From 783eb6da9b7a31d856fd13958560a6fbe73332b4 Mon Sep 17 00:00:00 2001
From: Gael Guennebaud <g.gael@free.fr>
Date: Tue, 8 Jul 2008 17:25:58 +0000
Subject: [PATCH] I forgot that the previous commit needed minor changes
 outside the bench folder

---
 Eigen/src/QR/EigenSolver.h |  4 +-
 disabled/ompbench.cxxlist  |  7 ++++
 disabled/ompbenchmark.cpp  | 81 ++++++++++++++++++++++++++++++++++++++
 3 files changed, 90 insertions(+), 2 deletions(-)
 create mode 100644 disabled/ompbench.cxxlist
 create mode 100644 disabled/ompbenchmark.cpp

diff --git a/Eigen/src/QR/EigenSolver.h b/Eigen/src/QR/EigenSolver.h
index ac4ad9147..6beba231e 100644
--- a/Eigen/src/QR/EigenSolver.h
+++ b/Eigen/src/QR/EigenSolver.h
@@ -60,10 +60,10 @@ template<typename _MatrixType> class EigenSolver
 
     EigenvalueType eigenvalues(void) const { return m_eivalues; }
 
-  private:
-
     void compute(const MatrixType& matrix);
 
+  private:
+
     void orthes(MatrixType& matH, RealVectorType& ort);
     void hqr2(MatrixType& matH);
 
diff --git a/disabled/ompbench.cxxlist b/disabled/ompbench.cxxlist
new file mode 100644
index 000000000..fc6681d33
--- /dev/null
+++ b/disabled/ompbench.cxxlist
@@ -0,0 +1,7 @@
+#!/bin/bash
+
+CLIST[((g++))]="g++-4.2 -O3 -DNDEBUG -finline-limit=10000 -fopenmp"
+
+# CLIST[((g++))]="g++-4.3 -O3 -DNDEBUG -finline-limit=10000 -fopenmp"
+
+CLIST[((g++))]="icpc -fast -DNDEBUG -fno-exceptions -no-inline-max-size -openmp"
\ No newline at end of file
diff --git a/disabled/ompbenchmark.cpp b/disabled/ompbenchmark.cpp
new file mode 100644
index 000000000..ac5155cb8
--- /dev/null
+++ b/disabled/ompbenchmark.cpp
@@ -0,0 +1,81 @@
+// g++ -O3 -DNDEBUG -I.. -fopenmp benchOpenMP.cpp -o benchOpenMP && ./benchOpenMP 2> /dev/null
+// icpc -fast -fno-exceptions -DNDEBUG -I.. -openmp  benchOpenMP.cpp -o benchOpenMP && ./benchOpenMP 2> /dev/null
+
+#include <omp.h>
+#include "BenchUtil.h"
+#include "basicbenchmark.h"
+
+// #include <Eigen/Core>
+// #include "BenchTimer.h"
+//
+// using namespace std;
+// USING_PART_OF_NAMESPACE_EIGEN
+//
+// enum {LazyEval, EarlyEval, OmpEval};
+//
+// template<int Mode, typename MatrixType>
+// double benchSingleProc(const MatrixType& mat, int iterations, int tries)  __attribute__((noinline));
+//
+// template<int Mode, typename MatrixType>
+// double benchBasic(const MatrixType& mat, int iterations, int tries)
+// {
+//   const int rows = mat.rows();
+//   const int cols = mat.cols();
+//
+//   Eigen::BenchTimer timer;
+//   for(uint t=0; t<tries; ++t)
+//   {
+//     MatrixType I = MatrixType::identity(rows, cols);
+//     MatrixType m = MatrixType::random(rows, cols);
+//
+//     timer.start();
+//     for(int a = 0; a < iterations; a++)
+//     {
+//       if(Mode==LazyEval)
+//         m = (I + 0.00005 * (m + m.lazyProduct(m))).eval();
+//       else if(Mode==OmpEval)
+//         m = (I + 0.00005 * (m + m.lazyProduct(m))).evalOMP();
+//       else
+//         m = I + 0.00005 * (m + m * m);
+//     }
+//     timer.stop();
+//     cerr << m;
+//   }
+//   return timer.value();
+// };
+
+int main(int argc, char *argv[])
+{
+  // disbale floating point exceptions
+  // this leads to more stable bench results
+  {
+    int aux;
+    asm(
+    "stmxcsr   %[aux]           \n\t"
+    "orl       $32832, %[aux]   \n\t"
+    "ldmxcsr   %[aux]           \n\t"
+    : : [aux] "m" (aux));
+  }
+
+  // commented since the default setting is use as many threads as processors
+  //omp_set_num_threads(omp_get_num_procs());
+
+  std::cout << "double, fixed-size 4x4: "
+    << benchBasic<LazyEval>(Matrix4d(), 10000, 10) << "s  "
+    << benchBasic<OmpEval>(Matrix4d(), 10000, 10) << "s  \n";
+
+  #define BENCH_MATRIX(TYPE, SIZE, ITERATIONS, TRIES) {\
+      double single = benchBasic<LazyEval>(Matrix<TYPE,Eigen::Dynamic,Eigen::Dynamic>(SIZE,SIZE), ITERATIONS, TRIES); \
+      double omp    = benchBasic<OmpEval> (Matrix<TYPE,Eigen::Dynamic,Eigen::Dynamic>(SIZE,SIZE), ITERATIONS, TRIES); \
+      std::cout << #TYPE << ", " << #SIZE << "x" << #SIZE << ": " << single << "s " << omp << "s " \
+        << " => x" << single/omp << " (" << omp_get_num_procs() << ")" << std::endl; \
+    }
+
+  BENCH_MATRIX(double,   32, 1000, 10);
+  BENCH_MATRIX(double,  128,   10, 10);
+  BENCH_MATRIX(double,  512,    1,  6);
+  BENCH_MATRIX(double, 1024,    1,  4);
+
+  return 0;
+}
+