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 PetscTruth active; /* The flag to activate logging */ 135 PetscTruth visible; /* The flag to print info in summary */ 136 int depth; /* The nesting depth of the event call */ 137 int count; /* The number of times this section was executed */ 138 PetscLogDouble flops; /* The flops used in this section */ 139 PetscLogDouble time; /* The time taken for this section */ 140 PetscLogDouble numMessages; /* The number of messages in this section */ 141 PetscLogDouble messageLength; /* The total message lengths in this section */ 142 PetscLogDouble numReductions; /* The number of reductions in this section */ 143 } PerfInfo; 144 145 /* The structure for logging events */ 146 typedef struct _EventLog *EventLog; 147 struct _EventLog { 148 int numEvents; /* The number of registered events */ 149 int maxEvents; /* The maximum number of events */ 150 PerfInfo *eventInfo; /* The performance information for each event */ 151 }; 152 153 /* The structure for logging class information */ 154 typedef struct _ClassInfo { 155 char *name; /* The class name */ 156 int cookie; /* The integer identifying this class */ 157 int creations; /* The number of objects of this class created */ 158 int destructions; /* The number of objects of this class destroyed */ 159 PetscLogDouble mem; /* The total memory allocated by objects of this class */ 160 PetscLogDouble descMem; /* The total memory allocated by descendents of these objects */ 161 } ClassInfo; 162 163 typedef struct _ClassLog *ClassLog; 164 struct _ClassLog { 165 int numClasses; /* The number of classes registered */ 166 int maxClasses; /* The maximum number of classes */ 167 ClassInfo *classInfo; /* The structure for classs information (cookies are monotonicly increasing) */ 168 }; 169 170 /* A simple stack (should replace) */ 171 typedef struct _IntStack *IntStack; 172 173 /* The structure for logging in stages */ 174 struct _StageLog { 175 /* Size information */ 176 int numStages; /* The number of registered stages */ 177 int maxStages; /* The maximum number of stages */ 178 /* Runtime information */ 179 IntStack stack; /* The stack for active stages */ 180 int curStage; /* The current stage (only used in macros so we don't call StackTop) */ 181 /* Stage specific information */ 182 PerfInfo *stageInfo; /* The performance information for each stage */ 183 EventLog *eventLog; /* The event log for each stage */ 184 ClassLog *classLog; /* The class information for each stage */ 185 }; 186 187 EXTERN int EventLogGetEvent(EventLog, int, int *); 188 189 #if defined(PETSC_HAVE_MPE) 190 #define PetscLogEventBarrierBegin(e,o1,o2,o3,o4,cm) \ 191 0; { int _1_ierr, _1_eventNum; \ 192 _1_ierr = EventLogGetEvent(_stageLog->eventLog[_stageLog->curStage], (e), &_1_eventNum); \ 193 if (_PetscLogPLB && _stageLog->eventLog[_stageLog->curStage]->eventInfo[_1_eventNum].active) { \ 194 _1_ierr = PetscLogEventBegin((e),o1,o2,o3,o4);CHKERRQ(_1_ierr); \ 195 if (UseMPE && PetscLogEventMPEFlags[(e)]) \ 196 MPE_Log_event(MPEBEGIN+2*(e),0,""); \ 197 _1_ierr = MPI_Barrier(cm);CHKERRQ(_1_ierr); \ 198 _1_ierr = PetscLogEventEnd((e),o1,o2,o3,o4);CHKERRQ(_1_ierr); \ 199 if (UseMPE && PetscLogEventMPEFlags[(e)]) \ 200 MPE_Log_event(MPEBEGIN+2*((e)+1),0,""); \ 201 } \ 202 _1_ierr = PetscLogEventBegin(e+1,o1,o2,o3,o4);CHKERRQ(_1_ierr); \ 203 if (UseMPE && PetscLogEventMPEFlags[(e)+1]) \ 204 MPE_Log_event(MPEBEGIN+2*((e)+1),0,""); \ 205 } 206 #define PetscLogEventBegin(e,o1,o2,o3,o4) \ 207 0; { int _3_ierr, _3_eventNum; \ 208 _3_ierr = EventLogGetEvent(_stageLog->eventLog[_stageLog->curStage], (e), &_3_eventNum); \ 209 if (_PetscLogPLB && _stageLog->eventLog[_stageLog->curStage]->eventInfo[_3_eventNum].active) {\ 210 (*_PetscLogPLB)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4));}\ 211 if (UseMPE && PetscLogEventMPEFlags[(e)])\ 212 MPE_Log_event(MPEBEGIN+2*(e),0,"");\ 213 } 214 #else 215 #define PetscLogEventBarrierBegin(e,o1,o2,o3,o4,cm) \ 216 0; { int _2_ierr, _2_eventNum;\ 217 _2_ierr = EventLogGetEvent(_stageLog->eventLog[_stageLog->curStage], (e), &_2_eventNum); \ 218 if (_PetscLogPLB && _stageLog->eventLog[_stageLog->curStage]->eventInfo[_2_eventNum].active) { \ 219 _2_ierr = PetscLogEventBegin((e),o1,o2,o3,o4);CHKERRQ(_2_ierr); \ 220 _2_ierr = MPI_Barrier(cm);CHKERRQ(_2_ierr); \ 221 _2_ierr = PetscLogEventEnd((e),o1,o2,o3,o4);CHKERRQ(_2_ierr); \ 222 } \ 223 _2_ierr = PetscLogEventBegin((e)+1,o1,o2,o3,o4);CHKERRQ(_2_ierr); \ 224 } 225 #define PetscLogEventBegin(e,o1,o2,o3,o4) \ 226 0; { int _4_ierr, _4_eventNum; \ 227 _4_ierr = EventLogGetEvent(_stageLog->eventLog[_stageLog->curStage], (e), &_4_eventNum); \ 228 if (_PetscLogPLB && _stageLog->eventLog[_stageLog->curStage]->eventInfo[_4_eventNum].active) {\ 229 (*_PetscLogPLB)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4));}\ 230 } 231 #endif 232 233 #if defined(PETSC_HAVE_MPE) 234 #define PetscLogEventBarrierEnd(e,o1,o2,o3,o4,cm) PetscLogEventEnd(e+1,o1,o2,o3,o4) 235 #define PetscLogEventEnd(e,o1,o2,o3,o4) \ 236 0; { int _5_ierr, _5_eventNum; \ 237 _5_ierr = EventLogGetEvent(_stageLog->eventLog[_stageLog->curStage], (e), &_5_eventNum); \ 238 if (_PetscLogPLE && _stageLog->eventLog[_stageLog->curStage]->eventInfo[_5_eventNum].active) {\ 239 (*_PetscLogPLE)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4));}\ 240 if (UseMPE && PetscLogEventMPEFlags[(e)])\ 241 MPE_Log_event(MPEBEGIN+2*(e)+1,0,"");\ 242 } 243 #else 244 #define PetscLogEventBarrierEnd(e,o1,o2,o3,o4,cm) PetscLogEventEnd(e+1,o1,o2,o3,o4) 245 #define PetscLogEventEnd(e,o1,o2,o3,o4) \ 246 0; { int _6_ierr, _6_eventNum; \ 247 _6_ierr = EventLogGetEvent(_stageLog->eventLog[_stageLog->curStage], (e), &_6_eventNum); \ 248 if (_PetscLogPLE && _stageLog->eventLog[_stageLog->curStage]->eventInfo[_6_eventNum].active) {\ 249 (*_PetscLogPLE)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4));}\ 250 } 251 #endif 252 253 /* 254 This does not work for MPI-Uni because our src/mpiuni/mpi.h file 255 uses macros to defined the MPI operations. 256 257 It does not work correctly from HP-UX because it processes the 258 macros in a way that sometimes it double counts, hence 259 PETSC_HAVE_BROKEN_RECURSIVE_MACRO 260 261 It does not work with Windows NT because winmpich lacks MPI_Type_size() 262 */ 263 #if !defined(HAVE_MPI_UNI) && !defined(PETSC_HAVE_BROKEN_RECURSIVE_MACRO) && !defined (PETSC_HAVE_MPI_MISSING_TYPESIZE) 264 /* 265 Logging of MPI activities 266 */ 267 268 #define TypeSize(buff,count,type) \ 269 (\ 270 MPI_Type_size(type,&PETSC_DUMMY_SIZE),buff += ((PetscLogDouble) ((count)*PETSC_DUMMY_SIZE)) \ 271 ) 272 273 #define MPI_Irecv(buf,count, datatype,source,tag,comm,request) \ 274 (\ 275 PETSC_DUMMY = MPI_Irecv(buf,count, datatype,source,tag,comm,request), \ 276 irecv_ct++,TypeSize(irecv_len,count,datatype),PETSC_DUMMY \ 277 ) 278 279 #define MPI_Isend(buf,count, datatype,dest,tag,comm,request) \ 280 (\ 281 PETSC_DUMMY = MPI_Isend(buf,count, datatype,dest,tag,comm,request), \ 282 isend_ct++, TypeSize(isend_len,count,datatype),PETSC_DUMMY \ 283 ) 284 285 #define MPI_Startall_irecv(count,number,requests) \ 286 (\ 287 PETSC_DUMMY = MPI_Startall(number,requests), \ 288 irecv_ct += (PetscLogDouble)(number),irecv_len += ((PetscLogDouble) ((count)*sizeof(PetscScalar))),PETSC_DUMMY \ 289 ) 290 291 #define MPI_Startall_isend(count,number,requests) \ 292 (\ 293 PETSC_DUMMY = MPI_Startall(number,requests), \ 294 isend_ct += (PetscLogDouble)(number),isend_len += ((PetscLogDouble) ((count)*sizeof(PetscScalar))),PETSC_DUMMY \ 295 ) 296 297 #define MPI_Start_isend(count, requests)\ 298 (\ 299 PETSC_DUMMY = MPI_Start(requests),\ 300 isend_ct++,isend_len += ((PetscLogDouble) ((count)*sizeof(PetscScalar))),PETSC_DUMMY\ 301 ) 302 303 #define MPI_Recv(buf,count, datatype,source,tag,comm,status) \ 304 (\ 305 PETSC_DUMMY = MPI_Recv(buf,count, datatype,source,tag,comm,status), \ 306 recv_ct++,TypeSize(recv_len,count,datatype),PETSC_DUMMY \ 307 ) 308 309 #define MPI_Send(buf,count, datatype,dest,tag,comm) \ 310 (\ 311 PETSC_DUMMY = MPI_Send(buf,count, datatype,dest,tag,comm), \ 312 send_ct++, TypeSize(send_len,count,datatype),PETSC_DUMMY \ 313 ) 314 315 #define MPI_Wait(request,status) \ 316 (\ 317 wait_ct++,sum_of_waits_ct++, \ 318 MPI_Wait(request,status) \ 319 ) 320 321 #define MPI_Waitany(a,b,c,d) \ 322 (\ 323 wait_any_ct++,sum_of_waits_ct++,\ 324 MPI_Waitany(a,b,c,d) \ 325 ) 326 327 #define MPI_Waitall(count,array_of_requests,array_of_statuses) \ 328 (\ 329 wait_all_ct++,sum_of_waits_ct += (PetscLogDouble) (count), \ 330 MPI_Waitall(count,array_of_requests,array_of_statuses) \ 331 ) 332 333 #define MPI_Allreduce(sendbuf, recvbuf,count,datatype,op,comm) \ 334 (\ 335 allreduce_ct++,MPI_Allreduce(sendbuf,recvbuf,count,datatype,op,comm)\ 336 ) 337 338 #else 339 340 #define MPI_Startall_irecv(count,number,requests) \ 341 (\ 342 MPI_Startall(number,requests) \ 343 ) 344 345 #define MPI_Startall_isend(count,number,requests) \ 346 (\ 347 MPI_Startall(number,requests) \ 348 ) 349 350 #define MPI_Start_isend(count, requests) \ 351 (\ 352 MPI_Start(requests) \ 353 ) 354 355 #endif /* !USING_MPIUNI && ! PETSC_HAVE_BROKEN_RECURSIVE_MACRO */ 356 357 #else /* ---Logging is turned off --------------------------------------------*/ 358 359 #define PetscLogFlops(n) 0 360 361 /* 362 With logging turned off, then MPE has to be turned off 363 */ 364 #define MPEBEGIN 1000 365 #define PetscLogMPEBegin() 0 366 #define PetscLogMPEDump(a) 0 367 #define PetscLogEventMPEActivate(a) 0 368 #define PetscLogEventMPEDeactivate(a) 0 369 370 #define PetscLogEventActivate(a) 0 371 #define PetscLogEventDeactivate(a) 0 372 373 #define PetscLogEventActivateClass(a) 0 374 #define PetscLogEventDeactivateClass(a) 0 375 376 #define _PetscLogPLB 0 377 #define _PetscLogPLE 0 378 #define _PetscLogPHC 0 379 #define _PetscLogPHD 0 380 #define PetscGetFlops(a) (*(a) = 0.0,0) 381 #define PetscLogEventBegin(e,o1,o2,o3,o4) 0 382 #define PetscLogEventEnd(e,o1,o2,o3,o4) 0 383 #define PetscLogEventBarrierBegin(e,o1,o2,o3,o4,cm) 0 384 #define PetscLogEventBarrierEnd(e,o1,o2,o3,o4,cm) 0 385 #define PetscLogObjectParent(p,c) 386 #define PetscLogObjectParents(p,n,c) 387 #define PetscLogObjectCreate(h) 388 #define PetscLogObjectDestroy(h) 389 #define PetscLogObjectMemory(p,m) 390 #define PetscLogDestroy() 0 391 #define PetscLogStagePush(a) 0 392 #define PetscLogStagePop() 0 393 #define PetscLogStageRegister(a,b) 0 394 #define PetscLogStagePrint(a,flg) 0 395 #define PetscLogPrintSummary(comm,file) 0 396 #define PetscLogBegin() 0 397 #define PetscLogTraceBegin(file) 0 398 #define PetscLogSet(lb,le) 0 399 #define PetscLogAllBegin() 0 400 #define PetscLogDump(c) 0 401 #define PetscLogEventRegister(a,b,c) 0 402 EXTERN int PetscLogObjectState(PetscObject,const char[],...); 403 404 /* If PETSC_USE_LOG is NOT defined, these still need to be! */ 405 #define MPI_Startall_irecv(count,number,requests) MPI_Startall(number,requests) 406 407 #define MPI_Startall_isend(count,number,requests) MPI_Startall(number,requests) 408 409 #define MPI_Start_isend(count,requests) MPI_Start(requests) 410 411 #endif /* PETSC_USE_LOG */ 412 413 extern PetscTruth PetscPreLoadingUsed; /* true if we are or have done preloading */ 414 extern PetscTruth PetscPreLoadingOn; /* true if we are currently in a preloading calculation */ 415 416 #define PreLoadBegin(flag,name) {PetscTruth PreLoading = flag; \ 417 int PreLoadMax,PreLoadIt,_stageNum,_3_ierr;\ 418 _3_ierr = PetscOptionsGetLogical(PETSC_NULL,"-preload",&PreLoading,PETSC_NULL);CHKERRQ(_3_ierr);\ 419 PreLoadMax = (int)(PreLoading);PetscPreLoadingUsed = PreLoading ? PETSC_TRUE : PetscPreLoadingUsed;\ 420 for (PreLoadIt=0; PreLoadIt<=PreLoadMax; PreLoadIt++) {\ 421 PetscPreLoadingOn = PreLoading;\ 422 _3_ierr = PetscBarrier(PETSC_NULL);CHKERRQ(_3_ierr);\ 423 if (PreLoadIt>0) {\ 424 _3_ierr = PetscLogStageGetId(name,&_stageNum);CHKERRQ(_3_ierr);\ 425 } else {\ 426 _3_ierr = PetscLogStageRegister(&_stageNum,name);CHKERRQ(_3_ierr);\ 427 _3_ierr = PetscLogStageSetVisible(_stageNum,(PetscTruth)(!PreLoadMax || PreLoadIt));\ 428 }\ 429 _3_ierr = PetscLogStagePush(_stageNum);CHKERRQ(_3_ierr); 430 431 #define PreLoadEnd() _3_ierr = PetscLogStagePop();CHKERRQ(_3_ierr);PreLoading = PETSC_FALSE;}} 432 433 #define PreLoadStage(name) _3_ierr = PetscLogStagePop();CHKERRQ(_3_ierr);\ 434 if (PreLoadIt>0) {\ 435 _3_ierr = PetscLogStageGetId(name,&_stageNum);CHKERRQ(_3_ierr);\ 436 } else {\ 437 _3_ierr = PetscLogStageRegister(&_stageNum,name);CHKERRQ(_3_ierr);\ 438 _3_ierr = PetscLogStageSetVisible(_stageNum,(PetscTruth)(!PreLoadMax || PreLoadIt));\ 439 }\ 440 _3_ierr = PetscLogStagePush(_stageNum);CHKERRQ(_3_ierr); 441 #endif 442