* fix aliasing checks when the lhs is also transposed. At the same time,

  significantly simplify the code of these checks while extending them
  to catch much more expressions!
* move the enabling/disabling of vectorized sin/cos to the architecture traits

7 files changed, 46 insertions, 78 deletions

diff --git a/Eigen/src/Array/Functors.h b/Eigen/src/Array/Functors.h
index fd259f7bc..120c56cc1 100644
--- a/Eigen/src/Array/Functors.h
+++ b/Eigen/src/Array/Functors.h
@@ -87,7 +87,7 @@ struct ei_functor_traits<ei_scalar_cos_op<Scalar> >
 {
   enum {
     Cost = 5 * NumTraits<Scalar>::MulCost,
-    PacketAccess = ei_packet_traits<Scalar>::HasCos && EIGEN_FAST_MATH
+    PacketAccess = ei_packet_traits<Scalar>::HasCos
   };
 };
 
@@ -109,7 +109,7 @@ struct ei_functor_traits<ei_scalar_sin_op<Scalar> >
 {
   enum {
     Cost = 5 * NumTraits<Scalar>::MulCost,
-    PacketAccess = ei_packet_traits<Scalar>::HasSin && EIGEN_FAST_MATH
+    PacketAccess = ei_packet_traits<Scalar>::HasSin
   };
 };
 
diff --git a/Eigen/src/Core/Assign.h b/Eigen/src/Core/Assign.h
index 00febdc5d..d544a6198 100644
--- a/Eigen/src/Core/Assign.h
+++ b/Eigen/src/Core/Assign.h
@@ -485,6 +485,9 @@ EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>
 #ifdef EIGEN_DEBUG_ASSIGN
   ei_assign_traits<Derived, OtherDerived>::debug();
 #endif
+#ifndef EIGEN_NO_DEBUG
+  checkTransposeAliasing(other.derived());
+#endif
   return derived();
 }
 
diff --git a/Eigen/src/Core/MatrixBase.h b/Eigen/src/Core/MatrixBase.h
index 03d8d5f55..593108522 100644
--- a/Eigen/src/Core/MatrixBase.h
+++ b/Eigen/src/Core/MatrixBase.h
@@ -436,21 +436,12 @@ template<typename Derived> class MatrixBase
     void transposeInPlace();
     const AdjointReturnType adjoint() const;
     void adjointInPlace();
-    #ifndef EIGEN_NO_DEBUG
-    template<typename OtherDerived>
-    Derived& lazyAssign(const Transpose<OtherDerived>& other);
-    template<typename DerivedA, typename DerivedB>
-    Derived& lazyAssign(const CwiseBinaryOp<ei_scalar_sum_op<Scalar>,Transpose<DerivedA>,DerivedB>& other);
-    template<typename DerivedA, typename DerivedB>
-    Derived& lazyAssign(const CwiseBinaryOp<ei_scalar_sum_op<Scalar>,DerivedA,Transpose<DerivedB> >& other);
-
+#ifndef EIGEN_NO_DEBUG
+  protected:
     template<typename OtherDerived>
-    Derived& lazyAssign(const CwiseUnaryOp<ei_scalar_conjugate_op<Scalar>, Eigen::Transpose<OtherDerived> >& other);
-    template<typename DerivedA, typename DerivedB>
-    Derived& lazyAssign(const CwiseBinaryOp<ei_scalar_sum_op<Scalar>,CwiseUnaryOp<ei_scalar_conjugate_op<Scalar>, Eigen::Transpose<DerivedA> >,DerivedB>& other);
-    template<typename DerivedA, typename DerivedB>
-    Derived& lazyAssign(const CwiseBinaryOp<ei_scalar_sum_op<Scalar>,DerivedA,CwiseUnaryOp<ei_scalar_conjugate_op<Scalar>, Eigen::Transpose<DerivedB> > >& other);
-    #endif
+    void checkTransposeAliasing(const OtherDerived& other) const;
+  public:
+#endif
 
     RowXpr row(int i);
     const RowXpr row(int i) const;
@@ -638,7 +629,7 @@ template<typename Derived> class MatrixBase
     const CwiseBinaryOp<CustomBinaryOp, Derived, OtherDerived>
     binaryExpr(const MatrixBase<OtherDerived> &other, const CustomBinaryOp& func = CustomBinaryOp()) const;
 
-    
+
     Scalar sum() const;
     Scalar mean() const;
     Scalar trace() const;
@@ -818,7 +809,7 @@ template<typename Derived> class MatrixBase
                           INVALID_MATRIXBASE_TEMPLATE_PARAMETERS)
 #endif
     }
-    
+
   private:
     explicit MatrixBase(int);
     MatrixBase(int,int);
diff --git a/Eigen/src/Core/Transpose.h b/Eigen/src/Core/Transpose.h
index 79aef9fa7..796ec8573 100644
--- a/Eigen/src/Core/Transpose.h
+++ b/Eigen/src/Core/Transpose.h
@@ -268,13 +268,7 @@ inline void MatrixBase<Derived>::adjointInPlace()
 
 #ifndef EIGEN_NO_DEBUG
 
-// The following is to detect aliasing problems in the following common cases:
-// a = a.transpose()
-// a = a.transpose() + X
-// a = X + a.transpose()
-// a = a.adjoint()
-// a = a.adjoint() + X
-// a = X + a.adjoint()
+// The following is to detect aliasing problems in most common cases.
 
 template<typename T, int Access=ei_blas_traits<T>::ActualAccess>
 struct ei_extract_data_selector {
@@ -294,63 +288,31 @@ template<typename T> typename T::Scalar* ei_extract_data(const T& m)
   return ei_extract_data_selector<T>::run(m);
 }
 
-template<typename Derived>
-template<typename OtherDerived>
-Derived& MatrixBase<Derived>::lazyAssign(const Transpose<OtherDerived>& other)
-{
-  ei_assert(ei_extract_data(other) != ei_extract_data(derived())
-            && "aliasing detected during tranposition, please use transposeInPlace()");
-  return lazyAssign(static_cast<const MatrixBase<Transpose<OtherDerived> >& >(other));
-}
-
-template<typename Derived>
-template<typename DerivedA, typename DerivedB>
-Derived& MatrixBase<Derived>::
-lazyAssign(const CwiseBinaryOp<ei_scalar_sum_op<Scalar>,Transpose<DerivedA>,DerivedB>& other)
-{
-  ei_assert(ei_extract_data(derived()) != ei_extract_data(other.lhs())
-            && "aliasing detected during tranposition, please evaluate your expression");
-  return lazyAssign(static_cast<const MatrixBase<CwiseBinaryOp<ei_scalar_sum_op<Scalar>,Transpose<DerivedA>,DerivedB> >& >(other));
-}
-
-template<typename Derived>
-template<typename DerivedA, typename DerivedB>
-Derived& MatrixBase<Derived>::
-lazyAssign(const CwiseBinaryOp<ei_scalar_sum_op<Scalar>,DerivedA,Transpose<DerivedB> >& other)
-{
-  ei_assert(ei_extract_data(derived()) != ei_extract_data(other.rhs())
-            && "aliasing detected during tranposition, please evaluate your expression");
-  return lazyAssign(static_cast<const MatrixBase<CwiseBinaryOp<ei_scalar_sum_op<Scalar>,DerivedA,Transpose<DerivedB> > >& >(other));
-}
-
-template<typename Derived>
-template<typename OtherDerived> Derived&
-MatrixBase<Derived>::
-lazyAssign(const CwiseUnaryOp<ei_scalar_conjugate_op<Scalar>, Eigen::Transpose<OtherDerived> >& other)
+template<typename Scalar, bool DestIsTranposed, typename OtherDerived>
+struct ei_check_transpose_aliasing_selector
 {
-  ei_assert(ei_extract_data(other) != ei_extract_data(derived())
-            && "aliasing detected during tranposition, please use adjointInPlace()");
-  return lazyAssign(static_cast<const MatrixBase<CwiseUnaryOp<ei_scalar_conjugate_op<Scalar>, Eigen::Transpose<OtherDerived> > >& >(other));
-}
+  static bool run(const Scalar* dest, const OtherDerived& src)
+  {
+    return (ei_blas_traits<OtherDerived>::IsTransposed != DestIsTranposed) && (dest==ei_extract_data(src));
+  }
+};
 
-template<typename Derived>
-template<typename DerivedA, typename DerivedB>
-Derived& MatrixBase<Derived>::
-lazyAssign(const CwiseBinaryOp<ei_scalar_sum_op<Scalar>,CwiseUnaryOp<ei_scalar_conjugate_op<Scalar>, Eigen::Transpose<DerivedA> >,DerivedB>& other)
+template<typename Scalar, bool DestIsTranposed, typename BinOp, typename DerivedA, typename DerivedB>
+struct ei_check_transpose_aliasing_selector<Scalar,DestIsTranposed,CwiseBinaryOp<BinOp,DerivedA,DerivedB> >
 {
-  ei_assert(ei_extract_data(derived()) != ei_extract_data(other.lhs())
-            && "aliasing detected during tranposition, please evaluate your expression");
-  return lazyAssign(static_cast<const MatrixBase<CwiseBinaryOp<ei_scalar_sum_op<Scalar>,CwiseUnaryOp<ei_scalar_conjugate_op<Scalar>, Eigen::Transpose<DerivedA> >,DerivedB> >& >(other));
-}
+  static bool run(const Scalar* dest, const CwiseBinaryOp<BinOp,DerivedA,DerivedB>& src)
+  {
+    return ((ei_blas_traits<DerivedA>::IsTransposed != DestIsTranposed) && dest==ei_extract_data(src.lhs()))
+        || ((ei_blas_traits<DerivedB>::IsTransposed != DestIsTranposed) && dest==ei_extract_data(src.rhs()));
+  }
+};
 
 template<typename Derived>
-template<typename DerivedA, typename DerivedB>
-Derived& MatrixBase<Derived>::
-lazyAssign(const CwiseBinaryOp<ei_scalar_sum_op<Scalar>,DerivedA,CwiseUnaryOp<ei_scalar_conjugate_op<Scalar>, Eigen::Transpose<DerivedB> > >& other)
+template<typename OtherDerived>
+void MatrixBase<Derived>::checkTransposeAliasing(const OtherDerived& other) const
 {
-  ei_assert(ei_extract_data(derived()) != ei_extract_data(other.rhs())
-            && "aliasing detected during tranposition, please evaluate your expression");
-  return lazyAssign(static_cast<const MatrixBase<CwiseBinaryOp<ei_scalar_sum_op<Scalar>,DerivedA,CwiseUnaryOp<ei_scalar_conjugate_op<Scalar>, Eigen::Transpose<DerivedB> > > >& >(other));
+  ei_assert((!ei_check_transpose_aliasing_selector<Scalar,ei_blas_traits<Derived>::IsTransposed,OtherDerived>::run(ei_extract_data(derived()), other))
+         && "aliasing detected during tranposition, use transposeInPlace() or evaluate the rhs into a temporary using .eval()");
 }
 #endif
 
diff --git a/Eigen/src/Core/arch/SSE/PacketMath.h b/Eigen/src/Core/arch/SSE/PacketMath.h
index 69f6979bd..dbdddb38d 100644
--- a/Eigen/src/Core/arch/SSE/PacketMath.h
+++ b/Eigen/src/Core/arch/SSE/PacketMath.h
@@ -58,8 +58,8 @@ template<> struct ei_packet_traits<float>  : ei_default_packet_traits
 {
   typedef Packet4f type; enum {size=4};
   enum {
-    HasSin  = 1,
-    HasCos  = 1,
+    HasSin  = EIGEN_FAST_MATH,
+    HasCos  = EIGEN_FAST_MATH,
     HasLog  = 1,
     HasExp  = 1,
     HasSqrt = 1
diff --git a/Eigen/src/Core/util/BlasUtil.h b/Eigen/src/Core/util/BlasUtil.h
index a012a6e12..fa21ceebb 100644
--- a/Eigen/src/Core/util/BlasUtil.h
+++ b/Eigen/src/Core/util/BlasUtil.h
@@ -160,6 +160,7 @@ template<typename XprType> struct ei_blas_traits
   typedef XprType _ExtractType;
   enum {
     IsComplex = NumTraits<Scalar>::IsComplex,
+    IsTransposed = false,
     NeedToConjugate = false,
     ActualAccess = int(ei_traits<XprType>::Flags)&DirectAccessBit ? HasDirectAccess : NoDirectAccess
   };
@@ -227,6 +228,9 @@ struct ei_blas_traits<Transpose<NestedXpr> >
     ExtractType,
     typename ExtractType::PlainMatrixType
     >::ret DirectLinearAccessType;
+  enum {
+    IsTransposed = Base::IsTransposed ? 0 : 1
+  };
   static inline const ExtractType extract(const XprType& x) { return Base::extract(x._expression()); }
   static inline Scalar extractScalarFactor(const XprType& x) { return Base::extractScalarFactor(x._expression()); }
 };
diff --git a/test/adjoint.cpp b/test/adjoint.cpp
index 344399257..b34112249 100644
--- a/test/adjoint.cpp
+++ b/test/adjoint.cpp
@@ -128,6 +128,14 @@ void test_adjoint()
     VERIFY_RAISES_ASSERT(a = a.adjoint());
     VERIFY_RAISES_ASSERT(a = a.adjoint() + b);
     VERIFY_RAISES_ASSERT(a = b + a.adjoint());
+
+    // no assertion should be triggered for these cases:
+    a.transpose() = a.transpose();
+    a.transpose() += a.transpose();
+    a.transpose() += a.transpose() + b;
+    a.transpose() = a.adjoint();
+    a.transpose() += a.adjoint();
+    a.transpose() += a.adjoint() + b;
   }
 #endif
 }