1 /* Using Modified Sparse Row (MSR) storage. 2 See page 85, "Iterative Methods ..." by Saad. */ 3 4 /*$Id: sbaijfact.c,v 1.25 2000/10/20 19:21:01 hzhang Exp bsmith $*/ 5 /* 6 Symbolic (-UT)*D*(-U) factorization for SBAIJ format. Modified from SSF of YSMP. 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 int *rip,ierr,i,mbs = a->mbs,*ai,*aj; 20 int *jutmp,bs = a->bs,bs2=a->bs2; 21 int m,nzi,realloc = 0; 22 int *jl,*q,jumin,jmin,jmax,juptr,nzk,qm,*iu,*ju,k,j,vj,umax,maxadd; 23 /* PetscTruth ident; */ 24 25 PetscFunctionBegin; 26 PetscValidHeaderSpecific(perm,IS_COOKIE); 27 if (A->M != A->N) SETERRQ(PETSC_ERR_ARG_WRONG,"matrix must be square"); 28 29 /* check whether perm is the identity mapping */ 30 /* 31 ierr = ISView(perm, VIEWER_STDOUT_SELF);CHKERRA(ierr); 32 ierr = ISIdentity(perm,&ident);CHKERRQ(ierr); 33 printf("ident = %d\n", ident); 34 */ 35 ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr); 36 for (i=0; i<mbs; i++){ 37 if (rip[i] != i){ 38 a->permute = PETSC_TRUE; 39 /* printf("non-trivial perm\n"); */ 40 break; 41 } 42 } 43 44 if (!a->permute){ /* without permutation */ 45 ai = a->i; aj = a->j; 46 } else { /* non-trivial permutation */ 47 ierr = MatReorderingSeqSBAIJ(A, perm);CHKERRA(ierr); 48 ai = a->inew; aj = a->jnew; 49 } 50 51 /* initialization */ 52 /* Don't know how many column pointers are needed so estimate. 53 Use Modified Sparse Row storage for u and ju, see Sasd pp.85 */ 54 iu = (int*)PetscMalloc((mbs+1)*sizeof(int));CHKPTRQ(iu); 55 umax = (int)(f*ai[mbs] + 1); umax += mbs + 1; 56 ju = (int*)PetscMalloc(umax*sizeof(int));CHKPTRQ(ju); 57 iu[0] = mbs+1; 58 juptr = mbs; 59 jl = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(jl); 60 q = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(q); 61 for (i=0; i<mbs; i++){ 62 jl[i] = mbs; q[i] = 0; 63 } 64 65 /* for each row k */ 66 for (k=0; k<mbs; k++){ 67 nzk = 0; /* num. of nz blocks in k-th block row with diagonal block excluded */ 68 q[k] = mbs; 69 /* initialize nonzero structure of k-th row to row rip[k] of A */ 70 jmin = ai[rip[k]]; 71 jmax = ai[rip[k]+1]; 72 for (j=jmin; j<jmax; j++){ 73 vj = rip[aj[j]]; /* col. value */ 74 if(vj > k){ 75 qm = k; 76 do { 77 m = qm; qm = q[m]; 78 } while(qm < vj); 79 if (qm == vj) { 80 printf(" error: duplicate entry in A\n"); break; 81 } 82 nzk++; 83 q[m] = vj; 84 q[vj] = qm; 85 } /* if(vj > k) */ 86 } /* for (j=jmin; j<jmax; j++) */ 87 88 /* modify nonzero structure of k-th row by computing fill-in 89 for each row i to be merged in */ 90 i = k; 91 i = jl[i]; /* next pivot row (== mbs for symbolic factorization) */ 92 /* printf(" next pivot row i=%d\n",i); */ 93 while (i < mbs){ 94 /* merge row i into k-th row */ 95 nzi = iu[i+1] - (iu[i]+1); 96 jmin = iu[i] + 1; jmax = iu[i] + nzi; 97 qm = k; 98 for (j=jmin; j<jmax+1; j++){ 99 vj = ju[j]; 100 do { 101 m = qm; qm = q[m]; 102 } while (qm < vj); 103 if (qm != vj){ 104 nzk++; q[m] = vj; q[vj] = qm; qm = vj; 105 } 106 } 107 i = jl[i]; /* next pivot row */ 108 } 109 110 /* add k to row list for first nonzero element in k-th row */ 111 if (nzk > 0){ 112 i = q[k]; /* col value of first nonzero element in U(k, k+1:mbs-1) */ 113 jl[k] = jl[i]; jl[i] = k; 114 } 115 iu[k+1] = iu[k] + nzk; /* printf(" iu[%d]=%d, umax=%d\n", k+1, iu[k+1],umax);*/ 116 117 /* allocate more space to ju if needed */ 118 if (iu[k+1] > umax) { printf("allocate more space, iu[%d]=%d > umax=%d\n",k+1, iu[k+1],umax); 119 /* estimate how much additional space we will need */ 120 /* use the strategy suggested by David Hysom <hysom@perch-t.icase.edu> */ 121 /* just double the memory each time */ 122 maxadd = umax; 123 if (maxadd < nzk) maxadd = (mbs-k)*(nzk+1)/2; 124 umax += maxadd; 125 126 /* allocate a longer ju */ 127 jutmp = (int*)PetscMalloc(umax*sizeof(int));CHKPTRQ(jutmp); 128 ierr = PetscMemcpy(jutmp,ju,iu[k]*sizeof(int));CHKERRQ(ierr); 129 ierr = PetscFree(ju);CHKERRQ(ierr); 130 ju = jutmp; 131 realloc++; /* count how many times we realloc */ 132 } 133 134 /* save nonzero structure of k-th row in ju */ 135 i=k; 136 jumin = juptr + 1; juptr += nzk; 137 for (j=jumin; j<juptr+1; j++){ 138 i=q[i]; 139 ju[j]=i; 140 } 141 } 142 143 if (ai[mbs] != 0) { 144 PetscReal af = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 145 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Reallocs %d Fill ratio:given %g needed %g\n",realloc,f,af); 146 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Run with -pc_lu_fill %g or use \n",af); 147 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:PCLUSetFill(pc,%g);\n",af); 148 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:for best performance.\n"); 149 } else { 150 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Empty matrix.\n"); 151 } 152 153 ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr); 154 ierr = PetscFree(q);CHKERRQ(ierr); 155 ierr = PetscFree(jl);CHKERRQ(ierr); 156 157 /* put together the new matrix */ 158 ierr = MatCreateSeqSBAIJ(A->comm,bs,bs*mbs,bs*mbs,0,PETSC_NULL,B);CHKERRQ(ierr); 159 /* PLogObjectParent(*B,iperm); */ 160 b = (Mat_SeqSBAIJ*)(*B)->data; 161 ierr = PetscFree(b->imax);CHKERRQ(ierr); 162 b->singlemalloc = PETSC_FALSE; 163 /* the next line frees the default space generated by the Create() */ 164 ierr = PetscFree(b->a);CHKERRQ(ierr); 165 ierr = PetscFree(b->ilen);CHKERRQ(ierr); 166 b->a = (MatScalar*)PetscMalloc((iu[mbs]+1)*sizeof(MatScalar)*bs2);CHKPTRQ(b->a); 167 b->j = ju; 168 b->i = iu; 169 b->diag = 0; 170 b->ilen = 0; 171 b->imax = 0; 172 b->row = perm; 173 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 174 b->icol = perm; 175 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 176 b->solve_work = (Scalar*)PetscMalloc((bs*mbs+bs)*sizeof(Scalar));CHKPTRQ(b->solve_work); 177 /* In b structure: Free imax, ilen, old a, old j. 178 Allocate idnew, solve_work, new a, new j */ 179 PLogObjectMemory(*B,(iu[mbs]-mbs)*(sizeof(int)+sizeof(MatScalar))); 180 b->s_maxnz = b->s_nz = iu[mbs]; 181 182 (*B)->factor = FACTOR_LU; 183 (*B)->info.factor_mallocs = realloc; 184 (*B)->info.fill_ratio_given = f; 185 if (ai[mbs] != 0) { 186 (*B)->info.fill_ratio_needed = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 187 } else { 188 (*B)->info.fill_ratio_needed = 0.0; 189 } 190 PetscFunctionReturn(0); 191 } 192 193 #undef __FUNC__ 194 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_N" 195 int MatCholeskyFactorNumeric_SeqSBAIJ_N(Mat A,Mat *B) 196 { 197 Mat C = *B; 198 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 199 IS isrow = b->row,isicol = b->icol; 200 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 201 int *ajtmpold,*ajtmp,nz,row,bslog,*ai=a->i,*aj=a->j,k,flg; 202 int *diag_offset=b->diag,diag,bs=a->bs,bs2 = a->bs2,*v_pivots,*pj; 203 MatScalar *ba = b->a,*aa = a->a,*pv,*v,*rtmp,*multiplier,*v_work,*pc,*w; 204 205 PetscFunctionBegin; 206 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 207 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 208 rtmp = (MatScalar*)PetscMalloc(bs2*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 209 ierr = PetscMemzero(rtmp,bs2*(n+1)*sizeof(MatScalar));CHKERRQ(ierr); 210 /* generate work space needed by dense LU factorization */ 211 v_work = (MatScalar*)PetscMalloc(bs*sizeof(int) + (bs+bs2)*sizeof(MatScalar));CHKPTRQ(v_work); 212 multiplier = v_work + bs; 213 v_pivots = (int*)(multiplier + bs2); 214 215 /* flops in while loop */ 216 bslog = 2*bs*bs2; 217 218 for (i=0; i<n; i++) { 219 nz = bi[i+1] - bi[i]; 220 ajtmp = bj + bi[i]; 221 for (j=0; j<nz; j++) { 222 ierr = PetscMemzero(rtmp+bs2*ajtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 223 } 224 /* load in initial (unfactored row) */ 225 nz = ai[r[i]+1] - ai[r[i]]; 226 ajtmpold = aj + ai[r[i]]; 227 v = aa + bs2*ai[r[i]]; 228 for (j=0; j<nz; j++) { 229 ierr = PetscMemcpy(rtmp+bs2*ic[ajtmpold[j]],v+bs2*j,bs2*sizeof(MatScalar));CHKERRQ(ierr); 230 } 231 row = *ajtmp++; 232 while (row < i) { 233 pc = rtmp + bs2*row; 234 /* if (*pc) { */ 235 for (flg=0,k=0; k<bs2; k++) { if (pc[k]!=0.0) { flg =1; break; }} 236 if (flg) { 237 pv = ba + bs2*diag_offset[row]; 238 pj = bj + diag_offset[row] + 1; 239 Kernel_A_gets_A_times_B(bs,pc,pv,multiplier); 240 nz = bi[row+1] - diag_offset[row] - 1; 241 pv += bs2; 242 for (j=0; j<nz; j++) { 243 Kernel_A_gets_A_minus_B_times_C(bs,rtmp+bs2*pj[j],pc,pv+bs2*j); 244 } 245 PLogFlops(bslog*(nz+1)-bs); 246 } 247 row = *ajtmp++; 248 } 249 /* finished row so stick it into b->a */ 250 pv = ba + bs2*bi[i]; 251 pj = bj + bi[i]; 252 nz = bi[i+1] - bi[i]; 253 for (j=0; j<nz; j++) { 254 ierr = PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 255 } 256 diag = diag_offset[i] - bi[i]; 257 /* invert diagonal block */ 258 w = pv + bs2*diag; 259 Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work); 260 } 261 262 ierr = PetscFree(rtmp);CHKERRQ(ierr); 263 ierr = PetscFree(v_work);CHKERRQ(ierr); 264 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 265 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 266 C->factor = FACTOR_LU; 267 C->assembled = PETSC_TRUE; 268 PLogFlops(1.3333*bs*bs2*b->mbs); /* from inverting diagonal blocks */ 269 PetscFunctionReturn(0); 270 } 271 272 /* 273 Version for when blocks are 7 by 7 274 */ 275 #undef __FUNC__ 276 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_7" 277 int MatCholeskyFactorNumeric_SeqSBAIJ_7(Mat A,Mat *B) 278 { 279 Mat C = *B; 280 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 281 IS isrow = b->row,isicol = b->icol; 282 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 283 int *ajtmpold,*ajtmp,nz,row; 284 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 285 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 286 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 287 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 288 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 289 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 290 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 291 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 292 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 293 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 294 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 295 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 296 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 297 MatScalar *ba = b->a,*aa = a->a; 298 299 PetscFunctionBegin; 300 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 301 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 302 rtmp = (MatScalar*)PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 303 304 for (i=0; i<n; i++) { 305 nz = bi[i+1] - bi[i]; 306 ajtmp = bj + bi[i]; 307 for (j=0; j<nz; j++) { 308 x = rtmp+49*ajtmp[j]; 309 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 310 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 311 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 312 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 313 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 314 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 315 } 316 /* load in initial (unfactored row) */ 317 idx = r[i]; 318 nz = ai[idx+1] - ai[idx]; 319 ajtmpold = aj + ai[idx]; 320 v = aa + 49*ai[idx]; 321 for (j=0; j<nz; j++) { 322 x = rtmp+49*ic[ajtmpold[j]]; 323 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 324 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 325 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 326 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 327 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 328 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 329 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 330 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 331 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 332 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 333 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 334 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 335 x[48] = v[48]; 336 v += 49; 337 } 338 row = *ajtmp++; 339 while (row < i) { 340 pc = rtmp + 49*row; 341 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 342 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 343 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 344 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 345 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 346 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 347 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 348 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 349 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 350 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 351 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 352 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 353 p49 = pc[48]; 354 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 355 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 356 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 357 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 358 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 359 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 360 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 361 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 362 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 363 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 364 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 365 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 366 p49 != 0.0) { 367 pv = ba + 49*diag_offset[row]; 368 pj = bj + diag_offset[row] + 1; 369 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 370 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 371 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 372 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 373 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 374 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 375 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 376 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 377 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 378 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 379 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 380 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 381 x49 = pv[48]; 382 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 383 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 384 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 385 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 386 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 387 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 388 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 389 390 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 391 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 392 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 393 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 394 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 395 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 396 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 397 398 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 399 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 400 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 401 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 402 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 403 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 404 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 405 406 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 407 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 408 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 409 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 410 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 411 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 412 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 413 414 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 415 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 416 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 417 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 418 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 419 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 420 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 421 422 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 423 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 424 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 425 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 426 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 427 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 428 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 429 430 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 431 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 432 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 433 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 434 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 435 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 436 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 437 438 nz = bi[row+1] - diag_offset[row] - 1; 439 pv += 49; 440 for (j=0; j<nz; j++) { 441 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 442 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 443 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 444 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 445 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 446 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 447 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 448 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 449 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 450 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 451 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 452 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 453 x49 = pv[48]; 454 x = rtmp + 49*pj[j]; 455 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 456 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 457 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 458 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 459 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 460 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 461 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 462 463 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 464 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 465 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 466 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 467 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 468 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 469 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 470 471 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 472 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 473 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 474 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 475 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 476 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 477 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 478 479 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 480 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 481 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 482 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 483 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 484 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 485 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 486 487 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 488 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 489 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 490 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 491 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 492 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 493 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 494 495 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 496 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 497 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 498 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 499 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 500 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 501 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 502 503 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 504 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 505 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 506 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 507 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 508 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 509 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 510 pv += 49; 511 } 512 PLogFlops(686*nz+637); 513 } 514 row = *ajtmp++; 515 } 516 /* finished row so stick it into b->a */ 517 pv = ba + 49*bi[i]; 518 pj = bj + bi[i]; 519 nz = bi[i+1] - bi[i]; 520 for (j=0; j<nz; j++) { 521 x = rtmp+49*pj[j]; 522 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 523 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 524 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 525 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 526 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 527 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 528 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 529 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 530 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 531 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 532 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 533 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 534 pv[48] = x[48]; 535 pv += 49; 536 } 537 /* invert diagonal block */ 538 w = ba + 49*diag_offset[i]; 539 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 540 } 541 542 ierr = PetscFree(rtmp);CHKERRQ(ierr); 543 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 544 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 545 C->factor = FACTOR_LU; 546 C->assembled = PETSC_TRUE; 547 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 548 PetscFunctionReturn(0); 549 } 550 551 /* 552 Version for when blocks are 7 by 7 Using natural ordering 553 */ 554 #undef __FUNC__ 555 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering" 556 int MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering(Mat A,Mat *B) 557 { 558 Mat C = *B; 559 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 560 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 561 int *ajtmpold,*ajtmp,nz,row; 562 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 563 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 564 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 565 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 566 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 567 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 568 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 569 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 570 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 571 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 572 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 573 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 574 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 575 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 576 MatScalar *ba = b->a,*aa = a->a; 577 578 PetscFunctionBegin; 579 rtmp = (MatScalar*)PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 580 for (i=0; i<n; i++) { 581 nz = bi[i+1] - bi[i]; 582 ajtmp = bj + bi[i]; 583 for (j=0; j<nz; j++) { 584 x = rtmp+49*ajtmp[j]; 585 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 586 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 587 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 588 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 589 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 590 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 591 } 592 /* load in initial (unfactored row) */ 593 nz = ai[i+1] - ai[i]; 594 ajtmpold = aj + ai[i]; 595 v = aa + 49*ai[i]; 596 for (j=0; j<nz; j++) { 597 x = rtmp+49*ajtmpold[j]; 598 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 599 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 600 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 601 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 602 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 603 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 604 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 605 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 606 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 607 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 608 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 609 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 610 x[48] = v[48]; 611 v += 49; 612 } 613 row = *ajtmp++; 614 while (row < i) { 615 pc = rtmp + 49*row; 616 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 617 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 618 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 619 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 620 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 621 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 622 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 623 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 624 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 625 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 626 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 627 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 628 p49 = pc[48]; 629 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 630 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 631 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 632 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 633 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 634 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 635 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 636 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 637 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 638 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 639 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 640 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 641 p49 != 0.0) { 642 pv = ba + 49*diag_offset[row]; 643 pj = bj + diag_offset[row] + 1; 644 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 645 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 646 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 647 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 648 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 649 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 650 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 651 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 652 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 653 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 654 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 655 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 656 x49 = pv[48]; 657 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 658 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 659 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 660 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 661 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 662 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 663 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 664 665 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 666 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 667 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 668 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 669 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 670 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 671 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 672 673 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 674 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 675 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 676 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 677 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 678 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 679 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 680 681 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 682 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 683 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 684 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 685 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 686 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 687 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 688 689 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 690 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 691 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 692 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 693 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 694 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 695 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 696 697 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 698 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 699 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 700 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 701 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 702 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 703 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 704 705 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 706 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 707 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 708 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 709 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 710 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 711 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 712 713 nz = bi[row+1] - diag_offset[row] - 1; 714 pv += 49; 715 for (j=0; j<nz; j++) { 716 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 717 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 718 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 719 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 720 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 721 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 722 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 723 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 724 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 725 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 726 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 727 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 728 x49 = pv[48]; 729 x = rtmp + 49*pj[j]; 730 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 731 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 732 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 733 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 734 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 735 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 736 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 737 738 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 739 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 740 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 741 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 742 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 743 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 744 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 745 746 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 747 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 748 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 749 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 750 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 751 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 752 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 753 754 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 755 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 756 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 757 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 758 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 759 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 760 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 761 762 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 763 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 764 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 765 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 766 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 767 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 768 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 769 770 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 771 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 772 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 773 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 774 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 775 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 776 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 777 778 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 779 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 780 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 781 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 782 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 783 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 784 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 785 pv += 49; 786 } 787 PLogFlops(686*nz+637); 788 } 789 row = *ajtmp++; 790 } 791 /* finished row so stick it into b->a */ 792 pv = ba + 49*bi[i]; 793 pj = bj + bi[i]; 794 nz = bi[i+1] - bi[i]; 795 for (j=0; j<nz; j++) { 796 x = rtmp+49*pj[j]; 797 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 798 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 799 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 800 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 801 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 802 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 803 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 804 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 805 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 806 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 807 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 808 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 809 pv[48] = x[48]; 810 pv += 49; 811 } 812 /* invert diagonal block */ 813 w = ba + 49*diag_offset[i]; 814 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 815 } 816 817 ierr = PetscFree(rtmp);CHKERRQ(ierr); 818 C->factor = FACTOR_LU; 819 C->assembled = PETSC_TRUE; 820 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 821 PetscFunctionReturn(0); 822 } 823 824 /* ------------------------------------------------------------*/ 825 /* 826 Version for when blocks are 6 by 6 827 */ 828 #undef __FUNC__ 829 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_6" 830 int MatCholeskyFactorNumeric_SeqSBAIJ_6(Mat A,Mat *B) 831 { 832 Mat C = *B; 833 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 834 IS isrow = b->row,isicol = b->icol; 835 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 836 int *ajtmpold,*ajtmp,nz,row; 837 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 838 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 839 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 840 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 841 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 842 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 843 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 844 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 845 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 846 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 847 MatScalar *ba = b->a,*aa = a->a; 848 849 PetscFunctionBegin; 850 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 851 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 852 rtmp = (MatScalar*)PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 853 854 for (i=0; i<n; i++) { 855 nz = bi[i+1] - bi[i]; 856 ajtmp = bj + bi[i]; 857 for (j=0; j<nz; j++) { 858 x = rtmp+36*ajtmp[j]; 859 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 860 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 861 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 862 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 863 x[34] = x[35] = 0.0 ; 864 } 865 /* load in initial (unfactored row) */ 866 idx = r[i]; 867 nz = ai[idx+1] - ai[idx]; 868 ajtmpold = aj + ai[idx]; 869 v = aa + 36*ai[idx]; 870 for (j=0; j<nz; j++) { 871 x = rtmp+36*ic[ajtmpold[j]]; 872 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 873 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 874 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 875 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 876 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 877 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 878 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 879 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 880 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 881 v += 36; 882 } 883 row = *ajtmp++; 884 while (row < i) { 885 pc = rtmp + 36*row; 886 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 887 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 888 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 889 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 890 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 891 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 892 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 893 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 894 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 895 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 896 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 897 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 898 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 899 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 900 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 901 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 902 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 903 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 904 pv = ba + 36*diag_offset[row]; 905 pj = bj + diag_offset[row] + 1; 906 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 907 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 908 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 909 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 910 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 911 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 912 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 913 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 914 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 915 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 916 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 917 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 918 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 919 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 920 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 921 922 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 923 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 924 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 925 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 926 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 927 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 928 929 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 930 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 931 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 932 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 933 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 934 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 935 936 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 937 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 938 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 939 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 940 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 941 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 942 943 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 944 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 945 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 946 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 947 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 948 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 949 950 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 951 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 952 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 953 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 954 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 955 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 956 957 nz = bi[row+1] - diag_offset[row] - 1; 958 pv += 36; 959 for (j=0; j<nz; j++) { 960 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 961 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 962 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 963 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 964 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 965 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 966 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 967 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 968 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 969 x = rtmp + 36*pj[j]; 970 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 971 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 972 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 973 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 974 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 975 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 976 977 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 978 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 979 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 980 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 981 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 982 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 983 984 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 985 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 986 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 987 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 988 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 989 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 990 991 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 992 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 993 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 994 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 995 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 996 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 997 998 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 999 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 1000 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 1001 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 1002 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 1003 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 1004 1005 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 1006 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 1007 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 1008 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 1009 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 1010 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 1011 1012 pv += 36; 1013 } 1014 PLogFlops(432*nz+396); 1015 } 1016 row = *ajtmp++; 1017 } 1018 /* finished row so stick it into b->a */ 1019 pv = ba + 36*bi[i]; 1020 pj = bj + bi[i]; 1021 nz = bi[i+1] - bi[i]; 1022 for (j=0; j<nz; j++) { 1023 x = rtmp+36*pj[j]; 1024 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1025 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1026 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1027 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1028 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1029 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1030 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1031 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1032 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1033 pv += 36; 1034 } 1035 /* invert diagonal block */ 1036 w = ba + 36*diag_offset[i]; 1037 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1038 } 1039 1040 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1041 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1042 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1043 C->factor = FACTOR_LU; 1044 C->assembled = PETSC_TRUE; 1045 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1046 PetscFunctionReturn(0); 1047 } 1048 /* 1049 Version for when blocks are 6 by 6 Using natural ordering 1050 */ 1051 #undef __FUNC__ 1052 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering" 1053 int MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering(Mat A,Mat *B) 1054 { 1055 Mat C = *B; 1056 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1057 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1058 int *ajtmpold,*ajtmp,nz,row; 1059 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1060 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1061 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1062 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1063 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1064 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1065 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1066 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1067 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 1068 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 1069 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 1070 MatScalar *ba = b->a,*aa = a->a; 1071 1072 PetscFunctionBegin; 1073 rtmp = (MatScalar*)PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1074 for (i=0; i<n; i++) { 1075 nz = bi[i+1] - bi[i]; 1076 ajtmp = bj + bi[i]; 1077 for (j=0; j<nz; j++) { 1078 x = rtmp+36*ajtmp[j]; 1079 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1080 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1081 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 1082 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 1083 x[34] = x[35] = 0.0 ; 1084 } 1085 /* load in initial (unfactored row) */ 1086 nz = ai[i+1] - ai[i]; 1087 ajtmpold = aj + ai[i]; 1088 v = aa + 36*ai[i]; 1089 for (j=0; j<nz; j++) { 1090 x = rtmp+36*ajtmpold[j]; 1091 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1092 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 1093 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 1094 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 1095 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 1096 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1097 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 1098 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 1099 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 1100 v += 36; 1101 } 1102 row = *ajtmp++; 1103 while (row < i) { 1104 pc = rtmp + 36*row; 1105 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1106 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 1107 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 1108 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 1109 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 1110 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1111 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 1112 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 1113 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 1114 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 1115 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 1116 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 1117 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 1118 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 1119 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 1120 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 1121 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 1122 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 1123 pv = ba + 36*diag_offset[row]; 1124 pj = bj + diag_offset[row] + 1; 1125 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1126 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1127 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1128 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1129 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1130 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1131 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1132 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1133 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1134 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 1135 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 1136 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 1137 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 1138 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 1139 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 1140 1141 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 1142 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 1143 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 1144 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 1145 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 1146 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 1147 1148 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 1149 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 1150 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 1151 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 1152 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 1153 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 1154 1155 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 1156 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 1157 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 1158 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 1159 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 1160 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 1161 1162 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 1163 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 1164 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 1165 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 1166 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 1167 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 1168 1169 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 1170 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 1171 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 1172 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 1173 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 1174 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 1175 1176 nz = bi[row+1] - diag_offset[row] - 1; 1177 pv += 36; 1178 for (j=0; j<nz; j++) { 1179 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1180 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1181 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1182 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1183 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1184 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1185 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1186 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1187 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1188 x = rtmp + 36*pj[j]; 1189 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 1190 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 1191 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 1192 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 1193 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 1194 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 1195 1196 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 1197 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 1198 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 1199 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 1200 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 1201 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 1202 1203 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 1204 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 1205 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 1206 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 1207 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 1208 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 1209 1210 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 1211 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 1212 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 1213 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 1214 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 1215 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 1216 1217 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 1218 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 1219 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 1220 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 1221 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 1222 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 1223 1224 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 1225 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 1226 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 1227 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 1228 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 1229 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 1230 1231 pv += 36; 1232 } 1233 PLogFlops(432*nz+396); 1234 } 1235 row = *ajtmp++; 1236 } 1237 /* finished row so stick it into b->a */ 1238 pv = ba + 36*bi[i]; 1239 pj = bj + bi[i]; 1240 nz = bi[i+1] - bi[i]; 1241 for (j=0; j<nz; j++) { 1242 x = rtmp+36*pj[j]; 1243 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1244 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1245 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1246 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1247 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1248 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1249 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1250 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1251 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1252 pv += 36; 1253 } 1254 /* invert diagonal block */ 1255 w = ba + 36*diag_offset[i]; 1256 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1257 } 1258 1259 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1260 C->factor = FACTOR_LU; 1261 C->assembled = PETSC_TRUE; 1262 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1263 PetscFunctionReturn(0); 1264 } 1265 1266 /* 1267 Version for when blocks are 5 by 5 1268 */ 1269 #undef __FUNC__ 1270 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_5" 1271 int MatCholeskyFactorNumeric_SeqSBAIJ_5(Mat A,Mat *B) 1272 { 1273 Mat C = *B; 1274 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1275 IS isrow = b->row,isicol = b->icol; 1276 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1277 int *ajtmpold,*ajtmp,nz,row; 1278 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 1279 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1280 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1281 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1282 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 1283 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 1284 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1285 MatScalar *ba = b->a,*aa = a->a; 1286 1287 PetscFunctionBegin; 1288 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1289 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1290 rtmp = (MatScalar*)PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1291 1292 for (i=0; i<n; i++) { 1293 nz = bi[i+1] - bi[i]; 1294 ajtmp = bj + bi[i]; 1295 for (j=0; j<nz; j++) { 1296 x = rtmp+25*ajtmp[j]; 1297 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1298 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1299 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1300 } 1301 /* load in initial (unfactored row) */ 1302 idx = r[i]; 1303 nz = ai[idx+1] - ai[idx]; 1304 ajtmpold = aj + ai[idx]; 1305 v = aa + 25*ai[idx]; 1306 for (j=0; j<nz; j++) { 1307 x = rtmp+25*ic[ajtmpold[j]]; 1308 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1309 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1310 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1311 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; 1312 x[18] = v[18]; x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; 1313 x[22] = v[22]; x[23] = v[23]; x[24] = v[24]; 1314 v += 25; 1315 } 1316 row = *ajtmp++; 1317 while (row < i) { 1318 pc = rtmp + 25*row; 1319 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1320 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1321 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1322 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; 1323 p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1324 p25 = pc[24]; 1325 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1326 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1327 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1328 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || 1329 p20 != 0.0 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || 1330 p24 != 0.0 || p25 != 0.0) { 1331 pv = ba + 25*diag_offset[row]; 1332 pj = bj + diag_offset[row] + 1; 1333 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1334 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1335 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1336 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1337 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; 1338 x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1339 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1340 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1341 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1342 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1343 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1344 1345 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1346 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1347 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1348 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1349 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1350 1351 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1352 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1353 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1354 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1355 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1356 1357 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1358 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1359 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1360 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1361 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1362 1363 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1364 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1365 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1366 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1367 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1368 1369 nz = bi[row+1] - diag_offset[row] - 1; 1370 pv += 25; 1371 for (j=0; j<nz; j++) { 1372 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1373 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1374 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1375 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; 1376 x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; 1377 x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1378 x = rtmp + 25*pj[j]; 1379 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1380 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1381 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1382 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1383 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1384 1385 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1386 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1387 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1388 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1389 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1390 1391 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1392 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1393 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1394 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1395 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1396 1397 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1398 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1399 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1400 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1401 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1402 1403 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1404 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1405 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1406 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1407 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1408 1409 pv += 25; 1410 } 1411 PLogFlops(250*nz+225); 1412 } 1413 row = *ajtmp++; 1414 } 1415 /* finished row so stick it into b->a */ 1416 pv = ba + 25*bi[i]; 1417 pj = bj + bi[i]; 1418 nz = bi[i+1] - bi[i]; 1419 for (j=0; j<nz; j++) { 1420 x = rtmp+25*pj[j]; 1421 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1422 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1423 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1424 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; 1425 pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; 1426 pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; pv[24] = x[24]; 1427 pv += 25; 1428 } 1429 /* invert diagonal block */ 1430 w = ba + 25*diag_offset[i]; 1431 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1432 } 1433 1434 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1435 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1436 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1437 C->factor = FACTOR_LU; 1438 C->assembled = PETSC_TRUE; 1439 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1440 PetscFunctionReturn(0); 1441 } 1442 /* 1443 Version for when blocks are 5 by 5 Using natural ordering 1444 */ 1445 #undef __FUNC__ 1446 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering" 1447 int MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering(Mat A,Mat *B) 1448 { 1449 Mat C = *B; 1450 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1451 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1452 int *ajtmpold,*ajtmp,nz,row; 1453 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1454 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1455 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1456 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1457 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1458 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1459 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1460 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1461 MatScalar *ba = b->a,*aa = a->a; 1462 1463 PetscFunctionBegin; 1464 rtmp = (MatScalar*)PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1465 for (i=0; i<n; i++) { 1466 nz = bi[i+1] - bi[i]; 1467 ajtmp = bj + bi[i]; 1468 for (j=0; j<nz; j++) { 1469 x = rtmp+25*ajtmp[j]; 1470 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1471 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1472 x[16] = x[17] = x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1473 } 1474 /* load in initial (unfactored row) */ 1475 nz = ai[i+1] - ai[i]; 1476 ajtmpold = aj + ai[i]; 1477 v = aa + 25*ai[i]; 1478 for (j=0; j<nz; j++) { 1479 x = rtmp+25*ajtmpold[j]; 1480 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1481 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1482 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1483 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; 1484 x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1485 x[24] = v[24]; 1486 v += 25; 1487 } 1488 row = *ajtmp++; 1489 while (row < i) { 1490 pc = rtmp + 25*row; 1491 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1492 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1493 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1494 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; 1495 p19 = pc[18]; p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; 1496 p24 = pc[23]; p25 = pc[24]; 1497 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1498 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1499 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1500 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 1501 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || p25 != 0.0) { 1502 pv = ba + 25*diag_offset[row]; 1503 pj = bj + diag_offset[row] + 1; 1504 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1505 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1506 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1507 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; 1508 x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1509 x25 = pv[24]; 1510 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1511 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1512 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1513 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1514 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1515 1516 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1517 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1518 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1519 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1520 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1521 1522 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1523 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1524 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1525 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1526 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1527 1528 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1529 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1530 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1531 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1532 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1533 1534 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1535 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1536 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1537 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1538 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1539 1540 nz = bi[row+1] - diag_offset[row] - 1; 1541 pv += 25; 1542 for (j=0; j<nz; j++) { 1543 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1544 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1545 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1546 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1547 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; 1548 x24 = pv[23]; x25 = pv[24]; 1549 x = rtmp + 25*pj[j]; 1550 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1551 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1552 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1553 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1554 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1555 1556 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1557 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1558 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1559 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1560 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1561 1562 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1563 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1564 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1565 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1566 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1567 1568 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1569 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1570 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1571 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1572 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1573 1574 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1575 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1576 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1577 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1578 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1579 pv += 25; 1580 } 1581 PLogFlops(250*nz+225); 1582 } 1583 row = *ajtmp++; 1584 } 1585 /* finished row so stick it into b->a */ 1586 pv = ba + 25*bi[i]; 1587 pj = bj + bi[i]; 1588 nz = bi[i+1] - bi[i]; 1589 for (j=0; j<nz; j++) { 1590 x = rtmp+25*pj[j]; 1591 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1592 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1593 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1594 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; pv[17] = x[17]; 1595 pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; 1596 pv[23] = x[23]; pv[24] = x[24]; 1597 pv += 25; 1598 } 1599 /* invert diagonal block */ 1600 w = ba + 25*diag_offset[i]; 1601 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1602 } 1603 1604 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1605 C->factor = FACTOR_LU; 1606 C->assembled = PETSC_TRUE; 1607 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1608 PetscFunctionReturn(0); 1609 } 1610 1611 /* 1612 Version for when blocks are 4 by 4 1613 */ 1614 #undef __FUNC__ 1615 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_4" 1616 int MatCholeskyFactorNumeric_SeqSBAIJ_4(Mat A,Mat *B) 1617 { 1618 Mat C = *B; 1619 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1620 IS isrow = b->row,isicol = b->icol; 1621 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1622 int *ajtmpold,*ajtmp,nz,row; 1623 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 1624 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1625 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1626 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1627 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1628 MatScalar m13,m14,m15,m16; 1629 MatScalar *ba = b->a,*aa = a->a; 1630 1631 PetscFunctionBegin; 1632 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1633 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1634 rtmp = (MatScalar*)PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1635 1636 for (i=0; i<n; i++) { 1637 nz = bi[i+1] - bi[i]; 1638 ajtmp = bj + bi[i]; 1639 for (j=0; j<nz; j++) { 1640 x = rtmp+16*ajtmp[j]; 1641 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1642 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1643 } 1644 /* load in initial (unfactored row) */ 1645 idx = r[i]; 1646 nz = ai[idx+1] - ai[idx]; 1647 ajtmpold = aj + ai[idx]; 1648 v = aa + 16*ai[idx]; 1649 for (j=0; j<nz; j++) { 1650 x = rtmp+16*ic[ajtmpold[j]]; 1651 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1652 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1653 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1654 x[14] = v[14]; x[15] = v[15]; 1655 v += 16; 1656 } 1657 row = *ajtmp++; 1658 while (row < i) { 1659 pc = rtmp + 16*row; 1660 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1661 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1662 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1663 p15 = pc[14]; p16 = pc[15]; 1664 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1665 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1666 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1667 || p16 != 0.0) { 1668 pv = ba + 16*diag_offset[row]; 1669 pj = bj + diag_offset[row] + 1; 1670 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1671 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1672 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1673 x15 = pv[14]; x16 = pv[15]; 1674 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1675 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1676 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1677 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1678 1679 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1680 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1681 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1682 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1683 1684 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1685 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1686 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1687 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1688 1689 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1690 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1691 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1692 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1693 1694 nz = bi[row+1] - diag_offset[row] - 1; 1695 pv += 16; 1696 for (j=0; j<nz; j++) { 1697 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1698 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1699 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1700 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1701 x = rtmp + 16*pj[j]; 1702 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1703 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1704 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1705 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1706 1707 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1708 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1709 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1710 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1711 1712 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1713 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1714 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1715 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1716 1717 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1718 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1719 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1720 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1721 1722 pv += 16; 1723 } 1724 PLogFlops(128*nz+112); 1725 } 1726 row = *ajtmp++; 1727 } 1728 /* finished row so stick it into b->a */ 1729 pv = ba + 16*bi[i]; 1730 pj = bj + bi[i]; 1731 nz = bi[i+1] - bi[i]; 1732 for (j=0; j<nz; j++) { 1733 x = rtmp+16*pj[j]; 1734 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1735 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1736 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1737 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1738 pv += 16; 1739 } 1740 /* invert diagonal block */ 1741 w = ba + 16*diag_offset[i]; 1742 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1743 } 1744 1745 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1746 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1747 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1748 C->factor = FACTOR_LU; 1749 C->assembled = PETSC_TRUE; 1750 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1751 PetscFunctionReturn(0); 1752 } 1753 /* 1754 Version for when blocks are 4 by 4 Using natural ordering 1755 */ 1756 #undef __FUNC__ 1757 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering" 1758 int MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering(Mat A,Mat *B) 1759 { 1760 Mat C = *B; 1761 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1762 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1763 int *ajtmpold,*ajtmp,nz,row; 1764 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1765 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1766 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1767 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1768 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1769 MatScalar m13,m14,m15,m16; 1770 MatScalar *ba = b->a,*aa = a->a; 1771 1772 PetscFunctionBegin; 1773 rtmp = (MatScalar*)PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1774 1775 for (i=0; i<n; i++) { 1776 nz = bi[i+1] - bi[i]; 1777 ajtmp = bj + bi[i]; 1778 for (j=0; j<nz; j++) { 1779 x = rtmp+16*ajtmp[j]; 1780 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1781 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1782 } 1783 /* load in initial (unfactored row) */ 1784 nz = ai[i+1] - ai[i]; 1785 ajtmpold = aj + ai[i]; 1786 v = aa + 16*ai[i]; 1787 for (j=0; j<nz; j++) { 1788 x = rtmp+16*ajtmpold[j]; 1789 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1790 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1791 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1792 x[14] = v[14]; x[15] = v[15]; 1793 v += 16; 1794 } 1795 row = *ajtmp++; 1796 while (row < i) { 1797 pc = rtmp + 16*row; 1798 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1799 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1800 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1801 p15 = pc[14]; p16 = pc[15]; 1802 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1803 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1804 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1805 || p16 != 0.0) { 1806 pv = ba + 16*diag_offset[row]; 1807 pj = bj + diag_offset[row] + 1; 1808 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1809 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1810 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1811 x15 = pv[14]; x16 = pv[15]; 1812 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1813 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1814 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1815 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1816 1817 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1818 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1819 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1820 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1821 1822 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1823 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1824 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1825 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1826 1827 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1828 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1829 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1830 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1831 1832 nz = bi[row+1] - diag_offset[row] - 1; 1833 pv += 16; 1834 for (j=0; j<nz; j++) { 1835 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1836 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1837 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1838 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1839 x = rtmp + 16*pj[j]; 1840 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1841 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1842 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1843 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1844 1845 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1846 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1847 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1848 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1849 1850 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1851 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1852 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1853 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1854 1855 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1856 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1857 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1858 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1859 1860 pv += 16; 1861 } 1862 PLogFlops(128*nz+112); 1863 } 1864 row = *ajtmp++; 1865 } 1866 /* finished row so stick it into b->a */ 1867 pv = ba + 16*bi[i]; 1868 pj = bj + bi[i]; 1869 nz = bi[i+1] - bi[i]; 1870 for (j=0; j<nz; j++) { 1871 x = rtmp+16*pj[j]; 1872 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1873 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1874 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1875 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1876 pv += 16; 1877 } 1878 /* invert diagonal block */ 1879 w = ba + 16*diag_offset[i]; 1880 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1881 } 1882 1883 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1884 C->factor = FACTOR_LU; 1885 C->assembled = PETSC_TRUE; 1886 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1887 PetscFunctionReturn(0); 1888 } 1889 1890 /* 1891 Version for when blocks are 3 by 3 1892 */ 1893 #undef __FUNC__ 1894 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_3" 1895 int MatCholeskyFactorNumeric_SeqSBAIJ_3(Mat A,Mat *B) 1896 { 1897 Mat C = *B; 1898 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1899 IS isrow = b->row,isicol = b->icol; 1900 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1901 int *ajtmpold,*ajtmp,nz,row,*ai=a->i,*aj=a->j; 1902 int *diag_offset = b->diag,idx,*pj; 1903 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1904 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1905 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 1906 MatScalar *ba = b->a,*aa = a->a; 1907 1908 PetscFunctionBegin; 1909 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1910 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1911 rtmp = (MatScalar*)PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1912 1913 for (i=0; i<n; i++) { 1914 nz = bi[i+1] - bi[i]; 1915 ajtmp = bj + bi[i]; 1916 for (j=0; j<nz; j++) { 1917 x = rtmp + 9*ajtmp[j]; 1918 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 1919 } 1920 /* load in initial (unfactored row) */ 1921 idx = r[i]; 1922 nz = ai[idx+1] - ai[idx]; 1923 ajtmpold = aj + ai[idx]; 1924 v = aa + 9*ai[idx]; 1925 for (j=0; j<nz; j++) { 1926 x = rtmp + 9*ic[ajtmpold[j]]; 1927 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1928 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1929 v += 9; 1930 } 1931 row = *ajtmp++; 1932 while (row < i) { 1933 pc = rtmp + 9*row; 1934 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1935 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1936 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1937 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 1938 pv = ba + 9*diag_offset[row]; 1939 pj = bj + diag_offset[row] + 1; 1940 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1941 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1942 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 1943 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 1944 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 1945 1946 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 1947 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 1948 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 1949 1950 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 1951 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 1952 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 1953 nz = bi[row+1] - diag_offset[row] - 1; 1954 pv += 9; 1955 for (j=0; j<nz; j++) { 1956 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1957 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1958 x = rtmp + 9*pj[j]; 1959 x[0] -= m1*x1 + m4*x2 + m7*x3; 1960 x[1] -= m2*x1 + m5*x2 + m8*x3; 1961 x[2] -= m3*x1 + m6*x2 + m9*x3; 1962 1963 x[3] -= m1*x4 + m4*x5 + m7*x6; 1964 x[4] -= m2*x4 + m5*x5 + m8*x6; 1965 x[5] -= m3*x4 + m6*x5 + m9*x6; 1966 1967 x[6] -= m1*x7 + m4*x8 + m7*x9; 1968 x[7] -= m2*x7 + m5*x8 + m8*x9; 1969 x[8] -= m3*x7 + m6*x8 + m9*x9; 1970 pv += 9; 1971 } 1972 PLogFlops(54*nz+36); 1973 } 1974 row = *ajtmp++; 1975 } 1976 /* finished row so stick it into b->a */ 1977 pv = ba + 9*bi[i]; 1978 pj = bj + bi[i]; 1979 nz = bi[i+1] - bi[i]; 1980 for (j=0; j<nz; j++) { 1981 x = rtmp + 9*pj[j]; 1982 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1983 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1984 pv += 9; 1985 } 1986 /* invert diagonal block */ 1987 w = ba + 9*diag_offset[i]; 1988 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 1989 } 1990 1991 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1992 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1993 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1994 C->factor = FACTOR_LU; 1995 C->assembled = PETSC_TRUE; 1996 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 1997 PetscFunctionReturn(0); 1998 } 1999 /* 2000 Version for when blocks are 3 by 3 Using natural ordering 2001 */ 2002 #undef __FUNC__ 2003 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering" 2004 int MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering(Mat A,Mat *B) 2005 { 2006 Mat C = *B; 2007 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 2008 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 2009 int *ajtmpold,*ajtmp,nz,row; 2010 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 2011 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2012 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 2013 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 2014 MatScalar *ba = b->a,*aa = a->a; 2015 2016 PetscFunctionBegin; 2017 rtmp = (MatScalar*)PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2018 2019 for (i=0; i<n; i++) { 2020 nz = bi[i+1] - bi[i]; 2021 ajtmp = bj + bi[i]; 2022 for (j=0; j<nz; j++) { 2023 x = rtmp+9*ajtmp[j]; 2024 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 2025 } 2026 /* load in initial (unfactored row) */ 2027 nz = ai[i+1] - ai[i]; 2028 ajtmpold = aj + ai[i]; 2029 v = aa + 9*ai[i]; 2030 for (j=0; j<nz; j++) { 2031 x = rtmp+9*ajtmpold[j]; 2032 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2033 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 2034 v += 9; 2035 } 2036 row = *ajtmp++; 2037 while (row < i) { 2038 pc = rtmp + 9*row; 2039 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2040 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 2041 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 2042 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 2043 pv = ba + 9*diag_offset[row]; 2044 pj = bj + diag_offset[row] + 1; 2045 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2046 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2047 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 2048 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 2049 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 2050 2051 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 2052 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 2053 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 2054 2055 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 2056 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 2057 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 2058 2059 nz = bi[row+1] - diag_offset[row] - 1; 2060 pv += 9; 2061 for (j=0; j<nz; j++) { 2062 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2063 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2064 x = rtmp + 9*pj[j]; 2065 x[0] -= m1*x1 + m4*x2 + m7*x3; 2066 x[1] -= m2*x1 + m5*x2 + m8*x3; 2067 x[2] -= m3*x1 + m6*x2 + m9*x3; 2068 2069 x[3] -= m1*x4 + m4*x5 + m7*x6; 2070 x[4] -= m2*x4 + m5*x5 + m8*x6; 2071 x[5] -= m3*x4 + m6*x5 + m9*x6; 2072 2073 x[6] -= m1*x7 + m4*x8 + m7*x9; 2074 x[7] -= m2*x7 + m5*x8 + m8*x9; 2075 x[8] -= m3*x7 + m6*x8 + m9*x9; 2076 pv += 9; 2077 } 2078 PLogFlops(54*nz+36); 2079 } 2080 row = *ajtmp++; 2081 } 2082 /* finished row so stick it into b->a */ 2083 pv = ba + 9*bi[i]; 2084 pj = bj + bi[i]; 2085 nz = bi[i+1] - bi[i]; 2086 for (j=0; j<nz; j++) { 2087 x = rtmp+9*pj[j]; 2088 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2089 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 2090 pv += 9; 2091 } 2092 /* invert diagonal block */ 2093 w = ba + 9*diag_offset[i]; 2094 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 2095 } 2096 2097 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2098 C->factor = FACTOR_LU; 2099 C->assembled = PETSC_TRUE; 2100 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 2101 PetscFunctionReturn(0); 2102 } 2103 2104 /* 2105 Numeric (-UT)*D*(-U) factorization for SBAIJ format. Modified from SNF of YSMP. 2106 Version for blocks are 2 by 2. 2107 */ 2108 #undef __FUNC__ 2109 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_2" 2110 int MatCholeskyFactorNumeric_SeqSBAIJ_2(Mat A,Mat *B) 2111 { 2112 Mat C = *B; 2113 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 2114 IS ip = b->row; 2115 int *rip,ierr,i,j,mbs = a->mbs,*bi = b->i,*bj = b->j; 2116 int *ai,*aj,*r,bs2 = a->bs2; 2117 MatScalar *rtmp; 2118 MatScalar *ba = b->a,*aa,*ak,*ap; 2119 MatScalar *dk,*uik; 2120 int k,jmin,jmax,*jl,*il,vj,nexti,juj,ili; 2121 int k1; 2122 MatScalar *up,*diag,*rtmp_ptr; 2123 2124 PetscFunctionBegin; 2125 printf("called factornum_2, bs2: %d\n",bs2); 2126 ierr = ISGetIndices(ip,&rip);CHKERRQ(ierr); 2127 2128 if (!a->permute){ 2129 ai = a->i; aj = a->j; aa = a->a; 2130 } else { 2131 ai = a->inew; aj = a->jnew; 2132 aa = (MatScalar*)PetscMalloc(bs2*ai[mbs]*sizeof(MatScalar));CHKPTRQ(aa); 2133 ierr = PetscMemcpy(aa,a->a,bs2*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr); 2134 r = (int*)PetscMalloc(ai[mbs]*sizeof(int));CHKPTRQ(r); 2135 ierr= PetscMemcpy(r,a->a2anew,(ai[mbs])*sizeof(int));CHKERRQ(ierr); 2136 2137 ak = (MatScalar*)PetscMalloc(bs2*sizeof(MatScalar));CHKPTRQ(ak); 2138 2139 for (i=0; i<mbs; i++){ 2140 jmin = ai[i]; jmax = ai[i+1]; 2141 for (j=jmin; j<jmax; j++){ 2142 while (r[j] != j){ 2143 k = r[j]; r[j] = r[k]; r[k] = k; 2144 for (k1=0; k1<bs2; k1++){ 2145 ak[k1] = aa[k*bs2+k1]; 2146 aa[k*bs2+k1] = aa[j*bs2+k1]; 2147 aa[j*bs2+k1] = ak[k1]; 2148 } 2149 } 2150 } 2151 } 2152 2153 /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks. 2154 should be combined into the above loops */ 2155 for (i=0; i<mbs; i++){ 2156 jmin = ai[i]; jmax = ai[i+1]; 2157 for (j=jmin; j<jmax; j++){ 2158 if (i > aj[j]){ 2159 ap = aa + j*bs2; /* ptr to the beginning of the block */ 2160 ak[1] = ap[1]; /* swap ap[1] and ap[2] */ 2161 ap[1] = ap[2]; 2162 ap[2] = ak[1]; 2163 } 2164 } 2165 } 2166 ierr = PetscFree(r);CHKERRA(ierr); 2167 ierr = PetscFree(ak);CHKERRA(ierr); 2168 } 2169 2170 /* initialization */ 2171 /* il and jl record the first nonzero element in each row of the accessing 2172 window U(0:k, k:mbs-1). 2173 jl: list of rows to be added to uneliminated rows 2174 i>= k: jl(i) is the first row to be added to row i 2175 i< k: jl(i) is the row following row i in some list of rows 2176 jl(i) = mbs indicates the end of a list 2177 il(i): points to the first nonzero element in columns k,...,mbs-1 of 2178 row i of U */ 2179 rtmp = (MatScalar*)PetscMalloc(bs2*mbs*sizeof(MatScalar));CHKPTRQ(rtmp); 2180 ierr = PetscMemzero(rtmp,bs2*sizeof(MatScalar));CHKERRQ(ierr); 2181 il = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(il); 2182 jl = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(jl); 2183 dk = (MatScalar*)PetscMalloc(bs2*sizeof(MatScalar));CHKPTRQ(dk); 2184 uik= (MatScalar*)PetscMalloc(bs2*sizeof(MatScalar));CHKPTRQ(uik); 2185 2186 for (i=0; i<mbs; i++) { 2187 jl[i] = mbs; il[0] = 0; 2188 } 2189 2190 /* for each row k */ 2191 for (k = 0; k<mbs; k++){ 2192 2193 /*initialize k-th row with elements nonzero in row perm(k) of A */ 2194 jmin = ai[rip[k]]; jmax = ai[rip[k]+1]; 2195 if (jmin < jmax) { 2196 for (j = jmin; j < jmax; j++){ 2197 vj = rip[aj[j]]; /* block col. index */ 2198 if (k <= vj) { 2199 ap = aa + j*bs2; /* ptr to the beginning of the block */ 2200 for (i=0; i<bs2; i++){ 2201 rtmp[vj*bs2 + i] = *ap; ap++; 2202 } 2203 } 2204 } 2205 } 2206 2207 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 2208 ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr); /* dk = rtmp[k]; */ 2209 i = jl[k]; /* first row to be added to k_th row */ 2210 /* printf(" k=%d, pivot row = %d\n",k,i); */ 2211 2212 while (i < mbs){ 2213 nexti = jl[i]; /* next row to be added to k_th row */ 2214 2215 /* compute multiplier */ 2216 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 2217 2218 /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */ 2219 diag = ba + i*bs2; 2220 up = ba + ili*bs2; 2221 uik[0] = -(diag[0]*up[0] + diag[2]*up[1]); 2222 uik[1] = -(diag[1]*up[0] + diag[3]*up[1]); 2223 uik[2] = -(diag[0]*up[2] + diag[2]*up[3]); 2224 uik[3] = -(diag[1]*up[2] + diag[3]*up[3]); 2225 2226 /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */ 2227 dk[0] += uik[0]*up[0] + uik[1]*up[1]; 2228 dk[1] += uik[2]*up[0] + uik[3]*up[1]; 2229 dk[2] += uik[0]*up[2] + uik[1]*up[3]; 2230 dk[3] += uik[2]*up[2] + uik[3]*up[3]; 2231 2232 /* update -U(i,k): ba[ili] = uik */ 2233 ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 2234 2235 /* add multiple of row i to k-th row ... */ 2236 jmin = ili + 1; jmax = bi[i+1]; 2237 if (jmin < jmax){ 2238 for (j=jmin; j<jmax; j++) { 2239 /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */ 2240 rtmp_ptr = rtmp + bj[j]*bs2; 2241 up = ba + j*bs2; 2242 rtmp_ptr[0] += uik[0]*up[0] + uik[1]*up[1]; 2243 rtmp_ptr[1] += uik[2]*up[0] + uik[3]*up[1]; 2244 rtmp_ptr[2] += uik[0]*up[2] + uik[1]*up[3]; 2245 rtmp_ptr[3] += uik[2]*up[2] + uik[3]*up[3]; 2246 } 2247 2248 /* ... add i to row list for next nonzero entry */ 2249 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 2250 j = bj[jmin]; 2251 jl[i] = jl[j]; jl[j] = i; /* update jl */ 2252 } 2253 i = nexti; /* printf(" pivot row i=%d\n",i); */ 2254 } 2255 #ifdef CONT 2256 /* check for zero pivot and save diagoanl element */ 2257 if (dk == 0.0){ 2258 SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot"); 2259 }else if (PetscRealPart(dk) < 0){ 2260 ierr = PetscPrintf(PETSC_COMM_SELF,"Negative pivot: d[%d] = %g\n",k,dk); 2261 } 2262 #endif 2263 /* save nonzero entries in k-th row of U ... */ 2264 2265 /* invert diagonal block: ba[k] = 1.0/dk */ 2266 diag = ba+k*bs2; 2267 ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr); 2268 ierr = Kernel_A_gets_inverse_A_2(diag);CHKERRQ(ierr); 2269 2270 jmin = bi[k]; jmax = bi[k+1]; 2271 if (jmin < jmax) { 2272 for (j=jmin; j<jmax; j++){ 2273 juj = bj[j]; 2274 2275 /* ba[j] = rtmp[juj]; rtmp[juj] = 0.0; */ 2276 up = ba + j*bs2; 2277 rtmp_ptr = rtmp + juj*bs2; 2278 for (k1=0; k1<bs2; k1++){ 2279 *up = *rtmp_ptr; *rtmp_ptr = 0.0; 2280 up++; rtmp_ptr++; 2281 } 2282 } 2283 /* ... add k to row list for first nonzero entry in k-th row */ 2284 il[k] = jmin; 2285 i = bj[jmin]; 2286 jl[k] = jl[i]; jl[i] = k; 2287 } 2288 } 2289 2290 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2291 ierr = PetscFree(il);CHKERRQ(ierr); 2292 ierr = PetscFree(jl);CHKERRQ(ierr); 2293 ierr = PetscFree(dk);CHKERRQ(ierr); 2294 ierr = PetscFree(uik);CHKERRQ(ierr); 2295 if (a->permute){ 2296 ierr = PetscFree(aa);CHKERRQ(ierr); 2297 } 2298 2299 ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr); 2300 C->factor = FACTOR_CHOLESKY; 2301 /* C->factor = FACTOR_LU; */ 2302 C->assembled = PETSC_TRUE; 2303 PLogFlops(b->mbs); 2304 PetscFunctionReturn(0); 2305 } 2306 2307 /* 2308 Version for when blocks are 2 by 2 Using natural ordering 2309 */ 2310 #undef __FUNC__ 2311 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering" 2312 int MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering(Mat A,Mat *B) 2313 { 2314 Mat C = *B; 2315 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 2316 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 2317 int *ajtmpold,*ajtmp,nz,row; 2318 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 2319 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2320 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,x1,x2,x3,x4; 2321 MatScalar *ba = b->a,*aa = a->a; 2322 2323 PetscFunctionBegin; 2324 rtmp = (MatScalar*)PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2325 2326 for (i=0; i<n; i++) { 2327 nz = bi[i+1] - bi[i]; 2328 ajtmp = bj + bi[i]; 2329 for (j=0; j<nz; j++) { 2330 x = rtmp+4*ajtmp[j]; 2331 x[0] = x[1] = x[2] = x[3] = 0.0; 2332 } 2333 /* load in initial (unfactored row) */ 2334 nz = ai[i+1] - ai[i]; 2335 ajtmpold = aj + ai[i]; 2336 v = aa + 4*ai[i]; 2337 for (j=0; j<nz; j++) { 2338 x = rtmp+4*ajtmpold[j]; 2339 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2340 v += 4; 2341 } 2342 row = *ajtmp++; 2343 while (row < i) { 2344 pc = rtmp + 4*row; 2345 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2346 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2347 pv = ba + 4*diag_offset[row]; 2348 pj = bj + diag_offset[row] + 1; 2349 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2350 pc[0] = m1 = p1*x1 + p3*x2; 2351 pc[1] = m2 = p2*x1 + p4*x2; 2352 pc[2] = m3 = p1*x3 + p3*x4; 2353 pc[3] = m4 = p2*x3 + p4*x4; 2354 nz = bi[row+1] - diag_offset[row] - 1; 2355 pv += 4; 2356 for (j=0; j<nz; j++) { 2357 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2358 x = rtmp + 4*pj[j]; 2359 x[0] -= m1*x1 + m3*x2; 2360 x[1] -= m2*x1 + m4*x2; 2361 x[2] -= m1*x3 + m3*x4; 2362 x[3] -= m2*x3 + m4*x4; 2363 pv += 4; 2364 } 2365 PLogFlops(16*nz+12); 2366 } 2367 row = *ajtmp++; 2368 } 2369 /* finished row so stick it into b->a */ 2370 pv = ba + 4*bi[i]; 2371 pj = bj + bi[i]; 2372 nz = bi[i+1] - bi[i]; 2373 for (j=0; j<nz; j++) { 2374 x = rtmp+4*pj[j]; 2375 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2376 pv += 4; 2377 } 2378 /* invert diagonal block */ 2379 w = ba + 4*diag_offset[i]; 2380 ierr = Kernel_A_gets_inverse_A_2(w);CHKERRQ(ierr); 2381 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2382 } 2383 2384 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2385 C->factor = FACTOR_LU; 2386 C->assembled = PETSC_TRUE; 2387 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2388 PetscFunctionReturn(0); 2389 } 2390 2391 /* 2392 Numeric (-UT)*D*(-U) factorization for SBAIJ format. Modified from SNF of YSMP. 2393 Version for blocks are 1 by 1. 2394 */ 2395 #undef __FUNC__ 2396 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_1" 2397 int MatCholeskyFactorNumeric_SeqSBAIJ_1(Mat A,Mat *B) 2398 { 2399 Mat C = *B; 2400 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 2401 IS ip = b->row; 2402 int *rip,ierr,i,j,mbs = a->mbs,*bi = b->i,*bj = b->j; 2403 int *ai,*aj,*r; 2404 MatScalar *rtmp; 2405 MatScalar *ba = b->a,*aa,ak; 2406 MatScalar dk,uikdi; 2407 int k,jmin,jmax,*jl,*il,vj,nexti,juj,ili; 2408 2409 PetscFunctionBegin; 2410 ierr = ISGetIndices(ip,&rip);CHKERRQ(ierr); 2411 2412 if (!a->permute){ 2413 ai = a->i; aj = a->j; aa = a->a; 2414 } else { 2415 ai = a->inew; aj = a->jnew; 2416 aa = (MatScalar*)PetscMalloc(ai[mbs]*sizeof(MatScalar));CHKPTRQ(aa); 2417 ierr = PetscMemcpy(aa,a->a,ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr); 2418 r = (int*)PetscMalloc(ai[mbs]*sizeof(int));CHKPTRQ(r); 2419 ierr= PetscMemcpy(r,a->a2anew,(ai[mbs])*sizeof(int));CHKERRQ(ierr); 2420 2421 jmin = ai[0]; jmax = ai[mbs]; 2422 for (j=jmin; j<jmax; j++){ 2423 while (r[j] != j){ 2424 k = r[j]; r[j] = r[k]; r[k] = k; 2425 ak = aa[k]; aa[k] = aa[j]; aa[j] = ak; 2426 } 2427 } 2428 ierr = PetscFree(r);CHKERRA(ierr); 2429 } 2430 2431 /* initialization */ 2432 /* il and jl record the first nonzero element in each row of the accessing 2433 window U(0:k, k:mbs-1). 2434 jl: list of rows to be added to uneliminated rows 2435 i>= k: jl(i) is the first row to be added to row i 2436 i< k: jl(i) is the row following row i in some list of rows 2437 jl(i) = mbs indicates the end of a list 2438 il(i): points to the first nonzero element in columns k,...,mbs-1 of 2439 row i of U */ 2440 rtmp = (MatScalar*)PetscMalloc(mbs*sizeof(MatScalar));CHKPTRQ(rtmp); 2441 il = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(il); 2442 jl = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(jl); 2443 for (i=0; i<mbs; i++) { 2444 rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0; 2445 } 2446 2447 /* for each row k */ 2448 for (k = 0; k<mbs; k++){ 2449 2450 /*initialize k-th row with elements nonzero in row perm(k) of A */ 2451 jmin = ai[rip[k]]; jmax = ai[rip[k]+1]; 2452 if (jmin < jmax) { 2453 for (j = jmin; j < jmax; j++){ 2454 vj = rip[aj[j]]; 2455 if (k <= vj) rtmp[vj] = aa[j]; 2456 } 2457 } 2458 2459 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 2460 dk = rtmp[k]; 2461 i = jl[k]; /* first row to be added to k_th row */ 2462 /* printf(" k=%d, pivot row = %d\n",k,i); */ 2463 2464 while (i < mbs){ 2465 nexti = jl[i]; /* next row to be added to k_th row */ 2466 2467 /* compute multiplier, update D(k) and U(i,k) */ 2468 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 2469 uikdi = - ba[ili]*ba[i]; 2470 dk += uikdi*ba[ili]; 2471 ba[ili] = uikdi; /* -U(i,k) */ 2472 2473 /* add multiple of row i to k-th row ... */ 2474 jmin = ili + 1; jmax = bi[i+1]; 2475 if (jmin < jmax){ 2476 for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j]; 2477 /* ... add i to row list for next nonzero entry */ 2478 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 2479 j = bj[jmin]; 2480 jl[i] = jl[j]; jl[j] = i; /* update jl */ 2481 } 2482 i = nexti; /* printf(" pivot row i=%d\n",i); */ 2483 } 2484 2485 /* check for zero pivot and save diagoanl element */ 2486 if (dk == 0.0){ 2487 SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot"); 2488 }else if (PetscRealPart(dk) < 0){ 2489 ierr = PetscPrintf(PETSC_COMM_SELF,"Negative pivot: d[%d] = %g\n",k,dk); 2490 } 2491 2492 /* save nonzero entries in k-th row of U ... */ 2493 ba[k] = 1.0/dk; 2494 jmin = bi[k]; jmax = bi[k+1]; 2495 if (jmin < jmax) { 2496 for (j=jmin; j<jmax; j++){ 2497 juj = bj[j]; ba[j] = rtmp[juj]; rtmp[juj] = 0.0; 2498 } 2499 /* ... add k to row list for first nonzero entry in k-th row */ 2500 il[k] = jmin; 2501 i = bj[jmin]; 2502 jl[k] = jl[i]; jl[i] = k; 2503 } 2504 } 2505 2506 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2507 ierr = PetscFree(il);CHKERRQ(ierr); 2508 ierr = PetscFree(jl);CHKERRQ(ierr); 2509 if (a->permute){ 2510 ierr = PetscFree(aa);CHKERRQ(ierr); 2511 } 2512 2513 ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr); 2514 C->factor = FACTOR_CHOLESKY; 2515 /* C->factor = FACTOR_LU; */ 2516 C->assembled = PETSC_TRUE; 2517 PLogFlops(b->mbs); 2518 PetscFunctionReturn(0); 2519 } 2520 2521 #undef __FUNC__ 2522 #define __FUNC__ "MatCholeskyFactor_SeqSBAIJ" 2523 int MatCholeskyFactor_SeqSBAIJ(Mat A,IS perm,PetscReal f) 2524 { 2525 int ierr; 2526 Mat C; 2527 2528 PetscFunctionBegin; 2529 ierr = MatCholeskyFactorSymbolic(A,perm,f,&C);CHKERRQ(ierr); 2530 ierr = MatCholeskyFactorNumeric(A,&C);CHKERRQ(ierr); 2531 ierr = MatHeaderCopy(A,C);CHKERRQ(ierr); 2532 PetscFunctionReturn(0); 2533 } 2534 2535 2536