Merge branch 'master' of ssh://git/gitroot/tau2

Former-commit-id: dbbd056b5099f7111611eb2f045e7f6df8df7003

examples/power/Makefile

@@ -0,0 +1,48 @@
+#****************************************************************************
+#*			TAU Portable Profiling Package			   **
+#*			http://www.cs.uoregon.edu/research/tau	           **
+#****************************************************************************
+#*    Copyright 1997  						   	   **
+#*    Department of Computer and Information Science, University of Oregon **
+#*    Advanced Computing Laboratory, Los Alamos National Laboratory        **
+#****************************************************************************
+#######################################################################
+##                  pC++/Sage++  Copyright (C) 1993,1995             ##
+##  Indiana University  University of Oregon  University of Rennes   ##
+#######################################################################
+
+
+TAU_MAKEFILE ?=../../include/Makefile
+
+include $(TAU_MAKEFILE)
+
+CC		= TAU_MAKEFILE=$(TAU_MAKEFILE) $(TAU_PREFIX_INSTALL_DIR)/$(CONFIG_ARCH)/bin/tau_cc.sh 
+
+PRINT	      	= pr
+
+RM 	      	= /bin/rm -rf
+
+TARGET	      	= matmult
+
+OBJS     	= matmult.o matmult_initialize.o
+CFLAGS=-g -O2
+LDFLAGS=-g -O2
+
+##############################################
+
+all:		$(TARGET)	
+
+install: 	$(TARGET)
+
+$(TARGET):	$(OBJS)
+	$(CC) $(LDFLAGS) $(OBJS) -o $@ $(LIBS)
+
+matmult.o: matmult.c
+	$(CC) $(CFLAGS) -c $< -o $@
+
+matmult_initialize.o: matmult_initialize.c matmult_initialize.h
+	$(CC) $(CFLAGS) -c $< -o $@
+
+clean: 	
+	$(RM) $(OBJS) $(TARGET) profile.* *.trc *.edf *.z MULT* *.inst.* *.pdb Comp_gnu.o
+##############################################

examples/power/README

@@ -0,0 +1,35 @@
+This example illustrates TAU's API for tracking Power. To track Power, simply insert:
+
+TAU_TRACK_POWER();
+
+in the code. An interrupt is generated every 10 seconds. To change the interrupt interval to say 1s, please insert:
+
+TAU_SET_INTERRUPT_INTERVAL(1);
+
+make
+
+To track Power, the following must be done:
+
+1. TAU must be configured to use PAPI configured with --with-component=rapl
+2. The executable must not rely on any LD_LIBRARY_PATH settings. So, we typically add -Wl,-rapth,/path/to/dso while linking 
+in the executable. So, we configure TAU using:
+
+ ./configure -pdt=/usr/local/packages/pdt -c++=mpicxx -cc=mpicc -fortran=mpif90 -mpilibrary=-lmpi_usempif08#-lmpi_usempi_ignore_tkr#-lmpi_mpifh#-lmpi#-Wl,-rpath,/usr/local/packages/openmpi/1.7_intel-14-tm/lib#-Wl,-rpath,/usr/local/packages/intel/14.0/compiler/lib/intel64/ -bfd=download -iowrapper -papi=/usr/local/packages/papi-5.2.0 -bfd=download -iowrapper
+
+make install
+
+3. The system on which you are running the executable must allow you to read the /dev/cpu/*/msr files:
+
+su
+# chmod o+r /dev/cpu/*/msr
+4. The executable must have special previleges to execute code. These are granted by root using:
+
+# setcap cap_sys_rawio=ep ./matmult
+
+5. The user then executes the code using:
+% mpirun -np 4 ./matmult
+
+6. Max and min marker events trigger when the usage is beyond a threshold of the value seen during the execution. To change 
+the event threshold, simply set:
+% export TAU_EVENT_THRESHOLD=0.01
+for capturing 1% variation in an atomic event. 

examples/power/matmult.c

@@ -0,0 +1,232 @@
+/******************************************************************************
+*   OpenMp Example - Matrix Multiply - C Version
+*   Demonstrates a matrix multiply using OpenMP. 
+*
+*   Modified from here:
+*   https://computing.llnl.gov/tutorials/openMP/samples/C/omp_mm.c
+*
+*   For  PAPI_FP_INS, the exclusive count for the event: 
+*   for (null) [OpenMP location: file:matmult.c ]
+*   should be  2E+06 / Number of Threads 
+******************************************************************************/
+#include <stdio.h>
+#include <stdlib.h>
+
+#include "matmult_initialize.h"
+
+#ifdef TAU_MPI
+int provided;
+#include <mpi.h>
+/* NOTE: MPI is just used to spawn multiple copies of the kernel to different ranks.
+This is not a parallel implementation */
+#endif /* TAU_MPI */
+
+#ifdef PTHREADS
+#include <pthread.h>
+#endif /* PTHREADS */
+
+#ifndef MATRIX_SIZE
+#define MATRIX_SIZE 1024
+#endif
+
+#define NRA MATRIX_SIZE                 /* number of rows in matrix A */
+#define NCA MATRIX_SIZE                 /* number of columns in matrix A */
+#define NCB MATRIX_SIZE                 /* number of columns in matrix B */
+
+double** allocateMatrix(int rows, int cols) {
+  int i;
+  double **matrix = (double**)malloc((sizeof(double*)) * rows);
+  for (i=0; i<rows; i++) {
+    matrix[i] = (double*)malloc((sizeof(double)) * cols);
+  }
+  return matrix;
+}
+
+#ifdef APP_USE_INLINE_MULTIPLY
+__inline double multiply(double a, double b) {
+	return a * b;
+}
+#endif /* APP_USE_INLINE_MULTIPLY */
+
+// cols_a and rows_b are the same value
+void compute_nested(double **a, double **b, double **c, int rows_a, int cols_a, int cols_b) {
+  int i,j,k;
+#pragma omp parallel private(i) shared(a,b,c) num_threads(2)
+  {
+    /*** Do matrix multiply sharing iterations on outer loop ***/
+    /*** Display who does which iterations for demonstration purposes ***/
+#pragma omp for nowait
+    for (i=0; i<rows_a; i++) {
+#pragma omp parallel private(i,j,k) shared(a,b,c) num_threads(2)
+      {
+#pragma omp for nowait
+      for(j=0; j<cols_b; j++) {
+        for (k=0; k<cols_a; k++) {
+#ifdef APP_USE_INLINE_MULTIPLY
+          c[i][j] += multiply(a[i][k], b[k][j]);
+#else 
+          c[i][j] += a[i][k] * b[k][j];
+#endif 
+        }
+      }
+      }
+    }
+  }   /*** End of parallel region ***/
+}
+
+// cols_a and rows_b are the same value
+void compute(double **a, double **b, double **c, int rows_a, int cols_a, int cols_b) {
+  int i,j,k;
+#pragma omp parallel private(i,j,k) shared(a,b,c)
+  {
+    /*** Do matrix multiply sharing iterations on outer loop ***/
+    /*** Display who does which iterations for demonstration purposes ***/
+#pragma omp for nowait
+    for (i=0; i<rows_a; i++) {
+      for(j=0; j<cols_b; j++) {
+        for (k=0; k<cols_a; k++) {
+#ifdef APP_USE_INLINE_MULTIPLY
+          c[i][j] += multiply(a[i][k], b[k][j]);
+#else /* APP_USE_INLINE_MULTIPLY */
+          c[i][j] += a[i][k] * b[k][j];
+#endif /* APP_USE_INLINE_MULTIPLY */
+        }
+      }
+    }
+  }   /*** End of parallel region ***/
+}
+
+void compute_interchange(double **a, double **b, double **c, int rows_a, int cols_a, int cols_b) {
+  int i,j,k;
+#pragma omp parallel private(i,j,k) shared(a,b,c)
+  {
+    /*** Do matrix multiply sharing iterations on outer loop ***/
+    /*** Display who does which iterations for demonstration purposes ***/
+#pragma omp for nowait
+    for (i=0; i<rows_a; i++) {
+      for (k=0; k<cols_a; k++) {
+        for(j=0; j<cols_b; j++) {
+#ifdef APP_USE_INLINE_MULTIPLY
+          c[i][j] += multiply(a[i][k], b[k][j]);
+#else /* APP_USE_INLINE_MULTIPLY */
+          c[i][j] += a[i][k] * b[k][j];
+#endif /* APP_USE_INLINE_MULTIPLY */
+        }
+      }
+    }
+  }   /*** End of parallel region ***/
+}
+
+double do_work(void) {
+  double **a,           /* matrix A to be multiplied */
+  **b,           /* matrix B to be multiplied */
+  **c;           /* result matrix C */
+  a = allocateMatrix(NRA, NCA);
+  b = allocateMatrix(NCA, NCB);
+  c = allocateMatrix(NRA, NCB);  
+
+/*** Spawn a parallel region explicitly scoping all variables ***/
+
+  initialize(a, NRA, NCA);
+  initialize(b, NCA, NCB);
+  initialize(c, NRA, NCB);
+
+  compute(a, b, c, NRA, NCA, NCB);
+#if defined(TAU_OPENMP)
+  if (omp_get_nested()) {
+    compute_nested(a, b, c, NRA, NCA, NCB);
+  }
+#endif
+#ifdef TAU_MPI
+  if (provided == MPI_THREAD_MULTIPLE)
+  { 
+    int rank;
+    MPI_Comm_rank(MPI_COMM_WORLD, &rank);
+//    printf("Rank: %d: provided is MPI_THREAD_MULTIPLE\n", rank);
+  }
+#endif /* TAU_MPI */
+  compute_interchange(a, b, c, NRA, NCA, NCB);
+
+  return c[0][1]; 
+}
+
+void * threaded_func(void *data)
+{
+  do_work();
+  return NULL;
+}
+
+int main (int argc, char *argv[]) 
+{
+
+#ifdef PTHREADS
+  int ret;
+  pthread_attr_t  attr;
+  pthread_t       tid1, tid2, tid3;
+#endif /* PTHREADS */
+
+
+#ifdef TAU_MPI
+#if (defined(PTHREADS) || defined(TAU_OPENMP))
+  MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &provided);
+  printf("MPI_Init_thread: provided = %d, MPI_THREAD_MULTIPLE=%d\n", provided, MPI_THREAD_MULTIPLE);
+#else
+  MPI_Init(&argc, &argv); 
+#endif /* THREADS */
+#endif /* TAU_MPI */
+
+  TAU_SET_INTERRUPT_INTERVAL(1);
+  TAU_TRACK_POWER();
+#ifdef PTHREADS
+  if (ret = pthread_create(&tid1, NULL, threaded_func, NULL) )
+  {
+    printf("Error: pthread_create (1) fails ret = %d\n", ret);
+    exit(1);
+  }   
+
+  if (ret = pthread_create(&tid2, NULL, threaded_func, NULL) )
+  {
+    printf("Error: pthread_create (2) fails ret = %d\n", ret);
+    exit(1);
+  }   
+
+  if (ret = pthread_create(&tid3, NULL, threaded_func, NULL) )
+  {
+    printf("Error: pthread_create (3) fails ret = %d\n", ret);
+    exit(1);
+  }   
+
+#endif /* PTHREADS */
+
+/* On thread 0: */
+  do_work();
+
+#ifdef PTHREADS 
+  if (ret = pthread_join(tid1, NULL) )
+  {
+    printf("Error: pthread_join (1) fails ret = %d\n", ret);
+    exit(1);
+  }   
+
+  if (ret = pthread_join(tid2, NULL) )
+  {
+    printf("Error: pthread_join (2) fails ret = %d\n", ret);
+    exit(1);
+  }   
+
+  if (ret = pthread_join(tid3, NULL) )
+  {
+    printf("Error: pthread_join (3) fails ret = %d\n", ret);
+    exit(1);
+  }   
+
+#endif /* PTHREADS */
+
+#ifdef TAU_MPI
+  MPI_Finalize();
+#endif /* TAU_MPI */
+  printf ("Done.\n");
+
+  return 0;
+}
+

examples/power/matmult_initialize.c

@@ -0,0 +1,17 @@
+#include "matmult_initialize.h"
+
+void initialize(double **matrix, int rows, int cols) {
+  int i,j;
+#pragma omp parallel private(i,j) shared(matrix)
+  {
+    //set_num_threads();
+    /*** Initialize matrices ***/
+#pragma omp for nowait
+    for (i=0; i<rows; i++) {
+      for (j=0; j<cols; j++) {
+        matrix[i][j]= i+j;
+      }
+    }
+  }
+}
+

examples/power/matmult_initialize.h

@@ -0,0 +1,14 @@
+#ifndef _INIT_H_
+#define _INIT_H_
+
+#ifdef __cplus_plus 
+extern "C" {
+#endif /* __cplus_plus */
+
+extern void initialize(double **matrix, int rows, int cols);
+
+#ifdef __cplus_plus 
+}
+#endif /* __cplus_plus */
+
+#endif /* _INIT_H */

include/Profile/TauEnv.h

@@ -34,7 +34,7 @@
 #define TAU_ACTION_DUMP_BACKTRACES 3
 
 #ifndef TAU_EVENT_THRESHOLD
-#define TAU_EVENT_THRESHOLD .5
+#define TAU_EVENT_THRESHOLD_DEFAULT .5
 #endif /* TAU_EVENT_THRESHOLD */
 
 #ifdef __cplusplus

include/TAU.h.default

@@ -89,6 +89,12 @@
 #define TAU_DISABLE_TRACKING_MEMORY()
 #define TAU_TRACK_MEMORY()
 #define TAU_TRACK_MEMORY_HERE()
+#define TAU_TRACK_POWER()                       
+#define TAU_TRACK_POWER_HERE()                 
+#define TAU_ENABLE_TRACKING_POWER()           
+#define TAU_DISABLE_TRACKING_POWER()         
+
+
 #define TAU_SET_INTERRUPT_INTERVAL(value)
 
 #define CT(obj)

src/Profile/TauEnv.cpp

@@ -1336,11 +1336,12 @@ void TauEnv_initialize()
     }
 
     const char *evt_threshold = getconf("TAU_EVENT_THRESHOLD");
-    env_evt_threshold = TAU_EVENT_THRESHOLD;
+    env_evt_threshold = TAU_EVENT_THRESHOLD_DEFAULT;
     if (evt_threshold) {
-      sscanf(evt_threshold,"%g",&env_evt_threshold);
+      double evt_value = 0.0;
+      sscanf(evt_threshold,"%g",&evt_value);
+      env_evt_threshold = evt_value;
       TAU_METADATA("TAU_EVENT_THRESHOLD", evt_threshold);
-      printf("TAU_EVENT_THRESHOLD set to %g\n", env_evt_threshold);
     }
 
     const char *numcalls = getconf("TAU_THROTTLE_NUMCALLS");

src/Profile/TauShmemOpenShmemC.c

@@ -2,6 +2,7 @@
 #include <pshmem.h>
 #include <Profile/Profiler.h>
 #include <stdio.h>
+#include <stddef.h>
 int TAUDECL tau_totalnodes(int set_or_get, int value);
 static int tau_shmem_tagid_f=0 ; 
 #define TAU_SHMEM_TAGID tau_shmem_tagid_f=tau_shmem_tagid_f%250