Actual source code: petsclog.h
1: /*
2: Defines profile/logging in PETSc.
3: */
5: #if !defined(PETSCLOG_H)
6: #define PETSCLOG_H
7: #include <petscsys.h>
8: #include <petsctime.h>
10: /* General logging of information; different from event logging */
11: PETSC_EXTERN PetscErrorCode PetscInfo_Private(const char[],PetscObject,const char[],...) PETSC_ATTRIBUTE_FORMAT(3,4);
12: #if defined(PETSC_USE_INFO)
13: #define PetscInfo(A,...) PetscInfo_Private(PETSC_FUNCTION_NAME,((PetscObject)A),__VA_ARGS__)
14: #else
15: #define PetscInfo(A,...) 0
16: #endif
18: #define PetscInfo1(...) PETSC_DEPRECATED_MACRO("GCC warning \"Use PetscInfo() (since version 3.17)\"") PetscInfo(__VA_ARGS__)
19: #define PetscInfo2(...) PETSC_DEPRECATED_MACRO("GCC warning \"Use PetscInfo() (since version 3.17)\"") PetscInfo(__VA_ARGS__)
20: #define PetscInfo3(...) PETSC_DEPRECATED_MACRO("GCC warning \"Use PetscInfo() (since version 3.17)\"") PetscInfo(__VA_ARGS__)
21: #define PetscInfo4(...) PETSC_DEPRECATED_MACRO("GCC warning \"Use PetscInfo() (since version 3.17)\"") PetscInfo(__VA_ARGS__)
22: #define PetscInfo5(...) PETSC_DEPRECATED_MACRO("GCC warning \"Use PetscInfo() (since version 3.17)\"") PetscInfo(__VA_ARGS__)
23: #define PetscInfo6(...) PETSC_DEPRECATED_MACRO("GCC warning \"Use PetscInfo() (since version 3.17)\"") PetscInfo(__VA_ARGS__)
24: #define PetscInfo7(...) PETSC_DEPRECATED_MACRO("GCC warning \"Use PetscInfo() (since version 3.17)\"") PetscInfo(__VA_ARGS__)
25: #define PetscInfo8(...) PETSC_DEPRECATED_MACRO("GCC warning \"Use PetscInfo() (since version 3.17)\"") PetscInfo(__VA_ARGS__)
26: #define PetscInfo9(...) PETSC_DEPRECATED_MACRO("GCC warning \"Use PetscInfo() (since version 3.17)\"") PetscInfo(__VA_ARGS__)
28: /*E
29: PetscInfoCommFlag - Describes the method by which to filter PetscInfo() by communicator size
31: Used as an input for PetscInfoSetFilterCommSelf()
33: $ PETSC_INFO_COMM_ALL - Default uninitialized value. PetscInfo() will not filter based on communicator size (i.e. will
34: print for all communicators)
35: $ PETSC_INFO_COMM_NO_SELF - PetscInfo() will NOT print for communicators with size = 1 (i.e. *_COMM_SELF)
36: $ PETSC_INFO_COMM_ONLY_SELF - PetscInfo will ONLY print for communicators with size = 1
38: Level: intermediate
40: .seealso: PetscInfo(), PetscInfoSetFromOptions(), PetscInfoSetFilterCommSelf()
41: E*/
42: typedef enum {
43: PETSC_INFO_COMM_ALL = -1,
44: PETSC_INFO_COMM_NO_SELF = 0,
45: PETSC_INFO_COMM_ONLY_SELF = 1
46: } PetscInfoCommFlag;
48: PETSC_EXTERN const char * const PetscInfoCommFlags[];
49: PETSC_EXTERN PetscErrorCode PetscInfoDeactivateClass(PetscClassId);
50: PETSC_EXTERN PetscErrorCode PetscInfoActivateClass(PetscClassId);
51: PETSC_EXTERN PetscErrorCode PetscInfoEnabled(PetscClassId, PetscBool *);
52: PETSC_EXTERN PetscErrorCode PetscInfoAllow(PetscBool);
53: PETSC_EXTERN PetscErrorCode PetscInfoSetFile(const char[],const char[]);
54: PETSC_EXTERN PetscErrorCode PetscInfoGetFile(char **,FILE **);
55: PETSC_EXTERN PetscErrorCode PetscInfoSetClasses(PetscBool,PetscInt,const char *const *);
56: PETSC_EXTERN PetscErrorCode PetscInfoGetClass(const char *, PetscBool *);
57: PETSC_EXTERN PetscErrorCode PetscInfoGetInfo(PetscBool *,PetscBool *,PetscBool *,PetscBool *,PetscInfoCommFlag *);
58: PETSC_EXTERN PetscErrorCode PetscInfoProcessClass(const char[],PetscInt,PetscClassId[]);
59: PETSC_EXTERN PetscErrorCode PetscInfoSetFilterCommSelf(PetscInfoCommFlag);
60: PETSC_EXTERN PetscErrorCode PetscInfoSetFromOptions(PetscOptions);
61: PETSC_EXTERN PetscErrorCode PetscInfoDestroy(void);
62: PETSC_EXTERN PetscBool PetscLogPrintInfo; /* if true, indicates PetscInfo() is turned on */
64: /*MC
65: PetscLogEvent - id used to identify PETSc or user events which timed portions (blocks of executable)
66: code.
68: Level: intermediate
70: .seealso: PetscLogEventRegister(), PetscLogEventBegin(), PetscLogEventEnd(), PetscLogStage
71: M*/
72: typedef int PetscLogEvent;
74: /*MC
75: PetscLogStage - id used to identify user stages (phases, sections) of runs - for logging
77: Level: intermediate
79: .seealso: PetscLogStageRegister(), PetscLogStagePush(), PetscLogStagePop(), PetscLogEvent
80: M*/
81: typedef int PetscLogStage;
83: #define PETSC_EVENT 1311311
84: PETSC_EXTERN PetscLogEvent PETSC_LARGEST_EVENT;
86: /* Global flop counter */
87: PETSC_EXTERN PetscLogDouble petsc_TotalFlops;
88: PETSC_EXTERN PetscLogDouble petsc_tmp_flops;
90: /* We must make the following structures available to access the event
91: activation flags in the PetscLogEventBegin/End() macros. These are not part of the PETSc public
92: API and are not intended to be used by other parts of PETSc or by users.
94: The code that manipulates these structures is in src/sys/logging/utils.
95: */
96: typedef struct _n_PetscIntStack *PetscIntStack;
98: /* -----------------------------------------------------------------------------------------------------*/
99: /*
100: PetscClassRegInfo, PetscClassPerfInfo - Each class has two data structures associated with it. The first has
101: static information about it, the second collects statistics on how many objects of the class are created,
102: how much memory they use, etc.
104: PetscClassRegLog, PetscClassPerfLog - arrays of the PetscClassRegInfo and PetscClassPerfInfo for all classes.
105: */
106: typedef struct {
107: char *name; /* The class name */
108: PetscClassId classid; /* The integer identifying this class */
109: } PetscClassRegInfo;
111: typedef struct {
112: PetscClassId id; /* The integer identifying this class */
113: int creations; /* The number of objects of this class created */
114: int destructions; /* The number of objects of this class destroyed */
115: PetscLogDouble mem; /* The total memory allocated by objects of this class */
116: PetscLogDouble descMem; /* The total memory allocated by descendents of these objects */
117: } PetscClassPerfInfo;
119: typedef struct _n_PetscClassRegLog *PetscClassRegLog;
120: struct _n_PetscClassRegLog {
121: int numClasses; /* The number of classes registered */
122: int maxClasses; /* The maximum number of classes */
123: PetscClassRegInfo *classInfo; /* The structure for class information (classids are monotonicly increasing) */
124: };
126: typedef struct _n_PetscClassPerfLog *PetscClassPerfLog;
127: struct _n_PetscClassPerfLog {
128: int numClasses; /* The number of logging classes */
129: int maxClasses; /* The maximum number of classes */
130: PetscClassPerfInfo *classInfo; /* The structure for class information (classids are monotonicly increasing) */
131: };
132: /* -----------------------------------------------------------------------------------------------------*/
133: /*
134: PetscEventRegInfo, PetscEventPerfInfo - Each event has two data structures associated with it. The first has
135: static information about it, the second collects statistics on how many times the event is used, how
136: much time it takes, etc.
138: PetscEventRegLog, PetscEventPerfLog - an array of all PetscEventRegInfo and PetscEventPerfInfo for all events. There is one
139: of these for each stage.
141: */
142: typedef struct {
143: char *name; /* The name of this event */
144: PetscClassId classid; /* The class the event is associated with */
145: PetscBool collective; /* Flag this event as collective */
146: #if defined (PETSC_HAVE_MPE)
147: int mpe_id_begin; /* MPE IDs that define the event */
148: int mpe_id_end;
149: #endif
150: } PetscEventRegInfo;
152: typedef struct {
153: int id; /* The integer identifying this event */
154: PetscBool active; /* The flag to activate logging */
155: PetscBool visible; /* The flag to print info in summary */
156: int depth; /* The nesting depth of the event call */
157: int count; /* The number of times this event was executed */
158: PetscLogDouble flops, flops2, flopsTmp; /* The flops and flops^2 used in this event */
159: PetscLogDouble time, time2, timeTmp; /* The time and time^2 taken for this event */
160: PetscLogDouble syncTime; /* The synchronization barrier time */
161: PetscLogDouble dof[8]; /* The number of degrees of freedom associated with this event */
162: PetscLogDouble errors[8]; /* The errors (user-defined) associated with this event */
163: PetscLogDouble numMessages; /* The number of messages in this event */
164: PetscLogDouble messageLength; /* The total message lengths in this event */
165: PetscLogDouble numReductions; /* The number of reductions in this event */
166: PetscLogDouble memIncrease; /* How much the resident memory has increased in this event */
167: PetscLogDouble mallocIncrease;/* How much the maximum malloced space has increased in this event */
168: PetscLogDouble mallocSpace; /* How much the space was malloced and kept during this event */
169: PetscLogDouble mallocIncreaseEvent; /* Maximum of the high water mark with in event minus memory available at the end of the event */
170: #if defined(PETSC_HAVE_DEVICE)
171: PetscLogDouble CpuToGpuCount; /* The total number of CPU to GPU copies */
172: PetscLogDouble GpuToCpuCount; /* The total number of GPU to CPU copies */
173: PetscLogDouble CpuToGpuSize; /* The total size of CPU to GPU copies */
174: PetscLogDouble GpuToCpuSize; /* The total size of GPU to CPU copies */
175: PetscLogDouble GpuFlops; /* The flops done on a GPU in this event */
176: PetscLogDouble GpuTime; /* The time spent on a GPU in this event */
177: #endif
178: } PetscEventPerfInfo;
180: typedef struct _n_PetscEventRegLog *PetscEventRegLog;
181: struct _n_PetscEventRegLog {
182: int numEvents; /* The number of registered events */
183: int maxEvents; /* The maximum number of events */
184: PetscEventRegInfo *eventInfo; /* The registration information for each event */
185: };
187: typedef struct _n_PetscEventPerfLog *PetscEventPerfLog;
188: struct _n_PetscEventPerfLog {
189: int numEvents; /* The number of logging events */
190: int maxEvents; /* The maximum number of events */
191: PetscEventPerfInfo *eventInfo; /* The performance information for each event */
192: };
193: /* ------------------------------------------------------------------------------------------------------------*/
194: /*
195: PetscStageInfo - Contains all the information about a particular stage.
197: PetscStageLog - An array of PetscStageInfo for each registered stage. There is a single one of these in the code.
198: */
199: typedef struct _PetscStageInfo {
200: char *name; /* The stage name */
201: PetscBool used; /* The stage was pushed on this processor */
202: PetscEventPerfInfo perfInfo; /* The stage performance information */
203: PetscEventPerfLog eventLog; /* The event information for this stage */
204: PetscClassPerfLog classLog; /* The class information for this stage */
205: } PetscStageInfo;
207: typedef struct _n_PetscStageLog *PetscStageLog;
208: struct _n_PetscStageLog {
209: int numStages; /* The number of registered stages */
210: int maxStages; /* The maximum number of stages */
211: PetscIntStack stack; /* The stack for active stages */
212: int curStage; /* The current stage (only used in macros so we don't call PetscIntStackTop) */
213: PetscStageInfo *stageInfo; /* The information for each stage */
214: PetscEventRegLog eventLog; /* The registered events */
215: PetscClassRegLog classLog; /* The registered classes */
216: };
217: /* -----------------------------------------------------------------------------------------------------*/
219: PETSC_EXTERN PetscErrorCode PetscLogObjectParent(PetscObject,PetscObject);
220: PETSC_EXTERN PetscErrorCode PetscLogObjectMemory(PetscObject,PetscLogDouble);
222: #if defined(PETSC_USE_LOG) /* --- Logging is turned on --------------------------------*/
223: PETSC_EXTERN PetscStageLog petsc_stageLog;
224: PETSC_EXTERN PetscErrorCode PetscLogGetStageLog(PetscStageLog*);
225: PETSC_EXTERN PetscErrorCode PetscStageLogGetCurrent(PetscStageLog,int*);
226: PETSC_EXTERN PetscErrorCode PetscStageLogGetEventPerfLog(PetscStageLog,int,PetscEventPerfLog*);
228: /*
229: Flop counting: We count each arithmetic operation (e.g., addition, multiplication) separately.
231: For the complex numbers version, note that
232: 1 complex addition = 2 flops
233: 1 complex multiplication = 6 flops,
234: where we define 1 flop as that for a double precision scalar. We roughly approximate
235: flop counting for complex numbers by multiplying the total flops by 4; this corresponds
236: to the assumption that we're counting mostly additions and multiplications -- and
237: roughly the same number of each. More accurate counting could be done by distinguishing
238: among the various arithmetic operations.
239: */
241: #if defined(PETSC_USE_COMPLEX)
242: #define PETSC_FLOPS_PER_OP 4.0
243: #else
244: #define PETSC_FLOPS_PER_OP 1.0
245: #endif
247: /*@C
248: PetscLogFlops - Log how many flops are performed in a calculation
250: Input Parameter:
251: . flops - the number of flops
253: Notes:
254: To limit the chance of integer overflow when multiplying by a constant, represent the constant as a double,
255: not an integer. Use PetscLogFlops(4.0*n) not PetscLogFlops(4*n)
257: Level: intermediate
259: .seealso: PetscLogView(), PetscLogGpuFlops()
260: @*/
262: static inline PetscErrorCode PetscLogFlops(PetscLogDouble n)
263: {
264: #if defined(PETSC_USE_DEBUG)
266: #endif
267: petsc_TotalFlops += PETSC_FLOPS_PER_OP*n;
268: return 0;
269: }
271: PETSC_EXTERN PetscErrorCode PetscGetFlops(PetscLogDouble *);
273: #if defined (PETSC_HAVE_MPE)
274: PETSC_EXTERN PetscErrorCode PetscLogMPEBegin(void);
275: PETSC_EXTERN PetscErrorCode PetscLogMPEDump(const char[]);
276: #endif
278: PETSC_EXTERN PetscErrorCode (*PetscLogPLB)(PetscLogEvent,int,PetscObject,PetscObject,PetscObject,PetscObject);
279: PETSC_EXTERN PetscErrorCode (*PetscLogPLE)(PetscLogEvent,int,PetscObject,PetscObject,PetscObject,PetscObject);
280: PETSC_EXTERN PetscErrorCode (*PetscLogPHC)(PetscObject);
281: PETSC_EXTERN PetscErrorCode (*PetscLogPHD)(PetscObject);
283: #define PetscLogObjectParents(p,n,d) PetscMacroReturnStandard(for (int _i=0; _i<(n); ++_i) PetscLogObjectParent((PetscObject)(p),(PetscObject)(d)[_i]);)
284: #define PetscLogObjectCreate(h) ((PetscLogPHC) ? (*PetscLogPHC)((PetscObject)(h)) : 0)
285: #define PetscLogObjectDestroy(h) ((PetscLogPHD) ? (*PetscLogPHD)((PetscObject)(h)) : 0)
286: PETSC_EXTERN PetscErrorCode PetscLogObjectState(PetscObject, const char[], ...) PETSC_ATTRIBUTE_FORMAT(2,3);
288: /* Initialization functions */
289: PETSC_EXTERN PetscErrorCode PetscLogDefaultBegin(void);
290: PETSC_EXTERN PetscErrorCode PetscLogAllBegin(void);
291: PETSC_EXTERN PetscErrorCode PetscLogNestedBegin(void);
292: PETSC_EXTERN PetscErrorCode PetscLogTraceBegin(FILE *);
293: PETSC_EXTERN PetscErrorCode PetscLogActions(PetscBool);
294: PETSC_EXTERN PetscErrorCode PetscLogObjects(PetscBool);
295: PETSC_EXTERN PetscErrorCode PetscLogSetThreshold(PetscLogDouble,PetscLogDouble*);
296: PETSC_EXTERN PetscErrorCode PetscLogSet(PetscErrorCode (*)(int, int, PetscObject, PetscObject, PetscObject, PetscObject),
297: PetscErrorCode (*)(int, int, PetscObject, PetscObject, PetscObject, PetscObject));
299: /* Output functions */
300: PETSC_EXTERN PetscErrorCode PetscLogView(PetscViewer);
301: PETSC_EXTERN PetscErrorCode PetscLogViewFromOptions(void);
302: PETSC_EXTERN PetscErrorCode PetscLogDump(const char[]);
304: /* Status checking functions */
305: PETSC_EXTERN PetscErrorCode PetscLogIsActive(PetscBool*);
307: /* Stage functions */
308: PETSC_EXTERN PetscErrorCode PetscLogStageRegister(const char[],PetscLogStage*);
309: PETSC_EXTERN PetscErrorCode PetscLogStagePush(PetscLogStage);
310: PETSC_EXTERN PetscErrorCode PetscLogStagePop(void);
311: PETSC_EXTERN PetscErrorCode PetscLogStageSetActive(PetscLogStage,PetscBool);
312: PETSC_EXTERN PetscErrorCode PetscLogStageGetActive(PetscLogStage,PetscBool*);
313: PETSC_EXTERN PetscErrorCode PetscLogStageSetVisible(PetscLogStage,PetscBool);
314: PETSC_EXTERN PetscErrorCode PetscLogStageGetVisible(PetscLogStage,PetscBool*);
315: PETSC_EXTERN PetscErrorCode PetscLogStageGetId(const char[],PetscLogStage*);
317: /* Event functions */
318: PETSC_EXTERN PetscErrorCode PetscLogEventRegister(const char[],PetscClassId,PetscLogEvent*);
319: PETSC_EXTERN PetscErrorCode PetscLogEventSetCollective(PetscLogEvent,PetscBool);
320: PETSC_EXTERN PetscErrorCode PetscLogEventIncludeClass(PetscClassId);
321: PETSC_EXTERN PetscErrorCode PetscLogEventExcludeClass(PetscClassId);
322: PETSC_EXTERN PetscErrorCode PetscLogEventActivate(PetscLogEvent);
323: PETSC_EXTERN PetscErrorCode PetscLogEventDeactivate(PetscLogEvent);
324: PETSC_EXTERN PetscErrorCode PetscLogEventDeactivatePush(PetscLogEvent);
325: PETSC_EXTERN PetscErrorCode PetscLogEventDeactivatePop(PetscLogEvent);
326: PETSC_EXTERN PetscErrorCode PetscLogEventSetActiveAll(PetscLogEvent,PetscBool);
327: PETSC_EXTERN PetscErrorCode PetscLogEventActivateClass(PetscClassId);
328: PETSC_EXTERN PetscErrorCode PetscLogEventDeactivateClass(PetscClassId);
329: PETSC_EXTERN PetscErrorCode PetscLogEventGetId(const char[],PetscLogEvent*);
330: PETSC_EXTERN PetscErrorCode PetscLogEventGetPerfInfo(int,PetscLogEvent,PetscEventPerfInfo*);
331: PETSC_EXTERN PetscErrorCode PetscLogEventSetDof(PetscLogEvent, PetscInt, PetscLogDouble);
332: PETSC_EXTERN PetscErrorCode PetscLogEventSetError(PetscLogEvent, PetscInt, PetscLogDouble);
334: /* Global counters */
335: PETSC_EXTERN PetscLogDouble petsc_irecv_ct;
336: PETSC_EXTERN PetscLogDouble petsc_isend_ct;
337: PETSC_EXTERN PetscLogDouble petsc_recv_ct;
338: PETSC_EXTERN PetscLogDouble petsc_send_ct;
339: PETSC_EXTERN PetscLogDouble petsc_irecv_len;
340: PETSC_EXTERN PetscLogDouble petsc_isend_len;
341: PETSC_EXTERN PetscLogDouble petsc_recv_len;
342: PETSC_EXTERN PetscLogDouble petsc_send_len;
343: PETSC_EXTERN PetscLogDouble petsc_allreduce_ct;
344: PETSC_EXTERN PetscLogDouble petsc_gather_ct;
345: PETSC_EXTERN PetscLogDouble petsc_scatter_ct;
346: PETSC_EXTERN PetscLogDouble petsc_wait_ct;
347: PETSC_EXTERN PetscLogDouble petsc_wait_any_ct;
348: PETSC_EXTERN PetscLogDouble petsc_wait_all_ct;
349: PETSC_EXTERN PetscLogDouble petsc_sum_of_waits_ct;
351: PETSC_EXTERN PetscBool PetscLogMemory;
353: PETSC_EXTERN PetscBool PetscLogSyncOn; /* true if logging synchronization is enabled */
354: PETSC_EXTERN PetscErrorCode PetscLogEventSynchronize(PetscLogEvent, MPI_Comm);
356: #define PetscLogEventSync(e,comm) \
357: (((PetscLogPLB && petsc_stageLog->stageInfo[petsc_stageLog->curStage].perfInfo.active && petsc_stageLog->stageInfo[petsc_stageLog->curStage].eventLog->eventInfo[e].active) ? \
358: PetscLogEventSynchronize((e),(comm)) : 0))
360: #define PetscLogEventBegin(e,o1,o2,o3,o4) \
361: (((PetscLogPLB && petsc_stageLog->stageInfo[petsc_stageLog->curStage].perfInfo.active && petsc_stageLog->stageInfo[petsc_stageLog->curStage].eventLog->eventInfo[e].active) ? \
362: (*PetscLogPLB)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4)) : 0))
364: #define PetscLogEventEnd(e,o1,o2,o3,o4) \
365: (((PetscLogPLE && petsc_stageLog->stageInfo[petsc_stageLog->curStage].perfInfo.active && petsc_stageLog->stageInfo[petsc_stageLog->curStage].eventLog->eventInfo[e].active) ? \
366: (*PetscLogPLE)((e),0,(PetscObject)(o1),(PetscObject)(o2),(PetscObject)(o3),(PetscObject)(o4)) : 0))
368: PETSC_EXTERN PetscErrorCode PetscLogEventGetFlops(PetscLogEvent,PetscLogDouble*);
369: PETSC_EXTERN PetscErrorCode PetscLogEventZeroFlops(PetscLogEvent);
371: /*
372: These are used internally in the PETSc routines to keep a count of MPI messages and
373: their sizes.
375: This does not work for MPI-Uni because our include/petsc/mpiuni/mpi.h file
376: uses macros to defined the MPI operations.
378: It does not work correctly from HP-UX because it processes the
379: macros in a way that sometimes it double counts, hence
380: PETSC_HAVE_BROKEN_RECURSIVE_MACRO
382: It does not work with Windows because winmpich lacks MPI_Type_size()
383: */
384: #if !defined(MPIUNI_H) && !defined(PETSC_HAVE_BROKEN_RECURSIVE_MACRO) && !defined (PETSC_HAVE_MPI_MISSING_TYPESIZE)
385: /*
386: Logging of MPI activities
387: */
388: static inline PetscErrorCode PetscMPITypeSize(PetscInt count,MPI_Datatype type,PetscLogDouble *length)
389: {
390: PetscMPIInt typesize;
392: if (type == MPI_DATATYPE_NULL) return 0;
393: MPI_Type_size(type,&typesize);
394: *length += (PetscLogDouble) (count*typesize);
395: return 0;
396: }
398: static inline PetscErrorCode PetscMPITypeSizeComm(MPI_Comm comm,const PetscMPIInt *counts,MPI_Datatype type,PetscLogDouble *length)
399: {
400: PetscMPIInt typesize,size,p;
402: if (type == MPI_DATATYPE_NULL) return 0;
403: MPI_Comm_size(comm,&size);
404: MPI_Type_size(type,&typesize);
405: for (p=0; p<size; ++p) *length += (PetscLogDouble)(counts[p]*typesize);
406: return 0;
407: }
409: static inline PetscErrorCode PetscMPITypeSizeCount(PetscInt n,const PetscMPIInt *counts,MPI_Datatype type,PetscLogDouble *length)
410: {
411: PetscMPIInt typesize,p;
413: if (type == MPI_DATATYPE_NULL) return 0;
414: MPI_Type_size(type,&typesize);
415: for (p=0; p<n; ++p) *length += (PetscLogDouble)(counts[p]*typesize);
416: return 0;
417: }
419: /*
420: Returns 1 if the communicator is parallel else zero
421: */
422: static inline int PetscMPIParallelComm(MPI_Comm comm)
423: {
424: PetscMPIInt size; MPI_Comm_size(comm,&size); return size > 1;
425: }
427: #define MPI_Irecv(buf,count,datatype,source,tag,comm,request) \
428: ((petsc_irecv_ct++,0) || PetscMPITypeSize((count),(datatype),&(petsc_irecv_len)) || MPI_Irecv((buf),(count),(datatype),(source),(tag),(comm),(request)))
430: #define MPI_Irecv_c(buf,count,datatype,source,tag,comm,request) \
431: ((petsc_irecv_ct++,0) || PetscMPITypeSize((count),(datatype),&(petsc_irecv_len)) || MPI_Irecv_c((buf),(count),(datatype),(source),(tag),(comm),(request)))
433: #define MPI_Isend(buf,count,datatype,dest,tag,comm,request) \
434: ((petsc_isend_ct++,0) || PetscMPITypeSize((count),(datatype),&(petsc_isend_len)) || MPI_Isend((buf),(count),(datatype),(dest),(tag),(comm),(request)))
436: #define MPI_Isend_c(buf,count,datatype,dest,tag,comm,request) \
437: ((petsc_isend_ct++,0) || PetscMPITypeSize((count),(datatype),&(petsc_isend_len)) || MPI_Isend_c((buf),(count),(datatype),(dest),(tag),(comm),(request)))
439: #define MPI_Startall_irecv(count,datatype,number,requests) \
440: ((petsc_irecv_ct += (PetscLogDouble)(number),0) || PetscMPITypeSize((count),(datatype),&(petsc_irecv_len)) || ((number) && MPI_Startall((number),(requests))))
442: #define MPI_Startall_isend(count,datatype,number,requests) \
443: ((petsc_isend_ct += (PetscLogDouble)(number),0) || PetscMPITypeSize((count),(datatype),&(petsc_isend_len)) || ((number) && MPI_Startall((number),(requests))))
445: #define MPI_Start_isend(count,datatype,requests) \
446: ((petsc_isend_ct++,0) || PetscMPITypeSize((count),(datatype),(&petsc_isend_len)) || MPI_Start((requests)))
448: #define MPI_Recv(buf,count,datatype,source,tag,comm,status) \
449: ((petsc_recv_ct++,0) || PetscMPITypeSize((count),(datatype),(&petsc_recv_len)) || MPI_Recv((buf),(count),(datatype),(source),(tag),(comm),(status)))
451: #define MPI_Recv_c(buf,count,datatype,source,tag,comm,status) \
452: ((petsc_recv_ct++,0) || PetscMPITypeSize((count),(datatype),(&petsc_recv_len)) || MPI_Recv_c((buf),(count),(datatype),(source),(tag),(comm),(status)))
454: #define MPI_Send(buf,count,datatype,dest,tag,comm) \
455: ((petsc_send_ct++,0) || PetscMPITypeSize((count),(datatype),(&petsc_send_len)) || MPI_Send((buf),(count),(datatype),(dest),(tag),(comm)))
457: #define MPI_Send_c(buf,count,datatype,dest,tag,comm) \
458: ((petsc_send_ct++,0) || PetscMPITypeSize((count),(datatype),(&petsc_send_len)) || MPI_Send_c((buf),(count),(datatype),(dest),(tag),(comm)))
460: #define MPI_Wait(request,status) \
461: ((petsc_wait_ct++,petsc_sum_of_waits_ct++,0) || MPI_Wait((request),(status)))
463: #define MPI_Waitany(a,b,c,d) \
464: ((petsc_wait_any_ct++,petsc_sum_of_waits_ct++,0) || MPI_Waitany((a),(b),(c),(d)))
466: #define MPI_Waitall(count,array_of_requests,array_of_statuses) \
467: ((petsc_wait_all_ct++,petsc_sum_of_waits_ct += (PetscLogDouble) (count),0) || MPI_Waitall((count),(array_of_requests),(array_of_statuses)))
469: #define MPI_Allreduce(sendbuf,recvbuf,count,datatype,op,comm) \
470: (petsc_allreduce_ct += PetscMPIParallelComm((comm)),MPI_Allreduce((sendbuf),(recvbuf),(count),(datatype),(op),(comm)))
472: #define MPI_Bcast(buffer,count,datatype,root,comm) \
473: ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || MPI_Bcast((buffer),(count),(datatype),(root),(comm)))
475: #define MPI_Reduce_scatter_block(sendbuf,recvbuf,recvcount,datatype,op,comm) \
476: ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || MPI_Reduce_scatter_block((sendbuf),(recvbuf),(recvcount),(datatype),(op),(comm)))
478: #define MPI_Alltoall(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,comm) \
479: ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Alltoall((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(comm)))
481: #define MPI_Alltoallv(sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm) \
482: ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || PetscMPITypeSizeComm((comm),(sendcnts),(sendtype),(&petsc_send_len)) || MPI_Alltoallv((sendbuf),(sendcnts),(sdispls),(sendtype),(recvbuf),(recvcnts),(rdispls),(recvtype),(comm)))
484: #define MPI_Allgather(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,comm) \
485: ((petsc_gather_ct += PetscMPIParallelComm((comm)),0) || MPI_Allgather((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(comm)))
487: #define MPI_Allgatherv(sendbuf,sendcount,sendtype,recvbuf,recvcount,displs,recvtype,comm) \
488: ((petsc_gather_ct += PetscMPIParallelComm((comm)),0) || MPI_Allgatherv((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(displs),(recvtype),(comm)))
490: #define MPI_Gather(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,root,comm) \
491: ((petsc_gather_ct++,0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Gather((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm)))
493: #define MPI_Gatherv(sendbuf,sendcount,sendtype,recvbuf,recvcount,displs,recvtype,root,comm) \
494: ((petsc_gather_ct++,0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Gatherv((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(displs),(recvtype),(root),(comm)))
496: #define MPI_Scatter(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,root,comm) \
497: ((petsc_scatter_ct++,0) || PetscMPITypeSize((recvcount),(recvtype),(&petsc_recv_len)) || MPI_Scatter((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm)))
499: #define MPI_Scatterv(sendbuf,sendcount,displs,sendtype,recvbuf,recvcount,recvtype,root,comm) \
500: ((petsc_scatter_ct++,0) || PetscMPITypeSize((recvcount),(recvtype),(&petsc_recv_len)) || MPI_Scatterv((sendbuf),(sendcount),(displs),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm)))
502: #define MPI_Ialltoall(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,comm,request) \
503: ((petsc_allreduce_ct += PetscMPIParallelComm((comm)),0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Ialltoall((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(comm),(request)))
505: #define MPI_Ialltoallv(sendbuf,sendcnts,sdispls,sendtype,recvbuf,recvcnts,rdispls,recvtype,comm,request) \
506: ((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)))
508: #define MPI_Iallgather(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,comm,request) \
509: ((petsc_gather_ct += PetscMPIParallelComm((comm)),0) || MPI_Iallgather((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(comm),(request)))
511: #define MPI_Iallgatherv(sendbuf,sendcount,sendtype,recvbuf,recvcount,displs,recvtype,comm,request) \
512: ((petsc_gather_ct += PetscMPIParallelComm((comm)),0) || MPI_Iallgatherv((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(displs),(recvtype),(comm),(request)))
514: #define MPI_Igather(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,root,comm,request) \
515: ((petsc_gather_ct++,0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Igather((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm),(request)))
517: #define MPI_Igatherv(sendbuf,sendcount,sendtype,recvbuf,recvcount,displs,recvtype,root,comm,request) \
518: ((petsc_gather_ct++,0) || PetscMPITypeSize((sendcount),(sendtype),(&petsc_send_len)) || MPI_Igatherv((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(displs),(recvtype),(root),(comm),(request)))
520: #define MPI_Iscatter(sendbuf,sendcount,sendtype,recvbuf,recvcount,recvtype,root,comm,request) \
521: ((petsc_scatter_ct++,0) || PetscMPITypeSize((recvcount),(recvtype),(&petsc_recv_len)) || MPI_Iscatter((sendbuf),(sendcount),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm),(request)))
523: #define MPI_Iscatterv(sendbuf,sendcount,displs,sendtype,recvbuf,recvcount,recvtype,root,comm,request) \
524: ((petsc_scatter_ct++,0) || PetscMPITypeSize((recvcount),(recvtype),(&petsc_recv_len)) || MPI_Iscatterv((sendbuf),(sendcount),(displs),(sendtype),(recvbuf),(recvcount),(recvtype),(root),(comm),(request)))
526: #else
528: #define MPI_Startall_irecv(count,datatype,number,requests) \
529: ((number) && MPI_Startall((number),(requests)))
531: #define MPI_Startall_isend(count,datatype,number,requests) \
532: ((number) && MPI_Startall((number),(requests)))
534: #define MPI_Start_isend(count,datatype,requests) \
535: (MPI_Start((requests)))
537: #endif /* !MPIUNI_H && ! PETSC_HAVE_BROKEN_RECURSIVE_MACRO */
539: #else /* ---Logging is turned off --------------------------------------------*/
541: #define PetscLogMemory PETSC_FALSE
543: #define PetscLogFlops(n) 0
544: #define PetscGetFlops(a) (*(a) = 0.0,0)
546: #define PetscLogStageRegister(a,b) 0
547: #define PetscLogStagePush(a) 0
548: #define PetscLogStagePop() 0
549: #define PetscLogStageSetActive(a,b) 0
550: #define PetscLogStageGetActive(a,b) 0
551: #define PetscLogStageGetVisible(a,b) 0
552: #define PetscLogStageSetVisible(a,b) 0
553: #define PetscLogStageGetId(a,b) (*(b)=0,0)
555: #define PetscLogEventRegister(a,b,c) 0
556: #define PetscLogEventSetCollective(a,b) 0
557: #define PetscLogEventIncludeClass(a) 0
558: #define PetscLogEventExcludeClass(a) 0
559: #define PetscLogEventActivate(a) 0
560: #define PetscLogEventDeactivate(a) 0
561: #define PetscLogEventDeactivatePush(a) 0
562: #define PetscLogEventDeactivatePop(a) 0
563: #define PetscLogEventActivateClass(a) 0
564: #define PetscLogEventDeactivateClass(a) 0
565: #define PetscLogEventSetActiveAll(a,b) 0
566: #define PetscLogEventGetId(a,b) (*(b)=0,0)
567: #define PetscLogEventGetPerfInfo(a,b,c) 0
568: #define PetscLogEventSetDof(a,b,c) 0
569: #define PetscLogEventSetError(a,b,c) 0
571: #define PetscLogPLB 0
572: #define PetscLogPLE 0
573: #define PetscLogPHC 0
574: #define PetscLogPHD 0
576: #define PetscLogObjectParents(p,n,c) 0
577: #define PetscLogObjectCreate(h) 0
578: #define PetscLogObjectDestroy(h) 0
579: PETSC_EXTERN PetscErrorCode PetscLogObjectState(PetscObject,const char[],...) PETSC_ATTRIBUTE_FORMAT(2,3);
581: #define PetscLogDefaultBegin() 0
582: #define PetscLogAllBegin() 0
583: #define PetscLogNestedBegin() 0
584: #define PetscLogTraceBegin(file) 0
585: #define PetscLogActions(a) 0
586: #define PetscLogObjects(a) 0
587: #define PetscLogSetThreshold(a,b) 0
588: #define PetscLogSet(lb,le) 0
589: #define PetscLogIsActive(flag) (*(flag) = PETSC_FALSE,0)
591: #define PetscLogView(viewer) 0
592: #define PetscLogViewFromOptions() 0
593: #define PetscLogDump(c) 0
595: #define PetscLogEventSync(e,comm) 0
596: #define PetscLogEventBegin(e,o1,o2,o3,o4) 0
597: #define PetscLogEventEnd(e,o1,o2,o3,o4) 0
599: /* If PETSC_USE_LOG is NOT defined, these still need to be! */
600: #define MPI_Startall_irecv(count,datatype,number,requests) ((number) && MPI_Startall(number,requests))
601: #define MPI_Startall_isend(count,datatype,number,requests) ((number) && MPI_Startall(number,requests))
602: #define MPI_Start_isend(count,datatype,requests) MPI_Start(requests)
604: #endif /* PETSC_USE_LOG */
606: #if defined (PETSC_USE_LOG) && defined(PETSC_HAVE_DEVICE)
608: /* Global GPU counters */
609: PETSC_EXTERN PetscLogDouble petsc_ctog_ct;
610: PETSC_EXTERN PetscLogDouble petsc_gtoc_ct;
611: PETSC_EXTERN PetscLogDouble petsc_ctog_sz;
612: PETSC_EXTERN PetscLogDouble petsc_gtoc_sz;
613: PETSC_EXTERN PetscLogDouble petsc_ctog_ct_scalar;
614: PETSC_EXTERN PetscLogDouble petsc_gtoc_ct_scalar;
615: PETSC_EXTERN PetscLogDouble petsc_ctog_sz_scalar;
616: PETSC_EXTERN PetscLogDouble petsc_gtoc_sz_scalar;
617: PETSC_EXTERN PetscLogDouble petsc_gflops;
618: PETSC_EXTERN PetscLogDouble petsc_gtime;
620: static inline PetscErrorCode PetscLogCpuToGpu(PetscLogDouble size)
621: {
622: petsc_ctog_ct += 1;
623: petsc_ctog_sz += size;
624: return 0;
625: }
627: static inline PetscErrorCode PetscLogGpuToCpu(PetscLogDouble size)
628: {
629: petsc_gtoc_ct += 1;
630: petsc_gtoc_sz += size;
631: return 0;
632: }
634: static inline PetscErrorCode PetscLogCpuToGpuScalar(PetscLogDouble size)
635: {
636: petsc_ctog_ct_scalar += 1;
637: petsc_ctog_sz_scalar += size;
638: return 0;
639: }
641: static inline PetscErrorCode PetscLogGpuToCpuScalar(PetscLogDouble size)
642: {
643: petsc_gtoc_ct_scalar += 1;
644: petsc_gtoc_sz_scalar += size;
645: return 0;
646: }
648: /*@C
649: PetscLogGpuFlops - Log how many flops are performed in a calculation on the device
651: Input Parameter:
652: . flops - the number of flops
654: Notes:
655: To limit the chance of integer overflow when multiplying by a constant, represent the constant as a double,
656: not an integer. Use PetscLogFlops(4.0*n) not PetscLogFlops(4*n)
658: Level: intermediate
660: .seealso: PetscLogView(), PetscLogFlops(), PetscLogGpuTimeBegin(), PetscLogGpuTimeEnd()
661: @*/
662: static inline PetscErrorCode PetscLogGpuFlops(PetscLogDouble n)
663: {
665: petsc_TotalFlops += PETSC_FLOPS_PER_OP*n;
666: petsc_gflops += PETSC_FLOPS_PER_OP*n;
667: return 0;
668: }
670: static inline PetscErrorCode PetscLogGpuTimeAdd(PetscLogDouble t)
671: {
672: petsc_gtime += t;
673: return 0;
674: }
676: PETSC_EXTERN PetscErrorCode PetscLogGpuTimeBegin(void);
677: PETSC_EXTERN PetscErrorCode PetscLogGpuTimeEnd(void);
679: #else
681: #define PetscLogCpuToGpu(a) 0
682: #define PetscLogGpuToCpu(a) 0
683: #define PetscLogCpuToGpuScalar(a) 0
684: #define PetscLogGpuToCpuScalar(a) 0
685: #define PetscLogGpuFlops(a) 0
686: #define PetscLogGpuTimeAdd(a) 0
687: #define PetscLogGpuTimeBegin() 0
688: #define PetscLogGpuTimeEnd() 0
690: #endif /* PETSC_USE_LOG && PETSC_HAVE_DEVICE */
692: #define PetscPreLoadBegin(flag,name) \
693: do {\
694: PetscBool PetscPreLoading = flag;\
695: int PetscPreLoadMax,PetscPreLoadIt;\
696: PetscLogStage _stageNum;\
697: PetscOptionsGetBool(NULL,NULL,"-preload",&PetscPreLoading,NULL); \
698: PetscPreLoadMax = (int)(PetscPreLoading);\
699: PetscPreLoadingUsed = PetscPreLoading ? PETSC_TRUE : PetscPreLoadingUsed;\
700: for (PetscPreLoadIt=0; PetscPreLoadIt<=PetscPreLoadMax; PetscPreLoadIt++) {\
701: PetscPreLoadingOn = PetscPreLoading;\
702: PetscBarrier(NULL);\
703: if (PetscPreLoadIt>0) PetscLogStageGetId(name,&_stageNum);\
704: else PetscLogStageRegister(name,&_stageNum);\
705: PetscLogStageSetActive(_stageNum,(PetscBool)(!PetscPreLoadMax || PetscPreLoadIt));\
706: PetscLogStagePush(_stageNum);
708: #define PetscPreLoadEnd() \
709: PetscLogStagePop();\
710: PetscPreLoading = PETSC_FALSE;\
711: }\
712: } while (0)
714: #define PetscPreLoadStage(name) do { \
715: PetscLogStagePop(); \
716: if (PetscPreLoadIt>0) PetscLogStageGetId(name,&_stageNum); \
717: else PetscLogStageRegister(name,&_stageNum); \
718: PetscLogStageSetActive(_stageNum,(PetscBool)(!PetscPreLoadMax || PetscPreLoadIt)); \
719: PetscLogStagePush(_stageNum); \
720: } while (0)
722: /* some vars for logging */
723: PETSC_EXTERN PetscBool PetscPreLoadingUsed; /* true if we are or have done preloading */
724: PETSC_EXTERN PetscBool PetscPreLoadingOn; /* true if we are currently in a preloading calculation */
726: #endif