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 */ 113ca90d2dSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfo_Private(const char[],PetscObject,const char[],...) PETSC_ATTRIBUTE_FORMAT(3,4); 12fa2bb9feSLisandro Dalcin #if defined(PETSC_USE_INFO) 137d3de750SJacob Faibussowitsch #define PetscInfo(A,...) PetscInfo_Private(PETSC_FUNCTION_NAME,((PetscObject)A),__VA_ARGS__) 14fa2bb9feSLisandro Dalcin #else 157d3de750SJacob Faibussowitsch #define PetscInfo(A,...) 0 16fa2bb9feSLisandro Dalcin #endif 17e94e781bSJacob Faibussowitsch 187d3de750SJacob Faibussowitsch #define PetscInfo1(...) PETSC_DEPRECATED_MACRO("GCC warning \"Use PetscInfo() (since version 3.17)\"") PetscInfo(__VA_ARGS__) 197d3de750SJacob Faibussowitsch #define PetscInfo2(...) PETSC_DEPRECATED_MACRO("GCC warning \"Use PetscInfo() (since version 3.17)\"") PetscInfo(__VA_ARGS__) 207d3de750SJacob Faibussowitsch #define PetscInfo3(...) PETSC_DEPRECATED_MACRO("GCC warning \"Use PetscInfo() (since version 3.17)\"") PetscInfo(__VA_ARGS_)_ 217d3de750SJacob Faibussowitsch #define PetscInfo4(...) PETSC_DEPRECATED_MACRO("GCC warning \"Use PetscInfo() (since version 3.17)\"") PetscInfo(__VA_ARGS__) 227d3de750SJacob Faibussowitsch #define PetscInfo5(...) PETSC_DEPRECATED_MACRO("GCC warning \"Use PetscInfo() (since version 3.17)\"") PetscInfo(__VA_ARGS__) 237d3de750SJacob Faibussowitsch #define PetscInfo6(...) PETSC_DEPRECATED_MACRO("GCC warning \"Use PetscInfo() (since version 3.17)\"") PetscInfo(__VA_ARGS__) 247d3de750SJacob Faibussowitsch #define PetscInfo7(...) PETSC_DEPRECATED_MACRO("GCC warning \"Use PetscInfo() (since version 3.17)\"") PetscInfo(__VA_ARGS__) 257d3de750SJacob Faibussowitsch #define PetscInfo8(...) PETSC_DEPRECATED_MACRO("GCC warning \"Use PetscInfo() (since version 3.17)\"") PetscInfo(__VA_ARGS__) 267d3de750SJacob Faibussowitsch #define PetscInfo9(...) PETSC_DEPRECATED_MACRO("GCC warning \"Use PetscInfo() (since version 3.17)\"") PetscInfo(__VA_ARGS__) 277d3de750SJacob Faibussowitsch 28e94e781bSJacob Faibussowitsch /*E 29e94e781bSJacob Faibussowitsch PetscInfoCommFlag - Describes the method by which to filter PetscInfo() by communicator size 30e94e781bSJacob Faibussowitsch 31e94e781bSJacob Faibussowitsch Used as an input for PetscInfoSetFilterCommSelf() 32e94e781bSJacob Faibussowitsch 33e94e781bSJacob Faibussowitsch $ PETSC_INFO_COMM_ALL - Default uninitialized value. PetscInfo() will not filter based on communicator size (i.e. will 34e94e781bSJacob Faibussowitsch print for all communicators) 35e94e781bSJacob Faibussowitsch $ PETSC_INFO_COMM_NO_SELF - PetscInfo() will NOT print for communicators with size = 1 (i.e. *_COMM_SELF) 36e94e781bSJacob Faibussowitsch $ PETSC_INFO_COMM_ONLY_SELF - PetscInfo will ONLY print for communicators with size = 1 37e94e781bSJacob Faibussowitsch 38e94e781bSJacob Faibussowitsch Level: intermediate 39e94e781bSJacob Faibussowitsch 40e94e781bSJacob Faibussowitsch .seealso: PetscInfo(), PetscInfoSetFromOptions(), PetscInfoSetFilterCommSelf() 41e94e781bSJacob Faibussowitsch E*/ 42e94e781bSJacob Faibussowitsch typedef enum { 43e94e781bSJacob Faibussowitsch PETSC_INFO_COMM_ALL = -1, 44e94e781bSJacob Faibussowitsch PETSC_INFO_COMM_NO_SELF = 0, 45e94e781bSJacob Faibussowitsch PETSC_INFO_COMM_ONLY_SELF = 1 46e94e781bSJacob Faibussowitsch } PetscInfoCommFlag; 47e94e781bSJacob Faibussowitsch 48e94e781bSJacob Faibussowitsch PETSC_EXTERN const char * const PetscInfoCommFlags[]; 49fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscInfoDeactivateClass(PetscClassId); 50fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscInfoActivateClass(PetscClassId); 51e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoEnabled(PetscClassId, PetscBool *); 52e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoAllow(PetscBool); 53e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoSetFile(const char[],const char[]); 54e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoGetFile(char **,FILE **); 55e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoSetClasses(PetscBool,PetscInt,const char *const *); 56e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoGetClass(const char *, PetscBool *); 57e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoGetInfo(PetscBool *,PetscBool *,PetscBool *,PetscBool *,PetscInfoCommFlag *); 58e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoProcessClass(const char[],PetscInt,PetscClassId[]); 59e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoSetFilterCommSelf(PetscInfoCommFlag); 60e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoSetFromOptions(PetscOptions); 61e94e781bSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscInfoDestroy(void); 62fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscBool PetscLogPrintInfo; /* if true, indicates PetscInfo() is turned on */ 63fa2bb9feSLisandro Dalcin 64b859824cSBarry Smith /*MC 65b859824cSBarry Smith PetscLogEvent - id used to identify PETSc or user events which timed portions (blocks of executable) 66b859824cSBarry Smith code. 67b859824cSBarry Smith 68b859824cSBarry Smith Level: intermediate 69b859824cSBarry Smith 70b859824cSBarry Smith .seealso: PetscLogEventRegister(), PetscLogEventBegin(), PetscLogEventEnd(), PetscLogStage 71b859824cSBarry Smith M*/ 72b859824cSBarry Smith typedef int PetscLogEvent; 73b859824cSBarry Smith 74b859824cSBarry Smith /*MC 75b859824cSBarry Smith PetscLogStage - id used to identify user stages (phases, sections) of runs - for logging 76b859824cSBarry Smith 77b859824cSBarry Smith Level: intermediate 78b859824cSBarry Smith 79522ce795SJed Brown .seealso: PetscLogStageRegister(), PetscLogStagePush(), PetscLogStagePop(), PetscLogEvent 80b859824cSBarry Smith M*/ 81b859824cSBarry Smith typedef int PetscLogStage; 82b859824cSBarry Smith 838ba1e511SMatthew Knepley #define PETSC_EVENT 1311311 84014dd563SJed Brown PETSC_EXTERN PetscLogEvent PETSC_LARGEST_EVENT; 8597bb86f7SLois Curfman McInnes 8619b02663SBarry Smith /* Global flop counter */ 87014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_TotalFlops; 88014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_tmp_flops; 8919b02663SBarry Smith 9031d06abdSBarry Smith /* We must make the following structures available to access the event 9131d06abdSBarry Smith activation flags in the PetscLogEventBegin/End() macros. These are not part of the PETSc public 9231d06abdSBarry Smith API and are not intended to be used by other parts of PETSc or by users. 93d49d4b11SBarry Smith 94fa2bb9feSLisandro Dalcin The code that manipulates these structures is in src/sys/logging/utils. 9531d06abdSBarry Smith */ 96aa213bdcSJed Brown typedef struct _n_PetscIntStack *PetscIntStack; 976a6a9b46SSatish Balay 98217044c2SLisandro Dalcin /* -----------------------------------------------------------------------------------------------------*/ 99c60ec953SBarry Smith /* 100aa213bdcSJed Brown PetscClassRegInfo, PetscClassPerfInfo - Each class has two data structures associated with it. The first has 101c60ec953SBarry Smith static information about it, the second collects statistics on how many objects of the class are created, 102c60ec953SBarry Smith how much memory they use, etc. 103c60ec953SBarry Smith 104aa213bdcSJed Brown PetscClassRegLog, PetscClassPerfLog - arrays of the PetscClassRegInfo and PetscClassPerfInfo for all classes. 105c60ec953SBarry Smith */ 1066a6a9b46SSatish Balay typedef struct { 107c60ec953SBarry Smith char *name; /* The class name */ 108c60ec953SBarry Smith PetscClassId classid; /* The integer identifying this class */ 109aa213bdcSJed Brown } PetscClassRegInfo; 1106a6a9b46SSatish Balay 1116a6a9b46SSatish Balay typedef struct { 1120700a824SBarry Smith PetscClassId id; /* The integer identifying this class */ 1136a6a9b46SSatish Balay int creations; /* The number of objects of this class created */ 1146a6a9b46SSatish Balay int destructions; /* The number of objects of this class destroyed */ 1156a6a9b46SSatish Balay PetscLogDouble mem; /* The total memory allocated by objects of this class */ 1166a6a9b46SSatish Balay PetscLogDouble descMem; /* The total memory allocated by descendents of these objects */ 117aa213bdcSJed Brown } PetscClassPerfInfo; 1186a6a9b46SSatish Balay 119aa213bdcSJed Brown typedef struct _n_PetscClassRegLog *PetscClassRegLog; 120aa213bdcSJed Brown struct _n_PetscClassRegLog { 121c60ec953SBarry Smith int numClasses; /* The number of classes registered */ 122c60ec953SBarry Smith int maxClasses; /* The maximum number of classes */ 123aa213bdcSJed Brown PetscClassRegInfo *classInfo; /* The structure for class information (classids are monotonicly increasing) */ 124c60ec953SBarry Smith }; 1256a6a9b46SSatish Balay 126aa213bdcSJed Brown typedef struct _n_PetscClassPerfLog *PetscClassPerfLog; 127aa213bdcSJed Brown struct _n_PetscClassPerfLog { 128c60ec953SBarry Smith int numClasses; /* The number of logging classes */ 129c60ec953SBarry Smith int maxClasses; /* The maximum number of classes */ 130aa213bdcSJed Brown PetscClassPerfInfo *classInfo; /* The structure for class information (classids are monotonicly increasing) */ 131c60ec953SBarry Smith }; 132c60ec953SBarry Smith /* -----------------------------------------------------------------------------------------------------*/ 133c60ec953SBarry Smith /* 134aa213bdcSJed Brown PetscEventRegInfo, PetscEventPerfInfo - Each event has two data structures associated with it. The first has 135c60ec953SBarry Smith static information about it, the second collects statistics on how many times the event is used, how 136c60ec953SBarry Smith much time it takes, etc. 137c60ec953SBarry Smith 138aa213bdcSJed Brown PetscEventRegLog, PetscEventPerfLog - an array of all PetscEventRegInfo and PetscEventPerfInfo for all events. There is one 139c60ec953SBarry Smith of these for each stage. 140c60ec953SBarry Smith 141c60ec953SBarry Smith */ 1426a6a9b46SSatish Balay typedef struct { 1436a6a9b46SSatish Balay char *name; /* The name of this event */ 144c60ec953SBarry Smith PetscClassId classid; /* The class the event is associated with */ 145217044c2SLisandro Dalcin PetscBool collective; /* Flag this event as collective */ 1466a6a9b46SSatish Balay #if defined (PETSC_HAVE_MPE) 1476a6a9b46SSatish Balay int mpe_id_begin; /* MPE IDs that define the event */ 1486a6a9b46SSatish Balay int mpe_id_end; 1496a6a9b46SSatish Balay #endif 150aa213bdcSJed Brown } PetscEventRegInfo; 1516a6a9b46SSatish Balay 152c60ec953SBarry Smith typedef struct { 153c60ec953SBarry Smith int id; /* The integer identifying this event */ 154ace3abfcSBarry Smith PetscBool active; /* The flag to activate logging */ 155ace3abfcSBarry Smith PetscBool visible; /* The flag to print info in summary */ 156c60ec953SBarry Smith int depth; /* The nesting depth of the event call */ 157c60ec953SBarry Smith int count; /* The number of times this event was executed */ 158a42e5b54SMatthew G Knepley PetscLogDouble flops, flops2, flopsTmp; /* The flops and flops^2 used in this event */ 159a42e5b54SMatthew G Knepley PetscLogDouble time, time2, timeTmp; /* The time and time^2 taken for this event */ 160217044c2SLisandro Dalcin PetscLogDouble syncTime; /* The synchronization barrier time */ 161891e75beSMatthew G. Knepley PetscLogDouble dof[8]; /* The number of degrees of freedom associated with this event */ 1625d68e14cSMatthew G. Knepley PetscLogDouble errors[8]; /* The errors (user-defined) associated with this event */ 163c60ec953SBarry Smith PetscLogDouble numMessages; /* The number of messages in this event */ 164c60ec953SBarry Smith PetscLogDouble messageLength; /* The total message lengths in this event */ 165c60ec953SBarry Smith PetscLogDouble numReductions; /* The number of reductions in this event */ 166e3ed9ee7SBarry Smith PetscLogDouble memIncrease; /* How much the resident memory has increased in this event */ 167e3ed9ee7SBarry Smith PetscLogDouble mallocIncrease;/* How much the maximum malloced space has increased in this event */ 168e3ed9ee7SBarry Smith PetscLogDouble mallocSpace; /* How much the space was malloced and kept during this event */ 169e3ed9ee7SBarry Smith PetscLogDouble mallocIncreaseEvent; /* Maximum of the high water mark with in event minus memory available at the end of the event */ 170091bc5acSJunchao Zhang #if defined(PETSC_HAVE_DEVICE) 171bec0b493Shannah_mairs PetscLogDouble CpuToGpuCount; /* The total number of CPU to GPU copies */ 172bec0b493Shannah_mairs PetscLogDouble GpuToCpuCount; /* The total number of GPU to CPU copies */ 173bec0b493Shannah_mairs PetscLogDouble CpuToGpuSize; /* The total size of CPU to GPU copies */ 174bec0b493Shannah_mairs PetscLogDouble GpuToCpuSize; /* The total size of GPU to CPU copies */ 175958c4211Shannah_mairs PetscLogDouble GpuFlops; /* The flops done on a GPU in this event */ 176958c4211Shannah_mairs PetscLogDouble GpuTime; /* The time spent on a GPU in this event */ 177bec0b493Shannah_mairs #endif 178aa213bdcSJed Brown } PetscEventPerfInfo; 179c60ec953SBarry Smith 180aa213bdcSJed Brown typedef struct _n_PetscEventRegLog *PetscEventRegLog; 181aa213bdcSJed Brown struct _n_PetscEventRegLog { 1826a6a9b46SSatish Balay int numEvents; /* The number of registered events */ 1836a6a9b46SSatish Balay int maxEvents; /* The maximum number of events */ 184aa213bdcSJed Brown PetscEventRegInfo *eventInfo; /* The registration information for each event */ 1856a6a9b46SSatish Balay }; 1866a6a9b46SSatish Balay 187aa213bdcSJed Brown typedef struct _n_PetscEventPerfLog *PetscEventPerfLog; 188aa213bdcSJed Brown struct _n_PetscEventPerfLog { 1896a6a9b46SSatish Balay int numEvents; /* The number of logging events */ 1906a6a9b46SSatish Balay int maxEvents; /* The maximum number of events */ 191aa213bdcSJed Brown PetscEventPerfInfo *eventInfo; /* The performance information for each event */ 1926a6a9b46SSatish Balay }; 193c60ec953SBarry Smith /* ------------------------------------------------------------------------------------------------------------*/ 194c60ec953SBarry Smith /* 195aa213bdcSJed Brown PetscStageInfo - Contains all the information about a particular stage. 1966a6a9b46SSatish Balay 197aa213bdcSJed Brown PetscStageLog - An array of PetscStageInfo for each registered stage. There is a single one of these in the code. 198c60ec953SBarry Smith */ 199aa213bdcSJed Brown typedef struct _PetscStageInfo { 2006a6a9b46SSatish Balay char *name; /* The stage name */ 201ace3abfcSBarry Smith PetscBool used; /* The stage was pushed on this processor */ 202aa213bdcSJed Brown PetscEventPerfInfo perfInfo; /* The stage performance information */ 203aa213bdcSJed Brown PetscEventPerfLog eventLog; /* The event information for this stage */ 204aa213bdcSJed Brown PetscClassPerfLog classLog; /* The class information for this stage */ 205aa213bdcSJed Brown } PetscStageInfo; 2066a6a9b46SSatish Balay 207aa213bdcSJed Brown typedef struct _n_PetscStageLog *PetscStageLog; 208aa213bdcSJed Brown struct _n_PetscStageLog { 2096a6a9b46SSatish Balay int numStages; /* The number of registered stages */ 2106a6a9b46SSatish Balay int maxStages; /* The maximum number of stages */ 211aa213bdcSJed Brown PetscIntStack stack; /* The stack for active stages */ 212aa213bdcSJed Brown int curStage; /* The current stage (only used in macros so we don't call PetscIntStackTop) */ 213aa213bdcSJed Brown PetscStageInfo *stageInfo; /* The information for each stage */ 214aa213bdcSJed Brown PetscEventRegLog eventLog; /* The registered events */ 215aa213bdcSJed Brown PetscClassRegLog classLog; /* The registered classes */ 2166a6a9b46SSatish Balay }; 217217044c2SLisandro Dalcin /* -----------------------------------------------------------------------------------------------------*/ 218fee518ebSJed Brown 2193bb1ff40SBarry Smith PETSC_EXTERN PetscErrorCode PetscLogObjectParent(PetscObject,PetscObject); 2203bb1ff40SBarry Smith PETSC_EXTERN PetscErrorCode PetscLogObjectMemory(PetscObject,PetscLogDouble); 2213bb1ff40SBarry Smith 222aa482453SBarry Smith #if defined(PETSC_USE_LOG) /* --- Logging is turned on --------------------------------*/ 223aeeda9f6SJohn Fettig PETSC_EXTERN PetscStageLog petsc_stageLog; 224f5d6ab90SLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogGetStageLog(PetscStageLog*); 225f5d6ab90SLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscStageLogGetCurrent(PetscStageLog,int*); 226f5d6ab90SLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscStageLogGetEventPerfLog(PetscStageLog,int,PetscEventPerfLog*); 227614700edSBarry Smith 228da63de55SLois Curfman McInnes /* 229da63de55SLois Curfman McInnes Flop counting: We count each arithmetic operation (e.g., addition, multiplication) separately. 230da63de55SLois Curfman McInnes 231da63de55SLois Curfman McInnes For the complex numbers version, note that 232da63de55SLois Curfman McInnes 1 complex addition = 2 flops 233da63de55SLois Curfman McInnes 1 complex multiplication = 6 flops, 234da63de55SLois Curfman McInnes where we define 1 flop as that for a double precision scalar. We roughly approximate 235da63de55SLois Curfman McInnes flop counting for complex numbers by multiplying the total flops by 4; this corresponds 236da63de55SLois Curfman McInnes to the assumption that we're counting mostly additions and multiplications -- and 237da63de55SLois Curfman McInnes roughly the same number of each. More accurate counting could be done by distinguishing 238da63de55SLois Curfman McInnes among the various arithmetic operations. 239da63de55SLois Curfman McInnes */ 240da63de55SLois Curfman McInnes 241aa482453SBarry Smith #if defined(PETSC_USE_COMPLEX) 242542d4b3fSSatish Balay #define PETSC_FLOPS_PER_OP 4.0 243da63de55SLois Curfman McInnes #else 244542d4b3fSSatish Balay #define PETSC_FLOPS_PER_OP 1.0 245bf3909cdSBarry Smith #endif 246bf3909cdSBarry Smith 247f8bfbb56SSatish Balay /*@C 248ca0c957dSBarry Smith PetscLogFlops - Log how many flops are performed in a calculation 249ca0c957dSBarry Smith 250c3b874c4SJose E. Roman Input Parameter: 251c3b874c4SJose E. Roman . flops - the number of flops 252ca0c957dSBarry Smith 253ca0c957dSBarry Smith Notes: 254ca0c957dSBarry Smith To limit the chance of integer overflow when multiplying by a constant, represent the constant as a double, 255ca0c957dSBarry Smith not an integer. Use PetscLogFlops(4.0*n) not PetscLogFlops(4*n) 256ca0c957dSBarry Smith 257ca0c957dSBarry Smith Level: intermediate 258ca0c957dSBarry Smith 25968481988SMark .seealso: PetscLogView(), PetscLogGpuFlops() 260ca0c957dSBarry Smith @*/ 261ca0c957dSBarry Smith 262*9fbee547SJacob Faibussowitsch static inline PetscErrorCode PetscLogFlops(PetscLogDouble n) 263dd19b90bSBarry Smith { 264dd19b90bSBarry Smith PetscFunctionBegin; 265bf3909cdSBarry Smith #if defined(PETSC_USE_DEBUG) 2669a202e32SJacob Faibussowitsch PetscAssert(n >= 0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Cannot log negative flops"); 267da63de55SLois Curfman McInnes #endif 268dd19b90bSBarry Smith petsc_TotalFlops += PETSC_FLOPS_PER_OP*n; 269dd19b90bSBarry Smith PetscFunctionReturn(0); 270dd19b90bSBarry Smith } 271fa2bb9feSLisandro Dalcin 272fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscGetFlops(PetscLogDouble *); 27377c4ece6SBarry Smith 274aa482453SBarry Smith #if defined (PETSC_HAVE_MPE) 275014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogMPEBegin(void); 276014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogMPEDump(const char[]); 27777c4ece6SBarry Smith #endif 27877c4ece6SBarry Smith 279014dd563SJed Brown PETSC_EXTERN PetscErrorCode (*PetscLogPLB)(PetscLogEvent,int,PetscObject,PetscObject,PetscObject,PetscObject); 280014dd563SJed Brown PETSC_EXTERN PetscErrorCode (*PetscLogPLE)(PetscLogEvent,int,PetscObject,PetscObject,PetscObject,PetscObject); 281014dd563SJed Brown PETSC_EXTERN PetscErrorCode (*PetscLogPHC)(PetscObject); 282014dd563SJed Brown PETSC_EXTERN PetscErrorCode (*PetscLogPHD)(PetscObject); 28377c4ece6SBarry Smith 284e83a5d19SLisandro 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) 285fa2bb9feSLisandro Dalcin #define PetscLogObjectCreate(h) ((PetscLogPHC) ? (*PetscLogPHC)((PetscObject)(h)) : 0) 286fa2bb9feSLisandro Dalcin #define PetscLogObjectDestroy(h) ((PetscLogPHD) ? (*PetscLogPHD)((PetscObject)(h)) : 0) 2873ca90d2dSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscLogObjectState(PetscObject, const char[], ...) PETSC_ATTRIBUTE_FORMAT(2,3); 288fa2bb9feSLisandro Dalcin 2898ba1e511SMatthew Knepley /* Initialization functions */ 290bb1d7374SBarry Smith PETSC_EXTERN PetscErrorCode PetscLogDefaultBegin(void); 291014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogAllBegin(void); 292bb1d7374SBarry Smith PETSC_EXTERN PetscErrorCode PetscLogNestedBegin(void); 293014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogTraceBegin(FILE *); 294014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogActions(PetscBool); 295014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogObjects(PetscBool); 296fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogSetThreshold(PetscLogDouble,PetscLogDouble*); 297014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogSet(PetscErrorCode (*)(int, int, PetscObject, PetscObject, PetscObject, PetscObject), 2986849ba73SBarry Smith PetscErrorCode (*)(int, int, PetscObject, PetscObject, PetscObject, PetscObject)); 299fa2bb9feSLisandro Dalcin 3008ba1e511SMatthew Knepley /* Output functions */ 301014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogView(PetscViewer); 302f14045dbSBarry Smith PETSC_EXTERN PetscErrorCode PetscLogViewFromOptions(void); 303014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogDump(const char[]); 30431d06abdSBarry Smith 305fa2bb9feSLisandro Dalcin /* Stage functions */ 306014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStageRegister(const char[],PetscLogStage*); 307014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStagePush(PetscLogStage); 308014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStagePop(void); 309014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStageSetActive(PetscLogStage,PetscBool); 310014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStageGetActive(PetscLogStage,PetscBool*); 311014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStageSetVisible(PetscLogStage,PetscBool); 312014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStageGetVisible(PetscLogStage,PetscBool*); 313014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStageGetId(const char[],PetscLogStage*); 314f5d6ab90SLisandro Dalcin 3158ba1e511SMatthew Knepley /* Event functions */ 316014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventRegister(const char[],PetscClassId,PetscLogEvent*); 317217044c2SLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogEventSetCollective(PetscLogEvent,PetscBool); 318fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogEventIncludeClass(PetscClassId); 319fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogEventExcludeClass(PetscClassId); 320014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventActivate(PetscLogEvent); 321014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventDeactivate(PetscLogEvent); 322c00cb57fSBarry Smith PETSC_EXTERN PetscErrorCode PetscLogEventDeactivatePush(PetscLogEvent); 323c00cb57fSBarry Smith PETSC_EXTERN PetscErrorCode PetscLogEventDeactivatePop(PetscLogEvent); 324014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventSetActiveAll(PetscLogEvent,PetscBool); 325014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventActivateClass(PetscClassId); 326014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventDeactivateClass(PetscClassId); 3278c49a8d9SBarry Smith PETSC_EXTERN PetscErrorCode PetscLogEventGetId(const char[],PetscLogEvent*); 32813230571SMichael Lange PETSC_EXTERN PetscErrorCode PetscLogEventGetPerfInfo(int,PetscLogEvent,PetscEventPerfInfo*); 329891e75beSMatthew G. Knepley PETSC_EXTERN PetscErrorCode PetscLogEventSetDof(PetscLogEvent, PetscInt, PetscLogDouble); 3305d68e14cSMatthew G. Knepley PETSC_EXTERN PetscErrorCode PetscLogEventSetError(PetscLogEvent, PetscInt, PetscLogDouble); 3318ba1e511SMatthew Knepley 3328ba1e511SMatthew Knepley /* Global counters */ 333014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_irecv_ct; 334014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_isend_ct; 335014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_recv_ct; 336014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_send_ct; 337014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_irecv_len; 338014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_isend_len; 339014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_recv_len; 340014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_send_len; 341014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_allreduce_ct; 342014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_gather_ct; 343014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_scatter_ct; 344014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_wait_ct; 345014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_wait_any_ct; 346014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_wait_all_ct; 347014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_sum_of_waits_ct; 3488ba1e511SMatthew Knepley 349e3ed9ee7SBarry Smith PETSC_EXTERN PetscBool PetscLogMemory; 350e3ed9ee7SBarry Smith 351fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscBool PetscLogSyncOn; /* true if logging synchronization is enabled */ 35262872c28SLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogEventSynchronize(PetscLogEvent, MPI_Comm); 35362872c28SLisandro Dalcin 35462872c28SLisandro Dalcin #define PetscLogEventSync(e,comm) \ 35562872c28SLisandro Dalcin (((PetscLogPLB && petsc_stageLog->stageInfo[petsc_stageLog->curStage].perfInfo.active && petsc_stageLog->stageInfo[petsc_stageLog->curStage].eventLog->eventInfo[e].active) ? \ 35662872c28SLisandro Dalcin PetscLogEventSynchronize((e),(comm)) : 0)) 35777c4ece6SBarry Smith 35852e6d16bSBarry Smith #define PetscLogEventBegin(e,o1,o2,o3,o4) \ 3598b5db460SBarry Smith (((PetscLogPLB && petsc_stageLog->stageInfo[petsc_stageLog->curStage].perfInfo.active && petsc_stageLog->stageInfo[petsc_stageLog->curStage].eventLog->eventInfo[e].active) ? \ 360495fc317SBarry Smith (*PetscLogPLB)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4)) : 0)) 361043328b6SSatish Balay 36252e6d16bSBarry Smith #define PetscLogEventEnd(e,o1,o2,o3,o4) \ 3638b5db460SBarry Smith (((PetscLogPLE && petsc_stageLog->stageInfo[petsc_stageLog->curStage].perfInfo.active && petsc_stageLog->stageInfo[petsc_stageLog->curStage].eventLog->eventInfo[e].active) ? \ 364495fc317SBarry Smith (*PetscLogPLE)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4)) : 0)) 36577c4ece6SBarry Smith 366014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventGetFlops(PetscLogEvent,PetscLogDouble*); 367014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventZeroFlops(PetscLogEvent); 368003131ecSBarry Smith 369ce85283eSBarry Smith /* 370f621e05eSBarry Smith These are used internally in the PETSc routines to keep a count of MPI messages and 371f621e05eSBarry Smith their sizes. 372f621e05eSBarry Smith 373af0996ceSBarry Smith This does not work for MPI-Uni because our include/petsc/mpiuni/mpi.h file 374ce85283eSBarry Smith uses macros to defined the MPI operations. 37515308354SBarry Smith 37615308354SBarry Smith It does not work correctly from HP-UX because it processes the 377bb4af37aSBarry Smith macros in a way that sometimes it double counts, hence 378b6410449SSatish Balay PETSC_HAVE_BROKEN_RECURSIVE_MACRO 3797c1e34a4SSatish Balay 380f569fd43SBarry Smith It does not work with Windows because winmpich lacks MPI_Type_size() 381ce85283eSBarry Smith */ 382c9b973beSBarry Smith #if !defined(MPIUNI_H) && !defined(PETSC_HAVE_BROKEN_RECURSIVE_MACRO) && !defined (PETSC_HAVE_MPI_MISSING_TYPESIZE) 38377a39924SBarry Smith /* 38477a39924SBarry Smith Logging of MPI activities 38577a39924SBarry Smith */ 386*9fbee547SJacob Faibussowitsch static inline PetscErrorCode PetscMPITypeSize(PetscInt count,MPI_Datatype type,PetscLogDouble *length) 387f95db71bSBarry Smith { 388f12d10ccSJunchao Zhang PetscMPIInt typesize; 389f12d10ccSJunchao Zhang PetscErrorCode ierr; 3907d30fa4aSBarry Smith if (type == MPI_DATATYPE_NULL) return 0; 39155b25c41SPierre Jolivet ierr = MPI_Type_size(type,&typesize);CHKERRMPI(ierr); 392f12d10ccSJunchao Zhang *length += (PetscLogDouble) (count*typesize); 3939c72ff1eSSatish Balay return 0; 394f95db71bSBarry Smith } 39577a39924SBarry Smith 396*9fbee547SJacob Faibussowitsch static inline PetscErrorCode PetscMPITypeSizeComm(MPI_Comm comm,const PetscMPIInt *counts,MPI_Datatype type,PetscLogDouble *length) 3973b9284c0SMatthew G Knepley { 398f12d10ccSJunchao Zhang PetscMPIInt typesize,size,p; 399f12d10ccSJunchao Zhang PetscErrorCode ierr; 4003b9284c0SMatthew G Knepley 4017d30fa4aSBarry Smith if (type == MPI_DATATYPE_NULL) return 0; 402ffc4695bSBarry Smith ierr = MPI_Comm_size(comm,&size);CHKERRMPI(ierr); 403ffc4695bSBarry Smith ierr = MPI_Type_size(type,&typesize);CHKERRMPI(ierr); 404f12d10ccSJunchao Zhang for (p=0; p<size; ++p) { 405f12d10ccSJunchao Zhang *length += (PetscLogDouble) (counts[p]*typesize); 406f12d10ccSJunchao Zhang } 407f12d10ccSJunchao Zhang return 0; 408f12d10ccSJunchao Zhang } 409f12d10ccSJunchao Zhang 410*9fbee547SJacob Faibussowitsch static inline PetscErrorCode PetscMPITypeSizeCount(PetscInt n,const PetscMPIInt *counts,MPI_Datatype type,PetscLogDouble *length) 411f12d10ccSJunchao Zhang { 412f12d10ccSJunchao Zhang PetscMPIInt typesize,p; 413f12d10ccSJunchao Zhang PetscErrorCode ierr; 414f12d10ccSJunchao Zhang 415f12d10ccSJunchao Zhang if (type == MPI_DATATYPE_NULL) return 0; 416ffc4695bSBarry Smith ierr = MPI_Type_size(type,&typesize);CHKERRMPI(ierr); 417f12d10ccSJunchao Zhang for (p=0; p<n; ++p) { 418f12d10ccSJunchao Zhang *length += (PetscLogDouble) (counts[p]*typesize); 4193b9284c0SMatthew G Knepley } 4203b9284c0SMatthew G Knepley return 0; 4213b9284c0SMatthew G Knepley } 4223b9284c0SMatthew G Knepley 42372b11a25SBarry Smith /* 42472b11a25SBarry Smith Returns 1 if the communicator is parallel else zero 42572b11a25SBarry Smith */ 426*9fbee547SJacob Faibussowitsch static inline int PetscMPIParallelComm(MPI_Comm comm) 42772b11a25SBarry Smith { 42872b11a25SBarry Smith PetscMPIInt size; MPI_Comm_size(comm,&size); return size > 1; 42972b11a25SBarry Smith } 43072b11a25SBarry Smith 43177a39924SBarry Smith #define MPI_Irecv(buf,count,datatype,source,tag,comm,request) \ 432f12d10ccSJunchao Zhang ((petsc_irecv_ct++,0) || PetscMPITypeSize((count),(datatype),&(petsc_irecv_len)) || MPI_Irecv((buf),(count),(datatype),(source),(tag),(comm),(request))) 43315308354SBarry Smith 434c87b50c4SJunchao Zhang #define MPI_Irecv_c(buf,count,datatype,source,tag,comm,request) \ 435c87b50c4SJunchao Zhang ((petsc_irecv_ct++,0) || PetscMPITypeSize((count),(datatype),&(petsc_irecv_len)) || MPI_Irecv_c((buf),(count),(datatype),(source),(tag),(comm),(request))) 436c87b50c4SJunchao Zhang 43777a39924SBarry Smith #define MPI_Isend(buf,count,datatype,dest,tag,comm,request) \ 438f12d10ccSJunchao Zhang ((petsc_isend_ct++,0) || PetscMPITypeSize((count),(datatype),&(petsc_isend_len)) || MPI_Isend((buf),(count),(datatype),(dest),(tag),(comm),(request))) 43915308354SBarry Smith 440c87b50c4SJunchao Zhang #define MPI_Isend_c(buf,count,datatype,dest,tag,comm,request) \ 441c87b50c4SJunchao Zhang ((petsc_isend_ct++,0) || PetscMPITypeSize((count),(datatype),&(petsc_isend_len)) || MPI_Isend_c((buf),(count),(datatype),(dest),(tag),(comm),(request))) 442c87b50c4SJunchao Zhang 443457b14b1SJunchao Zhang #define MPI_Startall_irecv(count,datatype,number,requests) \ 444f12d10ccSJunchao Zhang ((petsc_irecv_ct += (PetscLogDouble)(number),0) || PetscMPITypeSize((count),(datatype),&(petsc_irecv_len)) || ((number) && MPI_Startall((number),(requests)))) 4450d4b0b6cSBarry Smith 446457b14b1SJunchao Zhang #define MPI_Startall_isend(count,datatype,number,requests) \ 447f12d10ccSJunchao Zhang ((petsc_isend_ct += (PetscLogDouble)(number),0) || PetscMPITypeSize((count),(datatype),&(petsc_isend_len)) || ((number) && MPI_Startall((number),(requests)))) 4480d4b0b6cSBarry Smith 449457b14b1SJunchao Zhang #define MPI_Start_isend(count,datatype,requests) \ 450f12d10ccSJunchao Zhang ((petsc_isend_ct++,0) || PetscMPITypeSize((count),(datatype),(&petsc_isend_len)) || MPI_Start((requests))) 4510d4b0b6cSBarry Smith 452ce85283eSBarry Smith #define MPI_Recv(buf,count,datatype,source,tag,comm,status) \ 453f12d10ccSJunchao Zhang ((petsc_recv_ct++,0) || PetscMPITypeSize((count),(datatype),(&petsc_recv_len)) || MPI_Recv((buf),(count),(datatype),(source),(tag),(comm),(status))) 45415308354SBarry Smith 455c87b50c4SJunchao Zhang #define MPI_Recv_c(buf,count,datatype,source,tag,comm,status) \ 456c87b50c4SJunchao Zhang ((petsc_recv_ct++,0) || PetscMPITypeSize((count),(datatype),(&petsc_recv_len)) || MPI_Recv_c((buf),(count),(datatype),(source),(tag),(comm),(status))) 457c87b50c4SJunchao Zhang 45877a39924SBarry Smith #define MPI_Send(buf,count,datatype,dest,tag,comm) \ 459f12d10ccSJunchao Zhang ((petsc_send_ct++,0) || PetscMPITypeSize((count),(datatype),(&petsc_send_len)) || MPI_Send((buf),(count),(datatype),(dest),(tag),(comm))) 46077a39924SBarry Smith 461c87b50c4SJunchao Zhang #define MPI_Send_c(buf,count,datatype,dest,tag,comm) \ 462c87b50c4SJunchao Zhang ((petsc_send_ct++,0) || PetscMPITypeSize((count),(datatype),(&petsc_send_len)) || MPI_Send_c((buf),(count),(datatype),(dest),(tag),(comm))) 463c87b50c4SJunchao Zhang 46477a39924SBarry Smith #define MPI_Wait(request,status) \ 4655fa69175SJed Brown ((petsc_wait_ct++,petsc_sum_of_waits_ct++,0) || MPI_Wait((request),(status))) 46677a39924SBarry Smith 46777a39924SBarry Smith #define MPI_Waitany(a,b,c,d) \ 4685fa69175SJed Brown ((petsc_wait_any_ct++,petsc_sum_of_waits_ct++,0) || MPI_Waitany((a),(b),(c),(d))) 46977a39924SBarry Smith 47077a39924SBarry Smith #define MPI_Waitall(count,array_of_requests,array_of_statuses) \ 4715fa69175SJed Brown ((petsc_wait_all_ct++,petsc_sum_of_waits_ct += (PetscLogDouble) (count),0) || MPI_Waitall((count),(array_of_requests),(array_of_statuses))) 47277a39924SBarry Smith 47377a39924SBarry Smith #define MPI_Allreduce(sendbuf,recvbuf,count,datatype,op,comm) \ 474df05ca09SBarry Smith (petsc_allreduce_ct += PetscMPIParallelComm((comm)),MPI_Allreduce((sendbuf),(recvbuf),(count),(datatype),(op),(comm))) 4753914022bSBarry Smith 47654fe5c21SBarry Smith #define MPI_Bcast(buffer,count,datatype,root,comm) \ 4775fa69175SJed Brown ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || MPI_Bcast((buffer),(count),(datatype),(root),(comm))) 47854fe5c21SBarry Smith 479d6e4c47cSJed Brown #define MPI_Reduce_scatter_block(sendbuf,recvbuf,recvcount,datatype,op,comm) \ 4805fa69175SJed Brown ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || MPI_Reduce_scatter_block((sendbuf),(recvbuf),(recvcount),(datatype),(op),(comm))) 481d6e4c47cSJed Brown 4823b9284c0SMatthew G Knepley #define MPI_Alltoall(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,comm) \ 483f12d10ccSJunchao Zhang ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Alltoall((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(comm))) 4843b9284c0SMatthew G Knepley 4853b9284c0SMatthew G Knepley #define MPI_Alltoallv(sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm) \ 486f12d10ccSJunchao 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))) 4873b9284c0SMatthew G Knepley 48801faf4e4SMatthew Knepley #define MPI_Allgather(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,comm) \ 4895fa69175SJed Brown ((petsc_gather_ct += PetscMPIParallelComm((comm)),0) || MPI_Allgather((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(comm))) 49001faf4e4SMatthew Knepley 49101faf4e4SMatthew Knepley #define MPI_Allgatherv(sendbuf,sendcount,sendtype,recvbuf,recvcount,displs,recvtype,comm) \ 4925fa69175SJed Brown ((petsc_gather_ct += PetscMPIParallelComm((comm)),0) || MPI_Allgatherv((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(displs),(recvtype),(comm))) 49301faf4e4SMatthew Knepley 49401faf4e4SMatthew Knepley #define MPI_Gather(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,root,comm) \ 495f12d10ccSJunchao Zhang ((petsc_gather_ct++,0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Gather((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm))) 49601faf4e4SMatthew Knepley 49701faf4e4SMatthew Knepley #define MPI_Gatherv(sendbuf,sendcount,sendtype,recvbuf,recvcount,displs,recvtype,root,comm) \ 498f12d10ccSJunchao Zhang ((petsc_gather_ct++,0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Gatherv((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(displs),(recvtype),(root),(comm))) 49901faf4e4SMatthew Knepley 50001faf4e4SMatthew Knepley #define MPI_Scatter(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,root,comm) \ 501f12d10ccSJunchao Zhang ((petsc_scatter_ct++,0) || PetscMPITypeSize((recvcount),(recvtype),(&petsc_recv_len)) || MPI_Scatter((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm))) 50201faf4e4SMatthew Knepley 50301faf4e4SMatthew Knepley #define MPI_Scatterv(sendbuf,sendcount,displs,sendtype,recvbuf,recvcount,recvtype,root,comm) \ 504f12d10ccSJunchao Zhang ((petsc_scatter_ct++,0) || PetscMPITypeSize((recvcount),(recvtype),(&petsc_recv_len)) || MPI_Scatterv((sendbuf),(sendcount),(displs),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm))) 505f12d10ccSJunchao Zhang 506f12d10ccSJunchao Zhang #define MPI_Ialltoall(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,comm,request) \ 507f12d10ccSJunchao Zhang ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Ialltoall((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(comm),(request))) 508f12d10ccSJunchao Zhang 509f12d10ccSJunchao Zhang #define MPI_Ialltoallv(sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm,request) \ 510f12d10ccSJunchao 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))) 511f12d10ccSJunchao Zhang 512f12d10ccSJunchao Zhang #define MPI_Iallgather(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,comm,request) \ 513f12d10ccSJunchao Zhang ((petsc_gather_ct += PetscMPIParallelComm((comm)),0) || MPI_Iallgather((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(comm),(request))) 514f12d10ccSJunchao Zhang 515f12d10ccSJunchao Zhang #define MPI_Iallgatherv(sendbuf,sendcount,sendtype,recvbuf,recvcount,displs,recvtype,comm,request) \ 516f12d10ccSJunchao Zhang ((petsc_gather_ct += PetscMPIParallelComm((comm)),0) || MPI_Iallgatherv((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(displs),(recvtype),(comm),(request))) 517f12d10ccSJunchao Zhang 518f12d10ccSJunchao Zhang #define MPI_Igather(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,root,comm,request) \ 519f12d10ccSJunchao Zhang ((petsc_gather_ct++,0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Igather((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm),(request))) 520f12d10ccSJunchao Zhang 521f12d10ccSJunchao Zhang #define MPI_Igatherv(sendbuf,sendcount,sendtype,recvbuf,recvcount,displs,recvtype,root,comm,request) \ 522f12d10ccSJunchao Zhang ((petsc_gather_ct++,0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Igatherv((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(displs),(recvtype),(root),(comm),(request))) 523f12d10ccSJunchao Zhang 524f12d10ccSJunchao Zhang #define MPI_Iscatter(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,root,comm,request) \ 525f12d10ccSJunchao Zhang ((petsc_scatter_ct++,0) || PetscMPITypeSize((recvcount),(recvtype),(&petsc_recv_len)) || MPI_Iscatter((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm),(request))) 526f12d10ccSJunchao Zhang 527f12d10ccSJunchao Zhang #define MPI_Iscatterv(sendbuf,sendcount,displs,sendtype,recvbuf,recvcount,recvtype,root,comm,request) \ 528f12d10ccSJunchao Zhang ((petsc_scatter_ct++,0) || PetscMPITypeSize((recvcount),(recvtype),(&petsc_recv_len)) || MPI_Iscatterv((sendbuf),(sendcount),(displs),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm),(request))) 529f12d10ccSJunchao Zhang 5300d4b0b6cSBarry Smith #else 5310d4b0b6cSBarry Smith 532457b14b1SJunchao Zhang #define MPI_Startall_irecv(count,datatype,number,requests) \ 533ea1c87f7SJunchao Zhang ((number) && MPI_Startall((number),(requests))) 5340d4b0b6cSBarry Smith 535457b14b1SJunchao Zhang #define MPI_Startall_isend(count,datatype,number,requests) \ 536ea1c87f7SJunchao Zhang ((number) && MPI_Startall((number),(requests))) 5370d4b0b6cSBarry Smith 538457b14b1SJunchao Zhang #define MPI_Start_isend(count,datatype,requests) \ 5395fa69175SJed Brown (MPI_Start((requests))) 5400d4b0b6cSBarry Smith 541c9b973beSBarry Smith #endif /* !MPIUNI_H && ! PETSC_HAVE_BROKEN_RECURSIVE_MACRO */ 542614700edSBarry Smith 543df8cf0b5SBarry Smith #else /* ---Logging is turned off --------------------------------------------*/ 544614700edSBarry Smith 545e3ed9ee7SBarry Smith #define PetscLogMemory PETSC_FALSE 546e3ed9ee7SBarry Smith 547b0a32e0cSBarry Smith #define PetscLogFlops(n) 0 548fa2bb9feSLisandro Dalcin #define PetscGetFlops(a) (*(a) = 0.0,0) 549614700edSBarry Smith 550fa2bb9feSLisandro Dalcin #define PetscLogStageRegister(a,b) 0 551fa2bb9feSLisandro Dalcin #define PetscLogStagePush(a) 0 552fa2bb9feSLisandro Dalcin #define PetscLogStagePop() 0 553f5d6ab90SLisandro Dalcin #define PetscLogStageSetActive(a,b) 0 554f5d6ab90SLisandro Dalcin #define PetscLogStageGetActive(a,b) 0 555f5d6ab90SLisandro Dalcin #define PetscLogStageGetVisible(a,b) 0 556f5d6ab90SLisandro Dalcin #define PetscLogStageSetVisible(a,b) 0 557f5d6ab90SLisandro Dalcin #define PetscLogStageGetId(a,b) (*(b)=0,0) 558f5d6ab90SLisandro Dalcin 559fa2bb9feSLisandro Dalcin #define PetscLogEventRegister(a,b,c) 0 560fa2bb9feSLisandro Dalcin #define PetscLogEventSetCollective(a,b) 0 561fa2bb9feSLisandro Dalcin #define PetscLogEventIncludeClass(a) 0 562fa2bb9feSLisandro Dalcin #define PetscLogEventExcludeClass(a) 0 563b0a32e0cSBarry Smith #define PetscLogEventActivate(a) 0 564b0a32e0cSBarry Smith #define PetscLogEventDeactivate(a) 0 565c00cb57fSBarry Smith #define PetscLogEventDeactivatePush(a) 0 566c00cb57fSBarry Smith #define PetscLogEventDeactivatePop(a) 0 567b0a32e0cSBarry Smith #define PetscLogEventActivateClass(a) 0 568b0a32e0cSBarry Smith #define PetscLogEventDeactivateClass(a) 0 569e8e7597cSSatish Balay #define PetscLogEventSetActiveAll(a,b) 0 57046eb3cd7SSatish Balay #define PetscLogEventGetId(a,b) (*(b)=0,0) 571561b2a67SSatish Balay #define PetscLogEventGetPerfInfo(a,b,c) 0 5729c702cfcSMatthew G. Knepley #define PetscLogEventSetDof(a,b,c) 0 5739c702cfcSMatthew G. Knepley #define PetscLogEventSetError(a,b,c) 0 57477c4ece6SBarry Smith 5758b5db460SBarry Smith #define PetscLogPLB 0 5768b5db460SBarry Smith #define PetscLogPLE 0 5778b5db460SBarry Smith #define PetscLogPHC 0 5788b5db460SBarry Smith #define PetscLogPHD 0 579f5d6ab90SLisandro Dalcin 580efee365bSSatish Balay #define PetscLogObjectParents(p,n,c) 0 58152e6d16bSBarry Smith #define PetscLogObjectCreate(h) 0 58252e6d16bSBarry Smith #define PetscLogObjectDestroy(h) 0 5833ca90d2dSJacob Faibussowitsch PETSC_EXTERN PetscErrorCode PetscLogObjectState(PetscObject,const char[],...) PETSC_ATTRIBUTE_FORMAT(2,3); 584fa2bb9feSLisandro Dalcin 585bb1d7374SBarry Smith #define PetscLogDefaultBegin() 0 586b0a32e0cSBarry Smith #define PetscLogAllBegin() 0 587bb1d7374SBarry Smith #define PetscLogNestedBegin() 0 588fa2bb9feSLisandro Dalcin #define PetscLogTraceBegin(file) 0 58973fda44aSBarry Smith #define PetscLogActions(a) 0 590fa2bb9feSLisandro Dalcin #define PetscLogObjects(a) 0 591fa2bb9feSLisandro Dalcin #define PetscLogSetThreshold(a,b) 0 592fa2bb9feSLisandro Dalcin #define PetscLogSet(lb,le) 0 593fa2bb9feSLisandro Dalcin 594fa2bb9feSLisandro Dalcin #define PetscLogView(viewer) 0 595fa2bb9feSLisandro Dalcin #define PetscLogViewFromOptions() 0 596fa2bb9feSLisandro Dalcin #define PetscLogDump(c) 0 597fa2bb9feSLisandro Dalcin 59862872c28SLisandro Dalcin #define PetscLogEventSync(e,comm) 0 599fa2bb9feSLisandro Dalcin #define PetscLogEventBegin(e,o1,o2,o3,o4) 0 600fa2bb9feSLisandro Dalcin #define PetscLogEventEnd(e,o1,o2,o3,o4) 0 601ce6058e1SBarry Smith 60223c86b1bSSatish Balay /* If PETSC_USE_LOG is NOT defined, these still need to be! */ 60323c86b1bSSatish Balay #define MPI_Startall_irecv(count,datatype,number,requests) ((number) && MPI_Startall(number,requests)) 60423c86b1bSSatish Balay #define MPI_Startall_isend(count,datatype,number,requests) ((number) && MPI_Startall(number,requests)) 60523c86b1bSSatish Balay #define MPI_Start_isend(count,datatype,requests) MPI_Start(requests) 60623c86b1bSSatish Balay 60723c86b1bSSatish Balay #endif /* PETSC_USE_LOG */ 60823c86b1bSSatish Balay 60923c86b1bSSatish Balay #if defined (PETSC_USE_LOG) && defined(PETSC_HAVE_DEVICE) 61023c86b1bSSatish Balay 61123c86b1bSSatish Balay /* Global GPU counters */ 61223c86b1bSSatish Balay PETSC_EXTERN PetscLogDouble petsc_ctog_ct; 61323c86b1bSSatish Balay PETSC_EXTERN PetscLogDouble petsc_gtoc_ct; 61423c86b1bSSatish Balay PETSC_EXTERN PetscLogDouble petsc_ctog_sz; 61523c86b1bSSatish Balay PETSC_EXTERN PetscLogDouble petsc_gtoc_sz; 61623c86b1bSSatish Balay PETSC_EXTERN PetscLogDouble petsc_ctog_ct_scalar; 61723c86b1bSSatish Balay PETSC_EXTERN PetscLogDouble petsc_gtoc_ct_scalar; 61823c86b1bSSatish Balay PETSC_EXTERN PetscLogDouble petsc_ctog_sz_scalar; 61923c86b1bSSatish Balay PETSC_EXTERN PetscLogDouble petsc_gtoc_sz_scalar; 62023c86b1bSSatish Balay PETSC_EXTERN PetscLogDouble petsc_gflops; 62123c86b1bSSatish Balay PETSC_EXTERN PetscLogDouble petsc_gtime; 62223c86b1bSSatish Balay 623*9fbee547SJacob Faibussowitsch static inline PetscErrorCode PetscLogCpuToGpu(PetscLogDouble size) 62423c86b1bSSatish Balay { 62523c86b1bSSatish Balay PetscFunctionBegin; 62623c86b1bSSatish Balay petsc_ctog_ct += 1; 62723c86b1bSSatish Balay petsc_ctog_sz += size; 62823c86b1bSSatish Balay PetscFunctionReturn(0); 62923c86b1bSSatish Balay } 63023c86b1bSSatish Balay 631*9fbee547SJacob Faibussowitsch static inline PetscErrorCode PetscLogGpuToCpu(PetscLogDouble size) 63223c86b1bSSatish Balay { 63323c86b1bSSatish Balay PetscFunctionBegin; 63423c86b1bSSatish Balay petsc_gtoc_ct += 1; 63523c86b1bSSatish Balay petsc_gtoc_sz += size; 63623c86b1bSSatish Balay PetscFunctionReturn(0); 63723c86b1bSSatish Balay } 63823c86b1bSSatish Balay 639*9fbee547SJacob Faibussowitsch static inline PetscErrorCode PetscLogCpuToGpuScalar(PetscLogDouble size) 64023c86b1bSSatish Balay { 64123c86b1bSSatish Balay PetscFunctionBegin; 64223c86b1bSSatish Balay petsc_ctog_ct_scalar += 1; 64323c86b1bSSatish Balay petsc_ctog_sz_scalar += size; 64423c86b1bSSatish Balay PetscFunctionReturn(0); 64523c86b1bSSatish Balay } 64623c86b1bSSatish Balay 647*9fbee547SJacob Faibussowitsch static inline PetscErrorCode PetscLogGpuToCpuScalar(PetscLogDouble size) 64823c86b1bSSatish Balay { 64923c86b1bSSatish Balay PetscFunctionBegin; 65023c86b1bSSatish Balay petsc_gtoc_ct_scalar += 1; 65123c86b1bSSatish Balay petsc_gtoc_sz_scalar += size; 65223c86b1bSSatish Balay PetscFunctionReturn(0); 65323c86b1bSSatish Balay } 65423c86b1bSSatish Balay 65523c86b1bSSatish Balay /*@C 65623c86b1bSSatish Balay PetscLogGpuFlops - Log how many flops are performed in a calculation on the device 65723c86b1bSSatish Balay 65823c86b1bSSatish Balay Input Parameter: 65923c86b1bSSatish Balay . flops - the number of flops 66023c86b1bSSatish Balay 66123c86b1bSSatish Balay Notes: 66223c86b1bSSatish Balay To limit the chance of integer overflow when multiplying by a constant, represent the constant as a double, 66323c86b1bSSatish Balay not an integer. Use PetscLogFlops(4.0*n) not PetscLogFlops(4*n) 66423c86b1bSSatish Balay 66523c86b1bSSatish Balay Level: intermediate 66623c86b1bSSatish Balay 66723c86b1bSSatish Balay .seealso: PetscLogView(), PetscLogFlops(), PetscLogGpuTimeBegin(), PetscLogGpuTimeEnd() 66823c86b1bSSatish Balay @*/ 669*9fbee547SJacob Faibussowitsch static inline PetscErrorCode PetscLogGpuFlops(PetscLogDouble n) 67023c86b1bSSatish Balay { 67123c86b1bSSatish Balay PetscFunctionBegin; 6729ace16cdSJacob Faibussowitsch PetscAssert(n >= 0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Cannot log negative flops"); 67323c86b1bSSatish Balay petsc_TotalFlops += PETSC_FLOPS_PER_OP*n; 67423c86b1bSSatish Balay petsc_gflops += PETSC_FLOPS_PER_OP*n; 67523c86b1bSSatish Balay PetscFunctionReturn(0); 67623c86b1bSSatish Balay } 67723c86b1bSSatish Balay 678*9fbee547SJacob Faibussowitsch static inline PetscErrorCode PetscLogGpuTimeAdd(PetscLogDouble t) 67923c86b1bSSatish Balay { 68023c86b1bSSatish Balay PetscFunctionBegin; 68123c86b1bSSatish Balay petsc_gtime += t; 68223c86b1bSSatish Balay PetscFunctionReturn(0); 68323c86b1bSSatish Balay } 68423c86b1bSSatish Balay 68523c86b1bSSatish Balay PETSC_EXTERN PetscErrorCode PetscLogGpuTimeBegin(void); 68623c86b1bSSatish Balay PETSC_EXTERN PetscErrorCode PetscLogGpuTimeEnd(void); 68723c86b1bSSatish Balay 68823c86b1bSSatish Balay #else 68923c86b1bSSatish Balay 6908157aad8SJunchao Zhang #define PetscLogCpuToGpu(a) 0 6918157aad8SJunchao Zhang #define PetscLogGpuToCpu(a) 0 69245c4b7c1SBarry Smith #define PetscLogCpuToGpuScalar(a) 0 69345c4b7c1SBarry Smith #define PetscLogGpuToCpuScalar(a) 0 6948157aad8SJunchao Zhang #define PetscLogGpuFlops(a) 0 69553b96e71SJacob Faibussowitsch #define PetscLogGpuTimeAdd(a) 0 6968157aad8SJunchao Zhang #define PetscLogGpuTimeBegin() 0 6978157aad8SJunchao Zhang #define PetscLogGpuTimeEnd() 0 6988157aad8SJunchao Zhang 69923c86b1bSSatish Balay #endif /* PETSC_USE_LOG && PETSC_HAVE_DEVICE */ 7006daaf66cSBarry Smith 701204a6943SJed Brown #define PetscPreLoadBegin(flag,name) \ 702204a6943SJed Brown do {\ 703204a6943SJed Brown PetscBool PetscPreLoading = flag;\ 704204a6943SJed Brown int PetscPreLoadMax,PetscPreLoadIt;\ 705166c7f25SBarry Smith PetscLogStage _stageNum;\ 7068cbcd9ccSBarry Smith PetscErrorCode _3_ierr; \ 707c5929fdfSBarry Smith _3_ierr = PetscOptionsGetBool(NULL,NULL,"-preload",&PetscPreLoading,NULL);CHKERRQ(_3_ierr); \ 708204a6943SJed Brown PetscPreLoadMax = (int)(PetscPreLoading);\ 709204a6943SJed Brown PetscPreLoadingUsed = PetscPreLoading ? PETSC_TRUE : PetscPreLoadingUsed;\ 710204a6943SJed Brown for (PetscPreLoadIt=0; PetscPreLoadIt<=PetscPreLoadMax; PetscPreLoadIt++) {\ 711204a6943SJed Brown PetscPreLoadingOn = PetscPreLoading;\ 7120298fd71SBarry Smith _3_ierr = PetscBarrier(NULL);CHKERRQ(_3_ierr);\ 713204a6943SJed Brown if (PetscPreLoadIt>0) {\ 7148e58c17dSMatthew Knepley _3_ierr = PetscLogStageGetId(name,&_stageNum);CHKERRQ(_3_ierr);\ 7158e58c17dSMatthew Knepley } else {\ 716a3bc4eb9SBarry Smith _3_ierr = PetscLogStageRegister(name,&_stageNum);CHKERRQ(_3_ierr); \ 7178e58c17dSMatthew Knepley }\ 718204a6943SJed Brown _3_ierr = PetscLogStageSetActive(_stageNum,(PetscBool)(!PetscPreLoadMax || PetscPreLoadIt));\ 7198e58c17dSMatthew Knepley _3_ierr = PetscLogStagePush(_stageNum);CHKERRQ(_3_ierr); 7208e58c17dSMatthew Knepley 721204a6943SJed Brown #define PetscPreLoadEnd() \ 722043328b6SSatish Balay _3_ierr = PetscLogStagePop();CHKERRQ(_3_ierr);\ 723204a6943SJed Brown PetscPreLoading = PETSC_FALSE;\ 724043328b6SSatish Balay }\ 725204a6943SJed Brown } while (0) 7268e58c17dSMatthew Knepley 727204a6943SJed Brown #define PetscPreLoadStage(name) do { \ 728043328b6SSatish Balay _3_ierr = PetscLogStagePop();CHKERRQ(_3_ierr); \ 729204a6943SJed Brown if (PetscPreLoadIt>0) { \ 7308e58c17dSMatthew Knepley _3_ierr = PetscLogStageGetId(name,&_stageNum);CHKERRQ(_3_ierr); \ 7318e58c17dSMatthew Knepley } else { \ 732a3bc4eb9SBarry Smith _3_ierr = PetscLogStageRegister(name,&_stageNum);CHKERRQ(_3_ierr); \ 7338e58c17dSMatthew Knepley } \ 734204a6943SJed Brown _3_ierr = PetscLogStageSetActive(_stageNum,(PetscBool)(!PetscPreLoadMax || PetscPreLoadIt)); \ 735204a6943SJed Brown _3_ierr = PetscLogStagePush(_stageNum);CHKERRQ(_3_ierr); \ 736204a6943SJed Brown } while (0) 737e9fa29b7SSatish Balay 738ad14c47eSJed Brown /* some vars for logging */ 739ad14c47eSJed Brown PETSC_EXTERN PetscBool PetscPreLoadingUsed; /* true if we are or have done preloading */ 740ad14c47eSJed Brown PETSC_EXTERN PetscBool PetscPreLoadingOn; /* true if we are currently in a preloading calculation */ 741ad14c47eSJed Brown 74297bb86f7SLois Curfman McInnes #endif 743