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 9431d06abdSBarry Smith /* We must make the following structures available to access the event 9531d06abdSBarry Smith activation flags in the PetscLogEventBegin/End() macros. These are not part of the PETSc public 9631d06abdSBarry Smith API and are not intended to be used by other parts of PETSc or by users. 97d49d4b11SBarry Smith 98fa2bb9feSLisandro Dalcin The code that manipulates these structures is in src/sys/logging/utils. 9931d06abdSBarry Smith */ 100aa213bdcSJed Brown typedef struct _n_PetscIntStack *PetscIntStack; 1016a6a9b46SSatish Balay 102217044c2SLisandro Dalcin /* -----------------------------------------------------------------------------------------------------*/ 103c60ec953SBarry Smith /* 104aa213bdcSJed Brown PetscClassRegInfo, PetscClassPerfInfo - Each class has two data structures associated with it. The first has 105c60ec953SBarry Smith static information about it, the second collects statistics on how many objects of the class are created, 106c60ec953SBarry Smith how much memory they use, etc. 107c60ec953SBarry Smith 108aa213bdcSJed Brown PetscClassRegLog, PetscClassPerfLog - arrays of the PetscClassRegInfo and PetscClassPerfInfo for all classes. 109c60ec953SBarry Smith */ 1106a6a9b46SSatish Balay typedef struct { 111c60ec953SBarry Smith char *name; /* The class name */ 112c60ec953SBarry Smith PetscClassId classid; /* The integer identifying this class */ 113aa213bdcSJed Brown } PetscClassRegInfo; 1146a6a9b46SSatish Balay 1156a6a9b46SSatish Balay typedef struct { 1160700a824SBarry Smith PetscClassId id; /* The integer identifying this class */ 1176a6a9b46SSatish Balay int creations; /* The number of objects of this class created */ 1186a6a9b46SSatish Balay int destructions; /* The number of objects of this class destroyed */ 1196a6a9b46SSatish Balay PetscLogDouble mem; /* The total memory allocated by objects of this class */ 1206a6a9b46SSatish Balay PetscLogDouble descMem; /* The total memory allocated by descendents of these objects */ 121aa213bdcSJed Brown } PetscClassPerfInfo; 1226a6a9b46SSatish Balay 123aa213bdcSJed Brown typedef struct _n_PetscClassRegLog *PetscClassRegLog; 124aa213bdcSJed Brown struct _n_PetscClassRegLog { 125c60ec953SBarry Smith int numClasses; /* The number of classes registered */ 126c60ec953SBarry Smith int maxClasses; /* The maximum number of classes */ 127aa213bdcSJed Brown PetscClassRegInfo *classInfo; /* The structure for class information (classids are monotonicly increasing) */ 128c60ec953SBarry Smith }; 1296a6a9b46SSatish Balay 130aa213bdcSJed Brown typedef struct _n_PetscClassPerfLog *PetscClassPerfLog; 131aa213bdcSJed Brown struct _n_PetscClassPerfLog { 132c60ec953SBarry Smith int numClasses; /* The number of logging classes */ 133c60ec953SBarry Smith int maxClasses; /* The maximum number of classes */ 134aa213bdcSJed Brown PetscClassPerfInfo *classInfo; /* The structure for class information (classids are monotonicly increasing) */ 135c60ec953SBarry Smith }; 136c60ec953SBarry Smith /* -----------------------------------------------------------------------------------------------------*/ 137c60ec953SBarry Smith /* 138aa213bdcSJed Brown PetscEventRegInfo, PetscEventPerfInfo - Each event has two data structures associated with it. The first has 139c60ec953SBarry Smith static information about it, the second collects statistics on how many times the event is used, how 140c60ec953SBarry Smith much time it takes, etc. 141c60ec953SBarry Smith 142aa213bdcSJed Brown PetscEventRegLog, PetscEventPerfLog - an array of all PetscEventRegInfo and PetscEventPerfInfo for all events. There is one 143c60ec953SBarry Smith of these for each stage. 144c60ec953SBarry Smith 145c60ec953SBarry Smith */ 1466a6a9b46SSatish Balay typedef struct { 1476a6a9b46SSatish Balay char *name; /* The name of this event */ 148c60ec953SBarry Smith PetscClassId classid; /* The class the event is associated with */ 149217044c2SLisandro Dalcin PetscBool collective; /* Flag this event as collective */ 1506a6a9b46SSatish Balay #if defined (PETSC_HAVE_MPE) 1516a6a9b46SSatish Balay int mpe_id_begin; /* MPE IDs that define the event */ 1526a6a9b46SSatish Balay int mpe_id_end; 1536a6a9b46SSatish Balay #endif 154aa213bdcSJed Brown } PetscEventRegInfo; 1556a6a9b46SSatish Balay 156c60ec953SBarry Smith typedef struct { 157c60ec953SBarry Smith int id; /* The integer identifying this event */ 158ace3abfcSBarry Smith PetscBool active; /* The flag to activate logging */ 159ace3abfcSBarry Smith PetscBool visible; /* The flag to print info in summary */ 160c60ec953SBarry Smith int depth; /* The nesting depth of the event call */ 161c60ec953SBarry Smith int count; /* The number of times this event was executed */ 162a42e5b54SMatthew G Knepley PetscLogDouble flops, flops2, flopsTmp; /* The flops and flops^2 used in this event */ 163a42e5b54SMatthew G Knepley PetscLogDouble time, time2, timeTmp; /* The time and time^2 taken for this event */ 164217044c2SLisandro Dalcin PetscLogDouble syncTime; /* The synchronization barrier time */ 165891e75beSMatthew G. Knepley PetscLogDouble dof[8]; /* The number of degrees of freedom associated with this event */ 1665d68e14cSMatthew G. Knepley PetscLogDouble errors[8]; /* The errors (user-defined) associated with this event */ 167c60ec953SBarry Smith PetscLogDouble numMessages; /* The number of messages in this event */ 168c60ec953SBarry Smith PetscLogDouble messageLength; /* The total message lengths in this event */ 169c60ec953SBarry Smith PetscLogDouble numReductions; /* The number of reductions in this event */ 170e3ed9ee7SBarry Smith PetscLogDouble memIncrease; /* How much the resident memory has increased in this event */ 171e3ed9ee7SBarry Smith PetscLogDouble mallocIncrease;/* How much the maximum malloced space has increased in this event */ 172e3ed9ee7SBarry Smith PetscLogDouble mallocSpace; /* How much the space was malloced and kept during this event */ 173e3ed9ee7SBarry Smith PetscLogDouble mallocIncreaseEvent; /* Maximum of the high water mark with in event minus memory available at the end of the event */ 174bec0b493Shannah_mairs #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA) 175bec0b493Shannah_mairs PetscLogDouble CpuToGpuCount; /* The total number of CPU to GPU copies */ 176bec0b493Shannah_mairs PetscLogDouble GpuToCpuCount; /* The total number of GPU to CPU copies */ 177bec0b493Shannah_mairs PetscLogDouble CpuToGpuSize; /* The total size of CPU to GPU copies */ 178bec0b493Shannah_mairs PetscLogDouble GpuToCpuSize; /* The total size of GPU to CPU copies */ 179958c4211Shannah_mairs PetscLogDouble GpuFlops; /* The flops done on a GPU in this event */ 180958c4211Shannah_mairs PetscLogDouble GpuTime; /* The time spent on a GPU in this event */ 181bec0b493Shannah_mairs #endif 182aa213bdcSJed Brown } PetscEventPerfInfo; 183c60ec953SBarry Smith 184aa213bdcSJed Brown typedef struct _n_PetscEventRegLog *PetscEventRegLog; 185aa213bdcSJed Brown struct _n_PetscEventRegLog { 1866a6a9b46SSatish Balay int numEvents; /* The number of registered events */ 1876a6a9b46SSatish Balay int maxEvents; /* The maximum number of events */ 188aa213bdcSJed Brown PetscEventRegInfo *eventInfo; /* The registration information for each event */ 1896a6a9b46SSatish Balay }; 1906a6a9b46SSatish Balay 191aa213bdcSJed Brown typedef struct _n_PetscEventPerfLog *PetscEventPerfLog; 192aa213bdcSJed Brown struct _n_PetscEventPerfLog { 1936a6a9b46SSatish Balay int numEvents; /* The number of logging events */ 1946a6a9b46SSatish Balay int maxEvents; /* The maximum number of events */ 195aa213bdcSJed Brown PetscEventPerfInfo *eventInfo; /* The performance information for each event */ 1966a6a9b46SSatish Balay }; 197c60ec953SBarry Smith /* ------------------------------------------------------------------------------------------------------------*/ 198c60ec953SBarry Smith /* 199aa213bdcSJed Brown PetscStageInfo - Contains all the information about a particular stage. 2006a6a9b46SSatish Balay 201aa213bdcSJed Brown PetscStageLog - An array of PetscStageInfo for each registered stage. There is a single one of these in the code. 202c60ec953SBarry Smith */ 203aa213bdcSJed Brown typedef struct _PetscStageInfo { 2046a6a9b46SSatish Balay char *name; /* The stage name */ 205ace3abfcSBarry Smith PetscBool used; /* The stage was pushed on this processor */ 206aa213bdcSJed Brown PetscEventPerfInfo perfInfo; /* The stage performance information */ 207aa213bdcSJed Brown PetscEventPerfLog eventLog; /* The event information for this stage */ 208aa213bdcSJed Brown PetscClassPerfLog classLog; /* The class information for this stage */ 209aa213bdcSJed Brown } PetscStageInfo; 2106a6a9b46SSatish Balay 211aa213bdcSJed Brown typedef struct _n_PetscStageLog *PetscStageLog; 212aa213bdcSJed Brown struct _n_PetscStageLog { 2136a6a9b46SSatish Balay int numStages; /* The number of registered stages */ 2146a6a9b46SSatish Balay int maxStages; /* The maximum number of stages */ 215aa213bdcSJed Brown PetscIntStack stack; /* The stack for active stages */ 216aa213bdcSJed Brown int curStage; /* The current stage (only used in macros so we don't call PetscIntStackTop) */ 217aa213bdcSJed Brown PetscStageInfo *stageInfo; /* The information for each stage */ 218aa213bdcSJed Brown PetscEventRegLog eventLog; /* The registered events */ 219aa213bdcSJed Brown PetscClassRegLog classLog; /* The registered classes */ 2206a6a9b46SSatish Balay }; 221217044c2SLisandro Dalcin /* -----------------------------------------------------------------------------------------------------*/ 222fee518ebSJed Brown 2233bb1ff40SBarry Smith PETSC_EXTERN PetscErrorCode PetscLogObjectParent(PetscObject,PetscObject); 2243bb1ff40SBarry Smith PETSC_EXTERN PetscErrorCode PetscLogObjectMemory(PetscObject,PetscLogDouble); 2253bb1ff40SBarry Smith 226aa482453SBarry Smith #if defined(PETSC_USE_LOG) /* --- Logging is turned on --------------------------------*/ 227aeeda9f6SJohn Fettig PETSC_EXTERN PetscStageLog petsc_stageLog; 228f5d6ab90SLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogGetStageLog(PetscStageLog*); 229f5d6ab90SLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscStageLogGetCurrent(PetscStageLog,int*); 230f5d6ab90SLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscStageLogGetEventPerfLog(PetscStageLog,int,PetscEventPerfLog*); 231614700edSBarry Smith 232da63de55SLois Curfman McInnes /* 233da63de55SLois Curfman McInnes Flop counting: We count each arithmetic operation (e.g., addition, multiplication) separately. 234da63de55SLois Curfman McInnes 235da63de55SLois Curfman McInnes For the complex numbers version, note that 236da63de55SLois Curfman McInnes 1 complex addition = 2 flops 237da63de55SLois Curfman McInnes 1 complex multiplication = 6 flops, 238da63de55SLois Curfman McInnes where we define 1 flop as that for a double precision scalar. We roughly approximate 239da63de55SLois Curfman McInnes flop counting for complex numbers by multiplying the total flops by 4; this corresponds 240da63de55SLois Curfman McInnes to the assumption that we're counting mostly additions and multiplications -- and 241da63de55SLois Curfman McInnes roughly the same number of each. More accurate counting could be done by distinguishing 242da63de55SLois Curfman McInnes among the various arithmetic operations. 243da63de55SLois Curfman McInnes */ 244da63de55SLois Curfman McInnes 245aa482453SBarry Smith #if defined(PETSC_USE_COMPLEX) 246542d4b3fSSatish Balay #define PETSC_FLOPS_PER_OP 4.0 247da63de55SLois Curfman McInnes #else 248542d4b3fSSatish Balay #define PETSC_FLOPS_PER_OP 1.0 249bf3909cdSBarry Smith #endif 250bf3909cdSBarry Smith 251dd19b90bSBarry Smith PETSC_STATIC_INLINE PetscErrorCode PetscLogFlops(PetscLogDouble n) 252dd19b90bSBarry Smith { 253dd19b90bSBarry Smith PetscFunctionBegin; 254bf3909cdSBarry Smith #if defined(PETSC_USE_DEBUG) 255dd19b90bSBarry Smith if (n < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Cannot log negative flops"); 256da63de55SLois Curfman McInnes #endif 257dd19b90bSBarry Smith petsc_TotalFlops += PETSC_FLOPS_PER_OP*n; 258dd19b90bSBarry Smith PetscFunctionReturn(0); 259dd19b90bSBarry Smith } 260fa2bb9feSLisandro Dalcin 261bec0b493Shannah_mairs #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA) 262*8df2f0b1SJunchao Zhang /* Global GPU counters */ 263*8df2f0b1SJunchao Zhang PETSC_EXTERN PetscLogDouble petsc_ctog_ct; 264*8df2f0b1SJunchao Zhang PETSC_EXTERN PetscLogDouble petsc_gtoc_ct; 265*8df2f0b1SJunchao Zhang PETSC_EXTERN PetscLogDouble petsc_ctog_sz; 266*8df2f0b1SJunchao Zhang PETSC_EXTERN PetscLogDouble petsc_gtoc_sz; 267*8df2f0b1SJunchao Zhang PETSC_EXTERN PetscLogDouble petsc_gflops; 268*8df2f0b1SJunchao Zhang PETSC_EXTERN PetscLogDouble petsc_gtime; 269*8df2f0b1SJunchao Zhang #if defined(PETSC_USE_DEBUG) 270*8df2f0b1SJunchao Zhang PETSC_EXTERN PetscBool petsc_gtime_inuse; 271*8df2f0b1SJunchao Zhang #endif 272*8df2f0b1SJunchao Zhang 2734db21f4bSStefano Zampini PETSC_STATIC_INLINE PetscErrorCode PetscLogCpuToGpu(PetscLogDouble size) 2744db21f4bSStefano Zampini { 275958c4211Shannah_mairs PetscFunctionBegin; 276bec0b493Shannah_mairs petsc_ctog_ct += 1; 277bec0b493Shannah_mairs petsc_ctog_sz += size; 278bec0b493Shannah_mairs PetscFunctionReturn(0); 279bec0b493Shannah_mairs } 2804db21f4bSStefano Zampini 2814db21f4bSStefano Zampini PETSC_STATIC_INLINE PetscErrorCode PetscLogGpuToCpu(PetscLogDouble size) 2824db21f4bSStefano Zampini { 283958c4211Shannah_mairs PetscFunctionBegin; 284bec0b493Shannah_mairs petsc_gtoc_ct += 1; 285bec0b493Shannah_mairs petsc_gtoc_sz += size; 286bec0b493Shannah_mairs PetscFunctionReturn(0); 287bec0b493Shannah_mairs } 2884db21f4bSStefano Zampini 2894db21f4bSStefano Zampini PETSC_STATIC_INLINE PetscErrorCode PetscLogGpuFlops(PetscLogDouble n) 2904db21f4bSStefano Zampini { 291958c4211Shannah_mairs PetscFunctionBegin; 292958c4211Shannah_mairs #if defined(PETSC_USE_DEBUG) 293958c4211Shannah_mairs if (n < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Cannot log negative flops"); 294958c4211Shannah_mairs #endif 295958c4211Shannah_mairs petsc_TotalFlops += PETSC_FLOPS_PER_OP*n; 296958c4211Shannah_mairs petsc_gflops += PETSC_FLOPS_PER_OP*n; 297958c4211Shannah_mairs PetscFunctionReturn(0); 298958c4211Shannah_mairs } 2994db21f4bSStefano Zampini 3004db21f4bSStefano Zampini PETSC_STATIC_INLINE PetscErrorCode PetscLogGpuTimeBegin() 3014db21f4bSStefano Zampini { 302958c4211Shannah_mairs PetscErrorCode ierr; 303958c4211Shannah_mairs PetscFunctionBegin; 304703ce11fSStefano Zampini #if defined(PETSC_USE_DEBUG) 305703ce11fSStefano Zampini if (petsc_gtime_inuse) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Forgot to call PetscLogGpuTimeEnd()?"); 306703ce11fSStefano Zampini petsc_gtime_inuse = PETSC_TRUE; 307703ce11fSStefano Zampini #endif 308958c4211Shannah_mairs ierr = PetscTimeSubtract(&petsc_gtime);CHKERRQ(ierr); 309958c4211Shannah_mairs PetscFunctionReturn(0); 310958c4211Shannah_mairs } 3114db21f4bSStefano Zampini 3124db21f4bSStefano Zampini PETSC_STATIC_INLINE PetscErrorCode PetscLogGpuTimeEnd() 3134db21f4bSStefano Zampini { 314958c4211Shannah_mairs PetscErrorCode ierr; 315958c4211Shannah_mairs PetscFunctionBegin; 316703ce11fSStefano Zampini #if defined(PETSC_USE_DEBUG) 317703ce11fSStefano Zampini if (!petsc_gtime_inuse) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Forgot to call PetscLogGpuTimeBegin()?"); 318703ce11fSStefano Zampini petsc_gtime_inuse = PETSC_FALSE; 319703ce11fSStefano Zampini #endif 320958c4211Shannah_mairs ierr = PetscTimeAdd(&petsc_gtime);CHKERRQ(ierr); 321958c4211Shannah_mairs PetscFunctionReturn(0); 322958c4211Shannah_mairs } 3234db21f4bSStefano Zampini 3244db21f4bSStefano Zampini PETSC_STATIC_INLINE PetscErrorCode PetscLogGpuTimeAdd(PetscLogDouble t) 3254db21f4bSStefano Zampini { 3264db21f4bSStefano Zampini PetscFunctionBegin; 3274db21f4bSStefano Zampini petsc_gtime += t; 3284db21f4bSStefano Zampini PetscFunctionReturn(0); 3294db21f4bSStefano Zampini } 330bec0b493Shannah_mairs #endif 331bec0b493Shannah_mairs 332fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscGetFlops(PetscLogDouble *); 33377c4ece6SBarry Smith 334aa482453SBarry Smith #if defined (PETSC_HAVE_MPE) 335014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogMPEBegin(void); 336014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogMPEDump(const char[]); 33777c4ece6SBarry Smith #endif 33877c4ece6SBarry Smith 339014dd563SJed Brown PETSC_EXTERN PetscErrorCode (*PetscLogPLB)(PetscLogEvent,int,PetscObject,PetscObject,PetscObject,PetscObject); 340014dd563SJed Brown PETSC_EXTERN PetscErrorCode (*PetscLogPLE)(PetscLogEvent,int,PetscObject,PetscObject,PetscObject,PetscObject); 341014dd563SJed Brown PETSC_EXTERN PetscErrorCode (*PetscLogPHC)(PetscObject); 342014dd563SJed Brown PETSC_EXTERN PetscErrorCode (*PetscLogPHD)(PetscObject); 34377c4ece6SBarry Smith 344e83a5d19SLisandro 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) 345fa2bb9feSLisandro Dalcin #define PetscLogObjectCreate(h) ((PetscLogPHC) ? (*PetscLogPHC)((PetscObject)(h)) : 0) 346fa2bb9feSLisandro Dalcin #define PetscLogObjectDestroy(h) ((PetscLogPHD) ? (*PetscLogPHD)((PetscObject)(h)) : 0) 347fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogObjectState(PetscObject, const char[], ...); 348fa2bb9feSLisandro Dalcin 3498ba1e511SMatthew Knepley /* Initialization functions */ 350bb1d7374SBarry Smith PETSC_EXTERN PetscErrorCode PetscLogDefaultBegin(void); 351014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogAllBegin(void); 352bb1d7374SBarry Smith PETSC_EXTERN PetscErrorCode PetscLogNestedBegin(void); 353014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogTraceBegin(FILE *); 354014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogActions(PetscBool); 355014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogObjects(PetscBool); 356fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogSetThreshold(PetscLogDouble,PetscLogDouble*); 357014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogSet(PetscErrorCode (*)(int, int, PetscObject, PetscObject, PetscObject, PetscObject), 3586849ba73SBarry Smith PetscErrorCode (*)(int, int, PetscObject, PetscObject, PetscObject, PetscObject)); 359fa2bb9feSLisandro Dalcin 3608ba1e511SMatthew Knepley /* Output functions */ 361014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogView(PetscViewer); 362f14045dbSBarry Smith PETSC_EXTERN PetscErrorCode PetscLogViewFromOptions(void); 363014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogDump(const char[]); 36431d06abdSBarry Smith 365fa2bb9feSLisandro Dalcin /* Stage functions */ 366014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStageRegister(const char[],PetscLogStage*); 367014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStagePush(PetscLogStage); 368014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStagePop(void); 369014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStageSetActive(PetscLogStage,PetscBool); 370014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStageGetActive(PetscLogStage,PetscBool*); 371014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStageSetVisible(PetscLogStage,PetscBool); 372014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStageGetVisible(PetscLogStage,PetscBool*); 373014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogStageGetId(const char[],PetscLogStage*); 374f5d6ab90SLisandro Dalcin 3758ba1e511SMatthew Knepley /* Event functions */ 376014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventRegister(const char[],PetscClassId,PetscLogEvent*); 377217044c2SLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogEventSetCollective(PetscLogEvent,PetscBool); 378fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogEventIncludeClass(PetscClassId); 379fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogEventExcludeClass(PetscClassId); 380014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventActivate(PetscLogEvent); 381014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventDeactivate(PetscLogEvent); 382014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventSetActiveAll(PetscLogEvent,PetscBool); 383014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventActivateClass(PetscClassId); 384014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventDeactivateClass(PetscClassId); 3858c49a8d9SBarry Smith PETSC_EXTERN PetscErrorCode PetscLogEventGetId(const char[],PetscLogEvent*); 38613230571SMichael Lange PETSC_EXTERN PetscErrorCode PetscLogEventGetPerfInfo(int,PetscLogEvent,PetscEventPerfInfo*); 387891e75beSMatthew G. Knepley PETSC_EXTERN PetscErrorCode PetscLogEventSetDof(PetscLogEvent, PetscInt, PetscLogDouble); 3885d68e14cSMatthew G. Knepley PETSC_EXTERN PetscErrorCode PetscLogEventSetError(PetscLogEvent, PetscInt, PetscLogDouble); 3898ba1e511SMatthew Knepley 3908ba1e511SMatthew Knepley /* Global counters */ 391014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_irecv_ct; 392014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_isend_ct; 393014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_recv_ct; 394014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_send_ct; 395014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_irecv_len; 396014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_isend_len; 397014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_recv_len; 398014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_send_len; 399014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_allreduce_ct; 400014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_gather_ct; 401014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_scatter_ct; 402014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_wait_ct; 403014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_wait_any_ct; 404014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_wait_all_ct; 405014dd563SJed Brown PETSC_EXTERN PetscLogDouble petsc_sum_of_waits_ct; 4068ba1e511SMatthew Knepley 407e3ed9ee7SBarry Smith PETSC_EXTERN PetscBool PetscLogMemory; 408e3ed9ee7SBarry Smith 409fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscBool PetscLogSyncOn; /* true if logging synchronization is enabled */ 41062872c28SLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogEventSynchronize(PetscLogEvent, MPI_Comm); 41162872c28SLisandro Dalcin 41262872c28SLisandro Dalcin #define PetscLogEventSync(e,comm) \ 41362872c28SLisandro Dalcin (((PetscLogPLB && petsc_stageLog->stageInfo[petsc_stageLog->curStage].perfInfo.active && petsc_stageLog->stageInfo[petsc_stageLog->curStage].eventLog->eventInfo[e].active) ? \ 41462872c28SLisandro Dalcin PetscLogEventSynchronize((e),(comm)) : 0 )) 41577c4ece6SBarry Smith 41652e6d16bSBarry Smith #define PetscLogEventBegin(e,o1,o2,o3,o4) \ 4178b5db460SBarry Smith (((PetscLogPLB && petsc_stageLog->stageInfo[petsc_stageLog->curStage].perfInfo.active && petsc_stageLog->stageInfo[petsc_stageLog->curStage].eventLog->eventInfo[e].active) ? \ 418495fc317SBarry Smith (*PetscLogPLB)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4)) : 0 )) 419043328b6SSatish Balay 42052e6d16bSBarry Smith #define PetscLogEventEnd(e,o1,o2,o3,o4) \ 4218b5db460SBarry Smith (((PetscLogPLE && petsc_stageLog->stageInfo[petsc_stageLog->curStage].perfInfo.active && petsc_stageLog->stageInfo[petsc_stageLog->curStage].eventLog->eventInfo[e].active) ? \ 422495fc317SBarry Smith (*PetscLogPLE)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4)) : 0 )) 42377c4ece6SBarry Smith 424014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventGetFlops(PetscLogEvent,PetscLogDouble*); 425014dd563SJed Brown PETSC_EXTERN PetscErrorCode PetscLogEventZeroFlops(PetscLogEvent); 426003131ecSBarry Smith 427ce85283eSBarry Smith /* 428f621e05eSBarry Smith These are used internally in the PETSc routines to keep a count of MPI messages and 429f621e05eSBarry Smith their sizes. 430f621e05eSBarry Smith 431af0996ceSBarry Smith This does not work for MPI-Uni because our include/petsc/mpiuni/mpi.h file 432ce85283eSBarry Smith uses macros to defined the MPI operations. 43315308354SBarry Smith 43415308354SBarry Smith It does not work correctly from HP-UX because it processes the 435bb4af37aSBarry Smith macros in a way that sometimes it double counts, hence 436b6410449SSatish Balay PETSC_HAVE_BROKEN_RECURSIVE_MACRO 4377c1e34a4SSatish Balay 438f569fd43SBarry Smith It does not work with Windows because winmpich lacks MPI_Type_size() 439ce85283eSBarry Smith */ 440c9b973beSBarry Smith #if !defined(MPIUNI_H) && !defined(PETSC_HAVE_BROKEN_RECURSIVE_MACRO) && !defined (PETSC_HAVE_MPI_MISSING_TYPESIZE) 44177a39924SBarry Smith /* 44277a39924SBarry Smith Logging of MPI activities 44377a39924SBarry Smith */ 444b23bfdefSJunchao Zhang PETSC_STATIC_INLINE PetscErrorCode PetscMPITypeSize(PetscInt count,MPI_Datatype type,PetscLogDouble *length) 445f95db71bSBarry Smith { 446f12d10ccSJunchao Zhang PetscMPIInt typesize; 447f12d10ccSJunchao Zhang PetscErrorCode ierr; 4487d30fa4aSBarry Smith if (type == MPI_DATATYPE_NULL) return 0; 449f12d10ccSJunchao Zhang ierr = MPI_Type_size(type,&typesize);CHKERRQ(ierr); 450f12d10ccSJunchao Zhang *length += (PetscLogDouble) (count*typesize); 4519c72ff1eSSatish Balay return 0; 452f95db71bSBarry Smith } 45377a39924SBarry Smith 454f12d10ccSJunchao Zhang PETSC_STATIC_INLINE PetscErrorCode PetscMPITypeSizeComm(MPI_Comm comm,const PetscMPIInt *counts,MPI_Datatype type,PetscLogDouble *length) 4553b9284c0SMatthew G Knepley { 456f12d10ccSJunchao Zhang PetscMPIInt typesize,size,p; 457f12d10ccSJunchao Zhang PetscErrorCode ierr; 4583b9284c0SMatthew G Knepley 4597d30fa4aSBarry Smith if (type == MPI_DATATYPE_NULL) return 0; 460f12d10ccSJunchao Zhang ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 461f12d10ccSJunchao Zhang ierr = MPI_Type_size(type,&typesize);CHKERRQ(ierr); 462f12d10ccSJunchao Zhang for (p=0; p<size; ++p) { 463f12d10ccSJunchao Zhang *length += (PetscLogDouble) (counts[p]*typesize); 464f12d10ccSJunchao Zhang } 465f12d10ccSJunchao Zhang return 0; 466f12d10ccSJunchao Zhang } 467f12d10ccSJunchao Zhang 468f12d10ccSJunchao Zhang PETSC_STATIC_INLINE PetscErrorCode PetscMPITypeSizeCount(PetscInt n,const PetscMPIInt *counts,MPI_Datatype type,PetscLogDouble *length) 469f12d10ccSJunchao Zhang { 470f12d10ccSJunchao Zhang PetscMPIInt typesize,p; 471f12d10ccSJunchao Zhang PetscErrorCode ierr; 472f12d10ccSJunchao Zhang 473f12d10ccSJunchao Zhang if (type == MPI_DATATYPE_NULL) return 0; 474f12d10ccSJunchao Zhang ierr = MPI_Type_size(type,&typesize);CHKERRQ(ierr); 475f12d10ccSJunchao Zhang for (p=0; p<n; ++p) { 476f12d10ccSJunchao Zhang *length += (PetscLogDouble) (counts[p]*typesize); 4773b9284c0SMatthew G Knepley } 4783b9284c0SMatthew G Knepley return 0; 4793b9284c0SMatthew G Knepley } 4803b9284c0SMatthew G Knepley 48172b11a25SBarry Smith /* 48272b11a25SBarry Smith Returns 1 if the communicator is parallel else zero 48372b11a25SBarry Smith */ 48472b11a25SBarry Smith PETSC_STATIC_INLINE int PetscMPIParallelComm(MPI_Comm comm) 48572b11a25SBarry Smith { 48672b11a25SBarry Smith PetscMPIInt size; MPI_Comm_size(comm,&size); return size > 1; 48772b11a25SBarry Smith } 48872b11a25SBarry Smith 48977a39924SBarry Smith #define MPI_Irecv(buf,count,datatype,source,tag,comm,request) \ 490f12d10ccSJunchao Zhang ((petsc_irecv_ct++,0) || PetscMPITypeSize((count),(datatype),&(petsc_irecv_len)) || MPI_Irecv((buf),(count),(datatype),(source),(tag),(comm),(request))) 49115308354SBarry Smith 49277a39924SBarry Smith #define MPI_Isend(buf,count,datatype,dest,tag,comm,request) \ 493f12d10ccSJunchao Zhang ((petsc_isend_ct++,0) || PetscMPITypeSize((count),(datatype),&(petsc_isend_len)) || MPI_Isend((buf),(count),(datatype),(dest),(tag),(comm),(request))) 49415308354SBarry Smith 495457b14b1SJunchao Zhang #define MPI_Startall_irecv(count,datatype,number,requests) \ 496f12d10ccSJunchao Zhang ((petsc_irecv_ct += (PetscLogDouble)(number),0) || PetscMPITypeSize((count),(datatype),&(petsc_irecv_len)) || ((number) && MPI_Startall((number),(requests)))) 4970d4b0b6cSBarry Smith 498457b14b1SJunchao Zhang #define MPI_Startall_isend(count,datatype,number,requests) \ 499f12d10ccSJunchao Zhang ((petsc_isend_ct += (PetscLogDouble)(number),0) || PetscMPITypeSize((count),(datatype),&(petsc_isend_len)) || ((number) && MPI_Startall((number),(requests)))) 5000d4b0b6cSBarry Smith 501457b14b1SJunchao Zhang #define MPI_Start_isend(count,datatype,requests) \ 502f12d10ccSJunchao Zhang ((petsc_isend_ct++,0) || PetscMPITypeSize((count),(datatype),(&petsc_isend_len)) || MPI_Start((requests))) 5030d4b0b6cSBarry Smith 504ce85283eSBarry Smith #define MPI_Recv(buf,count,datatype,source,tag,comm,status) \ 505f12d10ccSJunchao Zhang ((petsc_recv_ct++,0) || PetscMPITypeSize((count),(datatype),(&petsc_recv_len)) || MPI_Recv((buf),(count),(datatype),(source),(tag),(comm),(status))) 50615308354SBarry Smith 50777a39924SBarry Smith #define MPI_Send(buf,count,datatype,dest,tag,comm) \ 508f12d10ccSJunchao Zhang ((petsc_send_ct++,0) || PetscMPITypeSize((count),(datatype),(&petsc_send_len)) || MPI_Send((buf),(count),(datatype),(dest),(tag),(comm))) 50977a39924SBarry Smith 51077a39924SBarry Smith #define MPI_Wait(request,status) \ 5115fa69175SJed Brown ((petsc_wait_ct++,petsc_sum_of_waits_ct++,0) || MPI_Wait((request),(status))) 51277a39924SBarry Smith 51377a39924SBarry Smith #define MPI_Waitany(a,b,c,d) \ 5145fa69175SJed Brown ((petsc_wait_any_ct++,petsc_sum_of_waits_ct++,0) || MPI_Waitany((a),(b),(c),(d))) 51577a39924SBarry Smith 51677a39924SBarry Smith #define MPI_Waitall(count,array_of_requests,array_of_statuses) \ 5175fa69175SJed Brown ((petsc_wait_all_ct++,petsc_sum_of_waits_ct += (PetscLogDouble) (count),0) || MPI_Waitall((count),(array_of_requests),(array_of_statuses))) 51877a39924SBarry Smith 51977a39924SBarry Smith #define MPI_Allreduce(sendbuf,recvbuf,count,datatype,op,comm) \ 5205fa69175SJed Brown ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || MPI_Allreduce((sendbuf),(recvbuf),(count),(datatype),(op),(comm))) 5213914022bSBarry Smith 52254fe5c21SBarry Smith #define MPI_Bcast(buffer,count,datatype,root,comm) \ 5235fa69175SJed Brown ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || MPI_Bcast((buffer),(count),(datatype),(root),(comm))) 52454fe5c21SBarry Smith 525d6e4c47cSJed Brown #define MPI_Reduce_scatter_block(sendbuf,recvbuf,recvcount,datatype,op,comm) \ 5265fa69175SJed Brown ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || MPI_Reduce_scatter_block((sendbuf),(recvbuf),(recvcount),(datatype),(op),(comm))) 527d6e4c47cSJed Brown 5283b9284c0SMatthew G Knepley #define MPI_Alltoall(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,comm) \ 529f12d10ccSJunchao Zhang ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Alltoall((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(comm))) 5303b9284c0SMatthew G Knepley 5313b9284c0SMatthew G Knepley #define MPI_Alltoallv(sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm) \ 532f12d10ccSJunchao 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))) 5333b9284c0SMatthew G Knepley 53401faf4e4SMatthew Knepley #define MPI_Allgather(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,comm) \ 5355fa69175SJed Brown ((petsc_gather_ct += PetscMPIParallelComm((comm)),0) || MPI_Allgather((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(comm))) 53601faf4e4SMatthew Knepley 53701faf4e4SMatthew Knepley #define MPI_Allgatherv(sendbuf,sendcount,sendtype,recvbuf,recvcount,displs,recvtype,comm) \ 5385fa69175SJed Brown ((petsc_gather_ct += PetscMPIParallelComm((comm)),0) || MPI_Allgatherv((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(displs),(recvtype),(comm))) 53901faf4e4SMatthew Knepley 54001faf4e4SMatthew Knepley #define MPI_Gather(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,root,comm) \ 541f12d10ccSJunchao Zhang ((petsc_gather_ct++,0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Gather((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm))) 54201faf4e4SMatthew Knepley 54301faf4e4SMatthew Knepley #define MPI_Gatherv(sendbuf,sendcount,sendtype,recvbuf,recvcount,displs,recvtype,root,comm) \ 544f12d10ccSJunchao Zhang ((petsc_gather_ct++,0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Gatherv((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(displs),(recvtype),(root),(comm))) 54501faf4e4SMatthew Knepley 54601faf4e4SMatthew Knepley #define MPI_Scatter(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,root,comm) \ 547f12d10ccSJunchao Zhang ((petsc_scatter_ct++,0) || PetscMPITypeSize((recvcount),(recvtype),(&petsc_recv_len)) || MPI_Scatter((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm))) 54801faf4e4SMatthew Knepley 54901faf4e4SMatthew Knepley #define MPI_Scatterv(sendbuf,sendcount,displs,sendtype,recvbuf,recvcount,recvtype,root,comm) \ 550f12d10ccSJunchao Zhang ((petsc_scatter_ct++,0) || PetscMPITypeSize((recvcount),(recvtype),(&petsc_recv_len)) || MPI_Scatterv((sendbuf),(sendcount),(displs),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm))) 551f12d10ccSJunchao Zhang 552f12d10ccSJunchao Zhang #define MPI_Ialltoall(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,comm,request) \ 553f12d10ccSJunchao Zhang ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Ialltoall((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(comm),(request))) 554f12d10ccSJunchao Zhang 555f12d10ccSJunchao Zhang #define MPI_Ialltoallv(sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm,request) \ 556f12d10ccSJunchao 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))) 557f12d10ccSJunchao Zhang 558f12d10ccSJunchao Zhang #define MPI_Iallgather(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,comm,request) \ 559f12d10ccSJunchao Zhang ((petsc_gather_ct += PetscMPIParallelComm((comm)),0) || MPI_Iallgather((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(comm),(request))) 560f12d10ccSJunchao Zhang 561f12d10ccSJunchao Zhang #define MPI_Iallgatherv(sendbuf,sendcount,sendtype,recvbuf,recvcount,displs,recvtype,comm,request) \ 562f12d10ccSJunchao Zhang ((petsc_gather_ct += PetscMPIParallelComm((comm)),0) || MPI_Iallgatherv((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(displs),(recvtype),(comm),(request))) 563f12d10ccSJunchao Zhang 564f12d10ccSJunchao Zhang #define MPI_Igather(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,root,comm,request) \ 565f12d10ccSJunchao Zhang ((petsc_gather_ct++,0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Igather((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm),(request))) 566f12d10ccSJunchao Zhang 567f12d10ccSJunchao Zhang #define MPI_Igatherv(sendbuf,sendcount,sendtype,recvbuf,recvcount,displs,recvtype,root,comm,request) \ 568f12d10ccSJunchao Zhang ((petsc_gather_ct++,0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Igatherv((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(displs),(recvtype),(root),(comm),(request))) 569f12d10ccSJunchao Zhang 570f12d10ccSJunchao Zhang #define MPI_Iscatter(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,root,comm,request) \ 571f12d10ccSJunchao Zhang ((petsc_scatter_ct++,0) || PetscMPITypeSize((recvcount),(recvtype),(&petsc_recv_len)) || MPI_Iscatter((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm),(request))) 572f12d10ccSJunchao Zhang 573f12d10ccSJunchao Zhang #define MPI_Iscatterv(sendbuf,sendcount,displs,sendtype,recvbuf,recvcount,recvtype,root,comm,request) \ 574f12d10ccSJunchao Zhang ((petsc_scatter_ct++,0) || PetscMPITypeSize((recvcount),(recvtype),(&petsc_recv_len)) || MPI_Iscatterv((sendbuf),(sendcount),(displs),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm),(request))) 575f12d10ccSJunchao Zhang 576f12d10ccSJunchao Zhang /* We treat MPI_Ineighbor_alltoallv as a set of isend/irecv instead of a traditional MPI collective. 577f12d10ccSJunchao Zhang OpenMPI-3.0 ran into error with outdegree = indegree = 0, so we use ((outdegree) || (indegree)) as a workaround. 578f12d10ccSJunchao Zhang */ 579f12d10ccSJunchao Zhang #define MPI_Start_ineighbor_alltoallv(outdegree,indegree,sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm,request) \ 580f12d10ccSJunchao 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)))) 581f12d10ccSJunchao Zhang 582f12d10ccSJunchao Zhang #define MPI_Start_neighbor_alltoallv(outdegree,indegree,sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm) \ 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_Neighbor_alltoallv((sendbuf),(sendcnts),(sdispls),(sendtype),(recvbuf),(recvcnts),(rdispls),(recvtype),(comm)))) 58401faf4e4SMatthew Knepley 5850d4b0b6cSBarry Smith #else 5860d4b0b6cSBarry Smith 587457b14b1SJunchao Zhang #define MPI_Startall_irecv(count,datatype,number,requests) \ 588ea1c87f7SJunchao Zhang ((number) && MPI_Startall((number),(requests))) 5890d4b0b6cSBarry Smith 590457b14b1SJunchao Zhang #define MPI_Startall_isend(count,datatype,number,requests) \ 591ea1c87f7SJunchao Zhang ((number) && MPI_Startall((number),(requests))) 5920d4b0b6cSBarry Smith 593457b14b1SJunchao Zhang #define MPI_Start_isend(count,datatype,requests) \ 5945fa69175SJed Brown (MPI_Start((requests))) 5950d4b0b6cSBarry Smith 596f12d10ccSJunchao Zhang #define MPI_Start_ineighbor_alltoallv(outdegree,indegree,sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm,request) \ 597f12d10ccSJunchao Zhang (((outdegree) || (indegree)) && MPI_Ineighbor_alltoallv((sendbuf),(sendcnts),(sdispls),(sendtype),(recvbuf),(recvcnts),(rdispls),(recvtype),(comm),(request))) 598f12d10ccSJunchao Zhang 599f12d10ccSJunchao Zhang #define MPI_Start_neighbor_alltoallv(outdegree,indegree,sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm) \ 600f12d10ccSJunchao Zhang (((outdegree) || (indegree)) && MPI_Neighbor_alltoallv((sendbuf),(sendcnts),(sdispls),(sendtype),(recvbuf),(recvcnts),(rdispls),(recvtype),(comm))) 601c9b973beSBarry Smith #endif /* !MPIUNI_H && ! PETSC_HAVE_BROKEN_RECURSIVE_MACRO */ 602614700edSBarry Smith 603df8cf0b5SBarry Smith #else /* ---Logging is turned off --------------------------------------------*/ 604614700edSBarry Smith 605e3ed9ee7SBarry Smith #define PetscLogMemory PETSC_FALSE 606e3ed9ee7SBarry Smith 607b0a32e0cSBarry Smith #define PetscLogFlops(n) 0 608fa2bb9feSLisandro Dalcin #define PetscGetFlops(a) (*(a) = 0.0,0) 609614700edSBarry Smith 610fa2bb9feSLisandro Dalcin #define PetscLogStageRegister(a,b) 0 611fa2bb9feSLisandro Dalcin #define PetscLogStagePush(a) 0 612fa2bb9feSLisandro Dalcin #define PetscLogStagePop() 0 613f5d6ab90SLisandro Dalcin #define PetscLogStageSetActive(a,b) 0 614f5d6ab90SLisandro Dalcin #define PetscLogStageGetActive(a,b) 0 615f5d6ab90SLisandro Dalcin #define PetscLogStageGetVisible(a,b) 0 616f5d6ab90SLisandro Dalcin #define PetscLogStageSetVisible(a,b) 0 617f5d6ab90SLisandro Dalcin #define PetscLogStageGetId(a,b) (*(b)=0,0) 618f5d6ab90SLisandro Dalcin 619fa2bb9feSLisandro Dalcin #define PetscLogEventRegister(a,b,c) 0 620fa2bb9feSLisandro Dalcin #define PetscLogEventSetCollective(a,b) 0 621fa2bb9feSLisandro Dalcin #define PetscLogEventIncludeClass(a) 0 622fa2bb9feSLisandro Dalcin #define PetscLogEventExcludeClass(a) 0 623b0a32e0cSBarry Smith #define PetscLogEventActivate(a) 0 624b0a32e0cSBarry Smith #define PetscLogEventDeactivate(a) 0 625b0a32e0cSBarry Smith #define PetscLogEventActivateClass(a) 0 626b0a32e0cSBarry Smith #define PetscLogEventDeactivateClass(a) 0 627e8e7597cSSatish Balay #define PetscLogEventSetActiveAll(a,b) 0 62846eb3cd7SSatish Balay #define PetscLogEventGetId(a,b) (*(b)=0,0) 629561b2a67SSatish Balay #define PetscLogEventGetPerfInfo(a,b,c) 0 6309c702cfcSMatthew G. Knepley #define PetscLogEventSetDof(a,b,c) 0 6319c702cfcSMatthew G. Knepley #define PetscLogEventSetError(a,b,c) 0 63277c4ece6SBarry Smith 6338b5db460SBarry Smith #define PetscLogPLB 0 6348b5db460SBarry Smith #define PetscLogPLE 0 6358b5db460SBarry Smith #define PetscLogPHC 0 6368b5db460SBarry Smith #define PetscLogPHD 0 637f5d6ab90SLisandro Dalcin 638efee365bSSatish Balay #define PetscLogObjectParents(p,n,c) 0 63952e6d16bSBarry Smith #define PetscLogObjectCreate(h) 0 64052e6d16bSBarry Smith #define PetscLogObjectDestroy(h) 0 641fa2bb9feSLisandro Dalcin PETSC_EXTERN PetscErrorCode PetscLogObjectState(PetscObject,const char[],...); 642fa2bb9feSLisandro Dalcin 643bb1d7374SBarry Smith #define PetscLogDefaultBegin() 0 644b0a32e0cSBarry Smith #define PetscLogAllBegin() 0 645bb1d7374SBarry Smith #define PetscLogNestedBegin() 0 646fa2bb9feSLisandro Dalcin #define PetscLogTraceBegin(file) 0 64773fda44aSBarry Smith #define PetscLogActions(a) 0 648fa2bb9feSLisandro Dalcin #define PetscLogObjects(a) 0 649fa2bb9feSLisandro Dalcin #define PetscLogSetThreshold(a,b) 0 650fa2bb9feSLisandro Dalcin #define PetscLogSet(lb,le) 0 651fa2bb9feSLisandro Dalcin 652fa2bb9feSLisandro Dalcin #define PetscLogView(viewer) 0 653fa2bb9feSLisandro Dalcin #define PetscLogViewFromOptions() 0 654fa2bb9feSLisandro Dalcin #define PetscLogDump(c) 0 655fa2bb9feSLisandro Dalcin 65662872c28SLisandro Dalcin #define PetscLogEventSync(e,comm) 0 657fa2bb9feSLisandro Dalcin #define PetscLogEventBegin(e,o1,o2,o3,o4) 0 658fa2bb9feSLisandro Dalcin #define PetscLogEventEnd(e,o1,o2,o3,o4) 0 659ce6058e1SBarry Smith 660*8df2f0b1SJunchao Zhang #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA) 6618157aad8SJunchao Zhang #define PetscLogCpuToGpu(a) 0 6628157aad8SJunchao Zhang #define PetscLogGpuToCpu(a) 0 6638157aad8SJunchao Zhang #define PetscLogGpuFlops(a) 0 6648157aad8SJunchao Zhang #define PetscLogGpuTimeBegin() 0 6658157aad8SJunchao Zhang #define PetscLogGpuTimeEnd() 0 6668157aad8SJunchao Zhang #define PetscLogGpuTimeAdd(a) 0 667*8df2f0b1SJunchao Zhang #endif 6688157aad8SJunchao Zhang 669aa482453SBarry Smith /* If PETSC_USE_LOG is NOT defined, these still need to be! */ 670ea1c87f7SJunchao Zhang #define MPI_Startall_irecv(count,datatype,number,requests) ((number) && MPI_Startall(number,requests)) 671ea1c87f7SJunchao Zhang #define MPI_Startall_isend(count,datatype,number,requests) ((number) && MPI_Startall(number,requests)) 672457b14b1SJunchao Zhang #define MPI_Start_isend(count,datatype,requests) MPI_Start(requests) 673f12d10ccSJunchao Zhang #define MPI_Start_ineighbor_alltoallv(outdegree,indegree,sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm,request) \ 674f12d10ccSJunchao Zhang (((outdegree) || (indegree)) && MPI_Ineighbor_alltoallv((sendbuf),(sendcnts),(sdispls),(sendtype),(recvbuf),(recvcnts),(rdispls),(recvtype),(comm),(request))) 675f12d10ccSJunchao Zhang #define MPI_Start_neighbor_alltoallv(outdegree,indegree,sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm) \ 676f12d10ccSJunchao Zhang (((outdegree) || (indegree)) && MPI_Neighbor_alltoallv((sendbuf),(sendcnts),(sdispls),(sendtype),(recvbuf),(recvcnts),(rdispls),(recvtype),(comm))) 677f141ce34SMatthew Knepley 678aa482453SBarry Smith #endif /* PETSC_USE_LOG */ 6796daaf66cSBarry Smith 680204a6943SJed Brown #define PetscPreLoadBegin(flag,name) \ 681204a6943SJed Brown do {\ 682204a6943SJed Brown PetscBool PetscPreLoading = flag;\ 683204a6943SJed Brown int PetscPreLoadMax,PetscPreLoadIt;\ 684166c7f25SBarry Smith PetscLogStage _stageNum;\ 6858cbcd9ccSBarry Smith PetscErrorCode _3_ierr; \ 686c5929fdfSBarry Smith _3_ierr = PetscOptionsGetBool(NULL,NULL,"-preload",&PetscPreLoading,NULL);CHKERRQ(_3_ierr); \ 687204a6943SJed Brown PetscPreLoadMax = (int)(PetscPreLoading);\ 688204a6943SJed Brown PetscPreLoadingUsed = PetscPreLoading ? PETSC_TRUE : PetscPreLoadingUsed;\ 689204a6943SJed Brown for (PetscPreLoadIt=0; PetscPreLoadIt<=PetscPreLoadMax; PetscPreLoadIt++) {\ 690204a6943SJed Brown PetscPreLoadingOn = PetscPreLoading;\ 6910298fd71SBarry Smith _3_ierr = PetscBarrier(NULL);CHKERRQ(_3_ierr);\ 692204a6943SJed Brown if (PetscPreLoadIt>0) {\ 6938e58c17dSMatthew Knepley _3_ierr = PetscLogStageGetId(name,&_stageNum);CHKERRQ(_3_ierr);\ 6948e58c17dSMatthew Knepley } else {\ 695a3bc4eb9SBarry Smith _3_ierr = PetscLogStageRegister(name,&_stageNum);CHKERRQ(_3_ierr); \ 6968e58c17dSMatthew Knepley }\ 697204a6943SJed Brown _3_ierr = PetscLogStageSetActive(_stageNum,(PetscBool)(!PetscPreLoadMax || PetscPreLoadIt));\ 6988e58c17dSMatthew Knepley _3_ierr = PetscLogStagePush(_stageNum);CHKERRQ(_3_ierr); 6998e58c17dSMatthew Knepley 700204a6943SJed Brown #define PetscPreLoadEnd() \ 701043328b6SSatish Balay _3_ierr = PetscLogStagePop();CHKERRQ(_3_ierr);\ 702204a6943SJed Brown PetscPreLoading = PETSC_FALSE;\ 703043328b6SSatish Balay }\ 704204a6943SJed Brown } while (0) 7058e58c17dSMatthew Knepley 706204a6943SJed Brown #define PetscPreLoadStage(name) do { \ 707043328b6SSatish Balay _3_ierr = PetscLogStagePop();CHKERRQ(_3_ierr); \ 708204a6943SJed Brown if (PetscPreLoadIt>0) { \ 7098e58c17dSMatthew Knepley _3_ierr = PetscLogStageGetId(name,&_stageNum);CHKERRQ(_3_ierr); \ 7108e58c17dSMatthew Knepley } else { \ 711a3bc4eb9SBarry Smith _3_ierr = PetscLogStageRegister(name,&_stageNum);CHKERRQ(_3_ierr); \ 7128e58c17dSMatthew Knepley } \ 713204a6943SJed Brown _3_ierr = PetscLogStageSetActive(_stageNum,(PetscBool)(!PetscPreLoadMax || PetscPreLoadIt)); \ 714204a6943SJed Brown _3_ierr = PetscLogStagePush(_stageNum);CHKERRQ(_3_ierr); \ 715204a6943SJed Brown } while (0) 716e9fa29b7SSatish Balay 717ad14c47eSJed Brown /* some vars for logging */ 718ad14c47eSJed Brown PETSC_EXTERN PetscBool PetscPreLoadingUsed; /* true if we are or have done preloading */ 719ad14c47eSJed Brown PETSC_EXTERN PetscBool PetscPreLoadingOn; /* true if we are currently in a preloading calculation */ 720ad14c47eSJed Brown 72197bb86f7SLois Curfman McInnes #endif 722