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[],...) PETSC_PRINTF_FORMAT_CHECK(2,3); 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[], ...) PETSC_PRINTF_FORMAT_CHECK(2,3); 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 typedef int PetscEvent; 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 IntStack stack; /* The stack for active stages */ 181 int curStage; /* The current stage (only used in macros so we don't call StackTop) */ 182 /* Stage specific information */ 183 PerfInfo *stageInfo; /* The performance information for each stage */ 184 EventLog *eventLog; /* The event log for each stage */ 185 ClassLog *classLog; /* The class information for each stage */ 186 }; 187 188 #if defined(PETSC_HAVE_MPE) 189 #define PetscLogEventBarrierBegin(e,o1,o2,o3,o4,cm) \ 190 0; { int _1_ierr; \ 191 if (_PetscLogPLB && _stageLog->eventLog[_stageLog->curStage]->eventInfo[e].active) { \ 192 _1_ierr = PetscLogEventBegin((e),o1,o2,o3,o4);CHKERRQ(_1_ierr); \ 193 if (UseMPE && PetscLogEventMPEFlags[(e)]) \ 194 MPE_Log_event(MPEBEGIN+2*(e),0,""); \ 195 _1_ierr = MPI_Barrier(cm);CHKERRQ(_1_ierr); \ 196 _1_ierr = PetscLogEventEnd((e),o1,o2,o3,o4);CHKERRQ(_1_ierr); \ 197 if (UseMPE && PetscLogEventMPEFlags[(e)]) \ 198 MPE_Log_event(MPEBEGIN+2*((e)+1),0,""); \ 199 } \ 200 _1_ierr = PetscLogEventBegin(e+1,o1,o2,o3,o4);CHKERRQ(_1_ierr); \ 201 if (UseMPE && PetscLogEventMPEFlags[(e)+1]) \ 202 MPE_Log_event(MPEBEGIN+2*((e)+1),0,""); \ 203 } 204 #define PetscLogEventBegin(e,o1,o2,o3,o4) \ 205 0; { \ 206 if (_PetscLogPLB && _stageLog->eventLog[_stageLog->curStage]->eventInfo[e].active) {\ 207 (*_PetscLogPLB)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4));}\ 208 if (UseMPE && PetscLogEventMPEFlags[(e)])\ 209 MPE_Log_event(MPEBEGIN+2*(e),0,"");\ 210 } 211 #else 212 #define PetscLogEventBarrierBegin(e,o1,o2,o3,o4,cm) \ 213 0; { int _2_ierr;\ 214 if (_PetscLogPLB && _stageLog->eventLog[_stageLog->curStage]->eventInfo[e].active) { \ 215 _2_ierr = PetscLogEventBegin((e),o1,o2,o3,o4);CHKERRQ(_2_ierr); \ 216 _2_ierr = MPI_Barrier(cm);CHKERRQ(_2_ierr); \ 217 _2_ierr = PetscLogEventEnd((e),o1,o2,o3,o4);CHKERRQ(_2_ierr); \ 218 } \ 219 _2_ierr = PetscLogEventBegin((e)+1,o1,o2,o3,o4);CHKERRQ(_2_ierr); \ 220 } 221 #define PetscLogEventBegin(e,o1,o2,o3,o4) \ 222 0; { \ 223 if (_PetscLogPLB && _stageLog->eventLog[_stageLog->curStage]->eventInfo[e].active) {\ 224 (*_PetscLogPLB)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4));}\ 225 } 226 #endif 227 228 #if defined(PETSC_HAVE_MPE) 229 #define PetscLogEventBarrierEnd(e,o1,o2,o3,o4,cm) PetscLogEventEnd(e+1,o1,o2,o3,o4) 230 #define PetscLogEventEnd(e,o1,o2,o3,o4) \ 231 0; { \ 232 if (_PetscLogPLE && _stageLog->eventLog[_stageLog->curStage]->eventInfo[e].active) {\ 233 (*_PetscLogPLE)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4));}\ 234 if (UseMPE && PetscLogEventMPEFlags[(e)])\ 235 MPE_Log_event(MPEBEGIN+2*(e)+1,0,"");\ 236 } 237 #else 238 #define PetscLogEventBarrierEnd(e,o1,o2,o3,o4,cm) PetscLogEventEnd(e+1,o1,o2,o3,o4) 239 #define PetscLogEventEnd(e,o1,o2,o3,o4) \ 240 0; { \ 241 if (_PetscLogPLE && _stageLog->eventLog[_stageLog->curStage]->eventInfo[e].active) {\ 242 (*_PetscLogPLE)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4));}\ 243 } 244 #endif 245 246 /* 247 This does not work for MPI-Uni because our src/mpiuni/mpi.h file 248 uses macros to defined the MPI operations. 249 250 It does not work correctly from HP-UX because it processes the 251 macros in a way that sometimes it double counts, hence 252 HAVE_BROKEN_RECURSIVE_MACRO 253 254 It does not work with Windows NT because winmpich lacks MPI_Type_size() 255 */ 256 #if !defined(PETSC_HAVE_MPI_UNI) && !defined(HAVE_BROKEN_RECURSIVE_MACRO) && !defined (PETSC_HAVE_MPI_MISSING_TYPESIZE) 257 /* 258 Logging of MPI activities 259 */ 260 261 #define TypeSize(buff,count,type) \ 262 (\ 263 MPI_Type_size(type,&PETSC_DUMMY_SIZE),buff += ((PetscLogDouble) ((count)*PETSC_DUMMY_SIZE)) \ 264 ) 265 266 #define MPI_Irecv(buf,count, datatype,source,tag,comm,request) \ 267 (\ 268 PETSC_DUMMY = MPI_Irecv(buf,count, datatype,source,tag,comm,request), \ 269 irecv_ct++,TypeSize(irecv_len,count,datatype),PETSC_DUMMY \ 270 ) 271 272 #define MPI_Isend(buf,count, datatype,dest,tag,comm,request) \ 273 (\ 274 PETSC_DUMMY = MPI_Isend(buf,count, datatype,dest,tag,comm,request), \ 275 isend_ct++, TypeSize(isend_len,count,datatype),PETSC_DUMMY \ 276 ) 277 278 #define MPI_Startall_irecv(count,number,requests) \ 279 (\ 280 PETSC_DUMMY = MPI_Startall(number,requests), \ 281 irecv_ct += (PetscLogDouble)(number),irecv_len += ((PetscLogDouble) ((count)*sizeof(PetscScalar))),PETSC_DUMMY \ 282 ) 283 284 #define MPI_Startall_isend(count,number,requests) \ 285 (\ 286 PETSC_DUMMY = MPI_Startall(number,requests), \ 287 isend_ct += (PetscLogDouble)(number),isend_len += ((PetscLogDouble) ((count)*sizeof(PetscScalar))),PETSC_DUMMY \ 288 ) 289 290 #define MPI_Start_isend(count, requests)\ 291 (\ 292 PETSC_DUMMY = MPI_Start(requests),\ 293 isend_ct++,isend_len += ((PetscLogDouble) ((count)*sizeof(PetscScalar))),PETSC_DUMMY\ 294 ) 295 296 #define MPI_Recv(buf,count, datatype,source,tag,comm,status) \ 297 (\ 298 PETSC_DUMMY = MPI_Recv(buf,count, datatype,source,tag,comm,status), \ 299 recv_ct++,TypeSize(recv_len,count,datatype),PETSC_DUMMY \ 300 ) 301 302 #define MPI_Send(buf,count, datatype,dest,tag,comm) \ 303 (\ 304 PETSC_DUMMY = MPI_Send(buf,count, datatype,dest,tag,comm), \ 305 send_ct++, TypeSize(send_len,count,datatype),PETSC_DUMMY \ 306 ) 307 308 #define MPI_Wait(request,status) \ 309 (\ 310 wait_ct++,sum_of_waits_ct++, \ 311 MPI_Wait(request,status) \ 312 ) 313 314 #define MPI_Waitany(a,b,c,d) \ 315 (\ 316 wait_any_ct++,sum_of_waits_ct++,\ 317 MPI_Waitany(a,b,c,d) \ 318 ) 319 320 #define MPI_Waitall(count,array_of_requests,array_of_statuses) \ 321 (\ 322 wait_all_ct++,sum_of_waits_ct += (PetscLogDouble) (count), \ 323 MPI_Waitall(count,array_of_requests,array_of_statuses) \ 324 ) 325 326 #define MPI_Allreduce(sendbuf, recvbuf,count,datatype,op,comm) \ 327 (\ 328 allreduce_ct++,MPI_Allreduce(sendbuf,recvbuf,count,datatype,op,comm)\ 329 ) 330 331 #else 332 333 #define MPI_Startall_irecv(count,number,requests) \ 334 (\ 335 MPI_Startall(number,requests) \ 336 ) 337 338 #define MPI_Startall_isend(count,number,requests) \ 339 (\ 340 MPI_Startall(number,requests) \ 341 ) 342 343 #define MPI_Start_isend(count, requests) \ 344 (\ 345 MPI_Start(requests) \ 346 ) 347 348 #endif /* !PETSC_HAVE_MPI_UNI && ! HAVE_BROKEN_RECURSIVE_MACRO */ 349 350 #else /* ---Logging is turned off --------------------------------------------*/ 351 352 #define PetscLogFlops(n) 0 353 354 /* 355 With logging turned off, then MPE has to be turned off 356 */ 357 #define MPEBEGIN 1000 358 #define PetscLogMPEBegin() 0 359 #define PetscLogMPEDump(a) 0 360 #define PetscLogEventMPEActivate(a) 0 361 #define PetscLogEventMPEDeactivate(a) 0 362 363 #define PetscLogEventActivate(a) 0 364 #define PetscLogEventDeactivate(a) 0 365 366 #define PetscLogEventActivateClass(a) 0 367 #define PetscLogEventDeactivateClass(a) 0 368 369 #define _PetscLogPLB 0 370 #define _PetscLogPLE 0 371 #define _PetscLogPHC 0 372 #define _PetscLogPHD 0 373 #define PetscGetFlops(a) (*(a) = 0.0,0) 374 #define PetscLogEventBegin(e,o1,o2,o3,o4) 0 375 #define PetscLogEventEnd(e,o1,o2,o3,o4) 0 376 #define PetscLogEventBarrierBegin(e,o1,o2,o3,o4,cm) 0 377 #define PetscLogEventBarrierEnd(e,o1,o2,o3,o4,cm) 0 378 #define PetscLogObjectParent(p,c) 379 #define PetscLogObjectParents(p,n,c) 380 #define PetscLogObjectCreate(h) 381 #define PetscLogObjectDestroy(h) 382 #define PetscLogObjectMemory(p,m) 383 #define PetscLogDestroy() 0 384 #define PetscLogStagePush(a) 0 385 #define PetscLogStagePop() 0 386 #define PetscLogStageRegister(a,b) 0 387 #define PetscLogStagePrint(a,flg) 0 388 #define PetscLogPrintSummary(comm,file) 0 389 #define PetscLogBegin() 0 390 #define PetscLogTraceBegin(file) 0 391 #define PetscLogSet(lb,le) 0 392 #define PetscLogAllBegin() 0 393 #define PetscLogDump(c) 0 394 #define PetscLogEventRegister(a,b,c,d) 0 395 EXTERN int PetscLogObjectState(PetscObject,const char[],...) PETSC_PRINTF_FORMAT_CHECK(2,3); 396 397 /* If PETSC_USE_LOG is NOT defined, these still need to be! */ 398 #define MPI_Startall_irecv(count,number,requests) MPI_Startall(number,requests) 399 400 #define MPI_Startall_isend(count,number,requests) MPI_Startall(number,requests) 401 402 #define MPI_Start_isend(count,requests) MPI_Start(requests) 403 404 #endif /* PETSC_USE_LOG */ 405 406 extern PetscTruth PetscPreLoadingUsed; /* true if we are or have done preloading */ 407 extern PetscTruth PetscPreLoadingOn; /* true if we are currently in a preloading calculation */ 408 409 #define PreLoadBegin(flag,name) {PetscTruth PreLoading = flag; \ 410 int PreLoadMax,PreLoadIt,_stageNum,_3_ierr;\ 411 _3_ierr = PetscOptionsGetLogical(PETSC_NULL,"-preload",&PreLoading,PETSC_NULL);CHKERRQ(_3_ierr);\ 412 PreLoadMax = (int)(PreLoading);PetscPreLoadingUsed = PreLoading ? PETSC_TRUE : PetscPreLoadingUsed;\ 413 for (PreLoadIt=0; PreLoadIt<=PreLoadMax; PreLoadIt++) {\ 414 PetscPreLoadingOn = PreLoading;\ 415 _3_ierr = PetscBarrier(PETSC_NULL);CHKERRQ(_3_ierr);\ 416 if (PreLoadIt>0) {\ 417 _3_ierr = PetscLogStageGetId(name,&_stageNum);CHKERRQ(_3_ierr);\ 418 } else {\ 419 _3_ierr = PetscLogStageRegister(&_stageNum,name);CHKERRQ(_3_ierr);\ 420 _3_ierr = PetscLogStageSetVisible(_stageNum,(PetscTruth)(!PreLoadMax || PreLoadIt));\ 421 }\ 422 _3_ierr = PetscLogStagePush(_stageNum);CHKERRQ(_3_ierr); 423 424 #define PreLoadEnd() _3_ierr = PetscLogStagePop();CHKERRQ(_3_ierr);PreLoading = PETSC_FALSE;}} 425 426 #define PreLoadStage(name) _3_ierr = PetscLogStagePop();CHKERRQ(_3_ierr);\ 427 if (PreLoadIt>0) {\ 428 _3_ierr = PetscLogStageGetId(name,&_stageNum);CHKERRQ(_3_ierr);\ 429 } else {\ 430 _3_ierr = PetscLogStageRegister(&_stageNum,name);CHKERRQ(_3_ierr);\ 431 _3_ierr = PetscLogStageSetVisible(_stageNum,(PetscTruth)(!PreLoadMax || PreLoadIt));\ 432 }\ 433 _3_ierr = PetscLogStagePush(_stageNum);CHKERRQ(_3_ierr); 434 #endif 435