* Start of the LU module, with matrix inversion already there and

  fully optimized.
* Even if LargeBit is set, only parallelize for large enough objects
  (controlled by EIGEN_PARALLELIZATION_TRESHOLD).

8 files changed, 40 insertions, 19 deletions

diff --git a/Eigen/src/Core/Assign.h b/Eigen/src/Core/Assign.h
index 37ad2c0cf..1b6e928d2 100644
--- a/Eigen/src/Core/Assign.h
+++ b/Eigen/src/Core/Assign.h
@@ -135,6 +135,11 @@ Derived& MatrixBase<Derived>
   }
 }
 
+template<typename T1, typename T2> bool ei_should_parallelize_assignment(const T1& t, const T2&)
+{
+  return (T1::Flags & T2::Flags & LargeBit) && t.size() >= EIGEN_PARALLELIZATION_TRESHOLD;
+}
+
 template <typename Derived, typename OtherDerived>
 struct ei_assignment_impl<Derived, OtherDerived, false>
 {
@@ -157,7 +162,7 @@ struct ei_assignment_impl<Derived, OtherDerived, false>
             for(int j = 0; j < dst.cols(); j++) \
               for(int i = 0; i < dst.rows(); i++) \
                 dst.coeffRef(i, j) = src.coeff(i, j);
-        EIGEN_RUN_PARALLELIZABLE_LOOP(Derived::Flags & OtherDerived::Flags & LargeBit)
+        EIGEN_RUN_PARALLELIZABLE_LOOP(ei_should_parallelize_assignment(dst, src))
         #undef EIGEN_THE_PARALLELIZABLE_LOOP
       }
       else
@@ -168,7 +173,7 @@ struct ei_assignment_impl<Derived, OtherDerived, false>
           for(int i = 0; i < dst.rows(); i++) \
             for(int j = 0; j < dst.cols(); j++) \
               dst.coeffRef(i, j) = src.coeff(i, j);
-        EIGEN_RUN_PARALLELIZABLE_LOOP(Derived::Flags & OtherDerived::Flags & LargeBit)
+        EIGEN_RUN_PARALLELIZABLE_LOOP(ei_should_parallelize_assignment(dst, src))
         #undef EIGEN_THE_PARALLELIZABLE_LOOP
       }
     }
@@ -198,7 +203,7 @@ struct ei_assignment_impl<Derived, OtherDerived, true>
         for(int i = 0; i < dst.rows(); i++) \
           for(int j = 0; j < dst.cols(); j+=ei_packet_traits<typename Derived::Scalar>::size) \
             dst.writePacketCoeff(i, j, src.packetCoeff(i, j));
-        EIGEN_RUN_PARALLELIZABLE_LOOP(Derived::Flags & OtherDerived::Flags & LargeBit)
+        EIGEN_RUN_PARALLELIZABLE_LOOP(ei_should_parallelize_assignment(dst, src))
         #undef EIGEN_THE_PARALLELIZABLE_LOOP
       }
       else
@@ -207,7 +212,7 @@ struct ei_assignment_impl<Derived, OtherDerived, true>
         for(int j = 0; j < dst.cols(); j++) \
           for(int i = 0; i < dst.rows(); i+=ei_packet_traits<typename Derived::Scalar>::size) \
             dst.writePacketCoeff(i, j, src.packetCoeff(i, j));
-        EIGEN_RUN_PARALLELIZABLE_LOOP(Derived::Flags & OtherDerived::Flags & LargeBit)
+        EIGEN_RUN_PARALLELIZABLE_LOOP(ei_should_parallelize_assignment(dst, src))
         #undef EIGEN_THE_PARALLELIZABLE_LOOP
       }
     }
diff --git a/Eigen/src/Core/Block.h b/Eigen/src/Core/Block.h
index c026b174c..0ff72075e 100644
--- a/Eigen/src/Core/Block.h
+++ b/Eigen/src/Core/Block.h
@@ -67,11 +67,11 @@ struct ei_traits<Block<MatrixType, BlockRows, BlockCols> >
       : (BlockRows==Dynamic ? MatrixType::MaxRowsAtCompileTime : BlockRows),
     MaxColsAtCompileTime = ColsAtCompileTime == 1 ? 1
       : (BlockCols==Dynamic ? MatrixType::MaxColsAtCompileTime : BlockCols),
-    FlagsLargeBit = ((RowsAtCompileTime != Dynamic && MatrixType::RowsAtCompileTime == Dynamic)
-                  || (ColsAtCompileTime != Dynamic && MatrixType::ColsAtCompileTime == Dynamic))
-                  ? ~LargeBit
-                  : ~(unsigned int)0,
-    Flags = MatrixType::Flags & FlagsLargeBit & ~VectorizableBit,
+    FlagsMaskLargeBit = ((RowsAtCompileTime != Dynamic && MatrixType::RowsAtCompileTime == Dynamic)
+                      || (ColsAtCompileTime != Dynamic && MatrixType::ColsAtCompileTime == Dynamic))
+                      ? ~LargeBit
+                      : ~(unsigned int)0,
+    Flags = MatrixType::Flags & FlagsMaskLargeBit & ~VectorizableBit,
     CoeffReadCost = MatrixType::CoeffReadCost
   };
 };
diff --git a/Eigen/src/Core/Matrix.h b/Eigen/src/Core/Matrix.h
index e79cbcbfb..2f65413da 100644
--- a/Eigen/src/Core/Matrix.h
+++ b/Eigen/src/Core/Matrix.h
@@ -95,11 +95,8 @@ struct ei_traits<Matrix<_Scalar, _Rows, _Cols, _Flags, _MaxRows, _MaxCols> >
   };
 };
 
-template<typename _Scalar, int _Rows, int _Cols,
-         unsigned int _Flags = EIGEN_DEFAULT_MATRIX_FLAGS,
-         int _MaxRows = _Rows, int _MaxCols = _Cols>
-class Matrix : public MatrixBase<Matrix<_Scalar, _Rows, _Cols,
-                                        _Flags, _MaxRows, _MaxCols> >
+template<typename _Scalar, int _Rows, int _Cols, unsigned int _Flags, int _MaxRows, int _MaxCols>
+class Matrix : public MatrixBase<Matrix<_Scalar, _Rows, _Cols, _Flags, _MaxRows, _MaxCols> >
 {
   public:
 
diff --git a/Eigen/src/Core/MatrixBase.h b/Eigen/src/Core/MatrixBase.h
index 151ee74d9..d5912c9e6 100644
--- a/Eigen/src/Core/MatrixBase.h
+++ b/Eigen/src/Core/MatrixBase.h
@@ -506,6 +506,15 @@ template<typename Derived> class MatrixBase
     { return *static_cast<Derived*>(const_cast<MatrixBase*>(this)); }
     //@}
 
+    /** \name LU module
+      *
+      * \code #include <Eigen/LU> \endcode
+      */
+    //@{
+    const Inverse<Derived, true> inverse() const;
+    const Inverse<Derived, false> quickInverse() const;
+    //@}
+
   private:
 
     PacketScalar _packetCoeff(int , int) const { ei_internal_assert(false && "_packetCoeff not defined"); }
diff --git a/Eigen/src/Core/Product.h b/Eigen/src/Core/Product.h
index 6771b5d30..b593825f8 100644
--- a/Eigen/src/Core/Product.h
+++ b/Eigen/src/Core/Product.h
@@ -280,6 +280,8 @@ void Product<Lhs,Rhs,EvalMode>::_cacheOptimalEval(DestDerived& res) const
 {
   res.setZero();
   const int cols4 = m_lhs.cols() & 0xfffffffC;
+  const bool should_parallelize = (Flags & DestDerived::Flags & LargeBit)
+                                && res.size() >= EIGEN_PARALLELIZATION_TRESHOLD;
   #ifdef EIGEN_VECTORIZE
   if( (Flags & VectorizableBit) && (!(Lhs::Flags & RowMajorBit)) )
   {
@@ -318,7 +320,7 @@ void Product<Lhs,Rhs,EvalMode>::_cacheOptimalEval(DestDerived& res) const
             res.writePacketCoeff(i,k,ei_pmul(tmp, m_lhs.packetCoeff(i,j))); \
         } \
       }
-    EIGEN_RUN_PARALLELIZABLE_LOOP(Flags & DestDerived::Flags & LargeBit)
+    EIGEN_RUN_PARALLELIZABLE_LOOP(should_parallelize)
     #undef EIGEN_THE_PARALLELIZABLE_LOOP
   }
   else
@@ -345,7 +347,7 @@ void Product<Lhs,Rhs,EvalMode>::_cacheOptimalEval(DestDerived& res) const
             res.coeffRef(i,k) += tmp * m_lhs.coeff(i,j); \
         } \
       }
-    EIGEN_RUN_PARALLELIZABLE_LOOP(Flags & DestDerived::Flags & LargeBit)
+    EIGEN_RUN_PARALLELIZABLE_LOOP(should_parallelize)
     #undef EIGEN_THE_PARALLELIZABLE_LOOP
   }
 }
diff --git a/Eigen/src/Core/util/ForwardDeclarations.h b/Eigen/src/Core/util/ForwardDeclarations.h
index 6167f13bd..86eaf4e54 100644
--- a/Eigen/src/Core/util/ForwardDeclarations.h
+++ b/Eigen/src/Core/util/ForwardDeclarations.h
@@ -29,7 +29,11 @@ template<typename T> struct ei_traits;
 template<typename Lhs, typename Rhs> struct ei_product_eval_mode;
 template<typename T> struct NumTraits;
 
-template<typename _Scalar, int _Rows, int _Cols, unsigned int _Flags, int _MaxRows, int _MaxCols> class Matrix;
+template<typename _Scalar, int _Rows, int _Cols,
+         unsigned int _Flags = EIGEN_DEFAULT_MATRIX_FLAGS,
+         int _MaxRows = _Rows, int _MaxCols = _Cols>
+class Matrix;
+
 template<typename ExpressionType> class Lazy;
 template<typename ExpressionType> class Temporary;
 template<typename MatrixType> class Minor;
@@ -47,7 +51,6 @@ 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;
 template<int Direction, typename UnaryOp, typename MatrixType> class PartialRedux;
 
 template<typename Scalar> struct ei_scalar_sum_op;
@@ -70,4 +73,6 @@ template<typename Scalar> struct ei_scalar_quotient1_op;
 template<typename Scalar> struct ei_scalar_min_op;
 template<typename Scalar> struct ei_scalar_max_op;
 
+template<typename ExpressionType, bool CheckExistence = true> class Inverse;
+
 #endif // EIGEN_FORWARDDECLARATIONS_H
diff --git a/Eigen/src/Core/util/Macros.h b/Eigen/src/Core/util/Macros.h
index 43d595451..fad046766 100644
--- a/Eigen/src/Core/util/Macros.h
+++ b/Eigen/src/Core/util/Macros.h
@@ -37,6 +37,10 @@
 #define EIGEN_UNROLLING_LIMIT 400
 #endif
 
+#ifndef EIGEN_PARALLELIZATION_TRESHOLD
+#define EIGEN_PARALLELIZATION_TRESHOLD 2000
+#endif
+
 #ifdef EIGEN_DEFAULT_TO_ROW_MAJOR
 #define EIGEN_DEFAULT_MATRIX_STORAGE_ORDER RowMajorBit
 #else
diff --git a/Eigen/src/Core/util/Meta.h b/Eigen/src/Core/util/Meta.h
index faf177473..4939128eb 100644
--- a/Eigen/src/Core/util/Meta.h
+++ b/Eigen/src/Core/util/Meta.h
@@ -182,5 +182,4 @@ template<typename T> struct ei_packet_traits
   enum {size=1};
 };
 
-
 #endif // EIGEN_META_H