From 3ba8a3ab1a77fd2f58448187c6881d13bf51f430 Mon Sep 17 00:00:00 2001
From: Gael Guennebaud <g.gael@free.fr>
Date: Sat, 30 Jan 2016 22:14:04 +0100
Subject: [PATCH 1/3] Disable underflow unit test on the i387 FPU.

---
 test/adjoint.cpp | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/test/adjoint.cpp b/test/adjoint.cpp
index b1e69c2e5..9c895e0ac 100644
--- a/test/adjoint.cpp
+++ b/test/adjoint.cpp
@@ -45,12 +45,14 @@ template<> struct adjoint_specific<false> {
 
     // check null inputs
     VERIFY_IS_APPROX((v1*0).normalized(), (v1*0));
+#if (!EIGEN_ARCH_i386) || defined(EIGEN_VECTORIZE)
     RealScalar very_small = (std::numeric_limits<RealScalar>::min)();
     VERIFY( (v1*very_small).norm() == 0 );
     VERIFY_IS_APPROX((v1*very_small).normalized(), (v1*very_small));
     v3 = v1*very_small;
     v3.normalize();
     VERIFY_IS_APPROX(v3, (v1*very_small));
+#endif
     
     // check compatibility of dot and adjoint
     ref = NumTraits<Scalar>::IsInteger ? 0 : (std::max)((std::max)(v1.norm(),v2.norm()),(std::max)((square * v2).norm(),(square.adjoint() * v1).norm()));

From a4e4542b89092eb2ed2984aae6f15bbcc43d7ed6 Mon Sep 17 00:00:00 2001
From: Gael Guennebaud <g.gael@free.fr>
Date: Sat, 30 Jan 2016 22:26:17 +0100
Subject: [PATCH 2/3] Avoid overflow in unit test.

---
 test/stable_norm.cpp | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/test/stable_norm.cpp b/test/stable_norm.cpp
index 9f12320e0..c3eb5ff31 100644
--- a/test/stable_norm.cpp
+++ b/test/stable_norm.cpp
@@ -174,7 +174,8 @@ template<typename MatrixType> void stable_norm(const MatrixType& m)
     VERIFY_IS_APPROX(vcopy.norm(), RealScalar(1));
     VERIFY_IS_APPROX((vbig.stableNormalized()).norm(), RealScalar(1));
     VERIFY_IS_APPROX((vsmall.stableNormalized()).norm(), RealScalar(1));
-    VERIFY_IS_APPROX(vbig, vbig.stableNorm() * vbig.stableNormalized());
+    RealScalar big_scaling = ((std::numeric_limits<RealScalar>::max)() * RealScalar(1e-4));
+    VERIFY_IS_APPROX(vbig/big_scaling, (vbig.stableNorm() * vbig.stableNormalized()).eval()/big_scaling);
     VERIFY_IS_APPROX(vsmall, vsmall.stableNorm() * vsmall.stableNormalized());
   }
 }

From d1421659427a5bb237bbba1c86781267b98ce235 Mon Sep 17 00:00:00 2001
From: Gael Guennebaud <g.gael@free.fr>
Date: Sun, 31 Jan 2016 16:34:10 +0100
Subject: [PATCH 3/3] bug #667: declare several critical functions as
 FORECE_INLINE to make ICC happier.

<g.gael@free.fr> HG: branch 'default' HG: changed Eigen/src/Core/ArrayBase.h HG: changed Eigen/src/Core/AssignEvaluator.h HG: changed
Eigen/src/Core/CoreEvaluators.h HG: changed Eigen/src/Core/CwiseUnaryOp.h HG: changed Eigen/src/Core/DenseBase.h HG: changed Eigen/src/Core/MatrixBase.h
---
 Eigen/src/Core/ArrayBase.h       |  16 +--
 Eigen/src/Core/AssignEvaluator.h |  34 ++++--
 Eigen/src/Core/CoreEvaluators.h  | 197 +++++++++++++++++++++----------
 Eigen/src/Core/CwiseUnaryOp.h    |  16 +--
 Eigen/src/Core/DenseBase.h       |   8 +-
 Eigen/src/Core/MatrixBase.h      |   8 +-
 6 files changed, 183 insertions(+), 96 deletions(-)

diff --git a/Eigen/src/Core/ArrayBase.h b/Eigen/src/Core/ArrayBase.h
index b4c24a27a..0443e3032 100644
--- a/Eigen/src/Core/ArrayBase.h
+++ b/Eigen/src/Core/ArrayBase.h
@@ -103,7 +103,7 @@ template<typename Derived> class ArrayBase
     /** Special case of the template operator=, in order to prevent the compiler
       * from generating a default operator= (issue hit with g++ 4.1)
       */
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const ArrayBase& other)
     {
       internal::call_assignment(derived(), other.derived());
@@ -112,28 +112,28 @@ template<typename Derived> class ArrayBase
     
     /** Set all the entries to \a value.
       * \sa DenseBase::setConstant(), DenseBase::fill() */
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const Scalar &value)
     { Base::setConstant(value); return derived(); }
 
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator+=(const Scalar& scalar);
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator-=(const Scalar& scalar);
 
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator+=(const ArrayBase<OtherDerived>& other);
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator-=(const ArrayBase<OtherDerived>& other);
 
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator*=(const ArrayBase<OtherDerived>& other);
 
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator/=(const ArrayBase<OtherDerived>& other);
 
   public:
diff --git a/Eigen/src/Core/AssignEvaluator.h b/Eigen/src/Core/AssignEvaluator.h
index f6632de69..5b65bfb0c 100755
--- a/Eigen/src/Core/AssignEvaluator.h
+++ b/Eigen/src/Core/AssignEvaluator.h
@@ -637,7 +637,7 @@ protected:
 ***************************************************************************/
 
 template<typename DstXprType, typename SrcXprType, typename Functor>
-EIGEN_DEVICE_FUNC void call_dense_assignment_loop(const DstXprType& dst, const SrcXprType& src, const Functor &func)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(const DstXprType& dst, const SrcXprType& src, const Functor &func)
 {
   eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
   
@@ -654,7 +654,7 @@ EIGEN_DEVICE_FUNC void call_dense_assignment_loop(const DstXprType& dst, const S
 }
 
 template<typename DstXprType, typename SrcXprType>
-EIGEN_DEVICE_FUNC void call_dense_assignment_loop(const DstXprType& dst, const SrcXprType& src)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(const DstXprType& dst, const SrcXprType& src)
 {
   call_dense_assignment_loop(dst, src, internal::assign_op<typename DstXprType::Scalar>());
 }
@@ -688,26 +688,30 @@ struct Assignment;
 // does not has to bother about these annoying details.
 
 template<typename Dst, typename Src>
-EIGEN_DEVICE_FUNC void call_assignment(Dst& dst, const Src& src)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(Dst& dst, const Src& src)
 {
   call_assignment(dst, src, internal::assign_op<typename Dst::Scalar>());
 }
 template<typename Dst, typename Src>
-EIGEN_DEVICE_FUNC void call_assignment(const Dst& dst, const Src& src)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(const Dst& dst, const Src& src)
 {
   call_assignment(dst, src, internal::assign_op<typename Dst::Scalar>());
 }
                      
 // Deal with "assume-aliasing"
 template<typename Dst, typename Src, typename Func>
-EIGEN_DEVICE_FUNC void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if< evaluator_assume_aliasing<Src>::value, void*>::type = 0)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if< evaluator_assume_aliasing<Src>::value, void*>::type = 0)
 {
   typename plain_matrix_type<Src>::type tmp(src);
   call_assignment_no_alias(dst, tmp, func);
 }
 
 template<typename Dst, typename Src, typename Func>
-EIGEN_DEVICE_FUNC void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if<!evaluator_assume_aliasing<Src>::value, void*>::type = 0)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if<!evaluator_assume_aliasing<Src>::value, void*>::type = 0)
 {
   call_assignment_no_alias(dst, src, func);
 }
@@ -715,14 +719,16 @@ EIGEN_DEVICE_FUNC void call_assignment(Dst& dst, const Src& src, const Func& fun
 // by-pass "assume-aliasing"
 // When there is no aliasing, we require that 'dst' has been properly resized
 template<typename Dst, template <typename> class StorageBase, typename Src, typename Func>
-EIGEN_DEVICE_FUNC void call_assignment(NoAlias<Dst,StorageBase>& dst, const Src& src, const Func& func)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(NoAlias<Dst,StorageBase>& dst, const Src& src, const Func& func)
 {
   call_assignment_no_alias(dst.expression(), src, func);
 }
 
 
 template<typename Dst, typename Src, typename Func>
-EIGEN_DEVICE_FUNC void call_assignment_no_alias(Dst& dst, const Src& src, const Func& func)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment_no_alias(Dst& dst, const Src& src, const Func& func)
 {
   enum {
     NeedToTranspose = (    (int(Dst::RowsAtCompileTime) == 1 && int(Src::ColsAtCompileTime) == 1)
@@ -747,13 +753,15 @@ EIGEN_DEVICE_FUNC void call_assignment_no_alias(Dst& dst, const Src& src, const
   Assignment<ActualDstTypeCleaned,Src,Func>::run(actualDst, src, func);
 }
 template<typename Dst, typename Src>
-EIGEN_DEVICE_FUNC void call_assignment_no_alias(Dst& dst, const Src& src)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment_no_alias(Dst& dst, const Src& src)
 {
   call_assignment_no_alias(dst, src, internal::assign_op<typename Dst::Scalar>());
 }
 
 template<typename Dst, typename Src, typename Func>
-EIGEN_DEVICE_FUNC void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src, const Func& func)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src, const Func& func)
 {
   Index dstRows = src.rows();
   Index dstCols = src.cols();
@@ -767,7 +775,8 @@ EIGEN_DEVICE_FUNC void call_assignment_no_alias_no_transpose(Dst& dst, const Src
   Assignment<Dst,Src,Func>::run(dst, src, func);
 }
 template<typename Dst, typename Src>
-EIGEN_DEVICE_FUNC void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src)
 {
   call_assignment_no_alias_no_transpose(dst, src, internal::assign_op<typename Dst::Scalar>());
 }
@@ -779,7 +788,8 @@ template<typename Dst, typename Src> void check_for_aliasing(const Dst &dst, con
 template< typename DstXprType, typename SrcXprType, typename Functor, typename Scalar>
 struct Assignment<DstXprType, SrcXprType, Functor, Dense2Dense, Scalar>
 {
-  EIGEN_DEVICE_FUNC static void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  static void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
   {
     eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
     
diff --git a/Eigen/src/Core/CoreEvaluators.h b/Eigen/src/Core/CoreEvaluators.h
index 7776948d1..a729e0454 100644
--- a/Eigen/src/Core/CoreEvaluators.h
+++ b/Eigen/src/Core/CoreEvaluators.h
@@ -148,7 +148,8 @@ struct evaluator<PlainObjectBase<Derived> >
     EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
   }
 
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index row, Index col) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
   {
     if (IsRowMajor)
       return m_data[row * m_outerStride.value() + col];
@@ -156,12 +157,14 @@ struct evaluator<PlainObjectBase<Derived> >
       return m_data[row + col * m_outerStride.value()];
   }
 
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index index) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
   {
     return m_data[index];
   }
 
-  EIGEN_DEVICE_FUNC Scalar& coeffRef(Index row, Index col)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
   {
     if (IsRowMajor)
       return const_cast<Scalar*>(m_data)[row * m_outerStride.value() + col];
@@ -169,12 +172,14 @@ struct evaluator<PlainObjectBase<Derived> >
       return const_cast<Scalar*>(m_data)[row + col * m_outerStride.value()];
   }
 
-  EIGEN_DEVICE_FUNC Scalar& coeffRef(Index index)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
   {
     return const_cast<Scalar*>(m_data)[index];
   }
 
   template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index row, Index col) const
   {
     if (IsRowMajor)
@@ -184,12 +189,14 @@ struct evaluator<PlainObjectBase<Derived> >
   }
 
   template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index index) const
   {
     return ploadt<PacketType, LoadMode>(m_data + index);
   }
 
   template<int StoreMode,typename PacketType>
+  EIGEN_STRONG_INLINE
   void writePacket(Index row, Index col, const PacketType& x)
   {
     if (IsRowMajor)
@@ -201,6 +208,7 @@ struct evaluator<PlainObjectBase<Derived> >
   }
 
   template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   void writePacket(Index index, const PacketType& x)
   {
     return pstoret<Scalar, PacketType, StoreMode>(const_cast<Scalar*>(m_data) + index, x);
@@ -260,45 +268,53 @@ struct unary_evaluator<Transpose<ArgType>, IndexBased>
   typedef typename XprType::Scalar Scalar;
   typedef typename XprType::CoeffReturnType CoeffReturnType;
 
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index row, Index col) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
   {
     return m_argImpl.coeff(col, row);
   }
 
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index index) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
   {
     return m_argImpl.coeff(index);
   }
 
-  EIGEN_DEVICE_FUNC Scalar& coeffRef(Index row, Index col)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
   {
     return m_argImpl.coeffRef(col, row);
   }
 
-  EIGEN_DEVICE_FUNC typename XprType::Scalar& coeffRef(Index index)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  typename XprType::Scalar& coeffRef(Index index)
   {
     return m_argImpl.coeffRef(index);
   }
 
   template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index row, Index col) const
   {
     return m_argImpl.template packet<LoadMode,PacketType>(col, row);
   }
 
   template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index index) const
   {
     return m_argImpl.template packet<LoadMode,PacketType>(index);
   }
 
-  template<int StoreMode, typename PacketType> 
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   void writePacket(Index row, Index col, const PacketType& x)
   {
     m_argImpl.template writePacket<StoreMode,PacketType>(col, row, x);
   }
 
-  template<int StoreMode, typename PacketType> 
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   void writePacket(Index index, const PacketType& x)
   {
     m_argImpl.template writePacket<StoreMode,PacketType>(index, x);
@@ -338,23 +354,27 @@ struct evaluator<CwiseNullaryOp<NullaryOp,PlainObjectType> >
 
   typedef typename XprType::CoeffReturnType CoeffReturnType;
 
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index row, Index col) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
   {
     return m_functor(row, col);
   }
 
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index index) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
   {
     return m_functor(index);
   }
 
   template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index row, Index col) const
   {
     return m_functor.template packetOp<Index,PacketType>(row, col);
   }
 
   template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index index) const
   {
     return m_functor.template packetOp<Index,PacketType>(index);
@@ -380,7 +400,8 @@ struct unary_evaluator<CwiseUnaryOp<UnaryOp, ArgType>, IndexBased >
     Alignment = evaluator<ArgType>::Alignment
   };
 
-  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  explicit unary_evaluator(const XprType& op)
     : m_functor(op.functor()), 
       m_argImpl(op.nestedExpression()) 
   {
@@ -390,23 +411,27 @@ struct unary_evaluator<CwiseUnaryOp<UnaryOp, ArgType>, IndexBased >
 
   typedef typename XprType::CoeffReturnType CoeffReturnType;
 
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index row, Index col) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
   {
     return m_functor(m_argImpl.coeff(row, col));
   }
 
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index index) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
   {
     return m_functor(m_argImpl.coeff(index));
   }
 
   template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index row, Index col) const
   {
     return m_functor.packetOp(m_argImpl.template packet<LoadMode, PacketType>(row, col));
   }
 
   template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index index) const
   {
     return m_functor.packetOp(m_argImpl.template packet<LoadMode, PacketType>(index));
@@ -466,17 +491,20 @@ struct binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs>, IndexBased, IndexBase
 
   typedef typename XprType::CoeffReturnType CoeffReturnType;
 
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index row, Index col) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
   {
     return m_functor(m_lhsImpl.coeff(row, col), m_rhsImpl.coeff(row, col));
   }
 
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index index) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
   {
     return m_functor(m_lhsImpl.coeff(index), m_rhsImpl.coeff(index));
   }
 
   template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index row, Index col) const
   {
     return m_functor.packetOp(m_lhsImpl.template packet<LoadMode,PacketType>(row, col),
@@ -484,6 +512,7 @@ struct binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs>, IndexBased, IndexBase
   }
 
   template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index index) const
   {
     return m_functor.packetOp(m_lhsImpl.template packet<LoadMode,PacketType>(index),
@@ -523,22 +552,26 @@ struct unary_evaluator<CwiseUnaryView<UnaryOp, ArgType>, IndexBased>
   typedef typename XprType::Scalar Scalar;
   typedef typename XprType::CoeffReturnType CoeffReturnType;
 
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index row, Index col) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
   {
     return m_unaryOp(m_argImpl.coeff(row, col));
   }
 
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index index) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
   {
     return m_unaryOp(m_argImpl.coeff(index));
   }
 
-  EIGEN_DEVICE_FUNC Scalar& coeffRef(Index row, Index col)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
   {
     return m_unaryOp(m_argImpl.coeffRef(row, col));
   }
 
-  EIGEN_DEVICE_FUNC Scalar& coeffRef(Index index)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
   {
     return m_unaryOp(m_argImpl.coeffRef(index));
   }
@@ -578,47 +611,55 @@ struct mapbase_evaluator : evaluator_base<Derived>
     EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
   }
  
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index row, Index col) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
   {
     return m_data[col * m_xpr.colStride() + row * m_xpr.rowStride()];
   }
   
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index index) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
   {
     return m_data[index * m_xpr.innerStride()];
   }
 
-  EIGEN_DEVICE_FUNC Scalar& coeffRef(Index row, Index col)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
   {
     return m_data[col * m_xpr.colStride() + row * m_xpr.rowStride()];
   }
   
-  EIGEN_DEVICE_FUNC Scalar& coeffRef(Index index)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
   {
     return m_data[index * m_xpr.innerStride()];
   }
  
-  template<int LoadMode, typename PacketType> 
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index row, Index col) const 
   {
     PointerType ptr = m_data + row * m_xpr.rowStride() + col * m_xpr.colStride();
     return internal::ploadt<PacketType, LoadMode>(ptr);
   }
 
-  template<int LoadMode, typename PacketType> 
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index index) const 
   {
     return internal::ploadt<PacketType, LoadMode>(m_data + index * m_xpr.innerStride());
   }
   
-  template<int StoreMode, typename PacketType> 
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   void writePacket(Index row, Index col, const PacketType& x) 
   {
     PointerType ptr = m_data + row * m_xpr.rowStride() + col * m_xpr.colStride();
     return internal::pstoret<Scalar, PacketType, StoreMode>(ptr, x);
   }
   
-  template<int StoreMode, typename PacketType> 
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   void writePacket(Index index, const PacketType& x) 
   {
     internal::pstoret<Scalar, PacketType, StoreMode>(m_data + index * m_xpr.innerStride(), x);
@@ -767,46 +808,54 @@ struct unary_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel>, IndexBa
     RowsAtCompileTime = XprType::RowsAtCompileTime
   };
  
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index row, Index col) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
   { 
     return m_argImpl.coeff(m_startRow.value() + row, m_startCol.value() + col); 
   }
   
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index index) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
   { 
     return coeff(RowsAtCompileTime == 1 ? 0 : index, RowsAtCompileTime == 1 ? index : 0);
   }
 
-  EIGEN_DEVICE_FUNC Scalar& coeffRef(Index row, Index col)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
   { 
     return m_argImpl.coeffRef(m_startRow.value() + row, m_startCol.value() + col); 
   }
   
-  EIGEN_DEVICE_FUNC Scalar& coeffRef(Index index)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
   { 
     return coeffRef(RowsAtCompileTime == 1 ? 0 : index, RowsAtCompileTime == 1 ? index : 0);
   }
  
-  template<int LoadMode, typename PacketType> 
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index row, Index col) const 
   { 
     return m_argImpl.template packet<LoadMode,PacketType>(m_startRow.value() + row, m_startCol.value() + col); 
   }
 
-  template<int LoadMode, typename PacketType> 
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index index) const 
   { 
     return packet<LoadMode,PacketType>(RowsAtCompileTime == 1 ? 0 : index,
                                        RowsAtCompileTime == 1 ? index : 0);
   }
   
-  template<int StoreMode, typename PacketType> 
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   void writePacket(Index row, Index col, const PacketType& x) 
   { 
     return m_argImpl.template writePacket<StoreMode,PacketType>(m_startRow.value() + row, m_startCol.value() + col, x); 
   }
   
-  template<int StoreMode, typename PacketType> 
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   void writePacket(Index index, const PacketType& x) 
   { 
     return writePacket<StoreMode,PacketType>(RowsAtCompileTime == 1 ? 0 : index,
@@ -859,7 +908,7 @@ struct evaluator<Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >
     Alignment = EIGEN_PLAIN_ENUM_MIN(evaluator<ThenMatrixType>::Alignment, evaluator<ElseMatrixType>::Alignment)
   };
 
-  inline EIGEN_DEVICE_FUNC  explicit evaluator(const XprType& select)
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& select)
     : m_conditionImpl(select.conditionMatrix()),
       m_thenImpl(select.thenMatrix()),
       m_elseImpl(select.elseMatrix())
@@ -869,7 +918,8 @@ struct evaluator<Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >
  
   typedef typename XprType::CoeffReturnType CoeffReturnType;
 
-  inline EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index row, Index col) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
   {
     if (m_conditionImpl.coeff(row, col))
       return m_thenImpl.coeff(row, col);
@@ -877,7 +927,8 @@ struct evaluator<Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >
       return m_elseImpl.coeff(row, col);
   }
 
-  inline EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index index) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
   {
     if (m_conditionImpl.coeff(index))
       return m_thenImpl.coeff(index);
@@ -921,7 +972,8 @@ struct unary_evaluator<Replicate<ArgType, RowFactor, ColFactor> >
       m_cols(replicate.nestedExpression().cols())
   {}
  
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index row, Index col) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
   {
     // try to avoid using modulo; this is a pure optimization strategy
     const Index actual_row = internal::traits<XprType>::RowsAtCompileTime==1 ? 0
@@ -934,7 +986,8 @@ struct unary_evaluator<Replicate<ArgType, RowFactor, ColFactor> >
     return m_argImpl.coeff(actual_row, actual_col);
   }
   
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index index) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
   {
     // try to avoid using modulo; this is a pure optimization strategy
     const Index actual_index = internal::traits<XprType>::RowsAtCompileTime==1
@@ -945,6 +998,7 @@ struct unary_evaluator<Replicate<ArgType, RowFactor, ColFactor> >
   }
 
   template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index row, Index col) const
   {
     const Index actual_row = internal::traits<XprType>::RowsAtCompileTime==1 ? 0
@@ -958,6 +1012,7 @@ struct unary_evaluator<Replicate<ArgType, RowFactor, ColFactor> >
   }
   
   template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index index) const
   {
     const Index actual_index = internal::traits<XprType>::RowsAtCompileTime==1
@@ -1008,7 +1063,8 @@ struct evaluator<PartialReduxExpr<ArgType, MemberOp, Direction> >
 
   typedef typename XprType::CoeffReturnType CoeffReturnType;
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar coeff(Index i, Index j) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  const Scalar coeff(Index i, Index j) const
   {
     if (Direction==Vertical)
       return m_functor(m_arg.col(j));
@@ -1016,7 +1072,8 @@ struct evaluator<PartialReduxExpr<ArgType, MemberOp, Direction> >
       return m_functor(m_arg.row(i));
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar coeff(Index index) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  const Scalar coeff(Index index) const
   {
     if (Direction==Vertical)
       return m_functor(m_arg.col(index));
@@ -1051,45 +1108,53 @@ struct evaluator_wrapper_base
   typedef typename ArgType::Scalar Scalar;
   typedef typename ArgType::CoeffReturnType CoeffReturnType;
 
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index row, Index col) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
   {
     return m_argImpl.coeff(row, col);
   }
 
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index index) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
   {
     return m_argImpl.coeff(index);
   }
 
-  EIGEN_DEVICE_FUNC Scalar& coeffRef(Index row, Index col)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
   {
     return m_argImpl.coeffRef(row, col);
   }
 
-  EIGEN_DEVICE_FUNC Scalar& coeffRef(Index index)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
   {
     return m_argImpl.coeffRef(index);
   }
 
-  template<int LoadMode, typename PacketType> 
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index row, Index col) const
   {
     return m_argImpl.template packet<LoadMode,PacketType>(row, col);
   }
 
-  template<int LoadMode, typename PacketType> 
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index index) const
   {
     return m_argImpl.template packet<LoadMode,PacketType>(index);
   }
 
-  template<int StoreMode, typename PacketType> 
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   void writePacket(Index row, Index col, const PacketType& x)
   {
     m_argImpl.template writePacket<StoreMode>(row, col, x);
   }
 
-  template<int StoreMode, typename PacketType> 
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   void writePacket(Index index, const PacketType& x)
   {
     m_argImpl.template writePacket<StoreMode>(index, x);
@@ -1164,29 +1229,34 @@ struct unary_evaluator<Reverse<ArgType, Direction> >
       m_cols(ReverseCol ? reverse.nestedExpression().cols() : 1)
   { }
  
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index row, Index col) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
   {
     return m_argImpl.coeff(ReverseRow ? m_rows.value() - row - 1 : row,
                            ReverseCol ? m_cols.value() - col - 1 : col);
   }
 
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index index) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
   {
     return m_argImpl.coeff(m_rows.value() * m_cols.value() - index - 1);
   }
 
-  EIGEN_DEVICE_FUNC Scalar& coeffRef(Index row, Index col)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
   {
     return m_argImpl.coeffRef(ReverseRow ? m_rows.value() - row - 1 : row,
                               ReverseCol ? m_cols.value() - col - 1 : col);
   }
 
-  EIGEN_DEVICE_FUNC Scalar& coeffRef(Index index)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
   {
     return m_argImpl.coeffRef(m_rows.value() * m_cols.value() - index - 1);
   }
 
   template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index row, Index col) const
   {
     enum {
@@ -1201,6 +1271,7 @@ struct unary_evaluator<Reverse<ArgType, Direction> >
   }
 
   template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   PacketType packet(Index index) const
   {
     enum { PacketSize = unpacket_traits<PacketType>::size };
@@ -1208,6 +1279,7 @@ struct unary_evaluator<Reverse<ArgType, Direction> >
   }
 
   template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   void writePacket(Index row, Index col, const PacketType& x)
   {
     // FIXME we could factorize some code with packet(i,j)
@@ -1224,6 +1296,7 @@ struct unary_evaluator<Reverse<ArgType, Direction> >
   }
 
   template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
   void writePacket(Index index, const PacketType& x)
   {
     enum { PacketSize = unpacket_traits<PacketType>::size };
@@ -1267,22 +1340,26 @@ struct evaluator<Diagonal<ArgType, DiagIndex> >
   typedef typename internal::conditional<!internal::is_same<typename ArgType::StorageKind,Sparse>::value,
                                          typename XprType::CoeffReturnType,Scalar>::type CoeffReturnType;
 
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index row, Index) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index) const
   {
     return m_argImpl.coeff(row + rowOffset(), row + colOffset());
   }
 
-  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index index) const
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
   {
     return m_argImpl.coeff(index + rowOffset(), index + colOffset());
   }
 
-  EIGEN_DEVICE_FUNC Scalar& coeffRef(Index row, Index)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index)
   {
     return m_argImpl.coeffRef(row + rowOffset(), row + colOffset());
   }
 
-  EIGEN_DEVICE_FUNC Scalar& coeffRef(Index index)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
   {
     return m_argImpl.coeffRef(index + rowOffset(), index + colOffset());
   }
diff --git a/Eigen/src/Core/CwiseUnaryOp.h b/Eigen/src/Core/CwiseUnaryOp.h
index 8c182303c..ff1391996 100644
--- a/Eigen/src/Core/CwiseUnaryOp.h
+++ b/Eigen/src/Core/CwiseUnaryOp.h
@@ -61,26 +61,26 @@ class CwiseUnaryOp : public CwiseUnaryOpImpl<UnaryOp, XprType, typename internal
     typedef typename internal::ref_selector<XprType>::type XprTypeNested;
     typedef typename internal::remove_all<XprType>::type NestedExpression;
 
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     explicit inline CwiseUnaryOp(const XprType& xpr, const UnaryOp& func = UnaryOp())
       : m_xpr(xpr), m_functor(func) {}
 
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE Index rows() const { return m_xpr.rows(); }
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE Index cols() const { return m_xpr.cols(); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Index rows() const { return m_xpr.rows(); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Index cols() const { return m_xpr.cols(); }
 
     /** \returns the functor representing the unary operation */
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     const UnaryOp& functor() const { return m_functor; }
 
     /** \returns the nested expression */
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     const typename internal::remove_all<XprTypeNested>::type&
     nestedExpression() const { return m_xpr; }
 
     /** \returns the nested expression */
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     typename internal::remove_all<XprTypeNested>::type&
     nestedExpression() { return m_xpr; }
 
diff --git a/Eigen/src/Core/DenseBase.h b/Eigen/src/Core/DenseBase.h
index 2c5c0ad28..ea8283cea 100644
--- a/Eigen/src/Core/DenseBase.h
+++ b/Eigen/src/Core/DenseBase.h
@@ -275,13 +275,13 @@ template<typename Derived> class DenseBase
 
     /** Copies \a other into *this. \returns a reference to *this. */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const DenseBase<OtherDerived>& other);
 
     /** Special case of the template operator=, in order to prevent the compiler
       * from generating a default operator= (issue hit with g++ 4.1)
       */
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const DenseBase& other);
 
     template<typename OtherDerived>
@@ -388,9 +388,9 @@ template<typename Derived> class DenseBase
     inline bool hasNaN() const;
     inline bool allFinite() const;
 
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     inline Derived& operator*=(const Scalar& other);
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     inline Derived& operator/=(const Scalar& other);
 
     typedef typename internal::add_const_on_value_type<typename internal::eval<Derived>::type>::type EvalReturnType;
diff --git a/Eigen/src/Core/MatrixBase.h b/Eigen/src/Core/MatrixBase.h
index 338879c73..3770ab257 100644
--- a/Eigen/src/Core/MatrixBase.h
+++ b/Eigen/src/Core/MatrixBase.h
@@ -135,14 +135,14 @@ template<typename Derived> class MatrixBase
     /** Special case of the template operator=, in order to prevent the compiler
       * from generating a default operator= (issue hit with g++ 4.1)
       */
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const MatrixBase& other);
 
     // We cannot inherit here via Base::operator= since it is causing
     // trouble with MSVC.
 
     template <typename OtherDerived>
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const DenseBase<OtherDerived>& other);
 
     template <typename OtherDerived>
@@ -154,10 +154,10 @@ template<typename Derived> class MatrixBase
     Derived& operator=(const ReturnByValue<OtherDerived>& other);
 
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator+=(const MatrixBase<OtherDerived>& other);
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator-=(const MatrixBase<OtherDerived>& other);
 
 #ifdef __CUDACC__