Set of fixes and workaround to make sun studio more happy.

Still remains the problem of alignment and vectorization.

9 files changed, 42 insertions, 26 deletions

diff --git a/Eigen/src/Core/Block.h b/Eigen/src/Core/Block.h
index d7af971f9..382518696 100644
--- a/Eigen/src/Core/Block.h
+++ b/Eigen/src/Core/Block.h
@@ -271,13 +271,19 @@ class Block<MatrixType,BlockRows,BlockCols,PacketAccess,HasDirectAccess>
 
     inline int stride(void) const { return m_matrix.stride(); }
 
+  #ifndef __SUNPRO_CC
+  // FIXME sunstudio is not friendly with the above friend...
   protected:
+  #endif
 
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** \internal used by allowAligned() */
     inline Block(const MatrixType& matrix, const Scalar* data, int blockRows, int blockCols)
       : Base(data, blockRows, blockCols), m_matrix(matrix)
     {}
+    #endif
 
+  protected:
     const typename MatrixType::Nested m_matrix;
 };
 
diff --git a/Eigen/src/Core/CwiseUnaryView.h b/Eigen/src/Core/CwiseUnaryView.h
index c8fd98ea1..8d8b51b4c 100644
--- a/Eigen/src/Core/CwiseUnaryView.h
+++ b/Eigen/src/Core/CwiseUnaryView.h
@@ -42,13 +42,13 @@ struct ei_traits<CwiseUnaryView<ViewOp, MatrixType> >
  : ei_traits<MatrixType>
 {
   typedef typename ei_result_of<
-                     ViewOp(typename MatrixType::Scalar)
+                     ViewOp(typename ei_traits<MatrixType>::Scalar)
                    >::type Scalar;
   typedef typename MatrixType::Nested MatrixTypeNested;
   typedef typename ei_unref<MatrixTypeNested>::type _MatrixTypeNested;
   enum {
-    Flags = (_MatrixTypeNested::Flags & (HereditaryBits | LinearAccessBit | AlignedBit)),
-    CoeffReadCost = _MatrixTypeNested::CoeffReadCost + ei_functor_traits<ViewOp>::Cost
+    Flags = (ei_traits<_MatrixTypeNested>::Flags & (HereditaryBits | LinearAccessBit | AlignedBit)),
+    CoeffReadCost = ei_traits<_MatrixTypeNested>::CoeffReadCost + ei_functor_traits<ViewOp>::Cost
   };
 };
 
@@ -62,7 +62,7 @@ class CwiseUnaryView : ei_no_assignment_operator,
 
     inline CwiseUnaryView(const MatrixType& mat, const ViewOp& func = ViewOp())
       : m_matrix(mat), m_functor(func) {}
-    
+
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(CwiseUnaryView)
 
     EIGEN_STRONG_INLINE int rows() const { return m_matrix.rows(); }
@@ -77,7 +77,7 @@ class CwiseUnaryView : ei_no_assignment_operator,
     {
       return m_functor(m_matrix.coeff(index));
     }
-    
+
     EIGEN_STRONG_INLINE Scalar& coeffRef(int row, int col)
     {
       return m_functor(m_matrix.const_cast_derived().coeffRef(row, col));
@@ -89,7 +89,8 @@ class CwiseUnaryView : ei_no_assignment_operator,
     }
 
   protected:
-    const typename MatrixType::Nested m_matrix;
+    // FIXME changed from MatrixType::Nested because of a weird compilation error with sun CC
+    const typename ei_nested<MatrixType>::type m_matrix;
     const ViewOp m_functor;
 };
 
diff --git a/Eigen/src/Core/MapBase.h b/Eigen/src/Core/MapBase.h
index 126974918..59bf69ad6 100644
--- a/Eigen/src/Core/MapBase.h
+++ b/Eigen/src/Core/MapBase.h
@@ -178,6 +178,7 @@ template<typename Derived> class MapBase
     }
 
     using Base::operator*=;
+    using Base::operator+=;
 
     template<typename OtherDerived>
     Derived& operator+=(const MatrixBase<OtherDerived>& other)
diff --git a/Eigen/src/Core/Matrix.h b/Eigen/src/Core/Matrix.h
index e099ba1e7..5301f4849 100644
--- a/Eigen/src/Core/Matrix.h
+++ b/Eigen/src/Core/Matrix.h
@@ -124,6 +124,7 @@ class Matrix
 {
   public:
     EIGEN_GENERIC_PUBLIC_INTERFACE(Matrix)
+
     enum { Options = _Options };
     friend class Eigen::Map<Matrix, Unaligned>;
     typedef class Eigen::Map<Matrix, Unaligned> UnalignedMapType;
@@ -217,7 +218,7 @@ class Matrix
       *
       * This method is intended for dynamic-size matrices, although it is legal to call it on any
       * matrix as long as fixed dimensions are left unchanged. If you only want to change the number
-      * of rows and/or of columns, you can use resize(NoChange_t, int), resize(int, NoChange_t). 
+      * of rows and/or of columns, you can use resize(NoChange_t, int), resize(int, NoChange_t).
       *
       * If the current number of coefficients of \c *this exactly matches the
       * product \a rows * \a cols, then no memory allocation is performed and
diff --git a/Eigen/src/Core/MatrixBase.h b/Eigen/src/Core/MatrixBase.h
index 941539214..2a49ae620 100644
--- a/Eigen/src/Core/MatrixBase.h
+++ b/Eigen/src/Core/MatrixBase.h
@@ -137,10 +137,14 @@ template<typename Derived> class MatrixBase
           * constructed from this one. See the \ref flags "list of flags".
           */
 
-      CoeffReadCost = ei_traits<Derived>::CoeffReadCost
+      CoeffReadCost = ei_traits<Derived>::CoeffReadCost,
         /**< This is a rough measure of how expensive it is to read one coefficient from
           * this expression.
           */
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+      _HasDirectAccess = (int(Flags)&DirectAccessBit) ? 1 : 0 // workaround sunCC
+#endif
     };
 
     /** Default constructor. Just checks at compile-time for self-consistency of the flags. */
@@ -204,7 +208,7 @@ template<typename Derived> class MatrixBase
 
     /** \internal the return type of coeff()
       */
-    typedef typename ei_meta_if<bool(int(Flags)&DirectAccessBit), const Scalar&, Scalar>::ret CoeffReturnType;
+    typedef typename ei_meta_if<_HasDirectAccess, const Scalar&, Scalar>::ret CoeffReturnType;
 
     /** \internal Represents a matrix with all coefficients equal to one another*/
     typedef CwiseNullaryOp<ei_scalar_constant_op<Scalar>,Derived> ConstantReturnType;
diff --git a/Eigen/src/Core/Product.h b/Eigen/src/Core/Product.h
index 353aec48e..2589d548e 100644
--- a/Eigen/src/Core/Product.h
+++ b/Eigen/src/Core/Product.h
@@ -83,15 +83,14 @@ struct ProductReturnType<Lhs,Rhs,CacheFriendlyProduct>
 template<typename Lhs, typename Rhs> struct ei_product_mode
 {
   enum{
-
     value = ei_is_diagonal<Rhs>::ret || ei_is_diagonal<Lhs>::ret
           ? DiagonalProduct
-          : Lhs::MaxColsAtCompileTime == Dynamic
-            && ( Lhs::MaxRowsAtCompileTime == Dynamic
-              || Rhs::MaxColsAtCompileTime == Dynamic )
-            && (!(Rhs::IsVectorAtCompileTime && (Lhs::Flags&RowMajorBit)  && (!(Lhs::Flags&DirectAccessBit))))
-            && (!(Lhs::IsVectorAtCompileTime && (!(Rhs::Flags&RowMajorBit)) && (!(Rhs::Flags&DirectAccessBit))))
-            && (ei_is_same_type<typename Lhs::Scalar, typename Rhs::Scalar>::ret)
+          : ei_traits<Lhs>::MaxColsAtCompileTime == Dynamic
+            && ( ei_traits<Lhs>::MaxRowsAtCompileTime == Dynamic
+              || ei_traits<Rhs>::MaxColsAtCompileTime == Dynamic )
+            && (!(ei_traits<Rhs>::IsVectorAtCompileTime && (ei_traits<Lhs>::Flags&RowMajorBit)  && (!(ei_traits<Lhs>::Flags&DirectAccessBit))))
+            && (!(ei_traits<Lhs>::IsVectorAtCompileTime && (!(ei_traits<Rhs>::Flags&RowMajorBit)) && (!(ei_traits<Rhs>::Flags&DirectAccessBit))))
+            && (ei_is_same_type<typename ei_traits<Lhs>::Scalar, typename ei_traits<Rhs>::Scalar>::ret)
           ? CacheFriendlyProduct
           : NormalProduct };
 };
@@ -215,7 +214,7 @@ template<typename LhsNested, typename RhsNested, int ProductMode> class Product
       */
     EIGEN_STRONG_INLINE bool _useCacheFriendlyProduct() const
     {
-      return  m_lhs.cols()>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
+      return m_lhs.cols()>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
               && (  rows()>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
                  || cols()>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD);
     }
diff --git a/Eigen/src/Core/util/Constants.h b/Eigen/src/Core/util/Constants.h
index c2641537a..572cf7caf 100644
--- a/Eigen/src/Core/util/Constants.h
+++ b/Eigen/src/Core/util/Constants.h
@@ -188,7 +188,7 @@ const unsigned int HereditaryBits = RowMajorBit
 
 // diagonal means both upper and lower triangular
 const unsigned DiagonalBits = UpperTriangularBit | LowerTriangularBit;
-    
+
 // Possible values for the Mode parameter of part()
 const unsigned int UpperTriangular = UpperTriangularBit;
 const unsigned int StrictlyUpperTriangular = UpperTriangularBit | ZeroDiagBit;
@@ -201,7 +201,7 @@ const unsigned int UnitLowerTriangular = LowerTriangularBit | UnitDiagBit;
 template<typename T> struct ei_is_diagonal
 {
   enum {
-    ret = ( (unsigned int)(T::Flags) & DiagonalBits ) == DiagonalBits
+    ret = ( int(ei_traits<T>::Flags) & DiagonalBits ) == DiagonalBits
   };
 };
 
diff --git a/Eigen/src/Core/util/ForwardDeclarations.h b/Eigen/src/Core/util/ForwardDeclarations.h
index a2105604a..ad7256635 100644
--- a/Eigen/src/Core/util/ForwardDeclarations.h
+++ b/Eigen/src/Core/util/ForwardDeclarations.h
@@ -97,7 +97,7 @@ template<typename Scalar1,typename Scalar2> struct ei_scalar_multiple2_op;
 struct IOFormat;
 
 template<typename Scalar>
-void ei_cache_friendly_product(
+static void ei_cache_friendly_product(
   int _rows, int _cols, int depth,
   bool _lhsRowMajor, const Scalar* _lhs, int _lhsStride,
   bool _rhsRowMajor, const Scalar* _rhs, int _rhsStride,
diff --git a/Eigen/src/Core/util/Macros.h b/Eigen/src/Core/util/Macros.h
index 8f62f9b69..2b2444ef0 100644
--- a/Eigen/src/Core/util/Macros.h
+++ b/Eigen/src/Core/util/Macros.h
@@ -51,7 +51,8 @@
 #define EIGEN_GCC3_OR_OLDER 0
 #endif
 
-#if !EIGEN_GCC_AND_ARCH_DOESNT_WANT_ALIGNMENT && !EIGEN_GCC3_OR_OLDER
+// FIXME vectorization + alignment is completely disabled with sun studio
+#if !EIGEN_GCC_AND_ARCH_DOESNT_WANT_ALIGNMENT && !EIGEN_GCC3_OR_OLDER && !defined(__SUNPRO_CC)
   #define EIGEN_ARCH_WANTS_ALIGNMENT 1
 #else
   #define EIGEN_ARCH_WANTS_ALIGNMENT 0
@@ -97,7 +98,7 @@
 /** Allows to disable some optimizations which might affect the accuracy of the result.
   * Such optimization are enabled by default, and set EIGEN_FAST_MATH to 0 to disable them.
   * They currently include:
-  *   - single precision Cwise::sin() and Cwise::cos() when SSE vectorization is enabled. 
+  *   - single precision Cwise::sin() and Cwise::cos() when SSE vectorization is enabled.
   */
 #ifndef EIGEN_FAST_MATH
 #define EIGEN_FAST_MATH 1
@@ -199,13 +200,16 @@ using Eigen::ei_cos;
  * vectorized and non-vectorized code.
  */
 #if !EIGEN_ALIGN
-#define EIGEN_ALIGN_128
+  #define EIGEN_ALIGN_128
 #elif (defined __GNUC__)
-#define EIGEN_ALIGN_128 __attribute__((aligned(16)))
+  #define EIGEN_ALIGN_128 __attribute__((aligned(16)))
 #elif (defined _MSC_VER)
-#define EIGEN_ALIGN_128 __declspec(align(16))
+  #define EIGEN_ALIGN_128 __declspec(align(16))
+#elif (defined __SUNPRO_CC)
+  // FIXME not sure about this one:
+  #define EIGEN_ALIGN_128 __attribute__((aligned(16)))
 #else
-#error Please tell me what is the equivalent of __attribute__((aligned(16))) for your compiler
+  #error Please tell me what is the equivalent of __attribute__((aligned(16))) for your compiler
 #endif
 
 #define EIGEN_RESTRICT __restrict