197bb86f7SLois Curfman McInnes /* 27588ac45SBarry Smith Defines profile/logging in PETSc. 397bb86f7SLois Curfman McInnes */ 497bb86f7SLois Curfman McInnes 526bd1501SBarry Smith #if !defined(PETSCLOG_H) 626bd1501SBarry Smith #define PETSCLOG_H 7aaa7dc30SBarry Smith #include <petscsys.h> 8958c4211Shannah_mairs #include <petsctime.h> 9c8d78d4dSSatish Balay 10fa2bb9feSLisandro Dalcin /* General logging of information; different from event logging */ 11fe9b927eSVaclav Hapla PETSC_EXTERN PetscErrorCode PetscInfo_Private(const char[],PetscObject,const char[],...); 12fa2bb9feSLisandro Dalcin #if defined(PETSC_USE_INFO) 13fe9b927eSVaclav Hapla #define PetscInfo(A,S) PetscInfo_Private(PETSC_FUNCTION_NAME,((PetscObject)A),S) 14fe9b927eSVaclav Hapla #define PetscInfo1(A,S,a1) PetscInfo_Private(PETSC_FUNCTION_NAME,((PetscObject)A),S,a1) 15fe9b927eSVaclav Hapla #define PetscInfo2(A,S,a1,a2) PetscInfo_Private(PETSC_FUNCTION_NAME,((PetscObject)A),S,a1,a2) 16fe9b927eSVaclav Hapla #define PetscInfo3(A,S,a1,a2,a3) PetscInfo_Private(PETSC_FUNCTION_NAME,((PetscObject)A),S,a1,a2,a3) 17fe9b927eSVaclav Hapla #define PetscInfo4(A,S,a1,a2,a3,a4) PetscInfo_Private(PETSC_FUNCTION_NAME,((PetscObject)A),S,a1,a2,a3,a4) 18fe9b927eSVaclav Hapla #define PetscInfo5(A,S,a1,a2,a3,a4,a5) PetscInfo_Private(PETSC_FUNCTION_NAME,((PetscObject)A),S,a1,a2,a3,a4,a5) 19fe9b927eSVaclav Hapla #define PetscInfo6(A,S,a1,a2,a3,a4,a5,a6) PetscInfo_Private(PETSC_FUNCTION_NAME,((PetscObject)A),S,a1,a2,a3,a4,a5,a6) 20fe9b927eSVaclav Hapla #define PetscInfo7(A,S,a1,a2,a3,a4,a5,a6,a7) PetscInfo_Private(PETSC_FUNCTION_NAME,((PetscObject)A),S,a1,a2,a3,a4,a5,a6,a7) 21fa2bb9feSLisandro Dalcin #else 22fa2bb9feSLisandro Dalcin #define PetscInfo(A,S) 0 23fa2bb9feSLisandro Dalcin #define PetscInfo1(A,S,a1) 0 24fa2bb9feSLisandro Dalcin #define PetscInfo2(A,S,a1,a2) 0 25fa2bb9feSLisandro Dalcin #define PetscInfo3(A,S,a1,a2,a3) 0 26fa2bb9feSLisandro Dalcin #define PetscInfo4(A,S,a1,a2,a3,a4) 0 27fa2bb9feSLisandro Dalcin #define PetscInfo5(A,S,a1,a2,a3,a4,a5) 0 28fa2bb9feSLisandro Dalcin #define PetscInfo6(A,S,a1,a2,a3,a4,a5,a6) 0 29fa2bb9feSLisandro Dalcin #define PetscInfo7(A,S,a1,a2,a3,a4,a5,a6,a7) 0 30fa2bb9feSLisandro Dalcin #endif 31e94e781bSJacob Faibussowitsch 32e94e781bSJacob Faibussowitsch /*E 33e94e781bSJacob Faibussowitsch PetscInfoCommFlag - Describes the method by which to filter PetscInfo() by communicator size 34e94e781bSJacob Faibussowitsch 35e94e781bSJacob Faibussowitsch Used as an input for PetscInfoSetFilterCommSelf() 36e94e781bSJacob Faibussowitsch 37e94e781bSJacob Faibussowitsch $ PETSC_INFO_COMM_ALL - Default uninitialized value. PetscInfo() will not filter based on communicator size (i.e. will 38e94e781bSJacob Faibussowitsch print for all communicators) 39e94e781bSJacob Faibussowitsch $ PETSC_INFO_COMM_NO_SELF - PetscInfo() will NOT print for communicators with size = 1 (i.e. *_COMM_SELF) 40e94e781bSJacob Faibussowitsch $ PETSC_INFO_COMM_ONLY_SELF - PetscInfo will ONLY print for communicators with size = 1 41e94e781bSJacob Faibussowitsch 42e94e781bSJacob Faibussowitsch Level: intermediate 43e94e781bSJacob Faibussowitsch 44e94e781bSJacob Faibussowitsch .seealso: PetscInfo(), PetscInfoSetFromOptions(), PetscInfoSetFilterCommSelf() 45e94e781bSJacob Faibussowitsch E*/ 46e94e781bSJacob Faibussowitsch typedef enum { 47e94e781bSJacob Faibussowitsch PETSC_INFO_COMM_ALL = -1, 48e94e781bSJacob Faibussowitsch PETSC_INFO_COMM_NO_SELF = 0, 49e94e781bSJacob Faibussowitsch PETSC_INFO_COMM_ONLY_SELF = 1 50e94e781bSJacob Faibussowitsch } PetscInfoCommFlag; 51e94e781bSJacob Faibussowitsch 52e94e781bSJacob Faibussowitsch PETSC_EXTERN const char * const PetscInfoCommFlags[]; 53fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscInfoDeactivateClass(PetscClassId); 54fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscInfoActivateClass(PetscClassId); 55e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoEnabled(PetscClassId, PetscBool *); 56e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoAllow(PetscBool); 57e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoSetFile(const char[],const char[]); 58e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoGetFile(char **,FILE **); 59e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoSetClasses(PetscBool,PetscInt,const char *const *); 60e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoGetClass(const char *, PetscBool *); 61e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoGetInfo(PetscBool *,PetscBool *,PetscBool *,PetscBool *,PetscInfoCommFlag *); 62e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoProcessClass(const char[],PetscInt,PetscClassId[]); 63e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoSetFilterCommSelf(PetscInfoCommFlag); 64e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoSetFromOptions(PetscOptions); 65e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoDestroy(void); 66fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscBool PetscLogPrintInfo; /* if true, indicates PetscInfo() is turned on */ 67fa2bb9feSLisandro Dalcin 68b859824cSBarry Smith /*MC 69b859824cSBarry Smith PetscLogEvent - id used to identify PETSc or user events which timed portions (blocks of executable) 70b859824cSBarry Smith code. 71b859824cSBarry Smith 72b859824cSBarry Smith Level: intermediate 73b859824cSBarry Smith 74b859824cSBarry Smith .seealso: PetscLogEventRegister(), PetscLogEventBegin(), PetscLogEventEnd(), PetscLogStage 75b859824cSBarry Smith M*/ 76b859824cSBarry Smith typedef int PetscLogEvent; 77b859824cSBarry Smith 78b859824cSBarry Smith /*MC 79b859824cSBarry Smith PetscLogStage - id used to identify user stages (phases, sections) of runs - for logging 80b859824cSBarry Smith 81b859824cSBarry Smith Level: intermediate 82b859824cSBarry Smith 83522ce795SJed Brown .seealso: PetscLogStageRegister(), PetscLogStagePush(), PetscLogStagePop(), PetscLogEvent 84b859824cSBarry Smith M*/ 85b859824cSBarry Smith typedef int PetscLogStage; 86b859824cSBarry Smith 878ba1e511SMatthew Knepley #define PETSC_EVENT 1311311 88014dd563SJed Brown PETSC_EXTERN PetscLogEvent PETSC_LARGEST_EVENT; 8997bb86f7SLois Curfman McInnes 9019b02663SBarry Smith /* Global flop counter */ 91014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_TotalFlops; 92014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_tmp_flops; 9319b02663SBarry Smith 94bec0b493Shannah_mairs /* Global GPU counters */ 95bec0b493Shannah_mairs #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA) 96bec0b493Shannah_mairs PETSC_EXTERN PetscLogDouble petsc_ctog_ct; 97bec0b493Shannah_mairs PETSC_EXTERN PetscLogDouble petsc_gtoc_ct; 98bec0b493Shannah_mairs PETSC_EXTERN PetscLogDouble petsc_ctog_sz; 99bec0b493Shannah_mairs PETSC_EXTERN PetscLogDouble petsc_gtoc_sz; 100958c4211Shannah_mairs PETSC_EXTERN PetscLogDouble petsc_gflops; 101958c4211Shannah_mairs PETSC_EXTERN PetscLogDouble petsc_gtime; 102bec0b493Shannah_mairs #endif 103bec0b493Shannah_mairs 10431d06abdSBarry Smith /* We must make the following structures available to access the event 10531d06abdSBarry Smith activation flags in the PetscLogEventBegin/End() macros. These are not part of the PETSc public 10631d06abdSBarry Smith API and are not intended to be used by other parts of PETSc or by users. 107d49d4b11SBarry Smith 108fa2bb9feSLisandro Dalcin The code that manipulates these structures is in src/sys/logging/utils. 10931d06abdSBarry Smith */ 110aa213bdcSJed Brown typedef struct _n_PetscIntStack *PetscIntStack; 1116a6a9b46SSatish Balay 112217044c2SLisandro Dalcin /* -----------------------------------------------------------------------------------------------------*/ 113c60ec953SBarry Smith /* 114aa213bdcSJed Brown PetscClassRegInfo, PetscClassPerfInfo - Each class has two data structures associated with it. The first has 115c60ec953SBarry Smith static information about it, the second collects statistics on how many objects of the class are created, 116c60ec953SBarry Smith how much memory they use, etc. 117c60ec953SBarry Smith 118aa213bdcSJed Brown PetscClassRegLog, PetscClassPerfLog - arrays of the PetscClassRegInfo and PetscClassPerfInfo for all classes. 119c60ec953SBarry Smith */ 1206a6a9b46SSatish Balay typedef struct { 121c60ec953SBarry Smith char *name; /* The class name */ 122c60ec953SBarry Smith PetscClassId classid; /* The integer identifying this class */ 123aa213bdcSJed Brown } PetscClassRegInfo; 1246a6a9b46SSatish Balay 1256a6a9b46SSatish Balay typedef struct { 1260700a824SBarry Smith PetscClassId id; /* The integer identifying this class */ 1276a6a9b46SSatish Balay int creations; /* The number of objects of this class created */ 1286a6a9b46SSatish Balay int destructions; /* The number of objects of this class destroyed */ 1296a6a9b46SSatish Balay PetscLogDouble mem; /* The total memory allocated by objects of this class */ 1306a6a9b46SSatish Balay PetscLogDouble descMem; /* The total memory allocated by descendents of these objects */ 131aa213bdcSJed Brown } PetscClassPerfInfo; 1326a6a9b46SSatish Balay 133aa213bdcSJed Brown typedef struct _n_PetscClassRegLog *PetscClassRegLog; 134aa213bdcSJed Brown struct _n_PetscClassRegLog { 135c60ec953SBarry Smith int numClasses; /* The number of classes registered */ 136c60ec953SBarry Smith int maxClasses; /* The maximum number of classes */ 137aa213bdcSJed Brown PetscClassRegInfo *classInfo; /* The structure for class information (classids are monotonicly increasing) */ 138c60ec953SBarry Smith }; 1396a6a9b46SSatish Balay 140aa213bdcSJed Brown typedef struct _n_PetscClassPerfLog *PetscClassPerfLog; 141aa213bdcSJed Brown struct _n_PetscClassPerfLog { 142c60ec953SBarry Smith int numClasses; /* The number of logging classes */ 143c60ec953SBarry Smith int maxClasses; /* The maximum number of classes */ 144aa213bdcSJed Brown PetscClassPerfInfo *classInfo; /* The structure for class information (classids are monotonicly increasing) */ 145c60ec953SBarry Smith }; 146c60ec953SBarry Smith /* -----------------------------------------------------------------------------------------------------*/ 147c60ec953SBarry Smith /* 148aa213bdcSJed Brown PetscEventRegInfo, PetscEventPerfInfo - Each event has two data structures associated with it. The first has 149c60ec953SBarry Smith static information about it, the second collects statistics on how many times the event is used, how 150c60ec953SBarry Smith much time it takes, etc. 151c60ec953SBarry Smith 152aa213bdcSJed Brown PetscEventRegLog, PetscEventPerfLog - an array of all PetscEventRegInfo and PetscEventPerfInfo for all events. There is one 153c60ec953SBarry Smith of these for each stage. 154c60ec953SBarry Smith 155c60ec953SBarry Smith */ 1566a6a9b46SSatish Balay typedef struct { 1576a6a9b46SSatish Balay char *name; /* The name of this event */ 158c60ec953SBarry Smith PetscClassId classid; /* The class the event is associated with */ 159217044c2SLisandro Dalcin PetscBool collective; /* Flag this event as collective */ 1606a6a9b46SSatish Balay #if defined (PETSC_HAVE_MPE) 1616a6a9b46SSatish Balay int mpe_id_begin; /* MPE IDs that define the event */ 1626a6a9b46SSatish Balay int mpe_id_end; 1636a6a9b46SSatish Balay #endif 164aa213bdcSJed Brown } PetscEventRegInfo; 1656a6a9b46SSatish Balay 166c60ec953SBarry Smith typedef struct { 167c60ec953SBarry Smith int id; /* The integer identifying this event */ 168ace3abfcSBarry Smith PetscBool active; /* The flag to activate logging */ 169ace3abfcSBarry Smith PetscBool visible; /* The flag to print info in summary */ 170c60ec953SBarry Smith int depth; /* The nesting depth of the event call */ 171c60ec953SBarry Smith int count; /* The number of times this event was executed */ 172a42e5b54SMatthew G Knepley PetscLogDouble flops, flops2, flopsTmp; /* The flops and flops^2 used in this event */ 173a42e5b54SMatthew G Knepley PetscLogDouble time, time2, timeTmp; /* The time and time^2 taken for this event */ 174217044c2SLisandro Dalcin PetscLogDouble syncTime; /* The synchronization barrier time */ 175891e75beSMatthew G. Knepley PetscLogDouble dof[8]; /* The number of degrees of freedom associated with this event */ 1765d68e14cSMatthew G. Knepley PetscLogDouble errors[8]; /* The errors (user-defined) associated with this event */ 177c60ec953SBarry Smith PetscLogDouble numMessages; /* The number of messages in this event */ 178c60ec953SBarry Smith PetscLogDouble messageLength; /* The total message lengths in this event */ 179c60ec953SBarry Smith PetscLogDouble numReductions; /* The number of reductions in this event */ 180e3ed9ee7SBarry Smith PetscLogDouble memIncrease; /* How much the resident memory has increased in this event */ 181e3ed9ee7SBarry Smith PetscLogDouble mallocIncrease;/* How much the maximum malloced space has increased in this event */ 182e3ed9ee7SBarry Smith PetscLogDouble mallocSpace; /* How much the space was malloced and kept during this event */ 183e3ed9ee7SBarry Smith PetscLogDouble mallocIncreaseEvent; /* Maximum of the high water mark with in event minus memory available at the end of the event */ 184bec0b493Shannah_mairs #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA) 185bec0b493Shannah_mairs PetscLogDouble CpuToGpuCount; /* The total number of CPU to GPU copies */ 186bec0b493Shannah_mairs PetscLogDouble GpuToCpuCount; /* The total number of GPU to CPU copies */ 187bec0b493Shannah_mairs PetscLogDouble CpuToGpuSize; /* The total size of CPU to GPU copies */ 188bec0b493Shannah_mairs PetscLogDouble GpuToCpuSize; /* The total size of GPU to CPU copies */ 189958c4211Shannah_mairs PetscLogDouble GpuFlops; /* The flops done on a GPU in this event */ 190958c4211Shannah_mairs PetscLogDouble GpuTime; /* The time spent on a GPU in this event */ 191bec0b493Shannah_mairs #endif 192aa213bdcSJed Brown } PetscEventPerfInfo; 193c60ec953SBarry Smith 194aa213bdcSJed Brown typedef struct _n_PetscEventRegLog *PetscEventRegLog; 195aa213bdcSJed Brown struct _n_PetscEventRegLog { 1966a6a9b46SSatish Balay int numEvents; /* The number of registered events */ 1976a6a9b46SSatish Balay int maxEvents; /* The maximum number of events */ 198aa213bdcSJed Brown PetscEventRegInfo *eventInfo; /* The registration information for each event */ 1996a6a9b46SSatish Balay }; 2006a6a9b46SSatish Balay 201aa213bdcSJed Brown typedef struct _n_PetscEventPerfLog *PetscEventPerfLog; 202aa213bdcSJed Brown struct _n_PetscEventPerfLog { 2036a6a9b46SSatish Balay int numEvents; /* The number of logging events */ 2046a6a9b46SSatish Balay int maxEvents; /* The maximum number of events */ 205aa213bdcSJed Brown PetscEventPerfInfo *eventInfo; /* The performance information for each event */ 2066a6a9b46SSatish Balay }; 207c60ec953SBarry Smith /* ------------------------------------------------------------------------------------------------------------*/ 208c60ec953SBarry Smith /* 209aa213bdcSJed Brown PetscStageInfo - Contains all the information about a particular stage. 2106a6a9b46SSatish Balay 211aa213bdcSJed Brown PetscStageLog - An array of PetscStageInfo for each registered stage. There is a single one of these in the code. 212c60ec953SBarry Smith */ 213aa213bdcSJed Brown typedef struct _PetscStageInfo { 2146a6a9b46SSatish Balay char *name; /* The stage name */ 215ace3abfcSBarry Smith PetscBool used; /* The stage was pushed on this processor */ 216aa213bdcSJed Brown PetscEventPerfInfo perfInfo; /* The stage performance information */ 217aa213bdcSJed Brown PetscEventPerfLog eventLog; /* The event information for this stage */ 218aa213bdcSJed Brown PetscClassPerfLog classLog; /* The class information for this stage */ 219aa213bdcSJed Brown } PetscStageInfo; 2206a6a9b46SSatish Balay 221aa213bdcSJed Brown typedef struct _n_PetscStageLog *PetscStageLog; 222aa213bdcSJed Brown struct _n_PetscStageLog { 2236a6a9b46SSatish Balay int numStages; /* The number of registered stages */ 2246a6a9b46SSatish Balay int maxStages; /* The maximum number of stages */ 225aa213bdcSJed Brown PetscIntStack stack; /* The stack for active stages */ 226aa213bdcSJed Brown int curStage; /* The current stage (only used in macros so we don't call PetscIntStackTop) */ 227aa213bdcSJed Brown PetscStageInfo *stageInfo; /* The information for each stage */ 228aa213bdcSJed Brown PetscEventRegLog eventLog; /* The registered events */ 229aa213bdcSJed Brown PetscClassRegLog classLog; /* The registered classes */ 2306a6a9b46SSatish Balay }; 231217044c2SLisandro Dalcin /* -----------------------------------------------------------------------------------------------------*/ 232fee518ebSJed Brown 2333bb1ff40SBarry Smith PETSC_EXTERN PetscErrorCode PetscLogObjectParent(PetscObject,PetscObject); 2343bb1ff40SBarry Smith PETSC_EXTERN PetscErrorCode PetscLogObjectMemory(PetscObject,PetscLogDouble); 2353bb1ff40SBarry Smith 236aa482453SBarry Smith #if defined(PETSC_USE_LOG) /* --- Logging is turned on --------------------------------*/ 237aeeda9f6SJohn Fettig PETSC_EXTERN PetscStageLog petsc_stageLog; 238f5d6ab90SLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogGetStageLog(PetscStageLog*); 239f5d6ab90SLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscStageLogGetCurrent(PetscStageLog,int*); 240f5d6ab90SLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscStageLogGetEventPerfLog(PetscStageLog,int,PetscEventPerfLog*); 241614700edSBarry Smith 242da63de55SLois Curfman McInnes /* 243da63de55SLois Curfman McInnes Flop counting: We count each arithmetic operation (e.g., addition, multiplication) separately. 244da63de55SLois Curfman McInnes 245da63de55SLois Curfman McInnes For the complex numbers version, note that 246da63de55SLois Curfman McInnes 1 complex addition = 2 flops 247da63de55SLois Curfman McInnes 1 complex multiplication = 6 flops, 248da63de55SLois Curfman McInnes where we define 1 flop as that for a double precision scalar. We roughly approximate 249da63de55SLois Curfman McInnes flop counting for complex numbers by multiplying the total flops by 4; this corresponds 250da63de55SLois Curfman McInnes to the assumption that we're counting mostly additions and multiplications -- and 251da63de55SLois Curfman McInnes roughly the same number of each. More accurate counting could be done by distinguishing 252da63de55SLois Curfman McInnes among the various arithmetic operations. 253da63de55SLois Curfman McInnes */ 254da63de55SLois Curfman McInnes 255aa482453SBarry Smith #if defined(PETSC_USE_COMPLEX) 256542d4b3fSSatish Balay #define PETSC_FLOPS_PER_OP 4.0 257da63de55SLois Curfman McInnes #else 258542d4b3fSSatish Balay #define PETSC_FLOPS_PER_OP 1.0 259bf3909cdSBarry Smith #endif 260bf3909cdSBarry Smith 261dd19b90bSBarry Smith PETSC_STATIC_INLINE PetscErrorCode PetscLogFlops(PetscLogDouble n) 262dd19b90bSBarry Smith { 263dd19b90bSBarry Smith PetscFunctionBegin; 264bf3909cdSBarry Smith #if defined(PETSC_USE_DEBUG) 265dd19b90bSBarry Smith if (n < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Cannot log negative flops"); 266da63de55SLois Curfman McInnes #endif 267dd19b90bSBarry Smith petsc_TotalFlops += PETSC_FLOPS_PER_OP*n; 268dd19b90bSBarry Smith PetscFunctionReturn(0); 269dd19b90bSBarry Smith } 270fa2bb9feSLisandro Dalcin 271bec0b493Shannah_mairs #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA) 2724db21f4bSStefano Zampini PETSC_STATIC_INLINE PetscErrorCode PetscLogCpuToGpu(PetscLogDouble size) 2734db21f4bSStefano Zampini { 274958c4211Shannah_mairs PetscFunctionBegin; 275bec0b493Shannah_mairs petsc_ctog_ct += 1; 276bec0b493Shannah_mairs petsc_ctog_sz += size; 277bec0b493Shannah_mairs PetscFunctionReturn(0); 278bec0b493Shannah_mairs } 2794db21f4bSStefano Zampini 2804db21f4bSStefano Zampini PETSC_STATIC_INLINE PetscErrorCode PetscLogGpuToCpu(PetscLogDouble size) 2814db21f4bSStefano Zampini { 282958c4211Shannah_mairs PetscFunctionBegin; 283bec0b493Shannah_mairs petsc_gtoc_ct += 1; 284bec0b493Shannah_mairs petsc_gtoc_sz += size; 285bec0b493Shannah_mairs PetscFunctionReturn(0); 286bec0b493Shannah_mairs } 2874db21f4bSStefano Zampini 2884db21f4bSStefano Zampini PETSC_STATIC_INLINE PetscErrorCode PetscLogGpuFlops(PetscLogDouble n) 2894db21f4bSStefano Zampini { 290958c4211Shannah_mairs PetscFunctionBegin; 291958c4211Shannah_mairs #if defined(PETSC_USE_DEBUG) 292958c4211Shannah_mairs if (n < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Cannot log negative flops"); 293958c4211Shannah_mairs #endif 294958c4211Shannah_mairs petsc_TotalFlops += PETSC_FLOPS_PER_OP*n; 295958c4211Shannah_mairs petsc_gflops += PETSC_FLOPS_PER_OP*n; 296958c4211Shannah_mairs PetscFunctionReturn(0); 297958c4211Shannah_mairs } 2984db21f4bSStefano Zampini 299*703ce11fSStefano Zampini #if defined(PETSC_USE_DEBUG) 300*703ce11fSStefano Zampini static PetscBool petsc_gtime_inuse = PETSC_FALSE; 301*703ce11fSStefano Zampini #endif 302*703ce11fSStefano Zampini 3034db21f4bSStefano Zampini PETSC_STATIC_INLINE PetscErrorCode PetscLogGpuTimeBegin() 3044db21f4bSStefano Zampini { 305958c4211Shannah_mairs PetscErrorCode ierr; 306958c4211Shannah_mairs PetscFunctionBegin; 307*703ce11fSStefano Zampini #if defined(PETSC_USE_DEBUG) 308*703ce11fSStefano Zampini if (petsc_gtime_inuse) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Forgot to call PetscLogGpuTimeEnd()?"); 309*703ce11fSStefano Zampini petsc_gtime_inuse = PETSC_TRUE; 310*703ce11fSStefano Zampini #endif 311958c4211Shannah_mairs ierr = PetscTimeSubtract(&petsc_gtime);CHKERRQ(ierr); 312958c4211Shannah_mairs PetscFunctionReturn(0); 313958c4211Shannah_mairs } 3144db21f4bSStefano Zampini 3154db21f4bSStefano Zampini PETSC_STATIC_INLINE PetscErrorCode PetscLogGpuTimeEnd() 3164db21f4bSStefano Zampini { 317958c4211Shannah_mairs PetscErrorCode ierr; 318958c4211Shannah_mairs PetscFunctionBegin; 319*703ce11fSStefano Zampini #if defined(PETSC_USE_DEBUG) 320*703ce11fSStefano Zampini if (!petsc_gtime_inuse) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Forgot to call PetscLogGpuTimeBegin()?"); 321*703ce11fSStefano Zampini petsc_gtime_inuse = PETSC_FALSE; 322*703ce11fSStefano Zampini #endif 323958c4211Shannah_mairs ierr = PetscTimeAdd(&petsc_gtime);CHKERRQ(ierr); 324958c4211Shannah_mairs PetscFunctionReturn(0); 325958c4211Shannah_mairs } 3264db21f4bSStefano Zampini 3274db21f4bSStefano Zampini PETSC_STATIC_INLINE PetscErrorCode PetscLogGpuTimeAdd(PetscLogDouble t) 3284db21f4bSStefano Zampini { 3294db21f4bSStefano Zampini PetscFunctionBegin; 3304db21f4bSStefano Zampini petsc_gtime += t; 3314db21f4bSStefano Zampini PetscFunctionReturn(0); 3324db21f4bSStefano Zampini } 333bec0b493Shannah_mairs #endif 334bec0b493Shannah_mairs 335fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscGetFlops(PetscLogDouble *); 33677c4ece6SBarry Smith 337aa482453SBarry Smith #if defined (PETSC_HAVE_MPE) 338014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogMPEBegin(void); 339014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogMPEDump(const char[]); 34077c4ece6SBarry Smith #endif 34177c4ece6SBarry Smith 342014dd563SJed Brown PETSC_EXTERN PetscErrorCode (*PetscLogPLB)(PetscLogEvent,int,PetscObject,PetscObject,PetscObject,PetscObject); 343014dd563SJed Brown PETSC_EXTERN PetscErrorCode (*PetscLogPLE)(PetscLogEvent,int,PetscObject,PetscObject,PetscObject,PetscObject); 344014dd563SJed Brown PETSC_EXTERN PetscErrorCode (*PetscLogPHC)(PetscObject); 345014dd563SJed Brown PETSC_EXTERN PetscErrorCode (*PetscLogPHD)(PetscObject); 34677c4ece6SBarry Smith 347e83a5d19SLisandro Dalcin #define PetscLogObjectParents(p,n,d) 0;do{int _i; for (_i=0; _i<(n); _i++) {ierr = PetscLogObjectParent((PetscObject)(p),(PetscObject)(d)[_i]);CHKERRQ(ierr);}}while(0) 348fa2bb9feSLisandro Dalcin #define PetscLogObjectCreate(h) ((PetscLogPHC) ? (*PetscLogPHC)((PetscObject)(h)) : 0) 349fa2bb9feSLisandro Dalcin #define PetscLogObjectDestroy(h) ((PetscLogPHD) ? (*PetscLogPHD)((PetscObject)(h)) : 0) 350fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogObjectState(PetscObject, const char[], ...); 351fa2bb9feSLisandro Dalcin 3528ba1e511SMatthew Knepley /* Initialization functions */ 353bb1d7374SBarry Smith PETSC_EXTERN PetscErrorCode PetscLogDefaultBegin(void); 354014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogAllBegin(void); 355bb1d7374SBarry Smith PETSC_EXTERN PetscErrorCode PetscLogNestedBegin(void); 356014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogTraceBegin(FILE *); 357014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogActions(PetscBool); 358014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogObjects(PetscBool); 359fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogSetThreshold(PetscLogDouble,PetscLogDouble*); 360014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogSet(PetscErrorCode (*)(int, int, PetscObject, PetscObject, PetscObject, PetscObject), 3616849ba73SBarry Smith PetscErrorCode (*)(int, int, PetscObject, PetscObject, PetscObject, PetscObject)); 362fa2bb9feSLisandro Dalcin 3638ba1e511SMatthew Knepley /* Output functions */ 364014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogView(PetscViewer); 365f14045dbSBarry Smith PETSC_EXTERN PetscErrorCode PetscLogViewFromOptions(void); 366014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogDump(const char[]); 36731d06abdSBarry Smith 368fa2bb9feSLisandro Dalcin /* Stage functions */ 369014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStageRegister(const char[],PetscLogStage*); 370014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStagePush(PetscLogStage); 371014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStagePop(void); 372014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStageSetActive(PetscLogStage,PetscBool); 373014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStageGetActive(PetscLogStage,PetscBool*); 374014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStageSetVisible(PetscLogStage,PetscBool); 375014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStageGetVisible(PetscLogStage,PetscBool*); 376014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStageGetId(const char[],PetscLogStage*); 377f5d6ab90SLisandro Dalcin 3788ba1e511SMatthew Knepley /* Event functions */ 379014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventRegister(const char[],PetscClassId,PetscLogEvent*); 380217044c2SLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogEventSetCollective(PetscLogEvent,PetscBool); 381fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogEventIncludeClass(PetscClassId); 382fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogEventExcludeClass(PetscClassId); 383014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventActivate(PetscLogEvent); 384014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventDeactivate(PetscLogEvent); 385014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventSetActiveAll(PetscLogEvent,PetscBool); 386014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventActivateClass(PetscClassId); 387014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventDeactivateClass(PetscClassId); 3888c49a8d9SBarry Smith PETSC_EXTERN PetscErrorCode PetscLogEventGetId(const char[],PetscLogEvent*); 38913230571SMichael Lange PETSC_EXTERN PetscErrorCode PetscLogEventGetPerfInfo(int,PetscLogEvent,PetscEventPerfInfo*); 390891e75beSMatthew G. Knepley PETSC_EXTERN PetscErrorCode PetscLogEventSetDof(PetscLogEvent, PetscInt, PetscLogDouble); 3915d68e14cSMatthew G. Knepley PETSC_EXTERN PetscErrorCode PetscLogEventSetError(PetscLogEvent, PetscInt, PetscLogDouble); 3928ba1e511SMatthew Knepley 3938ba1e511SMatthew Knepley /* Global counters */ 394014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_irecv_ct; 395014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_isend_ct; 396014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_recv_ct; 397014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_send_ct; 398014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_irecv_len; 399014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_isend_len; 400014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_recv_len; 401014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_send_len; 402014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_allreduce_ct; 403014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_gather_ct; 404014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_scatter_ct; 405014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_wait_ct; 406014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_wait_any_ct; 407014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_wait_all_ct; 408014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_sum_of_waits_ct; 4098ba1e511SMatthew Knepley 410e3ed9ee7SBarry Smith PETSC_EXTERN PetscBool PetscLogMemory; 411e3ed9ee7SBarry Smith 412fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscBool PetscLogSyncOn; /* true if logging synchronization is enabled */ 41362872c28SLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogEventSynchronize(PetscLogEvent, MPI_Comm); 41462872c28SLisandro Dalcin 41562872c28SLisandro Dalcin #define PetscLogEventSync(e,comm) \ 41662872c28SLisandro Dalcin (((PetscLogPLB && petsc_stageLog->stageInfo[petsc_stageLog->curStage].perfInfo.active && petsc_stageLog->stageInfo[petsc_stageLog->curStage].eventLog->eventInfo[e].active) ? \ 41762872c28SLisandro Dalcin PetscLogEventSynchronize((e),(comm)) : 0 )) 41877c4ece6SBarry Smith 41952e6d16bSBarry Smith #define PetscLogEventBegin(e,o1,o2,o3,o4) \ 4208b5db460SBarry Smith (((PetscLogPLB && petsc_stageLog->stageInfo[petsc_stageLog->curStage].perfInfo.active && petsc_stageLog->stageInfo[petsc_stageLog->curStage].eventLog->eventInfo[e].active) ? \ 421495fc317SBarry Smith (*PetscLogPLB)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4)) : 0 )) 422043328b6SSatish Balay 42352e6d16bSBarry Smith #define PetscLogEventEnd(e,o1,o2,o3,o4) \ 4248b5db460SBarry Smith (((PetscLogPLE && petsc_stageLog->stageInfo[petsc_stageLog->curStage].perfInfo.active && petsc_stageLog->stageInfo[petsc_stageLog->curStage].eventLog->eventInfo[e].active) ? \ 425495fc317SBarry Smith (*PetscLogPLE)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4)) : 0 )) 42677c4ece6SBarry Smith 427014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventGetFlops(PetscLogEvent,PetscLogDouble*); 428014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventZeroFlops(PetscLogEvent); 429003131ecSBarry Smith 430ce85283eSBarry Smith /* 431f621e05eSBarry Smith These are used internally in the PETSc routines to keep a count of MPI messages and 432f621e05eSBarry Smith their sizes. 433f621e05eSBarry Smith 434af0996ceSBarry Smith This does not work for MPI-Uni because our include/petsc/mpiuni/mpi.h file 435ce85283eSBarry Smith uses macros to defined the MPI operations. 43615308354SBarry Smith 43715308354SBarry Smith It does not work correctly from HP-UX because it processes the 438bb4af37aSBarry Smith macros in a way that sometimes it double counts, hence 439b6410449SSatish Balay PETSC_HAVE_BROKEN_RECURSIVE_MACRO 4407c1e34a4SSatish Balay 441f569fd43SBarry Smith It does not work with Windows because winmpich lacks MPI_Type_size() 442ce85283eSBarry Smith */ 443c9b973beSBarry Smith #if !defined(MPIUNI_H) && !defined(PETSC_HAVE_BROKEN_RECURSIVE_MACRO) && !defined (PETSC_HAVE_MPI_MISSING_TYPESIZE) 44477a39924SBarry Smith /* 44577a39924SBarry Smith Logging of MPI activities 44677a39924SBarry Smith */ 447b23bfdefSJunchao Zhang PETSC_STATIC_INLINE PetscErrorCode PetscMPITypeSize(PetscInt count,MPI_Datatype type,PetscLogDouble *length) 448f95db71bSBarry Smith { 449f12d10ccSJunchao Zhang PetscMPIInt typesize; 450f12d10ccSJunchao Zhang PetscErrorCode ierr; 4517d30fa4aSBarry Smith if (type == MPI_DATATYPE_NULL) return 0; 452f12d10ccSJunchao Zhang ierr = MPI_Type_size(type,&typesize);CHKERRQ(ierr); 453f12d10ccSJunchao Zhang *length += (PetscLogDouble) (count*typesize); 4549c72ff1eSSatish Balay return 0; 455f95db71bSBarry Smith } 45677a39924SBarry Smith 457f12d10ccSJunchao Zhang PETSC_STATIC_INLINE PetscErrorCode PetscMPITypeSizeComm(MPI_Comm comm,const PetscMPIInt *counts,MPI_Datatype type,PetscLogDouble *length) 4583b9284c0SMatthew G Knepley { 459f12d10ccSJunchao Zhang PetscMPIInt typesize,size,p; 460f12d10ccSJunchao Zhang PetscErrorCode ierr; 4613b9284c0SMatthew G Knepley 4627d30fa4aSBarry Smith if (type == MPI_DATATYPE_NULL) return 0; 463f12d10ccSJunchao Zhang ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 464f12d10ccSJunchao Zhang ierr = MPI_Type_size(type,&typesize);CHKERRQ(ierr); 465f12d10ccSJunchao Zhang for (p=0; p<size; ++p) { 466f12d10ccSJunchao Zhang *length += (PetscLogDouble) (counts[p]*typesize); 467f12d10ccSJunchao Zhang } 468f12d10ccSJunchao Zhang return 0; 469f12d10ccSJunchao Zhang } 470f12d10ccSJunchao Zhang 471f12d10ccSJunchao Zhang PETSC_STATIC_INLINE PetscErrorCode PetscMPITypeSizeCount(PetscInt n,const PetscMPIInt *counts,MPI_Datatype type,PetscLogDouble *length) 472f12d10ccSJunchao Zhang { 473f12d10ccSJunchao Zhang PetscMPIInt typesize,p; 474f12d10ccSJunchao Zhang PetscErrorCode ierr; 475f12d10ccSJunchao Zhang 476f12d10ccSJunchao Zhang if (type == MPI_DATATYPE_NULL) return 0; 477f12d10ccSJunchao Zhang ierr = MPI_Type_size(type,&typesize);CHKERRQ(ierr); 478f12d10ccSJunchao Zhang for (p=0; p<n; ++p) { 479f12d10ccSJunchao Zhang *length += (PetscLogDouble) (counts[p]*typesize); 4803b9284c0SMatthew G Knepley } 4813b9284c0SMatthew G Knepley return 0; 4823b9284c0SMatthew G Knepley } 4833b9284c0SMatthew G Knepley 48472b11a25SBarry Smith /* 48572b11a25SBarry Smith Returns 1 if the communicator is parallel else zero 48672b11a25SBarry Smith */ 48772b11a25SBarry Smith PETSC_STATIC_INLINE int PetscMPIParallelComm(MPI_Comm comm) 48872b11a25SBarry Smith { 48972b11a25SBarry Smith PetscMPIInt size; MPI_Comm_size(comm,&size); return size > 1; 49072b11a25SBarry Smith } 49172b11a25SBarry Smith 49277a39924SBarry Smith #define MPI_Irecv(buf,count,datatype,source,tag,comm,request) \ 493f12d10ccSJunchao Zhang ((petsc_irecv_ct++,0) || PetscMPITypeSize((count),(datatype),&(petsc_irecv_len)) || MPI_Irecv((buf),(count),(datatype),(source),(tag),(comm),(request))) 49415308354SBarry Smith 49577a39924SBarry Smith #define MPI_Isend(buf,count,datatype,dest,tag,comm,request) \ 496f12d10ccSJunchao Zhang ((petsc_isend_ct++,0) || PetscMPITypeSize((count),(datatype),&(petsc_isend_len)) || MPI_Isend((buf),(count),(datatype),(dest),(tag),(comm),(request))) 49715308354SBarry Smith 498457b14b1SJunchao Zhang #define MPI_Startall_irecv(count,datatype,number,requests) \ 499f12d10ccSJunchao Zhang ((petsc_irecv_ct += (PetscLogDouble)(number),0) || PetscMPITypeSize((count),(datatype),&(petsc_irecv_len)) || ((number) && MPI_Startall((number),(requests)))) 5000d4b0b6cSBarry Smith 501457b14b1SJunchao Zhang #define MPI_Startall_isend(count,datatype,number,requests) \ 502f12d10ccSJunchao Zhang ((petsc_isend_ct += (PetscLogDouble)(number),0) || PetscMPITypeSize((count),(datatype),&(petsc_isend_len)) || ((number) && MPI_Startall((number),(requests)))) 5030d4b0b6cSBarry Smith 504457b14b1SJunchao Zhang #define MPI_Start_isend(count,datatype,requests) \ 505f12d10ccSJunchao Zhang ((petsc_isend_ct++,0) || PetscMPITypeSize((count),(datatype),(&petsc_isend_len)) || MPI_Start((requests))) 5060d4b0b6cSBarry Smith 507ce85283eSBarry Smith #define MPI_Recv(buf,count,datatype,source,tag,comm,status) \ 508f12d10ccSJunchao Zhang ((petsc_recv_ct++,0) || PetscMPITypeSize((count),(datatype),(&petsc_recv_len)) || MPI_Recv((buf),(count),(datatype),(source),(tag),(comm),(status))) 50915308354SBarry Smith 51077a39924SBarry Smith #define MPI_Send(buf,count,datatype,dest,tag,comm) \ 511f12d10ccSJunchao Zhang ((petsc_send_ct++,0) || PetscMPITypeSize((count),(datatype),(&petsc_send_len)) || MPI_Send((buf),(count),(datatype),(dest),(tag),(comm))) 51277a39924SBarry Smith 51377a39924SBarry Smith #define MPI_Wait(request,status) \ 5145fa69175SJed Brown ((petsc_wait_ct++,petsc_sum_of_waits_ct++,0) || MPI_Wait((request),(status))) 51577a39924SBarry Smith 51677a39924SBarry Smith #define MPI_Waitany(a,b,c,d) \ 5175fa69175SJed Brown ((petsc_wait_any_ct++,petsc_sum_of_waits_ct++,0) || MPI_Waitany((a),(b),(c),(d))) 51877a39924SBarry Smith 51977a39924SBarry Smith #define MPI_Waitall(count,array_of_requests,array_of_statuses) \ 5205fa69175SJed Brown ((petsc_wait_all_ct++,petsc_sum_of_waits_ct += (PetscLogDouble) (count),0) || MPI_Waitall((count),(array_of_requests),(array_of_statuses))) 52177a39924SBarry Smith 52277a39924SBarry Smith #define MPI_Allreduce(sendbuf,recvbuf,count,datatype,op,comm) \ 5235fa69175SJed Brown ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || MPI_Allreduce((sendbuf),(recvbuf),(count),(datatype),(op),(comm))) 5243914022bSBarry Smith 52554fe5c21SBarry Smith #define MPI_Bcast(buffer,count,datatype,root,comm) \ 5265fa69175SJed Brown ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || MPI_Bcast((buffer),(count),(datatype),(root),(comm))) 52754fe5c21SBarry Smith 528d6e4c47cSJed Brown #define MPI_Reduce_scatter_block(sendbuf,recvbuf,recvcount,datatype,op,comm) \ 5295fa69175SJed Brown ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || MPI_Reduce_scatter_block((sendbuf),(recvbuf),(recvcount),(datatype),(op),(comm))) 530d6e4c47cSJed Brown 5313b9284c0SMatthew G Knepley #define MPI_Alltoall(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,comm) \ 532f12d10ccSJunchao Zhang ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Alltoall((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(comm))) 5333b9284c0SMatthew G Knepley 5343b9284c0SMatthew G Knepley #define MPI_Alltoallv(sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm) \ 535f12d10ccSJunchao Zhang ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || PetscMPITypeSizeComm((comm),(sendcnts),(sendtype),(&petsc_send_len)) || MPI_Alltoallv((sendbuf),(sendcnts),(sdispls),(sendtype),(recvbuf),(recvcnts),(rdispls),(recvtype),(comm))) 5363b9284c0SMatthew G Knepley 53701faf4e4SMatthew Knepley #define MPI_Allgather(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,comm) \ 5385fa69175SJed Brown ((petsc_gather_ct += PetscMPIParallelComm((comm)),0) || MPI_Allgather((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(comm))) 53901faf4e4SMatthew Knepley 54001faf4e4SMatthew Knepley #define MPI_Allgatherv(sendbuf,sendcount,sendtype,recvbuf,recvcount,displs,recvtype,comm) \ 5415fa69175SJed Brown ((petsc_gather_ct += PetscMPIParallelComm((comm)),0) || MPI_Allgatherv((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(displs),(recvtype),(comm))) 54201faf4e4SMatthew Knepley 54301faf4e4SMatthew Knepley #define MPI_Gather(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,root,comm) \ 544f12d10ccSJunchao Zhang ((petsc_gather_ct++,0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Gather((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm))) 54501faf4e4SMatthew Knepley 54601faf4e4SMatthew Knepley #define MPI_Gatherv(sendbuf,sendcount,sendtype,recvbuf,recvcount,displs,recvtype,root,comm) \ 547f12d10ccSJunchao Zhang ((petsc_gather_ct++,0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Gatherv((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(displs),(recvtype),(root),(comm))) 54801faf4e4SMatthew Knepley 54901faf4e4SMatthew Knepley #define MPI_Scatter(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,root,comm) \ 550f12d10ccSJunchao Zhang ((petsc_scatter_ct++,0) || PetscMPITypeSize((recvcount),(recvtype),(&petsc_recv_len)) || MPI_Scatter((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm))) 55101faf4e4SMatthew Knepley 55201faf4e4SMatthew Knepley #define MPI_Scatterv(sendbuf,sendcount,displs,sendtype,recvbuf,recvcount,recvtype,root,comm) \ 553f12d10ccSJunchao Zhang ((petsc_scatter_ct++,0) || PetscMPITypeSize((recvcount),(recvtype),(&petsc_recv_len)) || MPI_Scatterv((sendbuf),(sendcount),(displs),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm))) 554f12d10ccSJunchao Zhang 555f12d10ccSJunchao Zhang #define MPI_Ialltoall(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,comm,request) \ 556f12d10ccSJunchao Zhang ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Ialltoall((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(comm),(request))) 557f12d10ccSJunchao Zhang 558f12d10ccSJunchao Zhang #define MPI_Ialltoallv(sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm,request) \ 559f12d10ccSJunchao Zhang ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || PetscMPITypeSizeComm((comm),(sendcnts),(sendtype),(&petsc_send_len)) || MPI_Ialltoallv((sendbuf),(sendcnts),(sdispls),(sendtype),(recvbuf),(recvcnts),(rdispls),(recvtype),(comm),(request))) 560f12d10ccSJunchao Zhang 561f12d10ccSJunchao Zhang #define MPI_Iallgather(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,comm,request) \ 562f12d10ccSJunchao Zhang ((petsc_gather_ct += PetscMPIParallelComm((comm)),0) || MPI_Iallgather((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(comm),(request))) 563f12d10ccSJunchao Zhang 564f12d10ccSJunchao Zhang #define MPI_Iallgatherv(sendbuf,sendcount,sendtype,recvbuf,recvcount,displs,recvtype,comm,request) \ 565f12d10ccSJunchao Zhang ((petsc_gather_ct += PetscMPIParallelComm((comm)),0) || MPI_Iallgatherv((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(displs),(recvtype),(comm),(request))) 566f12d10ccSJunchao Zhang 567f12d10ccSJunchao Zhang #define MPI_Igather(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,root,comm,request) \ 568f12d10ccSJunchao Zhang ((petsc_gather_ct++,0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Igather((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm),(request))) 569f12d10ccSJunchao Zhang 570f12d10ccSJunchao Zhang #define MPI_Igatherv(sendbuf,sendcount,sendtype,recvbuf,recvcount,displs,recvtype,root,comm,request) \ 571f12d10ccSJunchao Zhang ((petsc_gather_ct++,0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Igatherv((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(displs),(recvtype),(root),(comm),(request))) 572f12d10ccSJunchao Zhang 573f12d10ccSJunchao Zhang #define MPI_Iscatter(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,root,comm,request) \ 574f12d10ccSJunchao Zhang ((petsc_scatter_ct++,0) || PetscMPITypeSize((recvcount),(recvtype),(&petsc_recv_len)) || MPI_Iscatter((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm),(request))) 575f12d10ccSJunchao Zhang 576f12d10ccSJunchao Zhang #define MPI_Iscatterv(sendbuf,sendcount,displs,sendtype,recvbuf,recvcount,recvtype,root,comm,request) \ 577f12d10ccSJunchao Zhang ((petsc_scatter_ct++,0) || PetscMPITypeSize((recvcount),(recvtype),(&petsc_recv_len)) || MPI_Iscatterv((sendbuf),(sendcount),(displs),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm),(request))) 578f12d10ccSJunchao Zhang 579f12d10ccSJunchao Zhang /* We treat MPI_Ineighbor_alltoallv as a set of isend/irecv instead of a traditional MPI collective. 580f12d10ccSJunchao Zhang OpenMPI-3.0 ran into error with outdegree = indegree = 0, so we use ((outdegree) || (indegree)) as a workaround. 581f12d10ccSJunchao Zhang */ 582f12d10ccSJunchao Zhang #define MPI_Start_ineighbor_alltoallv(outdegree,indegree,sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm,request) \ 583f12d10ccSJunchao Zhang ((petsc_isend_ct += (PetscLogDouble)(outdegree),0) || (petsc_irecv_ct += (PetscLogDouble)(indegree),0) || PetscMPITypeSizeCount((outdegree),(sendcnts),(sendtype),(&petsc_isend_len)) || PetscMPITypeSizeCount((indegree),(recvcnts),(recvtype),(&petsc_irecv_len)) || (((outdegree) || (indegree)) && MPI_Ineighbor_alltoallv((sendbuf),(sendcnts),(sdispls),(sendtype),(recvbuf),(recvcnts),(rdispls),(recvtype),(comm),(request)))) 584f12d10ccSJunchao Zhang 585f12d10ccSJunchao Zhang #define MPI_Start_neighbor_alltoallv(outdegree,indegree,sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm) \ 586f12d10ccSJunchao Zhang ((petsc_isend_ct += (PetscLogDouble)(outdegree),0) || (petsc_irecv_ct += (PetscLogDouble)(indegree),0) || PetscMPITypeSizeCount((outdegree),(sendcnts),(sendtype),(&petsc_isend_len)) || PetscMPITypeSizeCount((indegree),(recvcnts),(recvtype),(&petsc_irecv_len)) || (((outdegree) || (indegree)) && MPI_Neighbor_alltoallv((sendbuf),(sendcnts),(sdispls),(sendtype),(recvbuf),(recvcnts),(rdispls),(recvtype),(comm)))) 58701faf4e4SMatthew Knepley 5880d4b0b6cSBarry Smith #else 5890d4b0b6cSBarry Smith 590457b14b1SJunchao Zhang #define MPI_Startall_irecv(count,datatype,number,requests) \ 591ea1c87f7SJunchao Zhang ((number) && MPI_Startall((number),(requests))) 5920d4b0b6cSBarry Smith 593457b14b1SJunchao Zhang #define MPI_Startall_isend(count,datatype,number,requests) \ 594ea1c87f7SJunchao Zhang ((number) && MPI_Startall((number),(requests))) 5950d4b0b6cSBarry Smith 596457b14b1SJunchao Zhang #define MPI_Start_isend(count,datatype,requests) \ 5975fa69175SJed Brown (MPI_Start((requests))) 5980d4b0b6cSBarry Smith 599f12d10ccSJunchao Zhang #define MPI_Start_ineighbor_alltoallv(outdegree,indegree,sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm,request) \ 600f12d10ccSJunchao Zhang (((outdegree) || (indegree)) && MPI_Ineighbor_alltoallv((sendbuf),(sendcnts),(sdispls),(sendtype),(recvbuf),(recvcnts),(rdispls),(recvtype),(comm),(request))) 601f12d10ccSJunchao Zhang 602f12d10ccSJunchao Zhang #define MPI_Start_neighbor_alltoallv(outdegree,indegree,sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm) \ 603f12d10ccSJunchao Zhang (((outdegree) || (indegree)) && MPI_Neighbor_alltoallv((sendbuf),(sendcnts),(sdispls),(sendtype),(recvbuf),(recvcnts),(rdispls),(recvtype),(comm))) 604c9b973beSBarry Smith #endif /* !MPIUNI_H && ! PETSC_HAVE_BROKEN_RECURSIVE_MACRO */ 605614700edSBarry Smith 606df8cf0b5SBarry Smith #else /* ---Logging is turned off --------------------------------------------*/ 607614700edSBarry Smith 608e3ed9ee7SBarry Smith #define PetscLogMemory PETSC_FALSE 609e3ed9ee7SBarry Smith 610b0a32e0cSBarry Smith #define PetscLogFlops(n) 0 611fa2bb9feSLisandro Dalcin #define PetscGetFlops(a) (*(a) = 0.0,0) 612614700edSBarry Smith 613fa2bb9feSLisandro Dalcin #define PetscLogStageRegister(a,b) 0 614fa2bb9feSLisandro Dalcin #define PetscLogStagePush(a) 0 615fa2bb9feSLisandro Dalcin #define PetscLogStagePop() 0 616f5d6ab90SLisandro Dalcin #define PetscLogStageSetActive(a,b) 0 617f5d6ab90SLisandro Dalcin #define PetscLogStageGetActive(a,b) 0 618f5d6ab90SLisandro Dalcin #define PetscLogStageGetVisible(a,b) 0 619f5d6ab90SLisandro Dalcin #define PetscLogStageSetVisible(a,b) 0 620f5d6ab90SLisandro Dalcin #define PetscLogStageGetId(a,b) (*(b)=0,0) 621f5d6ab90SLisandro Dalcin 622fa2bb9feSLisandro Dalcin #define PetscLogEventRegister(a,b,c) 0 623fa2bb9feSLisandro Dalcin #define PetscLogEventSetCollective(a,b) 0 624fa2bb9feSLisandro Dalcin #define PetscLogEventIncludeClass(a) 0 625fa2bb9feSLisandro Dalcin #define PetscLogEventExcludeClass(a) 0 626b0a32e0cSBarry Smith #define PetscLogEventActivate(a) 0 627b0a32e0cSBarry Smith #define PetscLogEventDeactivate(a) 0 628b0a32e0cSBarry Smith #define PetscLogEventActivateClass(a) 0 629b0a32e0cSBarry Smith #define PetscLogEventDeactivateClass(a) 0 630e8e7597cSSatish Balay #define PetscLogEventSetActiveAll(a,b) 0 63146eb3cd7SSatish Balay #define PetscLogEventGetId(a,b) (*(b)=0,0) 632561b2a67SSatish Balay #define PetscLogEventGetPerfInfo(a,b,c) 0 6339c702cfcSMatthew G. Knepley #define PetscLogEventSetDof(a,b,c) 0 6349c702cfcSMatthew G. Knepley #define PetscLogEventSetError(a,b,c) 0 63577c4ece6SBarry Smith 6368b5db460SBarry Smith #define PetscLogPLB 0 6378b5db460SBarry Smith #define PetscLogPLE 0 6388b5db460SBarry Smith #define PetscLogPHC 0 6398b5db460SBarry Smith #define PetscLogPHD 0 640f5d6ab90SLisandro Dalcin 641efee365bSSatish Balay #define PetscLogObjectParents(p,n,c) 0 64252e6d16bSBarry Smith #define PetscLogObjectCreate(h) 0 64352e6d16bSBarry Smith #define PetscLogObjectDestroy(h) 0 644fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogObjectState(PetscObject,const char[],...); 645fa2bb9feSLisandro Dalcin 646bb1d7374SBarry Smith #define PetscLogDefaultBegin() 0 647b0a32e0cSBarry Smith #define PetscLogAllBegin() 0 648bb1d7374SBarry Smith #define PetscLogNestedBegin() 0 649fa2bb9feSLisandro Dalcin #define PetscLogTraceBegin(file) 0 65073fda44aSBarry Smith #define PetscLogActions(a) 0 651fa2bb9feSLisandro Dalcin #define PetscLogObjects(a) 0 652fa2bb9feSLisandro Dalcin #define PetscLogSetThreshold(a,b) 0 653fa2bb9feSLisandro Dalcin #define PetscLogSet(lb,le) 0 654fa2bb9feSLisandro Dalcin 655fa2bb9feSLisandro Dalcin #define PetscLogView(viewer) 0 656fa2bb9feSLisandro Dalcin #define PetscLogViewFromOptions() 0 657fa2bb9feSLisandro Dalcin #define PetscLogDump(c) 0 658fa2bb9feSLisandro Dalcin 65962872c28SLisandro Dalcin #define PetscLogEventSync(e,comm) 0 660fa2bb9feSLisandro Dalcin #define PetscLogEventBegin(e,o1,o2,o3,o4) 0 661fa2bb9feSLisandro Dalcin #define PetscLogEventEnd(e,o1,o2,o3,o4) 0 662ce6058e1SBarry Smith 6638157aad8SJunchao Zhang #define PetscLogCpuToGpu(a) 0 6648157aad8SJunchao Zhang #define PetscLogGpuToCpu(a) 0 6658157aad8SJunchao Zhang #define PetscLogGpuFlops(a) 0 6668157aad8SJunchao Zhang #define PetscLogGpuTimeBegin() 0 6678157aad8SJunchao Zhang #define PetscLogGpuTimeEnd() 0 6688157aad8SJunchao Zhang #define PetscLogGpuTimeAdd(a) 0 6698157aad8SJunchao Zhang 670aa482453SBarry Smith /* If PETSC_USE_LOG is NOT defined, these still need to be! */ 671ea1c87f7SJunchao Zhang #define MPI_Startall_irecv(count,datatype,number,requests) ((number) && MPI_Startall(number,requests)) 672ea1c87f7SJunchao Zhang #define MPI_Startall_isend(count,datatype,number,requests) ((number) && MPI_Startall(number,requests)) 673457b14b1SJunchao Zhang #define MPI_Start_isend(count,datatype,requests) MPI_Start(requests) 674f12d10ccSJunchao Zhang #define MPI_Start_ineighbor_alltoallv(outdegree,indegree,sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm,request) \ 675f12d10ccSJunchao Zhang (((outdegree) || (indegree)) && MPI_Ineighbor_alltoallv((sendbuf),(sendcnts),(sdispls),(sendtype),(recvbuf),(recvcnts),(rdispls),(recvtype),(comm),(request))) 676f12d10ccSJunchao Zhang #define MPI_Start_neighbor_alltoallv(outdegree,indegree,sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm) \ 677f12d10ccSJunchao Zhang (((outdegree) || (indegree)) && MPI_Neighbor_alltoallv((sendbuf),(sendcnts),(sdispls),(sendtype),(recvbuf),(recvcnts),(rdispls),(recvtype),(comm))) 678f141ce34SMatthew Knepley 679aa482453SBarry Smith #endif /* PETSC_USE_LOG */ 6806daaf66cSBarry Smith 681204a6943SJed Brown #define PetscPreLoadBegin(flag,name) \ 682204a6943SJed Brown do {\ 683204a6943SJed Brown PetscBool PetscPreLoading = flag;\ 684204a6943SJed Brown int PetscPreLoadMax,PetscPreLoadIt;\ 685166c7f25SBarry Smith PetscLogStage _stageNum;\ 6868cbcd9ccSBarry Smith PetscErrorCode _3_ierr; \ 687c5929fdfSBarry Smith _3_ierr = PetscOptionsGetBool(NULL,NULL,"-preload",&PetscPreLoading,NULL);CHKERRQ(_3_ierr); \ 688204a6943SJed Brown PetscPreLoadMax = (int)(PetscPreLoading);\ 689204a6943SJed Brown PetscPreLoadingUsed = PetscPreLoading ? PETSC_TRUE : PetscPreLoadingUsed;\ 690204a6943SJed Brown for (PetscPreLoadIt=0; PetscPreLoadIt<=PetscPreLoadMax; PetscPreLoadIt++) {\ 691204a6943SJed Brown PetscPreLoadingOn = PetscPreLoading;\ 6920298fd71SBarry Smith _3_ierr = PetscBarrier(NULL);CHKERRQ(_3_ierr);\ 693204a6943SJed Brown if (PetscPreLoadIt>0) {\ 6948e58c17dSMatthew Knepley _3_ierr = PetscLogStageGetId(name,&_stageNum);CHKERRQ(_3_ierr);\ 6958e58c17dSMatthew Knepley } else {\ 696a3bc4eb9SBarry Smith _3_ierr = PetscLogStageRegister(name,&_stageNum);CHKERRQ(_3_ierr); \ 6978e58c17dSMatthew Knepley }\ 698204a6943SJed Brown _3_ierr = PetscLogStageSetActive(_stageNum,(PetscBool)(!PetscPreLoadMax || PetscPreLoadIt));\ 6998e58c17dSMatthew Knepley _3_ierr = PetscLogStagePush(_stageNum);CHKERRQ(_3_ierr); 7008e58c17dSMatthew Knepley 701204a6943SJed Brown #define PetscPreLoadEnd() \ 702043328b6SSatish Balay _3_ierr = PetscLogStagePop();CHKERRQ(_3_ierr);\ 703204a6943SJed Brown PetscPreLoading = PETSC_FALSE;\ 704043328b6SSatish Balay }\ 705204a6943SJed Brown } while (0) 7068e58c17dSMatthew Knepley 707204a6943SJed Brown #define PetscPreLoadStage(name) do { \ 708043328b6SSatish Balay _3_ierr = PetscLogStagePop();CHKERRQ(_3_ierr); \ 709204a6943SJed Brown if (PetscPreLoadIt>0) { \ 7108e58c17dSMatthew Knepley _3_ierr = PetscLogStageGetId(name,&_stageNum);CHKERRQ(_3_ierr); \ 7118e58c17dSMatthew Knepley } else { \ 712a3bc4eb9SBarry Smith _3_ierr = PetscLogStageRegister(name,&_stageNum);CHKERRQ(_3_ierr); \ 7138e58c17dSMatthew Knepley } \ 714204a6943SJed Brown _3_ierr = PetscLogStageSetActive(_stageNum,(PetscBool)(!PetscPreLoadMax || PetscPreLoadIt)); \ 715204a6943SJed Brown _3_ierr = PetscLogStagePush(_stageNum);CHKERRQ(_3_ierr); \ 716204a6943SJed Brown } while (0) 717e9fa29b7SSatish Balay 718ad14c47eSJed Brown /* some vars for logging */ 719ad14c47eSJed Brown PETSC_EXTERN PetscBool PetscPreLoadingUsed; /* true if we are or have done preloading */ 720ad14c47eSJed Brown PETSC_EXTERN PetscBool PetscPreLoadingOn; /* true if we are currently in a preloading calculation */ 721ad14c47eSJed Brown 72297bb86f7SLois Curfman McInnes #endif 723