merge default and evaluator branches

9 files changed, 60 insertions, 32 deletions

diff --git a/Eigen/src/Core/ArrayWrapper.h b/Eigen/src/Core/ArrayWrapper.h
index 21830745b..4bb648024 100644
--- a/Eigen/src/Core/ArrayWrapper.h
+++ b/Eigen/src/Core/ArrayWrapper.h
@@ -49,7 +49,7 @@ class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
     typedef typename internal::nested<ExpressionType>::type NestedExpressionType;
 
     EIGEN_DEVICE_FUNC
-    inline ArrayWrapper(ExpressionType& matrix) : m_expression(matrix) {}
+    EIGEN_STRONG_INLINE ArrayWrapper(ExpressionType& matrix) : m_expression(matrix) {}
 
     EIGEN_DEVICE_FUNC
     inline Index rows() const { return m_expression.rows(); }
diff --git a/Eigen/src/Core/CommaInitializer.h b/Eigen/src/Core/CommaInitializer.h
index 2bbf74b05..70cbfeff5 100644
--- a/Eigen/src/Core/CommaInitializer.h
+++ b/Eigen/src/Core/CommaInitializer.h
@@ -45,6 +45,18 @@ struct CommaInitializer
     m_xpr.block(0, 0, other.rows(), other.cols()) = other;
   }
 
+  /* Copy/Move constructor which transfers ownership. This is crucial in 
+   * absence of return value optimization to avoid assertions during destruction. */
+  // FIXME in C++11 mode this could be replaced by a proper RValue constructor
+  EIGEN_DEVICE_FUNC
+  inline CommaInitializer(const CommaInitializer& o)
+  : m_xpr(o.m_xpr), m_row(o.m_row), m_col(o.m_col), m_currentBlockRows(o.m_currentBlockRows) {
+    // Mark original object as finished. In absence of R-value references we need to const_cast:
+    const_cast<CommaInitializer&>(o).m_row = m_xpr.rows();
+    const_cast<CommaInitializer&>(o).m_col = m_xpr.cols();
+    const_cast<CommaInitializer&>(o).m_currentBlockRows = 0;
+  }
+
   /* inserts a scalar value in the target matrix */
   EIGEN_DEVICE_FUNC
   CommaInitializer& operator,(const Scalar& s)
@@ -110,7 +122,7 @@ struct CommaInitializer
   EIGEN_DEVICE_FUNC
   inline XprType& finished() { return m_xpr; }
 
-  XprType& m_xpr;   // target expression
+  XprType& m_xpr;           // target expression
   Index m_row;              // current row id
   Index m_col;              // current col id
   Index m_currentBlockRows; // current block height
diff --git a/Eigen/src/Core/MatrixBase.h b/Eigen/src/Core/MatrixBase.h
index 1f9f295c9..b95b3e6d8 100644
--- a/Eigen/src/Core/MatrixBase.h
+++ b/Eigen/src/Core/MatrixBase.h
@@ -378,8 +378,6 @@ template<typename Derived> class MatrixBase
 
     Scalar trace() const;
 
-/////////// Array module ///////////
-
     template<int p> EIGEN_DEVICE_FUNC RealScalar lpNorm() const;
 
     EIGEN_DEVICE_FUNC MatrixBase<Derived>& matrix() { return *this; }
@@ -387,8 +385,10 @@ template<typename Derived> class MatrixBase
 
     /** \returns an \link Eigen::ArrayBase Array \endlink expression of this matrix
       * \sa ArrayBase::matrix() */
-    EIGEN_DEVICE_FUNC ArrayWrapper<Derived> array() { return derived(); }
-    EIGEN_DEVICE_FUNC const ArrayWrapper<const Derived> array() const { return derived(); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE ArrayWrapper<Derived> array() { return derived(); }
+    /** \returns a const \link Eigen::ArrayBase Array \endlink expression of this matrix
+      * \sa ArrayBase::matrix() */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const ArrayWrapper<const Derived> array() const { return derived(); }
 
 /////////// LU module ///////////
 
diff --git a/Eigen/src/Core/Transpose.h b/Eigen/src/Core/Transpose.h
index 11b0e45a8..079f21fd9 100644
--- a/Eigen/src/Core/Transpose.h
+++ b/Eigen/src/Core/Transpose.h
@@ -300,7 +300,8 @@ struct inplace_transpose_selector<MatrixType,false> { // non square matrix
   * Notice however that this method is only useful if you want to replace a matrix by its own transpose.
   * If you just need the transpose of a matrix, use transpose().
   *
-  * \note if the matrix is not square, then \c *this must be a resizable matrix.
+  * \note if the matrix is not square, then \c *this must be a resizable matrix. 
+  * This excludes (non-square) fixed-size matrices, block-expressions and maps.
   *
   * \sa transpose(), adjoint(), adjointInPlace() */
 template<typename Derived>
@@ -331,6 +332,7 @@ inline void DenseBase<Derived>::transposeInPlace()
   * If you just need the adjoint of a matrix, use adjoint().
   *
   * \note if the matrix is not square, then \c *this must be a resizable matrix.
+  * This excludes (non-square) fixed-size matrices, block-expressions and maps.
   *
   * \sa transpose(), adjoint(), transposeInPlace() */
 template<typename Derived>
diff --git a/Eigen/src/Core/products/GeneralBlockPanelKernel.h b/Eigen/src/Core/products/GeneralBlockPanelKernel.h
index 780fa74d3..686ff84f1 100644
--- a/Eigen/src/Core/products/GeneralBlockPanelKernel.h
+++ b/Eigen/src/Core/products/GeneralBlockPanelKernel.h
@@ -1128,6 +1128,8 @@ EIGEN_DONT_INLINE void gemm_pack_lhs<Scalar, Index, Pack1, Pack2, StorageOrder,
   enum { PacketSize = packet_traits<Scalar>::size };
 
   EIGEN_ASM_COMMENT("EIGEN PRODUCT PACK LHS");
+  EIGEN_UNUSED_VARIABLE(stride);
+  EIGEN_UNUSED_VARIABLE(offset);
   eigen_assert(((!PanelMode) && stride==0 && offset==0) || (PanelMode && stride>=depth && offset<=stride));
   eigen_assert( (StorageOrder==RowMajor) || ((Pack1%PacketSize)==0 && Pack1<=4*PacketSize) );
   conj_if<NumTraits<Scalar>::IsComplex && Conjugate> cj;
@@ -1215,6 +1217,8 @@ EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, nr, ColMajor, Conjugate, Pan
   ::operator()(Scalar* blockB, const Scalar* rhs, Index rhsStride, Index depth, Index cols, Index stride, Index offset)
 {
   EIGEN_ASM_COMMENT("EIGEN PRODUCT PACK RHS COLMAJOR");
+  EIGEN_UNUSED_VARIABLE(stride);
+  EIGEN_UNUSED_VARIABLE(offset);
   eigen_assert(((!PanelMode) && stride==0 && offset==0) || (PanelMode && stride>=depth && offset<=stride));
   conj_if<NumTraits<Scalar>::IsComplex && Conjugate> cj;
   Index packet_cols = (cols/nr) * nr;
@@ -1257,6 +1261,7 @@ EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, nr, ColMajor, Conjugate, Pan
 template<typename Scalar, typename Index, int nr, bool Conjugate, bool PanelMode>
 struct gemm_pack_rhs<Scalar, Index, nr, RowMajor, Conjugate, PanelMode>
 {
+  typedef typename packet_traits<Scalar>::type Packet;
   enum { PacketSize = packet_traits<Scalar>::size };
   EIGEN_DONT_INLINE void operator()(Scalar* blockB, const Scalar* rhs, Index rhsStride, Index depth, Index cols, Index stride=0, Index offset=0);
 };
@@ -1266,6 +1271,8 @@ EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, nr, RowMajor, Conjugate, Pan
   ::operator()(Scalar* blockB, const Scalar* rhs, Index rhsStride, Index depth, Index cols, Index stride, Index offset)
 {
   EIGEN_ASM_COMMENT("EIGEN PRODUCT PACK RHS ROWMAJOR");
+  EIGEN_UNUSED_VARIABLE(stride);
+  EIGEN_UNUSED_VARIABLE(offset);
   eigen_assert(((!PanelMode) && stride==0 && offset==0) || (PanelMode && stride>=depth && offset<=stride));
   conj_if<NumTraits<Scalar>::IsComplex && Conjugate> cj;
   Index packet_cols = (cols/nr) * nr;
@@ -1276,12 +1283,18 @@ EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, nr, RowMajor, Conjugate, Pan
     if(PanelMode) count += nr * offset;
     for(Index k=0; k<depth; k++)
     {
-      const Scalar* b0 = &rhs[k*rhsStride + j2];
-                blockB[count+0] = cj(b0[0]);
-                blockB[count+1] = cj(b0[1]);
-      if(nr==4) blockB[count+2] = cj(b0[2]);
-      if(nr==4) blockB[count+3] = cj(b0[3]);
-      count += nr;
+      if (nr == PacketSize) {
+        Packet A = ploadu<Packet>(&rhs[k*rhsStride + j2]);
+        pstoreu(blockB+count, cj.pconj(A));
+        count += PacketSize;
+      } else {
+        const Scalar* b0 = &rhs[k*rhsStride + j2];
+                  blockB[count+0] = cj(b0[0]);
+                  blockB[count+1] = cj(b0[1]);
+        if(nr==4) blockB[count+2] = cj(b0[2]);
+        if(nr==4) blockB[count+3] = cj(b0[3]);
+        count += nr;
+      }
     }
     // skip what we have after
     if(PanelMode) count += nr * (stride-offset-depth);
diff --git a/Eigen/src/Core/products/GeneralMatrixVector.h b/Eigen/src/Core/products/GeneralMatrixVector.h
index c8697c913..a73ce5ff0 100644
--- a/Eigen/src/Core/products/GeneralMatrixVector.h
+++ b/Eigen/src/Core/products/GeneralMatrixVector.h
@@ -80,11 +80,8 @@ EIGEN_DONT_INLINE static void run(
   Index rows, Index cols,
   const LhsScalar* lhs, Index lhsStride,
   const RhsScalar* rhs, Index rhsIncr,
-  ResScalar* res, Index
-  #ifdef EIGEN_INTERNAL_DEBUGGING
-    resIncr
-  #endif
-  , RhsScalar alpha);
+        ResScalar* res, Index resIncr,
+  RhsScalar alpha);
 };
 
 template<typename Index, typename LhsScalar, bool ConjugateLhs, typename RhsScalar, bool ConjugateRhs, int Version>
@@ -92,12 +89,10 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,ColMajor,Co
   Index rows, Index cols,
   const LhsScalar* lhs, Index lhsStride,
   const RhsScalar* rhs, Index rhsIncr,
-  ResScalar* res, Index
-  #ifdef EIGEN_INTERNAL_DEBUGGING
-    resIncr
-  #endif
-  , RhsScalar alpha)
+        ResScalar* res, Index resIncr,
+  RhsScalar alpha)
 {
+  EIGEN_UNUSED_VARIABLE(resIncr);
   eigen_internal_assert(resIncr==1);
   #ifdef _EIGEN_ACCUMULATE_PACKETS
   #error _EIGEN_ACCUMULATE_PACKETS has already been defined
@@ -350,7 +345,7 @@ EIGEN_DONT_INLINE static void run(
   Index rows, Index cols,
   const LhsScalar* lhs, Index lhsStride,
   const RhsScalar* rhs, Index rhsIncr,
-  ResScalar* res, Index resIncr,
+        ResScalar* res, Index resIncr,
   ResScalar alpha);
 };
 
@@ -364,6 +359,7 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,RowMajor,Co
 {
   EIGEN_UNUSED_VARIABLE(rhsIncr);
   eigen_internal_assert(rhsIncr==1);
+  
   #ifdef _EIGEN_ACCUMULATE_PACKETS
   #error _EIGEN_ACCUMULATE_PACKETS has already been defined
   #endif
diff --git a/Eigen/src/Core/products/SelfadjointMatrixVector.h b/Eigen/src/Core/products/SelfadjointMatrixVector.h
index ddc07d535..20b5f1627 100644
--- a/Eigen/src/Core/products/SelfadjointMatrixVector.h
+++ b/Eigen/src/Core/products/SelfadjointMatrixVector.h
@@ -113,9 +113,9 @@ EIGEN_DONT_INLINE void selfadjoint_matrix_vector_product<Scalar,Index,StorageOrd
 
     for (size_t i=starti; i<alignedStart; ++i)
     {
-      res[i] += t0 * A0[i] + t1 * A1[i];
-      t2 += numext::conj(A0[i]) * rhs[i];
-      t3 += numext::conj(A1[i]) * rhs[i];
+      res[i] += cj0.pmul(A0[i], t0) + cj0.pmul(A1[i],t1);
+      t2 += cj1.pmul(A0[i], rhs[i]);
+      t3 += cj1.pmul(A1[i], rhs[i]);
     }
     // Yes this an optimization for gcc 4.3 and 4.4 (=> huge speed up)
     // gcc 4.2 does this optimization automatically.
diff --git a/Eigen/src/Core/util/Macros.h b/Eigen/src/Core/util/Macros.h
index 3a1dec129..6aa716698 100644
--- a/Eigen/src/Core/util/Macros.h
+++ b/Eigen/src/Core/util/Macros.h
@@ -252,7 +252,12 @@
 #endif
 
 // Suppresses 'unused variable' warnings.
-#define EIGEN_UNUSED_VARIABLE(var) (void)var;
+namespace Eigen {
+  namespace internal {
+    template<typename T> void ignore_unused_variable(const T&) {}
+  }
+}
+#define EIGEN_UNUSED_VARIABLE(var) Eigen::internal::ignore_unused_variable(var);
 
 #if !defined(EIGEN_ASM_COMMENT)
   #if (defined __GNUC__) && ( defined(__i386__) || defined(__x86_64__) )
diff --git a/Eigen/src/Core/util/Memory.h b/Eigen/src/Core/util/Memory.h
index bc472d720..286e963d2 100644
--- a/Eigen/src/Core/util/Memory.h
+++ b/Eigen/src/Core/util/Memory.h
@@ -274,12 +274,12 @@ inline void* aligned_realloc(void *ptr, size_t new_size, size_t old_size)
   // The defined(_mm_free) is just here to verify that this MSVC version
   // implements _mm_malloc/_mm_free based on the corresponding _aligned_
   // functions. This may not always be the case and we just try to be safe.
-  #if defined(_MSC_VER) && defined(_mm_free)
+  #if defined(_MSC_VER) && (!defined(_WIN32_WCE)) && defined(_mm_free)
     result = _aligned_realloc(ptr,new_size,16);
   #else
     result = generic_aligned_realloc(ptr,new_size,old_size);
   #endif
-#elif defined(_MSC_VER)
+#elif defined(_MSC_VER) && (!defined(_WIN32_WCE))
   result = _aligned_realloc(ptr,new_size,16);
 #else
   result = handmade_aligned_realloc(ptr,new_size,old_size);
@@ -464,7 +464,7 @@ template<typename T, bool Align> inline void conditional_aligned_delete_auto(T *
   * There is also the variant first_aligned(const MatrixBase&) defined in DenseCoeffsBase.h.
   */
 template<typename Scalar, typename Index>
-static inline Index first_aligned(const Scalar* array, Index size)
+inline Index first_aligned(const Scalar* array, Index size)
 {
   enum { PacketSize = packet_traits<Scalar>::size,
          PacketAlignedMask = PacketSize-1
@@ -492,7 +492,7 @@ static inline Index first_aligned(const Scalar* array, Index size)
 /** \internal Returns the smallest integer multiple of \a base and greater or equal to \a size
   */ 
 template<typename Index> 
-inline static Index first_multiple(Index size, Index base)
+inline Index first_multiple(Index size, Index base)
 {
   return ((size+base-1)/base)*base;
 }