1236 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1237 } 1238 1239 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1240 C->factor = FACTOR_LU; 1241 C->assembled = PETSC_TRUE; 1242 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1243 PetscFunctionReturn(0); 1244 } 1245 1246 /* ------------------------------------------------------------*/ 1247 /* 1248 Version for when blocks are 5 by 5 1249 */ 1250 #undef __FUNC__ 1251 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_5" 1252 int MatLUFactorNumeric_SeqBAIJ_5(Mat A,Mat *B) 1253 { 1254 Mat C = *B; 1255 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1256 IS isrow = b->row, isicol = b->icol; 1257 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1258 int *ajtmpold, *ajtmp, nz, row; 1259 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 1260 register int *pj; 1261 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1262 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1263 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1264 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 1265 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 1266 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1267 MatScalar *ba = b->a,*aa = a->a; 1268 1269 PetscFunctionBegin; 1270 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1271 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1272 rtmp = (MatScalar *) PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1273 1274 for ( i=0; i<n; i++ ) { 1275 nz = bi[i+1] - bi[i]; 1276 ajtmp = bj + bi[i]; 1277 for ( j=0; j<nz; j++ ) { 1278 x = rtmp+25*ajtmp[j]; 1279 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1280 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1281 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1282 } 1283 /* load in initial (unfactored row) */ 1284 idx = r[i]; 1285 nz = ai[idx+1] - ai[idx]; 1286 ajtmpold = aj + ai[idx]; 1287 v = aa + 25*ai[idx]; 1288 for ( j=0; j<nz; j++ ) { 1289 x = rtmp+25*ic[ajtmpold[j]]; 1290 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1291 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1292 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1293 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; 1294 x[18] = v[18]; x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; 1295 x[22] = v[22]; x[23] = v[23]; x[24] = v[24]; 1296 v += 25; 1297 } 1298 row = *ajtmp++; 1299 while (row < i) { 1300 pc = rtmp + 25*row; 1301 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1302 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1303 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1304 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; 1305 p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1306 p25 = pc[24]; 1307 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1308 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1309 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1310 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || 1311 p20 != 0.0 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || 1312 p24 != 0.0 || p25 != 0.0) { 1313 pv = ba + 25*diag_offset[row]; 1314 pj = bj + diag_offset[row] + 1; 1315 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1316 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1317 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1318 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1319 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; 1320 x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1321 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1322 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1323 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1324 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1325 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1326 1327 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1328 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1329 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1330 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1331 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1332 1333 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1334 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1335 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1336 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1337 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1338 1339 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1340 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1341 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1342 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1343 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1344 1345 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1346 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1347 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1348 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1349 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1350 1351 nz = bi[row+1] - diag_offset[row] - 1; 1352 pv += 25; 1353 for (j=0; j<nz; j++) { 1354 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1355 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1356 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1357 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; 1358 x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; 1359 x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1360 x = rtmp + 25*pj[j]; 1361 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1362 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1363 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1364 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1365 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1366 1367 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1368 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1369 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1370 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1371 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1372 1373 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1374 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1375 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1376 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1377 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1378 1379 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1380 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1381 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1382 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1383 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1384 1385 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1386 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1387 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1388 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1389 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1390 1391 pv += 25; 1392 } 1393 PLogFlops(250*nz+225); 1394 } 1395 row = *ajtmp++; 1396 } 1397 /* finished row so stick it into b->a */ 1398 pv = ba + 25*bi[i]; 1399 pj = bj + bi[i]; 1400 nz = bi[i+1] - bi[i]; 1401 for ( j=0; j<nz; j++ ) { 1402 x = rtmp+25*pj[j]; 1403 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1404 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1405 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1406 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; 1407 pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; 1408 pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; pv[24] = x[24]; 1409 pv += 25; 1410 } 1411 /* invert diagonal block */ 1412 w = ba + 25*diag_offset[i]; 1413 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1414 } 1415 1416 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1417 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1418 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1419 C->factor = FACTOR_LU; 1420 C->assembled = PETSC_TRUE; 1421 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1422 PetscFunctionReturn(0); 1423 } 1424 /* 1425 Version for when blocks are 5 by 5 Using natural ordering 1426 */ 1427 #undef __FUNC__ 1428 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_5_NaturalOrdering" 1429 int MatLUFactorNumeric_SeqBAIJ_5_NaturalOrdering(Mat A,Mat *B) 1430 { 1431 Mat C = *B; 1432 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1433 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1434 int *ajtmpold, *ajtmp, nz, row; 1435 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 1436 register int *pj; 1437 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1438 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1439 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1440 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1441 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1442 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1443 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1444 MatScalar *ba = b->a,*aa = a->a; 1445 1446 PetscFunctionBegin; 1447 rtmp = (MatScalar *) PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1448 for ( i=0; i<n; i++ ) { 1449 nz = bi[i+1] - bi[i]; 1450 ajtmp = bj + bi[i]; 1451 for ( j=0; j<nz; j++ ) { 1452 x = rtmp+25*ajtmp[j]; 1453 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1454 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1455 x[16] = x[17] = x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1456 } 1457 /* load in initial (unfactored row) */ 1458 nz = ai[i+1] - ai[i]; 1459 ajtmpold = aj + ai[i]; 1460 v = aa + 25*ai[i]; 1461 for ( j=0; j<nz; j++ ) { 1462 x = rtmp+25*ajtmpold[j]; 1463 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1464 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1465 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1466 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; 1467 x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1468 x[24] = v[24]; 1469 v += 25; 1470 } 1471 row = *ajtmp++; 1472 while (row < i) { 1473 pc = rtmp + 25*row; 1474 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1475 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1476 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1477 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; 1478 p19 = pc[18]; p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; 1479 p24 = pc[23]; p25 = pc[24]; 1480 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1481 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1482 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1483 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 1484 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || p25 != 0.0) { 1485 pv = ba + 25*diag_offset[row]; 1486 pj = bj + diag_offset[row] + 1; 1487 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1488 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1489 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1490 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; 1491 x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1492 x25 = pv[24]; 1493 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1494 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1495 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1496 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1497 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1498 1499 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1500 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1501 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1502 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1503 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1504 1505 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1506 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1507 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1508 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1509 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1510 1511 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1512 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1513 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1514 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1515 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1516 1517 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1518 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1519 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1520 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1521 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1522 1523 nz = bi[row+1] - diag_offset[row] - 1; 1524 pv += 25; 1525 for (j=0; j<nz; j++) { 1526 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1527 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1528 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1529 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1530 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; 1531 x24 = pv[23]; x25 = pv[24]; 1532 x = rtmp + 25*pj[j]; 1533 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1534 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1535 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1536 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1537 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1538 1539 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1540 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1541 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1542 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1543 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1544 1545 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1546 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1547 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1548 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1549 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1550 1551 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1552 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1553 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1554 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1555 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1556 1557 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1558 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1559 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1560 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1561 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1562 pv += 25; 1563 } 1564 PLogFlops(250*nz+225); 1565 } 1566 row = *ajtmp++; 1567 } 1568 /* finished row so stick it into b->a */ 1569 pv = ba + 25*bi[i]; 1570 pj = bj + bi[i]; 1571 nz = bi[i+1] - bi[i]; 1572 for ( j=0; j<nz; j++ ) { 1573 x = rtmp+25*pj[j]; 1574 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1575 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1576 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1577 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; pv[17] = x[17]; 1578 pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; 1579 pv[23] = x[23]; pv[24] = x[24]; 1580 pv += 25; 1581 } 1582 /* invert diagonal block */ 1583 w = ba + 25*diag_offset[i]; 1584 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1585 } 1586 1587 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1588 C->factor = FACTOR_LU; 1589 C->assembled = PETSC_TRUE; 1590 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1591 PetscFunctionReturn(0); 1592 } 1593 1594 /* ------------------------------------------------------------*/ 1595 /* 1596 Version for when blocks are 4 by 4 1597 */ 1598 #undef __FUNC__ 1599 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_4" 1600 int MatLUFactorNumeric_SeqBAIJ_4(Mat A,Mat *B) 1601 { 1602 Mat C = *B; 1603 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1604 IS isrow = b->row, isicol = b->icol; 1605 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1606 int *ajtmpold, *ajtmp, nz, row; 1607 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j; 1608 register int *pj; 1609 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1610 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1611 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1612 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1613 MatScalar m13,m14,m15,m16; 1614 MatScalar *ba = b->a,*aa = a->a; 1615 1616 PetscFunctionBegin; 1617 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1618 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1619 rtmp = (MatScalar *) PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1620 1621 for ( i=0; i<n; i++ ) { 1622 nz = bi[i+1] - bi[i]; 1623 ajtmp = bj + bi[i]; 1624 for ( j=0; j<nz; j++ ) { 1625 x = rtmp+16*ajtmp[j]; 1626 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1627 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1628 } 1629 /* load in initial (unfactored row) */ 1630 idx = r[i]; 1631 nz = ai[idx+1] - ai[idx]; 1632 ajtmpold = aj + ai[idx]; 1633 v = aa + 16*ai[idx]; 1634 for ( j=0; j<nz; j++ ) { 1635 x = rtmp+16*ic[ajtmpold[j]]; 1636 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1637 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1638 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1639 x[14] = v[14]; x[15] = v[15]; 1640 v += 16; 1641 } 1642 row = *ajtmp++; 1643 while (row < i) { 1644 pc = rtmp + 16*row; 1645 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1646 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1647 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1648 p15 = pc[14]; p16 = pc[15]; 1649 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1650 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1651 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1652 || p16 != 0.0) { 1653 pv = ba + 16*diag_offset[row]; 1654 pj = bj + diag_offset[row] + 1; 1655 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1656 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1657 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1658 x15 = pv[14]; x16 = pv[15]; 1659 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1660 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1661 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1662 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1663 1664 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1665 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1666 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1667 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1668 1669 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1670 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1671 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1672 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1673 1674 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1675 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1676 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1677 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1678 1679 nz = bi[row+1] - diag_offset[row] - 1; 1680 pv += 16; 1681 for (j=0; j<nz; j++) { 1682 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1683 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1684 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1685 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1686 x = rtmp + 16*pj[j]; 1687 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1688 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1689 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1690 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1691 1692 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1693 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1694 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1695 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1696 1697 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1698 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1699 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1700 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1701 1702 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1703 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1704 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1705 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1706 1707 pv += 16; 1708 } 1709 PLogFlops(128*nz+112); 1710 } 1711 row = *ajtmp++; 1712 } 1713 /* finished row so stick it into b->a */ 1714 pv = ba + 16*bi[i]; 1715 pj = bj + bi[i]; 1716 nz = bi[i+1] - bi[i]; 1717 for ( j=0; j<nz; j++ ) { 1718 x = rtmp+16*pj[j]; 1719 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1720 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1721 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1722 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1723 pv += 16; 1724 } 1725 /* invert diagonal block */ 1726 w = ba + 16*diag_offset[i]; 1727 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1728 } 1729 1730 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1731 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1732 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1733 C->factor = FACTOR_LU; 1734 C->assembled = PETSC_TRUE; 1735 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1736 PetscFunctionReturn(0); 1737 } 1738 /* 1739 Version for when blocks are 4 by 4 Using natural ordering 1740 */ 1741 #undef __FUNC__ 1742 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_4_NaturalOrdering" 1743 int MatLUFactorNumeric_SeqBAIJ_4_NaturalOrdering(Mat A,Mat *B) 1744 { 1745 Mat C = *B; 1746 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1747 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1748 int *ajtmpold, *ajtmp, nz, row; 1749 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 1750 register int *pj; 1751 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1752 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1753 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1754 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1755 MatScalar m13,m14,m15,m16; 1756 MatScalar *ba = b->a,*aa = a->a; 1757 1758 PetscFunctionBegin; 1759 rtmp = (MatScalar *) PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1760 1761 for ( i=0; i<n; i++ ) { 1762 nz = bi[i+1] - bi[i]; 1763 ajtmp = bj + bi[i]; 1764 for ( j=0; j<nz; j++ ) { 1765 x = rtmp+16*ajtmp[j]; 1766 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1767 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1768 } 1769 /* load in initial (unfactored row) */ 1770 nz = ai[i+1] - ai[i]; 1771 ajtmpold = aj + ai[i]; 1772 v = aa + 16*ai[i]; 1773 for ( j=0; j<nz; j++ ) { 1774 x = rtmp+16*ajtmpold[j]; 1775 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1776 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1777 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1778 x[14] = v[14]; x[15] = v[15]; 1779 v += 16; 1780 } 1781 row = *ajtmp++; 1782 while (row < i) { 1783 pc = rtmp + 16*row; 1784 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1785 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1786 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1787 p15 = pc[14]; p16 = pc[15]; 1788 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1789 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1790 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1791 || p16 != 0.0) { 1792 pv = ba + 16*diag_offset[row]; 1793 pj = bj + diag_offset[row] + 1; 1794 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1795 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1796 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1797 x15 = pv[14]; x16 = pv[15]; 1798 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1799 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1800 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1801 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1802 1803 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1804 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1805 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1806 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1807 1808 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1809 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1810 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1811 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1812 1813 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1814 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1815 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1816 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1817 1818 nz = bi[row+1] - diag_offset[row] - 1; 1819 pv += 16; 1820 for (j=0; j<nz; j++) { 1821 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1822 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1823 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1824 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1825 x = rtmp + 16*pj[j]; 1826 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1827 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1828 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1829 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1830 1831 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1832 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1833 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1834 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1835 1836 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1837 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1838 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1839 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1840 1841 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1842 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1843 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1844 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1845 1846 pv += 16; 1847 } 1848 PLogFlops(128*nz+112); 1849 } 1850 row = *ajtmp++; 1851 } 1852 /* finished row so stick it into b->a */ 1853 pv = ba + 16*bi[i]; 1854 pj = bj + bi[i]; 1855 nz = bi[i+1] - bi[i]; 1856 for ( j=0; j<nz; j++ ) { 1857 x = rtmp+16*pj[j]; 1858 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1859 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1860 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1861 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1862 pv += 16; 1863 } 1864 /* invert diagonal block */ 1865 w = ba + 16*diag_offset[i]; 1866 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1867 } 1868 1869 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1870 C->factor = FACTOR_LU; 1871 C->assembled = PETSC_TRUE; 1872 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1873 PetscFunctionReturn(0); 1874 } 1875 1876 1877 /* ------------------------------------------------------------*/ 1878 /* 1879 Version for when blocks are 3 by 3 1880 */ 1881 #undef __FUNC__ 1882 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_3" 1883 int MatLUFactorNumeric_SeqBAIJ_3(Mat A,Mat *B) 1884 { 1885 Mat C = *B; 1886 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1887 IS isrow = b->row, isicol = b->icol; 1888 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1889 int *ajtmpold, *ajtmp, nz, row, *ai=a->i,*aj=a->j; 1890 int *diag_offset = b->diag,idx; 1891 register int *pj; 1892 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1893 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1894 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 1895 MatScalar *ba = b->a,*aa = a->a; 1896 1897 PetscFunctionBegin; 1898 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1899 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1900 rtmp = (MatScalar *) PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1901 1902 for ( i=0; i<n; i++ ) { 1903 nz = bi[i+1] - bi[i]; 1904 ajtmp = bj + bi[i]; 1905 for ( j=0; j<nz; j++ ) { 1906 x = rtmp + 9*ajtmp[j]; 1907 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 1908 } 1909 /* load in initial (unfactored row) */ 1910 idx = r[i]; 1911 nz = ai[idx+1] - ai[idx]; 1912 ajtmpold = aj + ai[idx]; 1913 v = aa + 9*ai[idx]; 1914 for ( j=0; j<nz; j++ ) { 1915 x = rtmp + 9*ic[ajtmpold[j]]; 1916 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1917 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1918 v += 9; 1919 } 1920 row = *ajtmp++; 1921 while (row < i) { 1922 pc = rtmp + 9*row; 1923 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1924 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1925 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1926 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 1927 pv = ba + 9*diag_offset[row]; 1928 pj = bj + diag_offset[row] + 1; 1929 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1930 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1931 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 1932 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 1933 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 1934 1935 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 1936 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 1937 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 1938 1939 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 1940 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 1941 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 1942 nz = bi[row+1] - diag_offset[row] - 1; 1943 pv += 9; 1944 for (j=0; j<nz; j++) { 1945 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1946 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1947 x = rtmp + 9*pj[j]; 1948 x[0] -= m1*x1 + m4*x2 + m7*x3; 1949 x[1] -= m2*x1 + m5*x2 + m8*x3; 1950 x[2] -= m3*x1 + m6*x2 + m9*x3; 1951 1952 x[3] -= m1*x4 + m4*x5 + m7*x6; 1953 x[4] -= m2*x4 + m5*x5 + m8*x6; 1954 x[5] -= m3*x4 + m6*x5 + m9*x6; 1955 1956 x[6] -= m1*x7 + m4*x8 + m7*x9; 1957 x[7] -= m2*x7 + m5*x8 + m8*x9; 1958 x[8] -= m3*x7 + m6*x8 + m9*x9; 1959 pv += 9; 1960 } 1961 PLogFlops(54*nz+36); 1962 } 1963 row = *ajtmp++; 1964 } 1965 /* finished row so stick it into b->a */ 1966 pv = ba + 9*bi[i]; 1967 pj = bj + bi[i]; 1968 nz = bi[i+1] - bi[i]; 1969 for ( j=0; j<nz; j++ ) { 1970 x = rtmp + 9*pj[j]; 1971 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1972 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1973 pv += 9; 1974 } 1975 /* invert diagonal block */ 1976 w = ba + 9*diag_offset[i]; 1977 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 1978 } 1979 1980 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1981 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1982 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1983 C->factor = FACTOR_LU; 1984 C->assembled = PETSC_TRUE; 1985 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 1986 PetscFunctionReturn(0); 1987 } 1988 /* 1989 Version for when blocks are 3 by 3 Using natural ordering 1990 */ 1991 #undef __FUNC__ 1992 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_3_NaturalOrdering" 1993 int MatLUFactorNumeric_SeqBAIJ_3_NaturalOrdering(Mat A,Mat *B) 1994 { 1995 Mat C = *B; 1996 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 1997 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 1998 int *ajtmpold, *ajtmp, nz, row; 1999 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 2000 register int *pj; 2001 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2002 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 2003 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 2004 MatScalar *ba = b->a,*aa = a->a; 2005 2006 PetscFunctionBegin; 2007 rtmp = (MatScalar *) PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2008 2009 for ( i=0; i<n; i++ ) { 2010 nz = bi[i+1] - bi[i]; 2011 ajtmp = bj + bi[i]; 2012 for ( j=0; j<nz; j++ ) { 2013 x = rtmp+9*ajtmp[j]; 2014 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 2015 } 2016 /* load in initial (unfactored row) */ 2017 nz = ai[i+1] - ai[i]; 2018 ajtmpold = aj + ai[i]; 2019 v = aa + 9*ai[i]; 2020 for ( j=0; j<nz; j++ ) { 2021 x = rtmp+9*ajtmpold[j]; 2022 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2023 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 2024 v += 9; 2025 } 2026 row = *ajtmp++; 2027 while (row < i) { 2028 pc = rtmp + 9*row; 2029 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2030 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 2031 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 2032 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 2033 pv = ba + 9*diag_offset[row]; 2034 pj = bj + diag_offset[row] + 1; 2035 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2036 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2037 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 2038 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 2039 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 2040 2041 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 2042 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 2043 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 2044 2045 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 2046 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 2047 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 2048 2049 nz = bi[row+1] - diag_offset[row] - 1; 2050 pv += 9; 2051 for (j=0; j<nz; j++) { 2052 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2053 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2054 x = rtmp + 9*pj[j]; 2055 x[0] -= m1*x1 + m4*x2 + m7*x3; 2056 x[1] -= m2*x1 + m5*x2 + m8*x3; 2057 x[2] -= m3*x1 + m6*x2 + m9*x3; 2058 2059 x[3] -= m1*x4 + m4*x5 + m7*x6; 2060 x[4] -= m2*x4 + m5*x5 + m8*x6; 2061 x[5] -= m3*x4 + m6*x5 + m9*x6; 2062 2063 x[6] -= m1*x7 + m4*x8 + m7*x9; 2064 x[7] -= m2*x7 + m5*x8 + m8*x9; 2065 x[8] -= m3*x7 + m6*x8 + m9*x9; 2066 pv += 9; 2067 } 2068 PLogFlops(54*nz+36); 2069 } 2070 row = *ajtmp++; 2071 } 2072 /* finished row so stick it into b->a */ 2073 pv = ba + 9*bi[i]; 2074 pj = bj + bi[i]; 2075 nz = bi[i+1] - bi[i]; 2076 for ( j=0; j<nz; j++ ) { 2077 x = rtmp+9*pj[j]; 2078 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2079 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 2080 pv += 9; 2081 } 2082 /* invert diagonal block */ 2083 w = ba + 9*diag_offset[i]; 2084 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 2085 } 2086 2087 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2088 C->factor = FACTOR_LU; 2089 C->assembled = PETSC_TRUE; 2090 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 2091 PetscFunctionReturn(0); 2092 } 2093 2094 /* ------------------------------------------------------------*/ 2095 /* 2096 Version for when blocks are 2 by 2 2097 */ 2098 #undef __FUNC__ 2099 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_2" 2100 int MatLUFactorNumeric_SeqBAIJ_2(Mat A,Mat *B) 2101 { 2102 Mat C = *B; 2103 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 2104 IS isrow = b->row, isicol = b->icol; 2105 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 2106 int *ajtmpold, *ajtmp, nz, row; 2107 int *diag_offset=b->diag,idx,*ai=a->i,*aj=a->j; 2108 register int *pj; 2109 register MatScalar *pv,*v,*rtmp,m1,m2,m3,m4,*pc,*w,*x,x1,x2,x3,x4; 2110 MatScalar p1,p2,p3,p4; 2111 MatScalar *ba = b->a,*aa = a->a; 2112 2113 PetscFunctionBegin; 2114 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 2115 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 2116 rtmp = (MatScalar *) PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2117 2118 for ( i=0; i<n; i++ ) { 2119 nz = bi[i+1] - bi[i]; 2120 ajtmp = bj + bi[i]; 2121 for ( j=0; j<nz; j++ ) { 2122 x = rtmp+4*ajtmp[j]; x[0] = x[1] = x[2] = x[3] = 0.0; 2123 } 2124 /* load in initial (unfactored row) */ 2125 idx = r[i]; 2126 nz = ai[idx+1] - ai[idx]; 2127 ajtmpold = aj + ai[idx]; 2128 v = aa + 4*ai[idx]; 2129 for ( j=0; j<nz; j++ ) { 2130 x = rtmp+4*ic[ajtmpold[j]]; 2131 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2132 v += 4; 2133 } 2134 row = *ajtmp++; 2135 while (row < i) { 2136 pc = rtmp + 4*row; 2137 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2138 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2139 pv = ba + 4*diag_offset[row]; 2140 pj = bj + diag_offset[row] + 1; 2141 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2142 pc[0] = m1 = p1*x1 + p3*x2; 2143 pc[1] = m2 = p2*x1 + p4*x2; 2144 pc[2] = m3 = p1*x3 + p3*x4; 2145 pc[3] = m4 = p2*x3 + p4*x4; 2146 nz = bi[row+1] - diag_offset[row] - 1; 2147 pv += 4; 2148 for (j=0; j<nz; j++) { 2149 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2150 x = rtmp + 4*pj[j]; 2151 x[0] -= m1*x1 + m3*x2; 2152 x[1] -= m2*x1 + m4*x2; 2153 x[2] -= m1*x3 + m3*x4; 2154 x[3] -= m2*x3 + m4*x4; 2155 pv += 4; 2156 } 2157 PLogFlops(16*nz+12); 2158 } 2159 row = *ajtmp++; 2160 } 2161 /* finished row so stick it into b->a */ 2162 pv = ba + 4*bi[i]; 2163 pj = bj + bi[i]; 2164 nz = bi[i+1] - bi[i]; 2165 for ( j=0; j<nz; j++ ) { 2166 x = rtmp+4*pj[j]; 2167 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2168 pv += 4; 2169 } 2170 /* invert diagonal block */ 2171 w = ba + 4*diag_offset[i]; 2172 ierr = Kernel_A_gets_inverse_A_2(w);CHKERRQ(ierr); 2173 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2174 } 2175 2176 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2177 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 2178 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 2179 C->factor = FACTOR_LU; 2180 C->assembled = PETSC_TRUE; 2181 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2182 PetscFunctionReturn(0); 2183 } 2184 /* 2185 Version for when blocks are 2 by 2 Using natural ordering 2186 */ 2187 #undef __FUNC__ 2188 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering" 2189 int MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering(Mat A,Mat *B) 2190 { 2191 Mat C = *B; 2192 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data,*b = (Mat_SeqBAIJ *)C->data; 2193 int ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 2194 int *ajtmpold, *ajtmp, nz, row; 2195 int *diag_offset = b->diag,*ai=a->i,*aj=a->j; 2196 register int *pj; 2197 register MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2198 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,x1,x2,x3,x4; 2199 MatScalar *ba = b->a,*aa = a->a; 2200 2201 PetscFunctionBegin; 2202 rtmp = (MatScalar *) PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2203 2204 for ( i=0; i<n; i++ ) { 2205 nz = bi[i+1] - bi[i]; 2206 ajtmp = bj + bi[i]; 2207 for ( j=0; j<nz; j++ ) { 2208 x = rtmp+4*ajtmp[j]; 2209 x[0] = x[1] = x[2] = x[3] = 0.0; 2210 } 2211 /* load in initial (unfactored row) */ 2212 nz = ai[i+1] - ai[i]; 2213 ajtmpold = aj + ai[i]; 2214 v = aa + 4*ai[i]; 2215 for ( j=0; j<nz; j++ ) { 2216 x = rtmp+4*ajtmpold[j]; 2217 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2218 v += 4; 2219 } 2220 row = *ajtmp++; 2221 while (row < i) { 2222 pc = rtmp + 4*row; 2223 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2224 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2225 pv = ba + 4*diag_offset[row]; 2226 pj = bj + diag_offset[row] + 1; 2227 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2228 pc[0] = m1 = p1*x1 + p3*x2; 2229 pc[1] = m2 = p2*x1 + p4*x2; 2230 pc[2] = m3 = p1*x3 + p3*x4; 2231 pc[3] = m4 = p2*x3 + p4*x4; 2232 nz = bi[row+1] - diag_offset[row] - 1; 2233 pv += 4; 2234 for (j=0; j<nz; j++) { 2235 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2236 x = rtmp + 4*pj[j]; 2237 x[0] -= m1*x1 + m3*x2; 2238 x[1] -= m2*x1 + m4*x2; 2239 x[2] -= m1*x3 + m3*x4; 2240 x[3] -= m2*x3 + m4*x4; 2241 pv += 4; 2242 } 2243 PLogFlops(16*nz+12); 2244 } 2245 row = *ajtmp++; 2246 } 2247 /* finished row so stick it into b->a */ 2248 pv = ba + 4*bi[i]; 2249 pj = bj + bi[i]; 2250 nz = bi[i+1] - bi[i]; 2251 for ( j=0; j<nz; j++ ) { 2252 x = rtmp+4*pj[j]; 2253 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2254 pv += 4; 2255 } 2256 /* invert diagonal block */ 2257 w = ba + 4*diag_offset[i]; 2258 ierr = Kernel_A_gets_inverse_A_2(w);CHKERRQ(ierr); 2259 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2260 } 2261 2262 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2263 C->factor = FACTOR_LU; 2264 C->assembled = PETSC_TRUE; 2265 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2266 PetscFunctionReturn(0); 2267 } 2268 2269 /* ----------------------------------------------------------- */ 2270 /* 2271 Version for when blocks are 1 by 1. 2272 */ 2273 #undef __FUNC__ 2274 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_1" 2275 int MatLUFactorNumeric_SeqBAIJ_1(Mat A,Mat *B) 2276 { 2277 Mat C = *B; 2278 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *) A->data, *b = (Mat_SeqBAIJ *)C->data; 2279 IS isrow = b->row, isicol = b->icol; 2280 int *r,*ic, ierr, i, j, n = a->mbs, *bi = b->i, *bj = b->j; 2281 int *ajtmpold, *ajtmp, nz, row,*ai = a->i,*aj = a->j; 2282 int *diag_offset = b->diag,diag; 2283 register int *pj; 2284 register MatScalar *pv,*v,*rtmp,multiplier,*pc; 2285 MatScalar *ba = b->a,*aa = a->a; 2286 2287 PetscFunctionBegin; 2288 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 2289 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 2290 rtmp = (MatScalar *) PetscMalloc((n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2291 2292 for ( i=0; i<n; i++ ) { 2293 nz = bi[i+1] - bi[i]; 2294 ajtmp = bj + bi[i]; 2295 for ( j=0; j<nz; j++ ) rtmp[ajtmp[j]] = 0.0; 2296 2297 /* load in initial (unfactored row) */ 2298 nz = ai[r[i]+1] - ai[r[i]]; 2299 ajtmpold = aj + ai[r[i]]; 2300 v = aa + ai[r[i]]; 2301 for ( j=0; j<nz; j++ ) rtmp[ic[ajtmpold[j]]] = v[j]; 2302 2303 row = *ajtmp++; 2304 while (row < i) { 2305 pc = rtmp + row; 2306 if (*pc != 0.0) { 2307 pv = ba + diag_offset[row]; 2308 pj = bj + diag_offset[row] + 1; 2309 multiplier = *pc * *pv++; 2310 *pc = multiplier; 2311 nz = bi[row+1] - diag_offset[row] - 1; 2312 for (j=0; j<nz; j++) rtmp[pj[j]] -= multiplier * pv[j]; 2313 PLogFlops(1+2*nz); 2314 } 2315 row = *ajtmp++; 2316 } 2317 /* finished row so stick it into b->a */ 2318 pv = ba + bi[i]; 2319 pj = bj + bi[i]; 2320 nz = bi[i+1] - bi[i]; 2321 for ( j=0; j<nz; j++ ) {pv[j] = rtmp[pj[j]];} 2322 diag = diag_offset[i] - bi[i]; 2323 /* check pivot entry for current row */ 2324 if (pv[diag] == 0.0) { 2325 SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,0,"Zero pivot"); 2326 } 2327 pv[diag] = 1.0/pv[diag]; 2328 } 2329 2330 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2331 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 2332 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 2333 C->factor = FACTOR_LU; 2334 C->assembled = PETSC_TRUE; 2335 PLogFlops(b->n); 2336 PetscFunctionReturn(0); 2337 } 2338 2339 /* ----------------------------------------------------------- */ 2340 #undef __FUNC__ 2341 #define __FUNC__ "MatLUFactor_SeqBAIJ" 2342 int MatLUFactor_SeqBAIJ(Mat A,IS row,IS col,double f) 2343 { 2344 Mat_SeqBAIJ *mat = (Mat_SeqBAIJ *) A->data; 2345 int ierr; 2346 Mat C; 2347 PetscOps *Abops; 2348 MatOps Aops; 2349 2350 PetscFunctionBegin; 2351 ierr = MatLUFactorSymbolic(A,row,col,f,&C);CHKERRQ(ierr); 2352 ierr = MatLUFactorNumeric(A,&C);CHKERRQ(ierr); 2353 2354 /* free all the data structures from mat */ 2355 ierr = PetscFree(mat->a);CHKERRQ(ierr); 2356 if (!mat->singlemalloc) { 2357 ierr = PetscFree(mat->i);CHKERRQ(ierr); 2358 ierr = PetscFree(mat->j);CHKERRQ(ierr); 2359 } 2360 if (mat->diag) {ierr = PetscFree(mat->diag);CHKERRQ(ierr);} 2361 if (mat->ilen) {ierr = PetscFree(mat->ilen);CHKERRQ(ierr);} 2362 if (mat->imax) {ierr = PetscFree(mat->imax);CHKERRQ(ierr);} 2363 if (mat->solve_work) {ierr = PetscFree(mat->solve_work);CHKERRQ(ierr);} 2364 if (mat->mult_work) {ierr = PetscFree(mat->mult_work);CHKERRQ(ierr);} 2365 if (mat->icol) {ierr = ISDestroy(mat->icol);CHKERRQ(ierr);} 2366 ierr = PetscFree(mat);CHKERRQ(ierr); 2367 2368 ierr = MapDestroy(A->rmap);CHKERRQ(ierr); 2369 ierr = MapDestroy(A->cmap);CHKERRQ(ierr); 2370 2371 /* 2372 This is horrible, horrible code. We need to keep the 2373 A pointers for the bops and ops but copy everything 2374 else from C. 2375 */ 2376 Abops = A->bops; 2377 Aops = A->ops; 2378 ierr = PetscMemcpy(A,C,sizeof(struct _p_Mat));CHKERRQ(ierr); 2379 mat = (Mat_SeqBAIJ *) A->data; 2380 PLogObjectParent(A,mat->icol); 2381 2382 A->bops = Abops; 2383 A->ops = Aops; 2384 A->qlist = 0; 2385 2386 PetscHeaderDestroy(C); 2387 PetscFunctionReturn(0); 2388 } 2389 2390 2391 2392