1 /* Using Modified Sparse Row (MSR) storage. 2 See page 85, "Iterative Methods ..." by Saad. */ 3 4 /*$Id: sbaijfact.c,v 1.15 2000/09/11 15:02:08 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 #include "include/petscis.h" 13 14 #undef __FUNC__ 15 #define __FUNC__ "MatCholeskyFactorSymbolic_SeqSBAIJ" 16 int MatCholeskyFactorSymbolic_SeqSBAIJ(Mat A,IS perm,PetscReal f,Mat *B) 17 { 18 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b; 19 IS iperm; 20 int *rip,*riip,ierr,i,mbs = a->mbs,*ai,*aj; 21 int *jutmp,bs = a->bs,bs2=a->bs2; 22 int m,nzi,realloc = 0; 23 int *jl,*q,jumin,jmin,jmax,juptr,nzk,qm,*iu,*ju,k,j,vj,umax,maxadd; 24 /* PetscTruth *ident; */ 25 PetscTruth flg_perm = PETSC_FALSE; /* non-trivial permutation */ 26 27 PetscFunctionBegin; 28 PetscValidHeaderSpecific(perm,IS_COOKIE); 29 if (A->M != A->N) SETERRQ(PETSC_ERR_ARG_WRONG,0,"matrix must be square"); 30 31 #ifdef ISIdentity 32 ierr = ISIdentity(perm,ident);CHKERRQ(ierr); 33 if (!*ident) { /* for a non-trivial perm, the matrix A in SBAIJ format needs to be 34 re-indexed so that A(perm(i),iperm(k)) is stored in the upper 35 triangle. */ 36 SETERRQ(PETSC_ERR_ARG_CORRUPT,0,"Call MatReIndexingSeqSBAIJ() to re-indexing (ai,aj,a)"); 37 } 38 #endif 39 40 if (!flg_perm){ /* no permutation */ 41 ai = a->i; aj = a->j; 42 } else { /* non-trivial permutation */ 43 /* 44 printf("2, called symbolic factor\n"); 45 ierr = ISView(perm, VIEWER_STDOUT_SELF);CHKERRA(ierr); 46 */ 47 ierr = MatReorderingSeqSBAIJ(A, perm);CHKERRA(ierr); 48 /* a = (Mat_SeqSBAIJ*)A->data; */ 49 50 ai = a->inew; aj = a->jnew; 51 printf("ainew= %d %d \n",a->inew[0],a->inew[mbs]); 52 printf("ajnew=%d %d\n",a->jnew[0],a->jnew[a->i[mbs]-1]); 53 } 54 55 ierr = ISInvertPermutation(perm,PETSC_DECIDE,&iperm);CHKERRQ(ierr); 56 ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr); 57 ierr = ISGetIndices(iperm,&riip);CHKERRQ(ierr); 58 59 /* initialization */ 60 /* Don't know how many column pointers are needed so estimate. 61 Use Modified Sparse Row storage for u and ju, see Sasd pp.85 */ 62 iu = (int*)PetscMalloc((mbs+1)*sizeof(int));CHKPTRQ(iu); 63 umax = (int)(f*ai[mbs] + 1); umax += mbs + 1; 64 ju = (int*)PetscMalloc(umax*sizeof(int));CHKPTRQ(ju); 65 iu[0] = mbs+1; 66 juptr = mbs; 67 jl = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(jl); 68 q = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(q); 69 for (i=0; i<mbs; i++){ 70 jl[i] = mbs; q[i] = 0; 71 } 72 73 /* for each row k */ 74 for (k=0; k<mbs; k++){ 75 nzk = 0; /* num. of nz blocks in k-th block row with diagonal block excluded */ 76 q[k] = mbs; 77 /* initialize nonzero structure of k-th row to row rip[k] of A */ 78 jmin = ai[rip[k]]; 79 jmax = ai[rip[k]+1]; 80 for (j=jmin; j<jmax; j++){ 81 vj = riip[aj[j]]; /* col. value */ 82 if(vj > k){ 83 qm = k; 84 do { 85 m = qm; qm = q[m]; 86 } while(qm < vj); 87 if (qm == vj) { 88 printf(" error: duplicate entry in A\n"); break; 89 } 90 nzk++; 91 q[m] = vj; 92 q[vj] = qm; 93 } /* if(vj > k) */ 94 } /* for (j=jmin; j<jmax; j++) */ 95 96 /* modify nonzero structure of k-th row by computing fill-in 97 for each row i to be merged in */ 98 i = k; 99 i = jl[i]; /* next pivot row (== mbs for symbolic factorization) */ 100 /* printf(" next pivot row i=%d\n",i); */ 101 while (i < mbs){ 102 /* merge row i into k-th row */ 103 nzi = iu[i+1] - (iu[i]+1); 104 jmin = iu[i] + 1; jmax = iu[i] + nzi; 105 qm = k; 106 for (j=jmin; j<jmax+1; j++){ 107 vj = ju[j]; 108 do { 109 m = qm; qm = q[m]; 110 } while (qm < vj); 111 if (qm != vj){ 112 nzk++; q[m] = vj; q[vj] = qm; qm = vj; 113 } 114 } 115 i = jl[i]; /* next pivot row */ 116 } 117 118 /* add k to row list for first nonzero element in k-th row */ 119 if (nzk > 0){ 120 i = q[k]; /* col value of first nonzero element in U(k, k+1:mbs-1) */ 121 jl[k] = jl[i]; jl[i] = k; 122 } 123 iu[k+1] = iu[k] + nzk; /* printf(" iu[%d]=%d, umax=%d\n", k+1, iu[k+1],umax);*/ 124 125 /* allocate more space to ju if needed */ 126 if (iu[k+1] > umax) { printf("allocate more space, iu[%d]=%d > umax=%d\n",k+1, iu[k+1],umax); 127 /* estimate how much additional space we will need */ 128 /* use the strategy suggested by David Hysom <hysom@perch-t.icase.edu> */ 129 /* just double the memory each time */ 130 maxadd = umax; 131 if (maxadd < nzk) maxadd = (mbs-k)*(nzk+1)/2; 132 umax += maxadd; 133 134 /* allocate a longer ju */ 135 jutmp = (int*)PetscMalloc(umax*sizeof(int));CHKPTRQ(jutmp); 136 ierr = PetscMemcpy(jutmp,ju,iu[k]*sizeof(int));CHKERRQ(ierr); 137 ierr = PetscFree(ju);CHKERRQ(ierr); 138 ju = jutmp; 139 realloc++; /* count how many times we realloc */ 140 } 141 142 /* save nonzero structure of k-th row in ju */ 143 i=k; 144 jumin = juptr + 1; juptr += nzk; 145 for (j=jumin; j<juptr+1; j++){ 146 i=q[i]; 147 ju[j]=i; 148 } 149 } 150 151 if (ai[mbs] != 0) { 152 PetscReal af = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 153 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Reallocs %d Fill ratio:given %g needed %g\n",realloc,f,af); 154 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Run with -pc_lu_fill %g or use \n",af); 155 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:PCLUSetFill(pc,%g);\n",af); 156 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:for best performance.\n"); 157 } else { 158 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Empty matrix.\n"); 159 } 160 161 ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr); 162 ierr = ISRestoreIndices(iperm,&riip);CHKERRQ(ierr); 163 164 ierr = PetscFree(q);CHKERRQ(ierr); 165 ierr = PetscFree(jl);CHKERRQ(ierr); 166 167 /* put together the new matrix */ 168 ierr = MatCreateSeqSBAIJ(A->comm,bs,bs*mbs,bs*mbs,0,PETSC_NULL,B);CHKERRQ(ierr); 169 PLogObjectParent(*B,iperm); 170 b = (Mat_SeqSBAIJ*)(*B)->data; 171 ierr = PetscFree(b->imax);CHKERRQ(ierr); 172 b->singlemalloc = PETSC_FALSE; 173 /* the next line frees the default space generated by the Create() */ 174 ierr = PetscFree(b->a);CHKERRQ(ierr); 175 ierr = PetscFree(b->ilen);CHKERRQ(ierr); 176 b->a = (MatScalar*)PetscMalloc((iu[mbs]+1)*sizeof(MatScalar)*bs2);CHKPTRQ(b->a); 177 b->j = ju; 178 b->i = iu; 179 b->diag = 0; 180 b->ilen = 0; 181 b->imax = 0; 182 b->row = perm; 183 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 184 b->icol = iperm; 185 b->solve_work = (Scalar*)PetscMalloc((bs*mbs+bs)*sizeof(Scalar));CHKPTRQ(b->solve_work); 186 /* In b structure: Free imax, ilen, old a, old j. 187 Allocate idnew, solve_work, new a, new j */ 188 PLogObjectMemory(*B,(iu[mbs]-mbs)*(sizeof(int)+sizeof(MatScalar))); 189 b->s_maxnz = b->s_nz = iu[mbs]; 190 191 (*B)->factor = FACTOR_LU; 192 (*B)->info.factor_mallocs = realloc; 193 (*B)->info.fill_ratio_given = f; 194 if (ai[mbs] != 0) { 195 (*B)->info.fill_ratio_needed = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 196 } else { 197 (*B)->info.fill_ratio_needed = 0.0; 198 } 199 200 201 PetscFunctionReturn(0); 202 } 203 204 #undef __FUNC__ 205 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_N" 206 int MatCholeskyFactorNumeric_SeqSBAIJ_N(Mat A,Mat *B) 207 { 208 Mat C = *B; 209 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 210 IS isrow = b->row,isicol = b->icol; 211 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 212 int *ajtmpold,*ajtmp,nz,row,bslog,*ai=a->i,*aj=a->j,k,flg; 213 int *diag_offset=b->diag,diag,bs=a->bs,bs2 = a->bs2,*v_pivots,*pj; 214 MatScalar *ba = b->a,*aa = a->a,*pv,*v,*rtmp,*multiplier,*v_work,*pc,*w; 215 216 PetscFunctionBegin; 217 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 218 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 219 rtmp = (MatScalar*)PetscMalloc(bs2*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 220 ierr = PetscMemzero(rtmp,bs2*(n+1)*sizeof(MatScalar));CHKERRQ(ierr); 221 /* generate work space needed by dense LU factorization */ 222 v_work = (MatScalar*)PetscMalloc(bs*sizeof(int) + (bs+bs2)*sizeof(MatScalar));CHKPTRQ(v_work); 223 multiplier = v_work + bs; 224 v_pivots = (int*)(multiplier + bs2); 225 226 /* flops in while loop */ 227 bslog = 2*bs*bs2; 228 229 for (i=0; i<n; i++) { 230 nz = bi[i+1] - bi[i]; 231 ajtmp = bj + bi[i]; 232 for (j=0; j<nz; j++) { 233 ierr = PetscMemzero(rtmp+bs2*ajtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 234 } 235 /* load in initial (unfactored row) */ 236 nz = ai[r[i]+1] - ai[r[i]]; 237 ajtmpold = aj + ai[r[i]]; 238 v = aa + bs2*ai[r[i]]; 239 for (j=0; j<nz; j++) { 240 ierr = PetscMemcpy(rtmp+bs2*ic[ajtmpold[j]],v+bs2*j,bs2*sizeof(MatScalar));CHKERRQ(ierr); 241 } 242 row = *ajtmp++; 243 while (row < i) { 244 pc = rtmp + bs2*row; 245 /* if (*pc) { */ 246 for (flg=0,k=0; k<bs2; k++) { if (pc[k]!=0.0) { flg =1; break; }} 247 if (flg) { 248 pv = ba + bs2*diag_offset[row]; 249 pj = bj + diag_offset[row] + 1; 250 Kernel_A_gets_A_times_B(bs,pc,pv,multiplier); 251 nz = bi[row+1] - diag_offset[row] - 1; 252 pv += bs2; 253 for (j=0; j<nz; j++) { 254 Kernel_A_gets_A_minus_B_times_C(bs,rtmp+bs2*pj[j],pc,pv+bs2*j); 255 } 256 PLogFlops(bslog*(nz+1)-bs); 257 } 258 row = *ajtmp++; 259 } 260 /* finished row so stick it into b->a */ 261 pv = ba + bs2*bi[i]; 262 pj = bj + bi[i]; 263 nz = bi[i+1] - bi[i]; 264 for (j=0; j<nz; j++) { 265 ierr = PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 266 } 267 diag = diag_offset[i] - bi[i]; 268 /* invert diagonal block */ 269 w = pv + bs2*diag; 270 Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work); 271 } 272 273 ierr = PetscFree(rtmp);CHKERRQ(ierr); 274 ierr = PetscFree(v_work);CHKERRQ(ierr); 275 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 276 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 277 C->factor = FACTOR_LU; 278 C->assembled = PETSC_TRUE; 279 PLogFlops(1.3333*bs*bs2*b->mbs); /* from inverting diagonal blocks */ 280 PetscFunctionReturn(0); 281 } 282 283 /* 284 Version for when blocks are 7 by 7 285 */ 286 #undef __FUNC__ 287 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_7" 288 int MatCholeskyFactorNumeric_SeqSBAIJ_7(Mat A,Mat *B) 289 { 290 Mat C = *B; 291 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 292 IS isrow = b->row,isicol = b->icol; 293 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 294 int *ajtmpold,*ajtmp,nz,row; 295 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 296 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 297 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 298 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 299 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 300 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 301 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 302 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 303 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 304 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 305 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 306 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 307 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 308 MatScalar *ba = b->a,*aa = a->a; 309 310 PetscFunctionBegin; 311 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 312 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 313 rtmp = (MatScalar*)PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 314 315 for (i=0; i<n; i++) { 316 nz = bi[i+1] - bi[i]; 317 ajtmp = bj + bi[i]; 318 for (j=0; j<nz; j++) { 319 x = rtmp+49*ajtmp[j]; 320 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 321 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 322 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 323 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 324 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 325 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 326 } 327 /* load in initial (unfactored row) */ 328 idx = r[i]; 329 nz = ai[idx+1] - ai[idx]; 330 ajtmpold = aj + ai[idx]; 331 v = aa + 49*ai[idx]; 332 for (j=0; j<nz; j++) { 333 x = rtmp+49*ic[ajtmpold[j]]; 334 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 335 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 336 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 337 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 338 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 339 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 340 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 341 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 342 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 343 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 344 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 345 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 346 x[48] = v[48]; 347 v += 49; 348 } 349 row = *ajtmp++; 350 while (row < i) { 351 pc = rtmp + 49*row; 352 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 353 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 354 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 355 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 356 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 357 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 358 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 359 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 360 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 361 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 362 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 363 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 364 p49 = pc[48]; 365 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 366 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 367 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 368 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 369 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 370 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 371 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 372 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 373 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 374 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 375 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 376 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 377 p49 != 0.0) { 378 pv = ba + 49*diag_offset[row]; 379 pj = bj + diag_offset[row] + 1; 380 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 381 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 382 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 383 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 384 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 385 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 386 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 387 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 388 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 389 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 390 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 391 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 392 x49 = pv[48]; 393 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 394 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 395 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 396 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 397 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 398 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 399 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 400 401 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 402 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 403 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 404 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 405 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 406 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 407 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 408 409 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 410 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 411 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 412 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 413 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 414 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 415 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 416 417 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 418 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 419 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 420 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 421 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 422 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 423 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 424 425 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 426 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 427 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 428 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 429 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 430 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 431 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 432 433 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 434 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 435 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 436 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 437 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 438 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 439 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 440 441 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 442 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 443 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 444 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 445 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 446 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 447 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 448 449 nz = bi[row+1] - diag_offset[row] - 1; 450 pv += 49; 451 for (j=0; j<nz; j++) { 452 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 453 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 454 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 455 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 456 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 457 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 458 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 459 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 460 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 461 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 462 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 463 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 464 x49 = pv[48]; 465 x = rtmp + 49*pj[j]; 466 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 467 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 468 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 469 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 470 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 471 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 472 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 473 474 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 475 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 476 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 477 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 478 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 479 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 480 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 481 482 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 483 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 484 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 485 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 486 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 487 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 488 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 489 490 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 491 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 492 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 493 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 494 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 495 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 496 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 497 498 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 499 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 500 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 501 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 502 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 503 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 504 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 505 506 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 507 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 508 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 509 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 510 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 511 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 512 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 513 514 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 515 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 516 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 517 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 518 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 519 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 520 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 521 pv += 49; 522 } 523 PLogFlops(686*nz+637); 524 } 525 row = *ajtmp++; 526 } 527 /* finished row so stick it into b->a */ 528 pv = ba + 49*bi[i]; 529 pj = bj + bi[i]; 530 nz = bi[i+1] - bi[i]; 531 for (j=0; j<nz; j++) { 532 x = rtmp+49*pj[j]; 533 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 534 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 535 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 536 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 537 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 538 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 539 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 540 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 541 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 542 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 543 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 544 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 545 pv[48] = x[48]; 546 pv += 49; 547 } 548 /* invert diagonal block */ 549 w = ba + 49*diag_offset[i]; 550 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 551 } 552 553 ierr = PetscFree(rtmp);CHKERRQ(ierr); 554 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 555 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 556 C->factor = FACTOR_LU; 557 C->assembled = PETSC_TRUE; 558 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 559 PetscFunctionReturn(0); 560 } 561 562 /* 563 Version for when blocks are 7 by 7 Using natural ordering 564 */ 565 #undef __FUNC__ 566 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering" 567 int MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering(Mat A,Mat *B) 568 { 569 Mat C = *B; 570 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 571 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 572 int *ajtmpold,*ajtmp,nz,row; 573 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 574 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 575 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 576 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 577 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 578 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 579 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 580 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 581 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 582 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 583 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 584 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 585 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 586 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 587 MatScalar *ba = b->a,*aa = a->a; 588 589 PetscFunctionBegin; 590 rtmp = (MatScalar*)PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 591 for (i=0; i<n; i++) { 592 nz = bi[i+1] - bi[i]; 593 ajtmp = bj + bi[i]; 594 for (j=0; j<nz; j++) { 595 x = rtmp+49*ajtmp[j]; 596 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 597 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 598 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 599 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 600 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 601 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 602 } 603 /* load in initial (unfactored row) */ 604 nz = ai[i+1] - ai[i]; 605 ajtmpold = aj + ai[i]; 606 v = aa + 49*ai[i]; 607 for (j=0; j<nz; j++) { 608 x = rtmp+49*ajtmpold[j]; 609 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 610 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 611 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 612 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 613 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 614 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 615 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 616 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 617 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 618 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 619 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 620 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 621 x[48] = v[48]; 622 v += 49; 623 } 624 row = *ajtmp++; 625 while (row < i) { 626 pc = rtmp + 49*row; 627 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 628 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 629 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 630 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 631 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 632 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 633 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 634 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 635 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 636 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 637 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 638 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 639 p49 = pc[48]; 640 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 641 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 642 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 643 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 644 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 645 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 646 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 647 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 648 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 649 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 650 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 651 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 652 p49 != 0.0) { 653 pv = ba + 49*diag_offset[row]; 654 pj = bj + diag_offset[row] + 1; 655 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 656 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 657 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 658 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 659 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 660 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 661 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 662 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 663 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 664 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 665 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 666 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 667 x49 = pv[48]; 668 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 669 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 670 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 671 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 672 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 673 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 674 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 675 676 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 677 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 678 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 679 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 680 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 681 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 682 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 683 684 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 685 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 686 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 687 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 688 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 689 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 690 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 691 692 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 693 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 694 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 695 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 696 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 697 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 698 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 699 700 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 701 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 702 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 703 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 704 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 705 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 706 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 707 708 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 709 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 710 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 711 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 712 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 713 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 714 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 715 716 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 717 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 718 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 719 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 720 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 721 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 722 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 723 724 nz = bi[row+1] - diag_offset[row] - 1; 725 pv += 49; 726 for (j=0; j<nz; j++) { 727 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 728 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 729 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 730 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 731 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 732 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 733 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 734 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 735 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 736 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 737 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 738 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 739 x49 = pv[48]; 740 x = rtmp + 49*pj[j]; 741 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 742 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 743 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 744 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 745 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 746 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 747 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 748 749 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 750 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 751 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 752 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 753 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 754 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 755 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 756 757 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 758 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 759 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 760 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 761 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 762 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 763 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 764 765 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 766 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 767 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 768 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 769 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 770 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 771 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 772 773 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 774 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 775 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 776 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 777 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 778 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 779 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 780 781 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 782 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 783 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 784 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 785 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 786 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 787 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 788 789 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 790 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 791 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 792 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 793 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 794 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 795 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 796 pv += 49; 797 } 798 PLogFlops(686*nz+637); 799 } 800 row = *ajtmp++; 801 } 802 /* finished row so stick it into b->a */ 803 pv = ba + 49*bi[i]; 804 pj = bj + bi[i]; 805 nz = bi[i+1] - bi[i]; 806 for (j=0; j<nz; j++) { 807 x = rtmp+49*pj[j]; 808 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 809 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 810 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 811 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 812 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 813 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 814 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 815 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 816 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 817 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 818 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 819 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 820 pv[48] = x[48]; 821 pv += 49; 822 } 823 /* invert diagonal block */ 824 w = ba + 49*diag_offset[i]; 825 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 826 } 827 828 ierr = PetscFree(rtmp);CHKERRQ(ierr); 829 C->factor = FACTOR_LU; 830 C->assembled = PETSC_TRUE; 831 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 832 PetscFunctionReturn(0); 833 } 834 835 /* ------------------------------------------------------------*/ 836 /* 837 Version for when blocks are 6 by 6 838 */ 839 #undef __FUNC__ 840 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_6" 841 int MatCholeskyFactorNumeric_SeqSBAIJ_6(Mat A,Mat *B) 842 { 843 Mat C = *B; 844 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 845 IS isrow = b->row,isicol = b->icol; 846 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 847 int *ajtmpold,*ajtmp,nz,row; 848 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 849 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 850 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 851 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 852 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 853 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 854 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 855 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 856 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 857 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 858 MatScalar *ba = b->a,*aa = a->a; 859 860 PetscFunctionBegin; 861 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 862 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 863 rtmp = (MatScalar*)PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 864 865 for (i=0; i<n; i++) { 866 nz = bi[i+1] - bi[i]; 867 ajtmp = bj + bi[i]; 868 for (j=0; j<nz; j++) { 869 x = rtmp+36*ajtmp[j]; 870 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 871 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 872 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 873 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 874 x[34] = x[35] = 0.0 ; 875 } 876 /* load in initial (unfactored row) */ 877 idx = r[i]; 878 nz = ai[idx+1] - ai[idx]; 879 ajtmpold = aj + ai[idx]; 880 v = aa + 36*ai[idx]; 881 for (j=0; j<nz; j++) { 882 x = rtmp+36*ic[ajtmpold[j]]; 883 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 884 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 885 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 886 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 887 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 888 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 889 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 890 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 891 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 892 v += 36; 893 } 894 row = *ajtmp++; 895 while (row < i) { 896 pc = rtmp + 36*row; 897 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 898 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 899 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 900 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 901 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 902 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 903 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 904 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 905 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 906 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 907 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 908 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 909 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 910 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 911 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 912 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 913 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 914 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 915 pv = ba + 36*diag_offset[row]; 916 pj = bj + diag_offset[row] + 1; 917 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 918 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 919 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 920 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 921 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 922 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 923 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 924 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 925 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 926 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 927 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 928 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 929 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 930 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 931 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 932 933 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 934 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 935 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 936 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 937 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 938 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 939 940 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 941 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 942 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 943 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 944 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 945 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 946 947 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 948 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 949 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 950 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 951 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 952 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 953 954 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 955 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 956 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 957 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 958 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 959 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 960 961 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 962 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 963 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 964 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 965 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 966 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 967 968 nz = bi[row+1] - diag_offset[row] - 1; 969 pv += 36; 970 for (j=0; j<nz; j++) { 971 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 972 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 973 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 974 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 975 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 976 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 977 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 978 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 979 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 980 x = rtmp + 36*pj[j]; 981 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 982 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 983 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 984 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 985 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 986 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 987 988 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 989 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 990 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 991 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 992 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 993 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 994 995 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 996 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 997 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 998 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 999 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 1000 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 1001 1002 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 1003 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 1004 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 1005 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 1006 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 1007 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 1008 1009 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 1010 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 1011 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 1012 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 1013 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 1014 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 1015 1016 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 1017 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 1018 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 1019 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 1020 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 1021 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 1022 1023 pv += 36; 1024 } 1025 PLogFlops(432*nz+396); 1026 } 1027 row = *ajtmp++; 1028 } 1029 /* finished row so stick it into b->a */ 1030 pv = ba + 36*bi[i]; 1031 pj = bj + bi[i]; 1032 nz = bi[i+1] - bi[i]; 1033 for (j=0; j<nz; j++) { 1034 x = rtmp+36*pj[j]; 1035 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1036 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1037 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1038 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1039 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1040 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1041 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1042 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1043 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1044 pv += 36; 1045 } 1046 /* invert diagonal block */ 1047 w = ba + 36*diag_offset[i]; 1048 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1049 } 1050 1051 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1052 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1053 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1054 C->factor = FACTOR_LU; 1055 C->assembled = PETSC_TRUE; 1056 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1057 PetscFunctionReturn(0); 1058 } 1059 /* 1060 Version for when blocks are 6 by 6 Using natural ordering 1061 */ 1062 #undef __FUNC__ 1063 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering" 1064 int MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering(Mat A,Mat *B) 1065 { 1066 Mat C = *B; 1067 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1068 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1069 int *ajtmpold,*ajtmp,nz,row; 1070 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1071 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1072 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1073 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1074 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1075 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1076 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1077 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1078 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 1079 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 1080 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 1081 MatScalar *ba = b->a,*aa = a->a; 1082 1083 PetscFunctionBegin; 1084 rtmp = (MatScalar*)PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1085 for (i=0; i<n; i++) { 1086 nz = bi[i+1] - bi[i]; 1087 ajtmp = bj + bi[i]; 1088 for (j=0; j<nz; j++) { 1089 x = rtmp+36*ajtmp[j]; 1090 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1091 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1092 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 1093 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 1094 x[34] = x[35] = 0.0 ; 1095 } 1096 /* load in initial (unfactored row) */ 1097 nz = ai[i+1] - ai[i]; 1098 ajtmpold = aj + ai[i]; 1099 v = aa + 36*ai[i]; 1100 for (j=0; j<nz; j++) { 1101 x = rtmp+36*ajtmpold[j]; 1102 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1103 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 1104 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 1105 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 1106 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 1107 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1108 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 1109 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 1110 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 1111 v += 36; 1112 } 1113 row = *ajtmp++; 1114 while (row < i) { 1115 pc = rtmp + 36*row; 1116 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1117 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 1118 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 1119 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 1120 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 1121 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1122 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 1123 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 1124 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 1125 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 1126 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 1127 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 1128 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 1129 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 1130 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 1131 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 1132 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 1133 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 1134 pv = ba + 36*diag_offset[row]; 1135 pj = bj + diag_offset[row] + 1; 1136 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1137 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1138 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1139 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1140 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1141 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1142 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1143 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1144 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1145 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 1146 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 1147 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 1148 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 1149 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 1150 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 1151 1152 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 1153 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 1154 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 1155 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 1156 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 1157 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 1158 1159 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 1160 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 1161 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 1162 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 1163 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 1164 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 1165 1166 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 1167 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 1168 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 1169 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 1170 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 1171 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 1172 1173 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 1174 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 1175 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 1176 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 1177 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 1178 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 1179 1180 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 1181 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 1182 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 1183 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 1184 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 1185 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 1186 1187 nz = bi[row+1] - diag_offset[row] - 1; 1188 pv += 36; 1189 for (j=0; j<nz; j++) { 1190 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1191 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1192 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1193 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1194 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1195 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1196 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1197 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1198 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1199 x = rtmp + 36*pj[j]; 1200 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 1201 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 1202 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 1203 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 1204 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 1205 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 1206 1207 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 1208 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 1209 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 1210 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 1211 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 1212 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 1213 1214 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 1215 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 1216 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 1217 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 1218 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 1219 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 1220 1221 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 1222 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 1223 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 1224 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 1225 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 1226 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 1227 1228 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 1229 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 1230 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 1231 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 1232 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 1233 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 1234 1235 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 1236 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 1237 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 1238 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 1239 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 1240 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 1241 1242 pv += 36; 1243 } 1244 PLogFlops(432*nz+396); 1245 } 1246 row = *ajtmp++; 1247 } 1248 /* finished row so stick it into b->a */ 1249 pv = ba + 36*bi[i]; 1250 pj = bj + bi[i]; 1251 nz = bi[i+1] - bi[i]; 1252 for (j=0; j<nz; j++) { 1253 x = rtmp+36*pj[j]; 1254 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1255 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1256 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1257 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1258 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1259 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1260 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1261 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1262 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1263 pv += 36; 1264 } 1265 /* invert diagonal block */ 1266 w = ba + 36*diag_offset[i]; 1267 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1268 } 1269 1270 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1271 C->factor = FACTOR_LU; 1272 C->assembled = PETSC_TRUE; 1273 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1274 PetscFunctionReturn(0); 1275 } 1276 1277 /* 1278 Version for when blocks are 5 by 5 1279 */ 1280 #undef __FUNC__ 1281 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_5" 1282 int MatCholeskyFactorNumeric_SeqSBAIJ_5(Mat A,Mat *B) 1283 { 1284 Mat C = *B; 1285 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1286 IS isrow = b->row,isicol = b->icol; 1287 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1288 int *ajtmpold,*ajtmp,nz,row; 1289 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 1290 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1291 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1292 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1293 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 1294 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 1295 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1296 MatScalar *ba = b->a,*aa = a->a; 1297 1298 PetscFunctionBegin; 1299 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1300 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1301 rtmp = (MatScalar*)PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1302 1303 for (i=0; i<n; i++) { 1304 nz = bi[i+1] - bi[i]; 1305 ajtmp = bj + bi[i]; 1306 for (j=0; j<nz; j++) { 1307 x = rtmp+25*ajtmp[j]; 1308 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1309 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1310 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1311 } 1312 /* load in initial (unfactored row) */ 1313 idx = r[i]; 1314 nz = ai[idx+1] - ai[idx]; 1315 ajtmpold = aj + ai[idx]; 1316 v = aa + 25*ai[idx]; 1317 for (j=0; j<nz; j++) { 1318 x = rtmp+25*ic[ajtmpold[j]]; 1319 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1320 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1321 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1322 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; 1323 x[18] = v[18]; x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; 1324 x[22] = v[22]; x[23] = v[23]; x[24] = v[24]; 1325 v += 25; 1326 } 1327 row = *ajtmp++; 1328 while (row < i) { 1329 pc = rtmp + 25*row; 1330 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1331 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1332 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1333 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; 1334 p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1335 p25 = pc[24]; 1336 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1337 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1338 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1339 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || 1340 p20 != 0.0 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || 1341 p24 != 0.0 || p25 != 0.0) { 1342 pv = ba + 25*diag_offset[row]; 1343 pj = bj + diag_offset[row] + 1; 1344 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1345 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1346 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1347 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1348 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; 1349 x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1350 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1351 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1352 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1353 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1354 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1355 1356 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1357 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1358 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1359 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1360 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1361 1362 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1363 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1364 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1365 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1366 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1367 1368 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1369 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1370 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1371 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1372 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1373 1374 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1375 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1376 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1377 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1378 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1379 1380 nz = bi[row+1] - diag_offset[row] - 1; 1381 pv += 25; 1382 for (j=0; j<nz; j++) { 1383 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1384 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1385 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1386 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; 1387 x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; 1388 x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1389 x = rtmp + 25*pj[j]; 1390 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1391 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1392 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1393 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1394 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1395 1396 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1397 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1398 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1399 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1400 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1401 1402 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1403 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1404 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1405 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1406 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1407 1408 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1409 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1410 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1411 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1412 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1413 1414 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1415 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1416 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1417 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1418 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1419 1420 pv += 25; 1421 } 1422 PLogFlops(250*nz+225); 1423 } 1424 row = *ajtmp++; 1425 } 1426 /* finished row so stick it into b->a */ 1427 pv = ba + 25*bi[i]; 1428 pj = bj + bi[i]; 1429 nz = bi[i+1] - bi[i]; 1430 for (j=0; j<nz; j++) { 1431 x = rtmp+25*pj[j]; 1432 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1433 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1434 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1435 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; 1436 pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; 1437 pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; pv[24] = x[24]; 1438 pv += 25; 1439 } 1440 /* invert diagonal block */ 1441 w = ba + 25*diag_offset[i]; 1442 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1443 } 1444 1445 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1446 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1447 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1448 C->factor = FACTOR_LU; 1449 C->assembled = PETSC_TRUE; 1450 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1451 PetscFunctionReturn(0); 1452 } 1453 /* 1454 Version for when blocks are 5 by 5 Using natural ordering 1455 */ 1456 #undef __FUNC__ 1457 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering" 1458 int MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering(Mat A,Mat *B) 1459 { 1460 Mat C = *B; 1461 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1462 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1463 int *ajtmpold,*ajtmp,nz,row; 1464 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1465 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1466 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1467 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1468 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1469 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1470 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1471 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1472 MatScalar *ba = b->a,*aa = a->a; 1473 1474 PetscFunctionBegin; 1475 rtmp = (MatScalar*)PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1476 for (i=0; i<n; i++) { 1477 nz = bi[i+1] - bi[i]; 1478 ajtmp = bj + bi[i]; 1479 for (j=0; j<nz; j++) { 1480 x = rtmp+25*ajtmp[j]; 1481 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1482 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1483 x[16] = x[17] = x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1484 } 1485 /* load in initial (unfactored row) */ 1486 nz = ai[i+1] - ai[i]; 1487 ajtmpold = aj + ai[i]; 1488 v = aa + 25*ai[i]; 1489 for (j=0; j<nz; j++) { 1490 x = rtmp+25*ajtmpold[j]; 1491 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1492 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1493 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1494 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; 1495 x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1496 x[24] = v[24]; 1497 v += 25; 1498 } 1499 row = *ajtmp++; 1500 while (row < i) { 1501 pc = rtmp + 25*row; 1502 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1503 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1504 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1505 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; 1506 p19 = pc[18]; p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; 1507 p24 = pc[23]; p25 = pc[24]; 1508 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1509 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1510 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1511 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 1512 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || p25 != 0.0) { 1513 pv = ba + 25*diag_offset[row]; 1514 pj = bj + diag_offset[row] + 1; 1515 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1516 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1517 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1518 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; 1519 x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1520 x25 = pv[24]; 1521 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1522 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1523 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1524 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1525 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1526 1527 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1528 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1529 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1530 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1531 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1532 1533 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1534 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1535 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1536 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1537 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1538 1539 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1540 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1541 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1542 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1543 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1544 1545 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1546 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1547 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1548 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1549 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1550 1551 nz = bi[row+1] - diag_offset[row] - 1; 1552 pv += 25; 1553 for (j=0; j<nz; j++) { 1554 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1555 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1556 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1557 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1558 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; 1559 x24 = pv[23]; x25 = pv[24]; 1560 x = rtmp + 25*pj[j]; 1561 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1562 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1563 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1564 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1565 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1566 1567 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1568 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1569 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1570 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1571 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1572 1573 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1574 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1575 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1576 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1577 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1578 1579 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1580 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1581 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1582 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1583 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1584 1585 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1586 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1587 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1588 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1589 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1590 pv += 25; 1591 } 1592 PLogFlops(250*nz+225); 1593 } 1594 row = *ajtmp++; 1595 } 1596 /* finished row so stick it into b->a */ 1597 pv = ba + 25*bi[i]; 1598 pj = bj + bi[i]; 1599 nz = bi[i+1] - bi[i]; 1600 for (j=0; j<nz; j++) { 1601 x = rtmp+25*pj[j]; 1602 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1603 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1604 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1605 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; pv[17] = x[17]; 1606 pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; 1607 pv[23] = x[23]; pv[24] = x[24]; 1608 pv += 25; 1609 } 1610 /* invert diagonal block */ 1611 w = ba + 25*diag_offset[i]; 1612 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1613 } 1614 1615 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1616 C->factor = FACTOR_LU; 1617 C->assembled = PETSC_TRUE; 1618 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1619 PetscFunctionReturn(0); 1620 } 1621 1622 /* 1623 Version for when blocks are 4 by 4 1624 */ 1625 #undef __FUNC__ 1626 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_4" 1627 int MatCholeskyFactorNumeric_SeqSBAIJ_4(Mat A,Mat *B) 1628 { 1629 Mat C = *B; 1630 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1631 IS isrow = b->row,isicol = b->icol; 1632 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1633 int *ajtmpold,*ajtmp,nz,row; 1634 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 1635 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1636 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1637 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1638 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1639 MatScalar m13,m14,m15,m16; 1640 MatScalar *ba = b->a,*aa = a->a; 1641 1642 PetscFunctionBegin; 1643 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1644 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1645 rtmp = (MatScalar*)PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1646 1647 for (i=0; i<n; i++) { 1648 nz = bi[i+1] - bi[i]; 1649 ajtmp = bj + bi[i]; 1650 for (j=0; j<nz; j++) { 1651 x = rtmp+16*ajtmp[j]; 1652 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1653 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1654 } 1655 /* load in initial (unfactored row) */ 1656 idx = r[i]; 1657 nz = ai[idx+1] - ai[idx]; 1658 ajtmpold = aj + ai[idx]; 1659 v = aa + 16*ai[idx]; 1660 for (j=0; j<nz; j++) { 1661 x = rtmp+16*ic[ajtmpold[j]]; 1662 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1663 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1664 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1665 x[14] = v[14]; x[15] = v[15]; 1666 v += 16; 1667 } 1668 row = *ajtmp++; 1669 while (row < i) { 1670 pc = rtmp + 16*row; 1671 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1672 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1673 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1674 p15 = pc[14]; p16 = pc[15]; 1675 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1676 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1677 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1678 || p16 != 0.0) { 1679 pv = ba + 16*diag_offset[row]; 1680 pj = bj + diag_offset[row] + 1; 1681 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1682 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1683 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1684 x15 = pv[14]; x16 = pv[15]; 1685 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1686 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1687 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1688 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1689 1690 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1691 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1692 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1693 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1694 1695 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1696 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1697 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1698 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1699 1700 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1701 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1702 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1703 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1704 1705 nz = bi[row+1] - diag_offset[row] - 1; 1706 pv += 16; 1707 for (j=0; j<nz; j++) { 1708 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1709 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1710 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1711 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1712 x = rtmp + 16*pj[j]; 1713 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1714 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1715 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1716 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1717 1718 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1719 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1720 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1721 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1722 1723 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1724 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1725 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1726 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1727 1728 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1729 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1730 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1731 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1732 1733 pv += 16; 1734 } 1735 PLogFlops(128*nz+112); 1736 } 1737 row = *ajtmp++; 1738 } 1739 /* finished row so stick it into b->a */ 1740 pv = ba + 16*bi[i]; 1741 pj = bj + bi[i]; 1742 nz = bi[i+1] - bi[i]; 1743 for (j=0; j<nz; j++) { 1744 x = rtmp+16*pj[j]; 1745 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1746 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1747 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1748 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1749 pv += 16; 1750 } 1751 /* invert diagonal block */ 1752 w = ba + 16*diag_offset[i]; 1753 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1754 } 1755 1756 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1757 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1758 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1759 C->factor = FACTOR_LU; 1760 C->assembled = PETSC_TRUE; 1761 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1762 PetscFunctionReturn(0); 1763 } 1764 /* 1765 Version for when blocks are 4 by 4 Using natural ordering 1766 */ 1767 #undef __FUNC__ 1768 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering" 1769 int MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering(Mat A,Mat *B) 1770 { 1771 Mat C = *B; 1772 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1773 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1774 int *ajtmpold,*ajtmp,nz,row; 1775 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1776 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1777 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1778 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1779 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1780 MatScalar m13,m14,m15,m16; 1781 MatScalar *ba = b->a,*aa = a->a; 1782 1783 PetscFunctionBegin; 1784 rtmp = (MatScalar*)PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1785 1786 for (i=0; i<n; i++) { 1787 nz = bi[i+1] - bi[i]; 1788 ajtmp = bj + bi[i]; 1789 for (j=0; j<nz; j++) { 1790 x = rtmp+16*ajtmp[j]; 1791 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1792 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1793 } 1794 /* load in initial (unfactored row) */ 1795 nz = ai[i+1] - ai[i]; 1796 ajtmpold = aj + ai[i]; 1797 v = aa + 16*ai[i]; 1798 for (j=0; j<nz; j++) { 1799 x = rtmp+16*ajtmpold[j]; 1800 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1801 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1802 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1803 x[14] = v[14]; x[15] = v[15]; 1804 v += 16; 1805 } 1806 row = *ajtmp++; 1807 while (row < i) { 1808 pc = rtmp + 16*row; 1809 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1810 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1811 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1812 p15 = pc[14]; p16 = pc[15]; 1813 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1814 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1815 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1816 || p16 != 0.0) { 1817 pv = ba + 16*diag_offset[row]; 1818 pj = bj + diag_offset[row] + 1; 1819 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1820 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1821 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1822 x15 = pv[14]; x16 = pv[15]; 1823 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1824 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1825 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1826 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1827 1828 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1829 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1830 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1831 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1832 1833 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1834 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1835 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1836 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1837 1838 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1839 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1840 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1841 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1842 1843 nz = bi[row+1] - diag_offset[row] - 1; 1844 pv += 16; 1845 for (j=0; j<nz; j++) { 1846 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1847 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1848 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1849 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1850 x = rtmp + 16*pj[j]; 1851 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1852 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1853 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1854 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1855 1856 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1857 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1858 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1859 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1860 1861 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1862 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1863 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1864 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1865 1866 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1867 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1868 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1869 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1870 1871 pv += 16; 1872 } 1873 PLogFlops(128*nz+112); 1874 } 1875 row = *ajtmp++; 1876 } 1877 /* finished row so stick it into b->a */ 1878 pv = ba + 16*bi[i]; 1879 pj = bj + bi[i]; 1880 nz = bi[i+1] - bi[i]; 1881 for (j=0; j<nz; j++) { 1882 x = rtmp+16*pj[j]; 1883 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1884 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1885 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1886 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1887 pv += 16; 1888 } 1889 /* invert diagonal block */ 1890 w = ba + 16*diag_offset[i]; 1891 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1892 } 1893 1894 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1895 C->factor = FACTOR_LU; 1896 C->assembled = PETSC_TRUE; 1897 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1898 PetscFunctionReturn(0); 1899 } 1900 1901 /* 1902 Version for when blocks are 3 by 3 1903 */ 1904 #undef __FUNC__ 1905 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_3" 1906 int MatCholeskyFactorNumeric_SeqSBAIJ_3(Mat A,Mat *B) 1907 { 1908 Mat C = *B; 1909 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1910 IS isrow = b->row,isicol = b->icol; 1911 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1912 int *ajtmpold,*ajtmp,nz,row,*ai=a->i,*aj=a->j; 1913 int *diag_offset = b->diag,idx,*pj; 1914 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1915 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1916 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 1917 MatScalar *ba = b->a,*aa = a->a; 1918 1919 PetscFunctionBegin; 1920 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1921 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1922 rtmp = (MatScalar*)PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1923 1924 for (i=0; i<n; i++) { 1925 nz = bi[i+1] - bi[i]; 1926 ajtmp = bj + bi[i]; 1927 for (j=0; j<nz; j++) { 1928 x = rtmp + 9*ajtmp[j]; 1929 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 1930 } 1931 /* load in initial (unfactored row) */ 1932 idx = r[i]; 1933 nz = ai[idx+1] - ai[idx]; 1934 ajtmpold = aj + ai[idx]; 1935 v = aa + 9*ai[idx]; 1936 for (j=0; j<nz; j++) { 1937 x = rtmp + 9*ic[ajtmpold[j]]; 1938 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1939 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1940 v += 9; 1941 } 1942 row = *ajtmp++; 1943 while (row < i) { 1944 pc = rtmp + 9*row; 1945 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1946 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1947 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1948 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 1949 pv = ba + 9*diag_offset[row]; 1950 pj = bj + diag_offset[row] + 1; 1951 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1952 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1953 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 1954 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 1955 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 1956 1957 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 1958 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 1959 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 1960 1961 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 1962 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 1963 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 1964 nz = bi[row+1] - diag_offset[row] - 1; 1965 pv += 9; 1966 for (j=0; j<nz; j++) { 1967 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1968 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1969 x = rtmp + 9*pj[j]; 1970 x[0] -= m1*x1 + m4*x2 + m7*x3; 1971 x[1] -= m2*x1 + m5*x2 + m8*x3; 1972 x[2] -= m3*x1 + m6*x2 + m9*x3; 1973 1974 x[3] -= m1*x4 + m4*x5 + m7*x6; 1975 x[4] -= m2*x4 + m5*x5 + m8*x6; 1976 x[5] -= m3*x4 + m6*x5 + m9*x6; 1977 1978 x[6] -= m1*x7 + m4*x8 + m7*x9; 1979 x[7] -= m2*x7 + m5*x8 + m8*x9; 1980 x[8] -= m3*x7 + m6*x8 + m9*x9; 1981 pv += 9; 1982 } 1983 PLogFlops(54*nz+36); 1984 } 1985 row = *ajtmp++; 1986 } 1987 /* finished row so stick it into b->a */ 1988 pv = ba + 9*bi[i]; 1989 pj = bj + bi[i]; 1990 nz = bi[i+1] - bi[i]; 1991 for (j=0; j<nz; j++) { 1992 x = rtmp + 9*pj[j]; 1993 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1994 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1995 pv += 9; 1996 } 1997 /* invert diagonal block */ 1998 w = ba + 9*diag_offset[i]; 1999 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 2000 } 2001 2002 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2003 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 2004 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 2005 C->factor = FACTOR_LU; 2006 C->assembled = PETSC_TRUE; 2007 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 2008 PetscFunctionReturn(0); 2009 } 2010 /* 2011 Version for when blocks are 3 by 3 Using natural ordering 2012 */ 2013 #undef __FUNC__ 2014 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering" 2015 int MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering(Mat A,Mat *B) 2016 { 2017 Mat C = *B; 2018 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 2019 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 2020 int *ajtmpold,*ajtmp,nz,row; 2021 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 2022 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2023 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 2024 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 2025 MatScalar *ba = b->a,*aa = a->a; 2026 2027 PetscFunctionBegin; 2028 rtmp = (MatScalar*)PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2029 2030 for (i=0; i<n; i++) { 2031 nz = bi[i+1] - bi[i]; 2032 ajtmp = bj + bi[i]; 2033 for (j=0; j<nz; j++) { 2034 x = rtmp+9*ajtmp[j]; 2035 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 2036 } 2037 /* load in initial (unfactored row) */ 2038 nz = ai[i+1] - ai[i]; 2039 ajtmpold = aj + ai[i]; 2040 v = aa + 9*ai[i]; 2041 for (j=0; j<nz; j++) { 2042 x = rtmp+9*ajtmpold[j]; 2043 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2044 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 2045 v += 9; 2046 } 2047 row = *ajtmp++; 2048 while (row < i) { 2049 pc = rtmp + 9*row; 2050 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2051 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 2052 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 2053 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 2054 pv = ba + 9*diag_offset[row]; 2055 pj = bj + diag_offset[row] + 1; 2056 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2057 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2058 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 2059 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 2060 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 2061 2062 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 2063 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 2064 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 2065 2066 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 2067 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 2068 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 2069 2070 nz = bi[row+1] - diag_offset[row] - 1; 2071 pv += 9; 2072 for (j=0; j<nz; j++) { 2073 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2074 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2075 x = rtmp + 9*pj[j]; 2076 x[0] -= m1*x1 + m4*x2 + m7*x3; 2077 x[1] -= m2*x1 + m5*x2 + m8*x3; 2078 x[2] -= m3*x1 + m6*x2 + m9*x3; 2079 2080 x[3] -= m1*x4 + m4*x5 + m7*x6; 2081 x[4] -= m2*x4 + m5*x5 + m8*x6; 2082 x[5] -= m3*x4 + m6*x5 + m9*x6; 2083 2084 x[6] -= m1*x7 + m4*x8 + m7*x9; 2085 x[7] -= m2*x7 + m5*x8 + m8*x9; 2086 x[8] -= m3*x7 + m6*x8 + m9*x9; 2087 pv += 9; 2088 } 2089 PLogFlops(54*nz+36); 2090 } 2091 row = *ajtmp++; 2092 } 2093 /* finished row so stick it into b->a */ 2094 pv = ba + 9*bi[i]; 2095 pj = bj + bi[i]; 2096 nz = bi[i+1] - bi[i]; 2097 for (j=0; j<nz; j++) { 2098 x = rtmp+9*pj[j]; 2099 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2100 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 2101 pv += 9; 2102 } 2103 /* invert diagonal block */ 2104 w = ba + 9*diag_offset[i]; 2105 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 2106 } 2107 2108 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2109 C->factor = FACTOR_LU; 2110 C->assembled = PETSC_TRUE; 2111 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 2112 PetscFunctionReturn(0); 2113 } 2114 2115 /* 2116 Version for when blocks are 2 by 2 2117 */ 2118 #undef __FUNC__ 2119 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_2" 2120 int MatCholeskyFactorNumeric_SeqSBAIJ_2(Mat A,Mat *B) 2121 { 2122 Mat C = *B; 2123 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 2124 IS isrow = b->row,isicol = b->icol; 2125 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 2126 int *ajtmpold,*ajtmp,nz,row; 2127 int *diag_offset=b->diag,idx,*ai=a->i,*aj=a->j,*pj; 2128 MatScalar *pv,*v,*rtmp,m1,m2,m3,m4,*pc,*w,*x,x1,x2,x3,x4; 2129 MatScalar p1,p2,p3,p4; 2130 MatScalar *ba = b->a,*aa = a->a; 2131 2132 PetscFunctionBegin; 2133 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 2134 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 2135 rtmp = (MatScalar*)PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2136 2137 for (i=0; i<n; i++) { 2138 nz = bi[i+1] - bi[i]; 2139 ajtmp = bj + bi[i]; 2140 for (j=0; j<nz; j++) { 2141 x = rtmp+4*ajtmp[j]; x[0] = x[1] = x[2] = x[3] = 0.0; 2142 } 2143 /* load in initial (unfactored row) */ 2144 idx = r[i]; 2145 nz = ai[idx+1] - ai[idx]; 2146 ajtmpold = aj + ai[idx]; 2147 v = aa + 4*ai[idx]; 2148 for (j=0; j<nz; j++) { 2149 x = rtmp+4*ic[ajtmpold[j]]; 2150 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2151 v += 4; 2152 } 2153 row = *ajtmp++; 2154 while (row < i) { 2155 pc = rtmp + 4*row; 2156 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2157 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2158 pv = ba + 4*diag_offset[row]; 2159 pj = bj + diag_offset[row] + 1; 2160 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2161 pc[0] = m1 = p1*x1 + p3*x2; 2162 pc[1] = m2 = p2*x1 + p4*x2; 2163 pc[2] = m3 = p1*x3 + p3*x4; 2164 pc[3] = m4 = p2*x3 + p4*x4; 2165 nz = bi[row+1] - diag_offset[row] - 1; 2166 pv += 4; 2167 for (j=0; j<nz; j++) { 2168 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2169 x = rtmp + 4*pj[j]; 2170 x[0] -= m1*x1 + m3*x2; 2171 x[1] -= m2*x1 + m4*x2; 2172 x[2] -= m1*x3 + m3*x4; 2173 x[3] -= m2*x3 + m4*x4; 2174 pv += 4; 2175 } 2176 PLogFlops(16*nz+12); 2177 } 2178 row = *ajtmp++; 2179 } 2180 /* finished row so stick it into b->a */ 2181 pv = ba + 4*bi[i]; 2182 pj = bj + bi[i]; 2183 nz = bi[i+1] - bi[i]; 2184 for (j=0; j<nz; j++) { 2185 x = rtmp+4*pj[j]; 2186 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2187 pv += 4; 2188 } 2189 /* invert diagonal block */ 2190 w = ba + 4*diag_offset[i]; 2191 ierr = Kernel_A_gets_inverse_A_2(w);CHKERRQ(ierr); 2192 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2193 } 2194 2195 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2196 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 2197 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 2198 C->factor = FACTOR_LU; 2199 C->assembled = PETSC_TRUE; 2200 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2201 PetscFunctionReturn(0); 2202 } 2203 /* 2204 Version for when blocks are 2 by 2 Using natural ordering 2205 */ 2206 #undef __FUNC__ 2207 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering" 2208 int MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering(Mat A,Mat *B) 2209 { 2210 Mat C = *B; 2211 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 2212 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 2213 int *ajtmpold,*ajtmp,nz,row; 2214 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 2215 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2216 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,x1,x2,x3,x4; 2217 MatScalar *ba = b->a,*aa = a->a; 2218 2219 PetscFunctionBegin; 2220 rtmp = (MatScalar*)PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2221 2222 for (i=0; i<n; i++) { 2223 nz = bi[i+1] - bi[i]; 2224 ajtmp = bj + bi[i]; 2225 for (j=0; j<nz; j++) { 2226 x = rtmp+4*ajtmp[j]; 2227 x[0] = x[1] = x[2] = x[3] = 0.0; 2228 } 2229 /* load in initial (unfactored row) */ 2230 nz = ai[i+1] - ai[i]; 2231 ajtmpold = aj + ai[i]; 2232 v = aa + 4*ai[i]; 2233 for (j=0; j<nz; j++) { 2234 x = rtmp+4*ajtmpold[j]; 2235 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2236 v += 4; 2237 } 2238 row = *ajtmp++; 2239 while (row < i) { 2240 pc = rtmp + 4*row; 2241 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2242 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2243 pv = ba + 4*diag_offset[row]; 2244 pj = bj + diag_offset[row] + 1; 2245 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2246 pc[0] = m1 = p1*x1 + p3*x2; 2247 pc[1] = m2 = p2*x1 + p4*x2; 2248 pc[2] = m3 = p1*x3 + p3*x4; 2249 pc[3] = m4 = p2*x3 + p4*x4; 2250 nz = bi[row+1] - diag_offset[row] - 1; 2251 pv += 4; 2252 for (j=0; j<nz; j++) { 2253 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2254 x = rtmp + 4*pj[j]; 2255 x[0] -= m1*x1 + m3*x2; 2256 x[1] -= m2*x1 + m4*x2; 2257 x[2] -= m1*x3 + m3*x4; 2258 x[3] -= m2*x3 + m4*x4; 2259 pv += 4; 2260 } 2261 PLogFlops(16*nz+12); 2262 } 2263 row = *ajtmp++; 2264 } 2265 /* finished row so stick it into b->a */ 2266 pv = ba + 4*bi[i]; 2267 pj = bj + bi[i]; 2268 nz = bi[i+1] - bi[i]; 2269 for (j=0; j<nz; j++) { 2270 x = rtmp+4*pj[j]; 2271 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2272 pv += 4; 2273 } 2274 /* invert diagonal block */ 2275 w = ba + 4*diag_offset[i]; 2276 ierr = Kernel_A_gets_inverse_A_2(w);CHKERRQ(ierr); 2277 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2278 } 2279 2280 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2281 C->factor = FACTOR_LU; 2282 C->assembled = PETSC_TRUE; 2283 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2284 PetscFunctionReturn(0); 2285 } 2286 2287 /* 2288 Version for when blocks are 1 by 1. 2289 */ 2290 #undef __FUNC__ 2291 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_1" 2292 int MatCholeskyFactorNumeric_SeqSBAIJ_1(Mat A,Mat *B) 2293 { 2294 Mat C = *B; 2295 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 2296 IS ip = b->row; 2297 int *rip,*riip,ierr,i,j,mbs = a->mbs,*bi = b->i,*bj = b->j; 2298 int *ai,*aj,*a2anew,*r; 2299 MatScalar *rtmp; 2300 MatScalar *ba = b->a,*aa,ak; 2301 MatScalar dk,uikdi; 2302 int k,jmin,jmax,*jl,*il,vj,nexti,juj,ili; 2303 PetscTruth flg_perm = PETSC_FALSE; /* non-trivial permutation */ 2304 2305 PetscFunctionBegin; 2306 ierr = ISGetIndices(ip,&rip);CHKERRQ(ierr); 2307 riip = rip; 2308 2309 if (!flg_perm){ 2310 ai = a->i; aj = a->j; aa = a->a; 2311 } else { 2312 2313 ai = a->inew; aj = a->jnew; 2314 2315 aa = (MatScalar*)PetscMalloc(ai[mbs]*sizeof(MatScalar));CHKPTRQ(aa); 2316 ierr = PetscMemcpy(aa,a->a,(ai[mbs])*sizeof(MatScalar));CHKERRQ(ierr); 2317 2318 r = (int*)PetscMalloc(ai[mbs]*sizeof(int));CHKPTRQ(r); 2319 ierr= PetscMemcpy(r,a->a2anew,(ai[mbs])*sizeof(int));CHKERRQ(ierr); 2320 2321 jmin = ai[0]; jmax = ai[mbs]; 2322 for (j=jmin; j<jmax; j++){ 2323 while (a2anew[j] != j){ 2324 k = r[j]; r[j] = r[k]; r[k] = k; 2325 ak = aa[k]; aa[k] = aa[j]; aa[j] = ak; 2326 } 2327 } 2328 a->anew = aa; 2329 ierr = PetscFree(r);CHKERRA(ierr); 2330 } 2331 2332 /* INITIALIZATION */ 2333 /* il and jl record the first nonzero element in each row of the accessing 2334 window U(0:k, k:mbs-1). 2335 jl: list of rows to be added to uneliminated rows 2336 i>= k: jl(i) is the first row to be added to row i 2337 i< k: jl(i) is the row following row i in some list of rows 2338 jl(i) = mbs indicates the end of a list 2339 il(i): points to the first nonzero element in columns k,...,mbs-1 of 2340 row i of U */ 2341 rtmp = (MatScalar*)PetscMalloc(mbs*sizeof(MatScalar));CHKPTRQ(rtmp); 2342 il = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(il); 2343 jl = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(jl); 2344 for (i=0; i<mbs; i++) { 2345 rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0; 2346 } 2347 2348 /* FOR EACH ROW K */ 2349 for (k = 0; k<mbs; k++){ 2350 2351 /* INITIALIZE K-TH ROW WITH ELEMENTS NONZERO IN ROW P(K) OF A */ 2352 jmin = ai[rip[k]]; jmax = ai[rip[k]+1]; 2353 if (jmin < jmax) { 2354 for (j = jmin; j < jmax; j++){ 2355 vj = riip[aj[j]]; 2356 if (k <= vj) rtmp[vj] = aa[j]; 2357 } 2358 } 2359 2360 /* MODIFY K-TH ROW BY ADDING IN THOSE ROWS I WITH U(I,K) NE 0 2361 FOR EACH ROW I TO BE ADDED IN */ 2362 dk = rtmp[k]; 2363 i = jl[k]; /* first row to be added to k_th row */ 2364 /* printf(" k=%d, pivot row = %d\n",k,i); */ 2365 2366 while (i < mbs){ 2367 nexti = jl[i]; /* next row to be added to k_th row */ 2368 /* printf(" pivot row = %d\n", nexti); */ 2369 2370 /* COMPUTE MULTIPLIER AND UPDATE DIAGONAL ELEMENT */ 2371 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 2372 uikdi = - ba[ili]*ba[i]; 2373 dk += uikdi*ba[ili]; 2374 ba[ili] = uikdi; /* update U(i,k) */ 2375 2376 /* ADD MULTIPLE OF ROW I TO K-TH ROW ... */ 2377 jmin = ili + 1; jmax = bi[i+1]; 2378 if (jmin < jmax){ 2379 for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j]; 2380 /* ... AND ADD I TO ROW LIST FOR NEXT NONZERO ENTRY */ 2381 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 2382 j = bj[jmin]; 2383 jl[i] = jl[j]; jl[j] = i; /* update jl */ 2384 } 2385 i = nexti; 2386 /* printf(" pivot row i=%d\n",i); */ 2387 } 2388 2389 /* CHECK FOR ZERO PIVOT AND SAVE DIAGONAL ELEMENT */ 2390 if (dk == 0.0){ 2391 SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,0,"Zero pivot"); 2392 } 2393 2394 /* SAVE NONZERO ENTRIES IN K-TH ROW OF U ... */ 2395 ba[k] = 1.0/dk; 2396 jmin = bi[k]; jmax = bi[k+1]; 2397 if (jmin < jmax) { 2398 for (j=jmin; j<jmax; j++){ 2399 juj = bj[j]; ba[j] = rtmp[juj]; rtmp[juj] = 0.0; 2400 } 2401 2402 /* ... AND ADD K TO ROW LIST FOR FIRST NONZERO ENTRY IN K-TH ROW */ 2403 il[k] = jmin; 2404 i = bj[jmin]; 2405 jl[k] = jl[i]; jl[i] = k; 2406 } 2407 } 2408 2409 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2410 ierr = PetscFree(il);CHKERRQ(ierr); 2411 ierr = PetscFree(jl);CHKERRQ(ierr); 2412 2413 ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr); 2414 C->factor = FACTOR_LU; 2415 C->assembled = PETSC_TRUE; 2416 PLogFlops(b->mbs); 2417 PetscFunctionReturn(0); 2418 } 2419 2420 #undef __FUNC__ 2421 #define __FUNC__ "MatCholeskyFactor_SeqSBAIJ" 2422 int MatCholeskyFactor_SeqSBAIJ(Mat A,IS perm,PetscReal f) 2423 { 2424 Mat_SeqSBAIJ *mat = (Mat_SeqSBAIJ*)A->data; 2425 int ierr,refct; 2426 Mat C; 2427 PetscOps *Abops; 2428 MatOps Aops; 2429 2430 PetscFunctionBegin; 2431 ierr = MatCholeskyFactorSymbolic(A,perm,f,&C);CHKERRQ(ierr); 2432 ierr = MatCholeskyFactorNumeric(A,&C);CHKERRQ(ierr); 2433 2434 /* free all the data structures from mat */ 2435 ierr = PetscFree(mat->a);CHKERRQ(ierr); 2436 if (!mat->singlemalloc) { 2437 ierr = PetscFree(mat->i);CHKERRQ(ierr); 2438 ierr = PetscFree(mat->j);CHKERRQ(ierr); 2439 } 2440 if (mat->diag) {ierr = PetscFree(mat->diag);CHKERRQ(ierr);} 2441 if (mat->ilen) {ierr = PetscFree(mat->ilen);CHKERRQ(ierr);} 2442 if (mat->imax) {ierr = PetscFree(mat->imax);CHKERRQ(ierr);} 2443 if (mat->solve_work) {ierr = PetscFree(mat->solve_work);CHKERRQ(ierr);} 2444 if (mat->mult_work) {ierr = PetscFree(mat->mult_work);CHKERRQ(ierr);} 2445 if (mat->icol) {ierr = ISDestroy(mat->icol);CHKERRQ(ierr);} 2446 ierr = PetscFree(mat);CHKERRQ(ierr); 2447 2448 ierr = MapDestroy(A->rmap);CHKERRQ(ierr); 2449 ierr = MapDestroy(A->cmap);CHKERRQ(ierr); 2450 2451 /* 2452 This is horrible,horrible code. We need to keep the 2453 A pointers for the bops and ops but copy everything 2454 else from C. 2455 */ 2456 Abops = A->bops; 2457 Aops = A->ops; 2458 refct = A->refct; 2459 ierr = PetscMemcpy(A,C,sizeof(struct _p_Mat));CHKERRQ(ierr); 2460 mat = (Mat_SeqSBAIJ*)A->data; 2461 PLogObjectParent(A,mat->icol); 2462 2463 A->bops = Abops; 2464 A->ops = Aops; 2465 A->qlist = 0; 2466 A->refct = refct; 2467 /* copy over the type_name and name */ 2468 ierr = PetscStrallocpy(C->type_name,&A->type_name);CHKERRQ(ierr); 2469 ierr = PetscStrallocpy(C->name,&A->name);CHKERRQ(ierr); 2470 2471 PetscHeaderDestroy(C); 2472 PetscFunctionReturn(0); 2473 } 2474 2475 2476