devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Eigen cost model part 1. This implements a basic recursive framework to estimate the cost of evaluating tensor expressions.

Browse Source
This commit is contained in:
Rasmus Munk Larsen
2016-04-14 13:57:35 -07:00
parent 3551dea887
commit 235e83aba6
28 changed files with 695 additions and 185 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -89,6 +89,12 @@ template<typename LeftArgType, typename RightArgType, typename Device>
struct TensorEvaluator<const TensorAssignOp<LeftArgType, RightArgType>, Device>
{
typedef TensorAssignOp<LeftArgType, RightArgType> XprType;
typedef typename XprType::Index Index;
typedef typename XprType::Scalar Scalar;
typedef typename XprType::CoeffReturnType CoeffReturnType;
typedef typename PacketType<CoeffReturnType, Device>::type PacketReturnType;
typedef typename TensorEvaluator<RightArgType, Device>::Dimensions Dimensions;
static const int PacketSize = internal::unpacket_traits<PacketReturnType>::size;
enum {
IsAligned = TensorEvaluator<LeftArgType, Device>::IsAligned & TensorEvaluator<RightArgType, Device>::IsAligned,
@@ -104,12 +110,6 @@ struct TensorEvaluator<const TensorAssignOp<LeftArgType, RightArgType>, Device>
EIGEN_STATIC_ASSERT((static_cast<int>(TensorEvaluator<LeftArgType, Device>::Layout) == static_cast<int>(TensorEvaluator<RightArgType, Device>::Layout)), YOU_MADE_A_PROGRAMMING_MISTAKE);
}
typedef typename XprType::Index Index;
typedef typename XprType::Scalar Scalar;
typedef typename XprType::CoeffReturnType CoeffReturnType;
typedef typename PacketType<CoeffReturnType, Device>::type PacketReturnType;
typedef typename TensorEvaluator<RightArgType, Device>::Dimensions Dimensions;
EIGEN_DEVICE_FUNC const Dimensions& dimensions() const
{
// The dimensions of the lhs and the rhs tensors should be equal to prevent
@@ -150,6 +150,19 @@ struct TensorEvaluator<const TensorAssignOp<LeftArgType, RightArgType>, Device>
return m_leftImpl.template packet<LoadMode>(index);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost
costPerCoeff(bool vectorized) const {
// We assume that evalPacket or evalScalar is called to perform the
// assignment and account for the cost of the write here, but reduce left
// cost by one load because we are using m_leftImpl.coeffRef.
TensorOpCost left = m_leftImpl.costPerCoeff(vectorized);
return m_rightImpl.costPerCoeff(vectorized) +
TensorOpCost(
numext::maxi(0.0, left.bytes_loaded() - sizeof(CoeffReturnType)),
left.bytes_stored(), left.compute_cycles()) +
TensorOpCost(0, sizeof(CoeffReturnType), 0, vectorized, PacketSize);
}
EIGEN_DEVICE_FUNC CoeffReturnType* data() const { return m_leftImpl.data(); }
private: