From b24fe22b1a4518f27ca064d496bfdb6c96d973ab Mon Sep 17 00:00:00 2001 From: Benoit Steiner Date: Wed, 3 Sep 2014 11:38:13 -0700 Subject: Improved the performance of the tensor convolution code by a factor of about 4. --- unsupported/Eigen/CXX11/src/Tensor/TensorEvalTo.h | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) (limited to 'unsupported/Eigen/CXX11/src/Tensor/TensorEvalTo.h') diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorEvalTo.h b/unsupported/Eigen/CXX11/src/Tensor/TensorEvalTo.h index db716a80e..587cbd5ca 100644 --- a/unsupported/Eigen/CXX11/src/Tensor/TensorEvalTo.h +++ b/unsupported/Eigen/CXX11/src/Tensor/TensorEvalTo.h @@ -108,8 +108,9 @@ struct TensorEvaluator, Device> EIGEN_DEVICE_FUNC const Dimensions& dimensions() const { return m_impl.dimensions(); } - EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeeded() { - m_impl.evalSubExprsIfNeeded(); + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(Scalar*) { + m_impl.evalSubExprsIfNeeded(NULL); + return true; } EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalScalar(Index i) { @@ -134,6 +135,8 @@ struct TensorEvaluator, Device> return internal::ploadt(m_buffer + index); } + Scalar* data() const { return NULL; } + private: TensorEvaluator m_impl; const Device& m_device; -- cgit v1.2.3