15c6c1daeSBarry Smith 25c6c1daeSBarry Smith /* 35c6c1daeSBarry Smith PETSc code to log object creation and destruction and PETSc events. 45c6c1daeSBarry Smith 55c6c1daeSBarry Smith This provides the public API used by the rest of PETSc and by users. 65c6c1daeSBarry Smith 75c6c1daeSBarry Smith These routines use a private API that is not used elsewhere in PETSc and is not 85c6c1daeSBarry Smith accessible to users. The private API is defined in logimpl.h and the utils directory. 95c6c1daeSBarry Smith 105c6c1daeSBarry Smith */ 11af0996ceSBarry Smith #include <petsc/private/logimpl.h> /*I "petscsys.h" I*/ 125c6c1daeSBarry Smith #include <petsctime.h> 13665c2dedSJed Brown #include <petscviewer.h> 148fe3844cSJunchao Zhang #include <petscdevice.h> 158fe3844cSJunchao Zhang #include <petsc/private/deviceimpl.h> 16a0c7f9aaSSamuel Khuvis #if defined(PETSC_HAVE_TAU_PERFSTUBS) 17a0c7f9aaSSamuel Khuvis #include <../src/sys/perfstubs/timer.h> 18a0c7f9aaSSamuel Khuvis #endif 195c6c1daeSBarry Smith 205c6c1daeSBarry Smith PetscLogEvent PETSC_LARGEST_EVENT = PETSC_EVENT; 215c6c1daeSBarry Smith 225c6c1daeSBarry Smith #if defined(PETSC_USE_LOG) 235c6c1daeSBarry Smith #include <petscmachineinfo.h> 245c6c1daeSBarry Smith #include <petscconfiginfo.h> 255c6c1daeSBarry Smith 26c708d6e3SStefano Zampini #if defined(PETSC_HAVE_THREADSAFETY) 27c708d6e3SStefano Zampini 28c708d6e3SStefano Zampini PetscInt petsc_log_gid = -1; /* Global threadId counter */ 29c708d6e3SStefano Zampini PETSC_TLS PetscInt petsc_log_tid = -1; /* Local threadId */ 30c708d6e3SStefano Zampini 31c708d6e3SStefano Zampini /* shared variables */ 32c708d6e3SStefano Zampini PetscSpinlock PetscLogSpinLock; 33c708d6e3SStefano Zampini PetscHMapEvent eventInfoMap_th = NULL; 34c708d6e3SStefano Zampini 35c708d6e3SStefano Zampini #endif 36c708d6e3SStefano Zampini 375c6c1daeSBarry Smith /* used in the MPI_XXX() count macros in petsclog.h */ 385c6c1daeSBarry Smith 395c6c1daeSBarry Smith /* Action and object logging variables */ 400298fd71SBarry Smith Action *petsc_actions = NULL; 410298fd71SBarry Smith Object *petsc_objects = NULL; 425c6c1daeSBarry Smith PetscBool petsc_logActions = PETSC_FALSE; 435c6c1daeSBarry Smith PetscBool petsc_logObjects = PETSC_FALSE; 445c6c1daeSBarry Smith int petsc_numActions = 0, petsc_maxActions = 100; 455c6c1daeSBarry Smith int petsc_numObjects = 0, petsc_maxObjects = 100; 465c6c1daeSBarry Smith int petsc_numObjectsDestroyed = 0; 475c6c1daeSBarry Smith 485c6c1daeSBarry Smith /* Global counters */ 495c6c1daeSBarry Smith PetscLogDouble petsc_BaseTime = 0.0; 505c6c1daeSBarry Smith PetscLogDouble petsc_TotalFlops = 0.0; /* The number of flops */ 515c6c1daeSBarry Smith PetscLogDouble petsc_send_ct = 0.0; /* The number of sends */ 525c6c1daeSBarry Smith PetscLogDouble petsc_recv_ct = 0.0; /* The number of receives */ 535c6c1daeSBarry Smith PetscLogDouble petsc_send_len = 0.0; /* The total length of all sent messages */ 545c6c1daeSBarry Smith PetscLogDouble petsc_recv_len = 0.0; /* The total length of all received messages */ 555c6c1daeSBarry Smith PetscLogDouble petsc_isend_ct = 0.0; /* The number of immediate sends */ 565c6c1daeSBarry Smith PetscLogDouble petsc_irecv_ct = 0.0; /* The number of immediate receives */ 575c6c1daeSBarry Smith PetscLogDouble petsc_isend_len = 0.0; /* The total length of all immediate send messages */ 585c6c1daeSBarry Smith PetscLogDouble petsc_irecv_len = 0.0; /* The total length of all immediate receive messages */ 595c6c1daeSBarry Smith PetscLogDouble petsc_wait_ct = 0.0; /* The number of waits */ 605c6c1daeSBarry Smith PetscLogDouble petsc_wait_any_ct = 0.0; /* The number of anywaits */ 615c6c1daeSBarry Smith PetscLogDouble petsc_wait_all_ct = 0.0; /* The number of waitalls */ 625c6c1daeSBarry Smith PetscLogDouble petsc_sum_of_waits_ct = 0.0; /* The total number of waits */ 635c6c1daeSBarry Smith PetscLogDouble petsc_allreduce_ct = 0.0; /* The number of reductions */ 645c6c1daeSBarry Smith PetscLogDouble petsc_gather_ct = 0.0; /* The number of gathers and gathervs */ 655c6c1daeSBarry Smith PetscLogDouble petsc_scatter_ct = 0.0; /* The number of scatters and scattervs */ 66c708d6e3SStefano Zampini 67c708d6e3SStefano Zampini /* Thread Local storage */ 68c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_TotalFlops_th = 0.0; 69c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_send_ct_th = 0.0; 70c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_recv_ct_th = 0.0; 71c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_send_len_th = 0.0; 72c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_recv_len_th = 0.0; 73c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_isend_ct_th = 0.0; 74c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_irecv_ct_th = 0.0; 75c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_isend_len_th = 0.0; 76c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_irecv_len_th = 0.0; 77c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_wait_ct_th = 0.0; 78c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_wait_any_ct_th = 0.0; 79c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_wait_all_ct_th = 0.0; 80c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_sum_of_waits_ct_th = 0.0; 81c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_allreduce_ct_th = 0.0; 82c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_gather_ct_th = 0.0; 83c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_scatter_ct_th = 0.0; 84c708d6e3SStefano Zampini 8520c24465SJunchao Zhang #if defined(PETSC_HAVE_DEVICE) 86bec0b493Shannah_mairs PetscLogDouble petsc_ctog_ct = 0.0; /* The total number of CPU to GPU copies */ 87bec0b493Shannah_mairs PetscLogDouble petsc_gtoc_ct = 0.0; /* The total number of GPU to CPU copies */ 88bec0b493Shannah_mairs PetscLogDouble petsc_ctog_sz = 0.0; /* The total size of CPU to GPU copies */ 89bec0b493Shannah_mairs PetscLogDouble petsc_gtoc_sz = 0.0; /* The total size of GPU to CPU copies */ 9045c4b7c1SBarry Smith PetscLogDouble petsc_ctog_ct_scalar = 0.0; /* The total number of CPU to GPU copies */ 9145c4b7c1SBarry Smith PetscLogDouble petsc_gtoc_ct_scalar = 0.0; /* The total number of GPU to CPU copies */ 9245c4b7c1SBarry Smith PetscLogDouble petsc_ctog_sz_scalar = 0.0; /* The total size of CPU to GPU copies */ 9345c4b7c1SBarry Smith PetscLogDouble petsc_gtoc_sz_scalar = 0.0; /* The total size of GPU to CPU copies */ 94958c4211Shannah_mairs PetscLogDouble petsc_gflops = 0.0; /* The flops done on a GPU */ 95958c4211Shannah_mairs PetscLogDouble petsc_gtime = 0.0; /* The time spent on a GPU */ 96c708d6e3SStefano Zampini 97c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_ctog_ct_th = 0.0; 98c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_gtoc_ct_th = 0.0; 99c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_ctog_sz_th = 0.0; 100c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_gtoc_sz_th = 0.0; 101c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_ctog_ct_scalar_th = 0.0; 102c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_gtoc_ct_scalar_th = 0.0; 103c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_ctog_sz_scalar_th = 0.0; 104c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_gtoc_sz_scalar_th = 0.0; 105c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_gflops_th = 0.0; 106c708d6e3SStefano Zampini PETSC_TLS PetscLogDouble petsc_gtime_th = 0.0; 107c708d6e3SStefano Zampini #endif 108c708d6e3SStefano Zampini 109c708d6e3SStefano Zampini #if defined(PETSC_HAVE_THREADSAFETY) 110c708d6e3SStefano Zampini PetscErrorCode PetscAddLogDouble(PetscLogDouble *tot, PetscLogDouble *tot_th, PetscLogDouble tmp) 111c708d6e3SStefano Zampini { 112c708d6e3SStefano Zampini *tot_th += tmp; 1133ba16761SJacob Faibussowitsch PetscCall(PetscSpinlockLock(&PetscLogSpinLock)); 114c708d6e3SStefano Zampini *tot += tmp; 1153ba16761SJacob Faibussowitsch PetscCall(PetscSpinlockUnlock(&PetscLogSpinLock)); 1163ba16761SJacob Faibussowitsch return PETSC_SUCCESS; 117c708d6e3SStefano Zampini } 118c708d6e3SStefano Zampini 119c708d6e3SStefano Zampini PetscErrorCode PetscAddLogDoubleCnt(PetscLogDouble *cnt, PetscLogDouble *tot, PetscLogDouble *cnt_th, PetscLogDouble *tot_th, PetscLogDouble tmp) 120c708d6e3SStefano Zampini { 121c708d6e3SStefano Zampini *cnt_th = *cnt_th + 1; 122c708d6e3SStefano Zampini *tot_th += tmp; 1233ba16761SJacob Faibussowitsch PetscCall(PetscSpinlockLock(&PetscLogSpinLock)); 124c708d6e3SStefano Zampini *tot += (PetscLogDouble)(tmp); 125c708d6e3SStefano Zampini *cnt += *cnt + 1; 1263ba16761SJacob Faibussowitsch PetscCall(PetscSpinlockUnlock(&PetscLogSpinLock)); 1273ba16761SJacob Faibussowitsch return PETSC_SUCCESS; 128c708d6e3SStefano Zampini } 129c708d6e3SStefano Zampini 130c708d6e3SStefano Zampini PetscInt PetscLogGetTid(void) 131c708d6e3SStefano Zampini { 132c708d6e3SStefano Zampini if (petsc_log_tid < 0) { 1333ba16761SJacob Faibussowitsch PetscCall(PetscSpinlockLock(&PetscLogSpinLock)); 134c708d6e3SStefano Zampini petsc_log_tid = ++petsc_log_gid; 1353ba16761SJacob Faibussowitsch PetscCall(PetscSpinlockUnlock(&PetscLogSpinLock)); 136c708d6e3SStefano Zampini } 137c708d6e3SStefano Zampini return petsc_log_tid; 138c708d6e3SStefano Zampini } 139c708d6e3SStefano Zampini 140bec0b493Shannah_mairs #endif 1415c6c1daeSBarry Smith 1425c6c1daeSBarry Smith /* Logging functions */ 1430298fd71SBarry Smith PetscErrorCode (*PetscLogPHC)(PetscObject) = NULL; 1440298fd71SBarry Smith PetscErrorCode (*PetscLogPHD)(PetscObject) = NULL; 1450298fd71SBarry Smith PetscErrorCode (*PetscLogPLB)(PetscLogEvent, int, PetscObject, PetscObject, PetscObject, PetscObject) = NULL; 1460298fd71SBarry Smith PetscErrorCode (*PetscLogPLE)(PetscLogEvent, int, PetscObject, PetscObject, PetscObject, PetscObject) = NULL; 1475c6c1daeSBarry Smith 1485c6c1daeSBarry Smith /* Tracing event logging variables */ 1490298fd71SBarry Smith FILE *petsc_tracefile = NULL; 1505c6c1daeSBarry Smith int petsc_tracelevel = 0; 1515c6c1daeSBarry Smith const char *petsc_traceblanks = " "; 1525c6c1daeSBarry Smith char petsc_tracespace[128] = " "; 1535c6c1daeSBarry Smith PetscLogDouble petsc_tracetime = 0.0; 154896238b9SBarry Smith static PetscBool PetscLogInitializeCalled = PETSC_FALSE; 1555c6c1daeSBarry Smith 1567a101e5eSJacob Faibussowitsch static PetscIntStack current_log_event_stack = NULL; 1577a101e5eSJacob Faibussowitsch 158d71ae5a4SJacob Faibussowitsch PETSC_INTERN PetscErrorCode PetscLogInitialize(void) 159d71ae5a4SJacob Faibussowitsch { 160fa2bb9feSLisandro Dalcin int stage; 161fa2bb9feSLisandro Dalcin PetscBool opt; 1625c6c1daeSBarry Smith 163fa2bb9feSLisandro Dalcin PetscFunctionBegin; 1643ba16761SJacob Faibussowitsch if (PetscLogInitializeCalled) PetscFunctionReturn(PETSC_SUCCESS); 165fa2bb9feSLisandro Dalcin PetscLogInitializeCalled = PETSC_TRUE; 1665c6c1daeSBarry Smith 1677a101e5eSJacob Faibussowitsch PetscCall(PetscIntStackCreate(¤t_log_event_stack)); 1689566063dSJacob Faibussowitsch PetscCall(PetscOptionsHasName(NULL, NULL, "-log_exclude_actions", &opt)); 169fa2bb9feSLisandro Dalcin if (opt) petsc_logActions = PETSC_FALSE; 1709566063dSJacob Faibussowitsch PetscCall(PetscOptionsHasName(NULL, NULL, "-log_exclude_objects", &opt)); 171fa2bb9feSLisandro Dalcin if (opt) petsc_logObjects = PETSC_FALSE; 17248a46eb9SPierre Jolivet if (petsc_logActions) PetscCall(PetscMalloc1(petsc_maxActions, &petsc_actions)); 17348a46eb9SPierre Jolivet if (petsc_logObjects) PetscCall(PetscMalloc1(petsc_maxObjects, &petsc_objects)); 174fa2bb9feSLisandro Dalcin PetscLogPHC = PetscLogObjCreateDefault; 175fa2bb9feSLisandro Dalcin PetscLogPHD = PetscLogObjDestroyDefault; 176fa2bb9feSLisandro Dalcin /* Setup default logging structures */ 1779566063dSJacob Faibussowitsch PetscCall(PetscStageLogCreate(&petsc_stageLog)); 1789566063dSJacob Faibussowitsch PetscCall(PetscStageLogRegister(petsc_stageLog, "Main Stage", &stage)); 1795c6c1daeSBarry Smith 180c708d6e3SStefano Zampini PetscCall(PetscSpinlockCreate(&PetscLogSpinLock)); 181c708d6e3SStefano Zampini #if defined(PETSC_HAVE_THREADSAFETY) 182c708d6e3SStefano Zampini petsc_log_tid = 0; 183c708d6e3SStefano Zampini petsc_log_gid = 0; 184c708d6e3SStefano Zampini PetscCall(PetscHMapEventCreate(&eventInfoMap_th)); 185c708d6e3SStefano Zampini #endif 186c708d6e3SStefano Zampini 187fa2bb9feSLisandro Dalcin /* All processors sync here for more consistent logging */ 1889566063dSJacob Faibussowitsch PetscCallMPI(MPI_Barrier(PETSC_COMM_WORLD)); 1893ba16761SJacob Faibussowitsch PetscCall(PetscTime(&petsc_BaseTime)); 1909566063dSJacob Faibussowitsch PetscCall(PetscLogStagePush(stage)); 191a0c7f9aaSSamuel Khuvis #if defined(PETSC_HAVE_TAU_PERFSTUBS) 192a0c7f9aaSSamuel Khuvis PetscStackCallExternalVoid("ps_initialize_", ps_initialize_()); 193a0c7f9aaSSamuel Khuvis #endif 1943ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 195fa2bb9feSLisandro Dalcin } 1965c6c1daeSBarry Smith 197d71ae5a4SJacob Faibussowitsch PETSC_INTERN PetscErrorCode PetscLogFinalize(void) 198d71ae5a4SJacob Faibussowitsch { 1995c6c1daeSBarry Smith PetscStageLog stageLog; 2005c6c1daeSBarry Smith 2015c6c1daeSBarry Smith PetscFunctionBegin; 202c708d6e3SStefano Zampini #if defined(PETSC_HAVE_THREADSAFETY) 203c708d6e3SStefano Zampini if (eventInfoMap_th) { 204c708d6e3SStefano Zampini PetscEventPerfInfo **array; 205c708d6e3SStefano Zampini PetscInt n, off = 0; 206c708d6e3SStefano Zampini 207c708d6e3SStefano Zampini PetscCall(PetscHMapEventGetSize(eventInfoMap_th, &n)); 208c708d6e3SStefano Zampini PetscCall(PetscMalloc1(n, &array)); 209c708d6e3SStefano Zampini PetscCall(PetscHMapEventGetVals(eventInfoMap_th, &off, array)); 210c708d6e3SStefano Zampini for (PetscInt i = 0; i < n; i++) PetscCall(PetscFree(array[i])); 211c708d6e3SStefano Zampini PetscCall(PetscFree(array)); 212c708d6e3SStefano Zampini PetscCall(PetscHMapEventDestroy(&eventInfoMap_th)); 213c708d6e3SStefano Zampini } 214c708d6e3SStefano Zampini #endif 2159566063dSJacob Faibussowitsch PetscCall(PetscFree(petsc_actions)); 2169566063dSJacob Faibussowitsch PetscCall(PetscFree(petsc_objects)); 2179566063dSJacob Faibussowitsch PetscCall(PetscLogNestedEnd()); 2189566063dSJacob Faibussowitsch PetscCall(PetscLogSet(NULL, NULL)); 2195c6c1daeSBarry Smith 2205c6c1daeSBarry Smith /* Resetting phase */ 2219566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 2229566063dSJacob Faibussowitsch PetscCall(PetscStageLogDestroy(stageLog)); 2237a101e5eSJacob Faibussowitsch PetscCall(PetscIntStackDestroy(current_log_event_stack)); 2247a101e5eSJacob Faibussowitsch current_log_event_stack = NULL; 225a297a907SKarl Rupp 2265c6c1daeSBarry Smith petsc_TotalFlops = 0.0; 2275c6c1daeSBarry Smith petsc_numActions = 0; 2285c6c1daeSBarry Smith petsc_numObjects = 0; 2295c6c1daeSBarry Smith petsc_numObjectsDestroyed = 0; 2305c6c1daeSBarry Smith petsc_maxActions = 100; 2315c6c1daeSBarry Smith petsc_maxObjects = 100; 2320298fd71SBarry Smith petsc_actions = NULL; 2330298fd71SBarry Smith petsc_objects = NULL; 2345c6c1daeSBarry Smith petsc_logActions = PETSC_FALSE; 2355c6c1daeSBarry Smith petsc_logObjects = PETSC_FALSE; 2365c6c1daeSBarry Smith petsc_BaseTime = 0.0; 2375c6c1daeSBarry Smith petsc_TotalFlops = 0.0; 2385c6c1daeSBarry Smith petsc_send_ct = 0.0; 2395c6c1daeSBarry Smith petsc_recv_ct = 0.0; 2405c6c1daeSBarry Smith petsc_send_len = 0.0; 2415c6c1daeSBarry Smith petsc_recv_len = 0.0; 2425c6c1daeSBarry Smith petsc_isend_ct = 0.0; 2435c6c1daeSBarry Smith petsc_irecv_ct = 0.0; 2445c6c1daeSBarry Smith petsc_isend_len = 0.0; 2455c6c1daeSBarry Smith petsc_irecv_len = 0.0; 2465c6c1daeSBarry Smith petsc_wait_ct = 0.0; 2475c6c1daeSBarry Smith petsc_wait_any_ct = 0.0; 2485c6c1daeSBarry Smith petsc_wait_all_ct = 0.0; 2495c6c1daeSBarry Smith petsc_sum_of_waits_ct = 0.0; 2505c6c1daeSBarry Smith petsc_allreduce_ct = 0.0; 2515c6c1daeSBarry Smith petsc_gather_ct = 0.0; 2525c6c1daeSBarry Smith petsc_scatter_ct = 0.0; 253c708d6e3SStefano Zampini petsc_TotalFlops_th = 0.0; 254c708d6e3SStefano Zampini petsc_send_ct_th = 0.0; 255c708d6e3SStefano Zampini petsc_recv_ct_th = 0.0; 256c708d6e3SStefano Zampini petsc_send_len_th = 0.0; 257c708d6e3SStefano Zampini petsc_recv_len_th = 0.0; 258c708d6e3SStefano Zampini petsc_isend_ct_th = 0.0; 259c708d6e3SStefano Zampini petsc_irecv_ct_th = 0.0; 260c708d6e3SStefano Zampini petsc_isend_len_th = 0.0; 261c708d6e3SStefano Zampini petsc_irecv_len_th = 0.0; 262c708d6e3SStefano Zampini petsc_wait_ct_th = 0.0; 263c708d6e3SStefano Zampini petsc_wait_any_ct_th = 0.0; 264c708d6e3SStefano Zampini petsc_wait_all_ct_th = 0.0; 265c708d6e3SStefano Zampini petsc_sum_of_waits_ct_th = 0.0; 266c708d6e3SStefano Zampini petsc_allreduce_ct_th = 0.0; 267c708d6e3SStefano Zampini petsc_gather_ct_th = 0.0; 268c708d6e3SStefano Zampini petsc_scatter_ct_th = 0.0; 269c708d6e3SStefano Zampini 270d9c66bfbSJunchao Zhang #if defined(PETSC_HAVE_DEVICE) 271bec0b493Shannah_mairs petsc_ctog_ct = 0.0; 272bec0b493Shannah_mairs petsc_gtoc_ct = 0.0; 273bec0b493Shannah_mairs petsc_ctog_sz = 0.0; 274bec0b493Shannah_mairs petsc_gtoc_sz = 0.0; 275958c4211Shannah_mairs petsc_gflops = 0.0; 276958c4211Shannah_mairs petsc_gtime = 0.0; 277c708d6e3SStefano Zampini petsc_ctog_ct_th = 0.0; 278c708d6e3SStefano Zampini petsc_gtoc_ct_th = 0.0; 279c708d6e3SStefano Zampini petsc_ctog_sz_th = 0.0; 280c708d6e3SStefano Zampini petsc_gtoc_sz_th = 0.0; 281c708d6e3SStefano Zampini petsc_gflops_th = 0.0; 282c708d6e3SStefano Zampini petsc_gtime_th = 0.0; 283bec0b493Shannah_mairs #endif 284c708d6e3SStefano Zampini 2855c6c1daeSBarry Smith PETSC_LARGEST_EVENT = PETSC_EVENT; 2860298fd71SBarry Smith PetscLogPHC = NULL; 2870298fd71SBarry Smith PetscLogPHD = NULL; 2880298fd71SBarry Smith petsc_tracefile = NULL; 2895c6c1daeSBarry Smith petsc_tracelevel = 0; 2905c6c1daeSBarry Smith petsc_traceblanks = " "; 2919371c9d4SSatish Balay petsc_tracespace[0] = ' '; 2929371c9d4SSatish Balay petsc_tracespace[1] = 0; 2935c6c1daeSBarry Smith petsc_tracetime = 0.0; 2945c6c1daeSBarry Smith PETSC_LARGEST_CLASSID = PETSC_SMALLEST_CLASSID; 2955c6c1daeSBarry Smith PETSC_OBJECT_CLASSID = 0; 29602c9f0b5SLisandro Dalcin petsc_stageLog = NULL; 297896238b9SBarry Smith PetscLogInitializeCalled = PETSC_FALSE; 2983ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 2995c6c1daeSBarry Smith } 3005c6c1daeSBarry Smith 3015c6c1daeSBarry Smith /*@C 3025c6c1daeSBarry Smith PetscLogSet - Sets the logging functions called at the beginning and ending of every event. 3035c6c1daeSBarry Smith 3045c6c1daeSBarry Smith Not Collective 3055c6c1daeSBarry Smith 3065c6c1daeSBarry Smith Input Parameters: 3075c6c1daeSBarry Smith + b - The function called at beginning of event 3085c6c1daeSBarry Smith - e - The function called at end of event 3095c6c1daeSBarry Smith 3105c6c1daeSBarry Smith Level: developer 3115c6c1daeSBarry Smith 312aec76313SJacob Faibussowitsch Developer Notes: 313811af0c4SBarry Smith The default loggers are `PetscLogEventBeginDefault()` and `PetscLogEventEndDefault()`. 314811af0c4SBarry Smith 315d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogDump()`, `PetscLogDefaultBegin()`, `PetscLogAllBegin()`, `PetscLogTraceBegin()`, `PetscLogEventBeginDefault()`, `PetscLogEventEndDefault()` 3165c6c1daeSBarry Smith @*/ 317d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogSet(PetscErrorCode (*b)(PetscLogEvent, int, PetscObject, PetscObject, PetscObject, PetscObject), PetscErrorCode (*e)(PetscLogEvent, int, PetscObject, PetscObject, PetscObject, PetscObject)) 318d71ae5a4SJacob Faibussowitsch { 3195c6c1daeSBarry Smith PetscFunctionBegin; 3205c6c1daeSBarry Smith PetscLogPLB = b; 3215c6c1daeSBarry Smith PetscLogPLE = e; 3223ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 3235c6c1daeSBarry Smith } 3245c6c1daeSBarry Smith 3255c6c1daeSBarry Smith /*@C 3264dd65854SConnor Ward PetscLogIsActive - Check if logging is currently in progress. 3274dd65854SConnor Ward 3284dd65854SConnor Ward Not Collective 3294dd65854SConnor Ward 3304dd65854SConnor Ward Output Parameter: 331811af0c4SBarry Smith . isActive - `PETSC_TRUE` if logging is in progress, `PETSC_FALSE` otherwise 3324dd65854SConnor Ward 3334dd65854SConnor Ward Level: beginner 3344dd65854SConnor Ward 335d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogDefaultBegin()`, `PetscLogAllBegin()`, `PetscLogSet()` 3364dd65854SConnor Ward @*/ 337d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogIsActive(PetscBool *isActive) 338d71ae5a4SJacob Faibussowitsch { 3394dd65854SConnor Ward PetscFunctionBegin; 3404dd65854SConnor Ward *isActive = (PetscLogPLB && PetscLogPLE) ? PETSC_TRUE : PETSC_FALSE; 3413ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 3424dd65854SConnor Ward } 3434dd65854SConnor Ward 3444dd65854SConnor Ward /*@C 345811af0c4SBarry Smith PetscLogDefaultBegin - Turns on logging of objects and events using the default logging functions `PetscLogEventBeginDefault()` and `PetscLogEventEndDefault()`. This logs flop 3465c6c1daeSBarry Smith rates and object creation and should not slow programs down too much. 3475c6c1daeSBarry Smith This routine may be called more than once. 3485c6c1daeSBarry Smith 349811af0c4SBarry Smith Logically Collective over `PETSC_COMM_WORLD` 3505c6c1daeSBarry Smith 351811af0c4SBarry Smith Options Database Key: 352a2553e36SBarry Smith . -log_view [viewertype:filename:viewerformat] - Prints summary of flop and timing information to the 353a2553e36SBarry Smith screen (for code configured with --with-log=1 (which is the default)) 3545c6c1daeSBarry Smith 355*10450e9eSJacob Faibussowitsch Example Usage: 3565c6c1daeSBarry Smith .vb 3575c6c1daeSBarry Smith PetscInitialize(...); 358bb1d7374SBarry Smith PetscLogDefaultBegin(); 3595c6c1daeSBarry Smith ... code ... 3605c6c1daeSBarry Smith PetscLogView(viewer); or PetscLogDump(); 3615c6c1daeSBarry Smith PetscFinalize(); 3625c6c1daeSBarry Smith .ve 3635c6c1daeSBarry Smith 364d1f92df0SBarry Smith Level: advanced 365d1f92df0SBarry Smith 366811af0c4SBarry Smith Note: 367811af0c4SBarry Smith `PetscLogView()` or `PetscLogDump()` actually cause the printing of 3685c6c1daeSBarry Smith the logging information. 3695c6c1daeSBarry Smith 370d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogDump()`, `PetscLogAllBegin()`, `PetscLogView()`, `PetscLogTraceBegin()` 3715c6c1daeSBarry Smith @*/ 372d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogDefaultBegin(void) 373d71ae5a4SJacob Faibussowitsch { 3745c6c1daeSBarry Smith PetscFunctionBegin; 3759566063dSJacob Faibussowitsch PetscCall(PetscLogSet(PetscLogEventBeginDefault, PetscLogEventEndDefault)); 3763ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 3775c6c1daeSBarry Smith } 3785c6c1daeSBarry Smith 3795c6c1daeSBarry Smith /*@C 3805c6c1daeSBarry Smith PetscLogAllBegin - Turns on extensive logging of objects and events. Logs 3815c6c1daeSBarry Smith all events. This creates large log files and slows the program down. 3825c6c1daeSBarry Smith 383811af0c4SBarry Smith Logically Collective on `PETSC_COMM_WORLD` 3845c6c1daeSBarry Smith 385811af0c4SBarry Smith Options Database Key: 386a2553e36SBarry Smith . -log_all - Prints extensive log information 3875c6c1daeSBarry Smith 388*10450e9eSJacob Faibussowitsch Example Usage: 3895c6c1daeSBarry Smith .vb 3905c6c1daeSBarry Smith PetscInitialize(...); 3915c6c1daeSBarry Smith PetscLogAllBegin(); 3925c6c1daeSBarry Smith ... code ... 3935c6c1daeSBarry Smith PetscLogDump(filename); 3945c6c1daeSBarry Smith PetscFinalize(); 3955c6c1daeSBarry Smith .ve 3965c6c1daeSBarry Smith 397d1f92df0SBarry Smith Level: advanced 398d1f92df0SBarry Smith 399811af0c4SBarry Smith Note: 400811af0c4SBarry Smith A related routine is `PetscLogDefaultBegin()` (with the options key -log_view), which is 4015c6c1daeSBarry Smith intended for production runs since it logs only flop rates and object 4025c6c1daeSBarry Smith creation (and shouldn't significantly slow the programs). 4035c6c1daeSBarry Smith 404d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogDump()`, `PetscLogDefaultBegin()`, `PetscLogTraceBegin()` 4055c6c1daeSBarry Smith @*/ 406d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogAllBegin(void) 407d71ae5a4SJacob Faibussowitsch { 4085c6c1daeSBarry Smith PetscFunctionBegin; 4099566063dSJacob Faibussowitsch PetscCall(PetscLogSet(PetscLogEventBeginComplete, PetscLogEventEndComplete)); 4103ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 4115c6c1daeSBarry Smith } 4125c6c1daeSBarry Smith 413956f8c0dSBarry Smith /*@C 4145c6c1daeSBarry Smith PetscLogTraceBegin - Activates trace logging. Every time a PETSc event 4155c6c1daeSBarry Smith begins or ends, the event name is printed. 4165c6c1daeSBarry Smith 417811af0c4SBarry Smith Logically Collective on `PETSC_COMM_WORLD` 4185c6c1daeSBarry Smith 4195c6c1daeSBarry Smith Input Parameter: 4205c6c1daeSBarry Smith . file - The file to print trace in (e.g. stdout) 4215c6c1daeSBarry Smith 4225c6c1daeSBarry Smith Options Database Key: 423811af0c4SBarry Smith . -log_trace [filename] - Activates `PetscLogTraceBegin()` 4245c6c1daeSBarry Smith 425d1f92df0SBarry Smith Level: intermediate 426d1f92df0SBarry Smith 4275c6c1daeSBarry Smith Notes: 428811af0c4SBarry Smith `PetscLogTraceBegin()` prints the processor number, the execution time (sec), 4295c6c1daeSBarry Smith then "Event begin:" or "Event end:" followed by the event name. 4305c6c1daeSBarry Smith 431811af0c4SBarry Smith `PetscLogTraceBegin()` allows tracing of all PETSc calls, which is useful 4325c6c1daeSBarry Smith to determine where a program is hanging without running in the 4335c6c1daeSBarry Smith debugger. Can be used in conjunction with the -info option. 4345c6c1daeSBarry Smith 435d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogDump()`, `PetscLogAllBegin()`, `PetscLogView()`, `PetscLogDefaultBegin()` 4365c6c1daeSBarry Smith @*/ 437d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogTraceBegin(FILE *file) 438d71ae5a4SJacob Faibussowitsch { 4395c6c1daeSBarry Smith PetscFunctionBegin; 4405c6c1daeSBarry Smith petsc_tracefile = file; 441a297a907SKarl Rupp 4429566063dSJacob Faibussowitsch PetscCall(PetscLogSet(PetscLogEventBeginTrace, PetscLogEventEndTrace)); 4433ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 4445c6c1daeSBarry Smith } 4455c6c1daeSBarry Smith 4465c6c1daeSBarry Smith /*@ 4475c6c1daeSBarry Smith PetscLogActions - Determines whether actions are logged for the graphical viewer. 4485c6c1daeSBarry Smith 4495c6c1daeSBarry Smith Not Collective 4505c6c1daeSBarry Smith 4515c6c1daeSBarry Smith Input Parameter: 452811af0c4SBarry Smith . flag - `PETSC_TRUE` if actions are to be logged 453811af0c4SBarry Smith 454811af0c4SBarry Smith Options Database Key: 455811af0c4SBarry Smith . -log_exclude_actions - Turns off actions logging 4565c6c1daeSBarry Smith 4575c6c1daeSBarry Smith Level: intermediate 4585c6c1daeSBarry Smith 459811af0c4SBarry Smith Note: 460811af0c4SBarry Smith Logging of actions continues to consume more memory as the program 4615c6c1daeSBarry Smith runs. Long running programs should consider turning this feature off. 462aec76313SJacob Faibussowitsch 463d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogStagePush()`, `PetscLogStagePop()` 4645c6c1daeSBarry Smith @*/ 465d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogActions(PetscBool flag) 466d71ae5a4SJacob Faibussowitsch { 4675c6c1daeSBarry Smith PetscFunctionBegin; 4685c6c1daeSBarry Smith petsc_logActions = flag; 4693ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 4705c6c1daeSBarry Smith } 4715c6c1daeSBarry Smith 4725c6c1daeSBarry Smith /*@ 4735c6c1daeSBarry Smith PetscLogObjects - Determines whether objects are logged for the graphical viewer. 4745c6c1daeSBarry Smith 4755c6c1daeSBarry Smith Not Collective 4765c6c1daeSBarry Smith 4775c6c1daeSBarry Smith Input Parameter: 478811af0c4SBarry Smith . flag - `PETSC_TRUE` if objects are to be logged 479811af0c4SBarry Smith 480811af0c4SBarry Smith Options Database Key: 481811af0c4SBarry Smith . -log_exclude_objects - Turns off objects logging 4825c6c1daeSBarry Smith 4835c6c1daeSBarry Smith Level: intermediate 4845c6c1daeSBarry Smith 485811af0c4SBarry Smith Note: 486811af0c4SBarry Smith Logging of objects continues to consume more memory as the program 4875c6c1daeSBarry Smith runs. Long running programs should consider turning this feature off. 4885c6c1daeSBarry Smith 489d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogStagePush()`, `PetscLogStagePop()` 4905c6c1daeSBarry Smith @*/ 491d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogObjects(PetscBool flag) 492d71ae5a4SJacob Faibussowitsch { 4935c6c1daeSBarry Smith PetscFunctionBegin; 4945c6c1daeSBarry Smith petsc_logObjects = flag; 4953ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 4965c6c1daeSBarry Smith } 4975c6c1daeSBarry Smith 4985c6c1daeSBarry Smith /*------------------------------------------------ Stage Functions --------------------------------------------------*/ 4995c6c1daeSBarry Smith /*@C 50074c0405dSRichard Tran Mills PetscLogStageRegister - Attaches a character string name to a logging stage. 5015c6c1daeSBarry Smith 5025c6c1daeSBarry Smith Not Collective 5035c6c1daeSBarry Smith 5045c6c1daeSBarry Smith Input Parameter: 5055c6c1daeSBarry Smith . sname - The name to associate with that stage 5065c6c1daeSBarry Smith 5075c6c1daeSBarry Smith Output Parameter: 5085c6c1daeSBarry Smith . stage - The stage number 5095c6c1daeSBarry Smith 5105c6c1daeSBarry Smith Level: intermediate 5115c6c1daeSBarry Smith 512d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogStagePush()`, `PetscLogStagePop()` 5135c6c1daeSBarry Smith @*/ 514d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogStageRegister(const char sname[], PetscLogStage *stage) 515d71ae5a4SJacob Faibussowitsch { 5165c6c1daeSBarry Smith PetscStageLog stageLog; 5175c6c1daeSBarry Smith PetscLogEvent event; 5185c6c1daeSBarry Smith 5195c6c1daeSBarry Smith PetscFunctionBegin; 5209566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 5219566063dSJacob Faibussowitsch PetscCall(PetscStageLogRegister(stageLog, sname, stage)); 5225c6c1daeSBarry Smith /* Copy events already changed in the main stage, this sucks */ 5239566063dSJacob Faibussowitsch PetscCall(PetscEventPerfLogEnsureSize(stageLog->stageInfo[*stage].eventLog, stageLog->eventLog->numEvents)); 52448a46eb9SPierre Jolivet for (event = 0; event < stageLog->eventLog->numEvents; event++) PetscCall(PetscEventPerfInfoCopy(&stageLog->stageInfo[0].eventLog->eventInfo[event], &stageLog->stageInfo[*stage].eventLog->eventInfo[event])); 5259566063dSJacob Faibussowitsch PetscCall(PetscClassPerfLogEnsureSize(stageLog->stageInfo[*stage].classLog, stageLog->classLog->numClasses)); 526a0c7f9aaSSamuel Khuvis #if defined(PETSC_HAVE_TAU_PERFSTUBS) 527a0c7f9aaSSamuel Khuvis if (perfstubs_initialized == PERFSTUBS_SUCCESS) PetscStackCallExternalVoid("ps_timer_create_", stageLog->stageInfo[*stage].timer = ps_timer_create_(sname)); 528a0c7f9aaSSamuel Khuvis #endif 5293ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 5305c6c1daeSBarry Smith } 5315c6c1daeSBarry Smith 5325c6c1daeSBarry Smith /*@C 533811af0c4SBarry Smith PetscLogStagePush - This function pushes a stage on the logging stack. Events started and stopped until `PetscLogStagePop()` will be associated with the stage 5345c6c1daeSBarry Smith 5355c6c1daeSBarry Smith Not Collective 5365c6c1daeSBarry Smith 5375c6c1daeSBarry Smith Input Parameter: 5385c6c1daeSBarry Smith . stage - The stage on which to log 5395c6c1daeSBarry Smith 540*10450e9eSJacob Faibussowitsch Example Usage: 541811af0c4SBarry Smith If the option -log_view is used to run the program containing the 5425c6c1daeSBarry Smith following code, then 2 sets of summary data will be printed during 5435c6c1daeSBarry Smith PetscFinalize(). 5445c6c1daeSBarry Smith .vb 5455c6c1daeSBarry Smith PetscInitialize(int *argc,char ***args,0,0); 5465c6c1daeSBarry Smith [stage 0 of code] 5475c6c1daeSBarry Smith PetscLogStagePush(1); 5485c6c1daeSBarry Smith [stage 1 of code] 5495c6c1daeSBarry Smith PetscLogStagePop(); 5505c6c1daeSBarry Smith PetscBarrier(...); 5515c6c1daeSBarry Smith [more stage 0 of code] 5525c6c1daeSBarry Smith PetscFinalize(); 5535c6c1daeSBarry Smith .ve 5545c6c1daeSBarry Smith 555d1f92df0SBarry Smith Level: intermediate 556d1f92df0SBarry Smith 557811af0c4SBarry Smith Note: 558811af0c4SBarry Smith Use `PetscLogStageRegister()` to register a stage. 5595c6c1daeSBarry Smith 560d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogStagePop()`, `PetscLogStageRegister()`, `PetscBarrier()` 5615c6c1daeSBarry Smith @*/ 562d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogStagePush(PetscLogStage stage) 563d71ae5a4SJacob Faibussowitsch { 5645c6c1daeSBarry Smith PetscStageLog stageLog; 5655c6c1daeSBarry Smith 5665c6c1daeSBarry Smith PetscFunctionBegin; 5679566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 5689566063dSJacob Faibussowitsch PetscCall(PetscStageLogPush(stageLog, stage)); 569a0c7f9aaSSamuel Khuvis #if defined(PETSC_HAVE_TAU_PERFSTUBS) 570a0c7f9aaSSamuel Khuvis if (perfstubs_initialized == PERFSTUBS_SUCCESS && stageLog->stageInfo[stage].timer != NULL) PetscStackCallExternalVoid("ps_timer_start_", ps_timer_start_(stageLog->stageInfo[stage].timer)); 571a0c7f9aaSSamuel Khuvis #endif 5723ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 5735c6c1daeSBarry Smith } 5745c6c1daeSBarry Smith 5755c6c1daeSBarry Smith /*@C 576811af0c4SBarry Smith PetscLogStagePop - This function pops a stage from the logging stack that was pushed with `PetscLogStagePush()` 5775c6c1daeSBarry Smith 5785c6c1daeSBarry Smith Not Collective 5795c6c1daeSBarry Smith 580*10450e9eSJacob Faibussowitsch Example Usage: 581811af0c4SBarry Smith If the option -log_view is used to run the program containing the 5825c6c1daeSBarry Smith following code, then 2 sets of summary data will be printed during 5835c6c1daeSBarry Smith PetscFinalize(). 5845c6c1daeSBarry Smith .vb 5855c6c1daeSBarry Smith PetscInitialize(int *argc,char ***args,0,0); 5865c6c1daeSBarry Smith [stage 0 of code] 5875c6c1daeSBarry Smith PetscLogStagePush(1); 5885c6c1daeSBarry Smith [stage 1 of code] 5895c6c1daeSBarry Smith PetscLogStagePop(); 5905c6c1daeSBarry Smith PetscBarrier(...); 5915c6c1daeSBarry Smith [more stage 0 of code] 5925c6c1daeSBarry Smith PetscFinalize(); 5935c6c1daeSBarry Smith .ve 5945c6c1daeSBarry Smith 5955c6c1daeSBarry Smith Level: intermediate 5965c6c1daeSBarry Smith 597d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogStagePush()`, `PetscLogStageRegister()`, `PetscBarrier()` 5985c6c1daeSBarry Smith @*/ 599d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogStagePop(void) 600d71ae5a4SJacob Faibussowitsch { 6015c6c1daeSBarry Smith PetscStageLog stageLog; 6025c6c1daeSBarry Smith 6035c6c1daeSBarry Smith PetscFunctionBegin; 6049566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 605a0c7f9aaSSamuel Khuvis #if defined(PETSC_HAVE_TAU_PERFSTUBS) 606a0c7f9aaSSamuel Khuvis if (perfstubs_initialized == PERFSTUBS_SUCCESS && stageLog->stageInfo[stageLog->curStage].timer != NULL) PetscStackCallExternalVoid("ps_timer_stop_", ps_timer_stop_(stageLog->stageInfo[stageLog->curStage].timer)); 607a0c7f9aaSSamuel Khuvis #endif 6089566063dSJacob Faibussowitsch PetscCall(PetscStageLogPop(stageLog)); 6093ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 6105c6c1daeSBarry Smith } 6115c6c1daeSBarry Smith 6125c6c1daeSBarry Smith /*@ 613811af0c4SBarry Smith PetscLogStageSetActive - Sets if a stage is used for `PetscLogEventBegin()` and `PetscLogEventEnd()`. 6145c6c1daeSBarry Smith 6155c6c1daeSBarry Smith Not Collective 6165c6c1daeSBarry Smith 6175c6c1daeSBarry Smith Input Parameters: 6185c6c1daeSBarry Smith + stage - The stage 619811af0c4SBarry Smith - isActive - The activity flag, `PETSC_TRUE` for logging, else `PETSC_FALSE` (defaults to `PETSC_TRUE`) 6205c6c1daeSBarry Smith 6215c6c1daeSBarry Smith Level: intermediate 6225c6c1daeSBarry Smith 623811af0c4SBarry Smith Note: 624811af0c4SBarry Smith If this is set to `PETSC_FALSE` the logging acts as if the stage did not exist 625811af0c4SBarry Smith 626d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogStageRegister()`, `PetscLogStagePush()`, `PetscLogStagePop()`, `PetscLogEventBegin()`, `PetscLogEventEnd()`, `PetscPreLoadBegin()`, `PetscPreLoadEnd()`, `PetscPreLoadStage()` 6275c6c1daeSBarry Smith @*/ 628d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogStageSetActive(PetscLogStage stage, PetscBool isActive) 629d71ae5a4SJacob Faibussowitsch { 6305c6c1daeSBarry Smith PetscStageLog stageLog; 6315c6c1daeSBarry Smith 6325c6c1daeSBarry Smith PetscFunctionBegin; 6339566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 6349566063dSJacob Faibussowitsch PetscCall(PetscStageLogSetActive(stageLog, stage, isActive)); 6353ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 6365c6c1daeSBarry Smith } 6375c6c1daeSBarry Smith 6385c6c1daeSBarry Smith /*@ 639811af0c4SBarry Smith PetscLogStageGetActive - Checks if a stage is used for `PetscLogEventBegin()` and `PetscLogEventEnd()`. 6405c6c1daeSBarry Smith 6415c6c1daeSBarry Smith Not Collective 6425c6c1daeSBarry Smith 6435c6c1daeSBarry Smith Input Parameter: 6445c6c1daeSBarry Smith . stage - The stage 6455c6c1daeSBarry Smith 6465c6c1daeSBarry Smith Output Parameter: 647811af0c4SBarry Smith . isActive - The activity flag, `PETSC_TRUE` for logging, else `PETSC_FALSE` (defaults to `PETSC_TRUE`) 6485c6c1daeSBarry Smith 6495c6c1daeSBarry Smith Level: intermediate 6505c6c1daeSBarry Smith 651d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogStageRegister()`, `PetscLogStagePush()`, `PetscLogStagePop()`, `PetscLogEventBegin()`, `PetscLogEventEnd()`, `PetscPreLoadBegin()`, `PetscPreLoadEnd()`, `PetscPreLoadStage()` 6525c6c1daeSBarry Smith @*/ 653d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogStageGetActive(PetscLogStage stage, PetscBool *isActive) 654d71ae5a4SJacob Faibussowitsch { 6555c6c1daeSBarry Smith PetscStageLog stageLog; 6565c6c1daeSBarry Smith 6575c6c1daeSBarry Smith PetscFunctionBegin; 6589566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 6599566063dSJacob Faibussowitsch PetscCall(PetscStageLogGetActive(stageLog, stage, isActive)); 6603ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 6615c6c1daeSBarry Smith } 6625c6c1daeSBarry Smith 6635c6c1daeSBarry Smith /*@ 664811af0c4SBarry Smith PetscLogStageSetVisible - Determines stage visibility in `PetscLogView()` 6655c6c1daeSBarry Smith 6665c6c1daeSBarry Smith Not Collective 6675c6c1daeSBarry Smith 6685c6c1daeSBarry Smith Input Parameters: 6695c6c1daeSBarry Smith + stage - The stage 670811af0c4SBarry Smith - isVisible - The visibility flag, `PETSC_TRUE` to print, else `PETSC_FALSE` (defaults to `PETSC_TRUE`) 6715c6c1daeSBarry Smith 6725c6c1daeSBarry Smith Level: intermediate 6735c6c1daeSBarry Smith 674aec76313SJacob Faibussowitsch Developer Notes: 675811af0c4SBarry Smith What does visible mean, needs to be documented. 676811af0c4SBarry Smith 677d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogStageRegister()`, `PetscLogStagePush()`, `PetscLogStagePop()`, `PetscLogView()` 6785c6c1daeSBarry Smith @*/ 679d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogStageSetVisible(PetscLogStage stage, PetscBool isVisible) 680d71ae5a4SJacob Faibussowitsch { 6815c6c1daeSBarry Smith PetscStageLog stageLog; 6825c6c1daeSBarry Smith 6835c6c1daeSBarry Smith PetscFunctionBegin; 6849566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 6859566063dSJacob Faibussowitsch PetscCall(PetscStageLogSetVisible(stageLog, stage, isVisible)); 6863ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 6875c6c1daeSBarry Smith } 6885c6c1daeSBarry Smith 6895c6c1daeSBarry Smith /*@ 690811af0c4SBarry Smith PetscLogStageGetVisible - Returns stage visibility in `PetscLogView()` 6915c6c1daeSBarry Smith 6925c6c1daeSBarry Smith Not Collective 6935c6c1daeSBarry Smith 6945c6c1daeSBarry Smith Input Parameter: 6955c6c1daeSBarry Smith . stage - The stage 6965c6c1daeSBarry Smith 6975c6c1daeSBarry Smith Output Parameter: 698811af0c4SBarry Smith . isVisible - The visibility flag, `PETSC_TRUE` to print, else `PETSC_FALSE` (defaults to `PETSC_TRUE`) 6995c6c1daeSBarry Smith 7005c6c1daeSBarry Smith Level: intermediate 7015c6c1daeSBarry Smith 702d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogStageRegister()`, `PetscLogStagePush()`, `PetscLogStagePop()`, `PetscLogView()` 7035c6c1daeSBarry Smith @*/ 704d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogStageGetVisible(PetscLogStage stage, PetscBool *isVisible) 705d71ae5a4SJacob Faibussowitsch { 7065c6c1daeSBarry Smith PetscStageLog stageLog; 7075c6c1daeSBarry Smith 7085c6c1daeSBarry Smith PetscFunctionBegin; 7099566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 7109566063dSJacob Faibussowitsch PetscCall(PetscStageLogGetVisible(stageLog, stage, isVisible)); 7113ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 7125c6c1daeSBarry Smith } 7135c6c1daeSBarry Smith 7145c6c1daeSBarry Smith /*@C 7155c6c1daeSBarry Smith PetscLogStageGetId - Returns the stage id when given the stage name. 7165c6c1daeSBarry Smith 7175c6c1daeSBarry Smith Not Collective 7185c6c1daeSBarry Smith 7195c6c1daeSBarry Smith Input Parameter: 7205c6c1daeSBarry Smith . name - The stage name 7215c6c1daeSBarry Smith 7225c6c1daeSBarry Smith Output Parameter: 7235a4a3fabSBarry Smith . stage - The stage, , or -1 if no stage with that name exists 7245c6c1daeSBarry Smith 7255c6c1daeSBarry Smith Level: intermediate 7265c6c1daeSBarry Smith 727d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogStageRegister()`, `PetscLogStagePush()`, `PetscLogStagePop()`, `PetscPreLoadBegin()`, `PetscPreLoadEnd()`, `PetscPreLoadStage()` 7285c6c1daeSBarry Smith @*/ 729d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogStageGetId(const char name[], PetscLogStage *stage) 730d71ae5a4SJacob Faibussowitsch { 7315c6c1daeSBarry Smith PetscStageLog stageLog; 7325c6c1daeSBarry Smith 7335c6c1daeSBarry Smith PetscFunctionBegin; 7349566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 7359566063dSJacob Faibussowitsch PetscCall(PetscStageLogGetStage(stageLog, name, stage)); 7363ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 7375c6c1daeSBarry Smith } 7385c6c1daeSBarry Smith 7395c6c1daeSBarry Smith /*------------------------------------------------ Event Functions --------------------------------------------------*/ 7407a101e5eSJacob Faibussowitsch 7415c6c1daeSBarry Smith /*@C 742811af0c4SBarry Smith PetscLogEventRegister - Registers an event name for logging operations 7435c6c1daeSBarry Smith 7445c6c1daeSBarry Smith Not Collective 7455c6c1daeSBarry Smith 746d8d19677SJose E. Roman Input Parameters: 7475c6c1daeSBarry Smith + name - The name associated with the event 7485c6c1daeSBarry Smith - classid - The classid associated to the class for this event, obtain either with 749811af0c4SBarry Smith `PetscClassIdRegister()` or use a predefined one such as `KSP_CLASSID`, `SNES_CLASSID`, the predefined ones 7505c6c1daeSBarry Smith are only available in C code 7515c6c1daeSBarry Smith 7525c6c1daeSBarry Smith Output Parameter: 753811af0c4SBarry Smith . event - The event id for use with `PetscLogEventBegin()` and `PetscLogEventEnd()`. 7545c6c1daeSBarry Smith 755*10450e9eSJacob Faibussowitsch Example Usage: 7565c6c1daeSBarry Smith .vb 7575c6c1daeSBarry Smith PetscLogEvent USER_EVENT; 7585c6c1daeSBarry Smith PetscClassId classid; 7595c6c1daeSBarry Smith PetscLogDouble user_event_flops; 7605c6c1daeSBarry Smith PetscClassIdRegister("class name",&classid); 7615c6c1daeSBarry Smith PetscLogEventRegister("User event name",classid,&USER_EVENT); 7625c6c1daeSBarry Smith PetscLogEventBegin(USER_EVENT,0,0,0,0); 7635c6c1daeSBarry Smith [code segment to monitor] 7645c6c1daeSBarry Smith PetscLogFlops(user_event_flops); 7655c6c1daeSBarry Smith PetscLogEventEnd(USER_EVENT,0,0,0,0); 7665c6c1daeSBarry Smith .ve 7675c6c1daeSBarry Smith 768d1f92df0SBarry Smith Level: intermediate 769d1f92df0SBarry Smith 7705c6c1daeSBarry Smith Notes: 7715c6c1daeSBarry Smith PETSc automatically logs library events if the code has been 772a2553e36SBarry Smith configured with --with-log (which is the default) and 773811af0c4SBarry Smith -log_view or -log_all is specified. `PetscLogEventRegister()` is 7745c6c1daeSBarry Smith intended for logging user events to supplement this PETSc 7755c6c1daeSBarry Smith information. 7765c6c1daeSBarry Smith 777495fc317SBarry Smith PETSc can gather data for use with the utilities Jumpshot 7785c6c1daeSBarry Smith (part of the MPICH distribution). If PETSc has been compiled 7795c6c1daeSBarry Smith with flag -DPETSC_HAVE_MPE (MPE is an additional utility within 7805c6c1daeSBarry Smith MPICH), the user can employ another command line option, -log_mpe, 7815c6c1daeSBarry Smith to create a logfile, "mpe.log", which can be visualized 782495fc317SBarry Smith Jumpshot. 7835c6c1daeSBarry Smith 7845c6c1daeSBarry Smith The classid is associated with each event so that classes of events 7855c6c1daeSBarry Smith can be disabled simultaneously, such as all matrix events. The user 786811af0c4SBarry Smith can either use an existing classid, such as `MAT_CLASSID`, or create 7875c6c1daeSBarry Smith their own as shown in the example. 7885c6c1daeSBarry Smith 789c5deb1d5SJed Brown If an existing event with the same name exists, its event handle is 790c5deb1d5SJed Brown returned instead of creating a new event. 791c5deb1d5SJed Brown 792d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogStageRegister()`, `PetscLogEventBegin()`, `PetscLogEventEnd()`, `PetscLogFlops()`, 793db781477SPatrick Sanan `PetscLogEventActivate()`, `PetscLogEventDeactivate()`, `PetscClassIdRegister()` 7945c6c1daeSBarry Smith @*/ 795d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogEventRegister(const char name[], PetscClassId classid, PetscLogEvent *event) 796d71ae5a4SJacob Faibussowitsch { 7975c6c1daeSBarry Smith PetscStageLog stageLog; 7985c6c1daeSBarry Smith int stage; 7995c6c1daeSBarry Smith 8005c6c1daeSBarry Smith PetscFunctionBegin; 8015c6c1daeSBarry Smith *event = PETSC_DECIDE; 8029566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 8039566063dSJacob Faibussowitsch PetscCall(PetscEventRegLogGetEvent(stageLog->eventLog, name, event)); 8043ba16761SJacob Faibussowitsch if (*event > 0) PetscFunctionReturn(PETSC_SUCCESS); 8059566063dSJacob Faibussowitsch PetscCall(PetscEventRegLogRegister(stageLog->eventLog, name, classid, event)); 8065c6c1daeSBarry Smith for (stage = 0; stage < stageLog->numStages; stage++) { 8079566063dSJacob Faibussowitsch PetscCall(PetscEventPerfLogEnsureSize(stageLog->stageInfo[stage].eventLog, stageLog->eventLog->numEvents)); 8089566063dSJacob Faibussowitsch PetscCall(PetscClassPerfLogEnsureSize(stageLog->stageInfo[stage].classLog, stageLog->classLog->numClasses)); 8095c6c1daeSBarry Smith } 8103ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 8115c6c1daeSBarry Smith } 8125c6c1daeSBarry Smith 8135c6c1daeSBarry Smith /*@ 814217044c2SLisandro Dalcin PetscLogEventSetCollective - Indicates that a particular event is collective. 815217044c2SLisandro Dalcin 816217044c2SLisandro Dalcin Not Collective 817217044c2SLisandro Dalcin 818d8d19677SJose E. Roman Input Parameters: 819217044c2SLisandro Dalcin + event - The event id 820d5b43468SJose E. Roman - collective - Boolean flag indicating whether a particular event is collective 821217044c2SLisandro Dalcin 822d1f92df0SBarry Smith Level: developer 823d1f92df0SBarry Smith 824811af0c4SBarry Smith Notes: 825811af0c4SBarry Smith New events returned from `PetscLogEventRegister()` are collective by default. 826811af0c4SBarry Smith 827811af0c4SBarry Smith Collective events are handled specially if the -log_sync is used. In that case the logging saves information about 828811af0c4SBarry Smith two parts of the event; the time for all the MPI ranks to synchronize and then the time for the actual computation/communication 829811af0c4SBarry Smith to be performed. This option is useful to debug imbalance within the computations or communications 830217044c2SLisandro Dalcin 831d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogEventBegin()`, `PetscLogEventEnd()`, `PetscLogEventRegister()` 832217044c2SLisandro Dalcin @*/ 833d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogEventSetCollective(PetscLogEvent event, PetscBool collective) 834d71ae5a4SJacob Faibussowitsch { 835217044c2SLisandro Dalcin PetscStageLog stageLog; 836217044c2SLisandro Dalcin PetscEventRegLog eventRegLog; 837217044c2SLisandro Dalcin 838217044c2SLisandro Dalcin PetscFunctionBegin; 8399566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 8409566063dSJacob Faibussowitsch PetscCall(PetscStageLogGetEventRegLog(stageLog, &eventRegLog)); 841cc73adaaSBarry Smith PetscCheck(event >= 0 && event <= eventRegLog->numEvents, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Invalid event id"); 842217044c2SLisandro Dalcin eventRegLog->eventInfo[event].collective = collective; 8433ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 844217044c2SLisandro Dalcin } 845217044c2SLisandro Dalcin 846217044c2SLisandro Dalcin /*@ 847fa2bb9feSLisandro Dalcin PetscLogEventIncludeClass - Activates event logging for a PETSc object class in every stage. 848fa2bb9feSLisandro Dalcin 849fa2bb9feSLisandro Dalcin Not Collective 850fa2bb9feSLisandro Dalcin 851fa2bb9feSLisandro Dalcin Input Parameter: 852811af0c4SBarry Smith . classid - The object class, for example `MAT_CLASSID`, `SNES_CLASSID`, etc. 853fa2bb9feSLisandro Dalcin 854fa2bb9feSLisandro Dalcin Level: developer 855fa2bb9feSLisandro Dalcin 856d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogEventActivateClass()`, `PetscLogEventDeactivateClass()`, `PetscLogEventActivate()`, `PetscLogEventDeactivate()` 857fa2bb9feSLisandro Dalcin @*/ 858d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogEventIncludeClass(PetscClassId classid) 859d71ae5a4SJacob Faibussowitsch { 860fa2bb9feSLisandro Dalcin PetscStageLog stageLog; 861fa2bb9feSLisandro Dalcin int stage; 862fa2bb9feSLisandro Dalcin 863fa2bb9feSLisandro Dalcin PetscFunctionBegin; 8649566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 86548a46eb9SPierre Jolivet for (stage = 0; stage < stageLog->numStages; stage++) PetscCall(PetscEventPerfLogActivateClass(stageLog->stageInfo[stage].eventLog, stageLog->eventLog, classid)); 8663ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 867fa2bb9feSLisandro Dalcin } 868fa2bb9feSLisandro Dalcin 869fa2bb9feSLisandro Dalcin /*@ 870fa2bb9feSLisandro Dalcin PetscLogEventExcludeClass - Deactivates event logging for a PETSc object class in every stage. 871fa2bb9feSLisandro Dalcin 872fa2bb9feSLisandro Dalcin Not Collective 873fa2bb9feSLisandro Dalcin 874fa2bb9feSLisandro Dalcin Input Parameter: 875811af0c4SBarry Smith . classid - The object class, for example `MAT_CLASSID`, `SNES_CLASSID`, etc. 876fa2bb9feSLisandro Dalcin 877fa2bb9feSLisandro Dalcin Level: developer 878fa2bb9feSLisandro Dalcin 879811af0c4SBarry Smith Note: 880811af0c4SBarry Smith If a class is excluded then events associated with that class are not logged. 881811af0c4SBarry Smith 882d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogEventDeactivateClass()`, `PetscLogEventActivateClass()`, `PetscLogEventDeactivate()`, `PetscLogEventActivate()` 883fa2bb9feSLisandro Dalcin @*/ 884d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogEventExcludeClass(PetscClassId classid) 885d71ae5a4SJacob Faibussowitsch { 886fa2bb9feSLisandro Dalcin PetscStageLog stageLog; 887fa2bb9feSLisandro Dalcin int stage; 888fa2bb9feSLisandro Dalcin 889fa2bb9feSLisandro Dalcin PetscFunctionBegin; 8909566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 89148a46eb9SPierre Jolivet for (stage = 0; stage < stageLog->numStages; stage++) PetscCall(PetscEventPerfLogDeactivateClass(stageLog->stageInfo[stage].eventLog, stageLog->eventLog, classid)); 8923ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 893fa2bb9feSLisandro Dalcin } 894fa2bb9feSLisandro Dalcin 895fa2bb9feSLisandro Dalcin /*@ 8965c6c1daeSBarry Smith PetscLogEventActivate - Indicates that a particular event should be logged. 8975c6c1daeSBarry Smith 8985c6c1daeSBarry Smith Not Collective 8995c6c1daeSBarry Smith 9005c6c1daeSBarry Smith Input Parameter: 9015c6c1daeSBarry Smith . event - The event id 9025c6c1daeSBarry Smith 903*10450e9eSJacob Faibussowitsch Example Usage: 9045c6c1daeSBarry Smith .vb 9055c6c1daeSBarry Smith PetscLogEventDeactivate(VEC_SetValues); 9065c6c1daeSBarry Smith [code where you do not want to log VecSetValues()] 9075c6c1daeSBarry Smith PetscLogEventActivate(VEC_SetValues); 9085c6c1daeSBarry Smith [code where you do want to log VecSetValues()] 9095c6c1daeSBarry Smith .ve 9105c6c1daeSBarry Smith 911d1f92df0SBarry Smith Level: advanced 912d1f92df0SBarry Smith 9135c6c1daeSBarry Smith Note: 9145c6c1daeSBarry Smith The event may be either a pre-defined PETSc event (found in include/petsclog.h) 915811af0c4SBarry Smith or an event number obtained with `PetscLogEventRegister()`. 9165c6c1daeSBarry Smith 917d1f92df0SBarry Smith .seealso: [](ch_profiling), `PlogEventDeactivate()`, `PlogEventDeactivatePush()`, `PetscLogEventDeactivatePop()` 9185c6c1daeSBarry Smith @*/ 919d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogEventActivate(PetscLogEvent event) 920d71ae5a4SJacob Faibussowitsch { 9215c6c1daeSBarry Smith PetscStageLog stageLog; 9225c6c1daeSBarry Smith int stage; 9235c6c1daeSBarry Smith 9245c6c1daeSBarry Smith PetscFunctionBegin; 9259566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 9269566063dSJacob Faibussowitsch PetscCall(PetscStageLogGetCurrent(stageLog, &stage)); 9279566063dSJacob Faibussowitsch PetscCall(PetscEventPerfLogActivate(stageLog->stageInfo[stage].eventLog, event)); 9283ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 9295c6c1daeSBarry Smith } 9305c6c1daeSBarry Smith 9315c6c1daeSBarry Smith /*@ 9325c6c1daeSBarry Smith PetscLogEventDeactivate - Indicates that a particular event should not be logged. 9335c6c1daeSBarry Smith 9345c6c1daeSBarry Smith Not Collective 9355c6c1daeSBarry Smith 9365c6c1daeSBarry Smith Input Parameter: 9375c6c1daeSBarry Smith . event - The event id 9385c6c1daeSBarry Smith 939*10450e9eSJacob Faibussowitsch Example Usage: 9405c6c1daeSBarry Smith .vb 9415c6c1daeSBarry Smith PetscLogEventDeactivate(VEC_SetValues); 9425c6c1daeSBarry Smith [code where you do not want to log VecSetValues()] 9435c6c1daeSBarry Smith PetscLogEventActivate(VEC_SetValues); 9445c6c1daeSBarry Smith [code where you do want to log VecSetValues()] 9455c6c1daeSBarry Smith .ve 9465c6c1daeSBarry Smith 947d1f92df0SBarry Smith Level: advanced 948d1f92df0SBarry Smith 9495c6c1daeSBarry Smith Note: 9505c6c1daeSBarry Smith The event may be either a pre-defined PETSc event (found in 951811af0c4SBarry Smith include/petsclog.h) or an event number obtained with `PetscLogEventRegister()`). 9525c6c1daeSBarry Smith 953d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogEventActivate()`, `PetscLogEventDeactivatePush()`, `PetscLogEventDeactivatePop()` 9545c6c1daeSBarry Smith @*/ 955d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogEventDeactivate(PetscLogEvent event) 956d71ae5a4SJacob Faibussowitsch { 9575c6c1daeSBarry Smith PetscStageLog stageLog; 9585c6c1daeSBarry Smith int stage; 9595c6c1daeSBarry Smith 9605c6c1daeSBarry Smith PetscFunctionBegin; 9619566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 9629566063dSJacob Faibussowitsch PetscCall(PetscStageLogGetCurrent(stageLog, &stage)); 9639566063dSJacob Faibussowitsch PetscCall(PetscEventPerfLogDeactivate(stageLog->stageInfo[stage].eventLog, event)); 9643ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 9655c6c1daeSBarry Smith } 9665c6c1daeSBarry Smith 9675c6c1daeSBarry Smith /*@ 968811af0c4SBarry Smith PetscLogEventDeactivatePush - Indicates that a particular event should not be logged until `PetscLogEventDeactivatePop()` is called 969c00cb57fSBarry Smith 970c00cb57fSBarry Smith Not Collective 971c00cb57fSBarry Smith 972c00cb57fSBarry Smith Input Parameter: 973c00cb57fSBarry Smith . event - The event id 974c00cb57fSBarry Smith 975*10450e9eSJacob Faibussowitsch Example Usage: 976c00cb57fSBarry Smith .vb 977c00cb57fSBarry Smith PetscLogEventDeactivatePush(VEC_SetValues); 978c00cb57fSBarry Smith [code where you do not want to log VecSetValues()] 979c00cb57fSBarry Smith PetscLogEventDeactivatePop(VEC_SetValues); 980c00cb57fSBarry Smith [code where you do want to log VecSetValues()] 981c00cb57fSBarry Smith .ve 982c00cb57fSBarry Smith 983d1f92df0SBarry Smith Level: advanced 984d1f92df0SBarry Smith 985c00cb57fSBarry Smith Note: 986c00cb57fSBarry Smith The event may be either a pre-defined PETSc event (found in 987811af0c4SBarry Smith include/petsclog.h) or an event number obtained with `PetscLogEventRegister()`). 988c00cb57fSBarry Smith 989d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogEventActivate()`, `PetscLogEventDeactivatePop()`, `PetscLogEventDeactivate()` 990c00cb57fSBarry Smith @*/ 991d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogEventDeactivatePush(PetscLogEvent event) 992d71ae5a4SJacob Faibussowitsch { 993c00cb57fSBarry Smith PetscStageLog stageLog; 994c00cb57fSBarry Smith int stage; 995c00cb57fSBarry Smith 996c00cb57fSBarry Smith PetscFunctionBegin; 9979566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 9989566063dSJacob Faibussowitsch PetscCall(PetscStageLogGetCurrent(stageLog, &stage)); 9999566063dSJacob Faibussowitsch PetscCall(PetscEventPerfLogDeactivatePush(stageLog->stageInfo[stage].eventLog, event)); 10003ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 1001c00cb57fSBarry Smith } 1002c00cb57fSBarry Smith 1003c00cb57fSBarry Smith /*@ 1004811af0c4SBarry Smith PetscLogEventDeactivatePop - Indicates that a particular event should again be logged after the logging was turned off with `PetscLogEventDeactivatePush()` 1005c00cb57fSBarry Smith 1006c00cb57fSBarry Smith Not Collective 1007c00cb57fSBarry Smith 1008c00cb57fSBarry Smith Input Parameter: 1009c00cb57fSBarry Smith . event - The event id 1010c00cb57fSBarry Smith 1011*10450e9eSJacob Faibussowitsch Example Usage: 1012c00cb57fSBarry Smith .vb 1013c00cb57fSBarry Smith PetscLogEventDeactivatePush(VEC_SetValues); 1014c00cb57fSBarry Smith [code where you do not want to log VecSetValues()] 1015c00cb57fSBarry Smith PetscLogEventDeactivatePop(VEC_SetValues); 1016c00cb57fSBarry Smith [code where you do want to log VecSetValues()] 1017c00cb57fSBarry Smith .ve 1018c00cb57fSBarry Smith 1019d1f92df0SBarry Smith Level: advanced 1020d1f92df0SBarry Smith 1021c00cb57fSBarry Smith Note: 1022c00cb57fSBarry Smith The event may be either a pre-defined PETSc event (found in 1023811af0c4SBarry Smith include/petsclog.h) or an event number obtained with `PetscLogEventRegister()`). 1024c00cb57fSBarry Smith 1025d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogEventActivate()`, `PetscLogEventDeactivatePush()` 1026c00cb57fSBarry Smith @*/ 1027d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogEventDeactivatePop(PetscLogEvent event) 1028d71ae5a4SJacob Faibussowitsch { 1029c00cb57fSBarry Smith PetscStageLog stageLog; 1030c00cb57fSBarry Smith int stage; 1031c00cb57fSBarry Smith 1032c00cb57fSBarry Smith PetscFunctionBegin; 10339566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 10349566063dSJacob Faibussowitsch PetscCall(PetscStageLogGetCurrent(stageLog, &stage)); 10359566063dSJacob Faibussowitsch PetscCall(PetscEventPerfLogDeactivatePop(stageLog->stageInfo[stage].eventLog, event)); 10363ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 1037c00cb57fSBarry Smith } 1038c00cb57fSBarry Smith 1039c00cb57fSBarry Smith /*@ 1040811af0c4SBarry Smith PetscLogEventSetActiveAll - Turns on logging of all events 10415c6c1daeSBarry Smith 10425c6c1daeSBarry Smith Not Collective 10435c6c1daeSBarry Smith 10445c6c1daeSBarry Smith Input Parameters: 10455c6c1daeSBarry Smith + event - The event id 10465c6c1daeSBarry Smith - isActive - The activity flag determining whether the event is logged 10475c6c1daeSBarry Smith 10485c6c1daeSBarry Smith Level: advanced 10495c6c1daeSBarry Smith 1050d1f92df0SBarry Smith .seealso: [](ch_profiling), `PlogEventActivate()`, `PlogEventDeactivate()` 10515c6c1daeSBarry Smith @*/ 1052d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogEventSetActiveAll(PetscLogEvent event, PetscBool isActive) 1053d71ae5a4SJacob Faibussowitsch { 10545c6c1daeSBarry Smith PetscStageLog stageLog; 10555c6c1daeSBarry Smith int stage; 10565c6c1daeSBarry Smith 10575c6c1daeSBarry Smith PetscFunctionBegin; 10589566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 10595c6c1daeSBarry Smith for (stage = 0; stage < stageLog->numStages; stage++) { 10605c6c1daeSBarry Smith if (isActive) { 10619566063dSJacob Faibussowitsch PetscCall(PetscEventPerfLogActivate(stageLog->stageInfo[stage].eventLog, event)); 10625c6c1daeSBarry Smith } else { 10639566063dSJacob Faibussowitsch PetscCall(PetscEventPerfLogDeactivate(stageLog->stageInfo[stage].eventLog, event)); 10645c6c1daeSBarry Smith } 10655c6c1daeSBarry Smith } 10663ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 10675c6c1daeSBarry Smith } 10685c6c1daeSBarry Smith 10695c6c1daeSBarry Smith /*@ 1070811af0c4SBarry Smith PetscLogEventActivateClass - Activates event logging for a PETSc object class for the current stage 10715c6c1daeSBarry Smith 10725c6c1daeSBarry Smith Not Collective 10735c6c1daeSBarry Smith 10745c6c1daeSBarry Smith Input Parameter: 1075811af0c4SBarry Smith . classid - The event class, for example `MAT_CLASSID`, `SNES_CLASSID`, etc. 10765c6c1daeSBarry Smith 10775c6c1daeSBarry Smith Level: developer 10785c6c1daeSBarry Smith 1079d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogEventIncludeClass()`, `PetscLogEventExcludeClass()`, `PetscLogEventDeactivateClass()`, `PetscLogEventActivate()`, `PetscLogEventDeactivate()` 10805c6c1daeSBarry Smith @*/ 1081d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogEventActivateClass(PetscClassId classid) 1082d71ae5a4SJacob Faibussowitsch { 10835c6c1daeSBarry Smith PetscStageLog stageLog; 10845c6c1daeSBarry Smith int stage; 10855c6c1daeSBarry Smith 10865c6c1daeSBarry Smith PetscFunctionBegin; 10879566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 10889566063dSJacob Faibussowitsch PetscCall(PetscStageLogGetCurrent(stageLog, &stage)); 10899566063dSJacob Faibussowitsch PetscCall(PetscEventPerfLogActivateClass(stageLog->stageInfo[stage].eventLog, stageLog->eventLog, classid)); 10903ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 10915c6c1daeSBarry Smith } 10925c6c1daeSBarry Smith 10935c6c1daeSBarry Smith /*@ 1094811af0c4SBarry Smith PetscLogEventDeactivateClass - Deactivates event logging for a PETSc object class for the current stage 10955c6c1daeSBarry Smith 10965c6c1daeSBarry Smith Not Collective 10975c6c1daeSBarry Smith 10985c6c1daeSBarry Smith Input Parameter: 1099811af0c4SBarry Smith . classid - The event class, for example `MAT_CLASSID`, `SNES_CLASSID`, etc. 11005c6c1daeSBarry Smith 11015c6c1daeSBarry Smith Level: developer 11025c6c1daeSBarry Smith 1103d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogEventIncludeClass()`, `PetscLogEventExcludeClass()`, `PetscLogEventActivateClass()`, `PetscLogEventActivate()`, `PetscLogEventDeactivate()` 11045c6c1daeSBarry Smith @*/ 1105d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogEventDeactivateClass(PetscClassId classid) 1106d71ae5a4SJacob Faibussowitsch { 11075c6c1daeSBarry Smith PetscStageLog stageLog; 11085c6c1daeSBarry Smith int stage; 11095c6c1daeSBarry Smith 11105c6c1daeSBarry Smith PetscFunctionBegin; 11119566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 11129566063dSJacob Faibussowitsch PetscCall(PetscStageLogGetCurrent(stageLog, &stage)); 11139566063dSJacob Faibussowitsch PetscCall(PetscEventPerfLogDeactivateClass(stageLog->stageInfo[stage].eventLog, stageLog->eventLog, classid)); 11143ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 11155c6c1daeSBarry Smith } 11165c6c1daeSBarry Smith 11175c6c1daeSBarry Smith /*MC 111862872c28SLisandro Dalcin PetscLogEventSync - Synchronizes the beginning of a user event. 111962872c28SLisandro Dalcin 112062872c28SLisandro Dalcin Synopsis: 112162872c28SLisandro Dalcin #include <petsclog.h> 112262872c28SLisandro Dalcin PetscErrorCode PetscLogEventSync(int e, MPI_Comm comm) 112362872c28SLisandro Dalcin 112462872c28SLisandro Dalcin Collective 112562872c28SLisandro Dalcin 112662872c28SLisandro Dalcin Input Parameters: 112762872c28SLisandro Dalcin + e - integer associated with the event obtained from PetscLogEventRegister() 112862872c28SLisandro Dalcin - comm - an MPI communicator 112962872c28SLisandro Dalcin 1130*10450e9eSJacob Faibussowitsch Example Usage: 113162872c28SLisandro Dalcin .vb 113262872c28SLisandro Dalcin PetscLogEvent USER_EVENT; 1133*10450e9eSJacob Faibussowitsch 113462872c28SLisandro Dalcin PetscLogEventRegister("User event", 0, &USER_EVENT); 113562872c28SLisandro Dalcin PetscLogEventSync(USER_EVENT, PETSC_COMM_WORLD); 113662872c28SLisandro Dalcin PetscLogEventBegin(USER_EVENT, 0, 0, 0, 0); 113762872c28SLisandro Dalcin [code segment to monitor] 113862872c28SLisandro Dalcin PetscLogEventEnd(USER_EVENT, 0, 0, 0 , 0); 113962872c28SLisandro Dalcin .ve 114062872c28SLisandro Dalcin 1141d1f92df0SBarry Smith Level: developer 1142d1f92df0SBarry Smith 1143811af0c4SBarry Smith Note: 1144*10450e9eSJacob Faibussowitsch This routine should be called only if there is not a `PetscObject` available to pass to 1145*10450e9eSJacob Faibussowitsch `PetscLogEventBegin()`. 114662872c28SLisandro Dalcin 1147d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogEventRegister()`, `PetscLogEventBegin()`, `PetscLogEventEnd()` 114862872c28SLisandro Dalcin M*/ 114962872c28SLisandro Dalcin 115062872c28SLisandro Dalcin /*MC 11515c6c1daeSBarry Smith PetscLogEventBegin - Logs the beginning of a user event. 11525c6c1daeSBarry Smith 11535c6c1daeSBarry Smith Synopsis: 1154aaa7dc30SBarry Smith #include <petsclog.h> 1155f2ba6396SBarry Smith PetscErrorCode PetscLogEventBegin(int e, PetscObject o1, PetscObject o2, PetscObject o3, PetscObject o4) 11565c6c1daeSBarry Smith 11575c6c1daeSBarry Smith Not Collective 11585c6c1daeSBarry Smith 11595c6c1daeSBarry Smith Input Parameters: 11605c6c1daeSBarry Smith + e - integer associated with the event obtained from PetscLogEventRegister() 11615c6c1daeSBarry Smith - o1,o2,o3,o4 - objects associated with the event, or 0 11625c6c1daeSBarry Smith 11635c6c1daeSBarry Smith Fortran Synopsis: 11645c6c1daeSBarry Smith void PetscLogEventBegin(int e, PetscErrorCode ierr) 11655c6c1daeSBarry Smith 1166*10450e9eSJacob Faibussowitsch Example Usage: 11675c6c1daeSBarry Smith .vb 11685c6c1daeSBarry Smith PetscLogEvent USER_EVENT; 1169*10450e9eSJacob Faibussowitsch 11705c6c1daeSBarry Smith PetscLogDouble user_event_flops; 11715c6c1daeSBarry Smith PetscLogEventRegister("User event",0, &USER_EVENT); 11725c6c1daeSBarry Smith PetscLogEventBegin(USER_EVENT, 0, 0, 0, 0); 11735c6c1daeSBarry Smith [code segment to monitor] 11745c6c1daeSBarry Smith PetscLogFlops(user_event_flops); 11755c6c1daeSBarry Smith PetscLogEventEnd(USER_EVENT, 0, 0, 0, 0); 11765c6c1daeSBarry Smith .ve 11775c6c1daeSBarry Smith 1178d1f92df0SBarry Smith Level: intermediate 1179d1f92df0SBarry Smith 1180811af0c4SBarry Smith Developer Note: 1181*10450e9eSJacob Faibussowitsch `PetscLogEventBegin()` and `PetscLogEventBegin()` return error codes instead of explicitly 1182*10450e9eSJacob Faibussowitsch handling the errors that occur in the macro directly because other packages that use this 1183*10450e9eSJacob Faibussowitsch macros have used them in their own functions or methods that do not return error codes and it 1184*10450e9eSJacob Faibussowitsch would be disruptive to change the current behavior. 1185d0609cedSBarry Smith 1186d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogEventRegister()`, `PetscLogEventEnd()`, `PetscLogFlops()` 11875c6c1daeSBarry Smith M*/ 11885c6c1daeSBarry Smith 11895c6c1daeSBarry Smith /*MC 11905c6c1daeSBarry Smith PetscLogEventEnd - Log the end of a user event. 11915c6c1daeSBarry Smith 11925c6c1daeSBarry Smith Synopsis: 1193aaa7dc30SBarry Smith #include <petsclog.h> 1194f2ba6396SBarry Smith PetscErrorCode PetscLogEventEnd(int e, PetscObject o1, PetscObject o2, PetscObject o3, PetscObject o4) 11955c6c1daeSBarry Smith 11965c6c1daeSBarry Smith Not Collective 11975c6c1daeSBarry Smith 11985c6c1daeSBarry Smith Input Parameters: 11995c6c1daeSBarry Smith + e - integer associated with the event obtained with PetscLogEventRegister() 12005c6c1daeSBarry Smith - o1,o2,o3,o4 - objects associated with the event, or 0 12015c6c1daeSBarry Smith 12025c6c1daeSBarry Smith Fortran Synopsis: 12035c6c1daeSBarry Smith void PetscLogEventEnd(int e, PetscErrorCode ierr) 12045c6c1daeSBarry Smith 1205*10450e9eSJacob Faibussowitsch Example Usage: 12065c6c1daeSBarry Smith .vb 12075c6c1daeSBarry Smith PetscLogEvent USER_EVENT; 1208*10450e9eSJacob Faibussowitsch 12095c6c1daeSBarry Smith PetscLogDouble user_event_flops; 1210*10450e9eSJacob Faibussowitsch PetscLogEventRegister("User event", 0, &USER_EVENT); 12115c6c1daeSBarry Smith PetscLogEventBegin(USER_EVENT, 0, 0, 0, 0); 12125c6c1daeSBarry Smith [code segment to monitor] 12135c6c1daeSBarry Smith PetscLogFlops(user_event_flops); 12145c6c1daeSBarry Smith PetscLogEventEnd(USER_EVENT, 0, 0, 0, 0); 12155c6c1daeSBarry Smith .ve 12165c6c1daeSBarry Smith 12175c6c1daeSBarry Smith Level: intermediate 12185c6c1daeSBarry Smith 1219d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogEventRegister()`, `PetscLogEventBegin()`, `PetscLogFlops()` 12205c6c1daeSBarry Smith M*/ 12215c6c1daeSBarry Smith 12225c6c1daeSBarry Smith /*@C 12235c6c1daeSBarry Smith PetscLogEventGetId - Returns the event id when given the event name. 12245c6c1daeSBarry Smith 12255c6c1daeSBarry Smith Not Collective 12265c6c1daeSBarry Smith 12275c6c1daeSBarry Smith Input Parameter: 12285c6c1daeSBarry Smith . name - The event name 12295c6c1daeSBarry Smith 12305c6c1daeSBarry Smith Output Parameter: 1231c5deb1d5SJed Brown . event - The event, or -1 if no event with that name exists 12325c6c1daeSBarry Smith 12335c6c1daeSBarry Smith Level: intermediate 12345c6c1daeSBarry Smith 1235d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogEventBegin()`, `PetscLogEventEnd()`, `PetscLogStageGetId()` 12365c6c1daeSBarry Smith @*/ 1237d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogEventGetId(const char name[], PetscLogEvent *event) 1238d71ae5a4SJacob Faibussowitsch { 12395c6c1daeSBarry Smith PetscStageLog stageLog; 12405c6c1daeSBarry Smith 12415c6c1daeSBarry Smith PetscFunctionBegin; 12429566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 12439566063dSJacob Faibussowitsch PetscCall(PetscEventRegLogGetEvent(stageLog->eventLog, name, event)); 12443ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 12455c6c1daeSBarry Smith } 12465c6c1daeSBarry Smith 1247d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogPushCurrentEvent_Internal(PetscLogEvent event) 1248d71ae5a4SJacob Faibussowitsch { 12497a101e5eSJacob Faibussowitsch PetscFunctionBegin; 1250c708d6e3SStefano Zampini if (!PetscDefined(HAVE_THREADSAFETY)) PetscCall(PetscIntStackPush(current_log_event_stack, event)); 12513ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 12527a101e5eSJacob Faibussowitsch } 12537a101e5eSJacob Faibussowitsch 1254d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogPopCurrentEvent_Internal(void) 1255d71ae5a4SJacob Faibussowitsch { 12567a101e5eSJacob Faibussowitsch PetscFunctionBegin; 1257c708d6e3SStefano Zampini if (!PetscDefined(HAVE_THREADSAFETY)) PetscCall(PetscIntStackPop(current_log_event_stack, NULL)); 12583ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 12597a101e5eSJacob Faibussowitsch } 12607a101e5eSJacob Faibussowitsch 1261d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogGetCurrentEvent_Internal(PetscLogEvent *event) 1262d71ae5a4SJacob Faibussowitsch { 12637a101e5eSJacob Faibussowitsch PetscBool empty; 12647a101e5eSJacob Faibussowitsch 12657a101e5eSJacob Faibussowitsch PetscFunctionBegin; 12667a101e5eSJacob Faibussowitsch PetscValidIntPointer(event, 1); 12677a101e5eSJacob Faibussowitsch *event = PETSC_DECIDE; 12687a101e5eSJacob Faibussowitsch PetscCall(PetscIntStackEmpty(current_log_event_stack, &empty)); 12697a101e5eSJacob Faibussowitsch if (!empty) PetscCall(PetscIntStackTop(current_log_event_stack, event)); 12703ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 12717a101e5eSJacob Faibussowitsch } 12727a101e5eSJacob Faibussowitsch 1273d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogEventPause_Internal(PetscLogEvent event) 1274d71ae5a4SJacob Faibussowitsch { 12757a101e5eSJacob Faibussowitsch PetscFunctionBegin; 12767a101e5eSJacob Faibussowitsch if (event != PETSC_DECIDE) PetscCall(PetscLogEventEnd(event, NULL, NULL, NULL, NULL)); 12773ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 12787a101e5eSJacob Faibussowitsch } 12797a101e5eSJacob Faibussowitsch 1280d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogEventResume_Internal(PetscLogEvent event) 1281d71ae5a4SJacob Faibussowitsch { 12827a101e5eSJacob Faibussowitsch PetscStageLog stageLog; 12837a101e5eSJacob Faibussowitsch PetscEventPerfLog eventLog; 12847a101e5eSJacob Faibussowitsch int stage; 12857a101e5eSJacob Faibussowitsch 12867a101e5eSJacob Faibussowitsch PetscFunctionBegin; 12873ba16761SJacob Faibussowitsch if (event == PETSC_DECIDE) PetscFunctionReturn(PETSC_SUCCESS); 12887a101e5eSJacob Faibussowitsch PetscCall(PetscLogEventBegin(event, NULL, NULL, NULL, NULL)); 12897a101e5eSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 12907a101e5eSJacob Faibussowitsch PetscCall(PetscStageLogGetCurrent(stageLog, &stage)); 12917a101e5eSJacob Faibussowitsch PetscCall(PetscStageLogGetEventPerfLog(stageLog, stage, &eventLog)); 12927a101e5eSJacob Faibussowitsch eventLog->eventInfo[event].count--; 12933ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 12947a101e5eSJacob Faibussowitsch } 12957a101e5eSJacob Faibussowitsch 12965c6c1daeSBarry Smith /*------------------------------------------------ Output Functions -------------------------------------------------*/ 12975c6c1daeSBarry Smith /*@C 12985c6c1daeSBarry Smith PetscLogDump - Dumps logs of objects to a file. This file is intended to 12995c6c1daeSBarry Smith be read by bin/petscview. This program no longer exists. 13005c6c1daeSBarry Smith 1301811af0c4SBarry Smith Collective on `PETSC_COMM_WORLD` 13025c6c1daeSBarry Smith 13035c6c1daeSBarry Smith Input Parameter: 1304aec76313SJacob Faibussowitsch . sname - an optional file name 13055c6c1daeSBarry Smith 1306*10450e9eSJacob Faibussowitsch Example Usage: 13075c6c1daeSBarry Smith .vb 13085c6c1daeSBarry Smith PetscInitialize(...); 1309bb1d7374SBarry Smith PetscLogDefaultBegin(); or PetscLogAllBegin(); 13105c6c1daeSBarry Smith ... code ... 13115c6c1daeSBarry Smith PetscLogDump(filename); 13125c6c1daeSBarry Smith PetscFinalize(); 13135c6c1daeSBarry Smith .ve 13145c6c1daeSBarry Smith 1315d1f92df0SBarry Smith Level: advanced 1316d1f92df0SBarry Smith 1317811af0c4SBarry Smith Note: 131837fdd005SBarry Smith The default file name is Log.<rank> where <rank> is the MPI process rank. If no name is specified, 13195c6c1daeSBarry Smith this file will be used. 13205c6c1daeSBarry Smith 1321d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogDefaultBegin()`, `PetscLogAllBegin()`, `PetscLogView()` 13225c6c1daeSBarry Smith @*/ 1323d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogDump(const char sname[]) 1324d71ae5a4SJacob Faibussowitsch { 13255c6c1daeSBarry Smith PetscStageLog stageLog; 13265c6c1daeSBarry Smith PetscEventPerfInfo *eventInfo; 13275c6c1daeSBarry Smith FILE *fd; 13285c6c1daeSBarry Smith char file[PETSC_MAX_PATH_LEN], fname[PETSC_MAX_PATH_LEN]; 13295c6c1daeSBarry Smith PetscLogDouble flops, _TotalTime; 13305c6c1daeSBarry Smith PetscMPIInt rank; 13315c6c1daeSBarry Smith int action, object, curStage; 13325c6c1daeSBarry Smith PetscLogEvent event; 13335c6c1daeSBarry Smith 13345c6c1daeSBarry Smith PetscFunctionBegin; 13355c6c1daeSBarry Smith /* Calculate the total elapsed time */ 13363ba16761SJacob Faibussowitsch PetscCall(PetscTime(&_TotalTime)); 13375c6c1daeSBarry Smith _TotalTime -= petsc_BaseTime; 13385c6c1daeSBarry Smith /* Open log file */ 13399566063dSJacob Faibussowitsch PetscCallMPI(MPI_Comm_rank(PETSC_COMM_WORLD, &rank)); 1340a364092eSJacob Faibussowitsch PetscCall(PetscSNPrintf(file, PETSC_STATIC_ARRAY_LENGTH(file), "%s.%d", sname && sname[0] ? sname : "Log", rank)); 13419566063dSJacob Faibussowitsch PetscCall(PetscFixFilename(file, fname)); 13429566063dSJacob Faibussowitsch PetscCall(PetscFOpen(PETSC_COMM_WORLD, fname, "w", &fd)); 134308401ef6SPierre Jolivet PetscCheck(!(rank == 0) || !(!fd), PETSC_COMM_SELF, PETSC_ERR_FILE_OPEN, "Cannot open file: %s", fname); 13445c6c1daeSBarry Smith /* Output totals */ 13459566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(PETSC_COMM_WORLD, fd, "Total Flop %14e %16.8e\n", petsc_TotalFlops, _TotalTime)); 13469566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(PETSC_COMM_WORLD, fd, "Clock Resolution %g\n", 0.0)); 13475c6c1daeSBarry Smith /* Output actions */ 13485c6c1daeSBarry Smith if (petsc_logActions) { 13499566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(PETSC_COMM_WORLD, fd, "Actions accomplished %d\n", petsc_numActions)); 13505c6c1daeSBarry Smith for (action = 0; action < petsc_numActions; action++) { 13519371c9d4SSatish Balay PetscCall(PetscFPrintf(PETSC_COMM_WORLD, fd, "%g %d %d %d %d %d %d %g %g %g\n", petsc_actions[action].time, petsc_actions[action].action, (int)petsc_actions[action].event, (int)petsc_actions[action].classid, petsc_actions[action].id1, 1352d0609cedSBarry Smith petsc_actions[action].id2, petsc_actions[action].id3, petsc_actions[action].flops, petsc_actions[action].mem, petsc_actions[action].maxmem)); 13535c6c1daeSBarry Smith } 13545c6c1daeSBarry Smith } 13555c6c1daeSBarry Smith /* Output objects */ 13565c6c1daeSBarry Smith if (petsc_logObjects) { 13579566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(PETSC_COMM_WORLD, fd, "Objects created %d destroyed %d\n", petsc_numObjects, petsc_numObjectsDestroyed)); 13585c6c1daeSBarry Smith for (object = 0; object < petsc_numObjects; object++) { 13599566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(PETSC_COMM_WORLD, fd, "Parent ID: %d Memory: %d\n", petsc_objects[object].parent, (int)petsc_objects[object].mem)); 13605c6c1daeSBarry Smith if (!petsc_objects[object].name[0]) { 13619566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(PETSC_COMM_WORLD, fd, "No Name\n")); 13625c6c1daeSBarry Smith } else { 13639566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(PETSC_COMM_WORLD, fd, "Name: %s\n", petsc_objects[object].name)); 13645c6c1daeSBarry Smith } 13655c6c1daeSBarry Smith if (petsc_objects[object].info[0] != 0) { 13669566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(PETSC_COMM_WORLD, fd, "No Info\n")); 13675c6c1daeSBarry Smith } else { 13689566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(PETSC_COMM_WORLD, fd, "Info: %s\n", petsc_objects[object].info)); 13695c6c1daeSBarry Smith } 13705c6c1daeSBarry Smith } 13715c6c1daeSBarry Smith } 13725c6c1daeSBarry Smith /* Output events */ 13739566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(PETSC_COMM_WORLD, fd, "Event log:\n")); 13749566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 13759566063dSJacob Faibussowitsch PetscCall(PetscIntStackTop(stageLog->stack, &curStage)); 13765c6c1daeSBarry Smith eventInfo = stageLog->stageInfo[curStage].eventLog->eventInfo; 13775c6c1daeSBarry Smith for (event = 0; event < stageLog->stageInfo[curStage].eventLog->numEvents; event++) { 1378a297a907SKarl Rupp if (eventInfo[event].time != 0.0) flops = eventInfo[event].flops / eventInfo[event].time; 1379a297a907SKarl Rupp else flops = 0.0; 1380d0609cedSBarry Smith PetscCall(PetscFPrintf(PETSC_COMM_WORLD, fd, "%d %16d %16g %16g %16g\n", event, eventInfo[event].count, eventInfo[event].flops, eventInfo[event].time, flops)); 13815c6c1daeSBarry Smith } 13829566063dSJacob Faibussowitsch PetscCall(PetscFClose(PETSC_COMM_WORLD, fd)); 13833ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 13845c6c1daeSBarry Smith } 13855c6c1daeSBarry Smith 1386f14045dbSBarry Smith /* 1387f14045dbSBarry Smith PetscLogView_Detailed - Each process prints the times for its own events 1388f14045dbSBarry Smith 1389f14045dbSBarry Smith */ 1390*10450e9eSJacob Faibussowitsch static PetscErrorCode PetscLogView_Detailed(PetscViewer viewer) 1391d71ae5a4SJacob Faibussowitsch { 13922dff6485SMatthew G. Knepley PetscStageLog stageLog; 1393217044c2SLisandro Dalcin PetscEventPerfInfo *eventInfo = NULL, *stageInfo = NULL; 1394217044c2SLisandro Dalcin PetscLogDouble locTotalTime, numRed, maxMem; 13952dff6485SMatthew G. Knepley int numStages, numEvents, stage, event; 1396217044c2SLisandro Dalcin MPI_Comm comm = PetscObjectComm((PetscObject)viewer); 13972dff6485SMatthew G. Knepley PetscMPIInt rank, size; 1398f14045dbSBarry Smith 1399f14045dbSBarry Smith PetscFunctionBegin; 14009566063dSJacob Faibussowitsch PetscCallMPI(MPI_Comm_size(comm, &size)); 14019566063dSJacob Faibussowitsch PetscCallMPI(MPI_Comm_rank(comm, &rank)); 14022dff6485SMatthew G. Knepley /* Must preserve reduction count before we go on */ 14032dff6485SMatthew G. Knepley numRed = petsc_allreduce_ct + petsc_gather_ct + petsc_scatter_ct; 14042d1ec343SBarry Smith /* Get the total elapsed time */ 14053ba16761SJacob Faibussowitsch PetscCall(PetscTime(&locTotalTime)); 14069371c9d4SSatish Balay locTotalTime -= petsc_BaseTime; 14079566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIIPrintf(viewer, "size = %d\n", size)); 14089566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIIPrintf(viewer, "LocalTimes = {}\n")); 14099566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIIPrintf(viewer, "LocalMessages = {}\n")); 14109566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIIPrintf(viewer, "LocalMessageLens = {}\n")); 14119566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIIPrintf(viewer, "LocalReductions = {}\n")); 14129566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIIPrintf(viewer, "LocalFlop = {}\n")); 14139566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIIPrintf(viewer, "LocalObjects = {}\n")); 14149566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIIPrintf(viewer, "LocalMemory = {}\n")); 14159566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 14169566063dSJacob Faibussowitsch PetscCallMPI(MPI_Allreduce(&stageLog->numStages, &numStages, 1, MPI_INT, MPI_MAX, comm)); 14179566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIIPrintf(viewer, "Stages = {}\n")); 14182d1ec343SBarry Smith for (stage = 0; stage < numStages; stage++) { 14199566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIIPrintf(viewer, "Stages[\"%s\"] = {}\n", stageLog->stageInfo[stage].name)); 14209566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIIPrintf(viewer, "Stages[\"%s\"][\"summary\"] = {}\n", stageLog->stageInfo[stage].name)); 14219566063dSJacob Faibussowitsch PetscCallMPI(MPI_Allreduce(&stageLog->stageInfo[stage].eventLog->numEvents, &numEvents, 1, MPI_INT, MPI_MAX, comm)); 142248a46eb9SPierre Jolivet for (event = 0; event < numEvents; event++) PetscCall(PetscViewerASCIIPrintf(viewer, "Stages[\"%s\"][\"%s\"] = {}\n", stageLog->stageInfo[stage].name, stageLog->eventLog->eventInfo[event].name)); 14232d1ec343SBarry Smith } 14249566063dSJacob Faibussowitsch PetscCall(PetscMallocGetMaximumUsage(&maxMem)); 14259566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIIPushSynchronized(viewer)); 14269566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "LocalTimes[%d] = %g\n", rank, locTotalTime)); 14279566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "LocalMessages[%d] = %g\n", rank, (petsc_irecv_ct + petsc_isend_ct + petsc_recv_ct + petsc_send_ct))); 14289566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "LocalMessageLens[%d] = %g\n", rank, (petsc_irecv_len + petsc_isend_len + petsc_recv_len + petsc_send_len))); 14299566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "LocalReductions[%d] = %g\n", rank, numRed)); 14309566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "LocalFlop[%d] = %g\n", rank, petsc_TotalFlops)); 14319566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "LocalObjects[%d] = %d\n", rank, petsc_numObjects)); 14329566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "LocalMemory[%d] = %g\n", rank, maxMem)); 14339566063dSJacob Faibussowitsch PetscCall(PetscViewerFlush(viewer)); 14342d1ec343SBarry Smith for (stage = 0; stage < numStages; stage++) { 1435217044c2SLisandro Dalcin stageInfo = &stageLog->stageInfo[stage].perfInfo; 14369371c9d4SSatish Balay PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "Stages[\"%s\"][\"summary\"][%d] = {\"time\" : %g, \"numMessages\" : %g, \"messageLength\" : %g, \"numReductions\" : %g, \"flop\" : %g}\n", stageLog->stageInfo[stage].name, rank, stageInfo->time, 14379371c9d4SSatish Balay stageInfo->numMessages, stageInfo->messageLength, stageInfo->numReductions, stageInfo->flops)); 14389566063dSJacob Faibussowitsch PetscCallMPI(MPI_Allreduce(&stageLog->stageInfo[stage].eventLog->numEvents, &numEvents, 1, MPI_INT, MPI_MAX, comm)); 14392d1ec343SBarry Smith for (event = 0; event < numEvents; event++) { 1440217044c2SLisandro Dalcin eventInfo = &stageLog->stageInfo[stage].eventLog->eventInfo[event]; 1441d0609cedSBarry Smith PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "Stages[\"%s\"][\"%s\"][%d] = {\"count\" : %d, \"time\" : %g, \"syncTime\" : %g, \"numMessages\" : %g, \"messageLength\" : %g, \"numReductions\" : %g, \"flop\" : %g", 14429371c9d4SSatish Balay stageLog->stageInfo[stage].name, stageLog->eventLog->eventInfo[event].name, rank, eventInfo->count, eventInfo->time, eventInfo->syncTime, eventInfo->numMessages, eventInfo->messageLength, eventInfo->numReductions, 14439371c9d4SSatish Balay eventInfo->flops)); 1444891e75beSMatthew G. Knepley if (eventInfo->dof[0] >= 0.) { 1445891e75beSMatthew G. Knepley PetscInt d, e; 14465d68e14cSMatthew G. Knepley 14479566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, ", \"dof\" : [")); 1448891e75beSMatthew G. Knepley for (d = 0; d < 8; ++d) { 14499566063dSJacob Faibussowitsch if (d > 0) PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, ", ")); 14509566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "%g", eventInfo->dof[d])); 1451891e75beSMatthew G. Knepley } 14529566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "]")); 14539566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, ", \"error\" : [")); 14545d68e14cSMatthew G. Knepley for (e = 0; e < 8; ++e) { 14559566063dSJacob Faibussowitsch if (e > 0) PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, ", ")); 14569566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "%g", eventInfo->errors[e])); 14575d68e14cSMatthew G. Knepley } 14589566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "]")); 14595d68e14cSMatthew G. Knepley } 14609566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "}\n")); 14612d1ec343SBarry Smith } 14622d1ec343SBarry Smith } 14639566063dSJacob Faibussowitsch PetscCall(PetscViewerFlush(viewer)); 14649566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIIPopSynchronized(viewer)); 14653ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 1466f14045dbSBarry Smith } 1467f14045dbSBarry Smith 146838144912Sdeepblu2718 /* 146938144912Sdeepblu2718 PetscLogView_CSV - Each process prints the times for its own events in Comma-Separated Value Format 147038144912Sdeepblu2718 */ 1471*10450e9eSJacob Faibussowitsch static PetscErrorCode PetscLogView_CSV(PetscViewer viewer) 1472d71ae5a4SJacob Faibussowitsch { 147338144912Sdeepblu2718 PetscStageLog stageLog; 1474669c5be0SSatish Balay PetscEventPerfInfo *eventInfo = NULL; 147538144912Sdeepblu2718 PetscLogDouble locTotalTime, maxMem; 147638144912Sdeepblu2718 int numStages, numEvents, stage, event; 147738144912Sdeepblu2718 MPI_Comm comm = PetscObjectComm((PetscObject)viewer); 147838144912Sdeepblu2718 PetscMPIInt rank, size; 147938144912Sdeepblu2718 148038144912Sdeepblu2718 PetscFunctionBegin; 14819566063dSJacob Faibussowitsch PetscCallMPI(MPI_Comm_size(comm, &size)); 14829566063dSJacob Faibussowitsch PetscCallMPI(MPI_Comm_rank(comm, &rank)); 148338144912Sdeepblu2718 /* Must preserve reduction count before we go on */ 148438144912Sdeepblu2718 /* Get the total elapsed time */ 14853ba16761SJacob Faibussowitsch PetscCall(PetscTime(&locTotalTime)); 14869371c9d4SSatish Balay locTotalTime -= petsc_BaseTime; 14879566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 14889566063dSJacob Faibussowitsch PetscCallMPI(MPI_Allreduce(&stageLog->numStages, &numStages, 1, MPI_INT, MPI_MAX, comm)); 14899566063dSJacob Faibussowitsch PetscCall(PetscMallocGetMaximumUsage(&maxMem)); 14909566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIIPushSynchronized(viewer)); 1491d0609cedSBarry Smith 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)); 14929566063dSJacob Faibussowitsch PetscCall(PetscViewerFlush(viewer)); 149338144912Sdeepblu2718 for (stage = 0; stage < numStages; stage++) { 14949e29573dSMatthew G. Knepley PetscEventPerfInfo *stageInfo = &stageLog->stageInfo[stage].perfInfo; 14959e29573dSMatthew G. Knepley 14969371c9d4SSatish Balay PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "%s,summary,%d,1,%g,%g,%g,%g,%g\n", stageLog->stageInfo[stage].name, rank, stageInfo->time, stageInfo->numMessages, stageInfo->messageLength, stageInfo->numReductions, stageInfo->flops)); 14979566063dSJacob Faibussowitsch PetscCallMPI(MPI_Allreduce(&stageLog->stageInfo[stage].eventLog->numEvents, &numEvents, 1, MPI_INT, MPI_MAX, comm)); 149838144912Sdeepblu2718 for (event = 0; event < numEvents; event++) { 149938144912Sdeepblu2718 eventInfo = &stageLog->stageInfo[stage].eventLog->eventInfo[event]; 15009371c9d4SSatish Balay PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "%s,%s,%d,%d,%g,%g,%g,%g,%g", stageLog->stageInfo[stage].name, stageLog->eventLog->eventInfo[event].name, rank, eventInfo->count, eventInfo->time, eventInfo->numMessages, eventInfo->messageLength, 15019371c9d4SSatish Balay eventInfo->numReductions, eventInfo->flops)); 150238144912Sdeepblu2718 if (eventInfo->dof[0] >= 0.) { 150338144912Sdeepblu2718 PetscInt d, e; 150438144912Sdeepblu2718 150548a46eb9SPierre Jolivet for (d = 0; d < 8; ++d) PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, ",%g", eventInfo->dof[d])); 150648a46eb9SPierre Jolivet for (e = 0; e < 8; ++e) PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, ",%g", eventInfo->errors[e])); 150738144912Sdeepblu2718 } 15089566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "\n")); 150938144912Sdeepblu2718 } 151038144912Sdeepblu2718 } 15119566063dSJacob Faibussowitsch PetscCall(PetscViewerFlush(viewer)); 15129566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIIPopSynchronized(viewer)); 15133ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 151438144912Sdeepblu2718 } 151538144912Sdeepblu2718 1516d71ae5a4SJacob Faibussowitsch static PetscErrorCode PetscLogViewWarnSync(MPI_Comm comm, FILE *fd) 1517d71ae5a4SJacob Faibussowitsch { 1518217044c2SLisandro Dalcin PetscFunctionBegin; 15193ba16761SJacob Faibussowitsch if (!PetscLogSyncOn) PetscFunctionReturn(PETSC_SUCCESS); 15209566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "\n\n")); 15219566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " ##########################################################\n")); 15229566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # #\n")); 15239566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # WARNING!!! #\n")); 15249566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # #\n")); 15259566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # This program was run with logging synchronization. #\n")); 15269566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # This option provides more meaningful imbalance #\n")); 15279566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # figures at the expense of slowing things down and #\n")); 15289566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # providing a distorted view of the overall runtime. #\n")); 15299566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # #\n")); 15309566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " ##########################################################\n\n\n")); 15313ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 1532217044c2SLisandro Dalcin } 1533217044c2SLisandro Dalcin 1534d71ae5a4SJacob Faibussowitsch static PetscErrorCode PetscLogViewWarnDebugging(MPI_Comm comm, FILE *fd) 1535d71ae5a4SJacob Faibussowitsch { 1536f4091ad2SBarry Smith PetscFunctionBegin; 153776bd3646SJed Brown if (PetscDefined(USE_DEBUG)) { 15389566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "\n\n")); 15399566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " ##########################################################\n")); 15409566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # #\n")); 15419566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # WARNING!!! #\n")); 15429566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # #\n")); 15439566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # This code was compiled with a debugging option. #\n")); 15449566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # To get timing results run ./configure #\n")); 15459566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # using --with-debugging=no, the performance will #\n")); 15469566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # be generally two or three times faster. #\n")); 15479566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # #\n")); 15489566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " ##########################################################\n\n\n")); 154976bd3646SJed Brown } 15503ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 1551f4091ad2SBarry Smith } 1552f4091ad2SBarry Smith 1553d71ae5a4SJacob Faibussowitsch static PetscErrorCode PetscLogViewWarnNoGpuAwareMpi(MPI_Comm comm, FILE *fd) 1554d71ae5a4SJacob Faibussowitsch { 1555009ab46cSJunchao Zhang #if defined(PETSC_HAVE_DEVICE) 1556009ab46cSJunchao Zhang PetscMPIInt size; 15578fe3844cSJunchao Zhang PetscBool deviceInitialized = PETSC_FALSE; 1558c2a741eeSJunchao Zhang 1559c2a741eeSJunchao Zhang PetscFunctionBegin; 15609566063dSJacob Faibussowitsch PetscCallMPI(MPI_Comm_size(comm, &size)); 15618fe3844cSJunchao Zhang for (int i = PETSC_DEVICE_HOST + 1; i < PETSC_DEVICE_MAX; ++i) { 15628fe3844cSJunchao Zhang const PetscDeviceType dtype = PetscDeviceTypeCast(i); 15638fe3844cSJunchao Zhang if (PetscDeviceInitialized(dtype)) { /* a non-host device was initialized */ 15648fe3844cSJunchao Zhang deviceInitialized = PETSC_TRUE; 15658fe3844cSJunchao Zhang break; 15668fe3844cSJunchao Zhang } 15678fe3844cSJunchao Zhang } 15688fe3844cSJunchao Zhang /* the last condition says petsc is configured with device but it is a pure CPU run, so don't print misleading warnings */ 15693ba16761SJacob Faibussowitsch if (use_gpu_aware_mpi || size == 1 || !deviceInitialized) PetscFunctionReturn(PETSC_SUCCESS); 15709566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "\n\n")); 15719566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " ##########################################################\n")); 15729566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # #\n")); 15739566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # WARNING!!! #\n")); 15749566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # #\n")); 15759566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # This code was compiled with GPU support and you've #\n")); 15769566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # created PETSc/GPU objects, but you intentionally #\n")); 15779566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # used -use_gpu_aware_mpi 0, requiring PETSc to copy #\n")); 15789566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # additional data between the GPU and CPU. To obtain #\n")); 15799566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # meaningful timing results on multi-rank runs, use #\n")); 15809566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # GPU-aware MPI instead. #\n")); 15819566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " # #\n")); 15829566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " ##########################################################\n\n\n")); 15833ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 1584c2a741eeSJunchao Zhang #else 1585*10450e9eSJacob Faibussowitsch (void)comm; 1586*10450e9eSJacob Faibussowitsch (void)fd; 15873ba16761SJacob Faibussowitsch return PETSC_SUCCESS; 1588c2a741eeSJunchao Zhang #endif 1589c2a741eeSJunchao Zhang } 1590c2a741eeSJunchao Zhang 1591d71ae5a4SJacob Faibussowitsch static PetscErrorCode PetscLogViewWarnGpuTime(MPI_Comm comm, FILE *fd) 1592d71ae5a4SJacob Faibussowitsch { 1593156b51fbSBarry Smith #if defined(PETSC_HAVE_DEVICE) 1594156b51fbSBarry Smith PetscFunctionBegin; 15953ba16761SJacob Faibussowitsch if (!PetscLogGpuTimeFlag || petsc_gflops == 0) PetscFunctionReturn(PETSC_SUCCESS); 1596156b51fbSBarry Smith PetscCall(PetscFPrintf(comm, fd, "\n\n")); 1597156b51fbSBarry Smith PetscCall(PetscFPrintf(comm, fd, " ##########################################################\n")); 1598156b51fbSBarry Smith PetscCall(PetscFPrintf(comm, fd, " # #\n")); 1599156b51fbSBarry Smith PetscCall(PetscFPrintf(comm, fd, " # WARNING!!! #\n")); 1600156b51fbSBarry Smith PetscCall(PetscFPrintf(comm, fd, " # #\n")); 1601156b51fbSBarry Smith PetscCall(PetscFPrintf(comm, fd, " # This code was run with -log_view_gpu_time #\n")); 1602156b51fbSBarry Smith PetscCall(PetscFPrintf(comm, fd, " # This provides accurate timing within the GPU kernels #\n")); 1603156b51fbSBarry Smith PetscCall(PetscFPrintf(comm, fd, " # but can slow down the entire computation by a #\n")); 1604156b51fbSBarry Smith PetscCall(PetscFPrintf(comm, fd, " # measurable amount. For fastest runs we recommend #\n")); 1605156b51fbSBarry Smith PetscCall(PetscFPrintf(comm, fd, " # not using this option. #\n")); 1606156b51fbSBarry Smith PetscCall(PetscFPrintf(comm, fd, " # #\n")); 1607156b51fbSBarry Smith PetscCall(PetscFPrintf(comm, fd, " ##########################################################\n\n\n")); 16083ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 1609156b51fbSBarry Smith #else 1610*10450e9eSJacob Faibussowitsch (void)comm; 1611*10450e9eSJacob Faibussowitsch (void)fd; 16123ba16761SJacob Faibussowitsch return PETSC_SUCCESS; 1613156b51fbSBarry Smith #endif 1614156b51fbSBarry Smith } 1615156b51fbSBarry Smith 1616*10450e9eSJacob Faibussowitsch static PetscErrorCode PetscLogView_Default(PetscViewer viewer) 1617d71ae5a4SJacob Faibussowitsch { 16185c6c1daeSBarry Smith FILE *fd; 16195c6c1daeSBarry Smith PetscLogDouble zero = 0.0; 16205c6c1daeSBarry Smith PetscStageLog stageLog; 16210298fd71SBarry Smith PetscStageInfo *stageInfo = NULL; 16220298fd71SBarry Smith PetscEventPerfInfo *eventInfo = NULL; 16235c6c1daeSBarry Smith PetscClassPerfInfo *classInfo; 16245c6c1daeSBarry Smith char arch[128], hostname[128], username[128], pname[PETSC_MAX_PATH_LEN], date[128]; 16255c6c1daeSBarry Smith const char *name; 16265c6c1daeSBarry Smith PetscLogDouble locTotalTime, TotalTime, TotalFlops; 16275c6c1daeSBarry Smith PetscLogDouble numMessages, messageLength, avgMessLen, numReductions; 16285c6c1daeSBarry Smith PetscLogDouble stageTime, flops, flopr, mem, mess, messLen, red; 16295c6c1daeSBarry Smith PetscLogDouble fracTime, fracFlops, fracMessages, fracLength, fracReductions, fracMess, fracMessLen, fracRed; 16305c6c1daeSBarry Smith PetscLogDouble fracStageTime, fracStageFlops, fracStageMess, fracStageMessLen, fracStageRed; 16315c6c1daeSBarry Smith PetscLogDouble min, max, tot, ratio, avg, x, y; 1632e3ed9ee7SBarry Smith PetscLogDouble minf, maxf, totf, ratf, mint, maxt, tott, ratt, ratC, totm, totml, totr, mal, malmax, emalmax; 1633d9c66bfbSJunchao Zhang #if defined(PETSC_HAVE_DEVICE) 1634156b51fbSBarry Smith PetscLogEvent KSP_Solve, SNES_Solve, TS_Step, TAO_Solve; /* These need to be fixed to be some events registered with certain objects */ 1635958c4211Shannah_mairs PetscLogDouble cct, gct, csz, gsz, gmaxt, gflops, gflopr, fracgflops; 1636bec0b493Shannah_mairs #endif 1637217044c2SLisandro Dalcin PetscMPIInt minC, maxC; 16385c6c1daeSBarry Smith PetscMPIInt size, rank; 16395c6c1daeSBarry Smith PetscBool *localStageUsed, *stageUsed; 16405c6c1daeSBarry Smith PetscBool *localStageVisible, *stageVisible; 16415c6c1daeSBarry Smith int numStages, localNumEvents, numEvents; 164237b78d16SBarry Smith int stage, oclass; 16435c6c1daeSBarry Smith PetscLogEvent event; 16445c6c1daeSBarry Smith char version[256]; 16455c6c1daeSBarry Smith MPI_Comm comm; 1646156b51fbSBarry Smith #if defined(PETSC_HAVE_DEVICE) 1647156b51fbSBarry Smith PetscLogEvent eventid; 1648156b51fbSBarry Smith PetscInt64 nas = 0x7FF0000000000002; 1649156b51fbSBarry Smith #endif 16505c6c1daeSBarry Smith 16515c6c1daeSBarry Smith PetscFunctionBegin; 1652156b51fbSBarry Smith PetscCall(PetscFPTrapPush(PETSC_FP_TRAP_OFF)); 16539566063dSJacob Faibussowitsch PetscCall(PetscObjectGetComm((PetscObject)viewer, &comm)); 16549566063dSJacob Faibussowitsch PetscCall(PetscViewerASCIIGetPointer(viewer, &fd)); 16559566063dSJacob Faibussowitsch PetscCallMPI(MPI_Comm_size(comm, &size)); 16569566063dSJacob Faibussowitsch PetscCallMPI(MPI_Comm_rank(comm, &rank)); 16575c6c1daeSBarry Smith /* Get the total elapsed time */ 16583ba16761SJacob Faibussowitsch PetscCall(PetscTime(&locTotalTime)); 16599371c9d4SSatish Balay locTotalTime -= petsc_BaseTime; 16605c6c1daeSBarry Smith 1661faad7b01SPierre Jolivet PetscCall(PetscFPrintf(comm, fd, "****************************************************************************************************************************************************************\n")); 16629566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "*** WIDEN YOUR WINDOW TO 160 CHARACTERS. Use 'enscript -r -fCourier9' to print this document ***\n")); 16639566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "****************************************************************************************************************************************************************\n")); 1664faad7b01SPierre Jolivet PetscCall(PetscFPrintf(comm, fd, "\n------------------------------------------------------------------ PETSc Performance Summary: ------------------------------------------------------------------\n\n")); 16659566063dSJacob Faibussowitsch PetscCall(PetscLogViewWarnSync(comm, fd)); 16669566063dSJacob Faibussowitsch PetscCall(PetscLogViewWarnDebugging(comm, fd)); 16679566063dSJacob Faibussowitsch PetscCall(PetscLogViewWarnNoGpuAwareMpi(comm, fd)); 1668156b51fbSBarry Smith PetscCall(PetscLogViewWarnGpuTime(comm, fd)); 16699566063dSJacob Faibussowitsch PetscCall(PetscGetArchType(arch, sizeof(arch))); 16709566063dSJacob Faibussowitsch PetscCall(PetscGetHostName(hostname, sizeof(hostname))); 16719566063dSJacob Faibussowitsch PetscCall(PetscGetUserName(username, sizeof(username))); 16729566063dSJacob Faibussowitsch PetscCall(PetscGetProgramName(pname, sizeof(pname))); 16739566063dSJacob Faibussowitsch PetscCall(PetscGetDate(date, sizeof(date))); 16749566063dSJacob Faibussowitsch PetscCall(PetscGetVersion(version, sizeof(version))); 16755c6c1daeSBarry Smith if (size == 1) { 16769566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "%s on a %s named %s with %d processor, by %s %s\n", pname, arch, hostname, size, username, date)); 16775c6c1daeSBarry Smith } else { 16789566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "%s on a %s named %s with %d processors, by %s %s\n", pname, arch, hostname, size, username, date)); 16795c6c1daeSBarry Smith } 1680f90b075cSBarry Smith #if defined(PETSC_HAVE_OPENMP) 16819566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Using %" PetscInt_FMT " OpenMP threads\n", PetscNumOMPThreads)); 1682f90b075cSBarry Smith #endif 16839566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Using %s\n", version)); 16845c6c1daeSBarry Smith 16855c6c1daeSBarry Smith /* Must preserve reduction count before we go on */ 16865c6c1daeSBarry Smith red = petsc_allreduce_ct + petsc_gather_ct + petsc_scatter_ct; 16875c6c1daeSBarry Smith 16885c6c1daeSBarry Smith /* Calculate summary information */ 16899566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "\n Max Max/Min Avg Total\n")); 16905c6c1daeSBarry Smith /* Time */ 1691712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&locTotalTime, &min, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1692712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&locTotalTime, &max, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1693712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&locTotalTime, &tot, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1694217044c2SLisandro Dalcin avg = tot / ((PetscLogDouble)size); 16959371c9d4SSatish Balay if (min != 0.0) ratio = max / min; 16969371c9d4SSatish Balay else ratio = 0.0; 16979566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Time (sec): %5.3e %7.3f %5.3e\n", max, ratio, avg)); 16985c6c1daeSBarry Smith TotalTime = tot; 16995c6c1daeSBarry Smith /* Objects */ 17005c6c1daeSBarry Smith avg = (PetscLogDouble)petsc_numObjects; 1701712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&avg, &min, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1702712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&avg, &max, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1703712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&avg, &tot, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1704217044c2SLisandro Dalcin avg = tot / ((PetscLogDouble)size); 17059371c9d4SSatish Balay if (min != 0.0) ratio = max / min; 17069371c9d4SSatish Balay else ratio = 0.0; 17079566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Objects: %5.3e %7.3f %5.3e\n", max, ratio, avg)); 17085c6c1daeSBarry Smith /* Flops */ 1709712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&petsc_TotalFlops, &min, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1710712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&petsc_TotalFlops, &max, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1711712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&petsc_TotalFlops, &tot, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1712217044c2SLisandro Dalcin avg = tot / ((PetscLogDouble)size); 17139371c9d4SSatish Balay if (min != 0.0) ratio = max / min; 17149371c9d4SSatish Balay else ratio = 0.0; 17159566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Flops: %5.3e %7.3f %5.3e %5.3e\n", max, ratio, avg, tot)); 17165c6c1daeSBarry Smith TotalFlops = tot; 17175c6c1daeSBarry Smith /* Flops/sec -- Must talk to Barry here */ 17189371c9d4SSatish Balay if (locTotalTime != 0.0) flops = petsc_TotalFlops / locTotalTime; 17199371c9d4SSatish Balay else flops = 0.0; 1720712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&flops, &min, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1721712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&flops, &max, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1722712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&flops, &tot, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1723217044c2SLisandro Dalcin avg = tot / ((PetscLogDouble)size); 17249371c9d4SSatish Balay if (min != 0.0) ratio = max / min; 17259371c9d4SSatish Balay else ratio = 0.0; 17269566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Flops/sec: %5.3e %7.3f %5.3e %5.3e\n", max, ratio, avg, tot)); 17275c6c1daeSBarry Smith /* Memory */ 17289566063dSJacob Faibussowitsch PetscCall(PetscMallocGetMaximumUsage(&mem)); 17295c6c1daeSBarry Smith if (mem > 0.0) { 1730712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&mem, &min, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1731712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&mem, &max, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1732712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&mem, &tot, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1733217044c2SLisandro Dalcin avg = tot / ((PetscLogDouble)size); 17349371c9d4SSatish Balay if (min != 0.0) ratio = max / min; 17359371c9d4SSatish Balay else ratio = 0.0; 17369566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Memory (bytes): %5.3e %7.3f %5.3e %5.3e\n", max, ratio, avg, tot)); 17375c6c1daeSBarry Smith } 17385c6c1daeSBarry Smith /* Messages */ 17395c6c1daeSBarry Smith mess = 0.5 * (petsc_irecv_ct + petsc_isend_ct + petsc_recv_ct + petsc_send_ct); 1740712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&mess, &min, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1741712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&mess, &max, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1742712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&mess, &tot, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1743217044c2SLisandro Dalcin avg = tot / ((PetscLogDouble)size); 17449371c9d4SSatish Balay if (min != 0.0) ratio = max / min; 17459371c9d4SSatish Balay else ratio = 0.0; 17469566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "MPI Msg Count: %5.3e %7.3f %5.3e %5.3e\n", max, ratio, avg, tot)); 17475c6c1daeSBarry Smith numMessages = tot; 17485c6c1daeSBarry Smith /* Message Lengths */ 17495c6c1daeSBarry Smith mess = 0.5 * (petsc_irecv_len + petsc_isend_len + petsc_recv_len + petsc_send_len); 1750712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&mess, &min, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1751712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&mess, &max, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1752712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&mess, &tot, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 17539371c9d4SSatish Balay if (numMessages != 0) avg = tot / numMessages; 17549371c9d4SSatish Balay else avg = 0.0; 17559371c9d4SSatish Balay if (min != 0.0) ratio = max / min; 17569371c9d4SSatish Balay else ratio = 0.0; 17579566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "MPI Msg Len (bytes): %5.3e %7.3f %5.3e %5.3e\n", max, ratio, avg, tot)); 17585c6c1daeSBarry Smith messageLength = tot; 17595c6c1daeSBarry Smith /* Reductions */ 1760712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&red, &min, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1761712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&red, &max, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1762712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&red, &tot, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 17639371c9d4SSatish Balay if (min != 0.0) ratio = max / min; 17649371c9d4SSatish Balay else ratio = 0.0; 17659566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "MPI Reductions: %5.3e %7.3f\n", max, ratio)); 17665c6c1daeSBarry Smith numReductions = red; /* wrong because uses count from process zero */ 17679566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "\nFlop counting convention: 1 flop = 1 real number operation of type (multiply/divide/add/subtract)\n")); 17689566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " e.g., VecAXPY() for real vectors of length N --> 2N flops\n")); 17699566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " and VecAXPY() for complex vectors of length N --> 8N flops\n")); 17705c6c1daeSBarry Smith 17715c6c1daeSBarry Smith /* Get total number of stages -- 17725c6c1daeSBarry Smith Currently, a single processor can register more stages than another, but stages must all be registered in order. 17735c6c1daeSBarry Smith We can removed this requirement if necessary by having a global stage numbering and indirection on the stage ID. 17745c6c1daeSBarry Smith This seems best accomplished by assoicating a communicator with each stage. 17755c6c1daeSBarry Smith */ 17769566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 17779566063dSJacob Faibussowitsch PetscCallMPI(MPI_Allreduce(&stageLog->numStages, &numStages, 1, MPI_INT, MPI_MAX, comm)); 17789566063dSJacob Faibussowitsch PetscCall(PetscMalloc1(numStages, &localStageUsed)); 17799566063dSJacob Faibussowitsch PetscCall(PetscMalloc1(numStages, &stageUsed)); 17809566063dSJacob Faibussowitsch PetscCall(PetscMalloc1(numStages, &localStageVisible)); 17819566063dSJacob Faibussowitsch PetscCall(PetscMalloc1(numStages, &stageVisible)); 17825c6c1daeSBarry Smith if (numStages > 0) { 17835c6c1daeSBarry Smith stageInfo = stageLog->stageInfo; 17845c6c1daeSBarry Smith for (stage = 0; stage < numStages; stage++) { 17855c6c1daeSBarry Smith if (stage < stageLog->numStages) { 17865c6c1daeSBarry Smith localStageUsed[stage] = stageInfo[stage].used; 17875c6c1daeSBarry Smith localStageVisible[stage] = stageInfo[stage].perfInfo.visible; 17885c6c1daeSBarry Smith } else { 17895c6c1daeSBarry Smith localStageUsed[stage] = PETSC_FALSE; 17905c6c1daeSBarry Smith localStageVisible[stage] = PETSC_TRUE; 17915c6c1daeSBarry Smith } 17925c6c1daeSBarry Smith } 1793712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(localStageUsed, stageUsed, numStages, MPIU_BOOL, MPI_LOR, comm)); 1794712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(localStageVisible, stageVisible, numStages, MPIU_BOOL, MPI_LAND, comm)); 17955c6c1daeSBarry Smith for (stage = 0; stage < numStages; stage++) { 17965c6c1daeSBarry Smith if (stageUsed[stage]) { 17979566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "\nSummary of Stages: ----- Time ------ ----- Flop ------ --- Messages --- -- Message Lengths -- -- Reductions --\n")); 17989566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " Avg %%Total Avg %%Total Count %%Total Avg %%Total Count %%Total\n")); 17995c6c1daeSBarry Smith break; 18005c6c1daeSBarry Smith } 18015c6c1daeSBarry Smith } 18025c6c1daeSBarry Smith for (stage = 0; stage < numStages; stage++) { 18035c6c1daeSBarry Smith if (!stageUsed[stage]) continue; 1804820f2d46SBarry Smith /* CANNOT use MPI_Allreduce() since it might fail the line number check */ 18055c6c1daeSBarry Smith if (localStageUsed[stage]) { 1806712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&stageInfo[stage].perfInfo.time, &stageTime, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1807712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&stageInfo[stage].perfInfo.flops, &flops, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1808712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&stageInfo[stage].perfInfo.numMessages, &mess, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1809712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&stageInfo[stage].perfInfo.messageLength, &messLen, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1810712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&stageInfo[stage].perfInfo.numReductions, &red, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 18115c6c1daeSBarry Smith name = stageInfo[stage].name; 18125c6c1daeSBarry Smith } else { 1813712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &stageTime, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1814712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &flops, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1815712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &mess, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1816712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &messLen, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1817712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &red, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 18185c6c1daeSBarry Smith name = ""; 18195c6c1daeSBarry Smith } 18209371c9d4SSatish Balay mess *= 0.5; 18219371c9d4SSatish Balay messLen *= 0.5; 18229371c9d4SSatish Balay red /= size; 18239371c9d4SSatish Balay if (TotalTime != 0.0) fracTime = stageTime / TotalTime; 18249371c9d4SSatish Balay else fracTime = 0.0; 18259371c9d4SSatish Balay if (TotalFlops != 0.0) fracFlops = flops / TotalFlops; 18269371c9d4SSatish Balay else fracFlops = 0.0; 18275c6c1daeSBarry Smith /* Talk to Barry if (stageTime != 0.0) flops = (size*flops)/stageTime; else flops = 0.0; */ 18289371c9d4SSatish Balay if (numMessages != 0.0) fracMessages = mess / numMessages; 18299371c9d4SSatish Balay else fracMessages = 0.0; 18309371c9d4SSatish Balay if (mess != 0.0) avgMessLen = messLen / mess; 18319371c9d4SSatish Balay else avgMessLen = 0.0; 18329371c9d4SSatish Balay if (messageLength != 0.0) fracLength = messLen / messageLength; 18339371c9d4SSatish Balay else fracLength = 0.0; 18349371c9d4SSatish Balay if (numReductions != 0.0) fracReductions = red / numReductions; 18359371c9d4SSatish Balay else fracReductions = 0.0; 18369371c9d4SSatish Balay PetscCall(PetscFPrintf(comm, fd, "%2d: %15s: %6.4e %5.1f%% %6.4e %5.1f%% %5.3e %5.1f%% %5.3e %5.1f%% %5.3e %5.1f%%\n", stage, name, stageTime / size, 100.0 * fracTime, flops, 100.0 * fracFlops, mess, 100.0 * fracMessages, avgMessLen, 100.0 * fracLength, red, 100.0 * fracReductions)); 18375c6c1daeSBarry Smith } 18385c6c1daeSBarry Smith } 18395c6c1daeSBarry Smith 18409566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "\n------------------------------------------------------------------------------------------------------------------------\n")); 18419566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "See the 'Profiling' chapter of the users' manual for details on interpreting output.\n")); 18429566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Phase summary info:\n")); 18439566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " Count: number of times phase was executed\n")); 18449566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " Time and Flop: Max - maximum over all processors\n")); 18459566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " Ratio - ratio of maximum to minimum over all processors\n")); 18469566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " Mess: number of messages sent\n")); 18479566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " AvgLen: average message length (bytes)\n")); 18489566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " Reduct: number of global reductions\n")); 18499566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " Global: entire computation\n")); 18509566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " Stage: stages of a computation. Set stages with PetscLogStagePush() and PetscLogStagePop().\n")); 18519566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " %%T - percent time in this phase %%F - percent flop in this phase\n")); 18529566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " %%M - percent messages in this phase %%L - percent message lengths in this phase\n")); 18539566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " %%R - percent reductions in this phase\n")); 18549566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " Total Mflop/s: 10e-6 * (sum of flop over all processors)/(max time over all processors)\n")); 1855e3ed9ee7SBarry Smith if (PetscLogMemory) { 185692d1ba04SBarry Smith PetscCall(PetscFPrintf(comm, fd, " Memory usage is summed over all MPI processes, it is given in mega-bytes\n")); 185754aabf2dSBarry Smith PetscCall(PetscFPrintf(comm, fd, " Malloc Mbytes: Memory allocated and kept during event (sum over all calls to event). May be negative\n")); 185854aabf2dSBarry Smith PetscCall(PetscFPrintf(comm, fd, " EMalloc Mbytes: extra memory allocated during event and then freed (maximum over all calls to events). Never negative\n")); 185954aabf2dSBarry Smith PetscCall(PetscFPrintf(comm, fd, " MMalloc Mbytes: Increase in high water mark of allocated memory (sum over all calls to event). Never negative\n")); 18609566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " RMI Mbytes: Increase in resident memory (sum over all calls to event)\n")); 1861e3ed9ee7SBarry Smith } 1862d9c66bfbSJunchao Zhang #if defined(PETSC_HAVE_DEVICE) 18639566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " GPU Mflop/s: 10e-6 * (sum of flop on GPU over all processors)/(max GPU time over all processors)\n")); 18649566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " CpuToGpu Count: total number of CPU to GPU copies per processor\n")); 18659566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " CpuToGpu Size (Mbytes): 10e-6 * (total size of CPU to GPU copies per processor)\n")); 18669566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " GpuToCpu Count: total number of GPU to CPU copies per processor\n")); 18679566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " GpuToCpu Size (Mbytes): 10e-6 * (total size of GPU to CPU copies per processor)\n")); 18689566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " GPU %%F: percent flops on GPU in this event\n")); 1869bec0b493Shannah_mairs #endif 18709566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "------------------------------------------------------------------------------------------------------------------------\n")); 18715c6c1daeSBarry Smith 18729566063dSJacob Faibussowitsch PetscCall(PetscLogViewWarnDebugging(comm, fd)); 18735c6c1daeSBarry Smith 18745c6c1daeSBarry Smith /* Report events */ 18759566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Event Count Time (sec) Flop --- Global --- --- Stage ---- Total")); 187648a46eb9SPierre Jolivet if (PetscLogMemory) PetscCall(PetscFPrintf(comm, fd, " Malloc EMalloc MMalloc RMI")); 1877d9c66bfbSJunchao Zhang #if defined(PETSC_HAVE_DEVICE) 18789566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " GPU - CpuToGpu - - GpuToCpu - GPU")); 1879bec0b493Shannah_mairs #endif 18809566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "\n")); 18819566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " Max Ratio Max Ratio Max Ratio Mess AvgLen Reduct %%T %%F %%M %%L %%R %%T %%F %%M %%L %%R Mflop/s")); 188248a46eb9SPierre Jolivet if (PetscLogMemory) PetscCall(PetscFPrintf(comm, fd, " Mbytes Mbytes Mbytes Mbytes")); 1883d9c66bfbSJunchao Zhang #if defined(PETSC_HAVE_DEVICE) 18849566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, " Mflop/s Count Size Count Size %%F")); 1885bec0b493Shannah_mairs #endif 18869566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "\n")); 18879566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "------------------------------------------------------------------------------------------------------------------------")); 188848a46eb9SPierre Jolivet if (PetscLogMemory) PetscCall(PetscFPrintf(comm, fd, "-----------------------------")); 1889d9c66bfbSJunchao Zhang #if defined(PETSC_HAVE_DEVICE) 18909566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "---------------------------------------")); 18914863603aSSatish Balay #endif 18929566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "\n")); 18935c6c1daeSBarry Smith 1894156b51fbSBarry Smith #if defined(PETSC_HAVE_DEVICE) 1895156b51fbSBarry Smith /* this indirect way of accessing these values is needed when PETSc is build with multiple libraries since the symbols are not in libpetscsys */ 1896156b51fbSBarry Smith PetscCall(PetscEventRegLogGetEvent(stageLog->eventLog, "TAOSolve", &TAO_Solve)); 1897156b51fbSBarry Smith PetscCall(PetscEventRegLogGetEvent(stageLog->eventLog, "TSStep", &TS_Step)); 1898156b51fbSBarry Smith PetscCall(PetscEventRegLogGetEvent(stageLog->eventLog, "SNESSolve", &SNES_Solve)); 1899156b51fbSBarry Smith PetscCall(PetscEventRegLogGetEvent(stageLog->eventLog, "KSPSolve", &KSP_Solve)); 1900156b51fbSBarry Smith #endif 1901156b51fbSBarry Smith 19025c6c1daeSBarry Smith /* Problem: The stage name will not show up unless the stage executed on proc 1 */ 19035c6c1daeSBarry Smith for (stage = 0; stage < numStages; stage++) { 19045c6c1daeSBarry Smith if (!stageVisible[stage]) continue; 1905820f2d46SBarry Smith /* CANNOT use MPI_Allreduce() since it might fail the line number check */ 19065c6c1daeSBarry Smith if (localStageUsed[stage]) { 19079566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "\n--- Event Stage %d: %s\n\n", stage, stageInfo[stage].name)); 1908712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&stageInfo[stage].perfInfo.time, &stageTime, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1909712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&stageInfo[stage].perfInfo.flops, &flops, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1910712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&stageInfo[stage].perfInfo.numMessages, &mess, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1911712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&stageInfo[stage].perfInfo.messageLength, &messLen, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1912712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&stageInfo[stage].perfInfo.numReductions, &red, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 19135c6c1daeSBarry Smith } else { 19149566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "\n--- Event Stage %d: Unknown\n\n", stage)); 1915712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &stageTime, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1916712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &flops, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1917712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &mess, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1918712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &messLen, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1919712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &red, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 19205c6c1daeSBarry Smith } 19219371c9d4SSatish Balay mess *= 0.5; 19229371c9d4SSatish Balay messLen *= 0.5; 19239371c9d4SSatish Balay red /= size; 19245c6c1daeSBarry Smith 19255c6c1daeSBarry Smith /* Get total number of events in this stage -- 19265c6c1daeSBarry Smith Currently, a single processor can register more events than another, but events must all be registered in order, 19275c6c1daeSBarry Smith just like stages. We can removed this requirement if necessary by having a global event numbering and indirection 1928217044c2SLisandro Dalcin on the event ID. This seems best accomplished by associating a communicator with each stage. 19295c6c1daeSBarry Smith 19305c6c1daeSBarry Smith Problem: If the event did not happen on proc 1, its name will not be available. 19315c6c1daeSBarry Smith Problem: Event visibility is not implemented 19325c6c1daeSBarry Smith */ 19335c6c1daeSBarry Smith if (localStageUsed[stage]) { 19345c6c1daeSBarry Smith eventInfo = stageLog->stageInfo[stage].eventLog->eventInfo; 19355c6c1daeSBarry Smith localNumEvents = stageLog->stageInfo[stage].eventLog->numEvents; 1936a297a907SKarl Rupp } else localNumEvents = 0; 19379566063dSJacob Faibussowitsch PetscCallMPI(MPI_Allreduce(&localNumEvents, &numEvents, 1, MPI_INT, MPI_MAX, comm)); 19385c6c1daeSBarry Smith for (event = 0; event < numEvents; event++) { 1939820f2d46SBarry Smith /* CANNOT use MPI_Allreduce() since it might fail the line number check */ 19405c6c1daeSBarry Smith if (localStageUsed[stage] && (event < stageLog->stageInfo[stage].eventLog->numEvents) && (eventInfo[event].depth == 0)) { 19419371c9d4SSatish Balay if ((eventInfo[event].count > 0) && (eventInfo[event].time > 0.0)) flopr = eventInfo[event].flops; 19429371c9d4SSatish Balay else flopr = 0.0; 1943712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&flopr, &minf, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1944712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&flopr, &maxf, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1945712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&eventInfo[event].flops, &totf, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1946712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&eventInfo[event].time, &mint, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1947712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&eventInfo[event].time, &maxt, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1948712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&eventInfo[event].time, &tott, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1949712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&eventInfo[event].numMessages, &totm, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1950712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&eventInfo[event].messageLength, &totml, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1951712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&eventInfo[event].numReductions, &totr, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 19529566063dSJacob Faibussowitsch PetscCallMPI(MPI_Allreduce(&eventInfo[event].count, &minC, 1, MPI_INT, MPI_MIN, comm)); 19539566063dSJacob Faibussowitsch PetscCallMPI(MPI_Allreduce(&eventInfo[event].count, &maxC, 1, MPI_INT, MPI_MAX, comm)); 1954e3ed9ee7SBarry Smith if (PetscLogMemory) { 1955712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&eventInfo[event].memIncrease, &mem, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1956712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&eventInfo[event].mallocSpace, &mal, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1957712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&eventInfo[event].mallocIncrease, &malmax, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1958712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&eventInfo[event].mallocIncreaseEvent, &emalmax, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1959e3ed9ee7SBarry Smith } 1960d9c66bfbSJunchao Zhang #if defined(PETSC_HAVE_DEVICE) 1961712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&eventInfo[event].CpuToGpuCount, &cct, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1962712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&eventInfo[event].GpuToCpuCount, &gct, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1963712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&eventInfo[event].CpuToGpuSize, &csz, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1964712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&eventInfo[event].GpuToCpuSize, &gsz, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1965712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&eventInfo[event].GpuFlops, &gflops, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1966712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&eventInfo[event].GpuTime, &gmaxt, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1967bec0b493Shannah_mairs #endif 19685c6c1daeSBarry Smith name = stageLog->eventLog->eventInfo[event].name; 19695c6c1daeSBarry Smith } else { 19703ba16761SJacob Faibussowitsch int ierr = 0; 19713ba16761SJacob Faibussowitsch 19725c6c1daeSBarry Smith flopr = 0.0; 1973712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&flopr, &minf, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1974712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&flopr, &maxf, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1975712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &totf, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1976712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &mint, 1, MPIU_PETSCLOGDOUBLE, MPI_MIN, comm)); 1977712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &maxt, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1978712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &tott, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1979712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &totm, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1980712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &totml, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1981712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &totr, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 19829566063dSJacob Faibussowitsch PetscCallMPI(MPI_Allreduce(&ierr, &minC, 1, MPI_INT, MPI_MIN, comm)); 19839566063dSJacob Faibussowitsch PetscCallMPI(MPI_Allreduce(&ierr, &maxC, 1, MPI_INT, MPI_MAX, comm)); 1984e3ed9ee7SBarry Smith if (PetscLogMemory) { 1985712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &mem, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1986712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &mal, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1987712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &malmax, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1988712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &emalmax, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1989e3ed9ee7SBarry Smith } 1990d9c66bfbSJunchao Zhang #if defined(PETSC_HAVE_DEVICE) 1991712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &cct, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1992712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &gct, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1993712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &csz, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1994712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &gsz, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1995712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &gflops, 1, MPIU_PETSCLOGDOUBLE, MPI_SUM, comm)); 1996712fec58SPierre Jolivet PetscCall(MPIU_Allreduce(&zero, &gmaxt, 1, MPIU_PETSCLOGDOUBLE, MPI_MAX, comm)); 1997bec0b493Shannah_mairs #endif 19985c6c1daeSBarry Smith name = ""; 19995c6c1daeSBarry Smith } 20005c6c1daeSBarry Smith if (mint < 0.0) { 2001d0609cedSBarry Smith PetscCall(PetscFPrintf(comm, fd, "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, name)); 20025c6c1daeSBarry Smith mint = 0; 20035c6c1daeSBarry Smith } 200408401ef6SPierre Jolivet PetscCheck(minf >= 0.0, PETSC_COMM_SELF, PETSC_ERR_PLIB, "Minimum flop %g over all processors for %s is negative! Not possible!", minf, name); 2005156b51fbSBarry Smith /* Put NaN into the time for all events that may not be time accurately since they may happen asynchronously on the GPU */ 2006156b51fbSBarry Smith #if defined(PETSC_HAVE_DEVICE) 2007156b51fbSBarry Smith if (!PetscLogGpuTimeFlag && petsc_gflops > 0) { 2008156b51fbSBarry Smith memcpy(&gmaxt, &nas, sizeof(PetscLogDouble)); 2009156b51fbSBarry Smith PetscCall(PetscEventRegLogGetEvent(stageLog->eventLog, name, &eventid)); 2010156b51fbSBarry Smith if (eventid != SNES_Solve && eventid != KSP_Solve && eventid != TS_Step && eventid != TAO_Solve) { 2011156b51fbSBarry Smith memcpy(&mint, &nas, sizeof(PetscLogDouble)); 2012156b51fbSBarry Smith memcpy(&maxt, &nas, sizeof(PetscLogDouble)); 2013156b51fbSBarry Smith } 2014156b51fbSBarry Smith } 2015156b51fbSBarry Smith #endif 20169371c9d4SSatish Balay totm *= 0.5; 20179371c9d4SSatish Balay totml *= 0.5; 20189371c9d4SSatish Balay totr /= size; 20195c6c1daeSBarry Smith 2020217044c2SLisandro Dalcin if (maxC != 0) { 20219371c9d4SSatish Balay if (minC != 0) ratC = ((PetscLogDouble)maxC) / minC; 20229371c9d4SSatish Balay else ratC = 0.0; 20239371c9d4SSatish Balay if (mint != 0.0) ratt = maxt / mint; 20249371c9d4SSatish Balay else ratt = 0.0; 20259371c9d4SSatish Balay if (minf != 0.0) ratf = maxf / minf; 20269371c9d4SSatish Balay else ratf = 0.0; 20279371c9d4SSatish Balay if (TotalTime != 0.0) fracTime = tott / TotalTime; 20289371c9d4SSatish Balay else fracTime = 0.0; 20299371c9d4SSatish Balay if (TotalFlops != 0.0) fracFlops = totf / TotalFlops; 20309371c9d4SSatish Balay else fracFlops = 0.0; 20319371c9d4SSatish Balay if (stageTime != 0.0) fracStageTime = tott / stageTime; 20329371c9d4SSatish Balay else fracStageTime = 0.0; 20339371c9d4SSatish Balay if (flops != 0.0) fracStageFlops = totf / flops; 20349371c9d4SSatish Balay else fracStageFlops = 0.0; 20359371c9d4SSatish Balay if (numMessages != 0.0) fracMess = totm / numMessages; 20369371c9d4SSatish Balay else fracMess = 0.0; 20379371c9d4SSatish Balay if (messageLength != 0.0) fracMessLen = totml / messageLength; 20389371c9d4SSatish Balay else fracMessLen = 0.0; 20399371c9d4SSatish Balay if (numReductions != 0.0) fracRed = totr / numReductions; 20409371c9d4SSatish Balay else fracRed = 0.0; 20419371c9d4SSatish Balay if (mess != 0.0) fracStageMess = totm / mess; 20429371c9d4SSatish Balay else fracStageMess = 0.0; 20439371c9d4SSatish Balay if (messLen != 0.0) fracStageMessLen = totml / messLen; 20449371c9d4SSatish Balay else fracStageMessLen = 0.0; 20459371c9d4SSatish Balay if (red != 0.0) fracStageRed = totr / red; 20469371c9d4SSatish Balay else fracStageRed = 0.0; 20479371c9d4SSatish Balay if (totm != 0.0) totml /= totm; 20489371c9d4SSatish Balay else totml = 0.0; 20499371c9d4SSatish Balay if (maxt != 0.0) flopr = totf / maxt; 20509371c9d4SSatish Balay else flopr = 0.0; 205168a21331SBarry Smith if (fracStageTime > 1.0 || fracStageFlops > 1.0 || fracStageMess > 1.0 || fracStageMessLen > 1.0 || fracStageRed > 1.0) 2052a76edefcSRicardo Jesus PetscCall(PetscFPrintf(comm, fd, "%-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", 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)); 205368a21331SBarry Smith else 2054a76edefcSRicardo Jesus PetscCall(PetscFPrintf(comm, fd, "%-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", 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)); 205548a46eb9SPierre Jolivet if (PetscLogMemory) PetscCall(PetscFPrintf(comm, fd, " %5.0f %5.0f %5.0f %5.0f", mal / 1.0e6, emalmax / 1.0e6, malmax / 1.0e6, mem / 1.0e6)); 2056d9c66bfbSJunchao Zhang #if defined(PETSC_HAVE_DEVICE) 20579371c9d4SSatish Balay if (totf != 0.0) fracgflops = gflops / totf; 20589371c9d4SSatish Balay else fracgflops = 0.0; 20599371c9d4SSatish Balay if (gmaxt != 0.0) gflopr = gflops / gmaxt; 20609371c9d4SSatish Balay else gflopr = 0.0; 2061a76edefcSRicardo Jesus PetscCall(PetscFPrintf(comm, fd, " %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)); 2062bec0b493Shannah_mairs #endif 20639566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "\n")); 20645c6c1daeSBarry Smith } 20655c6c1daeSBarry Smith } 20665c6c1daeSBarry Smith } 20675c6c1daeSBarry Smith 20685c6c1daeSBarry Smith /* Memory usage and object creation */ 20699566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "------------------------------------------------------------------------------------------------------------------------")); 207048a46eb9SPierre Jolivet if (PetscLogMemory) PetscCall(PetscFPrintf(comm, fd, "-----------------------------")); 2071d9c66bfbSJunchao Zhang #if defined(PETSC_HAVE_DEVICE) 20729566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "---------------------------------------")); 20734863603aSSatish Balay #endif 20749566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "\n")); 20759566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "\n")); 20765c6c1daeSBarry Smith 20775c6c1daeSBarry Smith /* Right now, only stages on the first processor are reported here, meaning only objects associated with 20785c6c1daeSBarry Smith the global communicator, or MPI_COMM_SELF for proc 1. We really should report global stats and then 20795c6c1daeSBarry Smith stats for stages local to processor sets. 20805c6c1daeSBarry Smith */ 20815c6c1daeSBarry Smith /* We should figure out the longest object name here (now 20 characters) */ 20824851f57dSBarry Smith PetscCall(PetscFPrintf(comm, fd, "Object Type Creations Destructions. Reports information only for process 0.\n")); 20835c6c1daeSBarry Smith for (stage = 0; stage < numStages; stage++) { 20845c6c1daeSBarry Smith if (localStageUsed[stage]) { 20855c6c1daeSBarry Smith classInfo = stageLog->stageInfo[stage].classLog->classInfo; 20869566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "\n--- Event Stage %d: %s\n\n", stage, stageInfo[stage].name)); 20875c6c1daeSBarry Smith for (oclass = 0; oclass < stageLog->stageInfo[stage].classLog->numClasses; oclass++) { 20885c6c1daeSBarry Smith if ((classInfo[oclass].creations > 0) || (classInfo[oclass].destructions > 0)) { 20894851f57dSBarry Smith PetscCall(PetscFPrintf(comm, fd, "%20s %5d %5d\n", stageLog->classLog->classInfo[oclass].name, classInfo[oclass].creations, classInfo[oclass].destructions)); 20905c6c1daeSBarry Smith } 20915c6c1daeSBarry Smith } 20925c6c1daeSBarry Smith } else { 2093cf019ec6SStefano Zampini if (!localStageVisible[stage]) continue; 20949566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "\n--- Event Stage %d: Unknown\n\n", stage)); 20955c6c1daeSBarry Smith } 20965c6c1daeSBarry Smith } 20975c6c1daeSBarry Smith 20989566063dSJacob Faibussowitsch PetscCall(PetscFree(localStageUsed)); 20999566063dSJacob Faibussowitsch PetscCall(PetscFree(stageUsed)); 21009566063dSJacob Faibussowitsch PetscCall(PetscFree(localStageVisible)); 21019566063dSJacob Faibussowitsch PetscCall(PetscFree(stageVisible)); 21025c6c1daeSBarry Smith 21035c6c1daeSBarry Smith /* Information unrelated to this particular run */ 21049566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "========================================================================================================================\n")); 21053ba16761SJacob Faibussowitsch PetscCall(PetscTime(&y)); 21063ba16761SJacob Faibussowitsch PetscCall(PetscTime(&x)); 21073ba16761SJacob Faibussowitsch PetscCall(PetscTime(&y)); 21083ba16761SJacob Faibussowitsch PetscCall(PetscTime(&y)); 21093ba16761SJacob Faibussowitsch PetscCall(PetscTime(&y)); 21103ba16761SJacob Faibussowitsch PetscCall(PetscTime(&y)); 21113ba16761SJacob Faibussowitsch PetscCall(PetscTime(&y)); 21123ba16761SJacob Faibussowitsch PetscCall(PetscTime(&y)); 21133ba16761SJacob Faibussowitsch PetscCall(PetscTime(&y)); 21143ba16761SJacob Faibussowitsch PetscCall(PetscTime(&y)); 21153ba16761SJacob Faibussowitsch PetscCall(PetscTime(&y)); 21163ba16761SJacob Faibussowitsch PetscCall(PetscTime(&y)); 21179566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Average time to get PetscTime(): %g\n", (y - x) / 10.0)); 21185c6c1daeSBarry Smith /* MPI information */ 21195c6c1daeSBarry Smith if (size > 1) { 21205c6c1daeSBarry Smith MPI_Status status; 21215c6c1daeSBarry Smith PetscMPIInt tag; 21225c6c1daeSBarry Smith MPI_Comm newcomm; 21235c6c1daeSBarry Smith 21249566063dSJacob Faibussowitsch PetscCallMPI(MPI_Barrier(comm)); 21253ba16761SJacob Faibussowitsch PetscCall(PetscTime(&x)); 21269566063dSJacob Faibussowitsch PetscCallMPI(MPI_Barrier(comm)); 21279566063dSJacob Faibussowitsch PetscCallMPI(MPI_Barrier(comm)); 21289566063dSJacob Faibussowitsch PetscCallMPI(MPI_Barrier(comm)); 21299566063dSJacob Faibussowitsch PetscCallMPI(MPI_Barrier(comm)); 21309566063dSJacob Faibussowitsch PetscCallMPI(MPI_Barrier(comm)); 21313ba16761SJacob Faibussowitsch PetscCall(PetscTime(&y)); 21329566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Average time for MPI_Barrier(): %g\n", (y - x) / 5.0)); 21339566063dSJacob Faibussowitsch PetscCall(PetscCommDuplicate(comm, &newcomm, &tag)); 21349566063dSJacob Faibussowitsch PetscCallMPI(MPI_Barrier(comm)); 21355c6c1daeSBarry Smith if (rank) { 21369566063dSJacob Faibussowitsch PetscCallMPI(MPI_Recv(NULL, 0, MPI_INT, rank - 1, tag, newcomm, &status)); 21379566063dSJacob Faibussowitsch PetscCallMPI(MPI_Send(NULL, 0, MPI_INT, (rank + 1) % size, tag, newcomm)); 21385c6c1daeSBarry Smith } else { 21393ba16761SJacob Faibussowitsch PetscCall(PetscTime(&x)); 21409566063dSJacob Faibussowitsch PetscCallMPI(MPI_Send(NULL, 0, MPI_INT, 1, tag, newcomm)); 21419566063dSJacob Faibussowitsch PetscCallMPI(MPI_Recv(NULL, 0, MPI_INT, size - 1, tag, newcomm, &status)); 21423ba16761SJacob Faibussowitsch PetscCall(PetscTime(&y)); 21439566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Average time for zero size MPI_Send(): %g\n", (y - x) / size)); 21445c6c1daeSBarry Smith } 21459566063dSJacob Faibussowitsch PetscCall(PetscCommDestroy(&newcomm)); 21465c6c1daeSBarry Smith } 21479566063dSJacob Faibussowitsch PetscCall(PetscOptionsView(NULL, viewer)); 21485c6c1daeSBarry Smith 21495c6c1daeSBarry Smith /* Machine and compile information */ 21505c6c1daeSBarry Smith #if defined(PETSC_USE_FORTRAN_KERNELS) 21519566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Compiled with FORTRAN kernels\n")); 21525c6c1daeSBarry Smith #else 21539566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Compiled without FORTRAN kernels\n")); 21545c6c1daeSBarry Smith #endif 2155217044c2SLisandro Dalcin #if defined(PETSC_USE_64BIT_INDICES) 21567de69702SBarry Smith PetscCall(PetscFPrintf(comm, fd, "Compiled with 64-bit PetscInt\n")); 2157217044c2SLisandro Dalcin #elif defined(PETSC_USE___FLOAT128) 21587de69702SBarry Smith PetscCall(PetscFPrintf(comm, fd, "Compiled with 32-bit PetscInt\n")); 2159217044c2SLisandro Dalcin #endif 21605c6c1daeSBarry Smith #if defined(PETSC_USE_REAL_SINGLE) 21619566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Compiled with single precision PetscScalar and PetscReal\n")); 21625f27b2e0SBarry Smith #elif defined(PETSC_USE___FLOAT128) 21639566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Compiled with 128 bit precision PetscScalar and PetscReal\n")); 21645c6c1daeSBarry Smith #endif 21655c6c1daeSBarry Smith #if defined(PETSC_USE_REAL_MAT_SINGLE) 21669566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Compiled with single precision matrices\n")); 21675c6c1daeSBarry Smith #else 21689566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Compiled with full precision matrices (default)\n")); 21695c6c1daeSBarry Smith #endif 21709371c9d4SSatish Balay PetscCall(PetscFPrintf(comm, fd, "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))); 21715c6c1daeSBarry Smith 21729566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "Configure options: %s", petscconfigureoptions)); 21739566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "%s", petscmachineinfo)); 21749566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "%s", petsccompilerinfo)); 21759566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "%s", petsccompilerflagsinfo)); 21769566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "%s", petsclinkerinfo)); 21775c6c1daeSBarry Smith 21785c6c1daeSBarry Smith /* Cleanup */ 21799566063dSJacob Faibussowitsch PetscCall(PetscFPrintf(comm, fd, "\n")); 21809566063dSJacob Faibussowitsch PetscCall(PetscLogViewWarnNoGpuAwareMpi(comm, fd)); 21819566063dSJacob Faibussowitsch PetscCall(PetscLogViewWarnDebugging(comm, fd)); 2182156b51fbSBarry Smith PetscCall(PetscFPTrapPop()); 21833ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 21845c6c1daeSBarry Smith } 21855c6c1daeSBarry Smith 21867d6c928cSSatish Balay /*@C 21877d6c928cSSatish Balay PetscLogView - Prints a summary of the logging. 21885c6c1daeSBarry Smith 21895c6c1daeSBarry Smith Collective over MPI_Comm 21905c6c1daeSBarry Smith 21915c6c1daeSBarry Smith Input Parameter: 2192f14045dbSBarry Smith . viewer - an ASCII viewer 21935c6c1daeSBarry Smith 21945c6c1daeSBarry Smith Options Database Keys: 2195bb1d7374SBarry Smith + -log_view [:filename] - Prints summary of log information 2196bb1d7374SBarry Smith . -log_view :filename.py:ascii_info_detail - Saves logging information from each process as a Python file 2197607d249eSBarry Smith . -log_view :filename.xml:ascii_xml - Saves a summary of the logging information in a nested format (see below for how to view it) 2198d0a29bd7SConnor Ward . -log_view :filename.txt:ascii_flamegraph - Saves logging information in a format suitable for visualising as a Flame Graph (see below for how to view it) 2199156b51fbSBarry Smith . -log_view_memory - Also display memory usage in each event 2200156b51fbSBarry Smith . -log_view_gpu_time - Also display time in each event for GPU kernels (Note this may slow the computation) 2201811af0c4SBarry Smith . -log_all - Saves a file Log.rank for each MPI rank with details of each step of the computation 2202bb1d7374SBarry Smith - -log_trace [filename] - Displays a trace of what each process is doing 22035c6c1daeSBarry Smith 2204d1f92df0SBarry Smith Level: beginner 2205d1f92df0SBarry Smith 22065c6c1daeSBarry Smith Notes: 2207da81f932SPierre Jolivet It is possible to control the logging programmatically but we recommend using the options database approach whenever possible 22085c6c1daeSBarry Smith By default the summary is printed to stdout. 22095c6c1daeSBarry Smith 2210bb1d7374SBarry Smith Before calling this routine you must have called either PetscLogDefaultBegin() or PetscLogNestedBegin() 2211bb1d7374SBarry Smith 2212bb1d7374SBarry Smith If PETSc is configured with --with-logging=0 then this functionality is not available 2213bb1d7374SBarry Smith 2214607d249eSBarry Smith To view the nested XML format filename.xml first copy ${PETSC_DIR}/share/petsc/xml/performance_xml2html.xsl to the current 2215607d249eSBarry Smith directory then open filename.xml with your browser. Specific notes for certain browsers 2216607d249eSBarry Smith $ Firefox and Internet explorer - simply open the file 2217607d249eSBarry Smith $ Google Chrome - you must start up Chrome with the option --allow-file-access-from-files 2218a8d69d7bSBarry Smith $ Safari - see https://ccm.net/faq/36342-safari-how-to-enable-local-file-access 2219607d249eSBarry Smith or one can use the package http://xmlsoft.org/XSLT/xsltproc2.html to translate the xml file to html and then open it with 2220607d249eSBarry Smith your browser. 22212add09c0SLisandro Dalcin Alternatively, use the script ${PETSC_DIR}/lib/petsc/bin/petsc-performance-view to automatically open a new browser 22222add09c0SLisandro Dalcin window and render the XML log file contents. 2223607d249eSBarry Smith 2224bb1d7374SBarry Smith The nested XML format was kindly donated by Koos Huijssen and Christiaan M. Klaij MARITIME RESEARCH INSTITUTE NETHERLANDS 2225bb1d7374SBarry Smith 2226d0a29bd7SConnor Ward The Flame Graph output can be visualised using either the original Flame Graph script (https://github.com/brendangregg/FlameGraph) 2227d0a29bd7SConnor Ward or using speedscope (https://www.speedscope.app). 2228d0a29bd7SConnor Ward Old XML profiles may be converted into this format using the script ${PETSC_DIR}/lib/petsc/bin/xml2flamegraph.py. 2229d0a29bd7SConnor Ward 2230d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogDefaultBegin()`, `PetscLogDump()` 22315c6c1daeSBarry Smith @*/ 2232d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogView(PetscViewer viewer) 2233d71ae5a4SJacob Faibussowitsch { 2234f14045dbSBarry Smith PetscBool isascii; 2235f14045dbSBarry Smith PetscViewerFormat format; 223637b78d16SBarry Smith int stage, lastStage; 223737b78d16SBarry Smith PetscStageLog stageLog; 22385c6c1daeSBarry Smith 22395c6c1daeSBarry Smith PetscFunctionBegin; 224028b400f6SJacob Faibussowitsch PetscCheck(PetscLogPLB, PETSC_COMM_SELF, PETSC_ERR_SUP, "Must use -log_view or PetscLogDefaultBegin() before calling this routine"); 224137b78d16SBarry Smith /* Pop off any stages the user forgot to remove */ 224237b78d16SBarry Smith lastStage = 0; 22439566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 22449566063dSJacob Faibussowitsch PetscCall(PetscStageLogGetCurrent(stageLog, &stage)); 224537b78d16SBarry Smith while (stage >= 0) { 224637b78d16SBarry Smith lastStage = stage; 22479566063dSJacob Faibussowitsch PetscCall(PetscStageLogPop(stageLog)); 22489566063dSJacob Faibussowitsch PetscCall(PetscStageLogGetCurrent(stageLog, &stage)); 224937b78d16SBarry Smith } 22509566063dSJacob Faibussowitsch PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &isascii)); 225128b400f6SJacob Faibussowitsch PetscCheck(isascii, PetscObjectComm((PetscObject)viewer), PETSC_ERR_SUP, "Currently can only view logging to ASCII"); 22529566063dSJacob Faibussowitsch PetscCall(PetscViewerGetFormat(viewer, &format)); 2253f14045dbSBarry Smith if (format == PETSC_VIEWER_DEFAULT || format == PETSC_VIEWER_ASCII_INFO) { 22549566063dSJacob Faibussowitsch PetscCall(PetscLogView_Default(viewer)); 2255f14045dbSBarry Smith } else if (format == PETSC_VIEWER_ASCII_INFO_DETAIL) { 22569566063dSJacob Faibussowitsch PetscCall(PetscLogView_Detailed(viewer)); 225738144912Sdeepblu2718 } else if (format == PETSC_VIEWER_ASCII_CSV) { 22589566063dSJacob Faibussowitsch PetscCall(PetscLogView_CSV(viewer)); 2259bb1d7374SBarry Smith } else if (format == PETSC_VIEWER_ASCII_XML) { 22609566063dSJacob Faibussowitsch PetscCall(PetscLogView_Nested(viewer)); 2261d0a29bd7SConnor Ward } else if (format == PETSC_VIEWER_ASCII_FLAMEGRAPH) { 22629566063dSJacob Faibussowitsch PetscCall(PetscLogView_Flamegraph(viewer)); 22635c6c1daeSBarry Smith } 22649566063dSJacob Faibussowitsch PetscCall(PetscStageLogPush(stageLog, lastStage)); 22653ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 22665c6c1daeSBarry Smith } 22675c6c1daeSBarry Smith 2268f14045dbSBarry Smith /*@C 2269811af0c4SBarry Smith PetscLogViewFromOptions - Processes command line options to determine if/how a `PetscLog` is to be viewed. 2270f14045dbSBarry Smith 2271811af0c4SBarry Smith Collective on `PETSC_COMM_WORLD` 2272f14045dbSBarry Smith 2273811af0c4SBarry Smith Level: developer 2274f14045dbSBarry Smith 2275d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogView()` 2276f14045dbSBarry Smith @*/ 2277d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogViewFromOptions(void) 2278d71ae5a4SJacob Faibussowitsch { 2279f14045dbSBarry Smith PetscViewer viewer; 2280f14045dbSBarry Smith PetscBool flg; 2281f14045dbSBarry Smith PetscViewerFormat format; 2282f14045dbSBarry Smith 2283f14045dbSBarry Smith PetscFunctionBegin; 22849566063dSJacob Faibussowitsch PetscCall(PetscOptionsGetViewer(PETSC_COMM_WORLD, NULL, NULL, "-log_view", &viewer, &format, &flg)); 2285f14045dbSBarry Smith if (flg) { 22869566063dSJacob Faibussowitsch PetscCall(PetscViewerPushFormat(viewer, format)); 22879566063dSJacob Faibussowitsch PetscCall(PetscLogView(viewer)); 22889566063dSJacob Faibussowitsch PetscCall(PetscViewerPopFormat(viewer)); 22899566063dSJacob Faibussowitsch PetscCall(PetscViewerDestroy(&viewer)); 2290f14045dbSBarry Smith } 22913ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 2292f14045dbSBarry Smith } 2293f14045dbSBarry Smith 22945c6c1daeSBarry Smith /*----------------------------------------------- Counter Functions -------------------------------------------------*/ 22955c6c1daeSBarry Smith /*@C 22965c6c1daeSBarry Smith PetscGetFlops - Returns the number of flops used on this processor 22975c6c1daeSBarry Smith since the program began. 22985c6c1daeSBarry Smith 22995c6c1daeSBarry Smith Not Collective 23005c6c1daeSBarry Smith 23015c6c1daeSBarry Smith Output Parameter: 2302*10450e9eSJacob Faibussowitsch . flops - number of floating point operations 23035c6c1daeSBarry Smith 2304d1f92df0SBarry Smith Level: intermediate 2305d1f92df0SBarry Smith 23065c6c1daeSBarry Smith Notes: 23075c6c1daeSBarry Smith A global counter logs all PETSc flop counts. The user can use 2308811af0c4SBarry Smith `PetscLogFlops()` to increment this counter to include flops for the 23095c6c1daeSBarry Smith application code. 23105c6c1daeSBarry Smith 2311811af0c4SBarry Smith A separate counter `PetscLogGPUFlops()` logs the flops that occur on any GPU associated with this MPI rank 2312811af0c4SBarry Smith 2313d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogGPUFlops()`, `PetscTime()`, `PetscLogFlops()` 23145c6c1daeSBarry Smith @*/ 2315d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscGetFlops(PetscLogDouble *flops) 2316d71ae5a4SJacob Faibussowitsch { 23175c6c1daeSBarry Smith PetscFunctionBegin; 23185c6c1daeSBarry Smith *flops = petsc_TotalFlops; 23193ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 23205c6c1daeSBarry Smith } 23215c6c1daeSBarry Smith 2322d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogObjectState(PetscObject obj, const char format[], ...) 2323d71ae5a4SJacob Faibussowitsch { 23245c6c1daeSBarry Smith size_t fullLength; 23255c6c1daeSBarry Smith va_list Argp; 23265c6c1daeSBarry Smith 23275c6c1daeSBarry Smith PetscFunctionBegin; 23283ba16761SJacob Faibussowitsch if (!petsc_logObjects) PetscFunctionReturn(PETSC_SUCCESS); 23295c6c1daeSBarry Smith va_start(Argp, format); 23309566063dSJacob Faibussowitsch PetscCall(PetscVSNPrintf(petsc_objects[obj->id].info, 64, format, &fullLength, Argp)); 23315c6c1daeSBarry Smith va_end(Argp); 23323ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 23335c6c1daeSBarry Smith } 23345c6c1daeSBarry Smith 23355c6c1daeSBarry Smith /*MC 23365c6c1daeSBarry Smith PetscLogFlops - Adds floating point operations to the global counter. 23375c6c1daeSBarry Smith 23385c6c1daeSBarry Smith Synopsis: 2339aaa7dc30SBarry Smith #include <petsclog.h> 23405c6c1daeSBarry Smith PetscErrorCode PetscLogFlops(PetscLogDouble f) 23415c6c1daeSBarry Smith 23425c6c1daeSBarry Smith Not Collective 23435c6c1daeSBarry Smith 23445c6c1daeSBarry Smith Input Parameter: 23455c6c1daeSBarry Smith . f - flop counter 23465c6c1daeSBarry Smith 2347*10450e9eSJacob Faibussowitsch Example Usage: 23485c6c1daeSBarry Smith .vb 23495c6c1daeSBarry Smith PetscLogEvent USER_EVENT; 2350*10450e9eSJacob Faibussowitsch 23515c6c1daeSBarry Smith PetscLogEventRegister("User event", 0, &USER_EVENT); 23525c6c1daeSBarry Smith PetscLogEventBegin(USER_EVENT, 0, 0, 0, 0); 23535c6c1daeSBarry Smith [code segment to monitor] 23545c6c1daeSBarry Smith PetscLogFlops(user_flops) 23555c6c1daeSBarry Smith PetscLogEventEnd(USER_EVENT, 0, 0, 0, 0); 23565c6c1daeSBarry Smith .ve 23575c6c1daeSBarry Smith 2358d1f92df0SBarry Smith Level: intermediate 2359d1f92df0SBarry Smith 2360811af0c4SBarry Smith Note: 2361*10450e9eSJacob Faibussowitsch A global counter logs all PETSc flop counts. The user can use PetscLogFlops() to increment 2362*10450e9eSJacob Faibussowitsch this counter to include flops for the application code. 23635c6c1daeSBarry Smith 2364d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogGPUFlops()`, `PetscLogEventRegister()`, `PetscLogEventBegin()`, `PetscLogEventEnd()`, `PetscGetFlops()` 23655c6c1daeSBarry Smith M*/ 23665c6c1daeSBarry Smith 23675c6c1daeSBarry Smith /*MC 2368*10450e9eSJacob Faibussowitsch PetscPreLoadBegin - Begin a segment of code that may be preloaded (run twice) to get accurate 2369*10450e9eSJacob Faibussowitsch timings 23705c6c1daeSBarry Smith 23715c6c1daeSBarry Smith Synopsis: 2372aaa7dc30SBarry Smith #include <petsclog.h> 23735c6c1daeSBarry Smith void PetscPreLoadBegin(PetscBool flag, char *name); 23745c6c1daeSBarry Smith 23755c6c1daeSBarry Smith Not Collective 23765c6c1daeSBarry Smith 2377d8d19677SJose E. Roman Input Parameters: 2378*10450e9eSJacob Faibussowitsch + flag - `PETSC_TRUE` to run twice, `PETSC_FALSE` to run once, may be overridden with command 2379*10450e9eSJacob Faibussowitsch line option `-preload true|false` 2380*10450e9eSJacob Faibussowitsch - name - name of first stage (lines of code timed separately with `-log_view`) to be preloaded 23815c6c1daeSBarry Smith 2382*10450e9eSJacob Faibussowitsch Example Usage: 23835c6c1daeSBarry Smith .vb 2384*10450e9eSJacob Faibussowitsch PetscPreLoadBegin(PETSC_TRUE, "first stage"); 2385*10450e9eSJacob Faibussowitsch // lines of code 23865c6c1daeSBarry Smith PetscPreLoadStage("second stage"); 2387*10450e9eSJacob Faibussowitsch // lines of code 23885c6c1daeSBarry Smith PetscPreLoadEnd(); 23895c6c1daeSBarry Smith .ve 23905c6c1daeSBarry Smith 2391d1f92df0SBarry Smith Level: intermediate 2392d1f92df0SBarry Smith 2393811af0c4SBarry Smith Note: 239495452b02SPatrick Sanan Only works in C/C++, not Fortran 23955c6c1daeSBarry Smith 2396*10450e9eSJacob Faibussowitsch Flags available within the macro\: 2397*10450e9eSJacob Faibussowitsch + PetscPreLoadingUsed - `PETSC_TRUE` if we are or have done preloading 2398*10450e9eSJacob Faibussowitsch . PetscPreLoadingOn - `PETSC_TRUE` if it is CURRENTLY doing preload 2399*10450e9eSJacob Faibussowitsch . PetscPreLoadIt - `0` for the first computation (with preloading turned off it is only 2400*10450e9eSJacob Faibussowitsch `0`) `1` for the second 2401*10450e9eSJacob Faibussowitsch - PetscPreLoadMax - number of times it will do the computation, only one when preloading is 2402*10450e9eSJacob Faibussowitsch turned on 2403*10450e9eSJacob Faibussowitsch 2404*10450e9eSJacob Faibussowitsch The first two variables are available throughout the program, the second two only between the 2405*10450e9eSJacob Faibussowitsch `PetscPreLoadBegin()` and `PetscPreLoadEnd()` 24065c6c1daeSBarry Smith 2407d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogEventRegister()`, `PetscLogEventBegin()`, `PetscLogEventEnd()`, `PetscPreLoadEnd()`, `PetscPreLoadStage()` 24085c6c1daeSBarry Smith M*/ 24095c6c1daeSBarry Smith 24105c6c1daeSBarry Smith /*MC 2411*10450e9eSJacob Faibussowitsch PetscPreLoadEnd - End a segment of code that may be preloaded (run twice) to get accurate 2412*10450e9eSJacob Faibussowitsch timings 24135c6c1daeSBarry Smith 24145c6c1daeSBarry Smith Synopsis: 2415aaa7dc30SBarry Smith #include <petsclog.h> 24165c6c1daeSBarry Smith void PetscPreLoadEnd(void); 24175c6c1daeSBarry Smith 24185c6c1daeSBarry Smith Not Collective 24195c6c1daeSBarry Smith 2420*10450e9eSJacob Faibussowitsch Example Usage: 24215c6c1daeSBarry Smith .vb 2422*10450e9eSJacob Faibussowitsch PetscPreLoadBegin(PETSC_TRUE, "first stage"); 2423*10450e9eSJacob Faibussowitsch // lines of code 24245c6c1daeSBarry Smith PetscPreLoadStage("second stage"); 2425*10450e9eSJacob Faibussowitsch // lines of code 24265c6c1daeSBarry Smith PetscPreLoadEnd(); 24275c6c1daeSBarry Smith .ve 24285c6c1daeSBarry Smith 2429d1f92df0SBarry Smith Level: intermediate 2430d1f92df0SBarry Smith 2431811af0c4SBarry Smith Note: 2432811af0c4SBarry Smith Only works in C/C++ not fortran 24335c6c1daeSBarry Smith 2434d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogEventRegister()`, `PetscLogEventBegin()`, `PetscLogEventEnd()`, `PetscPreLoadBegin()`, `PetscPreLoadStage()` 24355c6c1daeSBarry Smith M*/ 24365c6c1daeSBarry Smith 24375c6c1daeSBarry Smith /*MC 2438*10450e9eSJacob Faibussowitsch PetscPreLoadStage - Start a new segment of code to be timed separately to get accurate timings 24395c6c1daeSBarry Smith 24405c6c1daeSBarry Smith Synopsis: 2441aaa7dc30SBarry Smith #include <petsclog.h> 24425c6c1daeSBarry Smith void PetscPreLoadStage(char *name); 24435c6c1daeSBarry Smith 24445c6c1daeSBarry Smith Not Collective 24455c6c1daeSBarry Smith 2446*10450e9eSJacob Faibussowitsch Example Usage: 24475c6c1daeSBarry Smith .vb 2448*10450e9eSJacob Faibussowitsch PetscPreLoadBegin(PETSC_TRUE,"first stage"); 2449*10450e9eSJacob Faibussowitsch // lines of code 24505c6c1daeSBarry Smith PetscPreLoadStage("second stage"); 2451*10450e9eSJacob Faibussowitsch // lines of code 24525c6c1daeSBarry Smith PetscPreLoadEnd(); 24535c6c1daeSBarry Smith .ve 24545c6c1daeSBarry Smith 2455d1f92df0SBarry Smith Level: intermediate 2456d1f92df0SBarry Smith 2457811af0c4SBarry Smith Note: 2458811af0c4SBarry Smith Only works in C/C++ not fortran 24595c6c1daeSBarry Smith 2460d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogEventRegister()`, `PetscLogEventBegin()`, `PetscLogEventEnd()`, `PetscPreLoadBegin()`, `PetscPreLoadEnd()` 24615c6c1daeSBarry Smith M*/ 24625c6c1daeSBarry Smith 2463a4af0ceeSJacob Faibussowitsch #if PetscDefined(HAVE_DEVICE) 2464a4af0ceeSJacob Faibussowitsch #include <petsc/private/deviceimpl.h> 24659ffd0706SHong Zhang 2466156b51fbSBarry Smith PetscBool PetscLogGpuTimeFlag = PETSC_FALSE; 2467156b51fbSBarry Smith 2468156b51fbSBarry Smith /* 2469156b51fbSBarry Smith This cannot be called by users between PetscInitialize() and PetscFinalize() at any random location in the code 2470156b51fbSBarry Smith because it will result in timing results that cannot be interpreted. 2471156b51fbSBarry Smith */ 2472d71ae5a4SJacob Faibussowitsch static PetscErrorCode PetscLogGpuTime_Off(void) 2473d71ae5a4SJacob Faibussowitsch { 2474156b51fbSBarry Smith PetscLogGpuTimeFlag = PETSC_FALSE; 24753ba16761SJacob Faibussowitsch return PETSC_SUCCESS; 2476156b51fbSBarry Smith } 2477156b51fbSBarry Smith 2478156b51fbSBarry Smith /*@C 2479156b51fbSBarry Smith PetscLogGpuTime - turn on the logging of GPU time for GPU kernels 2480156b51fbSBarry Smith 2481811af0c4SBarry Smith Options Database Key: 2482*10450e9eSJacob Faibussowitsch . -log_view_gpu_time - provide the GPU times in the `-log_view` output 2483156b51fbSBarry Smith 2484d1f92df0SBarry Smith Level: advanced 2485d1f92df0SBarry Smith 2486156b51fbSBarry Smith Notes: 2487*10450e9eSJacob Faibussowitsch Turning on the timing of the GPU kernels can slow down the entire computation and should only 2488*10450e9eSJacob Faibussowitsch be used when studying the performance of operations on GPU such as vector operations and 2489*10450e9eSJacob Faibussowitsch matrix-vector operations. 2490156b51fbSBarry Smith 2491*10450e9eSJacob Faibussowitsch This routine should only be called once near the beginning of the program. Once it is started 2492*10450e9eSJacob Faibussowitsch it cannot be turned off. 2493156b51fbSBarry Smith 2494d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogView()`, `PetscLogGpuFlops()`, `PetscLogGpuTimeEnd()`, `PetscLogGpuTimeBegin()` 2495156b51fbSBarry Smith @*/ 2496d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogGpuTime(void) 2497d71ae5a4SJacob Faibussowitsch { 2498156b51fbSBarry Smith if (!PetscLogGpuTimeFlag) PetscCall(PetscRegisterFinalize(PetscLogGpuTime_Off)); 2499156b51fbSBarry Smith PetscLogGpuTimeFlag = PETSC_TRUE; 25003ba16761SJacob Faibussowitsch return PETSC_SUCCESS; 2501156b51fbSBarry Smith } 2502156b51fbSBarry Smith 25039ffd0706SHong Zhang /*@C 25049ffd0706SHong Zhang PetscLogGpuTimeBegin - Start timer for device 25059ffd0706SHong Zhang 2506d1f92df0SBarry Smith Level: intermediate 2507d1f92df0SBarry Smith 25089ffd0706SHong Zhang Notes: 2509*10450e9eSJacob Faibussowitsch When CUDA or HIP is enabled, the timer is run on the GPU, it is a separate logging of time 2510*10450e9eSJacob Faibussowitsch devoted to GPU computations (excluding kernel launch times). 2511811af0c4SBarry Smith 2512*10450e9eSJacob Faibussowitsch When CUDA or HIP is not available, the timer is run on the CPU, it is a separate logging of 2513*10450e9eSJacob Faibussowitsch time devoted to GPU computations (including kernel launch times). 2514811af0c4SBarry Smith 2515*10450e9eSJacob Faibussowitsch There is no need to call WaitForCUDA() or WaitForHIP() between `PetscLogGpuTimeBegin()` and 2516*10450e9eSJacob Faibussowitsch `PetscLogGpuTimeEnd()` 2517811af0c4SBarry Smith 2518*10450e9eSJacob Faibussowitsch This timer should NOT include times for data transfers between the GPU and CPU, nor setup 2519*10450e9eSJacob Faibussowitsch actions such as allocating space. 2520811af0c4SBarry Smith 2521*10450e9eSJacob Faibussowitsch The regular logging captures the time for data transfers and any CPU activities during the 2522*10450e9eSJacob Faibussowitsch event. It is used to compute the flop rate on the GPU as it is actively engaged in running a 2523*10450e9eSJacob Faibussowitsch kernel. 25249ffd0706SHong Zhang 25259ffd0706SHong Zhang Developer Notes: 2526*10450e9eSJacob Faibussowitsch The GPU event timer captures the execution time of all the kernels launched in the default 2527*10450e9eSJacob Faibussowitsch stream by the CPU between `PetscLogGpuTimeBegin()` and `PetsLogGpuTimeEnd()`. 2528811af0c4SBarry Smith 2529*10450e9eSJacob Faibussowitsch `PetscLogGpuTimeBegin()` and `PetsLogGpuTimeEnd()` insert the begin and end events into the 2530*10450e9eSJacob Faibussowitsch default stream (stream 0). The device will record a time stamp for the event when it reaches 2531*10450e9eSJacob Faibussowitsch that event in the stream. The function xxxEventSynchronize() is called in 2532*10450e9eSJacob Faibussowitsch `PetsLogGpuTimeEnd()` to block CPU execution, but not continued GPU execution, until the 2533*10450e9eSJacob Faibussowitsch timer event is recorded. 25349ffd0706SHong Zhang 2535d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogView()`, `PetscLogGpuFlops()`, `PetscLogGpuTimeEnd()`, `PetscLogGpuTime()` 25369ffd0706SHong Zhang @*/ 2537d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogGpuTimeBegin(void) 2538d71ae5a4SJacob Faibussowitsch { 25399ffd0706SHong Zhang PetscFunctionBegin; 25403ba16761SJacob Faibussowitsch if (!PetscLogPLB || !PetscLogGpuTimeFlag) PetscFunctionReturn(PETSC_SUCCESS); 2541744d70b0SJunchao Zhang if (PetscDefined(HAVE_DEVICE)) { 2542a4af0ceeSJacob Faibussowitsch PetscDeviceContext dctx; 2543a4af0ceeSJacob Faibussowitsch 25449566063dSJacob Faibussowitsch PetscCall(PetscDeviceContextGetCurrentContext(&dctx)); 25459566063dSJacob Faibussowitsch PetscCall(PetscDeviceContextBeginTimer_Internal(dctx)); 2546a4af0ceeSJacob Faibussowitsch } else { 25479566063dSJacob Faibussowitsch PetscCall(PetscTimeSubtract(&petsc_gtime)); 2548a4af0ceeSJacob Faibussowitsch } 25493ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 25509ffd0706SHong Zhang } 25519ffd0706SHong Zhang 25529ffd0706SHong Zhang /*@C 25539ffd0706SHong Zhang PetscLogGpuTimeEnd - Stop timer for device 25549ffd0706SHong Zhang 25559ffd0706SHong Zhang Level: intermediate 25569ffd0706SHong Zhang 2557d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogView()`, `PetscLogGpuFlops()`, `PetscLogGpuTimeBegin()` 25589ffd0706SHong Zhang @*/ 2559d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogGpuTimeEnd(void) 2560d71ae5a4SJacob Faibussowitsch { 25619ffd0706SHong Zhang PetscFunctionBegin; 25623ba16761SJacob Faibussowitsch if (!PetscLogPLE || !PetscLogGpuTimeFlag) PetscFunctionReturn(PETSC_SUCCESS); 2563744d70b0SJunchao Zhang if (PetscDefined(HAVE_DEVICE)) { 2564a4af0ceeSJacob Faibussowitsch PetscDeviceContext dctx; 2565a4af0ceeSJacob Faibussowitsch PetscLogDouble elapsed; 2566a4af0ceeSJacob Faibussowitsch 25679566063dSJacob Faibussowitsch PetscCall(PetscDeviceContextGetCurrentContext(&dctx)); 25689566063dSJacob Faibussowitsch PetscCall(PetscDeviceContextEndTimer_Internal(dctx, &elapsed)); 2569a4af0ceeSJacob Faibussowitsch petsc_gtime += (elapsed / 1000.0); 2570a4af0ceeSJacob Faibussowitsch } else { 25719566063dSJacob Faibussowitsch PetscCall(PetscTimeAdd(&petsc_gtime)); 2572a4af0ceeSJacob Faibussowitsch } 25733ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 25749ffd0706SHong Zhang } 2575c708d6e3SStefano Zampini 25769ffd0706SHong Zhang #endif /* end of PETSC_HAVE_DEVICE */ 25779ffd0706SHong Zhang 25785c6c1daeSBarry Smith #else /* end of -DPETSC_USE_LOG section */ 25795c6c1daeSBarry Smith 2580d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogObjectState(PetscObject obj, const char format[], ...) 2581d71ae5a4SJacob Faibussowitsch { 25825c6c1daeSBarry Smith PetscFunctionBegin; 25833ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 25845c6c1daeSBarry Smith } 25855c6c1daeSBarry Smith 25865c6c1daeSBarry Smith #endif /* PETSC_USE_LOG*/ 25875c6c1daeSBarry Smith 25885c6c1daeSBarry Smith PetscClassId PETSC_LARGEST_CLASSID = PETSC_SMALLEST_CLASSID; 25895c6c1daeSBarry Smith PetscClassId PETSC_OBJECT_CLASSID = 0; 25905c6c1daeSBarry Smith 25915c6c1daeSBarry Smith /*@C 25925c6c1daeSBarry Smith PetscClassIdRegister - Registers a new class name for objects and logging operations in an application code. 25935c6c1daeSBarry Smith 25945c6c1daeSBarry Smith Not Collective 25955c6c1daeSBarry Smith 25965c6c1daeSBarry Smith Input Parameter: 25975c6c1daeSBarry Smith . name - The class name 25985c6c1daeSBarry Smith 25995c6c1daeSBarry Smith Output Parameter: 26005c6c1daeSBarry Smith . oclass - The class id or classid 26015c6c1daeSBarry Smith 26025c6c1daeSBarry Smith Level: developer 26035c6c1daeSBarry Smith 2604d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogEventRegister()` 26055c6c1daeSBarry Smith @*/ 2606d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscClassIdRegister(const char name[], PetscClassId *oclass) 2607d71ae5a4SJacob Faibussowitsch { 26085c6c1daeSBarry Smith #if defined(PETSC_USE_LOG) 26095c6c1daeSBarry Smith PetscStageLog stageLog; 26105c6c1daeSBarry Smith PetscInt stage; 26115c6c1daeSBarry Smith #endif 26125c6c1daeSBarry Smith 26135c6c1daeSBarry Smith PetscFunctionBegin; 26145c6c1daeSBarry Smith *oclass = ++PETSC_LARGEST_CLASSID; 26155c6c1daeSBarry Smith #if defined(PETSC_USE_LOG) 26169566063dSJacob Faibussowitsch PetscCall(PetscLogGetStageLog(&stageLog)); 26179566063dSJacob Faibussowitsch PetscCall(PetscClassRegLogRegister(stageLog->classLog, name, *oclass)); 261848a46eb9SPierre Jolivet for (stage = 0; stage < stageLog->numStages; stage++) PetscCall(PetscClassPerfLogEnsureSize(stageLog->stageInfo[stage].classLog, stageLog->classLog->numClasses)); 26195c6c1daeSBarry Smith #endif 26203ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 26215c6c1daeSBarry Smith } 2622210b5426SBarry Smith 2623210b5426SBarry Smith #if defined(PETSC_USE_LOG) && defined(PETSC_HAVE_MPE) 2624210b5426SBarry Smith #include <mpe.h> 2625210b5426SBarry Smith 2626210b5426SBarry Smith PetscBool PetscBeganMPE = PETSC_FALSE; 2627210b5426SBarry Smith 2628495fc317SBarry Smith PETSC_INTERN PetscErrorCode PetscLogEventBeginMPE(PetscLogEvent, int, PetscObject, PetscObject, PetscObject, PetscObject); 2629495fc317SBarry Smith PETSC_INTERN PetscErrorCode PetscLogEventEndMPE(PetscLogEvent, int, PetscObject, PetscObject, PetscObject, PetscObject); 2630495fc317SBarry Smith 2631210b5426SBarry Smith /*@C 2632*10450e9eSJacob Faibussowitsch PetscLogMPEBegin - Turns on MPE logging of events. This creates large log files and slows the 2633*10450e9eSJacob Faibussowitsch program down. 2634210b5426SBarry Smith 2635811af0c4SBarry Smith Collective over `PETSC_COMM_WORLD` 2636210b5426SBarry Smith 2637811af0c4SBarry Smith Options Database Key: 2638a2553e36SBarry Smith . -log_mpe - Prints extensive log information 2639210b5426SBarry Smith 2640d1f92df0SBarry Smith Level: advanced 2641d1f92df0SBarry Smith 2642811af0c4SBarry Smith Note: 2643*10450e9eSJacob Faibussowitsch A related routine is `PetscLogDefaultBegin()` (with the options key `-log_view`), which is 2644*10450e9eSJacob Faibussowitsch intended for production runs since it logs only flop rates and object creation (and should 2645*10450e9eSJacob Faibussowitsch not significantly slow the programs). 2646210b5426SBarry Smith 2647d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogDump()`, `PetscLogDefaultBegin()`, `PetscLogAllBegin()`, `PetscLogEventActivate()`, 2648db781477SPatrick Sanan `PetscLogEventDeactivate()` 2649210b5426SBarry Smith @*/ 2650d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogMPEBegin(void) 2651d71ae5a4SJacob Faibussowitsch { 2652210b5426SBarry Smith PetscFunctionBegin; 2653210b5426SBarry Smith /* Do MPE initialization */ 2654210b5426SBarry Smith if (!MPE_Initialized_logging()) { /* This function exists in mpich 1.1.2 and higher */ 26559566063dSJacob Faibussowitsch PetscCall(PetscInfo(0, "Initializing MPE.\n")); 26569566063dSJacob Faibussowitsch PetscCall(MPE_Init_log()); 2657210b5426SBarry Smith 2658210b5426SBarry Smith PetscBeganMPE = PETSC_TRUE; 2659210b5426SBarry Smith } else { 26609566063dSJacob Faibussowitsch PetscCall(PetscInfo(0, "MPE already initialized. Not attempting to reinitialize.\n")); 2661210b5426SBarry Smith } 26629566063dSJacob Faibussowitsch PetscCall(PetscLogSet(PetscLogEventBeginMPE, PetscLogEventEndMPE)); 26633ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 2664210b5426SBarry Smith } 2665210b5426SBarry Smith 2666210b5426SBarry Smith /*@C 2667495fc317SBarry Smith PetscLogMPEDump - Dumps the MPE logging info to file for later use with Jumpshot. 2668210b5426SBarry Smith 2669*10450e9eSJacob Faibussowitsch Input Parameter: 2670*10450e9eSJacob Faibussowitsch . sname - The filename to dump to 2671*10450e9eSJacob Faibussowitsch 2672811af0c4SBarry Smith Collective over `PETSC_COMM_WORLD` 2673210b5426SBarry Smith 2674210b5426SBarry Smith Level: advanced 2675210b5426SBarry Smith 2676d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogDump()`, `PetscLogAllBegin()`, `PetscLogMPEBegin()` 2677210b5426SBarry Smith @*/ 2678d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogMPEDump(const char sname[]) 2679d71ae5a4SJacob Faibussowitsch { 2680210b5426SBarry Smith char name[PETSC_MAX_PATH_LEN]; 2681210b5426SBarry Smith 2682210b5426SBarry Smith PetscFunctionBegin; 2683210b5426SBarry Smith if (PetscBeganMPE) { 26849566063dSJacob Faibussowitsch PetscCall(PetscInfo(0, "Finalizing MPE.\n")); 2685210b5426SBarry Smith if (sname) { 2686c6a7a370SJeremy L Thompson PetscCall(PetscStrncpy(name, sname, sizeof(name))); 2687210b5426SBarry Smith } else { 26889566063dSJacob Faibussowitsch PetscCall(PetscGetProgramName(name, sizeof(name))); 2689210b5426SBarry Smith } 26909566063dSJacob Faibussowitsch PetscCall(MPE_Finish_log(name)); 2691210b5426SBarry Smith } else { 26929566063dSJacob Faibussowitsch PetscCall(PetscInfo(0, "Not finalizing MPE (not started by PETSc).\n")); 2693210b5426SBarry Smith } 26943ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 2695210b5426SBarry Smith } 2696210b5426SBarry Smith 2697210b5426SBarry Smith #define PETSC_RGB_COLORS_MAX 39 26989371c9d4SSatish Balay static const char *PetscLogMPERGBColors[PETSC_RGB_COLORS_MAX] = {"OliveDrab: ", "BlueViolet: ", "CadetBlue: ", "CornflowerBlue: ", "DarkGoldenrod: ", "DarkGreen: ", "DarkKhaki: ", "DarkOliveGreen: ", 26999371c9d4SSatish Balay "DarkOrange: ", "DarkOrchid: ", "DarkSeaGreen: ", "DarkSlateGray: ", "DarkTurquoise: ", "DeepPink: ", "DarkKhaki: ", "DimGray: ", 27009371c9d4SSatish Balay "DodgerBlue: ", "GreenYellow: ", "HotPink: ", "IndianRed: ", "LavenderBlush: ", "LawnGreen: ", "LemonChiffon: ", "LightCoral: ", 27019371c9d4SSatish Balay "LightCyan: ", "LightPink: ", "LightSalmon: ", "LightSlateGray: ", "LightYellow: ", "LimeGreen: ", "MediumPurple: ", "MediumSeaGreen: ", 27029371c9d4SSatish Balay "MediumSlateBlue:", "MidnightBlue: ", "MintCream: ", "MistyRose: ", "NavajoWhite: ", "NavyBlue: ", "OliveDrab: "}; 2703210b5426SBarry Smith 2704210b5426SBarry Smith /*@C 2705811af0c4SBarry Smith PetscLogMPEGetRGBColor - This routine returns a rgb color useable with `PetscLogEventRegister()` 2706210b5426SBarry Smith 2707210b5426SBarry Smith Not collective. Maybe it should be? 2708210b5426SBarry Smith 27097a7aea1fSJed Brown Output Parameter: 2710210b5426SBarry Smith . str - character string representing the color 2711210b5426SBarry Smith 2712210b5426SBarry Smith Level: developer 2713210b5426SBarry Smith 2714d1f92df0SBarry Smith .seealso: [](ch_profiling), `PetscLogEventRegister()` 2715210b5426SBarry Smith @*/ 2716d71ae5a4SJacob Faibussowitsch PetscErrorCode PetscLogMPEGetRGBColor(const char *str[]) 2717d71ae5a4SJacob Faibussowitsch { 2718210b5426SBarry Smith static int idx = 0; 2719210b5426SBarry Smith 2720210b5426SBarry Smith PetscFunctionBegin; 2721210b5426SBarry Smith *str = PetscLogMPERGBColors[idx]; 2722210b5426SBarry Smith idx = (idx + 1) % PETSC_RGB_COLORS_MAX; 27233ba16761SJacob Faibussowitsch PetscFunctionReturn(PETSC_SUCCESS); 2724210b5426SBarry Smith } 2725210b5426SBarry Smith 2726210b5426SBarry Smith #endif /* PETSC_USE_LOG && PETSC_HAVE_MPE */ 2727