task8

ex3/seq/skalar_stl/Makefile

@@ -0,0 +1,30 @@
+#
+# use GNU-Compiler tools
+COMPILER=GCC_
+# alternatively from the shell
+# export COMPILER=GCC_
+# or, alternatively from the shell
+# make COMPILER=GCC_
+
+# use Intel compilers
+#COMPILER=ICC_
+
+# use PGI compilers
+# COMPILER=PGI_
+
+
+SOURCES = main.cpp mylib.cpp
+OBJECTS = $(SOURCES:.cpp=.o)
+
+PROGRAM	= main.${COMPILER}
+
+# uncomment the next to lines for debugging and detailed performance analysis
+CXXFLAGS += -g
+LINKFLAGS += -g
+# do not use -pg with PGI compilers
+
+ifndef COMPILER
+  COMPILER=GCC_
+endif
+
+include ../${COMPILER}default.mk

ex3/seq/skalar_stl/compile.log

@@ -0,0 +1 @@
+

ex3/seq/skalar_stl/main.cpp

@@ -0,0 +1,118 @@
+#include "mylib.h"
+#include <cassert>
+#include <chrono>           // timing
+#include <cmath>            // sqrt()
+#include <cstdlib>          // atoi()
+#include <cstring>          // strncmp()
+#include <ctime>
+#include <iostream>
+#include <sstream>
+using namespace std;
+using namespace std::chrono;  // timing
+
+int main(int argc, char **argv)
+{
+    int const NLOOPS = 50;        // chose a value such that the benchmark runs at least 10 sec.
+    unsigned int N = 50000001;
+//##########################################################################
+//   Read Paramater from command line  (C++ style)
+    cout << "Checking command line parameters for: -n <number> " << endl;
+    for (int i = 1; i < argc; i++)
+    {
+        cout << " arg[" << i << "] = " << argv[i] << endl;
+        if (std::strncmp(argv[i], "-n", 2) == 0 && i + 1 < argc) // found "-n" followed by another parameter
+        {
+            N = static_cast<unsigned int>(atoi(argv[i + 1]));
+        }
+        else
+        {
+            cout << "Corect call: " << argv[0] << " -n  <number>\n";
+        }
+    }
+
+    cout << "\nN = " << N << endl;
+
+//##########################################################################
+//  Memory allocation
+    cout << "Memory allocation\n";
+
+    vector<double> x(N), y(N);
+
+    cout.precision(2);
+    cout << 2.0 * N *sizeof(x[0]) / 1024 / 1024 / 1024 << " GByte Memory allocated\n";
+    cout.precision(6);
+
+//##########################################################################
+//  Data initialization
+//  Special:  x_i = i+1;  y_i = 1/x_i  ==> <x,y> == N
+    for (unsigned int i = 0; i < N; ++i)
+    {
+        x[i] = i + 1;
+        y[i] = 1.0 / x[i];
+    }
+
+//##########################################################################
+    cout << "\nStart Benchmarking\n";
+
+    auto t1 = system_clock::now(); // start timer
+// Do calculation
+    double sk(0.0),ss(0.0);
+    for (int i = 0; i < NLOOPS; ++i)
+    {
+        sk = scalar(x, y);
+        //sk = scalar_cblas(x, y);
+      //sk = norm(x);
+        ss += sk;                   // prevents the optimizer from removing unused calculation results.
+    }
+
+    auto t2 = system_clock::now();  // stop timer
+    auto duration = duration_cast<microseconds>(t2 - t1);        // duration in microseconds
+    double t_diff = static_cast<double>(duration.count()) / 1e6; // overall duration in seconds
+    t_diff = t_diff/NLOOPS;                                      // duration per loop seconds
+    
+    //assert(std::abs(ss/NLOOPS-sk)<1e-5);  // avoids unsafe floating point comparison "==" 
+
+//##########################################################################
+// Check the correct result
+    cout << "\n <x,y> = " << sk << endl;
+    if (static_cast<unsigned int>(sk) != N)
+    {
+        cout << "  !!   W R O N G  result   !!\n";
+    }
+    cout << endl;
+
+//##########################################################################
+// Timings  and Performance
+    cout << endl;
+    cout.precision(2);
+    cout << "Timing in sec. : " << t_diff << endl;
+    cout << "GFLOPS         : " << 2.0 * N / t_diff / 1024 / 1024 / 1024 << endl;
+    cout << "GiByte/s       : " << 2.0 * N / t_diff / 1024 / 1024 / 1024 * sizeof(x[0]) << endl;
+
+//##########################################################################
+    cout << "\nStart Benchmarking norm\n";
+
+    auto t3 = system_clock::now(); // start timer
+// Do calculation
+    double ss2(0.0);
+    for (int i = 0; i < NLOOPS; ++i)
+    {
+        auto sk = sqrt(scalar(x, x));
+        //auto sk = norm(x);         // the same timing as scalar(x, x)
+        ss2 += sk;                   // prevents the optimizer from removing unused calculation results.
+    }
+
+    auto t4 = system_clock::now();  // stop timer
+    auto duration2 = duration_cast<microseconds>(t4 - t3);        // duration in microseconds
+    double t_diff2 = static_cast<double>(duration2.count()) / 1e6; // overall duration in seconds
+    t_diff2 = t_diff2/NLOOPS;                                      // duration per loop seconds
+    
+    //assert(std::abs(ss/NLOOPS-sk)<1e-5);  // avoids unsafe floating point comparison "==" 
+    cout << "ss(norm): " << ss2 << endl;
+    cout << "Timing in sec. : " << t_diff2 << endl;
+
+//##########################################################################
+
+
+    return 0;
+}  // memory for x and y will be deallocated by their destructors

ex3/seq/skalar_stl/mylib.cpp

@@ -0,0 +1,65 @@
+#include "mylib.h"
+#include <cassert>       // assert()
+#include <cmath>
+#include <vector>
+
+#ifdef __INTEL_CLANG_COMPILER
+#pragma message(" ##########  Use of MKL  ###############")
+#include <mkl.h>
+#else
+#pragma message(" ##########  Use of CBLAS  ###############")
+//extern "C"
+//{
+#include <cblas.h>               // cBLAS Library
+#include <lapacke.h>             // Lapack
+//}
+#endif
+
+using namespace std;
+
+double scalar(vector<double> const &x, vector<double> const &y)
+{
+    assert(x.size() == y.size()); // switch off via compile flag: -DNDEBUG
+    size_t const N = x.size();
+    double sum = 0.0;
+    for (size_t i = 0; i < N; ++i)
+    {
+        sum += x[i] * y[i];
+        //sum += exp(x[i])*log(y[i]);
+    }
+    return sum;
+}
+
+
+double scalar_cblas(vector<double> const &x, vector<double> const &y)
+{
+    int const asize = static_cast<int>(size(x));
+    int const bsize = static_cast<int>(size(y));
+    assert(asize == bsize); // switch off via compile flag: -DNDEBUG
+	return cblas_ddot(asize,x.data(),1,y.data(),1);    
+    //assert(x.size() == y.size()); // switch off via compile flag: -DNDEBUG
+	//return cblas_ddot(x.size(),x.data(),1,y.data(),1);
+}
+
+float scalar_cblas(vector<float> const &x, vector<float> const &y)
+{
+    int const asize = static_cast<int>(size(x));
+    int const bsize = static_cast<int>(size(y));
+    assert(asize == bsize); // switch off via compile flag: -DNDEBUG
+	return cblas_sdot(asize,x.data(),1,y.data(),1);    
+    //assert(x.size() == y.size()); // switch off via compile flag: -DNDEBUG
+	//return cblas_ddot(x.size(),x.data(),1,y.data(),1);
+}
+
+
+double norm(vector<double> const &x)
+{
+    size_t const N = x.size();
+    double sum = 0.0;
+    for (size_t i = 0; i < N; ++i)
+    {
+        sum += x[i] * x[i];
+    }
+    return std::sqrt(sum);
+}
+

ex3/seq/skalar_stl/mylib.h

@@ -0,0 +1,30 @@
+#ifndef FILE_MYLIB
+#define FILE_MYLIB
+#include <vector>
+
+/** 	Inner product
+	@param[in] x	vector
+	@param[in] y	vector
+	@return 	    resulting Euclidian inner product <x,y>
+*/
+double scalar(std::vector<double> const &x, std::vector<double> const &y);
+
+/** 	Inner product using BLAS routines
+	@param[in] x	vector
+	@param[in] y	vector
+	@return 	    resulting Euclidian inner product <x,y>
+*/
+double scalar_cblas(std::vector<double> const &x, std::vector<double> const &y);
+float scalar_cblas(std::vector<float> const &x, std::vector<float> const &y);
+
+
+/** 	L_2 Norm of a vector
+	@param[in] x	vector
+	@return 	    resulting Euclidian norm <x,y>
+*/
+double norm(std::vector<double> const &x);
+
+
+
+
+#endif