1 /* 2 Defines the basic matrix operations for the BAIJ (compressed row) 3 matrix storage format. 4 */ 5 #include <../src/mat/impls/baij/seq/baij.h> /*I "petscmat.h" I*/ 6 #include <petscblaslapack.h> 7 #include <petsc/private/kernels/blockinvert.h> 8 #include <petsc/private/kernels/blockmatmult.h> 9 10 /* defines MatSetValues_Seq_Hash(), MatAssemblyEnd_Seq_Hash(), MatSetUp_Seq_Hash() */ 11 #define TYPE BAIJ 12 #define TYPE_BS 13 #include "../src/mat/impls/aij/seq/seqhashmatsetvalues.h" 14 #undef TYPE_BS 15 #define TYPE_BS _BS 16 #define TYPE_BS_ON 17 #include "../src/mat/impls/aij/seq/seqhashmatsetvalues.h" 18 #undef TYPE_BS 19 #include "../src/mat/impls/aij/seq/seqhashmat.h" 20 #undef TYPE 21 #undef TYPE_BS_ON 22 23 #if defined(PETSC_HAVE_HYPRE) 24 PETSC_INTERN PetscErrorCode MatConvert_AIJ_HYPRE(Mat, MatType, MatReuse, Mat *); 25 #endif 26 27 #if defined(PETSC_HAVE_MKL_SPARSE_OPTIMIZE) 28 PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqBAIJMKL(Mat, MatType, MatReuse, Mat *); 29 #endif 30 PETSC_INTERN PetscErrorCode MatConvert_XAIJ_IS(Mat, MatType, MatReuse, Mat *); 31 32 static PetscErrorCode MatGetColumnReductions_SeqBAIJ(Mat A, PetscInt type, PetscReal *reductions) 33 { 34 Mat_SeqBAIJ *a_aij = (Mat_SeqBAIJ *)A->data; 35 PetscInt m, n, ib, jb, bs = A->rmap->bs; 36 MatScalar *a_val = a_aij->a; 37 38 PetscFunctionBegin; 39 PetscCall(MatGetSize(A, &m, &n)); 40 PetscCall(PetscArrayzero(reductions, n)); 41 if (type == NORM_2) { 42 for (PetscInt i = a_aij->i[0]; i < a_aij->i[A->rmap->n / bs]; i++) { 43 for (jb = 0; jb < bs; jb++) { 44 for (ib = 0; ib < bs; ib++) { 45 reductions[A->cmap->rstart + a_aij->j[i] * bs + jb] += PetscAbsScalar(*a_val * *a_val); 46 a_val++; 47 } 48 } 49 } 50 } else if (type == NORM_1) { 51 for (PetscInt i = a_aij->i[0]; i < a_aij->i[A->rmap->n / bs]; i++) { 52 for (jb = 0; jb < bs; jb++) { 53 for (ib = 0; ib < bs; ib++) { 54 reductions[A->cmap->rstart + a_aij->j[i] * bs + jb] += PetscAbsScalar(*a_val); 55 a_val++; 56 } 57 } 58 } 59 } else if (type == NORM_INFINITY) { 60 for (PetscInt i = a_aij->i[0]; i < a_aij->i[A->rmap->n / bs]; i++) { 61 for (jb = 0; jb < bs; jb++) { 62 for (ib = 0; ib < bs; ib++) { 63 int col = A->cmap->rstart + a_aij->j[i] * bs + jb; 64 reductions[col] = PetscMax(PetscAbsScalar(*a_val), reductions[col]); 65 a_val++; 66 } 67 } 68 } 69 } else if (type == REDUCTION_SUM_REALPART || type == REDUCTION_MEAN_REALPART) { 70 for (PetscInt i = a_aij->i[0]; i < a_aij->i[A->rmap->n / bs]; i++) { 71 for (jb = 0; jb < bs; jb++) { 72 for (ib = 0; ib < bs; ib++) { 73 reductions[A->cmap->rstart + a_aij->j[i] * bs + jb] += PetscRealPart(*a_val); 74 a_val++; 75 } 76 } 77 } 78 } else if (type == REDUCTION_SUM_IMAGINARYPART || type == REDUCTION_MEAN_IMAGINARYPART) { 79 for (PetscInt i = a_aij->i[0]; i < a_aij->i[A->rmap->n / bs]; i++) { 80 for (jb = 0; jb < bs; jb++) { 81 for (ib = 0; ib < bs; ib++) { 82 reductions[A->cmap->rstart + a_aij->j[i] * bs + jb] += PetscImaginaryPart(*a_val); 83 a_val++; 84 } 85 } 86 } 87 } else SETERRQ(PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONG, "Unknown reduction type"); 88 if (type == NORM_2) { 89 for (PetscInt i = 0; i < n; i++) reductions[i] = PetscSqrtReal(reductions[i]); 90 } else if (type == REDUCTION_MEAN_REALPART || type == REDUCTION_MEAN_IMAGINARYPART) { 91 for (PetscInt i = 0; i < n; i++) reductions[i] /= m; 92 } 93 PetscFunctionReturn(PETSC_SUCCESS); 94 } 95 96 static PetscErrorCode MatInvertBlockDiagonal_SeqBAIJ(Mat A, const PetscScalar **values) 97 { 98 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 99 PetscInt *diag_offset, i, bs = A->rmap->bs, mbs = a->mbs, ipvt[5], bs2 = bs * bs, *v_pivots; 100 MatScalar *v = a->a, *odiag, *diag, work[25], *v_work; 101 PetscReal shift = 0.0; 102 PetscBool allowzeropivot, zeropivotdetected = PETSC_FALSE; 103 104 PetscFunctionBegin; 105 allowzeropivot = PetscNot(A->erroriffailure); 106 107 if (a->idiagvalid) { 108 if (values) *values = a->idiag; 109 PetscFunctionReturn(PETSC_SUCCESS); 110 } 111 PetscCall(MatMarkDiagonal_SeqBAIJ(A)); 112 diag_offset = a->diag; 113 if (!a->idiag) { PetscCall(PetscMalloc1(bs2 * mbs, &a->idiag)); } 114 diag = a->idiag; 115 if (values) *values = a->idiag; 116 /* factor and invert each block */ 117 switch (bs) { 118 case 1: 119 for (i = 0; i < mbs; i++) { 120 odiag = v + 1 * diag_offset[i]; 121 diag[0] = odiag[0]; 122 123 if (PetscAbsScalar(diag[0] + shift) < PETSC_MACHINE_EPSILON) { 124 if (allowzeropivot) { 125 A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 126 A->factorerror_zeropivot_value = PetscAbsScalar(diag[0]); 127 A->factorerror_zeropivot_row = i; 128 PetscCall(PetscInfo(A, "Zero pivot, row %" PetscInt_FMT "\n", i)); 129 } else SETERRQ(PETSC_COMM_SELF, PETSC_ERR_MAT_LU_ZRPVT, "Zero pivot, row %" PetscInt_FMT " pivot value %g tolerance %g", i, (double)PetscAbsScalar(diag[0]), (double)PETSC_MACHINE_EPSILON); 130 } 131 132 diag[0] = (PetscScalar)1.0 / (diag[0] + shift); 133 diag += 1; 134 } 135 break; 136 case 2: 137 for (i = 0; i < mbs; i++) { 138 odiag = v + 4 * diag_offset[i]; 139 diag[0] = odiag[0]; 140 diag[1] = odiag[1]; 141 diag[2] = odiag[2]; 142 diag[3] = odiag[3]; 143 PetscCall(PetscKernel_A_gets_inverse_A_2(diag, shift, allowzeropivot, &zeropivotdetected)); 144 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 145 diag += 4; 146 } 147 break; 148 case 3: 149 for (i = 0; i < mbs; i++) { 150 odiag = v + 9 * diag_offset[i]; 151 diag[0] = odiag[0]; 152 diag[1] = odiag[1]; 153 diag[2] = odiag[2]; 154 diag[3] = odiag[3]; 155 diag[4] = odiag[4]; 156 diag[5] = odiag[5]; 157 diag[6] = odiag[6]; 158 diag[7] = odiag[7]; 159 diag[8] = odiag[8]; 160 PetscCall(PetscKernel_A_gets_inverse_A_3(diag, shift, allowzeropivot, &zeropivotdetected)); 161 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 162 diag += 9; 163 } 164 break; 165 case 4: 166 for (i = 0; i < mbs; i++) { 167 odiag = v + 16 * diag_offset[i]; 168 PetscCall(PetscArraycpy(diag, odiag, 16)); 169 PetscCall(PetscKernel_A_gets_inverse_A_4(diag, shift, allowzeropivot, &zeropivotdetected)); 170 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 171 diag += 16; 172 } 173 break; 174 case 5: 175 for (i = 0; i < mbs; i++) { 176 odiag = v + 25 * diag_offset[i]; 177 PetscCall(PetscArraycpy(diag, odiag, 25)); 178 PetscCall(PetscKernel_A_gets_inverse_A_5(diag, ipvt, work, shift, allowzeropivot, &zeropivotdetected)); 179 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 180 diag += 25; 181 } 182 break; 183 case 6: 184 for (i = 0; i < mbs; i++) { 185 odiag = v + 36 * diag_offset[i]; 186 PetscCall(PetscArraycpy(diag, odiag, 36)); 187 PetscCall(PetscKernel_A_gets_inverse_A_6(diag, shift, allowzeropivot, &zeropivotdetected)); 188 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 189 diag += 36; 190 } 191 break; 192 case 7: 193 for (i = 0; i < mbs; i++) { 194 odiag = v + 49 * diag_offset[i]; 195 PetscCall(PetscArraycpy(diag, odiag, 49)); 196 PetscCall(PetscKernel_A_gets_inverse_A_7(diag, shift, allowzeropivot, &zeropivotdetected)); 197 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 198 diag += 49; 199 } 200 break; 201 default: 202 PetscCall(PetscMalloc2(bs, &v_work, bs, &v_pivots)); 203 for (i = 0; i < mbs; i++) { 204 odiag = v + bs2 * diag_offset[i]; 205 PetscCall(PetscArraycpy(diag, odiag, bs2)); 206 PetscCall(PetscKernel_A_gets_inverse_A(bs, diag, v_pivots, v_work, allowzeropivot, &zeropivotdetected)); 207 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 208 diag += bs2; 209 } 210 PetscCall(PetscFree2(v_work, v_pivots)); 211 } 212 a->idiagvalid = PETSC_TRUE; 213 PetscFunctionReturn(PETSC_SUCCESS); 214 } 215 216 static PetscErrorCode MatSOR_SeqBAIJ(Mat A, Vec bb, PetscReal omega, MatSORType flag, PetscReal fshift, PetscInt its, PetscInt lits, Vec xx) 217 { 218 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 219 PetscScalar *x, *work, *w, *workt, *t; 220 const MatScalar *v, *aa = a->a, *idiag; 221 const PetscScalar *b, *xb; 222 PetscScalar s[7], xw[7] = {0}; /* avoid some compilers thinking xw is uninitialized */ 223 PetscInt m = a->mbs, i, i2, nz, bs = A->rmap->bs, bs2 = bs * bs, k, j, idx, it; 224 const PetscInt *diag, *ai = a->i, *aj = a->j, *vi; 225 226 PetscFunctionBegin; 227 its = its * lits; 228 PetscCheck(!(flag & SOR_EISENSTAT), PETSC_COMM_SELF, PETSC_ERR_SUP, "No support yet for Eisenstat"); 229 PetscCheck(its > 0, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "Relaxation requires global its %" PetscInt_FMT " and local its %" PetscInt_FMT " both positive", its, lits); 230 PetscCheck(!fshift, PETSC_COMM_SELF, PETSC_ERR_SUP, "No support for diagonal shift"); 231 PetscCheck(omega == 1.0, PETSC_COMM_SELF, PETSC_ERR_SUP, "No support for non-trivial relaxation factor"); 232 PetscCheck(!(flag & SOR_APPLY_UPPER) && !(flag & SOR_APPLY_LOWER), PETSC_COMM_SELF, PETSC_ERR_SUP, "No support for applying upper or lower triangular parts"); 233 234 if (!a->idiagvalid) PetscCall(MatInvertBlockDiagonal(A, NULL)); 235 236 if (!m) PetscFunctionReturn(PETSC_SUCCESS); 237 diag = a->diag; 238 idiag = a->idiag; 239 k = PetscMax(A->rmap->n, A->cmap->n); 240 if (!a->mult_work) PetscCall(PetscMalloc1(k + 1, &a->mult_work)); 241 if (!a->sor_workt) PetscCall(PetscMalloc1(k, &a->sor_workt)); 242 if (!a->sor_work) PetscCall(PetscMalloc1(bs, &a->sor_work)); 243 work = a->mult_work; 244 t = a->sor_workt; 245 w = a->sor_work; 246 247 PetscCall(VecGetArray(xx, &x)); 248 PetscCall(VecGetArrayRead(bb, &b)); 249 250 if (flag & SOR_ZERO_INITIAL_GUESS) { 251 if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) { 252 switch (bs) { 253 case 1: 254 PetscKernel_v_gets_A_times_w_1(x, idiag, b); 255 t[0] = b[0]; 256 i2 = 1; 257 idiag += 1; 258 for (i = 1; i < m; i++) { 259 v = aa + ai[i]; 260 vi = aj + ai[i]; 261 nz = diag[i] - ai[i]; 262 s[0] = b[i2]; 263 for (j = 0; j < nz; j++) { 264 xw[0] = x[vi[j]]; 265 PetscKernel_v_gets_v_minus_A_times_w_1(s, (v + j), xw); 266 } 267 t[i2] = s[0]; 268 PetscKernel_v_gets_A_times_w_1(xw, idiag, s); 269 x[i2] = xw[0]; 270 idiag += 1; 271 i2 += 1; 272 } 273 break; 274 case 2: 275 PetscKernel_v_gets_A_times_w_2(x, idiag, b); 276 t[0] = b[0]; 277 t[1] = b[1]; 278 i2 = 2; 279 idiag += 4; 280 for (i = 1; i < m; i++) { 281 v = aa + 4 * ai[i]; 282 vi = aj + ai[i]; 283 nz = diag[i] - ai[i]; 284 s[0] = b[i2]; 285 s[1] = b[i2 + 1]; 286 for (j = 0; j < nz; j++) { 287 idx = 2 * vi[j]; 288 it = 4 * j; 289 xw[0] = x[idx]; 290 xw[1] = x[1 + idx]; 291 PetscKernel_v_gets_v_minus_A_times_w_2(s, (v + it), xw); 292 } 293 t[i2] = s[0]; 294 t[i2 + 1] = s[1]; 295 PetscKernel_v_gets_A_times_w_2(xw, idiag, s); 296 x[i2] = xw[0]; 297 x[i2 + 1] = xw[1]; 298 idiag += 4; 299 i2 += 2; 300 } 301 break; 302 case 3: 303 PetscKernel_v_gets_A_times_w_3(x, idiag, b); 304 t[0] = b[0]; 305 t[1] = b[1]; 306 t[2] = b[2]; 307 i2 = 3; 308 idiag += 9; 309 for (i = 1; i < m; i++) { 310 v = aa + 9 * ai[i]; 311 vi = aj + ai[i]; 312 nz = diag[i] - ai[i]; 313 s[0] = b[i2]; 314 s[1] = b[i2 + 1]; 315 s[2] = b[i2 + 2]; 316 while (nz--) { 317 idx = 3 * (*vi++); 318 xw[0] = x[idx]; 319 xw[1] = x[1 + idx]; 320 xw[2] = x[2 + idx]; 321 PetscKernel_v_gets_v_minus_A_times_w_3(s, v, xw); 322 v += 9; 323 } 324 t[i2] = s[0]; 325 t[i2 + 1] = s[1]; 326 t[i2 + 2] = s[2]; 327 PetscKernel_v_gets_A_times_w_3(xw, idiag, s); 328 x[i2] = xw[0]; 329 x[i2 + 1] = xw[1]; 330 x[i2 + 2] = xw[2]; 331 idiag += 9; 332 i2 += 3; 333 } 334 break; 335 case 4: 336 PetscKernel_v_gets_A_times_w_4(x, idiag, b); 337 t[0] = b[0]; 338 t[1] = b[1]; 339 t[2] = b[2]; 340 t[3] = b[3]; 341 i2 = 4; 342 idiag += 16; 343 for (i = 1; i < m; i++) { 344 v = aa + 16 * ai[i]; 345 vi = aj + ai[i]; 346 nz = diag[i] - ai[i]; 347 s[0] = b[i2]; 348 s[1] = b[i2 + 1]; 349 s[2] = b[i2 + 2]; 350 s[3] = b[i2 + 3]; 351 while (nz--) { 352 idx = 4 * (*vi++); 353 xw[0] = x[idx]; 354 xw[1] = x[1 + idx]; 355 xw[2] = x[2 + idx]; 356 xw[3] = x[3 + idx]; 357 PetscKernel_v_gets_v_minus_A_times_w_4(s, v, xw); 358 v += 16; 359 } 360 t[i2] = s[0]; 361 t[i2 + 1] = s[1]; 362 t[i2 + 2] = s[2]; 363 t[i2 + 3] = s[3]; 364 PetscKernel_v_gets_A_times_w_4(xw, idiag, s); 365 x[i2] = xw[0]; 366 x[i2 + 1] = xw[1]; 367 x[i2 + 2] = xw[2]; 368 x[i2 + 3] = xw[3]; 369 idiag += 16; 370 i2 += 4; 371 } 372 break; 373 case 5: 374 PetscKernel_v_gets_A_times_w_5(x, idiag, b); 375 t[0] = b[0]; 376 t[1] = b[1]; 377 t[2] = b[2]; 378 t[3] = b[3]; 379 t[4] = b[4]; 380 i2 = 5; 381 idiag += 25; 382 for (i = 1; i < m; i++) { 383 v = aa + 25 * ai[i]; 384 vi = aj + ai[i]; 385 nz = diag[i] - ai[i]; 386 s[0] = b[i2]; 387 s[1] = b[i2 + 1]; 388 s[2] = b[i2 + 2]; 389 s[3] = b[i2 + 3]; 390 s[4] = b[i2 + 4]; 391 while (nz--) { 392 idx = 5 * (*vi++); 393 xw[0] = x[idx]; 394 xw[1] = x[1 + idx]; 395 xw[2] = x[2 + idx]; 396 xw[3] = x[3 + idx]; 397 xw[4] = x[4 + idx]; 398 PetscKernel_v_gets_v_minus_A_times_w_5(s, v, xw); 399 v += 25; 400 } 401 t[i2] = s[0]; 402 t[i2 + 1] = s[1]; 403 t[i2 + 2] = s[2]; 404 t[i2 + 3] = s[3]; 405 t[i2 + 4] = s[4]; 406 PetscKernel_v_gets_A_times_w_5(xw, idiag, s); 407 x[i2] = xw[0]; 408 x[i2 + 1] = xw[1]; 409 x[i2 + 2] = xw[2]; 410 x[i2 + 3] = xw[3]; 411 x[i2 + 4] = xw[4]; 412 idiag += 25; 413 i2 += 5; 414 } 415 break; 416 case 6: 417 PetscKernel_v_gets_A_times_w_6(x, idiag, b); 418 t[0] = b[0]; 419 t[1] = b[1]; 420 t[2] = b[2]; 421 t[3] = b[3]; 422 t[4] = b[4]; 423 t[5] = b[5]; 424 i2 = 6; 425 idiag += 36; 426 for (i = 1; i < m; i++) { 427 v = aa + 36 * ai[i]; 428 vi = aj + ai[i]; 429 nz = diag[i] - ai[i]; 430 s[0] = b[i2]; 431 s[1] = b[i2 + 1]; 432 s[2] = b[i2 + 2]; 433 s[3] = b[i2 + 3]; 434 s[4] = b[i2 + 4]; 435 s[5] = b[i2 + 5]; 436 while (nz--) { 437 idx = 6 * (*vi++); 438 xw[0] = x[idx]; 439 xw[1] = x[1 + idx]; 440 xw[2] = x[2 + idx]; 441 xw[3] = x[3 + idx]; 442 xw[4] = x[4 + idx]; 443 xw[5] = x[5 + idx]; 444 PetscKernel_v_gets_v_minus_A_times_w_6(s, v, xw); 445 v += 36; 446 } 447 t[i2] = s[0]; 448 t[i2 + 1] = s[1]; 449 t[i2 + 2] = s[2]; 450 t[i2 + 3] = s[3]; 451 t[i2 + 4] = s[4]; 452 t[i2 + 5] = s[5]; 453 PetscKernel_v_gets_A_times_w_6(xw, idiag, s); 454 x[i2] = xw[0]; 455 x[i2 + 1] = xw[1]; 456 x[i2 + 2] = xw[2]; 457 x[i2 + 3] = xw[3]; 458 x[i2 + 4] = xw[4]; 459 x[i2 + 5] = xw[5]; 460 idiag += 36; 461 i2 += 6; 462 } 463 break; 464 case 7: 465 PetscKernel_v_gets_A_times_w_7(x, idiag, b); 466 t[0] = b[0]; 467 t[1] = b[1]; 468 t[2] = b[2]; 469 t[3] = b[3]; 470 t[4] = b[4]; 471 t[5] = b[5]; 472 t[6] = b[6]; 473 i2 = 7; 474 idiag += 49; 475 for (i = 1; i < m; i++) { 476 v = aa + 49 * ai[i]; 477 vi = aj + ai[i]; 478 nz = diag[i] - ai[i]; 479 s[0] = b[i2]; 480 s[1] = b[i2 + 1]; 481 s[2] = b[i2 + 2]; 482 s[3] = b[i2 + 3]; 483 s[4] = b[i2 + 4]; 484 s[5] = b[i2 + 5]; 485 s[6] = b[i2 + 6]; 486 while (nz--) { 487 idx = 7 * (*vi++); 488 xw[0] = x[idx]; 489 xw[1] = x[1 + idx]; 490 xw[2] = x[2 + idx]; 491 xw[3] = x[3 + idx]; 492 xw[4] = x[4 + idx]; 493 xw[5] = x[5 + idx]; 494 xw[6] = x[6 + idx]; 495 PetscKernel_v_gets_v_minus_A_times_w_7(s, v, xw); 496 v += 49; 497 } 498 t[i2] = s[0]; 499 t[i2 + 1] = s[1]; 500 t[i2 + 2] = s[2]; 501 t[i2 + 3] = s[3]; 502 t[i2 + 4] = s[4]; 503 t[i2 + 5] = s[5]; 504 t[i2 + 6] = s[6]; 505 PetscKernel_v_gets_A_times_w_7(xw, idiag, s); 506 x[i2] = xw[0]; 507 x[i2 + 1] = xw[1]; 508 x[i2 + 2] = xw[2]; 509 x[i2 + 3] = xw[3]; 510 x[i2 + 4] = xw[4]; 511 x[i2 + 5] = xw[5]; 512 x[i2 + 6] = xw[6]; 513 idiag += 49; 514 i2 += 7; 515 } 516 break; 517 default: 518 PetscKernel_w_gets_Ar_times_v(bs, bs, b, idiag, x); 519 PetscCall(PetscArraycpy(t, b, bs)); 520 i2 = bs; 521 idiag += bs2; 522 for (i = 1; i < m; i++) { 523 v = aa + bs2 * ai[i]; 524 vi = aj + ai[i]; 525 nz = diag[i] - ai[i]; 526 527 PetscCall(PetscArraycpy(w, b + i2, bs)); 528 /* copy all rows of x that are needed into contiguous space */ 529 workt = work; 530 for (j = 0; j < nz; j++) { 531 PetscCall(PetscArraycpy(workt, x + bs * (*vi++), bs)); 532 workt += bs; 533 } 534 PetscKernel_w_gets_w_minus_Ar_times_v(bs, bs * nz, w, v, work); 535 PetscCall(PetscArraycpy(t + i2, w, bs)); 536 PetscKernel_w_gets_Ar_times_v(bs, bs, w, idiag, x + i2); 537 538 idiag += bs2; 539 i2 += bs; 540 } 541 break; 542 } 543 /* for logging purposes assume number of nonzero in lower half is 1/2 of total */ 544 PetscCall(PetscLogFlops(1.0 * bs2 * a->nz)); 545 xb = t; 546 } else xb = b; 547 if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) { 548 idiag = a->idiag + bs2 * (a->mbs - 1); 549 i2 = bs * (m - 1); 550 switch (bs) { 551 case 1: 552 s[0] = xb[i2]; 553 PetscKernel_v_gets_A_times_w_1(xw, idiag, s); 554 x[i2] = xw[0]; 555 i2 -= 1; 556 for (i = m - 2; i >= 0; i--) { 557 v = aa + (diag[i] + 1); 558 vi = aj + diag[i] + 1; 559 nz = ai[i + 1] - diag[i] - 1; 560 s[0] = xb[i2]; 561 for (j = 0; j < nz; j++) { 562 xw[0] = x[vi[j]]; 563 PetscKernel_v_gets_v_minus_A_times_w_1(s, (v + j), xw); 564 } 565 PetscKernel_v_gets_A_times_w_1(xw, idiag, s); 566 x[i2] = xw[0]; 567 idiag -= 1; 568 i2 -= 1; 569 } 570 break; 571 case 2: 572 s[0] = xb[i2]; 573 s[1] = xb[i2 + 1]; 574 PetscKernel_v_gets_A_times_w_2(xw, idiag, s); 575 x[i2] = xw[0]; 576 x[i2 + 1] = xw[1]; 577 i2 -= 2; 578 idiag -= 4; 579 for (i = m - 2; i >= 0; i--) { 580 v = aa + 4 * (diag[i] + 1); 581 vi = aj + diag[i] + 1; 582 nz = ai[i + 1] - diag[i] - 1; 583 s[0] = xb[i2]; 584 s[1] = xb[i2 + 1]; 585 for (j = 0; j < nz; j++) { 586 idx = 2 * vi[j]; 587 it = 4 * j; 588 xw[0] = x[idx]; 589 xw[1] = x[1 + idx]; 590 PetscKernel_v_gets_v_minus_A_times_w_2(s, (v + it), xw); 591 } 592 PetscKernel_v_gets_A_times_w_2(xw, idiag, s); 593 x[i2] = xw[0]; 594 x[i2 + 1] = xw[1]; 595 idiag -= 4; 596 i2 -= 2; 597 } 598 break; 599 case 3: 600 s[0] = xb[i2]; 601 s[1] = xb[i2 + 1]; 602 s[2] = xb[i2 + 2]; 603 PetscKernel_v_gets_A_times_w_3(xw, idiag, s); 604 x[i2] = xw[0]; 605 x[i2 + 1] = xw[1]; 606 x[i2 + 2] = xw[2]; 607 i2 -= 3; 608 idiag -= 9; 609 for (i = m - 2; i >= 0; i--) { 610 v = aa + 9 * (diag[i] + 1); 611 vi = aj + diag[i] + 1; 612 nz = ai[i + 1] - diag[i] - 1; 613 s[0] = xb[i2]; 614 s[1] = xb[i2 + 1]; 615 s[2] = xb[i2 + 2]; 616 while (nz--) { 617 idx = 3 * (*vi++); 618 xw[0] = x[idx]; 619 xw[1] = x[1 + idx]; 620 xw[2] = x[2 + idx]; 621 PetscKernel_v_gets_v_minus_A_times_w_3(s, v, xw); 622 v += 9; 623 } 624 PetscKernel_v_gets_A_times_w_3(xw, idiag, s); 625 x[i2] = xw[0]; 626 x[i2 + 1] = xw[1]; 627 x[i2 + 2] = xw[2]; 628 idiag -= 9; 629 i2 -= 3; 630 } 631 break; 632 case 4: 633 s[0] = xb[i2]; 634 s[1] = xb[i2 + 1]; 635 s[2] = xb[i2 + 2]; 636 s[3] = xb[i2 + 3]; 637 PetscKernel_v_gets_A_times_w_4(xw, idiag, s); 638 x[i2] = xw[0]; 639 x[i2 + 1] = xw[1]; 640 x[i2 + 2] = xw[2]; 641 x[i2 + 3] = xw[3]; 642 i2 -= 4; 643 idiag -= 16; 644 for (i = m - 2; i >= 0; i--) { 645 v = aa + 16 * (diag[i] + 1); 646 vi = aj + diag[i] + 1; 647 nz = ai[i + 1] - diag[i] - 1; 648 s[0] = xb[i2]; 649 s[1] = xb[i2 + 1]; 650 s[2] = xb[i2 + 2]; 651 s[3] = xb[i2 + 3]; 652 while (nz--) { 653 idx = 4 * (*vi++); 654 xw[0] = x[idx]; 655 xw[1] = x[1 + idx]; 656 xw[2] = x[2 + idx]; 657 xw[3] = x[3 + idx]; 658 PetscKernel_v_gets_v_minus_A_times_w_4(s, v, xw); 659 v += 16; 660 } 661 PetscKernel_v_gets_A_times_w_4(xw, idiag, s); 662 x[i2] = xw[0]; 663 x[i2 + 1] = xw[1]; 664 x[i2 + 2] = xw[2]; 665 x[i2 + 3] = xw[3]; 666 idiag -= 16; 667 i2 -= 4; 668 } 669 break; 670 case 5: 671 s[0] = xb[i2]; 672 s[1] = xb[i2 + 1]; 673 s[2] = xb[i2 + 2]; 674 s[3] = xb[i2 + 3]; 675 s[4] = xb[i2 + 4]; 676 PetscKernel_v_gets_A_times_w_5(xw, idiag, s); 677 x[i2] = xw[0]; 678 x[i2 + 1] = xw[1]; 679 x[i2 + 2] = xw[2]; 680 x[i2 + 3] = xw[3]; 681 x[i2 + 4] = xw[4]; 682 i2 -= 5; 683 idiag -= 25; 684 for (i = m - 2; i >= 0; i--) { 685 v = aa + 25 * (diag[i] + 1); 686 vi = aj + diag[i] + 1; 687 nz = ai[i + 1] - diag[i] - 1; 688 s[0] = xb[i2]; 689 s[1] = xb[i2 + 1]; 690 s[2] = xb[i2 + 2]; 691 s[3] = xb[i2 + 3]; 692 s[4] = xb[i2 + 4]; 693 while (nz--) { 694 idx = 5 * (*vi++); 695 xw[0] = x[idx]; 696 xw[1] = x[1 + idx]; 697 xw[2] = x[2 + idx]; 698 xw[3] = x[3 + idx]; 699 xw[4] = x[4 + idx]; 700 PetscKernel_v_gets_v_minus_A_times_w_5(s, v, xw); 701 v += 25; 702 } 703 PetscKernel_v_gets_A_times_w_5(xw, idiag, s); 704 x[i2] = xw[0]; 705 x[i2 + 1] = xw[1]; 706 x[i2 + 2] = xw[2]; 707 x[i2 + 3] = xw[3]; 708 x[i2 + 4] = xw[4]; 709 idiag -= 25; 710 i2 -= 5; 711 } 712 break; 713 case 6: 714 s[0] = xb[i2]; 715 s[1] = xb[i2 + 1]; 716 s[2] = xb[i2 + 2]; 717 s[3] = xb[i2 + 3]; 718 s[4] = xb[i2 + 4]; 719 s[5] = xb[i2 + 5]; 720 PetscKernel_v_gets_A_times_w_6(xw, idiag, s); 721 x[i2] = xw[0]; 722 x[i2 + 1] = xw[1]; 723 x[i2 + 2] = xw[2]; 724 x[i2 + 3] = xw[3]; 725 x[i2 + 4] = xw[4]; 726 x[i2 + 5] = xw[5]; 727 i2 -= 6; 728 idiag -= 36; 729 for (i = m - 2; i >= 0; i--) { 730 v = aa + 36 * (diag[i] + 1); 731 vi = aj + diag[i] + 1; 732 nz = ai[i + 1] - diag[i] - 1; 733 s[0] = xb[i2]; 734 s[1] = xb[i2 + 1]; 735 s[2] = xb[i2 + 2]; 736 s[3] = xb[i2 + 3]; 737 s[4] = xb[i2 + 4]; 738 s[5] = xb[i2 + 5]; 739 while (nz--) { 740 idx = 6 * (*vi++); 741 xw[0] = x[idx]; 742 xw[1] = x[1 + idx]; 743 xw[2] = x[2 + idx]; 744 xw[3] = x[3 + idx]; 745 xw[4] = x[4 + idx]; 746 xw[5] = x[5 + idx]; 747 PetscKernel_v_gets_v_minus_A_times_w_6(s, v, xw); 748 v += 36; 749 } 750 PetscKernel_v_gets_A_times_w_6(xw, idiag, s); 751 x[i2] = xw[0]; 752 x[i2 + 1] = xw[1]; 753 x[i2 + 2] = xw[2]; 754 x[i2 + 3] = xw[3]; 755 x[i2 + 4] = xw[4]; 756 x[i2 + 5] = xw[5]; 757 idiag -= 36; 758 i2 -= 6; 759 } 760 break; 761 case 7: 762 s[0] = xb[i2]; 763 s[1] = xb[i2 + 1]; 764 s[2] = xb[i2 + 2]; 765 s[3] = xb[i2 + 3]; 766 s[4] = xb[i2 + 4]; 767 s[5] = xb[i2 + 5]; 768 s[6] = xb[i2 + 6]; 769 PetscKernel_v_gets_A_times_w_7(x, idiag, b); 770 x[i2] = xw[0]; 771 x[i2 + 1] = xw[1]; 772 x[i2 + 2] = xw[2]; 773 x[i2 + 3] = xw[3]; 774 x[i2 + 4] = xw[4]; 775 x[i2 + 5] = xw[5]; 776 x[i2 + 6] = xw[6]; 777 i2 -= 7; 778 idiag -= 49; 779 for (i = m - 2; i >= 0; i--) { 780 v = aa + 49 * (diag[i] + 1); 781 vi = aj + diag[i] + 1; 782 nz = ai[i + 1] - diag[i] - 1; 783 s[0] = xb[i2]; 784 s[1] = xb[i2 + 1]; 785 s[2] = xb[i2 + 2]; 786 s[3] = xb[i2 + 3]; 787 s[4] = xb[i2 + 4]; 788 s[5] = xb[i2 + 5]; 789 s[6] = xb[i2 + 6]; 790 while (nz--) { 791 idx = 7 * (*vi++); 792 xw[0] = x[idx]; 793 xw[1] = x[1 + idx]; 794 xw[2] = x[2 + idx]; 795 xw[3] = x[3 + idx]; 796 xw[4] = x[4 + idx]; 797 xw[5] = x[5 + idx]; 798 xw[6] = x[6 + idx]; 799 PetscKernel_v_gets_v_minus_A_times_w_7(s, v, xw); 800 v += 49; 801 } 802 PetscKernel_v_gets_A_times_w_7(xw, idiag, s); 803 x[i2] = xw[0]; 804 x[i2 + 1] = xw[1]; 805 x[i2 + 2] = xw[2]; 806 x[i2 + 3] = xw[3]; 807 x[i2 + 4] = xw[4]; 808 x[i2 + 5] = xw[5]; 809 x[i2 + 6] = xw[6]; 810 idiag -= 49; 811 i2 -= 7; 812 } 813 break; 814 default: 815 PetscCall(PetscArraycpy(w, xb + i2, bs)); 816 PetscKernel_w_gets_Ar_times_v(bs, bs, w, idiag, x + i2); 817 i2 -= bs; 818 idiag -= bs2; 819 for (i = m - 2; i >= 0; i--) { 820 v = aa + bs2 * (diag[i] + 1); 821 vi = aj + diag[i] + 1; 822 nz = ai[i + 1] - diag[i] - 1; 823 824 PetscCall(PetscArraycpy(w, xb + i2, bs)); 825 /* copy all rows of x that are needed into contiguous space */ 826 workt = work; 827 for (j = 0; j < nz; j++) { 828 PetscCall(PetscArraycpy(workt, x + bs * (*vi++), bs)); 829 workt += bs; 830 } 831 PetscKernel_w_gets_w_minus_Ar_times_v(bs, bs * nz, w, v, work); 832 PetscKernel_w_gets_Ar_times_v(bs, bs, w, idiag, x + i2); 833 834 idiag -= bs2; 835 i2 -= bs; 836 } 837 break; 838 } 839 PetscCall(PetscLogFlops(1.0 * bs2 * (a->nz))); 840 } 841 its--; 842 } 843 while (its--) { 844 if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) { 845 idiag = a->idiag; 846 i2 = 0; 847 switch (bs) { 848 case 1: 849 for (i = 0; i < m; i++) { 850 v = aa + ai[i]; 851 vi = aj + ai[i]; 852 nz = ai[i + 1] - ai[i]; 853 s[0] = b[i2]; 854 for (j = 0; j < nz; j++) { 855 xw[0] = x[vi[j]]; 856 PetscKernel_v_gets_v_minus_A_times_w_1(s, (v + j), xw); 857 } 858 PetscKernel_v_gets_A_times_w_1(xw, idiag, s); 859 x[i2] += xw[0]; 860 idiag += 1; 861 i2 += 1; 862 } 863 break; 864 case 2: 865 for (i = 0; i < m; i++) { 866 v = aa + 4 * ai[i]; 867 vi = aj + ai[i]; 868 nz = ai[i + 1] - ai[i]; 869 s[0] = b[i2]; 870 s[1] = b[i2 + 1]; 871 for (j = 0; j < nz; j++) { 872 idx = 2 * vi[j]; 873 it = 4 * j; 874 xw[0] = x[idx]; 875 xw[1] = x[1 + idx]; 876 PetscKernel_v_gets_v_minus_A_times_w_2(s, (v + it), xw); 877 } 878 PetscKernel_v_gets_A_times_w_2(xw, idiag, s); 879 x[i2] += xw[0]; 880 x[i2 + 1] += xw[1]; 881 idiag += 4; 882 i2 += 2; 883 } 884 break; 885 case 3: 886 for (i = 0; i < m; i++) { 887 v = aa + 9 * ai[i]; 888 vi = aj + ai[i]; 889 nz = ai[i + 1] - ai[i]; 890 s[0] = b[i2]; 891 s[1] = b[i2 + 1]; 892 s[2] = b[i2 + 2]; 893 while (nz--) { 894 idx = 3 * (*vi++); 895 xw[0] = x[idx]; 896 xw[1] = x[1 + idx]; 897 xw[2] = x[2 + idx]; 898 PetscKernel_v_gets_v_minus_A_times_w_3(s, v, xw); 899 v += 9; 900 } 901 PetscKernel_v_gets_A_times_w_3(xw, idiag, s); 902 x[i2] += xw[0]; 903 x[i2 + 1] += xw[1]; 904 x[i2 + 2] += xw[2]; 905 idiag += 9; 906 i2 += 3; 907 } 908 break; 909 case 4: 910 for (i = 0; i < m; i++) { 911 v = aa + 16 * ai[i]; 912 vi = aj + ai[i]; 913 nz = ai[i + 1] - ai[i]; 914 s[0] = b[i2]; 915 s[1] = b[i2 + 1]; 916 s[2] = b[i2 + 2]; 917 s[3] = b[i2 + 3]; 918 while (nz--) { 919 idx = 4 * (*vi++); 920 xw[0] = x[idx]; 921 xw[1] = x[1 + idx]; 922 xw[2] = x[2 + idx]; 923 xw[3] = x[3 + idx]; 924 PetscKernel_v_gets_v_minus_A_times_w_4(s, v, xw); 925 v += 16; 926 } 927 PetscKernel_v_gets_A_times_w_4(xw, idiag, s); 928 x[i2] += xw[0]; 929 x[i2 + 1] += xw[1]; 930 x[i2 + 2] += xw[2]; 931 x[i2 + 3] += xw[3]; 932 idiag += 16; 933 i2 += 4; 934 } 935 break; 936 case 5: 937 for (i = 0; i < m; i++) { 938 v = aa + 25 * ai[i]; 939 vi = aj + ai[i]; 940 nz = ai[i + 1] - ai[i]; 941 s[0] = b[i2]; 942 s[1] = b[i2 + 1]; 943 s[2] = b[i2 + 2]; 944 s[3] = b[i2 + 3]; 945 s[4] = b[i2 + 4]; 946 while (nz--) { 947 idx = 5 * (*vi++); 948 xw[0] = x[idx]; 949 xw[1] = x[1 + idx]; 950 xw[2] = x[2 + idx]; 951 xw[3] = x[3 + idx]; 952 xw[4] = x[4 + idx]; 953 PetscKernel_v_gets_v_minus_A_times_w_5(s, v, xw); 954 v += 25; 955 } 956 PetscKernel_v_gets_A_times_w_5(xw, idiag, s); 957 x[i2] += xw[0]; 958 x[i2 + 1] += xw[1]; 959 x[i2 + 2] += xw[2]; 960 x[i2 + 3] += xw[3]; 961 x[i2 + 4] += xw[4]; 962 idiag += 25; 963 i2 += 5; 964 } 965 break; 966 case 6: 967 for (i = 0; i < m; i++) { 968 v = aa + 36 * ai[i]; 969 vi = aj + ai[i]; 970 nz = ai[i + 1] - ai[i]; 971 s[0] = b[i2]; 972 s[1] = b[i2 + 1]; 973 s[2] = b[i2 + 2]; 974 s[3] = b[i2 + 3]; 975 s[4] = b[i2 + 4]; 976 s[5] = b[i2 + 5]; 977 while (nz--) { 978 idx = 6 * (*vi++); 979 xw[0] = x[idx]; 980 xw[1] = x[1 + idx]; 981 xw[2] = x[2 + idx]; 982 xw[3] = x[3 + idx]; 983 xw[4] = x[4 + idx]; 984 xw[5] = x[5 + idx]; 985 PetscKernel_v_gets_v_minus_A_times_w_6(s, v, xw); 986 v += 36; 987 } 988 PetscKernel_v_gets_A_times_w_6(xw, idiag, s); 989 x[i2] += xw[0]; 990 x[i2 + 1] += xw[1]; 991 x[i2 + 2] += xw[2]; 992 x[i2 + 3] += xw[3]; 993 x[i2 + 4] += xw[4]; 994 x[i2 + 5] += xw[5]; 995 idiag += 36; 996 i2 += 6; 997 } 998 break; 999 case 7: 1000 for (i = 0; i < m; i++) { 1001 v = aa + 49 * ai[i]; 1002 vi = aj + ai[i]; 1003 nz = ai[i + 1] - ai[i]; 1004 s[0] = b[i2]; 1005 s[1] = b[i2 + 1]; 1006 s[2] = b[i2 + 2]; 1007 s[3] = b[i2 + 3]; 1008 s[4] = b[i2 + 4]; 1009 s[5] = b[i2 + 5]; 1010 s[6] = b[i2 + 6]; 1011 while (nz--) { 1012 idx = 7 * (*vi++); 1013 xw[0] = x[idx]; 1014 xw[1] = x[1 + idx]; 1015 xw[2] = x[2 + idx]; 1016 xw[3] = x[3 + idx]; 1017 xw[4] = x[4 + idx]; 1018 xw[5] = x[5 + idx]; 1019 xw[6] = x[6 + idx]; 1020 PetscKernel_v_gets_v_minus_A_times_w_7(s, v, xw); 1021 v += 49; 1022 } 1023 PetscKernel_v_gets_A_times_w_7(xw, idiag, s); 1024 x[i2] += xw[0]; 1025 x[i2 + 1] += xw[1]; 1026 x[i2 + 2] += xw[2]; 1027 x[i2 + 3] += xw[3]; 1028 x[i2 + 4] += xw[4]; 1029 x[i2 + 5] += xw[5]; 1030 x[i2 + 6] += xw[6]; 1031 idiag += 49; 1032 i2 += 7; 1033 } 1034 break; 1035 default: 1036 for (i = 0; i < m; i++) { 1037 v = aa + bs2 * ai[i]; 1038 vi = aj + ai[i]; 1039 nz = ai[i + 1] - ai[i]; 1040 1041 PetscCall(PetscArraycpy(w, b + i2, bs)); 1042 /* copy all rows of x that are needed into contiguous space */ 1043 workt = work; 1044 for (j = 0; j < nz; j++) { 1045 PetscCall(PetscArraycpy(workt, x + bs * (*vi++), bs)); 1046 workt += bs; 1047 } 1048 PetscKernel_w_gets_w_minus_Ar_times_v(bs, bs * nz, w, v, work); 1049 PetscKernel_w_gets_w_plus_Ar_times_v(bs, bs, w, idiag, x + i2); 1050 1051 idiag += bs2; 1052 i2 += bs; 1053 } 1054 break; 1055 } 1056 PetscCall(PetscLogFlops(2.0 * bs2 * a->nz)); 1057 } 1058 if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) { 1059 idiag = a->idiag + bs2 * (a->mbs - 1); 1060 i2 = bs * (m - 1); 1061 switch (bs) { 1062 case 1: 1063 for (i = m - 1; i >= 0; i--) { 1064 v = aa + ai[i]; 1065 vi = aj + ai[i]; 1066 nz = ai[i + 1] - ai[i]; 1067 s[0] = b[i2]; 1068 for (j = 0; j < nz; j++) { 1069 xw[0] = x[vi[j]]; 1070 PetscKernel_v_gets_v_minus_A_times_w_1(s, (v + j), xw); 1071 } 1072 PetscKernel_v_gets_A_times_w_1(xw, idiag, s); 1073 x[i2] += xw[0]; 1074 idiag -= 1; 1075 i2 -= 1; 1076 } 1077 break; 1078 case 2: 1079 for (i = m - 1; i >= 0; i--) { 1080 v = aa + 4 * ai[i]; 1081 vi = aj + ai[i]; 1082 nz = ai[i + 1] - ai[i]; 1083 s[0] = b[i2]; 1084 s[1] = b[i2 + 1]; 1085 for (j = 0; j < nz; j++) { 1086 idx = 2 * vi[j]; 1087 it = 4 * j; 1088 xw[0] = x[idx]; 1089 xw[1] = x[1 + idx]; 1090 PetscKernel_v_gets_v_minus_A_times_w_2(s, (v + it), xw); 1091 } 1092 PetscKernel_v_gets_A_times_w_2(xw, idiag, s); 1093 x[i2] += xw[0]; 1094 x[i2 + 1] += xw[1]; 1095 idiag -= 4; 1096 i2 -= 2; 1097 } 1098 break; 1099 case 3: 1100 for (i = m - 1; i >= 0; i--) { 1101 v = aa + 9 * ai[i]; 1102 vi = aj + ai[i]; 1103 nz = ai[i + 1] - ai[i]; 1104 s[0] = b[i2]; 1105 s[1] = b[i2 + 1]; 1106 s[2] = b[i2 + 2]; 1107 while (nz--) { 1108 idx = 3 * (*vi++); 1109 xw[0] = x[idx]; 1110 xw[1] = x[1 + idx]; 1111 xw[2] = x[2 + idx]; 1112 PetscKernel_v_gets_v_minus_A_times_w_3(s, v, xw); 1113 v += 9; 1114 } 1115 PetscKernel_v_gets_A_times_w_3(xw, idiag, s); 1116 x[i2] += xw[0]; 1117 x[i2 + 1] += xw[1]; 1118 x[i2 + 2] += xw[2]; 1119 idiag -= 9; 1120 i2 -= 3; 1121 } 1122 break; 1123 case 4: 1124 for (i = m - 1; i >= 0; i--) { 1125 v = aa + 16 * ai[i]; 1126 vi = aj + ai[i]; 1127 nz = ai[i + 1] - ai[i]; 1128 s[0] = b[i2]; 1129 s[1] = b[i2 + 1]; 1130 s[2] = b[i2 + 2]; 1131 s[3] = b[i2 + 3]; 1132 while (nz--) { 1133 idx = 4 * (*vi++); 1134 xw[0] = x[idx]; 1135 xw[1] = x[1 + idx]; 1136 xw[2] = x[2 + idx]; 1137 xw[3] = x[3 + idx]; 1138 PetscKernel_v_gets_v_minus_A_times_w_4(s, v, xw); 1139 v += 16; 1140 } 1141 PetscKernel_v_gets_A_times_w_4(xw, idiag, s); 1142 x[i2] += xw[0]; 1143 x[i2 + 1] += xw[1]; 1144 x[i2 + 2] += xw[2]; 1145 x[i2 + 3] += xw[3]; 1146 idiag -= 16; 1147 i2 -= 4; 1148 } 1149 break; 1150 case 5: 1151 for (i = m - 1; i >= 0; i--) { 1152 v = aa + 25 * ai[i]; 1153 vi = aj + ai[i]; 1154 nz = ai[i + 1] - ai[i]; 1155 s[0] = b[i2]; 1156 s[1] = b[i2 + 1]; 1157 s[2] = b[i2 + 2]; 1158 s[3] = b[i2 + 3]; 1159 s[4] = b[i2 + 4]; 1160 while (nz--) { 1161 idx = 5 * (*vi++); 1162 xw[0] = x[idx]; 1163 xw[1] = x[1 + idx]; 1164 xw[2] = x[2 + idx]; 1165 xw[3] = x[3 + idx]; 1166 xw[4] = x[4 + idx]; 1167 PetscKernel_v_gets_v_minus_A_times_w_5(s, v, xw); 1168 v += 25; 1169 } 1170 PetscKernel_v_gets_A_times_w_5(xw, idiag, s); 1171 x[i2] += xw[0]; 1172 x[i2 + 1] += xw[1]; 1173 x[i2 + 2] += xw[2]; 1174 x[i2 + 3] += xw[3]; 1175 x[i2 + 4] += xw[4]; 1176 idiag -= 25; 1177 i2 -= 5; 1178 } 1179 break; 1180 case 6: 1181 for (i = m - 1; i >= 0; i--) { 1182 v = aa + 36 * ai[i]; 1183 vi = aj + ai[i]; 1184 nz = ai[i + 1] - ai[i]; 1185 s[0] = b[i2]; 1186 s[1] = b[i2 + 1]; 1187 s[2] = b[i2 + 2]; 1188 s[3] = b[i2 + 3]; 1189 s[4] = b[i2 + 4]; 1190 s[5] = b[i2 + 5]; 1191 while (nz--) { 1192 idx = 6 * (*vi++); 1193 xw[0] = x[idx]; 1194 xw[1] = x[1 + idx]; 1195 xw[2] = x[2 + idx]; 1196 xw[3] = x[3 + idx]; 1197 xw[4] = x[4 + idx]; 1198 xw[5] = x[5 + idx]; 1199 PetscKernel_v_gets_v_minus_A_times_w_6(s, v, xw); 1200 v += 36; 1201 } 1202 PetscKernel_v_gets_A_times_w_6(xw, idiag, s); 1203 x[i2] += xw[0]; 1204 x[i2 + 1] += xw[1]; 1205 x[i2 + 2] += xw[2]; 1206 x[i2 + 3] += xw[3]; 1207 x[i2 + 4] += xw[4]; 1208 x[i2 + 5] += xw[5]; 1209 idiag -= 36; 1210 i2 -= 6; 1211 } 1212 break; 1213 case 7: 1214 for (i = m - 1; i >= 0; i--) { 1215 v = aa + 49 * ai[i]; 1216 vi = aj + ai[i]; 1217 nz = ai[i + 1] - ai[i]; 1218 s[0] = b[i2]; 1219 s[1] = b[i2 + 1]; 1220 s[2] = b[i2 + 2]; 1221 s[3] = b[i2 + 3]; 1222 s[4] = b[i2 + 4]; 1223 s[5] = b[i2 + 5]; 1224 s[6] = b[i2 + 6]; 1225 while (nz--) { 1226 idx = 7 * (*vi++); 1227 xw[0] = x[idx]; 1228 xw[1] = x[1 + idx]; 1229 xw[2] = x[2 + idx]; 1230 xw[3] = x[3 + idx]; 1231 xw[4] = x[4 + idx]; 1232 xw[5] = x[5 + idx]; 1233 xw[6] = x[6 + idx]; 1234 PetscKernel_v_gets_v_minus_A_times_w_7(s, v, xw); 1235 v += 49; 1236 } 1237 PetscKernel_v_gets_A_times_w_7(xw, idiag, s); 1238 x[i2] += xw[0]; 1239 x[i2 + 1] += xw[1]; 1240 x[i2 + 2] += xw[2]; 1241 x[i2 + 3] += xw[3]; 1242 x[i2 + 4] += xw[4]; 1243 x[i2 + 5] += xw[5]; 1244 x[i2 + 6] += xw[6]; 1245 idiag -= 49; 1246 i2 -= 7; 1247 } 1248 break; 1249 default: 1250 for (i = m - 1; i >= 0; i--) { 1251 v = aa + bs2 * ai[i]; 1252 vi = aj + ai[i]; 1253 nz = ai[i + 1] - ai[i]; 1254 1255 PetscCall(PetscArraycpy(w, b + i2, bs)); 1256 /* copy all rows of x that are needed into contiguous space */ 1257 workt = work; 1258 for (j = 0; j < nz; j++) { 1259 PetscCall(PetscArraycpy(workt, x + bs * (*vi++), bs)); 1260 workt += bs; 1261 } 1262 PetscKernel_w_gets_w_minus_Ar_times_v(bs, bs * nz, w, v, work); 1263 PetscKernel_w_gets_w_plus_Ar_times_v(bs, bs, w, idiag, x + i2); 1264 1265 idiag -= bs2; 1266 i2 -= bs; 1267 } 1268 break; 1269 } 1270 PetscCall(PetscLogFlops(2.0 * bs2 * (a->nz))); 1271 } 1272 } 1273 PetscCall(VecRestoreArray(xx, &x)); 1274 PetscCall(VecRestoreArrayRead(bb, &b)); 1275 PetscFunctionReturn(PETSC_SUCCESS); 1276 } 1277 1278 /* 1279 Special version for direct calls from Fortran (Used in PETSc-fun3d) 1280 */ 1281 #if defined(PETSC_HAVE_FORTRAN_CAPS) 1282 #define matsetvaluesblocked4_ MATSETVALUESBLOCKED4 1283 #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE) 1284 #define matsetvaluesblocked4_ matsetvaluesblocked4 1285 #endif 1286 1287 PETSC_EXTERN void matsetvaluesblocked4_(Mat *AA, PetscInt *mm, const PetscInt im[], PetscInt *nn, const PetscInt in[], const PetscScalar v[]) 1288 { 1289 Mat A = *AA; 1290 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 1291 PetscInt *rp, k, low, high, t, ii, jj, row, nrow, i, col, l, N, m = *mm, n = *nn; 1292 PetscInt *ai = a->i, *ailen = a->ilen; 1293 PetscInt *aj = a->j, stepval, lastcol = -1; 1294 const PetscScalar *value = v; 1295 MatScalar *ap, *aa = a->a, *bap; 1296 1297 PetscFunctionBegin; 1298 if (A->rmap->bs != 4) SETERRABORT(PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONG, "Can only be called with a block size of 4"); 1299 stepval = (n - 1) * 4; 1300 for (k = 0; k < m; k++) { /* loop over added rows */ 1301 row = im[k]; 1302 rp = aj + ai[row]; 1303 ap = aa + 16 * ai[row]; 1304 nrow = ailen[row]; 1305 low = 0; 1306 high = nrow; 1307 for (l = 0; l < n; l++) { /* loop over added columns */ 1308 col = in[l]; 1309 if (col <= lastcol) low = 0; 1310 else high = nrow; 1311 lastcol = col; 1312 value = v + k * (stepval + 4 + l) * 4; 1313 while (high - low > 7) { 1314 t = (low + high) / 2; 1315 if (rp[t] > col) high = t; 1316 else low = t; 1317 } 1318 for (i = low; i < high; i++) { 1319 if (rp[i] > col) break; 1320 if (rp[i] == col) { 1321 bap = ap + 16 * i; 1322 for (ii = 0; ii < 4; ii++, value += stepval) { 1323 for (jj = ii; jj < 16; jj += 4) bap[jj] += *value++; 1324 } 1325 goto noinsert2; 1326 } 1327 } 1328 N = nrow++ - 1; 1329 high++; /* added new column index thus must search to one higher than before */ 1330 /* shift up all the later entries in this row */ 1331 for (ii = N; ii >= i; ii--) { 1332 rp[ii + 1] = rp[ii]; 1333 PetscCallVoid(PetscArraycpy(ap + 16 * (ii + 1), ap + 16 * (ii), 16)); 1334 } 1335 if (N >= i) PetscCallVoid(PetscArrayzero(ap + 16 * i, 16)); 1336 rp[i] = col; 1337 bap = ap + 16 * i; 1338 for (ii = 0; ii < 4; ii++, value += stepval) { 1339 for (jj = ii; jj < 16; jj += 4) bap[jj] = *value++; 1340 } 1341 noinsert2:; 1342 low = i; 1343 } 1344 ailen[row] = nrow; 1345 } 1346 PetscFunctionReturnVoid(); 1347 } 1348 1349 #if defined(PETSC_HAVE_FORTRAN_CAPS) 1350 #define matsetvalues4_ MATSETVALUES4 1351 #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE) 1352 #define matsetvalues4_ matsetvalues4 1353 #endif 1354 1355 PETSC_EXTERN void matsetvalues4_(Mat *AA, PetscInt *mm, PetscInt *im, PetscInt *nn, PetscInt *in, PetscScalar *v) 1356 { 1357 Mat A = *AA; 1358 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 1359 PetscInt *rp, k, low, high, t, row, nrow, i, col, l, N, n = *nn, m = *mm; 1360 PetscInt *ai = a->i, *ailen = a->ilen; 1361 PetscInt *aj = a->j, brow, bcol; 1362 PetscInt ridx, cidx, lastcol = -1; 1363 MatScalar *ap, value, *aa = a->a, *bap; 1364 1365 PetscFunctionBegin; 1366 for (k = 0; k < m; k++) { /* loop over added rows */ 1367 row = im[k]; 1368 brow = row / 4; 1369 rp = aj + ai[brow]; 1370 ap = aa + 16 * ai[brow]; 1371 nrow = ailen[brow]; 1372 low = 0; 1373 high = nrow; 1374 for (l = 0; l < n; l++) { /* loop over added columns */ 1375 col = in[l]; 1376 bcol = col / 4; 1377 ridx = row % 4; 1378 cidx = col % 4; 1379 value = v[l + k * n]; 1380 if (col <= lastcol) low = 0; 1381 else high = nrow; 1382 lastcol = col; 1383 while (high - low > 7) { 1384 t = (low + high) / 2; 1385 if (rp[t] > bcol) high = t; 1386 else low = t; 1387 } 1388 for (i = low; i < high; i++) { 1389 if (rp[i] > bcol) break; 1390 if (rp[i] == bcol) { 1391 bap = ap + 16 * i + 4 * cidx + ridx; 1392 *bap += value; 1393 goto noinsert1; 1394 } 1395 } 1396 N = nrow++ - 1; 1397 high++; /* added new column thus must search to one higher than before */ 1398 /* shift up all the later entries in this row */ 1399 PetscCallVoid(PetscArraymove(rp + i + 1, rp + i, N - i + 1)); 1400 PetscCallVoid(PetscArraymove(ap + 16 * i + 16, ap + 16 * i, 16 * (N - i + 1))); 1401 PetscCallVoid(PetscArrayzero(ap + 16 * i, 16)); 1402 rp[i] = bcol; 1403 ap[16 * i + 4 * cidx + ridx] = value; 1404 noinsert1:; 1405 low = i; 1406 } 1407 ailen[brow] = nrow; 1408 } 1409 PetscFunctionReturnVoid(); 1410 } 1411 1412 /* 1413 Checks for missing diagonals 1414 */ 1415 PetscErrorCode MatMissingDiagonal_SeqBAIJ(Mat A, PetscBool *missing, PetscInt *d) 1416 { 1417 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 1418 PetscInt *diag, *ii = a->i, i; 1419 1420 PetscFunctionBegin; 1421 PetscCall(MatMarkDiagonal_SeqBAIJ(A)); 1422 *missing = PETSC_FALSE; 1423 if (A->rmap->n > 0 && !ii) { 1424 *missing = PETSC_TRUE; 1425 if (d) *d = 0; 1426 PetscCall(PetscInfo(A, "Matrix has no entries therefore is missing diagonal\n")); 1427 } else { 1428 PetscInt n; 1429 n = PetscMin(a->mbs, a->nbs); 1430 diag = a->diag; 1431 for (i = 0; i < n; i++) { 1432 if (diag[i] >= ii[i + 1]) { 1433 *missing = PETSC_TRUE; 1434 if (d) *d = i; 1435 PetscCall(PetscInfo(A, "Matrix is missing block diagonal number %" PetscInt_FMT "\n", i)); 1436 break; 1437 } 1438 } 1439 } 1440 PetscFunctionReturn(PETSC_SUCCESS); 1441 } 1442 1443 PetscErrorCode MatMarkDiagonal_SeqBAIJ(Mat A) 1444 { 1445 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 1446 PetscInt i, j, m = a->mbs; 1447 1448 PetscFunctionBegin; 1449 if (!a->diag) { 1450 PetscCall(PetscMalloc1(m, &a->diag)); 1451 a->free_diag = PETSC_TRUE; 1452 } 1453 for (i = 0; i < m; i++) { 1454 a->diag[i] = a->i[i + 1]; 1455 for (j = a->i[i]; j < a->i[i + 1]; j++) { 1456 if (a->j[j] == i) { 1457 a->diag[i] = j; 1458 break; 1459 } 1460 } 1461 } 1462 PetscFunctionReturn(PETSC_SUCCESS); 1463 } 1464 1465 static PetscErrorCode MatGetRowIJ_SeqBAIJ(Mat A, PetscInt oshift, PetscBool symmetric, PetscBool blockcompressed, PetscInt *nn, const PetscInt *inia[], const PetscInt *inja[], PetscBool *done) 1466 { 1467 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 1468 PetscInt i, j, n = a->mbs, nz = a->i[n], *tia, *tja, bs = A->rmap->bs, k, l, cnt; 1469 PetscInt **ia = (PetscInt **)inia, **ja = (PetscInt **)inja; 1470 1471 PetscFunctionBegin; 1472 *nn = n; 1473 if (!ia) PetscFunctionReturn(PETSC_SUCCESS); 1474 if (symmetric) { 1475 PetscCall(MatToSymmetricIJ_SeqAIJ(n, a->i, a->j, PETSC_TRUE, 0, 0, &tia, &tja)); 1476 nz = tia[n]; 1477 } else { 1478 tia = a->i; 1479 tja = a->j; 1480 } 1481 1482 if (!blockcompressed && bs > 1) { 1483 (*nn) *= bs; 1484 /* malloc & create the natural set of indices */ 1485 PetscCall(PetscMalloc1((n + 1) * bs, ia)); 1486 if (n) { 1487 (*ia)[0] = oshift; 1488 for (j = 1; j < bs; j++) (*ia)[j] = (tia[1] - tia[0]) * bs + (*ia)[j - 1]; 1489 } 1490 1491 for (i = 1; i < n; i++) { 1492 (*ia)[i * bs] = (tia[i] - tia[i - 1]) * bs + (*ia)[i * bs - 1]; 1493 for (j = 1; j < bs; j++) (*ia)[i * bs + j] = (tia[i + 1] - tia[i]) * bs + (*ia)[i * bs + j - 1]; 1494 } 1495 if (n) (*ia)[n * bs] = (tia[n] - tia[n - 1]) * bs + (*ia)[n * bs - 1]; 1496 1497 if (inja) { 1498 PetscCall(PetscMalloc1(nz * bs * bs, ja)); 1499 cnt = 0; 1500 for (i = 0; i < n; i++) { 1501 for (j = 0; j < bs; j++) { 1502 for (k = tia[i]; k < tia[i + 1]; k++) { 1503 for (l = 0; l < bs; l++) (*ja)[cnt++] = bs * tja[k] + l; 1504 } 1505 } 1506 } 1507 } 1508 1509 if (symmetric) { /* deallocate memory allocated in MatToSymmetricIJ_SeqAIJ() */ 1510 PetscCall(PetscFree(tia)); 1511 PetscCall(PetscFree(tja)); 1512 } 1513 } else if (oshift == 1) { 1514 if (symmetric) { 1515 nz = tia[A->rmap->n / bs]; 1516 /* add 1 to i and j indices */ 1517 for (i = 0; i < A->rmap->n / bs + 1; i++) tia[i] = tia[i] + 1; 1518 *ia = tia; 1519 if (ja) { 1520 for (i = 0; i < nz; i++) tja[i] = tja[i] + 1; 1521 *ja = tja; 1522 } 1523 } else { 1524 nz = a->i[A->rmap->n / bs]; 1525 /* malloc space and add 1 to i and j indices */ 1526 PetscCall(PetscMalloc1(A->rmap->n / bs + 1, ia)); 1527 for (i = 0; i < A->rmap->n / bs + 1; i++) (*ia)[i] = a->i[i] + 1; 1528 if (ja) { 1529 PetscCall(PetscMalloc1(nz, ja)); 1530 for (i = 0; i < nz; i++) (*ja)[i] = a->j[i] + 1; 1531 } 1532 } 1533 } else { 1534 *ia = tia; 1535 if (ja) *ja = tja; 1536 } 1537 PetscFunctionReturn(PETSC_SUCCESS); 1538 } 1539 1540 static PetscErrorCode MatRestoreRowIJ_SeqBAIJ(Mat A, PetscInt oshift, PetscBool symmetric, PetscBool blockcompressed, PetscInt *nn, const PetscInt *ia[], const PetscInt *ja[], PetscBool *done) 1541 { 1542 PetscFunctionBegin; 1543 if (!ia) PetscFunctionReturn(PETSC_SUCCESS); 1544 if ((!blockcompressed && A->rmap->bs > 1) || (symmetric || oshift == 1)) { 1545 PetscCall(PetscFree(*ia)); 1546 if (ja) PetscCall(PetscFree(*ja)); 1547 } 1548 PetscFunctionReturn(PETSC_SUCCESS); 1549 } 1550 1551 PetscErrorCode MatDestroy_SeqBAIJ(Mat A) 1552 { 1553 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 1554 1555 PetscFunctionBegin; 1556 if (A->hash_active) { 1557 PetscInt bs; 1558 A->ops[0] = a->cops; 1559 PetscCall(PetscHMapIJVDestroy(&a->ht)); 1560 PetscCall(MatGetBlockSize(A, &bs)); 1561 if (bs > 1) PetscCall(PetscHSetIJDestroy(&a->bht)); 1562 PetscCall(PetscFree(a->dnz)); 1563 PetscCall(PetscFree(a->bdnz)); 1564 A->hash_active = PETSC_FALSE; 1565 } 1566 PetscCall(PetscLogObjectState((PetscObject)A, "Rows=%" PetscInt_FMT ", Cols=%" PetscInt_FMT ", NZ=%" PetscInt_FMT, A->rmap->N, A->cmap->n, a->nz)); 1567 PetscCall(MatSeqXAIJFreeAIJ(A, &a->a, &a->j, &a->i)); 1568 PetscCall(ISDestroy(&a->row)); 1569 PetscCall(ISDestroy(&a->col)); 1570 if (a->free_diag) PetscCall(PetscFree(a->diag)); 1571 PetscCall(PetscFree(a->idiag)); 1572 if (a->free_imax_ilen) PetscCall(PetscFree2(a->imax, a->ilen)); 1573 PetscCall(PetscFree(a->solve_work)); 1574 PetscCall(PetscFree(a->mult_work)); 1575 PetscCall(PetscFree(a->sor_workt)); 1576 PetscCall(PetscFree(a->sor_work)); 1577 PetscCall(ISDestroy(&a->icol)); 1578 PetscCall(PetscFree(a->saved_values)); 1579 PetscCall(PetscFree2(a->compressedrow.i, a->compressedrow.rindex)); 1580 1581 PetscCall(MatDestroy(&a->sbaijMat)); 1582 PetscCall(MatDestroy(&a->parent)); 1583 PetscCall(PetscFree(A->data)); 1584 1585 PetscCall(PetscObjectChangeTypeName((PetscObject)A, NULL)); 1586 PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatSeqBAIJGetArray_C", NULL)); 1587 PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatSeqBAIJRestoreArray_C", NULL)); 1588 PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatStoreValues_C", NULL)); 1589 PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatRetrieveValues_C", NULL)); 1590 PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatSeqBAIJSetColumnIndices_C", NULL)); 1591 PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatConvert_seqbaij_seqaij_C", NULL)); 1592 PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatConvert_seqbaij_seqsbaij_C", NULL)); 1593 PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatSeqBAIJSetPreallocation_C", NULL)); 1594 PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatSeqBAIJSetPreallocationCSR_C", NULL)); 1595 PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatConvert_seqbaij_seqbstrm_C", NULL)); 1596 PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatIsTranspose_C", NULL)); 1597 #if defined(PETSC_HAVE_HYPRE) 1598 PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatConvert_seqbaij_hypre_C", NULL)); 1599 #endif 1600 PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatConvert_seqbaij_is_C", NULL)); 1601 PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatFactorGetSolverType_C", NULL)); 1602 PetscFunctionReturn(PETSC_SUCCESS); 1603 } 1604 1605 static PetscErrorCode MatSetOption_SeqBAIJ(Mat A, MatOption op, PetscBool flg) 1606 { 1607 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 1608 1609 PetscFunctionBegin; 1610 switch (op) { 1611 case MAT_ROW_ORIENTED: 1612 a->roworiented = flg; 1613 break; 1614 case MAT_KEEP_NONZERO_PATTERN: 1615 a->keepnonzeropattern = flg; 1616 break; 1617 case MAT_NEW_NONZERO_LOCATIONS: 1618 a->nonew = (flg ? 0 : 1); 1619 break; 1620 case MAT_NEW_NONZERO_LOCATION_ERR: 1621 a->nonew = (flg ? -1 : 0); 1622 break; 1623 case MAT_NEW_NONZERO_ALLOCATION_ERR: 1624 a->nonew = (flg ? -2 : 0); 1625 break; 1626 case MAT_UNUSED_NONZERO_LOCATION_ERR: 1627 a->nounused = (flg ? -1 : 0); 1628 break; 1629 case MAT_FORCE_DIAGONAL_ENTRIES: 1630 case MAT_IGNORE_OFF_PROC_ENTRIES: 1631 case MAT_USE_HASH_TABLE: 1632 case MAT_SORTED_FULL: 1633 PetscCall(PetscInfo(A, "Option %s ignored\n", MatOptions[op])); 1634 break; 1635 case MAT_SPD: 1636 case MAT_SYMMETRIC: 1637 case MAT_STRUCTURALLY_SYMMETRIC: 1638 case MAT_HERMITIAN: 1639 case MAT_SYMMETRY_ETERNAL: 1640 case MAT_STRUCTURAL_SYMMETRY_ETERNAL: 1641 case MAT_SUBMAT_SINGLEIS: 1642 case MAT_STRUCTURE_ONLY: 1643 case MAT_SPD_ETERNAL: 1644 /* if the diagonal matrix is square it inherits some of the properties above */ 1645 break; 1646 default: 1647 SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "unknown option %d", op); 1648 } 1649 PetscFunctionReturn(PETSC_SUCCESS); 1650 } 1651 1652 /* used for both SeqBAIJ and SeqSBAIJ matrices */ 1653 PetscErrorCode MatGetRow_SeqBAIJ_private(Mat A, PetscInt row, PetscInt *nz, PetscInt **idx, PetscScalar **v, PetscInt *ai, PetscInt *aj, PetscScalar *aa) 1654 { 1655 PetscInt itmp, i, j, k, M, bn, bp, *idx_i, bs, bs2; 1656 MatScalar *aa_i; 1657 PetscScalar *v_i; 1658 1659 PetscFunctionBegin; 1660 bs = A->rmap->bs; 1661 bs2 = bs * bs; 1662 PetscCheck(row >= 0 && row < A->rmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Row %" PetscInt_FMT " out of range", row); 1663 1664 bn = row / bs; /* Block number */ 1665 bp = row % bs; /* Block Position */ 1666 M = ai[bn + 1] - ai[bn]; 1667 *nz = bs * M; 1668 1669 if (v) { 1670 *v = NULL; 1671 if (*nz) { 1672 PetscCall(PetscMalloc1(*nz, v)); 1673 for (i = 0; i < M; i++) { /* for each block in the block row */ 1674 v_i = *v + i * bs; 1675 aa_i = aa + bs2 * (ai[bn] + i); 1676 for (j = bp, k = 0; j < bs2; j += bs, k++) v_i[k] = aa_i[j]; 1677 } 1678 } 1679 } 1680 1681 if (idx) { 1682 *idx = NULL; 1683 if (*nz) { 1684 PetscCall(PetscMalloc1(*nz, idx)); 1685 for (i = 0; i < M; i++) { /* for each block in the block row */ 1686 idx_i = *idx + i * bs; 1687 itmp = bs * aj[ai[bn] + i]; 1688 for (j = 0; j < bs; j++) idx_i[j] = itmp++; 1689 } 1690 } 1691 } 1692 PetscFunctionReturn(PETSC_SUCCESS); 1693 } 1694 1695 PetscErrorCode MatGetRow_SeqBAIJ(Mat A, PetscInt row, PetscInt *nz, PetscInt **idx, PetscScalar **v) 1696 { 1697 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 1698 1699 PetscFunctionBegin; 1700 PetscCall(MatGetRow_SeqBAIJ_private(A, row, nz, idx, v, a->i, a->j, a->a)); 1701 PetscFunctionReturn(PETSC_SUCCESS); 1702 } 1703 1704 PetscErrorCode MatRestoreRow_SeqBAIJ(Mat A, PetscInt row, PetscInt *nz, PetscInt **idx, PetscScalar **v) 1705 { 1706 PetscFunctionBegin; 1707 if (idx) PetscCall(PetscFree(*idx)); 1708 if (v) PetscCall(PetscFree(*v)); 1709 PetscFunctionReturn(PETSC_SUCCESS); 1710 } 1711 1712 static PetscErrorCode MatTranspose_SeqBAIJ(Mat A, MatReuse reuse, Mat *B) 1713 { 1714 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data, *at; 1715 Mat C; 1716 PetscInt i, j, k, *aj = a->j, *ai = a->i, bs = A->rmap->bs, mbs = a->mbs, nbs = a->nbs, *atfill; 1717 PetscInt bs2 = a->bs2, *ati, *atj, anzj, kr; 1718 MatScalar *ata, *aa = a->a; 1719 1720 PetscFunctionBegin; 1721 if (reuse == MAT_REUSE_MATRIX) PetscCall(MatTransposeCheckNonzeroState_Private(A, *B)); 1722 PetscCall(PetscCalloc1(1 + nbs, &atfill)); 1723 if (reuse == MAT_INITIAL_MATRIX || reuse == MAT_INPLACE_MATRIX) { 1724 for (i = 0; i < ai[mbs]; i++) atfill[aj[i]] += 1; /* count num of non-zeros in row aj[i] */ 1725 1726 PetscCall(MatCreate(PetscObjectComm((PetscObject)A), &C)); 1727 PetscCall(MatSetSizes(C, A->cmap->n, A->rmap->N, A->cmap->n, A->rmap->N)); 1728 PetscCall(MatSetType(C, ((PetscObject)A)->type_name)); 1729 PetscCall(MatSeqBAIJSetPreallocation(C, bs, 0, atfill)); 1730 1731 at = (Mat_SeqBAIJ *)C->data; 1732 ati = at->i; 1733 for (i = 0; i < nbs; i++) at->ilen[i] = at->imax[i] = ati[i + 1] - ati[i]; 1734 } else { 1735 C = *B; 1736 at = (Mat_SeqBAIJ *)C->data; 1737 ati = at->i; 1738 } 1739 1740 atj = at->j; 1741 ata = at->a; 1742 1743 /* Copy ati into atfill so we have locations of the next free space in atj */ 1744 PetscCall(PetscArraycpy(atfill, ati, nbs)); 1745 1746 /* Walk through A row-wise and mark nonzero entries of A^T. */ 1747 for (i = 0; i < mbs; i++) { 1748 anzj = ai[i + 1] - ai[i]; 1749 for (j = 0; j < anzj; j++) { 1750 atj[atfill[*aj]] = i; 1751 for (kr = 0; kr < bs; kr++) { 1752 for (k = 0; k < bs; k++) ata[bs2 * atfill[*aj] + k * bs + kr] = *aa++; 1753 } 1754 atfill[*aj++] += 1; 1755 } 1756 } 1757 PetscCall(MatAssemblyBegin(C, MAT_FINAL_ASSEMBLY)); 1758 PetscCall(MatAssemblyEnd(C, MAT_FINAL_ASSEMBLY)); 1759 1760 /* Clean up temporary space and complete requests. */ 1761 PetscCall(PetscFree(atfill)); 1762 1763 if (reuse == MAT_INITIAL_MATRIX || reuse == MAT_REUSE_MATRIX) { 1764 PetscCall(MatSetBlockSizes(C, PetscAbs(A->cmap->bs), PetscAbs(A->rmap->bs))); 1765 *B = C; 1766 } else { 1767 PetscCall(MatHeaderMerge(A, &C)); 1768 } 1769 PetscFunctionReturn(PETSC_SUCCESS); 1770 } 1771 1772 static PetscErrorCode MatIsTranspose_SeqBAIJ(Mat A, Mat B, PetscReal tol, PetscBool *f) 1773 { 1774 Mat Btrans; 1775 1776 PetscFunctionBegin; 1777 *f = PETSC_FALSE; 1778 PetscCall(MatTranspose(A, MAT_INITIAL_MATRIX, &Btrans)); 1779 PetscCall(MatEqual_SeqBAIJ(B, Btrans, f)); 1780 PetscCall(MatDestroy(&Btrans)); 1781 PetscFunctionReturn(PETSC_SUCCESS); 1782 } 1783 1784 /* Used for both SeqBAIJ and SeqSBAIJ matrices */ 1785 PetscErrorCode MatView_SeqBAIJ_Binary(Mat mat, PetscViewer viewer) 1786 { 1787 Mat_SeqBAIJ *A = (Mat_SeqBAIJ *)mat->data; 1788 PetscInt header[4], M, N, m, bs, nz, cnt, i, j, k, l; 1789 PetscInt *rowlens, *colidxs; 1790 PetscScalar *matvals; 1791 1792 PetscFunctionBegin; 1793 PetscCall(PetscViewerSetUp(viewer)); 1794 1795 M = mat->rmap->N; 1796 N = mat->cmap->N; 1797 m = mat->rmap->n; 1798 bs = mat->rmap->bs; 1799 nz = bs * bs * A->nz; 1800 1801 /* write matrix header */ 1802 header[0] = MAT_FILE_CLASSID; 1803 header[1] = M; 1804 header[2] = N; 1805 header[3] = nz; 1806 PetscCall(PetscViewerBinaryWrite(viewer, header, 4, PETSC_INT)); 1807 1808 /* store row lengths */ 1809 PetscCall(PetscMalloc1(m, &rowlens)); 1810 for (cnt = 0, i = 0; i < A->mbs; i++) 1811 for (j = 0; j < bs; j++) rowlens[cnt++] = bs * (A->i[i + 1] - A->i[i]); 1812 PetscCall(PetscViewerBinaryWrite(viewer, rowlens, m, PETSC_INT)); 1813 PetscCall(PetscFree(rowlens)); 1814 1815 /* store column indices */ 1816 PetscCall(PetscMalloc1(nz, &colidxs)); 1817 for (cnt = 0, i = 0; i < A->mbs; i++) 1818 for (k = 0; k < bs; k++) 1819 for (j = A->i[i]; j < A->i[i + 1]; j++) 1820 for (l = 0; l < bs; l++) colidxs[cnt++] = bs * A->j[j] + l; 1821 PetscCheck(cnt == nz, PETSC_COMM_SELF, PETSC_ERR_LIB, "Internal PETSc error: cnt = %" PetscInt_FMT " nz = %" PetscInt_FMT, cnt, nz); 1822 PetscCall(PetscViewerBinaryWrite(viewer, colidxs, nz, PETSC_INT)); 1823 PetscCall(PetscFree(colidxs)); 1824 1825 /* store nonzero values */ 1826 PetscCall(PetscMalloc1(nz, &matvals)); 1827 for (cnt = 0, i = 0; i < A->mbs; i++) 1828 for (k = 0; k < bs; k++) 1829 for (j = A->i[i]; j < A->i[i + 1]; j++) 1830 for (l = 0; l < bs; l++) matvals[cnt++] = A->a[bs * (bs * j + l) + k]; 1831 PetscCheck(cnt == nz, PETSC_COMM_SELF, PETSC_ERR_LIB, "Internal PETSc error: cnt = %" PetscInt_FMT " nz = %" PetscInt_FMT, cnt, nz); 1832 PetscCall(PetscViewerBinaryWrite(viewer, matvals, nz, PETSC_SCALAR)); 1833 PetscCall(PetscFree(matvals)); 1834 1835 /* write block size option to the viewer's .info file */ 1836 PetscCall(MatView_Binary_BlockSizes(mat, viewer)); 1837 PetscFunctionReturn(PETSC_SUCCESS); 1838 } 1839 1840 static PetscErrorCode MatView_SeqBAIJ_ASCII_structonly(Mat A, PetscViewer viewer) 1841 { 1842 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 1843 PetscInt i, bs = A->rmap->bs, k; 1844 1845 PetscFunctionBegin; 1846 PetscCall(PetscViewerASCIIUseTabs(viewer, PETSC_FALSE)); 1847 for (i = 0; i < a->mbs; i++) { 1848 PetscCall(PetscViewerASCIIPrintf(viewer, "row %" PetscInt_FMT "-%" PetscInt_FMT ":", i * bs, i * bs + bs - 1)); 1849 for (k = a->i[i]; k < a->i[i + 1]; k++) PetscCall(PetscViewerASCIIPrintf(viewer, " (%" PetscInt_FMT "-%" PetscInt_FMT ") ", bs * a->j[k], bs * a->j[k] + bs - 1)); 1850 PetscCall(PetscViewerASCIIPrintf(viewer, "\n")); 1851 } 1852 PetscCall(PetscViewerASCIIUseTabs(viewer, PETSC_TRUE)); 1853 PetscFunctionReturn(PETSC_SUCCESS); 1854 } 1855 1856 static PetscErrorCode MatView_SeqBAIJ_ASCII(Mat A, PetscViewer viewer) 1857 { 1858 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 1859 PetscInt i, j, bs = A->rmap->bs, k, l, bs2 = a->bs2; 1860 PetscViewerFormat format; 1861 1862 PetscFunctionBegin; 1863 if (A->structure_only) { 1864 PetscCall(MatView_SeqBAIJ_ASCII_structonly(A, viewer)); 1865 PetscFunctionReturn(PETSC_SUCCESS); 1866 } 1867 1868 PetscCall(PetscViewerGetFormat(viewer, &format)); 1869 if (format == PETSC_VIEWER_ASCII_INFO || format == PETSC_VIEWER_ASCII_INFO_DETAIL) { 1870 PetscCall(PetscViewerASCIIPrintf(viewer, " block size is %" PetscInt_FMT "\n", bs)); 1871 } else if (format == PETSC_VIEWER_ASCII_MATLAB) { 1872 const char *matname; 1873 Mat aij; 1874 PetscCall(MatConvert(A, MATSEQAIJ, MAT_INITIAL_MATRIX, &aij)); 1875 PetscCall(PetscObjectGetName((PetscObject)A, &matname)); 1876 PetscCall(PetscObjectSetName((PetscObject)aij, matname)); 1877 PetscCall(MatView(aij, viewer)); 1878 PetscCall(MatDestroy(&aij)); 1879 } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO) { 1880 PetscFunctionReturn(PETSC_SUCCESS); 1881 } else if (format == PETSC_VIEWER_ASCII_COMMON) { 1882 PetscCall(PetscViewerASCIIUseTabs(viewer, PETSC_FALSE)); 1883 for (i = 0; i < a->mbs; i++) { 1884 for (j = 0; j < bs; j++) { 1885 PetscCall(PetscViewerASCIIPrintf(viewer, "row %" PetscInt_FMT ":", i * bs + j)); 1886 for (k = a->i[i]; k < a->i[i + 1]; k++) { 1887 for (l = 0; l < bs; l++) { 1888 #if defined(PETSC_USE_COMPLEX) 1889 if (PetscImaginaryPart(a->a[bs2 * k + l * bs + j]) > 0.0 && PetscRealPart(a->a[bs2 * k + l * bs + j]) != 0.0) { 1890 PetscCall(PetscViewerASCIIPrintf(viewer, " (%" PetscInt_FMT ", %g + %gi) ", bs * a->j[k] + l, (double)PetscRealPart(a->a[bs2 * k + l * bs + j]), (double)PetscImaginaryPart(a->a[bs2 * k + l * bs + j]))); 1891 } else if (PetscImaginaryPart(a->a[bs2 * k + l * bs + j]) < 0.0 && PetscRealPart(a->a[bs2 * k + l * bs + j]) != 0.0) { 1892 PetscCall(PetscViewerASCIIPrintf(viewer, " (%" PetscInt_FMT ", %g - %gi) ", bs * a->j[k] + l, (double)PetscRealPart(a->a[bs2 * k + l * bs + j]), -(double)PetscImaginaryPart(a->a[bs2 * k + l * bs + j]))); 1893 } else if (PetscRealPart(a->a[bs2 * k + l * bs + j]) != 0.0) { 1894 PetscCall(PetscViewerASCIIPrintf(viewer, " (%" PetscInt_FMT ", %g) ", bs * a->j[k] + l, (double)PetscRealPart(a->a[bs2 * k + l * bs + j]))); 1895 } 1896 #else 1897 if (a->a[bs2 * k + l * bs + j] != 0.0) PetscCall(PetscViewerASCIIPrintf(viewer, " (%" PetscInt_FMT ", %g) ", bs * a->j[k] + l, (double)a->a[bs2 * k + l * bs + j])); 1898 #endif 1899 } 1900 } 1901 PetscCall(PetscViewerASCIIPrintf(viewer, "\n")); 1902 } 1903 } 1904 PetscCall(PetscViewerASCIIUseTabs(viewer, PETSC_TRUE)); 1905 } else { 1906 PetscCall(PetscViewerASCIIUseTabs(viewer, PETSC_FALSE)); 1907 for (i = 0; i < a->mbs; i++) { 1908 for (j = 0; j < bs; j++) { 1909 PetscCall(PetscViewerASCIIPrintf(viewer, "row %" PetscInt_FMT ":", i * bs + j)); 1910 for (k = a->i[i]; k < a->i[i + 1]; k++) { 1911 for (l = 0; l < bs; l++) { 1912 #if defined(PETSC_USE_COMPLEX) 1913 if (PetscImaginaryPart(a->a[bs2 * k + l * bs + j]) > 0.0) { 1914 PetscCall(PetscViewerASCIIPrintf(viewer, " (%" PetscInt_FMT ", %g + %g i) ", bs * a->j[k] + l, (double)PetscRealPart(a->a[bs2 * k + l * bs + j]), (double)PetscImaginaryPart(a->a[bs2 * k + l * bs + j]))); 1915 } else if (PetscImaginaryPart(a->a[bs2 * k + l * bs + j]) < 0.0) { 1916 PetscCall(PetscViewerASCIIPrintf(viewer, " (%" PetscInt_FMT ", %g - %g i) ", bs * a->j[k] + l, (double)PetscRealPart(a->a[bs2 * k + l * bs + j]), -(double)PetscImaginaryPart(a->a[bs2 * k + l * bs + j]))); 1917 } else { 1918 PetscCall(PetscViewerASCIIPrintf(viewer, " (%" PetscInt_FMT ", %g) ", bs * a->j[k] + l, (double)PetscRealPart(a->a[bs2 * k + l * bs + j]))); 1919 } 1920 #else 1921 PetscCall(PetscViewerASCIIPrintf(viewer, " (%" PetscInt_FMT ", %g) ", bs * a->j[k] + l, (double)a->a[bs2 * k + l * bs + j])); 1922 #endif 1923 } 1924 } 1925 PetscCall(PetscViewerASCIIPrintf(viewer, "\n")); 1926 } 1927 } 1928 PetscCall(PetscViewerASCIIUseTabs(viewer, PETSC_TRUE)); 1929 } 1930 PetscCall(PetscViewerFlush(viewer)); 1931 PetscFunctionReturn(PETSC_SUCCESS); 1932 } 1933 1934 #include <petscdraw.h> 1935 static PetscErrorCode MatView_SeqBAIJ_Draw_Zoom(PetscDraw draw, void *Aa) 1936 { 1937 Mat A = (Mat)Aa; 1938 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 1939 PetscInt row, i, j, k, l, mbs = a->mbs, color, bs = A->rmap->bs, bs2 = a->bs2; 1940 PetscReal xl, yl, xr, yr, x_l, x_r, y_l, y_r; 1941 MatScalar *aa; 1942 PetscViewer viewer; 1943 PetscViewerFormat format; 1944 1945 PetscFunctionBegin; 1946 PetscCall(PetscObjectQuery((PetscObject)A, "Zoomviewer", (PetscObject *)&viewer)); 1947 PetscCall(PetscViewerGetFormat(viewer, &format)); 1948 PetscCall(PetscDrawGetCoordinates(draw, &xl, &yl, &xr, &yr)); 1949 1950 /* loop over matrix elements drawing boxes */ 1951 1952 if (format != PETSC_VIEWER_DRAW_CONTOUR) { 1953 PetscDrawCollectiveBegin(draw); 1954 /* Blue for negative, Cyan for zero and Red for positive */ 1955 color = PETSC_DRAW_BLUE; 1956 for (i = 0, row = 0; i < mbs; i++, row += bs) { 1957 for (j = a->i[i]; j < a->i[i + 1]; j++) { 1958 y_l = A->rmap->N - row - 1.0; 1959 y_r = y_l + 1.0; 1960 x_l = a->j[j] * bs; 1961 x_r = x_l + 1.0; 1962 aa = a->a + j * bs2; 1963 for (k = 0; k < bs; k++) { 1964 for (l = 0; l < bs; l++) { 1965 if (PetscRealPart(*aa++) >= 0.) continue; 1966 PetscCall(PetscDrawRectangle(draw, x_l + k, y_l - l, x_r + k, y_r - l, color, color, color, color)); 1967 } 1968 } 1969 } 1970 } 1971 color = PETSC_DRAW_CYAN; 1972 for (i = 0, row = 0; i < mbs; i++, row += bs) { 1973 for (j = a->i[i]; j < a->i[i + 1]; j++) { 1974 y_l = A->rmap->N - row - 1.0; 1975 y_r = y_l + 1.0; 1976 x_l = a->j[j] * bs; 1977 x_r = x_l + 1.0; 1978 aa = a->a + j * bs2; 1979 for (k = 0; k < bs; k++) { 1980 for (l = 0; l < bs; l++) { 1981 if (PetscRealPart(*aa++) != 0.) continue; 1982 PetscCall(PetscDrawRectangle(draw, x_l + k, y_l - l, x_r + k, y_r - l, color, color, color, color)); 1983 } 1984 } 1985 } 1986 } 1987 color = PETSC_DRAW_RED; 1988 for (i = 0, row = 0; i < mbs; i++, row += bs) { 1989 for (j = a->i[i]; j < a->i[i + 1]; j++) { 1990 y_l = A->rmap->N - row - 1.0; 1991 y_r = y_l + 1.0; 1992 x_l = a->j[j] * bs; 1993 x_r = x_l + 1.0; 1994 aa = a->a + j * bs2; 1995 for (k = 0; k < bs; k++) { 1996 for (l = 0; l < bs; l++) { 1997 if (PetscRealPart(*aa++) <= 0.) continue; 1998 PetscCall(PetscDrawRectangle(draw, x_l + k, y_l - l, x_r + k, y_r - l, color, color, color, color)); 1999 } 2000 } 2001 } 2002 } 2003 PetscDrawCollectiveEnd(draw); 2004 } else { 2005 /* use contour shading to indicate magnitude of values */ 2006 /* first determine max of all nonzero values */ 2007 PetscReal minv = 0.0, maxv = 0.0; 2008 PetscDraw popup; 2009 2010 for (i = 0; i < a->nz * a->bs2; i++) { 2011 if (PetscAbsScalar(a->a[i]) > maxv) maxv = PetscAbsScalar(a->a[i]); 2012 } 2013 if (minv >= maxv) maxv = minv + PETSC_SMALL; 2014 PetscCall(PetscDrawGetPopup(draw, &popup)); 2015 PetscCall(PetscDrawScalePopup(popup, 0.0, maxv)); 2016 2017 PetscDrawCollectiveBegin(draw); 2018 for (i = 0, row = 0; i < mbs; i++, row += bs) { 2019 for (j = a->i[i]; j < a->i[i + 1]; j++) { 2020 y_l = A->rmap->N - row - 1.0; 2021 y_r = y_l + 1.0; 2022 x_l = a->j[j] * bs; 2023 x_r = x_l + 1.0; 2024 aa = a->a + j * bs2; 2025 for (k = 0; k < bs; k++) { 2026 for (l = 0; l < bs; l++) { 2027 MatScalar v = *aa++; 2028 color = PetscDrawRealToColor(PetscAbsScalar(v), minv, maxv); 2029 PetscCall(PetscDrawRectangle(draw, x_l + k, y_l - l, x_r + k, y_r - l, color, color, color, color)); 2030 } 2031 } 2032 } 2033 } 2034 PetscDrawCollectiveEnd(draw); 2035 } 2036 PetscFunctionReturn(PETSC_SUCCESS); 2037 } 2038 2039 static PetscErrorCode MatView_SeqBAIJ_Draw(Mat A, PetscViewer viewer) 2040 { 2041 PetscReal xl, yl, xr, yr, w, h; 2042 PetscDraw draw; 2043 PetscBool isnull; 2044 2045 PetscFunctionBegin; 2046 PetscCall(PetscViewerDrawGetDraw(viewer, 0, &draw)); 2047 PetscCall(PetscDrawIsNull(draw, &isnull)); 2048 if (isnull) PetscFunctionReturn(PETSC_SUCCESS); 2049 2050 xr = A->cmap->n; 2051 yr = A->rmap->N; 2052 h = yr / 10.0; 2053 w = xr / 10.0; 2054 xr += w; 2055 yr += h; 2056 xl = -w; 2057 yl = -h; 2058 PetscCall(PetscDrawSetCoordinates(draw, xl, yl, xr, yr)); 2059 PetscCall(PetscObjectCompose((PetscObject)A, "Zoomviewer", (PetscObject)viewer)); 2060 PetscCall(PetscDrawZoom(draw, MatView_SeqBAIJ_Draw_Zoom, A)); 2061 PetscCall(PetscObjectCompose((PetscObject)A, "Zoomviewer", NULL)); 2062 PetscCall(PetscDrawSave(draw)); 2063 PetscFunctionReturn(PETSC_SUCCESS); 2064 } 2065 2066 PetscErrorCode MatView_SeqBAIJ(Mat A, PetscViewer viewer) 2067 { 2068 PetscBool iascii, isbinary, isdraw; 2069 2070 PetscFunctionBegin; 2071 PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &iascii)); 2072 PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERBINARY, &isbinary)); 2073 PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERDRAW, &isdraw)); 2074 if (iascii) { 2075 PetscCall(MatView_SeqBAIJ_ASCII(A, viewer)); 2076 } else if (isbinary) { 2077 PetscCall(MatView_SeqBAIJ_Binary(A, viewer)); 2078 } else if (isdraw) { 2079 PetscCall(MatView_SeqBAIJ_Draw(A, viewer)); 2080 } else { 2081 Mat B; 2082 PetscCall(MatConvert(A, MATSEQAIJ, MAT_INITIAL_MATRIX, &B)); 2083 PetscCall(MatView(B, viewer)); 2084 PetscCall(MatDestroy(&B)); 2085 } 2086 PetscFunctionReturn(PETSC_SUCCESS); 2087 } 2088 2089 PetscErrorCode MatGetValues_SeqBAIJ(Mat A, PetscInt m, const PetscInt im[], PetscInt n, const PetscInt in[], PetscScalar v[]) 2090 { 2091 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 2092 PetscInt *rp, k, low, high, t, row, nrow, i, col, l, *aj = a->j; 2093 PetscInt *ai = a->i, *ailen = a->ilen; 2094 PetscInt brow, bcol, ridx, cidx, bs = A->rmap->bs, bs2 = a->bs2; 2095 MatScalar *ap, *aa = a->a; 2096 2097 PetscFunctionBegin; 2098 for (k = 0; k < m; k++) { /* loop over rows */ 2099 row = im[k]; 2100 brow = row / bs; 2101 if (row < 0) { 2102 v += n; 2103 continue; 2104 } /* negative row */ 2105 PetscCheck(row < A->rmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Row %" PetscInt_FMT " too large", row); 2106 rp = PetscSafePointerPlusOffset(aj, ai[brow]); 2107 ap = PetscSafePointerPlusOffset(aa, bs2 * ai[brow]); 2108 nrow = ailen[brow]; 2109 for (l = 0; l < n; l++) { /* loop over columns */ 2110 if (in[l] < 0) { 2111 v++; 2112 continue; 2113 } /* negative column */ 2114 PetscCheck(in[l] < A->cmap->n, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Column %" PetscInt_FMT " too large", in[l]); 2115 col = in[l]; 2116 bcol = col / bs; 2117 cidx = col % bs; 2118 ridx = row % bs; 2119 high = nrow; 2120 low = 0; /* assume unsorted */ 2121 while (high - low > 5) { 2122 t = (low + high) / 2; 2123 if (rp[t] > bcol) high = t; 2124 else low = t; 2125 } 2126 for (i = low; i < high; i++) { 2127 if (rp[i] > bcol) break; 2128 if (rp[i] == bcol) { 2129 *v++ = ap[bs2 * i + bs * cidx + ridx]; 2130 goto finished; 2131 } 2132 } 2133 *v++ = 0.0; 2134 finished:; 2135 } 2136 } 2137 PetscFunctionReturn(PETSC_SUCCESS); 2138 } 2139 2140 PetscErrorCode MatSetValuesBlocked_SeqBAIJ(Mat A, PetscInt m, const PetscInt im[], PetscInt n, const PetscInt in[], const PetscScalar v[], InsertMode is) 2141 { 2142 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 2143 PetscInt *rp, k, low, high, t, ii, jj, row, nrow, i, col, l, rmax, N, lastcol = -1; 2144 PetscInt *imax = a->imax, *ai = a->i, *ailen = a->ilen; 2145 PetscInt *aj = a->j, nonew = a->nonew, bs2 = a->bs2, bs = A->rmap->bs, stepval; 2146 PetscBool roworiented = a->roworiented; 2147 const PetscScalar *value = v; 2148 MatScalar *ap = NULL, *aa = a->a, *bap; 2149 2150 PetscFunctionBegin; 2151 if (roworiented) { 2152 stepval = (n - 1) * bs; 2153 } else { 2154 stepval = (m - 1) * bs; 2155 } 2156 for (k = 0; k < m; k++) { /* loop over added rows */ 2157 row = im[k]; 2158 if (row < 0) continue; 2159 PetscCheck(row < a->mbs, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Block row index too large %" PetscInt_FMT " max %" PetscInt_FMT, row, a->mbs - 1); 2160 rp = aj + ai[row]; 2161 if (!A->structure_only) ap = aa + bs2 * ai[row]; 2162 rmax = imax[row]; 2163 nrow = ailen[row]; 2164 low = 0; 2165 high = nrow; 2166 for (l = 0; l < n; l++) { /* loop over added columns */ 2167 if (in[l] < 0) continue; 2168 PetscCheck(in[l] < a->nbs, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Block column index too large %" PetscInt_FMT " max %" PetscInt_FMT, in[l], a->nbs - 1); 2169 col = in[l]; 2170 if (!A->structure_only) { 2171 if (roworiented) { 2172 value = v + (k * (stepval + bs) + l) * bs; 2173 } else { 2174 value = v + (l * (stepval + bs) + k) * bs; 2175 } 2176 } 2177 if (col <= lastcol) low = 0; 2178 else high = nrow; 2179 lastcol = col; 2180 while (high - low > 7) { 2181 t = (low + high) / 2; 2182 if (rp[t] > col) high = t; 2183 else low = t; 2184 } 2185 for (i = low; i < high; i++) { 2186 if (rp[i] > col) break; 2187 if (rp[i] == col) { 2188 if (A->structure_only) goto noinsert2; 2189 bap = ap + bs2 * i; 2190 if (roworiented) { 2191 if (is == ADD_VALUES) { 2192 for (ii = 0; ii < bs; ii++, value += stepval) { 2193 for (jj = ii; jj < bs2; jj += bs) bap[jj] += *value++; 2194 } 2195 } else { 2196 for (ii = 0; ii < bs; ii++, value += stepval) { 2197 for (jj = ii; jj < bs2; jj += bs) bap[jj] = *value++; 2198 } 2199 } 2200 } else { 2201 if (is == ADD_VALUES) { 2202 for (ii = 0; ii < bs; ii++, value += bs + stepval) { 2203 for (jj = 0; jj < bs; jj++) bap[jj] += value[jj]; 2204 bap += bs; 2205 } 2206 } else { 2207 for (ii = 0; ii < bs; ii++, value += bs + stepval) { 2208 for (jj = 0; jj < bs; jj++) bap[jj] = value[jj]; 2209 bap += bs; 2210 } 2211 } 2212 } 2213 goto noinsert2; 2214 } 2215 } 2216 if (nonew == 1) goto noinsert2; 2217 PetscCheck(nonew != -1, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Inserting a new blocked index new nonzero block (%" PetscInt_FMT ", %" PetscInt_FMT ") in the matrix", row, col); 2218 if (A->structure_only) { 2219 MatSeqXAIJReallocateAIJ_structure_only(A, a->mbs, bs2, nrow, row, col, rmax, ai, aj, rp, imax, nonew, MatScalar); 2220 } else { 2221 MatSeqXAIJReallocateAIJ(A, a->mbs, bs2, nrow, row, col, rmax, aa, ai, aj, rp, ap, imax, nonew, MatScalar); 2222 } 2223 N = nrow++ - 1; 2224 high++; 2225 /* shift up all the later entries in this row */ 2226 PetscCall(PetscArraymove(rp + i + 1, rp + i, N - i + 1)); 2227 rp[i] = col; 2228 if (!A->structure_only) { 2229 PetscCall(PetscArraymove(ap + bs2 * (i + 1), ap + bs2 * i, bs2 * (N - i + 1))); 2230 bap = ap + bs2 * i; 2231 if (roworiented) { 2232 for (ii = 0; ii < bs; ii++, value += stepval) { 2233 for (jj = ii; jj < bs2; jj += bs) bap[jj] = *value++; 2234 } 2235 } else { 2236 for (ii = 0; ii < bs; ii++, value += stepval) { 2237 for (jj = 0; jj < bs; jj++) *bap++ = *value++; 2238 } 2239 } 2240 } 2241 noinsert2:; 2242 low = i; 2243 } 2244 ailen[row] = nrow; 2245 } 2246 PetscFunctionReturn(PETSC_SUCCESS); 2247 } 2248 2249 PetscErrorCode MatAssemblyEnd_SeqBAIJ(Mat A, MatAssemblyType mode) 2250 { 2251 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 2252 PetscInt fshift = 0, i, *ai = a->i, *aj = a->j, *imax = a->imax; 2253 PetscInt m = A->rmap->N, *ip, N, *ailen = a->ilen; 2254 PetscInt mbs = a->mbs, bs2 = a->bs2, rmax = 0; 2255 MatScalar *aa = a->a, *ap; 2256 PetscReal ratio = 0.6; 2257 2258 PetscFunctionBegin; 2259 if (mode == MAT_FLUSH_ASSEMBLY || (A->was_assembled && A->ass_nonzerostate == A->nonzerostate)) PetscFunctionReturn(PETSC_SUCCESS); 2260 2261 if (m) rmax = ailen[0]; 2262 for (i = 1; i < mbs; i++) { 2263 /* move each row back by the amount of empty slots (fshift) before it*/ 2264 fshift += imax[i - 1] - ailen[i - 1]; 2265 rmax = PetscMax(rmax, ailen[i]); 2266 if (fshift) { 2267 ip = aj + ai[i]; 2268 ap = aa + bs2 * ai[i]; 2269 N = ailen[i]; 2270 PetscCall(PetscArraymove(ip - fshift, ip, N)); 2271 if (!A->structure_only) PetscCall(PetscArraymove(ap - bs2 * fshift, ap, bs2 * N)); 2272 } 2273 ai[i] = ai[i - 1] + ailen[i - 1]; 2274 } 2275 if (mbs) { 2276 fshift += imax[mbs - 1] - ailen[mbs - 1]; 2277 ai[mbs] = ai[mbs - 1] + ailen[mbs - 1]; 2278 } 2279 2280 /* reset ilen and imax for each row */ 2281 a->nonzerorowcnt = 0; 2282 if (A->structure_only) { 2283 PetscCall(PetscFree2(a->imax, a->ilen)); 2284 } else { /* !A->structure_only */ 2285 for (i = 0; i < mbs; i++) { 2286 ailen[i] = imax[i] = ai[i + 1] - ai[i]; 2287 a->nonzerorowcnt += ((ai[i + 1] - ai[i]) > 0); 2288 } 2289 } 2290 a->nz = ai[mbs]; 2291 2292 /* diagonals may have moved, so kill the diagonal pointers */ 2293 a->idiagvalid = PETSC_FALSE; 2294 if (fshift && a->diag) PetscCall(PetscFree(a->diag)); 2295 if (fshift) PetscCheck(a->nounused != -1, PETSC_COMM_SELF, PETSC_ERR_PLIB, "Unused space detected in matrix: %" PetscInt_FMT " X %" PetscInt_FMT " block size %" PetscInt_FMT ", %" PetscInt_FMT " unneeded", m, A->cmap->n, A->rmap->bs, fshift * bs2); 2296 PetscCall(PetscInfo(A, "Matrix size: %" PetscInt_FMT " X %" PetscInt_FMT ", block size %" PetscInt_FMT "; storage space: %" PetscInt_FMT " unneeded, %" PetscInt_FMT " used\n", m, A->cmap->n, A->rmap->bs, fshift * bs2, a->nz * bs2)); 2297 PetscCall(PetscInfo(A, "Number of mallocs during MatSetValues is %" PetscInt_FMT "\n", a->reallocs)); 2298 PetscCall(PetscInfo(A, "Most nonzeros blocks in any row is %" PetscInt_FMT "\n", rmax)); 2299 2300 A->info.mallocs += a->reallocs; 2301 a->reallocs = 0; 2302 A->info.nz_unneeded = (PetscReal)fshift * bs2; 2303 a->rmax = rmax; 2304 2305 if (!A->structure_only) PetscCall(MatCheckCompressedRow(A, a->nonzerorowcnt, &a->compressedrow, a->i, mbs, ratio)); 2306 PetscFunctionReturn(PETSC_SUCCESS); 2307 } 2308 2309 /* 2310 This function returns an array of flags which indicate the locations of contiguous 2311 blocks that should be zeroed. for eg: if bs = 3 and is = [0,1,2,3,5,6,7,8,9] 2312 then the resulting sizes = [3,1,1,3,1] corresponding to sets [(0,1,2),(3),(5),(6,7,8),(9)] 2313 Assume: sizes should be long enough to hold all the values. 2314 */ 2315 static PetscErrorCode MatZeroRows_SeqBAIJ_Check_Blocks(PetscInt idx[], PetscInt n, PetscInt bs, PetscInt sizes[], PetscInt *bs_max) 2316 { 2317 PetscInt j = 0; 2318 2319 PetscFunctionBegin; 2320 for (PetscInt i = 0; i < n; j++) { 2321 PetscInt row = idx[i]; 2322 if (row % bs != 0) { /* Not the beginning of a block */ 2323 sizes[j] = 1; 2324 i++; 2325 } else if (i + bs > n) { /* complete block doesn't exist (at idx end) */ 2326 sizes[j] = 1; /* Also makes sure at least 'bs' values exist for next else */ 2327 i++; 2328 } else { /* Beginning of the block, so check if the complete block exists */ 2329 PetscBool flg = PETSC_TRUE; 2330 for (PetscInt k = 1; k < bs; k++) { 2331 if (row + k != idx[i + k]) { /* break in the block */ 2332 flg = PETSC_FALSE; 2333 break; 2334 } 2335 } 2336 if (flg) { /* No break in the bs */ 2337 sizes[j] = bs; 2338 i += bs; 2339 } else { 2340 sizes[j] = 1; 2341 i++; 2342 } 2343 } 2344 } 2345 *bs_max = j; 2346 PetscFunctionReturn(PETSC_SUCCESS); 2347 } 2348 2349 PetscErrorCode MatZeroRows_SeqBAIJ(Mat A, PetscInt is_n, const PetscInt is_idx[], PetscScalar diag, Vec x, Vec b) 2350 { 2351 Mat_SeqBAIJ *baij = (Mat_SeqBAIJ *)A->data; 2352 PetscInt i, j, k, count, *rows; 2353 PetscInt bs = A->rmap->bs, bs2 = baij->bs2, *sizes, row, bs_max; 2354 PetscScalar zero = 0.0; 2355 MatScalar *aa; 2356 const PetscScalar *xx; 2357 PetscScalar *bb; 2358 2359 PetscFunctionBegin; 2360 /* fix right hand side if needed */ 2361 if (x && b) { 2362 PetscCall(VecGetArrayRead(x, &xx)); 2363 PetscCall(VecGetArray(b, &bb)); 2364 for (i = 0; i < is_n; i++) bb[is_idx[i]] = diag * xx[is_idx[i]]; 2365 PetscCall(VecRestoreArrayRead(x, &xx)); 2366 PetscCall(VecRestoreArray(b, &bb)); 2367 } 2368 2369 /* Make a copy of the IS and sort it */ 2370 /* allocate memory for rows,sizes */ 2371 PetscCall(PetscMalloc2(is_n, &rows, 2 * is_n, &sizes)); 2372 2373 /* copy IS values to rows, and sort them */ 2374 for (i = 0; i < is_n; i++) rows[i] = is_idx[i]; 2375 PetscCall(PetscSortInt(is_n, rows)); 2376 2377 if (baij->keepnonzeropattern) { 2378 for (i = 0; i < is_n; i++) sizes[i] = 1; 2379 bs_max = is_n; 2380 } else { 2381 PetscCall(MatZeroRows_SeqBAIJ_Check_Blocks(rows, is_n, bs, sizes, &bs_max)); 2382 A->nonzerostate++; 2383 } 2384 2385 for (i = 0, j = 0; i < bs_max; j += sizes[i], i++) { 2386 row = rows[j]; 2387 PetscCheck(row >= 0 && row <= A->rmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "row %" PetscInt_FMT " out of range", row); 2388 count = (baij->i[row / bs + 1] - baij->i[row / bs]) * bs; 2389 aa = ((MatScalar *)(baij->a)) + baij->i[row / bs] * bs2 + (row % bs); 2390 if (sizes[i] == bs && !baij->keepnonzeropattern) { 2391 if (diag != (PetscScalar)0.0) { 2392 if (baij->ilen[row / bs] > 0) { 2393 baij->ilen[row / bs] = 1; 2394 baij->j[baij->i[row / bs]] = row / bs; 2395 2396 PetscCall(PetscArrayzero(aa, count * bs)); 2397 } 2398 /* Now insert all the diagonal values for this bs */ 2399 for (k = 0; k < bs; k++) PetscUseTypeMethod(A, setvalues, 1, rows + j + k, 1, rows + j + k, &diag, INSERT_VALUES); 2400 } else { /* (diag == 0.0) */ 2401 baij->ilen[row / bs] = 0; 2402 } /* end (diag == 0.0) */ 2403 } else { /* (sizes[i] != bs) */ 2404 PetscAssert(sizes[i] == 1, PETSC_COMM_SELF, PETSC_ERR_PLIB, "Internal Error. Value should be 1"); 2405 for (k = 0; k < count; k++) { 2406 aa[0] = zero; 2407 aa += bs; 2408 } 2409 if (diag != (PetscScalar)0.0) PetscUseTypeMethod(A, setvalues, 1, rows + j, 1, rows + j, &diag, INSERT_VALUES); 2410 } 2411 } 2412 2413 PetscCall(PetscFree2(rows, sizes)); 2414 PetscCall(MatAssemblyEnd_SeqBAIJ(A, MAT_FINAL_ASSEMBLY)); 2415 PetscFunctionReturn(PETSC_SUCCESS); 2416 } 2417 2418 static PetscErrorCode MatZeroRowsColumns_SeqBAIJ(Mat A, PetscInt is_n, const PetscInt is_idx[], PetscScalar diag, Vec x, Vec b) 2419 { 2420 Mat_SeqBAIJ *baij = (Mat_SeqBAIJ *)A->data; 2421 PetscInt i, j, k, count; 2422 PetscInt bs = A->rmap->bs, bs2 = baij->bs2, row, col; 2423 PetscScalar zero = 0.0; 2424 MatScalar *aa; 2425 const PetscScalar *xx; 2426 PetscScalar *bb; 2427 PetscBool *zeroed, vecs = PETSC_FALSE; 2428 2429 PetscFunctionBegin; 2430 /* fix right hand side if needed */ 2431 if (x && b) { 2432 PetscCall(VecGetArrayRead(x, &xx)); 2433 PetscCall(VecGetArray(b, &bb)); 2434 vecs = PETSC_TRUE; 2435 } 2436 2437 /* zero the columns */ 2438 PetscCall(PetscCalloc1(A->rmap->n, &zeroed)); 2439 for (i = 0; i < is_n; i++) { 2440 PetscCheck(is_idx[i] >= 0 && is_idx[i] < A->rmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "row %" PetscInt_FMT " out of range", is_idx[i]); 2441 zeroed[is_idx[i]] = PETSC_TRUE; 2442 } 2443 for (i = 0; i < A->rmap->N; i++) { 2444 if (!zeroed[i]) { 2445 row = i / bs; 2446 for (j = baij->i[row]; j < baij->i[row + 1]; j++) { 2447 for (k = 0; k < bs; k++) { 2448 col = bs * baij->j[j] + k; 2449 if (zeroed[col]) { 2450 aa = ((MatScalar *)(baij->a)) + j * bs2 + (i % bs) + bs * k; 2451 if (vecs) bb[i] -= aa[0] * xx[col]; 2452 aa[0] = 0.0; 2453 } 2454 } 2455 } 2456 } else if (vecs) bb[i] = diag * xx[i]; 2457 } 2458 PetscCall(PetscFree(zeroed)); 2459 if (vecs) { 2460 PetscCall(VecRestoreArrayRead(x, &xx)); 2461 PetscCall(VecRestoreArray(b, &bb)); 2462 } 2463 2464 /* zero the rows */ 2465 for (i = 0; i < is_n; i++) { 2466 row = is_idx[i]; 2467 count = (baij->i[row / bs + 1] - baij->i[row / bs]) * bs; 2468 aa = ((MatScalar *)(baij->a)) + baij->i[row / bs] * bs2 + (row % bs); 2469 for (k = 0; k < count; k++) { 2470 aa[0] = zero; 2471 aa += bs; 2472 } 2473 if (diag != (PetscScalar)0.0) PetscUseTypeMethod(A, setvalues, 1, &row, 1, &row, &diag, INSERT_VALUES); 2474 } 2475 PetscCall(MatAssemblyEnd_SeqBAIJ(A, MAT_FINAL_ASSEMBLY)); 2476 PetscFunctionReturn(PETSC_SUCCESS); 2477 } 2478 2479 PetscErrorCode MatSetValues_SeqBAIJ(Mat A, PetscInt m, const PetscInt im[], PetscInt n, const PetscInt in[], const PetscScalar v[], InsertMode is) 2480 { 2481 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 2482 PetscInt *rp, k, low, high, t, ii, row, nrow, i, col, l, rmax, N, lastcol = -1; 2483 PetscInt *imax = a->imax, *ai = a->i, *ailen = a->ilen; 2484 PetscInt *aj = a->j, nonew = a->nonew, bs = A->rmap->bs, brow, bcol; 2485 PetscInt ridx, cidx, bs2 = a->bs2; 2486 PetscBool roworiented = a->roworiented; 2487 MatScalar *ap = NULL, value = 0.0, *aa = a->a, *bap; 2488 2489 PetscFunctionBegin; 2490 for (k = 0; k < m; k++) { /* loop over added rows */ 2491 row = im[k]; 2492 brow = row / bs; 2493 if (row < 0) continue; 2494 PetscCheck(row < A->rmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Row too large: row %" PetscInt_FMT " max %" PetscInt_FMT, row, A->rmap->N - 1); 2495 rp = PetscSafePointerPlusOffset(aj, ai[brow]); 2496 if (!A->structure_only) ap = PetscSafePointerPlusOffset(aa, bs2 * ai[brow]); 2497 rmax = imax[brow]; 2498 nrow = ailen[brow]; 2499 low = 0; 2500 high = nrow; 2501 for (l = 0; l < n; l++) { /* loop over added columns */ 2502 if (in[l] < 0) continue; 2503 PetscCheck(in[l] < A->cmap->n, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Column too large: col %" PetscInt_FMT " max %" PetscInt_FMT, in[l], A->cmap->n - 1); 2504 col = in[l]; 2505 bcol = col / bs; 2506 ridx = row % bs; 2507 cidx = col % bs; 2508 if (!A->structure_only) { 2509 if (roworiented) { 2510 value = v[l + k * n]; 2511 } else { 2512 value = v[k + l * m]; 2513 } 2514 } 2515 if (col <= lastcol) low = 0; 2516 else high = nrow; 2517 lastcol = col; 2518 while (high - low > 7) { 2519 t = (low + high) / 2; 2520 if (rp[t] > bcol) high = t; 2521 else low = t; 2522 } 2523 for (i = low; i < high; i++) { 2524 if (rp[i] > bcol) break; 2525 if (rp[i] == bcol) { 2526 bap = PetscSafePointerPlusOffset(ap, bs2 * i + bs * cidx + ridx); 2527 if (!A->structure_only) { 2528 if (is == ADD_VALUES) *bap += value; 2529 else *bap = value; 2530 } 2531 goto noinsert1; 2532 } 2533 } 2534 if (nonew == 1) goto noinsert1; 2535 PetscCheck(nonew != -1, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Inserting a new nonzero (%" PetscInt_FMT ", %" PetscInt_FMT ") in the matrix", row, col); 2536 if (A->structure_only) { 2537 MatSeqXAIJReallocateAIJ_structure_only(A, a->mbs, bs2, nrow, brow, bcol, rmax, ai, aj, rp, imax, nonew, MatScalar); 2538 } else { 2539 MatSeqXAIJReallocateAIJ(A, a->mbs, bs2, nrow, brow, bcol, rmax, aa, ai, aj, rp, ap, imax, nonew, MatScalar); 2540 } 2541 N = nrow++ - 1; 2542 high++; 2543 /* shift up all the later entries in this row */ 2544 PetscCall(PetscArraymove(rp + i + 1, rp + i, N - i + 1)); 2545 rp[i] = bcol; 2546 if (!A->structure_only) { 2547 PetscCall(PetscArraymove(ap + bs2 * (i + 1), ap + bs2 * i, bs2 * (N - i + 1))); 2548 PetscCall(PetscArrayzero(ap + bs2 * i, bs2)); 2549 ap[bs2 * i + bs * cidx + ridx] = value; 2550 } 2551 a->nz++; 2552 A->nonzerostate++; 2553 noinsert1:; 2554 low = i; 2555 } 2556 ailen[brow] = nrow; 2557 } 2558 PetscFunctionReturn(PETSC_SUCCESS); 2559 } 2560 2561 static PetscErrorCode MatILUFactor_SeqBAIJ(Mat inA, IS row, IS col, const MatFactorInfo *info) 2562 { 2563 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)inA->data; 2564 Mat outA; 2565 PetscBool row_identity, col_identity; 2566 2567 PetscFunctionBegin; 2568 PetscCheck(info->levels == 0, PETSC_COMM_SELF, PETSC_ERR_SUP, "Only levels = 0 supported for in-place ILU"); 2569 PetscCall(ISIdentity(row, &row_identity)); 2570 PetscCall(ISIdentity(col, &col_identity)); 2571 PetscCheck(row_identity && col_identity, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "Row and column permutations must be identity for in-place ILU"); 2572 2573 outA = inA; 2574 inA->factortype = MAT_FACTOR_LU; 2575 PetscCall(PetscFree(inA->solvertype)); 2576 PetscCall(PetscStrallocpy(MATSOLVERPETSC, &inA->solvertype)); 2577 2578 PetscCall(MatMarkDiagonal_SeqBAIJ(inA)); 2579 2580 PetscCall(PetscObjectReference((PetscObject)row)); 2581 PetscCall(ISDestroy(&a->row)); 2582 a->row = row; 2583 PetscCall(PetscObjectReference((PetscObject)col)); 2584 PetscCall(ISDestroy(&a->col)); 2585 a->col = col; 2586 2587 /* Create the invert permutation so that it can be used in MatLUFactorNumeric() */ 2588 PetscCall(ISDestroy(&a->icol)); 2589 PetscCall(ISInvertPermutation(col, PETSC_DECIDE, &a->icol)); 2590 2591 PetscCall(MatSeqBAIJSetNumericFactorization_inplace(inA, (PetscBool)(row_identity && col_identity))); 2592 if (!a->solve_work) PetscCall(PetscMalloc1(inA->rmap->N + inA->rmap->bs, &a->solve_work)); 2593 PetscCall(MatLUFactorNumeric(outA, inA, info)); 2594 PetscFunctionReturn(PETSC_SUCCESS); 2595 } 2596 2597 static PetscErrorCode MatSeqBAIJSetColumnIndices_SeqBAIJ(Mat mat, const PetscInt *indices) 2598 { 2599 Mat_SeqBAIJ *baij = (Mat_SeqBAIJ *)mat->data; 2600 2601 PetscFunctionBegin; 2602 baij->nz = baij->maxnz; 2603 PetscCall(PetscArraycpy(baij->j, indices, baij->nz)); 2604 PetscCall(PetscArraycpy(baij->ilen, baij->imax, baij->mbs)); 2605 PetscFunctionReturn(PETSC_SUCCESS); 2606 } 2607 2608 /*@ 2609 MatSeqBAIJSetColumnIndices - Set the column indices for all the block rows in the matrix. 2610 2611 Input Parameters: 2612 + mat - the `MATSEQBAIJ` matrix 2613 - indices - the block column indices 2614 2615 Level: advanced 2616 2617 Notes: 2618 This can be called if you have precomputed the nonzero structure of the 2619 matrix and want to provide it to the matrix object to improve the performance 2620 of the `MatSetValues()` operation. 2621 2622 You MUST have set the correct numbers of nonzeros per row in the call to 2623 `MatCreateSeqBAIJ()`, and the columns indices MUST be sorted. 2624 2625 MUST be called before any calls to `MatSetValues()` 2626 2627 .seealso: [](ch_matrices), `Mat`, `MATSEQBAIJ`, `MatSetValues()` 2628 @*/ 2629 PetscErrorCode MatSeqBAIJSetColumnIndices(Mat mat, PetscInt *indices) 2630 { 2631 PetscFunctionBegin; 2632 PetscValidHeaderSpecific(mat, MAT_CLASSID, 1); 2633 PetscAssertPointer(indices, 2); 2634 PetscUseMethod(mat, "MatSeqBAIJSetColumnIndices_C", (Mat, const PetscInt *), (mat, (const PetscInt *)indices)); 2635 PetscFunctionReturn(PETSC_SUCCESS); 2636 } 2637 2638 static PetscErrorCode MatGetRowMaxAbs_SeqBAIJ(Mat A, Vec v, PetscInt idx[]) 2639 { 2640 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 2641 PetscInt i, j, n, row, bs, *ai, *aj, mbs; 2642 PetscReal atmp; 2643 PetscScalar *x, zero = 0.0; 2644 MatScalar *aa; 2645 PetscInt ncols, brow, krow, kcol; 2646 2647 PetscFunctionBegin; 2648 /* why is this not a macro???????????????????????????????????????????????????????????????? */ 2649 PetscCheck(!A->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix"); 2650 bs = A->rmap->bs; 2651 aa = a->a; 2652 ai = a->i; 2653 aj = a->j; 2654 mbs = a->mbs; 2655 2656 PetscCall(VecSet(v, zero)); 2657 PetscCall(VecGetArray(v, &x)); 2658 PetscCall(VecGetLocalSize(v, &n)); 2659 PetscCheck(n == A->rmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Nonconforming matrix and vector"); 2660 for (i = 0; i < mbs; i++) { 2661 ncols = ai[1] - ai[0]; 2662 ai++; 2663 brow = bs * i; 2664 for (j = 0; j < ncols; j++) { 2665 for (kcol = 0; kcol < bs; kcol++) { 2666 for (krow = 0; krow < bs; krow++) { 2667 atmp = PetscAbsScalar(*aa); 2668 aa++; 2669 row = brow + krow; /* row index */ 2670 if (PetscAbsScalar(x[row]) < atmp) { 2671 x[row] = atmp; 2672 if (idx) idx[row] = bs * (*aj) + kcol; 2673 } 2674 } 2675 } 2676 aj++; 2677 } 2678 } 2679 PetscCall(VecRestoreArray(v, &x)); 2680 PetscFunctionReturn(PETSC_SUCCESS); 2681 } 2682 2683 static PetscErrorCode MatGetRowSumAbs_SeqBAIJ(Mat A, Vec v) 2684 { 2685 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 2686 PetscInt i, j, n, row, bs, *ai, mbs; 2687 PetscReal atmp; 2688 PetscScalar *x, zero = 0.0; 2689 MatScalar *aa; 2690 PetscInt ncols, brow, krow, kcol; 2691 2692 PetscFunctionBegin; 2693 PetscCheck(!A->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix"); 2694 bs = A->rmap->bs; 2695 aa = a->a; 2696 ai = a->i; 2697 mbs = a->mbs; 2698 2699 PetscCall(VecSet(v, zero)); 2700 PetscCall(VecGetArrayWrite(v, &x)); 2701 PetscCall(VecGetLocalSize(v, &n)); 2702 PetscCheck(n == A->rmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Nonconforming matrix and vector"); 2703 for (i = 0; i < mbs; i++) { 2704 ncols = ai[1] - ai[0]; 2705 ai++; 2706 brow = bs * i; 2707 for (j = 0; j < ncols; j++) { 2708 for (kcol = 0; kcol < bs; kcol++) { 2709 for (krow = 0; krow < bs; krow++) { 2710 atmp = PetscAbsScalar(*aa); 2711 aa++; 2712 row = brow + krow; /* row index */ 2713 x[row] += atmp; 2714 } 2715 } 2716 } 2717 } 2718 PetscCall(VecRestoreArrayWrite(v, &x)); 2719 PetscFunctionReturn(PETSC_SUCCESS); 2720 } 2721 2722 static PetscErrorCode MatCopy_SeqBAIJ(Mat A, Mat B, MatStructure str) 2723 { 2724 PetscFunctionBegin; 2725 /* If the two matrices have the same copy implementation, use fast copy. */ 2726 if (str == SAME_NONZERO_PATTERN && (A->ops->copy == B->ops->copy)) { 2727 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 2728 Mat_SeqBAIJ *b = (Mat_SeqBAIJ *)B->data; 2729 PetscInt ambs = a->mbs, bmbs = b->mbs, abs = A->rmap->bs, bbs = B->rmap->bs, bs2 = abs * abs; 2730 2731 PetscCheck(a->i[ambs] == b->i[bmbs], PETSC_COMM_SELF, PETSC_ERR_ARG_INCOMP, "Number of nonzero blocks in matrices A %" PetscInt_FMT " and B %" PetscInt_FMT " are different", a->i[ambs], b->i[bmbs]); 2732 PetscCheck(abs == bbs, PETSC_COMM_SELF, PETSC_ERR_ARG_INCOMP, "Block size A %" PetscInt_FMT " and B %" PetscInt_FMT " are different", abs, bbs); 2733 PetscCall(PetscArraycpy(b->a, a->a, bs2 * a->i[ambs])); 2734 PetscCall(PetscObjectStateIncrease((PetscObject)B)); 2735 } else { 2736 PetscCall(MatCopy_Basic(A, B, str)); 2737 } 2738 PetscFunctionReturn(PETSC_SUCCESS); 2739 } 2740 2741 static PetscErrorCode MatSeqBAIJGetArray_SeqBAIJ(Mat A, PetscScalar *array[]) 2742 { 2743 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 2744 2745 PetscFunctionBegin; 2746 *array = a->a; 2747 PetscFunctionReturn(PETSC_SUCCESS); 2748 } 2749 2750 static PetscErrorCode MatSeqBAIJRestoreArray_SeqBAIJ(Mat A, PetscScalar *array[]) 2751 { 2752 PetscFunctionBegin; 2753 *array = NULL; 2754 PetscFunctionReturn(PETSC_SUCCESS); 2755 } 2756 2757 PetscErrorCode MatAXPYGetPreallocation_SeqBAIJ(Mat Y, Mat X, PetscInt *nnz) 2758 { 2759 PetscInt bs = Y->rmap->bs, mbs = Y->rmap->N / bs; 2760 Mat_SeqBAIJ *x = (Mat_SeqBAIJ *)X->data; 2761 Mat_SeqBAIJ *y = (Mat_SeqBAIJ *)Y->data; 2762 2763 PetscFunctionBegin; 2764 /* Set the number of nonzeros in the new matrix */ 2765 PetscCall(MatAXPYGetPreallocation_SeqX_private(mbs, x->i, x->j, y->i, y->j, nnz)); 2766 PetscFunctionReturn(PETSC_SUCCESS); 2767 } 2768 2769 PetscErrorCode MatAXPY_SeqBAIJ(Mat Y, PetscScalar a, Mat X, MatStructure str) 2770 { 2771 Mat_SeqBAIJ *x = (Mat_SeqBAIJ *)X->data, *y = (Mat_SeqBAIJ *)Y->data; 2772 PetscInt bs = Y->rmap->bs, bs2 = bs * bs; 2773 PetscBLASInt one = 1; 2774 2775 PetscFunctionBegin; 2776 if (str == UNKNOWN_NONZERO_PATTERN || (PetscDefined(USE_DEBUG) && str == SAME_NONZERO_PATTERN)) { 2777 PetscBool e = x->nz == y->nz && x->mbs == y->mbs && bs == X->rmap->bs ? PETSC_TRUE : PETSC_FALSE; 2778 if (e) { 2779 PetscCall(PetscArraycmp(x->i, y->i, x->mbs + 1, &e)); 2780 if (e) { 2781 PetscCall(PetscArraycmp(x->j, y->j, x->i[x->mbs], &e)); 2782 if (e) str = SAME_NONZERO_PATTERN; 2783 } 2784 } 2785 if (!e) PetscCheck(str != SAME_NONZERO_PATTERN, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "MatStructure is not SAME_NONZERO_PATTERN"); 2786 } 2787 if (str == SAME_NONZERO_PATTERN) { 2788 PetscScalar alpha = a; 2789 PetscBLASInt bnz; 2790 PetscCall(PetscBLASIntCast(x->nz * bs2, &bnz)); 2791 PetscCallBLAS("BLASaxpy", BLASaxpy_(&bnz, &alpha, x->a, &one, y->a, &one)); 2792 PetscCall(PetscObjectStateIncrease((PetscObject)Y)); 2793 } else if (str == SUBSET_NONZERO_PATTERN) { /* nonzeros of X is a subset of Y's */ 2794 PetscCall(MatAXPY_Basic(Y, a, X, str)); 2795 } else { 2796 Mat B; 2797 PetscInt *nnz; 2798 PetscCheck(bs == X->rmap->bs, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Matrices must have same block size"); 2799 PetscCall(PetscMalloc1(Y->rmap->N, &nnz)); 2800 PetscCall(MatCreate(PetscObjectComm((PetscObject)Y), &B)); 2801 PetscCall(PetscObjectSetName((PetscObject)B, ((PetscObject)Y)->name)); 2802 PetscCall(MatSetSizes(B, Y->rmap->n, Y->cmap->n, Y->rmap->N, Y->cmap->N)); 2803 PetscCall(MatSetBlockSizesFromMats(B, Y, Y)); 2804 PetscCall(MatSetType(B, (MatType)((PetscObject)Y)->type_name)); 2805 PetscCall(MatAXPYGetPreallocation_SeqBAIJ(Y, X, nnz)); 2806 PetscCall(MatSeqBAIJSetPreallocation(B, bs, 0, nnz)); 2807 PetscCall(MatAXPY_BasicWithPreallocation(B, Y, a, X, str)); 2808 PetscCall(MatHeaderMerge(Y, &B)); 2809 PetscCall(PetscFree(nnz)); 2810 } 2811 PetscFunctionReturn(PETSC_SUCCESS); 2812 } 2813 2814 PETSC_INTERN PetscErrorCode MatConjugate_SeqBAIJ(Mat A) 2815 { 2816 #if PetscDefined(USE_COMPLEX) 2817 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 2818 PetscInt i, nz = a->bs2 * a->i[a->mbs]; 2819 MatScalar *aa = a->a; 2820 2821 PetscFunctionBegin; 2822 for (i = 0; i < nz; i++) aa[i] = PetscConj(aa[i]); 2823 PetscFunctionReturn(PETSC_SUCCESS); 2824 #else 2825 (void)A; 2826 return PETSC_SUCCESS; 2827 #endif 2828 } 2829 2830 static PetscErrorCode MatRealPart_SeqBAIJ(Mat A) 2831 { 2832 #if PetscDefined(USE_COMPLEX) 2833 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 2834 PetscInt i, nz = a->bs2 * a->i[a->mbs]; 2835 MatScalar *aa = a->a; 2836 2837 PetscFunctionBegin; 2838 for (i = 0; i < nz; i++) aa[i] = PetscRealPart(aa[i]); 2839 PetscFunctionReturn(PETSC_SUCCESS); 2840 #else 2841 (void)A; 2842 return PETSC_SUCCESS; 2843 #endif 2844 } 2845 2846 static PetscErrorCode MatImaginaryPart_SeqBAIJ(Mat A) 2847 { 2848 #if PetscDefined(USE_COMPLEX) 2849 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 2850 PetscInt i, nz = a->bs2 * a->i[a->mbs]; 2851 MatScalar *aa = a->a; 2852 2853 PetscFunctionBegin; 2854 for (i = 0; i < nz; i++) aa[i] = PetscImaginaryPart(aa[i]); 2855 PetscFunctionReturn(PETSC_SUCCESS); 2856 #else 2857 (void)A; 2858 return PETSC_SUCCESS; 2859 #endif 2860 } 2861 2862 /* 2863 Code almost identical to MatGetColumnIJ_SeqAIJ() should share common code 2864 */ 2865 static PetscErrorCode MatGetColumnIJ_SeqBAIJ(Mat A, PetscInt oshift, PetscBool symmetric, PetscBool inodecompressed, PetscInt *nn, const PetscInt *ia[], const PetscInt *ja[], PetscBool *done) 2866 { 2867 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 2868 PetscInt bs = A->rmap->bs, i, *collengths, *cia, *cja, n = A->cmap->n / bs, m = A->rmap->n / bs; 2869 PetscInt nz = a->i[m], row, *jj, mr, col; 2870 2871 PetscFunctionBegin; 2872 *nn = n; 2873 if (!ia) PetscFunctionReturn(PETSC_SUCCESS); 2874 PetscCheck(!symmetric, PETSC_COMM_SELF, PETSC_ERR_SUP, "Not for BAIJ matrices"); 2875 PetscCall(PetscCalloc1(n, &collengths)); 2876 PetscCall(PetscMalloc1(n + 1, &cia)); 2877 PetscCall(PetscMalloc1(nz, &cja)); 2878 jj = a->j; 2879 for (i = 0; i < nz; i++) collengths[jj[i]]++; 2880 cia[0] = oshift; 2881 for (i = 0; i < n; i++) cia[i + 1] = cia[i] + collengths[i]; 2882 PetscCall(PetscArrayzero(collengths, n)); 2883 jj = a->j; 2884 for (row = 0; row < m; row++) { 2885 mr = a->i[row + 1] - a->i[row]; 2886 for (i = 0; i < mr; i++) { 2887 col = *jj++; 2888 2889 cja[cia[col] + collengths[col]++ - oshift] = row + oshift; 2890 } 2891 } 2892 PetscCall(PetscFree(collengths)); 2893 *ia = cia; 2894 *ja = cja; 2895 PetscFunctionReturn(PETSC_SUCCESS); 2896 } 2897 2898 static PetscErrorCode MatRestoreColumnIJ_SeqBAIJ(Mat A, PetscInt oshift, PetscBool symmetric, PetscBool inodecompressed, PetscInt *n, const PetscInt *ia[], const PetscInt *ja[], PetscBool *done) 2899 { 2900 PetscFunctionBegin; 2901 if (!ia) PetscFunctionReturn(PETSC_SUCCESS); 2902 PetscCall(PetscFree(*ia)); 2903 PetscCall(PetscFree(*ja)); 2904 PetscFunctionReturn(PETSC_SUCCESS); 2905 } 2906 2907 /* 2908 MatGetColumnIJ_SeqBAIJ_Color() and MatRestoreColumnIJ_SeqBAIJ_Color() are customized from 2909 MatGetColumnIJ_SeqBAIJ() and MatRestoreColumnIJ_SeqBAIJ() by adding an output 2910 spidx[], index of a->a, to be used in MatTransposeColoringCreate() and MatFDColoringCreate() 2911 */ 2912 PetscErrorCode MatGetColumnIJ_SeqBAIJ_Color(Mat A, PetscInt oshift, PetscBool symmetric, PetscBool inodecompressed, PetscInt *nn, const PetscInt *ia[], const PetscInt *ja[], PetscInt *spidx[], PetscBool *done) 2913 { 2914 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 2915 PetscInt i, *collengths, *cia, *cja, n = a->nbs, m = a->mbs; 2916 PetscInt nz = a->i[m], row, *jj, mr, col; 2917 PetscInt *cspidx; 2918 2919 PetscFunctionBegin; 2920 *nn = n; 2921 if (!ia) PetscFunctionReturn(PETSC_SUCCESS); 2922 2923 PetscCall(PetscCalloc1(n, &collengths)); 2924 PetscCall(PetscMalloc1(n + 1, &cia)); 2925 PetscCall(PetscMalloc1(nz, &cja)); 2926 PetscCall(PetscMalloc1(nz, &cspidx)); 2927 jj = a->j; 2928 for (i = 0; i < nz; i++) collengths[jj[i]]++; 2929 cia[0] = oshift; 2930 for (i = 0; i < n; i++) cia[i + 1] = cia[i] + collengths[i]; 2931 PetscCall(PetscArrayzero(collengths, n)); 2932 jj = a->j; 2933 for (row = 0; row < m; row++) { 2934 mr = a->i[row + 1] - a->i[row]; 2935 for (i = 0; i < mr; i++) { 2936 col = *jj++; 2937 cspidx[cia[col] + collengths[col] - oshift] = a->i[row] + i; /* index of a->j */ 2938 cja[cia[col] + collengths[col]++ - oshift] = row + oshift; 2939 } 2940 } 2941 PetscCall(PetscFree(collengths)); 2942 *ia = cia; 2943 *ja = cja; 2944 *spidx = cspidx; 2945 PetscFunctionReturn(PETSC_SUCCESS); 2946 } 2947 2948 PetscErrorCode MatRestoreColumnIJ_SeqBAIJ_Color(Mat A, PetscInt oshift, PetscBool symmetric, PetscBool inodecompressed, PetscInt *n, const PetscInt *ia[], const PetscInt *ja[], PetscInt *spidx[], PetscBool *done) 2949 { 2950 PetscFunctionBegin; 2951 PetscCall(MatRestoreColumnIJ_SeqBAIJ(A, oshift, symmetric, inodecompressed, n, ia, ja, done)); 2952 PetscCall(PetscFree(*spidx)); 2953 PetscFunctionReturn(PETSC_SUCCESS); 2954 } 2955 2956 static PetscErrorCode MatShift_SeqBAIJ(Mat Y, PetscScalar a) 2957 { 2958 Mat_SeqBAIJ *aij = (Mat_SeqBAIJ *)Y->data; 2959 2960 PetscFunctionBegin; 2961 if (!Y->preallocated || !aij->nz) PetscCall(MatSeqBAIJSetPreallocation(Y, Y->rmap->bs, 1, NULL)); 2962 PetscCall(MatShift_Basic(Y, a)); 2963 PetscFunctionReturn(PETSC_SUCCESS); 2964 } 2965 2966 PetscErrorCode MatEliminateZeros_SeqBAIJ(Mat A, PetscBool keep) 2967 { 2968 Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data; 2969 PetscInt fshift = 0, fshift_prev = 0, i, *ai = a->i, *aj = a->j, *imax = a->imax, j, k; 2970 PetscInt m = A->rmap->N, *ailen = a->ilen; 2971 PetscInt mbs = a->mbs, bs2 = a->bs2, rmax = 0; 2972 MatScalar *aa = a->a, *ap; 2973 PetscBool zero; 2974 2975 PetscFunctionBegin; 2976 PetscCheck(A->assembled, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Cannot eliminate zeros for unassembled matrix"); 2977 if (m) rmax = ailen[0]; 2978 for (i = 1; i <= mbs; i++) { 2979 for (k = ai[i - 1]; k < ai[i]; k++) { 2980 zero = PETSC_TRUE; 2981 ap = aa + bs2 * k; 2982 for (j = 0; j < bs2 && zero; j++) { 2983 if (ap[j] != 0.0) zero = PETSC_FALSE; 2984 } 2985 if (zero && (aj[k] != i - 1 || !keep)) fshift++; 2986 else { 2987 if (zero && aj[k] == i - 1) PetscCall(PetscInfo(A, "Keep the diagonal block at row %" PetscInt_FMT "\n", i - 1)); 2988 aj[k - fshift] = aj[k]; 2989 PetscCall(PetscArraymove(ap - bs2 * fshift, ap, bs2)); 2990 } 2991 } 2992 ai[i - 1] -= fshift_prev; 2993 fshift_prev = fshift; 2994 ailen[i - 1] = imax[i - 1] = ai[i] - fshift - ai[i - 1]; 2995 a->nonzerorowcnt += ((ai[i] - fshift - ai[i - 1]) > 0); 2996 rmax = PetscMax(rmax, ailen[i - 1]); 2997 } 2998 if (fshift) { 2999 if (mbs) { 3000 ai[mbs] -= fshift; 3001 a->nz = ai[mbs]; 3002 } 3003 PetscCall(PetscInfo(A, "Matrix size: %" PetscInt_FMT " X %" PetscInt_FMT "; zeros eliminated: %" PetscInt_FMT "; nonzeros left: %" PetscInt_FMT "\n", m, A->cmap->n, fshift, a->nz)); 3004 A->nonzerostate++; 3005 A->info.nz_unneeded += (PetscReal)fshift; 3006 a->rmax = rmax; 3007 PetscCall(MatAssemblyBegin(A, MAT_FINAL_ASSEMBLY)); 3008 PetscCall(MatAssemblyEnd(A, MAT_FINAL_ASSEMBLY)); 3009 } 3010 PetscFunctionReturn(PETSC_SUCCESS); 3011 } 3012 3013 static struct _MatOps MatOps_Values = {MatSetValues_SeqBAIJ, 3014 MatGetRow_SeqBAIJ, 3015 MatRestoreRow_SeqBAIJ, 3016 MatMult_SeqBAIJ_N, 3017 /* 4*/ MatMultAdd_SeqBAIJ_N, 3018 MatMultTranspose_SeqBAIJ, 3019 MatMultTransposeAdd_SeqBAIJ, 3020 NULL, 3021 NULL, 3022 NULL, 3023 /* 10*/ NULL, 3024 MatLUFactor_SeqBAIJ, 3025 NULL, 3026 NULL, 3027 MatTranspose_SeqBAIJ, 3028 /* 15*/ MatGetInfo_SeqBAIJ, 3029 MatEqual_SeqBAIJ, 3030 MatGetDiagonal_SeqBAIJ, 3031 MatDiagonalScale_SeqBAIJ, 3032 MatNorm_SeqBAIJ, 3033 /* 20*/ NULL, 3034 MatAssemblyEnd_SeqBAIJ, 3035 MatSetOption_SeqBAIJ, 3036 MatZeroEntries_SeqBAIJ, 3037 /* 24*/ MatZeroRows_SeqBAIJ, 3038 NULL, 3039 NULL, 3040 NULL, 3041 NULL, 3042 /* 29*/ MatSetUp_Seq_Hash, 3043 NULL, 3044 NULL, 3045 NULL, 3046 NULL, 3047 /* 34*/ MatDuplicate_SeqBAIJ, 3048 NULL, 3049 NULL, 3050 MatILUFactor_SeqBAIJ, 3051 NULL, 3052 /* 39*/ MatAXPY_SeqBAIJ, 3053 MatCreateSubMatrices_SeqBAIJ, 3054 MatIncreaseOverlap_SeqBAIJ, 3055 MatGetValues_SeqBAIJ, 3056 MatCopy_SeqBAIJ, 3057 /* 44*/ NULL, 3058 MatScale_SeqBAIJ, 3059 MatShift_SeqBAIJ, 3060 NULL, 3061 MatZeroRowsColumns_SeqBAIJ, 3062 /* 49*/ NULL, 3063 MatGetRowIJ_SeqBAIJ, 3064 MatRestoreRowIJ_SeqBAIJ, 3065 MatGetColumnIJ_SeqBAIJ, 3066 MatRestoreColumnIJ_SeqBAIJ, 3067 /* 54*/ MatFDColoringCreate_SeqXAIJ, 3068 NULL, 3069 NULL, 3070 NULL, 3071 MatSetValuesBlocked_SeqBAIJ, 3072 /* 59*/ MatCreateSubMatrix_SeqBAIJ, 3073 MatDestroy_SeqBAIJ, 3074 MatView_SeqBAIJ, 3075 NULL, 3076 NULL, 3077 /* 64*/ NULL, 3078 NULL, 3079 NULL, 3080 NULL, 3081 NULL, 3082 /* 69*/ MatGetRowMaxAbs_SeqBAIJ, 3083 NULL, 3084 MatConvert_Basic, 3085 NULL, 3086 NULL, 3087 /* 74*/ NULL, 3088 MatFDColoringApply_BAIJ, 3089 NULL, 3090 NULL, 3091 NULL, 3092 /* 79*/ NULL, 3093 NULL, 3094 NULL, 3095 NULL, 3096 MatLoad_SeqBAIJ, 3097 /* 84*/ NULL, 3098 NULL, 3099 NULL, 3100 NULL, 3101 NULL, 3102 /* 89*/ NULL, 3103 NULL, 3104 NULL, 3105 NULL, 3106 NULL, 3107 /* 94*/ NULL, 3108 NULL, 3109 NULL, 3110 NULL, 3111 NULL, 3112 /* 99*/ NULL, 3113 NULL, 3114 NULL, 3115 MatConjugate_SeqBAIJ, 3116 NULL, 3117 /*104*/ NULL, 3118 MatRealPart_SeqBAIJ, 3119 MatImaginaryPart_SeqBAIJ, 3120 NULL, 3121 NULL, 3122 /*109*/ NULL, 3123 NULL, 3124 NULL, 3125 NULL, 3126 MatMissingDiagonal_SeqBAIJ, 3127 /*114*/ NULL, 3128 NULL, 3129 NULL, 3130 NULL, 3131 NULL, 3132 /*119*/ NULL, 3133 NULL, 3134 MatMultHermitianTranspose_SeqBAIJ, 3135 MatMultHermitianTransposeAdd_SeqBAIJ, 3136 NULL, 3137 /*124*/ NULL, 3138 MatGetColumnReductions_SeqBAIJ, 3139 MatInvertBlockDiagonal_SeqBAIJ, 3140 NULL, 3141 NULL, 3142 /*129*/ NULL, 3143 NULL, 3144 NULL, 3145 NULL, 3146 NULL, 3147 /*134*/ NULL, 3148 NULL, 3149 NULL, 3150 NULL, 3151 NULL, 3152 /*139*/ MatSetBlockSizes_Default, 3153 NULL, 3154 NULL, 3155 MatFDColoringSetUp_SeqXAIJ, 3156 NULL, 3157 /*144*/ MatCreateMPIMatConcatenateSeqMat_SeqBAIJ, 3158 MatDestroySubMatrices_SeqBAIJ, 3159 NULL, 3160 NULL, 3161 NULL, 3162 NULL, 3163 /*150*/ NULL, 3164 MatEliminateZeros_SeqBAIJ, 3165 MatGetRowSumAbs_SeqBAIJ}; 3166 3167 static PetscErrorCode MatStoreValues_SeqBAIJ(Mat mat) 3168 { 3169 Mat_SeqBAIJ *aij = (Mat_SeqBAIJ *)mat->data; 3170 PetscInt nz = aij->i[aij->mbs] * aij->bs2; 3171 3172 PetscFunctionBegin; 3173 PetscCheck(aij->nonew == 1, PETSC_COMM_SELF, PETSC_ERR_ORDER, "Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first"); 3174 3175 /* allocate space for values if not already there */ 3176 if (!aij->saved_values) PetscCall(PetscMalloc1(nz + 1, &aij->saved_values)); 3177 3178 /* copy values over */ 3179 PetscCall(PetscArraycpy(aij->saved_values, aij->a, nz)); 3180 PetscFunctionReturn(PETSC_SUCCESS); 3181 } 3182 3183 static PetscErrorCode MatRetrieveValues_SeqBAIJ(Mat mat) 3184 { 3185 Mat_SeqBAIJ *aij = (Mat_SeqBAIJ *)mat->data; 3186 PetscInt nz = aij->i[aij->mbs] * aij->bs2; 3187 3188 PetscFunctionBegin; 3189 PetscCheck(aij->nonew == 1, PETSC_COMM_SELF, PETSC_ERR_ORDER, "Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first"); 3190 PetscCheck(aij->saved_values, PETSC_COMM_SELF, PETSC_ERR_ORDER, "Must call MatStoreValues(A);first"); 3191 3192 /* copy values over */ 3193 PetscCall(PetscArraycpy(aij->a, aij->saved_values, nz)); 3194 PetscFunctionReturn(PETSC_SUCCESS); 3195 } 3196 3197 PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqAIJ(Mat, MatType, MatReuse, Mat *); 3198 PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqSBAIJ(Mat, MatType, MatReuse, Mat *); 3199 3200 PetscErrorCode MatSeqBAIJSetPreallocation_SeqBAIJ(Mat B, PetscInt bs, PetscInt nz, const PetscInt nnz[]) 3201 { 3202 Mat_SeqBAIJ *b = (Mat_SeqBAIJ *)B->data; 3203 PetscInt i, mbs, nbs, bs2; 3204 PetscBool flg = PETSC_FALSE, skipallocation = PETSC_FALSE, realalloc = PETSC_FALSE; 3205 3206 PetscFunctionBegin; 3207 if (B->hash_active) { 3208 PetscInt bs; 3209 B->ops[0] = b->cops; 3210 PetscCall(PetscHMapIJVDestroy(&b->ht)); 3211 PetscCall(MatGetBlockSize(B, &bs)); 3212 if (bs > 1) PetscCall(PetscHSetIJDestroy(&b->bht)); 3213 PetscCall(PetscFree(b->dnz)); 3214 PetscCall(PetscFree(b->bdnz)); 3215 B->hash_active = PETSC_FALSE; 3216 } 3217 if (nz >= 0 || nnz) realalloc = PETSC_TRUE; 3218 if (nz == MAT_SKIP_ALLOCATION) { 3219 skipallocation = PETSC_TRUE; 3220 nz = 0; 3221 } 3222 3223 PetscCall(MatSetBlockSize(B, PetscAbs(bs))); 3224 PetscCall(PetscLayoutSetUp(B->rmap)); 3225 PetscCall(PetscLayoutSetUp(B->cmap)); 3226 PetscCall(PetscLayoutGetBlockSize(B->rmap, &bs)); 3227 3228 B->preallocated = PETSC_TRUE; 3229 3230 mbs = B->rmap->n / bs; 3231 nbs = B->cmap->n / bs; 3232 bs2 = bs * bs; 3233 3234 PetscCheck(mbs * bs == B->rmap->n && nbs * bs == B->cmap->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Number rows %" PetscInt_FMT ", cols %" PetscInt_FMT " must be divisible by blocksize %" PetscInt_FMT, B->rmap->N, B->cmap->n, bs); 3235 3236 if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5; 3237 PetscCheck(nz >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "nz cannot be less than 0: value %" PetscInt_FMT, nz); 3238 if (nnz) { 3239 for (i = 0; i < mbs; i++) { 3240 PetscCheck(nnz[i] >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "nnz cannot be less than 0: local row %" PetscInt_FMT " value %" PetscInt_FMT, i, nnz[i]); 3241 PetscCheck(nnz[i] <= nbs, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "nnz cannot be greater than block row length: local row %" PetscInt_FMT " value %" PetscInt_FMT " rowlength %" PetscInt_FMT, i, nnz[i], nbs); 3242 } 3243 } 3244 3245 PetscOptionsBegin(PetscObjectComm((PetscObject)B), NULL, "Optimize options for SEQBAIJ matrix 2 ", "Mat"); 3246 PetscCall(PetscOptionsBool("-mat_no_unroll", "Do not optimize for block size (slow)", NULL, flg, &flg, NULL)); 3247 PetscOptionsEnd(); 3248 3249 if (!flg) { 3250 switch (bs) { 3251 case 1: 3252 B->ops->mult = MatMult_SeqBAIJ_1; 3253 B->ops->multadd = MatMultAdd_SeqBAIJ_1; 3254 break; 3255 case 2: 3256 B->ops->mult = MatMult_SeqBAIJ_2; 3257 B->ops->multadd = MatMultAdd_SeqBAIJ_2; 3258 break; 3259 case 3: 3260 B->ops->mult = MatMult_SeqBAIJ_3; 3261 B->ops->multadd = MatMultAdd_SeqBAIJ_3; 3262 break; 3263 case 4: 3264 B->ops->mult = MatMult_SeqBAIJ_4; 3265 B->ops->multadd = MatMultAdd_SeqBAIJ_4; 3266 break; 3267 case 5: 3268 B->ops->mult = MatMult_SeqBAIJ_5; 3269 B->ops->multadd = MatMultAdd_SeqBAIJ_5; 3270 break; 3271 case 6: 3272 B->ops->mult = MatMult_SeqBAIJ_6; 3273 B->ops->multadd = MatMultAdd_SeqBAIJ_6; 3274 break; 3275 case 7: 3276 B->ops->mult = MatMult_SeqBAIJ_7; 3277 B->ops->multadd = MatMultAdd_SeqBAIJ_7; 3278 break; 3279 case 9: { 3280 PetscInt version = 1; 3281 PetscCall(PetscOptionsGetInt(NULL, ((PetscObject)B)->prefix, "-mat_baij_mult_version", &version, NULL)); 3282 switch (version) { 3283 #if defined(PETSC_HAVE_IMMINTRIN_H) && defined(__AVX2__) && defined(__FMA__) && defined(PETSC_USE_REAL_DOUBLE) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_64BIT_INDICES) 3284 case 1: 3285 B->ops->mult = MatMult_SeqBAIJ_9_AVX2; 3286 B->ops->multadd = MatMultAdd_SeqBAIJ_9_AVX2; 3287 PetscCall(PetscInfo((PetscObject)B, "Using AVX2 for MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", bs)); 3288 break; 3289 #endif 3290 default: 3291 B->ops->mult = MatMult_SeqBAIJ_N; 3292 B->ops->multadd = MatMultAdd_SeqBAIJ_N; 3293 PetscCall(PetscInfo((PetscObject)B, "Using BLAS for MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", bs)); 3294 break; 3295 } 3296 break; 3297 } 3298 case 11: 3299 B->ops->mult = MatMult_SeqBAIJ_11; 3300 B->ops->multadd = MatMultAdd_SeqBAIJ_11; 3301 break; 3302 case 12: { 3303 PetscInt version = 1; 3304 PetscCall(PetscOptionsGetInt(NULL, ((PetscObject)B)->prefix, "-mat_baij_mult_version", &version, NULL)); 3305 switch (version) { 3306 case 1: 3307 B->ops->mult = MatMult_SeqBAIJ_12_ver1; 3308 B->ops->multadd = MatMultAdd_SeqBAIJ_12_ver1; 3309 PetscCall(PetscInfo((PetscObject)B, "Using version %" PetscInt_FMT " of MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", version, bs)); 3310 break; 3311 case 2: 3312 B->ops->mult = MatMult_SeqBAIJ_12_ver2; 3313 B->ops->multadd = MatMultAdd_SeqBAIJ_12_ver2; 3314 PetscCall(PetscInfo((PetscObject)B, "Using version %" PetscInt_FMT " of MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", version, bs)); 3315 break; 3316 #if defined(PETSC_HAVE_IMMINTRIN_H) && defined(__AVX2__) && defined(__FMA__) && defined(PETSC_USE_REAL_DOUBLE) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_64BIT_INDICES) 3317 case 3: 3318 B->ops->mult = MatMult_SeqBAIJ_12_AVX2; 3319 B->ops->multadd = MatMultAdd_SeqBAIJ_12_ver1; 3320 PetscCall(PetscInfo((PetscObject)B, "Using AVX2 for MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", bs)); 3321 break; 3322 #endif 3323 default: 3324 B->ops->mult = MatMult_SeqBAIJ_N; 3325 B->ops->multadd = MatMultAdd_SeqBAIJ_N; 3326 PetscCall(PetscInfo((PetscObject)B, "Using BLAS for MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", bs)); 3327 break; 3328 } 3329 break; 3330 } 3331 case 15: { 3332 PetscInt version = 1; 3333 PetscCall(PetscOptionsGetInt(NULL, ((PetscObject)B)->prefix, "-mat_baij_mult_version", &version, NULL)); 3334 switch (version) { 3335 case 1: 3336 B->ops->mult = MatMult_SeqBAIJ_15_ver1; 3337 PetscCall(PetscInfo((PetscObject)B, "Using version %" PetscInt_FMT " of MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", version, bs)); 3338 break; 3339 case 2: 3340 B->ops->mult = MatMult_SeqBAIJ_15_ver2; 3341 PetscCall(PetscInfo((PetscObject)B, "Using version %" PetscInt_FMT " of MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", version, bs)); 3342 break; 3343 case 3: 3344 B->ops->mult = MatMult_SeqBAIJ_15_ver3; 3345 PetscCall(PetscInfo((PetscObject)B, "Using version %" PetscInt_FMT " of MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", version, bs)); 3346 break; 3347 case 4: 3348 B->ops->mult = MatMult_SeqBAIJ_15_ver4; 3349 PetscCall(PetscInfo((PetscObject)B, "Using version %" PetscInt_FMT " of MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", version, bs)); 3350 break; 3351 default: 3352 B->ops->mult = MatMult_SeqBAIJ_N; 3353 PetscCall(PetscInfo((PetscObject)B, "Using BLAS for MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", bs)); 3354 break; 3355 } 3356 B->ops->multadd = MatMultAdd_SeqBAIJ_N; 3357 break; 3358 } 3359 default: 3360 B->ops->mult = MatMult_SeqBAIJ_N; 3361 B->ops->multadd = MatMultAdd_SeqBAIJ_N; 3362 PetscCall(PetscInfo((PetscObject)B, "Using BLAS for MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", bs)); 3363 break; 3364 } 3365 } 3366 B->ops->sor = MatSOR_SeqBAIJ; 3367 b->mbs = mbs; 3368 b->nbs = nbs; 3369 if (!skipallocation) { 3370 if (!b->imax) { 3371 PetscCall(PetscMalloc2(mbs, &b->imax, mbs, &b->ilen)); 3372 3373 b->free_imax_ilen = PETSC_TRUE; 3374 } 3375 /* b->ilen will count nonzeros in each block row so far. */ 3376 for (i = 0; i < mbs; i++) b->ilen[i] = 0; 3377 if (!nnz) { 3378 if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5; 3379 else if (nz < 0) nz = 1; 3380 nz = PetscMin(nz, nbs); 3381 for (i = 0; i < mbs; i++) b->imax[i] = nz; 3382 PetscCall(PetscIntMultError(nz, mbs, &nz)); 3383 } else { 3384 PetscInt64 nz64 = 0; 3385 for (i = 0; i < mbs; i++) { 3386 b->imax[i] = nnz[i]; 3387 nz64 += nnz[i]; 3388 } 3389 PetscCall(PetscIntCast(nz64, &nz)); 3390 } 3391 3392 /* allocate the matrix space */ 3393 PetscCall(MatSeqXAIJFreeAIJ(B, &b->a, &b->j, &b->i)); 3394 if (B->structure_only) { 3395 PetscCall(PetscMalloc1(nz, &b->j)); 3396 PetscCall(PetscMalloc1(B->rmap->N + 1, &b->i)); 3397 } else { 3398 PetscInt nzbs2 = 0; 3399 PetscCall(PetscIntMultError(nz, bs2, &nzbs2)); 3400 PetscCall(PetscMalloc3(nzbs2, &b->a, nz, &b->j, B->rmap->N + 1, &b->i)); 3401 PetscCall(PetscArrayzero(b->a, nz * bs2)); 3402 } 3403 PetscCall(PetscArrayzero(b->j, nz)); 3404 3405 if (B->structure_only) { 3406 b->singlemalloc = PETSC_FALSE; 3407 b->free_a = PETSC_FALSE; 3408 } else { 3409 b->singlemalloc = PETSC_TRUE; 3410 b->free_a = PETSC_TRUE; 3411 } 3412 b->free_ij = PETSC_TRUE; 3413 3414 b->i[0] = 0; 3415 for (i = 1; i < mbs + 1; i++) b->i[i] = b->i[i - 1] + b->imax[i - 1]; 3416 3417 } else { 3418 b->free_a = PETSC_FALSE; 3419 b->free_ij = PETSC_FALSE; 3420 } 3421 3422 b->bs2 = bs2; 3423 b->mbs = mbs; 3424 b->nz = 0; 3425 b->maxnz = nz; 3426 B->info.nz_unneeded = (PetscReal)b->maxnz * bs2; 3427 B->was_assembled = PETSC_FALSE; 3428 B->assembled = PETSC_FALSE; 3429 if (realalloc) PetscCall(MatSetOption(B, MAT_NEW_NONZERO_ALLOCATION_ERR, PETSC_TRUE)); 3430 PetscFunctionReturn(PETSC_SUCCESS); 3431 } 3432 3433 static PetscErrorCode MatSeqBAIJSetPreallocationCSR_SeqBAIJ(Mat B, PetscInt bs, const PetscInt ii[], const PetscInt jj[], const PetscScalar V[]) 3434 { 3435 PetscInt i, m, nz, nz_max = 0, *nnz; 3436 PetscScalar *values = NULL; 3437 PetscBool roworiented = ((Mat_SeqBAIJ *)B->data)->roworiented; 3438 3439 PetscFunctionBegin; 3440 PetscCheck(bs >= 1, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Invalid block size specified, must be positive but it is %" PetscInt_FMT, bs); 3441 PetscCall(PetscLayoutSetBlockSize(B->rmap, bs)); 3442 PetscCall(PetscLayoutSetBlockSize(B->cmap, bs)); 3443 PetscCall(PetscLayoutSetUp(B->rmap)); 3444 PetscCall(PetscLayoutSetUp(B->cmap)); 3445 PetscCall(PetscLayoutGetBlockSize(B->rmap, &bs)); 3446 m = B->rmap->n / bs; 3447 3448 PetscCheck(ii[0] == 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "ii[0] must be 0 but it is %" PetscInt_FMT, ii[0]); 3449 PetscCall(PetscMalloc1(m + 1, &nnz)); 3450 for (i = 0; i < m; i++) { 3451 nz = ii[i + 1] - ii[i]; 3452 PetscCheck(nz >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Local row %" PetscInt_FMT " has a negative number of columns %" PetscInt_FMT, i, nz); 3453 nz_max = PetscMax(nz_max, nz); 3454 nnz[i] = nz; 3455 } 3456 PetscCall(MatSeqBAIJSetPreallocation(B, bs, 0, nnz)); 3457 PetscCall(PetscFree(nnz)); 3458 3459 values = (PetscScalar *)V; 3460 if (!values) PetscCall(PetscCalloc1(bs * bs * (nz_max + 1), &values)); 3461 for (i = 0; i < m; i++) { 3462 PetscInt ncols = ii[i + 1] - ii[i]; 3463 const PetscInt *icols = jj + ii[i]; 3464 if (bs == 1 || !roworiented) { 3465 const PetscScalar *svals = values + (V ? (bs * bs * ii[i]) : 0); 3466 PetscCall(MatSetValuesBlocked_SeqBAIJ(B, 1, &i, ncols, icols, svals, INSERT_VALUES)); 3467 } else { 3468 PetscInt j; 3469 for (j = 0; j < ncols; j++) { 3470 const PetscScalar *svals = values + (V ? (bs * bs * (ii[i] + j)) : 0); 3471 PetscCall(MatSetValuesBlocked_SeqBAIJ(B, 1, &i, 1, &icols[j], svals, INSERT_VALUES)); 3472 } 3473 } 3474 } 3475 if (!V) PetscCall(PetscFree(values)); 3476 PetscCall(MatAssemblyBegin(B, MAT_FINAL_ASSEMBLY)); 3477 PetscCall(MatAssemblyEnd(B, MAT_FINAL_ASSEMBLY)); 3478 PetscCall(MatSetOption(B, MAT_NEW_NONZERO_LOCATION_ERR, PETSC_TRUE)); 3479 PetscFunctionReturn(PETSC_SUCCESS); 3480 } 3481 3482 /*@C 3483 MatSeqBAIJGetArray - gives read/write access to the array where the data for a `MATSEQBAIJ` matrix is stored 3484 3485 Not Collective 3486 3487 Input Parameter: 3488 . A - a `MATSEQBAIJ` matrix 3489 3490 Output Parameter: 3491 . array - pointer to the data 3492 3493 Level: intermediate 3494 3495 .seealso: [](ch_matrices), `Mat`, `MATSEQBAIJ`, `MatSeqBAIJRestoreArray()`, `MatSeqAIJGetArray()`, `MatSeqAIJRestoreArray()` 3496 @*/ 3497 PetscErrorCode MatSeqBAIJGetArray(Mat A, PetscScalar **array) 3498 { 3499 PetscFunctionBegin; 3500 PetscUseMethod(A, "MatSeqBAIJGetArray_C", (Mat, PetscScalar **), (A, array)); 3501 PetscFunctionReturn(PETSC_SUCCESS); 3502 } 3503 3504 /*@C 3505 MatSeqBAIJRestoreArray - returns access to the array where the data for a `MATSEQBAIJ` matrix is stored obtained by `MatSeqBAIJGetArray()` 3506 3507 Not Collective 3508 3509 Input Parameters: 3510 + A - a `MATSEQBAIJ` matrix 3511 - array - pointer to the data 3512 3513 Level: intermediate 3514 3515 .seealso: [](ch_matrices), `Mat`, `MatSeqBAIJGetArray()`, `MatSeqAIJGetArray()`, `MatSeqAIJRestoreArray()` 3516 @*/ 3517 PetscErrorCode MatSeqBAIJRestoreArray(Mat A, PetscScalar **array) 3518 { 3519 PetscFunctionBegin; 3520 PetscUseMethod(A, "MatSeqBAIJRestoreArray_C", (Mat, PetscScalar **), (A, array)); 3521 PetscFunctionReturn(PETSC_SUCCESS); 3522 } 3523 3524 /*MC 3525 MATSEQBAIJ - MATSEQBAIJ = "seqbaij" - A matrix type to be used for sequential block sparse matrices, based on 3526 block sparse compressed row format. 3527 3528 Options Database Keys: 3529 + -mat_type seqbaij - sets the matrix type to `MATSEQBAIJ` during a call to `MatSetFromOptions()` 3530 - -mat_baij_mult_version version - indicate the version of the matrix-vector product to use (0 often indicates using BLAS) 3531 3532 Level: beginner 3533 3534 Notes: 3535 `MatSetOptions`(,`MAT_STRUCTURE_ONLY`,`PETSC_TRUE`) may be called for this matrix type. In this no 3536 space is allocated for the nonzero entries and any entries passed with `MatSetValues()` are ignored 3537 3538 Run with `-info` to see what version of the matrix-vector product is being used 3539 3540 .seealso: [](ch_matrices), `Mat`, `MatCreateSeqBAIJ()` 3541 M*/ 3542 3543 PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqBSTRM(Mat, MatType, MatReuse, Mat *); 3544 3545 PETSC_EXTERN PetscErrorCode MatCreate_SeqBAIJ(Mat B) 3546 { 3547 PetscMPIInt size; 3548 Mat_SeqBAIJ *b; 3549 3550 PetscFunctionBegin; 3551 PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)B), &size)); 3552 PetscCheck(size == 1, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "Comm must be of size 1"); 3553 3554 PetscCall(PetscNew(&b)); 3555 B->data = (void *)b; 3556 B->ops[0] = MatOps_Values; 3557 3558 b->row = NULL; 3559 b->col = NULL; 3560 b->icol = NULL; 3561 b->reallocs = 0; 3562 b->saved_values = NULL; 3563 3564 b->roworiented = PETSC_TRUE; 3565 b->nonew = 0; 3566 b->diag = NULL; 3567 B->spptr = NULL; 3568 B->info.nz_unneeded = (PetscReal)b->maxnz * b->bs2; 3569 b->keepnonzeropattern = PETSC_FALSE; 3570 3571 PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatSeqBAIJGetArray_C", MatSeqBAIJGetArray_SeqBAIJ)); 3572 PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatSeqBAIJRestoreArray_C", MatSeqBAIJRestoreArray_SeqBAIJ)); 3573 PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatStoreValues_C", MatStoreValues_SeqBAIJ)); 3574 PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatRetrieveValues_C", MatRetrieveValues_SeqBAIJ)); 3575 PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatSeqBAIJSetColumnIndices_C", MatSeqBAIJSetColumnIndices_SeqBAIJ)); 3576 PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatConvert_seqbaij_seqaij_C", MatConvert_SeqBAIJ_SeqAIJ)); 3577 PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatConvert_seqbaij_seqsbaij_C", MatConvert_SeqBAIJ_SeqSBAIJ)); 3578 PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatSeqBAIJSetPreallocation_C", MatSeqBAIJSetPreallocation_SeqBAIJ)); 3579 PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatSeqBAIJSetPreallocationCSR_C", MatSeqBAIJSetPreallocationCSR_SeqBAIJ)); 3580 PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatIsTranspose_C", MatIsTranspose_SeqBAIJ)); 3581 #if defined(PETSC_HAVE_HYPRE) 3582 PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatConvert_seqbaij_hypre_C", MatConvert_AIJ_HYPRE)); 3583 #endif 3584 PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatConvert_seqbaij_is_C", MatConvert_XAIJ_IS)); 3585 PetscCall(PetscObjectChangeTypeName((PetscObject)B, MATSEQBAIJ)); 3586 PetscFunctionReturn(PETSC_SUCCESS); 3587 } 3588 3589 PetscErrorCode MatDuplicateNoCreate_SeqBAIJ(Mat C, Mat A, MatDuplicateOption cpvalues, PetscBool mallocmatspace) 3590 { 3591 Mat_SeqBAIJ *c = (Mat_SeqBAIJ *)C->data, *a = (Mat_SeqBAIJ *)A->data; 3592 PetscInt i, mbs = a->mbs, nz = a->nz, bs2 = a->bs2; 3593 3594 PetscFunctionBegin; 3595 PetscCheck(A->assembled, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Cannot duplicate unassembled matrix"); 3596 PetscCheck(a->i[mbs] == nz, PETSC_COMM_SELF, PETSC_ERR_PLIB, "Corrupt matrix"); 3597 3598 if (cpvalues == MAT_SHARE_NONZERO_PATTERN) { 3599 c->imax = a->imax; 3600 c->ilen = a->ilen; 3601 c->free_imax_ilen = PETSC_FALSE; 3602 } else { 3603 PetscCall(PetscMalloc2(mbs, &c->imax, mbs, &c->ilen)); 3604 for (i = 0; i < mbs; i++) { 3605 c->imax[i] = a->imax[i]; 3606 c->ilen[i] = a->ilen[i]; 3607 } 3608 c->free_imax_ilen = PETSC_TRUE; 3609 } 3610 3611 /* allocate the matrix space */ 3612 if (mallocmatspace) { 3613 if (cpvalues == MAT_SHARE_NONZERO_PATTERN) { 3614 PetscCall(PetscCalloc1(bs2 * nz, &c->a)); 3615 3616 c->i = a->i; 3617 c->j = a->j; 3618 c->singlemalloc = PETSC_FALSE; 3619 c->free_a = PETSC_TRUE; 3620 c->free_ij = PETSC_FALSE; 3621 c->parent = A; 3622 C->preallocated = PETSC_TRUE; 3623 C->assembled = PETSC_TRUE; 3624 3625 PetscCall(PetscObjectReference((PetscObject)A)); 3626 PetscCall(MatSetOption(A, MAT_NEW_NONZERO_LOCATION_ERR, PETSC_TRUE)); 3627 PetscCall(MatSetOption(C, MAT_NEW_NONZERO_LOCATION_ERR, PETSC_TRUE)); 3628 } else { 3629 PetscCall(PetscMalloc3(bs2 * nz, &c->a, nz, &c->j, mbs + 1, &c->i)); 3630 3631 c->singlemalloc = PETSC_TRUE; 3632 c->free_a = PETSC_TRUE; 3633 c->free_ij = PETSC_TRUE; 3634 3635 PetscCall(PetscArraycpy(c->i, a->i, mbs + 1)); 3636 if (mbs > 0) { 3637 PetscCall(PetscArraycpy(c->j, a->j, nz)); 3638 if (cpvalues == MAT_COPY_VALUES) { 3639 PetscCall(PetscArraycpy(c->a, a->a, bs2 * nz)); 3640 } else { 3641 PetscCall(PetscArrayzero(c->a, bs2 * nz)); 3642 } 3643 } 3644 C->preallocated = PETSC_TRUE; 3645 C->assembled = PETSC_TRUE; 3646 } 3647 } 3648 3649 c->roworiented = a->roworiented; 3650 c->nonew = a->nonew; 3651 3652 PetscCall(PetscLayoutReference(A->rmap, &C->rmap)); 3653 PetscCall(PetscLayoutReference(A->cmap, &C->cmap)); 3654 3655 c->bs2 = a->bs2; 3656 c->mbs = a->mbs; 3657 c->nbs = a->nbs; 3658 3659 if (a->diag) { 3660 if (cpvalues == MAT_SHARE_NONZERO_PATTERN) { 3661 c->diag = a->diag; 3662 c->free_diag = PETSC_FALSE; 3663 } else { 3664 PetscCall(PetscMalloc1(mbs + 1, &c->diag)); 3665 for (i = 0; i < mbs; i++) c->diag[i] = a->diag[i]; 3666 c->free_diag = PETSC_TRUE; 3667 } 3668 } else c->diag = NULL; 3669 3670 c->nz = a->nz; 3671 c->maxnz = a->nz; /* Since we allocate exactly the right amount */ 3672 c->solve_work = NULL; 3673 c->mult_work = NULL; 3674 c->sor_workt = NULL; 3675 c->sor_work = NULL; 3676 3677 c->compressedrow.use = a->compressedrow.use; 3678 c->compressedrow.nrows = a->compressedrow.nrows; 3679 if (a->compressedrow.use) { 3680 i = a->compressedrow.nrows; 3681 PetscCall(PetscMalloc2(i + 1, &c->compressedrow.i, i + 1, &c->compressedrow.rindex)); 3682 PetscCall(PetscArraycpy(c->compressedrow.i, a->compressedrow.i, i + 1)); 3683 PetscCall(PetscArraycpy(c->compressedrow.rindex, a->compressedrow.rindex, i)); 3684 } else { 3685 c->compressedrow.use = PETSC_FALSE; 3686 c->compressedrow.i = NULL; 3687 c->compressedrow.rindex = NULL; 3688 } 3689 c->nonzerorowcnt = a->nonzerorowcnt; 3690 C->nonzerostate = A->nonzerostate; 3691 3692 PetscCall(PetscFunctionListDuplicate(((PetscObject)A)->qlist, &((PetscObject)C)->qlist)); 3693 PetscFunctionReturn(PETSC_SUCCESS); 3694 } 3695 3696 PetscErrorCode MatDuplicate_SeqBAIJ(Mat A, MatDuplicateOption cpvalues, Mat *B) 3697 { 3698 PetscFunctionBegin; 3699 PetscCall(MatCreate(PetscObjectComm((PetscObject)A), B)); 3700 PetscCall(MatSetSizes(*B, A->rmap->N, A->cmap->n, A->rmap->N, A->cmap->n)); 3701 PetscCall(MatSetType(*B, MATSEQBAIJ)); 3702 PetscCall(MatDuplicateNoCreate_SeqBAIJ(*B, A, cpvalues, PETSC_TRUE)); 3703 PetscFunctionReturn(PETSC_SUCCESS); 3704 } 3705 3706 /* Used for both SeqBAIJ and SeqSBAIJ matrices */ 3707 PetscErrorCode MatLoad_SeqBAIJ_Binary(Mat mat, PetscViewer viewer) 3708 { 3709 PetscInt header[4], M, N, nz, bs, m, n, mbs, nbs, rows, cols, sum, i, j, k; 3710 PetscInt *rowidxs, *colidxs; 3711 PetscScalar *matvals; 3712 3713 PetscFunctionBegin; 3714 PetscCall(PetscViewerSetUp(viewer)); 3715 3716 /* read matrix header */ 3717 PetscCall(PetscViewerBinaryRead(viewer, header, 4, NULL, PETSC_INT)); 3718 PetscCheck(header[0] == MAT_FILE_CLASSID, PETSC_COMM_SELF, PETSC_ERR_FILE_UNEXPECTED, "Not a matrix object in file"); 3719 M = header[1]; 3720 N = header[2]; 3721 nz = header[3]; 3722 PetscCheck(M >= 0, PetscObjectComm((PetscObject)viewer), PETSC_ERR_FILE_UNEXPECTED, "Matrix row size (%" PetscInt_FMT ") in file is negative", M); 3723 PetscCheck(N >= 0, PetscObjectComm((PetscObject)viewer), PETSC_ERR_FILE_UNEXPECTED, "Matrix column size (%" PetscInt_FMT ") in file is negative", N); 3724 PetscCheck(nz >= 0, PETSC_COMM_SELF, PETSC_ERR_FILE_UNEXPECTED, "Matrix stored in special format on disk, cannot load as SeqBAIJ"); 3725 3726 /* set block sizes from the viewer's .info file */ 3727 PetscCall(MatLoad_Binary_BlockSizes(mat, viewer)); 3728 /* set local and global sizes if not set already */ 3729 if (mat->rmap->n < 0) mat->rmap->n = M; 3730 if (mat->cmap->n < 0) mat->cmap->n = N; 3731 if (mat->rmap->N < 0) mat->rmap->N = M; 3732 if (mat->cmap->N < 0) mat->cmap->N = N; 3733 PetscCall(PetscLayoutSetUp(mat->rmap)); 3734 PetscCall(PetscLayoutSetUp(mat->cmap)); 3735 3736 /* check if the matrix sizes are correct */ 3737 PetscCall(MatGetSize(mat, &rows, &cols)); 3738 PetscCheck(M == rows && N == cols, PETSC_COMM_SELF, PETSC_ERR_FILE_UNEXPECTED, "Matrix in file of different sizes (%" PetscInt_FMT ", %" PetscInt_FMT ") than the input matrix (%" PetscInt_FMT ", %" PetscInt_FMT ")", M, N, rows, cols); 3739 PetscCall(MatGetBlockSize(mat, &bs)); 3740 PetscCall(MatGetLocalSize(mat, &m, &n)); 3741 mbs = m / bs; 3742 nbs = n / bs; 3743 3744 /* read in row lengths, column indices and nonzero values */ 3745 PetscCall(PetscMalloc1(m + 1, &rowidxs)); 3746 PetscCall(PetscViewerBinaryRead(viewer, rowidxs + 1, m, NULL, PETSC_INT)); 3747 rowidxs[0] = 0; 3748 for (i = 0; i < m; i++) rowidxs[i + 1] += rowidxs[i]; 3749 sum = rowidxs[m]; 3750 PetscCheck(sum == nz, PETSC_COMM_SELF, PETSC_ERR_FILE_UNEXPECTED, "Inconsistent matrix data in file: nonzeros = %" PetscInt_FMT ", sum-row-lengths = %" PetscInt_FMT, nz, sum); 3751 3752 /* read in column indices and nonzero values */ 3753 PetscCall(PetscMalloc2(rowidxs[m], &colidxs, nz, &matvals)); 3754 PetscCall(PetscViewerBinaryRead(viewer, colidxs, rowidxs[m], NULL, PETSC_INT)); 3755 PetscCall(PetscViewerBinaryRead(viewer, matvals, rowidxs[m], NULL, PETSC_SCALAR)); 3756 3757 { /* preallocate matrix storage */ 3758 PetscBT bt; /* helper bit set to count nonzeros */ 3759 PetscInt *nnz; 3760 PetscBool sbaij; 3761 3762 PetscCall(PetscBTCreate(nbs, &bt)); 3763 PetscCall(PetscCalloc1(mbs, &nnz)); 3764 PetscCall(PetscObjectTypeCompare((PetscObject)mat, MATSEQSBAIJ, &sbaij)); 3765 for (i = 0; i < mbs; i++) { 3766 PetscCall(PetscBTMemzero(nbs, bt)); 3767 for (k = 0; k < bs; k++) { 3768 PetscInt row = bs * i + k; 3769 for (j = rowidxs[row]; j < rowidxs[row + 1]; j++) { 3770 PetscInt col = colidxs[j]; 3771 if (!sbaij || col >= row) 3772 if (!PetscBTLookupSet(bt, col / bs)) nnz[i]++; 3773 } 3774 } 3775 } 3776 PetscCall(PetscBTDestroy(&bt)); 3777 PetscCall(MatSeqBAIJSetPreallocation(mat, bs, 0, nnz)); 3778 PetscCall(MatSeqSBAIJSetPreallocation(mat, bs, 0, nnz)); 3779 PetscCall(PetscFree(nnz)); 3780 } 3781 3782 /* store matrix values */ 3783 for (i = 0; i < m; i++) { 3784 PetscInt row = i, s = rowidxs[i], e = rowidxs[i + 1]; 3785 PetscUseTypeMethod(mat, setvalues, 1, &row, e - s, colidxs + s, matvals + s, INSERT_VALUES); 3786 } 3787 3788 PetscCall(PetscFree(rowidxs)); 3789 PetscCall(PetscFree2(colidxs, matvals)); 3790 PetscCall(MatAssemblyBegin(mat, MAT_FINAL_ASSEMBLY)); 3791 PetscCall(MatAssemblyEnd(mat, MAT_FINAL_ASSEMBLY)); 3792 PetscFunctionReturn(PETSC_SUCCESS); 3793 } 3794 3795 PetscErrorCode MatLoad_SeqBAIJ(Mat mat, PetscViewer viewer) 3796 { 3797 PetscBool isbinary; 3798 3799 PetscFunctionBegin; 3800 PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERBINARY, &isbinary)); 3801 PetscCheck(isbinary, PetscObjectComm((PetscObject)viewer), PETSC_ERR_SUP, "Viewer type %s not yet supported for reading %s matrices", ((PetscObject)viewer)->type_name, ((PetscObject)mat)->type_name); 3802 PetscCall(MatLoad_SeqBAIJ_Binary(mat, viewer)); 3803 PetscFunctionReturn(PETSC_SUCCESS); 3804 } 3805 3806 /*@C 3807 MatCreateSeqBAIJ - Creates a sparse matrix in `MATSEQAIJ` (block 3808 compressed row) format. For good matrix assembly performance the 3809 user should preallocate the matrix storage by setting the parameter `nz` 3810 (or the array `nnz`). 3811 3812 Collective 3813 3814 Input Parameters: 3815 + comm - MPI communicator, set to `PETSC_COMM_SELF` 3816 . bs - size of block, the blocks are ALWAYS square. One can use `MatSetBlockSizes()` to set a different row and column blocksize but the row 3817 blocksize always defines the size of the blocks. The column blocksize sets the blocksize of the vectors obtained with `MatCreateVecs()` 3818 . m - number of rows 3819 . n - number of columns 3820 . nz - number of nonzero blocks per block row (same for all rows) 3821 - nnz - array containing the number of nonzero blocks in the various block rows 3822 (possibly different for each block row) or `NULL` 3823 3824 Output Parameter: 3825 . A - the matrix 3826 3827 Options Database Keys: 3828 + -mat_no_unroll - uses code that does not unroll the loops in the block calculations (much slower) 3829 - -mat_block_size - size of the blocks to use 3830 3831 Level: intermediate 3832 3833 Notes: 3834 It is recommended that one use `MatCreateFromOptions()` or the `MatCreate()`, `MatSetType()` and/or `MatSetFromOptions()`, 3835 MatXXXXSetPreallocation() paradigm instead of this routine directly. 3836 [MatXXXXSetPreallocation() is, for example, `MatSeqAIJSetPreallocation()`] 3837 3838 The number of rows and columns must be divisible by blocksize. 3839 3840 If the `nnz` parameter is given then the `nz` parameter is ignored 3841 3842 A nonzero block is any block that as 1 or more nonzeros in it 3843 3844 The `MATSEQBAIJ` format is fully compatible with standard Fortran 3845 storage. That is, the stored row and column indices can begin at 3846 either one (as in Fortran) or zero. 3847 3848 Specify the preallocated storage with either `nz` or `nnz` (not both). 3849 Set `nz` = `PETSC_DEFAULT` and `nnz` = `NULL` for PETSc to control dynamic memory 3850 allocation. See [Sparse Matrices](sec_matsparse) for details. 3851 matrices. 3852 3853 .seealso: [](ch_matrices), `Mat`, [Sparse Matrices](sec_matsparse), `MatCreate()`, `MatCreateSeqAIJ()`, `MatSetValues()`, `MatCreateBAIJ()` 3854 @*/ 3855 PetscErrorCode MatCreateSeqBAIJ(MPI_Comm comm, PetscInt bs, PetscInt m, PetscInt n, PetscInt nz, const PetscInt nnz[], Mat *A) 3856 { 3857 PetscFunctionBegin; 3858 PetscCall(MatCreate(comm, A)); 3859 PetscCall(MatSetSizes(*A, m, n, m, n)); 3860 PetscCall(MatSetType(*A, MATSEQBAIJ)); 3861 PetscCall(MatSeqBAIJSetPreallocation(*A, bs, nz, (PetscInt *)nnz)); 3862 PetscFunctionReturn(PETSC_SUCCESS); 3863 } 3864 3865 /*@C 3866 MatSeqBAIJSetPreallocation - Sets the block size and expected nonzeros 3867 per row in the matrix. For good matrix assembly performance the 3868 user should preallocate the matrix storage by setting the parameter `nz` 3869 (or the array `nnz`). 3870 3871 Collective 3872 3873 Input Parameters: 3874 + B - the matrix 3875 . bs - size of block, the blocks are ALWAYS square. One can use `MatSetBlockSizes()` to set a different row and column blocksize but the row 3876 blocksize always defines the size of the blocks. The column blocksize sets the blocksize of the vectors obtained with `MatCreateVecs()` 3877 . nz - number of block nonzeros per block row (same for all rows) 3878 - nnz - array containing the number of block nonzeros in the various block rows 3879 (possibly different for each block row) or `NULL` 3880 3881 Options Database Keys: 3882 + -mat_no_unroll - uses code that does not unroll the loops in the block calculations (much slower) 3883 - -mat_block_size - size of the blocks to use 3884 3885 Level: intermediate 3886 3887 Notes: 3888 If the `nnz` parameter is given then the `nz` parameter is ignored 3889 3890 You can call `MatGetInfo()` to get information on how effective the preallocation was; 3891 for example the fields mallocs,nz_allocated,nz_used,nz_unneeded; 3892 You can also run with the option `-info` and look for messages with the string 3893 malloc in them to see if additional memory allocation was needed. 3894 3895 The `MATSEQBAIJ` format is fully compatible with standard Fortran 3896 storage. That is, the stored row and column indices can begin at 3897 either one (as in Fortran) or zero. 3898 3899 Specify the preallocated storage with either `nz` or `nnz` (not both). 3900 Set `nz` = `PETSC_DEFAULT` and `nnz` = `NULL` for PETSc to control dynamic memory 3901 allocation. See [Sparse Matrices](sec_matsparse) for details. 3902 3903 .seealso: [](ch_matrices), `Mat`, [Sparse Matrices](sec_matsparse), `MatCreate()`, `MatCreateSeqAIJ()`, `MatSetValues()`, `MatCreateBAIJ()`, `MatGetInfo()` 3904 @*/ 3905 PetscErrorCode MatSeqBAIJSetPreallocation(Mat B, PetscInt bs, PetscInt nz, const PetscInt nnz[]) 3906 { 3907 PetscFunctionBegin; 3908 PetscValidHeaderSpecific(B, MAT_CLASSID, 1); 3909 PetscValidType(B, 1); 3910 PetscValidLogicalCollectiveInt(B, bs, 2); 3911 PetscTryMethod(B, "MatSeqBAIJSetPreallocation_C", (Mat, PetscInt, PetscInt, const PetscInt[]), (B, bs, nz, nnz)); 3912 PetscFunctionReturn(PETSC_SUCCESS); 3913 } 3914 3915 /*@C 3916 MatSeqBAIJSetPreallocationCSR - Creates a sparse sequential matrix in `MATSEQBAIJ` format using the given nonzero structure and (optional) numerical values 3917 3918 Collective 3919 3920 Input Parameters: 3921 + B - the matrix 3922 . bs - the blocksize 3923 . i - the indices into `j` for the start of each local row (indices start with zero) 3924 . j - the column indices for each local row (indices start with zero) these must be sorted for each row 3925 - v - optional values in the matrix, use `NULL` if not provided 3926 3927 Level: advanced 3928 3929 Notes: 3930 The `i`,`j`,`v` values are COPIED with this routine; to avoid the copy use `MatCreateSeqBAIJWithArrays()` 3931 3932 The order of the entries in values is specified by the `MatOption` `MAT_ROW_ORIENTED`. For example, C programs 3933 may want to use the default `MAT_ROW_ORIENTED` of `PETSC_TRUE` and use an array v[nnz][bs][bs] where the second index is 3934 over rows within a block and the last index is over columns within a block row. Fortran programs will likely set 3935 `MAT_ROW_ORIENTED` of `PETSC_FALSE` and use a Fortran array v(bs,bs,nnz) in which the first index is over rows within a 3936 block column and the second index is over columns within a block. 3937 3938 Though this routine has Preallocation() in the name it also sets the exact nonzero locations of the matrix entries and usually the numerical values as well 3939 3940 .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `MatCreateSeqBAIJ()`, `MatSetValues()`, `MatSeqBAIJSetPreallocation()`, `MATSEQBAIJ` 3941 @*/ 3942 PetscErrorCode MatSeqBAIJSetPreallocationCSR(Mat B, PetscInt bs, const PetscInt i[], const PetscInt j[], const PetscScalar v[]) 3943 { 3944 PetscFunctionBegin; 3945 PetscValidHeaderSpecific(B, MAT_CLASSID, 1); 3946 PetscValidType(B, 1); 3947 PetscValidLogicalCollectiveInt(B, bs, 2); 3948 PetscTryMethod(B, "MatSeqBAIJSetPreallocationCSR_C", (Mat, PetscInt, const PetscInt[], const PetscInt[], const PetscScalar[]), (B, bs, i, j, v)); 3949 PetscFunctionReturn(PETSC_SUCCESS); 3950 } 3951 3952 /*@ 3953 MatCreateSeqBAIJWithArrays - Creates a `MATSEQBAIJ` matrix using matrix elements provided by the user. 3954 3955 Collective 3956 3957 Input Parameters: 3958 + comm - must be an MPI communicator of size 1 3959 . bs - size of block 3960 . m - number of rows 3961 . n - number of columns 3962 . i - row indices; that is i[0] = 0, i[row] = i[row-1] + number of elements in that row block row of the matrix 3963 . j - column indices 3964 - a - matrix values 3965 3966 Output Parameter: 3967 . mat - the matrix 3968 3969 Level: advanced 3970 3971 Notes: 3972 The `i`, `j`, and `a` arrays are not copied by this routine, the user must free these arrays 3973 once the matrix is destroyed 3974 3975 You cannot set new nonzero locations into this matrix, that will generate an error. 3976 3977 The `i` and `j` indices are 0 based 3978 3979 When block size is greater than 1 the matrix values must be stored using the `MATSEQBAIJ` storage format 3980 3981 The order of the entries in values is the same as the block compressed sparse row storage format; that is, it is 3982 the same as a three dimensional array in Fortran values(bs,bs,nnz) that contains the first column of the first 3983 block, followed by the second column of the first block etc etc. That is, the blocks are contiguous in memory 3984 with column-major ordering within blocks. 3985 3986 .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `MatCreateBAIJ()`, `MatCreateSeqBAIJ()` 3987 @*/ 3988 PetscErrorCode MatCreateSeqBAIJWithArrays(MPI_Comm comm, PetscInt bs, PetscInt m, PetscInt n, PetscInt i[], PetscInt j[], PetscScalar a[], Mat *mat) 3989 { 3990 Mat_SeqBAIJ *baij; 3991 3992 PetscFunctionBegin; 3993 PetscCheck(bs == 1, PETSC_COMM_SELF, PETSC_ERR_SUP, "block size %" PetscInt_FMT " > 1 is not supported yet", bs); 3994 if (m > 0) PetscCheck(i[0] == 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "i (row indices) must start with 0"); 3995 3996 PetscCall(MatCreate(comm, mat)); 3997 PetscCall(MatSetSizes(*mat, m, n, m, n)); 3998 PetscCall(MatSetType(*mat, MATSEQBAIJ)); 3999 PetscCall(MatSeqBAIJSetPreallocation(*mat, bs, MAT_SKIP_ALLOCATION, NULL)); 4000 baij = (Mat_SeqBAIJ *)(*mat)->data; 4001 PetscCall(PetscMalloc2(m, &baij->imax, m, &baij->ilen)); 4002 4003 baij->i = i; 4004 baij->j = j; 4005 baij->a = a; 4006 4007 baij->singlemalloc = PETSC_FALSE; 4008 baij->nonew = -1; /*this indicates that inserting a new value in the matrix that generates a new nonzero is an error*/ 4009 baij->free_a = PETSC_FALSE; 4010 baij->free_ij = PETSC_FALSE; 4011 baij->free_imax_ilen = PETSC_TRUE; 4012 4013 for (PetscInt ii = 0; ii < m; ii++) { 4014 const PetscInt row_len = i[ii + 1] - i[ii]; 4015 4016 baij->ilen[ii] = baij->imax[ii] = row_len; 4017 PetscCheck(row_len >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Negative row length in i (row indices) row = %" PetscInt_FMT " length = %" PetscInt_FMT, ii, row_len); 4018 } 4019 if (PetscDefined(USE_DEBUG)) { 4020 for (PetscInt ii = 0; ii < baij->i[m]; ii++) { 4021 PetscCheck(j[ii] >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Negative column index at location = %" PetscInt_FMT " index = %" PetscInt_FMT, ii, j[ii]); 4022 PetscCheck(j[ii] <= n - 1, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Column index to large at location = %" PetscInt_FMT " index = %" PetscInt_FMT, ii, j[ii]); 4023 } 4024 } 4025 4026 PetscCall(MatAssemblyBegin(*mat, MAT_FINAL_ASSEMBLY)); 4027 PetscCall(MatAssemblyEnd(*mat, MAT_FINAL_ASSEMBLY)); 4028 PetscFunctionReturn(PETSC_SUCCESS); 4029 } 4030 4031 PetscErrorCode MatCreateMPIMatConcatenateSeqMat_SeqBAIJ(MPI_Comm comm, Mat inmat, PetscInt n, MatReuse scall, Mat *outmat) 4032 { 4033 PetscFunctionBegin; 4034 PetscCall(MatCreateMPIMatConcatenateSeqMat_MPIBAIJ(comm, inmat, n, scall, outmat)); 4035 PetscFunctionReturn(PETSC_SUCCESS); 4036 } 4037