GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-06-22 08:52:15 +08:00
Code Issues Packages Projects Releases Wiki Activity

Added suppor for in place evaluation to simple tensor expressions.

Browse Source 
Use mempy to speedup tensor copies whenever possible.
This commit is contained in:
Benoit Steiner 2014-08-13 08:26:44 -07:00
parent 439feca139
commit b1892ab14d
2 changed files with 50 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
unsupported/Eigen/CXX11/src/Tensor/TensorAssign.h
View File

@ -102,6 +102,7 @@ struct TensorEvaluator<const TensorAssignOp<LeftArgType, RightArgType>, Device>
{ } { }
typedef typename XprType::Index Index; typedef typename XprType::Index Index;
typedef typename XprType::Scalar Scalar;
typedef typename XprType::CoeffReturnType CoeffReturnType; typedef typename XprType::CoeffReturnType CoeffReturnType;
typedef typename XprType::PacketReturnType PacketReturnType; typedef typename XprType::PacketReturnType PacketReturnType;
typedef typename TensorEvaluator<RightArgType, Device>::Dimensions Dimensions; typedef typename TensorEvaluator<RightArgType, Device>::Dimensions Dimensions;
@ -112,9 +113,14 @@ struct TensorEvaluator<const TensorAssignOp<LeftArgType, RightArgType>, Device>
return m_rightImpl.dimensions(); return m_rightImpl.dimensions();
} }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeeded() { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(Scalar*) {
m_leftImpl.evalSubExprsIfNeeded(); eigen_assert(internal::dimensions_match(m_leftImpl.dimensions(), m_rightImpl.dimensions()));
m_rightImpl.evalSubExprsIfNeeded(); m_leftImpl.evalSubExprsIfNeeded(NULL);
// If the lhs provides raw access to its storage area (i.e. if m_leftImpl.data() returns a non
// null value), attempt to evaluate the rhs expression in place. Returns true iff in place
// evaluation isn't supported and the caller still needs to manually assign the values generated
// by the rhs to the lhs.
return m_rightImpl.evalSubExprsIfNeeded(m_leftImpl.data());
} }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
m_leftImpl.cleanup(); m_leftImpl.cleanup();

55
unsupported/Eigen/CXX11/src/Tensor/TensorEvaluator.h
View File

@ -39,13 +39,20 @@ struct TensorEvaluator
PacketAccess = Derived::PacketAccess, PacketAccess = Derived::PacketAccess,
}; };
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const Derived& m, const Device&) EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const Derived& m, const Device& device)
: m_data(const_cast<Scalar*>(m.data())), m_dims(m.dimensions()) : m_data(const_cast<Scalar*>(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 const Dimensions& dimensions() const { return m_dims; }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeeded() { } EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(Scalar* dest) {
if (dest) {
m_device.memcpy((void*)dest, m_data, sizeof(Scalar) * m_dims.TotalSize());
return false;
}
return true;
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() { } EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() { }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType coeff(Index index) const { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType coeff(Index index) const {
@ -70,9 +77,12 @@ struct TensorEvaluator
return internal::pstoret<Scalar, Packet, StoreMode>(m_data + index, x); return internal::pstoret<Scalar, Packet, StoreMode>(m_data + index, x);
} }
Scalar* data() const { return m_data; }
protected: protected:
Scalar* m_data; Scalar* m_data;
Dimensions m_dims; Dimensions m_dims;
const Device& m_device;
}; };
@ -98,7 +108,7 @@ struct TensorEvaluator<const Derived, Device>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dims; } EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dims; }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeeded() { } EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(Scalar*) { return true; }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() { } EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() { }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType coeff(Index index) const { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType coeff(Index index) const {
@ -112,6 +122,8 @@ struct TensorEvaluator<const Derived, Device>
return internal::ploadt<Packet, LoadMode>(m_data + index); return internal::ploadt<Packet, LoadMode>(m_data + index);
} }
const Scalar* data() const { return m_data; }
protected: protected:
const Scalar* m_data; const Scalar* m_data;
Dimensions m_dims; Dimensions m_dims;
@ -138,13 +150,14 @@ struct TensorEvaluator<const TensorCwiseNullaryOp<NullaryOp, ArgType>, Device>
{ } { }
typedef typename XprType::Index Index; typedef typename XprType::Index Index;
typedef typename XprType::Scalar Scalar;
typedef typename XprType::CoeffReturnType CoeffReturnType; typedef typename XprType::CoeffReturnType CoeffReturnType;
typedef typename XprType::PacketReturnType PacketReturnType; typedef typename XprType::PacketReturnType PacketReturnType;
typedef typename TensorEvaluator<ArgType, Device>::Dimensions Dimensions; typedef typename TensorEvaluator<ArgType, Device>::Dimensions Dimensions;
EIGEN_DEVICE_FUNC const Dimensions& dimensions() const { return m_argImpl.dimensions(); } EIGEN_DEVICE_FUNC const Dimensions& dimensions() const { return m_argImpl.dimensions(); }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeeded() { } EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(Scalar*) { return true; }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() { } EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() { }
EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index index) const EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index index) const
@ -158,6 +171,8 @@ struct TensorEvaluator<const TensorCwiseNullaryOp<NullaryOp, ArgType>, Device>
return m_functor.packetOp(index); return m_functor.packetOp(index);
} }
Scalar* data() const { return NULL; }
private: private:
const NullaryOp m_functor; const NullaryOp m_functor;
TensorEvaluator<ArgType, Device> m_argImpl; TensorEvaluator<ArgType, Device> m_argImpl;
@ -183,14 +198,16 @@ struct TensorEvaluator<const TensorCwiseUnaryOp<UnaryOp, ArgType>, Device>
{ } { }
typedef typename XprType::Index Index; typedef typename XprType::Index Index;
typedef typename XprType::Scalar Scalar;
typedef typename XprType::CoeffReturnType CoeffReturnType; typedef typename XprType::CoeffReturnType CoeffReturnType;
typedef typename XprType::PacketReturnType PacketReturnType; typedef typename XprType::PacketReturnType PacketReturnType;
typedef typename TensorEvaluator<ArgType, Device>::Dimensions Dimensions; typedef typename TensorEvaluator<ArgType, Device>::Dimensions Dimensions;
EIGEN_DEVICE_FUNC const Dimensions& dimensions() const { return m_argImpl.dimensions(); } EIGEN_DEVICE_FUNC const Dimensions& dimensions() const { return m_argImpl.dimensions(); }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeeded() { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(Scalar*) {
m_argImpl.evalSubExprsIfNeeded(); m_argImpl.evalSubExprsIfNeeded(NULL);
return true;
} }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
m_argImpl.cleanup(); m_argImpl.cleanup();
@ -207,6 +224,8 @@ struct TensorEvaluator<const TensorCwiseUnaryOp<UnaryOp, ArgType>, Device>
return m_functor.packetOp(m_argImpl.template packet<LoadMode>(index)); return m_functor.packetOp(m_argImpl.template packet<LoadMode>(index));
} }
Scalar* data() const { return NULL; }
private: private:
const UnaryOp m_functor; const UnaryOp m_functor;
TensorEvaluator<ArgType, Device> m_argImpl; TensorEvaluator<ArgType, Device> m_argImpl;
@ -233,6 +252,7 @@ struct TensorEvaluator<const TensorCwiseBinaryOp<BinaryOp, LeftArgType, RightArg
{ } { }
typedef typename XprType::Index Index; typedef typename XprType::Index Index;
typedef typename XprType::Scalar Scalar;
typedef typename XprType::CoeffReturnType CoeffReturnType; typedef typename XprType::CoeffReturnType CoeffReturnType;
typedef typename XprType::PacketReturnType PacketReturnType; typedef typename XprType::PacketReturnType PacketReturnType;
typedef typename TensorEvaluator<LeftArgType, Device>::Dimensions Dimensions; typedef typename TensorEvaluator<LeftArgType, Device>::Dimensions Dimensions;
@ -243,9 +263,10 @@ struct TensorEvaluator<const TensorCwiseBinaryOp<BinaryOp, LeftArgType, RightArg
return m_leftImpl.dimensions(); return m_leftImpl.dimensions();
} }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeeded() { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(Scalar*) {
m_leftImpl.evalSubExprsIfNeeded(); m_leftImpl.evalSubExprsIfNeeded(NULL);
m_rightImpl.evalSubExprsIfNeeded(); m_rightImpl.evalSubExprsIfNeeded(NULL);
return true;
} }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
m_leftImpl.cleanup(); m_leftImpl.cleanup();
@ -262,6 +283,8 @@ struct TensorEvaluator<const TensorCwiseBinaryOp<BinaryOp, LeftArgType, RightArg
return m_functor.packetOp(m_leftImpl.template packet<LoadMode>(index), m_rightImpl.template packet<LoadMode>(index)); return m_functor.packetOp(m_leftImpl.template packet<LoadMode>(index), m_rightImpl.template packet<LoadMode>(index));
} }
Scalar* data() const { return NULL; }
private: private:
const BinaryOp m_functor; const BinaryOp m_functor;
TensorEvaluator<LeftArgType, Device> m_leftImpl; TensorEvaluator<LeftArgType, Device> m_leftImpl;
@ -289,6 +312,7 @@ struct TensorEvaluator<const TensorSelectOp<IfArgType, ThenArgType, ElseArgType>
{ } { }
typedef typename XprType::Index Index; typedef typename XprType::Index Index;
typedef typename XprType::Scalar Scalar;
typedef typename XprType::CoeffReturnType CoeffReturnType; typedef typename XprType::CoeffReturnType CoeffReturnType;
typedef typename XprType::PacketReturnType PacketReturnType; typedef typename XprType::PacketReturnType PacketReturnType;
typedef typename TensorEvaluator<IfArgType, Device>::Dimensions Dimensions; typedef typename TensorEvaluator<IfArgType, Device>::Dimensions Dimensions;
@ -299,10 +323,11 @@ struct TensorEvaluator<const TensorSelectOp<IfArgType, ThenArgType, ElseArgType>
return m_condImpl.dimensions(); return m_condImpl.dimensions();
} }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeeded() { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(Scalar*) {
m_condImpl.evalSubExprsIfNeeded(); m_condImpl.evalSubExprsIfNeeded(NULL);
m_thenImpl.evalSubExprsIfNeeded(); m_thenImpl.evalSubExprsIfNeeded(NULL);
m_elseImpl.evalSubExprsIfNeeded(); m_elseImpl.evalSubExprsIfNeeded(NULL);
return true;
} }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
m_condImpl.cleanup(); m_condImpl.cleanup();
@ -327,6 +352,8 @@ struct TensorEvaluator<const TensorSelectOp<IfArgType, ThenArgType, ElseArgType>
m_elseImpl.template packet<LoadMode>(index)); m_elseImpl.template packet<LoadMode>(index));
} }
Scalar* data() const { return NULL; }
private: private:
TensorEvaluator<IfArgType, Device> m_condImpl; TensorEvaluator<IfArgType, Device> m_condImpl;
TensorEvaluator<ThenArgType, Device> m_thenImpl; TensorEvaluator<ThenArgType, Device> m_thenImpl;