Merge pull request #4 from fmihpc/performance_improvement

Performance improvement + fix to #2

example/cpp_test/Makefile

@@ -7,7 +7,7 @@ INCLUDES =
 
 # C++ compiler flags (-g -O2 -Wall)
 CCFLAGS = -O3 -I../../include 
-LDFLAGS = -L../../lib -lphiprof
+LDFLAGS = -L../../lib -lphiprof -lgomp
 # compiler
 CCC = mpiCC
 

example/cpp_test/test.cpp

@@ -96,6 +96,7 @@ int main(int argc,char **argv){
 
    if(rank==0)
       cout << "Measuring accuracy" <<endl;
+/*
    phiprof::start("Test accuracy");
    if(rank==0)
       cout << "  1/2" <<endl;
@@ -121,14 +122,13 @@ int main(int argc,char **argv){
    phiprof::stop("100x0.1s computations + logprofile"); 
 
    phiprof::stop("Test accuracy");
-
+*/   
    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);
-
    MPI_Finalize();
 }

include/phiprof.hpp

@@ -196,8 +196,16 @@ namespace phiprof
    * \overload  bool phiprof::stop(const std::string &label,double workUnits=-1.0,const std::string &workUnitLabel="")
    */
    bool stop (int id,
-              double workUnits=-1.0,
-              const std::string &workUnitLabel="");
+              double workUnits,
+              const std::string &workUnitLabel);
+
+
+
+   /**
+   * Fastest stop routine for cases when no workunits are defined.
+   */
+   bool stop (int id);
+
 
 
    /**

src/Makefile

@@ -9,8 +9,14 @@ OUT_NO = ../lib/libnophiprof.a
 # include directories
 INCLUDES = -I../include
 
+#Set clock.
+#CLOCK_MONOTONIC_COARSE is (much) faster, but has very poor
+# resolution. Also only available on newer linux kernels see man
+# clock_gettime
+CLOCK_ID = CLOCK_MONOTONIC
+
 # C++ compiler flags 
-CCFLAGS = -O3 -fopenmp # -W -Wall -Wextra -pedantic 
+CCFLAGS = -O3 -fopenmp -DCLOCK_ID=$(CLOCK_ID) # -W -Wall -Wextra -pedantic 
 
 # compiler
 CCC = mpiCC