Merge pull request #7 from fmihpc/improved-thread-support

Phiprof 2.0: Improved thread support

Author: galfthan

.gitignore

@@ -0,0 +1,11 @@
+*.dat
+*.vlsv
+*.silo
+*.o
+version.cpp
+\#*\#
+*~
+nbproject
+vlasiator
+vlasiator+pat
+nbproject

README

@@ -1,26 +1,29 @@
 Phiprof – Parallel Hierarchical Profiler
 
-Copyright (c) 2011,2012 Finnish Meteorological Institute
-Author: Sebastian von Alfthan (sebastian.von.alfthan at fmi.fi)
+Copyright (c) 2011 - 2015 Finnish Meteorological Institute
+Copyright (c) 2016   CSC - IT Center for Science
+
+Licence: LGPL 3
+
+Author: Sebastian von Alfthan (sebastian.von.alfthan at csc.fi)
+
 
 Phiprof a simple library that can be used to profile parallel MPI
-programs developed at the Finnish Meteorological Institute. It can be
-used to produce a hierarchical report of time (average, max, min)
-spent in different timer regions. A log format can also be written,
-where the performance as a function of time may be reported.
+programs developed originally at the Finnish Meteorological
+Institute. It can be used to produce a hierarchical report of time
+(average, max, min) spent in different timer regions. A log format can
+also be written, where the performance as a function of time may be
+reported.
 
 
 Features
   * Supports C++ and C (C programs have to be linked with a C++ compiler)
   * Low overhead (Less than 1 us)
   * Can print the total time as a human-readable hierarchical report
-  * Can print the evolution of time metrics in a easy-to-parse format to track how the performance develops over time
   * Automatically handles cases where groups of processes execute different codepaths
-  * (Cray specific) All timer regions are also defined as Craypat regions if Craypat module is loaded
 
 Not supported
   * No Fortran interface
-  * Phiprof is not threadsafe, it should only be called by the master thread, or outside of threaded regions.
   * Phiprof does not automatically add regions, these have to be added manually into the code.
 
 

example/c_test/Makefile

@@ -6,17 +6,17 @@ OBJ = $(SRC:.c=.o)
 INCLUDES = 
 
 # C compiler flags (-g -O2 -Wall)
-CFLAGS = -O3 -I../../include 
-LDFLAGS = -L../../lib -lphiprof
+CFLAGS = -O3 -I../../include -fopenmp
+LDFLAGS = -L../../lib -lphiprof  -lgomp -ldl
 # compiler
-CCC = mpicc
+CCC = cc
 # linker (c++ compiler)
-LNK = mpiCC
+LNK = CC
 
 .SUFFIXES: .c
 
 default: $(OBJ) 
-	$(LNK) -o test  $(OBJ) $(LDFLAGS) 
+	$(LNK) -o test   $(OBJ)  $(LDFLAGS) 
 
 .c.o:
 	$(CCC) $(INCLUDES) $(CFLAGS) -c $< -o $@

example/c_test/test.c

@@ -56,7 +56,9 @@ int main(int argc,char **argv){
    if(rank==0)
       printf( "Measuring performance of start-stop calls\n");
 
+   phiprof_initialize();
 
+
    phiprof_start("Benchmarking phiprof"); 
    phiprof_start("Initalized timers using ID");
 
@@ -97,33 +99,21 @@ int main(int argc,char **argv){
    phiprof_start("Test accuracy");
    if(rank==0)
       printf( "  1/2\n");
-   phiprof_start("100 computations"); 
+   phiprof_start("100 computations x 0.1s"); 
    for(i=0;i<100;i++){
       phiprof_start("compute");
       compute(0.1);
       phiprof_stop("compute");
    }
-   phiprof_stop("100 computations");
+   phiprof_stop("100 computations x 0.1s");
 
-   if(rank==0)
-      printf( "  2/2\n" );
-   MPI_Barrier(MPI_COMM_WORLD);
-   phiprof_start("100 computations + logprofile"); 
-   for(i=0;i<100;i++){
-      phiprof_start("compute");
-      compute(0.1);
-      phiprof_printLogProfile(MPI_COMM_WORLD,i,"profile_log_alllev"," ",0);
-      phiprof_printLogProfile(MPI_COMM_WORLD,i,"profile_log_maxlev1"," ",1);
-      phiprof_stop("compute");
-   }
-   phiprof_stop("100 computations + logprofile"); 
 
    phiprof_stop("Test accuracy");
 
    MPI_Barrier(MPI_COMM_WORLD);
    double t1=MPI_Wtime();
-   phiprof_print(MPI_COMM_WORLD,"profile_full",0.0);
-   phiprof_print(MPI_COMM_WORLD,"profile_minfrac0.01",0.01);
+   phiprof_print(MPI_COMM_WORLD,"profile");
+
    if(rank==0)   
       printf( "Print time is %g\n",MPI_Wtime()-t1);
 

example/cpp_test/Makefile

@@ -6,16 +6,16 @@ OBJ = $(SRC:.cpp=.o)
 INCLUDES = 
 
 # C++ compiler flags (-g -O2 -Wall)
-CCFLAGS = -O3 -I../../include 
-LDFLAGS = -L../../lib -lphiprof -lgomp
+CCFLAGS = -O3 -I../../include  -fopenmp
+LDFLAGS = -L../../lib -lphiprof  -lgomp -ldl
 # compiler
-CCC = mpiCC
+CCC = CC
 
 
 .SUFFIXES: .cpp
 
 default: $(OBJ) 
-	$(CCC) -o test  $(OBJ) $(LDFLAGS) 
+	$(CCC) -o test  $(OBJ) $(LDFLAGS)   
 
 .cpp.o:
 	$(CCC) $(INCLUDES) $(CCFLAGS) -c $< -o $@

example/cpp_test/test.cpp

@@ -17,13 +17,12 @@
   License along with this library.  If not, see <http://www.gnu.org/licenses/>.
 
 */
-
-
-#include "mpi.h"
 #include <iostream>
 #include <iomanip> 
 #include <string>    
 #include <vector>     
+#include "mpi.h"
+#include "omp.h"
 #include "phiprof.hpp"
 
 using namespace std;
@@ -58,9 +57,11 @@ int main(int argc,char **argv){
    const int nIterations=1000000;
    if(rank==0)
       cout << "Measuring performance of start-stop calls" <<endl;
-
 
-   phiprof::start("Benchmarking phiprofr"); 
+   phiprof::initialize();
+   
+   phiprof::start("Benchmarking phiprof"); 
+
    phiprof::start("Initalized timers using ID");
 
    if(rank==0)
@@ -74,6 +75,7 @@ int main(int argc,char **argv){
 
    if(rank==0)
       cout << "  2/3" <<endl;
+
    phiprof::start("Re-initialized timers using ID");
    for(int i=0;i<nIterations;i++){
       id_a=phiprof::initializeTimer("a","A with ID");
@@ -89,44 +91,68 @@ int main(int argc,char **argv){
       phiprof::start("a");
       phiprof::stop("a");
    }
-   phiprof::stop("Timers using labels",nIterations*2,"start-stop");
-   phiprof::stop("Benchmarking phiprofr"); 
+   phiprof::stop("Timers using labels", nIterations*2, "start-stop");
+   phiprof::stop("Benchmarking phiprof"); 
 
    MPI_Barrier(MPI_COMM_WORLD);
 
    if(rank==0)
       cout << "Measuring accuracy" <<endl;
-/*
+
+
    phiprof::start("Test accuracy");
+
    if(rank==0)
-      cout << "  1/2" <<endl;
-   phiprof::start("100x0.1s computations"); 
+      cout << "  1/3" <<endl;
+   phiprof::start("100x0.01s computations"); 
    for(int i=0;i<100;i++){
       phiprof::start("compute");
-      compute(0.1);
+      compute(0.01);
       phiprof::stop("compute");
    }
-   phiprof::stop("100x0.1s computations");
+   phiprof::stop("100x0.01s computations");
+
+   if(rank==0)
+      cout << "  2/3" <<endl;
+   MPI_Barrier(MPI_COMM_WORLD);
+
+   phiprof::start("100 x 0.01 (threadId + 1)s, ID");
+   int id = phiprof::initializeTimer("compute");
+#pragma omp parallel
+   for(int i=0;i<100;i++){
+      phiprof::start(id);
+      compute(0.01 * (omp_get_thread_num() + 1));
+      phiprof::stop(id);
+   }
+   phiprof::stop("100 x 0.01 (threadId + 1)s, ID"); 
 
    if(rank==0)
-      cout << "  2/2" <<endl;
+      cout << "  3/3" <<endl;
    MPI_Barrier(MPI_COMM_WORLD);
-   phiprof::start("100x0.1s computations + logprofile"); 
+
+   phiprof::start("100 x 0.01 (threadId + 1)s, String");
+#pragma omp parallel
    for(int i=0;i<100;i++){
       phiprof::start("compute");
-      compute(0.1);
-      phiprof::printLogProfile(MPI_COMM_WORLD,i);
-      phiprof::printLogProfile(MPI_COMM_WORLD,i,"profile_log_maxlev1"," ",1);
+      compute(0.01 * (omp_get_thread_num() + 1));
       phiprof::stop("compute");
    }
-   phiprof::stop("100x0.1s computations + logprofile"); 
+   phiprof::stop("100 x 0.01 (threadId + 1)s, String");
+
+
 
    phiprof::stop("Test accuracy");
-*/   
+
+   if(rank%2 == 1) {
+      
+      phiprof::start("Test-profile-groups");
+      phiprof::stop("Test-profile-groups");
+   }
+   
+
    MPI_Barrier(MPI_COMM_WORLD);
    double t1=MPI_Wtime();
    phiprof::print(MPI_COMM_WORLD);
-   phiprof::print(MPI_COMM_WORLD,"profile_minfrac0.01",0.01);
    if(rank==0)   
       cout<< "Print time is "<<MPI_Wtime()-t1<<endl;
 //   phiprof::print(MPI_COMM_WORLD,0.1);

include/phiprof.h

@@ -1,7 +1,8 @@
 /*
 This file is part of the phiprof library
 
-Copyright 2010, 2011, 2012 Finnish Meteorological Institute
+Copyright 2012, 2013, 2014, 2015 Finnish Meteorological Institute
+Copyright 2015, 2016 CSC - IT Center for Science 
 
 Phiprof is free software: you can redistribute it and/or modify
 it under the terms of the GNU Lesser General Public License as
@@ -34,9 +35,9 @@ License along with this library.  If not, see <http://www.gnu.org/licenses/>.
 #define GET_MACRO(_1,_2,NAME,...) NAME
 #define phiprof_assert(...) GET_MACRO(__VA_ARGS__, phiprof_assert2, phiprof_assert1)(__VA_ARGS__)
 
-
+int phiprof_initialize();
 int phiprof_initializeTimer(char *label,int nGroups, ... );
-int phiprof_getId(char *label);
+int phiprof_getChildId(char *label);
 
 int phiprof_start(char *label);
 int phiprof_stop(char *label);
@@ -45,8 +46,7 @@ int phiprof_startId(int id);
 int phiprof_stopId(int id);
 int phiprof_stopIdUnits(int id,double units,char *unitName);
 
-int phiprof_print(MPI_Comm comm, char *fileNamePrefix, double minFraction);
-int phiprof_printLogProfile(MPI_Comm comm, double simulationTime, char *fileNamePrefix, char *separator, int maxLevel);
+int phiprof_print(MPI_Comm comm, char *fileNamePrefix);
 
 void phiprof_phiprofAssert(int condition, char* errorMessage, char* fileName, int line);