performance test sources

performance/Makefile

@@ -0,0 +1,39 @@
+# Copyright 2011-2014 Mario Mulansky
+# Copyright 2011-2014 Karsten Ahnert
+#
+# Distributed under the Boost Software License, Version 1.0.
+# (See accompanying file LICENSE_1_0.txt or
+# copy at http://www.boost.org/LICENSE_1_0.txt)
+
+# make sure BOOST_ROOT is pointing to your boost directory
+# otherwise, set it here:
+# BOOST_ROOT = /path/to/boost
+
+INCLUDES += -I../../include/ -I$(BOOST_ROOT)
+GCCFLAGS = -O3 -ffast-math -DNDEBUG
+# disabling -ffast-math might give slightly better performance
+ICCFLAGS = -fast -ip -inline-forceinline -DNDEBUG
+# Possible options: -fp-model source -no-fma
+GFORTFLAGS = -Ofast
+
+bin/gcc:
+	mkdir -p bin/gcc
+
+bin/intel:
+	mkdir -p bin/intel
+
+bin/gfort:
+	mkdir -p bin/gfort
+
+bin/gcc/odeint_rk4_array: odeint_rk4_array.cpp bin/gcc
+	g++ ${GCCFLAGS} ${INCLUDES} -o bin/gcc/odeint_rk4_array odeint_rk4_array.cpp
+
+
+bin/intel/odeint_rk4_array: odeint_rk4_array.cpp bin/intel
+	icpc ${ICCFLAGS} ${INCLUDES} -o bin/intel/odeint_rk4_array odeint_rk4_array.cpp
+
+
+bin/gfort/fortran_lorenz: fortran_lorenz.f90 bin/gfort
+	gfortran ${GFORTFLAGS} fortran_lorenz.f90 -o bin/gfort/fortran_lorenz
+
+all: bin/gcc/odeint_rk4_array bin/intel/odeint_rk4_array bin/gfort/fortran_lorenz

performance/fortran_lorenz.f90

@@ -0,0 +1,57 @@
+program main
+    implicit none
+
+    integer, parameter :: dp = 8
+    real(dp), dimension(1:3) :: x
+    integer, parameter :: nstep = 20000000
+    real(dp) :: t = 0.0_dp
+    real(dp) :: h = 1.0e-10_dp
+    integer, parameter :: nb_loops = 21
+    integer, parameter :: n = 3
+    integer :: k
+    integer :: time_begin
+    integer :: time_end
+    integer :: count_rate
+    real(dp) :: time
+
+    do k = 1, nb_loops
+        x = [ 8.5_dp, 3.1_dp, 1.2_dp ]
+        call system_clock(time_begin, count_rate)
+        call rk4sys(n, t, x, h, nstep)
+        call system_clock(time_end, count_rate)
+        time = real(time_end - time_begin, dp) / real(count_rate, dp)
+        write (*,*) time, x(1)
+    end do
+contains
+    subroutine xpsys(x,f)
+        real(dp), dimension(1:3), intent(in) :: x
+        real(dp), dimension(1:3), intent(out) :: f
+        f(1) = 10.0_dp * ( x(2) - x(1) )
+        f(2) = 28.0_dp * x(1) - x(2) - x(1) * x(3)
+        f(3) = x(1) * x(2) - (8.0_dp / 3.0_dp) * x(3)
+    end subroutine xpsys
+
+    subroutine rk4sys(n, t, x, h, nstep)
+        integer, intent(in) :: n
+        real(dp), intent(in) :: t
+        real(dp), dimension(1:n), intent(inout) :: x
+        real(dp), intent(in) :: h
+        integer, intent(in) :: nstep
+        ! Local variables
+        real(dp) :: h2
+        real(dp), dimension(1:n) :: y, f1, f2, f3, f4
+        integer :: i, k
+
+        h2 = 0.5_dp * h
+        do k = 1, nstep
+            call xpsys(x, f1)
+            y = x + h2 * f1
+            call xpsys(y, f2)
+            y = x + h2 * f2
+            call xpsys(y, f3)
+            y = x + h * f3
+            call xpsys(y, f4)
+            x = x + h * (f1 + 2.0_dp * (f2 + f3) + f4) / 6.0_dp
+        end do
+    end subroutine rk4sys
+end program main

performance/odeint_rk4_array.cpp

@@ -0,0 +1,55 @@
+/*
+ * odeint_rk4_lorenz_def_alg.cpp
+ *
+ * Copyright 2011 Mario Mulansky
+ * Copyright 2012 Karsten Ahnert
+ *
+ * Distributed under the Boost Software License, Version 1.0.
+ * (See accompanying file LICENSE_1_0.txt or
+ * copy at http://www.boost.org/LICENSE_1_0.txt)
+ */
+
+
+#include <boost/timer.hpp>
+#include <boost/array.hpp>
+
+#include <boost/numeric/odeint/stepper/runge_kutta4_classic.hpp>
+#include <boost/numeric/odeint/stepper/runge_kutta4.hpp>
+#include <boost/numeric/odeint/algebra/array_algebra.hpp>
+
+#include "lorenz.hpp"
+
+typedef boost::timer timer_type;
+
+typedef boost::array< double , 3 > state_type;
+
+using namespace boost::numeric::odeint;
+//typedef boost::numeric::odeint::runge_kutta4_classic< state_type > rk4_odeint_type;
+typedef runge_kutta4_classic< state_type , double , state_type , double ,
+                              range_algebra, default_operations, never_resizer > rk4_odeint_type;
+
+
+const int loops = 21;
+const int num_of_steps = 20000000;
+const double dt = 1E-10;
+
+
+int main()
+{
+    double min_time = 1E6; // something big
+    rk4_odeint_type stepper;
+    for( int n=0; n<loops; n++ )
+    {
+        state_type x = {{ 8.5, 3.1, 1.2 }};
+        double t = 0.0;
+        timer_type timer;
+        for( size_t i = 0 ; i < num_of_steps ; ++i )
+        {
+            stepper.do_step( lorenz(), x, t, dt );
+            t += dt;
+        }
+        min_time = std::min( timer.elapsed() , min_time );
+        std::clog << timer.elapsed() << '\t' << x[0] << std::endl;
+    }
+    std::cout << "Minimal Runtime: " << min_time << std::endl;
+}