GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-10-13 00:21:29 +08:00
Code Issues Packages Projects Releases Wiki Activity

Optimized the performance of broadcasting of scalars.

Browse Source
This commit is contained in:
Benoit Steiner 2016-01-06 18:47:45 -08:00
parent ee738321aa
commit 213459d818
1 changed files with 24 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

25
unsupported/Eigen/CXX11/src/Tensor/TensorBroadcasting.h
View File

@ -46,6 +46,21 @@ struct nested<TensorBroadcastingOp<Broadcast, XprType>, 1, typename eval<TensorB
typedef TensorBroadcastingOp<Broadcast, XprType> type; typedef TensorBroadcastingOp<Broadcast, XprType> type;
}; };
template <typename Dims>
struct is_input_scalar {
static const bool value = false;
};
template <>
struct is_input_scalar<Sizes<> > {
static const bool value = true;
};
#ifndef EIGEN_EMULATE_CXX11_META_H
template <typename std::size_t... Indices>
struct is_input_scalar<Sizes<Indices...> > {
static const bool value = (Sizes<Indices...>::total_size == 1);
};
#endif
} // end namespace internal } // end namespace internal
@ -103,7 +118,7 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
// and store the result in a scalar. Instead one should reshape the scalar into a a N-D // and store the result in a scalar. Instead one should reshape the scalar into a a N-D
// tensor with N >= 1 of 1 element first and then broadcast. // tensor with N >= 1 of 1 element first and then broadcast.
EIGEN_STATIC_ASSERT(NumDims > 0, YOU_MADE_A_PROGRAMMING_MISTAKE); EIGEN_STATIC_ASSERT(NumDims > 0, YOU_MADE_A_PROGRAMMING_MISTAKE);
const typename TensorEvaluator<ArgType, Device>::Dimensions& input_dims = m_impl.dimensions(); const InputDimensions& input_dims = m_impl.dimensions();
const Broadcast& broadcast = op.broadcast(); const Broadcast& broadcast = op.broadcast();
for (int i = 0; i < NumDims; ++i) { for (int i = 0; i < NumDims; ++i) {
eigen_assert(input_dims[i] > 0); eigen_assert(input_dims[i] > 0);
@ -143,6 +158,10 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE CoeffReturnType coeff(Index index) const EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE CoeffReturnType coeff(Index index) const
{ {
if (internal::is_input_scalar<typename internal::remove_all<InputDimensions>::type>::value) {
return m_impl.coeff(0);
}
if (static_cast<int>(Layout) == static_cast<int>(ColMajor)) { if (static_cast<int>(Layout) == static_cast<int>(ColMajor)) {
return coeffColMajor(index); return coeffColMajor(index);
} else { } else {
@ -214,6 +233,10 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
template<int LoadMode> template<int LoadMode>
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE PacketReturnType packet(Index index) const EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE PacketReturnType packet(Index index) const
{ {
if (internal::is_input_scalar<typename internal::remove_all<InputDimensions>::type>::value) {
return m_impl.coeff(0);
}
if (static_cast<int>(Layout) == static_cast<int>(ColMajor)) { if (static_cast<int>(Layout) == static_cast<int>(ColMajor)) {
return packetColMajor<LoadMode>(index); return packetColMajor<LoadMode>(index);
} else { } else {