1 /*$Id: baijfact.c,v 1.78 2000/01/11 21:00:52 bsmith Exp bsmith $*/ 2 /* 3 Factorization code for BAIJ format. 4 */ 5 #include "src/mat/impls/baij/seq/baij.h" 6 #include "src/vec/vecimpl.h" 7 #include "src/inline/ilu.h" 8 9 /* 10 The symbolic factorization code is identical to that for AIJ format, 11 except for very small changes since this is now a SeqBAIJ datastructure. 12 NOT good code reuse. 13 */ 14 #undef __FUNC__ 15 #define __FUNC__ "MatLUFactorSymbolic_SeqBAIJ" 16 int MatLUFactorSymbolic_SeqBAIJ(Mat A,IS isrow,IS iscol,double f,Mat *B) 17 { 18 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b; 19 IS isicol; 20 int *r,*ic,ierr,i,n = a->mbs,*ai = a->i,*aj = a->j; 21 int *ainew,*ajnew,jmax,*fill,*ajtmp,nz,bs = a->bs,bs2=a->bs2; 22 int *idnew,idx,row,m,fm,nnz,nzi,realloc = 0,nzbd,*im; 23 24 PetscFunctionBegin; 25 PetscValidHeaderSpecific(isrow,IS_COOKIE); 26 PetscValidHeaderSpecific(iscol,IS_COOKIE); 27 if (A->M != A->N) SETERRQ(PETSC_ERR_ARG_WRONG,0,"matrix must be square"); 28 ierr = ISInvertPermutation(iscol,PETSC_DECIDE,&isicol);CHKERRQ(ierr); 29 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 30 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 31 32 /* get new row pointers */ 33 ainew = (int*)PetscMalloc((n+1)*sizeof(int));CHKPTRQ(ainew); 34 ainew[0] = 0; 35 /* don't know how many column pointers are needed so estimate */ 36 jmax = (int)(f*ai[n] + 1); 37 ajnew = (int*)PetscMalloc((jmax)*sizeof(int));CHKPTRQ(ajnew); 38 /* fill is a linked list of nonzeros in active row */ 39 fill = (int*)PetscMalloc((2*n+1)*sizeof(int));CHKPTRQ(fill); 40 im = fill + n + 1; 41 /* idnew is location of diagonal in factor */ 42 idnew = (int*)PetscMalloc((n+1)*sizeof(int));CHKPTRQ(idnew); 43 idnew[0] = 0; 44 45 for (i=0; i<n; i++) { 46 /* first copy previous fill into linked list */ 47 nnz = nz = ai[r[i]+1] - ai[r[i]]; 48 if (!nz) SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,1,"Empty row in matrix"); 49 ajtmp = aj + ai[r[i]]; 50 fill[n] = n; 51 while (nz--) { 52 fm = n; 53 idx = ic[*ajtmp++]; 54 do { 55 m = fm; 56 fm = fill[m]; 57 } while (fm < idx); 58 fill[m] = idx; 59 fill[idx] = fm; 60 } 61 row = fill[n]; 62 while (row < i) { 63 ajtmp = ajnew + idnew[row] + 1; 64 nzbd = 1 + idnew[row] - ainew[row]; 65 nz = im[row] - nzbd; 66 fm = row; 67 while (nz-- > 0) { 68 idx = *ajtmp++; 69 nzbd++; 70 if (idx == i) im[row] = nzbd; 71 do { 72 m = fm; 73 fm = fill[m]; 74 } while (fm < idx); 75 if (fm != idx) { 76 fill[m] = idx; 77 fill[idx] = fm; 78 fm = idx; 79 nnz++; 80 } 81 } 82 row = fill[row]; 83 } 84 /* copy new filled row into permanent storage */ 85 ainew[i+1] = ainew[i] + nnz; 86 if (ainew[i+1] > jmax) { 87 88 /* estimate how much additional space we will need */ 89 /* use the strategy suggested by David Hysom <hysom@perch-t.icase.edu> */ 90 /* just double the memory each time */ 91 int maxadd = jmax; 92 /* maxadd = (int)((f*(ai[n]+1)*(n-i+5))/n); */ 93 if (maxadd < nnz) maxadd = (n-i)*(nnz+1); 94 jmax += maxadd; 95 96 /* allocate a longer ajnew */ 97 ajtmp = (int*)PetscMalloc(jmax*sizeof(int));CHKPTRQ(ajtmp); 98 ierr = PetscMemcpy(ajtmp,ajnew,ainew[i]*sizeof(int));CHKERRQ(ierr); 99 ierr = PetscFree(ajnew);CHKERRQ(ierr); 100 ajnew = ajtmp; 101 realloc++; /* count how many times we realloc */ 102 } 103 ajtmp = ajnew + ainew[i]; 104 fm = fill[n]; 105 nzi = 0; 106 im[i] = nnz; 107 while (nnz--) { 108 if (fm < i) nzi++; 109 *ajtmp++ = fm; 110 fm = fill[fm]; 111 } 112 idnew[i] = ainew[i] + nzi; 113 } 114 115 if (ai[n] != 0) { 116 PetscReal af = ((PetscReal)ainew[n])/((PetscReal)ai[n]); 117 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:Reallocs %d Fill ratio:given %g needed %g\n",realloc,f,af); 118 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:Run with -pc_lu_fill %g or use \n",af); 119 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:PCLUSetFill(pc,%g);\n",af); 120 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:for best performance.\n"); 121 } else { 122 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:Empty matrix.\n"); 123 } 124 125 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 126 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 127 128 ierr = PetscFree(fill);CHKERRQ(ierr); 129 130 /* put together the new matrix */ 131 ierr = MatCreateSeqBAIJ(A->comm,bs,bs*n,bs*n,0,PETSC_NULL,B);CHKERRQ(ierr); 132 PLogObjectParent(*B,isicol); 133 b = (Mat_SeqBAIJ*)(*B)->data; 134 ierr = PetscFree(b->imax);CHKERRQ(ierr); 135 b->singlemalloc = PETSC_FALSE; 136 /* the next line frees the default space generated by the Create() */ 137 ierr = PetscFree(b->a);CHKERRQ(ierr); 138 ierr = PetscFree(b->ilen);CHKERRQ(ierr); 139 b->a = (MatScalar*)PetscMalloc((ainew[n]+1)*sizeof(MatScalar)*bs2);CHKPTRQ(b->a); 140 b->j = ajnew; 141 b->i = ainew; 142 b->diag = idnew; 143 b->ilen = 0; 144 b->imax = 0; 145 b->row = isrow; 146 b->col = iscol; 147 ierr = PetscObjectReference((PetscObject)isrow);CHKERRQ(ierr); 148 ierr = PetscObjectReference((PetscObject)iscol);CHKERRQ(ierr); 149 b->icol = isicol; 150 b->solve_work = (Scalar*)PetscMalloc((bs*n+bs)*sizeof(Scalar));CHKPTRQ(b->solve_work); 151 /* In b structure: Free imax, ilen, old a, old j. 152 Allocate idnew, solve_work, new a, new j */ 153 PLogObjectMemory(*B,(ainew[n]-n)*(sizeof(int)+sizeof(MatScalar))); 154 b->maxnz = b->nz = ainew[n]; 155 156 (*B)->factor = FACTOR_LU; 157 (*B)->info.factor_mallocs = realloc; 158 (*B)->info.fill_ratio_given = f; 159 if (ai[n] != 0) { 160 (*B)->info.fill_ratio_needed = ((PetscReal)ainew[n])/((PetscReal)ai[n]); 161 } else { 162 (*B)->info.fill_ratio_needed = 0.0; 163 } 164 165 166 PetscFunctionReturn(0); 167 } 168 169 /* ----------------------------------------------------------- */ 170 #undef __FUNC__ 171 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_N" 172 int MatLUFactorNumeric_SeqBAIJ_N(Mat A,Mat *B) 173 { 174 Mat C = *B; 175 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 176 IS isrow = b->row,isicol = b->icol; 177 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 178 int *ajtmpold,*ajtmp,nz,row,bslog,*ai=a->i,*aj=a->j,k,flg; 179 int *diag_offset=b->diag,diag,bs=a->bs,bs2 = a->bs2,*v_pivots,*pj; 180 MatScalar *ba = b->a,*aa = a->a,*pv,*v,*rtmp,*multiplier,*v_work,*pc,*w; 181 182 PetscFunctionBegin; 183 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 184 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 185 rtmp = (MatScalar*)PetscMalloc(bs2*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 186 ierr = PetscMemzero(rtmp,bs2*(n+1)*sizeof(MatScalar));CHKERRQ(ierr); 187 /* generate work space needed by dense LU factorization */ 188 v_work = (MatScalar*)PetscMalloc(bs*sizeof(int) + (bs+bs2)*sizeof(MatScalar));CHKPTRQ(v_work); 189 multiplier = v_work + bs; 190 v_pivots = (int*)(multiplier + bs2); 191 192 /* flops in while loop */ 193 bslog = 2*bs*bs2; 194 195 for (i=0; i<n; i++) { 196 nz = bi[i+1] - bi[i]; 197 ajtmp = bj + bi[i]; 198 for (j=0; j<nz; j++) { 199 ierr = PetscMemzero(rtmp+bs2*ajtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 200 } 201 /* load in initial (unfactored row) */ 202 nz = ai[r[i]+1] - ai[r[i]]; 203 ajtmpold = aj + ai[r[i]]; 204 v = aa + bs2*ai[r[i]]; 205 for (j=0; j<nz; j++) { 206 ierr = PetscMemcpy(rtmp+bs2*ic[ajtmpold[j]],v+bs2*j,bs2*sizeof(MatScalar));CHKERRQ(ierr); 207 } 208 row = *ajtmp++; 209 while (row < i) { 210 pc = rtmp + bs2*row; 211 /* if (*pc) { */ 212 for (flg=0,k=0; k<bs2; k++) { if (pc[k]!=0.0) { flg =1; break; }} 213 if (flg) { 214 pv = ba + bs2*diag_offset[row]; 215 pj = bj + diag_offset[row] + 1; 216 Kernel_A_gets_A_times_B(bs,pc,pv,multiplier); 217 nz = bi[row+1] - diag_offset[row] - 1; 218 pv += bs2; 219 for (j=0; j<nz; j++) { 220 Kernel_A_gets_A_minus_B_times_C(bs,rtmp+bs2*pj[j],pc,pv+bs2*j); 221 } 222 PLogFlops(bslog*(nz+1)-bs); 223 } 224 row = *ajtmp++; 225 } 226 /* finished row so stick it into b->a */ 227 pv = ba + bs2*bi[i]; 228 pj = bj + bi[i]; 229 nz = bi[i+1] - bi[i]; 230 for (j=0; j<nz; j++) { 231 ierr = PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 232 } 233 diag = diag_offset[i] - bi[i]; 234 /* invert diagonal block */ 235 w = pv + bs2*diag; 236 Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work); 237 } 238 239 ierr = PetscFree(rtmp);CHKERRQ(ierr); 240 ierr = PetscFree(v_work);CHKERRQ(ierr); 241 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 242 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 243 C->factor = FACTOR_LU; 244 C->assembled = PETSC_TRUE; 245 PLogFlops(1.3333*bs*bs2*b->mbs); /* from inverting diagonal blocks */ 246 PetscFunctionReturn(0); 247 } 248 /* ------------------------------------------------------------*/ 249 /* 250 Version for when blocks are 7 by 7 251 */ 252 #undef __FUNC__ 253 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_7" 254 int MatLUFactorNumeric_SeqBAIJ_7(Mat A,Mat *B) 255 { 256 Mat C = *B; 257 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 258 IS isrow = b->row,isicol = b->icol; 259 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 260 int *ajtmpold,*ajtmp,nz,row; 261 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 262 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,*pj; 540 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 541 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 542 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 543 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 544 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 545 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 546 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 547 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 548 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 549 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 550 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 551 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 552 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 553 MatScalar *ba = b->a,*aa = a->a; 554 555 PetscFunctionBegin; 556 rtmp = (MatScalar*)PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 557 for (i=0; i<n; i++) { 558 nz = bi[i+1] - bi[i]; 559 ajtmp = bj + bi[i]; 560 for (j=0; j<nz; j++) { 561 x = rtmp+49*ajtmp[j]; 562 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 563 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 564 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 565 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 566 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 567 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 568 } 569 /* load in initial (unfactored row) */ 570 nz = ai[i+1] - ai[i]; 571 ajtmpold = aj + ai[i]; 572 v = aa + 49*ai[i]; 573 for (j=0; j<nz; j++) { 574 x = rtmp+49*ajtmpold[j]; 575 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 576 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 577 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 578 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 579 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 580 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 581 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 582 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 583 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 584 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 585 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 586 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 587 x[48] = v[48]; 588 v += 49; 589 } 590 row = *ajtmp++; 591 while (row < i) { 592 pc = rtmp + 49*row; 593 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 594 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 595 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 596 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 597 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 598 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 599 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 600 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 601 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 602 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 603 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 604 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 605 p49 = pc[48]; 606 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 607 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 608 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 609 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 610 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 611 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 612 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 613 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 614 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 615 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 616 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 617 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 618 p49 != 0.0) { 619 pv = ba + 49*diag_offset[row]; 620 pj = bj + diag_offset[row] + 1; 621 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 622 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 623 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 624 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 625 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 626 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 627 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 628 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 629 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 630 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 631 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 632 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 633 x49 = pv[48]; 634 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 635 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 636 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 637 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 638 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 639 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 640 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 641 642 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 643 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 644 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 645 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 646 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 647 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 648 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 649 650 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 651 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 652 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 653 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 654 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 655 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 656 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 657 658 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 659 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 660 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 661 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 662 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 663 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 664 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 665 666 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 667 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 668 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 669 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 670 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 671 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 672 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 673 674 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 675 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 676 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 677 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 678 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 679 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 680 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 681 682 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 683 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 684 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 685 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 686 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 687 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 688 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 689 690 nz = bi[row+1] - diag_offset[row] - 1; 691 pv += 49; 692 for (j=0; j<nz; j++) { 693 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 694 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 695 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 696 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 697 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 698 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 699 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 700 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 701 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 702 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 703 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 704 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 705 x49 = pv[48]; 706 x = rtmp + 49*pj[j]; 707 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 708 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 709 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 710 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 711 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 712 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 713 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 714 715 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 716 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 717 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 718 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 719 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 720 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 721 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 722 723 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 724 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 725 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 726 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 727 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 728 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 729 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 730 731 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 732 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 733 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 734 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 735 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 736 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 737 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 738 739 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 740 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 741 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 742 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 743 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 744 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 745 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 746 747 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 748 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 749 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 750 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 751 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 752 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 753 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 754 755 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 756 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 757 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 758 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 759 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 760 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 761 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 762 pv += 49; 763 } 764 PLogFlops(686*nz+637); 765 } 766 row = *ajtmp++; 767 } 768 /* finished row so stick it into b->a */ 769 pv = ba + 49*bi[i]; 770 pj = bj + bi[i]; 771 nz = bi[i+1] - bi[i]; 772 for (j=0; j<nz; j++) { 773 x = rtmp+49*pj[j]; 774 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 775 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 776 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 777 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 778 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 779 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 780 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 781 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 782 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 783 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 784 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 785 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 786 pv[48] = x[48]; 787 pv += 49; 788 } 789 /* invert diagonal block */ 790 w = ba + 49*diag_offset[i]; 791 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 792 } 793 794 ierr = PetscFree(rtmp);CHKERRQ(ierr); 795 C->factor = FACTOR_LU; 796 C->assembled = PETSC_TRUE; 797 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 798 PetscFunctionReturn(0); 799 } 800 801 /* ------------------------------------------------------------*/ 802 /* 803 Version for when blocks are 6 by 6 804 */ 805 #undef __FUNC__ 806 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_6" 807 int MatLUFactorNumeric_SeqBAIJ_6(Mat A,Mat *B) 808 { 809 Mat C = *B; 810 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 811 IS isrow = b->row,isicol = b->icol; 812 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 813 int *ajtmpold,*ajtmp,nz,row; 814 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 815 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 816 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 817 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 818 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 819 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 820 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 821 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 822 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 823 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 824 MatScalar *ba = b->a,*aa = a->a; 825 826 PetscFunctionBegin; 827 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 828 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 829 rtmp = (MatScalar*)PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 830 831 for (i=0; i<n; i++) { 832 nz = bi[i+1] - bi[i]; 833 ajtmp = bj + bi[i]; 834 for (j=0; j<nz; j++) { 835 x = rtmp+36*ajtmp[j]; 836 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 837 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 838 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 839 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 840 x[34] = x[35] = 0.0 ; 841 } 842 /* load in initial (unfactored row) */ 843 idx = r[i]; 844 nz = ai[idx+1] - ai[idx]; 845 ajtmpold = aj + ai[idx]; 846 v = aa + 36*ai[idx]; 847 for (j=0; j<nz; j++) { 848 x = rtmp+36*ic[ajtmpold[j]]; 849 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 850 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 851 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 852 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 853 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 854 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 855 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 856 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 857 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 858 v += 36; 859 } 860 row = *ajtmp++; 861 while (row < i) { 862 pc = rtmp + 36*row; 863 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 864 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 865 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 866 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 867 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 868 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 869 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 870 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 871 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 872 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 873 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 874 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 875 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 876 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 877 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 878 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 879 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 880 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 881 pv = ba + 36*diag_offset[row]; 882 pj = bj + diag_offset[row] + 1; 883 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 884 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 885 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 886 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 887 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 888 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 889 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 890 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 891 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 892 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 893 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 894 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 895 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 896 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 897 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 898 899 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 900 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 901 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 902 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 903 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 904 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 905 906 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 907 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 908 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 909 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 910 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 911 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 912 913 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 914 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 915 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 916 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 917 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 918 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 919 920 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 921 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 922 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 923 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 924 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 925 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 926 927 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 928 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 929 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 930 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 931 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 932 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 933 934 nz = bi[row+1] - diag_offset[row] - 1; 935 pv += 36; 936 for (j=0; j<nz; j++) { 937 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 938 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 939 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 940 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 941 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 942 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 943 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 944 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 945 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 946 x = rtmp + 36*pj[j]; 947 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 948 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 949 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 950 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 951 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 952 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 953 954 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 955 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 956 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 957 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 958 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 959 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 960 961 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 962 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 963 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 964 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 965 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 966 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 967 968 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 969 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 970 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 971 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 972 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 973 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 974 975 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 976 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 977 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 978 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 979 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 980 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 981 982 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 983 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 984 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 985 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 986 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 987 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 988 989 pv += 36; 990 } 991 PLogFlops(432*nz+396); 992 } 993 row = *ajtmp++; 994 } 995 /* finished row so stick it into b->a */ 996 pv = ba + 36*bi[i]; 997 pj = bj + bi[i]; 998 nz = bi[i+1] - bi[i]; 999 for (j=0; j<nz; j++) { 1000 x = rtmp+36*pj[j]; 1001 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1002 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1003 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1004 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1005 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1006 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1007 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1008 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1009 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1010 pv += 36; 1011 } 1012 /* invert diagonal block */ 1013 w = ba + 36*diag_offset[i]; 1014 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1015 } 1016 1017 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1018 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1019 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1020 C->factor = FACTOR_LU; 1021 C->assembled = PETSC_TRUE; 1022 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1023 PetscFunctionReturn(0); 1024 } 1025 /* 1026 Version for when blocks are 6 by 6 Using natural ordering 1027 */ 1028 #undef __FUNC__ 1029 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_6_NaturalOrdering" 1030 int MatLUFactorNumeric_SeqBAIJ_6_NaturalOrdering(Mat A,Mat *B) 1031 { 1032 Mat C = *B; 1033 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1034 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1035 int *ajtmpold,*ajtmp,nz,row; 1036 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1037 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1038 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1039 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1040 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1041 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1042 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1043 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1044 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 1045 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 1046 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 1047 MatScalar *ba = b->a,*aa = a->a; 1048 1049 PetscFunctionBegin; 1050 rtmp = (MatScalar*)PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1051 for (i=0; i<n; i++) { 1052 nz = bi[i+1] - bi[i]; 1053 ajtmp = bj + bi[i]; 1054 for (j=0; j<nz; j++) { 1055 x = rtmp+36*ajtmp[j]; 1056 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1057 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1058 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 1059 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 1060 x[34] = x[35] = 0.0 ; 1061 } 1062 /* load in initial (unfactored row) */ 1063 nz = ai[i+1] - ai[i]; 1064 ajtmpold = aj + ai[i]; 1065 v = aa + 36*ai[i]; 1066 for (j=0; j<nz; j++) { 1067 x = rtmp+36*ajtmpold[j]; 1068 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1069 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 1070 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 1071 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 1072 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 1073 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1074 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 1075 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 1076 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 1077 v += 36; 1078 } 1079 row = *ajtmp++; 1080 while (row < i) { 1081 pc = rtmp + 36*row; 1082 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1083 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 1084 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 1085 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 1086 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 1087 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1088 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 1089 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 1090 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 1091 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 1092 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 1093 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 1094 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 1095 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 1096 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 1097 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 1098 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 1099 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 1100 pv = ba + 36*diag_offset[row]; 1101 pj = bj + diag_offset[row] + 1; 1102 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1103 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1104 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1105 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1106 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1107 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1108 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1109 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1110 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1111 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 1112 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 1113 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 1114 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 1115 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 1116 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 1117 1118 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 1119 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 1120 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 1121 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 1122 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 1123 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 1124 1125 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 1126 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 1127 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 1128 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 1129 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 1130 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 1131 1132 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 1133 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 1134 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 1135 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 1136 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 1137 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 1138 1139 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 1140 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 1141 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 1142 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 1143 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 1144 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 1145 1146 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 1147 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 1148 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 1149 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 1150 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 1151 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 1152 1153 nz = bi[row+1] - diag_offset[row] - 1; 1154 pv += 36; 1155 for (j=0; j<nz; j++) { 1156 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1157 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1158 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1159 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1160 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1161 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1162 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1163 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1164 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1165 x = rtmp + 36*pj[j]; 1166 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 1167 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 1168 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 1169 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 1170 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 1171 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 1172 1173 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 1174 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 1175 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 1176 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 1177 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 1178 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 1179 1180 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 1181 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 1182 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 1183 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 1184 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 1185 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 1186 1187 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 1188 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 1189 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 1190 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 1191 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 1192 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 1193 1194 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 1195 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 1196 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 1197 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 1198 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 1199 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 1200 1201 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 1202 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 1203 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 1204 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 1205 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 1206 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 1207 1208 pv += 36; 1209 } 1210 PLogFlops(432*nz+396); 1211 } 1212 row = *ajtmp++; 1213 } 1214 /* finished row so stick it into b->a */ 1215 pv = ba + 36*bi[i]; 1216 pj = bj + bi[i]; 1217 nz = bi[i+1] - bi[i]; 1218 for (j=0; j<nz; j++) { 1219 x = rtmp+36*pj[j]; 1220 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1221 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1222 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1223 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1224 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1225 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1226 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1227 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1228 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1229 pv += 36; 1230 } 1231 /* invert diagonal block */ 1232 w = ba + 36*diag_offset[i]; 1233 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1234 } 1235 1236 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1237 C->factor = FACTOR_LU; 1238 C->assembled = PETSC_TRUE; 1239 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1240 PetscFunctionReturn(0); 1241 } 1242 1243 /* ------------------------------------------------------------*/ 1244 /* 1245 Version for when blocks are 5 by 5 1246 */ 1247 #undef __FUNC__ 1248 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_5" 1249 int MatLUFactorNumeric_SeqBAIJ_5(Mat A,Mat *B) 1250 { 1251 Mat C = *B; 1252 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1253 IS isrow = b->row,isicol = b->icol; 1254 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1255 int *ajtmpold,*ajtmp,nz,row; 1256 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 1257 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1258 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1259 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1260 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 1261 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 1262 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1263 MatScalar *ba = b->a,*aa = a->a; 1264 1265 PetscFunctionBegin; 1266 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1267 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1268 rtmp = (MatScalar*)PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1269 1270 for (i=0; i<n; i++) { 1271 nz = bi[i+1] - bi[i]; 1272 ajtmp = bj + bi[i]; 1273 for (j=0; j<nz; j++) { 1274 x = rtmp+25*ajtmp[j]; 1275 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1276 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1277 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1278 } 1279 /* load in initial (unfactored row) */ 1280 idx = r[i]; 1281 nz = ai[idx+1] - ai[idx]; 1282 ajtmpold = aj + ai[idx]; 1283 v = aa + 25*ai[idx]; 1284 for (j=0; j<nz; j++) { 1285 x = rtmp+25*ic[ajtmpold[j]]; 1286 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1287 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1288 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1289 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; 1290 x[18] = v[18]; x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; 1291 x[22] = v[22]; x[23] = v[23]; x[24] = v[24]; 1292 v += 25; 1293 } 1294 row = *ajtmp++; 1295 while (row < i) { 1296 pc = rtmp + 25*row; 1297 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1298 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1299 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1300 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; 1301 p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1302 p25 = pc[24]; 1303 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1304 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1305 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1306 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || 1307 p20 != 0.0 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || 1308 p24 != 0.0 || p25 != 0.0) { 1309 pv = ba + 25*diag_offset[row]; 1310 pj = bj + diag_offset[row] + 1; 1311 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1312 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1313 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1314 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1315 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; 1316 x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1317 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1318 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1319 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1320 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1321 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1322 1323 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1324 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1325 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1326 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1327 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1328 1329 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1330 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1331 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1332 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1333 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1334 1335 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1336 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1337 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1338 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1339 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1340 1341 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1342 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1343 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1344 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1345 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1346 1347 nz = bi[row+1] - diag_offset[row] - 1; 1348 pv += 25; 1349 for (j=0; j<nz; j++) { 1350 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1351 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1352 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1353 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; 1354 x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; 1355 x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1356 x = rtmp + 25*pj[j]; 1357 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1358 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1359 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1360 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1361 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1362 1363 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1364 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1365 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1366 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1367 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1368 1369 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1370 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1371 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1372 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1373 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1374 1375 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1376 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1377 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1378 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1379 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1380 1381 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1382 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1383 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1384 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1385 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1386 1387 pv += 25; 1388 } 1389 PLogFlops(250*nz+225); 1390 } 1391 row = *ajtmp++; 1392 } 1393 /* finished row so stick it into b->a */ 1394 pv = ba + 25*bi[i]; 1395 pj = bj + bi[i]; 1396 nz = bi[i+1] - bi[i]; 1397 for (j=0; j<nz; j++) { 1398 x = rtmp+25*pj[j]; 1399 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1400 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1401 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1402 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; 1403 pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; 1404 pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; pv[24] = x[24]; 1405 pv += 25; 1406 } 1407 /* invert diagonal block */ 1408 w = ba + 25*diag_offset[i]; 1409 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1410 } 1411 1412 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1413 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1414 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1415 C->factor = FACTOR_LU; 1416 C->assembled = PETSC_TRUE; 1417 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1418 PetscFunctionReturn(0); 1419 } 1420 /* 1421 Version for when blocks are 5 by 5 Using natural ordering 1422 */ 1423 #undef __FUNC__ 1424 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_5_NaturalOrdering" 1425 int MatLUFactorNumeric_SeqBAIJ_5_NaturalOrdering(Mat A,Mat *B) 1426 { 1427 Mat C = *B; 1428 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1429 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1430 int *ajtmpold,*ajtmp,nz,row; 1431 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1432 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1433 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1434 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1435 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1436 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1437 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1438 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1439 MatScalar *ba = b->a,*aa = a->a; 1440 1441 PetscFunctionBegin; 1442 rtmp = (MatScalar*)PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1443 for (i=0; i<n; i++) { 1444 nz = bi[i+1] - bi[i]; 1445 ajtmp = bj + bi[i]; 1446 for (j=0; j<nz; j++) { 1447 x = rtmp+25*ajtmp[j]; 1448 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1449 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1450 x[16] = x[17] = x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1451 } 1452 /* load in initial (unfactored row) */ 1453 nz = ai[i+1] - ai[i]; 1454 ajtmpold = aj + ai[i]; 1455 v = aa + 25*ai[i]; 1456 for (j=0; j<nz; j++) { 1457 x = rtmp+25*ajtmpold[j]; 1458 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1459 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1460 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1461 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; 1462 x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1463 x[24] = v[24]; 1464 v += 25; 1465 } 1466 row = *ajtmp++; 1467 while (row < i) { 1468 pc = rtmp + 25*row; 1469 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1470 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1471 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1472 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; 1473 p19 = pc[18]; p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; 1474 p24 = pc[23]; p25 = pc[24]; 1475 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1476 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1477 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1478 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 1479 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || p25 != 0.0) { 1480 pv = ba + 25*diag_offset[row]; 1481 pj = bj + diag_offset[row] + 1; 1482 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1483 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1484 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1485 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; 1486 x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1487 x25 = pv[24]; 1488 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1489 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1490 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1491 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1492 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1493 1494 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1495 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1496 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1497 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1498 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1499 1500 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1501 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1502 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1503 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1504 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1505 1506 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1507 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1508 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1509 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1510 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1511 1512 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1513 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1514 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1515 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1516 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1517 1518 nz = bi[row+1] - diag_offset[row] - 1; 1519 pv += 25; 1520 for (j=0; j<nz; j++) { 1521 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1522 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1523 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1524 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1525 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; 1526 x24 = pv[23]; x25 = pv[24]; 1527 x = rtmp + 25*pj[j]; 1528 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1529 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1530 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1531 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1532 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1533 1534 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1535 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1536 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1537 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1538 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1539 1540 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1541 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1542 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1543 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1544 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1545 1546 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1547 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1548 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1549 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1550 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1551 1552 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1553 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1554 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1555 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1556 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1557 pv += 25; 1558 } 1559 PLogFlops(250*nz+225); 1560 } 1561 row = *ajtmp++; 1562 } 1563 /* finished row so stick it into b->a */ 1564 pv = ba + 25*bi[i]; 1565 pj = bj + bi[i]; 1566 nz = bi[i+1] - bi[i]; 1567 for (j=0; j<nz; j++) { 1568 x = rtmp+25*pj[j]; 1569 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1570 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1571 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1572 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; pv[17] = x[17]; 1573 pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; 1574 pv[23] = x[23]; pv[24] = x[24]; 1575 pv += 25; 1576 } 1577 /* invert diagonal block */ 1578 w = ba + 25*diag_offset[i]; 1579 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1580 } 1581 1582 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1583 C->factor = FACTOR_LU; 1584 C->assembled = PETSC_TRUE; 1585 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1586 PetscFunctionReturn(0); 1587 } 1588 1589 /* ------------------------------------------------------------*/ 1590 /* 1591 Version for when blocks are 4 by 4 1592 */ 1593 #undef __FUNC__ 1594 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_4" 1595 int MatLUFactorNumeric_SeqBAIJ_4(Mat A,Mat *B) 1596 { 1597 Mat C = *B; 1598 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1599 IS isrow = b->row,isicol = b->icol; 1600 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1601 int *ajtmpold,*ajtmp,nz,row; 1602 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 1603 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1604 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1605 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1606 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1607 MatScalar m13,m14,m15,m16; 1608 MatScalar *ba = b->a,*aa = a->a; 1609 1610 PetscFunctionBegin; 1611 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1612 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1613 rtmp = (MatScalar*)PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1614 1615 for (i=0; i<n; i++) { 1616 nz = bi[i+1] - bi[i]; 1617 ajtmp = bj + bi[i]; 1618 for (j=0; j<nz; j++) { 1619 x = rtmp+16*ajtmp[j]; 1620 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1621 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1622 } 1623 /* load in initial (unfactored row) */ 1624 idx = r[i]; 1625 nz = ai[idx+1] - ai[idx]; 1626 ajtmpold = aj + ai[idx]; 1627 v = aa + 16*ai[idx]; 1628 for (j=0; j<nz; j++) { 1629 x = rtmp+16*ic[ajtmpold[j]]; 1630 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1631 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1632 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1633 x[14] = v[14]; x[15] = v[15]; 1634 v += 16; 1635 } 1636 row = *ajtmp++; 1637 while (row < i) { 1638 pc = rtmp + 16*row; 1639 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1640 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1641 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1642 p15 = pc[14]; p16 = pc[15]; 1643 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1644 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1645 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1646 || p16 != 0.0) { 1647 pv = ba + 16*diag_offset[row]; 1648 pj = bj + diag_offset[row] + 1; 1649 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1650 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1651 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1652 x15 = pv[14]; x16 = pv[15]; 1653 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1654 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1655 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1656 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1657 1658 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1659 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1660 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1661 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1662 1663 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1664 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1665 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1666 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1667 1668 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1669 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1670 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1671 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1672 1673 nz = bi[row+1] - diag_offset[row] - 1; 1674 pv += 16; 1675 for (j=0; j<nz; j++) { 1676 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1677 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1678 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1679 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1680 x = rtmp + 16*pj[j]; 1681 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1682 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1683 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1684 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1685 1686 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1687 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1688 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1689 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1690 1691 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1692 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1693 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1694 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1695 1696 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1697 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1698 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1699 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1700 1701 pv += 16; 1702 } 1703 PLogFlops(128*nz+112); 1704 } 1705 row = *ajtmp++; 1706 } 1707 /* finished row so stick it into b->a */ 1708 pv = ba + 16*bi[i]; 1709 pj = bj + bi[i]; 1710 nz = bi[i+1] - bi[i]; 1711 for (j=0; j<nz; j++) { 1712 x = rtmp+16*pj[j]; 1713 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1714 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1715 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1716 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1717 pv += 16; 1718 } 1719 /* invert diagonal block */ 1720 w = ba + 16*diag_offset[i]; 1721 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1722 } 1723 1724 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1725 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1726 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1727 C->factor = FACTOR_LU; 1728 C->assembled = PETSC_TRUE; 1729 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1730 PetscFunctionReturn(0); 1731 } 1732 /* 1733 Version for when blocks are 4 by 4 Using natural ordering 1734 */ 1735 #undef __FUNC__ 1736 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_4_NaturalOrdering" 1737 int MatLUFactorNumeric_SeqBAIJ_4_NaturalOrdering(Mat A,Mat *B) 1738 { 1739 Mat C = *B; 1740 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1741 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1742 int *ajtmpold,*ajtmp,nz,row; 1743 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1744 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1745 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1746 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1747 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1748 MatScalar m13,m14,m15,m16; 1749 MatScalar *ba = b->a,*aa = a->a; 1750 1751 PetscFunctionBegin; 1752 rtmp = (MatScalar*)PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1753 1754 for (i=0; i<n; i++) { 1755 nz = bi[i+1] - bi[i]; 1756 ajtmp = bj + bi[i]; 1757 for (j=0; j<nz; j++) { 1758 x = rtmp+16*ajtmp[j]; 1759 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1760 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1761 } 1762 /* load in initial (unfactored row) */ 1763 nz = ai[i+1] - ai[i]; 1764 ajtmpold = aj + ai[i]; 1765 v = aa + 16*ai[i]; 1766 for (j=0; j<nz; j++) { 1767 x = rtmp+16*ajtmpold[j]; 1768 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1769 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1770 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1771 x[14] = v[14]; x[15] = v[15]; 1772 v += 16; 1773 } 1774 row = *ajtmp++; 1775 while (row < i) { 1776 pc = rtmp + 16*row; 1777 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1778 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1779 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1780 p15 = pc[14]; p16 = pc[15]; 1781 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1782 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1783 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1784 || p16 != 0.0) { 1785 pv = ba + 16*diag_offset[row]; 1786 pj = bj + diag_offset[row] + 1; 1787 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1788 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1789 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1790 x15 = pv[14]; x16 = pv[15]; 1791 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1792 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1793 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1794 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1795 1796 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1797 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1798 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1799 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1800 1801 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1802 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1803 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1804 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1805 1806 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1807 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1808 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1809 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1810 1811 nz = bi[row+1] - diag_offset[row] - 1; 1812 pv += 16; 1813 for (j=0; j<nz; j++) { 1814 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1815 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1816 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1817 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1818 x = rtmp + 16*pj[j]; 1819 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1820 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1821 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1822 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1823 1824 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1825 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1826 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1827 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1828 1829 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1830 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1831 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1832 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1833 1834 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1835 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1836 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1837 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1838 1839 pv += 16; 1840 } 1841 PLogFlops(128*nz+112); 1842 } 1843 row = *ajtmp++; 1844 } 1845 /* finished row so stick it into b->a */ 1846 pv = ba + 16*bi[i]; 1847 pj = bj + bi[i]; 1848 nz = bi[i+1] - bi[i]; 1849 for (j=0; j<nz; j++) { 1850 x = rtmp+16*pj[j]; 1851 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1852 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1853 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1854 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1855 pv += 16; 1856 } 1857 /* invert diagonal block */ 1858 w = ba + 16*diag_offset[i]; 1859 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1860 } 1861 1862 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1863 C->factor = FACTOR_LU; 1864 C->assembled = PETSC_TRUE; 1865 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1866 PetscFunctionReturn(0); 1867 } 1868 1869 1870 /* ------------------------------------------------------------*/ 1871 /* 1872 Version for when blocks are 3 by 3 1873 */ 1874 #undef __FUNC__ 1875 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_3" 1876 int MatLUFactorNumeric_SeqBAIJ_3(Mat A,Mat *B) 1877 { 1878 Mat C = *B; 1879 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1880 IS isrow = b->row,isicol = b->icol; 1881 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1882 int *ajtmpold,*ajtmp,nz,row,*ai=a->i,*aj=a->j; 1883 int *diag_offset = b->diag,idx,*pj; 1884 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1885 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1886 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 1887 MatScalar *ba = b->a,*aa = a->a; 1888 1889 PetscFunctionBegin; 1890 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1891 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1892 rtmp = (MatScalar*)PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1893 1894 for (i=0; i<n; i++) { 1895 nz = bi[i+1] - bi[i]; 1896 ajtmp = bj + bi[i]; 1897 for (j=0; j<nz; j++) { 1898 x = rtmp + 9*ajtmp[j]; 1899 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 1900 } 1901 /* load in initial (unfactored row) */ 1902 idx = r[i]; 1903 nz = ai[idx+1] - ai[idx]; 1904 ajtmpold = aj + ai[idx]; 1905 v = aa + 9*ai[idx]; 1906 for (j=0; j<nz; j++) { 1907 x = rtmp + 9*ic[ajtmpold[j]]; 1908 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1909 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1910 v += 9; 1911 } 1912 row = *ajtmp++; 1913 while (row < i) { 1914 pc = rtmp + 9*row; 1915 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1916 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1917 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1918 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 1919 pv = ba + 9*diag_offset[row]; 1920 pj = bj + diag_offset[row] + 1; 1921 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1922 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1923 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 1924 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 1925 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 1926 1927 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 1928 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 1929 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 1930 1931 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 1932 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 1933 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 1934 nz = bi[row+1] - diag_offset[row] - 1; 1935 pv += 9; 1936 for (j=0; j<nz; j++) { 1937 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1938 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1939 x = rtmp + 9*pj[j]; 1940 x[0] -= m1*x1 + m4*x2 + m7*x3; 1941 x[1] -= m2*x1 + m5*x2 + m8*x3; 1942 x[2] -= m3*x1 + m6*x2 + m9*x3; 1943 1944 x[3] -= m1*x4 + m4*x5 + m7*x6; 1945 x[4] -= m2*x4 + m5*x5 + m8*x6; 1946 x[5] -= m3*x4 + m6*x5 + m9*x6; 1947 1948 x[6] -= m1*x7 + m4*x8 + m7*x9; 1949 x[7] -= m2*x7 + m5*x8 + m8*x9; 1950 x[8] -= m3*x7 + m6*x8 + m9*x9; 1951 pv += 9; 1952 } 1953 PLogFlops(54*nz+36); 1954 } 1955 row = *ajtmp++; 1956 } 1957 /* finished row so stick it into b->a */ 1958 pv = ba + 9*bi[i]; 1959 pj = bj + bi[i]; 1960 nz = bi[i+1] - bi[i]; 1961 for (j=0; j<nz; j++) { 1962 x = rtmp + 9*pj[j]; 1963 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1964 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1965 pv += 9; 1966 } 1967 /* invert diagonal block */ 1968 w = ba + 9*diag_offset[i]; 1969 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 1970 } 1971 1972 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1973 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1974 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1975 C->factor = FACTOR_LU; 1976 C->assembled = PETSC_TRUE; 1977 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 1978 PetscFunctionReturn(0); 1979 } 1980 /* 1981 Version for when blocks are 3 by 3 Using natural ordering 1982 */ 1983 #undef __FUNC__ 1984 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_3_NaturalOrdering" 1985 int MatLUFactorNumeric_SeqBAIJ_3_NaturalOrdering(Mat A,Mat *B) 1986 { 1987 Mat C = *B; 1988 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1989 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1990 int *ajtmpold,*ajtmp,nz,row; 1991 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1992 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1993 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1994 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 1995 MatScalar *ba = b->a,*aa = a->a; 1996 1997 PetscFunctionBegin; 1998 rtmp = (MatScalar*)PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1999 2000 for (i=0; i<n; i++) { 2001 nz = bi[i+1] - bi[i]; 2002 ajtmp = bj + bi[i]; 2003 for (j=0; j<nz; j++) { 2004 x = rtmp+9*ajtmp[j]; 2005 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 2006 } 2007 /* load in initial (unfactored row) */ 2008 nz = ai[i+1] - ai[i]; 2009 ajtmpold = aj + ai[i]; 2010 v = aa + 9*ai[i]; 2011 for (j=0; j<nz; j++) { 2012 x = rtmp+9*ajtmpold[j]; 2013 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2014 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 2015 v += 9; 2016 } 2017 row = *ajtmp++; 2018 while (row < i) { 2019 pc = rtmp + 9*row; 2020 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2021 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 2022 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 2023 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 2024 pv = ba + 9*diag_offset[row]; 2025 pj = bj + diag_offset[row] + 1; 2026 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2027 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2028 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 2029 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 2030 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 2031 2032 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 2033 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 2034 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 2035 2036 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 2037 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 2038 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 2039 2040 nz = bi[row+1] - diag_offset[row] - 1; 2041 pv += 9; 2042 for (j=0; j<nz; j++) { 2043 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2044 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2045 x = rtmp + 9*pj[j]; 2046 x[0] -= m1*x1 + m4*x2 + m7*x3; 2047 x[1] -= m2*x1 + m5*x2 + m8*x3; 2048 x[2] -= m3*x1 + m6*x2 + m9*x3; 2049 2050 x[3] -= m1*x4 + m4*x5 + m7*x6; 2051 x[4] -= m2*x4 + m5*x5 + m8*x6; 2052 x[5] -= m3*x4 + m6*x5 + m9*x6; 2053 2054 x[6] -= m1*x7 + m4*x8 + m7*x9; 2055 x[7] -= m2*x7 + m5*x8 + m8*x9; 2056 x[8] -= m3*x7 + m6*x8 + m9*x9; 2057 pv += 9; 2058 } 2059 PLogFlops(54*nz+36); 2060 } 2061 row = *ajtmp++; 2062 } 2063 /* finished row so stick it into b->a */ 2064 pv = ba + 9*bi[i]; 2065 pj = bj + bi[i]; 2066 nz = bi[i+1] - bi[i]; 2067 for (j=0; j<nz; j++) { 2068 x = rtmp+9*pj[j]; 2069 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2070 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 2071 pv += 9; 2072 } 2073 /* invert diagonal block */ 2074 w = ba + 9*diag_offset[i]; 2075 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 2076 } 2077 2078 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2079 C->factor = FACTOR_LU; 2080 C->assembled = PETSC_TRUE; 2081 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 2082 PetscFunctionReturn(0); 2083 } 2084 2085 /* ------------------------------------------------------------*/ 2086 /* 2087 Version for when blocks are 2 by 2 2088 */ 2089 #undef __FUNC__ 2090 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_2" 2091 int MatLUFactorNumeric_SeqBAIJ_2(Mat A,Mat *B) 2092 { 2093 Mat C = *B; 2094 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 2095 IS isrow = b->row,isicol = b->icol; 2096 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 2097 int *ajtmpold,*ajtmp,nz,row; 2098 int *diag_offset=b->diag,idx,*ai=a->i,*aj=a->j,*pj; 2099 MatScalar *pv,*v,*rtmp,m1,m2,m3,m4,*pc,*w,*x,x1,x2,x3,x4; 2100 MatScalar p1,p2,p3,p4; 2101 MatScalar *ba = b->a,*aa = a->a; 2102 2103 PetscFunctionBegin; 2104 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 2105 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 2106 rtmp = (MatScalar*)PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2107 2108 for (i=0; i<n; i++) { 2109 nz = bi[i+1] - bi[i]; 2110 ajtmp = bj + bi[i]; 2111 for (j=0; j<nz; j++) { 2112 x = rtmp+4*ajtmp[j]; x[0] = x[1] = x[2] = x[3] = 0.0; 2113 } 2114 /* load in initial (unfactored row) */ 2115 idx = r[i]; 2116 nz = ai[idx+1] - ai[idx]; 2117 ajtmpold = aj + ai[idx]; 2118 v = aa + 4*ai[idx]; 2119 for (j=0; j<nz; j++) { 2120 x = rtmp+4*ic[ajtmpold[j]]; 2121 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2122 v += 4; 2123 } 2124 row = *ajtmp++; 2125 while (row < i) { 2126 pc = rtmp + 4*row; 2127 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2128 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2129 pv = ba + 4*diag_offset[row]; 2130 pj = bj + diag_offset[row] + 1; 2131 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2132 pc[0] = m1 = p1*x1 + p3*x2; 2133 pc[1] = m2 = p2*x1 + p4*x2; 2134 pc[2] = m3 = p1*x3 + p3*x4; 2135 pc[3] = m4 = p2*x3 + p4*x4; 2136 nz = bi[row+1] - diag_offset[row] - 1; 2137 pv += 4; 2138 for (j=0; j<nz; j++) { 2139 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2140 x = rtmp + 4*pj[j]; 2141 x[0] -= m1*x1 + m3*x2; 2142 x[1] -= m2*x1 + m4*x2; 2143 x[2] -= m1*x3 + m3*x4; 2144 x[3] -= m2*x3 + m4*x4; 2145 pv += 4; 2146 } 2147 PLogFlops(16*nz+12); 2148 } 2149 row = *ajtmp++; 2150 } 2151 /* finished row so stick it into b->a */ 2152 pv = ba + 4*bi[i]; 2153 pj = bj + bi[i]; 2154 nz = bi[i+1] - bi[i]; 2155 for (j=0; j<nz; j++) { 2156 x = rtmp+4*pj[j]; 2157 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2158 pv += 4; 2159 } 2160 /* invert diagonal block */ 2161 w = ba + 4*diag_offset[i]; 2162 ierr = Kernel_A_gets_inverse_A_2(w);CHKERRQ(ierr); 2163 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2164 } 2165 2166 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2167 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 2168 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 2169 C->factor = FACTOR_LU; 2170 C->assembled = PETSC_TRUE; 2171 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2172 PetscFunctionReturn(0); 2173 } 2174 /* 2175 Version for when blocks are 2 by 2 Using natural ordering 2176 */ 2177 #undef __FUNC__ 2178 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering" 2179 int MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering(Mat A,Mat *B) 2180 { 2181 Mat C = *B; 2182 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 2183 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 2184 int *ajtmpold,*ajtmp,nz,row; 2185 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 2186 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2187 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,x1,x2,x3,x4; 2188 MatScalar *ba = b->a,*aa = a->a; 2189 2190 PetscFunctionBegin; 2191 rtmp = (MatScalar*)PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2192 2193 for (i=0; i<n; i++) { 2194 nz = bi[i+1] - bi[i]; 2195 ajtmp = bj + bi[i]; 2196 for (j=0; j<nz; j++) { 2197 x = rtmp+4*ajtmp[j]; 2198 x[0] = x[1] = x[2] = x[3] = 0.0; 2199 } 2200 /* load in initial (unfactored row) */ 2201 nz = ai[i+1] - ai[i]; 2202 ajtmpold = aj + ai[i]; 2203 v = aa + 4*ai[i]; 2204 for (j=0; j<nz; j++) { 2205 x = rtmp+4*ajtmpold[j]; 2206 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2207 v += 4; 2208 } 2209 row = *ajtmp++; 2210 while (row < i) { 2211 pc = rtmp + 4*row; 2212 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2213 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2214 pv = ba + 4*diag_offset[row]; 2215 pj = bj + diag_offset[row] + 1; 2216 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2217 pc[0] = m1 = p1*x1 + p3*x2; 2218 pc[1] = m2 = p2*x1 + p4*x2; 2219 pc[2] = m3 = p1*x3 + p3*x4; 2220 pc[3] = m4 = p2*x3 + p4*x4; 2221 nz = bi[row+1] - diag_offset[row] - 1; 2222 pv += 4; 2223 for (j=0; j<nz; j++) { 2224 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2225 x = rtmp + 4*pj[j]; 2226 x[0] -= m1*x1 + m3*x2; 2227 x[1] -= m2*x1 + m4*x2; 2228 x[2] -= m1*x3 + m3*x4; 2229 x[3] -= m2*x3 + m4*x4; 2230 pv += 4; 2231 } 2232 PLogFlops(16*nz+12); 2233 } 2234 row = *ajtmp++; 2235 } 2236 /* finished row so stick it into b->a */ 2237 pv = ba + 4*bi[i]; 2238 pj = bj + bi[i]; 2239 nz = bi[i+1] - bi[i]; 2240 for (j=0; j<nz; j++) { 2241 x = rtmp+4*pj[j]; 2242 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2243 pv += 4; 2244 } 2245 /* invert diagonal block */ 2246 w = ba + 4*diag_offset[i]; 2247 ierr = Kernel_A_gets_inverse_A_2(w);CHKERRQ(ierr); 2248 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2249 } 2250 2251 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2252 C->factor = FACTOR_LU; 2253 C->assembled = PETSC_TRUE; 2254 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2255 PetscFunctionReturn(0); 2256 } 2257 2258 /* ----------------------------------------------------------- */ 2259 /* 2260 Version for when blocks are 1 by 1. 2261 */ 2262 #undef __FUNC__ 2263 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_1" 2264 int MatLUFactorNumeric_SeqBAIJ_1(Mat A,Mat *B) 2265 { 2266 Mat C = *B; 2267 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 2268 IS isrow = b->row,isicol = b->icol; 2269 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 2270 int *ajtmpold,*ajtmp,nz,row,*ai = a->i,*aj = a->j; 2271 int *diag_offset = b->diag,diag,*pj; 2272 MatScalar *pv,*v,*rtmp,multiplier,*pc; 2273 MatScalar *ba = b->a,*aa = a->a; 2274 2275 PetscFunctionBegin; 2276 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 2277 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 2278 rtmp = (MatScalar*)PetscMalloc((n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2279 2280 for (i=0; i<n; i++) { 2281 nz = bi[i+1] - bi[i]; 2282 ajtmp = bj + bi[i]; 2283 for (j=0; j<nz; j++) rtmp[ajtmp[j]] = 0.0; 2284 2285 /* load in initial (unfactored row) */ 2286 nz = ai[r[i]+1] - ai[r[i]]; 2287 ajtmpold = aj + ai[r[i]]; 2288 v = aa + ai[r[i]]; 2289 for (j=0; j<nz; j++) rtmp[ic[ajtmpold[j]]] = v[j]; 2290 2291 row = *ajtmp++; 2292 while (row < i) { 2293 pc = rtmp + row; 2294 if (*pc != 0.0) { 2295 pv = ba + diag_offset[row]; 2296 pj = bj + diag_offset[row] + 1; 2297 multiplier = *pc * *pv++; 2298 *pc = multiplier; 2299 nz = bi[row+1] - diag_offset[row] - 1; 2300 for (j=0; j<nz; j++) rtmp[pj[j]] -= multiplier * pv[j]; 2301 PLogFlops(1+2*nz); 2302 } 2303 row = *ajtmp++; 2304 } 2305 /* finished row so stick it into b->a */ 2306 pv = ba + bi[i]; 2307 pj = bj + bi[i]; 2308 nz = bi[i+1] - bi[i]; 2309 for (j=0; j<nz; j++) {pv[j] = rtmp[pj[j]];} 2310 diag = diag_offset[i] - bi[i]; 2311 /* check pivot entry for current row */ 2312 if (pv[diag] == 0.0) { 2313 SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,0,"Zero pivot"); 2314 } 2315 pv[diag] = 1.0/pv[diag]; 2316 } 2317 2318 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2319 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 2320 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 2321 C->factor = FACTOR_LU; 2322 C->assembled = PETSC_TRUE; 2323 PLogFlops(b->n); 2324 PetscFunctionReturn(0); 2325 } 2326 2327 /* ----------------------------------------------------------- */ 2328 #undef __FUNC__ 2329 #define __FUNC__ "MatLUFactor_SeqBAIJ" 2330 int MatLUFactor_SeqBAIJ(Mat A,IS row,IS col,PetscReal f) 2331 { 2332 Mat_SeqBAIJ *mat = (Mat_SeqBAIJ*)A->data; 2333 int ierr,refct; 2334 Mat C; 2335 PetscOps *Abops; 2336 MatOps Aops; 2337 2338 PetscFunctionBegin; 2339 ierr = MatLUFactorSymbolic(A,row,col,f,&C);CHKERRQ(ierr); 2340 ierr = MatLUFactorNumeric(A,&C);CHKERRQ(ierr); 2341 2342 /* free all the data structures from mat */ 2343 ierr = PetscFree(mat->a);CHKERRQ(ierr); 2344 if (!mat->singlemalloc) { 2345 ierr = PetscFree(mat->i);CHKERRQ(ierr); 2346 ierr = PetscFree(mat->j);CHKERRQ(ierr); 2347 } 2348 if (mat->diag) {ierr = PetscFree(mat->diag);CHKERRQ(ierr);} 2349 if (mat->ilen) {ierr = PetscFree(mat->ilen);CHKERRQ(ierr);} 2350 if (mat->imax) {ierr = PetscFree(mat->imax);CHKERRQ(ierr);} 2351 if (mat->solve_work) {ierr = PetscFree(mat->solve_work);CHKERRQ(ierr);} 2352 if (mat->mult_work) {ierr = PetscFree(mat->mult_work);CHKERRQ(ierr);} 2353 if (mat->icol) {ierr = ISDestroy(mat->icol);CHKERRQ(ierr);} 2354 ierr = PetscFree(mat);CHKERRQ(ierr); 2355 2356 ierr = MapDestroy(A->rmap);CHKERRQ(ierr); 2357 ierr = MapDestroy(A->cmap);CHKERRQ(ierr); 2358 2359 /* 2360 This is horrible,horrible code. We need to keep the 2361 A pointers for the bops and ops but copy everything 2362 else from C. 2363 */ 2364 Abops = A->bops; 2365 Aops = A->ops; 2366 refct = A->refct; 2367 ierr = PetscMemcpy(A,C,sizeof(struct _p_Mat));CHKERRQ(ierr); 2368 mat = (Mat_SeqBAIJ*)A->data; 2369 PLogObjectParent(A,mat->icol); 2370 2371 A->bops = Abops; 2372 A->ops = Aops; 2373 A->qlist = 0; 2374 A->refct = refct; 2375 /* copy over the type_name and name */ 2376 ierr = PetscStrallocpy(C->type_name,&A->type_name);CHKERRQ(ierr); 2377 ierr = PetscStrallocpy(C->name,&A->name);CHKERRQ(ierr); 2378 2379 PetscHeaderDestroy(C); 2380 PetscFunctionReturn(0); 2381 } 2382 2383 2384 2385