mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2024-12-21 07:19:46 +08:00
Code Issues Packages Projects Releases Wiki Activity

Refined the cost of the striding operation.

Browse Source
This commit is contained in:
Benoit Steiner 2016-04-25 09:16:08 -07:00
parent 5c372d19e3
commit fd9401f260
1 changed files with 11 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
unsupported/Eigen/CXX11/src/Tensor/TensorStriding.h
View File

@ -209,11 +209,17 @@ struct TensorEvaluator<const TensorStridingOp<Strides, ArgType>, Device>
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const {
const double compute_cost = NumDims * (2 * TensorOpCost::AddCost<Index>() +
2 * TensorOpCost::MulCost<Index>() +
TensorOpCost::DivCost<Index>());
return m_impl.costPerCoeff(vectorized) +
TensorOpCost(0, 0, compute_cost, false /* vectorized */, PacketSize);
double compute_cost = (NumDims - 1) * (TensorOpCost::AddCost<Index>() +
TensorOpCost::MulCost<Index>() +
TensorOpCost::DivCost<Index>()) +
TensorOpCost::MulCost<Index>();
if (vectorized) {
compute_cost *= 2; // packet() computes two indices
}
const int innerDim = (static_cast<int>(Layout) == static_cast<int>(ColMajor)) ? 0 : (NumDims - 1);
return m_impl.costPerCoeff(vectorized && m_inputStrides[innerDim] == 1) +
// Computation is not vectorized per se, but it is done once per packet.
TensorOpCost(0, 0, compute_cost, vectorized, PacketSize);
}
EIGEN_DEVICE_FUNC Scalar* data() const { return NULL; }