Actual source code: ex52.c
2: static char help[] = "Solves a linear system in parallel with KSP. Modified from ex2.c \n\
3: Illustrate how to use external packages MUMPS, SUPERLU and STRUMPACK \n\
4: Input parameters include:\n\
5: -random_exact_sol : use a random exact solution vector\n\
6: -view_exact_sol : write exact solution vector to stdout\n\
7: -m <mesh_x> : number of mesh points in x-direction\n\
8: -n <mesh_y> : number of mesh points in y-direction\n\n";
10: #include <petscksp.h>
12: #if defined(PETSC_HAVE_MUMPS)
13: /* Subroutine contributed by Varun Hiremath */
14: PetscErrorCode printMumpsMemoryInfo(Mat F)
15: {
16: PetscInt maxMem, sumMem;
18: PetscFunctionBeginUser;
19: PetscCall(MatMumpsGetInfog(F, 16, &maxMem));
20: PetscCall(MatMumpsGetInfog(F, 17, &sumMem));
21: PetscCall(PetscPrintf(PETSC_COMM_WORLD, "\n MUMPS INFOG(16) :: Max memory in MB = %" PetscInt_FMT, maxMem));
22: PetscCall(PetscPrintf(PETSC_COMM_WORLD, "\n MUMPS INFOG(17) :: Sum memory in MB = %" PetscInt_FMT "\n", sumMem));
23: PetscFunctionReturn(PETSC_SUCCESS);
24: }
25: #endif
27: int main(int argc, char **args)
28: {
29: Vec x, b, u; /* approx solution, RHS, exact solution */
30: Mat A, F;
31: KSP ksp; /* linear solver context */
32: PC pc;
33: PetscRandom rctx; /* random number generator context */
34: PetscReal norm; /* norm of solution error */
35: PetscInt i, j, Ii, J, Istart, Iend, m = 8, n = 7, its;
36: PetscBool flg = PETSC_FALSE, flg_ilu = PETSC_FALSE, flg_ch = PETSC_FALSE;
37: #if defined(PETSC_HAVE_MUMPS)
38: PetscBool flg_mumps = PETSC_FALSE, flg_mumps_ch = PETSC_FALSE;
39: #endif
40: #if defined(PETSC_HAVE_SUPERLU) || defined(PETSC_HAVE_SUPERLU_DIST)
41: PetscBool flg_superlu = PETSC_FALSE;
42: #endif
43: #if defined(PETSC_HAVE_STRUMPACK)
44: PetscBool flg_strumpack = PETSC_FALSE;
45: #endif
46: PetscScalar v;
47: PetscMPIInt rank, size;
48: #if defined(PETSC_USE_LOG)
49: PetscLogStage stage;
50: #endif
52: PetscFunctionBeginUser;
53: PetscCall(PetscInitialize(&argc, &args, (char *)0, help));
54: PetscCallMPI(MPI_Comm_rank(PETSC_COMM_WORLD, &rank));
55: PetscCallMPI(MPI_Comm_size(PETSC_COMM_WORLD, &size));
56: PetscCall(PetscOptionsGetInt(NULL, NULL, "-m", &m, NULL));
57: PetscCall(PetscOptionsGetInt(NULL, NULL, "-n", &n, NULL));
58: /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
59: Compute the matrix and right-hand-side vector that define
60: the linear system, Ax = b.
61: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
62: PetscCall(MatCreate(PETSC_COMM_WORLD, &A));
63: PetscCall(MatSetSizes(A, PETSC_DECIDE, PETSC_DECIDE, m * n, m * n));
64: PetscCall(MatSetFromOptions(A));
65: PetscCall(MatMPIAIJSetPreallocation(A, 5, NULL, 5, NULL));
66: PetscCall(MatSeqAIJSetPreallocation(A, 5, NULL));
67: PetscCall(MatSetUp(A));
69: /*
70: Currently, all PETSc parallel matrix formats are partitioned by
71: contiguous chunks of rows across the processors. Determine which
72: rows of the matrix are locally owned.
73: */
74: PetscCall(MatGetOwnershipRange(A, &Istart, &Iend));
76: /*
77: Set matrix elements for the 2-D, five-point stencil in parallel.
78: - Each processor needs to insert only elements that it owns
79: locally (but any non-local elements will be sent to the
80: appropriate processor during matrix assembly).
81: - Always specify global rows and columns of matrix entries.
83: Note: this uses the less common natural ordering that orders first
84: all the unknowns for x = h then for x = 2h etc; Hence you see J = Ii +- n
85: instead of J = I +- m as you might expect. The more standard ordering
86: would first do all variables for y = h, then y = 2h etc.
88: */
89: PetscCall(PetscLogStageRegister("Assembly", &stage));
90: PetscCall(PetscLogStagePush(stage));
91: for (Ii = Istart; Ii < Iend; Ii++) {
92: v = -1.0;
93: i = Ii / n;
94: j = Ii - i * n;
95: if (i > 0) {
96: J = Ii - n;
97: PetscCall(MatSetValues(A, 1, &Ii, 1, &J, &v, INSERT_VALUES));
98: }
99: if (i < m - 1) {
100: J = Ii + n;
101: PetscCall(MatSetValues(A, 1, &Ii, 1, &J, &v, INSERT_VALUES));
102: }
103: if (j > 0) {
104: J = Ii - 1;
105: PetscCall(MatSetValues(A, 1, &Ii, 1, &J, &v, INSERT_VALUES));
106: }
107: if (j < n - 1) {
108: J = Ii + 1;
109: PetscCall(MatSetValues(A, 1, &Ii, 1, &J, &v, INSERT_VALUES));
110: }
111: v = 4.0;
112: PetscCall(MatSetValues(A, 1, &Ii, 1, &Ii, &v, INSERT_VALUES));
113: }
115: /*
116: Assemble matrix, using the 2-step process:
117: MatAssemblyBegin(), MatAssemblyEnd()
118: Computations can be done while messages are in transition
119: by placing code between these two statements.
120: */
121: PetscCall(MatAssemblyBegin(A, MAT_FINAL_ASSEMBLY));
122: PetscCall(MatAssemblyEnd(A, MAT_FINAL_ASSEMBLY));
123: PetscCall(PetscLogStagePop());
125: /* A is symmetric. Set symmetric flag to enable ICC/Cholesky preconditioner */
126: PetscCall(MatSetOption(A, MAT_SYMMETRIC, PETSC_TRUE));
128: /* Create parallel vectors */
129: PetscCall(MatCreateVecs(A, &u, &b));
130: PetscCall(VecDuplicate(u, &x));
132: /*
133: Set exact solution; then compute right-hand-side vector.
134: By default we use an exact solution of a vector with all
135: elements of 1.0; Alternatively, using the runtime option
136: -random_sol forms a solution vector with random components.
137: */
138: PetscCall(PetscOptionsGetBool(NULL, NULL, "-random_exact_sol", &flg, NULL));
139: if (flg) {
140: PetscCall(PetscRandomCreate(PETSC_COMM_WORLD, &rctx));
141: PetscCall(PetscRandomSetFromOptions(rctx));
142: PetscCall(VecSetRandom(u, rctx));
143: PetscCall(PetscRandomDestroy(&rctx));
144: } else {
145: PetscCall(VecSet(u, 1.0));
146: }
147: PetscCall(MatMult(A, u, b));
149: /*
150: View the exact solution vector if desired
151: */
152: flg = PETSC_FALSE;
153: PetscCall(PetscOptionsGetBool(NULL, NULL, "-view_exact_sol", &flg, NULL));
154: if (flg) PetscCall(VecView(u, PETSC_VIEWER_STDOUT_WORLD));
156: /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
157: Create the linear solver and set various options
158: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
160: /*
161: Create linear solver context
162: */
163: PetscCall(KSPCreate(PETSC_COMM_WORLD, &ksp));
164: PetscCall(KSPSetOperators(ksp, A, A));
166: /*
167: Example of how to use external package MUMPS
168: Note: runtime options
169: '-ksp_type preonly -pc_type lu -pc_factor_mat_solver_type mumps -mat_mumps_icntl_7 3 -mat_mumps_icntl_1 0.0'
170: are equivalent to these procedural calls
171: */
172: #if defined(PETSC_HAVE_MUMPS)
173: flg_mumps = PETSC_FALSE;
174: flg_mumps_ch = PETSC_FALSE;
175: PetscCall(PetscOptionsGetBool(NULL, NULL, "-use_mumps_lu", &flg_mumps, NULL));
176: PetscCall(PetscOptionsGetBool(NULL, NULL, "-use_mumps_ch", &flg_mumps_ch, NULL));
177: if (flg_mumps || flg_mumps_ch) {
178: PetscCall(KSPSetType(ksp, KSPPREONLY));
179: PetscInt ival, icntl;
180: PetscReal val;
181: PetscCall(KSPGetPC(ksp, &pc));
182: if (flg_mumps) {
183: PetscCall(PCSetType(pc, PCLU));
184: } else if (flg_mumps_ch) {
185: PetscCall(MatSetOption(A, MAT_SPD, PETSC_TRUE)); /* set MUMPS id%SYM=1 */
186: PetscCall(PCSetType(pc, PCCHOLESKY));
187: }
188: PetscCall(PCFactorSetMatSolverType(pc, MATSOLVERMUMPS));
189: PetscCall(PCFactorSetUpMatSolverType(pc)); /* call MatGetFactor() to create F */
190: PetscCall(PCFactorGetMatrix(pc, &F));
191: PetscCall(MatMumpsSetIcntl(F, 24, 1));
192: PetscCall(MatMumpsGetIcntl(F, 24, &ival));
193: PetscCheck(ival == 1, PetscObjectComm((PetscObject)F), PETSC_ERR_LIB, "ICNTL(24) = %" PetscInt_FMT " (!= 1)", ival);
194: PetscCall(MatMumpsSetCntl(F, 3, 1e-6));
195: PetscCall(MatMumpsGetCntl(F, 3, &val));
196: PetscCheck(PetscEqualReal(val, 1e-6), PetscObjectComm((PetscObject)F), PETSC_ERR_LIB, "CNTL(3) = %g (!= %g)", (double)val, 1e-6);
197: if (flg_mumps) {
198: /* Zero the first and last rows in the rank, they should then show up in corresponding null pivot rows output via
199: MatMumpsGetNullPivots */
200: flg = PETSC_FALSE;
201: PetscCall(PetscOptionsGetBool(NULL, NULL, "-zero_first_and_last_rows", &flg, NULL));
202: if (flg) {
203: PetscInt rows[2];
204: rows[0] = Istart; /* first row of the rank */
205: rows[1] = Iend - 1; /* last row of the rank */
206: PetscCall(MatZeroRows(A, 2, rows, 0.0, NULL, NULL));
207: }
208: /* Get memory estimates from MUMPS' MatLUFactorSymbolic(), e.g. INFOG(16), INFOG(17).
209: KSPSetUp() below will do nothing inside MatLUFactorSymbolic() */
210: MatFactorInfo info;
211: PetscCall(MatLUFactorSymbolic(F, A, NULL, NULL, &info));
212: flg = PETSC_FALSE;
213: PetscCall(PetscOptionsGetBool(NULL, NULL, "-print_mumps_memory", &flg, NULL));
214: if (flg) PetscCall(printMumpsMemoryInfo(F));
215: }
217: /* sequential ordering */
218: icntl = 7;
219: ival = 2;
220: PetscCall(MatMumpsSetIcntl(F, icntl, ival));
222: /* threshold for row pivot detection */
223: PetscCall(MatMumpsGetIcntl(F, 24, &ival));
224: PetscCheck(ival == 1, PetscObjectComm((PetscObject)F), PETSC_ERR_LIB, "ICNTL(24) = %" PetscInt_FMT " (!= 1)", ival);
225: icntl = 3;
226: PetscCall(MatMumpsGetCntl(F, icntl, &val));
227: PetscCheck(PetscEqualReal(val, 1e-6), PetscObjectComm((PetscObject)F), PETSC_ERR_LIB, "CNTL(3) = %g (!= %g)", (double)val, 1e-6);
229: /* compute determinant of A */
230: PetscCall(MatMumpsSetIcntl(F, 33, 1));
231: }
232: #endif
234: /*
235: Example of how to use external package SuperLU
236: Note: runtime options
237: '-ksp_type preonly -pc_type ilu -pc_factor_mat_solver_type superlu -mat_superlu_ilu_droptol 1.e-8'
238: are equivalent to these procedual calls
239: */
240: #if defined(PETSC_HAVE_SUPERLU) || defined(PETSC_HAVE_SUPERLU_DIST)
241: flg_ilu = PETSC_FALSE;
242: flg_superlu = PETSC_FALSE;
243: PetscCall(PetscOptionsGetBool(NULL, NULL, "-use_superlu_lu", &flg_superlu, NULL));
244: PetscCall(PetscOptionsGetBool(NULL, NULL, "-use_superlu_ilu", &flg_ilu, NULL));
245: if (flg_superlu || flg_ilu) {
246: PetscCall(KSPSetType(ksp, KSPPREONLY));
247: PetscCall(KSPGetPC(ksp, &pc));
248: if (flg_superlu) PetscCall(PCSetType(pc, PCLU));
249: else if (flg_ilu) PetscCall(PCSetType(pc, PCILU));
250: if (size == 1) {
251: #if !defined(PETSC_HAVE_SUPERLU)
252: SETERRQ(PETSC_COMM_WORLD, PETSC_ERR_SUP, "This test requires SUPERLU");
253: #else
254: PetscCall(PCFactorSetMatSolverType(pc, MATSOLVERSUPERLU));
255: #endif
256: } else {
257: #if !defined(PETSC_HAVE_SUPERLU_DIST)
258: SETERRQ(PETSC_COMM_WORLD, PETSC_ERR_SUP, "This test requires SUPERLU_DIST");
259: #else
260: PetscCall(PCFactorSetMatSolverType(pc, MATSOLVERSUPERLU_DIST));
261: #endif
262: }
263: PetscCall(PCFactorSetUpMatSolverType(pc)); /* call MatGetFactor() to create F */
264: PetscCall(PCFactorGetMatrix(pc, &F));
265: #if defined(PETSC_HAVE_SUPERLU)
266: if (size == 1) PetscCall(MatSuperluSetILUDropTol(F, 1.e-8));
267: #endif
268: }
269: #endif
271: /*
272: Example of how to use external package STRUMPACK
273: Note: runtime options
274: '-pc_type lu/ilu \
275: -pc_factor_mat_solver_type strumpack \
276: -mat_strumpack_reordering METIS \
277: -mat_strumpack_colperm 0 \
278: -mat_strumpack_hss_rel_tol 1.e-3 \
279: -mat_strumpack_hss_min_sep_size 50 \
280: -mat_strumpack_max_rank 100 \
281: -mat_strumpack_leaf_size 4'
282: are equivalent to these procedural calls
284: We refer to the STRUMPACK-sparse manual, section 5, for more info on
285: how to tune the preconditioner.
286: */
287: #if defined(PETSC_HAVE_STRUMPACK)
288: flg_ilu = PETSC_FALSE;
289: flg_strumpack = PETSC_FALSE;
290: PetscCall(PetscOptionsGetBool(NULL, NULL, "-use_strumpack_lu", &flg_strumpack, NULL));
291: PetscCall(PetscOptionsGetBool(NULL, NULL, "-use_strumpack_ilu", &flg_ilu, NULL));
292: if (flg_strumpack || flg_ilu) {
293: PetscCall(KSPSetType(ksp, KSPPREONLY));
294: PetscCall(KSPGetPC(ksp, &pc));
295: if (flg_strumpack) PetscCall(PCSetType(pc, PCLU));
296: else if (flg_ilu) PetscCall(PCSetType(pc, PCILU));
297: #if !defined(PETSC_HAVE_STRUMPACK)
298: SETERRQ(PETSC_COMM_WORLD, PETSC_ERR_SUP, "This test requires STRUMPACK");
299: #endif
300: PetscCall(PCFactorSetMatSolverType(pc, MATSOLVERSTRUMPACK));
301: PetscCall(PCFactorSetUpMatSolverType(pc)); /* call MatGetFactor() to create F */
302: PetscCall(PCFactorGetMatrix(pc, &F));
303: #if defined(PETSC_HAVE_STRUMPACK)
304: /* Set the fill-reducing reordering. */
305: PetscCall(MatSTRUMPACKSetReordering(F, MAT_STRUMPACK_METIS));
306: /* Since this is a simple discretization, the diagonal is always */
307: /* nonzero, and there is no need for the extra MC64 permutation. */
308: PetscCall(MatSTRUMPACKSetColPerm(F, PETSC_FALSE));
309: /* The compression tolerance used when doing low-rank compression */
310: /* in the preconditioner. This is problem specific! */
311: PetscCall(MatSTRUMPACKSetHSSRelTol(F, 1.e-3));
312: /* Set minimum matrix size for HSS compression to 15 in order to */
313: /* demonstrate preconditioner on small problems. For performance */
314: /* a value of say 500 is better. */
315: PetscCall(MatSTRUMPACKSetHSSMinSepSize(F, 15));
316: /* You can further limit the fill in the preconditioner by */
317: /* setting a maximum rank */
318: PetscCall(MatSTRUMPACKSetHSSMaxRank(F, 100));
319: /* Set the size of the diagonal blocks (the leafs) in the HSS */
320: /* approximation. The default value should be better for real */
321: /* problems. This is mostly for illustration on a small problem. */
322: PetscCall(MatSTRUMPACKSetHSSLeafSize(F, 4));
323: #endif
324: }
325: #endif
327: /*
328: Example of how to use procedural calls that are equivalent to
329: '-ksp_type preonly -pc_type lu/ilu -pc_factor_mat_solver_type petsc'
330: */
331: flg = PETSC_FALSE;
332: flg_ilu = PETSC_FALSE;
333: flg_ch = PETSC_FALSE;
334: PetscCall(PetscOptionsGetBool(NULL, NULL, "-use_petsc_lu", &flg, NULL));
335: PetscCall(PetscOptionsGetBool(NULL, NULL, "-use_petsc_ilu", &flg_ilu, NULL));
336: PetscCall(PetscOptionsGetBool(NULL, NULL, "-use_petsc_ch", &flg_ch, NULL));
337: if (flg || flg_ilu || flg_ch) {
338: Vec diag;
340: PetscCall(KSPSetType(ksp, KSPPREONLY));
341: PetscCall(KSPGetPC(ksp, &pc));
342: if (flg) PetscCall(PCSetType(pc, PCLU));
343: else if (flg_ilu) PetscCall(PCSetType(pc, PCILU));
344: else if (flg_ch) PetscCall(PCSetType(pc, PCCHOLESKY));
345: PetscCall(PCFactorSetMatSolverType(pc, MATSOLVERPETSC));
346: PetscCall(PCFactorSetUpMatSolverType(pc)); /* call MatGetFactor() to create F */
347: PetscCall(PCFactorGetMatrix(pc, &F));
349: /* Test MatGetDiagonal() */
350: PetscCall(KSPSetUp(ksp));
351: PetscCall(VecDuplicate(x, &diag));
352: PetscCall(MatGetDiagonal(F, diag));
353: /* PetscCall(VecView(diag,PETSC_VIEWER_STDOUT_WORLD)); */
354: PetscCall(VecDestroy(&diag));
355: }
357: PetscCall(KSPSetFromOptions(ksp));
359: /* Get info from matrix factors */
360: PetscCall(KSPSetUp(ksp));
362: #if defined(PETSC_HAVE_MUMPS)
363: if (flg_mumps || flg_mumps_ch) {
364: PetscInt icntl, infog34, num_null_pivots, *null_pivots;
365: PetscReal cntl, rinfo12, rinfo13;
366: icntl = 3;
367: PetscCall(MatMumpsGetCntl(F, icntl, &cntl));
369: /* compute determinant and check for any null pivots*/
370: if (rank == 0) {
371: PetscCall(MatMumpsGetInfog(F, 34, &infog34));
372: PetscCall(MatMumpsGetRinfog(F, 12, &rinfo12));
373: PetscCall(MatMumpsGetRinfog(F, 13, &rinfo13));
374: PetscCall(MatMumpsGetNullPivots(F, &num_null_pivots, &null_pivots));
375: PetscCall(PetscPrintf(PETSC_COMM_SELF, " Mumps row pivot threshold = %g\n", cntl));
376: PetscCall(PetscPrintf(PETSC_COMM_SELF, " Mumps determinant = (%g, %g) * 2^%" PetscInt_FMT " \n", (double)rinfo12, (double)rinfo13, infog34));
377: if (num_null_pivots > 0) {
378: PetscCall(PetscPrintf(PETSC_COMM_SELF, " Mumps num of null pivots detected = %" PetscInt_FMT "\n", num_null_pivots));
379: PetscCall(PetscSortInt(num_null_pivots, null_pivots)); /* just make the printf deterministic */
380: for (j = 0; j < num_null_pivots; j++) PetscCall(PetscPrintf(PETSC_COMM_SELF, " Mumps row with null pivots is = %" PetscInt_FMT "\n", null_pivots[j]));
381: }
382: PetscCall(PetscFree(null_pivots));
383: }
384: }
385: #endif
387: /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
388: Solve the linear system
389: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
390: PetscCall(KSPSolve(ksp, b, x));
392: /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
393: Check solution and clean up
394: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
395: PetscCall(VecAXPY(x, -1.0, u));
396: PetscCall(VecNorm(x, NORM_2, &norm));
397: PetscCall(KSPGetIterationNumber(ksp, &its));
399: /*
400: Print convergence information. PetscPrintf() produces a single
401: print statement from all processes that share a communicator.
402: An alternative is PetscFPrintf(), which prints to a file.
403: */
404: if (norm < 1.e-12) {
405: PetscCall(PetscPrintf(PETSC_COMM_WORLD, "Norm of error < 1.e-12 iterations %" PetscInt_FMT "\n", its));
406: } else {
407: PetscCall(PetscPrintf(PETSC_COMM_WORLD, "Norm of error %g iterations %" PetscInt_FMT "\n", (double)norm, its));
408: }
410: /*
411: Free work space. All PETSc objects should be destroyed when they
412: are no longer needed.
413: */
414: PetscCall(KSPDestroy(&ksp));
415: PetscCall(VecDestroy(&u));
416: PetscCall(VecDestroy(&x));
417: PetscCall(VecDestroy(&b));
418: PetscCall(MatDestroy(&A));
420: /*
421: Always call PetscFinalize() before exiting a program. This routine
422: - finalizes the PETSc libraries as well as MPI
423: - provides summary and diagnostic information if certain runtime
424: options are chosen (e.g., -log_view).
425: */
426: PetscCall(PetscFinalize());
427: return 0;
428: }
430: /*TEST
432: test:
433: args: -use_petsc_lu
434: output_file: output/ex52_2.out
436: test:
437: suffix: mumps
438: nsize: 3
439: requires: mumps
440: args: -use_mumps_lu
441: output_file: output/ex52_1.out
443: test:
444: suffix: mumps_2
445: nsize: 3
446: requires: mumps
447: args: -use_mumps_ch
448: output_file: output/ex52_1.out
450: test:
451: suffix: mumps_3
452: nsize: 3
453: requires: mumps
454: args: -use_mumps_ch -mat_type sbaij
455: output_file: output/ex52_1.out
457: test:
458: suffix: mumps_4
459: nsize: 3
460: requires: mumps !complex !single
461: args: -use_mumps_lu -m 50 -n 50 -print_mumps_memory
462: output_file: output/ex52_4.out
464: test:
465: suffix: mumps_5
466: nsize: 3
467: requires: mumps !complex !single
468: args: -use_mumps_lu -m 50 -n 50 -zero_first_and_last_rows
469: output_file: output/ex52_5.out
471: test:
472: suffix: mumps_omp_2
473: nsize: 4
474: requires: mumps hwloc openmp pthread defined(PETSC_HAVE_MPI_PROCESS_SHARED_MEMORY)
475: args: -use_mumps_lu -mat_mumps_use_omp_threads 2
476: output_file: output/ex52_1.out
478: test:
479: suffix: mumps_omp_3
480: nsize: 4
481: requires: mumps hwloc openmp pthread defined(PETSC_HAVE_MPI_PROCESS_SHARED_MEMORY)
482: args: -use_mumps_ch -mat_mumps_use_omp_threads 3
483: # Ignore the warning since we are intentionally testing the imbalanced case
484: filter: grep -v "Warning: number of OpenMP threads"
485: output_file: output/ex52_1.out
487: test:
488: suffix: mumps_omp_4
489: nsize: 4
490: requires: mumps hwloc openmp pthread defined(PETSC_HAVE_MPI_PROCESS_SHARED_MEMORY)
491: # let petsc guess a proper number for threads
492: args: -use_mumps_ch -mat_type sbaij -mat_mumps_use_omp_threads
493: output_file: output/ex52_1.out
495: testset:
496: suffix: strumpack_2
497: nsize: {{1 2}}
498: requires: strumpack
499: args: -use_strumpack_lu
500: output_file: output/ex52_3.out
502: test:
503: suffix: aij
504: args: -mat_type aij
506: test:
507: requires: kokkos_kernels
508: suffix: kok
509: args: -mat_type aijkokkos
511: test:
512: requires: cuda
513: suffix: cuda
514: args: -mat_type aijcusparse
516: test:
517: requires: hip
518: suffix: hip
519: args: -mat_type aijhipsparse
521: test:
522: suffix: strumpack_ilu
523: nsize: {{1 2}}
524: requires: strumpack
525: args: -use_strumpack_ilu
526: output_file: output/ex52_3.out
528: testset:
529: suffix: superlu_dist
530: nsize: {{1 2}}
531: requires: superlu superlu_dist
532: args: -use_superlu_lu
533: output_file: output/ex52_2.out
535: test:
536: suffix: aij
537: args: -mat_type aij
539: test:
540: requires: kokkos_kernels
541: suffix: kok
542: args: -mat_type aijkokkos
544: test:
545: requires: cuda
546: suffix: cuda
547: args: -mat_type aijcusparse
549: test:
550: requires: hip
551: suffix: hip
552: args: -mat_type aijhipsparse
554: test:
555: suffix: superlu_ilu
556: requires: superlu superlu_dist
557: args: -use_superlu_ilu
558: output_file: output/ex52_2.out
560: TEST*/