diff options
Diffstat (limited to 'unsupported/Eigen/CXX11/src/Tensor/TensorInflation.h')
-rw-r--r-- | unsupported/Eigen/CXX11/src/Tensor/TensorInflation.h | 28 |
1 files changed, 23 insertions, 5 deletions
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorInflation.h b/unsupported/Eigen/CXX11/src/Tensor/TensorInflation.h index f391fb9ee..c5cb61af5 100644 --- a/unsupported/Eigen/CXX11/src/Tensor/TensorInflation.h +++ b/unsupported/Eigen/CXX11/src/Tensor/TensorInflation.h @@ -31,6 +31,7 @@ struct traits<TensorInflationOp<Strides, XprType> > : public traits<XprType> typedef typename remove_reference<Nested>::type _Nested; static const int NumDimensions = XprTraits::NumDimensions; static const int Layout = XprTraits::Layout; + typedef typename XprTraits::PointerType PointerType; }; template<typename Strides, typename XprType> @@ -84,18 +85,25 @@ struct TensorEvaluator<const TensorInflationOp<Strides, ArgType>, Device> typedef typename XprType::Scalar Scalar; typedef typename XprType::CoeffReturnType CoeffReturnType; typedef typename PacketType<CoeffReturnType, Device>::type PacketReturnType; - static const int PacketSize = internal::unpacket_traits<PacketReturnType>::size; + static const int PacketSize = PacketType<CoeffReturnType, Device>::size; + typedef StorageMemory<CoeffReturnType, Device> Storage; + typedef typename Storage::Type EvaluatorPointerType; enum { IsAligned = /*TensorEvaluator<ArgType, Device>::IsAligned*/ false, PacketAccess = TensorEvaluator<ArgType, Device>::PacketAccess, BlockAccess = false, + PreferBlockAccess = false, Layout = TensorEvaluator<ArgType, Device>::Layout, CoordAccess = false, // to be implemented RawAccess = false }; - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device) + //===- Tensor block evaluation strategy (see TensorBlock.h) -------------===// + typedef internal::TensorBlockNotImplemented TensorBlock; + //===--------------------------------------------------------------------===// + + EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device) : m_impl(op.expression(), device), m_strides(op.strides()) { m_dimensions = m_impl.dimensions(); @@ -129,11 +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(Scalar* /*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(); } @@ -144,6 +152,7 @@ struct TensorEvaluator<const TensorInflationOp<Strides, ArgType>, Device> eigen_assert(index < dimensions().TotalSize()); *inputIndex = 0; if (static_cast<int>(Layout) == static_cast<int>(ColMajor)) { + EIGEN_UNROLL_LOOP for (int i = NumDims - 1; i > 0; --i) { const Index idx = index / m_outputStrides[i]; if (idx != idx / m_fastStrides[i] * m_strides[i]) { @@ -158,6 +167,7 @@ struct TensorEvaluator<const TensorInflationOp<Strides, ArgType>, Device> *inputIndex += index / m_strides[0]; return true; } else { + EIGEN_UNROLL_LOOP for (int i = 0; i < NumDims - 1; ++i) { const Index idx = index / m_outputStrides[i]; if (idx != idx / m_fastStrides[i] * m_strides[i]) { @@ -193,6 +203,7 @@ struct TensorEvaluator<const TensorInflationOp<Strides, ArgType>, Device> eigen_assert(index+PacketSize-1 < dimensions().TotalSize()); EIGEN_ALIGN_MAX typename internal::remove_const<CoeffReturnType>::type values[PacketSize]; + EIGEN_UNROLL_LOOP for (int i = 0; i < PacketSize; ++i) { values[i] = coeff(index+i); } @@ -213,7 +224,14 @@ struct TensorEvaluator<const TensorInflationOp<Strides, ArgType>, Device> compute_cost, vectorized, PacketSize); } - EIGEN_DEVICE_FUNC Scalar* data() const { return NULL; } + EIGEN_DEVICE_FUNC EvaluatorPointerType data() const { return NULL; } + +#ifdef EIGEN_USE_SYCL + // binding placeholder accessors to a command group handler for SYCL + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void bind(cl::sycl::handler &cgh) const { + m_impl.bind(cgh); + } +#endif protected: Dimensions m_dimensions; |