diff options
| author |  Rasmus Munk Larsen <rmlarsen@google.com> | 2019-08-02 11:18:13 -0700 |
|---|---|---|
| committer | Rasmus Munk Larsen <rmlarsen@google.com> | 2019-08-02 11:18:13 -0700 |
| commit | e2999d4c388f3bc556a556befdcb51b1139e9d92 (patch) | |
| tree | dd0d03cd89ce8dbc2b74b22d741a21b9e6bc0752 /unsupported/Eigen/CXX11/src/Tensor/TensorEvaluator.h | |
| parent | f22b7283a3822524d03e4d93e6144bc1c9dd13a5 (diff) | |
Fix performance regressions due to https://bitbucket.org/eigen/eigen/pull-requests/662.
The change caused the device struct to be copied for each expression evaluation, and caused, e.g., a 10% regression in the TensorFlow multinomial op on GPU:
Benchmark Time(ns) CPU(ns) Iterations
----------------------------------------------------------------------
BM_Multinomial_gpu_1_100000_4 128173 231326 2922 1.610G items/s
VS
Benchmark Time(ns) CPU(ns) Iterations
----------------------------------------------------------------------
BM_Multinomial_gpu_1_100000_4 146683 246914 2719 1.509G items/s
Diffstat (limited to 'unsupported/Eigen/CXX11/src/Tensor/TensorEvaluator.h')
| -rw-r--r-- | unsupported/Eigen/CXX11/src/Tensor/TensorEvaluator.h | 8 |
1 files changed, 4 insertions, 4 deletions
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorEvaluator.h b/unsupported/Eigen/CXX11/src/Tensor/TensorEvaluator.h index 1d48b5eed..a3a79d4e9 100644 --- a/unsupported/Eigen/CXX11/src/Tensor/TensorEvaluator.h +++ b/unsupported/Eigen/CXX11/src/Tensor/TensorEvaluator.h @@ -164,7 +164,7 @@ struct TensorEvaluator protected: EvaluatorPointerType m_data; Dimensions m_dims; - const Device m_device; + const Device EIGEN_DEVICE_REF m_device; }; namespace { @@ -302,7 +302,7 @@ struct TensorEvaluator<const Derived, Device> protected: EvaluatorPointerType m_data; Dimensions m_dims; - const Device m_device; + const Device EIGEN_DEVICE_REF m_device; }; @@ -480,7 +480,7 @@ struct TensorEvaluator<const TensorCwiseUnaryOp<UnaryOp, ArgType>, Device> private: - const Device m_device; + const Device EIGEN_DEVICE_REF m_device; const UnaryOp m_functor; TensorEvaluator<ArgType, Device> m_argImpl; }; @@ -603,7 +603,7 @@ struct TensorEvaluator<const TensorCwiseBinaryOp<BinaryOp, LeftArgType, RightArg } #endif private: - const Device m_device; + const Device EIGEN_DEVICE_REF m_device; const BinaryOp m_functor; TensorEvaluator<LeftArgType, Device> m_leftImpl; TensorEvaluator<RightArgType, Device> m_rightImpl; |