Actual source code: chwirut2.c
1: /*
2: Include "petsctao.h" so that we can use TAO solvers. Note that this
3: file automatically includes libraries such as:
4: petsc.h - base PETSc routines petscvec.h - vectors
5: petscsys.h - system routines petscmat.h - matrices
6: petscis.h - index sets petscksp.h - Krylov subspace methods
7: petscviewer.h - viewers petscpc.h - preconditioners
9: */
11: #include <petsctao.h>
13: /*
14: Description: These data are the result of a NIST study involving
15: ultrasonic calibration. The response variable is
16: ultrasonic response, and the predictor variable is
17: metal distance.
19: Reference: Chwirut, D., NIST (197?).
20: Ultrasonic Reference Block Study.
21: */
23: static char help[] = "Finds the nonlinear least-squares solution to the model \n\
24: y = exp[-b1*x]/(b2+b3*x) + e \n";
26: #define NOBSERVATIONS 214
27: #define NPARAMETERS 3
29: #define DIE_TAG 2000
30: #define IDLE_TAG 1000
32: /* User-defined application context */
33: typedef struct {
34: /* Working space */
35: PetscReal t[NOBSERVATIONS]; /* array of independent variables of observation */
36: PetscReal y[NOBSERVATIONS]; /* array of dependent variables */
37: PetscMPIInt size, rank;
38: } AppCtx;
40: /* User provided Routines */
41: PetscErrorCode InitializeData(AppCtx *user);
42: PetscErrorCode FormStartingPoint(Vec);
43: PetscErrorCode EvaluateFunction(Tao, Vec, Vec, void *);
44: PetscErrorCode TaskWorker(AppCtx *user);
45: PetscErrorCode StopWorkers(AppCtx *user);
46: PetscErrorCode RunSimulation(PetscReal *x, PetscInt i, PetscReal *f, AppCtx *user);
48: /*--------------------------------------------------------------------*/
49: int main(int argc, char **argv)
50: {
51: Vec x, f; /* solution, function */
52: Tao tao; /* Tao solver context */
53: AppCtx user; /* user-defined work context */
55: /* Initialize TAO and PETSc */
56: PetscFunctionBeginUser;
57: PetscCall(PetscInitialize(&argc, &argv, NULL, help));
58: MPI_Comm_size(MPI_COMM_WORLD, &user.size);
59: MPI_Comm_rank(MPI_COMM_WORLD, &user.rank);
60: PetscCall(InitializeData(&user));
62: /* Run optimization on rank 0 */
63: if (user.rank == 0) {
64: /* Allocate vectors */
65: PetscCall(VecCreateSeq(PETSC_COMM_SELF, NPARAMETERS, &x));
66: PetscCall(VecCreateSeq(PETSC_COMM_SELF, NOBSERVATIONS, &f));
68: /* TAO code begins here */
70: /* Create TAO solver and set desired solution method */
71: PetscCall(TaoCreate(PETSC_COMM_SELF, &tao));
72: PetscCall(TaoSetType(tao, TAOPOUNDERS));
74: /* Set the function and Jacobian routines. */
75: PetscCall(FormStartingPoint(x));
76: PetscCall(TaoSetSolution(tao, x));
77: PetscCall(TaoSetResidualRoutine(tao, f, EvaluateFunction, (void *)&user));
79: /* Check for any TAO command line arguments */
80: PetscCall(TaoSetFromOptions(tao));
82: /* Perform the Solve */
83: PetscCall(TaoSolve(tao));
85: /* Free TAO data structures */
86: PetscCall(TaoDestroy(&tao));
88: /* Free PETSc data structures */
89: PetscCall(VecDestroy(&x));
90: PetscCall(VecDestroy(&f));
91: PetscCall(StopWorkers(&user));
92: } else {
93: PetscCall(TaskWorker(&user));
94: }
95: PetscCall(PetscFinalize());
96: return 0;
97: }
99: /*--------------------------------------------------------------------*/
100: PetscErrorCode EvaluateFunction(Tao tao, Vec X, Vec F, void *ptr)
101: {
102: AppCtx *user = (AppCtx *)ptr;
103: PetscInt i;
104: PetscReal *x, *f;
106: PetscFunctionBegin;
107: PetscCall(VecGetArray(X, &x));
108: PetscCall(VecGetArray(F, &f));
109: if (user->size == 1) {
110: /* Single processor */
111: for (i = 0; i < NOBSERVATIONS; i++) PetscCall(RunSimulation(x, i, &f[i], user));
112: } else {
113: /* Multiprocessor main */
114: PetscMPIInt tag;
115: PetscInt finishedtasks, next_task, checkedin;
116: PetscReal f_i = 0.0;
117: MPI_Status status;
119: next_task = 0;
120: finishedtasks = 0;
121: checkedin = 0;
123: while (finishedtasks < NOBSERVATIONS || checkedin < user->size - 1) {
124: PetscCallMPI(MPI_Recv(&f_i, 1, MPIU_REAL, MPI_ANY_SOURCE, MPI_ANY_TAG, PETSC_COMM_WORLD, &status));
125: if (status.MPI_TAG == IDLE_TAG) {
126: checkedin++;
127: } else {
128: tag = status.MPI_TAG;
129: f[tag] = (PetscReal)f_i;
130: finishedtasks++;
131: }
133: if (next_task < NOBSERVATIONS) {
134: PetscCallMPI(MPI_Send(x, NPARAMETERS, MPIU_REAL, status.MPI_SOURCE, (PetscMPIInt)next_task, PETSC_COMM_WORLD));
135: next_task++;
137: } else {
138: /* Send idle message */
139: PetscCallMPI(MPI_Send(x, NPARAMETERS, MPIU_REAL, status.MPI_SOURCE, IDLE_TAG, PETSC_COMM_WORLD));
140: }
141: }
142: }
143: PetscCall(VecRestoreArray(X, &x));
144: PetscCall(VecRestoreArray(F, &f));
145: PetscCall(PetscLogFlops(6 * NOBSERVATIONS));
146: PetscFunctionReturn(PETSC_SUCCESS);
147: }
149: /* ------------------------------------------------------------ */
150: PetscErrorCode FormStartingPoint(Vec X)
151: {
152: PetscReal *x;
154: PetscFunctionBegin;
155: PetscCall(VecGetArray(X, &x));
156: x[0] = 0.15;
157: x[1] = 0.008;
158: x[2] = 0.010;
159: PetscCall(VecRestoreArray(X, &x));
160: PetscFunctionReturn(PETSC_SUCCESS);
161: }
163: /* ---------------------------------------------------------------------- */
164: PetscErrorCode InitializeData(AppCtx *user)
165: {
166: PetscReal *t = user->t, *y = user->y;
167: PetscInt i = 0;
169: PetscFunctionBegin;
170: y[i] = 92.9000;
171: t[i++] = 0.5000;
172: y[i] = 78.7000;
173: t[i++] = 0.6250;
174: y[i] = 64.2000;
175: t[i++] = 0.7500;
176: y[i] = 64.9000;
177: t[i++] = 0.8750;
178: y[i] = 57.1000;
179: t[i++] = 1.0000;
180: y[i] = 43.3000;
181: t[i++] = 1.2500;
182: y[i] = 31.1000;
183: t[i++] = 1.7500;
184: y[i] = 23.6000;
185: t[i++] = 2.2500;
186: y[i] = 31.0500;
187: t[i++] = 1.7500;
188: y[i] = 23.7750;
189: t[i++] = 2.2500;
190: y[i] = 17.7375;
191: t[i++] = 2.7500;
192: y[i] = 13.8000;
193: t[i++] = 3.2500;
194: y[i] = 11.5875;
195: t[i++] = 3.7500;
196: y[i] = 9.4125;
197: t[i++] = 4.2500;
198: y[i] = 7.7250;
199: t[i++] = 4.7500;
200: y[i] = 7.3500;
201: t[i++] = 5.2500;
202: y[i] = 8.0250;
203: t[i++] = 5.7500;
204: y[i] = 90.6000;
205: t[i++] = 0.5000;
206: y[i] = 76.9000;
207: t[i++] = 0.6250;
208: y[i] = 71.6000;
209: t[i++] = 0.7500;
210: y[i] = 63.6000;
211: t[i++] = 0.8750;
212: y[i] = 54.0000;
213: t[i++] = 1.0000;
214: y[i] = 39.2000;
215: t[i++] = 1.2500;
216: y[i] = 29.3000;
217: t[i++] = 1.7500;
218: y[i] = 21.4000;
219: t[i++] = 2.2500;
220: y[i] = 29.1750;
221: t[i++] = 1.7500;
222: y[i] = 22.1250;
223: t[i++] = 2.2500;
224: y[i] = 17.5125;
225: t[i++] = 2.7500;
226: y[i] = 14.2500;
227: t[i++] = 3.2500;
228: y[i] = 9.4500;
229: t[i++] = 3.7500;
230: y[i] = 9.1500;
231: t[i++] = 4.2500;
232: y[i] = 7.9125;
233: t[i++] = 4.7500;
234: y[i] = 8.4750;
235: t[i++] = 5.2500;
236: y[i] = 6.1125;
237: t[i++] = 5.7500;
238: y[i] = 80.0000;
239: t[i++] = 0.5000;
240: y[i] = 79.0000;
241: t[i++] = 0.6250;
242: y[i] = 63.8000;
243: t[i++] = 0.7500;
244: y[i] = 57.2000;
245: t[i++] = 0.8750;
246: y[i] = 53.2000;
247: t[i++] = 1.0000;
248: y[i] = 42.5000;
249: t[i++] = 1.2500;
250: y[i] = 26.8000;
251: t[i++] = 1.7500;
252: y[i] = 20.4000;
253: t[i++] = 2.2500;
254: y[i] = 26.8500;
255: t[i++] = 1.7500;
256: y[i] = 21.0000;
257: t[i++] = 2.2500;
258: y[i] = 16.4625;
259: t[i++] = 2.7500;
260: y[i] = 12.5250;
261: t[i++] = 3.2500;
262: y[i] = 10.5375;
263: t[i++] = 3.7500;
264: y[i] = 8.5875;
265: t[i++] = 4.2500;
266: y[i] = 7.1250;
267: t[i++] = 4.7500;
268: y[i] = 6.1125;
269: t[i++] = 5.2500;
270: y[i] = 5.9625;
271: t[i++] = 5.7500;
272: y[i] = 74.1000;
273: t[i++] = 0.5000;
274: y[i] = 67.3000;
275: t[i++] = 0.6250;
276: y[i] = 60.8000;
277: t[i++] = 0.7500;
278: y[i] = 55.5000;
279: t[i++] = 0.8750;
280: y[i] = 50.3000;
281: t[i++] = 1.0000;
282: y[i] = 41.0000;
283: t[i++] = 1.2500;
284: y[i] = 29.4000;
285: t[i++] = 1.7500;
286: y[i] = 20.4000;
287: t[i++] = 2.2500;
288: y[i] = 29.3625;
289: t[i++] = 1.7500;
290: y[i] = 21.1500;
291: t[i++] = 2.2500;
292: y[i] = 16.7625;
293: t[i++] = 2.7500;
294: y[i] = 13.2000;
295: t[i++] = 3.2500;
296: y[i] = 10.8750;
297: t[i++] = 3.7500;
298: y[i] = 8.1750;
299: t[i++] = 4.2500;
300: y[i] = 7.3500;
301: t[i++] = 4.7500;
302: y[i] = 5.9625;
303: t[i++] = 5.2500;
304: y[i] = 5.6250;
305: t[i++] = 5.7500;
306: y[i] = 81.5000;
307: t[i++] = .5000;
308: y[i] = 62.4000;
309: t[i++] = .7500;
310: y[i] = 32.5000;
311: t[i++] = 1.5000;
312: y[i] = 12.4100;
313: t[i++] = 3.0000;
314: y[i] = 13.1200;
315: t[i++] = 3.0000;
316: y[i] = 15.5600;
317: t[i++] = 3.0000;
318: y[i] = 5.6300;
319: t[i++] = 6.0000;
320: y[i] = 78.0000;
321: t[i++] = .5000;
322: y[i] = 59.9000;
323: t[i++] = .7500;
324: y[i] = 33.2000;
325: t[i++] = 1.5000;
326: y[i] = 13.8400;
327: t[i++] = 3.0000;
328: y[i] = 12.7500;
329: t[i++] = 3.0000;
330: y[i] = 14.6200;
331: t[i++] = 3.0000;
332: y[i] = 3.9400;
333: t[i++] = 6.0000;
334: y[i] = 76.8000;
335: t[i++] = .5000;
336: y[i] = 61.0000;
337: t[i++] = .7500;
338: y[i] = 32.9000;
339: t[i++] = 1.5000;
340: y[i] = 13.8700;
341: t[i++] = 3.0000;
342: y[i] = 11.8100;
343: t[i++] = 3.0000;
344: y[i] = 13.3100;
345: t[i++] = 3.0000;
346: y[i] = 5.4400;
347: t[i++] = 6.0000;
348: y[i] = 78.0000;
349: t[i++] = .5000;
350: y[i] = 63.5000;
351: t[i++] = .7500;
352: y[i] = 33.8000;
353: t[i++] = 1.5000;
354: y[i] = 12.5600;
355: t[i++] = 3.0000;
356: y[i] = 5.6300;
357: t[i++] = 6.0000;
358: y[i] = 12.7500;
359: t[i++] = 3.0000;
360: y[i] = 13.1200;
361: t[i++] = 3.0000;
362: y[i] = 5.4400;
363: t[i++] = 6.0000;
364: y[i] = 76.8000;
365: t[i++] = .5000;
366: y[i] = 60.0000;
367: t[i++] = .7500;
368: y[i] = 47.8000;
369: t[i++] = 1.0000;
370: y[i] = 32.0000;
371: t[i++] = 1.5000;
372: y[i] = 22.2000;
373: t[i++] = 2.0000;
374: y[i] = 22.5700;
375: t[i++] = 2.0000;
376: y[i] = 18.8200;
377: t[i++] = 2.5000;
378: y[i] = 13.9500;
379: t[i++] = 3.0000;
380: y[i] = 11.2500;
381: t[i++] = 4.0000;
382: y[i] = 9.0000;
383: t[i++] = 5.0000;
384: y[i] = 6.6700;
385: t[i++] = 6.0000;
386: y[i] = 75.8000;
387: t[i++] = .5000;
388: y[i] = 62.0000;
389: t[i++] = .7500;
390: y[i] = 48.8000;
391: t[i++] = 1.0000;
392: y[i] = 35.2000;
393: t[i++] = 1.5000;
394: y[i] = 20.0000;
395: t[i++] = 2.0000;
396: y[i] = 20.3200;
397: t[i++] = 2.0000;
398: y[i] = 19.3100;
399: t[i++] = 2.5000;
400: y[i] = 12.7500;
401: t[i++] = 3.0000;
402: y[i] = 10.4200;
403: t[i++] = 4.0000;
404: y[i] = 7.3100;
405: t[i++] = 5.0000;
406: y[i] = 7.4200;
407: t[i++] = 6.0000;
408: y[i] = 70.5000;
409: t[i++] = .5000;
410: y[i] = 59.5000;
411: t[i++] = .7500;
412: y[i] = 48.5000;
413: t[i++] = 1.0000;
414: y[i] = 35.8000;
415: t[i++] = 1.5000;
416: y[i] = 21.0000;
417: t[i++] = 2.0000;
418: y[i] = 21.6700;
419: t[i++] = 2.0000;
420: y[i] = 21.0000;
421: t[i++] = 2.5000;
422: y[i] = 15.6400;
423: t[i++] = 3.0000;
424: y[i] = 8.1700;
425: t[i++] = 4.0000;
426: y[i] = 8.5500;
427: t[i++] = 5.0000;
428: y[i] = 10.1200;
429: t[i++] = 6.0000;
430: y[i] = 78.0000;
431: t[i++] = .5000;
432: y[i] = 66.0000;
433: t[i++] = .6250;
434: y[i] = 62.0000;
435: t[i++] = .7500;
436: y[i] = 58.0000;
437: t[i++] = .8750;
438: y[i] = 47.7000;
439: t[i++] = 1.0000;
440: y[i] = 37.8000;
441: t[i++] = 1.2500;
442: y[i] = 20.2000;
443: t[i++] = 2.2500;
444: y[i] = 21.0700;
445: t[i++] = 2.2500;
446: y[i] = 13.8700;
447: t[i++] = 2.7500;
448: y[i] = 9.6700;
449: t[i++] = 3.2500;
450: y[i] = 7.7600;
451: t[i++] = 3.7500;
452: y[i] = 5.4400;
453: t[i++] = 4.2500;
454: y[i] = 4.8700;
455: t[i++] = 4.7500;
456: y[i] = 4.0100;
457: t[i++] = 5.2500;
458: y[i] = 3.7500;
459: t[i++] = 5.7500;
460: y[i] = 24.1900;
461: t[i++] = 3.0000;
462: y[i] = 25.7600;
463: t[i++] = 3.0000;
464: y[i] = 18.0700;
465: t[i++] = 3.0000;
466: y[i] = 11.8100;
467: t[i++] = 3.0000;
468: y[i] = 12.0700;
469: t[i++] = 3.0000;
470: y[i] = 16.1200;
471: t[i++] = 3.0000;
472: y[i] = 70.8000;
473: t[i++] = .5000;
474: y[i] = 54.7000;
475: t[i++] = .7500;
476: y[i] = 48.0000;
477: t[i++] = 1.0000;
478: y[i] = 39.8000;
479: t[i++] = 1.5000;
480: y[i] = 29.8000;
481: t[i++] = 2.0000;
482: y[i] = 23.7000;
483: t[i++] = 2.5000;
484: y[i] = 29.6200;
485: t[i++] = 2.0000;
486: y[i] = 23.8100;
487: t[i++] = 2.5000;
488: y[i] = 17.7000;
489: t[i++] = 3.0000;
490: y[i] = 11.5500;
491: t[i++] = 4.0000;
492: y[i] = 12.0700;
493: t[i++] = 5.0000;
494: y[i] = 8.7400;
495: t[i++] = 6.0000;
496: y[i] = 80.7000;
497: t[i++] = .5000;
498: y[i] = 61.3000;
499: t[i++] = .7500;
500: y[i] = 47.5000;
501: t[i++] = 1.0000;
502: y[i] = 29.0000;
503: t[i++] = 1.5000;
504: y[i] = 24.0000;
505: t[i++] = 2.0000;
506: y[i] = 17.7000;
507: t[i++] = 2.5000;
508: y[i] = 24.5600;
509: t[i++] = 2.0000;
510: y[i] = 18.6700;
511: t[i++] = 2.5000;
512: y[i] = 16.2400;
513: t[i++] = 3.0000;
514: y[i] = 8.7400;
515: t[i++] = 4.0000;
516: y[i] = 7.8700;
517: t[i++] = 5.0000;
518: y[i] = 8.5100;
519: t[i++] = 6.0000;
520: y[i] = 66.7000;
521: t[i++] = .5000;
522: y[i] = 59.2000;
523: t[i++] = .7500;
524: y[i] = 40.8000;
525: t[i++] = 1.0000;
526: y[i] = 30.7000;
527: t[i++] = 1.5000;
528: y[i] = 25.7000;
529: t[i++] = 2.0000;
530: y[i] = 16.3000;
531: t[i++] = 2.5000;
532: y[i] = 25.9900;
533: t[i++] = 2.0000;
534: y[i] = 16.9500;
535: t[i++] = 2.5000;
536: y[i] = 13.3500;
537: t[i++] = 3.0000;
538: y[i] = 8.6200;
539: t[i++] = 4.0000;
540: y[i] = 7.2000;
541: t[i++] = 5.0000;
542: y[i] = 6.6400;
543: t[i++] = 6.0000;
544: y[i] = 13.6900;
545: t[i++] = 3.0000;
546: y[i] = 81.0000;
547: t[i++] = .5000;
548: y[i] = 64.5000;
549: t[i++] = .7500;
550: y[i] = 35.5000;
551: t[i++] = 1.5000;
552: y[i] = 13.3100;
553: t[i++] = 3.0000;
554: y[i] = 4.8700;
555: t[i++] = 6.0000;
556: y[i] = 12.9400;
557: t[i++] = 3.0000;
558: y[i] = 5.0600;
559: t[i++] = 6.0000;
560: y[i] = 15.1900;
561: t[i++] = 3.0000;
562: y[i] = 14.6200;
563: t[i++] = 3.0000;
564: y[i] = 15.6400;
565: t[i++] = 3.0000;
566: y[i] = 25.5000;
567: t[i++] = 1.7500;
568: y[i] = 25.9500;
569: t[i++] = 1.7500;
570: y[i] = 81.7000;
571: t[i++] = .5000;
572: y[i] = 61.6000;
573: t[i++] = .7500;
574: y[i] = 29.8000;
575: t[i++] = 1.7500;
576: y[i] = 29.8100;
577: t[i++] = 1.7500;
578: y[i] = 17.1700;
579: t[i++] = 2.7500;
580: y[i] = 10.3900;
581: t[i++] = 3.7500;
582: y[i] = 28.4000;
583: t[i++] = 1.7500;
584: y[i] = 28.6900;
585: t[i++] = 1.7500;
586: y[i] = 81.3000;
587: t[i++] = .5000;
588: y[i] = 60.9000;
589: t[i++] = .7500;
590: y[i] = 16.6500;
591: t[i++] = 2.7500;
592: y[i] = 10.0500;
593: t[i++] = 3.7500;
594: y[i] = 28.9000;
595: t[i++] = 1.7500;
596: y[i] = 28.9500;
597: t[i++] = 1.7500;
598: PetscFunctionReturn(PETSC_SUCCESS);
599: }
601: PetscErrorCode TaskWorker(AppCtx *user)
602: {
603: PetscReal x[NPARAMETERS], f = 0.0;
604: PetscMPIInt tag = IDLE_TAG;
605: PetscInt index;
606: MPI_Status status;
608: PetscFunctionBegin;
609: /* Send check-in message to rank-0 */
611: PetscCallMPI(MPI_Send(&f, 1, MPIU_REAL, 0, IDLE_TAG, PETSC_COMM_WORLD));
612: while (tag != DIE_TAG) {
613: PetscCallMPI(MPI_Recv(x, NPARAMETERS, MPIU_REAL, 0, MPI_ANY_TAG, PETSC_COMM_WORLD, &status));
614: tag = status.MPI_TAG;
615: if (tag == IDLE_TAG) {
616: PetscCallMPI(MPI_Send(&f, 1, MPIU_REAL, 0, IDLE_TAG, PETSC_COMM_WORLD));
617: } else if (tag != DIE_TAG) {
618: index = (PetscInt)tag;
619: PetscCall(RunSimulation(x, index, &f, user));
620: PetscCallMPI(MPI_Send(&f, 1, MPIU_REAL, 0, tag, PETSC_COMM_WORLD));
621: }
622: }
623: PetscFunctionReturn(PETSC_SUCCESS);
624: }
626: PetscErrorCode RunSimulation(PetscReal *x, PetscInt i, PetscReal *f, AppCtx *user)
627: {
628: PetscReal *t = user->t;
629: PetscReal *y = user->y;
631: PetscFunctionBeginUser;
632: #if defined(PETSC_USE_REAL_SINGLE)
633: *f = y[i] - exp(-x[0] * t[i]) / (x[1] + x[2] * t[i]); /* expf() for single-precision breaks this example on Freebsd, Valgrind errors on Linux */
634: #else
635: *f = y[i] - PetscExpScalar(-x[0] * t[i]) / (x[1] + x[2] * t[i]);
636: #endif
637: PetscFunctionReturn(PETSC_SUCCESS);
638: }
640: PetscErrorCode StopWorkers(AppCtx *user)
641: {
642: PetscInt checkedin;
643: MPI_Status status;
644: PetscReal f, x[NPARAMETERS];
646: PetscFunctionBeginUser;
647: checkedin = 0;
648: while (checkedin < user->size - 1) {
649: PetscCallMPI(MPI_Recv(&f, 1, MPIU_REAL, MPI_ANY_SOURCE, MPI_ANY_TAG, PETSC_COMM_WORLD, &status));
650: checkedin++;
651: PetscCall(PetscArrayzero(x, NPARAMETERS));
652: PetscCallMPI(MPI_Send(x, NPARAMETERS, MPIU_REAL, status.MPI_SOURCE, DIE_TAG, PETSC_COMM_WORLD));
653: }
654: PetscFunctionReturn(PETSC_SUCCESS);
655: }
657: /*TEST
659: build:
660: requires: !complex
662: test:
663: nsize: 3
664: requires: !single
665: args: -tao_monitor_short -tao_max_it 100 -tao_type pounders -tao_gatol 1.e-5
667: TEST*/