178f1b9b4SToby Isaac #include <petscviewer.h> 278f1b9b4SToby Isaac #include <petscdevice.h> 378f1b9b4SToby Isaac #include <petsc/private/logimpl.h> /*I "petscsys.h" I*/ 478f1b9b4SToby Isaac #include <petsc/private/loghandlerimpl.h> 578f1b9b4SToby Isaac #include <petsc/private/deviceimpl.h> 678f1b9b4SToby Isaac #include <petscconfiginfo.h> 778f1b9b4SToby Isaac #include <petscmachineinfo.h> 878f1b9b4SToby Isaac #include "logdefault.h" 978f1b9b4SToby Isaac 1078f1b9b4SToby Isaac static PetscErrorCode PetscEventPerfInfoInit(PetscEventPerfInfo *eventInfo) 1178f1b9b4SToby Isaac { 1278f1b9b4SToby Isaac PetscFunctionBegin; 1378f1b9b4SToby Isaac PetscCall(PetscMemzero(eventInfo, sizeof(*eventInfo))); 1478f1b9b4SToby Isaac eventInfo->visible = PETSC_TRUE; 1578f1b9b4SToby Isaac eventInfo->id = -1; 1678f1b9b4SToby Isaac eventInfo->dof[0] = -1.0; 1778f1b9b4SToby Isaac eventInfo->dof[1] = -1.0; 1878f1b9b4SToby Isaac eventInfo->dof[2] = -1.0; 1978f1b9b4SToby Isaac eventInfo->dof[3] = -1.0; 2078f1b9b4SToby Isaac eventInfo->dof[4] = -1.0; 2178f1b9b4SToby Isaac eventInfo->dof[5] = -1.0; 2278f1b9b4SToby Isaac eventInfo->dof[6] = -1.0; 2378f1b9b4SToby Isaac eventInfo->dof[7] = -1.0; 2478f1b9b4SToby Isaac eventInfo->errors[0] = -1.0; 2578f1b9b4SToby Isaac eventInfo->errors[1] = -1.0; 2678f1b9b4SToby Isaac eventInfo->errors[2] = -1.0; 2778f1b9b4SToby Isaac eventInfo->errors[3] = -1.0; 2878f1b9b4SToby Isaac eventInfo->errors[4] = -1.0; 2978f1b9b4SToby Isaac eventInfo->errors[5] = -1.0; 3078f1b9b4SToby Isaac eventInfo->errors[6] = -1.0; 3178f1b9b4SToby Isaac eventInfo->errors[7] = -1.0; 3278f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 3378f1b9b4SToby Isaac } 3478f1b9b4SToby Isaac 3578f1b9b4SToby Isaac static PetscErrorCode PetscEventPerfInfoTic_Internal(PetscEventPerfInfo *eventInfo, PetscLogDouble time, PetscBool logMemory, int event, PetscBool resume) 3678f1b9b4SToby Isaac { 3778f1b9b4SToby Isaac PetscFunctionBegin; 3878f1b9b4SToby Isaac if (resume) { 3978f1b9b4SToby Isaac eventInfo->timeTmp -= time; 4078f1b9b4SToby Isaac eventInfo->flopsTmp -= petsc_TotalFlops_th; 4178f1b9b4SToby Isaac } else { 4278f1b9b4SToby Isaac eventInfo->timeTmp = -time; 4378f1b9b4SToby Isaac eventInfo->flopsTmp = -petsc_TotalFlops_th; 4478f1b9b4SToby Isaac } 4578f1b9b4SToby Isaac eventInfo->numMessages -= petsc_irecv_ct_th + petsc_isend_ct_th + petsc_recv_ct_th + petsc_send_ct_th; 4678f1b9b4SToby Isaac eventInfo->messageLength -= petsc_irecv_len_th + petsc_isend_len_th + petsc_recv_len_th + petsc_send_len_th; 4778f1b9b4SToby Isaac eventInfo->numReductions -= petsc_allreduce_ct_th + petsc_gather_ct_th + petsc_scatter_ct_th; 4878f1b9b4SToby Isaac #if defined(PETSC_HAVE_DEVICE) 4978f1b9b4SToby Isaac eventInfo->CpuToGpuCount -= petsc_ctog_ct_th; 5078f1b9b4SToby Isaac eventInfo->GpuToCpuCount -= petsc_gtoc_ct_th; 5178f1b9b4SToby Isaac eventInfo->CpuToGpuSize -= petsc_ctog_sz_th; 5278f1b9b4SToby Isaac eventInfo->GpuToCpuSize -= petsc_gtoc_sz_th; 5378f1b9b4SToby Isaac eventInfo->GpuFlops -= petsc_gflops_th; 5478f1b9b4SToby Isaac eventInfo->GpuTime -= petsc_gtime; 5578f1b9b4SToby Isaac #endif 5678f1b9b4SToby Isaac if (logMemory) { 5778f1b9b4SToby Isaac PetscLogDouble usage; 5878f1b9b4SToby Isaac PetscCall(PetscMemoryGetCurrentUsage(&usage)); 5978f1b9b4SToby Isaac eventInfo->memIncrease -= usage; 6078f1b9b4SToby Isaac PetscCall(PetscMallocGetCurrentUsage(&usage)); 6178f1b9b4SToby Isaac eventInfo->mallocSpace -= usage; 6278f1b9b4SToby Isaac PetscCall(PetscMallocGetMaximumUsage(&usage)); 6378f1b9b4SToby Isaac eventInfo->mallocIncrease -= usage; 6478f1b9b4SToby Isaac PetscCall(PetscMallocPushMaximumUsage(event)); 6578f1b9b4SToby Isaac } 6678f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 6778f1b9b4SToby Isaac } 6878f1b9b4SToby Isaac 6978f1b9b4SToby Isaac static PetscErrorCode PetscEventPerfInfoTic(PetscEventPerfInfo *eventInfo, PetscLogDouble time, PetscBool logMemory, int event) 7078f1b9b4SToby Isaac { 7178f1b9b4SToby Isaac PetscFunctionBegin; 7278f1b9b4SToby Isaac PetscCall(PetscEventPerfInfoTic_Internal(eventInfo, time, logMemory, event, PETSC_FALSE)); 7378f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 7478f1b9b4SToby Isaac } 7578f1b9b4SToby Isaac 7678f1b9b4SToby Isaac static PetscErrorCode PetscEventPerfInfoResume(PetscEventPerfInfo *eventInfo, PetscLogDouble time, PetscBool logMemory, int event) 7778f1b9b4SToby Isaac { 7878f1b9b4SToby Isaac PetscFunctionBegin; 7978f1b9b4SToby Isaac PetscCall(PetscEventPerfInfoTic_Internal(eventInfo, time, logMemory, event, PETSC_TRUE)); 8078f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 8178f1b9b4SToby Isaac } 8278f1b9b4SToby Isaac 8378f1b9b4SToby Isaac static PetscErrorCode PetscEventPerfInfoToc_Internal(PetscEventPerfInfo *eventInfo, PetscLogDouble time, PetscBool logMemory, int event, PetscBool pause) 8478f1b9b4SToby Isaac { 8578f1b9b4SToby Isaac PetscFunctionBegin; 8678f1b9b4SToby Isaac eventInfo->timeTmp += time; 8778f1b9b4SToby Isaac eventInfo->flopsTmp += petsc_TotalFlops_th; 8878f1b9b4SToby Isaac if (!pause) { 8978f1b9b4SToby Isaac eventInfo->time += eventInfo->timeTmp; 9078f1b9b4SToby Isaac eventInfo->time2 += eventInfo->timeTmp * eventInfo->timeTmp; 9178f1b9b4SToby Isaac eventInfo->flops += eventInfo->flopsTmp; 9278f1b9b4SToby Isaac eventInfo->flops2 += eventInfo->flopsTmp * eventInfo->flopsTmp; 9378f1b9b4SToby Isaac } 9478f1b9b4SToby Isaac eventInfo->numMessages += petsc_irecv_ct_th + petsc_isend_ct_th + petsc_recv_ct_th + petsc_send_ct_th; 9578f1b9b4SToby Isaac eventInfo->messageLength += petsc_irecv_len_th + petsc_isend_len_th + petsc_recv_len + petsc_send_len_th; 9678f1b9b4SToby Isaac eventInfo->numReductions += petsc_allreduce_ct_th + petsc_gather_ct_th + petsc_scatter_ct_th; 9778f1b9b4SToby Isaac #if defined(PETSC_HAVE_DEVICE) 9878f1b9b4SToby Isaac eventInfo->CpuToGpuCount += petsc_ctog_ct_th; 9978f1b9b4SToby Isaac eventInfo->GpuToCpuCount += petsc_gtoc_ct_th; 10078f1b9b4SToby Isaac eventInfo->CpuToGpuSize += petsc_ctog_sz_th; 10178f1b9b4SToby Isaac eventInfo->GpuToCpuSize += petsc_gtoc_sz_th; 10278f1b9b4SToby Isaac eventInfo->GpuFlops += petsc_gflops_th; 10378f1b9b4SToby Isaac eventInfo->GpuTime += petsc_gtime; 10478f1b9b4SToby Isaac #endif 10578f1b9b4SToby Isaac if (logMemory) { 10678f1b9b4SToby Isaac PetscLogDouble usage, musage; 10778f1b9b4SToby Isaac PetscCall(PetscMemoryGetCurrentUsage(&usage)); /* the comments below match the column labels printed in PetscLogView_Default() */ 10878f1b9b4SToby Isaac eventInfo->memIncrease += usage; /* RMI */ 10978f1b9b4SToby Isaac PetscCall(PetscMallocGetCurrentUsage(&usage)); 11078f1b9b4SToby Isaac eventInfo->mallocSpace += usage; /* Malloc */ 111835f2295SStefano Zampini PetscCall(PetscMallocPopMaximumUsage(event, &musage)); 11278f1b9b4SToby Isaac eventInfo->mallocIncreaseEvent = PetscMax(musage - usage, eventInfo->mallocIncreaseEvent); /* EMalloc */ 11378f1b9b4SToby Isaac PetscCall(PetscMallocGetMaximumUsage(&usage)); 11478f1b9b4SToby Isaac eventInfo->mallocIncrease += usage; /* MMalloc */ 11578f1b9b4SToby Isaac } 11678f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 11778f1b9b4SToby Isaac } 11878f1b9b4SToby Isaac 11978f1b9b4SToby Isaac static PetscErrorCode PetscEventPerfInfoToc(PetscEventPerfInfo *eventInfo, PetscLogDouble time, PetscBool logMemory, int event) 12078f1b9b4SToby Isaac { 12178f1b9b4SToby Isaac PetscFunctionBegin; 12278f1b9b4SToby Isaac PetscCall(PetscEventPerfInfoToc_Internal(eventInfo, time, logMemory, event, PETSC_FALSE)); 12378f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 12478f1b9b4SToby Isaac } 12578f1b9b4SToby Isaac 12678f1b9b4SToby Isaac static PetscErrorCode PetscEventPerfInfoPause(PetscEventPerfInfo *eventInfo, PetscLogDouble time, PetscBool logMemory, int event) 12778f1b9b4SToby Isaac { 12878f1b9b4SToby Isaac PetscFunctionBegin; 1293e446a5eSToby Isaac PetscCall(PetscEventPerfInfoToc_Internal(eventInfo, time, logMemory, event, PETSC_TRUE)); 13078f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 13178f1b9b4SToby Isaac } 13278f1b9b4SToby Isaac 13378f1b9b4SToby Isaac static PetscErrorCode PetscEventPerfInfoAdd_Internal(const PetscEventPerfInfo *eventInfo, PetscEventPerfInfo *outInfo) 13478f1b9b4SToby Isaac { 13578f1b9b4SToby Isaac PetscFunctionBegin; 13678f1b9b4SToby Isaac outInfo->count += eventInfo->count; 13778f1b9b4SToby Isaac outInfo->time += eventInfo->time; 13878f1b9b4SToby Isaac outInfo->time2 += eventInfo->time2; 13978f1b9b4SToby Isaac outInfo->flops += eventInfo->flops; 14078f1b9b4SToby Isaac outInfo->flops2 += eventInfo->flops2; 14178f1b9b4SToby Isaac outInfo->numMessages += eventInfo->numMessages; 14278f1b9b4SToby Isaac outInfo->messageLength += eventInfo->messageLength; 14378f1b9b4SToby Isaac outInfo->numReductions += eventInfo->numReductions; 14478f1b9b4SToby Isaac #if defined(PETSC_HAVE_DEVICE) 14578f1b9b4SToby Isaac outInfo->CpuToGpuCount += eventInfo->CpuToGpuCount; 14678f1b9b4SToby Isaac outInfo->GpuToCpuCount += eventInfo->GpuToCpuCount; 14778f1b9b4SToby Isaac outInfo->CpuToGpuSize += eventInfo->CpuToGpuSize; 14878f1b9b4SToby Isaac outInfo->GpuToCpuSize += eventInfo->GpuToCpuSize; 14978f1b9b4SToby Isaac outInfo->GpuFlops += eventInfo->GpuFlops; 15078f1b9b4SToby Isaac outInfo->GpuTime += eventInfo->GpuTime; 15178f1b9b4SToby Isaac #endif 15278f1b9b4SToby Isaac outInfo->memIncrease += eventInfo->memIncrease; 15378f1b9b4SToby Isaac outInfo->mallocSpace += eventInfo->mallocSpace; 15478f1b9b4SToby Isaac outInfo->mallocIncreaseEvent += eventInfo->mallocIncreaseEvent; 15578f1b9b4SToby Isaac outInfo->mallocIncrease += eventInfo->mallocIncrease; 15678f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 15778f1b9b4SToby Isaac } 15878f1b9b4SToby Isaac 15978f1b9b4SToby Isaac PETSC_LOG_RESIZABLE_ARRAY(EventPerfArray, PetscEventPerfInfo, PetscLogEvent, PetscEventPerfInfoInit, NULL, NULL) 16078f1b9b4SToby Isaac 16178f1b9b4SToby Isaac /* --- PetscClassPerf --- */ 16278f1b9b4SToby Isaac 16378f1b9b4SToby Isaac typedef struct { 16478f1b9b4SToby Isaac int creations; /* The number of objects of this class created */ 16578f1b9b4SToby Isaac int destructions; /* The number of objects of this class destroyed */ 16678f1b9b4SToby Isaac PetscLogDouble mem; /* The total memory allocated by objects of this class; this is completely wrong and should possibly be removed */ 16778f1b9b4SToby Isaac PetscLogDouble descMem; /* The total memory allocated by descendents of these objects; this is completely wrong and should possibly be removed */ 16878f1b9b4SToby Isaac } PetscClassPerf; 16978f1b9b4SToby Isaac 17078f1b9b4SToby Isaac static PetscErrorCode PetscClassPerfInit(PetscClassPerf *classInfo) 17178f1b9b4SToby Isaac { 17278f1b9b4SToby Isaac PetscFunctionBegin; 17378f1b9b4SToby Isaac PetscCall(PetscMemzero(classInfo, sizeof(*classInfo))); 17478f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 17578f1b9b4SToby Isaac } 17678f1b9b4SToby Isaac 17778f1b9b4SToby Isaac PETSC_LOG_RESIZABLE_ARRAY(ClassPerfArray, PetscClassPerf, PetscLogClass, PetscClassPerfInit, NULL, NULL) 17878f1b9b4SToby Isaac 17978f1b9b4SToby Isaac /* --- PetscStagePerf --- */ 18078f1b9b4SToby Isaac 18178f1b9b4SToby Isaac typedef struct _PetscStagePerf { 18278f1b9b4SToby Isaac PetscBool used; /* The stage was pushed on this processor */ 18378f1b9b4SToby Isaac PetscEventPerfInfo perfInfo; /* The stage performance information */ 18478f1b9b4SToby Isaac PetscLogEventPerfArray eventLog; /* The event information for this stage */ 18578f1b9b4SToby Isaac PetscLogClassPerfArray classLog; /* The class information for this stage */ 18678f1b9b4SToby Isaac } PetscStagePerf; 18778f1b9b4SToby Isaac 18878f1b9b4SToby Isaac static PetscErrorCode PetscStageInfoInit(PetscStagePerf *stageInfo) 18978f1b9b4SToby Isaac { 19078f1b9b4SToby Isaac PetscFunctionBegin; 19178f1b9b4SToby Isaac PetscCall(PetscMemzero(stageInfo, sizeof(*stageInfo))); 192f4f49eeaSPierre Jolivet PetscCall(PetscLogEventPerfArrayCreate(128, &stageInfo->eventLog)); 193f4f49eeaSPierre Jolivet PetscCall(PetscLogClassPerfArrayCreate(128, &stageInfo->classLog)); 19478f1b9b4SToby Isaac PetscCall(PetscEventPerfInfoInit(&stageInfo->perfInfo)); 19578f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 19678f1b9b4SToby Isaac } 19778f1b9b4SToby Isaac 19878f1b9b4SToby Isaac static PetscErrorCode PetscStageInfoReset(PetscStagePerf *stageInfo) 19978f1b9b4SToby Isaac { 20078f1b9b4SToby Isaac PetscFunctionBegin; 201f4f49eeaSPierre Jolivet PetscCall(PetscLogEventPerfArrayDestroy(&stageInfo->eventLog)); 202f4f49eeaSPierre Jolivet PetscCall(PetscLogClassPerfArrayDestroy(&stageInfo->classLog)); 20378f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 20478f1b9b4SToby Isaac } 20578f1b9b4SToby Isaac 20678f1b9b4SToby Isaac PETSC_LOG_RESIZABLE_ARRAY(StageInfoArray, PetscStagePerf, PetscLogStage, PetscStageInfoInit, PetscStageInfoReset, NULL) 20778f1b9b4SToby Isaac 20878f1b9b4SToby Isaac /* --- Action --- */ 20978f1b9b4SToby Isaac 21078f1b9b4SToby Isaac /* The structure for action logging */ 21178f1b9b4SToby Isaac typedef enum { 21278f1b9b4SToby Isaac PETSC_LOG_ACTION_CREATE, 21378f1b9b4SToby Isaac PETSC_LOG_ACTION_DESTROY, 21478f1b9b4SToby Isaac PETSC_LOG_ACTION_BEGIN, 21578f1b9b4SToby Isaac PETSC_LOG_ACTION_END, 21678f1b9b4SToby Isaac } PetscLogActionType; 21778f1b9b4SToby Isaac 218b665b14eSToby Isaac typedef struct _Action { 219b665b14eSToby Isaac PetscLogActionType action; /* The type of execution */ 220b665b14eSToby Isaac PetscLogEvent event; /* The event number */ 221b665b14eSToby Isaac PetscClassId classid; /* The event class id */ 222b665b14eSToby Isaac PetscLogDouble time; /* The time of occurrence */ 223b665b14eSToby Isaac PetscLogDouble flops; /* The cumulative flops */ 224b665b14eSToby Isaac PetscLogDouble mem; /* The current memory usage */ 225b665b14eSToby Isaac PetscLogDouble maxmem; /* The maximum memory usage */ 2266497c311SBarry Smith PetscObjectId id1, id2, id3; /* The ids of associated objects */ 227b665b14eSToby Isaac } Action; 228b665b14eSToby Isaac 22978f1b9b4SToby Isaac PETSC_LOG_RESIZABLE_ARRAY(ActionArray, Action, PetscLogEvent, NULL, NULL, NULL) 23078f1b9b4SToby Isaac 23178f1b9b4SToby Isaac /* --- Object --- */ 23278f1b9b4SToby Isaac 233b665b14eSToby Isaac /* The structure for object logging */ 234b665b14eSToby Isaac typedef struct _Object { 235b665b14eSToby Isaac PetscObject obj; /* The associated PetscObject */ 236b665b14eSToby Isaac int parent; /* The parent id */ 237b665b14eSToby Isaac PetscLogDouble mem; /* The memory associated with the object */ 238b665b14eSToby Isaac char name[64]; /* The object name */ 239b665b14eSToby Isaac char info[64]; /* The information string */ 240b665b14eSToby Isaac } Object; 241b665b14eSToby Isaac 24278f1b9b4SToby Isaac PETSC_LOG_RESIZABLE_ARRAY(ObjectArray, Object, PetscObject, NULL, NULL, NULL) 24378f1b9b4SToby Isaac 24478f1b9b4SToby Isaac /* Map from (threadid,stage,event) to perfInfo data struct */ 24578f1b9b4SToby Isaac #include <petsc/private/hashmapijk.h> 24678f1b9b4SToby Isaac 24778f1b9b4SToby Isaac PETSC_HASH_MAP(HMapEvent, PetscHashIJKKey, PetscEventPerfInfo *, PetscHashIJKKeyHash, PetscHashIJKKeyEqual, NULL) 24878f1b9b4SToby Isaac 24978f1b9b4SToby Isaac typedef struct _n_PetscLogHandler_Default *PetscLogHandler_Default; 25078f1b9b4SToby Isaac struct _n_PetscLogHandler_Default { 25178f1b9b4SToby Isaac PetscLogStageInfoArray stages; 25278f1b9b4SToby Isaac PetscSpinlock lock; 25378f1b9b4SToby Isaac PetscLogActionArray petsc_actions; 25478f1b9b4SToby Isaac PetscLogObjectArray petsc_objects; 25578f1b9b4SToby Isaac PetscBool petsc_logActions; 25678f1b9b4SToby Isaac PetscBool petsc_logObjects; 25778f1b9b4SToby Isaac int petsc_numObjectsCreated; 25878f1b9b4SToby Isaac int petsc_numObjectsDestroyed; 25978f1b9b4SToby Isaac PetscHMapEvent eventInfoMap_th; 26078f1b9b4SToby Isaac int pause_depth; 261461318ebSToby Isaac PetscBool use_threadsafe; 26278f1b9b4SToby Isaac }; 26378f1b9b4SToby Isaac 26478f1b9b4SToby Isaac /* --- PetscLogHandler_Default --- */ 26578f1b9b4SToby Isaac 26678f1b9b4SToby Isaac static PetscErrorCode PetscLogHandlerContextCreate_Default(PetscLogHandler_Default *def_p) 26778f1b9b4SToby Isaac { 26878f1b9b4SToby Isaac PetscLogHandler_Default def; 26978f1b9b4SToby Isaac 27078f1b9b4SToby Isaac PetscFunctionBegin; 27178f1b9b4SToby Isaac PetscCall(PetscNew(def_p)); 27278f1b9b4SToby Isaac def = *def_p; 273f4f49eeaSPierre Jolivet PetscCall(PetscLogStageInfoArrayCreate(8, &def->stages)); 274f4f49eeaSPierre Jolivet PetscCall(PetscLogActionArrayCreate(64, &def->petsc_actions)); 275f4f49eeaSPierre Jolivet PetscCall(PetscLogObjectArrayCreate(64, &def->petsc_objects)); 2763bbd2b04SJunchao Zhang PetscCall(PetscSpinlockCreate(&def->lock)); 27778f1b9b4SToby Isaac 27878f1b9b4SToby Isaac PetscCall(PetscOptionsGetBool(NULL, NULL, "-log_include_actions", &def->petsc_logActions, NULL)); 27978f1b9b4SToby Isaac PetscCall(PetscOptionsGetBool(NULL, NULL, "-log_include_objects", &def->petsc_logObjects, NULL)); 280461318ebSToby Isaac PetscCall(PetscOptionsGetBool(NULL, NULL, "-log_handler_default_use_threadsafe_events", &def->use_threadsafe, NULL)); 281461318ebSToby Isaac if (PetscDefined(HAVE_THREADSAFETY) || def->use_threadsafe) { PetscCall(PetscHMapEventCreate(&def->eventInfoMap_th)); } 28278f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 28378f1b9b4SToby Isaac } 28478f1b9b4SToby Isaac 28578f1b9b4SToby Isaac static PetscErrorCode PetscLogHandlerDestroy_Default(PetscLogHandler h) 28678f1b9b4SToby Isaac { 28778f1b9b4SToby Isaac PetscLogHandler_Default def = (PetscLogHandler_Default)h->data; 28878f1b9b4SToby Isaac 28978f1b9b4SToby Isaac PetscFunctionBegin; 29078f1b9b4SToby Isaac PetscCall(PetscLogStageInfoArrayDestroy(&def->stages)); 29178f1b9b4SToby Isaac PetscCall(PetscLogActionArrayDestroy(&def->petsc_actions)); 29278f1b9b4SToby Isaac PetscCall(PetscLogObjectArrayDestroy(&def->petsc_objects)); 2933bbd2b04SJunchao Zhang PetscCall(PetscSpinlockDestroy(&def->lock)); 29478f1b9b4SToby Isaac if (def->eventInfoMap_th) { 29578f1b9b4SToby Isaac PetscEventPerfInfo **array; 29678f1b9b4SToby Isaac PetscInt n, off = 0; 29778f1b9b4SToby Isaac 29878f1b9b4SToby Isaac PetscCall(PetscHMapEventGetSize(def->eventInfoMap_th, &n)); 29978f1b9b4SToby Isaac PetscCall(PetscMalloc1(n, &array)); 30078f1b9b4SToby Isaac PetscCall(PetscHMapEventGetVals(def->eventInfoMap_th, &off, array)); 30178f1b9b4SToby Isaac for (PetscInt i = 0; i < n; i++) PetscCall(PetscFree(array[i])); 30278f1b9b4SToby Isaac PetscCall(PetscFree(array)); 30378f1b9b4SToby Isaac PetscCall(PetscHMapEventDestroy(&def->eventInfoMap_th)); 30478f1b9b4SToby Isaac } 305dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)h, "PetscLogHandlerGetEventPerfInfo_C", NULL)); 3068b08f494SToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)h, "PetscLogHandlerGetStagePerfInfo_C", NULL)); 307dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)h, "PetscLogHandlerSetLogActions_C", NULL)); 308dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)h, "PetscLogHandlerSetLogObjects_C", NULL)); 309dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)h, "PetscLogHandlerLogObjectState_C", NULL)); 310dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)h, "PetscLogHandlerGetNumObjects_C", NULL)); 311dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)h, "PetscLogHandlerEventDeactivatePush_C", NULL)); 312dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)h, "PetscLogHandlerEventDeactivatePop_C", NULL)); 313dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)h, "PetscLogHandlerEventsPause_C", NULL)); 314dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)h, "PetscLogHandlerEventsResume_C", NULL)); 315dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)h, "PetscLogHandlerDump_C", NULL)); 316dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)h, "PetscLogHandlerStageSetVisible_C", NULL)); 317dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)h, "PetscLogHandlerStageGetVisible_C", NULL)); 31878f1b9b4SToby Isaac PetscCall(PetscFree(def)); 31978f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 32078f1b9b4SToby Isaac } 32178f1b9b4SToby Isaac 32278f1b9b4SToby Isaac static PetscErrorCode PetscLogHandlerDefaultGetStageInfo(PetscLogHandler handler, PetscLogStage stage, PetscStagePerf **stage_info_p) 32378f1b9b4SToby Isaac { 32478f1b9b4SToby Isaac PetscStagePerf *stage_info = NULL; 32578f1b9b4SToby Isaac PetscLogHandler_Default def = (PetscLogHandler_Default)handler->data; 32678f1b9b4SToby Isaac 32778f1b9b4SToby Isaac PetscFunctionBegin; 32878f1b9b4SToby Isaac PetscCall(PetscLogStageInfoArrayResize(def->stages, stage + 1)); 32978f1b9b4SToby Isaac PetscCall(PetscLogStageInfoArrayGetRef(def->stages, stage, &stage_info)); 33078f1b9b4SToby Isaac *stage_info_p = stage_info; 33178f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 33278f1b9b4SToby Isaac } 33378f1b9b4SToby Isaac 334dff009beSToby Isaac static PetscErrorCode PetscLogHandlerGetEventPerfInfo_Default(PetscLogHandler handler, PetscLogStage stage, PetscLogEvent event, PetscEventPerfInfo **event_info_p) 33578f1b9b4SToby Isaac { 33678f1b9b4SToby Isaac PetscEventPerfInfo *event_info = NULL; 33778f1b9b4SToby Isaac PetscStagePerf *stage_info = NULL; 33878f1b9b4SToby Isaac PetscLogEventPerfArray event_log; 33978f1b9b4SToby Isaac 34078f1b9b4SToby Isaac PetscFunctionBegin; 34178f1b9b4SToby Isaac if (stage < 0) PetscCall(PetscLogStateGetCurrentStage(handler->state, &stage)); 34278f1b9b4SToby Isaac PetscCall(PetscLogHandlerDefaultGetStageInfo(handler, stage, &stage_info)); 34378f1b9b4SToby Isaac event_log = stage_info->eventLog; 34478f1b9b4SToby Isaac PetscCall(PetscLogEventPerfArrayResize(event_log, event + 1)); 34578f1b9b4SToby Isaac PetscCall(PetscLogEventPerfArrayGetRef(event_log, event, &event_info)); 34678f1b9b4SToby Isaac event_info->id = event; 34778f1b9b4SToby Isaac *event_info_p = event_info; 34878f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 34978f1b9b4SToby Isaac } 35078f1b9b4SToby Isaac 3518b08f494SToby Isaac static PetscErrorCode PetscLogHandlerGetStagePerfInfo_Default(PetscLogHandler handler, PetscLogStage stage, PetscEventPerfInfo **stage_info_p) 3528b08f494SToby Isaac { 3538b08f494SToby Isaac PetscStagePerf *stage_perf_info = NULL; 3548b08f494SToby Isaac 3558b08f494SToby Isaac PetscFunctionBegin; 3568b08f494SToby Isaac if (stage < 0) PetscCall(PetscLogStateGetCurrentStage(handler->state, &stage)); 3578b08f494SToby Isaac PetscCall(PetscLogHandlerDefaultGetStageInfo(handler, stage, &stage_perf_info)); 3588b08f494SToby Isaac *stage_info_p = &stage_perf_info->perfInfo; 3598b08f494SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 3608b08f494SToby Isaac } 3618b08f494SToby Isaac 36278f1b9b4SToby Isaac static PetscErrorCode PetscLogHandlerDefaultGetClassPerf(PetscLogHandler handler, PetscLogStage stage, PetscLogClass clss, PetscClassPerf **class_info) 36378f1b9b4SToby Isaac { 36478f1b9b4SToby Isaac PetscLogClassPerfArray class_log; 36578f1b9b4SToby Isaac PetscStagePerf *stage_info; 36678f1b9b4SToby Isaac 36778f1b9b4SToby Isaac PetscFunctionBegin; 36878f1b9b4SToby Isaac PetscCall(PetscLogHandlerDefaultGetStageInfo(handler, stage, &stage_info)); 36978f1b9b4SToby Isaac class_log = stage_info->classLog; 37078f1b9b4SToby Isaac PetscCall(PetscLogClassPerfArrayResize(class_log, clss + 1)); 37178f1b9b4SToby Isaac PetscCall(PetscLogClassPerfArrayGetRef(class_log, clss, class_info)); 37278f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 37378f1b9b4SToby Isaac } 37478f1b9b4SToby Isaac 37578f1b9b4SToby Isaac static PetscErrorCode PetscLogHandlerObjectCreate_Default(PetscLogHandler h, PetscObject obj) 37678f1b9b4SToby Isaac { 37778f1b9b4SToby Isaac PetscLogHandler_Default def = (PetscLogHandler_Default)h->data; 37878f1b9b4SToby Isaac PetscLogState state; 37978f1b9b4SToby Isaac PetscLogStage stage; 38078f1b9b4SToby Isaac PetscClassPerf *classInfo; 38178f1b9b4SToby Isaac int oclass = 0; 38278f1b9b4SToby Isaac 38378f1b9b4SToby Isaac PetscFunctionBegin; 38478f1b9b4SToby Isaac PetscCall(PetscLogHandlerGetState(h, &state)); 38578f1b9b4SToby Isaac PetscCall(PetscSpinlockLock(&def->lock)); 38678f1b9b4SToby Isaac /* Record stage info */ 38778f1b9b4SToby Isaac PetscCall(PetscLogStateGetCurrentStage(state, &stage)); 38878f1b9b4SToby Isaac PetscCall(PetscLogStateGetClassFromClassId(state, obj->classid, &oclass)); 38978f1b9b4SToby Isaac PetscCall(PetscLogHandlerDefaultGetClassPerf(h, stage, oclass, &classInfo)); 39078f1b9b4SToby Isaac classInfo->creations++; 39178f1b9b4SToby Isaac /* Record the creation action */ 39278f1b9b4SToby Isaac if (def->petsc_logActions) { 39378f1b9b4SToby Isaac Action new_action; 39478f1b9b4SToby Isaac 39578f1b9b4SToby Isaac PetscCall(PetscTime(&new_action.time)); 39678f1b9b4SToby Isaac new_action.time -= petsc_BaseTime; 39778f1b9b4SToby Isaac new_action.action = PETSC_LOG_ACTION_CREATE; 39878f1b9b4SToby Isaac new_action.event = -1; 39978f1b9b4SToby Isaac new_action.classid = obj->classid; 40078f1b9b4SToby Isaac new_action.id1 = obj->id; 40178f1b9b4SToby Isaac new_action.id2 = -1; 40278f1b9b4SToby Isaac new_action.id3 = -1; 40378f1b9b4SToby Isaac new_action.flops = petsc_TotalFlops; 40478f1b9b4SToby Isaac PetscCall(PetscMallocGetCurrentUsage(&new_action.mem)); 40578f1b9b4SToby Isaac PetscCall(PetscMallocGetMaximumUsage(&new_action.maxmem)); 40678f1b9b4SToby Isaac PetscCall(PetscLogActionArrayPush(def->petsc_actions, new_action)); 40778f1b9b4SToby Isaac } 40878f1b9b4SToby Isaac /* We don't just use obj->id to count all objects that are created 40978f1b9b4SToby Isaac because PetscLogHandlers are objects and PetscLogObjectDestroy() will not 41078f1b9b4SToby Isaac be called for them: the number of objects created and destroyed as counted 41178f1b9b4SToby Isaac here and below would have an imbalance */ 41278f1b9b4SToby Isaac def->petsc_numObjectsCreated++; 41378f1b9b4SToby Isaac /* Record the object */ 41478f1b9b4SToby Isaac if (def->petsc_logObjects) { 41578f1b9b4SToby Isaac Object new_object; 4166497c311SBarry Smith PetscInt objid; 41778f1b9b4SToby Isaac 41878f1b9b4SToby Isaac new_object.parent = -1; 41978f1b9b4SToby Isaac new_object.obj = obj; 42078f1b9b4SToby Isaac new_object.mem = 0; 42178f1b9b4SToby Isaac 42278f1b9b4SToby Isaac PetscCall(PetscMemzero(new_object.name, sizeof(new_object.name))); 42378f1b9b4SToby Isaac PetscCall(PetscMemzero(new_object.info, sizeof(new_object.info))); 42478f1b9b4SToby Isaac PetscAssert(obj->id >= 1, PETSC_COMM_SELF, PETSC_ERR_PLIB, "Object ids from PetscObjectNewId_Internal() start at 1"); 4256497c311SBarry Smith PetscCall(PetscIntCast(obj->id, &objid)); 4266497c311SBarry Smith PetscCall(PetscLogObjectArrayResize(def->petsc_objects, objid)); 4276497c311SBarry Smith PetscCall(PetscLogObjectArraySet(def->petsc_objects, objid - 1, new_object)); 42878f1b9b4SToby Isaac } 42978f1b9b4SToby Isaac PetscCall(PetscSpinlockUnlock(&def->lock)); 43078f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 43178f1b9b4SToby Isaac } 43278f1b9b4SToby Isaac 43378f1b9b4SToby Isaac static PetscErrorCode PetscLogHandlerObjectDestroy_Default(PetscLogHandler h, PetscObject obj) 43478f1b9b4SToby Isaac { 43578f1b9b4SToby Isaac PetscLogHandler_Default def = (PetscLogHandler_Default)h->data; 43678f1b9b4SToby Isaac PetscLogState state; 43778f1b9b4SToby Isaac PetscLogStage stage; 43878f1b9b4SToby Isaac PetscClassPerf *classInfo; 43978f1b9b4SToby Isaac int oclass = 0; 44078f1b9b4SToby Isaac 44178f1b9b4SToby Isaac PetscFunctionBegin; 44278f1b9b4SToby Isaac PetscCall(PetscLogHandlerGetState(h, &state)); 44378f1b9b4SToby Isaac /* Record stage info */ 44478f1b9b4SToby Isaac PetscCall(PetscSpinlockLock(&def->lock)); 44578f1b9b4SToby Isaac PetscCall(PetscLogStateGetCurrentStage(state, &stage)); 44678f1b9b4SToby Isaac if (stage >= 0) { 44778f1b9b4SToby Isaac /* stage < 0 can happen if the log summary is output before some things are destroyed */ 44878f1b9b4SToby Isaac PetscCall(PetscLogStateGetClassFromClassId(state, obj->classid, &oclass)); 44978f1b9b4SToby Isaac PetscCall(PetscLogHandlerDefaultGetClassPerf(h, stage, oclass, &classInfo)); 45078f1b9b4SToby Isaac classInfo->destructions++; 45178f1b9b4SToby Isaac } 45278f1b9b4SToby Isaac /* Cannot Credit all ancestors with your memory because they may have already been destroyed*/ 45378f1b9b4SToby Isaac def->petsc_numObjectsDestroyed++; 45478f1b9b4SToby Isaac /* Dynamically enlarge logging structures */ 45578f1b9b4SToby Isaac /* Record the destruction action */ 45678f1b9b4SToby Isaac if (def->petsc_logActions) { 45778f1b9b4SToby Isaac Action new_action; 45878f1b9b4SToby Isaac 45978f1b9b4SToby Isaac PetscCall(PetscTime(&new_action.time)); 46078f1b9b4SToby Isaac new_action.time -= petsc_BaseTime; 46178f1b9b4SToby Isaac new_action.event = -1; 46278f1b9b4SToby Isaac new_action.action = PETSC_LOG_ACTION_DESTROY; 46378f1b9b4SToby Isaac new_action.classid = obj->classid; 46478f1b9b4SToby Isaac new_action.id1 = obj->id; 46578f1b9b4SToby Isaac new_action.id2 = -1; 46678f1b9b4SToby Isaac new_action.id3 = -1; 46778f1b9b4SToby Isaac new_action.flops = petsc_TotalFlops; 46878f1b9b4SToby Isaac PetscCall(PetscMallocGetCurrentUsage(&new_action.mem)); 46978f1b9b4SToby Isaac PetscCall(PetscMallocGetMaximumUsage(&new_action.maxmem)); 47078f1b9b4SToby Isaac PetscCall(PetscLogActionArrayPush(def->petsc_actions, new_action)); 47178f1b9b4SToby Isaac } 47278f1b9b4SToby Isaac if (def->petsc_logObjects) { 47378f1b9b4SToby Isaac Object *obj_entry = NULL; 4746497c311SBarry Smith PetscInt objid; 47578f1b9b4SToby Isaac 47678f1b9b4SToby Isaac PetscAssert(obj->id >= 1, PETSC_COMM_SELF, PETSC_ERR_PLIB, "Object ids from PetscObjectNewId_Internal() start at 1"); 4776497c311SBarry Smith PetscCall(PetscIntCast(obj->id - 1, &objid)); 4786497c311SBarry Smith PetscCall(PetscLogObjectArrayGetRef(def->petsc_objects, objid, &obj_entry)); 47978f1b9b4SToby Isaac if (obj->name) PetscCall(PetscStrncpy(obj_entry->name, obj->name, 64)); 48078f1b9b4SToby Isaac obj_entry->obj = NULL; 48178f1b9b4SToby Isaac } 48278f1b9b4SToby Isaac PetscCall(PetscSpinlockUnlock(&def->lock)); 48378f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 48478f1b9b4SToby Isaac } 48578f1b9b4SToby Isaac 48678f1b9b4SToby Isaac static PetscErrorCode PetscLogHandlerEventSync_Default(PetscLogHandler h, PetscLogEvent event, MPI_Comm comm) 48778f1b9b4SToby Isaac { 48878f1b9b4SToby Isaac PetscLogState state; 48978f1b9b4SToby Isaac PetscLogEventInfo event_info; 49078f1b9b4SToby Isaac PetscEventPerfInfo *event_perf_info; 49178f1b9b4SToby Isaac int stage; 49278f1b9b4SToby Isaac PetscLogDouble time = 0.0; 49378f1b9b4SToby Isaac 49478f1b9b4SToby Isaac PetscFunctionBegin; 4955aefd447SStefano Zampini if (!PetscLogSyncOn || comm == MPI_COMM_NULL) PetscFunctionReturn(PETSC_SUCCESS); 49678f1b9b4SToby Isaac PetscCall(PetscLogHandlerGetState(h, &state)); 49778f1b9b4SToby Isaac PetscCall(PetscLogStateEventGetInfo(state, event, &event_info)); 49878f1b9b4SToby Isaac if (!event_info.collective) PetscFunctionReturn(PETSC_SUCCESS); 49978f1b9b4SToby Isaac PetscCall(PetscLogStateGetCurrentStage(state, &stage)); 500dff009beSToby Isaac PetscCall(PetscLogHandlerGetEventPerfInfo_Default(h, stage, event, &event_perf_info)); 50178f1b9b4SToby Isaac if (event_perf_info->depth > 0) PetscFunctionReturn(PETSC_SUCCESS); 50278f1b9b4SToby Isaac 50378f1b9b4SToby Isaac PetscCall(PetscTimeSubtract(&time)); 50478f1b9b4SToby Isaac PetscCallMPI(MPI_Barrier(comm)); 50578f1b9b4SToby Isaac PetscCall(PetscTimeAdd(&time)); 50678f1b9b4SToby Isaac event_perf_info->syncTime += time; 50778f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 50878f1b9b4SToby Isaac } 50978f1b9b4SToby Isaac 51078f1b9b4SToby Isaac static PetscErrorCode PetscLogGetStageEventPerfInfo_threaded(PetscLogHandler_Default def, PetscLogStage stage, PetscLogEvent event, PetscEventPerfInfo **eventInfo) 51178f1b9b4SToby Isaac { 51278f1b9b4SToby Isaac PetscEventPerfInfo *leventInfo = NULL; 51378f1b9b4SToby Isaac PetscHashIJKKey key; 51478f1b9b4SToby Isaac 51578f1b9b4SToby Isaac PetscFunctionBegin; 51678f1b9b4SToby Isaac #if PetscDefined(HAVE_THREADSAFETY) 51778f1b9b4SToby Isaac key.i = PetscLogGetTid(); 51878f1b9b4SToby Isaac #else 51978f1b9b4SToby Isaac key.i = 0; 52078f1b9b4SToby Isaac #endif 52178f1b9b4SToby Isaac key.j = stage; 52278f1b9b4SToby Isaac key.k = event; 52378f1b9b4SToby Isaac PetscCall(PetscSpinlockLock(&def->lock)); 52478f1b9b4SToby Isaac PetscCall(PetscHMapEventGet(def->eventInfoMap_th, key, &leventInfo)); 52578f1b9b4SToby Isaac if (!leventInfo) { 52678f1b9b4SToby Isaac PetscCall(PetscNew(&leventInfo)); 52778f1b9b4SToby Isaac leventInfo->id = event; 52878f1b9b4SToby Isaac PetscCall(PetscHMapEventSet(def->eventInfoMap_th, key, leventInfo)); 52978f1b9b4SToby Isaac } 53078f1b9b4SToby Isaac PetscCall(PetscSpinlockUnlock(&def->lock)); 53178f1b9b4SToby Isaac *eventInfo = leventInfo; 53278f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 53378f1b9b4SToby Isaac } 53478f1b9b4SToby Isaac 53578f1b9b4SToby Isaac static PetscErrorCode PetscLogHandlerEventBegin_Default(PetscLogHandler h, PetscLogEvent event, PetscObject o1, PetscObject o2, PetscObject o3, PetscObject o4) 53678f1b9b4SToby Isaac { 53778f1b9b4SToby Isaac PetscLogHandler_Default def = (PetscLogHandler_Default)h->data; 53878f1b9b4SToby Isaac PetscEventPerfInfo *event_perf_info = NULL; 53978f1b9b4SToby Isaac PetscLogEventInfo event_info; 54078f1b9b4SToby Isaac PetscLogDouble time; 54178f1b9b4SToby Isaac PetscLogState state; 54278f1b9b4SToby Isaac PetscLogStage stage; 54378f1b9b4SToby Isaac 54478f1b9b4SToby Isaac PetscFunctionBegin; 54578f1b9b4SToby Isaac PetscCall(PetscLogHandlerGetState(h, &state)); 5465aefd447SStefano Zampini if (PetscDefined(USE_DEBUG)) { 5475aefd447SStefano Zampini PetscCall(PetscLogStateEventGetInfo(state, event, &event_info)); 5485aefd447SStefano Zampini if (PetscUnlikely(o1)) PetscValidHeader(o1, 3); 5495aefd447SStefano Zampini if (PetscUnlikely(o2)) PetscValidHeader(o2, 4); 5505aefd447SStefano Zampini if (PetscUnlikely(o3)) PetscValidHeader(o3, 5); 5515aefd447SStefano Zampini if (PetscUnlikely(o4)) PetscValidHeader(o4, 6); 5525aefd447SStefano Zampini if (event_info.collective && o1) { 5535aefd447SStefano Zampini PetscInt64 b1[2], b2[2]; 5545aefd447SStefano Zampini 5555aefd447SStefano Zampini b1[0] = -o1->cidx; 5565aefd447SStefano Zampini b1[1] = o1->cidx; 557462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(b1, b2, 2, MPIU_INT64, MPI_MAX, PetscObjectComm(o1))); 5589649dde3SJunchao Zhang PetscCheck(-b2[0] == b2[1], PETSC_COMM_SELF, PETSC_ERR_PLIB, "Collective event %s not called collectively %" PetscInt64_FMT " != %" PetscInt64_FMT, event_info.name, -b2[0], b2[1]); 5595aefd447SStefano Zampini } 5605aefd447SStefano Zampini } 56178f1b9b4SToby Isaac /* Synchronization */ 56278f1b9b4SToby Isaac PetscCall(PetscLogHandlerEventSync_Default(h, event, PetscObjectComm(o1))); 56378f1b9b4SToby Isaac PetscCall(PetscLogStateGetCurrentStage(state, &stage)); 56478f1b9b4SToby Isaac if (def->pause_depth > 0) stage = 0; // in pause-mode, all events run on the main stage 565461318ebSToby Isaac if (PetscDefined(HAVE_THREADSAFETY) || def->use_threadsafe) { 56678f1b9b4SToby Isaac PetscCall(PetscLogGetStageEventPerfInfo_threaded(def, stage, event, &event_perf_info)); 56778f1b9b4SToby Isaac if (event_perf_info->depth == 0) { PetscCall(PetscEventPerfInfoInit(event_perf_info)); } 56878f1b9b4SToby Isaac } else { 569dff009beSToby Isaac PetscCall(PetscLogHandlerGetEventPerfInfo_Default(h, stage, event, &event_perf_info)); 57078f1b9b4SToby Isaac } 57178f1b9b4SToby Isaac PetscCheck(event_perf_info->depth >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Trying to begin a paused event, this is not allowed"); 57278f1b9b4SToby Isaac event_perf_info->depth++; 57378f1b9b4SToby Isaac /* Check for double counting */ 57478f1b9b4SToby Isaac if (event_perf_info->depth > 1) PetscFunctionReturn(PETSC_SUCCESS); 57578f1b9b4SToby Isaac PetscCall(PetscLogStateEventGetInfo(state, event, &event_info)); 57678f1b9b4SToby Isaac /* Log the performance info */ 57778f1b9b4SToby Isaac event_perf_info->count++; 57878f1b9b4SToby Isaac PetscCall(PetscTime(&time)); 579835f2295SStefano Zampini PetscCall(PetscEventPerfInfoTic(event_perf_info, time, PetscLogMemory, event)); 58078f1b9b4SToby Isaac if (def->petsc_logActions) { 58178f1b9b4SToby Isaac PetscLogDouble curTime; 58278f1b9b4SToby Isaac Action new_action; 58378f1b9b4SToby Isaac 58478f1b9b4SToby Isaac PetscCall(PetscTime(&curTime)); 58578f1b9b4SToby Isaac new_action.time = curTime - petsc_BaseTime; 58678f1b9b4SToby Isaac new_action.action = PETSC_LOG_ACTION_BEGIN; 58778f1b9b4SToby Isaac new_action.event = event; 58878f1b9b4SToby Isaac new_action.classid = event_info.classid; 58978f1b9b4SToby Isaac new_action.id1 = o1 ? o1->id : -1; 59078f1b9b4SToby Isaac new_action.id2 = o2 ? o2->id : -1; 59178f1b9b4SToby Isaac new_action.id3 = o3 ? o3->id : -1; 59278f1b9b4SToby Isaac new_action.flops = petsc_TotalFlops; 59378f1b9b4SToby Isaac PetscCall(PetscMallocGetCurrentUsage(&new_action.mem)); 59478f1b9b4SToby Isaac PetscCall(PetscMallocGetMaximumUsage(&new_action.maxmem)); 59578f1b9b4SToby Isaac PetscCall(PetscLogActionArrayPush(def->petsc_actions, new_action)); 59678f1b9b4SToby Isaac } 59778f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 59878f1b9b4SToby Isaac } 59978f1b9b4SToby Isaac 60078f1b9b4SToby Isaac static PetscErrorCode PetscLogHandlerEventEnd_Default(PetscLogHandler h, PetscLogEvent event, PetscObject o1, PetscObject o2, PetscObject o3, PetscObject o4) 60178f1b9b4SToby Isaac { 60278f1b9b4SToby Isaac PetscLogHandler_Default def = (PetscLogHandler_Default)h->data; 60378f1b9b4SToby Isaac PetscEventPerfInfo *event_perf_info = NULL; 60478f1b9b4SToby Isaac PetscLogDouble time; 60578f1b9b4SToby Isaac PetscLogState state; 60678f1b9b4SToby Isaac int stage; 6075aefd447SStefano Zampini PetscLogEventInfo event_info; 60878f1b9b4SToby Isaac 60978f1b9b4SToby Isaac PetscFunctionBegin; 61078f1b9b4SToby Isaac PetscCall(PetscLogHandlerGetState(h, &state)); 6115aefd447SStefano Zampini if (PetscDefined(USE_DEBUG)) { 6125aefd447SStefano Zampini PetscCall(PetscLogStateEventGetInfo(state, event, &event_info)); 6135aefd447SStefano Zampini if (PetscUnlikely(o1)) PetscValidHeader(o1, 3); 6145aefd447SStefano Zampini if (PetscUnlikely(o2)) PetscValidHeader(o2, 4); 6155aefd447SStefano Zampini if (PetscUnlikely(o3)) PetscValidHeader(o3, 5); 6165aefd447SStefano Zampini if (PetscUnlikely(o4)) PetscValidHeader(o4, 6); 6175aefd447SStefano Zampini if (event_info.collective && o1) { 6185aefd447SStefano Zampini PetscInt64 b1[2], b2[2]; 6195aefd447SStefano Zampini 6205aefd447SStefano Zampini b1[0] = -o1->cidx; 6215aefd447SStefano Zampini b1[1] = o1->cidx; 622462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(b1, b2, 2, MPIU_INT64, MPI_MAX, PetscObjectComm(o1))); 6239649dde3SJunchao Zhang PetscCheck(-b2[0] == b2[1], PETSC_COMM_SELF, PETSC_ERR_PLIB, "Collective event %s not called collectively %" PetscInt64_FMT " != %" PetscInt64_FMT, event_info.name, -b2[0], b2[1]); 6245aefd447SStefano Zampini } 6255aefd447SStefano Zampini } 62678f1b9b4SToby Isaac if (def->petsc_logActions) { 62778f1b9b4SToby Isaac PetscLogDouble curTime; 62878f1b9b4SToby Isaac Action new_action; 62978f1b9b4SToby Isaac 63078f1b9b4SToby Isaac PetscCall(PetscLogStateEventGetInfo(state, event, &event_info)); 63178f1b9b4SToby Isaac PetscCall(PetscTime(&curTime)); 63278f1b9b4SToby Isaac new_action.time = curTime - petsc_BaseTime; 63378f1b9b4SToby Isaac new_action.action = PETSC_LOG_ACTION_END; 63478f1b9b4SToby Isaac new_action.event = event; 63578f1b9b4SToby Isaac new_action.classid = event_info.classid; 63678f1b9b4SToby Isaac new_action.id1 = o1 ? o1->id : -1; 63778f1b9b4SToby Isaac new_action.id2 = o2 ? o2->id : -2; 63878f1b9b4SToby Isaac new_action.id3 = o3 ? o3->id : -3; 63978f1b9b4SToby Isaac new_action.flops = petsc_TotalFlops; 64078f1b9b4SToby Isaac PetscCall(PetscMallocGetCurrentUsage(&new_action.mem)); 64178f1b9b4SToby Isaac PetscCall(PetscMallocGetMaximumUsage(&new_action.maxmem)); 64278f1b9b4SToby Isaac PetscCall(PetscLogActionArrayPush(def->petsc_actions, new_action)); 64378f1b9b4SToby Isaac } 64478f1b9b4SToby Isaac PetscCall(PetscLogStateGetCurrentStage(state, &stage)); 64578f1b9b4SToby Isaac if (def->pause_depth > 0) stage = 0; // all events run on the main stage in pause-mode 646461318ebSToby Isaac if (PetscDefined(HAVE_THREADSAFETY) || def->use_threadsafe) { 64778f1b9b4SToby Isaac PetscCall(PetscLogGetStageEventPerfInfo_threaded(def, stage, event, &event_perf_info)); 64878f1b9b4SToby Isaac } else { 649dff009beSToby Isaac PetscCall(PetscLogHandlerGetEventPerfInfo_Default(h, stage, event, &event_perf_info)); 65078f1b9b4SToby Isaac } 65178f1b9b4SToby Isaac PetscCheck(event_perf_info->depth > 0, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Trying to end paused event, not allowed"); 65278f1b9b4SToby Isaac event_perf_info->depth--; 65378f1b9b4SToby Isaac /* Check for double counting */ 65478f1b9b4SToby Isaac if (event_perf_info->depth > 0) PetscFunctionReturn(PETSC_SUCCESS); 65578f1b9b4SToby Isaac else PetscCheck(event_perf_info->depth == 0, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Logging event had unbalanced begin/end pairs"); 65678f1b9b4SToby Isaac 65778f1b9b4SToby Isaac /* Log performance info */ 65878f1b9b4SToby Isaac PetscCall(PetscTime(&time)); 659835f2295SStefano Zampini PetscCall(PetscEventPerfInfoToc(event_perf_info, time, PetscLogMemory, event)); 660461318ebSToby Isaac if (PetscDefined(HAVE_THREADSAFETY) || def->use_threadsafe) { 66178f1b9b4SToby Isaac PetscEventPerfInfo *event_perf_info_global; 66278f1b9b4SToby Isaac PetscCall(PetscSpinlockLock(&def->lock)); 663dff009beSToby Isaac PetscCall(PetscLogHandlerGetEventPerfInfo_Default(h, stage, event, &event_perf_info_global)); 66478f1b9b4SToby Isaac PetscCall(PetscEventPerfInfoAdd_Internal(event_perf_info, event_perf_info_global)); 66578f1b9b4SToby Isaac PetscCall(PetscSpinlockUnlock(&def->lock)); 66678f1b9b4SToby Isaac } 66778f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 66878f1b9b4SToby Isaac } 66978f1b9b4SToby Isaac 670dff009beSToby Isaac static PetscErrorCode PetscLogHandlerEventDeactivatePush_Default(PetscLogHandler h, PetscLogStage stage, PetscLogEvent event) 67178f1b9b4SToby Isaac { 67278f1b9b4SToby Isaac PetscEventPerfInfo *event_perf_info; 67378f1b9b4SToby Isaac 67478f1b9b4SToby Isaac PetscFunctionBegin; 67578f1b9b4SToby Isaac if (stage < 0) PetscCall(PetscLogStateGetCurrentStage(h->state, &stage)); 676dff009beSToby Isaac PetscCall(PetscLogHandlerGetEventPerfInfo_Default(h, stage, event, &event_perf_info)); 67778f1b9b4SToby Isaac event_perf_info->depth++; 67878f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 67978f1b9b4SToby Isaac } 68078f1b9b4SToby Isaac 681dff009beSToby Isaac static PetscErrorCode PetscLogHandlerEventDeactivatePop_Default(PetscLogHandler h, PetscLogStage stage, PetscLogEvent event) 68278f1b9b4SToby Isaac { 68378f1b9b4SToby Isaac PetscEventPerfInfo *event_perf_info; 68478f1b9b4SToby Isaac 68578f1b9b4SToby Isaac PetscFunctionBegin; 68678f1b9b4SToby Isaac if (stage < 0) PetscCall(PetscLogStateGetCurrentStage(h->state, &stage)); 687dff009beSToby Isaac PetscCall(PetscLogHandlerGetEventPerfInfo_Default(h, stage, event, &event_perf_info)); 68878f1b9b4SToby Isaac event_perf_info->depth--; 68978f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 69078f1b9b4SToby Isaac } 69178f1b9b4SToby Isaac 692dff009beSToby Isaac static PetscErrorCode PetscLogHandlerEventsPause_Default(PetscLogHandler h) 69378f1b9b4SToby Isaac { 69478f1b9b4SToby Isaac PetscLogHandler_Default def = (PetscLogHandler_Default)h->data; 69578f1b9b4SToby Isaac PetscLogDouble time; 69678f1b9b4SToby Isaac PetscInt num_stages; 69778f1b9b4SToby Isaac 69878f1b9b4SToby Isaac PetscFunctionBegin; 69978f1b9b4SToby Isaac if (def->pause_depth++ > 0) PetscFunctionReturn(PETSC_SUCCESS); 70078f1b9b4SToby Isaac PetscCall(PetscLogStageInfoArrayGetSize(def->stages, &num_stages, NULL)); 70178f1b9b4SToby Isaac PetscCall(PetscTime(&time)); 70278f1b9b4SToby Isaac for (PetscInt stage = 0; stage < num_stages; stage++) { 70378f1b9b4SToby Isaac PetscStagePerf *stage_info = NULL; 70478f1b9b4SToby Isaac PetscInt num_events; 70578f1b9b4SToby Isaac 70678f1b9b4SToby Isaac PetscCall(PetscLogStageInfoArrayGetRef(def->stages, stage, &stage_info)); 70778f1b9b4SToby Isaac PetscCall(PetscLogEventPerfArrayGetSize(stage_info->eventLog, &num_events, NULL)); 70878f1b9b4SToby Isaac for (PetscInt event = 0; event < num_events; event++) { 70978f1b9b4SToby Isaac PetscEventPerfInfo *event_info = NULL; 71078f1b9b4SToby Isaac PetscCall(PetscLogEventPerfArrayGetRef(stage_info->eventLog, event, &event_info)); 71178f1b9b4SToby Isaac if (event_info->depth > 0) { 71278f1b9b4SToby Isaac event_info->depth *= -1; 71378f1b9b4SToby Isaac PetscCall(PetscEventPerfInfoPause(event_info, time, PetscLogMemory, event)); 71478f1b9b4SToby Isaac } 71578f1b9b4SToby Isaac } 71678f1b9b4SToby Isaac if (stage > 0 && stage_info->perfInfo.depth > 0) { 71778f1b9b4SToby Isaac stage_info->perfInfo.depth *= -1; 71878f1b9b4SToby Isaac PetscCall(PetscEventPerfInfoPause(&stage_info->perfInfo, time, PetscLogMemory, -(stage + 2))); 71978f1b9b4SToby Isaac } 72078f1b9b4SToby Isaac } 72178f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 72278f1b9b4SToby Isaac } 72378f1b9b4SToby Isaac 724dff009beSToby Isaac static PetscErrorCode PetscLogHandlerEventsResume_Default(PetscLogHandler h) 72578f1b9b4SToby Isaac { 72678f1b9b4SToby Isaac PetscLogHandler_Default def = (PetscLogHandler_Default)h->data; 72778f1b9b4SToby Isaac PetscLogDouble time; 72878f1b9b4SToby Isaac PetscInt num_stages; 72978f1b9b4SToby Isaac 73078f1b9b4SToby Isaac PetscFunctionBegin; 73178f1b9b4SToby Isaac if (--def->pause_depth > 0) PetscFunctionReturn(PETSC_SUCCESS); 73278f1b9b4SToby Isaac PetscCall(PetscLogStageInfoArrayGetSize(def->stages, &num_stages, NULL)); 73378f1b9b4SToby Isaac PetscCall(PetscTime(&time)); 73478f1b9b4SToby Isaac for (PetscInt stage = 0; stage < num_stages; stage++) { 73578f1b9b4SToby Isaac PetscStagePerf *stage_info = NULL; 73678f1b9b4SToby Isaac PetscInt num_events; 73778f1b9b4SToby Isaac 73878f1b9b4SToby Isaac PetscCall(PetscLogStageInfoArrayGetRef(def->stages, stage, &stage_info)); 73978f1b9b4SToby Isaac PetscCall(PetscLogEventPerfArrayGetSize(stage_info->eventLog, &num_events, NULL)); 74078f1b9b4SToby Isaac for (PetscInt event = 0; event < num_events; event++) { 74178f1b9b4SToby Isaac PetscEventPerfInfo *event_info = NULL; 74278f1b9b4SToby Isaac PetscCall(PetscLogEventPerfArrayGetRef(stage_info->eventLog, event, &event_info)); 74378f1b9b4SToby Isaac if (event_info->depth < 0) { 74478f1b9b4SToby Isaac event_info->depth *= -1; 74578f1b9b4SToby Isaac PetscCall(PetscEventPerfInfoResume(event_info, time, PetscLogMemory, event)); 74678f1b9b4SToby Isaac } 74778f1b9b4SToby Isaac } 74878f1b9b4SToby Isaac if (stage > 0 && stage_info->perfInfo.depth < 0) { 74978f1b9b4SToby Isaac stage_info->perfInfo.depth *= -1; 75078f1b9b4SToby Isaac PetscCall(PetscEventPerfInfoResume(&stage_info->perfInfo, time, PetscLogMemory, -(stage + 2))); 75178f1b9b4SToby Isaac } 75278f1b9b4SToby Isaac } 75378f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 75478f1b9b4SToby Isaac } 75578f1b9b4SToby Isaac 75678f1b9b4SToby Isaac static PetscErrorCode PetscLogHandlerStagePush_Default(PetscLogHandler h, PetscLogStage new_stage) 75778f1b9b4SToby Isaac { 75878f1b9b4SToby Isaac PetscLogHandler_Default def = (PetscLogHandler_Default)h->data; 75978f1b9b4SToby Isaac PetscLogDouble time; 76078f1b9b4SToby Isaac PetscLogState state; 76178f1b9b4SToby Isaac PetscLogStage current_stage; 76278f1b9b4SToby Isaac PetscStagePerf *new_stage_info; 76378f1b9b4SToby Isaac 76478f1b9b4SToby Isaac PetscFunctionBegin; 76578f1b9b4SToby Isaac if (def->pause_depth > 0) PetscFunctionReturn(PETSC_SUCCESS); 76678f1b9b4SToby Isaac PetscCall(PetscLogHandlerGetState(h, &state)); 76778f1b9b4SToby Isaac current_stage = state->current_stage; 76878f1b9b4SToby Isaac PetscCall(PetscLogHandlerDefaultGetStageInfo(h, new_stage, &new_stage_info)); 76978f1b9b4SToby Isaac PetscCall(PetscTime(&time)); 77078f1b9b4SToby Isaac 77178f1b9b4SToby Isaac /* Record flops/time of previous stage */ 77278f1b9b4SToby Isaac if (current_stage >= 0) { 77378f1b9b4SToby Isaac if (PetscBTLookup(state->active, current_stage)) { 77478f1b9b4SToby Isaac PetscStagePerf *current_stage_info; 77578f1b9b4SToby Isaac PetscCall(PetscLogHandlerDefaultGetStageInfo(h, current_stage, ¤t_stage_info)); 776835f2295SStefano Zampini PetscCall(PetscEventPerfInfoToc(¤t_stage_info->perfInfo, time, PetscLogMemory, -(current_stage + 2))); 77778f1b9b4SToby Isaac } 77878f1b9b4SToby Isaac } 77978f1b9b4SToby Isaac new_stage_info->used = PETSC_TRUE; 78078f1b9b4SToby Isaac new_stage_info->perfInfo.count++; 78178f1b9b4SToby Isaac new_stage_info->perfInfo.depth++; 78278f1b9b4SToby Isaac /* Subtract current quantities so that we obtain the difference when we pop */ 783835f2295SStefano Zampini if (PetscBTLookup(state->active, new_stage)) PetscCall(PetscEventPerfInfoTic(&new_stage_info->perfInfo, time, PetscLogMemory, -(new_stage + 2))); 78478f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 78578f1b9b4SToby Isaac } 78678f1b9b4SToby Isaac 78778f1b9b4SToby Isaac static PetscErrorCode PetscLogHandlerStagePop_Default(PetscLogHandler h, PetscLogStage old_stage) 78878f1b9b4SToby Isaac { 78978f1b9b4SToby Isaac PetscLogHandler_Default def = (PetscLogHandler_Default)h->data; 79078f1b9b4SToby Isaac PetscLogStage current_stage; 79178f1b9b4SToby Isaac PetscStagePerf *old_stage_info; 79278f1b9b4SToby Isaac PetscLogState state; 79378f1b9b4SToby Isaac PetscLogDouble time; 79478f1b9b4SToby Isaac 79578f1b9b4SToby Isaac PetscFunctionBegin; 79678f1b9b4SToby Isaac if (def->pause_depth > 0) PetscFunctionReturn(PETSC_SUCCESS); 79778f1b9b4SToby Isaac PetscCall(PetscLogHandlerGetState(h, &state)); 79878f1b9b4SToby Isaac current_stage = state->current_stage; 79978f1b9b4SToby Isaac PetscCall(PetscLogHandlerDefaultGetStageInfo(h, old_stage, &old_stage_info)); 80078f1b9b4SToby Isaac PetscCall(PetscTime(&time)); 80178f1b9b4SToby Isaac old_stage_info->perfInfo.depth--; 802835f2295SStefano Zampini if (PetscBTLookup(state->active, old_stage)) { PetscCall(PetscEventPerfInfoToc(&old_stage_info->perfInfo, time, PetscLogMemory, -(old_stage + 2))); } 80378f1b9b4SToby Isaac if (current_stage >= 0) { 80478f1b9b4SToby Isaac if (PetscBTLookup(state->active, current_stage)) { 80578f1b9b4SToby Isaac PetscStagePerf *current_stage_info; 80678f1b9b4SToby Isaac PetscCall(PetscLogHandlerDefaultGetStageInfo(h, current_stage, ¤t_stage_info)); 807835f2295SStefano Zampini PetscCall(PetscEventPerfInfoTic(¤t_stage_info->perfInfo, time, PetscLogMemory, -(current_stage + 2))); 80878f1b9b4SToby Isaac } 80978f1b9b4SToby Isaac } 81078f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 81178f1b9b4SToby Isaac } 81278f1b9b4SToby Isaac 813dff009beSToby Isaac static PetscErrorCode PetscLogHandlerStageSetVisible_Default(PetscLogHandler h, PetscLogStage stage, PetscBool is_visible) 81478f1b9b4SToby Isaac { 81578f1b9b4SToby Isaac PetscStagePerf *stage_info; 81678f1b9b4SToby Isaac 81778f1b9b4SToby Isaac PetscFunctionBegin; 81878f1b9b4SToby Isaac if (stage < 0) PetscCall(PetscLogStateGetCurrentStage(h->state, &stage)); 81978f1b9b4SToby Isaac PetscCall(PetscLogHandlerDefaultGetStageInfo(h, stage, &stage_info)); 82078f1b9b4SToby Isaac stage_info->perfInfo.visible = is_visible; 82178f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 82278f1b9b4SToby Isaac } 82378f1b9b4SToby Isaac 824dff009beSToby Isaac static PetscErrorCode PetscLogHandlerStageGetVisible_Default(PetscLogHandler h, PetscLogStage stage, PetscBool *is_visible) 82578f1b9b4SToby Isaac { 82678f1b9b4SToby Isaac PetscStagePerf *stage_info; 82778f1b9b4SToby Isaac 82878f1b9b4SToby Isaac PetscFunctionBegin; 82978f1b9b4SToby Isaac if (stage < 0) PetscCall(PetscLogStateGetCurrentStage(h->state, &stage)); 83078f1b9b4SToby Isaac PetscCall(PetscLogHandlerDefaultGetStageInfo(h, stage, &stage_info)); 83178f1b9b4SToby Isaac *is_visible = stage_info->perfInfo.visible; 83278f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 83378f1b9b4SToby Isaac } 83478f1b9b4SToby Isaac 835dff009beSToby Isaac static PetscErrorCode PetscLogHandlerSetLogActions_Default(PetscLogHandler handler, PetscBool flag) 83678f1b9b4SToby Isaac { 83778f1b9b4SToby Isaac PetscLogHandler_Default def = (PetscLogHandler_Default)handler->data; 83878f1b9b4SToby Isaac 83978f1b9b4SToby Isaac PetscFunctionBegin; 84078f1b9b4SToby Isaac def->petsc_logActions = flag; 84178f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 84278f1b9b4SToby Isaac } 84378f1b9b4SToby Isaac 844dff009beSToby Isaac static PetscErrorCode PetscLogHandlerSetLogObjects_Default(PetscLogHandler handler, PetscBool flag) 84578f1b9b4SToby Isaac { 84678f1b9b4SToby Isaac PetscLogHandler_Default def = (PetscLogHandler_Default)handler->data; 84778f1b9b4SToby Isaac 84878f1b9b4SToby Isaac PetscFunctionBegin; 84978f1b9b4SToby Isaac def->petsc_logObjects = flag; 85078f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 85178f1b9b4SToby Isaac } 85278f1b9b4SToby Isaac 853dff009beSToby Isaac static PetscErrorCode PetscLogHandlerLogObjectState_Default(PetscLogHandler handler, PetscObject obj, const char format[], va_list Argp) 85478f1b9b4SToby Isaac { 85578f1b9b4SToby Isaac PetscLogHandler_Default def = (PetscLogHandler_Default)handler->data; 85678f1b9b4SToby Isaac size_t fullLength; 85778f1b9b4SToby Isaac 85878f1b9b4SToby Isaac PetscFunctionBegin; 85978f1b9b4SToby Isaac if (def->petsc_logObjects) { 86078f1b9b4SToby Isaac Object *obj_entry = NULL; 8616497c311SBarry Smith PetscInt objid; 86278f1b9b4SToby Isaac 8636497c311SBarry Smith PetscCall(PetscIntCast(obj->id - 1, &objid)); 8646497c311SBarry Smith PetscCall(PetscLogObjectArrayGetRef(def->petsc_objects, objid, &obj_entry)); 86578f1b9b4SToby Isaac PetscCall(PetscVSNPrintf(obj_entry->info, 64, format, &fullLength, Argp)); 86678f1b9b4SToby Isaac } 86778f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 86878f1b9b4SToby Isaac } 86978f1b9b4SToby Isaac 870dff009beSToby Isaac static PetscErrorCode PetscLogHandlerGetNumObjects_Default(PetscLogHandler handler, PetscInt *num_objects) 87178f1b9b4SToby Isaac { 87278f1b9b4SToby Isaac PetscLogHandler_Default def = (PetscLogHandler_Default)handler->data; 87378f1b9b4SToby Isaac 87478f1b9b4SToby Isaac PetscFunctionBegin; 87578f1b9b4SToby Isaac PetscCall(PetscLogObjectArrayGetSize(def->petsc_objects, num_objects, NULL)); 87678f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 87778f1b9b4SToby Isaac } 87878f1b9b4SToby Isaac 879dff009beSToby Isaac static PetscErrorCode PetscLogHandlerDump_Default(PetscLogHandler handler, const char sname[]) 88078f1b9b4SToby Isaac { 88178f1b9b4SToby Isaac PetscLogHandler_Default def = (PetscLogHandler_Default)handler->data; 88278f1b9b4SToby Isaac FILE *fd; 88378f1b9b4SToby Isaac char file[PETSC_MAX_PATH_LEN], fname[PETSC_MAX_PATH_LEN]; 88478f1b9b4SToby Isaac PetscLogDouble flops, _TotalTime; 88578f1b9b4SToby Isaac PetscMPIInt rank; 88678f1b9b4SToby Isaac int curStage; 88778f1b9b4SToby Isaac PetscLogState state; 88878f1b9b4SToby Isaac PetscInt num_events; 88978f1b9b4SToby Isaac PetscLogEvent event; 89078f1b9b4SToby Isaac 89178f1b9b4SToby Isaac PetscFunctionBegin; 89278f1b9b4SToby Isaac /* Calculate the total elapsed time */ 89378f1b9b4SToby Isaac PetscCall(PetscTime(&_TotalTime)); 89478f1b9b4SToby Isaac _TotalTime -= petsc_BaseTime; 89578f1b9b4SToby Isaac /* Open log file */ 89678f1b9b4SToby Isaac PetscCallMPI(MPI_Comm_rank(PetscObjectComm((PetscObject)handler), &rank)); 89778f1b9b4SToby Isaac PetscCall(PetscSNPrintf(file, PETSC_STATIC_ARRAY_LENGTH(file), "%s.%d", sname && sname[0] ? sname : "Log", rank)); 89878f1b9b4SToby Isaac PetscCall(PetscFixFilename(file, fname)); 89978f1b9b4SToby Isaac PetscCall(PetscFOpen(PETSC_COMM_SELF, fname, "w", &fd)); 90078f1b9b4SToby Isaac PetscCheck(!(rank == 0) || !(!fd), PETSC_COMM_SELF, PETSC_ERR_FILE_OPEN, "Cannot open file: %s", fname); 90178f1b9b4SToby Isaac /* Output totals */ 90278f1b9b4SToby Isaac PetscCall(PetscFPrintf(PETSC_COMM_SELF, fd, "Total Flop %14e %16.8e\n", petsc_TotalFlops, _TotalTime)); 90378f1b9b4SToby Isaac PetscCall(PetscFPrintf(PETSC_COMM_SELF, fd, "Clock Resolution %g\n", 0.0)); 90478f1b9b4SToby Isaac /* Output actions */ 90578f1b9b4SToby Isaac if (def->petsc_logActions) { 90678f1b9b4SToby Isaac PetscInt num_actions; 90778f1b9b4SToby Isaac PetscCall(PetscLogActionArrayGetSize(def->petsc_actions, &num_actions, NULL)); 908835f2295SStefano Zampini PetscCall(PetscFPrintf(PETSC_COMM_SELF, fd, "Actions accomplished %" PetscInt_FMT "\n", num_actions)); 90978f1b9b4SToby Isaac for (int a = 0; a < num_actions; a++) { 91078f1b9b4SToby Isaac Action *action; 91178f1b9b4SToby Isaac 91278f1b9b4SToby Isaac PetscCall(PetscLogActionArrayGetRef(def->petsc_actions, a, &action)); 913835f2295SStefano Zampini PetscCall(PetscFPrintf(PETSC_COMM_SELF, fd, "%g %d %d %d %" PetscInt64_FMT " %" PetscInt64_FMT " %" PetscInt64_FMT " %g %g %g\n", action->time, action->action, action->event, action->classid, action->id1, action->id2, action->id3, action->flops, 914835f2295SStefano Zampini action->mem, action->maxmem)); 91578f1b9b4SToby Isaac } 91678f1b9b4SToby Isaac } 91778f1b9b4SToby Isaac /* Output objects */ 91878f1b9b4SToby Isaac if (def->petsc_logObjects) { 91978f1b9b4SToby Isaac PetscInt num_objects; 92078f1b9b4SToby Isaac 92178f1b9b4SToby Isaac PetscCall(PetscLogObjectArrayGetSize(def->petsc_objects, &num_objects, NULL)); 92278f1b9b4SToby Isaac PetscCall(PetscFPrintf(PETSC_COMM_SELF, fd, "Objects created %d destroyed %d\n", def->petsc_numObjectsCreated, def->petsc_numObjectsDestroyed)); 92378f1b9b4SToby Isaac for (int o = 0; o < num_objects; o++) { 92478f1b9b4SToby Isaac Object *object = NULL; 92578f1b9b4SToby Isaac 92678f1b9b4SToby Isaac PetscCall(PetscLogObjectArrayGetRef(def->petsc_objects, o, &object)); 92778f1b9b4SToby Isaac if (object->parent != -1) continue; // object with this id wasn't logged, probably a PetscLogHandler 92878f1b9b4SToby Isaac PetscCall(PetscFPrintf(PETSC_COMM_SELF, fd, "Parent ID: %d Memory: %d\n", object->parent, (int)object->mem)); 92978f1b9b4SToby Isaac if (!object->name[0]) { 93078f1b9b4SToby Isaac PetscCall(PetscFPrintf(PETSC_COMM_SELF, fd, "No Name\n")); 93178f1b9b4SToby Isaac } else { 93278f1b9b4SToby Isaac PetscCall(PetscFPrintf(PETSC_COMM_SELF, fd, "Name: %s\n", object->name)); 93378f1b9b4SToby Isaac } 93478f1b9b4SToby Isaac if (!object->info[0]) { 93578f1b9b4SToby Isaac PetscCall(PetscFPrintf(PETSC_COMM_SELF, fd, "No Info\n")); 93678f1b9b4SToby Isaac } else { 93778f1b9b4SToby Isaac PetscCall(PetscFPrintf(PETSC_COMM_SELF, fd, "Info: %s\n", object->info)); 93878f1b9b4SToby Isaac } 93978f1b9b4SToby Isaac } 94078f1b9b4SToby Isaac } 94178f1b9b4SToby Isaac /* Output events */ 94278f1b9b4SToby Isaac PetscCall(PetscFPrintf(PETSC_COMM_SELF, fd, "Event log:\n")); 94378f1b9b4SToby Isaac PetscCall(PetscLogHandlerGetState(handler, &state)); 94478f1b9b4SToby Isaac PetscCall(PetscLogStateGetNumEvents(state, &num_events)); 94578f1b9b4SToby Isaac PetscCall(PetscLogStateGetCurrentStage(state, &curStage)); 94678f1b9b4SToby Isaac for (event = 0; event < num_events; event++) { 94778f1b9b4SToby Isaac PetscEventPerfInfo *event_info; 94878f1b9b4SToby Isaac 949dff009beSToby Isaac PetscCall(PetscLogHandlerGetEventPerfInfo_Default(handler, curStage, event, &event_info)); 95078f1b9b4SToby Isaac if (event_info->time != 0.0) flops = event_info->flops / event_info->time; 95178f1b9b4SToby Isaac else flops = 0.0; 95278f1b9b4SToby Isaac PetscCall(PetscFPrintf(PETSC_COMM_SELF, fd, "%d %16d %16g %16g %16g\n", event, event_info->count, event_info->flops, event_info->time, flops)); 95378f1b9b4SToby Isaac } 95478f1b9b4SToby Isaac PetscCall(PetscFClose(PETSC_COMM_SELF, fd)); 95578f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 95678f1b9b4SToby Isaac } 95778f1b9b4SToby Isaac 95878f1b9b4SToby Isaac /* 95978f1b9b4SToby Isaac PetscLogView_Detailed - Each process prints the times for its own events 96078f1b9b4SToby Isaac 96178f1b9b4SToby Isaac */ 96278f1b9b4SToby Isaac static PetscErrorCode PetscLogHandlerView_Default_Detailed(PetscLogHandler handler, PetscViewer viewer) 96378f1b9b4SToby Isaac { 96478f1b9b4SToby Isaac PetscLogHandler_Default def = (PetscLogHandler_Default)handler->data; 96578f1b9b4SToby Isaac PetscLogDouble locTotalTime, numRed, maxMem; 96678f1b9b4SToby Isaac PetscInt numStages, numEvents; 96778f1b9b4SToby Isaac MPI_Comm comm = PetscObjectComm((PetscObject)viewer); 96878f1b9b4SToby Isaac PetscMPIInt rank, size; 96978f1b9b4SToby Isaac PetscLogGlobalNames global_stages, global_events; 97078f1b9b4SToby Isaac PetscLogState state; 97178f1b9b4SToby Isaac PetscEventPerfInfo zero_info; 97278f1b9b4SToby Isaac 97378f1b9b4SToby Isaac PetscFunctionBegin; 97478f1b9b4SToby Isaac PetscCall(PetscLogHandlerGetState(handler, &state)); 97578f1b9b4SToby Isaac PetscCallMPI(MPI_Comm_size(comm, &size)); 97678f1b9b4SToby Isaac PetscCallMPI(MPI_Comm_rank(comm, &rank)); 97778f1b9b4SToby Isaac /* Must preserve reduction count before we go on */ 97878f1b9b4SToby Isaac numRed = petsc_allreduce_ct + petsc_gather_ct + petsc_scatter_ct; 97978f1b9b4SToby Isaac /* Get the total elapsed time */ 98078f1b9b4SToby Isaac PetscCall(PetscTime(&locTotalTime)); 98178f1b9b4SToby Isaac locTotalTime -= petsc_BaseTime; 98278f1b9b4SToby Isaac PetscCall(PetscViewerASCIIPrintf(viewer, "size = %d\n", size)); 98378f1b9b4SToby Isaac PetscCall(PetscViewerASCIIPrintf(viewer, "LocalTimes = {}\n")); 98478f1b9b4SToby Isaac PetscCall(PetscViewerASCIIPrintf(viewer, "LocalMessages = {}\n")); 98578f1b9b4SToby Isaac PetscCall(PetscViewerASCIIPrintf(viewer, "LocalMessageLens = {}\n")); 98678f1b9b4SToby Isaac PetscCall(PetscViewerASCIIPrintf(viewer, "LocalReductions = {}\n")); 98778f1b9b4SToby Isaac PetscCall(PetscViewerASCIIPrintf(viewer, "LocalFlop = {}\n")); 98878f1b9b4SToby Isaac PetscCall(PetscViewerASCIIPrintf(viewer, "LocalObjects = {}\n")); 98978f1b9b4SToby Isaac PetscCall(PetscViewerASCIIPrintf(viewer, "LocalMemory = {}\n")); 99078f1b9b4SToby Isaac PetscCall(PetscLogRegistryCreateGlobalStageNames(comm, state->registry, &global_stages)); 99178f1b9b4SToby Isaac PetscCall(PetscLogRegistryCreateGlobalEventNames(comm, state->registry, &global_events)); 99278f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGetSize(global_stages, NULL, &numStages)); 99378f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGetSize(global_events, NULL, &numEvents)); 99478f1b9b4SToby Isaac PetscCall(PetscMemzero(&zero_info, sizeof(zero_info))); 99578f1b9b4SToby Isaac PetscCall(PetscViewerASCIIPrintf(viewer, "Stages = {}\n")); 99678f1b9b4SToby Isaac for (PetscInt stage = 0; stage < numStages; stage++) { 99778f1b9b4SToby Isaac PetscInt stage_id; 99878f1b9b4SToby Isaac const char *stage_name; 99978f1b9b4SToby Isaac 100078f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetLocal(global_stages, stage, &stage_id)); 100178f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetName(global_stages, stage, &stage_name)); 100278f1b9b4SToby Isaac PetscCall(PetscViewerASCIIPrintf(viewer, "Stages[\"%s\"] = {}\n", stage_name)); 100378f1b9b4SToby Isaac PetscCall(PetscViewerASCIIPrintf(viewer, "Stages[\"%s\"][\"summary\"] = {}\n", stage_name)); 100478f1b9b4SToby Isaac for (PetscInt event = 0; event < numEvents; event++) { 100578f1b9b4SToby Isaac PetscEventPerfInfo *eventInfo = &zero_info; 100678f1b9b4SToby Isaac PetscBool is_zero = PETSC_FALSE; 100778f1b9b4SToby Isaac PetscInt event_id; 100878f1b9b4SToby Isaac const char *event_name; 100978f1b9b4SToby Isaac 101078f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetLocal(global_events, event, &event_id)); 101178f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetName(global_events, event, &event_name)); 1012dff009beSToby Isaac if (event_id >= 0 && stage_id >= 0) PetscCall(PetscLogHandlerGetEventPerfInfo_Default(handler, stage_id, event_id, &eventInfo)); 101378f1b9b4SToby Isaac is_zero = eventInfo->count == 0 ? PETSC_TRUE : PETSC_FALSE; 1014462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(MPI_IN_PLACE, &is_zero, 1, MPIU_BOOL, MPI_LAND, comm)); 101578f1b9b4SToby Isaac if (!is_zero) { PetscCall(PetscViewerASCIIPrintf(viewer, "Stages[\"%s\"][\"%s\"] = {}\n", stage_name, event_name)); } 101678f1b9b4SToby Isaac } 101778f1b9b4SToby Isaac } 101878f1b9b4SToby Isaac PetscCall(PetscMallocGetMaximumUsage(&maxMem)); 101978f1b9b4SToby Isaac PetscCall(PetscViewerASCIIPushSynchronized(viewer)); 102078f1b9b4SToby Isaac PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "LocalTimes[%d] = %g\n", rank, locTotalTime)); 102157508eceSPierre Jolivet PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "LocalMessages[%d] = %g\n", rank, petsc_irecv_ct + petsc_isend_ct + petsc_recv_ct + petsc_send_ct)); 102257508eceSPierre Jolivet PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "LocalMessageLens[%d] = %g\n", rank, petsc_irecv_len + petsc_isend_len + petsc_recv_len + petsc_send_len)); 102378f1b9b4SToby Isaac PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "LocalReductions[%d] = %g\n", rank, numRed)); 102478f1b9b4SToby Isaac PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "LocalFlop[%d] = %g\n", rank, petsc_TotalFlops)); 102578f1b9b4SToby Isaac { 102678f1b9b4SToby Isaac PetscInt num_objects; 102778f1b9b4SToby Isaac 102878f1b9b4SToby Isaac PetscCall(PetscLogObjectArrayGetSize(def->petsc_objects, &num_objects, NULL)); 1029835f2295SStefano Zampini PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "LocalObjects[%d] = %" PetscInt_FMT "\n", rank, num_objects)); 103078f1b9b4SToby Isaac } 103178f1b9b4SToby Isaac PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "LocalMemory[%d] = %g\n", rank, maxMem)); 103278f1b9b4SToby Isaac PetscCall(PetscViewerFlush(viewer)); 103378f1b9b4SToby Isaac for (PetscInt stage = 0; stage < numStages; stage++) { 103478f1b9b4SToby Isaac PetscEventPerfInfo *stage_perf_info = &zero_info; 103578f1b9b4SToby Isaac PetscInt stage_id; 103678f1b9b4SToby Isaac const char *stage_name; 103778f1b9b4SToby Isaac 103878f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetLocal(global_stages, stage, &stage_id)); 103978f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetName(global_stages, stage, &stage_name)); 104078f1b9b4SToby Isaac if (stage_id >= 0) { 104178f1b9b4SToby Isaac PetscStagePerf *stage_info; 104278f1b9b4SToby Isaac PetscCall(PetscLogHandlerDefaultGetStageInfo(handler, stage_id, &stage_info)); 104378f1b9b4SToby Isaac stage_perf_info = &stage_info->perfInfo; 104478f1b9b4SToby Isaac } 104578f1b9b4SToby Isaac PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "Stages[\"%s\"][\"summary\"][%d] = {\"time\" : %g, \"numMessages\" : %g, \"messageLength\" : %g, \"numReductions\" : %g, \"flop\" : %g}\n", stage_name, rank, stage_perf_info->time, 104678f1b9b4SToby Isaac stage_perf_info->numMessages, stage_perf_info->messageLength, stage_perf_info->numReductions, stage_perf_info->flops)); 104778f1b9b4SToby Isaac for (PetscInt event = 0; event < numEvents; event++) { 104878f1b9b4SToby Isaac PetscEventPerfInfo *eventInfo = &zero_info; 104978f1b9b4SToby Isaac PetscBool is_zero = PETSC_FALSE; 105078f1b9b4SToby Isaac PetscInt event_id; 105178f1b9b4SToby Isaac const char *event_name; 105278f1b9b4SToby Isaac 105378f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetLocal(global_events, event, &event_id)); 105478f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetName(global_events, event, &event_name)); 1055dff009beSToby Isaac if (event_id >= 0 && stage_id >= 0) PetscCall(PetscLogHandlerGetEventPerfInfo_Default(handler, stage_id, event_id, &eventInfo)); 105678f1b9b4SToby Isaac is_zero = eventInfo->count == 0 ? PETSC_TRUE : PETSC_FALSE; 105778f1b9b4SToby Isaac PetscCall(PetscMemcmp(eventInfo, &zero_info, sizeof(zero_info), &is_zero)); 1058462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(MPI_IN_PLACE, &is_zero, 1, MPIU_BOOL, MPI_LAND, comm)); 105978f1b9b4SToby Isaac if (!is_zero) { 106078f1b9b4SToby Isaac PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "Stages[\"%s\"][\"%s\"][%d] = {\"count\" : %d, \"time\" : %g, \"syncTime\" : %g, \"numMessages\" : %g, \"messageLength\" : %g, \"numReductions\" : %g, \"flop\" : %g", stage_name, event_name, rank, 106178f1b9b4SToby Isaac eventInfo->count, eventInfo->time, eventInfo->syncTime, eventInfo->numMessages, eventInfo->messageLength, eventInfo->numReductions, eventInfo->flops)); 106278f1b9b4SToby Isaac if (eventInfo->dof[0] >= 0.) { 106378f1b9b4SToby Isaac PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, ", \"dof\" : [")); 106478f1b9b4SToby Isaac for (PetscInt d = 0; d < 8; ++d) { 106578f1b9b4SToby Isaac if (d > 0) PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, ", ")); 106678f1b9b4SToby Isaac PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "%g", eventInfo->dof[d])); 106778f1b9b4SToby Isaac } 106878f1b9b4SToby Isaac PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "]")); 106978f1b9b4SToby Isaac PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, ", \"error\" : [")); 107078f1b9b4SToby Isaac for (PetscInt e = 0; e < 8; ++e) { 107178f1b9b4SToby Isaac if (e > 0) PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, ", ")); 107278f1b9b4SToby Isaac PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "%g", eventInfo->errors[e])); 107378f1b9b4SToby Isaac } 107478f1b9b4SToby Isaac PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "]")); 107578f1b9b4SToby Isaac } 107678f1b9b4SToby Isaac } 107778f1b9b4SToby Isaac PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "}\n")); 107878f1b9b4SToby Isaac } 107978f1b9b4SToby Isaac } 108078f1b9b4SToby Isaac PetscCall(PetscViewerFlush(viewer)); 108178f1b9b4SToby Isaac PetscCall(PetscViewerASCIIPopSynchronized(viewer)); 108278f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesDestroy(&global_events)); 108378f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesDestroy(&global_stages)); 108478f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 108578f1b9b4SToby Isaac } 108678f1b9b4SToby Isaac 108778f1b9b4SToby Isaac /* 108878f1b9b4SToby Isaac PetscLogView_CSV - Each process prints the times for its own events in Comma-Separated Value Format 108978f1b9b4SToby Isaac */ 109078f1b9b4SToby Isaac static PetscErrorCode PetscLogHandlerView_Default_CSV(PetscLogHandler handler, PetscViewer viewer) 109178f1b9b4SToby Isaac { 109278f1b9b4SToby Isaac PetscLogDouble locTotalTime, maxMem; 109378f1b9b4SToby Isaac PetscInt numStages, numEvents, stage, event; 109478f1b9b4SToby Isaac MPI_Comm comm = PetscObjectComm((PetscObject)viewer); 109578f1b9b4SToby Isaac PetscMPIInt rank, size; 109678f1b9b4SToby Isaac PetscLogGlobalNames global_stages, global_events; 109778f1b9b4SToby Isaac PetscLogState state; 109878f1b9b4SToby Isaac PetscEventPerfInfo zero_info; 109978f1b9b4SToby Isaac 110078f1b9b4SToby Isaac PetscFunctionBegin; 110178f1b9b4SToby Isaac PetscCall(PetscLogHandlerGetState(handler, &state)); 110278f1b9b4SToby Isaac PetscCallMPI(MPI_Comm_size(comm, &size)); 110378f1b9b4SToby Isaac PetscCallMPI(MPI_Comm_rank(comm, &rank)); 110478f1b9b4SToby Isaac /* Must preserve reduction count before we go on */ 110578f1b9b4SToby Isaac /* Get the total elapsed time */ 110678f1b9b4SToby Isaac PetscCall(PetscTime(&locTotalTime)); 110778f1b9b4SToby Isaac locTotalTime -= petsc_BaseTime; 110878f1b9b4SToby Isaac PetscCall(PetscMallocGetMaximumUsage(&maxMem)); 110978f1b9b4SToby Isaac PetscCall(PetscLogRegistryCreateGlobalStageNames(comm, state->registry, &global_stages)); 111078f1b9b4SToby Isaac PetscCall(PetscLogRegistryCreateGlobalEventNames(comm, state->registry, &global_events)); 111178f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGetSize(global_stages, NULL, &numStages)); 111278f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGetSize(global_events, NULL, &numEvents)); 111378f1b9b4SToby Isaac PetscCall(PetscMemzero(&zero_info, sizeof(zero_info))); 111478f1b9b4SToby Isaac PetscCall(PetscViewerASCIIPushSynchronized(viewer)); 111578f1b9b4SToby Isaac PetscCall(PetscViewerASCIIPrintf(viewer, "Stage Name,Event Name,Rank,Count,Time,Num Messages,Message Length,Num Reductions,FLOP,dof0,dof1,dof2,dof3,dof4,dof5,dof6,dof7,e0,e1,e2,e3,e4,e5,e6,e7,%d\n", size)); 111678f1b9b4SToby Isaac PetscCall(PetscViewerFlush(viewer)); 111778f1b9b4SToby Isaac for (stage = 0; stage < numStages; stage++) { 111878f1b9b4SToby Isaac PetscEventPerfInfo *stage_perf_info; 111978f1b9b4SToby Isaac PetscInt stage_id; 112078f1b9b4SToby Isaac const char *stage_name; 112178f1b9b4SToby Isaac 112278f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetLocal(global_stages, stage, &stage_id)); 112378f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetName(global_stages, stage, &stage_name)); 112478f1b9b4SToby Isaac stage_perf_info = &zero_info; 112578f1b9b4SToby Isaac if (stage_id >= 0) { 112678f1b9b4SToby Isaac PetscStagePerf *stage_info; 112778f1b9b4SToby Isaac PetscCall(PetscLogHandlerDefaultGetStageInfo(handler, stage_id, &stage_info)); 112878f1b9b4SToby Isaac stage_perf_info = &stage_info->perfInfo; 112978f1b9b4SToby Isaac } 113078f1b9b4SToby Isaac PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "%s,summary,%d,1,%g,%g,%g,%g,%g\n", stage_name, rank, stage_perf_info->time, stage_perf_info->numMessages, stage_perf_info->messageLength, stage_perf_info->numReductions, stage_perf_info->flops)); 113178f1b9b4SToby Isaac for (event = 0; event < numEvents; event++) { 113278f1b9b4SToby Isaac PetscEventPerfInfo *eventInfo = &zero_info; 113378f1b9b4SToby Isaac PetscBool is_zero = PETSC_FALSE; 113478f1b9b4SToby Isaac PetscInt event_id; 113578f1b9b4SToby Isaac const char *event_name; 113678f1b9b4SToby Isaac 113778f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetLocal(global_events, event, &event_id)); 113878f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetName(global_events, event, &event_name)); 1139dff009beSToby Isaac if (event_id >= 0 && stage_id >= 0) PetscCall(PetscLogHandlerGetEventPerfInfo_Default(handler, stage_id, event_id, &eventInfo)); 114078f1b9b4SToby Isaac PetscCall(PetscMemcmp(eventInfo, &zero_info, sizeof(zero_info), &is_zero)); 1141462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(MPI_IN_PLACE, &is_zero, 1, MPIU_BOOL, MPI_LAND, comm)); 114278f1b9b4SToby Isaac if (!is_zero) { 114378f1b9b4SToby Isaac PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "%s,%s,%d,%d,%g,%g,%g,%g,%g", stage_name, event_name, rank, eventInfo->count, eventInfo->time, eventInfo->numMessages, eventInfo->messageLength, eventInfo->numReductions, eventInfo->flops)); 114478f1b9b4SToby Isaac if (eventInfo->dof[0] >= 0.) { 114578f1b9b4SToby Isaac for (PetscInt d = 0; d < 8; ++d) PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, ",%g", eventInfo->dof[d])); 114678f1b9b4SToby Isaac for (PetscInt e = 0; e < 8; ++e) PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, ",%g", eventInfo->errors[e])); 114778f1b9b4SToby Isaac } 114878f1b9b4SToby Isaac } 114978f1b9b4SToby Isaac PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "\n")); 115078f1b9b4SToby Isaac } 115178f1b9b4SToby Isaac } 115278f1b9b4SToby Isaac PetscCall(PetscViewerFlush(viewer)); 115378f1b9b4SToby Isaac PetscCall(PetscViewerASCIIPopSynchronized(viewer)); 115478f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesDestroy(&global_stages)); 115578f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesDestroy(&global_events)); 115678f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 115778f1b9b4SToby Isaac } 115878f1b9b4SToby Isaac 11591d031f67SBarry Smith static PetscErrorCode PetscLogViewWarnSync(PetscViewer viewer) 116078f1b9b4SToby Isaac { 116178f1b9b4SToby Isaac PetscFunctionBegin; 116278f1b9b4SToby Isaac if (!PetscLogSyncOn) PetscFunctionReturn(PETSC_SUCCESS); 11631d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "\n\n")); 11641d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " ##########################################################\n")); 11651d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # #\n")); 11661d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # WARNING!!! #\n")); 11671d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # #\n")); 11681d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # This program was run with logging synchronization. #\n")); 11691d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # This option provides more meaningful imbalance #\n")); 11701d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # figures at the expense of slowing things down and #\n")); 11711d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # providing a distorted view of the overall runtime. #\n")); 11721d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # #\n")); 11731d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " ##########################################################\n\n\n")); 117478f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 117578f1b9b4SToby Isaac } 117678f1b9b4SToby Isaac 11771d031f67SBarry Smith static PetscErrorCode PetscLogViewWarnDebugging(PetscViewer viewer) 117878f1b9b4SToby Isaac { 117978f1b9b4SToby Isaac PetscFunctionBegin; 118078f1b9b4SToby Isaac if (PetscDefined(USE_DEBUG)) { 11811d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "\n\n")); 11821d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " ##########################################################\n")); 11831d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # #\n")); 11841d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # WARNING!!! #\n")); 11851d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # #\n")); 11861d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # This code was compiled with a debugging option. #\n")); 11871d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # To get timing results run ./configure #\n")); 11881d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # using --with-debugging=no, the performance will #\n")); 11891d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # be generally two or three times faster. #\n")); 11901d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # #\n")); 11911d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " ##########################################################\n\n\n")); 119278f1b9b4SToby Isaac } 119378f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 119478f1b9b4SToby Isaac } 119578f1b9b4SToby Isaac 11961d031f67SBarry Smith static PetscErrorCode PetscLogViewWarnNoGpuAwareMpi(PetscViewer viewer) 119778f1b9b4SToby Isaac { 119878f1b9b4SToby Isaac #if defined(PETSC_HAVE_DEVICE) 119978f1b9b4SToby Isaac PetscMPIInt size; 120078f1b9b4SToby Isaac PetscBool deviceInitialized = PETSC_FALSE; 120178f1b9b4SToby Isaac 120278f1b9b4SToby Isaac PetscFunctionBegin; 1203c621c6acSBarry Smith PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)viewer), &size)); 120478f1b9b4SToby Isaac for (int i = PETSC_DEVICE_HOST + 1; i < PETSC_DEVICE_MAX; ++i) { 120578f1b9b4SToby Isaac const PetscDeviceType dtype = PetscDeviceTypeCast(i); 120678f1b9b4SToby Isaac if (PetscDeviceInitialized(dtype)) { /* a non-host device was initialized */ 120778f1b9b4SToby Isaac deviceInitialized = PETSC_TRUE; 120878f1b9b4SToby Isaac break; 120978f1b9b4SToby Isaac } 121078f1b9b4SToby Isaac } 1211f0b74427SPierre Jolivet /* the last condition says PETSc is configured with device but it is a pure CPU run, so don't print misleading warnings */ 121278f1b9b4SToby Isaac if (use_gpu_aware_mpi || size == 1 || !deviceInitialized) PetscFunctionReturn(PETSC_SUCCESS); 12131d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "\n\n")); 12141d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " ##########################################################\n")); 12151d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # #\n")); 12161d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # WARNING!!! #\n")); 12171d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # #\n")); 12181d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # This code was compiled with GPU support and you've #\n")); 12191d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # created PETSc/GPU objects, but you intentionally #\n")); 12201d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # used -use_gpu_aware_mpi 0, requiring PETSc to copy #\n")); 12211d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # additional data between the GPU and CPU. To obtain #\n")); 12221d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # meaningful timing results on multi-rank runs, use #\n")); 12231d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # GPU-aware MPI instead. #\n")); 12241d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # #\n")); 12251d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " ##########################################################\n\n\n")); 122678f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 122778f1b9b4SToby Isaac #else 122878f1b9b4SToby Isaac return PETSC_SUCCESS; 122978f1b9b4SToby Isaac #endif 123078f1b9b4SToby Isaac } 123178f1b9b4SToby Isaac 12321d031f67SBarry Smith static PetscErrorCode PetscLogViewWarnGpuTime(PetscViewer viewer) 123378f1b9b4SToby Isaac { 12347d766218SJunchao Zhang #if defined(PETSC_HAVE_DEVICE) && !defined(PETSC_HAVE_KOKKOS_WITHOUT_GPU) 123578f1b9b4SToby Isaac 123678f1b9b4SToby Isaac PetscFunctionBegin; 123778f1b9b4SToby Isaac if (!PetscLogGpuTimeFlag || petsc_gflops == 0) PetscFunctionReturn(PETSC_SUCCESS); 12381d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "\n\n")); 12391d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " ##########################################################\n")); 12401d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # #\n")); 12411d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # WARNING!!! #\n")); 12421d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # #\n")); 12431d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # This code was run with -log_view_gpu_time #\n")); 12441d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # This provides accurate timing within the GPU kernels #\n")); 12451d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # but can slow down the entire computation by a #\n")); 12461d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # measurable amount. For fastest runs we recommend #\n")); 12471d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # not using this option. #\n")); 12481d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " # #\n")); 12491d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " ##########################################################\n\n\n")); 125078f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 125178f1b9b4SToby Isaac #else 125278f1b9b4SToby Isaac return PETSC_SUCCESS; 125378f1b9b4SToby Isaac #endif 125478f1b9b4SToby Isaac } 125578f1b9b4SToby Isaac 125678f1b9b4SToby Isaac static PetscErrorCode PetscLogHandlerView_Default_Info(PetscLogHandler handler, PetscViewer viewer) 125778f1b9b4SToby Isaac { 125878f1b9b4SToby Isaac PetscLogHandler_Default def = (PetscLogHandler_Default)handler->data; 125978f1b9b4SToby Isaac char arch[128], hostname[128], username[128], pname[PETSC_MAX_PATH_LEN], date[128]; 126078f1b9b4SToby Isaac PetscLogDouble locTotalTime, TotalTime, TotalFlops; 126178f1b9b4SToby Isaac PetscLogDouble numMessages, messageLength, avgMessLen, numReductions; 126278f1b9b4SToby Isaac PetscLogDouble stageTime, flops, flopr, mem, mess, messLen, red; 126378f1b9b4SToby Isaac PetscLogDouble fracTime, fracFlops, fracMessages, fracLength, fracReductions, fracMess, fracMessLen, fracRed; 126478f1b9b4SToby Isaac PetscLogDouble fracStageTime, fracStageFlops, fracStageMess, fracStageMessLen, fracStageRed; 126578f1b9b4SToby Isaac PetscLogDouble min, max, tot, ratio, avg, x, y; 126678f1b9b4SToby Isaac PetscLogDouble minf, maxf, totf, ratf, mint, maxt, tott, ratt, ratC, totm, totml, totr, mal, malmax, emalmax; 126778f1b9b4SToby Isaac #if defined(PETSC_HAVE_DEVICE) 126878f1b9b4SToby Isaac PetscLogEvent KSP_Solve, SNES_Solve, TS_Step, TAO_Solve; /* These need to be fixed to be some events registered with certain objects */ 126978f1b9b4SToby Isaac PetscLogDouble cct, gct, csz, gsz, gmaxt, gflops, gflopr, fracgflops; 127078f1b9b4SToby Isaac #endif 127178f1b9b4SToby Isaac PetscMPIInt minC, maxC; 127278f1b9b4SToby Isaac PetscMPIInt size, rank; 127378f1b9b4SToby Isaac PetscBool *localStageUsed, *stageUsed; 127478f1b9b4SToby Isaac PetscBool *localStageVisible, *stageVisible; 127578f1b9b4SToby Isaac PetscInt numStages, numEvents; 127678f1b9b4SToby Isaac int stage, oclass; 127778f1b9b4SToby Isaac PetscLogEvent event; 127878f1b9b4SToby Isaac char version[256]; 127978f1b9b4SToby Isaac MPI_Comm comm; 12807d766218SJunchao Zhang #if defined(PETSC_HAVE_DEVICE) && !defined(PETSC_HAVE_KOKKOS_WITHOUT_GPU) 128178f1b9b4SToby Isaac PetscInt64 nas = 0x7FF0000000000002; 128278f1b9b4SToby Isaac #endif 128378f1b9b4SToby Isaac PetscLogGlobalNames global_stages, global_events; 128478f1b9b4SToby Isaac PetscEventPerfInfo zero_info; 128578f1b9b4SToby Isaac PetscLogState state; 128678f1b9b4SToby Isaac 128778f1b9b4SToby Isaac PetscFunctionBegin; 128878f1b9b4SToby Isaac PetscCall(PetscLogHandlerGetState(handler, &state)); 128978f1b9b4SToby Isaac PetscCall(PetscFPTrapPush(PETSC_FP_TRAP_OFF)); 129078f1b9b4SToby Isaac PetscCall(PetscObjectGetComm((PetscObject)viewer, &comm)); 129178f1b9b4SToby Isaac PetscCallMPI(MPI_Comm_size(comm, &size)); 129278f1b9b4SToby Isaac PetscCallMPI(MPI_Comm_rank(comm, &rank)); 129378f1b9b4SToby Isaac /* Get the total elapsed time */ 129478f1b9b4SToby Isaac PetscCall(PetscTime(&locTotalTime)); 129578f1b9b4SToby Isaac locTotalTime -= petsc_BaseTime; 129678f1b9b4SToby Isaac 12971d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "****************************************************************************************************************************************************************\n")); 12981d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "*** WIDEN YOUR WINDOW TO 160 CHARACTERS. Use 'enscript -r -fCourier9' to print this document ***\n")); 12991d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "****************************************************************************************************************************************************************\n")); 13001d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "\n------------------------------------------------------------------ PETSc Performance Summary: ------------------------------------------------------------------\n\n")); 13011d031f67SBarry Smith PetscCall(PetscLogViewWarnSync(viewer)); 13021d031f67SBarry Smith PetscCall(PetscLogViewWarnDebugging(viewer)); 13031d031f67SBarry Smith PetscCall(PetscLogViewWarnNoGpuAwareMpi(viewer)); 13041d031f67SBarry Smith PetscCall(PetscLogViewWarnGpuTime(viewer)); 130578f1b9b4SToby Isaac PetscCall(PetscGetArchType(arch, sizeof(arch))); 130678f1b9b4SToby Isaac PetscCall(PetscGetHostName(hostname, sizeof(hostname))); 130778f1b9b4SToby Isaac PetscCall(PetscGetUserName(username, sizeof(username))); 130878f1b9b4SToby Isaac PetscCall(PetscGetProgramName(pname, sizeof(pname))); 130978f1b9b4SToby Isaac PetscCall(PetscGetDate(date, sizeof(date))); 131078f1b9b4SToby Isaac PetscCall(PetscGetVersion(version, sizeof(version))); 1311456dbbcdSJunchao Zhang 1312456dbbcdSJunchao Zhang #if defined(PETSC_HAVE_CUPM) 1313456dbbcdSJunchao Zhang const char *cupm = PetscDefined(HAVE_CUDA) ? "CUDA" : "HIP"; 13145c78b439SPierre Jolivet if (PetscDeviceCUPMRuntimeArch) 13155c78b439SPierre Jolivet PetscCall(PetscViewerASCIIPrintf(viewer, "%s on a %s named %s with %d process%s and %s architecture %d, by %s on %s\n", pname, arch, hostname, size, size > 1 ? "es" : "", cupm, PetscDeviceCUPMRuntimeArch, username, date)); 1316456dbbcdSJunchao Zhang else 1317456dbbcdSJunchao Zhang #endif 13185c78b439SPierre Jolivet PetscCall(PetscViewerASCIIPrintf(viewer, "%s on a %s named %s with %d process%s, by %s on %s\n", pname, arch, hostname, size, size > 1 ? "es" : "", username, date)); 1319456dbbcdSJunchao Zhang 132078f1b9b4SToby Isaac #if defined(PETSC_HAVE_OPENMP) 13211d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Using %" PetscInt_FMT " OpenMP threads\n", PetscNumOMPThreads)); 132278f1b9b4SToby Isaac #endif 13231d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Using %s\n", version)); 132478f1b9b4SToby Isaac 132578f1b9b4SToby Isaac /* Must preserve reduction count before we go on */ 132678f1b9b4SToby Isaac red = petsc_allreduce_ct + petsc_gather_ct + petsc_scatter_ct; 132778f1b9b4SToby Isaac 132878f1b9b4SToby Isaac /* Calculate summary information */ 13291d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "\n Max Max/Min Avg Total\n")); 133078f1b9b4SToby Isaac /* Time */ 1331462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&locTotalTime, &min, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1332462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&locTotalTime, &max, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1333462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&locTotalTime, &tot, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 133478f1b9b4SToby Isaac avg = tot / ((PetscLogDouble)size); 133578f1b9b4SToby Isaac if (min != 0.0) ratio = max / min; 133678f1b9b4SToby Isaac else ratio = 0.0; 13371d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Time (sec): %5.3e %7.3f %5.3e\n", max, ratio, avg)); 133878f1b9b4SToby Isaac TotalTime = tot; 133978f1b9b4SToby Isaac /* Objects */ 134078f1b9b4SToby Isaac { 134178f1b9b4SToby Isaac PetscInt num_objects; 134278f1b9b4SToby Isaac 134378f1b9b4SToby Isaac PetscCall(PetscLogObjectArrayGetSize(def->petsc_objects, &num_objects, NULL)); 134478f1b9b4SToby Isaac avg = (PetscLogDouble)num_objects; 134578f1b9b4SToby Isaac } 1346462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&avg, &min, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1347462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&avg, &max, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1348462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&avg, &tot, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 134978f1b9b4SToby Isaac avg = tot / ((PetscLogDouble)size); 135078f1b9b4SToby Isaac if (min != 0.0) ratio = max / min; 135178f1b9b4SToby Isaac else ratio = 0.0; 13521d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Objects: %5.3e %7.3f %5.3e\n", max, ratio, avg)); 135378f1b9b4SToby Isaac /* Flops */ 1354462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&petsc_TotalFlops, &min, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1355462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&petsc_TotalFlops, &max, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1356462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&petsc_TotalFlops, &tot, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 135778f1b9b4SToby Isaac avg = tot / ((PetscLogDouble)size); 135878f1b9b4SToby Isaac if (min != 0.0) ratio = max / min; 135978f1b9b4SToby Isaac else ratio = 0.0; 13601d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Flops: %5.3e %7.3f %5.3e %5.3e\n", max, ratio, avg, tot)); 136178f1b9b4SToby Isaac TotalFlops = tot; 136278f1b9b4SToby Isaac /* Flops/sec -- Must talk to Barry here */ 136378f1b9b4SToby Isaac if (locTotalTime != 0.0) flops = petsc_TotalFlops / locTotalTime; 136478f1b9b4SToby Isaac else flops = 0.0; 1365462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&flops, &min, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1366462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&flops, &max, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1367462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&flops, &tot, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 136878f1b9b4SToby Isaac avg = tot / ((PetscLogDouble)size); 136978f1b9b4SToby Isaac if (min != 0.0) ratio = max / min; 137078f1b9b4SToby Isaac else ratio = 0.0; 13711d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Flops/sec: %5.3e %7.3f %5.3e %5.3e\n", max, ratio, avg, tot)); 137278f1b9b4SToby Isaac /* Memory */ 137378f1b9b4SToby Isaac PetscCall(PetscMallocGetMaximumUsage(&mem)); 137478f1b9b4SToby Isaac if (mem > 0.0) { 1375462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&mem, &min, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1376462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&mem, &max, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1377462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&mem, &tot, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 137878f1b9b4SToby Isaac avg = tot / ((PetscLogDouble)size); 137978f1b9b4SToby Isaac if (min != 0.0) ratio = max / min; 138078f1b9b4SToby Isaac else ratio = 0.0; 13811d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Memory (bytes): %5.3e %7.3f %5.3e %5.3e\n", max, ratio, avg, tot)); 138278f1b9b4SToby Isaac } 138378f1b9b4SToby Isaac /* Messages */ 138478f1b9b4SToby Isaac mess = 0.5 * (petsc_irecv_ct + petsc_isend_ct + petsc_recv_ct + petsc_send_ct); 1385462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&mess, &min, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1386462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&mess, &max, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1387462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&mess, &tot, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 138878f1b9b4SToby Isaac avg = tot / ((PetscLogDouble)size); 138978f1b9b4SToby Isaac if (min != 0.0) ratio = max / min; 139078f1b9b4SToby Isaac else ratio = 0.0; 13911d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "MPI Msg Count: %5.3e %7.3f %5.3e %5.3e\n", max, ratio, avg, tot)); 139278f1b9b4SToby Isaac numMessages = tot; 139378f1b9b4SToby Isaac /* Message Lengths */ 139478f1b9b4SToby Isaac mess = 0.5 * (petsc_irecv_len + petsc_isend_len + petsc_recv_len + petsc_send_len); 1395462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&mess, &min, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1396462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&mess, &max, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1397462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&mess, &tot, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 139878f1b9b4SToby Isaac if (numMessages != 0) avg = tot / numMessages; 139978f1b9b4SToby Isaac else avg = 0.0; 140078f1b9b4SToby Isaac if (min != 0.0) ratio = max / min; 140178f1b9b4SToby Isaac else ratio = 0.0; 14021d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "MPI Msg Len (bytes): %5.3e %7.3f %5.3e %5.3e\n", max, ratio, avg, tot)); 140378f1b9b4SToby Isaac messageLength = tot; 140478f1b9b4SToby Isaac /* Reductions */ 1405462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&red, &min, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1406462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&red, &max, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1407462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&red, &tot, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 140878f1b9b4SToby Isaac if (min != 0.0) ratio = max / min; 140978f1b9b4SToby Isaac else ratio = 0.0; 14101d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "MPI Reductions: %5.3e %7.3f\n", max, ratio)); 141178f1b9b4SToby Isaac numReductions = red; /* wrong because uses count from process zero */ 14121d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "\nFlop counting convention: 1 flop = 1 real number operation of type (multiply/divide/add/subtract)\n")); 14131d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " e.g., VecAXPY() for real vectors of length N --> 2N flops\n")); 14141d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " and VecAXPY() for complex vectors of length N --> 8N flops\n")); 141578f1b9b4SToby Isaac 141678f1b9b4SToby Isaac PetscCall(PetscLogRegistryCreateGlobalStageNames(comm, state->registry, &global_stages)); 141778f1b9b4SToby Isaac PetscCall(PetscLogRegistryCreateGlobalEventNames(comm, state->registry, &global_events)); 141878f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGetSize(global_stages, NULL, &numStages)); 141978f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGetSize(global_events, NULL, &numEvents)); 142078f1b9b4SToby Isaac PetscCall(PetscMemzero(&zero_info, sizeof(zero_info))); 142178f1b9b4SToby Isaac PetscCall(PetscMalloc1(numStages, &localStageUsed)); 142278f1b9b4SToby Isaac PetscCall(PetscMalloc1(numStages, &stageUsed)); 142378f1b9b4SToby Isaac PetscCall(PetscMalloc1(numStages, &localStageVisible)); 142478f1b9b4SToby Isaac PetscCall(PetscMalloc1(numStages, &stageVisible)); 142578f1b9b4SToby Isaac if (numStages > 0) { 142678f1b9b4SToby Isaac for (stage = 0; stage < numStages; stage++) { 142778f1b9b4SToby Isaac PetscInt stage_id; 142878f1b9b4SToby Isaac 142978f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetLocal(global_stages, stage, &stage_id)); 143078f1b9b4SToby Isaac if (stage_id >= 0) { 143178f1b9b4SToby Isaac PetscStagePerf *stage_info; 143278f1b9b4SToby Isaac 143378f1b9b4SToby Isaac PetscCall(PetscLogHandlerDefaultGetStageInfo(handler, stage, &stage_info)); 143478f1b9b4SToby Isaac localStageUsed[stage] = stage_info->used; 143578f1b9b4SToby Isaac localStageVisible[stage] = stage_info->perfInfo.visible; 143678f1b9b4SToby Isaac } else { 143778f1b9b4SToby Isaac localStageUsed[stage] = PETSC_FALSE; 143878f1b9b4SToby Isaac localStageVisible[stage] = PETSC_TRUE; 143978f1b9b4SToby Isaac } 144078f1b9b4SToby Isaac } 1441462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(localStageUsed, stageUsed, numStages, MPIU_BOOL, MPI_LOR, comm)); 1442462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(localStageVisible, stageVisible, numStages, MPIU_BOOL, MPI_LAND, comm)); 144378f1b9b4SToby Isaac for (stage = 0; stage < numStages; stage++) { 144478f1b9b4SToby Isaac if (stageUsed[stage] && stageVisible[stage]) { 14451d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "\nSummary of Stages: ----- Time ------ ----- Flop ------ --- Messages --- -- Message Lengths -- -- Reductions --\n")); 14461d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " Avg %%Total Avg %%Total Count %%Total Avg %%Total Count %%Total\n")); 144778f1b9b4SToby Isaac break; 144878f1b9b4SToby Isaac } 144978f1b9b4SToby Isaac } 145078f1b9b4SToby Isaac for (stage = 0; stage < numStages; stage++) { 145178f1b9b4SToby Isaac PetscInt stage_id; 145278f1b9b4SToby Isaac PetscEventPerfInfo *stage_info; 145378f1b9b4SToby Isaac const char *stage_name; 145478f1b9b4SToby Isaac 145578f1b9b4SToby Isaac if (!(stageUsed[stage] && stageVisible[stage])) continue; 145678f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetLocal(global_stages, stage, &stage_id)); 145778f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetName(global_stages, stage, &stage_name)); 145878f1b9b4SToby Isaac stage_info = &zero_info; 145978f1b9b4SToby Isaac if (localStageUsed[stage]) { 146078f1b9b4SToby Isaac PetscStagePerf *stage_perf_info; 146178f1b9b4SToby Isaac 146278f1b9b4SToby Isaac PetscCall(PetscLogHandlerDefaultGetStageInfo(handler, stage, &stage_perf_info)); 146378f1b9b4SToby Isaac stage_info = &stage_perf_info->perfInfo; 146478f1b9b4SToby Isaac } 1465462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&stage_info->time, &stageTime, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1466462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&stage_info->flops, &flops, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1467462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&stage_info->numMessages, &mess, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1468462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&stage_info->messageLength, &messLen, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1469462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&stage_info->numReductions, &red, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 147078f1b9b4SToby Isaac mess *= 0.5; 147178f1b9b4SToby Isaac messLen *= 0.5; 147278f1b9b4SToby Isaac red /= size; 147378f1b9b4SToby Isaac if (TotalTime != 0.0) fracTime = stageTime / TotalTime; 147478f1b9b4SToby Isaac else fracTime = 0.0; 147578f1b9b4SToby Isaac if (TotalFlops != 0.0) fracFlops = flops / TotalFlops; 147678f1b9b4SToby Isaac else fracFlops = 0.0; 147778f1b9b4SToby Isaac /* Talk to Barry if (stageTime != 0.0) flops = (size*flops)/stageTime; else flops = 0.0; */ 147878f1b9b4SToby Isaac if (numMessages != 0.0) fracMessages = mess / numMessages; 147978f1b9b4SToby Isaac else fracMessages = 0.0; 148078f1b9b4SToby Isaac if (mess != 0.0) avgMessLen = messLen / mess; 148178f1b9b4SToby Isaac else avgMessLen = 0.0; 148278f1b9b4SToby Isaac if (messageLength != 0.0) fracLength = messLen / messageLength; 148378f1b9b4SToby Isaac else fracLength = 0.0; 148478f1b9b4SToby Isaac if (numReductions != 0.0) fracReductions = red / numReductions; 148578f1b9b4SToby Isaac else fracReductions = 0.0; 14861d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "%2d: %15s: %6.4e %5.1f%% %6.4e %5.1f%% %5.3e %5.1f%% %5.3e %5.1f%% %5.3e %5.1f%%\n", stage, stage_name, stageTime / size, 100.0 * fracTime, flops, 100.0 * fracFlops, mess, 100.0 * fracMessages, avgMessLen, 100.0 * fracLength, red, 100.0 * fracReductions)); 148778f1b9b4SToby Isaac } 148878f1b9b4SToby Isaac } 148978f1b9b4SToby Isaac 14901d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "\n------------------------------------------------------------------------------------------------------------------------\n")); 14911d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "See the 'Profiling' chapter of the users' manual for details on interpreting output.\n")); 14921d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Phase summary info:\n")); 14931d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " Count: number of times phase was executed\n")); 14941d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " Time and Flop: Max - maximum over all processors\n")); 14951d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " Ratio - ratio of maximum to minimum over all processors\n")); 14961d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " Mess: number of messages sent\n")); 14971d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " AvgLen: average message length (bytes)\n")); 14981d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " Reduct: number of global reductions\n")); 14991d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " Global: entire computation\n")); 15001d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " Stage: stages of a computation. Set stages with PetscLogStagePush() and PetscLogStagePop().\n")); 15011d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " %%T - percent time in this phase %%F - percent flop in this phase\n")); 15021d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " %%M - percent messages in this phase %%L - percent message lengths in this phase\n")); 15031d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " %%R - percent reductions in this phase\n")); 1504*69fd5feaSJose E. Roman PetscCall(PetscViewerASCIIPrintf(viewer, " Total Mflop/s: 1e-6 * (sum of flop over all processors)/(max time over all processors)\n")); 150578f1b9b4SToby Isaac if (PetscLogMemory) { 15061d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " Memory usage is summed over all MPI processes, it is given in mega-bytes\n")); 15071d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " Malloc Mbytes: Memory allocated and kept during event (sum over all calls to event). May be negative\n")); 15081d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " EMalloc Mbytes: extra memory allocated during event and then freed (maximum over all calls to events). Never negative\n")); 15091d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " MMalloc Mbytes: Increase in high water mark of allocated memory (sum over all calls to event). Never negative\n")); 15101d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " RMI Mbytes: Increase in resident memory (sum over all calls to event)\n")); 151178f1b9b4SToby Isaac } 151278f1b9b4SToby Isaac #if defined(PETSC_HAVE_DEVICE) 1513*69fd5feaSJose E. Roman PetscCall(PetscViewerASCIIPrintf(viewer, " GPU Mflop/s: 1e-6 * (sum of flop on GPU over all processors)/(max GPU time over all processors)\n")); 15141d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " CpuToGpu Count: total number of CPU to GPU copies per processor\n")); 1515*69fd5feaSJose E. Roman PetscCall(PetscViewerASCIIPrintf(viewer, " CpuToGpu Size (Mbytes): 1e-6 * (total size of CPU to GPU copies per processor)\n")); 15161d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " GpuToCpu Count: total number of GPU to CPU copies per processor\n")); 1517*69fd5feaSJose E. Roman PetscCall(PetscViewerASCIIPrintf(viewer, " GpuToCpu Size (Mbytes): 1e-6 * (total size of GPU to CPU copies per processor)\n")); 15181d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " GPU %%F: percent flops on GPU in this event\n")); 151978f1b9b4SToby Isaac #endif 15201d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "------------------------------------------------------------------------------------------------------------------------\n")); 152178f1b9b4SToby Isaac 15221d031f67SBarry Smith PetscCall(PetscLogViewWarnDebugging(viewer)); 152378f1b9b4SToby Isaac 152478f1b9b4SToby Isaac /* Report events */ 15251d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Event Count Time (sec) Flop --- Global --- --- Stage ---- Total")); 15261d031f67SBarry Smith if (PetscLogMemory) PetscCall(PetscViewerASCIIPrintf(viewer, " Malloc EMalloc MMalloc RMI")); 152778f1b9b4SToby Isaac #if defined(PETSC_HAVE_DEVICE) 15281d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " GPU - CpuToGpu - - GpuToCpu - GPU")); 152978f1b9b4SToby Isaac #endif 15301d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "\n")); 15311d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " Max Ratio Max Ratio Max Ratio Mess AvgLen Reduct %%T %%F %%M %%L %%R %%T %%F %%M %%L %%R Mflop/s")); 15321d031f67SBarry Smith if (PetscLogMemory) PetscCall(PetscViewerASCIIPrintf(viewer, " Mbytes Mbytes Mbytes Mbytes")); 153378f1b9b4SToby Isaac #if defined(PETSC_HAVE_DEVICE) 15341d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " Mflop/s Count Size Count Size %%F")); 153578f1b9b4SToby Isaac #endif 15361d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "\n")); 15371d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "------------------------------------------------------------------------------------------------------------------------")); 15381d031f67SBarry Smith if (PetscLogMemory) PetscCall(PetscViewerASCIIPrintf(viewer, "-----------------------------")); 153978f1b9b4SToby Isaac #if defined(PETSC_HAVE_DEVICE) 15401d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "---------------------------------------")); 154178f1b9b4SToby Isaac #endif 15421d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "\n")); 154378f1b9b4SToby Isaac 154478f1b9b4SToby Isaac #if defined(PETSC_HAVE_DEVICE) 154578f1b9b4SToby Isaac /* this indirect way of accessing these values is needed when PETSc is build with multiple libraries since the symbols are not in libpetscsys */ 15468025bc25SToby Isaac PetscCall(PetscLogStateGetEventFromName(state, "TaoSolve", &TAO_Solve)); 154778f1b9b4SToby Isaac PetscCall(PetscLogStateGetEventFromName(state, "TSStep", &TS_Step)); 154878f1b9b4SToby Isaac PetscCall(PetscLogStateGetEventFromName(state, "SNESSolve", &SNES_Solve)); 154978f1b9b4SToby Isaac PetscCall(PetscLogStateGetEventFromName(state, "KSPSolve", &KSP_Solve)); 155078f1b9b4SToby Isaac #endif 155178f1b9b4SToby Isaac 155278f1b9b4SToby Isaac for (stage = 0; stage < numStages; stage++) { 155378f1b9b4SToby Isaac PetscInt stage_id; 155478f1b9b4SToby Isaac PetscEventPerfInfo *stage_info; 155578f1b9b4SToby Isaac const char *stage_name; 155678f1b9b4SToby Isaac 155778f1b9b4SToby Isaac if (!(stageVisible[stage] && stageUsed[stage])) continue; 155878f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetLocal(global_stages, stage, &stage_id)); 155978f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetName(global_stages, stage, &stage_name)); 15601d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "\n--- Event Stage %d: %s\n\n", stage, stage_name)); 156178f1b9b4SToby Isaac stage_info = &zero_info; 156278f1b9b4SToby Isaac if (localStageUsed[stage]) { 156378f1b9b4SToby Isaac PetscStagePerf *stage_perf_info; 156478f1b9b4SToby Isaac 156578f1b9b4SToby Isaac PetscCall(PetscLogHandlerDefaultGetStageInfo(handler, stage_id, &stage_perf_info)); 156678f1b9b4SToby Isaac stage_info = &stage_perf_info->perfInfo; 156778f1b9b4SToby Isaac } 1568462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&stage_info->time, &stageTime, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1569462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&stage_info->flops, &flops, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1570462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&stage_info->numMessages, &mess, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1571462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&stage_info->messageLength, &messLen, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1572462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&stage_info->numReductions, &red, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 157378f1b9b4SToby Isaac mess *= 0.5; 157478f1b9b4SToby Isaac messLen *= 0.5; 157578f1b9b4SToby Isaac red /= size; 157678f1b9b4SToby Isaac 157778f1b9b4SToby Isaac for (event = 0; event < numEvents; event++) { 157878f1b9b4SToby Isaac PetscInt event_id; 157978f1b9b4SToby Isaac PetscEventPerfInfo *event_info = &zero_info; 158078f1b9b4SToby Isaac PetscBool is_zero = PETSC_FALSE; 158178f1b9b4SToby Isaac const char *event_name; 158278f1b9b4SToby Isaac 158378f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetLocal(global_events, event, &event_id)); 158478f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetName(global_events, event, &event_name)); 1585dff009beSToby Isaac if (event_id >= 0 && stage_id >= 0) { PetscCall(PetscLogHandlerGetEventPerfInfo_Default(handler, stage_id, event_id, &event_info)); } 158678f1b9b4SToby Isaac PetscCall(PetscMemcmp(event_info, &zero_info, sizeof(zero_info), &is_zero)); 1587462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(MPI_IN_PLACE, &is_zero, 1, MPIU_BOOL, MPI_LAND, comm)); 158878f1b9b4SToby Isaac if (!is_zero) { 158978f1b9b4SToby Isaac flopr = event_info->flops; 1590462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&flopr, &minf, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1591462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&flopr, &maxf, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1592462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->flops, &totf, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1593462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->time, &mint, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1594462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->time, &maxt, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1595462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->time, &tott, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1596462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->numMessages, &totm, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1597462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->messageLength, &totml, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1598462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->numReductions, &totr, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1599462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->count, &minC, 1, MPI_INT, MPI_MIN, comm)); 1600462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->count, &maxC, 1, MPI_INT, MPI_MAX, comm)); 160178f1b9b4SToby Isaac if (PetscLogMemory) { 1602462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->memIncrease, &mem, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1603462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->mallocSpace, &mal, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1604462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->mallocIncrease, &malmax, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1605462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->mallocIncreaseEvent, &emalmax, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 160678f1b9b4SToby Isaac } 160778f1b9b4SToby Isaac #if defined(PETSC_HAVE_DEVICE) 1608462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->CpuToGpuCount, &cct, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1609462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->GpuToCpuCount, &gct, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1610462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->CpuToGpuSize, &csz, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1611462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->GpuToCpuSize, &gsz, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1612462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->GpuFlops, &gflops, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1613462c564dSBarry Smith PetscCallMPI(MPIU_Allreduce(&event_info->GpuTime, &gmaxt, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 161478f1b9b4SToby Isaac #endif 161578f1b9b4SToby Isaac if (mint < 0.0) { 16161d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "WARNING!!! Minimum time %g over all processors for %s is negative! This happens\n on some machines whose times cannot handle too rapid calls.!\n artificially changing minimum to zero.\n", mint, event_name)); 161778f1b9b4SToby Isaac mint = 0; 161878f1b9b4SToby Isaac } 161978f1b9b4SToby Isaac PetscCheck(minf >= 0.0, PETSC_COMM_SELF, PETSC_ERR_PLIB, "Minimum flop %g over all processors for %s is negative! Not possible!", minf, event_name); 16207d766218SJunchao Zhang #if defined(PETSC_HAVE_DEVICE) && !defined(PETSC_HAVE_KOKKOS_WITHOUT_GPU) 162178f1b9b4SToby Isaac /* Put NaN into the time for all events that may not be time accurately since they may happen asynchronously on the GPU */ 162278f1b9b4SToby Isaac if (!PetscLogGpuTimeFlag && petsc_gflops > 0) { 162378f1b9b4SToby Isaac memcpy(&gmaxt, &nas, sizeof(PetscLogDouble)); 162478f1b9b4SToby Isaac if (event_id != SNES_Solve && event_id != KSP_Solve && event_id != TS_Step && event_id != TAO_Solve) { 162578f1b9b4SToby Isaac memcpy(&mint, &nas, sizeof(PetscLogDouble)); 162678f1b9b4SToby Isaac memcpy(&maxt, &nas, sizeof(PetscLogDouble)); 162778f1b9b4SToby Isaac } 162878f1b9b4SToby Isaac } 162978f1b9b4SToby Isaac #endif 163078f1b9b4SToby Isaac totm *= 0.5; 163178f1b9b4SToby Isaac totml *= 0.5; 163278f1b9b4SToby Isaac totr /= size; 163378f1b9b4SToby Isaac 163478f1b9b4SToby Isaac if (maxC != 0) { 163578f1b9b4SToby Isaac if (minC != 0) ratC = ((PetscLogDouble)maxC) / minC; 163678f1b9b4SToby Isaac else ratC = 0.0; 163778f1b9b4SToby Isaac if (mint != 0.0) ratt = maxt / mint; 163878f1b9b4SToby Isaac else ratt = 0.0; 163978f1b9b4SToby Isaac if (minf != 0.0) ratf = maxf / minf; 164078f1b9b4SToby Isaac else ratf = 0.0; 164178f1b9b4SToby Isaac if (TotalTime != 0.0) fracTime = tott / TotalTime; 164278f1b9b4SToby Isaac else fracTime = 0.0; 164378f1b9b4SToby Isaac if (TotalFlops != 0.0) fracFlops = totf / TotalFlops; 164478f1b9b4SToby Isaac else fracFlops = 0.0; 164578f1b9b4SToby Isaac if (stageTime != 0.0) fracStageTime = tott / stageTime; 164678f1b9b4SToby Isaac else fracStageTime = 0.0; 164778f1b9b4SToby Isaac if (flops != 0.0) fracStageFlops = totf / flops; 164878f1b9b4SToby Isaac else fracStageFlops = 0.0; 164978f1b9b4SToby Isaac if (numMessages != 0.0) fracMess = totm / numMessages; 165078f1b9b4SToby Isaac else fracMess = 0.0; 165178f1b9b4SToby Isaac if (messageLength != 0.0) fracMessLen = totml / messageLength; 165278f1b9b4SToby Isaac else fracMessLen = 0.0; 165378f1b9b4SToby Isaac if (numReductions != 0.0) fracRed = totr / numReductions; 165478f1b9b4SToby Isaac else fracRed = 0.0; 165578f1b9b4SToby Isaac if (mess != 0.0) fracStageMess = totm / mess; 165678f1b9b4SToby Isaac else fracStageMess = 0.0; 165778f1b9b4SToby Isaac if (messLen != 0.0) fracStageMessLen = totml / messLen; 165878f1b9b4SToby Isaac else fracStageMessLen = 0.0; 165978f1b9b4SToby Isaac if (red != 0.0) fracStageRed = totr / red; 166078f1b9b4SToby Isaac else fracStageRed = 0.0; 166178f1b9b4SToby Isaac if (totm != 0.0) totml /= totm; 166278f1b9b4SToby Isaac else totml = 0.0; 166378f1b9b4SToby Isaac if (maxt != 0.0) flopr = totf / maxt; 166478f1b9b4SToby Isaac else flopr = 0.0; 16650a82db20SJunchao Zhang if (fracStageTime > 1.0 || fracStageFlops > 1.0 || fracStageMess > 1.0 || fracStageMessLen > 1.0 || fracStageRed > 1.0) { 16660a82db20SJunchao Zhang if (PetscIsNanReal(maxt)) 16670a82db20SJunchao Zhang PetscCall(PetscViewerASCIIPrintf(viewer, "%-16s %7d %3.1f n/a n/a %3.2e %3.1f %2.1e %2.1e %2.1e %2.0f %2.0f %2.0f %2.0f %2.0f Multiple stages n/a", event_name, maxC, ratC, maxf, ratf, totm, totml, totr, 100.0 * fracTime, 100.0 * fracFlops, 100.0 * fracMess, 100.0 * fracMessLen, 100.0 * fracRed)); 16680a82db20SJunchao Zhang else 16691d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "%-16s %7d %3.1f %5.4e %3.1f %3.2e %3.1f %2.1e %2.1e %2.1e %2.0f %2.0f %2.0f %2.0f %2.0f Multiple stages %5.0f", event_name, maxC, ratC, maxt, ratt, maxf, ratf, totm, totml, totr, 100.0 * fracTime, 100.0 * fracFlops, 100.0 * fracMess, 100.0 * fracMessLen, 100.0 * fracRed, PetscAbs(flopr) / 1.0e6)); 16700a82db20SJunchao Zhang } else { 16710a82db20SJunchao Zhang if (PetscIsNanReal(maxt)) { // when maxt, ratt, flopr are NaN (i.e., run with GPUs but without -log_view_gpu_time), replace the confusing "nan" with "n/a" 16721d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "%-16s %7d %3.1f n/a n/a %3.2e %3.1f %2.1e %2.1e %2.1e %2.0f %2.0f %2.0f %2.0f %2.0f %3.0f %2.0f %2.0f %2.0f %2.0f n/a", event_name, maxC, ratC, maxf, ratf, totm, totml, totr, 100.0 * fracTime, 100.0 * fracFlops, 100.0 * fracMess, 100.0 * fracMessLen, 100.0 * fracRed, 100.0 * fracStageTime, 100.0 * fracStageFlops, 100.0 * fracStageMess, 100.0 * fracStageMessLen, 100.0 * fracStageRed)); 16733141c872SJunchao Zhang } else { 16741d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "%-16s %7d %3.1f %5.4e %3.1f %3.2e %3.1f %2.1e %2.1e %2.1e %2.0f %2.0f %2.0f %2.0f %2.0f %3.0f %2.0f %2.0f %2.0f %2.0f %5.0f", event_name, maxC, ratC, maxt, ratt, maxf, ratf, totm, totml, totr, 100.0 * fracTime, 100.0 * fracFlops, 100.0 * fracMess, 100.0 * fracMessLen, 100.0 * fracRed, 100.0 * fracStageTime, 100.0 * fracStageFlops, 100.0 * fracStageMess, 100.0 * fracStageMessLen, 100.0 * fracStageRed, PetscAbs(flopr) / 1.0e6)); 16753141c872SJunchao Zhang } 16763141c872SJunchao Zhang } 16771d031f67SBarry Smith if (PetscLogMemory) PetscCall(PetscViewerASCIIPrintf(viewer, " %5.0f %5.0f %5.0f %5.0f", mal / 1.0e6, emalmax / 1.0e6, malmax / 1.0e6, mem / 1.0e6)); 167878f1b9b4SToby Isaac #if defined(PETSC_HAVE_DEVICE) 167978f1b9b4SToby Isaac if (totf != 0.0) fracgflops = gflops / totf; 168078f1b9b4SToby Isaac else fracgflops = 0.0; 168178f1b9b4SToby Isaac if (gmaxt != 0.0) gflopr = gflops / gmaxt; 168278f1b9b4SToby Isaac else gflopr = 0.0; 16830a82db20SJunchao Zhang if (PetscIsNanReal(gflopr)) { 16841d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " n/a %4.0f %3.2e %4.0f %3.2e % 2.0f", cct / size, csz / (1.0e6 * size), gct / size, gsz / (1.0e6 * size), 100.0 * fracgflops)); 16853141c872SJunchao Zhang } else { 16861d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, " %5.0f %4.0f %3.2e %4.0f %3.2e % 2.0f", PetscAbs(gflopr) / 1.0e6, cct / size, csz / (1.0e6 * size), gct / size, gsz / (1.0e6 * size), 100.0 * fracgflops)); 16873141c872SJunchao Zhang } 168878f1b9b4SToby Isaac #endif 16891d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "\n")); 169078f1b9b4SToby Isaac } 169178f1b9b4SToby Isaac } 169278f1b9b4SToby Isaac } 169378f1b9b4SToby Isaac } 169478f1b9b4SToby Isaac 169578f1b9b4SToby Isaac /* Memory usage and object creation */ 16961d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "------------------------------------------------------------------------------------------------------------------------")); 16971d031f67SBarry Smith if (PetscLogMemory) PetscCall(PetscViewerASCIIPrintf(viewer, "-----------------------------")); 169878f1b9b4SToby Isaac #if defined(PETSC_HAVE_DEVICE) 16991d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "---------------------------------------")); 170078f1b9b4SToby Isaac #endif 17011d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "\n")); 17021d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "\n")); 170378f1b9b4SToby Isaac 170478f1b9b4SToby Isaac /* Right now, only stages on the first processor are reported here, meaning only objects associated with 170578f1b9b4SToby Isaac the global communicator, or MPI_COMM_SELF for proc 1. We really should report global stats and then 170678f1b9b4SToby Isaac stats for stages local to processor sets. 170778f1b9b4SToby Isaac */ 170878f1b9b4SToby Isaac /* We should figure out the longest object name here (now 20 characters) */ 17091d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Object Type Creations Destructions. Reports information only for process 0.\n")); 171078f1b9b4SToby Isaac for (stage = 0; stage < numStages; stage++) { 171178f1b9b4SToby Isaac const char *stage_name; 171278f1b9b4SToby Isaac 171378f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesGlobalGetName(global_stages, stage, &stage_name)); 17141d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "\n--- Event Stage %d: %s\n\n", stage, stage_name)); 171578f1b9b4SToby Isaac if (localStageUsed[stage]) { 171678f1b9b4SToby Isaac PetscInt num_classes; 171778f1b9b4SToby Isaac 171878f1b9b4SToby Isaac PetscCall(PetscLogStateGetNumClasses(state, &num_classes)); 171978f1b9b4SToby Isaac for (oclass = 0; oclass < num_classes; oclass++) { 172078f1b9b4SToby Isaac PetscClassPerf *class_perf_info; 172178f1b9b4SToby Isaac 172278f1b9b4SToby Isaac PetscCall(PetscLogHandlerDefaultGetClassPerf(handler, stage, oclass, &class_perf_info)); 172378f1b9b4SToby Isaac if ((class_perf_info->creations > 0) || (class_perf_info->destructions > 0)) { 172478f1b9b4SToby Isaac PetscLogClassInfo class_reg_info; 172507e942bbSPierre Jolivet PetscBool flg; 172678f1b9b4SToby Isaac 172778f1b9b4SToby Isaac PetscCall(PetscLogStateClassGetInfo(state, oclass, &class_reg_info)); 172807e942bbSPierre Jolivet if (stage == 0 && oclass == num_classes - 1) { 172907e942bbSPierre Jolivet PetscCall(PetscStrcmp(class_reg_info.name, "Viewer", &flg)); 17300bc1c2e8SToby Isaac PetscCheck(flg && class_perf_info->creations == PetscLogNumViewersCreated && class_perf_info->destructions == PetscLogNumViewersDestroyed, PETSC_COMM_SELF, PETSC_ERR_PLIB, "The last PetscObject type of the main PetscLogStage should be PetscViewer with only %" PetscInt_FMT " log viewers created and %" PetscInt_FMT "destroyed", PetscLogNumViewersCreated, PetscLogNumViewersDestroyed); 17311d031f67SBarry Smith } else PetscCall(PetscViewerASCIIPrintf(viewer, "%20s %5d %5d\n", class_reg_info.name, class_perf_info->creations, class_perf_info->destructions)); 173278f1b9b4SToby Isaac } 173378f1b9b4SToby Isaac } 173478f1b9b4SToby Isaac } 173578f1b9b4SToby Isaac } 173678f1b9b4SToby Isaac 173778f1b9b4SToby Isaac PetscCall(PetscFree(localStageUsed)); 173878f1b9b4SToby Isaac PetscCall(PetscFree(stageUsed)); 173978f1b9b4SToby Isaac PetscCall(PetscFree(localStageVisible)); 174078f1b9b4SToby Isaac PetscCall(PetscFree(stageVisible)); 174178f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesDestroy(&global_stages)); 174278f1b9b4SToby Isaac PetscCall(PetscLogGlobalNamesDestroy(&global_events)); 174378f1b9b4SToby Isaac 174478f1b9b4SToby Isaac /* Information unrelated to this particular run */ 17451d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "========================================================================================================================\n")); 174678f1b9b4SToby Isaac PetscCall(PetscTime(&y)); 174778f1b9b4SToby Isaac PetscCall(PetscTime(&x)); 174878f1b9b4SToby Isaac PetscCall(PetscTime(&y)); 174978f1b9b4SToby Isaac PetscCall(PetscTime(&y)); 175078f1b9b4SToby Isaac PetscCall(PetscTime(&y)); 175178f1b9b4SToby Isaac PetscCall(PetscTime(&y)); 175278f1b9b4SToby Isaac PetscCall(PetscTime(&y)); 175378f1b9b4SToby Isaac PetscCall(PetscTime(&y)); 175478f1b9b4SToby Isaac PetscCall(PetscTime(&y)); 175578f1b9b4SToby Isaac PetscCall(PetscTime(&y)); 175678f1b9b4SToby Isaac PetscCall(PetscTime(&y)); 175778f1b9b4SToby Isaac PetscCall(PetscTime(&y)); 17581d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Average time to get PetscTime(): %g\n", (y - x) / 10.0)); 175978f1b9b4SToby Isaac /* MPI information */ 176078f1b9b4SToby Isaac if (size > 1) { 176178f1b9b4SToby Isaac MPI_Status status; 176278f1b9b4SToby Isaac PetscMPIInt tag; 176378f1b9b4SToby Isaac MPI_Comm newcomm; 176478f1b9b4SToby Isaac 176578f1b9b4SToby Isaac PetscCallMPI(MPI_Barrier(comm)); 176678f1b9b4SToby Isaac PetscCall(PetscTime(&x)); 176778f1b9b4SToby Isaac PetscCallMPI(MPI_Barrier(comm)); 176878f1b9b4SToby Isaac PetscCallMPI(MPI_Barrier(comm)); 176978f1b9b4SToby Isaac PetscCallMPI(MPI_Barrier(comm)); 177078f1b9b4SToby Isaac PetscCallMPI(MPI_Barrier(comm)); 177178f1b9b4SToby Isaac PetscCallMPI(MPI_Barrier(comm)); 177278f1b9b4SToby Isaac PetscCall(PetscTime(&y)); 17731d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Average time for MPI_Barrier(): %g\n", (y - x) / 5.0)); 177478f1b9b4SToby Isaac PetscCall(PetscCommDuplicate(comm, &newcomm, &tag)); 177578f1b9b4SToby Isaac PetscCallMPI(MPI_Barrier(comm)); 177678f1b9b4SToby Isaac if (rank) { 177778f1b9b4SToby Isaac PetscCallMPI(MPI_Recv(NULL, 0, MPI_INT, rank - 1, tag, newcomm, &status)); 177878f1b9b4SToby Isaac PetscCallMPI(MPI_Send(NULL, 0, MPI_INT, (rank + 1) % size, tag, newcomm)); 177978f1b9b4SToby Isaac } else { 178078f1b9b4SToby Isaac PetscCall(PetscTime(&x)); 178178f1b9b4SToby Isaac PetscCallMPI(MPI_Send(NULL, 0, MPI_INT, 1, tag, newcomm)); 178278f1b9b4SToby Isaac PetscCallMPI(MPI_Recv(NULL, 0, MPI_INT, size - 1, tag, newcomm, &status)); 178378f1b9b4SToby Isaac PetscCall(PetscTime(&y)); 17841d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Average time for zero size MPI_Send(): %g\n", (y - x) / size)); 178578f1b9b4SToby Isaac } 178678f1b9b4SToby Isaac PetscCall(PetscCommDestroy(&newcomm)); 178778f1b9b4SToby Isaac } 178878f1b9b4SToby Isaac PetscCall(PetscOptionsView(NULL, viewer)); 178978f1b9b4SToby Isaac 179078f1b9b4SToby Isaac /* Machine and compile information */ 179178f1b9b4SToby Isaac if (PetscDefined(USE_FORTRAN_KERNELS)) { 17921d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Compiled with FORTRAN kernels\n")); 179378f1b9b4SToby Isaac } else { 17941d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Compiled without FORTRAN kernels\n")); 179578f1b9b4SToby Isaac } 179678f1b9b4SToby Isaac if (PetscDefined(USE_64BIT_INDICES)) { 17971d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Compiled with 64-bit PetscInt\n")); 179878f1b9b4SToby Isaac } else if (PetscDefined(USE___FLOAT128)) { 17991d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Compiled with 32-bit PetscInt\n")); 180078f1b9b4SToby Isaac } 180178f1b9b4SToby Isaac if (PetscDefined(USE_REAL_SINGLE)) { 18021d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Compiled with single precision PetscScalar and PetscReal\n")); 180378f1b9b4SToby Isaac } else if (PetscDefined(USE___FLOAT128)) { 18041d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Compiled with 128 bit precision PetscScalar and PetscReal\n")); 180578f1b9b4SToby Isaac } 180678f1b9b4SToby Isaac if (PetscDefined(USE_REAL_MAT_SINGLE)) { 18071d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Compiled with single precision matrices\n")); 180878f1b9b4SToby Isaac } else { 18091d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Compiled with full precision matrices (default)\n")); 181078f1b9b4SToby Isaac } 18111d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "sizeof(short) %d sizeof(int) %d sizeof(long) %d sizeof(void*) %d sizeof(PetscScalar) %d sizeof(PetscInt) %d\n", (int)sizeof(short), (int)sizeof(int), (int)sizeof(long), (int)sizeof(void *), (int)sizeof(PetscScalar), (int)sizeof(PetscInt))); 181278f1b9b4SToby Isaac 18131d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "Configure options: %s", petscconfigureoptions)); 18141d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "%s", petscmachineinfo)); 18151d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "%s", petsccompilerinfo)); 18161d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "%s", petsccompilerflagsinfo)); 18171d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "%s", petsclinkerinfo)); 181878f1b9b4SToby Isaac 181978f1b9b4SToby Isaac /* Cleanup */ 18201d031f67SBarry Smith PetscCall(PetscViewerASCIIPrintf(viewer, "\n")); 18211d031f67SBarry Smith PetscCall(PetscLogViewWarnNoGpuAwareMpi(viewer)); 18221d031f67SBarry Smith PetscCall(PetscLogViewWarnDebugging(viewer)); 182378f1b9b4SToby Isaac PetscCall(PetscFPTrapPop()); 182478f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 182578f1b9b4SToby Isaac } 182678f1b9b4SToby Isaac 182778f1b9b4SToby Isaac static PetscErrorCode PetscLogHandlerView_Default(PetscLogHandler handler, PetscViewer viewer) 182878f1b9b4SToby Isaac { 182978f1b9b4SToby Isaac PetscViewerFormat format; 18304d86920dSPierre Jolivet 183178f1b9b4SToby Isaac PetscFunctionBegin; 183278f1b9b4SToby Isaac PetscCall(PetscViewerGetFormat(viewer, &format)); 183378f1b9b4SToby Isaac if (format == PETSC_VIEWER_DEFAULT || format == PETSC_VIEWER_ASCII_INFO) { 183478f1b9b4SToby Isaac PetscCall(PetscLogHandlerView_Default_Info(handler, viewer)); 183578f1b9b4SToby Isaac } else if (format == PETSC_VIEWER_ASCII_INFO_DETAIL) { 183678f1b9b4SToby Isaac PetscCall(PetscLogHandlerView_Default_Detailed(handler, viewer)); 183778f1b9b4SToby Isaac } else if (format == PETSC_VIEWER_ASCII_CSV) { 183878f1b9b4SToby Isaac PetscCall(PetscLogHandlerView_Default_CSV(handler, viewer)); 183978f1b9b4SToby Isaac } 184078f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 184178f1b9b4SToby Isaac } 184278f1b9b4SToby Isaac 184378f1b9b4SToby Isaac /*MC 1844294de794SToby Isaac PETSCLOGHANDLERDEFAULT - PETSCLOGHANDLERDEFAULT = "default" - A `PetscLogHandler` that collects data for PETSc 1845b665b14eSToby Isaac default profiling log viewers (`PetscLogView()` and `PetscLogDump()`). A log handler of this type is 1846b665b14eSToby Isaac created and started (`PetscLogHandlerStart()`) by `PetscLogDefaultBegin()`. 184778f1b9b4SToby Isaac 184878f1b9b4SToby Isaac Options Database Keys: 184978f1b9b4SToby Isaac + -log_include_actions - include a growing list of actions (event beginnings and endings, object creations and destructions) in `PetscLogDump()` (`PetscLogActions()`). 185078f1b9b4SToby Isaac - -log_include_objects - include a growing list of object creations and destructions in `PetscLogDump()` (`PetscLogObjects()`). 185178f1b9b4SToby Isaac 185278f1b9b4SToby Isaac Level: developer 185378f1b9b4SToby Isaac 185478f1b9b4SToby Isaac .seealso: [](ch_profiling), `PetscLogHandler` 185578f1b9b4SToby Isaac M*/ 185678f1b9b4SToby Isaac 185778f1b9b4SToby Isaac PETSC_INTERN PetscErrorCode PetscLogHandlerCreate_Default(PetscLogHandler handler) 185878f1b9b4SToby Isaac { 185978f1b9b4SToby Isaac PetscFunctionBegin; 186078f1b9b4SToby Isaac PetscCall(PetscLogHandlerContextCreate_Default((PetscLogHandler_Default *)&handler->data)); 186178f1b9b4SToby Isaac handler->ops->destroy = PetscLogHandlerDestroy_Default; 186278f1b9b4SToby Isaac handler->ops->eventbegin = PetscLogHandlerEventBegin_Default; 186378f1b9b4SToby Isaac handler->ops->eventend = PetscLogHandlerEventEnd_Default; 186478f1b9b4SToby Isaac handler->ops->eventsync = PetscLogHandlerEventSync_Default; 186578f1b9b4SToby Isaac handler->ops->objectcreate = PetscLogHandlerObjectCreate_Default; 186678f1b9b4SToby Isaac handler->ops->objectdestroy = PetscLogHandlerObjectDestroy_Default; 186778f1b9b4SToby Isaac handler->ops->stagepush = PetscLogHandlerStagePush_Default; 186878f1b9b4SToby Isaac handler->ops->stagepop = PetscLogHandlerStagePop_Default; 186978f1b9b4SToby Isaac handler->ops->view = PetscLogHandlerView_Default; 1870dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)handler, "PetscLogHandlerGetEventPerfInfo_C", PetscLogHandlerGetEventPerfInfo_Default)); 18718b08f494SToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)handler, "PetscLogHandlerGetStagePerfInfo_C", PetscLogHandlerGetStagePerfInfo_Default)); 1872dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)handler, "PetscLogHandlerSetLogActions_C", PetscLogHandlerSetLogActions_Default)); 1873dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)handler, "PetscLogHandlerSetLogObjects_C", PetscLogHandlerSetLogObjects_Default)); 1874dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)handler, "PetscLogHandlerLogObjectState_C", PetscLogHandlerLogObjectState_Default)); 1875dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)handler, "PetscLogHandlerGetNumObjects_C", PetscLogHandlerGetNumObjects_Default)); 1876dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)handler, "PetscLogHandlerEventDeactivatePush_C", PetscLogHandlerEventDeactivatePush_Default)); 1877dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)handler, "PetscLogHandlerEventDeactivatePop_C", PetscLogHandlerEventDeactivatePop_Default)); 1878dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)handler, "PetscLogHandlerEventsPause_C", PetscLogHandlerEventsPause_Default)); 1879dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)handler, "PetscLogHandlerEventsResume_C", PetscLogHandlerEventsResume_Default)); 1880dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)handler, "PetscLogHandlerDump_C", PetscLogHandlerDump_Default)); 1881dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)handler, "PetscLogHandlerStageSetVisible_C", PetscLogHandlerStageSetVisible_Default)); 1882dff009beSToby Isaac PetscCall(PetscObjectComposeFunction((PetscObject)handler, "PetscLogHandlerStageGetVisible_C", PetscLogHandlerStageGetVisible_Default)); 188378f1b9b4SToby Isaac PetscFunctionReturn(PETSC_SUCCESS); 188478f1b9b4SToby Isaac } 1885