1 /* Using Modified Sparse Row (MSR) storage. 2 See page 85, "Iterative Methods ..." by Saad. */ 3 4 /*$Id: sbaijfact.c,v 1.33 2000/10/27 16:32:23 hzhang Exp hzhang $*/ 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_CHOLESKY; 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 #ifdef TEMP 191 for (k=0; k<mbs; k++){ 192 nzk = b->i[k+1] - b->i[k]; 193 printf("\n b->i[%d]: %d, nzk: %d\n",k,b->i[k],nzk); 194 jmin = b->i[k]; jmax = b->i[k+1]; 195 for (j=jmin; j<jmax; j++){ 196 printf(" %d,",b->j[j]); 197 } 198 } 199 #endif 200 PetscFunctionReturn(0); 201 } 202 203 #undef __FUNC__ 204 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_N" 205 int MatCholeskyFactorNumeric_SeqSBAIJ_N(Mat A,Mat *B) 206 { 207 Mat C = *B; 208 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 209 IS isrow = b->row,isicol = b->icol; 210 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 211 int *ajtmpold,*ajtmp,nz,row,bslog,*ai=a->i,*aj=a->j,k,flg; 212 int *diag_offset=b->diag,diag,bs=a->bs,bs2 = a->bs2,*v_pivots,*pj; 213 MatScalar *ba = b->a,*aa = a->a,*pv,*v,*rtmp,*multiplier,*v_work,*pc,*w; 214 215 PetscFunctionBegin; 216 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 217 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 218 rtmp = (MatScalar*)PetscMalloc(bs2*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 219 ierr = PetscMemzero(rtmp,bs2*(n+1)*sizeof(MatScalar));CHKERRQ(ierr); 220 /* generate work space needed by dense LU factorization */ 221 v_work = (MatScalar*)PetscMalloc(bs*sizeof(int) + (bs+bs2)*sizeof(MatScalar));CHKPTRQ(v_work); 222 multiplier = v_work + bs; 223 v_pivots = (int*)(multiplier + bs2); 224 225 /* flops in while loop */ 226 bslog = 2*bs*bs2; 227 228 for (i=0; i<n; i++) { 229 nz = bi[i+1] - bi[i]; 230 ajtmp = bj + bi[i]; 231 for (j=0; j<nz; j++) { 232 ierr = PetscMemzero(rtmp+bs2*ajtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 233 } 234 /* load in initial (unfactored row) */ 235 nz = ai[r[i]+1] - ai[r[i]]; 236 ajtmpold = aj + ai[r[i]]; 237 v = aa + bs2*ai[r[i]]; 238 for (j=0; j<nz; j++) { 239 ierr = PetscMemcpy(rtmp+bs2*ic[ajtmpold[j]],v+bs2*j,bs2*sizeof(MatScalar));CHKERRQ(ierr); 240 } 241 row = *ajtmp++; 242 while (row < i) { 243 pc = rtmp + bs2*row; 244 /* if (*pc) { */ 245 for (flg=0,k=0; k<bs2; k++) { if (pc[k]!=0.0) { flg =1; break; }} 246 if (flg) { 247 pv = ba + bs2*diag_offset[row]; 248 pj = bj + diag_offset[row] + 1; 249 Kernel_A_gets_A_times_B(bs,pc,pv,multiplier); 250 nz = bi[row+1] - diag_offset[row] - 1; 251 pv += bs2; 252 for (j=0; j<nz; j++) { 253 Kernel_A_gets_A_minus_B_times_C(bs,rtmp+bs2*pj[j],pc,pv+bs2*j); 254 } 255 PLogFlops(bslog*(nz+1)-bs); 256 } 257 row = *ajtmp++; 258 } 259 /* finished row so stick it into b->a */ 260 pv = ba + bs2*bi[i]; 261 pj = bj + bi[i]; 262 nz = bi[i+1] - bi[i]; 263 for (j=0; j<nz; j++) { 264 ierr = PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 265 } 266 diag = diag_offset[i] - bi[i]; 267 /* invert diagonal block */ 268 w = pv + bs2*diag; 269 Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work); 270 } 271 272 ierr = PetscFree(rtmp);CHKERRQ(ierr); 273 ierr = PetscFree(v_work);CHKERRQ(ierr); 274 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 275 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 276 C->factor = FACTOR_LU; 277 C->assembled = PETSC_TRUE; 278 PLogFlops(1.3333*bs*bs2*b->mbs); /* from inverting diagonal blocks */ 279 PetscFunctionReturn(0); 280 } 281 282 /* 283 Version for when blocks are 7 by 7 284 */ 285 #undef __FUNC__ 286 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_7" 287 int MatCholeskyFactorNumeric_SeqSBAIJ_7(Mat A,Mat *B) 288 { 289 Mat C = *B; 290 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 291 IS isrow = b->row,isicol = b->icol; 292 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 293 int *ajtmpold,*ajtmp,nz,row; 294 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 295 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 296 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 297 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 298 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 299 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 300 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 301 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 302 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 303 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 304 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 305 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 306 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 307 MatScalar *ba = b->a,*aa = a->a; 308 309 PetscFunctionBegin; 310 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 311 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 312 rtmp = (MatScalar*)PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 313 314 for (i=0; i<n; i++) { 315 nz = bi[i+1] - bi[i]; 316 ajtmp = bj + bi[i]; 317 for (j=0; j<nz; j++) { 318 x = rtmp+49*ajtmp[j]; 319 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 320 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 321 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 322 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 323 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 324 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 325 } 326 /* load in initial (unfactored row) */ 327 idx = r[i]; 328 nz = ai[idx+1] - ai[idx]; 329 ajtmpold = aj + ai[idx]; 330 v = aa + 49*ai[idx]; 331 for (j=0; j<nz; j++) { 332 x = rtmp+49*ic[ajtmpold[j]]; 333 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 334 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 335 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 336 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 337 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 338 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 339 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 340 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 341 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 342 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 343 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 344 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 345 x[48] = v[48]; 346 v += 49; 347 } 348 row = *ajtmp++; 349 while (row < i) { 350 pc = rtmp + 49*row; 351 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 352 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 353 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 354 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 355 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 356 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 357 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 358 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 359 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 360 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 361 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 362 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 363 p49 = pc[48]; 364 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 365 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 366 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 367 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 368 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 369 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 370 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 371 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 372 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 373 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 374 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 375 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 376 p49 != 0.0) { 377 pv = ba + 49*diag_offset[row]; 378 pj = bj + diag_offset[row] + 1; 379 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 380 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 381 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 382 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 383 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 384 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 385 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 386 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 387 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 388 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 389 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 390 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 391 x49 = pv[48]; 392 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 393 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 394 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 395 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 396 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 397 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 398 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 399 400 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 401 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 402 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 403 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 404 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 405 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 406 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 407 408 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 409 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 410 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 411 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 412 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 413 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 414 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 415 416 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 417 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 418 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 419 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 420 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 421 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 422 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 423 424 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 425 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 426 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 427 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 428 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 429 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 430 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 431 432 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 433 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 434 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 435 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 436 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 437 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 438 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 439 440 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 441 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 442 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 443 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 444 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 445 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 446 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 447 448 nz = bi[row+1] - diag_offset[row] - 1; 449 pv += 49; 450 for (j=0; j<nz; j++) { 451 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 452 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 453 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 454 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 455 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 456 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 457 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 458 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 459 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 460 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 461 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 462 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 463 x49 = pv[48]; 464 x = rtmp + 49*pj[j]; 465 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 466 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 467 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 468 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 469 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 470 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 471 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 472 473 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 474 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 475 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 476 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 477 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 478 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 479 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 480 481 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 482 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 483 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 484 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 485 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 486 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 487 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 488 489 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 490 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 491 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 492 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 493 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 494 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 495 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 496 497 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 498 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 499 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 500 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 501 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 502 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 503 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 504 505 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 506 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 507 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 508 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 509 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 510 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 511 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 512 513 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 514 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 515 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 516 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 517 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 518 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 519 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 520 pv += 49; 521 } 522 PLogFlops(686*nz+637); 523 } 524 row = *ajtmp++; 525 } 526 /* finished row so stick it into b->a */ 527 pv = ba + 49*bi[i]; 528 pj = bj + bi[i]; 529 nz = bi[i+1] - bi[i]; 530 for (j=0; j<nz; j++) { 531 x = rtmp+49*pj[j]; 532 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 533 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 534 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 535 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 536 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 537 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 538 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 539 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 540 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 541 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 542 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 543 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 544 pv[48] = x[48]; 545 pv += 49; 546 } 547 /* invert diagonal block */ 548 w = ba + 49*diag_offset[i]; 549 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 550 } 551 552 ierr = PetscFree(rtmp);CHKERRQ(ierr); 553 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 554 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 555 C->factor = FACTOR_LU; 556 C->assembled = PETSC_TRUE; 557 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 558 PetscFunctionReturn(0); 559 } 560 561 /* 562 Version for when blocks are 7 by 7 Using natural ordering 563 */ 564 #undef __FUNC__ 565 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering" 566 int MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering(Mat A,Mat *B) 567 { 568 Mat C = *B; 569 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 570 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 571 int *ajtmpold,*ajtmp,nz,row; 572 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 573 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 574 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 575 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 576 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 577 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 578 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 579 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 580 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 581 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 582 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 583 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 584 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 585 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 586 MatScalar *ba = b->a,*aa = a->a; 587 588 PetscFunctionBegin; 589 rtmp = (MatScalar*)PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 590 for (i=0; i<n; i++) { 591 nz = bi[i+1] - bi[i]; 592 ajtmp = bj + bi[i]; 593 for (j=0; j<nz; j++) { 594 x = rtmp+49*ajtmp[j]; 595 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 596 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 597 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 598 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 599 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 600 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 601 } 602 /* load in initial (unfactored row) */ 603 nz = ai[i+1] - ai[i]; 604 ajtmpold = aj + ai[i]; 605 v = aa + 49*ai[i]; 606 for (j=0; j<nz; j++) { 607 x = rtmp+49*ajtmpold[j]; 608 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 609 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 610 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 611 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 612 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 613 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 614 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 615 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 616 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 617 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 618 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 619 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 620 x[48] = v[48]; 621 v += 49; 622 } 623 row = *ajtmp++; 624 while (row < i) { 625 pc = rtmp + 49*row; 626 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 627 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 628 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 629 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 630 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 631 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 632 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 633 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 634 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 635 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 636 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 637 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 638 p49 = pc[48]; 639 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 640 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 641 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 642 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 643 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 644 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 645 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 646 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 647 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 648 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 649 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 650 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 651 p49 != 0.0) { 652 pv = ba + 49*diag_offset[row]; 653 pj = bj + diag_offset[row] + 1; 654 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 655 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 656 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 657 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 658 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 659 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 660 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 661 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 662 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 663 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 664 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 665 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 666 x49 = pv[48]; 667 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 668 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 669 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 670 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 671 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 672 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 673 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 674 675 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 676 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 677 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 678 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 679 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 680 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 681 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 682 683 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 684 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 685 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 686 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 687 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 688 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 689 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 690 691 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 692 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 693 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 694 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 695 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 696 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 697 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 698 699 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 700 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 701 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 702 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 703 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 704 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 705 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 706 707 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 708 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 709 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 710 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 711 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 712 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 713 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 714 715 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 716 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 717 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 718 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 719 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 720 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 721 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 722 723 nz = bi[row+1] - diag_offset[row] - 1; 724 pv += 49; 725 for (j=0; j<nz; j++) { 726 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 727 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 728 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 729 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 730 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 731 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 732 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 733 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 734 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 735 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 736 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 737 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 738 x49 = pv[48]; 739 x = rtmp + 49*pj[j]; 740 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 741 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 742 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 743 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 744 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 745 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 746 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 747 748 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 749 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 750 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 751 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 752 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 753 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 754 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 755 756 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 757 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 758 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 759 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 760 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 761 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 762 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 763 764 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 765 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 766 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 767 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 768 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 769 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 770 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 771 772 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 773 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 774 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 775 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 776 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 777 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 778 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 779 780 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 781 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 782 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 783 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 784 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 785 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 786 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 787 788 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 789 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 790 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 791 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 792 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 793 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 794 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 795 pv += 49; 796 } 797 PLogFlops(686*nz+637); 798 } 799 row = *ajtmp++; 800 } 801 /* finished row so stick it into b->a */ 802 pv = ba + 49*bi[i]; 803 pj = bj + bi[i]; 804 nz = bi[i+1] - bi[i]; 805 for (j=0; j<nz; j++) { 806 x = rtmp+49*pj[j]; 807 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 808 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 809 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 810 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 811 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 812 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 813 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 814 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 815 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 816 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 817 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 818 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 819 pv[48] = x[48]; 820 pv += 49; 821 } 822 /* invert diagonal block */ 823 w = ba + 49*diag_offset[i]; 824 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 825 } 826 827 ierr = PetscFree(rtmp);CHKERRQ(ierr); 828 C->factor = FACTOR_LU; 829 C->assembled = PETSC_TRUE; 830 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 831 PetscFunctionReturn(0); 832 } 833 834 /* ------------------------------------------------------------*/ 835 /* 836 Version for when blocks are 6 by 6 837 */ 838 #undef __FUNC__ 839 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_6" 840 int MatCholeskyFactorNumeric_SeqSBAIJ_6(Mat A,Mat *B) 841 { 842 Mat C = *B; 843 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 844 IS isrow = b->row,isicol = b->icol; 845 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 846 int *ajtmpold,*ajtmp,nz,row; 847 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 848 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 849 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 850 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 851 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 852 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 853 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 854 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 855 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 856 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 857 MatScalar *ba = b->a,*aa = a->a; 858 859 PetscFunctionBegin; 860 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 861 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 862 rtmp = (MatScalar*)PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 863 864 for (i=0; i<n; i++) { 865 nz = bi[i+1] - bi[i]; 866 ajtmp = bj + bi[i]; 867 for (j=0; j<nz; j++) { 868 x = rtmp+36*ajtmp[j]; 869 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 870 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 871 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 872 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 873 x[34] = x[35] = 0.0 ; 874 } 875 /* load in initial (unfactored row) */ 876 idx = r[i]; 877 nz = ai[idx+1] - ai[idx]; 878 ajtmpold = aj + ai[idx]; 879 v = aa + 36*ai[idx]; 880 for (j=0; j<nz; j++) { 881 x = rtmp+36*ic[ajtmpold[j]]; 882 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 883 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 884 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 885 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 886 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 887 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 888 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 889 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 890 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 891 v += 36; 892 } 893 row = *ajtmp++; 894 while (row < i) { 895 pc = rtmp + 36*row; 896 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 897 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 898 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 899 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 900 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 901 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 902 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 903 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 904 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 905 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 906 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 907 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 908 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 909 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 910 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 911 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 912 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 913 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 914 pv = ba + 36*diag_offset[row]; 915 pj = bj + diag_offset[row] + 1; 916 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 917 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 918 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 919 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 920 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 921 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 922 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 923 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 924 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 925 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 926 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 927 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 928 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 929 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 930 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 931 932 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 933 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 934 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 935 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 936 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 937 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 938 939 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 940 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 941 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 942 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 943 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 944 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 945 946 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 947 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 948 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 949 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 950 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 951 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 952 953 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 954 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 955 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 956 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 957 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 958 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 959 960 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 961 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 962 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 963 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 964 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 965 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 966 967 nz = bi[row+1] - diag_offset[row] - 1; 968 pv += 36; 969 for (j=0; j<nz; j++) { 970 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 971 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 972 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 973 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 974 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 975 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 976 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 977 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 978 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 979 x = rtmp + 36*pj[j]; 980 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 981 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 982 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 983 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 984 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 985 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 986 987 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 988 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 989 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 990 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 991 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 992 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 993 994 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 995 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 996 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 997 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 998 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 999 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 1000 1001 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 1002 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 1003 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 1004 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 1005 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 1006 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 1007 1008 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 1009 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 1010 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 1011 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 1012 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 1013 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 1014 1015 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 1016 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 1017 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 1018 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 1019 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 1020 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 1021 1022 pv += 36; 1023 } 1024 PLogFlops(432*nz+396); 1025 } 1026 row = *ajtmp++; 1027 } 1028 /* finished row so stick it into b->a */ 1029 pv = ba + 36*bi[i]; 1030 pj = bj + bi[i]; 1031 nz = bi[i+1] - bi[i]; 1032 for (j=0; j<nz; j++) { 1033 x = rtmp+36*pj[j]; 1034 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1035 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1036 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1037 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1038 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1039 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1040 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1041 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1042 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1043 pv += 36; 1044 } 1045 /* invert diagonal block */ 1046 w = ba + 36*diag_offset[i]; 1047 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1048 } 1049 1050 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1051 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1052 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1053 C->factor = FACTOR_LU; 1054 C->assembled = PETSC_TRUE; 1055 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1056 PetscFunctionReturn(0); 1057 } 1058 /* 1059 Version for when blocks are 6 by 6 Using natural ordering 1060 */ 1061 #undef __FUNC__ 1062 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering" 1063 int MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering(Mat A,Mat *B) 1064 { 1065 Mat C = *B; 1066 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1067 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1068 int *ajtmpold,*ajtmp,nz,row; 1069 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1070 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1071 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1072 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1073 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1074 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1075 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1076 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1077 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 1078 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 1079 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 1080 MatScalar *ba = b->a,*aa = a->a; 1081 1082 PetscFunctionBegin; 1083 rtmp = (MatScalar*)PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1084 for (i=0; i<n; i++) { 1085 nz = bi[i+1] - bi[i]; 1086 ajtmp = bj + bi[i]; 1087 for (j=0; j<nz; j++) { 1088 x = rtmp+36*ajtmp[j]; 1089 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1090 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1091 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 1092 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 1093 x[34] = x[35] = 0.0 ; 1094 } 1095 /* load in initial (unfactored row) */ 1096 nz = ai[i+1] - ai[i]; 1097 ajtmpold = aj + ai[i]; 1098 v = aa + 36*ai[i]; 1099 for (j=0; j<nz; j++) { 1100 x = rtmp+36*ajtmpold[j]; 1101 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1102 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 1103 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 1104 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 1105 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 1106 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1107 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 1108 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 1109 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 1110 v += 36; 1111 } 1112 row = *ajtmp++; 1113 while (row < i) { 1114 pc = rtmp + 36*row; 1115 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1116 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 1117 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 1118 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 1119 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 1120 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1121 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 1122 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 1123 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 1124 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 1125 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 1126 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 1127 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 1128 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 1129 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 1130 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 1131 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 1132 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 1133 pv = ba + 36*diag_offset[row]; 1134 pj = bj + diag_offset[row] + 1; 1135 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1136 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1137 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1138 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1139 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1140 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1141 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1142 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1143 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1144 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 1145 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 1146 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 1147 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 1148 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 1149 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 1150 1151 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 1152 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 1153 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 1154 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 1155 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 1156 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 1157 1158 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 1159 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 1160 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 1161 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 1162 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 1163 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 1164 1165 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 1166 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 1167 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 1168 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 1169 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 1170 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 1171 1172 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 1173 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 1174 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 1175 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 1176 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 1177 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 1178 1179 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 1180 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 1181 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 1182 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 1183 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 1184 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 1185 1186 nz = bi[row+1] - diag_offset[row] - 1; 1187 pv += 36; 1188 for (j=0; j<nz; j++) { 1189 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1190 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1191 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1192 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1193 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1194 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1195 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1196 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1197 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1198 x = rtmp + 36*pj[j]; 1199 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 1200 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 1201 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 1202 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 1203 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 1204 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 1205 1206 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 1207 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 1208 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 1209 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 1210 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 1211 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 1212 1213 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 1214 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 1215 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 1216 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 1217 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 1218 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 1219 1220 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 1221 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 1222 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 1223 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 1224 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 1225 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 1226 1227 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 1228 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 1229 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 1230 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 1231 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 1232 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 1233 1234 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 1235 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 1236 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 1237 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 1238 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 1239 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 1240 1241 pv += 36; 1242 } 1243 PLogFlops(432*nz+396); 1244 } 1245 row = *ajtmp++; 1246 } 1247 /* finished row so stick it into b->a */ 1248 pv = ba + 36*bi[i]; 1249 pj = bj + bi[i]; 1250 nz = bi[i+1] - bi[i]; 1251 for (j=0; j<nz; j++) { 1252 x = rtmp+36*pj[j]; 1253 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1254 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1255 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1256 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1257 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1258 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1259 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1260 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1261 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1262 pv += 36; 1263 } 1264 /* invert diagonal block */ 1265 w = ba + 36*diag_offset[i]; 1266 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1267 } 1268 1269 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1270 C->factor = FACTOR_LU; 1271 C->assembled = PETSC_TRUE; 1272 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1273 PetscFunctionReturn(0); 1274 } 1275 1276 /* 1277 Version for when blocks are 5 by 5 1278 */ 1279 #undef __FUNC__ 1280 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_5" 1281 int MatCholeskyFactorNumeric_SeqSBAIJ_5(Mat A,Mat *B) 1282 { 1283 Mat C = *B; 1284 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1285 IS isrow = b->row,isicol = b->icol; 1286 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1287 int *ajtmpold,*ajtmp,nz,row; 1288 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 1289 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1290 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1291 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1292 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 1293 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 1294 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1295 MatScalar *ba = b->a,*aa = a->a; 1296 1297 PetscFunctionBegin; 1298 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1299 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1300 rtmp = (MatScalar*)PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1301 1302 for (i=0; i<n; i++) { 1303 nz = bi[i+1] - bi[i]; 1304 ajtmp = bj + bi[i]; 1305 for (j=0; j<nz; j++) { 1306 x = rtmp+25*ajtmp[j]; 1307 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1308 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1309 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1310 } 1311 /* load in initial (unfactored row) */ 1312 idx = r[i]; 1313 nz = ai[idx+1] - ai[idx]; 1314 ajtmpold = aj + ai[idx]; 1315 v = aa + 25*ai[idx]; 1316 for (j=0; j<nz; j++) { 1317 x = rtmp+25*ic[ajtmpold[j]]; 1318 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1319 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1320 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1321 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; 1322 x[18] = v[18]; x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; 1323 x[22] = v[22]; x[23] = v[23]; x[24] = v[24]; 1324 v += 25; 1325 } 1326 row = *ajtmp++; 1327 while (row < i) { 1328 pc = rtmp + 25*row; 1329 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1330 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1331 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1332 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; 1333 p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1334 p25 = pc[24]; 1335 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1336 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1337 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1338 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || 1339 p20 != 0.0 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || 1340 p24 != 0.0 || p25 != 0.0) { 1341 pv = ba + 25*diag_offset[row]; 1342 pj = bj + diag_offset[row] + 1; 1343 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1344 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1345 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1346 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1347 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; 1348 x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1349 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1350 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1351 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1352 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1353 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1354 1355 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1356 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1357 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1358 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1359 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1360 1361 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1362 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1363 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1364 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1365 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1366 1367 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1368 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1369 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1370 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1371 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1372 1373 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1374 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1375 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1376 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1377 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1378 1379 nz = bi[row+1] - diag_offset[row] - 1; 1380 pv += 25; 1381 for (j=0; j<nz; j++) { 1382 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1383 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1384 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1385 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; 1386 x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; 1387 x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1388 x = rtmp + 25*pj[j]; 1389 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1390 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1391 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1392 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1393 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1394 1395 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1396 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1397 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1398 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1399 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1400 1401 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1402 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1403 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1404 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1405 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1406 1407 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1408 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1409 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1410 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1411 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1412 1413 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1414 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1415 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1416 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1417 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1418 1419 pv += 25; 1420 } 1421 PLogFlops(250*nz+225); 1422 } 1423 row = *ajtmp++; 1424 } 1425 /* finished row so stick it into b->a */ 1426 pv = ba + 25*bi[i]; 1427 pj = bj + bi[i]; 1428 nz = bi[i+1] - bi[i]; 1429 for (j=0; j<nz; j++) { 1430 x = rtmp+25*pj[j]; 1431 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1432 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1433 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1434 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; 1435 pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; 1436 pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; pv[24] = x[24]; 1437 pv += 25; 1438 } 1439 /* invert diagonal block */ 1440 w = ba + 25*diag_offset[i]; 1441 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1442 } 1443 1444 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1445 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1446 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1447 C->factor = FACTOR_LU; 1448 C->assembled = PETSC_TRUE; 1449 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1450 PetscFunctionReturn(0); 1451 } 1452 /* 1453 Version for when blocks are 5 by 5 Using natural ordering 1454 */ 1455 #undef __FUNC__ 1456 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering" 1457 int MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering(Mat A,Mat *B) 1458 { 1459 Mat C = *B; 1460 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1461 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1462 int *ajtmpold,*ajtmp,nz,row; 1463 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1464 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1465 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1466 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1467 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1468 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1469 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1470 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1471 MatScalar *ba = b->a,*aa = a->a; 1472 1473 PetscFunctionBegin; 1474 rtmp = (MatScalar*)PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1475 for (i=0; i<n; i++) { 1476 nz = bi[i+1] - bi[i]; 1477 ajtmp = bj + bi[i]; 1478 for (j=0; j<nz; j++) { 1479 x = rtmp+25*ajtmp[j]; 1480 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1481 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1482 x[16] = x[17] = x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1483 } 1484 /* load in initial (unfactored row) */ 1485 nz = ai[i+1] - ai[i]; 1486 ajtmpold = aj + ai[i]; 1487 v = aa + 25*ai[i]; 1488 for (j=0; j<nz; j++) { 1489 x = rtmp+25*ajtmpold[j]; 1490 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1491 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1492 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1493 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; 1494 x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1495 x[24] = v[24]; 1496 v += 25; 1497 } 1498 row = *ajtmp++; 1499 while (row < i) { 1500 pc = rtmp + 25*row; 1501 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1502 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1503 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1504 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; 1505 p19 = pc[18]; p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; 1506 p24 = pc[23]; p25 = pc[24]; 1507 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1508 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1509 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1510 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 1511 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || p25 != 0.0) { 1512 pv = ba + 25*diag_offset[row]; 1513 pj = bj + diag_offset[row] + 1; 1514 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1515 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1516 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1517 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; 1518 x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1519 x25 = pv[24]; 1520 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1521 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1522 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1523 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1524 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1525 1526 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1527 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1528 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1529 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1530 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1531 1532 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1533 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1534 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1535 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1536 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1537 1538 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1539 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1540 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1541 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1542 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1543 1544 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1545 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1546 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1547 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1548 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1549 1550 nz = bi[row+1] - diag_offset[row] - 1; 1551 pv += 25; 1552 for (j=0; j<nz; j++) { 1553 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1554 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1555 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1556 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1557 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; 1558 x24 = pv[23]; x25 = pv[24]; 1559 x = rtmp + 25*pj[j]; 1560 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1561 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1562 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1563 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1564 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1565 1566 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1567 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1568 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1569 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1570 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1571 1572 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1573 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1574 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1575 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1576 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1577 1578 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1579 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1580 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1581 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1582 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1583 1584 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1585 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1586 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1587 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1588 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1589 pv += 25; 1590 } 1591 PLogFlops(250*nz+225); 1592 } 1593 row = *ajtmp++; 1594 } 1595 /* finished row so stick it into b->a */ 1596 pv = ba + 25*bi[i]; 1597 pj = bj + bi[i]; 1598 nz = bi[i+1] - bi[i]; 1599 for (j=0; j<nz; j++) { 1600 x = rtmp+25*pj[j]; 1601 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1602 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1603 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1604 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; pv[17] = x[17]; 1605 pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; 1606 pv[23] = x[23]; pv[24] = x[24]; 1607 pv += 25; 1608 } 1609 /* invert diagonal block */ 1610 w = ba + 25*diag_offset[i]; 1611 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1612 } 1613 1614 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1615 C->factor = FACTOR_LU; 1616 C->assembled = PETSC_TRUE; 1617 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1618 PetscFunctionReturn(0); 1619 } 1620 1621 /* 1622 Version for when blocks are 4 by 4 1623 */ 1624 #undef __FUNC__ 1625 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_4" 1626 int MatCholeskyFactorNumeric_SeqSBAIJ_4(Mat A,Mat *B) 1627 { 1628 Mat C = *B; 1629 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1630 IS isrow = b->row,isicol = b->icol; 1631 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1632 int *ajtmpold,*ajtmp,nz,row; 1633 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 1634 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1635 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1636 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1637 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1638 MatScalar m13,m14,m15,m16; 1639 MatScalar *ba = b->a,*aa = a->a; 1640 1641 PetscFunctionBegin; 1642 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1643 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1644 rtmp = (MatScalar*)PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1645 1646 for (i=0; i<n; i++) { 1647 nz = bi[i+1] - bi[i]; 1648 ajtmp = bj + bi[i]; 1649 for (j=0; j<nz; j++) { 1650 x = rtmp+16*ajtmp[j]; 1651 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1652 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1653 } 1654 /* load in initial (unfactored row) */ 1655 idx = r[i]; 1656 nz = ai[idx+1] - ai[idx]; 1657 ajtmpold = aj + ai[idx]; 1658 v = aa + 16*ai[idx]; 1659 for (j=0; j<nz; j++) { 1660 x = rtmp+16*ic[ajtmpold[j]]; 1661 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1662 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1663 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1664 x[14] = v[14]; x[15] = v[15]; 1665 v += 16; 1666 } 1667 row = *ajtmp++; 1668 while (row < i) { 1669 pc = rtmp + 16*row; 1670 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1671 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1672 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1673 p15 = pc[14]; p16 = pc[15]; 1674 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1675 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1676 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1677 || p16 != 0.0) { 1678 pv = ba + 16*diag_offset[row]; 1679 pj = bj + diag_offset[row] + 1; 1680 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1681 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1682 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1683 x15 = pv[14]; x16 = pv[15]; 1684 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1685 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1686 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1687 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1688 1689 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1690 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1691 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1692 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1693 1694 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1695 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1696 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1697 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1698 1699 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1700 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1701 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1702 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1703 1704 nz = bi[row+1] - diag_offset[row] - 1; 1705 pv += 16; 1706 for (j=0; j<nz; j++) { 1707 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1708 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1709 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1710 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1711 x = rtmp + 16*pj[j]; 1712 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1713 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1714 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1715 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1716 1717 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1718 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1719 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1720 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1721 1722 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1723 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1724 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1725 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1726 1727 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1728 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1729 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1730 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1731 1732 pv += 16; 1733 } 1734 PLogFlops(128*nz+112); 1735 } 1736 row = *ajtmp++; 1737 } 1738 /* finished row so stick it into b->a */ 1739 pv = ba + 16*bi[i]; 1740 pj = bj + bi[i]; 1741 nz = bi[i+1] - bi[i]; 1742 for (j=0; j<nz; j++) { 1743 x = rtmp+16*pj[j]; 1744 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1745 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1746 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1747 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1748 pv += 16; 1749 } 1750 /* invert diagonal block */ 1751 w = ba + 16*diag_offset[i]; 1752 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1753 } 1754 1755 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1756 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1757 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1758 C->factor = FACTOR_LU; 1759 C->assembled = PETSC_TRUE; 1760 C->preallocated = PETSC_TRUE; 1761 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1762 PetscFunctionReturn(0); 1763 } 1764 /* 1765 Version for when blocks are 4 by 4 Using natural ordering 1766 */ 1767 #undef __FUNC__ 1768 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering" 1769 int MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering(Mat A,Mat *B) 1770 { 1771 Mat C = *B; 1772 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1773 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1774 int *ajtmpold,*ajtmp,nz,row; 1775 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1776 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1777 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1778 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1779 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1780 MatScalar m13,m14,m15,m16; 1781 MatScalar *ba = b->a,*aa = a->a; 1782 1783 PetscFunctionBegin; 1784 rtmp = (MatScalar*)PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1785 1786 for (i=0; i<n; i++) { 1787 nz = bi[i+1] - bi[i]; 1788 ajtmp = bj + bi[i]; 1789 for (j=0; j<nz; j++) { 1790 x = rtmp+16*ajtmp[j]; 1791 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1792 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1793 } 1794 /* load in initial (unfactored row) */ 1795 nz = ai[i+1] - ai[i]; 1796 ajtmpold = aj + ai[i]; 1797 v = aa + 16*ai[i]; 1798 for (j=0; j<nz; j++) { 1799 x = rtmp+16*ajtmpold[j]; 1800 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1801 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1802 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1803 x[14] = v[14]; x[15] = v[15]; 1804 v += 16; 1805 } 1806 row = *ajtmp++; 1807 while (row < i) { 1808 pc = rtmp + 16*row; 1809 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1810 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1811 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1812 p15 = pc[14]; p16 = pc[15]; 1813 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1814 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1815 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1816 || p16 != 0.0) { 1817 pv = ba + 16*diag_offset[row]; 1818 pj = bj + diag_offset[row] + 1; 1819 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1820 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1821 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1822 x15 = pv[14]; x16 = pv[15]; 1823 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1824 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1825 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1826 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1827 1828 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1829 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1830 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1831 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1832 1833 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1834 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1835 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1836 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1837 1838 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1839 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1840 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1841 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1842 1843 nz = bi[row+1] - diag_offset[row] - 1; 1844 pv += 16; 1845 for (j=0; j<nz; j++) { 1846 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1847 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1848 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1849 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1850 x = rtmp + 16*pj[j]; 1851 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1852 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1853 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1854 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1855 1856 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1857 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1858 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1859 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1860 1861 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1862 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1863 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1864 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1865 1866 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1867 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1868 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1869 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1870 1871 pv += 16; 1872 } 1873 PLogFlops(128*nz+112); 1874 } 1875 row = *ajtmp++; 1876 } 1877 /* finished row so stick it into b->a */ 1878 pv = ba + 16*bi[i]; 1879 pj = bj + bi[i]; 1880 nz = bi[i+1] - bi[i]; 1881 for (j=0; j<nz; j++) { 1882 x = rtmp+16*pj[j]; 1883 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1884 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1885 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1886 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1887 pv += 16; 1888 } 1889 /* invert diagonal block */ 1890 w = ba + 16*diag_offset[i]; 1891 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1892 } 1893 1894 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1895 C->factor = FACTOR_LU; 1896 C->assembled = PETSC_TRUE; 1897 C->preallocated = PETSC_TRUE; 1898 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1899 PetscFunctionReturn(0); 1900 } 1901 1902 /* Version for when blocks are 3 by 3 */ 1903 #undef __FUNC__ 1904 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_3" 1905 int MatCholeskyFactorNumeric_SeqSBAIJ_3(Mat A,Mat *B) 1906 { 1907 Mat C = *B; 1908 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 1909 IS perm = b->row; 1910 int *perm_ptr,ierr,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 1911 int *ai,*aj,*a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 1912 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 1913 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 1914 1915 PetscFunctionBegin; 1916 /* initialization */ 1917 rtmp = (MatScalar*)PetscMalloc(9*mbs*sizeof(MatScalar));CHKPTRQ(rtmp); 1918 ierr = PetscMemzero(rtmp,9*mbs*sizeof(MatScalar));CHKERRQ(ierr); 1919 il = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(il); 1920 jl = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(jl); 1921 for (i=0; i<mbs; i++) { 1922 jl[i] = mbs; il[0] = 0; 1923 } 1924 dk = (MatScalar*)PetscMalloc(9*sizeof(MatScalar));CHKPTRQ(dk); 1925 uik = (MatScalar*)PetscMalloc(9*sizeof(MatScalar));CHKPTRQ(uik); 1926 ierr = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr); 1927 1928 /* check permutation */ 1929 if (!a->permute){ 1930 ai = a->i; aj = a->j; aa = a->a; 1931 } else { 1932 ai = a->inew; aj = a->jnew; 1933 aa = (MatScalar*)PetscMalloc(9*ai[mbs]*sizeof(MatScalar));CHKPTRQ(aa); 1934 ierr = PetscMemcpy(aa,a->a,9*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr); 1935 a2anew = (int*)PetscMalloc(ai[mbs]*sizeof(int));CHKPTRQ(a2anew); 1936 ierr= PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(int));CHKERRQ(ierr); 1937 1938 for (i=0; i<mbs; i++){ 1939 jmin = ai[i]; jmax = ai[i+1]; 1940 for (j=jmin; j<jmax; j++){ 1941 while (a2anew[j] != j){ 1942 k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k; 1943 for (k1=0; k1<9; k1++){ 1944 dk[k1] = aa[k*9+k1]; 1945 aa[k*9+k1] = aa[j*9+k1]; 1946 aa[j*9+k1] = dk[k1]; 1947 } 1948 } 1949 /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */ 1950 if (i > aj[j]){ 1951 /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */ 1952 ap = aa + j*9; /* ptr to the beginning of j-th block of aa */ 1953 for (k=0; k<9; k++) dk[k] = ap[k]; /* dk <- j-th block of aa */ 1954 for (k=0; k<3; k++){ /* j-th block of aa <- dk^T */ 1955 for (k1=0; k1<3; k1++) *ap++ = dk[k + 3*k1]; 1956 } 1957 } 1958 } 1959 } 1960 ierr = PetscFree(a2anew);CHKERRA(ierr); 1961 } 1962 1963 /* for each row k */ 1964 for (k = 0; k<mbs; k++){ 1965 1966 /*initialize k-th row with elements nonzero in row perm(k) of A */ 1967 jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1]; 1968 if (jmin < jmax) { 1969 ap = aa + jmin*9; 1970 for (j = jmin; j < jmax; j++){ 1971 vj = perm_ptr[aj[j]]; /* block col. index */ 1972 rtmp_ptr = rtmp + vj*9; 1973 for (i=0; i<9; i++) *rtmp_ptr++ = *ap++; 1974 } 1975 } 1976 1977 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 1978 ierr = PetscMemcpy(dk,rtmp+k*9,9*sizeof(MatScalar));CHKERRQ(ierr); 1979 i = jl[k]; /* first row to be added to k_th row */ 1980 1981 while (i < mbs){ 1982 nexti = jl[i]; /* next row to be added to k_th row */ 1983 1984 /* compute multiplier */ 1985 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1986 1987 /* uik = -inv(Di)*U_bar(i,k) */ 1988 diag = ba + i*9; 1989 u = ba + ili*9; 1990 1991 uik[0] = -(diag[0]*u[0] + diag[3]*u[1] + diag[6]*u[2]); 1992 uik[1] = -(diag[1]*u[0] + diag[4]*u[1] + diag[7]*u[2]); 1993 uik[2] = -(diag[2]*u[0] + diag[5]*u[1] + diag[8]*u[2]); 1994 1995 uik[3] = -(diag[0]*u[3] + diag[3]*u[4] + diag[6]*u[5]); 1996 uik[4] = -(diag[1]*u[3] + diag[4]*u[4] + diag[7]*u[5]); 1997 uik[5] = -(diag[2]*u[3] + diag[5]*u[4] + diag[8]*u[5]); 1998 1999 uik[6] = -(diag[0]*u[6] + diag[3]*u[7] + diag[6]*u[8]); 2000 uik[7] = -(diag[1]*u[6] + diag[4]*u[7] + diag[7]*u[8]); 2001 uik[8] = -(diag[2]*u[6] + diag[5]*u[7] + diag[8]*u[8]); 2002 2003 /* update D(k) += -U(i,k)^T * U_bar(i,k) */ 2004 dk[0] += uik[0]*u[0] + uik[1]*u[1] + uik[2]*u[2]; 2005 dk[1] += uik[3]*u[0] + uik[4]*u[1] + uik[5]*u[2]; 2006 dk[2] += uik[6]*u[0] + uik[7]*u[1] + uik[8]*u[2]; 2007 2008 dk[3] += uik[0]*u[3] + uik[1]*u[4] + uik[2]*u[5]; 2009 dk[4] += uik[3]*u[3] + uik[4]*u[4] + uik[5]*u[5]; 2010 dk[5] += uik[6]*u[3] + uik[7]*u[4] + uik[8]*u[5]; 2011 2012 dk[6] += uik[0]*u[6] + uik[1]*u[7] + uik[2]*u[8]; 2013 dk[7] += uik[3]*u[6] + uik[4]*u[7] + uik[5]*u[8]; 2014 dk[8] += uik[6]*u[6] + uik[7]*u[7] + uik[8]*u[8]; 2015 2016 /* update -U(i,k) */ 2017 ierr = PetscMemcpy(ba+ili*9,uik,9*sizeof(MatScalar));CHKERRQ(ierr); 2018 2019 /* add multiple of row i to k-th row ... */ 2020 jmin = ili + 1; jmax = bi[i+1]; 2021 if (jmin < jmax){ 2022 for (j=jmin; j<jmax; j++) { 2023 /* rtmp += -U(i,k)^T * U_bar(i,j) */ 2024 rtmp_ptr = rtmp + bj[j]*9; 2025 u = ba + j*9; 2026 rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1] + uik[2]*u[2]; 2027 rtmp_ptr[1] += uik[3]*u[0] + uik[4]*u[1] + uik[5]*u[2]; 2028 rtmp_ptr[2] += uik[6]*u[0] + uik[7]*u[1] + uik[8]*u[2]; 2029 2030 rtmp_ptr[3] += uik[0]*u[3] + uik[1]*u[4] + uik[2]*u[5]; 2031 rtmp_ptr[4] += uik[3]*u[3] + uik[4]*u[4] + uik[5]*u[5]; 2032 rtmp_ptr[5] += uik[6]*u[3] + uik[7]*u[4] + uik[8]*u[5]; 2033 2034 rtmp_ptr[6] += uik[0]*u[6] + uik[1]*u[7] + uik[2]*u[8]; 2035 rtmp_ptr[7] += uik[3]*u[6] + uik[4]*u[7] + uik[5]*u[8]; 2036 rtmp_ptr[8] += uik[6]*u[6] + uik[7]*u[7] + uik[8]*u[8]; 2037 } 2038 2039 /* ... add i to row list for next nonzero entry */ 2040 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 2041 j = bj[jmin]; 2042 jl[i] = jl[j]; jl[j] = i; /* update jl */ 2043 } 2044 i = nexti; 2045 } 2046 2047 /* save nonzero entries in k-th row of U ... */ 2048 2049 /* invert diagonal block */ 2050 diag = ba+k*9; 2051 ierr = PetscMemcpy(diag,dk,9*sizeof(MatScalar));CHKERRQ(ierr); 2052 ierr = Kernel_A_gets_inverse_A_3(diag);CHKERRQ(ierr); 2053 2054 jmin = bi[k]; jmax = bi[k+1]; 2055 if (jmin < jmax) { 2056 for (j=jmin; j<jmax; j++){ 2057 vj = bj[j]; /* block col. index of U */ 2058 u = ba + j*9; 2059 rtmp_ptr = rtmp + vj*9; 2060 for (k1=0; k1<9; k1++){ 2061 *u++ = *rtmp_ptr; 2062 *rtmp_ptr++ = 0.0; 2063 } 2064 } 2065 2066 /* ... add k to row list for first nonzero entry in k-th row */ 2067 il[k] = jmin; 2068 i = bj[jmin]; 2069 jl[k] = jl[i]; jl[i] = k; 2070 } 2071 } 2072 2073 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2074 ierr = PetscFree(il);CHKERRQ(ierr); 2075 ierr = PetscFree(jl);CHKERRQ(ierr); 2076 ierr = PetscFree(dk);CHKERRQ(ierr); 2077 ierr = PetscFree(uik);CHKERRQ(ierr); 2078 if (a->permute){ 2079 ierr = PetscFree(aa);CHKERRQ(ierr); 2080 } 2081 2082 ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr); 2083 C->factor = FACTOR_CHOLESKY; 2084 C->assembled = PETSC_TRUE; 2085 C->preallocated = PETSC_TRUE; 2086 PLogFlops(b->mbs); 2087 PetscFunctionReturn(0); 2088 } 2089 2090 /* 2091 Version for when blocks are 3 by 3 Using natural ordering 2092 */ 2093 #undef __FUNC__ 2094 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering" 2095 int MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering(Mat A,Mat *B) 2096 { 2097 Mat C = *B; 2098 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 2099 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 2100 int *ajtmpold,*ajtmp,nz,row; 2101 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 2102 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2103 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 2104 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 2105 MatScalar *ba = b->a,*aa = a->a; 2106 2107 PetscFunctionBegin; 2108 rtmp = (MatScalar*)PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2109 2110 for (i=0; i<n; i++) { 2111 nz = bi[i+1] - bi[i]; 2112 ajtmp = bj + bi[i]; 2113 for (j=0; j<nz; j++) { 2114 x = rtmp+9*ajtmp[j]; 2115 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 2116 } 2117 /* load in initial (unfactored row) */ 2118 nz = ai[i+1] - ai[i]; 2119 ajtmpold = aj + ai[i]; 2120 v = aa + 9*ai[i]; 2121 for (j=0; j<nz; j++) { 2122 x = rtmp+9*ajtmpold[j]; 2123 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2124 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 2125 v += 9; 2126 } 2127 row = *ajtmp++; 2128 while (row < i) { 2129 pc = rtmp + 9*row; 2130 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2131 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 2132 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 2133 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 2134 pv = ba + 9*diag_offset[row]; 2135 pj = bj + diag_offset[row] + 1; 2136 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2137 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2138 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 2139 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 2140 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 2141 2142 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 2143 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 2144 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 2145 2146 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 2147 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 2148 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 2149 2150 nz = bi[row+1] - diag_offset[row] - 1; 2151 pv += 9; 2152 for (j=0; j<nz; j++) { 2153 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2154 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2155 x = rtmp + 9*pj[j]; 2156 x[0] -= m1*x1 + m4*x2 + m7*x3; 2157 x[1] -= m2*x1 + m5*x2 + m8*x3; 2158 x[2] -= m3*x1 + m6*x2 + m9*x3; 2159 2160 x[3] -= m1*x4 + m4*x5 + m7*x6; 2161 x[4] -= m2*x4 + m5*x5 + m8*x6; 2162 x[5] -= m3*x4 + m6*x5 + m9*x6; 2163 2164 x[6] -= m1*x7 + m4*x8 + m7*x9; 2165 x[7] -= m2*x7 + m5*x8 + m8*x9; 2166 x[8] -= m3*x7 + m6*x8 + m9*x9; 2167 pv += 9; 2168 } 2169 PLogFlops(54*nz+36); 2170 } 2171 row = *ajtmp++; 2172 } 2173 /* finished row so stick it into b->a */ 2174 pv = ba + 9*bi[i]; 2175 pj = bj + bi[i]; 2176 nz = bi[i+1] - bi[i]; 2177 for (j=0; j<nz; j++) { 2178 x = rtmp+9*pj[j]; 2179 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2180 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 2181 pv += 9; 2182 } 2183 /* invert diagonal block */ 2184 w = ba + 9*diag_offset[i]; 2185 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 2186 } 2187 2188 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2189 C->factor = FACTOR_LU; 2190 C->assembled = PETSC_TRUE; 2191 C->preallocated = PETSC_TRUE; 2192 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 2193 PetscFunctionReturn(0); 2194 } 2195 2196 /* 2197 Numeric U^T*D*U factorization for SBAIJ format. Modified from SNF of YSMP. 2198 Version for blocks 2 by 2. 2199 */ 2200 #undef __FUNC__ 2201 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_2" 2202 int MatCholeskyFactorNumeric_SeqSBAIJ_2(Mat A,Mat *B) 2203 { 2204 Mat C = *B; 2205 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 2206 IS perm = b->row; 2207 int *perm_ptr,ierr,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 2208 int *ai,*aj,*a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 2209 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 2210 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 2211 2212 PetscFunctionBegin; 2213 2214 /* initialization */ 2215 /* il and jl record the first nonzero element in each row of the accessing 2216 window U(0:k, k:mbs-1). 2217 jl: list of rows to be added to uneliminated rows 2218 i>= k: jl(i) is the first row to be added to row i 2219 i< k: jl(i) is the row following row i in some list of rows 2220 jl(i) = mbs indicates the end of a list 2221 il(i): points to the first nonzero element in columns k,...,mbs-1 of 2222 row i of U */ 2223 rtmp = (MatScalar*)PetscMalloc(4*mbs*sizeof(MatScalar));CHKPTRQ(rtmp); 2224 ierr = PetscMemzero(rtmp,4*mbs*sizeof(MatScalar));CHKERRQ(ierr); 2225 il = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(il); 2226 jl = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(jl); 2227 for (i=0; i<mbs; i++) { 2228 jl[i] = mbs; il[0] = 0; 2229 } 2230 dk = (MatScalar*)PetscMalloc(4*sizeof(MatScalar));CHKPTRQ(dk); 2231 uik = (MatScalar*)PetscMalloc(4*sizeof(MatScalar));CHKPTRQ(uik); 2232 ierr = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr); 2233 2234 /* check permutation */ 2235 if (!a->permute){ 2236 ai = a->i; aj = a->j; aa = a->a; 2237 } else { 2238 ai = a->inew; aj = a->jnew; 2239 aa = (MatScalar*)PetscMalloc(4*ai[mbs]*sizeof(MatScalar));CHKPTRQ(aa); 2240 ierr = PetscMemcpy(aa,a->a,4*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr); 2241 a2anew = (int*)PetscMalloc(ai[mbs]*sizeof(int));CHKPTRQ(a2anew); 2242 ierr= PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(int));CHKERRQ(ierr); 2243 2244 for (i=0; i<mbs; i++){ 2245 jmin = ai[i]; jmax = ai[i+1]; 2246 for (j=jmin; j<jmax; j++){ 2247 while (a2anew[j] != j){ 2248 k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k; 2249 for (k1=0; k1<4; k1++){ 2250 dk[k1] = aa[k*4+k1]; 2251 aa[k*4+k1] = aa[j*4+k1]; 2252 aa[j*4+k1] = dk[k1]; 2253 } 2254 } 2255 /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */ 2256 if (i > aj[j]){ 2257 /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */ 2258 ap = aa + j*4; /* ptr to the beginning of the block */ 2259 dk[1] = ap[1]; /* swap ap[1] and ap[2] */ 2260 ap[1] = ap[2]; 2261 ap[2] = dk[1]; 2262 } 2263 } 2264 } 2265 ierr = PetscFree(a2anew);CHKERRA(ierr); 2266 } 2267 2268 /* for each row k */ 2269 for (k = 0; k<mbs; k++){ 2270 2271 /*initialize k-th row with elements nonzero in row perm(k) of A */ 2272 jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1]; 2273 if (jmin < jmax) { 2274 ap = aa + jmin*4; 2275 for (j = jmin; j < jmax; j++){ 2276 vj = perm_ptr[aj[j]]; /* block col. index */ 2277 rtmp_ptr = rtmp + vj*4; 2278 for (i=0; i<4; i++) *rtmp_ptr++ = *ap++; 2279 } 2280 } 2281 2282 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 2283 ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr); 2284 i = jl[k]; /* first row to be added to k_th row */ 2285 2286 while (i < mbs){ 2287 nexti = jl[i]; /* next row to be added to k_th row */ 2288 2289 /* compute multiplier */ 2290 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 2291 2292 /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */ 2293 diag = ba + i*4; 2294 u = ba + ili*4; 2295 uik[0] = -(diag[0]*u[0] + diag[2]*u[1]); 2296 uik[1] = -(diag[1]*u[0] + diag[3]*u[1]); 2297 uik[2] = -(diag[0]*u[2] + diag[2]*u[3]); 2298 uik[3] = -(diag[1]*u[2] + diag[3]*u[3]); 2299 2300 /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */ 2301 dk[0] += uik[0]*u[0] + uik[1]*u[1]; 2302 dk[1] += uik[2]*u[0] + uik[3]*u[1]; 2303 dk[2] += uik[0]*u[2] + uik[1]*u[3]; 2304 dk[3] += uik[2]*u[2] + uik[3]*u[3]; 2305 2306 /* update -U(i,k): ba[ili] = uik */ 2307 ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr); 2308 2309 /* add multiple of row i to k-th row ... */ 2310 jmin = ili + 1; jmax = bi[i+1]; 2311 if (jmin < jmax){ 2312 for (j=jmin; j<jmax; j++) { 2313 /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */ 2314 rtmp_ptr = rtmp + bj[j]*4; 2315 u = ba + j*4; 2316 rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1]; 2317 rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1]; 2318 rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3]; 2319 rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3]; 2320 } 2321 2322 /* ... add i to row list for next nonzero entry */ 2323 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 2324 j = bj[jmin]; 2325 jl[i] = jl[j]; jl[j] = i; /* update jl */ 2326 } 2327 i = nexti; 2328 } 2329 2330 /* save nonzero entries in k-th row of U ... */ 2331 2332 /* invert diagonal block */ 2333 diag = ba+k*4; 2334 ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr); 2335 ierr = Kernel_A_gets_inverse_A_2(diag);CHKERRQ(ierr); 2336 2337 jmin = bi[k]; jmax = bi[k+1]; 2338 if (jmin < jmax) { 2339 for (j=jmin; j<jmax; j++){ 2340 vj = bj[j]; /* block col. index of U */ 2341 u = ba + j*4; 2342 rtmp_ptr = rtmp + vj*4; 2343 for (k1=0; k1<4; k1++){ 2344 *u++ = *rtmp_ptr; 2345 *rtmp_ptr++ = 0.0; 2346 } 2347 } 2348 2349 /* ... add k to row list for first nonzero entry in k-th row */ 2350 il[k] = jmin; 2351 i = bj[jmin]; 2352 jl[k] = jl[i]; jl[i] = k; 2353 } 2354 } 2355 2356 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2357 ierr = PetscFree(il);CHKERRQ(ierr); 2358 ierr = PetscFree(jl);CHKERRQ(ierr); 2359 ierr = PetscFree(dk);CHKERRQ(ierr); 2360 ierr = PetscFree(uik);CHKERRQ(ierr); 2361 if (a->permute){ 2362 ierr = PetscFree(aa);CHKERRQ(ierr); 2363 } 2364 2365 ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr); 2366 C->factor = FACTOR_CHOLESKY; 2367 C->assembled = PETSC_TRUE; 2368 C->preallocated = PETSC_TRUE; 2369 PLogFlops(b->mbs); 2370 PetscFunctionReturn(0); 2371 } 2372 2373 /* 2374 Version for when blocks are 2 by 2 Using natural ordering 2375 */ 2376 #undef __FUNC__ 2377 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering" 2378 int MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering(Mat A,Mat *B) 2379 { 2380 Mat C = *B; 2381 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 2382 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 2383 int *ajtmpold,*ajtmp,nz,row; 2384 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 2385 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2386 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,x1,x2,x3,x4; 2387 MatScalar *ba = b->a,*aa = a->a; 2388 2389 PetscFunctionBegin; 2390 rtmp = (MatScalar*)PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2391 2392 for (i=0; i<n; i++) { 2393 nz = bi[i+1] - bi[i]; 2394 ajtmp = bj + bi[i]; 2395 for (j=0; j<nz; j++) { 2396 x = rtmp+4*ajtmp[j]; 2397 x[0] = x[1] = x[2] = x[3] = 0.0; 2398 } 2399 /* load in initial (unfactored row) */ 2400 nz = ai[i+1] - ai[i]; 2401 ajtmpold = aj + ai[i]; 2402 v = aa + 4*ai[i]; 2403 for (j=0; j<nz; j++) { 2404 x = rtmp+4*ajtmpold[j]; 2405 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2406 v += 4; 2407 } 2408 row = *ajtmp++; 2409 while (row < i) { 2410 pc = rtmp + 4*row; 2411 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2412 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2413 pv = ba + 4*diag_offset[row]; 2414 pj = bj + diag_offset[row] + 1; 2415 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2416 pc[0] = m1 = p1*x1 + p3*x2; 2417 pc[1] = m2 = p2*x1 + p4*x2; 2418 pc[2] = m3 = p1*x3 + p3*x4; 2419 pc[3] = m4 = p2*x3 + p4*x4; 2420 nz = bi[row+1] - diag_offset[row] - 1; 2421 pv += 4; 2422 for (j=0; j<nz; j++) { 2423 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2424 x = rtmp + 4*pj[j]; 2425 x[0] -= m1*x1 + m3*x2; 2426 x[1] -= m2*x1 + m4*x2; 2427 x[2] -= m1*x3 + m3*x4; 2428 x[3] -= m2*x3 + m4*x4; 2429 pv += 4; 2430 } 2431 PLogFlops(16*nz+12); 2432 } 2433 row = *ajtmp++; 2434 } 2435 /* finished row so stick it into b->a */ 2436 pv = ba + 4*bi[i]; 2437 pj = bj + bi[i]; 2438 nz = bi[i+1] - bi[i]; 2439 for (j=0; j<nz; j++) { 2440 x = rtmp+4*pj[j]; 2441 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2442 pv += 4; 2443 } 2444 /* invert diagonal block */ 2445 w = ba + 4*diag_offset[i]; 2446 ierr = Kernel_A_gets_inverse_A_2(w);CHKERRQ(ierr); 2447 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2448 } 2449 2450 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2451 C->factor = FACTOR_LU; 2452 C->assembled = PETSC_TRUE; 2453 C->preallocated = PETSC_TRUE; 2454 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2455 PetscFunctionReturn(0); 2456 } 2457 2458 /* 2459 Numeric U^T*D*U factorization for SBAIJ format. Modified from SNF of YSMP. 2460 Version for blocks are 1 by 1. 2461 */ 2462 #undef __FUNC__ 2463 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_1" 2464 int MatCholeskyFactorNumeric_SeqSBAIJ_1(Mat A,Mat *B) 2465 { 2466 Mat C = *B; 2467 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 2468 IS ip = b->row; 2469 int *rip,ierr,i,j,mbs = a->mbs,*bi = b->i,*bj = b->j; 2470 int *ai,*aj,*r; 2471 MatScalar *rtmp; 2472 MatScalar *ba = b->a,*aa,ak; 2473 MatScalar dk,uikdi; 2474 int k,jmin,jmax,*jl,*il,vj,nexti,ili; 2475 2476 PetscFunctionBegin; 2477 ierr = ISGetIndices(ip,&rip);CHKERRQ(ierr); 2478 if (!a->permute){ 2479 ai = a->i; aj = a->j; aa = a->a; 2480 } else { 2481 ai = a->inew; aj = a->jnew; 2482 aa = (MatScalar*)PetscMalloc(ai[mbs]*sizeof(MatScalar));CHKPTRQ(aa); 2483 ierr = PetscMemcpy(aa,a->a,ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr); 2484 r = (int*)PetscMalloc(ai[mbs]*sizeof(int));CHKPTRQ(r); 2485 ierr= PetscMemcpy(r,a->a2anew,(ai[mbs])*sizeof(int));CHKERRQ(ierr); 2486 2487 jmin = ai[0]; jmax = ai[mbs]; 2488 for (j=jmin; j<jmax; j++){ 2489 while (r[j] != j){ 2490 k = r[j]; r[j] = r[k]; r[k] = k; 2491 ak = aa[k]; aa[k] = aa[j]; aa[j] = ak; 2492 } 2493 } 2494 ierr = PetscFree(r);CHKERRA(ierr); 2495 } 2496 2497 /* initialization */ 2498 /* il and jl record the first nonzero element in each row of the accessing 2499 window U(0:k, k:mbs-1). 2500 jl: list of rows to be added to uneliminated rows 2501 i>= k: jl(i) is the first row to be added to row i 2502 i< k: jl(i) is the row following row i in some list of rows 2503 jl(i) = mbs indicates the end of a list 2504 il(i): points to the first nonzero element in columns k,...,mbs-1 of 2505 row i of U */ 2506 rtmp = (MatScalar*)PetscMalloc(mbs*sizeof(MatScalar));CHKPTRQ(rtmp); 2507 il = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(il); 2508 jl = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(jl); 2509 for (i=0; i<mbs; i++) { 2510 rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0; 2511 } 2512 2513 /* for each row k */ 2514 for (k = 0; k<mbs; k++){ 2515 2516 /*initialize k-th row with elements nonzero in row perm(k) of A */ 2517 jmin = ai[rip[k]]; jmax = ai[rip[k]+1]; 2518 if (jmin < jmax) { 2519 for (j = jmin; j < jmax; j++){ 2520 vj = rip[aj[j]]; 2521 /* if (k <= vj)*/ rtmp[vj] = aa[j]; 2522 } 2523 } 2524 2525 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 2526 dk = rtmp[k]; 2527 i = jl[k]; /* first row to be added to k_th row */ 2528 /* printf(" k=%d, pivot row = %d\n",k,i); */ 2529 2530 while (i < mbs){ 2531 nexti = jl[i]; /* next row to be added to k_th row */ 2532 2533 /* compute multiplier, update D(k) and U(i,k) */ 2534 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 2535 uikdi = - ba[ili]*ba[i]; 2536 dk += uikdi*ba[ili]; 2537 ba[ili] = uikdi; /* -U(i,k) */ 2538 2539 /* add multiple of row i to k-th row ... */ 2540 jmin = ili + 1; jmax = bi[i+1]; 2541 if (jmin < jmax){ 2542 for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j]; 2543 /* ... add i to row list for next nonzero entry */ 2544 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 2545 j = bj[jmin]; 2546 jl[i] = jl[j]; jl[j] = i; /* update jl */ 2547 } 2548 i = nexti; /* printf(" pivot row i=%d\n",i); */ 2549 } 2550 2551 /* check for zero pivot and save diagoanl element */ 2552 if (dk == 0.0){ 2553 SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot"); 2554 }else if (PetscRealPart(dk) < 0){ 2555 ierr = PetscPrintf(PETSC_COMM_SELF,"Negative pivot: d[%d] = %g\n",k,dk); 2556 } 2557 2558 /* save nonzero entries in k-th row of U ... */ 2559 ba[k] = 1.0/dk; 2560 jmin = bi[k]; jmax = bi[k+1]; 2561 if (jmin < jmax) { 2562 for (j=jmin; j<jmax; j++){ 2563 vj = bj[j]; ba[j] = rtmp[vj]; rtmp[vj] = 0.0; 2564 } 2565 /* ... add k to row list for first nonzero entry in k-th row */ 2566 il[k] = jmin; 2567 i = bj[jmin]; 2568 jl[k] = jl[i]; jl[i] = k; 2569 } 2570 } 2571 2572 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2573 ierr = PetscFree(il);CHKERRQ(ierr); 2574 ierr = PetscFree(jl);CHKERRQ(ierr); 2575 if (a->permute){ 2576 ierr = PetscFree(aa);CHKERRQ(ierr); 2577 } 2578 2579 ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr); 2580 C->factor = FACTOR_CHOLESKY; 2581 C->assembled = PETSC_TRUE; 2582 C->preallocated = PETSC_TRUE; 2583 PLogFlops(b->mbs); 2584 #ifdef TEMP 2585 printf("in factnum_1\n"); 2586 for (k=0; k<mbs; k++){ 2587 i = b->i[k+1] - b->i[k]; 2588 printf("\n b->i[%d]: %d, nzk: %d, diag: %g\n",k,b->i[k],i,b->a[k]); 2589 jmin = b->i[k]; jmax = b->i[k+1]; 2590 for (j=jmin; j<jmax; j++){ 2591 printf(" %d %g, ",b->j[j],b->a[j]); 2592 } 2593 } 2594 #endif 2595 PetscFunctionReturn(0); 2596 } 2597 2598 #undef __FUNC__ 2599 #define __FUNC__ "MatCholeskyFactor_SeqSBAIJ" 2600 int MatCholeskyFactor_SeqSBAIJ(Mat A,IS perm,PetscReal f) 2601 { 2602 int ierr; 2603 Mat C; 2604 2605 PetscFunctionBegin; 2606 ierr = MatCholeskyFactorSymbolic(A,perm,f,&C);CHKERRQ(ierr); 2607 ierr = MatCholeskyFactorNumeric(A,&C);CHKERRQ(ierr); 2608 ierr = MatHeaderCopy(A,C);CHKERRQ(ierr); 2609 PetscFunctionReturn(0); 2610 } 2611 2612 2613