From 972cf0c28a8d2ee0808c1277dea2c5c206591ce6 Mon Sep 17 00:00:00 2001
From: Nathan Luehr <nluehr@nvidia.com>
Date: Tue, 11 May 2021 22:47:49 +0000
Subject: Fix calls to device functions from host code

---
 unsupported/Eigen/CXX11/src/Tensor/TensorArgMax.h  | 12 ++---
 unsupported/Eigen/CXX11/src/Tensor/TensorAssign.h  |  8 ++--
 .../Eigen/CXX11/src/Tensor/TensorBroadcasting.h    |  9 ++--
 .../Eigen/CXX11/src/Tensor/TensorChipping.h        |  8 ++--
 .../Eigen/CXX11/src/Tensor/TensorConcatenation.h   |  8 ++--
 .../Eigen/CXX11/src/Tensor/TensorContraction.h     | 10 ++--
 .../Eigen/CXX11/src/Tensor/TensorContractionGpu.h  |  4 +-
 .../Eigen/CXX11/src/Tensor/TensorContractionSycl.h |  6 +--
 .../Eigen/CXX11/src/Tensor/TensorConversion.h      | 18 ++++----
 .../Eigen/CXX11/src/Tensor/TensorConvolution.h     |  8 ++--
 .../Eigen/CXX11/src/Tensor/TensorConvolutionSycl.h |  6 +--
 .../Eigen/CXX11/src/Tensor/TensorCustomOp.h        | 16 +++----
 unsupported/Eigen/CXX11/src/Tensor/TensorEvalTo.h  | 10 ++--
 .../Eigen/CXX11/src/Tensor/TensorEvaluator.h       | 53 +++++++++++-----------
 unsupported/Eigen/CXX11/src/Tensor/TensorFFT.h     |  6 +--
 .../Eigen/CXX11/src/Tensor/TensorForcedEval.h      |  9 ++--
 .../Eigen/CXX11/src/Tensor/TensorGenerator.h       |  6 +--
 .../Eigen/CXX11/src/Tensor/TensorImagePatch.h      | 28 ++++++------
 .../Eigen/CXX11/src/Tensor/TensorInflation.h       |  6 +--
 .../Eigen/CXX11/src/Tensor/TensorLayoutSwap.h      |  8 ++--
 .../Eigen/CXX11/src/Tensor/TensorMorphing.h        | 28 ++++++------
 unsupported/Eigen/CXX11/src/Tensor/TensorPadding.h |  8 ++--
 unsupported/Eigen/CXX11/src/Tensor/TensorPatch.h   |  6 +--
 .../Eigen/CXX11/src/Tensor/TensorReduction.h       | 25 ++--------
 unsupported/Eigen/CXX11/src/Tensor/TensorRef.h     |  8 ++--
 unsupported/Eigen/CXX11/src/Tensor/TensorReverse.h | 12 ++---
 unsupported/Eigen/CXX11/src/Tensor/TensorScan.h    |  5 +-
 .../Eigen/CXX11/src/Tensor/TensorShuffling.h       | 11 ++---
 .../Eigen/CXX11/src/Tensor/TensorStriding.h        |  8 ++--
 unsupported/Eigen/CXX11/src/Tensor/TensorTrace.h   |  6 +--
 .../Eigen/CXX11/src/Tensor/TensorVolumePatch.h     | 36 +++++++--------
 31 files changed, 182 insertions(+), 210 deletions(-)

diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorArgMax.h b/unsupported/Eigen/CXX11/src/Tensor/TensorArgMax.h
index 91a6f8d6c..8b8fb9235 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorArgMax.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorArgMax.h
@@ -99,18 +99,18 @@ struct TensorEvaluator<const TensorIndexTupleOp<ArgType>, Device>
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_impl(op.expression(), device) { }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const {
     return m_impl.dimensions();
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
     m_impl.evalSubExprsIfNeeded(NULL);
     return true;
   }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
   }
 
@@ -240,7 +240,7 @@ struct TensorEvaluator<const TensorTupleReducerOp<ReduceOp, Dims, ArgType>, Devi
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_orig_impl(op.expression(), device),
         m_impl(op.expression().index_tuples().reduce(op.reduce_dims(), op.reduce_op()), device),
         m_return_dim(op.return_dim())
@@ -263,11 +263,11 @@ struct TensorEvaluator<const TensorTupleReducerOp<ReduceOp, Dims, ArgType>, Devi
     return m_impl.dimensions();
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
     m_impl.evalSubExprsIfNeeded(NULL);
     return true;
   }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
   }
 
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorAssign.h b/unsupported/Eigen/CXX11/src/Tensor/TensorAssign.h
index 72f072cf2..ca0453f79 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorAssign.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorAssign.h
@@ -124,7 +124,7 @@ struct TensorEvaluator<const TensorAssignOp<LeftArgType, RightArgType>, Device>
       RightTensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC TensorEvaluator(const XprType& op, const Device& device) :
+  TensorEvaluator(const XprType& op, const Device& device) :
       m_leftImpl(op.lhsExpression(), device),
       m_rightImpl(op.rhsExpression(), device)
   {
@@ -142,7 +142,7 @@ struct TensorEvaluator<const TensorAssignOp<LeftArgType, RightArgType>, Device>
     return m_rightImpl.dimensions();
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
     eigen_assert(dimensions_match(m_leftImpl.dimensions(), m_rightImpl.dimensions()));
     m_leftImpl.evalSubExprsIfNeeded(NULL);
     // If the lhs provides raw access to its storage area (i.e. if m_leftImpl.data() returns a non
@@ -154,7 +154,7 @@ struct TensorEvaluator<const TensorAssignOp<LeftArgType, RightArgType>, Device>
 
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
+  EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
       EvaluatorPointerType, EvalSubExprsCallback done) {
     m_leftImpl.evalSubExprsIfNeededAsync(nullptr, [this, done](bool) {
       m_rightImpl.evalSubExprsIfNeededAsync(
@@ -163,7 +163,7 @@ struct TensorEvaluator<const TensorAssignOp<LeftArgType, RightArgType>, Device>
   }
 #endif  // EIGEN_USE_THREADS
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_leftImpl.cleanup();
     m_rightImpl.cleanup();
   }
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorBroadcasting.h b/unsupported/Eigen/CXX11/src/Tensor/TensorBroadcasting.h
index fc75c8d9a..a354132f6 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorBroadcasting.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorBroadcasting.h
@@ -138,8 +138,7 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
       TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op,
-                                                        const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : isCopy(false), nByOne(false), oneByN(false),
         m_device(device), m_broadcast(op.broadcast()), m_impl(op.expression(), device)
   {
@@ -211,20 +210,20 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
     m_impl.evalSubExprsIfNeeded(NULL);
     return true;
   }
 
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
+  EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
       EvaluatorPointerType, EvalSubExprsCallback done) {
     m_impl.evalSubExprsIfNeededAsync(nullptr, [done](bool) { done(true); });
   }
 #endif  // EIGEN_USE_THREADS
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
   }
 
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorChipping.h b/unsupported/Eigen/CXX11/src/Tensor/TensorChipping.h
index 7c6bbd180..376457341 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorChipping.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorChipping.h
@@ -164,7 +164,7 @@ struct TensorEvaluator<const TensorChippingOp<DimId, ArgType>, Device>
       TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_impl(op.expression(), device), m_dim(op.dim()), m_device(device)
   {
     EIGEN_STATIC_ASSERT((NumInputDims >= 1), YOU_MADE_A_PROGRAMMING_MISTAKE);
@@ -200,12 +200,12 @@ struct TensorEvaluator<const TensorChippingOp<DimId, ArgType>, Device>
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
     m_impl.evalSubExprsIfNeeded(NULL);
     return true;
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
   }
 
@@ -433,7 +433,7 @@ struct TensorEvaluator<TensorChippingOp<DimId, ArgType>, Device>
   typedef internal::TensorBlockDescriptor<NumDims, Index> TensorBlockDesc;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
     : Base(op, device)
     { }
 
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorConcatenation.h b/unsupported/Eigen/CXX11/src/Tensor/TensorConcatenation.h
index 0dfe21604..5235a8e6f 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorConcatenation.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorConcatenation.h
@@ -119,7 +119,7 @@ struct TensorEvaluator<const TensorConcatenationOp<Axis, LeftArgType, RightArgTy
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
     : m_leftImpl(op.lhsExpression(), device), m_rightImpl(op.rhsExpression(), device), m_axis(op.axis())
   {
     EIGEN_STATIC_ASSERT((static_cast<int>(TensorEvaluator<LeftArgType, Device>::Layout) == static_cast<int>(TensorEvaluator<RightArgType, Device>::Layout) || NumDims == 1), YOU_MADE_A_PROGRAMMING_MISTAKE);
@@ -172,14 +172,14 @@ struct TensorEvaluator<const TensorConcatenationOp<Axis, LeftArgType, RightArgTy
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
   // TODO(phli): Add short-circuit memcpy evaluation if underlying data are linear?
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType)
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType)
   {
     m_leftImpl.evalSubExprsIfNeeded(NULL);
     m_rightImpl.evalSubExprsIfNeeded(NULL);
     return true;
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup()
+  EIGEN_STRONG_INLINE void cleanup()
   {
     m_leftImpl.cleanup();
     m_rightImpl.cleanup();
@@ -318,7 +318,7 @@ template<typename Axis, typename LeftArgType, typename RightArgType, typename De
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(XprType& op, const Device& device)
     : Base(op, device)
   {
     EIGEN_STATIC_ASSERT((static_cast<int>(Layout) == static_cast<int>(ColMajor)), YOU_MADE_A_PROGRAMMING_MISTAKE);
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorContraction.h b/unsupported/Eigen/CXX11/src/Tensor/TensorContraction.h
index 424caced1..d442c782c 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorContraction.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorContraction.h
@@ -417,7 +417,7 @@ struct TensorContractionEvaluatorBase : internal::no_assignment_operator
 
   typedef DSizes<Index, NumDims> Dimensions;
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_STRONG_INLINE
   TensorContractionEvaluatorBase(const XprType& op, const Device& device)
       : m_leftImpl(choose(Cond<static_cast<int>(Layout) == static_cast<int>(ColMajor)>(),
                           op.lhsExpression(), op.rhsExpression()), device),
@@ -602,7 +602,7 @@ struct TensorContractionEvaluatorBase : internal::no_assignment_operator
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
     m_leftImpl.evalSubExprsIfNeeded(NULL);
     m_rightImpl.evalSubExprsIfNeeded(NULL);
     if (data) {
@@ -617,7 +617,7 @@ struct TensorContractionEvaluatorBase : internal::no_assignment_operator
 
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
+  EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
       EvaluatorPointerType dest, EvalSubExprsCallback done) {
     m_leftImpl.evalSubExprsIfNeededAsync(nullptr, [this, done, dest](bool) {
       m_rightImpl.evalSubExprsIfNeededAsync(nullptr, [this, done, dest](bool) {
@@ -908,7 +908,7 @@ struct TensorContractionEvaluatorBase : internal::no_assignment_operator
     kernel.deallocate(this->m_device, packed_mem);
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_leftImpl.cleanup();
     m_rightImpl.cleanup();
 
@@ -1005,7 +1005,7 @@ struct TensorEvaluator<const TensorContractionOp<Indices, LeftArgType, RightArgT
   // Could we use NumDimensions here?
   typedef DSizes<Index, NumDims> Dimensions;
 
-  EIGEN_DEVICE_FUNC TensorEvaluator(const XprType& op, const Device& device) :
+  TensorEvaluator(const XprType& op, const Device& device) :
       Base(op, device) { }
 
   template <int Alignment>
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorContractionGpu.h b/unsupported/Eigen/CXX11/src/Tensor/TensorContractionGpu.h
index bb990b378..c81803827 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorContractionGpu.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorContractionGpu.h
@@ -1270,7 +1270,7 @@ struct TensorEvaluator<const TensorContractionOp<Indices, LeftArgType, RightArgT
   typedef typename LeftEvaluator::Dimensions LeftDimensions;
   typedef typename RightEvaluator::Dimensions RightDimensions;
 
-  EIGEN_DEVICE_FUNC TensorEvaluator(const XprType& op, const Device& device) :
+  TensorEvaluator(const XprType& op, const Device& device) :
       Base(op, device)
   {
     EIGEN_STATIC_ASSERT( (internal::is_same<OutputKernelType, const NoOpOutputKernel>::value),
@@ -1278,7 +1278,7 @@ struct TensorEvaluator<const TensorContractionOp<Indices, LeftArgType, RightArgT
   }
 
   // We need to redefine this method to make nvcc happy
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(Scalar* data) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(Scalar* data) {
     this->m_leftImpl.evalSubExprsIfNeeded(NULL);
     this->m_rightImpl.evalSubExprsIfNeeded(NULL);
     if (data) {
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorContractionSycl.h b/unsupported/Eigen/CXX11/src/Tensor/TensorContractionSycl.h
index a6ca1777a..473c22849 100755
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorContractionSycl.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorContractionSycl.h
@@ -1340,10 +1340,10 @@ struct TensorEvaluator<const TensorContractionOp<Indices, LeftArgType, RightArgT
         (RDims == 2 && ContractDims == 1) || (rhs_inner_dim_contiguous && !rhs_inner_dim_reordered);
   };
 
-  EIGEN_DEVICE_FUNC TensorEvaluator(const XprType &op, const Device &device) : Base(op, device) {}
+  TensorEvaluator(const XprType &op, const Device &device) : Base(op, device) {}
 
   // We need to redefine this method to make nvcc happy
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(typename Base::EvaluatorPointerType data) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(typename Base::EvaluatorPointerType data) {
     this->m_leftImpl.evalSubExprsIfNeeded(NULL);
     this->m_rightImpl.evalSubExprsIfNeeded(NULL);
     if (!data) {
@@ -1630,7 +1630,7 @@ struct TensorEvaluator<const TensorContractionOp<Indices, LeftArgType, RightArgT
   }
 #endif
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     this->m_leftImpl.cleanup();
     this->m_rightImpl.cleanup();
 
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorConversion.h b/unsupported/Eigen/CXX11/src/Tensor/TensorConversion.h
index 44493906d..4968babc1 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorConversion.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorConversion.h
@@ -195,14 +195,14 @@ class TensorConversionOp : public TensorBase<TensorConversionOp<TargetType, XprT
 };
 
 template <bool SameType, typename Eval, typename EvalPointerType> struct ConversionSubExprEval {
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool run(Eval& impl, EvalPointerType) {
+  static EIGEN_STRONG_INLINE bool run(Eval& impl, EvalPointerType) {
     impl.evalSubExprsIfNeeded(NULL);
     return true;
   }
 };
 
 template <typename Eval, typename EvalPointerType> struct ConversionSubExprEval<true, Eval, EvalPointerType> {
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool run(Eval& impl, EvalPointerType data) {
+  static EIGEN_STRONG_INLINE bool run(Eval& impl, EvalPointerType data) {
     return impl.evalSubExprsIfNeeded(data);
   }
 };
@@ -211,8 +211,7 @@ template <typename Eval, typename EvalPointerType> struct ConversionSubExprEval<
 template <bool SameType, typename Eval, typename EvalPointerType,
           typename EvalSubExprsCallback>
 struct ConversionSubExprEvalAsync {
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(
-      Eval& impl, EvalPointerType, EvalSubExprsCallback done) {
+  static EIGEN_STRONG_INLINE void run(Eval& impl, EvalPointerType, EvalSubExprsCallback done) {
     impl.evalSubExprsIfNeededAsync(nullptr, std::move(done));
   }
 };
@@ -221,8 +220,7 @@ template <typename Eval, typename EvalPointerType,
           typename EvalSubExprsCallback>
 struct ConversionSubExprEvalAsync<true, Eval, EvalPointerType,
                                   EvalSubExprsCallback> {
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(
-      Eval& impl, EvalPointerType data, EvalSubExprsCallback done) {
+  static EIGEN_STRONG_INLINE void run(Eval& impl, EvalPointerType data, EvalSubExprsCallback done) {
     impl.evalSubExprsIfNeededAsync(data, std::move(done));
   }
 };
@@ -363,21 +361,21 @@ struct TensorEvaluator<const TensorConversionOp<TargetType, ArgType>, Device>
       TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
     : m_impl(op.expression(), device)
   {
   }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_impl.dimensions(); }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data)
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data)
   {
     return ConversionSubExprEval<IsSameType, TensorEvaluator<ArgType, Device>, EvaluatorPointerType>::run(m_impl, data);
   }
 
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
+  EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
       EvaluatorPointerType data, EvalSubExprsCallback done) {
     ConversionSubExprEvalAsync<IsSameType, TensorEvaluator<ArgType, Device>,
                                EvaluatorPointerType,
@@ -385,7 +383,7 @@ struct TensorEvaluator<const TensorConversionOp<TargetType, ArgType>, Device>
   }
 #endif
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup()
+  EIGEN_STRONG_INLINE void cleanup()
   {
     m_impl.cleanup();
   }
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorConvolution.h b/unsupported/Eigen/CXX11/src/Tensor/TensorConvolution.h
index df289e2c0..1b71023c0 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorConvolution.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorConvolution.h
@@ -320,7 +320,7 @@ struct TensorEvaluator<const TensorConvolutionOp<Indices, InputArgType, KernelAr
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_inputImpl(op.inputExpression(), device), m_kernelImpl(op.kernelExpression(), device), m_kernelArg(op.kernelExpression()), m_kernel(NULL), m_local_kernel(false), m_device(device)
   {
     EIGEN_STATIC_ASSERT((static_cast<int>(TensorEvaluator<InputArgType, Device>::Layout) == static_cast<int>(TensorEvaluator<KernelArgType, Device>::Layout)), YOU_MADE_A_PROGRAMMING_MISTAKE);
@@ -384,12 +384,12 @@ struct TensorEvaluator<const TensorConvolutionOp<Indices, InputArgType, KernelAr
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(Scalar*) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(Scalar*) {
     m_inputImpl.evalSubExprsIfNeeded(NULL);
     preloadKernel();
     return true;
   }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_inputImpl.cleanup();
     if (m_local_kernel) {
       m_device.deallocate((void*)m_kernel);
@@ -797,7 +797,7 @@ struct TensorEvaluator<const TensorConvolutionOp<Indices, InputArgType, KernelAr
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC TensorEvaluator(const XprType& op, const GpuDevice& device)
+  TensorEvaluator(const XprType& op, const GpuDevice& device)
       : m_inputImpl(op.inputExpression(), device), m_kernelImpl(op.kernelExpression(), device), m_kernelArg(op.kernelExpression()), m_indices(op.indices()), m_buf(NULL), m_kernel(NULL), m_local_kernel(false), m_device(device)
   {
     EIGEN_STATIC_ASSERT((static_cast<int>(TensorEvaluator<InputArgType, GpuDevice>::Layout) == static_cast<int>(TensorEvaluator<KernelArgType, GpuDevice>::Layout)), YOU_MADE_A_PROGRAMMING_MISTAKE);
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorConvolutionSycl.h b/unsupported/Eigen/CXX11/src/Tensor/TensorConvolutionSycl.h
index 92003c766..033318fdc 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorConvolutionSycl.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorConvolutionSycl.h
@@ -305,7 +305,7 @@ struct TensorEvaluator<const TensorConvolutionOp<Indices, InputArgType, KernelAr
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC TensorEvaluator(const XprType &op, const Eigen::SyclDevice &device)
+  TensorEvaluator(const XprType &op, const Eigen::SyclDevice &device)
       : m_inputImpl(op.inputExpression(), device),
         m_kernelArg(op.kernelExpression()),
         m_kernelImpl(op.kernelExpression(), device),
@@ -334,7 +334,7 @@ struct TensorEvaluator<const TensorConvolutionOp<Indices, InputArgType, KernelAr
 
   EIGEN_DEVICE_FUNC const Dimensions &dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
     preloadKernel();
     m_inputImpl.evalSubExprsIfNeeded(NULL);
     if (data) {
@@ -348,7 +348,7 @@ struct TensorEvaluator<const TensorConvolutionOp<Indices, InputArgType, KernelAr
     }
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_inputImpl.cleanup();
     if (m_buf) {
       m_device.deallocate_temp(m_buf);
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorCustomOp.h b/unsupported/Eigen/CXX11/src/Tensor/TensorCustomOp.h
index 476b2282a..95a8a84ee 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorCustomOp.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorCustomOp.h
@@ -106,7 +106,7 @@ struct TensorEvaluator<const TensorCustomUnaryOp<CustomUnaryFunc, XprType>, Devi
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const ArgType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const ArgType& op, const Device& device)
       : m_op(op), m_device(device), m_result(NULL)
   {
     m_dimensions = op.func().dimensions(op.expression());
@@ -114,7 +114,7 @@ struct TensorEvaluator<const TensorCustomUnaryOp<CustomUnaryFunc, XprType>, Devi
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
     if (data) {
       evalTo(data);
       return false;
@@ -126,7 +126,7 @@ struct TensorEvaluator<const TensorCustomUnaryOp<CustomUnaryFunc, XprType>, Devi
     }
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     if (m_result) {
       m_device.deallocate_temp(m_result);
       m_result = NULL;
@@ -157,7 +157,7 @@ struct TensorEvaluator<const TensorCustomUnaryOp<CustomUnaryFunc, XprType>, Devi
 #endif
 
  protected:
-  EIGEN_DEVICE_FUNC void evalTo(EvaluatorPointerType data) {
+  void evalTo(EvaluatorPointerType data) {
     TensorMap<Tensor<CoeffReturnType, NumDims, Layout, Index> > result(m_device.get(data), m_dimensions);
     m_op.func().eval(m_op.expression(), result, m_device);
   }
@@ -279,7 +279,7 @@ struct TensorEvaluator<const TensorCustomBinaryOp<CustomBinaryFunc, LhsXprType,
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_op(op), m_device(device), m_result(NULL)
   {
     m_dimensions = op.func().dimensions(op.lhsExpression(), op.rhsExpression());
@@ -287,7 +287,7 @@ struct TensorEvaluator<const TensorCustomBinaryOp<CustomBinaryFunc, LhsXprType,
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
     if (data) {
       evalTo(data);
       return false;
@@ -299,7 +299,7 @@ struct TensorEvaluator<const TensorCustomBinaryOp<CustomBinaryFunc, LhsXprType,
     }
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     if (m_result != NULL) {
       m_device.deallocate_temp(m_result);
       m_result = NULL;
@@ -330,7 +330,7 @@ struct TensorEvaluator<const TensorCustomBinaryOp<CustomBinaryFunc, LhsXprType,
 #endif
 
  protected:
-  EIGEN_DEVICE_FUNC void evalTo(EvaluatorPointerType data) {
+  void evalTo(EvaluatorPointerType data) {
     TensorMap<Tensor<CoeffReturnType, NumDims, Layout> > result(m_device.get(data), m_dimensions);
     m_op.func().eval(m_op.lhsExpression(), m_op.rhsExpression(), result, m_device);
   }
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorEvalTo.h b/unsupported/Eigen/CXX11/src/Tensor/TensorEvalTo.h
index 4689b0230..a48d035f5 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorEvalTo.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorEvalTo.h
@@ -131,17 +131,17 @@ struct TensorEvaluator<const TensorEvalToOp<ArgType, MakePointer_>, Device>
       TensorBlockAssignment;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_impl(op.expression(), device), m_buffer(device.get(op.buffer())), m_expression(op.expression()){}
 
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE ~TensorEvaluator() {
+  EIGEN_STRONG_INLINE ~TensorEvaluator() {
   }
 
 
   EIGEN_DEVICE_FUNC const Dimensions& dimensions() const { return m_impl.dimensions(); }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType scalar) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType scalar) {
     EIGEN_UNUSED_VARIABLE(scalar);
     eigen_assert(scalar == NULL);
     return m_impl.evalSubExprsIfNeeded(m_buffer);
@@ -149,7 +149,7 @@ struct TensorEvaluator<const TensorEvalToOp<ArgType, MakePointer_>, Device>
 
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
+  EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
       EvaluatorPointerType scalar, EvalSubExprsCallback done) {
     EIGEN_UNUSED_VARIABLE(scalar);
     eigen_assert(scalar == NULL);
@@ -191,7 +191,7 @@ struct TensorEvaluator<const TensorEvalToOp<ArgType, MakePointer_>, Device>
     block.cleanup();
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
   }
 
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorEvaluator.h b/unsupported/Eigen/CXX11/src/Tensor/TensorEvaluator.h
index d4532b72c..35fe643ea 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorEvaluator.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorEvaluator.h
@@ -63,7 +63,7 @@ struct TensorEvaluator
       TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const Derived& m, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const Derived& m, const Device& device)
       : m_data(device.get((const_cast<TensorPointerType>(m.data())))),
         m_dims(m.dimensions()),
         m_device(device)
@@ -72,7 +72,7 @@ struct TensorEvaluator
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dims; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType dest) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType dest) {
     if (!NumTraits<typename internal::remove_const<Scalar>::type>::RequireInitialization && dest) {
       m_device.memcpy((void*)(m_device.get(dest)), m_device.get(m_data), m_dims.TotalSize() * sizeof(Scalar));
       return false;
@@ -82,14 +82,14 @@ struct TensorEvaluator
 
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
+  EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
       EvaluatorPointerType dest, EvalSubExprsCallback done) {
     // TODO(ezhulenev): ThreadPoolDevice memcpy is blockign operation.
     done(evalSubExprsIfNeeded(dest));
   }
 #endif  // EIGEN_USE_THREADS
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {}
+  EIGEN_STRONG_INLINE void cleanup() {}
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType coeff(Index index) const {
     eigen_assert(m_data != NULL);
@@ -262,13 +262,13 @@ struct TensorEvaluator<const Derived, Device>
       TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const Derived& m, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const Derived& m, const Device& device)
       : m_data(device.get(m.data())), m_dims(m.dimensions()), m_device(device)
   { }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dims; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
     if (!NumTraits<typename internal::remove_const<Scalar>::type>::RequireInitialization && data) {
       m_device.memcpy((void*)(m_device.get(data)),m_device.get(m_data), m_dims.TotalSize() * sizeof(Scalar));
       return false;
@@ -278,14 +278,14 @@ struct TensorEvaluator<const Derived, Device>
 
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
+  EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
       EvaluatorPointerType dest, EvalSubExprsCallback done) {
     // TODO(ezhulenev): ThreadPoolDevice memcpy is a blockign operation.
     done(evalSubExprsIfNeeded(dest));
   }
 #endif  // EIGEN_USE_THREADS
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() { }
+  EIGEN_STRONG_INLINE void cleanup() { }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType coeff(Index index) const {
     eigen_assert(m_data != NULL);
@@ -357,7 +357,6 @@ struct TensorEvaluator<const TensorCwiseNullaryOp<NullaryOp, ArgType>, Device>
 {
   typedef TensorCwiseNullaryOp<NullaryOp, ArgType> XprType;
 
-  EIGEN_DEVICE_FUNC
   TensorEvaluator(const XprType& op, const Device& device)
       : m_functor(op.functor()), m_argImpl(op.nestedExpression(), device), m_wrapper()
   { }
@@ -391,17 +390,17 @@ struct TensorEvaluator<const TensorCwiseNullaryOp<NullaryOp, ArgType>, Device>
 
   EIGEN_DEVICE_FUNC const Dimensions& dimensions() const { return m_argImpl.dimensions(); }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) { return true; }
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) { return true; }
 
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
+  EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
       EvaluatorPointerType, EvalSubExprsCallback done) {
     done(true);
   }
 #endif  // EIGEN_USE_THREADS
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() { }
+  EIGEN_STRONG_INLINE void cleanup() { }
 
   EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index index) const
   {
@@ -455,7 +454,7 @@ struct TensorEvaluator<const TensorCwiseUnaryOp<UnaryOp, ArgType>, Device>
     RawAccess          = false
   };
 
-  EIGEN_DEVICE_FUNC TensorEvaluator(const XprType& op, const Device& device)
+  TensorEvaluator(const XprType& op, const Device& device)
     : m_device(device),
       m_functor(op.functor()),
       m_argImpl(op.nestedExpression(), device)
@@ -485,20 +484,20 @@ struct TensorEvaluator<const TensorCwiseUnaryOp<UnaryOp, ArgType>, Device>
 
   EIGEN_DEVICE_FUNC const Dimensions& dimensions() const { return m_argImpl.dimensions(); }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
     m_argImpl.evalSubExprsIfNeeded(NULL);
     return true;
   }
 
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
+  EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
       EvaluatorPointerType, EvalSubExprsCallback done) {
     m_argImpl.evalSubExprsIfNeededAsync(nullptr, [done](bool) { done(true); });
   }
 #endif  // EIGEN_USE_THREADS
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_argImpl.cleanup();
   }
 
@@ -571,7 +570,7 @@ struct TensorEvaluator<const TensorCwiseBinaryOp<BinaryOp, LeftArgType, RightArg
     RawAccess         = false
   };
 
-  EIGEN_DEVICE_FUNC TensorEvaluator(const XprType& op, const Device& device)
+  TensorEvaluator(const XprType& op, const Device& device)
     : m_device(device),
       m_functor(op.functor()),
       m_leftImpl(op.lhsExpression(), device),
@@ -613,7 +612,7 @@ struct TensorEvaluator<const TensorCwiseBinaryOp<BinaryOp, LeftArgType, RightArg
     return m_leftImpl.dimensions();
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
     m_leftImpl.evalSubExprsIfNeeded(NULL);
     m_rightImpl.evalSubExprsIfNeeded(NULL);
     return true;
@@ -621,7 +620,7 @@ struct TensorEvaluator<const TensorCwiseBinaryOp<BinaryOp, LeftArgType, RightArg
 
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
+  EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
       EvaluatorPointerType, EvalSubExprsCallback done) {
     // TODO(ezhulenev): Evaluate two expression in parallel?
     m_leftImpl.evalSubExprsIfNeededAsync(nullptr, [this, done](bool) {
@@ -631,7 +630,7 @@ struct TensorEvaluator<const TensorCwiseBinaryOp<BinaryOp, LeftArgType, RightArg
   }
 #endif  // EIGEN_USE_THREADS
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_leftImpl.cleanup();
     m_rightImpl.cleanup();
   }
@@ -709,7 +708,7 @@ struct TensorEvaluator<const TensorCwiseTernaryOp<TernaryOp, Arg1Type, Arg2Type,
     RawAccess         = false
   };
 
-  EIGEN_DEVICE_FUNC TensorEvaluator(const XprType& op, const Device& device)
+  TensorEvaluator(const XprType& op, const Device& device)
     : m_functor(op.functor()),
       m_arg1Impl(op.arg1Expression(), device),
       m_arg2Impl(op.arg2Expression(), device),
@@ -752,13 +751,13 @@ struct TensorEvaluator<const TensorCwiseTernaryOp<TernaryOp, Arg1Type, Arg2Type,
     return m_arg1Impl.dimensions();
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
     m_arg1Impl.evalSubExprsIfNeeded(NULL);
     m_arg2Impl.evalSubExprsIfNeeded(NULL);
     m_arg3Impl.evalSubExprsIfNeeded(NULL);
     return true;
   }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_arg1Impl.cleanup();
     m_arg2Impl.cleanup();
     m_arg3Impl.cleanup();
@@ -829,7 +828,7 @@ struct TensorEvaluator<const TensorSelectOp<IfArgType, ThenArgType, ElseArgType>
     RawAccess         = false
   };
 
-  EIGEN_DEVICE_FUNC TensorEvaluator(const XprType& op, const Device& device)
+  TensorEvaluator(const XprType& op, const Device& device)
     : m_condImpl(op.ifExpression(), device),
       m_thenImpl(op.thenExpression(), device),
       m_elseImpl(op.elseExpression(), device)
@@ -886,7 +885,7 @@ struct TensorEvaluator<const TensorSelectOp<IfArgType, ThenArgType, ElseArgType>
     return m_condImpl.dimensions();
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
     m_condImpl.evalSubExprsIfNeeded(NULL);
     m_thenImpl.evalSubExprsIfNeeded(NULL);
     m_elseImpl.evalSubExprsIfNeeded(NULL);
@@ -895,7 +894,7 @@ struct TensorEvaluator<const TensorSelectOp<IfArgType, ThenArgType, ElseArgType>
 
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
+  EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
       EvaluatorPointerType, EvalSubExprsCallback done) {
     m_condImpl.evalSubExprsIfNeeded(nullptr, [this, done](bool) {
       m_thenImpl.evalSubExprsIfNeeded(nullptr, [this, done](bool) {
@@ -905,7 +904,7 @@ struct TensorEvaluator<const TensorSelectOp<IfArgType, ThenArgType, ElseArgType>
   }
 #endif  // EIGEN_USE_THREADS
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_condImpl.cleanup();
     m_thenImpl.cleanup();
     m_elseImpl.cleanup();
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorFFT.h b/unsupported/Eigen/CXX11/src/Tensor/TensorFFT.h
index c62bc5fa9..4a1a0687c 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorFFT.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorFFT.h
@@ -144,7 +144,7 @@ struct TensorEvaluator<const TensorFFTOp<FFT, ArgType, FFTResultType, FFTDir>, D
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device) : m_fft(op.fft()), m_impl(op.expression(), device), m_data(NULL), m_device(device) {
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device) : m_fft(op.fft()), m_impl(op.expression(), device), m_data(NULL), m_device(device) {
     const typename TensorEvaluator<ArgType, Device>::Dimensions& input_dims = m_impl.dimensions();
     for (int i = 0; i < NumDims; ++i) {
       eigen_assert(input_dims[i] > 0);
@@ -169,7 +169,7 @@ struct TensorEvaluator<const TensorFFTOp<FFT, ArgType, FFTResultType, FFTDir>, D
     return m_dimensions;
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
     m_impl.evalSubExprsIfNeeded(NULL);
     if (data) {
       evalToBuf(data);
@@ -181,7 +181,7 @@ struct TensorEvaluator<const TensorFFTOp<FFT, ArgType, FFTResultType, FFTDir>, D
     }
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     if (m_data) {
       m_device.deallocate(m_data);
       m_data = NULL;
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorForcedEval.h b/unsupported/Eigen/CXX11/src/Tensor/TensorForcedEval.h
index 14020aa68..e800dedc6 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorForcedEval.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorForcedEval.h
@@ -135,16 +135,13 @@ struct TensorEvaluator<const TensorForcedEvalOp<ArgType_>, Device>
       TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC TensorEvaluator(const XprType& op, const Device& device)
+  TensorEvaluator(const XprType& op, const Device& device)
       : m_impl(op.expression(), device), m_op(op.expression()),
       m_device(device), m_buffer(NULL)
   { }
 
   EIGEN_DEVICE_FUNC const Dimensions& dimensions() const { return m_impl.dimensions(); }
 
-  #if !defined(EIGEN_HIPCC)
-  EIGEN_DEVICE_FUNC
-  #endif
   EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
     const Index numValues =  internal::array_prod(m_impl.dimensions());
     m_buffer = m_device.get((CoeffReturnType*)m_device.allocate_temp(numValues * sizeof(CoeffReturnType)));
@@ -165,7 +162,7 @@ struct TensorEvaluator<const TensorForcedEvalOp<ArgType_>, Device>
 
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
-  EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC void evalSubExprsIfNeededAsync(
+  EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
       EvaluatorPointerType, EvalSubExprsCallback done) {
     const Index numValues = internal::array_prod(m_impl.dimensions());
     m_buffer = m_device.get((CoeffReturnType*)m_device.allocate_temp(
@@ -185,7 +182,7 @@ struct TensorEvaluator<const TensorForcedEvalOp<ArgType_>, Device>
   }
 #endif
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_device.deallocate_temp(m_buffer);
     m_buffer = NULL;
   }
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorGenerator.h b/unsupported/Eigen/CXX11/src/Tensor/TensorGenerator.h
index b1ff1d8b1..174bf0683 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorGenerator.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorGenerator.h
@@ -111,7 +111,7 @@ struct TensorEvaluator<const TensorGeneratorOp<Generator, ArgType>, Device>
       TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       :  m_device(device), m_generator(op.generator())
   {
     TensorEvaluator<ArgType, Device> argImpl(op.expression(), device);
@@ -136,10 +136,10 @@ struct TensorEvaluator<const TensorGeneratorOp<Generator, ArgType>, Device>
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
     return true;
   }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
   }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType coeff(Index index) const
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorImagePatch.h b/unsupported/Eigen/CXX11/src/Tensor/TensorImagePatch.h
index 49d1004f3..dd51850b7 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorImagePatch.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorImagePatch.h
@@ -242,7 +242,7 @@ struct TensorEvaluator<const TensorImagePatchOp<Rows, Cols, ArgType>, Device>
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator( const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator( const XprType& op, const Device& device)
       : m_device(device), m_impl(op.expression(), device)
   {
     EIGEN_STATIC_ASSERT((NumDims >= 4), YOU_MADE_A_PROGRAMMING_MISTAKE);
@@ -389,20 +389,20 @@ struct TensorEvaluator<const TensorImagePatchOp<Rows, Cols, ArgType>, Device>
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
     m_impl.evalSubExprsIfNeeded(NULL);
     return true;
   }
 
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
+  EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
       EvaluatorPointerType, EvalSubExprsCallback done) {
     m_impl.evalSubExprsIfNeededAsync(nullptr, [done](bool) { done(true); });
   }
 #endif  // EIGEN_USE_THREADS
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
   }
 
@@ -514,16 +514,16 @@ struct TensorEvaluator<const TensorImagePatchOp<Rows, Cols, ArgType>, Device>
   }
 #endif
 
-  Index rowPaddingTop() const { return m_rowPaddingTop; }
-  Index colPaddingLeft() const { return m_colPaddingLeft; }
-  Index outputRows() const { return m_outputRows; }
-  Index outputCols() const { return m_outputCols; }
-  Index userRowStride() const { return m_row_strides; }
-  Index userColStride() const { return m_col_strides; }
-  Index userInRowStride() const { return m_in_row_strides; }
-  Index userInColStride() const { return m_in_col_strides; }
-  Index rowInflateStride() const { return m_row_inflate_strides; }
-  Index colInflateStride() const { return m_col_inflate_strides; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index rowPaddingTop() const { return m_rowPaddingTop; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index colPaddingLeft() const { return m_colPaddingLeft; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index outputRows() const { return m_outputRows; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index outputCols() const { return m_outputCols; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index userRowStride() const { return m_row_strides; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index userColStride() const { return m_col_strides; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index userInRowStride() const { return m_in_row_strides; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index userInColStride() const { return m_in_col_strides; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index rowInflateStride() const { return m_row_inflate_strides; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index colInflateStride() const { return m_col_inflate_strides; }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost
   costPerCoeff(bool vectorized) const {
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorInflation.h b/unsupported/Eigen/CXX11/src/Tensor/TensorInflation.h
index 7dadec7fb..c5cb61af5 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorInflation.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorInflation.h
@@ -103,7 +103,7 @@ struct TensorEvaluator<const TensorInflationOp<Strides, ArgType>, Device>
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_impl(op.expression(), device), m_strides(op.strides())
   {
     m_dimensions = m_impl.dimensions();
@@ -137,11 +137,11 @@ struct TensorEvaluator<const TensorInflationOp<Strides, ArgType>, Device>
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
     m_impl.evalSubExprsIfNeeded(NULL);
     return true;
   }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
   }
 
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorLayoutSwap.h b/unsupported/Eigen/CXX11/src/Tensor/TensorLayoutSwap.h
index f159db1b9..80106c1a0 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorLayoutSwap.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorLayoutSwap.h
@@ -113,7 +113,7 @@ struct TensorEvaluator<const TensorLayoutSwapOp<ArgType>, Device>
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_impl(op.expression(), device)
   {
     for(int i = 0; i < NumDims; ++i) {
@@ -136,10 +136,10 @@ struct TensorEvaluator<const TensorLayoutSwapOp<ArgType>, Device>
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
     return m_impl.evalSubExprsIfNeeded(data);
   }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
   }
 
@@ -191,7 +191,7 @@ template<typename ArgType, typename Device>
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
     : Base(op, device)
   { }
 
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorMorphing.h b/unsupported/Eigen/CXX11/src/Tensor/TensorMorphing.h
index ef79c8567..ea97cf185 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorMorphing.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorMorphing.h
@@ -142,7 +142,7 @@ struct TensorEvaluator<const TensorReshapingOp<NewDimensions, ArgType>, Device>
           TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_impl(op.expression(), device), m_dimensions(op.dimensions())
   {
     // The total size of the reshaped tensor must be equal to the total size
@@ -154,16 +154,16 @@ struct TensorEvaluator<const TensorReshapingOp<NewDimensions, ArgType>, Device>
 
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
+  EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
       EvaluatorPointerType data, EvalSubExprsCallback done) {
     m_impl.evalSubExprsIfNeededAsync(data, std::move(done));
   }
 #endif
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
     return m_impl.evalSubExprsIfNeeded(data);
   }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
   }
 
@@ -255,7 +255,7 @@ template<typename NewDimensions, typename ArgType, typename Device>
     RawAccess         = TensorEvaluator<ArgType, Device>::RawAccess
   };
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
     : Base(op, device)
   { }
 
@@ -443,7 +443,7 @@ struct TensorEvaluator<const TensorSlicingOp<StartIndices, Sizes, ArgType>, Devi
       TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_impl(op.expression(), device), m_device(device), m_dimensions(op.sizes()), m_offsets(op.startIndices())
   {
     for (Index i = 0; i < internal::array_size<Dimensions>::value; ++i) {
@@ -498,7 +498,7 @@ struct TensorEvaluator<const TensorSlicingOp<StartIndices, Sizes, ArgType>, Devi
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
     m_impl.evalSubExprsIfNeeded(NULL);
     if (!NumTraits<typename internal::remove_const<Scalar>::type>::RequireInitialization
         && data && m_impl.data()) {
@@ -534,13 +534,13 @@ struct TensorEvaluator<const TensorSlicingOp<StartIndices, Sizes, ArgType>, Devi
 
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
+  EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
       EvaluatorPointerType /*data*/, EvalSubExprsCallback done) {
     m_impl.evalSubExprsIfNeededAsync(nullptr, [done](bool) { done(true); });
   }
 #endif  // EIGEN_USE_THREADS
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
   }
 
@@ -738,7 +738,7 @@ struct TensorEvaluator<TensorSlicingOp<StartIndices, Sizes, ArgType>, Device>
   typedef internal::TensorBlockScratchAllocator<Device> TensorBlockScratch;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
     : Base(op, device)
     { }
 
@@ -906,7 +906,7 @@ struct TensorEvaluator<const TensorStridingSlicingOp<StartIndices, StopIndices,
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_impl(op.expression(), device),
         m_device(device),
         m_strides(op.strides())
@@ -992,12 +992,12 @@ struct TensorEvaluator<const TensorStridingSlicingOp<StartIndices, StopIndices,
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
     m_impl.evalSubExprsIfNeeded(NULL);
     return true;
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
   }
 
@@ -1088,7 +1088,7 @@ struct TensorEvaluator<TensorStridingSlicingOp<StartIndices, StopIndices, Stride
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
     : Base(op, device)
     { }
 
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorPadding.h b/unsupported/Eigen/CXX11/src/Tensor/TensorPadding.h
index 561666c6f..ee44382cf 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorPadding.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorPadding.h
@@ -116,7 +116,7 @@ struct TensorEvaluator<const TensorPaddingOp<PaddingDimensions, ArgType>, Device
       TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_impl(op.expression(), device), m_padding(op.padding()), m_paddingValue(op.padding_value()), m_device(device)
   {
     // The padding op doesn't change the rank of the tensor. Directly padding a scalar would lead
@@ -151,20 +151,20 @@ struct TensorEvaluator<const TensorPaddingOp<PaddingDimensions, ArgType>, Device
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
     m_impl.evalSubExprsIfNeeded(NULL);
     return true;
   }
 
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
+  EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
       EvaluatorPointerType, EvalSubExprsCallback done) {
     m_impl.evalSubExprsIfNeededAsync(nullptr, [done](bool) { done(true); });
   }
 #endif  // EIGEN_USE_THREADS
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
   }
 
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorPatch.h b/unsupported/Eigen/CXX11/src/Tensor/TensorPatch.h
index 64a436e50..413d25dd4 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorPatch.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorPatch.h
@@ -107,7 +107,7 @@ struct TensorEvaluator<const TensorPatchOp<PatchDim, ArgType>, Device>
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_impl(op.expression(), device)
   {
     Index num_patches = 1;
@@ -152,12 +152,12 @@ struct TensorEvaluator<const TensorPatchOp<PatchDim, ArgType>, Device>
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
     m_impl.evalSubExprsIfNeeded(NULL);
     return true;
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
   }
 
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h b/unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h
index 0a65591e6..583f46256 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h
@@ -549,7 +549,7 @@ struct TensorReductionEvaluatorBase<const TensorReductionOp<Op, Dims, ArgType, M
   static const bool PreservingInnerMostDims = internal::preserve_inner_most_dims<Dims, NumInputDims, Layout>::value;
   static const bool RunningFullReduction = (NumOutputDims==0);
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorReductionEvaluatorBase(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorReductionEvaluatorBase(const XprType& op, const Device& device)
       : m_impl(op.expression(), device), m_reducer(op.reducer()), m_result(NULL), m_device(device)
   {
     EIGEN_STATIC_ASSERT((NumInputDims >= NumReducedDims), YOU_MADE_A_PROGRAMMING_MISTAKE);
@@ -631,13 +631,6 @@ struct TensorReductionEvaluatorBase<const TensorReductionOp<Op, Dims, ArgType, M
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
   EIGEN_STRONG_INLINE
-#if !defined(EIGEN_HIPCC)
-  // Marking this as EIGEN_DEVICE_FUNC for HIPCC requires also doing the same
-  // for all the functions being called within here, which then leads to
-  // proliferation of EIGEN_DEVICE_FUNC markings, one of which will eventually
-  // result in an NVCC error
-  EIGEN_DEVICE_FUNC
-#endif
   bool evalSubExprsIfNeededCommon(EvaluatorPointerType data) {
     // Use the FullReducer if possible.
     if ((RunningFullReduction && RunningOnSycl) ||(RunningFullReduction &&
@@ -746,9 +739,6 @@ struct TensorReductionEvaluatorBase<const TensorReductionOp<Op, Dims, ArgType, M
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
   EIGEN_STRONG_INLINE
-#if !defined(EIGEN_HIPCC)
-      EIGEN_DEVICE_FUNC
-#endif
       void
       evalSubExprsIfNeededAsync(EvaluatorPointerType data,
                                 EvalSubExprsCallback done) {
@@ -759,19 +749,12 @@ struct TensorReductionEvaluatorBase<const TensorReductionOp<Op, Dims, ArgType, M
 #endif
 
   EIGEN_STRONG_INLINE
-#if !defined(EIGEN_HIPCC)
-  // Marking this as EIGEN_DEVICE_FUNC for HIPCC requires also doing the same
-  // for all the functions being called within here, which then leads to
-  // proliferation of EIGEN_DEVICE_FUNC markings, one of which will eventually
-  // result in an NVCC error
-  EIGEN_DEVICE_FUNC
-#endif
   bool evalSubExprsIfNeeded(EvaluatorPointerType data) {
     m_impl.evalSubExprsIfNeeded(NULL);
     return evalSubExprsIfNeededCommon(data);
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
     if (m_result) {
       m_device.deallocate_temp(m_result);
@@ -987,7 +970,7 @@ template<typename Op, typename Dims, typename ArgType, template <class> class Ma
 struct TensorEvaluator<const TensorReductionOp<Op, Dims, ArgType, MakePointer_>, Device>
 : public TensorReductionEvaluatorBase<const TensorReductionOp<Op, Dims, ArgType, MakePointer_>, Device> {
   typedef TensorReductionEvaluatorBase<const TensorReductionOp<Op, Dims, ArgType, MakePointer_>, Device> Base;
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const typename Base::XprType& op, const Device& device) : Base(op, device){}
+  EIGEN_STRONG_INLINE TensorEvaluator(const typename Base::XprType& op, const Device& device) : Base(op, device){}
 };
 
 
@@ -996,7 +979,7 @@ struct TensorEvaluator<const TensorReductionOp<Op, Dims, ArgType, MakePointer_>,
 : public TensorReductionEvaluatorBase<const TensorReductionOp<Op, Dims, ArgType, MakePointer_>, Eigen::SyclDevice> {
 
   typedef TensorReductionEvaluatorBase<const TensorReductionOp<Op, Dims, ArgType, MakePointer_>, Eigen::SyclDevice> Base;
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const typename Base::XprType& op, const Eigen::SyclDevice& device) : Base(op, device){}
+  EIGEN_STRONG_INLINE TensorEvaluator(const typename Base::XprType& op, const Eigen::SyclDevice& device) : Base(op, device){}
   // The coeff function in the base the recursive method which is not an standard layout and cannot be used in the SYCL kernel
   //Therefore the coeff function should be overridden by for SYCL kernel
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename Base::CoeffReturnType coeff(typename Base::Index index) const {
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorRef.h b/unsupported/Eigen/CXX11/src/Tensor/TensorRef.h
index 030d19844..a27d3646d 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorRef.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorRef.h
@@ -388,17 +388,17 @@ struct TensorEvaluator<const TensorRef<Derived>, Device>
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const TensorRef<Derived>& m, const Device&)
+  EIGEN_STRONG_INLINE TensorEvaluator(const TensorRef<Derived>& m, const Device&)
       : m_ref(m)
   { }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_ref.dimensions(); }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
     return true;
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() { }
+  EIGEN_STRONG_INLINE void cleanup() { }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType coeff(Index index) const {
     return m_ref.coeff(index);
@@ -439,7 +439,7 @@ struct TensorEvaluator<TensorRef<Derived>, Device> : public TensorEvaluator<cons
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(TensorRef<Derived>& m, const Device& d) : Base(m, d)
+  EIGEN_STRONG_INLINE TensorEvaluator(TensorRef<Derived>& m, const Device& d) : Base(m, d)
   { }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar& coeffRef(Index index) {
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorReverse.h b/unsupported/Eigen/CXX11/src/Tensor/TensorReverse.h
index 3b1fca59b..586ce68ab 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorReverse.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorReverse.h
@@ -121,8 +121,7 @@ struct TensorEvaluator<const TensorReverseOp<ReverseDimensions, ArgType>, Device
       TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op,
-                                                        const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_impl(op.expression(), device),
         m_reverse(op.reverse()),
         m_device(device)
@@ -150,20 +149,20 @@ struct TensorEvaluator<const TensorReverseOp<ReverseDimensions, ArgType>, Device
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   const Dimensions& dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType) {
     m_impl.evalSubExprsIfNeeded(NULL);
     return true;
   }
 
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
+  EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
       EvaluatorPointerType, EvalSubExprsCallback done) {
     m_impl.evalSubExprsIfNeededAsync(nullptr, [done](bool) { done(true); });
   }
 #endif  // EIGEN_USE_THREADS
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
   }
 
@@ -426,8 +425,7 @@ struct TensorEvaluator<TensorReverseOp<ReverseDimensions, ArgType>, Device>
     CoordAccess = false,  // to be implemented
     RawAccess = false
   };
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op,
-                                                        const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : Base(op, device) {}
 
   typedef typename XprType::Scalar Scalar;
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorScan.h b/unsupported/Eigen/CXX11/src/Tensor/TensorScan.h
index a06c4a9f3..beae854dd 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorScan.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorScan.h
@@ -402,8 +402,7 @@ struct TensorEvaluator<const TensorScanOp<Op, ArgType>, Device> {
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op,
-                                                        const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_impl(op.expression(), device),
         m_device(device),
         m_exclusive(op.exclusive()),
@@ -498,7 +497,7 @@ struct TensorEvaluator<const TensorScanOp<Op, ArgType>, Device> {
     return TensorOpCost(sizeof(CoeffReturnType), 0, 0);
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     if (m_output) {
       m_device.deallocate_temp(m_output);
       m_output = NULL;
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorShuffling.h b/unsupported/Eigen/CXX11/src/Tensor/TensorShuffling.h
index e6fed3d0b..0999815d7 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorShuffling.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorShuffling.h
@@ -118,8 +118,7 @@ struct TensorEvaluator<const TensorShufflingOp<Shuffle, ArgType>, Device>
       TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op,
-                                                        const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_device(device),
         m_impl(op.expression(), device)
   {
@@ -163,20 +162,20 @@ struct TensorEvaluator<const TensorShufflingOp<Shuffle, ArgType>, Device>
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
     m_impl.evalSubExprsIfNeeded(NULL);
     return true;
   }
 
 #ifdef EIGEN_USE_THREADS
   template <typename EvalSubExprsCallback>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
+  EIGEN_STRONG_INLINE void evalSubExprsIfNeededAsync(
       EvaluatorPointerType, EvalSubExprsCallback done) {
     m_impl.evalSubExprsIfNeededAsync(nullptr, [done](bool) { done(true); });
   }
 #endif  // EIGEN_USE_THREADS
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
   }
 
@@ -384,7 +383,7 @@ struct TensorEvaluator<TensorShufflingOp<Shuffle, ArgType>, Device>
   typedef internal::TensorBlockDescriptor<NumDims, Index> TensorBlockDesc;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : Base(op, device)
   { }
 
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorStriding.h b/unsupported/Eigen/CXX11/src/Tensor/TensorStriding.h
index 64bf3f139..2f62a668f 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorStriding.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorStriding.h
@@ -109,7 +109,7 @@ struct TensorEvaluator<const TensorStridingOp<Strides, ArgType>, Device>
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_impl(op.expression(), device)
   {
     m_dimensions = m_impl.dimensions();
@@ -142,11 +142,11 @@ struct TensorEvaluator<const TensorStridingOp<Strides, ArgType>, Device>
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType/*data*/) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType/*data*/) {
     m_impl.evalSubExprsIfNeeded(NULL);
     return true;
   }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
   }
 
@@ -277,7 +277,7 @@ struct TensorEvaluator<TensorStridingOp<Strides, ArgType>, Device>
     RawAccess = false
   };
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : Base(op, device) { }
 
   typedef typename XprType::Index Index;
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorTrace.h b/unsupported/Eigen/CXX11/src/Tensor/TensorTrace.h
index 24d22c189..926ecdd38 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorTrace.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorTrace.h
@@ -108,7 +108,7 @@ struct TensorEvaluator<const TensorTraceOp<Dims, ArgType>, Device>
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
     : m_impl(op.expression(), device), m_traceDim(1), m_device(device)
   {
 
@@ -211,12 +211,12 @@ struct TensorEvaluator<const TensorTraceOp<Dims, ArgType>, Device>
     return m_dimensions;
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
     m_impl.evalSubExprsIfNeeded(NULL);
     return true;
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
   }
 
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorVolumePatch.h b/unsupported/Eigen/CXX11/src/Tensor/TensorVolumePatch.h
index 81bed57f3..0beb9ff09 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorVolumePatch.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorVolumePatch.h
@@ -194,7 +194,7 @@ struct TensorEvaluator<const TensorVolumePatchOp<Planes, Rows, Cols, ArgType>, D
   typedef internal::TensorBlockNotImplemented TensorBlock;
   //===--------------------------------------------------------------------===//
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device) :
+  EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device) :
  m_impl(op.expression(), device)
   {
     EIGEN_STATIC_ASSERT((NumDims >= 5), YOU_MADE_A_PROGRAMMING_MISTAKE);
@@ -352,12 +352,12 @@ struct TensorEvaluator<const TensorVolumePatchOp<Planes, Rows, Cols, ArgType>, D
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
+  EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
     m_impl.evalSubExprsIfNeeded(NULL);
     return true;
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+  EIGEN_STRONG_INLINE void cleanup() {
     m_impl.cleanup();
   }
 
@@ -518,21 +518,21 @@ struct TensorEvaluator<const TensorVolumePatchOp<Planes, Rows, Cols, ArgType>, D
   const TensorEvaluator<ArgType, Device>& impl() const { return m_impl; }
 
 
-  Index planePaddingTop() const { return m_planePaddingTop; }
-  Index rowPaddingTop() const { return m_rowPaddingTop; }
-  Index colPaddingLeft() const { return m_colPaddingLeft; }
-  Index outputPlanes() const { return m_outputPlanes; }
-  Index outputRows() const { return m_outputRows; }
-  Index outputCols() const { return m_outputCols; }
-  Index userPlaneStride() const { return m_plane_strides; }
-  Index userRowStride() const { return m_row_strides; }
-  Index userColStride() const { return m_col_strides; }
-  Index userInPlaneStride() const { return m_in_plane_strides; }
-  Index userInRowStride() const { return m_in_row_strides; }
-  Index userInColStride() const { return m_in_col_strides; }
-  Index planeInflateStride() const { return m_plane_inflate_strides; }
-  Index rowInflateStride() const { return m_row_inflate_strides; }
-  Index colInflateStride() const { return m_col_inflate_strides; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index planePaddingTop() const { return m_planePaddingTop; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index rowPaddingTop() const { return m_rowPaddingTop; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index colPaddingLeft() const { return m_colPaddingLeft; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index outputPlanes() const { return m_outputPlanes; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index outputRows() const { return m_outputRows; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index outputCols() const { return m_outputCols; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index userPlaneStride() const { return m_plane_strides; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index userRowStride() const { return m_row_strides; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index userColStride() const { return m_col_strides; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index userInPlaneStride() const { return m_in_plane_strides; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index userInRowStride() const { return m_in_row_strides; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index userInColStride() const { return m_in_col_strides; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index planeInflateStride() const { return m_plane_inflate_strides; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index rowInflateStride() const { return m_row_inflate_strides; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index colInflateStride() const { return m_col_inflate_strides; }
 
 #ifdef EIGEN_USE_SYCL
   // binding placeholder accessors to a command group handler for SYCL
-- 
cgit v1.2.3