1 #define PETSCMAT_DLL 2 3 /* 4 Factorization code for BAIJ format. 5 */ 6 #include "../src/mat/impls/baij/seq/baij.h" 7 #include "../src/mat/blockinvert.h" 8 9 /* ------------------------------------------------------------*/ 10 /* 11 Version for when blocks are 2 by 2 12 */ 13 /* MatLUFactorNumeric_SeqBAIJ_2_newdatastruct - 14 copied from MatLUFactorNumeric_SeqBAIJ_N_newdatastruct() and manually re-implemented 15 Kernel_A_gets_A_times_B() 16 Kernel_A_gets_A_minus_B_times_C() 17 Kernel_A_gets_inverse_A() 18 */ 19 #undef __FUNCT__ 20 #define __FUNCT__ "MatLUFactorNumeric_SeqBAIJ_2_newdatastruct" 21 PetscErrorCode MatLUFactorNumeric_SeqBAIJ_2_newdatastruct(Mat B,Mat A,const MatFactorInfo *info) 22 { 23 Mat C=B; 24 Mat_SeqBAIJ *a=(Mat_SeqBAIJ*)A->data,*b=(Mat_SeqBAIJ *)C->data; 25 IS isrow = b->row,isicol = b->icol; 26 PetscErrorCode ierr; 27 const PetscInt *r,*ic,*ics; 28 PetscInt i,j,k,n=a->mbs,*ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j; 29 PetscInt *ajtmp,*bjtmp,nz,nzL,row,*bdiag=b->diag,*pj; 30 MatScalar *rtmp,*pc,*mwork,*v,*pv,*aa=a->a; 31 PetscInt bs2 = a->bs2,flg; 32 PetscReal shift = info->shiftinblocks; 33 34 PetscFunctionBegin; 35 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 36 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 37 38 /* generate work space needed by the factorization */ 39 ierr = PetscMalloc((bs2*n+bs2+1)*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 40 mwork = rtmp + bs2*n; 41 ierr = PetscMemzero(rtmp,bs2*n*sizeof(MatScalar));CHKERRQ(ierr); 42 ics = ic; 43 44 for (i=0; i<n; i++){ 45 /* zero rtmp */ 46 /* L part */ 47 nz = bi[i+1] - bi[i]; 48 bjtmp = bj + bi[i]; 49 for (j=0; j<nz; j++){ 50 ierr = PetscMemzero(rtmp+bs2*bjtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 51 } 52 53 /* U part */ 54 nz = bi[2*n-i+1] - bi[2*n-i]; 55 bjtmp = bj + bi[2*n-i]; 56 for (j=0; j<nz; j++){ 57 ierr = PetscMemzero(rtmp+bs2*bjtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 58 } 59 60 /* load in initial (unfactored row) */ 61 nz = ai[r[i]+1] - ai[r[i]]; 62 ajtmp = aj + ai[r[i]]; 63 v = aa + bs2*ai[r[i]]; 64 for (j=0; j<nz; j++) { 65 ierr = PetscMemcpy(rtmp+bs2*ic[ajtmp[j]],v+bs2*j,bs2*sizeof(MatScalar));CHKERRQ(ierr); 66 } 67 68 /* elimination */ 69 bjtmp = bj + bi[i]; 70 nzL = bi[i+1] - bi[i]; 71 for(k=0;k < nzL;k++) { 72 row = bjtmp[k]; 73 pc = rtmp + bs2*row; 74 for (flg=0,j=0; j<bs2; j++) { if (pc[j]!=0.0) { flg = 1; break; }} 75 if (flg) { 76 pv = b->a + bs2*bdiag[row]; 77 /* Kernel_A_gets_A_times_B(bs,pc,pv,mwork); *pc = *pc * (*pv); */ 78 ierr = Kernel_A_gets_A_times_B_2(pc,pv,mwork);CHKERRQ(ierr); 79 80 pj = b->j + bi[2*n-row]; /* begining of U(row,:) */ 81 pv = b->a + bs2*bi[2*n-row]; 82 nz = bi[2*n-row+1] - bi[2*n-row] - 1; /* num of entries inU(row,:), excluding diag */ 83 for (j=0; j<nz; j++) { 84 /* Kernel_A_gets_A_minus_B_times_C(bs,rtmp+bs2*pj[j],pc,pv+bs2*j); */ 85 /* rtmp+bs2*pj[j] = rtmp+bs2*pj[j] - (*pc)*(pv+bs2*j) */ 86 v = rtmp + 4*pj[j]; 87 ierr = Kernel_A_gets_A_minus_B_times_C_2(v,pc,pv);CHKERRQ(ierr); 88 pv += 4; 89 } 90 ierr = PetscLogFlops(16*nz+12);CHKERRQ(ierr); /* flops = 2*bs^3*nz + 2*bs^3 - bs2) */ 91 } 92 } 93 94 /* finished row so stick it into b->a */ 95 /* L part */ 96 pv = b->a + bs2*bi[i] ; 97 pj = b->j + bi[i] ; 98 nz = bi[i+1] - bi[i]; 99 for (j=0; j<nz; j++) { 100 ierr = PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 101 } 102 103 /* Mark diagonal and invert diagonal for simplier triangular solves */ 104 pv = b->a + bs2*bdiag[i]; 105 pj = b->j + bdiag[i]; 106 ierr = PetscMemcpy(pv,rtmp+bs2*pj[0],bs2*sizeof(MatScalar));CHKERRQ(ierr); 107 /* ierr = Kernel_A_gets_inverse_A(bs,pv,v_pivots,v_work);CHKERRQ(ierr); */ 108 ierr = Kernel_A_gets_inverse_A_2(pv,shift);CHKERRQ(ierr); 109 110 /* U part */ 111 pv = b->a + bs2*bi[2*n-i]; 112 pj = b->j + bi[2*n-i]; 113 nz = bi[2*n-i+1] - bi[2*n-i] - 1; 114 for (j=0; j<nz; j++){ 115 ierr = PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 116 } 117 } 118 119 ierr = PetscFree(rtmp);CHKERRQ(ierr); 120 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 121 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 122 123 C->assembled = PETSC_TRUE; 124 ierr = PetscLogFlops(1.3333*bs2*n);CHKERRQ(ierr); /* from inverting diagonal blocks */ 125 PetscFunctionReturn(0); 126 } 127 128 /* 129 MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering_newdatastruct - 130 copied from MatLUFactorNumeric_SeqBAIJ_2_newdatastruct(), then remove ordering 131 */ 132 #undef __FUNCT__ 133 #define __FUNCT__ "MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering_newdatastruct" 134 PetscErrorCode MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering_newdatastruct(Mat B,Mat A,const MatFactorInfo *info) 135 { 136 Mat C=B; 137 Mat_SeqBAIJ *a=(Mat_SeqBAIJ*)A->data,*b=(Mat_SeqBAIJ *)C->data; 138 PetscErrorCode ierr; 139 PetscInt i,j,k,n=a->mbs,*ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j; 140 PetscInt *ajtmp,*bjtmp,nz,nzL,row,*bdiag=b->diag,*pj; 141 MatScalar *rtmp,*pc,*mwork,*v,*pv,*aa=a->a; 142 PetscInt bs2 = a->bs2,flg; 143 PetscReal shift = info->shiftinblocks; 144 145 PetscFunctionBegin; 146 /* generate work space needed by the factorization */ 147 ierr = PetscMalloc((bs2*n+bs2+1)*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 148 mwork = rtmp + bs2*n; 149 ierr = PetscMemzero(rtmp,bs2*n*sizeof(MatScalar));CHKERRQ(ierr); 150 151 for (i=0; i<n; i++){ 152 /* zero rtmp */ 153 /* L part */ 154 nz = bi[i+1] - bi[i]; 155 bjtmp = bj + bi[i]; 156 for (j=0; j<nz; j++){ 157 ierr = PetscMemzero(rtmp+bs2*bjtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 158 } 159 160 /* U part */ 161 nz = bi[2*n-i+1] - bi[2*n-i]; 162 bjtmp = bj + bi[2*n-i]; 163 for (j=0; j<nz; j++){ 164 ierr = PetscMemzero(rtmp+bs2*bjtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 165 } 166 167 /* load in initial (unfactored row) */ 168 nz = ai[i+1] - ai[i]; 169 ajtmp = aj + ai[i]; 170 v = aa + bs2*ai[i]; 171 for (j=0; j<nz; j++) { 172 ierr = PetscMemcpy(rtmp+bs2*ajtmp[j],v+bs2*j,bs2*sizeof(MatScalar));CHKERRQ(ierr); 173 } 174 175 /* elimination */ 176 bjtmp = bj + bi[i]; 177 nzL = bi[i+1] - bi[i]; 178 for(k=0;k < nzL;k++) { 179 row = bjtmp[k]; 180 pc = rtmp + bs2*row; 181 for (flg=0,j=0; j<bs2; j++) { if (pc[j]!=0.0) { flg = 1; break; }} 182 if (flg) { 183 pv = b->a + bs2*bdiag[row]; 184 /* Kernel_A_gets_A_times_B(bs,pc,pv,mwork); *pc = *pc * (*pv); */ 185 ierr = Kernel_A_gets_A_times_B_2(pc,pv,mwork);CHKERRQ(ierr); 186 187 pj = b->j + bi[2*n-row]; /* begining of U(row,:) */ 188 pv = b->a + bs2*bi[2*n-row]; 189 nz = bi[2*n-row+1] - bi[2*n-row] - 1; /* num of entries inU(row,:), excluding diag */ 190 for (j=0; j<nz; j++) { 191 /* Kernel_A_gets_A_minus_B_times_C(bs,rtmp+bs2*pj[j],pc,pv+bs2*j); */ 192 /* rtmp+bs2*pj[j] = rtmp+bs2*pj[j] - (*pc)*(pv+bs2*j) */ 193 v = rtmp + 4*pj[j]; 194 ierr = Kernel_A_gets_A_minus_B_times_C_2(v,pc,pv);CHKERRQ(ierr); 195 pv += 4; 196 } 197 ierr = PetscLogFlops(16*nz+12);CHKERRQ(ierr); /* flops = 2*bs^3*nz + 2*bs^3 - bs2) */ 198 } 199 } 200 201 /* finished row so stick it into b->a */ 202 /* L part */ 203 pv = b->a + bs2*bi[i] ; 204 pj = b->j + bi[i] ; 205 nz = bi[i+1] - bi[i]; 206 for (j=0; j<nz; j++) { 207 ierr = PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 208 } 209 210 /* Mark diagonal and invert diagonal for simplier triangular solves */ 211 pv = b->a + bs2*bdiag[i]; 212 pj = b->j + bdiag[i]; 213 ierr = PetscMemcpy(pv,rtmp+bs2*pj[0],bs2*sizeof(MatScalar));CHKERRQ(ierr); 214 /* ierr = Kernel_A_gets_inverse_A(bs,pv,v_pivots,v_work);CHKERRQ(ierr); */ 215 ierr = Kernel_A_gets_inverse_A_2(pv,shift);CHKERRQ(ierr); 216 217 /* U part */ 218 pv = b->a + bs2*bi[2*n-i]; 219 pj = b->j + bi[2*n-i]; 220 nz = bi[2*n-i+1] - bi[2*n-i] - 1; 221 for (j=0; j<nz; j++){ 222 ierr = PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 223 } 224 } 225 226 ierr = PetscFree(rtmp);CHKERRQ(ierr); 227 C->assembled = PETSC_TRUE; 228 ierr = PetscLogFlops(1.3333*bs2*n);CHKERRQ(ierr); /* from inverting diagonal blocks */ 229 PetscFunctionReturn(0); 230 } 231 232 #undef __FUNCT__ 233 #define __FUNCT__ "MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering_newdatastruct_v2" 234 PetscErrorCode MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering_newdatastruct_v2(Mat B,Mat A,const MatFactorInfo *info) 235 { 236 Mat C=B; 237 Mat_SeqBAIJ *a=(Mat_SeqBAIJ*)A->data,*b=(Mat_SeqBAIJ *)C->data; 238 PetscErrorCode ierr; 239 PetscInt i,j,k,n=a->mbs,*ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j; 240 PetscInt *ajtmp,*bjtmp,nz,nzL,row,*bdiag=b->diag,*pj; 241 MatScalar *rtmp,*pc,*mwork,*v,*pv,*aa=a->a; 242 PetscInt bs2 = a->bs2,flg; 243 PetscReal shift = info->shiftinblocks; 244 245 PetscFunctionBegin; 246 /* generate work space needed by the factorization */ 247 ierr = PetscMalloc((bs2*n+bs2+1)*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 248 mwork = rtmp + bs2*n; 249 ierr = PetscMemzero(rtmp,bs2*n*sizeof(MatScalar));CHKERRQ(ierr); 250 251 for (i=0; i<n; i++){ 252 /* zero rtmp */ 253 /* L part */ 254 nz = bi[i+1] - bi[i]; 255 bjtmp = bj + bi[i]; 256 for (j=0; j<nz; j++){ 257 ierr = PetscMemzero(rtmp+bs2*bjtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 258 } 259 260 /* U part */ 261 nz = bdiag[i] - bdiag[i+1]; 262 bjtmp = bj + bdiag[i+1]+1; 263 for (j=0; j<nz; j++){ 264 ierr = PetscMemzero(rtmp+bs2*bjtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 265 } 266 267 /* load in initial (unfactored row) */ 268 nz = ai[i+1] - ai[i]; 269 ajtmp = aj + ai[i]; 270 v = aa + bs2*ai[i]; 271 for (j=0; j<nz; j++) { 272 ierr = PetscMemcpy(rtmp+bs2*ajtmp[j],v+bs2*j,bs2*sizeof(MatScalar));CHKERRQ(ierr); 273 } 274 275 /* elimination */ 276 bjtmp = bj + bi[i]; 277 nzL = bi[i+1] - bi[i]; 278 for(k=0;k < nzL;k++) { 279 row = bjtmp[k]; 280 pc = rtmp + bs2*row; 281 for (flg=0,j=0; j<bs2; j++) { if (pc[j]!=0.0) { flg = 1; break; }} 282 if (flg) { 283 pv = b->a + bs2*bdiag[row]; 284 /* Kernel_A_gets_A_times_B(bs,pc,pv,mwork); *pc = *pc * (*pv); */ 285 ierr = Kernel_A_gets_A_times_B_2(pc,pv,mwork);CHKERRQ(ierr); 286 287 pj = b->j + bdiag[row+1]+1; /* beginning of U(row,:) */ 288 pv = b->a + bs2*(bdiag[row+1]+1); 289 nz = bdiag[row]-bdiag[row+1] - 1; /* num of entries in U(row,:) excluding diag */ 290 for (j=0; j<nz; j++) { 291 /* Kernel_A_gets_A_minus_B_times_C(bs,rtmp+bs2*pj[j],pc,pv+bs2*j); */ 292 /* rtmp+bs2*pj[j] = rtmp+bs2*pj[j] - (*pc)*(pv+bs2*j) */ 293 v = rtmp + 4*pj[j]; 294 ierr = Kernel_A_gets_A_minus_B_times_C_2(v,pc,pv);CHKERRQ(ierr); 295 pv += 4; 296 } 297 ierr = PetscLogFlops(16*nz+12);CHKERRQ(ierr); /* flops = 2*bs^3*nz + 2*bs^3 - bs2) */ 298 } 299 } 300 301 /* finished row so stick it into b->a */ 302 /* L part */ 303 pv = b->a + bs2*bi[i] ; 304 pj = b->j + bi[i] ; 305 nz = bi[i+1] - bi[i]; 306 for (j=0; j<nz; j++) { 307 ierr = PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 308 } 309 310 /* Mark diagonal and invert diagonal for simplier triangular solves */ 311 pv = b->a + bs2*bdiag[i]; 312 pj = b->j + bdiag[i]; 313 ierr = PetscMemcpy(pv,rtmp+bs2*pj[0],bs2*sizeof(MatScalar));CHKERRQ(ierr); 314 /* ierr = Kernel_A_gets_inverse_A(bs,pv,v_pivots,v_work);CHKERRQ(ierr); */ 315 ierr = Kernel_A_gets_inverse_A_2(pv,shift);CHKERRQ(ierr); 316 317 /* U part */ 318 /* 319 pv = b->a + bs2*bi[2*n-i]; 320 pj = b->j + bi[2*n-i]; 321 nz = bi[2*n-i+1] - bi[2*n-i] - 1; 322 */ 323 pv = b->a + bs2*(bdiag[i+1]+1); 324 pj = b->j + bdiag[i+1]+1; 325 nz = bdiag[i] - bdiag[i+1] - 1; 326 for (j=0; j<nz; j++){ 327 ierr = PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 328 } 329 } 330 331 ierr = PetscFree(rtmp);CHKERRQ(ierr); 332 C->assembled = PETSC_TRUE; 333 ierr = PetscLogFlops(1.3333*bs2*n);CHKERRQ(ierr); /* from inverting diagonal blocks */ 334 PetscFunctionReturn(0); 335 } 336 337 #undef __FUNCT__ 338 #define __FUNCT__ "MatLUFactorNumeric_SeqBAIJ_2" 339 PetscErrorCode MatLUFactorNumeric_SeqBAIJ_2(Mat B,Mat A,const MatFactorInfo *info) 340 { 341 Mat C = B; 342 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 343 IS isrow = b->row,isicol = b->icol; 344 PetscErrorCode ierr; 345 const PetscInt *r,*ic; 346 PetscInt i,j,n = a->mbs,*bi = b->i,*bj = b->j; 347 PetscInt *ajtmpold,*ajtmp,nz,row; 348 PetscInt *diag_offset=b->diag,idx,*ai=a->i,*aj=a->j,*pj; 349 MatScalar *pv,*v,*rtmp,m1,m2,m3,m4,*pc,*w,*x,x1,x2,x3,x4; 350 MatScalar p1,p2,p3,p4; 351 MatScalar *ba = b->a,*aa = a->a; 352 PetscReal shift = info->shiftinblocks; 353 354 PetscFunctionBegin; 355 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 356 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 357 ierr = PetscMalloc(4*(n+1)*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 358 359 for (i=0; i<n; i++) { 360 nz = bi[i+1] - bi[i]; 361 ajtmp = bj + bi[i]; 362 for (j=0; j<nz; j++) { 363 x = rtmp+4*ajtmp[j]; x[0] = x[1] = x[2] = x[3] = 0.0; 364 } 365 /* load in initial (unfactored row) */ 366 idx = r[i]; 367 nz = ai[idx+1] - ai[idx]; 368 ajtmpold = aj + ai[idx]; 369 v = aa + 4*ai[idx]; 370 for (j=0; j<nz; j++) { 371 x = rtmp+4*ic[ajtmpold[j]]; 372 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 373 v += 4; 374 } 375 row = *ajtmp++; 376 while (row < i) { 377 pc = rtmp + 4*row; 378 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 379 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 380 pv = ba + 4*diag_offset[row]; 381 pj = bj + diag_offset[row] + 1; 382 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 383 pc[0] = m1 = p1*x1 + p3*x2; 384 pc[1] = m2 = p2*x1 + p4*x2; 385 pc[2] = m3 = p1*x3 + p3*x4; 386 pc[3] = m4 = p2*x3 + p4*x4; 387 nz = bi[row+1] - diag_offset[row] - 1; 388 pv += 4; 389 for (j=0; j<nz; j++) { 390 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 391 x = rtmp + 4*pj[j]; 392 x[0] -= m1*x1 + m3*x2; 393 x[1] -= m2*x1 + m4*x2; 394 x[2] -= m1*x3 + m3*x4; 395 x[3] -= m2*x3 + m4*x4; 396 pv += 4; 397 } 398 ierr = PetscLogFlops(16.0*nz+12.0);CHKERRQ(ierr); 399 } 400 row = *ajtmp++; 401 } 402 /* finished row so stick it into b->a */ 403 pv = ba + 4*bi[i]; 404 pj = bj + bi[i]; 405 nz = bi[i+1] - bi[i]; 406 for (j=0; j<nz; j++) { 407 x = rtmp+4*pj[j]; 408 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 409 pv += 4; 410 } 411 /* invert diagonal block */ 412 w = ba + 4*diag_offset[i]; 413 ierr = Kernel_A_gets_inverse_A_2(w,shift);CHKERRQ(ierr); 414 } 415 416 ierr = PetscFree(rtmp);CHKERRQ(ierr); 417 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 418 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 419 C->ops->solve = MatSolve_SeqBAIJ_2; 420 C->ops->solvetranspose = MatSolveTranspose_SeqBAIJ_2; 421 C->assembled = PETSC_TRUE; 422 ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 423 PetscFunctionReturn(0); 424 } 425 /* 426 Version for when blocks are 2 by 2 Using natural ordering 427 */ 428 #undef __FUNCT__ 429 #define __FUNCT__ "MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering" 430 PetscErrorCode MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering(Mat C,Mat A,const MatFactorInfo *info) 431 { 432 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 433 PetscErrorCode ierr; 434 PetscInt i,j,n = a->mbs,*bi = b->i,*bj = b->j; 435 PetscInt *ajtmpold,*ajtmp,nz,row; 436 PetscInt *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj; 437 MatScalar *pv,*v,*rtmp,*pc,*w,*x; 438 MatScalar p1,p2,p3,p4,m1,m2,m3,m4,x1,x2,x3,x4; 439 MatScalar *ba = b->a,*aa = a->a; 440 PetscReal shift = info->shiftinblocks; 441 442 PetscFunctionBegin; 443 ierr = PetscMalloc(4*(n+1)*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 444 for (i=0; i<n; i++) { 445 nz = bi[i+1] - bi[i]; 446 ajtmp = bj + bi[i]; 447 for (j=0; j<nz; j++) { 448 x = rtmp+4*ajtmp[j]; 449 x[0] = x[1] = x[2] = x[3] = 0.0; 450 } 451 /* load in initial (unfactored row) */ 452 nz = ai[i+1] - ai[i]; 453 ajtmpold = aj + ai[i]; 454 v = aa + 4*ai[i]; 455 for (j=0; j<nz; j++) { 456 x = rtmp+4*ajtmpold[j]; 457 x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3]; 458 v += 4; 459 } 460 row = *ajtmp++; 461 while (row < i) { 462 pc = rtmp + 4*row; 463 p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3]; 464 if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) { 465 pv = ba + 4*diag_offset[row]; 466 pj = bj + diag_offset[row] + 1; 467 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 468 pc[0] = m1 = p1*x1 + p3*x2; 469 pc[1] = m2 = p2*x1 + p4*x2; 470 pc[2] = m3 = p1*x3 + p3*x4; 471 pc[3] = m4 = p2*x3 + p4*x4; 472 nz = bi[row+1] - diag_offset[row] - 1; 473 pv += 4; 474 for (j=0; j<nz; j++) { 475 x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3]; 476 x = rtmp + 4*pj[j]; 477 x[0] -= m1*x1 + m3*x2; 478 x[1] -= m2*x1 + m4*x2; 479 x[2] -= m1*x3 + m3*x4; 480 x[3] -= m2*x3 + m4*x4; 481 pv += 4; 482 } 483 ierr = PetscLogFlops(16.0*nz+12.0);CHKERRQ(ierr); 484 } 485 row = *ajtmp++; 486 } 487 /* finished row so stick it into b->a */ 488 pv = ba + 4*bi[i]; 489 pj = bj + bi[i]; 490 nz = bi[i+1] - bi[i]; 491 for (j=0; j<nz; j++) { 492 x = rtmp+4*pj[j]; 493 pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3]; 494 /* 495 printf(" col %d:",pj[j]); 496 PetscInt j1; 497 for (j1=0; j1<4; j1++) printf(" %g,",*(pv+j1)); 498 printf("\n"); 499 */ 500 pv += 4; 501 } 502 /* invert diagonal block */ 503 w = ba + 4*diag_offset[i]; 504 /* 505 printf(" \n%d -th: diag: ",i); 506 for (j=0; j<4; j++){ 507 printf(" %g,",w[j]); 508 } 509 printf("\n----------------------------\n"); 510 */ 511 ierr = Kernel_A_gets_inverse_A_2(w,shift);CHKERRQ(ierr); 512 } 513 514 ierr = PetscFree(rtmp);CHKERRQ(ierr); 515 C->ops->solve = MatSolve_SeqBAIJ_2_NaturalOrdering; 516 C->ops->solvetranspose = MatSolveTranspose_SeqBAIJ_2_NaturalOrdering; 517 C->assembled = PETSC_TRUE; 518 ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 519 PetscFunctionReturn(0); 520 } 521 522 /* ----------------------------------------------------------- */ 523 /* 524 Version for when blocks are 1 by 1. 525 */ 526 #undef __FUNCT__ 527 #define __FUNCT__ "MatLUFactorNumeric_SeqBAIJ_1" 528 PetscErrorCode MatLUFactorNumeric_SeqBAIJ_1(Mat C,Mat A,const MatFactorInfo *info) 529 { 530 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data; 531 IS isrow = b->row,isicol = b->icol; 532 PetscErrorCode ierr; 533 const PetscInt *r,*ic; 534 PetscInt i,j,n = a->mbs,*bi = b->i,*bj = b->j; 535 PetscInt *ajtmpold,*ajtmp,nz,row,*ai = a->i,*aj = a->j; 536 PetscInt *diag_offset = b->diag,diag,*pj; 537 MatScalar *pv,*v,*rtmp,multiplier,*pc; 538 MatScalar *ba = b->a,*aa = a->a; 539 PetscTruth row_identity, col_identity; 540 541 PetscFunctionBegin; 542 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 543 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 544 ierr = PetscMalloc((n+1)*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 545 546 for (i=0; i<n; i++) { 547 nz = bi[i+1] - bi[i]; 548 ajtmp = bj + bi[i]; 549 for (j=0; j<nz; j++) rtmp[ajtmp[j]] = 0.0; 550 551 /* load in initial (unfactored row) */ 552 nz = ai[r[i]+1] - ai[r[i]]; 553 ajtmpold = aj + ai[r[i]]; 554 v = aa + ai[r[i]]; 555 for (j=0; j<nz; j++) rtmp[ic[ajtmpold[j]]] = v[j]; 556 557 row = *ajtmp++; 558 while (row < i) { 559 pc = rtmp + row; 560 if (*pc != 0.0) { 561 pv = ba + diag_offset[row]; 562 pj = bj + diag_offset[row] + 1; 563 multiplier = *pc * *pv++; 564 *pc = multiplier; 565 nz = bi[row+1] - diag_offset[row] - 1; 566 for (j=0; j<nz; j++) rtmp[pj[j]] -= multiplier * pv[j]; 567 ierr = PetscLogFlops(1.0+2.0*nz);CHKERRQ(ierr); 568 } 569 row = *ajtmp++; 570 } 571 /* finished row so stick it into b->a */ 572 pv = ba + bi[i]; 573 pj = bj + bi[i]; 574 nz = bi[i+1] - bi[i]; 575 for (j=0; j<nz; j++) {pv[j] = rtmp[pj[j]];} 576 diag = diag_offset[i] - bi[i]; 577 /* check pivot entry for current row */ 578 if (pv[diag] == 0.0) { 579 SETERRQ2(PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot: row in original ordering %D in permuted ordering %D",r[i],i); 580 } 581 pv[diag] = 1.0/pv[diag]; 582 } 583 584 ierr = PetscFree(rtmp);CHKERRQ(ierr); 585 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 586 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 587 ierr = ISIdentity(isrow,&row_identity);CHKERRQ(ierr); 588 ierr = ISIdentity(isicol,&col_identity);CHKERRQ(ierr); 589 if (row_identity && col_identity) { 590 C->ops->solve = MatSolve_SeqBAIJ_1_NaturalOrdering; 591 C->ops->solvetranspose = MatSolveTranspose_SeqBAIJ_1_NaturalOrdering; 592 } else { 593 C->ops->solve = MatSolve_SeqBAIJ_1; 594 C->ops->solvetranspose = MatSolveTranspose_SeqBAIJ_1; 595 } 596 C->assembled = PETSC_TRUE; 597 ierr = PetscLogFlops(C->cmap->n);CHKERRQ(ierr); 598 PetscFunctionReturn(0); 599 } 600 601 EXTERN_C_BEGIN 602 #undef __FUNCT__ 603 #define __FUNCT__ "MatGetFactor_seqbaij_petsc" 604 PetscErrorCode MatGetFactor_seqbaij_petsc(Mat A,MatFactorType ftype,Mat *B) 605 { 606 PetscInt n = A->rmap->n; 607 PetscErrorCode ierr; 608 609 PetscFunctionBegin; 610 ierr = MatCreate(((PetscObject)A)->comm,B);CHKERRQ(ierr); 611 ierr = MatSetSizes(*B,n,n,n,n);CHKERRQ(ierr); 612 if (ftype == MAT_FACTOR_LU || ftype == MAT_FACTOR_ILU || ftype == MAT_FACTOR_ILUDT) { 613 ierr = MatSetType(*B,MATSEQBAIJ);CHKERRQ(ierr); 614 (*B)->ops->lufactorsymbolic = MatLUFactorSymbolic_SeqBAIJ; 615 (*B)->ops->ilufactorsymbolic = MatILUFactorSymbolic_SeqBAIJ; 616 (*B)->ops->iludtfactor = MatILUDTFactor_SeqBAIJ; 617 } else if (ftype == MAT_FACTOR_CHOLESKY || ftype == MAT_FACTOR_ICC) { 618 ierr = MatSetType(*B,MATSEQSBAIJ);CHKERRQ(ierr); 619 ierr = MatSeqSBAIJSetPreallocation(*B,1,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr); 620 (*B)->ops->iccfactorsymbolic = MatICCFactorSymbolic_SeqBAIJ; 621 (*B)->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_SeqBAIJ; 622 } else SETERRQ(PETSC_ERR_SUP,"Factor type not supported"); 623 (*B)->factor = ftype; 624 PetscFunctionReturn(0); 625 } 626 EXTERN_C_END 627 628 EXTERN_C_BEGIN 629 #undef __FUNCT__ 630 #define __FUNCT__ "MatGetFactorAvailable_seqaij_petsc" 631 PetscErrorCode MatGetFactorAvailable_seqbaij_petsc(Mat A,MatFactorType ftype,PetscTruth *flg) 632 { 633 PetscFunctionBegin; 634 *flg = PETSC_TRUE; 635 PetscFunctionReturn(0); 636 } 637 EXTERN_C_END 638 639 /* ----------------------------------------------------------- */ 640 #undef __FUNCT__ 641 #define __FUNCT__ "MatLUFactor_SeqBAIJ" 642 PetscErrorCode MatLUFactor_SeqBAIJ(Mat A,IS row,IS col,const MatFactorInfo *info) 643 { 644 PetscErrorCode ierr; 645 Mat C; 646 647 PetscFunctionBegin; 648 ierr = MatGetFactor(A,MAT_SOLVER_PETSC,MAT_FACTOR_LU,&C);CHKERRQ(ierr); 649 ierr = MatLUFactorSymbolic(C,A,row,col,info);CHKERRQ(ierr); 650 ierr = MatLUFactorNumeric(C,A,info);CHKERRQ(ierr); 651 A->ops->solve = C->ops->solve; 652 A->ops->solvetranspose = C->ops->solvetranspose; 653 ierr = MatHeaderCopy(A,C);CHKERRQ(ierr); 654 ierr = PetscLogObjectParent(A,((Mat_SeqBAIJ*)(A->data))->icol);CHKERRQ(ierr); 655 PetscFunctionReturn(0); 656 } 657 658 #include "../src/mat/impls/sbaij/seq/sbaij.h" 659 #undef __FUNCT__ 660 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqBAIJ_N" 661 PetscErrorCode MatCholeskyFactorNumeric_SeqBAIJ_N(Mat C,Mat A,const MatFactorInfo *info) 662 { 663 PetscErrorCode ierr; 664 Mat_SeqBAIJ *a=(Mat_SeqBAIJ*)A->data; 665 Mat_SeqSBAIJ *b=(Mat_SeqSBAIJ*)C->data; 666 IS ip=b->row; 667 const PetscInt *rip; 668 PetscInt i,j,mbs=a->mbs,bs=A->rmap->bs,*bi=b->i,*bj=b->j,*bcol; 669 PetscInt *ai=a->i,*aj=a->j; 670 PetscInt k,jmin,jmax,*jl,*il,col,nexti,ili,nz; 671 MatScalar *rtmp,*ba=b->a,*bval,*aa=a->a,dk,uikdi; 672 PetscReal zeropivot,rs,shiftnz; 673 PetscReal shiftpd; 674 ChShift_Ctx sctx; 675 PetscInt newshift; 676 677 PetscFunctionBegin; 678 if (bs > 1) { 679 if (!a->sbaijMat){ 680 ierr = MatConvert(A,MATSEQSBAIJ,MAT_INITIAL_MATRIX,&a->sbaijMat);CHKERRQ(ierr); 681 } 682 ierr = (a->sbaijMat)->ops->choleskyfactornumeric(C,a->sbaijMat,info);CHKERRQ(ierr); 683 ierr = MatDestroy(a->sbaijMat);CHKERRQ(ierr); 684 a->sbaijMat = PETSC_NULL; 685 PetscFunctionReturn(0); 686 } 687 688 /* initialization */ 689 shiftnz = info->shiftnz; 690 shiftpd = info->shiftpd; 691 zeropivot = info->zeropivot; 692 693 ierr = ISGetIndices(ip,&rip);CHKERRQ(ierr); 694 nz = (2*mbs+1)*sizeof(PetscInt)+mbs*sizeof(MatScalar); 695 ierr = PetscMalloc(nz,&il);CHKERRQ(ierr); 696 jl = il + mbs; 697 rtmp = (MatScalar*)(jl + mbs); 698 699 sctx.shift_amount = 0; 700 sctx.nshift = 0; 701 do { 702 sctx.chshift = PETSC_FALSE; 703 for (i=0; i<mbs; i++) { 704 rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0; 705 } 706 707 for (k = 0; k<mbs; k++){ 708 bval = ba + bi[k]; 709 /* initialize k-th row by the perm[k]-th row of A */ 710 jmin = ai[rip[k]]; jmax = ai[rip[k]+1]; 711 for (j = jmin; j < jmax; j++){ 712 col = rip[aj[j]]; 713 if (col >= k){ /* only take upper triangular entry */ 714 rtmp[col] = aa[j]; 715 *bval++ = 0.0; /* for in-place factorization */ 716 } 717 } 718 719 /* shift the diagonal of the matrix */ 720 if (sctx.nshift) rtmp[k] += sctx.shift_amount; 721 722 /* modify k-th row by adding in those rows i with U(i,k)!=0 */ 723 dk = rtmp[k]; 724 i = jl[k]; /* first row to be added to k_th row */ 725 726 while (i < k){ 727 nexti = jl[i]; /* next row to be added to k_th row */ 728 729 /* compute multiplier, update diag(k) and U(i,k) */ 730 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 731 uikdi = - ba[ili]*ba[bi[i]]; /* diagonal(k) */ 732 dk += uikdi*ba[ili]; 733 ba[ili] = uikdi; /* -U(i,k) */ 734 735 /* add multiple of row i to k-th row */ 736 jmin = ili + 1; jmax = bi[i+1]; 737 if (jmin < jmax){ 738 for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j]; 739 /* update il and jl for row i */ 740 il[i] = jmin; 741 j = bj[jmin]; jl[i] = jl[j]; jl[j] = i; 742 } 743 i = nexti; 744 } 745 746 /* shift the diagonals when zero pivot is detected */ 747 /* compute rs=sum of abs(off-diagonal) */ 748 rs = 0.0; 749 jmin = bi[k]+1; 750 nz = bi[k+1] - jmin; 751 if (nz){ 752 bcol = bj + jmin; 753 while (nz--){ 754 rs += PetscAbsScalar(rtmp[*bcol]); 755 bcol++; 756 } 757 } 758 759 sctx.rs = rs; 760 sctx.pv = dk; 761 ierr = MatCholeskyCheckShift_inline(info,sctx,k,newshift);CHKERRQ(ierr); 762 if (newshift == 1) break; 763 764 /* copy data into U(k,:) */ 765 ba[bi[k]] = 1.0/dk; /* U(k,k) */ 766 jmin = bi[k]+1; jmax = bi[k+1]; 767 if (jmin < jmax) { 768 for (j=jmin; j<jmax; j++){ 769 col = bj[j]; ba[j] = rtmp[col]; rtmp[col] = 0.0; 770 } 771 /* add the k-th row into il and jl */ 772 il[k] = jmin; 773 i = bj[jmin]; jl[k] = jl[i]; jl[i] = k; 774 } 775 } 776 } while (sctx.chshift); 777 ierr = PetscFree(il);CHKERRQ(ierr); 778 779 ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr); 780 C->assembled = PETSC_TRUE; 781 C->preallocated = PETSC_TRUE; 782 ierr = PetscLogFlops(C->rmap->N);CHKERRQ(ierr); 783 if (sctx.nshift){ 784 if (shiftpd) { 785 ierr = PetscInfo2(A,"number of shiftpd tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 786 } else if (shiftnz) { 787 ierr = PetscInfo2(A,"number of shiftnz tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 788 } 789 } 790 PetscFunctionReturn(0); 791 } 792 793 #undef __FUNCT__ 794 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqBAIJ_N_NaturalOrdering" 795 PetscErrorCode MatCholeskyFactorNumeric_SeqBAIJ_N_NaturalOrdering(Mat C,Mat A,const MatFactorInfo *info) 796 { 797 Mat_SeqBAIJ *a=(Mat_SeqBAIJ*)A->data; 798 Mat_SeqSBAIJ *b=(Mat_SeqSBAIJ*)C->data; 799 PetscErrorCode ierr; 800 PetscInt i,j,am=a->mbs; 801 PetscInt *ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j; 802 PetscInt k,jmin,*jl,*il,nexti,ili,*acol,*bcol,nz; 803 MatScalar *rtmp,*ba=b->a,*aa=a->a,dk,uikdi,*aval,*bval; 804 PetscReal zeropivot,rs,shiftnz; 805 PetscReal shiftpd; 806 ChShift_Ctx sctx; 807 PetscInt newshift; 808 809 PetscFunctionBegin; 810 /* initialization */ 811 shiftnz = info->shiftnz; 812 shiftpd = info->shiftpd; 813 zeropivot = info->zeropivot; 814 815 nz = (2*am+1)*sizeof(PetscInt)+am*sizeof(MatScalar); 816 ierr = PetscMalloc(nz,&il);CHKERRQ(ierr); 817 jl = il + am; 818 rtmp = (MatScalar*)(jl + am); 819 820 sctx.shift_amount = 0; 821 sctx.nshift = 0; 822 do { 823 sctx.chshift = PETSC_FALSE; 824 for (i=0; i<am; i++) { 825 rtmp[i] = 0.0; jl[i] = am; il[0] = 0; 826 } 827 828 for (k = 0; k<am; k++){ 829 /* initialize k-th row with elements nonzero in row perm(k) of A */ 830 nz = ai[k+1] - ai[k]; 831 acol = aj + ai[k]; 832 aval = aa + ai[k]; 833 bval = ba + bi[k]; 834 while (nz -- ){ 835 if (*acol < k) { /* skip lower triangular entries */ 836 acol++; aval++; 837 } else { 838 rtmp[*acol++] = *aval++; 839 *bval++ = 0.0; /* for in-place factorization */ 840 } 841 } 842 843 /* shift the diagonal of the matrix */ 844 if (sctx.nshift) rtmp[k] += sctx.shift_amount; 845 846 /* modify k-th row by adding in those rows i with U(i,k)!=0 */ 847 dk = rtmp[k]; 848 i = jl[k]; /* first row to be added to k_th row */ 849 850 while (i < k){ 851 nexti = jl[i]; /* next row to be added to k_th row */ 852 /* compute multiplier, update D(k) and U(i,k) */ 853 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 854 uikdi = - ba[ili]*ba[bi[i]]; 855 dk += uikdi*ba[ili]; 856 ba[ili] = uikdi; /* -U(i,k) */ 857 858 /* add multiple of row i to k-th row ... */ 859 jmin = ili + 1; 860 nz = bi[i+1] - jmin; 861 if (nz > 0){ 862 bcol = bj + jmin; 863 bval = ba + jmin; 864 while (nz --) rtmp[*bcol++] += uikdi*(*bval++); 865 /* update il and jl for i-th row */ 866 il[i] = jmin; 867 j = bj[jmin]; jl[i] = jl[j]; jl[j] = i; 868 } 869 i = nexti; 870 } 871 872 /* shift the diagonals when zero pivot is detected */ 873 /* compute rs=sum of abs(off-diagonal) */ 874 rs = 0.0; 875 jmin = bi[k]+1; 876 nz = bi[k+1] - jmin; 877 if (nz){ 878 bcol = bj + jmin; 879 while (nz--){ 880 rs += PetscAbsScalar(rtmp[*bcol]); 881 bcol++; 882 } 883 } 884 885 sctx.rs = rs; 886 sctx.pv = dk; 887 ierr = MatCholeskyCheckShift_inline(info,sctx,k,newshift);CHKERRQ(ierr); 888 if (newshift == 1) break; /* sctx.shift_amount is updated */ 889 890 /* copy data into U(k,:) */ 891 ba[bi[k]] = 1.0/dk; 892 jmin = bi[k]+1; 893 nz = bi[k+1] - jmin; 894 if (nz){ 895 bcol = bj + jmin; 896 bval = ba + jmin; 897 while (nz--){ 898 *bval++ = rtmp[*bcol]; 899 rtmp[*bcol++] = 0.0; 900 } 901 /* add k-th row into il and jl */ 902 il[k] = jmin; 903 i = bj[jmin]; jl[k] = jl[i]; jl[i] = k; 904 } 905 } 906 } while (sctx.chshift); 907 ierr = PetscFree(il);CHKERRQ(ierr); 908 909 C->ops->solve = MatSolve_SeqSBAIJ_1_NaturalOrdering; 910 C->ops->solvetranspose = MatSolve_SeqSBAIJ_1_NaturalOrdering; 911 C->assembled = PETSC_TRUE; 912 C->preallocated = PETSC_TRUE; 913 ierr = PetscLogFlops(C->rmap->N);CHKERRQ(ierr); 914 if (sctx.nshift){ 915 if (shiftnz) { 916 ierr = PetscInfo2(A,"number of shiftnz tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 917 } else if (shiftpd) { 918 ierr = PetscInfo2(A,"number of shiftpd tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 919 } 920 } 921 PetscFunctionReturn(0); 922 } 923 924 #include "petscbt.h" 925 #include "../src/mat/utils/freespace.h" 926 #undef __FUNCT__ 927 #define __FUNCT__ "MatICCFactorSymbolic_SeqBAIJ" 928 PetscErrorCode MatICCFactorSymbolic_SeqBAIJ(Mat fact,Mat A,IS perm,const MatFactorInfo *info) 929 { 930 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data; 931 Mat_SeqSBAIJ *b; 932 Mat B; 933 PetscErrorCode ierr; 934 PetscTruth perm_identity; 935 PetscInt reallocs=0,i,*ai=a->i,*aj=a->j,am=a->mbs,bs=A->rmap->bs,*ui; 936 const PetscInt *rip; 937 PetscInt jmin,jmax,nzk,k,j,*jl,prow,*il,nextprow; 938 PetscInt nlnk,*lnk,*lnk_lvl=PETSC_NULL,ncols,ncols_upper,*cols,*cols_lvl,*uj,**uj_ptr,**uj_lvl_ptr; 939 PetscReal fill=info->fill,levels=info->levels; 940 PetscFreeSpaceList free_space=PETSC_NULL,current_space=PETSC_NULL; 941 PetscFreeSpaceList free_space_lvl=PETSC_NULL,current_space_lvl=PETSC_NULL; 942 PetscBT lnkbt; 943 944 PetscFunctionBegin; 945 if (bs > 1){ 946 if (!a->sbaijMat){ 947 ierr = MatConvert(A,MATSEQSBAIJ,MAT_INITIAL_MATRIX,&a->sbaijMat);CHKERRQ(ierr); 948 } 949 (fact)->ops->iccfactorsymbolic = MatICCFactorSymbolic_SeqSBAIJ; /* undue the change made in MatGetFactor_seqbaij_petsc */ 950 ierr = MatICCFactorSymbolic(fact,a->sbaijMat,perm,info);CHKERRQ(ierr); 951 PetscFunctionReturn(0); 952 } 953 954 ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr); 955 ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr); 956 957 /* special case that simply copies fill pattern */ 958 if (!levels && perm_identity) { 959 ierr = MatMarkDiagonal_SeqBAIJ(A);CHKERRQ(ierr); 960 ierr = PetscMalloc((am+1)*sizeof(PetscInt),&ui);CHKERRQ(ierr); 961 for (i=0; i<am; i++) { 962 ui[i] = ai[i+1] - a->diag[i]; /* ui: rowlengths - changes when !perm_identity */ 963 } 964 B = fact; 965 ierr = MatSeqSBAIJSetPreallocation(B,1,0,ui);CHKERRQ(ierr); 966 967 968 b = (Mat_SeqSBAIJ*)B->data; 969 uj = b->j; 970 for (i=0; i<am; i++) { 971 aj = a->j + a->diag[i]; 972 for (j=0; j<ui[i]; j++){ 973 *uj++ = *aj++; 974 } 975 b->ilen[i] = ui[i]; 976 } 977 ierr = PetscFree(ui);CHKERRQ(ierr); 978 B->factor = MAT_FACTOR_NONE; 979 ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 980 ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 981 B->factor = MAT_FACTOR_ICC; 982 983 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqBAIJ_N_NaturalOrdering; 984 PetscFunctionReturn(0); 985 } 986 987 /* initialization */ 988 ierr = PetscMalloc((am+1)*sizeof(PetscInt),&ui);CHKERRQ(ierr); 989 ui[0] = 0; 990 ierr = PetscMalloc((2*am+1)*sizeof(PetscInt),&cols_lvl);CHKERRQ(ierr); 991 992 /* jl: linked list for storing indices of the pivot rows 993 il: il[i] points to the 1st nonzero entry of U(i,k:am-1) */ 994 ierr = PetscMalloc((2*am+1)*sizeof(PetscInt)+2*am*sizeof(PetscInt*),&jl);CHKERRQ(ierr); 995 il = jl + am; 996 uj_ptr = (PetscInt**)(il + am); 997 uj_lvl_ptr = (PetscInt**)(uj_ptr + am); 998 for (i=0; i<am; i++){ 999 jl[i] = am; il[i] = 0; 1000 } 1001 1002 /* create and initialize a linked list for storing column indices of the active row k */ 1003 nlnk = am + 1; 1004 ierr = PetscIncompleteLLCreate(am,am,nlnk,lnk,lnk_lvl,lnkbt);CHKERRQ(ierr); 1005 1006 /* initial FreeSpace size is fill*(ai[am]+1) */ 1007 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+1)),&free_space);CHKERRQ(ierr); 1008 current_space = free_space; 1009 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+1)),&free_space_lvl);CHKERRQ(ierr); 1010 current_space_lvl = free_space_lvl; 1011 1012 for (k=0; k<am; k++){ /* for each active row k */ 1013 /* initialize lnk by the column indices of row rip[k] of A */ 1014 nzk = 0; 1015 ncols = ai[rip[k]+1] - ai[rip[k]]; 1016 ncols_upper = 0; 1017 cols = cols_lvl + am; 1018 for (j=0; j<ncols; j++){ 1019 i = rip[*(aj + ai[rip[k]] + j)]; 1020 if (i >= k){ /* only take upper triangular entry */ 1021 cols[ncols_upper] = i; 1022 cols_lvl[ncols_upper] = -1; /* initialize level for nonzero entries */ 1023 ncols_upper++; 1024 } 1025 } 1026 ierr = PetscIncompleteLLAdd(ncols_upper,cols,levels,cols_lvl,am,nlnk,lnk,lnk_lvl,lnkbt);CHKERRQ(ierr); 1027 nzk += nlnk; 1028 1029 /* update lnk by computing fill-in for each pivot row to be merged in */ 1030 prow = jl[k]; /* 1st pivot row */ 1031 1032 while (prow < k){ 1033 nextprow = jl[prow]; 1034 1035 /* merge prow into k-th row */ 1036 jmin = il[prow] + 1; /* index of the 2nd nzero entry in U(prow,k:am-1) */ 1037 jmax = ui[prow+1]; 1038 ncols = jmax-jmin; 1039 i = jmin - ui[prow]; 1040 cols = uj_ptr[prow] + i; /* points to the 2nd nzero entry in U(prow,k:am-1) */ 1041 for (j=0; j<ncols; j++) cols_lvl[j] = *(uj_lvl_ptr[prow] + i + j); 1042 ierr = PetscIncompleteLLAddSorted(ncols,cols,levels,cols_lvl,am,nlnk,lnk,lnk_lvl,lnkbt);CHKERRQ(ierr); 1043 nzk += nlnk; 1044 1045 /* update il and jl for prow */ 1046 if (jmin < jmax){ 1047 il[prow] = jmin; 1048 j = *cols; jl[prow] = jl[j]; jl[j] = prow; 1049 } 1050 prow = nextprow; 1051 } 1052 1053 /* if free space is not available, make more free space */ 1054 if (current_space->local_remaining<nzk) { 1055 i = am - k + 1; /* num of unfactored rows */ 1056 i = PetscMin(i*nzk, i*(i-1)); /* i*nzk, i*(i-1): estimated and max additional space needed */ 1057 ierr = PetscFreeSpaceGet(i,¤t_space);CHKERRQ(ierr); 1058 ierr = PetscFreeSpaceGet(i,¤t_space_lvl);CHKERRQ(ierr); 1059 reallocs++; 1060 } 1061 1062 /* copy data into free_space and free_space_lvl, then initialize lnk */ 1063 ierr = PetscIncompleteLLClean(am,am,nzk,lnk,lnk_lvl,current_space->array,current_space_lvl->array,lnkbt);CHKERRQ(ierr); 1064 1065 /* add the k-th row into il and jl */ 1066 if (nzk-1 > 0){ 1067 i = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:am-1) */ 1068 jl[k] = jl[i]; jl[i] = k; 1069 il[k] = ui[k] + 1; 1070 } 1071 uj_ptr[k] = current_space->array; 1072 uj_lvl_ptr[k] = current_space_lvl->array; 1073 1074 current_space->array += nzk; 1075 current_space->local_used += nzk; 1076 current_space->local_remaining -= nzk; 1077 1078 current_space_lvl->array += nzk; 1079 current_space_lvl->local_used += nzk; 1080 current_space_lvl->local_remaining -= nzk; 1081 1082 ui[k+1] = ui[k] + nzk; 1083 } 1084 1085 #if defined(PETSC_USE_INFO) 1086 if (ai[am] != 0) { 1087 PetscReal af = ((PetscReal)(2*ui[am]-am))/((PetscReal)ai[am]); 1088 ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %G needed %G\n",reallocs,fill,af);CHKERRQ(ierr); 1089 ierr = PetscInfo1(A,"Run with -pc_factor_fill %G or use \n",af);CHKERRQ(ierr); 1090 ierr = PetscInfo1(A,"PCFactorSetFill(pc,%G) for best performance.\n",af);CHKERRQ(ierr); 1091 } else { 1092 ierr = PetscInfo(A,"Empty matrix.\n");CHKERRQ(ierr); 1093 } 1094 #endif 1095 1096 ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr); 1097 ierr = PetscFree(jl);CHKERRQ(ierr); 1098 ierr = PetscFree(cols_lvl);CHKERRQ(ierr); 1099 1100 /* destroy list of free space and other temporary array(s) */ 1101 ierr = PetscMalloc((ui[am]+1)*sizeof(PetscInt),&uj);CHKERRQ(ierr); 1102 ierr = PetscFreeSpaceContiguous(&free_space,uj);CHKERRQ(ierr); 1103 ierr = PetscIncompleteLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 1104 ierr = PetscFreeSpaceDestroy(free_space_lvl);CHKERRQ(ierr); 1105 1106 /* put together the new matrix in MATSEQSBAIJ format */ 1107 B = fact; 1108 ierr = MatSeqSBAIJSetPreallocation(B,1,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr); 1109 1110 b = (Mat_SeqSBAIJ*)B->data; 1111 b->singlemalloc = PETSC_FALSE; 1112 b->free_a = PETSC_TRUE; 1113 b->free_ij = PETSC_TRUE; 1114 ierr = PetscMalloc((ui[am]+1)*sizeof(MatScalar),&b->a);CHKERRQ(ierr); 1115 b->j = uj; 1116 b->i = ui; 1117 b->diag = 0; 1118 b->ilen = 0; 1119 b->imax = 0; 1120 b->row = perm; 1121 b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */ 1122 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 1123 b->icol = perm; 1124 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 1125 ierr = PetscMalloc((am+1)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr); 1126 ierr = PetscLogObjectMemory(B,(ui[am]-am)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr); 1127 b->maxnz = b->nz = ui[am]; 1128 1129 B->info.factor_mallocs = reallocs; 1130 B->info.fill_ratio_given = fill; 1131 if (ai[am] != 0) { 1132 B->info.fill_ratio_needed = ((PetscReal)ui[am])/((PetscReal)ai[am]); 1133 } else { 1134 B->info.fill_ratio_needed = 0.0; 1135 } 1136 if (perm_identity){ 1137 B->ops->solve = MatSolve_SeqSBAIJ_1_NaturalOrdering; 1138 B->ops->solvetranspose = MatSolve_SeqSBAIJ_1_NaturalOrdering; 1139 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqBAIJ_N_NaturalOrdering; 1140 } else { 1141 (fact)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqBAIJ_N; 1142 } 1143 PetscFunctionReturn(0); 1144 } 1145 1146 #undef __FUNCT__ 1147 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqBAIJ" 1148 PetscErrorCode MatCholeskyFactorSymbolic_SeqBAIJ(Mat fact,Mat A,IS perm,const MatFactorInfo *info) 1149 { 1150 Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data; 1151 Mat_SeqSBAIJ *b; 1152 Mat B; 1153 PetscErrorCode ierr; 1154 PetscTruth perm_identity; 1155 PetscReal fill = info->fill; 1156 const PetscInt *rip; 1157 PetscInt i,mbs=a->mbs,bs=A->rmap->bs,*ai=a->i,*aj=a->j,reallocs=0,prow; 1158 PetscInt *jl,jmin,jmax,nzk,*ui,k,j,*il,nextprow; 1159 PetscInt nlnk,*lnk,ncols,ncols_upper,*cols,*uj,**ui_ptr,*uj_ptr; 1160 PetscFreeSpaceList free_space=PETSC_NULL,current_space=PETSC_NULL; 1161 PetscBT lnkbt; 1162 1163 PetscFunctionBegin; 1164 if (bs > 1) { /* convert to seqsbaij */ 1165 if (!a->sbaijMat){ 1166 ierr = MatConvert(A,MATSEQSBAIJ,MAT_INITIAL_MATRIX,&a->sbaijMat);CHKERRQ(ierr); 1167 } 1168 (fact)->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_SeqSBAIJ; /* undue the change made in MatGetFactor_seqbaij_petsc */ 1169 ierr = MatCholeskyFactorSymbolic(fact,a->sbaijMat,perm,info);CHKERRQ(ierr); 1170 PetscFunctionReturn(0); 1171 } 1172 1173 /* check whether perm is the identity mapping */ 1174 ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr); 1175 if (!perm_identity) SETERRQ(PETSC_ERR_SUP,"Matrix reordering is not supported"); 1176 ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr); 1177 1178 /* initialization */ 1179 ierr = PetscMalloc((mbs+1)*sizeof(PetscInt),&ui);CHKERRQ(ierr); 1180 ui[0] = 0; 1181 1182 /* jl: linked list for storing indices of the pivot rows 1183 il: il[i] points to the 1st nonzero entry of U(i,k:mbs-1) */ 1184 ierr = PetscMalloc((3*mbs+1)*sizeof(PetscInt)+mbs*sizeof(PetscInt*),&jl);CHKERRQ(ierr); 1185 il = jl + mbs; 1186 cols = il + mbs; 1187 ui_ptr = (PetscInt**)(cols + mbs); 1188 for (i=0; i<mbs; i++){ 1189 jl[i] = mbs; il[i] = 0; 1190 } 1191 1192 /* create and initialize a linked list for storing column indices of the active row k */ 1193 nlnk = mbs + 1; 1194 ierr = PetscLLCreate(mbs,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 1195 1196 /* initial FreeSpace size is fill*(ai[mbs]+1) */ 1197 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[mbs]+1)),&free_space);CHKERRQ(ierr); 1198 current_space = free_space; 1199 1200 for (k=0; k<mbs; k++){ /* for each active row k */ 1201 /* initialize lnk by the column indices of row rip[k] of A */ 1202 nzk = 0; 1203 ncols = ai[rip[k]+1] - ai[rip[k]]; 1204 ncols_upper = 0; 1205 for (j=0; j<ncols; j++){ 1206 i = rip[*(aj + ai[rip[k]] + j)]; 1207 if (i >= k){ /* only take upper triangular entry */ 1208 cols[ncols_upper] = i; 1209 ncols_upper++; 1210 } 1211 } 1212 ierr = PetscLLAdd(ncols_upper,cols,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 1213 nzk += nlnk; 1214 1215 /* update lnk by computing fill-in for each pivot row to be merged in */ 1216 prow = jl[k]; /* 1st pivot row */ 1217 1218 while (prow < k){ 1219 nextprow = jl[prow]; 1220 /* merge prow into k-th row */ 1221 jmin = il[prow] + 1; /* index of the 2nd nzero entry in U(prow,k:mbs-1) */ 1222 jmax = ui[prow+1]; 1223 ncols = jmax-jmin; 1224 uj_ptr = ui_ptr[prow] + jmin - ui[prow]; /* points to the 2nd nzero entry in U(prow,k:mbs-1) */ 1225 ierr = PetscLLAddSorted(ncols,uj_ptr,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 1226 nzk += nlnk; 1227 1228 /* update il and jl for prow */ 1229 if (jmin < jmax){ 1230 il[prow] = jmin; 1231 j = *uj_ptr; jl[prow] = jl[j]; jl[j] = prow; 1232 } 1233 prow = nextprow; 1234 } 1235 1236 /* if free space is not available, make more free space */ 1237 if (current_space->local_remaining<nzk) { 1238 i = mbs - k + 1; /* num of unfactored rows */ 1239 i = PetscMin(i*nzk, i*(i-1)); /* i*nzk, i*(i-1): estimated and max additional space needed */ 1240 ierr = PetscFreeSpaceGet(i,¤t_space);CHKERRQ(ierr); 1241 reallocs++; 1242 } 1243 1244 /* copy data into free space, then initialize lnk */ 1245 ierr = PetscLLClean(mbs,mbs,nzk,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 1246 1247 /* add the k-th row into il and jl */ 1248 if (nzk-1 > 0){ 1249 i = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:mbs-1) */ 1250 jl[k] = jl[i]; jl[i] = k; 1251 il[k] = ui[k] + 1; 1252 } 1253 ui_ptr[k] = current_space->array; 1254 current_space->array += nzk; 1255 current_space->local_used += nzk; 1256 current_space->local_remaining -= nzk; 1257 1258 ui[k+1] = ui[k] + nzk; 1259 } 1260 1261 #if defined(PETSC_USE_INFO) 1262 if (ai[mbs] != 0) { 1263 PetscReal af = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]); 1264 ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %G needed %G\n",reallocs,fill,af);CHKERRQ(ierr); 1265 ierr = PetscInfo1(A,"Run with -pc_factor_fill %G or use \n",af);CHKERRQ(ierr); 1266 ierr = PetscInfo1(A,"PCFactorSetFill(pc,%G) for best performance.\n",af);CHKERRQ(ierr); 1267 } else { 1268 ierr = PetscInfo(A,"Empty matrix.\n");CHKERRQ(ierr); 1269 } 1270 #endif 1271 1272 ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr); 1273 ierr = PetscFree(jl);CHKERRQ(ierr); 1274 1275 /* destroy list of free space and other temporary array(s) */ 1276 ierr = PetscMalloc((ui[mbs]+1)*sizeof(PetscInt),&uj);CHKERRQ(ierr); 1277 ierr = PetscFreeSpaceContiguous(&free_space,uj);CHKERRQ(ierr); 1278 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 1279 1280 /* put together the new matrix in MATSEQSBAIJ format */ 1281 B = fact; 1282 ierr = MatSeqSBAIJSetPreallocation(B,bs,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr); 1283 1284 b = (Mat_SeqSBAIJ*)B->data; 1285 b->singlemalloc = PETSC_FALSE; 1286 b->free_a = PETSC_TRUE; 1287 b->free_ij = PETSC_TRUE; 1288 ierr = PetscMalloc((ui[mbs]+1)*sizeof(MatScalar),&b->a);CHKERRQ(ierr); 1289 b->j = uj; 1290 b->i = ui; 1291 b->diag = 0; 1292 b->ilen = 0; 1293 b->imax = 0; 1294 b->row = perm; 1295 b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */ 1296 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 1297 b->icol = perm; 1298 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 1299 ierr = PetscMalloc((mbs+1)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr); 1300 ierr = PetscLogObjectMemory(B,(ui[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr); 1301 b->maxnz = b->nz = ui[mbs]; 1302 1303 B->info.factor_mallocs = reallocs; 1304 B->info.fill_ratio_given = fill; 1305 if (ai[mbs] != 0) { 1306 B->info.fill_ratio_needed = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]); 1307 } else { 1308 B->info.fill_ratio_needed = 0.0; 1309 } 1310 if (perm_identity){ 1311 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqBAIJ_N_NaturalOrdering; 1312 } else { 1313 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqBAIJ_N; 1314 } 1315 PetscFunctionReturn(0); 1316 } 1317 1318 /* --------------------------------------------------------- */ 1319 #undef __FUNCT__ 1320 #define __FUNCT__ "MatSolve_SeqBAIJ_N_NaturalOrdering_newdatastruct" 1321 PetscErrorCode MatSolve_SeqBAIJ_N_NaturalOrdering_newdatastruct(Mat A,Vec bb,Vec xx) 1322 { 1323 Mat_SeqBAIJ *a=(Mat_SeqBAIJ *)A->data; 1324 PetscErrorCode ierr; 1325 const PetscInt *ai=a->i,*aj=a->j,*vi; 1326 PetscInt i,k,n=a->mbs; 1327 PetscInt nz,bs=A->rmap->bs,bs2=a->bs2,*adiag=a->diag; 1328 MatScalar *aa=a->a,*v; 1329 PetscScalar *x,*b,*s,*t,*ls; 1330 1331 PetscFunctionBegin; 1332 /* printf("MatSolve_SeqBAIJ_NaturalOrdering_iludt..bs %d\n",bs); */ 1333 ierr = VecGetArray(bb,&b);CHKERRQ(ierr); 1334 ierr = VecGetArray(xx,&x);CHKERRQ(ierr); 1335 t = a->solve_work; 1336 1337 /* forward solve the lower triangular */ 1338 ierr = PetscMemcpy(t,b,bs*sizeof(PetscScalar));CHKERRQ(ierr); /* copy 1st block of b to t */ 1339 1340 for (i=1; i<n; i++) { 1341 v = aa + bs2*ai[i]; 1342 vi = aj + ai[i]; 1343 nz = ai[i+1] - ai[i]; 1344 s = t + bs*i; 1345 ierr = PetscMemcpy(s,b+bs*i,bs*sizeof(PetscScalar));CHKERRQ(ierr); /* copy i_th block of b to t */ 1346 for(k=0;k<nz;k++){ 1347 Kernel_v_gets_v_minus_A_times_w(bs,s,v,t+bs*vi[k]); 1348 v += bs2; 1349 } 1350 } 1351 1352 /* backward solve the upper triangular */ 1353 ls = a->solve_work + A->cmap->n; 1354 for (i=n-1; i>=0; i--){ 1355 v = aa + bs2*ai[2*n-i]; 1356 vi = aj + ai[2*n-i]; 1357 nz = ai[2*n-i +1] - ai[2*n-i]-1; 1358 ierr = PetscMemcpy(ls,t+i*bs,bs*sizeof(PetscScalar));CHKERRQ(ierr); 1359 for(k=0;k<nz;k++){ 1360 Kernel_v_gets_v_minus_A_times_w(bs,ls,v,t+bs*vi[k]); 1361 v += bs2; 1362 } 1363 Kernel_w_gets_A_times_v(bs,ls,aa+bs2*adiag[i],t+i*bs); /* *inv(diagonal[i]) */ 1364 ierr = PetscMemcpy(x+i*bs,t+i*bs,bs*sizeof(PetscScalar));CHKERRQ(ierr); 1365 } 1366 1367 ierr = VecRestoreArray(bb,&b);CHKERRQ(ierr); 1368 ierr = VecRestoreArray(xx,&x);CHKERRQ(ierr); 1369 ierr = PetscLogFlops(2.0*(a->bs2)*(a->nz) - A->rmap->bs*A->cmap->n);CHKERRQ(ierr); 1370 PetscFunctionReturn(0); 1371 } 1372 1373 #undef __FUNCT__ 1374 #define __FUNCT__ "MatSolve_SeqBAIJ_N_newdatastruct" 1375 PetscErrorCode MatSolve_SeqBAIJ_N_newdatastruct(Mat A,Vec bb,Vec xx) 1376 { 1377 Mat_SeqBAIJ *a=(Mat_SeqBAIJ *)A->data; 1378 IS iscol=a->col,isrow=a->row; 1379 PetscErrorCode ierr; 1380 const PetscInt *r,*c,*rout,*cout,*ai=a->i,*aj=a->j,*vi; 1381 PetscInt i,m,n=a->mbs; 1382 PetscInt nz,bs=A->rmap->bs,bs2=a->bs2,k; 1383 MatScalar *aa=a->a,*v; 1384 PetscScalar *x,*b,*s,*t,*ls; 1385 1386 PetscFunctionBegin; 1387 ierr = VecGetArray(bb,&b);CHKERRQ(ierr); 1388 ierr = VecGetArray(xx,&x);CHKERRQ(ierr); 1389 t = a->solve_work; 1390 1391 ierr = ISGetIndices(isrow,&rout);CHKERRQ(ierr); r = rout; 1392 ierr = ISGetIndices(iscol,&cout);CHKERRQ(ierr); c = cout; 1393 1394 /* forward solve the lower triangular */ 1395 ierr = PetscMemcpy(t,b+bs*r[0],bs*sizeof(PetscScalar));CHKERRQ(ierr); 1396 for (i=1; i<n; i++) { 1397 v = aa + bs2*ai[i]; 1398 vi = aj + ai[i]; 1399 nz = ai[i+1] - ai[i]; 1400 s = t + bs*i; 1401 ierr = PetscMemcpy(s,b+bs*r[i],bs*sizeof(PetscScalar));CHKERRQ(ierr); 1402 for(m=0;m<nz;m++){ 1403 Kernel_v_gets_v_minus_A_times_w(bs,s,v,t+bs*vi[m]); 1404 v += bs2; 1405 } 1406 } 1407 1408 /* backward solve the upper triangular */ 1409 ls = a->solve_work + A->cmap->n; 1410 for (i=n-1; i>=0; i--){ 1411 k = 2*n-i; 1412 v = aa + bs2*ai[k]; 1413 vi = aj + ai[k]; 1414 nz = ai[k+1] - ai[k] - 1; 1415 ierr = PetscMemcpy(ls,t+i*bs,bs*sizeof(PetscScalar));CHKERRQ(ierr); 1416 for(m=0;m<nz;m++){ 1417 Kernel_v_gets_v_minus_A_times_w(bs,ls,v,t+bs*vi[m]); 1418 v += bs2; 1419 } 1420 Kernel_w_gets_A_times_v(bs,ls,v,t+i*bs); /* *inv(diagonal[i]) */ 1421 ierr = PetscMemcpy(x + bs*c[i],t+i*bs,bs*sizeof(PetscScalar));CHKERRQ(ierr); 1422 } 1423 ierr = ISRestoreIndices(isrow,&rout);CHKERRQ(ierr); 1424 ierr = ISRestoreIndices(iscol,&cout);CHKERRQ(ierr); 1425 ierr = VecRestoreArray(bb,&b);CHKERRQ(ierr); 1426 ierr = VecRestoreArray(xx,&x);CHKERRQ(ierr); 1427 ierr = PetscLogFlops(2.0*(a->bs2)*(a->nz) - A->rmap->bs*A->cmap->n);CHKERRQ(ierr); 1428 PetscFunctionReturn(0); 1429 } 1430 1431 #undef __FUNCT__ 1432 #define __FUNCT__ "BlockAbs_privat" 1433 PetscErrorCode BlockAbs_private(PetscInt nbs,PetscInt bs2,PetscScalar *blockarray,PetscReal *absarray) 1434 { 1435 PetscErrorCode ierr; 1436 PetscInt i,j; 1437 PetscFunctionBegin; 1438 ierr = PetscMemzero(absarray,(nbs+1)*sizeof(PetscReal));CHKERRQ(ierr); 1439 for (i=0; i<nbs; i++){ 1440 for (j=0; j<bs2; j++){ 1441 if (absarray[i] < PetscAbsScalar(blockarray[i*nbs+j])) absarray[i] = PetscAbsScalar(blockarray[i*nbs+j]); 1442 } 1443 } 1444 PetscFunctionReturn(0); 1445 } 1446 1447 #undef __FUNCT__ 1448 #define __FUNCT__ "MatILUDTFactor_SeqBAIJ" 1449 PetscErrorCode MatILUDTFactor_SeqBAIJ(Mat A,IS isrow,IS iscol,const MatFactorInfo *info,Mat *fact) 1450 { 1451 Mat B = *fact; 1452 Mat_SeqBAIJ *a=(Mat_SeqBAIJ*)A->data,*b; 1453 IS isicol; 1454 PetscErrorCode ierr; 1455 const PetscInt *r,*ic; 1456 PetscInt i,mbs=a->mbs,bs=A->rmap->bs,bs2=a->bs2,*ai=a->i,*aj=a->j,*ajtmp,*adiag; 1457 PetscInt *bi,*bj,*bdiag; 1458 1459 PetscInt row,nzi,nzi_bl,nzi_bu,*im,dtcount,nzi_al,nzi_au; 1460 PetscInt nlnk,*lnk; 1461 PetscBT lnkbt; 1462 PetscTruth row_identity,icol_identity,both_identity; 1463 MatScalar *aatmp,*pv,*batmp,*ba,*rtmp,*pc,*multiplier,*vtmp; 1464 const PetscInt *ics; 1465 PetscInt j,nz,*pj,*bjtmp,k,ncut,*jtmp; 1466 1467 PetscReal dt=info->dt; /* shift=info->shiftinblocks; */ 1468 PetscInt nnz_max; 1469 PetscTruth missing; 1470 PetscReal *vtmp_abs; 1471 MatScalar *v_work; 1472 PetscInt *v_pivots; 1473 1474 PetscFunctionBegin; 1475 /* ------- symbolic factorization, can be reused ---------*/ 1476 ierr = MatMissingDiagonal(A,&missing,&i);CHKERRQ(ierr); 1477 if (missing) SETERRQ1(PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",i); 1478 adiag=a->diag; 1479 1480 ierr = ISInvertPermutation(iscol,PETSC_DECIDE,&isicol);CHKERRQ(ierr); 1481 1482 /* bdiag is location of diagonal in factor */ 1483 ierr = PetscMalloc((mbs+1)*sizeof(PetscInt),&bdiag);CHKERRQ(ierr); 1484 1485 /* allocate row pointers bi */ 1486 ierr = PetscMalloc((2*mbs+2)*sizeof(PetscInt),&bi);CHKERRQ(ierr); 1487 1488 /* allocate bj and ba; max num of nonzero entries is (ai[n]+2*n*dtcount+2) */ 1489 dtcount = (PetscInt)info->dtcount; 1490 if (dtcount > mbs-1) dtcount = mbs-1; 1491 nnz_max = ai[mbs]+2*mbs*dtcount +2; 1492 /* printf("MatILUDTFactor_SeqBAIJ, bs %d, ai[mbs] %d, nnz_max %d, dtcount %d\n",bs,ai[mbs],nnz_max,dtcount); */ 1493 ierr = PetscMalloc(nnz_max*sizeof(PetscInt),&bj);CHKERRQ(ierr); 1494 nnz_max = nnz_max*bs2; 1495 ierr = PetscMalloc(nnz_max*sizeof(MatScalar),&ba);CHKERRQ(ierr); 1496 1497 /* put together the new matrix */ 1498 ierr = MatSeqBAIJSetPreallocation_SeqBAIJ(B,bs,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr); 1499 ierr = PetscLogObjectParent(B,isicol);CHKERRQ(ierr); 1500 b = (Mat_SeqBAIJ*)(B)->data; 1501 b->free_a = PETSC_TRUE; 1502 b->free_ij = PETSC_TRUE; 1503 b->singlemalloc = PETSC_FALSE; 1504 b->a = ba; 1505 b->j = bj; 1506 b->i = bi; 1507 b->diag = bdiag; 1508 b->ilen = 0; 1509 b->imax = 0; 1510 b->row = isrow; 1511 b->col = iscol; 1512 ierr = PetscObjectReference((PetscObject)isrow);CHKERRQ(ierr); 1513 ierr = PetscObjectReference((PetscObject)iscol);CHKERRQ(ierr); 1514 b->icol = isicol; 1515 ierr = PetscMalloc((bs*(mbs+1))*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr); 1516 1517 ierr = PetscLogObjectMemory(B,nnz_max*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr); 1518 b->maxnz = nnz_max/bs2; 1519 1520 (B)->factor = MAT_FACTOR_ILUDT; 1521 (B)->info.factor_mallocs = 0; 1522 (B)->info.fill_ratio_given = ((PetscReal)nnz_max)/((PetscReal)(ai[mbs]*bs2)); 1523 CHKMEMQ; 1524 /* ------- end of symbolic factorization ---------*/ 1525 ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr); 1526 ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr); 1527 ics = ic; 1528 1529 /* linked list for storing column indices of the active row */ 1530 nlnk = mbs + 1; 1531 ierr = PetscLLCreate(mbs,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 1532 1533 /* im: used by PetscLLAddSortedLU(); jtmp: working array for column indices of active row */ 1534 ierr = PetscMalloc((2*mbs+1)*sizeof(PetscInt),&im);CHKERRQ(ierr); 1535 jtmp = im + mbs; 1536 /* rtmp, vtmp: working arrays for sparse and contiguous row entries of active row */ 1537 ierr = PetscMalloc((2*mbs*bs2+1)*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 1538 vtmp = rtmp + bs2*mbs; 1539 ierr = PetscMalloc((mbs+1)*sizeof(PetscReal),&vtmp_abs);CHKERRQ(ierr); 1540 1541 ierr = PetscMalloc(bs*sizeof(PetscInt) + (bs+bs2)*sizeof(MatScalar),&v_work);CHKERRQ(ierr); 1542 multiplier = v_work + bs; 1543 v_pivots = (PetscInt*)(multiplier + bs2); 1544 1545 bi[0] = 0; 1546 bdiag[0] = (nnz_max/bs2)-1; /* location of diagonal in factor B */ 1547 bi[2*mbs+1] = bdiag[0]+1; /* endof bj and ba array */ 1548 for (i=0; i<mbs; i++) { 1549 /* copy initial fill into linked list */ 1550 nzi = 0; /* nonzeros for active row i */ 1551 nzi = ai[r[i]+1] - ai[r[i]]; 1552 if (!nzi) SETERRQ2(PETSC_ERR_MAT_LU_ZRPVT,"Empty row in matrix: row in original ordering %D in permuted ordering %D",r[i],i); 1553 nzi_al = adiag[r[i]] - ai[r[i]]; 1554 nzi_au = ai[r[i]+1] - adiag[r[i]] -1; 1555 /* printf("row %d, nzi_al/au %d %d\n",i,nzi_al,nzi_au); */ 1556 1557 /* load in initial unfactored row */ 1558 ajtmp = aj + ai[r[i]]; 1559 ierr = PetscLLAddPerm(nzi,ajtmp,ic,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 1560 ierr = PetscMemzero(rtmp,(mbs*bs2+1)*sizeof(PetscScalar));CHKERRQ(ierr); 1561 aatmp = a->a + bs2*ai[r[i]]; 1562 for (j=0; j<nzi; j++) { 1563 ierr = PetscMemcpy(rtmp+bs2*ic[ajtmp[j]],aatmp+bs2*j,bs2*sizeof(MatScalar));CHKERRQ(ierr); 1564 } 1565 1566 /* add pivot rows into linked list */ 1567 row = lnk[mbs]; 1568 while (row < i) { 1569 nzi_bl = bi[row+1] - bi[row] + 1; 1570 bjtmp = bj + bdiag[row+1]+1; /* points to 1st column next to the diagonal in U */ 1571 ierr = PetscLLAddSortedLU(bjtmp,row,nlnk,lnk,lnkbt,i,nzi_bl,im);CHKERRQ(ierr); 1572 nzi += nlnk; 1573 row = lnk[row]; 1574 } 1575 1576 /* copy data from lnk into jtmp, then initialize lnk */ 1577 ierr = PetscLLClean(mbs,mbs,nzi,lnk,jtmp,lnkbt);CHKERRQ(ierr); 1578 1579 /* numerical factorization */ 1580 bjtmp = jtmp; 1581 row = *bjtmp++; /* 1st pivot row */ 1582 1583 while (row < i) { 1584 pc = rtmp + bs2*row; 1585 pv = ba + bs2*bdiag[row]; /* inv(diag) of the pivot row */ 1586 Kernel_A_gets_A_times_B(bs,pc,pv,multiplier); /* pc= multiplier = pc*inv(diag[row]) */ 1587 ierr = BlockAbs_private(1,bs2,pc,vtmp_abs);CHKERRQ(ierr); 1588 if (vtmp_abs[0] > dt){ /* apply tolerance dropping rule */ 1589 pj = bj + bdiag[row+1] + 1; /* point to 1st entry of U(row,:) */ 1590 pv = ba + bs2*(bdiag[row+1] + 1); 1591 nz = bdiag[row] - bdiag[row+1] - 1; /* num of entries in U(row,:), excluding diagonal */ 1592 for (j=0; j<nz; j++){ 1593 Kernel_A_gets_A_minus_B_times_C(bs,rtmp+bs2*pj[j],pc,pv+bs2*j); 1594 } 1595 /* ierr = PetscLogFlops(bslog*(nz+1.0)-bs);CHKERRQ(ierr); */ 1596 } 1597 row = *bjtmp++; 1598 } 1599 1600 /* copy sparse rtmp into contiguous vtmp; separate L and U part */ 1601 nzi_bl = 0; j = 0; 1602 while (jtmp[j] < i){ /* L-part. Note: jtmp is sorted */ 1603 ierr = PetscMemcpy(vtmp+bs2*j,rtmp+bs2*jtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 1604 nzi_bl++; j++; 1605 } 1606 nzi_bu = nzi - nzi_bl -1; 1607 /* printf("nzi %d, nzi_bl %d, nzi_bu %d\n",nzi,nzi_bl,nzi_bu); */ 1608 1609 while (j < nzi){ /* U-part */ 1610 ierr = PetscMemcpy(vtmp+bs2*j,rtmp+bs2*jtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 1611 /* 1612 printf(" col %d: ",jtmp[j]); 1613 for (j1=0; j1<bs2; j1++) printf(" %g",*(vtmp+bs2*j+j1)); 1614 printf(" \n"); 1615 */ 1616 j++; 1617 } 1618 1619 ierr = BlockAbs_private(nzi,bs2,vtmp,vtmp_abs);CHKERRQ(ierr); 1620 /* 1621 printf(" row %d, nzi %d, vtmp_abs\n",i,nzi); 1622 for (j1=0; j1<nzi; j1++) printf(" (%d %g),",jtmp[j1],vtmp_abs[j1]); 1623 printf(" \n"); 1624 */ 1625 bjtmp = bj + bi[i]; 1626 batmp = ba + bs2*bi[i]; 1627 /* apply level dropping rule to L part */ 1628 ncut = nzi_al + dtcount; 1629 if (ncut < nzi_bl){ 1630 ierr = PetscSortSplitReal(ncut,nzi_bl,vtmp_abs,jtmp);CHKERRQ(ierr); 1631 ierr = PetscSortIntWithScalarArray(ncut,jtmp,vtmp);CHKERRQ(ierr); 1632 } else { 1633 ncut = nzi_bl; 1634 } 1635 for (j=0; j<ncut; j++){ 1636 bjtmp[j] = jtmp[j]; 1637 ierr = PetscMemcpy(batmp+bs2*j,rtmp+bs2*bjtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr); 1638 /* 1639 printf(" col %d: ",bjtmp[j]); 1640 for (j1=0; j1<bs2; j1++) printf(" %g,",*(batmp+bs2*j+j1)); 1641 printf("\n"); 1642 */ 1643 } 1644 bi[i+1] = bi[i] + ncut; 1645 nzi = ncut + 1; 1646 1647 /* apply level dropping rule to U part */ 1648 ncut = nzi_au + dtcount; 1649 if (ncut < nzi_bu){ 1650 ierr = PetscSortSplitReal(ncut,nzi_bu,vtmp_abs+nzi_bl+1,jtmp+nzi_bl+1);CHKERRQ(ierr); 1651 ierr = PetscSortIntWithScalarArray(ncut,jtmp+nzi_bl+1,vtmp+nzi_bl+1);CHKERRQ(ierr); 1652 } else { 1653 ncut = nzi_bu; 1654 } 1655 nzi += ncut; 1656 1657 /* mark bdiagonal */ 1658 bdiag[i+1] = bdiag[i] - (ncut + 1); 1659 bi[2*mbs - i] = bi[2*mbs - i +1] - (ncut + 1); 1660 1661 bjtmp = bj + bdiag[i]; 1662 batmp = ba + bs2*bdiag[i]; 1663 ierr = PetscMemcpy(batmp,rtmp+bs2*i,bs2*sizeof(MatScalar));CHKERRQ(ierr); 1664 *bjtmp = i; 1665 /* 1666 printf(" diag %d: ",*bjtmp); 1667 for (j=0; j<bs2; j++){ 1668 printf(" %g,",batmp[j]); 1669 } 1670 printf("\n"); 1671 */ 1672 bjtmp = bj + bdiag[i+1]+1; 1673 batmp = ba + (bdiag[i+1]+1)*bs2; 1674 1675 for (k=0; k<ncut; k++){ 1676 bjtmp[k] = jtmp[nzi_bl+1+k]; 1677 ierr = PetscMemcpy(batmp+bs2*k,rtmp+bs2*bjtmp[k],bs2*sizeof(MatScalar));CHKERRQ(ierr); 1678 /* 1679 printf(" col %d:",bjtmp[k]); 1680 for (j1=0; j1<bs2; j1++) printf(" %g,",*(batmp+bs2*k+j1)); 1681 printf("\n"); 1682 */ 1683 } 1684 1685 im[i] = nzi; /* used by PetscLLAddSortedLU() */ 1686 1687 /* invert diagonal block for simplier triangular solves - add shift??? */ 1688 batmp = ba + bs2*bdiag[i]; 1689 ierr = Kernel_A_gets_inverse_A(bs,batmp,v_pivots,v_work);CHKERRQ(ierr); 1690 } /* for (i=0; i<mbs; i++) */ 1691 1692 /* printf("end of L %d, beginning of U %d\n",bi[mbs],bdiag[mbs]); */ 1693 if (bi[mbs] >= bdiag[mbs]) SETERRQ2(PETSC_ERR_ARG_SIZ,"end of L array %d cannot >= the beginning of U array %d",bi[mbs],bdiag[mbs]); 1694 1695 ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr); 1696 ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr); 1697 1698 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 1699 1700 ierr = PetscFree(im);CHKERRQ(ierr); 1701 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1702 1703 ierr = PetscLogFlops(bs2*B->cmap->n);CHKERRQ(ierr); 1704 b->maxnz = b->nz = bi[mbs] + bdiag[0] - bdiag[mbs]; 1705 1706 ierr = ISIdentity(isrow,&row_identity);CHKERRQ(ierr); 1707 ierr = ISIdentity(isicol,&icol_identity);CHKERRQ(ierr); 1708 both_identity = (PetscTruth) (row_identity && icol_identity); 1709 if (row_identity && icol_identity) { 1710 B->ops->solve = MatSolve_SeqBAIJ_N_NaturalOrdering_newdatastruct; 1711 } else { 1712 B->ops->solve = MatSolve_SeqBAIJ_N_newdatastruct; 1713 } 1714 1715 B->ops->solveadd = 0; 1716 B->ops->solvetranspose = 0; 1717 B->ops->solvetransposeadd = 0; 1718 B->ops->matsolve = 0; 1719 B->assembled = PETSC_TRUE; 1720 B->preallocated = PETSC_TRUE; 1721 PetscFunctionReturn(0); 1722 } 1723