Task 5, 5* some fixes and cleanup

clean_all.sh

@@ -0,0 +1,42 @@
+#!/usr/bin/env bash
+
+set -e
+
+echo "Searching for Makefiles..."
+
+# Find all directories containing a Makefile
+find . -type f \( -iname "makefile" -o -iname "Makefile" \) | while read -r mk; do
+    dir=$(dirname "$mk")
+
+    # Check if Makefile contains a clean target
+    if grep -qE '^[[:space:]]*clean[: ]' "$mk"; then
+        echo "→ Found clean target in: $dir"
+        echo "   Running make clean..."
+
+        (cd "$dir" && make clean)
+
+        echo "   Done."
+    else
+        echo "→ No clean target in: $dir"
+    fi
+done
+
+echo ""
+echo "─────────────────────────────"
+echo "Deleting extra files"
+echo "─────────────────────────────"
+
+echo "→ Deleting .Identifier files"
+find . -type f -name ".Identifier" -print -delete
+
+echo "→ Deleting *:Zone.Identifier files"
+find . -type f -name "*:Zone.Identifier" -print -delete
+
+echo "→ Deleting all .o object files"
+find . -type f -name "*.o" -print -delete
+
+echo "→ Deleting all .GCC_ files"
+find . -type f -name "*.GCC_" -print -delete
+
+echo ""
+echo "Cleanup complete."

sheet1/F/mayer_primes.h:Zone.Identifier

@@ -1,3 +0,0 @@
-[ZoneTransfer]
-ZoneId=3
-HostUrl=https://imsc.uni-graz.at/haasegu/Lectures/Math2CPP/Examples/goldbach/mayer_primes.h

sheet5/1/main.cpp

@@ -10,7 +10,7 @@
 #include <cmath>
 using namespace std;
 
-void benchmark(vector<double> &x, vector<double> &y, unsigned int N, unsigned int NLOOPS)
+void benchmark(vector<double> &x, vector<double> &y, unsigned int NLOOPS)
 {
     double sk = 0.0;
     for (int i = 0; i < NLOOPS; ++i)
@@ -92,8 +92,7 @@ int main(int argc, char const *argv[])
     for (int i = 0; i < NLOOPS; ++i)
     {
         sk = scalar(x, y);
-        sk = scalar_trans(x, y);
-        //sk = norm(x);
+        sk = norm(x);
     }
 
     double t1 = omp_get_wtime() - tstart;             // OpenMP
@@ -124,7 +123,7 @@ int main(int argc, char const *argv[])
     cout << "done\n";
     cout << vr << endl;
 
-    N=2;
+    N=200;
     //Data (re-)inizialiion
     for (unsigned int i = 0; i < N; ++i)
     {
@@ -147,9 +146,9 @@ int main(int argc, char const *argv[])
 
          omp_set_schedule(omp_sched_static, 0);
         tstart = omp_get_wtime();
-        benchmark(x, y, N, NLOOPS);
-        t1 = omp_get_wtime()/NLOOPS;
-        cout << "static (chunk 0) "<< (t1-tstart) << endl;
+        benchmark(x, y, NLOOPS);
+        t1 = (omp_get_wtime()-tstart)/NLOOPS;
+        cout << "static (chunk 0) "<< t1 << endl;
         for(int i=0; i<= 5; i++)
         {
             
@@ -159,30 +158,30 @@ int main(int argc, char const *argv[])
             // STATIC
             omp_set_schedule(omp_sched_static, chunk);
             tstart = omp_get_wtime();
-            benchmark(x, y, N, NLOOPS);
-            t1 = omp_get_wtime()/NLOOPS;
-            std::cout << "static:  " << (t1 - tstart) << " s\n";
+            benchmark(x, y, NLOOPS);
+            t1 = (omp_get_wtime()-tstart)/NLOOPS;
+            std::cout << "static:  "<< t1 << " s\n";
 
             // DYNAMIC
             omp_set_schedule(omp_sched_dynamic, chunk);
             tstart = omp_get_wtime();
-            benchmark(x, y, N, NLOOPS);
-            t1 = omp_get_wtime()/NLOOPS;
-            std::cout << "dynamic: " << (t1 - tstart) << " s\n";
+            benchmark(x, y, NLOOPS);
+            t1 = (omp_get_wtime()-tstart)/NLOOPS;
+            std::cout << "dynamic: "<< t1 << " s\n";
 
             // GUIDED
             omp_set_schedule(omp_sched_guided, chunk);
             tstart = omp_get_wtime();
-            benchmark(x, y, N, NLOOPS);
-            t1 = omp_get_wtime()/NLOOPS;
-            std::cout << "guided:  " << (t1 - tstart) << " s\n";
+            benchmark(x, y, NLOOPS);
+            t1 = (omp_get_wtime()-tstart)/NLOOPS;
+            std::cout << "guided:  "<< t1 << " s\n";
 
             // AUTO
             omp_set_schedule(omp_sched_auto, chunk);
             tstart = omp_get_wtime();
-            benchmark(x, y, N, NLOOPS);
-            t1 = omp_get_wtime()/NLOOPS;
-            std::cout << "auto:    " << (t1 - tstart) << " s\n";
+            benchmark(x, y, NLOOPS);
+            t1 = (omp_get_wtime()-tstart)/NLOOPS;
+            std::cout << "auto:    "<< t1 << " s\n";
             cout << endl;
         }
         cout << endl;
@@ -194,7 +193,9 @@ int main(int argc, char const *argv[])
      
     
     vector<int> vec = reduction_vec_append(N);
-    for(int i=0; i< N; i++)
+    unsigned int n = vec.size();
+    cout << "vec.size() = " << n << "\n";
+    for(int i=0; i< n; i++)
     {
         cout << vec[i] << ", ";
     }

sheet5/1/mylib.cpp

@@ -91,7 +91,7 @@ vector<int> reduction_vec(int n)
 
 vector<int> reduction_vec_append(int n)
 {
-    vector<int> vec(n);
+    vector<int> vec;
     #pragma omp parallel default(none) shared(cout,n) reduction(VecAppend:vec)
     {
         int tid = omp_get_thread_num();
@@ -99,9 +99,6 @@ vector<int> reduction_vec_append(int n)
         vector<int> local(n);
         iota(local.begin(), local.end(), tid);
 
-        #pragma omp critical
-        cout << tid << " : " << local.size() << endl;
-
         vec = local;
     }
 

sheet5/3/mayer_primes.h:Zone.Identifier

@@ -1,3 +0,0 @@
-[ZoneTransfer]
-ZoneId=3
-HostUrl=https://imsc.uni-graz.at/haasegu/Lectures/Math2CPP/Examples/goldbach/mayer_primes.h

sheet5/4/main.cpp

@@ -183,50 +183,63 @@ int main(int argc, char **argv)
 
 
 
+const int NLOOPS = 20;   
+
 for (int k = 3; k <= 8; ++k)
+{
+    unsigned int n = pow(10.0, k);
+
+    vector<double> x(n), y(n);
+    for (unsigned int  i = 0; i < n; ++i)
     {
-        size_t n = (size_t)pow(10, k);
-
-        vector<double> x(n), y(n);
-        for (unsigned int i = 0; i < n; ++i)
-        {
-            double xi= (i % 219) + 1;
-            x[i] = xi;
-            y[i] = 1.0 / xi;
-        }
-
-        // ---- SUM benchmark (sequential) ----
-        double t0 = omp_get_wtime();
-        double s1 = benchmark_A_sum_old(x);
-        double t_sum_seq = omp_get_wtime() - t0;
-
-        // ---- SUM benchmark (parallel) ----
-        t0 = omp_get_wtime();
-        double s2 = benchmark_A_sum(x);
-        double t_sum_omp = omp_get_wtime() - t0;
-
-        double sum_speedup = t_sum_seq / t_sum_omp;
-
-
-        // ---- INNER PRODUCT benchmark (sequential) ----
-        t0 = omp_get_wtime();
-        double ip1 = benchmark_A_old(x, y);
-        double t_inner_seq = omp_get_wtime() - t0;
-
-        // ---- INNER PRODUCT benchmark (parallel) ----
-        t0 = omp_get_wtime();
-        double ip2 = benchmark_A(x, y);
-        double t_inner_omp = omp_get_wtime() - t0;
-
-        double inner_speedup = t_inner_seq / t_inner_omp;
-
-        // ---- Print results ----
-        cout << k << endl;
-            cout << t_sum_seq << ", " << t_sum_omp << ", " << sum_speedup << endl;
-           cout << t_inner_seq << ", " << t_inner_omp << ", " << inner_speedup << endl;
-        cout << endl;
+        double xi = (i % 219) + 1;
+        x[i] = xi;
+        y[i] = 1.0 / xi;
     }
 
+    double s1_guard = 0.0, s2_guard = 0.0;
+    double ip1_guard = 0.0, ip2_guard = 0.0;
+
+    // ---- SUM benchmark (sequential) ----
+    double t0 = omp_get_wtime();
+    for (int r = 0; r < NLOOPS; ++r)
+        s1_guard += benchmark_A_sum_old(x);
+    double t_sum_seq = (omp_get_wtime() - t0) / NLOOPS;
+
+    // ---- SUM benchmark (parallel) ----
+    t0 = omp_get_wtime();
+    for (int r = 0; r < NLOOPS; ++r)
+        s2_guard += benchmark_A_sum(x);
+    double t_sum_omp = (omp_get_wtime() - t0) / NLOOPS;
+
+    double sum_speedup = t_sum_seq / t_sum_omp;
+
+    // ---- INNER PRODUCT benchmark (sequential) ----
+    t0 = omp_get_wtime();
+    for (int r = 0; r < NLOOPS; ++r)
+        ip1_guard += benchmark_A_old(x, y);
+    double t_inner_seq = (omp_get_wtime() - t0) / NLOOPS;
+
+    // ---- INNER PRODUCT benchmark (parallel) ----
+    t0 = omp_get_wtime();
+    for (int r = 0; r < NLOOPS; ++r)
+        ip2_guard += benchmark_A(x, y);
+    double t_inner_omp = (omp_get_wtime() - t0) / NLOOPS;
+
+    double inner_speedup = t_inner_seq / t_inner_omp;
+
+    // ---- Print results ----
+    std::cout << "k = " << k << " (n = 10^" << k << " = " << n << ")\n";
+    std::cout << "SUM    seq: " << t_sum_seq   << " s,  omp: " << t_sum_omp
+              << " s,  speedup = " << sum_speedup   << '\n';
+    std::cout << "INNER  seq: " << t_inner_seq << " s,  omp: " << t_inner_omp
+              << " s,  speedup = " << inner_speedup << '\n';
+    std::cout << "guards: "
+              << s1_guard << ", " << s2_guard << ", "
+              << ip1_guard << ", " << ip2_guard << "\n\n";
+}
+
+
 
 
 return 0;

sheet5/5/getmatrix.cpp

@@ -8,6 +8,7 @@
 #include <iostream>
 #include <list>
 #include <vector>
+#include <omp.h>
 using namespace std;
 
 
@@ -168,33 +169,64 @@ void CRS_Matrix::Debug() const
 
 void CRS_Matrix::CalculateLaplace(vector<double> &f)
 {
+    double t_start = omp_get_wtime();
+
+    double t0 ,t1,t2;
     assert(_mesh.NdofsElement() == 3);               // only for triangular, linear elements
     //cout << _nnz << " vs. " << _id[_nrows] << "  " << _nrows<< endl;
     assert(_nnz == _id[_nrows]);
 
-    for (int k = 0; k < _nrows; ++k)
+    #pragma omp parallel
     {
-        _sk[k] = 0.0;
-    }
-    for (int k = 0; k < _nrows; ++k)
-    {
-        f[k] = 0.0;
-    }
+        #pragma omp for
+        for (int k = 0; k < _nrows; ++k)
+            {
+                _sk[k] = 0.0;
+            }
 
-    double ske[3][3], fe[3];
-    //  Loop over all elements
-    auto const nelem = _mesh.Nelems();
-    auto const &ia   = _mesh.GetConnectivity();
-    auto const &xc   = _mesh.GetCoords();
+        #pragma omp barrier
+        #pragma omp single
+        t0 = omp_get_wtime();
+        
+
+        #pragma omp for   
+        for (int k = 0; k < _nrows; ++k)
+        {
+            f[k] = 0.0;
+        }
+
+        #pragma omp barrier
+        #pragma omp single
+        t1 = omp_get_wtime();
+       
+
+        double ske[3][3], fe[3];
+        //  Loop over all elements
+        auto const nelem = _mesh.Nelems();
+        auto const &ia   = _mesh.GetConnectivity();
+        auto const &xc   = _mesh.GetCoords();
+
+        #pragma omp barrier
+        #pragma omp single
+        t2 = omp_get_wtime();
+        
+        #pragma omp for
+        for (int i = 0; i < nelem; ++i)
+        {
+            CalcElem(ia.data() + 3 * i, xc.data(), ske, fe);
+            AddElem_3(ia.data() + 3 * i, ske, fe, f);
+        }
+        //Debug();
 
-    for (int i = 0; i < nelem; ++i)
-    {
-        CalcElem(ia.data() + 3 * i, xc.data(), ske, fe);
-        AddElem_3(ia.data() + 3 * i, ske, fe, f);
     }
-    //Debug();
+    double t3 = omp_get_wtime();
+    cout << "Zero matrix: " << (t0 - t_start) << " sec\n";
+    cout << "Zero RHS: " << (t1 - t0)     << " sec\n";
+    cout << "Element loop: " << (t3 - t2)    << " sec\n";
+    cout << "Total assembly:" << (t3 - t_start) << " sec\n";
+   
+
     
-    return;
 }
 
 void CRS_Matrix::ApplyDirichletBC(std::vector<double> const &u, std::vector<double> &f)
@@ -287,6 +319,7 @@ void CRS_Matrix::Defect(vector<double> &w,
     assert( _nrows == static_cast<int>(w.size()) );
     assert( w.size() == u.size() && u.size() == f.size() );
     
+    #pragma omp parallel for
     for (int row = 0; row < _nrows; ++row)
     {
         double wi = f[row];
@@ -340,8 +373,10 @@ void CRS_Matrix::AddElem_3(int const ial[3], double const ske[3][3], double cons
                 assert(ip >= 0);
             }
 #endif
+            #pragma omp atomic
             _sk[ip] += ske[i][j];
         }
+        #pragma omp atomic
         f[ii] += fe[i];
     }
 }

sheet5/5/jacsolve.cpp

@@ -23,6 +23,7 @@ void JacobiSolve(CRS_Matrix const &SK, vector<double> const &f, vector<double> &
 
     cout << endl << " Start Jacobi solver for " << nrows << " d.o.f.s"  << endl;
     //  Choose initial guess
+    #pragma omp parallel for
     for (int k = 0; k < nrows; ++k)
     {
         u[k] = 0.0;                          //  u := 0
@@ -44,6 +45,7 @@ void JacobiSolve(CRS_Matrix const &SK, vector<double> const &f, vector<double> &
     // Iteration sweeps
     int iter  = 0;
     double sigma = sigma0;
+    // i dont parallelize this cause its iteration dependent
     while ( sigma > tol2 * sigma0 && maxiter > iter)
     {
         ++iter;

sheet5/5/main.cpp

@@ -11,6 +11,7 @@
 #include <chrono>             // timing
 #include <cmath>
 #include <iostream>
+#include <omp.h>
 using namespace std;
 using namespace std::chrono;  // timing
 
@@ -19,7 +20,6 @@ int main(int, char ** )
 {
     const int numprocs = 1;
     const int myrank   = 0;
-
     if (myrank == 0)
     {
         cout << "\n There are " << numprocs << " processes running.\n \n";
@@ -121,6 +121,7 @@ int main(int, char ** )
 
             //mesh.Write_ascii_matlab("uv.txt", uv);
             //mesh.Visualize(uv);
+            
         }
         return 0;
     }