1 /* Using Modified Sparse Row (MSR) storage. 2 See page 85, "Iterative Methods ..." by Saad. */ 3 4 /*$Id: sbaijfact.c,v 1.11 2000/08/01 14:14:52 hzhang Exp hzhang $*/ 5 /* 6 Factorization code for SBAIJ format. 7 */ 8 #include "sbaij.h" 9 #include "src/mat/impls/baij/seq/baij.h" 10 #include "src/vec/vecimpl.h" 11 #include "src/inline/ilu.h" 12 13 #undef __FUNC__ 14 #define __FUNC__ "MatCholeskyFactorSymbolic_SeqSBAIJ" 15 int MatCholeskyFactorSymbolic_SeqSBAIJ(Mat A,IS perm,PetscReal f,Mat *B) 16 { 17 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b; 18 IS iperm; 19 int *rip,*riip,ierr,i,mbs = a->mbs,*ai = a->i,*aj = a->j; 20 int *jutmp,bs = a->bs,bs2=a->bs2; 21 int m,nzi,realloc = 0; 22 int *jl,*q,jumin,jmin,jmax,juptr,nzk,qm,*iu,*ju,k,j,vj,umax,maxadd; 23 24 PetscFunctionBegin; 25 PetscValidHeaderSpecific(perm,IS_COOKIE); 26 if (A->M != A->N) SETERRQ(PETSC_ERR_ARG_WRONG,0,"matrix must be square"); 27 ierr = ISInvertPermutation(perm,PETSC_DECIDE,&iperm);CHKERRQ(ierr); 28 ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr); 29 ierr = ISGetIndices(iperm,&riip);CHKERRQ(ierr); 30 31 /* initialization */ 32 /* Don't know how many column pointers are needed so estimate. 33 Use Modified Sparse Row storage for u and ju, see Sasd pp.85 */ 34 iu = (int*)PetscMalloc((mbs+1)*sizeof(int));CHKPTRQ(iu); 35 umax = (int)(f*ai[mbs] + 1); umax += mbs + 1; 36 ju = (int*)PetscMalloc(umax*sizeof(int));CHKPTRQ(ju); 37 iu[0] = mbs+1; 38 juptr = mbs; 39 jl = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(jl); 40 q = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(q); 41 for (i=0; i<mbs; i++){ 42 jl[i] = mbs; q[i] = 0; 43 } 44 45 /* for each row k */ 46 for (k=0; k<mbs; k++){ 47 nzk = 0; /* num. of nz blocks in k-th block row with diagonal block excluded */ 48 q[k] = mbs; 49 /* initialize nonzero structure of k-th row to row rip[k] of A */ 50 jmin = ai[rip[k]]; 51 jmax = ai[rip[k]+1]; 52 for (j=jmin; j<jmax; j++){ 53 vj = riip[aj[j]]; /* col. value */ 54 if(vj > k){ 55 qm = k; 56 do { 57 m = qm; qm = q[m]; 58 } while(qm < vj); 59 if (qm == vj) { 60 printf(" error: duplicate entry in A\n"); break; 61 } 62 nzk++; 63 q[m] = vj; 64 q[vj] = qm; 65 } /* if(vj > k) */ 66 } /* for (j=jmin; j<jmax; j++) */ 67 68 /* modify nonzero structure of k-th row by computing fill-in 69 for each row i to be merged in */ 70 i = k; 71 i = jl[i]; /* next pivot row (== mbs for symbolic factorization) */ 72 /* printf(" next pivot row i=%d\n",i); */ 73 while (i < mbs){ 74 /* merge row i into k-th row */ 75 nzi = iu[i+1] - (iu[i]+1); 76 jmin = iu[i] + 1; jmax = iu[i] + nzi; 77 qm = k; 78 for (j=jmin; j<jmax+1; j++){ 79 vj = ju[j]; 80 do { 81 m = qm; qm = q[m]; 82 } while (qm < vj); 83 if (qm != vj){ 84 nzk++; q[m] = vj; q[vj] = qm; qm = vj; 85 } 86 } 87 i = jl[i]; /* next pivot row */ 88 } 89 90 /* add k to row list for first nonzero element in k-th row */ 91 if (nzk > 0){ 92 i = q[k]; /* col value of first nonzero element in U(k, k+1:mbs-1) */ 93 jl[k] = jl[i]; jl[i] = k; 94 } 95 iu[k+1] = iu[k] + nzk; /* printf(" iu[%d]=%d, umax=%d\n", k+1, iu[k+1],umax);*/ 96 97 /* allocate more space to ju if needed */ 98 if (iu[k+1] > umax) { printf("allocate more space, iu[%d]=%d > umax=%d\n",k+1, iu[k+1],umax); 99 /* estimate how much additional space we will need */ 100 /* use the strategy suggested by David Hysom <hysom@perch-t.icase.edu> */ 101 /* just double the memory each time */ 102 maxadd = umax; 103 if (maxadd < nzk) maxadd = (mbs-k)*(nzk+1)/2; 104 umax += maxadd; 105 106 /* allocate a longer ju */ 107 jutmp = (int*)PetscMalloc(umax*sizeof(int));CHKPTRQ(jutmp); 108 ierr = PetscMemcpy(jutmp,ju,iu[k]*sizeof(int));CHKERRQ(ierr); 109 ierr = PetscFree(ju);CHKERRQ(ierr); 110 ju = jutmp; 111 realloc++; /* count how many times we realloc */ 112 } 113 114 /* save nonzero structure of k-th row in ju */ 115 i=k; 116 jumin = juptr + 1; juptr += nzk; 117 for (j=jumin; j<juptr+1; j++){ 118 i=q[i]; 119 ju[j]=i; 120 } 121 } 122 123 if (ai[mbs] != 0) { 124 PetscReal af = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 125 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Reallocs %d Fill ratio:given %g needed %g\n",realloc,f,af); 126 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Run with -pc_lu_fill %g or use \n",af); 127 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:PCLUSetFill(pc,%g);\n",af); 128 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:for best performance.\n"); 129 } else { 130 PLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Empty matrix.\n"); 131 } 132 133 ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr); 134 ierr = ISRestoreIndices(iperm,&riip);CHKERRQ(ierr); 135 136 ierr = PetscFree(q);CHKERRQ(ierr); 137 ierr = PetscFree(jl);CHKERRQ(ierr); 138 139 /* put together the new matrix */ 140 ierr = MatCreateSeqSBAIJ(A->comm,bs,bs*mbs,bs*mbs,0,PETSC_NULL,B);CHKERRQ(ierr); 141 PLogObjectParent(*B,iperm); 142 b = (Mat_SeqSBAIJ*)(*B)->data; 143 ierr = PetscFree(b->imax);CHKERRQ(ierr); 144 b->singlemalloc = PETSC_FALSE; 145 /* the next line frees the default space generated by the Create() */ 146 ierr = PetscFree(b->a);CHKERRQ(ierr); 147 ierr = PetscFree(b->ilen);CHKERRQ(ierr); 148 b->a = (MatScalar*)PetscMalloc((iu[mbs]+1)*sizeof(MatScalar)*bs2);CHKPTRQ(b->a); 149 b->j = ju; 150 b->i = iu; 151 b->diag = 0; 152 b->ilen = 0; 153 b->imax = 0; 154 b->row = perm; 155 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 156 b->icol = iperm; 157 b->solve_work = (Scalar*)PetscMalloc((bs*mbs+bs)*sizeof(Scalar));CHKPTRQ(b->solve_work); 158 /* In b structure: Free imax, ilen, old a, old j. 159 Allocate idnew, solve_work, new a, new j */ 160 PLogObjectMemory(*B,(iu[mbs]-mbs)*(sizeof(int)+sizeof(MatScalar))); 161 b->s_maxnz = b->s_nz = iu[mbs]; 162 163 (*B)->factor = FACTOR_LU; 164 (*B)->info.factor_mallocs = realloc; 165 (*B)->info.fill_ratio_given = f; 166 if (ai[mbs] != 0) { 167 (*B)->info.fill_ratio_needed = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 168 } else { 169 (*B)->info.fill_ratio_needed = 0.0; 170 } 171 172 173 PetscFunctionReturn(0); 174 } 175 176 #undef __FUNC__ 177 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_N" 178 int MatCholeskyFactorNumeric_SeqSBAIJ_N(Mat A,Mat *B) 179 { 180 Mat C = *B; 181 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 182 IS isrow = b->row,isicol = b->icol; 183 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 184 int *ajtmpold,*ajtmp,nz,row,bslog,*ai=a->i,*aj=a->j,k,flg; 185 int *diag_offset=b->diag,diag,bs=a->bs,bs2 = a->bs2,*v_pivots,*pj; 186 MatScalar *ba = b->a,*aa = a->a,*pv,*v,*rtmp,*multiplier,*v_work,*pc,*w; 187 188 PetscFunctionBegin; 189 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 190 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 191 rtmp = (MatScalar*)PetscMalloc(bs2*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 192 ierr = PetscMemzero(rtmp,bs2*(n+1)*sizeof(MatScalar));CHKERRQ(ierr); 193 /* generate work space needed by dense LU factorization */ 194 v_work = (MatScalar*)PetscMalloc(bs*sizeof(int) + (bs+bs2)*sizeof(MatScalar));CHKPTRQ(v_work); 195 multiplier = v_work + bs; 196 v_pivots = (int*)(multiplier + bs2); 197 198 /* flops in while loop */ 199 bslog = 2*bs*bs2; 200 201 for (i=0; i<n; i++) { 202 nz = bi[i+1] - bi[i]; 203 ajtmp = bj + bi[i]; 204 for (j=0; j<nz; j++) { 205 ierr = PetscMemzero(rtmp+bs2*ajtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 206 } 207 /* load in initial (unfactored row) */ 208 nz = ai[r[i]+1] - ai[r[i]]; 209 ajtmpold = aj + ai[r[i]]; 210 v = aa + bs2*ai[r[i]]; 211 for (j=0; j<nz; j++) { 212 ierr = PetscMemcpy(rtmp+bs2*ic[ajtmpold[j]],v+bs2*j,bs2*sizeof(MatScalar));CHKERRQ(ierr); 213 } 214 row = *ajtmp++; 215 while (row < i) { 216 pc = rtmp + bs2*row; 217 /* if (*pc) { */ 218 for (flg=0,k=0; k<bs2; k++) { if (pc[k]!=0.0) { flg =1; break; }} 219 if (flg) { 220 pv = ba + bs2*diag_offset[row]; 221 pj = bj + diag_offset[row] + 1; 222 Kernel_A_gets_A_times_B(bs,pc,pv,multiplier); 223 nz = bi[row+1] - diag_offset[row] - 1; 224 pv += bs2; 225 for (j=0; j<nz; j++) { 226 Kernel_A_gets_A_minus_B_times_C(bs,rtmp+bs2*pj[j],pc,pv+bs2*j); 227 } 228 PLogFlops(bslog*(nz+1)-bs); 229 } 230 row = *ajtmp++; 231 } 232 /* finished row so stick it into b->a */ 233 pv = ba + bs2*bi[i]; 234 pj = bj + bi[i]; 235 nz = bi[i+1] - bi[i]; 236 for (j=0; j<nz; j++) { 237 ierr = PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 238 } 239 diag = diag_offset[i] - bi[i]; 240 /* invert diagonal block */ 241 w = pv + bs2*diag; 242 Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work); 243 } 244 245 ierr = PetscFree(rtmp);CHKERRQ(ierr); 246 ierr = PetscFree(v_work);CHKERRQ(ierr); 247 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 248 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 249 C->factor = FACTOR_LU; 250 C->assembled = PETSC_TRUE; 251 PLogFlops(1.3333*bs*bs2*b->mbs); /* from inverting diagonal blocks */ 252 PetscFunctionReturn(0); 253 } 254 255 /* 256 Version for when blocks are 7 by 7 257 */ 258 #undef __FUNC__ 259 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_7" 260 int MatCholeskyFactorNumeric_SeqSBAIJ_7(Mat A,Mat *B) 261 { 262 Mat C = *B; 263 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 264 IS isrow = b->row,isicol = b->icol; 265 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 266 int *ajtmpold,*ajtmp,nz,row; 267 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 268 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 269 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 270 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 271 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 272 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 273 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 274 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 275 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 276 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 277 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 278 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 279 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 280 MatScalar *ba = b->a,*aa = a->a; 281 282 PetscFunctionBegin; 283 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 284 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 285 rtmp = (MatScalar*)PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 286 287 for (i=0; i<n; i++) { 288 nz = bi[i+1] - bi[i]; 289 ajtmp = bj + bi[i]; 290 for (j=0; j<nz; j++) { 291 x = rtmp+49*ajtmp[j]; 292 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 293 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 294 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 295 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 296 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 297 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 298 } 299 /* load in initial (unfactored row) */ 300 idx = r[i]; 301 nz = ai[idx+1] - ai[idx]; 302 ajtmpold = aj + ai[idx]; 303 v = aa + 49*ai[idx]; 304 for (j=0; j<nz; j++) { 305 x = rtmp+49*ic[ajtmpold[j]]; 306 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 307 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 308 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 309 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 310 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 311 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 312 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 313 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 314 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 315 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 316 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 317 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 318 x[48] = v[48]; 319 v += 49; 320 } 321 row = *ajtmp++; 322 while (row < i) { 323 pc = rtmp + 49*row; 324 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 325 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 326 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 327 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 328 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 329 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 330 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 331 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 332 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 333 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 334 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 335 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 336 p49 = pc[48]; 337 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 338 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 339 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 340 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 341 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 342 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 343 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 344 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 345 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 346 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 347 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 348 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 349 p49 != 0.0) { 350 pv = ba + 49*diag_offset[row]; 351 pj = bj + diag_offset[row] + 1; 352 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 353 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 354 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 355 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 356 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 357 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 358 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 359 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 360 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 361 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 362 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 363 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 364 x49 = pv[48]; 365 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 366 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 367 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 368 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 369 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 370 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 371 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 372 373 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 374 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 375 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 376 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 377 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 378 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 379 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 380 381 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 382 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 383 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 384 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 385 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 386 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 387 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 388 389 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 390 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 391 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 392 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 393 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 394 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 395 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 396 397 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 398 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 399 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 400 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 401 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 402 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 403 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 404 405 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 406 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 407 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 408 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 409 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 410 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 411 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 412 413 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 414 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 415 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 416 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 417 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 418 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 419 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 420 421 nz = bi[row+1] - diag_offset[row] - 1; 422 pv += 49; 423 for (j=0; j<nz; j++) { 424 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 425 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 426 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 427 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 428 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 429 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 430 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 431 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 432 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 433 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 434 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 435 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 436 x49 = pv[48]; 437 x = rtmp + 49*pj[j]; 438 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 439 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 440 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 441 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 442 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 443 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 444 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 445 446 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 447 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 448 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 449 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 450 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 451 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 452 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 453 454 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 455 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 456 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 457 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 458 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 459 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 460 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 461 462 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 463 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 464 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 465 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 466 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 467 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 468 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 469 470 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 471 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 472 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 473 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 474 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 475 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 476 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 477 478 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 479 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 480 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 481 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 482 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 483 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 484 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 485 486 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 487 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 488 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 489 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 490 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 491 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 492 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 493 pv += 49; 494 } 495 PLogFlops(686*nz+637); 496 } 497 row = *ajtmp++; 498 } 499 /* finished row so stick it into b->a */ 500 pv = ba + 49*bi[i]; 501 pj = bj + bi[i]; 502 nz = bi[i+1] - bi[i]; 503 for (j=0; j<nz; j++) { 504 x = rtmp+49*pj[j]; 505 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 506 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 507 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 508 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 509 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 510 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 511 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 512 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 513 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 514 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 515 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 516 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 517 pv[48] = x[48]; 518 pv += 49; 519 } 520 /* invert diagonal block */ 521 w = ba + 49*diag_offset[i]; 522 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 523 } 524 525 ierr = PetscFree(rtmp);CHKERRQ(ierr); 526 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 527 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 528 C->factor = FACTOR_LU; 529 C->assembled = PETSC_TRUE; 530 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 531 PetscFunctionReturn(0); 532 } 533 534 /* 535 Version for when blocks are 7 by 7 Using natural ordering 536 */ 537 #undef __FUNC__ 538 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering" 539 int MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering(Mat A,Mat *B) 540 { 541 Mat C = *B; 542 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 543 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 544 int *ajtmpold,*ajtmp,nz,row; 545 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 546 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 547 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 548 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 549 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 550 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 551 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 552 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 553 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 554 MatScalar p37,p38,p39,p40,p41,p42,p43,p44,p45,p46,p47,p48,p49; 555 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 556 MatScalar x37,x38,x39,x40,x41,x42,x43,x44,x45,x46,x47,x48,x49; 557 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 558 MatScalar m37,m38,m39,m40,m41,m42,m43,m44,m45,m46,m47,m48,m49; 559 MatScalar *ba = b->a,*aa = a->a; 560 561 PetscFunctionBegin; 562 rtmp = (MatScalar*)PetscMalloc(49*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 563 for (i=0; i<n; i++) { 564 nz = bi[i+1] - bi[i]; 565 ajtmp = bj + bi[i]; 566 for (j=0; j<nz; j++) { 567 x = rtmp+49*ajtmp[j]; 568 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 569 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 570 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 571 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 572 x[34] = x[35] = x[36] = x[37] = x[38] = x[39] = x[40] = x[41] = 0.0 ; 573 x[42] = x[43] = x[44] = x[45] = x[46] = x[47] = x[48] = 0.0 ; 574 } 575 /* load in initial (unfactored row) */ 576 nz = ai[i+1] - ai[i]; 577 ajtmpold = aj + ai[i]; 578 v = aa + 49*ai[i]; 579 for (j=0; j<nz; j++) { 580 x = rtmp+49*ajtmpold[j]; 581 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 582 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 583 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 584 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 585 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 586 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 587 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 588 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 589 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 590 x[36] = v[36]; x[37] = v[37]; x[38] = v[38]; x[39] = v[39]; 591 x[40] = v[40]; x[41] = v[41]; x[42] = v[42]; x[43] = v[43]; 592 x[44] = v[44]; x[45] = v[45]; x[46] = v[46]; x[47] = v[47]; 593 x[48] = v[48]; 594 v += 49; 595 } 596 row = *ajtmp++; 597 while (row < i) { 598 pc = rtmp + 49*row; 599 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 600 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 601 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 602 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 603 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 604 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 605 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 606 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 607 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 608 p37 = pc[36]; p38 = pc[37]; p39 = pc[38]; p40 = pc[39]; 609 p41 = pc[40]; p42 = pc[41]; p43 = pc[42]; p44 = pc[43]; 610 p45 = pc[44]; p46 = pc[45]; p47 = pc[46]; p48 = pc[47]; 611 p49 = pc[48]; 612 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 613 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 614 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 615 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 616 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 617 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 618 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 619 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 620 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0 || 621 p37 != 0.0 || p38 != 0.0 || p39 != 0.0 || p40 != 0.0 || 622 p41 != 0.0 || p42 != 0.0 || p43 != 0.0 || p44 != 0.0 || 623 p45 != 0.0 || p46 != 0.0 || p47 != 0.0 || p48 != 0.0 || 624 p49 != 0.0) { 625 pv = ba + 49*diag_offset[row]; 626 pj = bj + diag_offset[row] + 1; 627 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 628 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 629 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 630 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 631 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 632 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 633 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 634 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 635 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 636 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 637 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 638 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 639 x49 = pv[48]; 640 pc[0] = m1 = p1*x1 + p8*x2 + p15*x3 + p22*x4 + p29*x5 + p36*x6 + p43*x7; 641 pc[1] = m2 = p2*x1 + p9*x2 + p16*x3 + p23*x4 + p30*x5 + p37*x6 + p44*x7; 642 pc[2] = m3 = p3*x1 + p10*x2 + p17*x3 + p24*x4 + p31*x5 + p38*x6 + p45*x7; 643 pc[3] = m4 = p4*x1 + p11*x2 + p18*x3 + p25*x4 + p32*x5 + p39*x6 + p46*x7; 644 pc[4] = m5 = p5*x1 + p12*x2 + p19*x3 + p26*x4 + p33*x5 + p40*x6 + p47*x7; 645 pc[5] = m6 = p6*x1 + p13*x2 + p20*x3 + p27*x4 + p34*x5 + p41*x6 + p48*x7; 646 pc[6] = m7 = p7*x1 + p14*x2 + p21*x3 + p28*x4 + p35*x5 + p42*x6 + p49*x7; 647 648 pc[7] = m8 = p1*x8 + p8*x9 + p15*x10 + p22*x11 + p29*x12 + p36*x13 + p43*x14; 649 pc[8] = m9 = p2*x8 + p9*x9 + p16*x10 + p23*x11 + p30*x12 + p37*x13 + p44*x14; 650 pc[9] = m10 = p3*x8 + p10*x9 + p17*x10 + p24*x11 + p31*x12 + p38*x13 + p45*x14; 651 pc[10] = m11 = p4*x8 + p11*x9 + p18*x10 + p25*x11 + p32*x12 + p39*x13 + p46*x14; 652 pc[11] = m12 = p5*x8 + p12*x9 + p19*x10 + p26*x11 + p33*x12 + p40*x13 + p47*x14; 653 pc[12] = m13 = p6*x8 + p13*x9 + p20*x10 + p27*x11 + p34*x12 + p41*x13 + p48*x14; 654 pc[13] = m14 = p7*x8 + p14*x9 + p21*x10 + p28*x11 + p35*x12 + p42*x13 + p49*x14; 655 656 pc[14] = m15 = p1*x15 + p8*x16 + p15*x17 + p22*x18 + p29*x19 + p36*x20 + p43*x21; 657 pc[15] = m16 = p2*x15 + p9*x16 + p16*x17 + p23*x18 + p30*x19 + p37*x20 + p44*x21; 658 pc[16] = m17 = p3*x15 + p10*x16 + p17*x17 + p24*x18 + p31*x19 + p38*x20 + p45*x21; 659 pc[17] = m18 = p4*x15 + p11*x16 + p18*x17 + p25*x18 + p32*x19 + p39*x20 + p46*x21; 660 pc[18] = m19 = p5*x15 + p12*x16 + p19*x17 + p26*x18 + p33*x19 + p40*x20 + p47*x21; 661 pc[19] = m20 = p6*x15 + p13*x16 + p20*x17 + p27*x18 + p34*x19 + p41*x20 + p48*x21; 662 pc[20] = m21 = p7*x15 + p14*x16 + p21*x17 + p28*x18 + p35*x19 + p42*x20 + p49*x21; 663 664 pc[21] = m22 = p1*x22 + p8*x23 + p15*x24 + p22*x25 + p29*x26 + p36*x27 + p43*x28; 665 pc[22] = m23 = p2*x22 + p9*x23 + p16*x24 + p23*x25 + p30*x26 + p37*x27 + p44*x28; 666 pc[23] = m24 = p3*x22 + p10*x23 + p17*x24 + p24*x25 + p31*x26 + p38*x27 + p45*x28; 667 pc[24] = m25 = p4*x22 + p11*x23 + p18*x24 + p25*x25 + p32*x26 + p39*x27 + p46*x28; 668 pc[25] = m26 = p5*x22 + p12*x23 + p19*x24 + p26*x25 + p33*x26 + p40*x27 + p47*x28; 669 pc[26] = m27 = p6*x22 + p13*x23 + p20*x24 + p27*x25 + p34*x26 + p41*x27 + p48*x28; 670 pc[27] = m28 = p7*x22 + p14*x23 + p21*x24 + p28*x25 + p35*x26 + p42*x27 + p49*x28; 671 672 pc[28] = m29 = p1*x29 + p8*x30 + p15*x31 + p22*x32 + p29*x33 + p36*x34 + p43*x35; 673 pc[29] = m30 = p2*x29 + p9*x30 + p16*x31 + p23*x32 + p30*x33 + p37*x34 + p44*x35; 674 pc[30] = m31 = p3*x29 + p10*x30 + p17*x31 + p24*x32 + p31*x33 + p38*x34 + p45*x35; 675 pc[31] = m32 = p4*x29 + p11*x30 + p18*x31 + p25*x32 + p32*x33 + p39*x34 + p46*x35; 676 pc[32] = m33 = p5*x29 + p12*x30 + p19*x31 + p26*x32 + p33*x33 + p40*x34 + p47*x35; 677 pc[33] = m34 = p6*x29 + p13*x30 + p20*x31 + p27*x32 + p34*x33 + p41*x34 + p48*x35; 678 pc[34] = m35 = p7*x29 + p14*x30 + p21*x31 + p28*x32 + p35*x33 + p42*x34 + p49*x35; 679 680 pc[35] = m36 = p1*x36 + p8*x37 + p15*x38 + p22*x39 + p29*x40 + p36*x41 + p43*x42; 681 pc[36] = m37 = p2*x36 + p9*x37 + p16*x38 + p23*x39 + p30*x40 + p37*x41 + p44*x42; 682 pc[37] = m38 = p3*x36 + p10*x37 + p17*x38 + p24*x39 + p31*x40 + p38*x41 + p45*x42; 683 pc[38] = m39 = p4*x36 + p11*x37 + p18*x38 + p25*x39 + p32*x40 + p39*x41 + p46*x42; 684 pc[39] = m40 = p5*x36 + p12*x37 + p19*x38 + p26*x39 + p33*x40 + p40*x41 + p47*x42; 685 pc[40] = m41 = p6*x36 + p13*x37 + p20*x38 + p27*x39 + p34*x40 + p41*x41 + p48*x42; 686 pc[41] = m42 = p7*x36 + p14*x37 + p21*x38 + p28*x39 + p35*x40 + p42*x41 + p49*x42; 687 688 pc[42] = m43 = p1*x43 + p8*x44 + p15*x45 + p22*x46 + p29*x47 + p36*x48 + p43*x49; 689 pc[43] = m44 = p2*x43 + p9*x44 + p16*x45 + p23*x46 + p30*x47 + p37*x48 + p44*x49; 690 pc[44] = m45 = p3*x43 + p10*x44 + p17*x45 + p24*x46 + p31*x47 + p38*x48 + p45*x49; 691 pc[45] = m46 = p4*x43 + p11*x44 + p18*x45 + p25*x46 + p32*x47 + p39*x48 + p46*x49; 692 pc[46] = m47 = p5*x43 + p12*x44 + p19*x45 + p26*x46 + p33*x47 + p40*x48 + p47*x49; 693 pc[47] = m48 = p6*x43 + p13*x44 + p20*x45 + p27*x46 + p34*x47 + p41*x48 + p48*x49; 694 pc[48] = m49 = p7*x43 + p14*x44 + p21*x45 + p28*x46 + p35*x47 + p42*x48 + p49*x49; 695 696 nz = bi[row+1] - diag_offset[row] - 1; 697 pv += 49; 698 for (j=0; j<nz; j++) { 699 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 700 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 701 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 702 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 703 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 704 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 705 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 706 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 707 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 708 x37 = pv[36]; x38 = pv[37]; x39 = pv[38]; x40 = pv[39]; 709 x41 = pv[40]; x42 = pv[41]; x43 = pv[42]; x44 = pv[43]; 710 x45 = pv[44]; x46 = pv[45]; x47 = pv[46]; x48 = pv[47]; 711 x49 = pv[48]; 712 x = rtmp + 49*pj[j]; 713 x[0] -= m1*x1 + m8*x2 + m15*x3 + m22*x4 + m29*x5 + m36*x6 + m43*x7; 714 x[1] -= m2*x1 + m9*x2 + m16*x3 + m23*x4 + m30*x5 + m37*x6 + m44*x7; 715 x[2] -= m3*x1 + m10*x2 + m17*x3 + m24*x4 + m31*x5 + m38*x6 + m45*x7; 716 x[3] -= m4*x1 + m11*x2 + m18*x3 + m25*x4 + m32*x5 + m39*x6 + m46*x7; 717 x[4] -= m5*x1 + m12*x2 + m19*x3 + m26*x4 + m33*x5 + m40*x6 + m47*x7; 718 x[5] -= m6*x1 + m13*x2 + m20*x3 + m27*x4 + m34*x5 + m41*x6 + m48*x7; 719 x[6] -= m7*x1 + m14*x2 + m21*x3 + m28*x4 + m35*x5 + m42*x6 + m49*x7; 720 721 x[7] -= m1*x8 + m8*x9 + m15*x10 + m22*x11 + m29*x12 + m36*x13 + m43*x14; 722 x[8] -= m2*x8 + m9*x9 + m16*x10 + m23*x11 + m30*x12 + m37*x13 + m44*x14; 723 x[9] -= m3*x8 + m10*x9 + m17*x10 + m24*x11 + m31*x12 + m38*x13 + m45*x14; 724 x[10] -= m4*x8 + m11*x9 + m18*x10 + m25*x11 + m32*x12 + m39*x13 + m46*x14; 725 x[11] -= m5*x8 + m12*x9 + m19*x10 + m26*x11 + m33*x12 + m40*x13 + m47*x14; 726 x[12] -= m6*x8 + m13*x9 + m20*x10 + m27*x11 + m34*x12 + m41*x13 + m48*x14; 727 x[13] -= m7*x8 + m14*x9 + m21*x10 + m28*x11 + m35*x12 + m42*x13 + m49*x14; 728 729 x[14] -= m1*x15 + m8*x16 + m15*x17 + m22*x18 + m29*x19 + m36*x20 + m43*x21; 730 x[15] -= m2*x15 + m9*x16 + m16*x17 + m23*x18 + m30*x19 + m37*x20 + m44*x21; 731 x[16] -= m3*x15 + m10*x16 + m17*x17 + m24*x18 + m31*x19 + m38*x20 + m45*x21; 732 x[17] -= m4*x15 + m11*x16 + m18*x17 + m25*x18 + m32*x19 + m39*x20 + m46*x21; 733 x[18] -= m5*x15 + m12*x16 + m19*x17 + m26*x18 + m33*x19 + m40*x20 + m47*x21; 734 x[19] -= m6*x15 + m13*x16 + m20*x17 + m27*x18 + m34*x19 + m41*x20 + m48*x21; 735 x[20] -= m7*x15 + m14*x16 + m21*x17 + m28*x18 + m35*x19 + m42*x20 + m49*x21; 736 737 x[21] -= m1*x22 + m8*x23 + m15*x24 + m22*x25 + m29*x26 + m36*x27 + m43*x28; 738 x[22] -= m2*x22 + m9*x23 + m16*x24 + m23*x25 + m30*x26 + m37*x27 + m44*x28; 739 x[23] -= m3*x22 + m10*x23 + m17*x24 + m24*x25 + m31*x26 + m38*x27 + m45*x28; 740 x[24] -= m4*x22 + m11*x23 + m18*x24 + m25*x25 + m32*x26 + m39*x27 + m46*x28; 741 x[25] -= m5*x22 + m12*x23 + m19*x24 + m26*x25 + m33*x26 + m40*x27 + m47*x28; 742 x[26] -= m6*x22 + m13*x23 + m20*x24 + m27*x25 + m34*x26 + m41*x27 + m48*x28; 743 x[27] -= m7*x22 + m14*x23 + m21*x24 + m28*x25 + m35*x26 + m42*x27 + m49*x28; 744 745 x[28] -= m1*x29 + m8*x30 + m15*x31 + m22*x32 + m29*x33 + m36*x34 + m43*x35; 746 x[29] -= m2*x29 + m9*x30 + m16*x31 + m23*x32 + m30*x33 + m37*x34 + m44*x35; 747 x[30] -= m3*x29 + m10*x30 + m17*x31 + m24*x32 + m31*x33 + m38*x34 + m45*x35; 748 x[31] -= m4*x29 + m11*x30 + m18*x31 + m25*x32 + m32*x33 + m39*x34 + m46*x35; 749 x[32] -= m5*x29 + m12*x30 + m19*x31 + m26*x32 + m33*x33 + m40*x34 + m47*x35; 750 x[33] -= m6*x29 + m13*x30 + m20*x31 + m27*x32 + m34*x33 + m41*x34 + m48*x35; 751 x[34] -= m7*x29 + m14*x30 + m21*x31 + m28*x32 + m35*x33 + m42*x34 + m49*x35; 752 753 x[35] -= m1*x36 + m8*x37 + m15*x38 + m22*x39 + m29*x40 + m36*x41 + m43*x42; 754 x[36] -= m2*x36 + m9*x37 + m16*x38 + m23*x39 + m30*x40 + m37*x41 + m44*x42; 755 x[37] -= m3*x36 + m10*x37 + m17*x38 + m24*x39 + m31*x40 + m38*x41 + m45*x42; 756 x[38] -= m4*x36 + m11*x37 + m18*x38 + m25*x39 + m32*x40 + m39*x41 + m46*x42; 757 x[39] -= m5*x36 + m12*x37 + m19*x38 + m26*x39 + m33*x40 + m40*x41 + m47*x42; 758 x[40] -= m6*x36 + m13*x37 + m20*x38 + m27*x39 + m34*x40 + m41*x41 + m48*x42; 759 x[41] -= m7*x36 + m14*x37 + m21*x38 + m28*x39 + m35*x40 + m42*x41 + m49*x42; 760 761 x[42] -= m1*x43 + m8*x44 + m15*x45 + m22*x46 + m29*x47 + m36*x48 + m43*x49; 762 x[43] -= m2*x43 + m9*x44 + m16*x45 + m23*x46 + m30*x47 + m37*x48 + m44*x49; 763 x[44] -= m3*x43 + m10*x44 + m17*x45 + m24*x46 + m31*x47 + m38*x48 + m45*x49; 764 x[45] -= m4*x43 + m11*x44 + m18*x45 + m25*x46 + m32*x47 + m39*x48 + m46*x49; 765 x[46] -= m5*x43 + m12*x44 + m19*x45 + m26*x46 + m33*x47 + m40*x48 + m47*x49; 766 x[47] -= m6*x43 + m13*x44 + m20*x45 + m27*x46 + m34*x47 + m41*x48 + m48*x49; 767 x[48] -= m7*x43 + m14*x44 + m21*x45 + m28*x46 + m35*x47 + m42*x48 + m49*x49; 768 pv += 49; 769 } 770 PLogFlops(686*nz+637); 771 } 772 row = *ajtmp++; 773 } 774 /* finished row so stick it into b->a */ 775 pv = ba + 49*bi[i]; 776 pj = bj + bi[i]; 777 nz = bi[i+1] - bi[i]; 778 for (j=0; j<nz; j++) { 779 x = rtmp+49*pj[j]; 780 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 781 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 782 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 783 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 784 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 785 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 786 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 787 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 788 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 789 pv[36] = x[36]; pv[37] = x[37]; pv[38] = x[38]; pv[39] = x[39]; 790 pv[40] = x[40]; pv[41] = x[41]; pv[42] = x[42]; pv[43] = x[43]; 791 pv[44] = x[44]; pv[45] = x[45]; pv[46] = x[46]; pv[47] = x[47]; 792 pv[48] = x[48]; 793 pv += 49; 794 } 795 /* invert diagonal block */ 796 w = ba + 49*diag_offset[i]; 797 ierr = Kernel_A_gets_inverse_A_7(w);CHKERRQ(ierr); 798 } 799 800 ierr = PetscFree(rtmp);CHKERRQ(ierr); 801 C->factor = FACTOR_LU; 802 C->assembled = PETSC_TRUE; 803 PLogFlops(1.3333*343*b->mbs); /* from inverting diagonal blocks */ 804 PetscFunctionReturn(0); 805 } 806 807 /* ------------------------------------------------------------*/ 808 /* 809 Version for when blocks are 6 by 6 810 */ 811 #undef __FUNC__ 812 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_6" 813 int MatCholeskyFactorNumeric_SeqSBAIJ_6(Mat A,Mat *B) 814 { 815 Mat C = *B; 816 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 817 IS isrow = b->row,isicol = b->icol; 818 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 819 int *ajtmpold,*ajtmp,nz,row; 820 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 821 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 822 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 823 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 824 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 825 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 826 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 827 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 828 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 829 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 830 MatScalar *ba = b->a,*aa = a->a; 831 832 PetscFunctionBegin; 833 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 834 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 835 rtmp = (MatScalar*)PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 836 837 for (i=0; i<n; i++) { 838 nz = bi[i+1] - bi[i]; 839 ajtmp = bj + bi[i]; 840 for (j=0; j<nz; j++) { 841 x = rtmp+36*ajtmp[j]; 842 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 843 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 844 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 845 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 846 x[34] = x[35] = 0.0 ; 847 } 848 /* load in initial (unfactored row) */ 849 idx = r[i]; 850 nz = ai[idx+1] - ai[idx]; 851 ajtmpold = aj + ai[idx]; 852 v = aa + 36*ai[idx]; 853 for (j=0; j<nz; j++) { 854 x = rtmp+36*ic[ajtmpold[j]]; 855 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 856 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 857 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 858 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 859 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 860 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 861 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 862 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 863 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 864 v += 36; 865 } 866 row = *ajtmp++; 867 while (row < i) { 868 pc = rtmp + 36*row; 869 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 870 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 871 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 872 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 873 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 874 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 875 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 876 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 877 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 878 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 879 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 880 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 881 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 882 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 883 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 884 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 885 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 886 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 887 pv = ba + 36*diag_offset[row]; 888 pj = bj + diag_offset[row] + 1; 889 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 890 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 891 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 892 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 893 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 894 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 895 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 896 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 897 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 898 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 899 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 900 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 901 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 902 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 903 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 904 905 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 906 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 907 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 908 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 909 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 910 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 911 912 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 913 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 914 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 915 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 916 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 917 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 918 919 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 920 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 921 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 922 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 923 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 924 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 925 926 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 927 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 928 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 929 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 930 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 931 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 932 933 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 934 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 935 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 936 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 937 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 938 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 939 940 nz = bi[row+1] - diag_offset[row] - 1; 941 pv += 36; 942 for (j=0; j<nz; j++) { 943 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 944 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 945 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 946 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 947 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 948 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 949 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 950 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 951 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 952 x = rtmp + 36*pj[j]; 953 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 954 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 955 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 956 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 957 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 958 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 959 960 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 961 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 962 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 963 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 964 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 965 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 966 967 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 968 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 969 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 970 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 971 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 972 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 973 974 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 975 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 976 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 977 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 978 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 979 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 980 981 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 982 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 983 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 984 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 985 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 986 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 987 988 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 989 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 990 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 991 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 992 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 993 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 994 995 pv += 36; 996 } 997 PLogFlops(432*nz+396); 998 } 999 row = *ajtmp++; 1000 } 1001 /* finished row so stick it into b->a */ 1002 pv = ba + 36*bi[i]; 1003 pj = bj + bi[i]; 1004 nz = bi[i+1] - bi[i]; 1005 for (j=0; j<nz; j++) { 1006 x = rtmp+36*pj[j]; 1007 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1008 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1009 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1010 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1011 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1012 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1013 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1014 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1015 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1016 pv += 36; 1017 } 1018 /* invert diagonal block */ 1019 w = ba + 36*diag_offset[i]; 1020 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1021 } 1022 1023 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1024 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1025 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1026 C->factor = FACTOR_LU; 1027 C->assembled = PETSC_TRUE; 1028 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1029 PetscFunctionReturn(0); 1030 } 1031 /* 1032 Version for when blocks are 6 by 6 Using natural ordering 1033 */ 1034 #undef __FUNC__ 1035 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering" 1036 int MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering(Mat A,Mat *B) 1037 { 1038 Mat C = *B; 1039 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1040 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1041 int *ajtmpold,*ajtmp,nz,row; 1042 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1043 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1044 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1045 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1046 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1047 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1048 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1049 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1050 MatScalar p26,p27,p28,p29,p30,p31,p32,p33,p34,p35,p36; 1051 MatScalar x26,x27,x28,x29,x30,x31,x32,x33,x34,x35,x36; 1052 MatScalar m26,m27,m28,m29,m30,m31,m32,m33,m34,m35,m36; 1053 MatScalar *ba = b->a,*aa = a->a; 1054 1055 PetscFunctionBegin; 1056 rtmp = (MatScalar*)PetscMalloc(36*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1057 for (i=0; i<n; i++) { 1058 nz = bi[i+1] - bi[i]; 1059 ajtmp = bj + bi[i]; 1060 for (j=0; j<nz; j++) { 1061 x = rtmp+36*ajtmp[j]; 1062 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1063 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1064 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = x[25] = 0.0 ; 1065 x[26] = x[27] = x[28] = x[29] = x[30] = x[31] = x[32] = x[33] = 0.0 ; 1066 x[34] = x[35] = 0.0 ; 1067 } 1068 /* load in initial (unfactored row) */ 1069 nz = ai[i+1] - ai[i]; 1070 ajtmpold = aj + ai[i]; 1071 v = aa + 36*ai[i]; 1072 for (j=0; j<nz; j++) { 1073 x = rtmp+36*ajtmpold[j]; 1074 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1075 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; 1076 x[8] = v[8]; x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; 1077 x[12] = v[12]; x[13] = v[13]; x[14] = v[14]; x[15] = v[15]; 1078 x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; x[19] = v[19]; 1079 x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1080 x[24] = v[24]; x[25] = v[25]; x[26] = v[26]; x[27] = v[27]; 1081 x[28] = v[28]; x[29] = v[29]; x[30] = v[30]; x[31] = v[31]; 1082 x[32] = v[32]; x[33] = v[33]; x[34] = v[34]; x[35] = v[35]; 1083 v += 36; 1084 } 1085 row = *ajtmp++; 1086 while (row < i) { 1087 pc = rtmp + 36*row; 1088 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1089 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; 1090 p9 = pc[8]; p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; 1091 p13 = pc[12]; p14 = pc[13]; p15 = pc[14]; p16 = pc[15]; 1092 p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; p20 = pc[19]; 1093 p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1094 p25 = pc[24]; p26 = pc[25]; p27 = pc[26]; p28 = pc[27]; 1095 p29 = pc[28]; p30 = pc[29]; p31 = pc[30]; p32 = pc[31]; 1096 p33 = pc[32]; p34 = pc[33]; p35 = pc[34]; p36 = pc[35]; 1097 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || 1098 p5 != 0.0 || p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || 1099 p9 != 0.0 || p10 != 0.0 || p11 != 0.0 || p12 != 0.0 || 1100 p13 != 0.0 || p14 != 0.0 || p15 != 0.0 || p16 != 0.0 || 1101 p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 || 1102 p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || 1103 p25 != 0.0 || p26 != 0.0 || p27 != 0.0 || p28 != 0.0 || 1104 p29 != 0.0 || p30 != 0.0 || p31 != 0.0 || p32 != 0.0 || 1105 p33 != 0.0 || p34 != 0.0 || p35 != 0.0 || p36 != 0.0) { 1106 pv = ba + 36*diag_offset[row]; 1107 pj = bj + diag_offset[row] + 1; 1108 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1109 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1110 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1111 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1112 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1113 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1114 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1115 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1116 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1117 pc[0] = m1 = p1*x1 + p7*x2 + p13*x3 + p19*x4 + p25*x5 + p31*x6; 1118 pc[1] = m2 = p2*x1 + p8*x2 + p14*x3 + p20*x4 + p26*x5 + p32*x6; 1119 pc[2] = m3 = p3*x1 + p9*x2 + p15*x3 + p21*x4 + p27*x5 + p33*x6; 1120 pc[3] = m4 = p4*x1 + p10*x2 + p16*x3 + p22*x4 + p28*x5 + p34*x6; 1121 pc[4] = m5 = p5*x1 + p11*x2 + p17*x3 + p23*x4 + p29*x5 + p35*x6; 1122 pc[5] = m6 = p6*x1 + p12*x2 + p18*x3 + p24*x4 + p30*x5 + p36*x6; 1123 1124 pc[6] = m7 = p1*x7 + p7*x8 + p13*x9 + p19*x10 + p25*x11 + p31*x12; 1125 pc[7] = m8 = p2*x7 + p8*x8 + p14*x9 + p20*x10 + p26*x11 + p32*x12; 1126 pc[8] = m9 = p3*x7 + p9*x8 + p15*x9 + p21*x10 + p27*x11 + p33*x12; 1127 pc[9] = m10 = p4*x7 + p10*x8 + p16*x9 + p22*x10 + p28*x11 + p34*x12; 1128 pc[10] = m11 = p5*x7 + p11*x8 + p17*x9 + p23*x10 + p29*x11 + p35*x12; 1129 pc[11] = m12 = p6*x7 + p12*x8 + p18*x9 + p24*x10 + p30*x11 + p36*x12; 1130 1131 pc[12] = m13 = p1*x13 + p7*x14 + p13*x15 + p19*x16 + p25*x17 + p31*x18; 1132 pc[13] = m14 = p2*x13 + p8*x14 + p14*x15 + p20*x16 + p26*x17 + p32*x18; 1133 pc[14] = m15 = p3*x13 + p9*x14 + p15*x15 + p21*x16 + p27*x17 + p33*x18; 1134 pc[15] = m16 = p4*x13 + p10*x14 + p16*x15 + p22*x16 + p28*x17 + p34*x18; 1135 pc[16] = m17 = p5*x13 + p11*x14 + p17*x15 + p23*x16 + p29*x17 + p35*x18; 1136 pc[17] = m18 = p6*x13 + p12*x14 + p18*x15 + p24*x16 + p30*x17 + p36*x18; 1137 1138 pc[18] = m19 = p1*x19 + p7*x20 + p13*x21 + p19*x22 + p25*x23 + p31*x24; 1139 pc[19] = m20 = p2*x19 + p8*x20 + p14*x21 + p20*x22 + p26*x23 + p32*x24; 1140 pc[20] = m21 = p3*x19 + p9*x20 + p15*x21 + p21*x22 + p27*x23 + p33*x24; 1141 pc[21] = m22 = p4*x19 + p10*x20 + p16*x21 + p22*x22 + p28*x23 + p34*x24; 1142 pc[22] = m23 = p5*x19 + p11*x20 + p17*x21 + p23*x22 + p29*x23 + p35*x24; 1143 pc[23] = m24 = p6*x19 + p12*x20 + p18*x21 + p24*x22 + p30*x23 + p36*x24; 1144 1145 pc[24] = m25 = p1*x25 + p7*x26 + p13*x27 + p19*x28 + p25*x29 + p31*x30; 1146 pc[25] = m26 = p2*x25 + p8*x26 + p14*x27 + p20*x28 + p26*x29 + p32*x30; 1147 pc[26] = m27 = p3*x25 + p9*x26 + p15*x27 + p21*x28 + p27*x29 + p33*x30; 1148 pc[27] = m28 = p4*x25 + p10*x26 + p16*x27 + p22*x28 + p28*x29 + p34*x30; 1149 pc[28] = m29 = p5*x25 + p11*x26 + p17*x27 + p23*x28 + p29*x29 + p35*x30; 1150 pc[29] = m30 = p6*x25 + p12*x26 + p18*x27 + p24*x28 + p30*x29 + p36*x30; 1151 1152 pc[30] = m31 = p1*x31 + p7*x32 + p13*x33 + p19*x34 + p25*x35 + p31*x36; 1153 pc[31] = m32 = p2*x31 + p8*x32 + p14*x33 + p20*x34 + p26*x35 + p32*x36; 1154 pc[32] = m33 = p3*x31 + p9*x32 + p15*x33 + p21*x34 + p27*x35 + p33*x36; 1155 pc[33] = m34 = p4*x31 + p10*x32 + p16*x33 + p22*x34 + p28*x35 + p34*x36; 1156 pc[34] = m35 = p5*x31 + p11*x32 + p17*x33 + p23*x34 + p29*x35 + p35*x36; 1157 pc[35] = m36 = p6*x31 + p12*x32 + p18*x33 + p24*x34 + p30*x35 + p36*x36; 1158 1159 nz = bi[row+1] - diag_offset[row] - 1; 1160 pv += 36; 1161 for (j=0; j<nz; j++) { 1162 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1163 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; 1164 x9 = pv[8]; x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; 1165 x13 = pv[12]; x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1166 x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; 1167 x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1168 x25 = pv[24]; x26 = pv[25]; x27 = pv[26]; x28 = pv[27]; 1169 x29 = pv[28]; x30 = pv[29]; x31 = pv[30]; x32 = pv[31]; 1170 x33 = pv[32]; x34 = pv[33]; x35 = pv[34]; x36 = pv[35]; 1171 x = rtmp + 36*pj[j]; 1172 x[0] -= m1*x1 + m7*x2 + m13*x3 + m19*x4 + m25*x5 + m31*x6; 1173 x[1] -= m2*x1 + m8*x2 + m14*x3 + m20*x4 + m26*x5 + m32*x6; 1174 x[2] -= m3*x1 + m9*x2 + m15*x3 + m21*x4 + m27*x5 + m33*x6; 1175 x[3] -= m4*x1 + m10*x2 + m16*x3 + m22*x4 + m28*x5 + m34*x6; 1176 x[4] -= m5*x1 + m11*x2 + m17*x3 + m23*x4 + m29*x5 + m35*x6; 1177 x[5] -= m6*x1 + m12*x2 + m18*x3 + m24*x4 + m30*x5 + m36*x6; 1178 1179 x[6] -= m1*x7 + m7*x8 + m13*x9 + m19*x10 + m25*x11 + m31*x12; 1180 x[7] -= m2*x7 + m8*x8 + m14*x9 + m20*x10 + m26*x11 + m32*x12; 1181 x[8] -= m3*x7 + m9*x8 + m15*x9 + m21*x10 + m27*x11 + m33*x12; 1182 x[9] -= m4*x7 + m10*x8 + m16*x9 + m22*x10 + m28*x11 + m34*x12; 1183 x[10] -= m5*x7 + m11*x8 + m17*x9 + m23*x10 + m29*x11 + m35*x12; 1184 x[11] -= m6*x7 + m12*x8 + m18*x9 + m24*x10 + m30*x11 + m36*x12; 1185 1186 x[12] -= m1*x13 + m7*x14 + m13*x15 + m19*x16 + m25*x17 + m31*x18; 1187 x[13] -= m2*x13 + m8*x14 + m14*x15 + m20*x16 + m26*x17 + m32*x18; 1188 x[14] -= m3*x13 + m9*x14 + m15*x15 + m21*x16 + m27*x17 + m33*x18; 1189 x[15] -= m4*x13 + m10*x14 + m16*x15 + m22*x16 + m28*x17 + m34*x18; 1190 x[16] -= m5*x13 + m11*x14 + m17*x15 + m23*x16 + m29*x17 + m35*x18; 1191 x[17] -= m6*x13 + m12*x14 + m18*x15 + m24*x16 + m30*x17 + m36*x18; 1192 1193 x[18] -= m1*x19 + m7*x20 + m13*x21 + m19*x22 + m25*x23 + m31*x24; 1194 x[19] -= m2*x19 + m8*x20 + m14*x21 + m20*x22 + m26*x23 + m32*x24; 1195 x[20] -= m3*x19 + m9*x20 + m15*x21 + m21*x22 + m27*x23 + m33*x24; 1196 x[21] -= m4*x19 + m10*x20 + m16*x21 + m22*x22 + m28*x23 + m34*x24; 1197 x[22] -= m5*x19 + m11*x20 + m17*x21 + m23*x22 + m29*x23 + m35*x24; 1198 x[23] -= m6*x19 + m12*x20 + m18*x21 + m24*x22 + m30*x23 + m36*x24; 1199 1200 x[24] -= m1*x25 + m7*x26 + m13*x27 + m19*x28 + m25*x29 + m31*x30; 1201 x[25] -= m2*x25 + m8*x26 + m14*x27 + m20*x28 + m26*x29 + m32*x30; 1202 x[26] -= m3*x25 + m9*x26 + m15*x27 + m21*x28 + m27*x29 + m33*x30; 1203 x[27] -= m4*x25 + m10*x26 + m16*x27 + m22*x28 + m28*x29 + m34*x30; 1204 x[28] -= m5*x25 + m11*x26 + m17*x27 + m23*x28 + m29*x29 + m35*x30; 1205 x[29] -= m6*x25 + m12*x26 + m18*x27 + m24*x28 + m30*x29 + m36*x30; 1206 1207 x[30] -= m1*x31 + m7*x32 + m13*x33 + m19*x34 + m25*x35 + m31*x36; 1208 x[31] -= m2*x31 + m8*x32 + m14*x33 + m20*x34 + m26*x35 + m32*x36; 1209 x[32] -= m3*x31 + m9*x32 + m15*x33 + m21*x34 + m27*x35 + m33*x36; 1210 x[33] -= m4*x31 + m10*x32 + m16*x33 + m22*x34 + m28*x35 + m34*x36; 1211 x[34] -= m5*x31 + m11*x32 + m17*x33 + m23*x34 + m29*x35 + m35*x36; 1212 x[35] -= m6*x31 + m12*x32 + m18*x33 + m24*x34 + m30*x35 + m36*x36; 1213 1214 pv += 36; 1215 } 1216 PLogFlops(432*nz+396); 1217 } 1218 row = *ajtmp++; 1219 } 1220 /* finished row so stick it into b->a */ 1221 pv = ba + 36*bi[i]; 1222 pj = bj + bi[i]; 1223 nz = bi[i+1] - bi[i]; 1224 for (j=0; j<nz; j++) { 1225 x = rtmp+36*pj[j]; 1226 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1227 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; 1228 pv[8] = x[8]; pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; 1229 pv[12] = x[12]; pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1230 pv[16] = x[16]; pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; 1231 pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; 1232 pv[24] = x[24]; pv[25] = x[25]; pv[26] = x[26]; pv[27] = x[27]; 1233 pv[28] = x[28]; pv[29] = x[29]; pv[30] = x[30]; pv[31] = x[31]; 1234 pv[32] = x[32]; pv[33] = x[33]; pv[34] = x[34]; pv[35] = x[35]; 1235 pv += 36; 1236 } 1237 /* invert diagonal block */ 1238 w = ba + 36*diag_offset[i]; 1239 ierr = Kernel_A_gets_inverse_A_6(w);CHKERRQ(ierr); 1240 } 1241 1242 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1243 C->factor = FACTOR_LU; 1244 C->assembled = PETSC_TRUE; 1245 PLogFlops(1.3333*216*b->mbs); /* from inverting diagonal blocks */ 1246 PetscFunctionReturn(0); 1247 } 1248 1249 /* 1250 Version for when blocks are 5 by 5 1251 */ 1252 #undef __FUNC__ 1253 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_5" 1254 int MatCholeskyFactorNumeric_SeqSBAIJ_5(Mat A,Mat *B) 1255 { 1256 Mat C = *B; 1257 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1258 IS isrow = b->row,isicol = b->icol; 1259 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1260 int *ajtmpold,*ajtmp,nz,row; 1261 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 1262 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1263 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1264 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1265 MatScalar x17,x18,x19,x20,x21,x22,x23,x24,x25,p10,p11,p12,p13,p14; 1266 MatScalar p15,p16,p17,p18,p19,p20,p21,p22,p23,p24,p25,m10,m11,m12; 1267 MatScalar m13,m14,m15,m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1268 MatScalar *ba = b->a,*aa = a->a; 1269 1270 PetscFunctionBegin; 1271 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1272 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1273 rtmp = (MatScalar*)PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1274 1275 for (i=0; i<n; i++) { 1276 nz = bi[i+1] - bi[i]; 1277 ajtmp = bj + bi[i]; 1278 for (j=0; j<nz; j++) { 1279 x = rtmp+25*ajtmp[j]; 1280 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1281 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = x[16] = x[17] = 0.0; 1282 x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1283 } 1284 /* load in initial (unfactored row) */ 1285 idx = r[i]; 1286 nz = ai[idx+1] - ai[idx]; 1287 ajtmpold = aj + ai[idx]; 1288 v = aa + 25*ai[idx]; 1289 for (j=0; j<nz; j++) { 1290 x = rtmp+25*ic[ajtmpold[j]]; 1291 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1292 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1293 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1294 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; 1295 x[18] = v[18]; x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; 1296 x[22] = v[22]; x[23] = v[23]; x[24] = v[24]; 1297 v += 25; 1298 } 1299 row = *ajtmp++; 1300 while (row < i) { 1301 pc = rtmp + 25*row; 1302 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1303 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1304 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1305 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; p19 = pc[18]; 1306 p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; p24 = pc[23]; 1307 p25 = pc[24]; 1308 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1309 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1310 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1311 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || 1312 p20 != 0.0 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || 1313 p24 != 0.0 || p25 != 0.0) { 1314 pv = ba + 25*diag_offset[row]; 1315 pj = bj + diag_offset[row] + 1; 1316 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1317 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1318 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1319 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1320 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; 1321 x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1322 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1323 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1324 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1325 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1326 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1327 1328 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1329 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1330 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1331 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1332 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1333 1334 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1335 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1336 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1337 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1338 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1339 1340 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1341 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1342 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1343 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1344 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1345 1346 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1347 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1348 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1349 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1350 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1351 1352 nz = bi[row+1] - diag_offset[row] - 1; 1353 pv += 25; 1354 for (j=0; j<nz; j++) { 1355 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1356 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1357 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1358 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; 1359 x18 = pv[17]; x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; 1360 x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; x25 = pv[24]; 1361 x = rtmp + 25*pj[j]; 1362 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1363 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1364 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1365 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1366 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1367 1368 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1369 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1370 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1371 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1372 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1373 1374 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1375 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1376 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1377 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1378 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1379 1380 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1381 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1382 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1383 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1384 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1385 1386 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1387 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1388 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1389 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1390 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1391 1392 pv += 25; 1393 } 1394 PLogFlops(250*nz+225); 1395 } 1396 row = *ajtmp++; 1397 } 1398 /* finished row so stick it into b->a */ 1399 pv = ba + 25*bi[i]; 1400 pj = bj + bi[i]; 1401 nz = bi[i+1] - bi[i]; 1402 for (j=0; j<nz; j++) { 1403 x = rtmp+25*pj[j]; 1404 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1405 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1406 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1407 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; 1408 pv[17] = x[17]; pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; 1409 pv[21] = x[21]; pv[22] = x[22]; pv[23] = x[23]; pv[24] = x[24]; 1410 pv += 25; 1411 } 1412 /* invert diagonal block */ 1413 w = ba + 25*diag_offset[i]; 1414 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1415 } 1416 1417 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1418 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1419 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1420 C->factor = FACTOR_LU; 1421 C->assembled = PETSC_TRUE; 1422 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1423 PetscFunctionReturn(0); 1424 } 1425 /* 1426 Version for when blocks are 5 by 5 Using natural ordering 1427 */ 1428 #undef __FUNC__ 1429 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering" 1430 int MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering(Mat A,Mat *B) 1431 { 1432 Mat C = *B; 1433 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1434 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1435 int *ajtmpold,*ajtmp,nz,row; 1436 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1437 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1438 MatScalar x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15; 1439 MatScalar x16,x17,x18,x19,x20,x21,x22,x23,x24,x25; 1440 MatScalar p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15; 1441 MatScalar p16,p17,p18,p19,p20,p21,p22,p23,p24,p25; 1442 MatScalar m1,m2,m3,m4,m5,m6,m7,m8,m9,m10,m11,m12,m13,m14,m15; 1443 MatScalar m16,m17,m18,m19,m20,m21,m22,m23,m24,m25; 1444 MatScalar *ba = b->a,*aa = a->a; 1445 1446 PetscFunctionBegin; 1447 rtmp = (MatScalar*)PetscMalloc(25*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1448 for (i=0; i<n; i++) { 1449 nz = bi[i+1] - bi[i]; 1450 ajtmp = bj + bi[i]; 1451 for (j=0; j<nz; j++) { 1452 x = rtmp+25*ajtmp[j]; 1453 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1454 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1455 x[16] = x[17] = x[18] = x[19] = x[20] = x[21] = x[22] = x[23] = x[24] = 0.0; 1456 } 1457 /* load in initial (unfactored row) */ 1458 nz = ai[i+1] - ai[i]; 1459 ajtmpold = aj + ai[i]; 1460 v = aa + 25*ai[i]; 1461 for (j=0; j<nz; j++) { 1462 x = rtmp+25*ajtmpold[j]; 1463 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1464 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1465 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1466 x[14] = v[14]; x[15] = v[15]; x[16] = v[16]; x[17] = v[17]; x[18] = v[18]; 1467 x[19] = v[19]; x[20] = v[20]; x[21] = v[21]; x[22] = v[22]; x[23] = v[23]; 1468 x[24] = v[24]; 1469 v += 25; 1470 } 1471 row = *ajtmp++; 1472 while (row < i) { 1473 pc = rtmp + 25*row; 1474 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1475 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1476 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1477 p15 = pc[14]; p16 = pc[15]; p17 = pc[16]; p18 = pc[17]; 1478 p19 = pc[18]; p20 = pc[19]; p21 = pc[20]; p22 = pc[21]; p23 = pc[22]; 1479 p24 = pc[23]; p25 = pc[24]; 1480 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1481 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1482 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1483 || p16 != 0.0 || p17 != 0.0 || p18 != 0.0 || p19 != 0.0 || p20 != 0.0 1484 || p21 != 0.0 || p22 != 0.0 || p23 != 0.0 || p24 != 0.0 || p25 != 0.0) { 1485 pv = ba + 25*diag_offset[row]; 1486 pj = bj + diag_offset[row] + 1; 1487 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1488 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1489 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1490 x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; x19 = pv[18]; 1491 x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; x24 = pv[23]; 1492 x25 = pv[24]; 1493 pc[0] = m1 = p1*x1 + p6*x2 + p11*x3 + p16*x4 + p21*x5; 1494 pc[1] = m2 = p2*x1 + p7*x2 + p12*x3 + p17*x4 + p22*x5; 1495 pc[2] = m3 = p3*x1 + p8*x2 + p13*x3 + p18*x4 + p23*x5; 1496 pc[3] = m4 = p4*x1 + p9*x2 + p14*x3 + p19*x4 + p24*x5; 1497 pc[4] = m5 = p5*x1 + p10*x2 + p15*x3 + p20*x4 + p25*x5; 1498 1499 pc[5] = m6 = p1*x6 + p6*x7 + p11*x8 + p16*x9 + p21*x10; 1500 pc[6] = m7 = p2*x6 + p7*x7 + p12*x8 + p17*x9 + p22*x10; 1501 pc[7] = m8 = p3*x6 + p8*x7 + p13*x8 + p18*x9 + p23*x10; 1502 pc[8] = m9 = p4*x6 + p9*x7 + p14*x8 + p19*x9 + p24*x10; 1503 pc[9] = m10 = p5*x6 + p10*x7 + p15*x8 + p20*x9 + p25*x10; 1504 1505 pc[10] = m11 = p1*x11 + p6*x12 + p11*x13 + p16*x14 + p21*x15; 1506 pc[11] = m12 = p2*x11 + p7*x12 + p12*x13 + p17*x14 + p22*x15; 1507 pc[12] = m13 = p3*x11 + p8*x12 + p13*x13 + p18*x14 + p23*x15; 1508 pc[13] = m14 = p4*x11 + p9*x12 + p14*x13 + p19*x14 + p24*x15; 1509 pc[14] = m15 = p5*x11 + p10*x12 + p15*x13 + p20*x14 + p25*x15; 1510 1511 pc[15] = m16 = p1*x16 + p6*x17 + p11*x18 + p16*x19 + p21*x20; 1512 pc[16] = m17 = p2*x16 + p7*x17 + p12*x18 + p17*x19 + p22*x20; 1513 pc[17] = m18 = p3*x16 + p8*x17 + p13*x18 + p18*x19 + p23*x20; 1514 pc[18] = m19 = p4*x16 + p9*x17 + p14*x18 + p19*x19 + p24*x20; 1515 pc[19] = m20 = p5*x16 + p10*x17 + p15*x18 + p20*x19 + p25*x20; 1516 1517 pc[20] = m21 = p1*x21 + p6*x22 + p11*x23 + p16*x24 + p21*x25; 1518 pc[21] = m22 = p2*x21 + p7*x22 + p12*x23 + p17*x24 + p22*x25; 1519 pc[22] = m23 = p3*x21 + p8*x22 + p13*x23 + p18*x24 + p23*x25; 1520 pc[23] = m24 = p4*x21 + p9*x22 + p14*x23 + p19*x24 + p24*x25; 1521 pc[24] = m25 = p5*x21 + p10*x22 + p15*x23 + p20*x24 + p25*x25; 1522 1523 nz = bi[row+1] - diag_offset[row] - 1; 1524 pv += 25; 1525 for (j=0; j<nz; j++) { 1526 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1527 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1528 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1529 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; x17 = pv[16]; x18 = pv[17]; 1530 x19 = pv[18]; x20 = pv[19]; x21 = pv[20]; x22 = pv[21]; x23 = pv[22]; 1531 x24 = pv[23]; x25 = pv[24]; 1532 x = rtmp + 25*pj[j]; 1533 x[0] -= m1*x1 + m6*x2 + m11*x3 + m16*x4 + m21*x5; 1534 x[1] -= m2*x1 + m7*x2 + m12*x3 + m17*x4 + m22*x5; 1535 x[2] -= m3*x1 + m8*x2 + m13*x3 + m18*x4 + m23*x5; 1536 x[3] -= m4*x1 + m9*x2 + m14*x3 + m19*x4 + m24*x5; 1537 x[4] -= m5*x1 + m10*x2 + m15*x3 + m20*x4 + m25*x5; 1538 1539 x[5] -= m1*x6 + m6*x7 + m11*x8 + m16*x9 + m21*x10; 1540 x[6] -= m2*x6 + m7*x7 + m12*x8 + m17*x9 + m22*x10; 1541 x[7] -= m3*x6 + m8*x7 + m13*x8 + m18*x9 + m23*x10; 1542 x[8] -= m4*x6 + m9*x7 + m14*x8 + m19*x9 + m24*x10; 1543 x[9] -= m5*x6 + m10*x7 + m15*x8 + m20*x9 + m25*x10; 1544 1545 x[10] -= m1*x11 + m6*x12 + m11*x13 + m16*x14 + m21*x15; 1546 x[11] -= m2*x11 + m7*x12 + m12*x13 + m17*x14 + m22*x15; 1547 x[12] -= m3*x11 + m8*x12 + m13*x13 + m18*x14 + m23*x15; 1548 x[13] -= m4*x11 + m9*x12 + m14*x13 + m19*x14 + m24*x15; 1549 x[14] -= m5*x11 + m10*x12 + m15*x13 + m20*x14 + m25*x15; 1550 1551 x[15] -= m1*x16 + m6*x17 + m11*x18 + m16*x19 + m21*x20; 1552 x[16] -= m2*x16 + m7*x17 + m12*x18 + m17*x19 + m22*x20; 1553 x[17] -= m3*x16 + m8*x17 + m13*x18 + m18*x19 + m23*x20; 1554 x[18] -= m4*x16 + m9*x17 + m14*x18 + m19*x19 + m24*x20; 1555 x[19] -= m5*x16 + m10*x17 + m15*x18 + m20*x19 + m25*x20; 1556 1557 x[20] -= m1*x21 + m6*x22 + m11*x23 + m16*x24 + m21*x25; 1558 x[21] -= m2*x21 + m7*x22 + m12*x23 + m17*x24 + m22*x25; 1559 x[22] -= m3*x21 + m8*x22 + m13*x23 + m18*x24 + m23*x25; 1560 x[23] -= m4*x21 + m9*x22 + m14*x23 + m19*x24 + m24*x25; 1561 x[24] -= m5*x21 + m10*x22 + m15*x23 + m20*x24 + m25*x25; 1562 pv += 25; 1563 } 1564 PLogFlops(250*nz+225); 1565 } 1566 row = *ajtmp++; 1567 } 1568 /* finished row so stick it into b->a */ 1569 pv = ba + 25*bi[i]; 1570 pj = bj + bi[i]; 1571 nz = bi[i+1] - bi[i]; 1572 for (j=0; j<nz; j++) { 1573 x = rtmp+25*pj[j]; 1574 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1575 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1576 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1577 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; pv[16] = x[16]; pv[17] = x[17]; 1578 pv[18] = x[18]; pv[19] = x[19]; pv[20] = x[20]; pv[21] = x[21]; pv[22] = x[22]; 1579 pv[23] = x[23]; pv[24] = x[24]; 1580 pv += 25; 1581 } 1582 /* invert diagonal block */ 1583 w = ba + 25*diag_offset[i]; 1584 ierr = Kernel_A_gets_inverse_A_5(w);CHKERRQ(ierr); 1585 } 1586 1587 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1588 C->factor = FACTOR_LU; 1589 C->assembled = PETSC_TRUE; 1590 PLogFlops(1.3333*125*b->mbs); /* from inverting diagonal blocks */ 1591 PetscFunctionReturn(0); 1592 } 1593 1594 /* 1595 Version for when blocks are 4 by 4 1596 */ 1597 #undef __FUNC__ 1598 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_4" 1599 int MatCholeskyFactorNumeric_SeqSBAIJ_4(Mat A,Mat *B) 1600 { 1601 Mat C = *B; 1602 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1603 IS isrow = b->row,isicol = b->icol; 1604 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1605 int *ajtmpold,*ajtmp,nz,row; 1606 int *diag_offset = b->diag,idx,*ai=a->i,*aj=a->j,*pj; 1607 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1608 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1609 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1610 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1611 MatScalar m13,m14,m15,m16; 1612 MatScalar *ba = b->a,*aa = a->a; 1613 1614 PetscFunctionBegin; 1615 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1616 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1617 rtmp = (MatScalar*)PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1618 1619 for (i=0; i<n; i++) { 1620 nz = bi[i+1] - bi[i]; 1621 ajtmp = bj + bi[i]; 1622 for (j=0; j<nz; j++) { 1623 x = rtmp+16*ajtmp[j]; 1624 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1625 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1626 } 1627 /* load in initial (unfactored row) */ 1628 idx = r[i]; 1629 nz = ai[idx+1] - ai[idx]; 1630 ajtmpold = aj + ai[idx]; 1631 v = aa + 16*ai[idx]; 1632 for (j=0; j<nz; j++) { 1633 x = rtmp+16*ic[ajtmpold[j]]; 1634 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1635 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1636 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1637 x[14] = v[14]; x[15] = v[15]; 1638 v += 16; 1639 } 1640 row = *ajtmp++; 1641 while (row < i) { 1642 pc = rtmp + 16*row; 1643 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1644 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1645 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1646 p15 = pc[14]; p16 = pc[15]; 1647 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1648 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1649 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1650 || p16 != 0.0) { 1651 pv = ba + 16*diag_offset[row]; 1652 pj = bj + diag_offset[row] + 1; 1653 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1654 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1655 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1656 x15 = pv[14]; x16 = pv[15]; 1657 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1658 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1659 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1660 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1661 1662 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1663 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1664 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1665 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1666 1667 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1668 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1669 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1670 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1671 1672 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1673 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1674 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1675 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1676 1677 nz = bi[row+1] - diag_offset[row] - 1; 1678 pv += 16; 1679 for (j=0; j<nz; j++) { 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]; 1683 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1684 x = rtmp + 16*pj[j]; 1685 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1686 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1687 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1688 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1689 1690 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1691 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1692 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1693 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1694 1695 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1696 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1697 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1698 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1699 1700 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1701 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1702 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1703 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1704 1705 pv += 16; 1706 } 1707 PLogFlops(128*nz+112); 1708 } 1709 row = *ajtmp++; 1710 } 1711 /* finished row so stick it into b->a */ 1712 pv = ba + 16*bi[i]; 1713 pj = bj + bi[i]; 1714 nz = bi[i+1] - bi[i]; 1715 for (j=0; j<nz; j++) { 1716 x = rtmp+16*pj[j]; 1717 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1718 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1719 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1720 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1721 pv += 16; 1722 } 1723 /* invert diagonal block */ 1724 w = ba + 16*diag_offset[i]; 1725 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1726 } 1727 1728 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1729 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1730 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1731 C->factor = FACTOR_LU; 1732 C->assembled = PETSC_TRUE; 1733 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1734 PetscFunctionReturn(0); 1735 } 1736 /* 1737 Version for when blocks are 4 by 4 Using natural ordering 1738 */ 1739 #undef __FUNC__ 1740 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering" 1741 int MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering(Mat A,Mat *B) 1742 { 1743 Mat C = *B; 1744 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1745 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1746 int *ajtmpold,*ajtmp,nz,row; 1747 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1748 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1749 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1750 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,x16; 1751 MatScalar p10,p11,p12,p13,p14,p15,p16,m10,m11,m12; 1752 MatScalar m13,m14,m15,m16; 1753 MatScalar *ba = b->a,*aa = a->a; 1754 1755 PetscFunctionBegin; 1756 rtmp = (MatScalar*)PetscMalloc(16*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1757 1758 for (i=0; i<n; i++) { 1759 nz = bi[i+1] - bi[i]; 1760 ajtmp = bj + bi[i]; 1761 for (j=0; j<nz; j++) { 1762 x = rtmp+16*ajtmp[j]; 1763 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = x[9] = 0.0; 1764 x[10] = x[11] = x[12] = x[13] = x[14] = x[15] = 0.0; 1765 } 1766 /* load in initial (unfactored row) */ 1767 nz = ai[i+1] - ai[i]; 1768 ajtmpold = aj + ai[i]; 1769 v = aa + 16*ai[i]; 1770 for (j=0; j<nz; j++) { 1771 x = rtmp+16*ajtmpold[j]; 1772 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1773 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1774 x[9] = v[9]; x[10] = v[10]; x[11] = v[11]; x[12] = v[12]; x[13] = v[13]; 1775 x[14] = v[14]; x[15] = v[15]; 1776 v += 16; 1777 } 1778 row = *ajtmp++; 1779 while (row < i) { 1780 pc = rtmp + 16*row; 1781 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1782 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1783 p10 = pc[9]; p11 = pc[10]; p12 = pc[11]; p13 = pc[12]; p14 = pc[13]; 1784 p15 = pc[14]; p16 = pc[15]; 1785 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1786 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0 || p10 != 0.0 || 1787 p11 != 0.0 || p12 != 0.0 || p13 != 0.0 || p14 != 0.0 || p15 != 0.0 1788 || p16 != 0.0) { 1789 pv = ba + 16*diag_offset[row]; 1790 pj = bj + diag_offset[row] + 1; 1791 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1792 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1793 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; x14 = pv[13]; 1794 x15 = pv[14]; x16 = pv[15]; 1795 pc[0] = m1 = p1*x1 + p5*x2 + p9*x3 + p13*x4; 1796 pc[1] = m2 = p2*x1 + p6*x2 + p10*x3 + p14*x4; 1797 pc[2] = m3 = p3*x1 + p7*x2 + p11*x3 + p15*x4; 1798 pc[3] = m4 = p4*x1 + p8*x2 + p12*x3 + p16*x4; 1799 1800 pc[4] = m5 = p1*x5 + p5*x6 + p9*x7 + p13*x8; 1801 pc[5] = m6 = p2*x5 + p6*x6 + p10*x7 + p14*x8; 1802 pc[6] = m7 = p3*x5 + p7*x6 + p11*x7 + p15*x8; 1803 pc[7] = m8 = p4*x5 + p8*x6 + p12*x7 + p16*x8; 1804 1805 pc[8] = m9 = p1*x9 + p5*x10 + p9*x11 + p13*x12; 1806 pc[9] = m10 = p2*x9 + p6*x10 + p10*x11 + p14*x12; 1807 pc[10] = m11 = p3*x9 + p7*x10 + p11*x11 + p15*x12; 1808 pc[11] = m12 = p4*x9 + p8*x10 + p12*x11 + p16*x12; 1809 1810 pc[12] = m13 = p1*x13 + p5*x14 + p9*x15 + p13*x16; 1811 pc[13] = m14 = p2*x13 + p6*x14 + p10*x15 + p14*x16; 1812 pc[14] = m15 = p3*x13 + p7*x14 + p11*x15 + p15*x16; 1813 pc[15] = m16 = p4*x13 + p8*x14 + p12*x15 + p16*x16; 1814 1815 nz = bi[row+1] - diag_offset[row] - 1; 1816 pv += 16; 1817 for (j=0; j<nz; j++) { 1818 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1819 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1820 x10 = pv[9]; x11 = pv[10]; x12 = pv[11]; x13 = pv[12]; 1821 x14 = pv[13]; x15 = pv[14]; x16 = pv[15]; 1822 x = rtmp + 16*pj[j]; 1823 x[0] -= m1*x1 + m5*x2 + m9*x3 + m13*x4; 1824 x[1] -= m2*x1 + m6*x2 + m10*x3 + m14*x4; 1825 x[2] -= m3*x1 + m7*x2 + m11*x3 + m15*x4; 1826 x[3] -= m4*x1 + m8*x2 + m12*x3 + m16*x4; 1827 1828 x[4] -= m1*x5 + m5*x6 + m9*x7 + m13*x8; 1829 x[5] -= m2*x5 + m6*x6 + m10*x7 + m14*x8; 1830 x[6] -= m3*x5 + m7*x6 + m11*x7 + m15*x8; 1831 x[7] -= m4*x5 + m8*x6 + m12*x7 + m16*x8; 1832 1833 x[8] -= m1*x9 + m5*x10 + m9*x11 + m13*x12; 1834 x[9] -= m2*x9 + m6*x10 + m10*x11 + m14*x12; 1835 x[10] -= m3*x9 + m7*x10 + m11*x11 + m15*x12; 1836 x[11] -= m4*x9 + m8*x10 + m12*x11 + m16*x12; 1837 1838 x[12] -= m1*x13 + m5*x14 + m9*x15 + m13*x16; 1839 x[13] -= m2*x13 + m6*x14 + m10*x15 + m14*x16; 1840 x[14] -= m3*x13 + m7*x14 + m11*x15 + m15*x16; 1841 x[15] -= m4*x13 + m8*x14 + m12*x15 + m16*x16; 1842 1843 pv += 16; 1844 } 1845 PLogFlops(128*nz+112); 1846 } 1847 row = *ajtmp++; 1848 } 1849 /* finished row so stick it into b->a */ 1850 pv = ba + 16*bi[i]; 1851 pj = bj + bi[i]; 1852 nz = bi[i+1] - bi[i]; 1853 for (j=0; j<nz; j++) { 1854 x = rtmp+16*pj[j]; 1855 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1856 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1857 pv[9] = x[9]; pv[10] = x[10]; pv[11] = x[11]; pv[12] = x[12]; 1858 pv[13] = x[13]; pv[14] = x[14]; pv[15] = x[15]; 1859 pv += 16; 1860 } 1861 /* invert diagonal block */ 1862 w = ba + 16*diag_offset[i]; 1863 ierr = Kernel_A_gets_inverse_A_4(w);CHKERRQ(ierr); 1864 } 1865 1866 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1867 C->factor = FACTOR_LU; 1868 C->assembled = PETSC_TRUE; 1869 PLogFlops(1.3333*64*b->mbs); /* from inverting diagonal blocks */ 1870 PetscFunctionReturn(0); 1871 } 1872 1873 /* 1874 Version for when blocks are 3 by 3 1875 */ 1876 #undef __FUNC__ 1877 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_3" 1878 int MatCholeskyFactorNumeric_SeqSBAIJ_3(Mat A,Mat *B) 1879 { 1880 Mat C = *B; 1881 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1882 IS isrow = b->row,isicol = b->icol; 1883 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1884 int *ajtmpold,*ajtmp,nz,row,*ai=a->i,*aj=a->j; 1885 int *diag_offset = b->diag,idx,*pj; 1886 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1887 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1888 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 1889 MatScalar *ba = b->a,*aa = a->a; 1890 1891 PetscFunctionBegin; 1892 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1893 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1894 rtmp = (MatScalar*)PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 1895 1896 for (i=0; i<n; i++) { 1897 nz = bi[i+1] - bi[i]; 1898 ajtmp = bj + bi[i]; 1899 for (j=0; j<nz; j++) { 1900 x = rtmp + 9*ajtmp[j]; 1901 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 1902 } 1903 /* load in initial (unfactored row) */ 1904 idx = r[i]; 1905 nz = ai[idx+1] - ai[idx]; 1906 ajtmpold = aj + ai[idx]; 1907 v = aa + 9*ai[idx]; 1908 for (j=0; j<nz; j++) { 1909 x = rtmp + 9*ic[ajtmpold[j]]; 1910 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 1911 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 1912 v += 9; 1913 } 1914 row = *ajtmp++; 1915 while (row < i) { 1916 pc = rtmp + 9*row; 1917 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 1918 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 1919 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 1920 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 1921 pv = ba + 9*diag_offset[row]; 1922 pj = bj + diag_offset[row] + 1; 1923 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1924 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1925 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 1926 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 1927 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 1928 1929 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 1930 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 1931 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 1932 1933 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 1934 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 1935 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 1936 nz = bi[row+1] - diag_offset[row] - 1; 1937 pv += 9; 1938 for (j=0; j<nz; j++) { 1939 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 1940 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 1941 x = rtmp + 9*pj[j]; 1942 x[0] -= m1*x1 + m4*x2 + m7*x3; 1943 x[1] -= m2*x1 + m5*x2 + m8*x3; 1944 x[2] -= m3*x1 + m6*x2 + m9*x3; 1945 1946 x[3] -= m1*x4 + m4*x5 + m7*x6; 1947 x[4] -= m2*x4 + m5*x5 + m8*x6; 1948 x[5] -= m3*x4 + m6*x5 + m9*x6; 1949 1950 x[6] -= m1*x7 + m4*x8 + m7*x9; 1951 x[7] -= m2*x7 + m5*x8 + m8*x9; 1952 x[8] -= m3*x7 + m6*x8 + m9*x9; 1953 pv += 9; 1954 } 1955 PLogFlops(54*nz+36); 1956 } 1957 row = *ajtmp++; 1958 } 1959 /* finished row so stick it into b->a */ 1960 pv = ba + 9*bi[i]; 1961 pj = bj + bi[i]; 1962 nz = bi[i+1] - bi[i]; 1963 for (j=0; j<nz; j++) { 1964 x = rtmp + 9*pj[j]; 1965 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 1966 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 1967 pv += 9; 1968 } 1969 /* invert diagonal block */ 1970 w = ba + 9*diag_offset[i]; 1971 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 1972 } 1973 1974 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1975 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1976 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1977 C->factor = FACTOR_LU; 1978 C->assembled = PETSC_TRUE; 1979 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 1980 PetscFunctionReturn(0); 1981 } 1982 /* 1983 Version for when blocks are 3 by 3 Using natural ordering 1984 */ 1985 #undef __FUNC__ 1986 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering" 1987 int MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering(Mat A,Mat *B) 1988 { 1989 Mat C = *B; 1990 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 1991 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 1992 int *ajtmpold,*ajtmp,nz,row; 1993 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 1994 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 1995 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,m5,m6,m7,m8,m9,x1,x2,x3,x4; 1996 MatScalar p5,p6,p7,p8,p9,x5,x6,x7,x8,x9; 1997 MatScalar *ba = b->a,*aa = a->a; 1998 1999 PetscFunctionBegin; 2000 rtmp = (MatScalar*)PetscMalloc(9*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2001 2002 for (i=0; i<n; i++) { 2003 nz = bi[i+1] - bi[i]; 2004 ajtmp = bj + bi[i]; 2005 for (j=0; j<nz; j++) { 2006 x = rtmp+9*ajtmp[j]; 2007 x[0] = x[1] = x[2] = x[3] = x[4] = x[5] = x[6] = x[7] = x[8] = 0.0; 2008 } 2009 /* load in initial (unfactored row) */ 2010 nz = ai[i+1] - ai[i]; 2011 ajtmpold = aj + ai[i]; 2012 v = aa + 9*ai[i]; 2013 for (j=0; j<nz; j++) { 2014 x = rtmp+9*ajtmpold[j]; 2015 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2016 x[4] = v[4]; x[5] = v[5]; x[6] = v[6]; x[7] = v[7]; x[8] = v[8]; 2017 v += 9; 2018 } 2019 row = *ajtmp++; 2020 while (row < i) { 2021 pc = rtmp + 9*row; 2022 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2023 p5 = pc[4]; p6 = pc[5]; p7 = pc[6]; p8 = pc[7]; p9 = pc[8]; 2024 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0 || p5 != 0.0 || 2025 p6 != 0.0 || p7 != 0.0 || p8 != 0.0 || p9 != 0.0) { 2026 pv = ba + 9*diag_offset[row]; 2027 pj = bj + diag_offset[row] + 1; 2028 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2029 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2030 pc[0] = m1 = p1*x1 + p4*x2 + p7*x3; 2031 pc[1] = m2 = p2*x1 + p5*x2 + p8*x3; 2032 pc[2] = m3 = p3*x1 + p6*x2 + p9*x3; 2033 2034 pc[3] = m4 = p1*x4 + p4*x5 + p7*x6; 2035 pc[4] = m5 = p2*x4 + p5*x5 + p8*x6; 2036 pc[5] = m6 = p3*x4 + p6*x5 + p9*x6; 2037 2038 pc[6] = m7 = p1*x7 + p4*x8 + p7*x9; 2039 pc[7] = m8 = p2*x7 + p5*x8 + p8*x9; 2040 pc[8] = m9 = p3*x7 + p6*x8 + p9*x9; 2041 2042 nz = bi[row+1] - diag_offset[row] - 1; 2043 pv += 9; 2044 for (j=0; j<nz; j++) { 2045 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2046 x5 = pv[4]; x6 = pv[5]; x7 = pv[6]; x8 = pv[7]; x9 = pv[8]; 2047 x = rtmp + 9*pj[j]; 2048 x[0] -= m1*x1 + m4*x2 + m7*x3; 2049 x[1] -= m2*x1 + m5*x2 + m8*x3; 2050 x[2] -= m3*x1 + m6*x2 + m9*x3; 2051 2052 x[3] -= m1*x4 + m4*x5 + m7*x6; 2053 x[4] -= m2*x4 + m5*x5 + m8*x6; 2054 x[5] -= m3*x4 + m6*x5 + m9*x6; 2055 2056 x[6] -= m1*x7 + m4*x8 + m7*x9; 2057 x[7] -= m2*x7 + m5*x8 + m8*x9; 2058 x[8] -= m3*x7 + m6*x8 + m9*x9; 2059 pv += 9; 2060 } 2061 PLogFlops(54*nz+36); 2062 } 2063 row = *ajtmp++; 2064 } 2065 /* finished row so stick it into b->a */ 2066 pv = ba + 9*bi[i]; 2067 pj = bj + bi[i]; 2068 nz = bi[i+1] - bi[i]; 2069 for (j=0; j<nz; j++) { 2070 x = rtmp+9*pj[j]; 2071 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2072 pv[4] = x[4]; pv[5] = x[5]; pv[6] = x[6]; pv[7] = x[7]; pv[8] = x[8]; 2073 pv += 9; 2074 } 2075 /* invert diagonal block */ 2076 w = ba + 9*diag_offset[i]; 2077 ierr = Kernel_A_gets_inverse_A_3(w);CHKERRQ(ierr); 2078 } 2079 2080 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2081 C->factor = FACTOR_LU; 2082 C->assembled = PETSC_TRUE; 2083 PLogFlops(1.3333*27*b->mbs); /* from inverting diagonal blocks */ 2084 PetscFunctionReturn(0); 2085 } 2086 2087 /* 2088 Version for when blocks are 2 by 2 2089 */ 2090 #undef __FUNC__ 2091 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_2" 2092 int MatCholeskyFactorNumeric_SeqSBAIJ_2(Mat A,Mat *B) 2093 { 2094 Mat C = *B; 2095 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 2096 IS isrow = b->row,isicol = b->icol; 2097 int *r,*ic,ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 2098 int *ajtmpold,*ajtmp,nz,row; 2099 int *diag_offset=b->diag,idx,*ai=a->i,*aj=a->j,*pj; 2100 MatScalar *pv,*v,*rtmp,m1,m2,m3,m4,*pc,*w,*x,x1,x2,x3,x4; 2101 MatScalar p1,p2,p3,p4; 2102 MatScalar *ba = b->a,*aa = a->a; 2103 2104 PetscFunctionBegin; 2105 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 2106 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 2107 rtmp = (MatScalar*)PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2108 2109 for (i=0; i<n; i++) { 2110 nz = bi[i+1] - bi[i]; 2111 ajtmp = bj + bi[i]; 2112 for (j=0; j<nz; j++) { 2113 x = rtmp+4*ajtmp[j]; x[0] = x[1] = x[2] = x[3] = 0.0; 2114 } 2115 /* load in initial (unfactored row) */ 2116 idx = r[i]; 2117 nz = ai[idx+1] - ai[idx]; 2118 ajtmpold = aj + ai[idx]; 2119 v = aa + 4*ai[idx]; 2120 for (j=0; j<nz; j++) { 2121 x = rtmp+4*ic[ajtmpold[j]]; 2122 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2123 v += 4; 2124 } 2125 row = *ajtmp++; 2126 while (row < i) { 2127 pc = rtmp + 4*row; 2128 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2129 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2130 pv = ba + 4*diag_offset[row]; 2131 pj = bj + diag_offset[row] + 1; 2132 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2133 pc[0] = m1 = p1*x1 + p3*x2; 2134 pc[1] = m2 = p2*x1 + p4*x2; 2135 pc[2] = m3 = p1*x3 + p3*x4; 2136 pc[3] = m4 = p2*x3 + p4*x4; 2137 nz = bi[row+1] - diag_offset[row] - 1; 2138 pv += 4; 2139 for (j=0; j<nz; j++) { 2140 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2141 x = rtmp + 4*pj[j]; 2142 x[0] -= m1*x1 + m3*x2; 2143 x[1] -= m2*x1 + m4*x2; 2144 x[2] -= m1*x3 + m3*x4; 2145 x[3] -= m2*x3 + m4*x4; 2146 pv += 4; 2147 } 2148 PLogFlops(16*nz+12); 2149 } 2150 row = *ajtmp++; 2151 } 2152 /* finished row so stick it into b->a */ 2153 pv = ba + 4*bi[i]; 2154 pj = bj + bi[i]; 2155 nz = bi[i+1] - bi[i]; 2156 for (j=0; j<nz; j++) { 2157 x = rtmp+4*pj[j]; 2158 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2159 pv += 4; 2160 } 2161 /* invert diagonal block */ 2162 w = ba + 4*diag_offset[i]; 2163 ierr = Kernel_A_gets_inverse_A_2(w);CHKERRQ(ierr); 2164 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2165 } 2166 2167 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2168 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 2169 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 2170 C->factor = FACTOR_LU; 2171 C->assembled = PETSC_TRUE; 2172 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2173 PetscFunctionReturn(0); 2174 } 2175 /* 2176 Version for when blocks are 2 by 2 Using natural ordering 2177 */ 2178 #undef __FUNC__ 2179 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering" 2180 int MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering(Mat A,Mat *B) 2181 { 2182 Mat C = *B; 2183 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 2184 int ierr,i,j,n = a->mbs,*bi = b->i,*bj = b->j; 2185 int *ajtmpold,*ajtmp,nz,row; 2186 int *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 2187 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 2188 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,x1,x2,x3,x4; 2189 MatScalar *ba = b->a,*aa = a->a; 2190 2191 PetscFunctionBegin; 2192 rtmp = (MatScalar*)PetscMalloc(4*(n+1)*sizeof(MatScalar));CHKPTRQ(rtmp); 2193 2194 for (i=0; i<n; i++) { 2195 nz = bi[i+1] - bi[i]; 2196 ajtmp = bj + bi[i]; 2197 for (j=0; j<nz; j++) { 2198 x = rtmp+4*ajtmp[j]; 2199 x[0] = x[1] = x[2] = x[3] = 0.0; 2200 } 2201 /* load in initial (unfactored row) */ 2202 nz = ai[i+1] - ai[i]; 2203 ajtmpold = aj + ai[i]; 2204 v = aa + 4*ai[i]; 2205 for (j=0; j<nz; j++) { 2206 x = rtmp+4*ajtmpold[j]; 2207 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 2208 v += 4; 2209 } 2210 row = *ajtmp++; 2211 while (row < i) { 2212 pc = rtmp + 4*row; 2213 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 2214 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 2215 pv = ba + 4*diag_offset[row]; 2216 pj = bj + diag_offset[row] + 1; 2217 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2218 pc[0] = m1 = p1*x1 + p3*x2; 2219 pc[1] = m2 = p2*x1 + p4*x2; 2220 pc[2] = m3 = p1*x3 + p3*x4; 2221 pc[3] = m4 = p2*x3 + p4*x4; 2222 nz = bi[row+1] - diag_offset[row] - 1; 2223 pv += 4; 2224 for (j=0; j<nz; j++) { 2225 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 2226 x = rtmp + 4*pj[j]; 2227 x[0] -= m1*x1 + m3*x2; 2228 x[1] -= m2*x1 + m4*x2; 2229 x[2] -= m1*x3 + m3*x4; 2230 x[3] -= m2*x3 + m4*x4; 2231 pv += 4; 2232 } 2233 PLogFlops(16*nz+12); 2234 } 2235 row = *ajtmp++; 2236 } 2237 /* finished row so stick it into b->a */ 2238 pv = ba + 4*bi[i]; 2239 pj = bj + bi[i]; 2240 nz = bi[i+1] - bi[i]; 2241 for (j=0; j<nz; j++) { 2242 x = rtmp+4*pj[j]; 2243 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 2244 pv += 4; 2245 } 2246 /* invert diagonal block */ 2247 w = ba + 4*diag_offset[i]; 2248 ierr = Kernel_A_gets_inverse_A_2(w);CHKERRQ(ierr); 2249 /*Kernel_A_gets_inverse_A(bs,w,v_pivots,v_work);*/ 2250 } 2251 2252 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2253 C->factor = FACTOR_LU; 2254 C->assembled = PETSC_TRUE; 2255 PLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 2256 PetscFunctionReturn(0); 2257 } 2258 2259 /* 2260 Version for when blocks are 1 by 1. 2261 */ 2262 #undef __FUNC__ 2263 #define __FUNC__ "MatCholeskyFactorNumeric_SeqSBAIJ_1" 2264 int MatCholeskyFactorNumeric_SeqSBAIJ_1(Mat A,Mat *B) 2265 { 2266 Mat C = *B; 2267 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 2268 IS ip = b->row; 2269 int *rip,*riip,ierr,i,j,mbs = a->mbs,*bi = b->i,*bj = b->j; 2270 int *ai = a->i,*aj = a->j; 2271 MatScalar *rtmp; 2272 MatScalar *ba = b->a,*aa = a->a; 2273 MatScalar dk,uikdi; 2274 int k,jmin,jmax,*jl,*il,vj,nexti,juj,ili; 2275 2276 PetscFunctionBegin; 2277 ierr = ISGetIndices(ip,&rip);CHKERRQ(ierr); 2278 riip = rip; 2279 2280 /* INITIALIZATION */ 2281 /* il and jl record the first nonzero element in each row of the accessing 2282 window U(0:k, k:mbs-1). 2283 jl: list of rows to be added to uneliminated rows 2284 i>= k: jl(i) is the first row to be added to row i 2285 i< k: jl(i) is the row following row i in some list of rows 2286 jl(i) = mbs indicates the end of a list 2287 il(i): points to the first nonzero element in columns k,...,mbs-1 of 2288 row i of U */ 2289 rtmp = (MatScalar*)PetscMalloc(mbs*sizeof(MatScalar));CHKPTRQ(rtmp); 2290 il = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(il); 2291 jl = (int*)PetscMalloc(mbs*sizeof(int));CHKPTRQ(jl); 2292 for (i=0; i<mbs; i++) { 2293 rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0; 2294 } 2295 2296 /* FOR EACH ROW K */ 2297 for (k = 0; k<mbs; k++){ 2298 2299 /* INITIALIZE K-TH ROW WITH ELEMENTS NONZERO IN ROW P(K) OF A */ 2300 jmin = ai[rip[k]]; jmax = ai[rip[k]+1]; 2301 if (jmin < jmax) { 2302 for (j = jmin; j < jmax; j++){ 2303 vj = riip[aj[j]]; 2304 if (k <= vj) rtmp[vj] = aa[j]; 2305 } 2306 } 2307 2308 /* MODIFY K-TH ROW BY ADDING IN THOSE ROWS I WITH U(I,K) NE 0 2309 FOR EACH ROW I TO BE ADDED IN */ 2310 dk = rtmp[k]; 2311 i = jl[k]; /* first row to be added to k_th row */ 2312 /* printf(" k=%d, pivot row = %d\n",k,i); */ 2313 2314 while (i < mbs){ 2315 nexti = jl[i]; /* next row to be added to k_th row */ 2316 /* printf(" pivot row = %d\n", nexti); */ 2317 2318 /* COMPUTE MULTIPLIER AND UPDATE DIAGONAL ELEMENT */ 2319 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 2320 uikdi = - ba[ili]*ba[i]; 2321 dk += uikdi*ba[ili]; 2322 ba[ili] = uikdi; /* update U(i,k) */ 2323 2324 /* ADD MULTIPLE OF ROW I TO K-TH ROW ... */ 2325 jmin = ili + 1; jmax = bi[i+1]; 2326 if (jmin < jmax){ 2327 for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j]; 2328 /* ... AND ADD I TO ROW LIST FOR NEXT NONZERO ENTRY */ 2329 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 2330 j = bj[jmin]; 2331 jl[i] = jl[j]; jl[j] = i; /* update jl */ 2332 } 2333 i = nexti; 2334 /* printf(" pivot row i=%d\n",i); */ 2335 } 2336 2337 /* CHECK FOR ZERO PIVOT AND SAVE DIAGONAL ELEMENT */ 2338 if (dk == 0.0){ 2339 SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,0,"Zero pivot"); 2340 } 2341 2342 /* SAVE NONZERO ENTRIES IN K-TH ROW OF U ... */ 2343 ba[k] = 1.0/dk; 2344 jmin = bi[k]; jmax = bi[k+1]; 2345 if (jmin < jmax) { 2346 for (j=jmin; j<jmax; j++){ 2347 juj = bj[j]; ba[j] = rtmp[juj]; rtmp[juj] = 0.0; 2348 } 2349 2350 /* ... AND ADD K TO ROW LIST FOR FIRST NONZERO ENTRY IN K-TH ROW */ 2351 il[k] = jmin; 2352 i = bj[jmin]; 2353 jl[k] = jl[i]; jl[i] = k; 2354 } 2355 } 2356 2357 ierr = PetscFree(rtmp);CHKERRQ(ierr); 2358 ierr = PetscFree(il);CHKERRQ(ierr); 2359 ierr = PetscFree(jl);CHKERRQ(ierr); 2360 2361 ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr); 2362 C->factor = FACTOR_LU; 2363 C->assembled = PETSC_TRUE; 2364 PLogFlops(b->mbs); 2365 PetscFunctionReturn(0); 2366 } 2367 2368 #undef __FUNC__ 2369 #define __FUNC__ "MatCholeskyFactor_SeqSBAIJ" 2370 int MatCholeskyFactor_SeqSBAIJ(Mat A,IS perm,PetscReal f) 2371 { 2372 Mat_SeqSBAIJ *mat = (Mat_SeqSBAIJ*)A->data; 2373 int ierr,refct; 2374 Mat C; 2375 PetscOps *Abops; 2376 MatOps Aops; 2377 2378 PetscFunctionBegin; 2379 ierr = MatCholeskyFactorSymbolic(A,perm,f,&C);CHKERRQ(ierr); 2380 ierr = MatCholeskyFactorNumeric(A,&C);CHKERRQ(ierr); 2381 2382 /* free all the data structures from mat */ 2383 ierr = PetscFree(mat->a);CHKERRQ(ierr); 2384 if (!mat->singlemalloc) { 2385 ierr = PetscFree(mat->i);CHKERRQ(ierr); 2386 ierr = PetscFree(mat->j);CHKERRQ(ierr); 2387 } 2388 if (mat->diag) {ierr = PetscFree(mat->diag);CHKERRQ(ierr);} 2389 if (mat->ilen) {ierr = PetscFree(mat->ilen);CHKERRQ(ierr);} 2390 if (mat->imax) {ierr = PetscFree(mat->imax);CHKERRQ(ierr);} 2391 if (mat->solve_work) {ierr = PetscFree(mat->solve_work);CHKERRQ(ierr);} 2392 if (mat->mult_work) {ierr = PetscFree(mat->mult_work);CHKERRQ(ierr);} 2393 if (mat->icol) {ierr = ISDestroy(mat->icol);CHKERRQ(ierr);} 2394 ierr = PetscFree(mat);CHKERRQ(ierr); 2395 2396 ierr = MapDestroy(A->rmap);CHKERRQ(ierr); 2397 ierr = MapDestroy(A->cmap);CHKERRQ(ierr); 2398 2399 /* 2400 This is horrible,horrible code. We need to keep the 2401 A pointers for the bops and ops but copy everything 2402 else from C. 2403 */ 2404 Abops = A->bops; 2405 Aops = A->ops; 2406 refct = A->refct; 2407 ierr = PetscMemcpy(A,C,sizeof(struct _p_Mat));CHKERRQ(ierr); 2408 mat = (Mat_SeqSBAIJ*)A->data; 2409 PLogObjectParent(A,mat->icol); 2410 2411 A->bops = Abops; 2412 A->ops = Aops; 2413 A->qlist = 0; 2414 A->refct = refct; 2415 /* copy over the type_name and name */ 2416 ierr = PetscStrallocpy(C->type_name,&A->type_name);CHKERRQ(ierr); 2417 ierr = PetscStrallocpy(C->name,&A->name);CHKERRQ(ierr); 2418 2419 PetscHeaderDestroy(C); 2420 PetscFunctionReturn(0); 2421 } 2422 2423 2424