1 /* Using Modified Sparse Row (MSR) storage. 2 See page 85, "Iterative Methods ..." by Saad. */ 3 4 /*$Id: sbaijfact.c,v 1.8 2000/07/26 15:42:02 hzhang Exp hzhang $*/ 5 /* 6 Factorization code for SBAIJ format. 7 */ 8 #include "sbaij.h" 9 #include "src/mat/impls/baij/seq/baij.h" 10 #include "src/vec/vecimpl.h" 11 #include "src/inline/ilu.h" 12 13 #undef __FUNC__ 14 #define __FUNC__ "MatCholeskyFactorSymbolic_SeqSBAIJ" 15 int MatCholeskyFactorSymbolic_SeqSBAIJ(Mat A,IS iscol,PetscReal f,Mat *B) 16 { 17 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b; 18 IS isicol,isrow; 19 int *rip,*riip,ierr,i,mbs = a->mbs,*ai = a->i,*aj = a->j; 20 int *jutmp,bs = a->bs,bs2=a->bs2; 21 int m,nzi,realloc = 0; 22 int *jl,*q,jumin,jmin,jmax,juptr,nzk,qm,*iu,*ju,k,j,vj,umax,maxadd; 23 /* PetscReal f = 1.0; */ 24 25 PetscFunctionBegin; 26 isrow = iscol; /* remove isrow later! */ 27 PetscValidHeaderSpecific(isrow,IS_COOKIE); 28 PetscValidHeaderSpecific(iscol,IS_COOKIE); 29 /* if (A->M != A->N) SETERRQ(PETSC_ERR_ARG_WRONG,0,"matrix must be square");*/ 30 ierr = ISInvertPermutation(iscol,PETSC_DECIDE,&isicol);CHKERRQ(ierr); 31 ierr = ISGetIndices(isrow,&rip);CHKERRQ(ierr); 32 ierr = ISGetIndices(isicol,&riip);CHKERRQ(ierr); 33 34 for (k=0; k<mbs; k++) { 35 if ( rip[k] - riip[k] != 0 ) { 36 printf("Non-symm. permutation, use symm. permutation or general matrix format\n"); 37 break; 38 } 39 } 40 41 /* initialization */ 42 /* Don't know how many column pointers are needed so estimate. 43 Use Modified Sparse Row storage for u and ju, see Sasd pp.85 */ 44 /* if (info) f = info->fill; */ 45 iu = (int*)PetscMalloc((mbs+1)*sizeof(int));CHKPTRQ(iu); 46 umax = (int)(f*ai[mbs] + 1); umax += mbs + 1; 47 ju = (int*)PetscMalloc(umax*sizeof(int));CHKPTRQ(ju); 48 iu[0] = mbs+1; 49 juptr = mbs; 50 jl = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(jl); 51 q = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(q); 52 for (i=0; i<mbs; i++){ 53 jl[i] = mbs; q[i] = 0; 54 } 55 56 /* for each row k */ 57 for (k=0; k<mbs; k++){ 58 nzk = 0; /* num. of nz blocks in k-th block row with diagonal block excluded */ 59 q[k] = mbs; 60 /* initialize nonzero structure of k-th row to row rip[k] of A */ 61 jmin = ai[rip[k]]; 62 jmax = ai[rip[k]+1]; 63 for (j=jmin; j<jmax; j++){ 64 vj = riip[aj[j]]; /* col. value */ 65 if(vj > k){ 66 qm = k; 67 do { 68 m = qm; qm = q[m]; 69 } while(qm < vj); 70 if (qm == vj) { 71 printf(" error: duplicate entry in A\n"); break; 72 } 73 nzk++; 74 q[m] = vj; 75 q[vj] = qm; 76 } /* if(vj > k) */ 77 } /* for (j=jmin; j<jmax; j++) */ 78 79 /* modify nonzero structure of k-th row by computing fill-in 80 for each row i to be merged in */ 81 i = k; 82 i = jl[i]; /* next pivot row (== mbs for symbolic factorization) */ 83 /* printf(" next pivot row i=%d\n",i); */ 84 while (i < mbs){ 85 /* merge row i into k-th row */ 86 nzi = iu[i+1] - (iu[i]+1); 87 jmin = iu[i] + 1; jmax = iu[i] + nzi; 88 qm = k; 89 for (j=jmin; j<jmax+1; j++){ 90 vj = ju[j]; 91 do { 92 m = qm; qm = q[m]; 93 } while (qm < vj); 94 if (qm != vj){ 95 nzk++; q[m] = vj; q[vj] = qm; qm = vj; 96 } 97 } 98 i = jl[i]; /* next pivot row */ 99 } 100 101 /* add k to row list for first nonzero element in k-th row */ 102 if (nzk > 0){ 103 i = q[k]; /* col value of first nonzero element in U(k, k+1:mbs-1) */ 104 jl[k] = jl[i]; jl[i] = k; 105 } 106 iu[k+1] = iu[k] + nzk; /* printf(" iu[%d]=%d, umax=%d\n", k+1, iu[k+1],umax);*/ 107 108 /* allocate more space to ju if needed */ 109 if (iu[k+1] > umax) { printf("allocate more space, iu[%d]=%d > umax=%d\n",k+1, iu[k+1],umax); 110 /* estimate how much additional space we will need */ 111 /* use the strategy suggested by David Hysom <hysom@perch-t.icase.edu> */ 112 /* just double the memory each time */ 113 maxadd = umax; 114 if (maxadd < nzk) maxadd = (mbs-k)*(nzk+1)/2; 115 umax += maxadd; 116 117 /* allocate a longer ju */ 118 jutmp = (int*)PetscMalloc(umax*sizeof(int));CHKPTRQ(jutmp); 119 ierr = PetscMemcpy(jutmp,ju,iu[k]*sizeof(int));CHKERRQ(ierr); 120 ierr = PetscFree(ju);CHKERRQ(ierr); 121 ju = jutmp; 122 realloc++; /* count how many times we realloc */ 123 } 124 125 /* save nonzero structure of k-th row in ju */ 126 i=k; 127 jumin = juptr + 1; juptr += nzk; 128 for (j=jumin; j<juptr+1; j++){ 129 i=q[i]; 130 ju[j]=i; 131 /* printf(" k=%d, ju[%d]=%d\n",k,j,ju[j]);*/ 132 } 133 /* printf("\n"); */ 134 } /* for (k=0; k<mbs; k++) */ 135 136 if (ai[mbs] != 0) { 137 PetscReal af = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 138 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Reallocs %d Fill ratio:given %g needed %g\n",realloc,f,af); 139 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Run with -pc_lu_fill %g or use \n",af); 140 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:PCLUSetFill(pc,%g);\n",af); 141 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:for best performance.\n"); 142 } else { 143 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Empty matrix.\n"); 144 } 145 146 ierr = ISRestoreIndices(isrow,&rip);CHKERRQ(ierr); 147 ierr = ISRestoreIndices(isicol,&riip);CHKERRQ(ierr); 148 149 ierr = PetscFree(q);CHKERRQ(ierr); 150 ierr = PetscFree(jl);CHKERRQ(ierr); 151 152 /* put together the new matrix */ 153 ierr = MatCreateSeqSBAIJ(A->comm,bs,bs*mbs,bs*mbs,0,PETSC_NULL,B);CHKERRQ(ierr); 154 PLogObjectParent(*B,isicol); 155 b = (Mat_SeqSBAIJ*)(*B)->data; 156 ierr = PetscFree(b->imax);CHKERRQ(ierr); 157 b->singlemalloc = PETSC_FALSE; 158 /* the next line frees the default space generated by the Create() */ 159 ierr = PetscFree(b->a);CHKERRQ(ierr); 160 ierr = PetscFree(b->ilen);CHKERRQ(ierr); 161 b->a = (MatScalar*)PetscMalloc((iu[mbs]+1)*sizeof(MatScalar)*bs2);CHKPTRQ(b->a); 162 b->j = ju; 163 b->i = iu; 164 b->diag = 0; 165 b->ilen = 0; 166 b->imax = 0; 167 b->row = isrow; 168 b->col = iscol; 169 ierr = PetscObjectReference((PetscObject)isrow);CHKERRQ(ierr); 170 ierr = PetscObjectReference((PetscObject)iscol);CHKERRQ(ierr); 171 b->icol = isicol; 172 b->solve_work = (Scalar*)PetscMalloc((bs*mbs+bs)*sizeof(Scalar));CHKPTRQ(b->solve_work); 173 /* In b structure: Free imax, ilen, old a, old j. 174 Allocate idnew, solve_work, new a, new j */ 175 PLogObjectMemory(*B,(iu[mbs]-mbs)*(sizeof(int)+sizeof(MatScalar))); 176 b->s_maxnz = b->s_nz = iu[mbs]; 177 178 (*B)->factor = FACTOR_LU; 179 (*B)->info.factor_mallocs = realloc; 180 (*B)->info.fill_ratio_given = f; 181 if (ai[mbs] != 0) { 182 (*B)->info.fill_ratio_needed = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 183 } else { 184 (*B)->info.fill_ratio_needed = 0.0; 185 } 186 187 188 PetscFunctionReturn(0); 189 } 190 191 /* ----------------------------------------------------------- */ 192 #undef __FUNC__ 193 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_N" 194 int MatCholeskyFactorNumeric_SeqSBAIJ_N(Mat A,Mat *B) 195 { 196 Mat C = *B; 197 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 198 IS isrow = b->row,isicol = b->icol; 199 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 200 int *ajtmpold,*ajtmp,nz,row,bslog,*ai=a->i,*aj=a->j,k,flg; 201 int *diag_offset=b->diag,diag,bs=a->bs,bs2 = a->bs2,*v_pivots,*pj; 202 MatScalar *ba = b->a,*aa = a->a,*pv,*v,*rtmp,*multiplier,*v_work,*pc,*w; 203 204 PetscFunctionBegin; 205 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 206 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 207 rtmp = (MatScalar*)PetscMalloc(bs2*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 208 ierr = PetscMemzero(rtmp,bs2*(n+1)*sizeof(MatScalar));CHKERRQ(ierr); 209 /* generate work space needed by dense LU factorization */ 210 v_work = (MatScalar*)PetscMalloc(bs*sizeof(int) + (bs+bs2)*sizeof(MatScalar));CHKPTRQ(v_work); 211 multiplier = v_work + bs; 212 v_pivots = (int*)(multiplier + bs2); 213 214 /* flops in while loop */ 215 bslog = 2*bs*bs2; 216 217 for (i=0; i<n; i++) { 218 nz = bi[i+1] - bi[i]; 219 ajtmp = bj + bi[i]; 220 for (j=0; j<nz; j++) { 221 ierr = PetscMemzero(rtmp+bs2*ajtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 222 } 223 /* load in initial (unfactored row) */ 224 nz = ai[r[i]+1] - ai[r[i]]; 225 ajtmpold = aj + ai[r[i]]; 226 v = aa + bs2*ai[r[i]]; 227 for (j=0; j<nz; j++) { 228 ierr = PetscMemcpy(rtmp+bs2*ic[ajtmpold[j]],v+bs2*j,bs2*sizeof(MatScalar));CHKERRQ(ierr); 229 } 230 row = *ajtmp++; 231 while (row < i) { 232 pc = rtmp + bs2*row; 233 /* if (*pc) { */ 234 for (flg=0,k=0; k<bs2; k++) { if (pc[k]!=0.0) { flg =1; break; }} 235 if (flg) { 236 pv = ba + bs2*diag_offset[row]; 237 pj = bj + diag_offset[row] + 1; 238 Kernel_A_gets_A_times_B(bs,pc,pv,multiplier); 239 nz = bi[row+1] - diag_offset[row] - 1; 240 pv += bs2; 241 for (j=0; j<nz; j++) { 242 Kernel_A_gets_A_minus_B_times_C(bs,rtmp+bs2*pj[j],pc,pv+bs2*j); 243 } 244 PLogFlops(bslog*(nz+1)-bs); 245 } 246 row = *ajtmp++; 247 } 248 /* finished row so stick it into b->a */ 249 pv = ba + bs2*bi[i]; 250 pj = bj + bi[i]; 251 nz = bi[i+1] - bi[i]; 252 for (j=0; j<nz; j++) { 253 ierr = PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 254 } 255 diag = diag_offset[i] - bi[i]; 256 /* invert diagonal block */ 257 w = pv + bs2*diag; 258 Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work); 259 } 260 261 ierr = PetscFree(rtmp);CHKERRQ(ierr); 262 ierr = PetscFree(v_work);CHKERRQ(ierr); 263 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 264 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 265 C->factor = FACTOR_LU; 266 C->assembled = PETSC_TRUE; 267 PLogFlops(1.3333*bs*bs2*b->mbs); /* from inverting diagonal blocks */ 268 PetscFunctionReturn(0); 269 } 270 /* ------------------------------------------------------------*/ 271 /* 272 Version for when blocks are 7 by 7 273 */ 274 #undef __FUNC__ 275 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_7" 276 int MatCholeskyFactorNumeric_SeqSBAIJ_7(Mat A,Mat *B) 277 { 278 Mat C = *B; 279 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 280 IS isrow = b->row,isicol = b->icol; 281 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 282 int *ajtmpold,*ajtmp,nz,row; 283 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 284 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 285 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 286 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 287 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 288 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 289 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 290 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 291 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 292 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 293 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 294 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 295 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 296 MatScalar *ba = b->a,*aa = a->a; 297 298 PetscFunctionBegin; 299 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 300 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 301 rtmp = (MatScalar*)PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 302 303 for (i=0; i<n; i++) { 304 nz = bi[i+1] - bi[i]; 305 ajtmp = bj + bi[i]; 306 for (j=0; j<nz; j++) { 307 x = rtmp+49*ajtmp[j]; 308 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 309 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 310 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 311 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 312 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 313 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 314 } 315 /* load in initial (unfactored row) */ 316 idx = r[i]; 317 nz = ai[idx+1] - ai[idx]; 318 ajtmpold = aj + ai[idx]; 319 v = aa + 49*ai[idx]; 320 for (j=0; j<nz; j++) { 321 x = rtmp+49*ic[ajtmpold[j]]; 322 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 323 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 324 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 325 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 326 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 327 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 328 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 329 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 330 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 331 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 332 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 333 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 334 x[48] = v[48]; 335 v += 49; 336 } 337 row = *ajtmp++; 338 while (row < i) { 339 pc = rtmp + 49*row; 340 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 341 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 342 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 343 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 344 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 345 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 346 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 347 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 348 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 349 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 350 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 351 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 352 p49 = pc[48]; 353 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 354 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 355 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 356 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 357 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 358 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 359 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 360 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 361 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 362 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 363 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 364 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 365 p49 != 0.0) { 366 pv = ba + 49*diag_offset[row]; 367 pj = bj + diag_offset[row] + 1; 368 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 369 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 370 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 371 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 372 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 373 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 374 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 375 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 376 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 377 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 378 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 379 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 380 x49 = pv[48]; 381 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 382 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 383 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 384 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 385 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 386 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 387 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 388 389 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 390 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 391 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 392 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 393 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 394 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 395 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 396 397 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 398 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 399 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 400 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 401 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 402 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 403 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 404 405 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 406 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 407 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 408 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 409 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 410 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 411 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 412 413 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 414 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 415 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 416 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 417 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 418 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 419 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 420 421 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 422 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 423 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 424 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 425 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 426 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 427 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 428 429 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 430 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 431 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 432 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 433 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 434 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 435 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 436 437 nz = bi[row+1] - diag_offset[row] - 1; 438 pv += 49; 439 for (j=0; j<nz; j++) { 440 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 441 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 442 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 443 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 444 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 445 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 446 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 447 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 448 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 449 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 450 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 451 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 452 x49 = pv[48]; 453 x = rtmp + 49*pj[j]; 454 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 455 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 456 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 457 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 458 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 459 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 460 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 461 462 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 463 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 464 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 465 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 466 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 467 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 468 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 469 470 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 471 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 472 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 473 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 474 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 475 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 476 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 477 478 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 479 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 480 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 481 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 482 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 483 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 484 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 485 486 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 487 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 488 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 489 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 490 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 491 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 492 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 493 494 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 495 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 496 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 497 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 498 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 499 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 500 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 501 502 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 503 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 504 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 505 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 506 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 507 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 508 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 509 pv += 49; 510 } 511 PLogFlops(686*nz+637); 512 } 513 row = *ajtmp++; 514 } 515 /* finished row so stick it into b->a */ 516 pv = ba + 49*bi[i]; 517 pj = bj + bi[i]; 518 nz = bi[i+1] - bi[i]; 519 for (j=0; j<nz; j++) { 520 x = rtmp+49*pj[j]; 521 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 522 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 523 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 524 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 525 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 526 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 527 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 528 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 529 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 530 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 531 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 532 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 533 pv[48] = x[48]; 534 pv += 49; 535 } 536 /* invert diagonal block */ 537 w = ba + 49*diag_offset[i]; 538 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 539 } 540 541 ierr = PetscFree(rtmp);CHKERRQ(ierr); 542 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 543 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 544 C->factor = FACTOR_LU; 545 C->assembled = PETSC_TRUE; 546 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 547 PetscFunctionReturn(0); 548 } 549 550 /* 551 Version for when blocks are 7 by 7 Using natural ordering 552 */ 553 #undef __FUNC__ 554 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering" 555 int MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering(Mat A,Mat *B) 556 { 557 Mat C = *B; 558 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 559 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 560 int *ajtmpold,*ajtmp,nz,row; 561 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 562 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 563 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 564 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 565 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 566 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 567 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 568 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 569 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 570 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 571 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 572 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 573 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 574 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 575 MatScalar *ba = b->a,*aa = a->a; 576 577 PetscFunctionBegin; 578 rtmp = (MatScalar*)PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 579 for (i=0; i<n; i++) { 580 nz = bi[i+1] - bi[i]; 581 ajtmp = bj + bi[i]; 582 for (j=0; j<nz; j++) { 583 x = rtmp+49*ajtmp[j]; 584 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 585 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 586 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 587 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 588 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 589 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 590 } 591 /* load in initial (unfactored row) */ 592 nz = ai[i+1] - ai[i]; 593 ajtmpold = aj + ai[i]; 594 v = aa + 49*ai[i]; 595 for (j=0; j<nz; j++) { 596 x = rtmp+49*ajtmpold[j]; 597 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 598 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 599 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 600 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 601 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 602 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 603 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 604 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 605 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 606 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 607 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 608 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 609 x[48] = v[48]; 610 v += 49; 611 } 612 row = *ajtmp++; 613 while (row < i) { 614 pc = rtmp + 49*row; 615 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 616 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 617 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 618 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 619 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 620 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 621 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 622 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 623 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 624 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 625 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 626 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 627 p49 = pc[48]; 628 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 629 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 630 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 631 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 632 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 633 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 634 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 635 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 636 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 637 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 638 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 639 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 640 p49 != 0.0) { 641 pv = ba + 49*diag_offset[row]; 642 pj = bj + diag_offset[row] + 1; 643 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 644 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 645 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 646 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 647 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 648 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 649 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 650 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 651 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 652 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 653 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 654 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 655 x49 = pv[48]; 656 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 657 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 658 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 659 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 660 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 661 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 662 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 663 664 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 665 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 666 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 667 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 668 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 669 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 670 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 671 672 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 673 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 674 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 675 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 676 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 677 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 678 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 679 680 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 681 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 682 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 683 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 684 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 685 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 686 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 687 688 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 689 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 690 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 691 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 692 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 693 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 694 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 695 696 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 697 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 698 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 699 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 700 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 701 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 702 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 703 704 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 705 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 706 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 707 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 708 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 709 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 710 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 711 712 nz = bi[row+1] - diag_offset[row] - 1; 713 pv += 49; 714 for (j=0; j<nz; j++) { 715 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 716 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 717 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 718 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 719 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 720 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 721 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 722 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 723 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 724 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 725 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 726 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 727 x49 = pv[48]; 728 x = rtmp + 49*pj[j]; 729 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 730 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 731 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 732 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 733 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 734 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 735 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 736 737 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 738 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 739 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 740 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 741 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 742 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 743 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 744 745 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 746 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 747 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 748 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 749 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 750 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 751 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 752 753 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 754 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 755 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 756 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 757 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 758 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 759 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 760 761 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 762 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 763 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 764 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 765 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 766 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 767 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 768 769 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 770 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 771 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 772 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 773 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 774 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 775 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 776 777 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 778 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 779 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 780 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 781 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 782 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 783 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 784 pv += 49; 785 } 786 PLogFlops(686*nz+637); 787 } 788 row = *ajtmp++; 789 } 790 /* finished row so stick it into b->a */ 791 pv = ba + 49*bi[i]; 792 pj = bj + bi[i]; 793 nz = bi[i+1] - bi[i]; 794 for (j=0; j<nz; j++) { 795 x = rtmp+49*pj[j]; 796 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 797 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 798 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 799 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 800 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 801 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 802 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 803 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 804 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 805 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 806 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 807 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 808 pv[48] = x[48]; 809 pv += 49; 810 } 811 /* invert diagonal block */ 812 w = ba + 49*diag_offset[i]; 813 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 814 } 815 816 ierr = PetscFree(rtmp);CHKERRQ(ierr); 817 C->factor = FACTOR_LU; 818 C->assembled = PETSC_TRUE; 819 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 820 PetscFunctionReturn(0); 821 } 822 823 /* ------------------------------------------------------------*/ 824 /* 825 Version for when blocks are 6 by 6 826 */ 827 #undef __FUNC__ 828 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_6" 829 int MatCholeskyFactorNumeric_SeqSBAIJ_6(Mat A,Mat *B) 830 { 831 Mat C = *B; 832 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 833 IS isrow = b->row,isicol = b->icol; 834 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 835 int *ajtmpold,*ajtmp,nz,row; 836 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 837 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 838 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 839 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 840 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 841 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 842 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 843 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 844 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 845 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 846 MatScalar *ba = b->a,*aa = a->a; 847 848 PetscFunctionBegin; 849 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 850 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 851 rtmp = (MatScalar*)PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 852 853 for (i=0; i<n; i++) { 854 nz = bi[i+1] - bi[i]; 855 ajtmp = bj + bi[i]; 856 for (j=0; j<nz; j++) { 857 x = rtmp+36*ajtmp[j]; 858 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 859 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 860 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 861 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 862 x[34] = x[35] = 0.0 ; 863 } 864 /* load in initial (unfactored row) */ 865 idx = r[i]; 866 nz = ai[idx+1] - ai[idx]; 867 ajtmpold = aj + ai[idx]; 868 v = aa + 36*ai[idx]; 869 for (j=0; j<nz; j++) { 870 x = rtmp+36*ic[ajtmpold[j]]; 871 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 872 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 873 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 874 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 875 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 876 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 877 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 878 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 879 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 880 v += 36; 881 } 882 row = *ajtmp++; 883 while (row < i) { 884 pc = rtmp + 36*row; 885 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 886 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 887 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 888 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 889 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 890 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 891 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 892 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 893 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 894 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 895 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 896 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 897 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 898 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 899 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 900 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 901 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 902 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 903 pv = ba + 36*diag_offset[row]; 904 pj = bj + diag_offset[row] + 1; 905 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 906 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 907 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 908 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 909 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 910 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 911 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 912 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 913 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 914 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 915 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 916 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 917 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 918 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 919 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 920 921 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 922 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 923 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 924 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 925 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 926 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 927 928 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 929 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 930 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 931 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 932 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 933 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 934 935 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 936 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 937 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 938 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 939 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 940 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 941 942 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 943 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 944 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 945 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 946 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 947 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 948 949 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 950 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 951 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 952 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 953 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 954 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 955 956 nz = bi[row+1] - diag_offset[row] - 1; 957 pv += 36; 958 for (j=0; j<nz; j++) { 959 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 960 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 961 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 962 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 963 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 964 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 965 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 966 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 967 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 968 x = rtmp + 36*pj[j]; 969 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 970 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 971 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 972 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 973 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 974 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 975 976 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 977 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 978 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 979 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 980 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 981 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 982 983 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 984 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 985 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 986 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 987 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 988 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 989 990 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 991 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 992 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 993 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 994 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 995 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 996 997 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 998 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 999 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 1000 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 1001 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 1002 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 1003 1004 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 1005 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 1006 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 1007 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 1008 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 1009 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 1010 1011 pv += 36; 1012 } 1013 PLogFlops(432*nz+396); 1014 } 1015 row = *ajtmp++; 1016 } 1017 /* finished row so stick it into b->a */ 1018 pv = ba + 36*bi[i]; 1019 pj = bj + bi[i]; 1020 nz = bi[i+1] - bi[i]; 1021 for (j=0; j<nz; j++) { 1022 x = rtmp+36*pj[j]; 1023 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1024 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1025 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1026 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1027 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1028 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1029 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1030 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1031 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1032 pv += 36; 1033 } 1034 /* invert diagonal block */ 1035 w = ba + 36*diag_offset[i]; 1036 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1037 } 1038 1039 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1040 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1041 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1042 C->factor = FACTOR_LU; 1043 C->assembled = PETSC_TRUE; 1044 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1045 PetscFunctionReturn(0); 1046 } 1047 /* 1048 Version for when blocks are 6 by 6 Using natural ordering 1049 */ 1050 #undef __FUNC__ 1051 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering" 1052 int MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering(Mat A,Mat *B) 1053 { 1054 Mat C = *B; 1055 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1056 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1057 int *ajtmpold,*ajtmp,nz,row; 1058 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1059 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1060 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1061 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1062 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1063 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1064 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1065 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1066 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 1067 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 1068 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 1069 MatScalar *ba = b->a,*aa = a->a; 1070 1071 PetscFunctionBegin; 1072 rtmp = (MatScalar*)PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1073 for (i=0; i<n; i++) { 1074 nz = bi[i+1] - bi[i]; 1075 ajtmp = bj + bi[i]; 1076 for (j=0; j<nz; j++) { 1077 x = rtmp+36*ajtmp[j]; 1078 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1079 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1080 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 1081 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 1082 x[34] = x[35] = 0.0 ; 1083 } 1084 /* load in initial (unfactored row) */ 1085 nz = ai[i+1] - ai[i]; 1086 ajtmpold = aj + ai[i]; 1087 v = aa + 36*ai[i]; 1088 for (j=0; j<nz; j++) { 1089 x = rtmp+36*ajtmpold[j]; 1090 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1091 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 1092 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 1093 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 1094 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 1095 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1096 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 1097 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 1098 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 1099 v += 36; 1100 } 1101 row = *ajtmp++; 1102 while (row < i) { 1103 pc = rtmp + 36*row; 1104 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1105 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 1106 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 1107 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 1108 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 1109 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1110 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 1111 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 1112 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 1113 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 1114 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 1115 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 1116 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 1117 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 1118 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 1119 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 1120 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 1121 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 1122 pv = ba + 36*diag_offset[row]; 1123 pj = bj + diag_offset[row] + 1; 1124 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1125 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1126 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1127 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1128 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1129 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1130 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1131 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1132 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1133 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 1134 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 1135 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 1136 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 1137 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 1138 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 1139 1140 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 1141 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 1142 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 1143 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 1144 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 1145 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 1146 1147 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 1148 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 1149 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 1150 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 1151 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 1152 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 1153 1154 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 1155 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 1156 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 1157 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 1158 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 1159 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 1160 1161 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 1162 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 1163 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 1164 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 1165 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 1166 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 1167 1168 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 1169 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 1170 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 1171 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 1172 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 1173 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 1174 1175 nz = bi[row+1] - diag_offset[row] - 1; 1176 pv += 36; 1177 for (j=0; j<nz; j++) { 1178 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1179 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1180 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1181 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1182 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1183 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1184 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1185 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1186 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1187 x = rtmp + 36*pj[j]; 1188 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 1189 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 1190 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 1191 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 1192 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 1193 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 1194 1195 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 1196 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 1197 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 1198 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 1199 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 1200 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 1201 1202 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 1203 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 1204 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 1205 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 1206 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 1207 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 1208 1209 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 1210 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 1211 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 1212 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 1213 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 1214 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 1215 1216 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 1217 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 1218 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 1219 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 1220 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 1221 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 1222 1223 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 1224 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 1225 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 1226 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 1227 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 1228 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 1229 1230 pv += 36; 1231 } 1232 PLogFlops(432*nz+396); 1233 } 1234 row = *ajtmp++; 1235 } 1236 /* finished row so stick it into b->a */ 1237 pv = ba + 36*bi[i]; 1238 pj = bj + bi[i]; 1239 nz = bi[i+1] - bi[i]; 1240 for (j=0; j<nz; j++) { 1241 x = rtmp+36*pj[j]; 1242 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1243 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1244 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1245 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1246 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1247 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1248 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1249 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1250 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1251 pv += 36; 1252 } 1253 /* invert diagonal block */ 1254 w = ba + 36*diag_offset[i]; 1255 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1256 } 1257 1258 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1259 C->factor = FACTOR_LU; 1260 C->assembled = PETSC_TRUE; 1261 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1262 PetscFunctionReturn(0); 1263 } 1264 1265 /* ------------------------------------------------------------*/ 1266 /* 1267 Version for when blocks are 5 by 5 1268 */ 1269 #undef __FUNC__ 1270 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_5" 1271 int MatCholeskyFactorNumeric_SeqSBAIJ_5(Mat A,Mat *B) 1272 { 1273 Mat C = *B; 1274 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1275 IS isrow = b->row,isicol = b->icol; 1276 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1277 int *ajtmpold,*ajtmp,nz,row; 1278 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 1279 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1280 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1281 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1282 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 1283 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 1284 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1285 MatScalar *ba = b->a,*aa = a->a; 1286 1287 PetscFunctionBegin; 1288 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1289 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1290 rtmp = (MatScalar*)PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1291 1292 for (i=0; i<n; i++) { 1293 nz = bi[i+1] - bi[i]; 1294 ajtmp = bj + bi[i]; 1295 for (j=0; j<nz; j++) { 1296 x = rtmp+25*ajtmp[j]; 1297 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1298 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1299 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1300 } 1301 /* load in initial (unfactored row) */ 1302 idx = r[i]; 1303 nz = ai[idx+1] - ai[idx]; 1304 ajtmpold = aj + ai[idx]; 1305 v = aa + 25*ai[idx]; 1306 for (j=0; j<nz; j++) { 1307 x = rtmp+25*ic[ajtmpold[j]]; 1308 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1309 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1310 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1311 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; 1312 x[18] = v[18]; x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; 1313 x[22] = v[22]; x[23] = v[23]; x[24] = v[24]; 1314 v += 25; 1315 } 1316 row = *ajtmp++; 1317 while (row < i) { 1318 pc = rtmp + 25*row; 1319 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1320 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1321 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1322 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; 1323 p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1324 p25 = pc[24]; 1325 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1326 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1327 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1328 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || 1329 p20 != 0.0 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || 1330 p24 != 0.0 || p25 != 0.0) { 1331 pv = ba + 25*diag_offset[row]; 1332 pj = bj + diag_offset[row] + 1; 1333 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1334 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1335 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1336 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1337 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; 1338 x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1339 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1340 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1341 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1342 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1343 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1344 1345 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1346 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1347 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1348 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1349 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1350 1351 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1352 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1353 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1354 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1355 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1356 1357 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1358 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1359 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1360 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1361 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1362 1363 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1364 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1365 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1366 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1367 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1368 1369 nz = bi[row+1] - diag_offset[row] - 1; 1370 pv += 25; 1371 for (j=0; j<nz; j++) { 1372 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1373 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1374 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1375 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; 1376 x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; 1377 x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1378 x = rtmp + 25*pj[j]; 1379 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1380 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1381 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1382 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1383 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1384 1385 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1386 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1387 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1388 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1389 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1390 1391 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1392 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1393 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1394 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1395 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1396 1397 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1398 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1399 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1400 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1401 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1402 1403 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1404 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1405 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1406 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1407 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1408 1409 pv += 25; 1410 } 1411 PLogFlops(250*nz+225); 1412 } 1413 row = *ajtmp++; 1414 } 1415 /* finished row so stick it into b->a */ 1416 pv = ba + 25*bi[i]; 1417 pj = bj + bi[i]; 1418 nz = bi[i+1] - bi[i]; 1419 for (j=0; j<nz; j++) { 1420 x = rtmp+25*pj[j]; 1421 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1422 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1423 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1424 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; 1425 pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; 1426 pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; pv[24] = x[24]; 1427 pv += 25; 1428 } 1429 /* invert diagonal block */ 1430 w = ba + 25*diag_offset[i]; 1431 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1432 } 1433 1434 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1435 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1436 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1437 C->factor = FACTOR_LU; 1438 C->assembled = PETSC_TRUE; 1439 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1440 PetscFunctionReturn(0); 1441 } 1442 /* 1443 Version for when blocks are 5 by 5 Using natural ordering 1444 */ 1445 #undef __FUNC__ 1446 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering" 1447 int MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering(Mat A,Mat *B) 1448 { 1449 Mat C = *B; 1450 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1451 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1452 int *ajtmpold,*ajtmp,nz,row; 1453 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1454 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1455 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1456 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1457 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1458 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1459 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1460 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1461 MatScalar *ba = b->a,*aa = a->a; 1462 1463 PetscFunctionBegin; 1464 rtmp = (MatScalar*)PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1465 for (i=0; i<n; i++) { 1466 nz = bi[i+1] - bi[i]; 1467 ajtmp = bj + bi[i]; 1468 for (j=0; j<nz; j++) { 1469 x = rtmp+25*ajtmp[j]; 1470 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1471 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1472 x[16] = x[17] = x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1473 } 1474 /* load in initial (unfactored row) */ 1475 nz = ai[i+1] - ai[i]; 1476 ajtmpold = aj + ai[i]; 1477 v = aa + 25*ai[i]; 1478 for (j=0; j<nz; j++) { 1479 x = rtmp+25*ajtmpold[j]; 1480 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1481 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1482 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1483 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; 1484 x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1485 x[24] = v[24]; 1486 v += 25; 1487 } 1488 row = *ajtmp++; 1489 while (row < i) { 1490 pc = rtmp + 25*row; 1491 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1492 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1493 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1494 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; 1495 p19 = pc[18]; p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; 1496 p24 = pc[23]; p25 = pc[24]; 1497 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1498 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1499 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1500 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 1501 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || p25 != 0.0) { 1502 pv = ba + 25*diag_offset[row]; 1503 pj = bj + diag_offset[row] + 1; 1504 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1505 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1506 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1507 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; 1508 x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1509 x25 = pv[24]; 1510 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1511 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1512 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1513 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1514 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1515 1516 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1517 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1518 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1519 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1520 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1521 1522 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1523 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1524 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1525 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1526 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1527 1528 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1529 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1530 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1531 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1532 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1533 1534 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1535 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1536 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1537 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1538 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1539 1540 nz = bi[row+1] - diag_offset[row] - 1; 1541 pv += 25; 1542 for (j=0; j<nz; j++) { 1543 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1544 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1545 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1546 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1547 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; 1548 x24 = pv[23]; x25 = pv[24]; 1549 x = rtmp + 25*pj[j]; 1550 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1551 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1552 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1553 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1554 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1555 1556 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1557 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1558 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1559 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1560 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1561 1562 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1563 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1564 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1565 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1566 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1567 1568 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1569 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1570 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1571 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1572 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1573 1574 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1575 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1576 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1577 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1578 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1579 pv += 25; 1580 } 1581 PLogFlops(250*nz+225); 1582 } 1583 row = *ajtmp++; 1584 } 1585 /* finished row so stick it into b->a */ 1586 pv = ba + 25*bi[i]; 1587 pj = bj + bi[i]; 1588 nz = bi[i+1] - bi[i]; 1589 for (j=0; j<nz; j++) { 1590 x = rtmp+25*pj[j]; 1591 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1592 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1593 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1594 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; pv[17] = x[17]; 1595 pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; 1596 pv[23] = x[23]; pv[24] = x[24]; 1597 pv += 25; 1598 } 1599 /* invert diagonal block */ 1600 w = ba + 25*diag_offset[i]; 1601 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1602 } 1603 1604 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1605 C->factor = FACTOR_LU; 1606 C->assembled = PETSC_TRUE; 1607 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1608 PetscFunctionReturn(0); 1609 } 1610 1611 /* ------------------------------------------------------------*/ 1612 /* 1613 Version for when blocks are 4 by 4 1614 */ 1615 #undef __FUNC__ 1616 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_4" 1617 int MatCholeskyFactorNumeric_SeqSBAIJ_4(Mat A,Mat *B) 1618 { 1619 Mat C = *B; 1620 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1621 IS isrow = b->row,isicol = b->icol; 1622 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1623 int *ajtmpold,*ajtmp,nz,row; 1624 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 1625 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1626 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1627 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1628 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1629 MatScalar m13,m14,m15,m16; 1630 MatScalar *ba = b->a,*aa = a->a; 1631 1632 PetscFunctionBegin; 1633 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1634 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1635 rtmp = (MatScalar*)PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1636 1637 for (i=0; i<n; i++) { 1638 nz = bi[i+1] - bi[i]; 1639 ajtmp = bj + bi[i]; 1640 for (j=0; j<nz; j++) { 1641 x = rtmp+16*ajtmp[j]; 1642 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1643 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1644 } 1645 /* load in initial (unfactored row) */ 1646 idx = r[i]; 1647 nz = ai[idx+1] - ai[idx]; 1648 ajtmpold = aj + ai[idx]; 1649 v = aa + 16*ai[idx]; 1650 for (j=0; j<nz; j++) { 1651 x = rtmp+16*ic[ajtmpold[j]]; 1652 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1653 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1654 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1655 x[14] = v[14]; x[15] = v[15]; 1656 v += 16; 1657 } 1658 row = *ajtmp++; 1659 while (row < i) { 1660 pc = rtmp + 16*row; 1661 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1662 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1663 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1664 p15 = pc[14]; p16 = pc[15]; 1665 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1666 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1667 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1668 || p16 != 0.0) { 1669 pv = ba + 16*diag_offset[row]; 1670 pj = bj + diag_offset[row] + 1; 1671 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1672 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1673 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1674 x15 = pv[14]; x16 = pv[15]; 1675 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1676 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1677 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1678 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1679 1680 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1681 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1682 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1683 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1684 1685 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1686 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1687 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1688 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1689 1690 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1691 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1692 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1693 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1694 1695 nz = bi[row+1] - diag_offset[row] - 1; 1696 pv += 16; 1697 for (j=0; j<nz; j++) { 1698 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1699 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1700 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1701 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1702 x = rtmp + 16*pj[j]; 1703 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1704 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1705 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1706 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1707 1708 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1709 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1710 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1711 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1712 1713 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1714 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1715 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1716 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1717 1718 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1719 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1720 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1721 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1722 1723 pv += 16; 1724 } 1725 PLogFlops(128*nz+112); 1726 } 1727 row = *ajtmp++; 1728 } 1729 /* finished row so stick it into b->a */ 1730 pv = ba + 16*bi[i]; 1731 pj = bj + bi[i]; 1732 nz = bi[i+1] - bi[i]; 1733 for (j=0; j<nz; j++) { 1734 x = rtmp+16*pj[j]; 1735 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1736 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1737 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1738 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1739 pv += 16; 1740 } 1741 /* invert diagonal block */ 1742 w = ba + 16*diag_offset[i]; 1743 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1744 } 1745 1746 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1747 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1748 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1749 C->factor = FACTOR_LU; 1750 C->assembled = PETSC_TRUE; 1751 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1752 PetscFunctionReturn(0); 1753 } 1754 /* 1755 Version for when blocks are 4 by 4 Using natural ordering 1756 */ 1757 #undef __FUNC__ 1758 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering" 1759 int MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering(Mat A,Mat *B) 1760 { 1761 Mat C = *B; 1762 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1763 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1764 int *ajtmpold,*ajtmp,nz,row; 1765 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1766 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1767 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1768 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1769 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1770 MatScalar m13,m14,m15,m16; 1771 MatScalar *ba = b->a,*aa = a->a; 1772 1773 PetscFunctionBegin; 1774 rtmp = (MatScalar*)PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1775 1776 for (i=0; i<n; i++) { 1777 nz = bi[i+1] - bi[i]; 1778 ajtmp = bj + bi[i]; 1779 for (j=0; j<nz; j++) { 1780 x = rtmp+16*ajtmp[j]; 1781 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1782 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1783 } 1784 /* load in initial (unfactored row) */ 1785 nz = ai[i+1] - ai[i]; 1786 ajtmpold = aj + ai[i]; 1787 v = aa + 16*ai[i]; 1788 for (j=0; j<nz; j++) { 1789 x = rtmp+16*ajtmpold[j]; 1790 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1791 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1792 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1793 x[14] = v[14]; x[15] = v[15]; 1794 v += 16; 1795 } 1796 row = *ajtmp++; 1797 while (row < i) { 1798 pc = rtmp + 16*row; 1799 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1800 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1801 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1802 p15 = pc[14]; p16 = pc[15]; 1803 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1804 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1805 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1806 || p16 != 0.0) { 1807 pv = ba + 16*diag_offset[row]; 1808 pj = bj + diag_offset[row] + 1; 1809 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1810 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1811 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1812 x15 = pv[14]; x16 = pv[15]; 1813 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1814 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1815 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1816 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1817 1818 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1819 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1820 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1821 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1822 1823 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1824 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1825 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1826 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1827 1828 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1829 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1830 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1831 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1832 1833 nz = bi[row+1] - diag_offset[row] - 1; 1834 pv += 16; 1835 for (j=0; j<nz; j++) { 1836 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1837 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1838 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1839 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1840 x = rtmp + 16*pj[j]; 1841 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1842 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1843 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1844 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1845 1846 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1847 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1848 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1849 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1850 1851 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1852 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1853 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1854 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1855 1856 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1857 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1858 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1859 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1860 1861 pv += 16; 1862 } 1863 PLogFlops(128*nz+112); 1864 } 1865 row = *ajtmp++; 1866 } 1867 /* finished row so stick it into b->a */ 1868 pv = ba + 16*bi[i]; 1869 pj = bj + bi[i]; 1870 nz = bi[i+1] - bi[i]; 1871 for (j=0; j<nz; j++) { 1872 x = rtmp+16*pj[j]; 1873 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1874 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1875 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1876 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1877 pv += 16; 1878 } 1879 /* invert diagonal block */ 1880 w = ba + 16*diag_offset[i]; 1881 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1882 } 1883 1884 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1885 C->factor = FACTOR_LU; 1886 C->assembled = PETSC_TRUE; 1887 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1888 PetscFunctionReturn(0); 1889 } 1890 1891 1892 /* ------------------------------------------------------------*/ 1893 /* 1894 Version for when blocks are 3 by 3 1895 */ 1896 #undef __FUNC__ 1897 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_3" 1898 int MatCholeskyFactorNumeric_SeqSBAIJ_3(Mat A,Mat *B) 1899 { 1900 Mat C = *B; 1901 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1902 IS isrow = b->row,isicol = b->icol; 1903 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1904 int *ajtmpold,*ajtmp,nz,row,*ai=a->i,*aj=a->j; 1905 int *diag_offset = b->diag,idx,*pj; 1906 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1907 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1908 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 1909 MatScalar *ba = b->a,*aa = a->a; 1910 1911 PetscFunctionBegin; 1912 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1913 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1914 rtmp = (MatScalar*)PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1915 1916 for (i=0; i<n; i++) { 1917 nz = bi[i+1] - bi[i]; 1918 ajtmp = bj + bi[i]; 1919 for (j=0; j<nz; j++) { 1920 x = rtmp + 9*ajtmp[j]; 1921 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 1922 } 1923 /* load in initial (unfactored row) */ 1924 idx = r[i]; 1925 nz = ai[idx+1] - ai[idx]; 1926 ajtmpold = aj + ai[idx]; 1927 v = aa + 9*ai[idx]; 1928 for (j=0; j<nz; j++) { 1929 x = rtmp + 9*ic[ajtmpold[j]]; 1930 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1931 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1932 v += 9; 1933 } 1934 row = *ajtmp++; 1935 while (row < i) { 1936 pc = rtmp + 9*row; 1937 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1938 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1939 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1940 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 1941 pv = ba + 9*diag_offset[row]; 1942 pj = bj + diag_offset[row] + 1; 1943 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1944 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1945 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 1946 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 1947 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 1948 1949 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 1950 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 1951 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 1952 1953 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 1954 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 1955 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 1956 nz = bi[row+1] - diag_offset[row] - 1; 1957 pv += 9; 1958 for (j=0; j<nz; j++) { 1959 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1960 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1961 x = rtmp + 9*pj[j]; 1962 x[0] -= m1*x1 + m4*x2 + m7*x3; 1963 x[1] -= m2*x1 + m5*x2 + m8*x3; 1964 x[2] -= m3*x1 + m6*x2 + m9*x3; 1965 1966 x[3] -= m1*x4 + m4*x5 + m7*x6; 1967 x[4] -= m2*x4 + m5*x5 + m8*x6; 1968 x[5] -= m3*x4 + m6*x5 + m9*x6; 1969 1970 x[6] -= m1*x7 + m4*x8 + m7*x9; 1971 x[7] -= m2*x7 + m5*x8 + m8*x9; 1972 x[8] -= m3*x7 + m6*x8 + m9*x9; 1973 pv += 9; 1974 } 1975 PLogFlops(54*nz+36); 1976 } 1977 row = *ajtmp++; 1978 } 1979 /* finished row so stick it into b->a */ 1980 pv = ba + 9*bi[i]; 1981 pj = bj + bi[i]; 1982 nz = bi[i+1] - bi[i]; 1983 for (j=0; j<nz; j++) { 1984 x = rtmp + 9*pj[j]; 1985 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1986 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1987 pv += 9; 1988 } 1989 /* invert diagonal block */ 1990 w = ba + 9*diag_offset[i]; 1991 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 1992 } 1993 1994 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1995 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1996 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1997 C->factor = FACTOR_LU; 1998 C->assembled = PETSC_TRUE; 1999 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 2000 PetscFunctionReturn(0); 2001 } 2002 /* 2003 Version for when blocks are 3 by 3 Using natural ordering 2004 */ 2005 #undef __FUNC__ 2006 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering" 2007 int MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering(Mat A,Mat *B) 2008 { 2009 Mat C = *B; 2010 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 2011 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 2012 int *ajtmpold,*ajtmp,nz,row; 2013 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 2014 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2015 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 2016 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 2017 MatScalar *ba = b->a,*aa = a->a; 2018 2019 PetscFunctionBegin; 2020 rtmp = (MatScalar*)PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2021 2022 for (i=0; i<n; i++) { 2023 nz = bi[i+1] - bi[i]; 2024 ajtmp = bj + bi[i]; 2025 for (j=0; j<nz; j++) { 2026 x = rtmp+9*ajtmp[j]; 2027 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 2028 } 2029 /* load in initial (unfactored row) */ 2030 nz = ai[i+1] - ai[i]; 2031 ajtmpold = aj + ai[i]; 2032 v = aa + 9*ai[i]; 2033 for (j=0; j<nz; j++) { 2034 x = rtmp+9*ajtmpold[j]; 2035 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2036 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 2037 v += 9; 2038 } 2039 row = *ajtmp++; 2040 while (row < i) { 2041 pc = rtmp + 9*row; 2042 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2043 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 2044 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 2045 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 2046 pv = ba + 9*diag_offset[row]; 2047 pj = bj + diag_offset[row] + 1; 2048 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2049 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2050 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 2051 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 2052 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 2053 2054 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 2055 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 2056 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 2057 2058 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 2059 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 2060 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 2061 2062 nz = bi[row+1] - diag_offset[row] - 1; 2063 pv += 9; 2064 for (j=0; j<nz; j++) { 2065 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2066 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2067 x = rtmp + 9*pj[j]; 2068 x[0] -= m1*x1 + m4*x2 + m7*x3; 2069 x[1] -= m2*x1 + m5*x2 + m8*x3; 2070 x[2] -= m3*x1 + m6*x2 + m9*x3; 2071 2072 x[3] -= m1*x4 + m4*x5 + m7*x6; 2073 x[4] -= m2*x4 + m5*x5 + m8*x6; 2074 x[5] -= m3*x4 + m6*x5 + m9*x6; 2075 2076 x[6] -= m1*x7 + m4*x8 + m7*x9; 2077 x[7] -= m2*x7 + m5*x8 + m8*x9; 2078 x[8] -= m3*x7 + m6*x8 + m9*x9; 2079 pv += 9; 2080 } 2081 PLogFlops(54*nz+36); 2082 } 2083 row = *ajtmp++; 2084 } 2085 /* finished row so stick it into b->a */ 2086 pv = ba + 9*bi[i]; 2087 pj = bj + bi[i]; 2088 nz = bi[i+1] - bi[i]; 2089 for (j=0; j<nz; j++) { 2090 x = rtmp+9*pj[j]; 2091 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2092 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 2093 pv += 9; 2094 } 2095 /* invert diagonal block */ 2096 w = ba + 9*diag_offset[i]; 2097 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 2098 } 2099 2100 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2101 C->factor = FACTOR_LU; 2102 C->assembled = PETSC_TRUE; 2103 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 2104 PetscFunctionReturn(0); 2105 } 2106 2107 /* ------------------------------------------------------------*/ 2108 /* 2109 Version for when blocks are 2 by 2 2110 */ 2111 #undef __FUNC__ 2112 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_2" 2113 int MatCholeskyFactorNumeric_SeqSBAIJ_2(Mat A,Mat *B) 2114 { 2115 Mat C = *B; 2116 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 2117 IS isrow = b->row,isicol = b->icol; 2118 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 2119 int *ajtmpold,*ajtmp,nz,row; 2120 int *diag_offset=b->diag,idx,*ai=a->i,*aj=a->j,*pj; 2121 MatScalar *pv,*v,*rtmp,m1,m2,m3,m4,*pc,*w,*x,x1,x2,x3,x4; 2122 MatScalar p1,p2,p3,p4; 2123 MatScalar *ba = b->a,*aa = a->a; 2124 2125 PetscFunctionBegin; 2126 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 2127 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 2128 rtmp = (MatScalar*)PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2129 2130 for (i=0; i<n; i++) { 2131 nz = bi[i+1] - bi[i]; 2132 ajtmp = bj + bi[i]; 2133 for (j=0; j<nz; j++) { 2134 x = rtmp+4*ajtmp[j]; x[0] = x[1] = x[2] = x[3] = 0.0; 2135 } 2136 /* load in initial (unfactored row) */ 2137 idx = r[i]; 2138 nz = ai[idx+1] - ai[idx]; 2139 ajtmpold = aj + ai[idx]; 2140 v = aa + 4*ai[idx]; 2141 for (j=0; j<nz; j++) { 2142 x = rtmp+4*ic[ajtmpold[j]]; 2143 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2144 v += 4; 2145 } 2146 row = *ajtmp++; 2147 while (row < i) { 2148 pc = rtmp + 4*row; 2149 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2150 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2151 pv = ba + 4*diag_offset[row]; 2152 pj = bj + diag_offset[row] + 1; 2153 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2154 pc[0] = m1 = p1*x1 + p3*x2; 2155 pc[1] = m2 = p2*x1 + p4*x2; 2156 pc[2] = m3 = p1*x3 + p3*x4; 2157 pc[3] = m4 = p2*x3 + p4*x4; 2158 nz = bi[row+1] - diag_offset[row] - 1; 2159 pv += 4; 2160 for (j=0; j<nz; j++) { 2161 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2162 x = rtmp + 4*pj[j]; 2163 x[0] -= m1*x1 + m3*x2; 2164 x[1] -= m2*x1 + m4*x2; 2165 x[2] -= m1*x3 + m3*x4; 2166 x[3] -= m2*x3 + m4*x4; 2167 pv += 4; 2168 } 2169 PLogFlops(16*nz+12); 2170 } 2171 row = *ajtmp++; 2172 } 2173 /* finished row so stick it into b->a */ 2174 pv = ba + 4*bi[i]; 2175 pj = bj + bi[i]; 2176 nz = bi[i+1] - bi[i]; 2177 for (j=0; j<nz; j++) { 2178 x = rtmp+4*pj[j]; 2179 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2180 pv += 4; 2181 } 2182 /* invert diagonal block */ 2183 w = ba + 4*diag_offset[i]; 2184 ierr = Kernel_A_gets_inverse_A_2(w);CHKERRQ(ierr); 2185 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2186 } 2187 2188 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2189 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 2190 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 2191 C->factor = FACTOR_LU; 2192 C->assembled = PETSC_TRUE; 2193 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2194 PetscFunctionReturn(0); 2195 } 2196 /* 2197 Version for when blocks are 2 by 2 Using natural ordering 2198 */ 2199 #undef __FUNC__ 2200 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering" 2201 int MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering(Mat A,Mat *B) 2202 { 2203 Mat C = *B; 2204 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 2205 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 2206 int *ajtmpold,*ajtmp,nz,row; 2207 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 2208 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2209 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,x1,x2,x3,x4; 2210 MatScalar *ba = b->a,*aa = a->a; 2211 2212 PetscFunctionBegin; 2213 rtmp = (MatScalar*)PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2214 2215 for (i=0; i<n; i++) { 2216 nz = bi[i+1] - bi[i]; 2217 ajtmp = bj + bi[i]; 2218 for (j=0; j<nz; j++) { 2219 x = rtmp+4*ajtmp[j]; 2220 x[0] = x[1] = x[2] = x[3] = 0.0; 2221 } 2222 /* load in initial (unfactored row) */ 2223 nz = ai[i+1] - ai[i]; 2224 ajtmpold = aj + ai[i]; 2225 v = aa + 4*ai[i]; 2226 for (j=0; j<nz; j++) { 2227 x = rtmp+4*ajtmpold[j]; 2228 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2229 v += 4; 2230 } 2231 row = *ajtmp++; 2232 while (row < i) { 2233 pc = rtmp + 4*row; 2234 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2235 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2236 pv = ba + 4*diag_offset[row]; 2237 pj = bj + diag_offset[row] + 1; 2238 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2239 pc[0] = m1 = p1*x1 + p3*x2; 2240 pc[1] = m2 = p2*x1 + p4*x2; 2241 pc[2] = m3 = p1*x3 + p3*x4; 2242 pc[3] = m4 = p2*x3 + p4*x4; 2243 nz = bi[row+1] - diag_offset[row] - 1; 2244 pv += 4; 2245 for (j=0; j<nz; j++) { 2246 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2247 x = rtmp + 4*pj[j]; 2248 x[0] -= m1*x1 + m3*x2; 2249 x[1] -= m2*x1 + m4*x2; 2250 x[2] -= m1*x3 + m3*x4; 2251 x[3] -= m2*x3 + m4*x4; 2252 pv += 4; 2253 } 2254 PLogFlops(16*nz+12); 2255 } 2256 row = *ajtmp++; 2257 } 2258 /* finished row so stick it into b->a */ 2259 pv = ba + 4*bi[i]; 2260 pj = bj + bi[i]; 2261 nz = bi[i+1] - bi[i]; 2262 for (j=0; j<nz; j++) { 2263 x = rtmp+4*pj[j]; 2264 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2265 pv += 4; 2266 } 2267 /* invert diagonal block */ 2268 w = ba + 4*diag_offset[i]; 2269 ierr = Kernel_A_gets_inverse_A_2(w);CHKERRQ(ierr); 2270 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2271 } 2272 2273 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2274 C->factor = FACTOR_LU; 2275 C->assembled = PETSC_TRUE; 2276 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2277 PetscFunctionReturn(0); 2278 } 2279 2280 /* ----------------------------------------------------------- */ 2281 /* 2282 Version for when blocks are 1 by 1. 2283 */ 2284 #undef __FUNC__ 2285 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_1" 2286 int MatCholeskyFactorNumeric_SeqSBAIJ_1(Mat A,Mat *B) 2287 { 2288 Mat C = *B; 2289 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 2290 IS ip = b->row; 2291 int *rip,*riip,ierr,i,j,mbs = a->mbs,*bi = b->i,*bj = b->j; 2292 int *ai = a->i,*aj = a->j; 2293 MatScalar *rtmp; 2294 MatScalar *ba = b->a,*aa = a->a; 2295 MatScalar dk,uikdi; 2296 int k,jmin,jmax,*jl,*il,vj,nexti,juj,ili; 2297 2298 PetscFunctionBegin; 2299 ierr = ISGetIndices(ip,&rip);CHKERRQ(ierr); 2300 riip = rip; 2301 2302 /* INITIALIZATION */ 2303 /* il and jl record the first nonzero element in each row of the accessing 2304 window U(0:k, k:mbs-1). 2305 jl: list of rows to be added to uneliminated rows 2306 i>= k: jl(i) is the first row to be added to row i 2307 i< k: jl(i) is the row following row i in some list of rows 2308 jl(i) = mbs indicates the end of a list 2309 il(i): points to the first nonzero element in columns k,...,mbs-1 of 2310 row i of U */ 2311 rtmp = (MatScalar*)PetscMalloc(mbs*sizeof(MatScalar));CHKPTRQ(rtmp); 2312 il = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(il); 2313 jl = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(jl); 2314 for (i=0; i<mbs; i++) { 2315 rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0; 2316 } 2317 2318 /* FOR EACH ROW K */ 2319 for (k = 0; k<mbs; k++){ 2320 2321 /* INITIALIZE K-TH ROW WITH ELEMENTS NONZERO IN ROW P(K) OF A */ 2322 jmin = ai[rip[k]]; jmax = ai[rip[k]+1]; 2323 if (jmin < jmax) { 2324 for (j = jmin; j < jmax; j++){ 2325 vj = riip[aj[j]]; 2326 if (k <= vj) rtmp[vj] = aa[j]; 2327 } 2328 } 2329 2330 /* MODIFY K-TH ROW BY ADDING IN THOSE ROWS I WITH U(I,K) NE 0 2331 FOR EACH ROW I TO BE ADDED IN */ 2332 dk = rtmp[k]; 2333 i = jl[k]; /* first row to be added to k_th row */ 2334 /* printf(" k=%d, pivot row = %d\n",k,i); */ 2335 2336 while (i < mbs){ 2337 nexti = jl[i]; /* next row to be added to k_th row */ 2338 /* printf(" pivot row = %d\n", nexti); */ 2339 2340 /* COMPUTE MULTIPLIER AND UPDATE DIAGONAL ELEMENT */ 2341 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 2342 uikdi = - ba[ili]*ba[i]; 2343 dk += uikdi*ba[ili]; 2344 ba[ili] = uikdi; /* update U(i,k) */ 2345 2346 /* ADD MULTIPLE OF ROW I TO K-TH ROW ... */ 2347 jmin = ili + 1; jmax = bi[i+1]; 2348 if (jmin < jmax){ 2349 for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j]; 2350 /* ... AND ADD I TO ROW LIST FOR NEXT NONZERO ENTRY */ 2351 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 2352 j = bj[jmin]; 2353 jl[i] = jl[j]; jl[j] = i; /* update jl */ 2354 } 2355 i = nexti; 2356 /* printf(" pivot row i=%d\n",i); */ 2357 } 2358 2359 /* CHECK FOR ZERO PIVOT AND SAVE DIAGONAL ELEMENT */ 2360 if (dk == 0.0){ 2361 SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,0,"Zero pivot"); 2362 } 2363 2364 /* SAVE NONZERO ENTRIES IN K-TH ROW OF U ... */ 2365 ba[k] = 1.0/dk; 2366 jmin = bi[k]; jmax = bi[k+1]; 2367 if (jmin < jmax) { 2368 for (j=jmin; j<jmax; j++){ 2369 juj = bj[j]; ba[j] = rtmp[juj]; rtmp[juj] = 0.0; 2370 } 2371 2372 /* ... AND ADD K TO ROW LIST FOR FIRST NONZERO ENTRY IN K-TH ROW */ 2373 il[k] = jmin; 2374 i = bj[jmin]; 2375 jl[k] = jl[i]; jl[i] = k; 2376 } 2377 } 2378 2379 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2380 ierr = PetscFree(il);CHKERRQ(ierr); 2381 ierr = PetscFree(jl);CHKERRQ(ierr); 2382 2383 ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr); 2384 C->factor = FACTOR_LU; 2385 C->assembled = PETSC_TRUE; 2386 PLogFlops(b->mbs); 2387 PetscFunctionReturn(0); 2388 } 2389 2390 #undef __FUNC__ 2391 #define __FUNC__ "MatCholeskyFactor_SeqSBAIJ" 2392 int MatCholeskyFactor_SeqSBAIJ(Mat A,IS perm,PetscReal f) 2393 { 2394 Mat_SeqSBAIJ *mat = (Mat_SeqSBAIJ*)A->data; 2395 int ierr,refct; 2396 Mat C; 2397 PetscOps *Abops; 2398 MatOps Aops; 2399 2400 PetscFunctionBegin; 2401 ierr = MatCholeskyFactorSymbolic(A,perm,f,&C);CHKERRQ(ierr); 2402 ierr = MatCholeskyFactorNumeric(A,&C);CHKERRQ(ierr); 2403 2404 /* free all the data structures from mat */ 2405 ierr = PetscFree(mat->a);CHKERRQ(ierr); 2406 if (!mat->singlemalloc) { 2407 ierr = PetscFree(mat->i);CHKERRQ(ierr); 2408 ierr = PetscFree(mat->j);CHKERRQ(ierr); 2409 } 2410 if (mat->diag) {ierr = PetscFree(mat->diag);CHKERRQ(ierr);} 2411 if (mat->ilen) {ierr = PetscFree(mat->ilen);CHKERRQ(ierr);} 2412 if (mat->imax) {ierr = PetscFree(mat->imax);CHKERRQ(ierr);} 2413 if (mat->solve_work) {ierr = PetscFree(mat->solve_work);CHKERRQ(ierr);} 2414 if (mat->mult_work) {ierr = PetscFree(mat->mult_work);CHKERRQ(ierr);} 2415 if (mat->icol) {ierr = ISDestroy(mat->icol);CHKERRQ(ierr);} 2416 ierr = PetscFree(mat);CHKERRQ(ierr); 2417 2418 ierr = MapDestroy(A->rmap);CHKERRQ(ierr); 2419 ierr = MapDestroy(A->cmap);CHKERRQ(ierr); 2420 2421 /* 2422 This is horrible,horrible code. We need to keep the 2423 A pointers for the bops and ops but copy everything 2424 else from C. 2425 */ 2426 Abops = A->bops; 2427 Aops = A->ops; 2428 refct = A->refct; 2429 ierr = PetscMemcpy(A,C,sizeof(struct _p_Mat));CHKERRQ(ierr); 2430 mat = (Mat_SeqSBAIJ*)A->data; 2431 PLogObjectParent(A,mat->icol); 2432 2433 A->bops = Abops; 2434 A->ops = Aops; 2435 A->qlist = 0; 2436 A->refct = refct; 2437 /* copy over the type_name and name */ 2438 ierr = PetscStrallocpy(C->type_name,&A->type_name);CHKERRQ(ierr); 2439 ierr = PetscStrallocpy(C->name,&A->name);CHKERRQ(ierr); 2440 2441 PetscHeaderDestroy(C); 2442 PetscFunctionReturn(0); 2443 } 2444 2445 2446