* basic support for multicore CPU via a .evalOMP() which

internaly uses OpenMP if enabled at compile time.
 * added a bench/ folder with a couple benchmarks and benchmark tools.

Eigen/src/Core/EvalOMP.h

@@ -0,0 +1,118 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra. Eigen itself is part of the KDE project.
+//
+// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
+// Copyright (C) 2006-2008 Benoit Jacob <jacob@math.jussieu.fr>
+//
+// Eigen is free software; you can redistribute it and/or
+// modify it under the terms of the GNU Lesser General Public
+// License as published by the Free Software Foundation; either
+// version 3 of the License, or (at your option) any later version.
+//
+// Alternatively, you can redistribute it and/or
+// modify it under the terms of the GNU General Public License as
+// published by the Free Software Foundation; either version 2 of
+// the License, or (at your option) any later version.
+//
+// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
+// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
+// GNU General Public License for more details.
+//
+// You should have received a copy of the GNU Lesser General Public
+// License and a copy of the GNU General Public License along with
+// Eigen. If not, see <http://www.gnu.org/licenses/>.
+
+#ifndef EIGEN_EVAL_OMP_H
+#define EIGEN_EVAL_OMP_H
+
+/** \class EvalOMP
+  *
+  * \brief Parallel evaluation of an expression using OpenMP
+  *
+  * The template parameter Expression is the type of the expression that we are evaluating.
+  *
+  * This class is the return type of MatrixBase::evalOMP() and most of the time this is the
+  * only way it is used.
+  *
+  * Note that if OpenMP is not enabled, then this class is equivalent to Eval.
+  *
+  * \sa MatrixBase::evalOMP(), class Eval, MatrixBase::eval()
+  */
+template<typename ExpressionType> class EvalOMP : NoOperatorEquals,
+  public Matrix< typename ExpressionType::Scalar,
+                 ExpressionType::Traits::RowsAtCompileTime,
+                 ExpressionType::Traits::ColsAtCompileTime,
+                 EIGEN_DEFAULT_MATRIX_STORAGE_ORDER,
+                 ExpressionType::Traits::MaxRowsAtCompileTime,
+                 ExpressionType::Traits::MaxColsAtCompileTime>
+{
+  public:
+    typedef typename ExpressionType::Scalar Scalar;
+
+    /** The actual matrix type to evaluate to. This type can be used independently
+      * of the rest of this class to get the actual matrix type to evaluate and store
+      * the value of an expression.
+      */
+    typedef Matrix<Scalar,
+                   ExpressionType::Traits::RowsAtCompileTime,
+                   ExpressionType::Traits::ColsAtCompileTime,
+                   EIGEN_DEFAULT_MATRIX_STORAGE_ORDER,
+                   ExpressionType::Traits::MaxRowsAtCompileTime,
+                   ExpressionType::Traits::MaxColsAtCompileTime> MatrixType;
+
+    #ifdef _OPENMP
+    explicit EvalOMP(const ExpressionType& other)
+      : MatrixType(other.rows(), other.cols())
+    {
+      #ifdef __INTEL_COMPILER
+      #pragma omp parallel default(none) shared(other)
+      #else
+      #pragma omp parallel default(none)
+      #endif
+      {
+        if (this->cols()>this->rows())
+        {
+          #pragma omp for
+          for(int j = 0; j < this->cols(); j++)
+            for(int i = 0; i < this->rows(); i++)
+              this->coeffRef(i, j) = other.coeff(i, j);
+        }
+        else
+        {
+          #pragma omp for
+          for(int i = 0; i < this->rows(); i++)
+            for(int j = 0; j < this->cols(); j++)
+              this->coeffRef(i, j) = other.coeff(i, j);
+        }
+      }
+    }
+    #else
+    explicit EvalOMP(const ExpressionType& other) : MatrixType(other) {}
+    #endif
+};
+
+/** Evaluates *this in a parallel fashion using OpenMP and returns the obtained matrix.
+  *
+  * Of course, it only makes sense to call this function for complex expressions, and/or
+  * large matrices (>32x32), \b and if there is no outer loop which can be parallelized.
+  *
+  * It is the responsibility of the user manage the OpenMP parameters, for instance:
+  * \code
+  * #include <omp.h>
+  * // ...
+  * omp_set_num_threads(omp_get_num_procs());
+  * \endcode
+  * You also need to enable OpenMP on your compiler (e.g., -fopenmp) during both compilation and linking.
+  *
+  * Note that if OpenMP is not enabled, then evalOMP() is equivalent to eval().
+  *
+  * \sa class EvalOMP, eval()
+  */
+template<typename Scalar, typename Derived>
+const EvalOMP<Derived> MatrixBase<Scalar, Derived>::evalOMP() const
+{
+  return EvalOMP<Derived>(*static_cast<const Derived*>(this));
+}
+
+#endif // EIGEN_EVAL_OMP_H

Eigen/src/Core/ForwardDeclarations.h

@@ -44,6 +44,7 @@ template<typename MatrixType> class DiagonalCoeffs;
 template<typename MatrixType> class Identity;
 template<typename MatrixType> class Map;
 template<typename Derived> class Eval;
+template<typename Derived> class EvalOMP;
 
 struct ScalarProductOp;
 struct ScalarQuotientOp;

Eigen/src/Core/MatrixBase.h

@@ -359,6 +359,7 @@ template<typename Scalar, typename Derived> class MatrixBase
     /// \name special functions
     //@{
     const Eval<Derived> eval() const EIGEN_ALWAYS_INLINE;
+    const EvalOMP<Derived> evalOMP() const EIGEN_ALWAYS_INLINE;
 
     template<typename CustomUnaryOp>
     const CwiseUnaryOp<CustomUnaryOp, Derived> cwise(const CustomUnaryOp& func = CustomUnaryOp()) const;