1 /* $Id: petsclog.h,v 1.155 2001/09/06 14:51:20 bsmith Exp $ */ 2 3 /* 4 Defines profile/logging in PETSc. 5 */ 6 7 #if !defined(__PetscLog_H) 8 #define __PetscLog_H 9 #include "petsc.h" 10 11 /* 12 Each PETSc object class has it's own cookie (internal integer in the 13 data structure used for error checking). These are all defined by an offset 14 from the lowest one, PETSC_COOKIE. 15 */ 16 #define PETSC_COOKIE 1211211 17 extern int PETSC_LARGEST_COOKIE; 18 #define PETSC_EVENT 1311311 19 extern int PETSC_LARGEST_EVENT; 20 21 /* Events for the Petsc standard library */ 22 extern int PETSC_Barrier; 23 24 /* Global flop counter */ 25 extern PetscLogDouble _TotalFlops; 26 27 /* General logging of information; different from event logging */ 28 EXTERN int PetscLogInfo(void*,const char[],...); 29 EXTERN int PetscLogInfoDeactivateClass(int); 30 EXTERN int PetscLogInfoActivateClass(int); 31 extern PetscTruth PetscLogPrintInfo; /* if true, indicates PetscLogInfo() is turned on */ 32 33 #if defined(PETSC_USE_LOG) /* --- Logging is turned on --------------------------------*/ 34 35 /* 36 Flop counting: We count each arithmetic operation (e.g., addition, multiplication) separately. 37 38 For the complex numbers version, note that 39 1 complex addition = 2 flops 40 1 complex multiplication = 6 flops, 41 where we define 1 flop as that for a double precision scalar. We roughly approximate 42 flop counting for complex numbers by multiplying the total flops by 4; this corresponds 43 to the assumption that we're counting mostly additions and multiplications -- and 44 roughly the same number of each. More accurate counting could be done by distinguishing 45 among the various arithmetic operations. 46 */ 47 48 #if defined(PETSC_USE_COMPLEX) 49 #define PetscLogFlops(n) (_TotalFlops += (4*n),0) 50 #else 51 #define PetscLogFlops(n) (_TotalFlops += (n),0) 52 #endif 53 54 #if defined (PETSC_HAVE_MPE) 55 #include "mpe.h" 56 #define MPEBEGIN 1000 57 EXTERN int PetscLogMPEBegin(void); 58 EXTERN int PetscLogMPEDump(const char[]); 59 extern PetscTruth UseMPE; 60 extern int PetscLogEventMPEFlags[]; 61 EXTERN int PetscLogEventMPEActivate(int); 62 EXTERN int PetscLogEventMPEDeactivate(int); 63 #else 64 #define PetscLogEventMPEActivate(a) 0 65 #define PetscLogEventMPEDeactivate(a) 0 66 #endif 67 68 EXTERN int (*_PetscLogPLB)(int,int,PetscObject,PetscObject,PetscObject,PetscObject); 69 EXTERN int (*_PetscLogPLE)(int,int,PetscObject,PetscObject,PetscObject,PetscObject); 70 EXTERN int (*_PetscLogPHC)(PetscObject); 71 EXTERN int (*_PetscLogPHD)(PetscObject); 72 73 #define PetscLogObjectParent(p,c) if (c) {PetscValidHeader((PetscObject)(c)); \ 74 PetscValidHeader((PetscObject)(p));\ 75 ((PetscObject)(c))->parent = (PetscObject)(p);\ 76 ((PetscObject)(c))->parentid = ((PetscObject)p)->id;} 77 #define PetscLogObjectParents(p,n,d) {int _i; for (_i=0; _i<n; _i++) \ 78 PetscLogObjectParent(p,(d)[_i]);} 79 #define PetscLogObjectCreate(h) {if (_PetscLogPHC) (*_PetscLogPHC)((PetscObject)h);} 80 #define PetscLogObjectDestroy(h) {if (_PetscLogPHD) (*_PetscLogPHD)((PetscObject)h);} 81 #define PetscLogObjectMemory(p,m) {PetscValidHeader((PetscObject)p);\ 82 ((PetscObject)(p))->mem += (m);} 83 /* Initialization functions */ 84 EXTERN int PetscLogBegin(void); 85 EXTERN int PetscLogAllBegin(void); 86 EXTERN int PetscLogTraceBegin(FILE *); 87 /* General functions */ 88 EXTERN int PetscLogDestroy(void); 89 EXTERN int PetscLogSet(int (*)(int, int, PetscObject, PetscObject, PetscObject, PetscObject), 90 int (*)(int, int, PetscObject, PetscObject, PetscObject, PetscObject)); 91 EXTERN int PetscLogObjectState(PetscObject, const char[], ...); 92 /* Output functions */ 93 EXTERN int PetscLogPrintSummary(MPI_Comm, const char[]); 94 EXTERN int PetscLogDump(const char[]); 95 /* Counter functions */ 96 EXTERN int PetscGetFlops(PetscLogDouble *); 97 /* Stage functions */ 98 EXTERN int PetscLogStageRegister(int *, const char[]); 99 EXTERN int PetscLogStagePush(int); 100 EXTERN int PetscLogStagePop(void); 101 EXTERN int PetscLogStageSetVisible(int, PetscTruth); 102 EXTERN int PetscLogStageGetVisible(int, PetscTruth *); 103 EXTERN int PetscLogStageGetId(const char [], int *); 104 /* Event functions */ 105 EXTERN int PetscLogEventRegister(int *, const char[], const char[], int); 106 EXTERN int PetscLogEventActivate(int); 107 EXTERN int PetscLogEventDeactivate(int); 108 EXTERN int PetscLogEventActivateClass(int); 109 EXTERN int PetscLogEventDeactivateClass(int); 110 /* Class functions */ 111 EXTERN int PetscLogClassRegister(int *, const char []); 112 113 /* Default log */ 114 typedef struct _StageLog *StageLog; 115 extern StageLog _stageLog; 116 117 /* Global counters */ 118 extern PetscLogDouble irecv_ct, isend_ct, recv_ct, send_ct; 119 extern PetscLogDouble irecv_len, isend_len, recv_len, send_len; 120 extern PetscLogDouble allreduce_ct; 121 extern PetscLogDouble wait_ct, wait_any_ct, wait_all_ct, sum_of_waits_ct; 122 extern int PETSC_DUMMY, PETSC_DUMMY_SIZE; 123 124 /* We must make these structures available if we are to access the event 125 activation flags in the PetscLogEventBegin/End() macros. If we forced a 126 function call each time, we could leave these structures in plog.h 127 */ 128 /* The structure for logging performance */ 129 typedef struct _PerfInfo { 130 char *name; /* The name of this section */ 131 char *color; /* The color of this section */ 132 int id; /* The integer identifying this event */ 133 int cookie; /* The class id for this section */ 134 int depth; /* The nesting depth of the event call */ 135 int count; /* The number of times this section was executed */ 136 PetscLogDouble flops; /* The flops used in this section */ 137 PetscLogDouble time; /* The time taken for this section */ 138 PetscLogDouble numMessages; /* The number of messages in this section */ 139 PetscLogDouble messageLength; /* The total message lengths in this section */ 140 PetscLogDouble numReductions; /* The number of reductions in this section */ 141 } PerfInfo; 142 143 /* The structure for logging events */ 144 typedef struct _EventLog *EventLog; 145 struct _EventLog { 146 /* Size information */ 147 int numEvents; /* The number of registered events */ 148 int maxEvents; /* The maximum number of events */ 149 /* Event specific information */ 150 int *eventActive; /* The flag for active events */ 151 PerfInfo *eventInfo; /* The performance information for each event */ 152 }; 153 154 /* The structure for logging class information */ 155 typedef struct _ClassInfo { 156 char *name; /* The class name */ 157 int cookie; /* The integer identifying this class */ 158 int creations; /* The number of objects of this class created */ 159 int destructions; /* The number of objects of this class destroyed */ 160 PetscLogDouble mem; /* The total memory allocated by objects of this class */ 161 PetscLogDouble descMem; /* The total memory allocated by descendents of these objects */ 162 } ClassInfo; 163 164 typedef struct _ClassLog *ClassLog; 165 struct _ClassLog { 166 int numClasses; /* The number of classes registered */ 167 int maxClasses; /* The maximum number of classes */ 168 ClassInfo *classInfo; /* The structure for classs information (cookies are monotonicly increasing) */ 169 }; 170 171 /* A simple stack (should replace) */ 172 typedef struct _IntStack *IntStack; 173 174 /* The structure for logging in stages */ 175 struct _StageLog { 176 /* Size information */ 177 int numStages; /* The number of registered stages */ 178 int maxStages; /* The maximum number of stages */ 179 /* Runtime information */ 180 int *stageUsed; /* The flags for stages which were executed */ 181 PetscTruth *stageVisible; /* The flags for stages which are visible to PetscLogPrintSummary() */ 182 IntStack stack; /* The stack for active stages */ 183 int curStage; /* The current stage (only used in macros so we don't call StackTop) */ 184 /* Stage specific information */ 185 PerfInfo *stageInfo; /* The performance information for each stage */ 186 EventLog *eventLog; /* The event log for each stage */ 187 ClassLog *classLog; /* The class information for each stage */ 188 }; 189 190 EXTERN int EventLogGetEvent(EventLog, int, int *); 191 192 #if defined(PETSC_HAVE_MPE) 193 #define PetscLogEventBarrierBegin(e,o1,o2,o3,o4,cm) \ 194 0; { int _1_ierr, _1_eventNum; \ 195 _1_ierr = EventLogGetEvent(_stageLog->eventLog[_stageLog->curStage], (e), &_1_eventNum); \ 196 if (_PetscLogPLB && _stageLog->eventLog[_stageLog->curStage]->eventActive[_1_eventNum]) { \ 197 _1_ierr = PetscLogEventBegin((e),o1,o2,o3,o4);CHKERRQ(_1_ierr); \ 198 if (UseMPE && PetscLogEventMPEFlags[(e)]) \ 199 MPE_Log_event(MPEBEGIN+2*(e),0,""); \ 200 _1_ierr = MPI_Barrier(cm);CHKERRQ(_1_ierr); \ 201 _1_ierr = PetscLogEventEnd((e),o1,o2,o3,o4);CHKERRQ(_1_ierr); \ 202 if (UseMPE && PetscLogEventMPEFlags[(e)]) \ 203 MPE_Log_event(MPEBEGIN+2*((e)+1),0,""); \ 204 } \ 205 _1_ierr = PetscLogEventBegin(e+1,o1,o2,o3,o4);CHKERRQ(_1_ierr); \ 206 if (UseMPE && PetscLogEventMPEFlags[(e)+1]) \ 207 MPE_Log_event(MPEBEGIN+2*((e)+1),0,""); \ 208 } 209 #define PetscLogEventBegin(e,o1,o2,o3,o4) \ 210 0; { int _3_ierr, _3_eventNum; \ 211 _3_ierr = EventLogGetEvent(_stageLog->eventLog[_stageLog->curStage], (e), &_3_eventNum); \ 212 if (_PetscLogPLB && _stageLog->eventLog[_stageLog->curStage]->eventActive[_3_eventNum]) {\ 213 (*_PetscLogPLB)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4));}\ 214 if (UseMPE && PetscLogEventMPEFlags[(e)])\ 215 MPE_Log_event(MPEBEGIN+2*(e),0,"");\ 216 } 217 #else 218 #define PetscLogEventBarrierBegin(e,o1,o2,o3,o4,cm) \ 219 0; { int _2_ierr, _2_eventNum;\ 220 _2_ierr = EventLogGetEvent(_stageLog->eventLog[_stageLog->curStage], (e), &_2_eventNum); \ 221 if (_PetscLogPLB && _stageLog->eventLog[_stageLog->curStage]->eventActive[_2_eventNum]) { \ 222 _2_ierr = PetscLogEventBegin((e),o1,o2,o3,o4);CHKERRQ(_2_ierr); \ 223 _2_ierr = MPI_Barrier(cm);CHKERRQ(_2_ierr); \ 224 _2_ierr = PetscLogEventEnd((e),o1,o2,o3,o4);CHKERRQ(_2_ierr); \ 225 } \ 226 _2_ierr = PetscLogEventBegin((e)+1,o1,o2,o3,o4);CHKERRQ(_2_ierr); \ 227 } 228 #define PetscLogEventBegin(e,o1,o2,o3,o4) \ 229 0; { int _4_ierr, _4_eventNum; \ 230 _4_ierr = EventLogGetEvent(_stageLog->eventLog[_stageLog->curStage], (e), &_4_eventNum); \ 231 if (_PetscLogPLB && _stageLog->eventLog[_stageLog->curStage]->eventActive[_4_eventNum]) {\ 232 (*_PetscLogPLB)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4));}\ 233 } 234 #endif 235 236 #if defined(PETSC_HAVE_MPE) 237 #define PetscLogEventBarrierEnd(e,o1,o2,o3,o4,cm) PetscLogEventEnd(e+1,o1,o2,o3,o4) 238 #define PetscLogEventEnd(e,o1,o2,o3,o4) \ 239 0; { int _5_ierr, _5_eventNum; \ 240 _5_ierr = EventLogGetEvent(_stageLog->eventLog[_stageLog->curStage], (e), &_5_eventNum); \ 241 if (_PetscLogPLE && _stageLog->eventLog[_stageLog->curStage]->eventActive[_5_eventNum]) {\ 242 (*_PetscLogPLE)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4));}\ 243 if (UseMPE && PetscLogEventMPEFlags[(e)])\ 244 MPE_Log_event(MPEBEGIN+2*(e)+1,0,"");\ 245 } 246 #else 247 #define PetscLogEventBarrierEnd(e,o1,o2,o3,o4,cm) PetscLogEventEnd(e+1,o1,o2,o3,o4) 248 #define PetscLogEventEnd(e,o1,o2,o3,o4) \ 249 0; { int _6_ierr, _6_eventNum; \ 250 _6_ierr = EventLogGetEvent(_stageLog->eventLog[_stageLog->curStage], (e), &_6_eventNum); \ 251 if (_PetscLogPLE && _stageLog->eventLog[_stageLog->curStage]->eventActive[_6_eventNum]) {\ 252 (*_PetscLogPLE)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4));}\ 253 } 254 #endif 255 256 /* 257 This does not work for MPI-Uni because our src/mpiuni/mpi.h file 258 uses macros to defined the MPI operations. 259 260 It does not work correctly from HP-UX because it processes the 261 macros in a way that sometimes it double counts, hence 262 PETSC_HAVE_BROKEN_RECURSIVE_MACRO 263 264 It does not work with Windows NT because winmpich lacks MPI_Type_size() 265 */ 266 #if !defined(HAVE_MPI_UNI) && !defined(PETSC_HAVE_BROKEN_RECURSIVE_MACRO) && !defined (PETSC_HAVE_MPI_MISSING_TYPESIZE) 267 /* 268 Logging of MPI activities 269 */ 270 271 #define TypeSize(buff,count,type) \ 272 (\ 273 MPI_Type_size(type,&PETSC_DUMMY_SIZE),buff += ((PetscLogDouble) ((count)*PETSC_DUMMY_SIZE)) \ 274 ) 275 276 #define MPI_Irecv(buf,count, datatype,source,tag,comm,request) \ 277 (\ 278 PETSC_DUMMY = MPI_Irecv(buf,count, datatype,source,tag,comm,request), \ 279 irecv_ct++,TypeSize(irecv_len,count,datatype),PETSC_DUMMY \ 280 ) 281 282 #define MPI_Isend(buf,count, datatype,dest,tag,comm,request) \ 283 (\ 284 PETSC_DUMMY = MPI_Isend(buf,count, datatype,dest,tag,comm,request), \ 285 isend_ct++, TypeSize(isend_len,count,datatype),PETSC_DUMMY \ 286 ) 287 288 #define MPI_Startall_irecv(count,number,requests) \ 289 (\ 290 PETSC_DUMMY = MPI_Startall(number,requests), \ 291 irecv_ct += (PetscLogDouble)(number),irecv_len += ((PetscLogDouble) ((count)*sizeof(PetscScalar))),PETSC_DUMMY \ 292 ) 293 294 #define MPI_Startall_isend(count,number,requests) \ 295 (\ 296 PETSC_DUMMY = MPI_Startall(number,requests), \ 297 isend_ct += (PetscLogDouble)(number),isend_len += ((PetscLogDouble) ((count)*sizeof(PetscScalar))),PETSC_DUMMY \ 298 ) 299 300 #define MPI_Start_isend(count, requests)\ 301 (\ 302 PETSC_DUMMY = MPI_Start(requests),\ 303 isend_ct++,isend_len += ((PetscLogDouble) ((count)*sizeof(PetscScalar))),PETSC_DUMMY\ 304 ) 305 306 #define MPI_Recv(buf,count, datatype,source,tag,comm,status) \ 307 (\ 308 PETSC_DUMMY = MPI_Recv(buf,count, datatype,source,tag,comm,status), \ 309 recv_ct++,TypeSize(recv_len,count,datatype),PETSC_DUMMY \ 310 ) 311 312 #define MPI_Send(buf,count, datatype,dest,tag,comm) \ 313 (\ 314 PETSC_DUMMY = MPI_Send(buf,count, datatype,dest,tag,comm), \ 315 send_ct++, TypeSize(send_len,count,datatype),PETSC_DUMMY \ 316 ) 317 318 #define MPI_Wait(request,status) \ 319 (\ 320 wait_ct++,sum_of_waits_ct++, \ 321 MPI_Wait(request,status) \ 322 ) 323 324 #define MPI_Waitany(a,b,c,d) \ 325 (\ 326 wait_any_ct++,sum_of_waits_ct++,\ 327 MPI_Waitany(a,b,c,d) \ 328 ) 329 330 #define MPI_Waitall(count,array_of_requests,array_of_statuses) \ 331 (\ 332 wait_all_ct++,sum_of_waits_ct += (PetscLogDouble) (count), \ 333 MPI_Waitall(count,array_of_requests,array_of_statuses) \ 334 ) 335 336 #define MPI_Allreduce(sendbuf, recvbuf,count,datatype,op,comm) \ 337 (\ 338 allreduce_ct++,MPI_Allreduce(sendbuf,recvbuf,count,datatype,op,comm)\ 339 ) 340 341 #else 342 343 #define MPI_Startall_irecv(count,number,requests) \ 344 (\ 345 MPI_Startall(number,requests) \ 346 ) 347 348 #define MPI_Startall_isend(count,number,requests) \ 349 (\ 350 MPI_Startall(number,requests) \ 351 ) 352 353 #define MPI_Start_isend(count, requests) \ 354 (\ 355 MPI_Start(requests) \ 356 ) 357 358 #endif /* !USING_MPIUNI && ! PETSC_HAVE_BROKEN_RECURSIVE_MACRO */ 359 360 #else /* ---Logging is turned off --------------------------------------------*/ 361 362 #define PetscLogFlops(n) 0 363 364 /* 365 With logging turned off, then MPE has to be turned off 366 */ 367 #define MPEBEGIN 1000 368 #define PetscLogMPEBegin() 0 369 #define PetscLogMPEDump(a) 0 370 #define PetscLogEventMPEActivate(a) 0 371 #define PetscLogEventMPEDeactivate(a) 0 372 373 #define PetscLogEventActivate(a) 0 374 #define PetscLogEventDeactivate(a) 0 375 376 #define PetscLogEventActivateClass(a) 0 377 #define PetscLogEventDeactivateClass(a) 0 378 379 #define _PetscLogPLB 0 380 #define _PetscLogPLE 0 381 #define _PetscLogPHC 0 382 #define _PetscLogPHD 0 383 #define PetscGetFlops(a) (*(a) = 0.0,0) 384 #define PetscLogEventBegin(e,o1,o2,o3,o4) 0 385 #define PetscLogEventEnd(e,o1,o2,o3,o4) 0 386 #define PetscLogEventBarrierBegin(e,o1,o2,o3,o4,cm) 0 387 #define PetscLogEventBarrierEnd(e,o1,o2,o3,o4,cm) 0 388 #define PetscLogObjectParent(p,c) 389 #define PetscLogObjectParents(p,n,c) 390 #define PetscLogObjectCreate(h) 391 #define PetscLogObjectDestroy(h) 392 #define PetscLogObjectMemory(p,m) 393 #define PetscLogDestroy() 0 394 #define PetscLogStagePush(a) 0 395 #define PetscLogStagePop() 0 396 #define PetscLogStageRegister(a,b) 0 397 #define PetscLogStagePrint(a,flg) 0 398 #define PetscLogPrintSummary(comm,file) 0 399 #define PetscLogBegin() 0 400 #define PetscLogTraceBegin(file) 0 401 #define PetscLogSet(lb,le) 0 402 #define PetscLogAllBegin() 0 403 #define PetscLogDump(c) 0 404 #define PetscLogEventRegister(a,b,c) 0 405 EXTERN int PetscLogObjectState(PetscObject,const char[],...); 406 407 /* If PETSC_USE_LOG is NOT defined, these still need to be! */ 408 #define MPI_Startall_irecv(count,number,requests) MPI_Startall(number,requests) 409 410 #define MPI_Startall_isend(count,number,requests) MPI_Startall(number,requests) 411 412 #define MPI_Start_isend(count,requests) MPI_Start(requests) 413 414 #endif /* PETSC_USE_LOG */ 415 416 extern PetscTruth PetscPreLoadingUsed; /* true if we are or have done preloading */ 417 extern PetscTruth PetscPreLoadingOn; /* true if we are currently in a preloading calculation */ 418 419 #define PreLoadBegin(flag,name) {PetscTruth PreLoading = flag; \ 420 int PreLoadMax,PreLoadIt,_stageNum,_3_ierr;\ 421 _3_ierr = PetscOptionsGetLogical(PETSC_NULL,"-preload",&PreLoading,PETSC_NULL);CHKERRQ(_3_ierr);\ 422 PreLoadMax = (int)(PreLoading);PetscPreLoadingUsed = PreLoading ? PETSC_TRUE : PetscPreLoadingUsed;\ 423 for (PreLoadIt=0; PreLoadIt<=PreLoadMax; PreLoadIt++) {\ 424 PetscPreLoadingOn = PreLoading;\ 425 _3_ierr = PetscBarrier(PETSC_NULL);CHKERRQ(_3_ierr);\ 426 if (PreLoadIt>0) {\ 427 _3_ierr = PetscLogStageGetId(name,&_stageNum);CHKERRQ(_3_ierr);\ 428 } else {\ 429 _3_ierr = PetscLogStageRegister(&_stageNum,name);CHKERRQ(_3_ierr);\ 430 _3_ierr = PetscLogStageSetVisible(_stageNum,(PetscTruth)(!PreLoadMax || PreLoadIt));\ 431 }\ 432 _3_ierr = PetscLogStagePush(_stageNum);CHKERRQ(_3_ierr); 433 434 #define PreLoadEnd() _3_ierr = PetscLogStagePop();CHKERRQ(_3_ierr);PreLoading = PETSC_FALSE;}} 435 436 #define PreLoadStage(name) _3_ierr = PetscLogStagePop();CHKERRQ(_3_ierr);\ 437 if (PreLoadIt>0) {\ 438 _3_ierr = PetscLogStageGetId(name,&_stageNum);CHKERRQ(_3_ierr);\ 439 } else {\ 440 _3_ierr = PetscLogStageRegister(&_stageNum,name);CHKERRQ(_3_ierr);\ 441 _3_ierr = PetscLogStageSetVisible(_stageNum,(PetscTruth)(!PreLoadMax || PreLoadIt));\ 442 }\ 443 _3_ierr = PetscLogStagePush(_stageNum);CHKERRQ(_3_ierr); 444 #endif 445