1 /*$Id: baijfact.c,v 1.79 2000/02/02 20:09:09 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 29 if (!isrow) { 30 ierr = ISCreateStride(PETSC_COMM_SELF,A->M,0,1,&isrow);CHKERRQ(ierr); 31 } 32 if (!iscol) { 33 ierr = ISCreateStride(PETSC_COMM_SELF,A->M,0,1,&iscol);CHKERRQ(ierr); 34 } 35 ierr = ISInvertPermutation(iscol,PETSC_DECIDE,&isicol);CHKERRQ(ierr); 36 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 37 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 38 39 /* get new row pointers */ 40 ainew = (int*)PetscMalloc((n+1)*sizeof(int));CHKPTRQ(ainew); 41 ainew[0] = 0; 42 /* don't know how many column pointers are needed so estimate */ 43 jmax = (int)(f*ai[n] + 1); 44 ajnew = (int*)PetscMalloc((jmax)*sizeof(int));CHKPTRQ(ajnew); 45 /* fill is a linked list of nonzeros in active row */ 46 fill = (int*)PetscMalloc((2*n+1)*sizeof(int));CHKPTRQ(fill); 47 im = fill + n + 1; 48 /* idnew is location of diagonal in factor */ 49 idnew = (int*)PetscMalloc((n+1)*sizeof(int));CHKPTRQ(idnew); 50 idnew[0] = 0; 51 52 for (i=0; i<n; i++) { 53 /* first copy previous fill into linked list */ 54 nnz = nz = ai[r[i]+1] - ai[r[i]]; 55 if (!nz) SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,1,"Empty row in matrix"); 56 ajtmp = aj + ai[r[i]]; 57 fill[n] = n; 58 while (nz--) { 59 fm = n; 60 idx = ic[*ajtmp++]; 61 do { 62 m = fm; 63 fm = fill[m]; 64 } while (fm < idx); 65 fill[m] = idx; 66 fill[idx] = fm; 67 } 68 row = fill[n]; 69 while (row < i) { 70 ajtmp = ajnew + idnew[row] + 1; 71 nzbd = 1 + idnew[row] - ainew[row]; 72 nz = im[row] - nzbd; 73 fm = row; 74 while (nz-- > 0) { 75 idx = *ajtmp++; 76 nzbd++; 77 if (idx == i) im[row] = nzbd; 78 do { 79 m = fm; 80 fm = fill[m]; 81 } while (fm < idx); 82 if (fm != idx) { 83 fill[m] = idx; 84 fill[idx] = fm; 85 fm = idx; 86 nnz++; 87 } 88 } 89 row = fill[row]; 90 } 91 /* copy new filled row into permanent storage */ 92 ainew[i+1] = ainew[i] + nnz; 93 if (ainew[i+1] > jmax) { 94 95 /* estimate how much additional space we will need */ 96 /* use the strategy suggested by David Hysom <hysom@perch-t.icase.edu> */ 97 /* just double the memory each time */ 98 int maxadd = jmax; 99 /* maxadd = (int)((f*(ai[n]+1)*(n-i+5))/n); */ 100 if (maxadd < nnz) maxadd = (n-i)*(nnz+1); 101 jmax += maxadd; 102 103 /* allocate a longer ajnew */ 104 ajtmp = (int*)PetscMalloc(jmax*sizeof(int));CHKPTRQ(ajtmp); 105 ierr = PetscMemcpy(ajtmp,ajnew,ainew[i]*sizeof(int));CHKERRQ(ierr); 106 ierr = PetscFree(ajnew);CHKERRQ(ierr); 107 ajnew = ajtmp; 108 realloc++; /* count how many times we realloc */ 109 } 110 ajtmp = ajnew + ainew[i]; 111 fm = fill[n]; 112 nzi = 0; 113 im[i] = nnz; 114 while (nnz--) { 115 if (fm < i) nzi++; 116 *ajtmp++ = fm; 117 fm = fill[fm]; 118 } 119 idnew[i] = ainew[i] + nzi; 120 } 121 122 if (ai[n] != 0) { 123 PetscReal af = ((PetscReal)ainew[n])/((PetscReal)ai[n]); 124 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:Reallocs %d Fill ratio:given %g needed %g\n",realloc,f,af); 125 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:Run with -pc_lu_fill %g or use \n",af); 126 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:PCLUSetFill(pc,%g);\n",af); 127 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:for best performance.\n"); 128 } else { 129 PLogInfo(A,"MatLUFactorSymbolic_SeqBAIJ:Empty matrix.\n"); 130 } 131 132 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 133 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 134 135 ierr = PetscFree(fill);CHKERRQ(ierr); 136 137 /* put together the new matrix */ 138 ierr = MatCreateSeqBAIJ(A->comm,bs,bs*n,bs*n,0,PETSC_NULL,B);CHKERRQ(ierr); 139 PLogObjectParent(*B,isicol); 140 b = (Mat_SeqBAIJ*)(*B)->data; 141 ierr = PetscFree(b->imax);CHKERRQ(ierr); 142 b->singlemalloc = PETSC_FALSE; 143 /* the next line frees the default space generated by the Create() */ 144 ierr = PetscFree(b->a);CHKERRQ(ierr); 145 ierr = PetscFree(b->ilen);CHKERRQ(ierr); 146 b->a = (MatScalar*)PetscMalloc((ainew[n]+1)*sizeof(MatScalar)*bs2);CHKPTRQ(b->a); 147 b->j = ajnew; 148 b->i = ainew; 149 b->diag = idnew; 150 b->ilen = 0; 151 b->imax = 0; 152 b->row = isrow; 153 b->col = iscol; 154 ierr = PetscObjectReference((PetscObject)isrow);CHKERRQ(ierr); 155 ierr = PetscObjectReference((PetscObject)iscol);CHKERRQ(ierr); 156 b->icol = isicol; 157 b->solve_work = (Scalar*)PetscMalloc((bs*n+bs)*sizeof(Scalar));CHKPTRQ(b->solve_work); 158 /* In b structure: Free imax, ilen, old a, old j. 159 Allocate idnew, solve_work, new a, new j */ 160 PLogObjectMemory(*B,(ainew[n]-n)*(sizeof(int)+sizeof(MatScalar))); 161 b->maxnz = b->nz = ainew[n]; 162 163 (*B)->factor = FACTOR_LU; 164 (*B)->info.factor_mallocs = realloc; 165 (*B)->info.fill_ratio_given = f; 166 if (ai[n] != 0) { 167 (*B)->info.fill_ratio_needed = ((PetscReal)ainew[n])/((PetscReal)ai[n]); 168 } else { 169 (*B)->info.fill_ratio_needed = 0.0; 170 } 171 172 173 PetscFunctionReturn(0); 174 } 175 176 /* ----------------------------------------------------------- */ 177 #undef __FUNC__ 178 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_N" 179 int MatLUFactorNumeric_SeqBAIJ_N(Mat A,Mat *B) 180 { 181 Mat C = *B; 182 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 183 IS isrow = b->row,isicol = b->icol; 184 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 185 int *ajtmpold,*ajtmp,nz,row,bslog,*ai=a->i,*aj=a->j,k,flg; 186 int *diag_offset=b->diag,diag,bs=a->bs,bs2 = a->bs2,*v_pivots,*pj; 187 MatScalar *ba = b->a,*aa = a->a,*pv,*v,*rtmp,*multiplier,*v_work,*pc,*w; 188 189 PetscFunctionBegin; 190 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 191 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 192 rtmp = (MatScalar*)PetscMalloc(bs2*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 193 ierr = PetscMemzero(rtmp,bs2*(n+1)*sizeof(MatScalar));CHKERRQ(ierr); 194 /* generate work space needed by dense LU factorization */ 195 v_work = (MatScalar*)PetscMalloc(bs*sizeof(int) + (bs+bs2)*sizeof(MatScalar));CHKPTRQ(v_work); 196 multiplier = v_work + bs; 197 v_pivots = (int*)(multiplier + bs2); 198 199 /* flops in while loop */ 200 bslog = 2*bs*bs2; 201 202 for (i=0; i<n; i++) { 203 nz = bi[i+1] - bi[i]; 204 ajtmp = bj + bi[i]; 205 for (j=0; j<nz; j++) { 206 ierr = PetscMemzero(rtmp+bs2*ajtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 207 } 208 /* load in initial (unfactored row) */ 209 nz = ai[r[i]+1] - ai[r[i]]; 210 ajtmpold = aj + ai[r[i]]; 211 v = aa + bs2*ai[r[i]]; 212 for (j=0; j<nz; j++) { 213 ierr = PetscMemcpy(rtmp+bs2*ic[ajtmpold[j]],v+bs2*j,bs2*sizeof(MatScalar));CHKERRQ(ierr); 214 } 215 row = *ajtmp++; 216 while (row < i) { 217 pc = rtmp + bs2*row; 218 /* if (*pc) { */ 219 for (flg=0,k=0; k<bs2; k++) { if (pc[k]!=0.0) { flg =1; break; }} 220 if (flg) { 221 pv = ba + bs2*diag_offset[row]; 222 pj = bj + diag_offset[row] + 1; 223 Kernel_A_gets_A_times_B(bs,pc,pv,multiplier); 224 nz = bi[row+1] - diag_offset[row] - 1; 225 pv += bs2; 226 for (j=0; j<nz; j++) { 227 Kernel_A_gets_A_minus_B_times_C(bs,rtmp+bs2*pj[j],pc,pv+bs2*j); 228 } 229 PLogFlops(bslog*(nz+1)-bs); 230 } 231 row = *ajtmp++; 232 } 233 /* finished row so stick it into b->a */ 234 pv = ba + bs2*bi[i]; 235 pj = bj + bi[i]; 236 nz = bi[i+1] - bi[i]; 237 for (j=0; j<nz; j++) { 238 ierr = PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 239 } 240 diag = diag_offset[i] - bi[i]; 241 /* invert diagonal block */ 242 w = pv + bs2*diag; 243 Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work); 244 } 245 246 ierr = PetscFree(rtmp);CHKERRQ(ierr); 247 ierr = PetscFree(v_work);CHKERRQ(ierr); 248 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 249 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 250 C->factor = FACTOR_LU; 251 C->assembled = PETSC_TRUE; 252 PLogFlops(1.3333*bs*bs2*b->mbs); /* from inverting diagonal blocks */ 253 PetscFunctionReturn(0); 254 } 255 /* ------------------------------------------------------------*/ 256 /* 257 Version for when blocks are 7 by 7 258 */ 259 #undef __FUNC__ 260 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_7" 261 int MatLUFactorNumeric_SeqBAIJ_7(Mat A,Mat *B) 262 { 263 Mat C = *B; 264 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 265 IS isrow = b->row,isicol = b->icol; 266 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 267 int *ajtmpold,*ajtmp,nz,row; 268 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 269 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 270 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 271 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 272 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 273 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 274 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 275 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 276 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 277 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 278 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 279 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 280 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 281 MatScalar *ba = b->a,*aa = a->a; 282 283 PetscFunctionBegin; 284 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 285 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 286 rtmp = (MatScalar*)PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 287 288 for (i=0; i<n; i++) { 289 nz = bi[i+1] - bi[i]; 290 ajtmp = bj + bi[i]; 291 for (j=0; j<nz; j++) { 292 x = rtmp+49*ajtmp[j]; 293 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 294 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 295 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 296 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 297 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 298 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 299 } 300 /* load in initial (unfactored row) */ 301 idx = r[i]; 302 nz = ai[idx+1] - ai[idx]; 303 ajtmpold = aj + ai[idx]; 304 v = aa + 49*ai[idx]; 305 for (j=0; j<nz; j++) { 306 x = rtmp+49*ic[ajtmpold[j]]; 307 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 308 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 309 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 310 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 311 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 312 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 313 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 314 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 315 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 316 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 317 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 318 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 319 x[48] = v[48]; 320 v += 49; 321 } 322 row = *ajtmp++; 323 while (row < i) { 324 pc = rtmp + 49*row; 325 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 326 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 327 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 328 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 329 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 330 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 331 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 332 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 333 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 334 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 335 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 336 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 337 p49 = pc[48]; 338 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 339 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 340 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 341 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 342 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 343 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 344 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 345 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 346 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 347 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 348 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 349 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 350 p49 != 0.0) { 351 pv = ba + 49*diag_offset[row]; 352 pj = bj + diag_offset[row] + 1; 353 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 354 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 355 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 356 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 357 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 358 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 359 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 360 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 361 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 362 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 363 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 364 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 365 x49 = pv[48]; 366 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 367 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 368 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 369 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 370 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 371 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 372 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 373 374 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 375 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 376 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 377 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 378 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 379 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 380 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 381 382 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 383 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 384 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 385 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 386 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 387 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 388 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 389 390 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 391 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 392 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 393 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 394 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 395 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 396 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 397 398 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 399 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 400 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 401 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 402 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 403 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 404 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 405 406 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 407 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 408 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 409 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 410 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 411 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 412 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 413 414 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 415 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 416 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 417 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 418 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 419 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 420 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 421 422 nz = bi[row+1] - diag_offset[row] - 1; 423 pv += 49; 424 for (j=0; j<nz; j++) { 425 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 426 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 427 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 428 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 429 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 430 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 431 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 432 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 433 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 434 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 435 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 436 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 437 x49 = pv[48]; 438 x = rtmp + 49*pj[j]; 439 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 440 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 441 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 442 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 443 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 444 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 445 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 446 447 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 448 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 449 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 450 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 451 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 452 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 453 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 454 455 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 456 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 457 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 458 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 459 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 460 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 461 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 462 463 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 464 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 465 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 466 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 467 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 468 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 469 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 470 471 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 472 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 473 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 474 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 475 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 476 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 477 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 478 479 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 480 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 481 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 482 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 483 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 484 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 485 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 486 487 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 488 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 489 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 490 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 491 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 492 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 493 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 494 pv += 49; 495 } 496 PLogFlops(686*nz+637); 497 } 498 row = *ajtmp++; 499 } 500 /* finished row so stick it into b->a */ 501 pv = ba + 49*bi[i]; 502 pj = bj + bi[i]; 503 nz = bi[i+1] - bi[i]; 504 for (j=0; j<nz; j++) { 505 x = rtmp+49*pj[j]; 506 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 507 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 508 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 509 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 510 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 511 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 512 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 513 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 514 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 515 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 516 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 517 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 518 pv[48] = x[48]; 519 pv += 49; 520 } 521 /* invert diagonal block */ 522 w = ba + 49*diag_offset[i]; 523 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 524 } 525 526 ierr = PetscFree(rtmp);CHKERRQ(ierr); 527 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 528 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 529 C->factor = FACTOR_LU; 530 C->assembled = PETSC_TRUE; 531 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 532 PetscFunctionReturn(0); 533 } 534 535 /* 536 Version for when blocks are 7 by 7 Using natural ordering 537 */ 538 #undef __FUNC__ 539 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_7_NaturalOrdering" 540 int MatLUFactorNumeric_SeqBAIJ_7_NaturalOrdering(Mat A,Mat *B) 541 { 542 Mat C = *B; 543 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 544 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 545 int *ajtmpold,*ajtmp,nz,row; 546 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 547 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 548 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 549 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 550 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 551 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 552 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 553 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 554 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 555 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 556 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 557 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 558 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 559 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 560 MatScalar *ba = b->a,*aa = a->a; 561 562 PetscFunctionBegin; 563 rtmp = (MatScalar*)PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 564 for (i=0; i<n; i++) { 565 nz = bi[i+1] - bi[i]; 566 ajtmp = bj + bi[i]; 567 for (j=0; j<nz; j++) { 568 x = rtmp+49*ajtmp[j]; 569 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 570 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 571 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 572 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 573 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 574 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 575 } 576 /* load in initial (unfactored row) */ 577 nz = ai[i+1] - ai[i]; 578 ajtmpold = aj + ai[i]; 579 v = aa + 49*ai[i]; 580 for (j=0; j<nz; j++) { 581 x = rtmp+49*ajtmpold[j]; 582 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 583 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 584 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 585 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 586 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 587 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 588 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 589 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 590 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 591 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 592 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 593 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 594 x[48] = v[48]; 595 v += 49; 596 } 597 row = *ajtmp++; 598 while (row < i) { 599 pc = rtmp + 49*row; 600 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 601 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 602 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 603 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 604 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 605 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 606 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 607 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 608 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 609 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 610 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 611 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 612 p49 = pc[48]; 613 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 614 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 615 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 616 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 617 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 618 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 619 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 620 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 621 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 622 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 623 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 624 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 625 p49 != 0.0) { 626 pv = ba + 49*diag_offset[row]; 627 pj = bj + diag_offset[row] + 1; 628 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 629 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 630 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 631 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 632 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 633 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 634 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 635 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 636 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 637 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 638 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 639 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 640 x49 = pv[48]; 641 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 642 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 643 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 644 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 645 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 646 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 647 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 648 649 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 650 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 651 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 652 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 653 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 654 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 655 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 656 657 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 658 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 659 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 660 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 661 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 662 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 663 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 664 665 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 666 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 667 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 668 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 669 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 670 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 671 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 672 673 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 674 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 675 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 676 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 677 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 678 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 679 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 680 681 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 682 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 683 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 684 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 685 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 686 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 687 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 688 689 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 690 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 691 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 692 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 693 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 694 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 695 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 696 697 nz = bi[row+1] - diag_offset[row] - 1; 698 pv += 49; 699 for (j=0; j<nz; j++) { 700 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 701 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 702 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 703 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 704 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 705 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 706 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 707 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 708 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 709 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 710 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 711 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 712 x49 = pv[48]; 713 x = rtmp + 49*pj[j]; 714 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 715 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 716 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 717 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 718 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 719 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 720 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 721 722 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 723 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 724 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 725 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 726 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 727 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 728 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 729 730 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 731 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 732 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 733 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 734 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 735 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 736 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 737 738 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 739 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 740 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 741 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 742 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 743 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 744 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 745 746 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 747 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 748 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 749 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 750 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 751 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 752 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 753 754 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 755 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 756 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 757 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 758 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 759 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 760 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 761 762 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 763 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 764 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 765 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 766 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 767 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 768 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 769 pv += 49; 770 } 771 PLogFlops(686*nz+637); 772 } 773 row = *ajtmp++; 774 } 775 /* finished row so stick it into b->a */ 776 pv = ba + 49*bi[i]; 777 pj = bj + bi[i]; 778 nz = bi[i+1] - bi[i]; 779 for (j=0; j<nz; j++) { 780 x = rtmp+49*pj[j]; 781 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 782 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 783 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 784 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 785 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 786 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 787 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 788 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 789 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 790 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 791 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 792 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 793 pv[48] = x[48]; 794 pv += 49; 795 } 796 /* invert diagonal block */ 797 w = ba + 49*diag_offset[i]; 798 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 799 } 800 801 ierr = PetscFree(rtmp);CHKERRQ(ierr); 802 C->factor = FACTOR_LU; 803 C->assembled = PETSC_TRUE; 804 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 805 PetscFunctionReturn(0); 806 } 807 808 /* ------------------------------------------------------------*/ 809 /* 810 Version for when blocks are 6 by 6 811 */ 812 #undef __FUNC__ 813 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_6" 814 int MatLUFactorNumeric_SeqBAIJ_6(Mat A,Mat *B) 815 { 816 Mat C = *B; 817 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 818 IS isrow = b->row,isicol = b->icol; 819 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 820 int *ajtmpold,*ajtmp,nz,row; 821 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 822 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 823 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 824 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 825 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 826 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 827 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 828 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 829 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 830 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 831 MatScalar *ba = b->a,*aa = a->a; 832 833 PetscFunctionBegin; 834 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 835 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 836 rtmp = (MatScalar*)PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 837 838 for (i=0; i<n; i++) { 839 nz = bi[i+1] - bi[i]; 840 ajtmp = bj + bi[i]; 841 for (j=0; j<nz; j++) { 842 x = rtmp+36*ajtmp[j]; 843 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 844 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 845 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 846 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 847 x[34] = x[35] = 0.0 ; 848 } 849 /* load in initial (unfactored row) */ 850 idx = r[i]; 851 nz = ai[idx+1] - ai[idx]; 852 ajtmpold = aj + ai[idx]; 853 v = aa + 36*ai[idx]; 854 for (j=0; j<nz; j++) { 855 x = rtmp+36*ic[ajtmpold[j]]; 856 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 857 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 858 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 859 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 860 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 861 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 862 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 863 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 864 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 865 v += 36; 866 } 867 row = *ajtmp++; 868 while (row < i) { 869 pc = rtmp + 36*row; 870 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 871 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 872 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 873 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 874 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 875 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 876 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 877 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 878 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 879 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 880 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 881 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 882 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 883 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 884 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 885 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 886 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 887 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 888 pv = ba + 36*diag_offset[row]; 889 pj = bj + diag_offset[row] + 1; 890 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 891 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 892 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 893 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 894 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 895 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 896 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 897 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 898 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 899 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 900 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 901 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 902 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 903 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 904 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 905 906 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 907 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 908 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 909 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 910 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 911 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 912 913 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 914 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 915 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 916 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 917 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 918 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 919 920 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 921 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 922 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 923 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 924 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 925 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 926 927 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 928 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 929 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 930 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 931 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 932 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 933 934 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 935 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 936 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 937 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 938 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 939 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 940 941 nz = bi[row+1] - diag_offset[row] - 1; 942 pv += 36; 943 for (j=0; j<nz; j++) { 944 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 945 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 946 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 947 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 948 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 949 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 950 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 951 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 952 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 953 x = rtmp + 36*pj[j]; 954 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 955 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 956 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 957 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 958 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 959 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 960 961 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 962 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 963 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 964 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 965 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 966 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 967 968 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 969 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 970 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 971 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 972 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 973 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 974 975 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 976 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 977 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 978 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 979 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 980 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 981 982 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 983 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 984 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 985 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 986 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 987 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 988 989 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 990 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 991 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 992 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 993 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 994 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 995 996 pv += 36; 997 } 998 PLogFlops(432*nz+396); 999 } 1000 row = *ajtmp++; 1001 } 1002 /* finished row so stick it into b->a */ 1003 pv = ba + 36*bi[i]; 1004 pj = bj + bi[i]; 1005 nz = bi[i+1] - bi[i]; 1006 for (j=0; j<nz; j++) { 1007 x = rtmp+36*pj[j]; 1008 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1009 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1010 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1011 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1012 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1013 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1014 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1015 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1016 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1017 pv += 36; 1018 } 1019 /* invert diagonal block */ 1020 w = ba + 36*diag_offset[i]; 1021 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1022 } 1023 1024 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1025 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1026 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1027 C->factor = FACTOR_LU; 1028 C->assembled = PETSC_TRUE; 1029 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1030 PetscFunctionReturn(0); 1031 } 1032 /* 1033 Version for when blocks are 6 by 6 Using natural ordering 1034 */ 1035 #undef __FUNC__ 1036 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_6_NaturalOrdering" 1037 int MatLUFactorNumeric_SeqBAIJ_6_NaturalOrdering(Mat A,Mat *B) 1038 { 1039 Mat C = *B; 1040 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1041 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1042 int *ajtmpold,*ajtmp,nz,row; 1043 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1044 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1045 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1046 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1047 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1048 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1049 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1050 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1051 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 1052 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 1053 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 1054 MatScalar *ba = b->a,*aa = a->a; 1055 1056 PetscFunctionBegin; 1057 rtmp = (MatScalar*)PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1058 for (i=0; i<n; i++) { 1059 nz = bi[i+1] - bi[i]; 1060 ajtmp = bj + bi[i]; 1061 for (j=0; j<nz; j++) { 1062 x = rtmp+36*ajtmp[j]; 1063 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1064 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1065 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 1066 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 1067 x[34] = x[35] = 0.0 ; 1068 } 1069 /* load in initial (unfactored row) */ 1070 nz = ai[i+1] - ai[i]; 1071 ajtmpold = aj + ai[i]; 1072 v = aa + 36*ai[i]; 1073 for (j=0; j<nz; j++) { 1074 x = rtmp+36*ajtmpold[j]; 1075 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1076 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 1077 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 1078 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 1079 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 1080 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1081 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 1082 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 1083 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 1084 v += 36; 1085 } 1086 row = *ajtmp++; 1087 while (row < i) { 1088 pc = rtmp + 36*row; 1089 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1090 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 1091 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 1092 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 1093 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 1094 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1095 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 1096 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 1097 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 1098 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 1099 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 1100 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 1101 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 1102 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 1103 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 1104 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 1105 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 1106 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 1107 pv = ba + 36*diag_offset[row]; 1108 pj = bj + diag_offset[row] + 1; 1109 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1110 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1111 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1112 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1113 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1114 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1115 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1116 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1117 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1118 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 1119 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 1120 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 1121 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 1122 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 1123 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 1124 1125 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 1126 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 1127 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 1128 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 1129 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 1130 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 1131 1132 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 1133 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 1134 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 1135 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 1136 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 1137 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 1138 1139 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 1140 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 1141 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 1142 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 1143 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 1144 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 1145 1146 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 1147 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 1148 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 1149 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 1150 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 1151 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 1152 1153 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 1154 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 1155 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 1156 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 1157 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 1158 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 1159 1160 nz = bi[row+1] - diag_offset[row] - 1; 1161 pv += 36; 1162 for (j=0; j<nz; j++) { 1163 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1164 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1165 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1166 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1167 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1168 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1169 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1170 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1171 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1172 x = rtmp + 36*pj[j]; 1173 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 1174 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 1175 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 1176 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 1177 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 1178 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 1179 1180 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 1181 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 1182 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 1183 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 1184 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 1185 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 1186 1187 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 1188 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 1189 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 1190 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 1191 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 1192 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 1193 1194 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 1195 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 1196 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 1197 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 1198 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 1199 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 1200 1201 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 1202 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 1203 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 1204 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 1205 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 1206 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 1207 1208 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 1209 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 1210 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 1211 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 1212 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 1213 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 1214 1215 pv += 36; 1216 } 1217 PLogFlops(432*nz+396); 1218 } 1219 row = *ajtmp++; 1220 } 1221 /* finished row so stick it into b->a */ 1222 pv = ba + 36*bi[i]; 1223 pj = bj + bi[i]; 1224 nz = bi[i+1] - bi[i]; 1225 for (j=0; j<nz; j++) { 1226 x = rtmp+36*pj[j]; 1227 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1228 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1229 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1230 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1231 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1232 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1233 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1234 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1235 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1236 pv += 36; 1237 } 1238 /* invert diagonal block */ 1239 w = ba + 36*diag_offset[i]; 1240 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1241 } 1242 1243 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1244 C->factor = FACTOR_LU; 1245 C->assembled = PETSC_TRUE; 1246 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1247 PetscFunctionReturn(0); 1248 } 1249 1250 /* ------------------------------------------------------------*/ 1251 /* 1252 Version for when blocks are 5 by 5 1253 */ 1254 #undef __FUNC__ 1255 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_5" 1256 int MatLUFactorNumeric_SeqBAIJ_5(Mat A,Mat *B) 1257 { 1258 Mat C = *B; 1259 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1260 IS isrow = b->row,isicol = b->icol; 1261 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1262 int *ajtmpold,*ajtmp,nz,row; 1263 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 1264 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1265 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1266 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1267 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 1268 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 1269 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1270 MatScalar *ba = b->a,*aa = a->a; 1271 1272 PetscFunctionBegin; 1273 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1274 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1275 rtmp = (MatScalar*)PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1276 1277 for (i=0; i<n; i++) { 1278 nz = bi[i+1] - bi[i]; 1279 ajtmp = bj + bi[i]; 1280 for (j=0; j<nz; j++) { 1281 x = rtmp+25*ajtmp[j]; 1282 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1283 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1284 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1285 } 1286 /* load in initial (unfactored row) */ 1287 idx = r[i]; 1288 nz = ai[idx+1] - ai[idx]; 1289 ajtmpold = aj + ai[idx]; 1290 v = aa + 25*ai[idx]; 1291 for (j=0; j<nz; j++) { 1292 x = rtmp+25*ic[ajtmpold[j]]; 1293 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1294 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1295 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1296 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; 1297 x[18] = v[18]; x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; 1298 x[22] = v[22]; x[23] = v[23]; x[24] = v[24]; 1299 v += 25; 1300 } 1301 row = *ajtmp++; 1302 while (row < i) { 1303 pc = rtmp + 25*row; 1304 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1305 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1306 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1307 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; 1308 p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1309 p25 = pc[24]; 1310 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1311 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1312 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1313 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || 1314 p20 != 0.0 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || 1315 p24 != 0.0 || p25 != 0.0) { 1316 pv = ba + 25*diag_offset[row]; 1317 pj = bj + diag_offset[row] + 1; 1318 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1319 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1320 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1321 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1322 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; 1323 x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1324 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1325 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1326 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1327 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1328 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1329 1330 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1331 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1332 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1333 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1334 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1335 1336 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1337 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1338 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1339 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1340 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1341 1342 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1343 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1344 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1345 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1346 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1347 1348 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1349 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1350 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1351 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1352 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1353 1354 nz = bi[row+1] - diag_offset[row] - 1; 1355 pv += 25; 1356 for (j=0; j<nz; j++) { 1357 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1358 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1359 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1360 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; 1361 x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; 1362 x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1363 x = rtmp + 25*pj[j]; 1364 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1365 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1366 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1367 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1368 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1369 1370 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1371 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1372 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1373 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1374 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1375 1376 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1377 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1378 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1379 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1380 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1381 1382 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1383 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1384 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1385 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1386 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1387 1388 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1389 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1390 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1391 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1392 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1393 1394 pv += 25; 1395 } 1396 PLogFlops(250*nz+225); 1397 } 1398 row = *ajtmp++; 1399 } 1400 /* finished row so stick it into b->a */ 1401 pv = ba + 25*bi[i]; 1402 pj = bj + bi[i]; 1403 nz = bi[i+1] - bi[i]; 1404 for (j=0; j<nz; j++) { 1405 x = rtmp+25*pj[j]; 1406 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1407 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1408 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1409 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; 1410 pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; 1411 pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; pv[24] = x[24]; 1412 pv += 25; 1413 } 1414 /* invert diagonal block */ 1415 w = ba + 25*diag_offset[i]; 1416 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1417 } 1418 1419 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1420 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1421 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1422 C->factor = FACTOR_LU; 1423 C->assembled = PETSC_TRUE; 1424 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1425 PetscFunctionReturn(0); 1426 } 1427 /* 1428 Version for when blocks are 5 by 5 Using natural ordering 1429 */ 1430 #undef __FUNC__ 1431 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_5_NaturalOrdering" 1432 int MatLUFactorNumeric_SeqBAIJ_5_NaturalOrdering(Mat A,Mat *B) 1433 { 1434 Mat C = *B; 1435 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1436 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1437 int *ajtmpold,*ajtmp,nz,row; 1438 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1439 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1440 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1441 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1442 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1443 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1444 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1445 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1446 MatScalar *ba = b->a,*aa = a->a; 1447 1448 PetscFunctionBegin; 1449 rtmp = (MatScalar*)PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1450 for (i=0; i<n; i++) { 1451 nz = bi[i+1] - bi[i]; 1452 ajtmp = bj + bi[i]; 1453 for (j=0; j<nz; j++) { 1454 x = rtmp+25*ajtmp[j]; 1455 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1456 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1457 x[16] = x[17] = x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1458 } 1459 /* load in initial (unfactored row) */ 1460 nz = ai[i+1] - ai[i]; 1461 ajtmpold = aj + ai[i]; 1462 v = aa + 25*ai[i]; 1463 for (j=0; j<nz; j++) { 1464 x = rtmp+25*ajtmpold[j]; 1465 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1466 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1467 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1468 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; 1469 x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1470 x[24] = v[24]; 1471 v += 25; 1472 } 1473 row = *ajtmp++; 1474 while (row < i) { 1475 pc = rtmp + 25*row; 1476 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1477 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1478 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1479 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; 1480 p19 = pc[18]; p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; 1481 p24 = pc[23]; p25 = pc[24]; 1482 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1483 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1484 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1485 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 1486 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || p25 != 0.0) { 1487 pv = ba + 25*diag_offset[row]; 1488 pj = bj + diag_offset[row] + 1; 1489 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1490 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1491 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1492 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; 1493 x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1494 x25 = pv[24]; 1495 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1496 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1497 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1498 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1499 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1500 1501 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1502 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1503 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1504 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1505 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1506 1507 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1508 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1509 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1510 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1511 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1512 1513 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1514 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1515 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1516 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1517 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1518 1519 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1520 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1521 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1522 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1523 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1524 1525 nz = bi[row+1] - diag_offset[row] - 1; 1526 pv += 25; 1527 for (j=0; j<nz; j++) { 1528 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1529 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1530 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1531 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1532 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; 1533 x24 = pv[23]; x25 = pv[24]; 1534 x = rtmp + 25*pj[j]; 1535 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1536 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1537 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1538 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1539 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1540 1541 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1542 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1543 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1544 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1545 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1546 1547 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1548 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1549 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1550 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1551 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1552 1553 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1554 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1555 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1556 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1557 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1558 1559 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1560 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1561 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1562 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1563 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1564 pv += 25; 1565 } 1566 PLogFlops(250*nz+225); 1567 } 1568 row = *ajtmp++; 1569 } 1570 /* finished row so stick it into b->a */ 1571 pv = ba + 25*bi[i]; 1572 pj = bj + bi[i]; 1573 nz = bi[i+1] - bi[i]; 1574 for (j=0; j<nz; j++) { 1575 x = rtmp+25*pj[j]; 1576 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1577 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1578 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1579 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; pv[17] = x[17]; 1580 pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; 1581 pv[23] = x[23]; pv[24] = x[24]; 1582 pv += 25; 1583 } 1584 /* invert diagonal block */ 1585 w = ba + 25*diag_offset[i]; 1586 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1587 } 1588 1589 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1590 C->factor = FACTOR_LU; 1591 C->assembled = PETSC_TRUE; 1592 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1593 PetscFunctionReturn(0); 1594 } 1595 1596 /* ------------------------------------------------------------*/ 1597 /* 1598 Version for when blocks are 4 by 4 1599 */ 1600 #undef __FUNC__ 1601 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_4" 1602 int MatLUFactorNumeric_SeqBAIJ_4(Mat A,Mat *B) 1603 { 1604 Mat C = *B; 1605 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1606 IS isrow = b->row,isicol = b->icol; 1607 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1608 int *ajtmpold,*ajtmp,nz,row; 1609 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 1610 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1611 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1612 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1613 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1614 MatScalar m13,m14,m15,m16; 1615 MatScalar *ba = b->a,*aa = a->a; 1616 1617 PetscFunctionBegin; 1618 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1619 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1620 rtmp = (MatScalar*)PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1621 1622 for (i=0; i<n; i++) { 1623 nz = bi[i+1] - bi[i]; 1624 ajtmp = bj + bi[i]; 1625 for (j=0; j<nz; j++) { 1626 x = rtmp+16*ajtmp[j]; 1627 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1628 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1629 } 1630 /* load in initial (unfactored row) */ 1631 idx = r[i]; 1632 nz = ai[idx+1] - ai[idx]; 1633 ajtmpold = aj + ai[idx]; 1634 v = aa + 16*ai[idx]; 1635 for (j=0; j<nz; j++) { 1636 x = rtmp+16*ic[ajtmpold[j]]; 1637 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1638 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1639 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1640 x[14] = v[14]; x[15] = v[15]; 1641 v += 16; 1642 } 1643 row = *ajtmp++; 1644 while (row < i) { 1645 pc = rtmp + 16*row; 1646 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1647 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1648 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1649 p15 = pc[14]; p16 = pc[15]; 1650 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1651 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1652 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1653 || p16 != 0.0) { 1654 pv = ba + 16*diag_offset[row]; 1655 pj = bj + diag_offset[row] + 1; 1656 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1657 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1658 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1659 x15 = pv[14]; x16 = pv[15]; 1660 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1661 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1662 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1663 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1664 1665 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1666 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1667 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1668 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1669 1670 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1671 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1672 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1673 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1674 1675 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1676 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1677 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1678 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1679 1680 nz = bi[row+1] - diag_offset[row] - 1; 1681 pv += 16; 1682 for (j=0; j<nz; j++) { 1683 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1684 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1685 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1686 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1687 x = rtmp + 16*pj[j]; 1688 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1689 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1690 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1691 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1692 1693 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1694 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1695 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1696 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1697 1698 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1699 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1700 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1701 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1702 1703 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1704 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1705 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1706 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1707 1708 pv += 16; 1709 } 1710 PLogFlops(128*nz+112); 1711 } 1712 row = *ajtmp++; 1713 } 1714 /* finished row so stick it into b->a */ 1715 pv = ba + 16*bi[i]; 1716 pj = bj + bi[i]; 1717 nz = bi[i+1] - bi[i]; 1718 for (j=0; j<nz; j++) { 1719 x = rtmp+16*pj[j]; 1720 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1721 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1722 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1723 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1724 pv += 16; 1725 } 1726 /* invert diagonal block */ 1727 w = ba + 16*diag_offset[i]; 1728 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1729 } 1730 1731 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1732 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1733 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1734 C->factor = FACTOR_LU; 1735 C->assembled = PETSC_TRUE; 1736 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1737 PetscFunctionReturn(0); 1738 } 1739 /* 1740 Version for when blocks are 4 by 4 Using natural ordering 1741 */ 1742 #undef __FUNC__ 1743 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_4_NaturalOrdering" 1744 int MatLUFactorNumeric_SeqBAIJ_4_NaturalOrdering(Mat A,Mat *B) 1745 { 1746 Mat C = *B; 1747 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1748 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1749 int *ajtmpold,*ajtmp,nz,row; 1750 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1751 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1752 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1753 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1754 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1755 MatScalar m13,m14,m15,m16; 1756 MatScalar *ba = b->a,*aa = a->a; 1757 1758 PetscFunctionBegin; 1759 rtmp = (MatScalar*)PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1760 1761 for (i=0; i<n; i++) { 1762 nz = bi[i+1] - bi[i]; 1763 ajtmp = bj + bi[i]; 1764 for (j=0; j<nz; j++) { 1765 x = rtmp+16*ajtmp[j]; 1766 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1767 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1768 } 1769 /* load in initial (unfactored row) */ 1770 nz = ai[i+1] - ai[i]; 1771 ajtmpold = aj + ai[i]; 1772 v = aa + 16*ai[i]; 1773 for (j=0; j<nz; j++) { 1774 x = rtmp+16*ajtmpold[j]; 1775 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1776 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1777 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1778 x[14] = v[14]; x[15] = v[15]; 1779 v += 16; 1780 } 1781 row = *ajtmp++; 1782 while (row < i) { 1783 pc = rtmp + 16*row; 1784 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1785 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1786 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1787 p15 = pc[14]; p16 = pc[15]; 1788 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1789 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1790 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1791 || p16 != 0.0) { 1792 pv = ba + 16*diag_offset[row]; 1793 pj = bj + diag_offset[row] + 1; 1794 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1795 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1796 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1797 x15 = pv[14]; x16 = pv[15]; 1798 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1799 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1800 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1801 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1802 1803 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1804 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1805 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1806 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1807 1808 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1809 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1810 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1811 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1812 1813 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1814 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1815 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1816 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1817 1818 nz = bi[row+1] - diag_offset[row] - 1; 1819 pv += 16; 1820 for (j=0; j<nz; j++) { 1821 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1822 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1823 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1824 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1825 x = rtmp + 16*pj[j]; 1826 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1827 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1828 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1829 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1830 1831 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1832 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1833 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1834 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1835 1836 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1837 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1838 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1839 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1840 1841 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1842 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1843 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1844 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1845 1846 pv += 16; 1847 } 1848 PLogFlops(128*nz+112); 1849 } 1850 row = *ajtmp++; 1851 } 1852 /* finished row so stick it into b->a */ 1853 pv = ba + 16*bi[i]; 1854 pj = bj + bi[i]; 1855 nz = bi[i+1] - bi[i]; 1856 for (j=0; j<nz; j++) { 1857 x = rtmp+16*pj[j]; 1858 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1859 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1860 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1861 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1862 pv += 16; 1863 } 1864 /* invert diagonal block */ 1865 w = ba + 16*diag_offset[i]; 1866 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1867 } 1868 1869 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1870 C->factor = FACTOR_LU; 1871 C->assembled = PETSC_TRUE; 1872 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1873 PetscFunctionReturn(0); 1874 } 1875 1876 1877 /* ------------------------------------------------------------*/ 1878 /* 1879 Version for when blocks are 3 by 3 1880 */ 1881 #undef __FUNC__ 1882 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_3" 1883 int MatLUFactorNumeric_SeqBAIJ_3(Mat A,Mat *B) 1884 { 1885 Mat C = *B; 1886 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1887 IS isrow = b->row,isicol = b->icol; 1888 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1889 int *ajtmpold,*ajtmp,nz,row,*ai=a->i,*aj=a->j; 1890 int *diag_offset = b->diag,idx,*pj; 1891 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1892 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1893 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 1894 MatScalar *ba = b->a,*aa = a->a; 1895 1896 PetscFunctionBegin; 1897 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1898 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1899 rtmp = (MatScalar*)PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1900 1901 for (i=0; i<n; i++) { 1902 nz = bi[i+1] - bi[i]; 1903 ajtmp = bj + bi[i]; 1904 for (j=0; j<nz; j++) { 1905 x = rtmp + 9*ajtmp[j]; 1906 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 1907 } 1908 /* load in initial (unfactored row) */ 1909 idx = r[i]; 1910 nz = ai[idx+1] - ai[idx]; 1911 ajtmpold = aj + ai[idx]; 1912 v = aa + 9*ai[idx]; 1913 for (j=0; j<nz; j++) { 1914 x = rtmp + 9*ic[ajtmpold[j]]; 1915 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1916 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1917 v += 9; 1918 } 1919 row = *ajtmp++; 1920 while (row < i) { 1921 pc = rtmp + 9*row; 1922 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1923 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1924 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1925 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 1926 pv = ba + 9*diag_offset[row]; 1927 pj = bj + diag_offset[row] + 1; 1928 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1929 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1930 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 1931 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 1932 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 1933 1934 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 1935 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 1936 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 1937 1938 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 1939 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 1940 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 1941 nz = bi[row+1] - diag_offset[row] - 1; 1942 pv += 9; 1943 for (j=0; j<nz; j++) { 1944 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1945 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1946 x = rtmp + 9*pj[j]; 1947 x[0] -= m1*x1 + m4*x2 + m7*x3; 1948 x[1] -= m2*x1 + m5*x2 + m8*x3; 1949 x[2] -= m3*x1 + m6*x2 + m9*x3; 1950 1951 x[3] -= m1*x4 + m4*x5 + m7*x6; 1952 x[4] -= m2*x4 + m5*x5 + m8*x6; 1953 x[5] -= m3*x4 + m6*x5 + m9*x6; 1954 1955 x[6] -= m1*x7 + m4*x8 + m7*x9; 1956 x[7] -= m2*x7 + m5*x8 + m8*x9; 1957 x[8] -= m3*x7 + m6*x8 + m9*x9; 1958 pv += 9; 1959 } 1960 PLogFlops(54*nz+36); 1961 } 1962 row = *ajtmp++; 1963 } 1964 /* finished row so stick it into b->a */ 1965 pv = ba + 9*bi[i]; 1966 pj = bj + bi[i]; 1967 nz = bi[i+1] - bi[i]; 1968 for (j=0; j<nz; j++) { 1969 x = rtmp + 9*pj[j]; 1970 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1971 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1972 pv += 9; 1973 } 1974 /* invert diagonal block */ 1975 w = ba + 9*diag_offset[i]; 1976 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 1977 } 1978 1979 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1980 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1981 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1982 C->factor = FACTOR_LU; 1983 C->assembled = PETSC_TRUE; 1984 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 1985 PetscFunctionReturn(0); 1986 } 1987 /* 1988 Version for when blocks are 3 by 3 Using natural ordering 1989 */ 1990 #undef __FUNC__ 1991 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_3_NaturalOrdering" 1992 int MatLUFactorNumeric_SeqBAIJ_3_NaturalOrdering(Mat A,Mat *B) 1993 { 1994 Mat C = *B; 1995 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1996 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1997 int *ajtmpold,*ajtmp,nz,row; 1998 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1999 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2000 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 2001 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 2002 MatScalar *ba = b->a,*aa = a->a; 2003 2004 PetscFunctionBegin; 2005 rtmp = (MatScalar*)PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2006 2007 for (i=0; i<n; i++) { 2008 nz = bi[i+1] - bi[i]; 2009 ajtmp = bj + bi[i]; 2010 for (j=0; j<nz; j++) { 2011 x = rtmp+9*ajtmp[j]; 2012 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 2013 } 2014 /* load in initial (unfactored row) */ 2015 nz = ai[i+1] - ai[i]; 2016 ajtmpold = aj + ai[i]; 2017 v = aa + 9*ai[i]; 2018 for (j=0; j<nz; j++) { 2019 x = rtmp+9*ajtmpold[j]; 2020 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2021 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 2022 v += 9; 2023 } 2024 row = *ajtmp++; 2025 while (row < i) { 2026 pc = rtmp + 9*row; 2027 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2028 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 2029 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 2030 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 2031 pv = ba + 9*diag_offset[row]; 2032 pj = bj + diag_offset[row] + 1; 2033 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2034 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2035 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 2036 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 2037 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 2038 2039 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 2040 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 2041 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 2042 2043 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 2044 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 2045 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 2046 2047 nz = bi[row+1] - diag_offset[row] - 1; 2048 pv += 9; 2049 for (j=0; j<nz; j++) { 2050 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2051 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2052 x = rtmp + 9*pj[j]; 2053 x[0] -= m1*x1 + m4*x2 + m7*x3; 2054 x[1] -= m2*x1 + m5*x2 + m8*x3; 2055 x[2] -= m3*x1 + m6*x2 + m9*x3; 2056 2057 x[3] -= m1*x4 + m4*x5 + m7*x6; 2058 x[4] -= m2*x4 + m5*x5 + m8*x6; 2059 x[5] -= m3*x4 + m6*x5 + m9*x6; 2060 2061 x[6] -= m1*x7 + m4*x8 + m7*x9; 2062 x[7] -= m2*x7 + m5*x8 + m8*x9; 2063 x[8] -= m3*x7 + m6*x8 + m9*x9; 2064 pv += 9; 2065 } 2066 PLogFlops(54*nz+36); 2067 } 2068 row = *ajtmp++; 2069 } 2070 /* finished row so stick it into b->a */ 2071 pv = ba + 9*bi[i]; 2072 pj = bj + bi[i]; 2073 nz = bi[i+1] - bi[i]; 2074 for (j=0; j<nz; j++) { 2075 x = rtmp+9*pj[j]; 2076 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2077 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 2078 pv += 9; 2079 } 2080 /* invert diagonal block */ 2081 w = ba + 9*diag_offset[i]; 2082 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 2083 } 2084 2085 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2086 C->factor = FACTOR_LU; 2087 C->assembled = PETSC_TRUE; 2088 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 2089 PetscFunctionReturn(0); 2090 } 2091 2092 /* ------------------------------------------------------------*/ 2093 /* 2094 Version for when blocks are 2 by 2 2095 */ 2096 #undef __FUNC__ 2097 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_2" 2098 int MatLUFactorNumeric_SeqBAIJ_2(Mat A,Mat *B) 2099 { 2100 Mat C = *B; 2101 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 2102 IS isrow = b->row,isicol = b->icol; 2103 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 2104 int *ajtmpold,*ajtmp,nz,row; 2105 int *diag_offset=b->diag,idx,*ai=a->i,*aj=a->j,*pj; 2106 MatScalar *pv,*v,*rtmp,m1,m2,m3,m4,*pc,*w,*x,x1,x2,x3,x4; 2107 MatScalar p1,p2,p3,p4; 2108 MatScalar *ba = b->a,*aa = a->a; 2109 2110 PetscFunctionBegin; 2111 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 2112 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 2113 rtmp = (MatScalar*)PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2114 2115 for (i=0; i<n; i++) { 2116 nz = bi[i+1] - bi[i]; 2117 ajtmp = bj + bi[i]; 2118 for (j=0; j<nz; j++) { 2119 x = rtmp+4*ajtmp[j]; x[0] = x[1] = x[2] = x[3] = 0.0; 2120 } 2121 /* load in initial (unfactored row) */ 2122 idx = r[i]; 2123 nz = ai[idx+1] - ai[idx]; 2124 ajtmpold = aj + ai[idx]; 2125 v = aa + 4*ai[idx]; 2126 for (j=0; j<nz; j++) { 2127 x = rtmp+4*ic[ajtmpold[j]]; 2128 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2129 v += 4; 2130 } 2131 row = *ajtmp++; 2132 while (row < i) { 2133 pc = rtmp + 4*row; 2134 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2135 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2136 pv = ba + 4*diag_offset[row]; 2137 pj = bj + diag_offset[row] + 1; 2138 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2139 pc[0] = m1 = p1*x1 + p3*x2; 2140 pc[1] = m2 = p2*x1 + p4*x2; 2141 pc[2] = m3 = p1*x3 + p3*x4; 2142 pc[3] = m4 = p2*x3 + p4*x4; 2143 nz = bi[row+1] - diag_offset[row] - 1; 2144 pv += 4; 2145 for (j=0; j<nz; j++) { 2146 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2147 x = rtmp + 4*pj[j]; 2148 x[0] -= m1*x1 + m3*x2; 2149 x[1] -= m2*x1 + m4*x2; 2150 x[2] -= m1*x3 + m3*x4; 2151 x[3] -= m2*x3 + m4*x4; 2152 pv += 4; 2153 } 2154 PLogFlops(16*nz+12); 2155 } 2156 row = *ajtmp++; 2157 } 2158 /* finished row so stick it into b->a */ 2159 pv = ba + 4*bi[i]; 2160 pj = bj + bi[i]; 2161 nz = bi[i+1] - bi[i]; 2162 for (j=0; j<nz; j++) { 2163 x = rtmp+4*pj[j]; 2164 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2165 pv += 4; 2166 } 2167 /* invert diagonal block */ 2168 w = ba + 4*diag_offset[i]; 2169 ierr = Kernel_A_gets_inverse_A_2(w);CHKERRQ(ierr); 2170 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2171 } 2172 2173 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2174 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 2175 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 2176 C->factor = FACTOR_LU; 2177 C->assembled = PETSC_TRUE; 2178 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2179 PetscFunctionReturn(0); 2180 } 2181 /* 2182 Version for when blocks are 2 by 2 Using natural ordering 2183 */ 2184 #undef __FUNC__ 2185 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering" 2186 int MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering(Mat A,Mat *B) 2187 { 2188 Mat C = *B; 2189 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 2190 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 2191 int *ajtmpold,*ajtmp,nz,row; 2192 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 2193 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2194 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,x1,x2,x3,x4; 2195 MatScalar *ba = b->a,*aa = a->a; 2196 2197 PetscFunctionBegin; 2198 rtmp = (MatScalar*)PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2199 2200 for (i=0; i<n; i++) { 2201 nz = bi[i+1] - bi[i]; 2202 ajtmp = bj + bi[i]; 2203 for (j=0; j<nz; j++) { 2204 x = rtmp+4*ajtmp[j]; 2205 x[0] = x[1] = x[2] = x[3] = 0.0; 2206 } 2207 /* load in initial (unfactored row) */ 2208 nz = ai[i+1] - ai[i]; 2209 ajtmpold = aj + ai[i]; 2210 v = aa + 4*ai[i]; 2211 for (j=0; j<nz; j++) { 2212 x = rtmp+4*ajtmpold[j]; 2213 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2214 v += 4; 2215 } 2216 row = *ajtmp++; 2217 while (row < i) { 2218 pc = rtmp + 4*row; 2219 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2220 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2221 pv = ba + 4*diag_offset[row]; 2222 pj = bj + diag_offset[row] + 1; 2223 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2224 pc[0] = m1 = p1*x1 + p3*x2; 2225 pc[1] = m2 = p2*x1 + p4*x2; 2226 pc[2] = m3 = p1*x3 + p3*x4; 2227 pc[3] = m4 = p2*x3 + p4*x4; 2228 nz = bi[row+1] - diag_offset[row] - 1; 2229 pv += 4; 2230 for (j=0; j<nz; j++) { 2231 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2232 x = rtmp + 4*pj[j]; 2233 x[0] -= m1*x1 + m3*x2; 2234 x[1] -= m2*x1 + m4*x2; 2235 x[2] -= m1*x3 + m3*x4; 2236 x[3] -= m2*x3 + m4*x4; 2237 pv += 4; 2238 } 2239 PLogFlops(16*nz+12); 2240 } 2241 row = *ajtmp++; 2242 } 2243 /* finished row so stick it into b->a */ 2244 pv = ba + 4*bi[i]; 2245 pj = bj + bi[i]; 2246 nz = bi[i+1] - bi[i]; 2247 for (j=0; j<nz; j++) { 2248 x = rtmp+4*pj[j]; 2249 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2250 pv += 4; 2251 } 2252 /* invert diagonal block */ 2253 w = ba + 4*diag_offset[i]; 2254 ierr = Kernel_A_gets_inverse_A_2(w);CHKERRQ(ierr); 2255 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2256 } 2257 2258 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2259 C->factor = FACTOR_LU; 2260 C->assembled = PETSC_TRUE; 2261 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2262 PetscFunctionReturn(0); 2263 } 2264 2265 /* ----------------------------------------------------------- */ 2266 /* 2267 Version for when blocks are 1 by 1. 2268 */ 2269 #undef __FUNC__ 2270 #define __FUNC__ "MatLUFactorNumeric_SeqBAIJ_1" 2271 int MatLUFactorNumeric_SeqBAIJ_1(Mat A,Mat *B) 2272 { 2273 Mat C = *B; 2274 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 2275 IS isrow = b->row,isicol = b->icol; 2276 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 2277 int *ajtmpold,*ajtmp,nz,row,*ai = a->i,*aj = a->j; 2278 int *diag_offset = b->diag,diag,*pj; 2279 MatScalar *pv,*v,*rtmp,multiplier,*pc; 2280 MatScalar *ba = b->a,*aa = a->a; 2281 2282 PetscFunctionBegin; 2283 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 2284 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 2285 rtmp = (MatScalar*)PetscMalloc((n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2286 2287 for (i=0; i<n; i++) { 2288 nz = bi[i+1] - bi[i]; 2289 ajtmp = bj + bi[i]; 2290 for (j=0; j<nz; j++) rtmp[ajtmp[j]] = 0.0; 2291 2292 /* load in initial (unfactored row) */ 2293 nz = ai[r[i]+1] - ai[r[i]]; 2294 ajtmpold = aj + ai[r[i]]; 2295 v = aa + ai[r[i]]; 2296 for (j=0; j<nz; j++) rtmp[ic[ajtmpold[j]]] = v[j]; 2297 2298 row = *ajtmp++; 2299 while (row < i) { 2300 pc = rtmp + row; 2301 if (*pc != 0.0) { 2302 pv = ba + diag_offset[row]; 2303 pj = bj + diag_offset[row] + 1; 2304 multiplier = *pc * *pv++; 2305 *pc = multiplier; 2306 nz = bi[row+1] - diag_offset[row] - 1; 2307 for (j=0; j<nz; j++) rtmp[pj[j]] -= multiplier * pv[j]; 2308 PLogFlops(1+2*nz); 2309 } 2310 row = *ajtmp++; 2311 } 2312 /* finished row so stick it into b->a */ 2313 pv = ba + bi[i]; 2314 pj = bj + bi[i]; 2315 nz = bi[i+1] - bi[i]; 2316 for (j=0; j<nz; j++) {pv[j] = rtmp[pj[j]];} 2317 diag = diag_offset[i] - bi[i]; 2318 /* check pivot entry for current row */ 2319 if (pv[diag] == 0.0) { 2320 SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,0,"Zero pivot"); 2321 } 2322 pv[diag] = 1.0/pv[diag]; 2323 } 2324 2325 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2326 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 2327 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 2328 C->factor = FACTOR_LU; 2329 C->assembled = PETSC_TRUE; 2330 PLogFlops(b->n); 2331 PetscFunctionReturn(0); 2332 } 2333 2334 /* ----------------------------------------------------------- */ 2335 #undef __FUNC__ 2336 #define __FUNC__ "MatLUFactor_SeqBAIJ" 2337 int MatLUFactor_SeqBAIJ(Mat A,IS row,IS col,PetscReal f) 2338 { 2339 Mat_SeqBAIJ *mat = (Mat_SeqBAIJ*)A->data; 2340 int ierr,refct; 2341 Mat C; 2342 PetscOps *Abops; 2343 MatOps Aops; 2344 2345 PetscFunctionBegin; 2346 ierr = MatLUFactorSymbolic(A,row,col,f,&C);CHKERRQ(ierr); 2347 ierr = MatLUFactorNumeric(A,&C);CHKERRQ(ierr); 2348 2349 /* free all the data structures from mat */ 2350 ierr = PetscFree(mat->a);CHKERRQ(ierr); 2351 if (!mat->singlemalloc) { 2352 ierr = PetscFree(mat->i);CHKERRQ(ierr); 2353 ierr = PetscFree(mat->j);CHKERRQ(ierr); 2354 } 2355 if (mat->diag) {ierr = PetscFree(mat->diag);CHKERRQ(ierr);} 2356 if (mat->ilen) {ierr = PetscFree(mat->ilen);CHKERRQ(ierr);} 2357 if (mat->imax) {ierr = PetscFree(mat->imax);CHKERRQ(ierr);} 2358 if (mat->solve_work) {ierr = PetscFree(mat->solve_work);CHKERRQ(ierr);} 2359 if (mat->mult_work) {ierr = PetscFree(mat->mult_work);CHKERRQ(ierr);} 2360 if (mat->icol) {ierr = ISDestroy(mat->icol);CHKERRQ(ierr);} 2361 ierr = PetscFree(mat);CHKERRQ(ierr); 2362 2363 ierr = MapDestroy(A->rmap);CHKERRQ(ierr); 2364 ierr = MapDestroy(A->cmap);CHKERRQ(ierr); 2365 2366 /* 2367 This is horrible,horrible code. We need to keep the 2368 A pointers for the bops and ops but copy everything 2369 else from C. 2370 */ 2371 Abops = A->bops; 2372 Aops = A->ops; 2373 refct = A->refct; 2374 ierr = PetscMemcpy(A,C,sizeof(struct _p_Mat));CHKERRQ(ierr); 2375 mat = (Mat_SeqBAIJ*)A->data; 2376 PLogObjectParent(A,mat->icol); 2377 2378 A->bops = Abops; 2379 A->ops = Aops; 2380 A->qlist = 0; 2381 A->refct = refct; 2382 /* copy over the type_name and name */ 2383 ierr = PetscStrallocpy(C->type_name,&A->type_name);CHKERRQ(ierr); 2384 ierr = PetscStrallocpy(C->name,&A->name);CHKERRQ(ierr); 2385 2386 PetscHeaderDestroy(C); 2387 PetscFunctionReturn(0); 2388 } 2389 2390 2391 2392