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

47
unsupported/Eigen/CXX11/src/Tensor/TensorStriding.h
View File

@@ -103,6 +103,10 @@ struct TensorEvaluator<const TensorStridingOp<Strides, ArgType>, Device>
typedef typename XprType::Index Index;
static const int NumDims = internal::array_size<typename TensorEvaluator<ArgType, Device>::Dimensions>::value;
typedef DSizes<Index, NumDims> Dimensions;
typedef typename XprType::Scalar Scalar;
typedef typename XprType::CoeffReturnType CoeffReturnType;
typedef typename PacketType<CoeffReturnType, Device>::type PacketReturnType;
static const int PacketSize = internal::unpacket_traits<PacketReturnType>::size;
enum {
IsAligned = /*TensorEvaluator<ArgType, Device>::IsAligned*/false,
@@ -142,10 +146,6 @@ struct TensorEvaluator<const TensorStridingOp<Strides, ArgType>, Device>
}
}
typedef typename XprType::Scalar Scalar;
typedef typename XprType::CoeffReturnType CoeffReturnType;
typedef typename PacketType<CoeffReturnType, Device>::type PacketReturnType;
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(Scalar* /*data*/) {
@@ -164,12 +164,11 @@ struct TensorEvaluator<const TensorStridingOp<Strides, ArgType>, Device>
template<int LoadMode>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PacketReturnType packet(Index index) const
{
const int packetSize = internal::unpacket_traits<PacketReturnType>::size;
EIGEN_STATIC_ASSERT(packetSize > 1, YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index+packetSize-1 < dimensions().TotalSize());
EIGEN_STATIC_ASSERT(PacketSize > 1, YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index+PacketSize-1 < dimensions().TotalSize());
Index inputIndices[] = {0, 0};
Index indices[] = {index, index + packetSize - 1};
Index indices[] = {index, index + PacketSize - 1};
if (static_cast<int>(Layout) == static_cast<int>(ColMajor)) {
for (int i = NumDims - 1; i > 0; --i) {
const Index idx0 = indices[0] / m_outputStrides[i];
@@ -193,15 +192,15 @@ struct TensorEvaluator<const TensorStridingOp<Strides, ArgType>, Device>
inputIndices[0] += indices[0] * m_inputStrides[NumDims-1];
inputIndices[1] += indices[1] * m_inputStrides[NumDims-1];
}
if (inputIndices[1] - inputIndices[0] == packetSize - 1) {
if (inputIndices[1] - inputIndices[0] == PacketSize - 1) {
PacketReturnType rslt = m_impl.template packet<Unaligned>(inputIndices[0]);
return rslt;
}
else {
EIGEN_ALIGN_MAX typename internal::remove_const<CoeffReturnType>::type values[packetSize];
EIGEN_ALIGN_MAX typename internal::remove_const<CoeffReturnType>::type values[PacketSize];
values[0] = m_impl.coeff(inputIndices[0]);
values[packetSize-1] = m_impl.coeff(inputIndices[1]);
for (int i = 1; i < packetSize-1; ++i) {
values[PacketSize-1] = m_impl.coeff(inputIndices[1]);
for (int i = 1; i < PacketSize-1; ++i) {
values[i] = coeff(index+i);
}
PacketReturnType rslt = internal::pload<PacketReturnType>(values);
@@ -209,6 +208,14 @@ 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);
}
EIGEN_DEVICE_FUNC Scalar* data() const { return NULL; }
protected:
@@ -266,6 +273,7 @@ struct TensorEvaluator<TensorStridingOp<Strides, ArgType>, Device>
typedef typename XprType::Scalar Scalar;
typedef typename XprType::CoeffReturnType CoeffReturnType;
typedef typename PacketType<CoeffReturnType, Device>::type PacketReturnType;
static const int PacketSize = internal::unpacket_traits<PacketReturnType>::size;
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar& coeffRef(Index index)
{
@@ -275,12 +283,11 @@ struct TensorEvaluator<TensorStridingOp<Strides, ArgType>, Device>
template <int StoreMode> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void writePacket(Index index, const PacketReturnType& x)
{
const int packetSize = internal::unpacket_traits<PacketReturnType>::size;
EIGEN_STATIC_ASSERT(packetSize > 1, YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index+packetSize-1 < this->dimensions().TotalSize());
EIGEN_STATIC_ASSERT(PacketSize > 1, YOU_MADE_A_PROGRAMMING_MISTAKE)
eigen_assert(index+PacketSize-1 < this->dimensions().TotalSize());
Index inputIndices[] = {0, 0};
Index indices[] = {index, index + packetSize - 1};
Index indices[] = {index, index + PacketSize - 1};
if (static_cast<int>(Layout) == static_cast<int>(ColMajor)) {
for (int i = NumDims - 1; i > 0; --i) {
const Index idx0 = indices[0] / this->m_outputStrides[i];
@@ -304,15 +311,15 @@ struct TensorEvaluator<TensorStridingOp<Strides, ArgType>, Device>
inputIndices[0] += indices[0] * this->m_inputStrides[NumDims-1];
inputIndices[1] += indices[1] * this->m_inputStrides[NumDims-1];
}
if (inputIndices[1] - inputIndices[0] == packetSize - 1) {
if (inputIndices[1] - inputIndices[0] == PacketSize - 1) {
this->m_impl.template writePacket<Unaligned>(inputIndices[0], x);
}
else {
EIGEN_ALIGN_MAX Scalar values[packetSize];
EIGEN_ALIGN_MAX Scalar values[PacketSize];
internal::pstore<Scalar, PacketReturnType>(values, x);
this->m_impl.coeffRef(inputIndices[0]) = values[0];
this->m_impl.coeffRef(inputIndices[1]) = values[packetSize-1];
for (int i = 1; i < packetSize-1; ++i) {
this->m_impl.coeffRef(inputIndices[1]) = values[PacketSize-1];
for (int i = 1; i < PacketSize-1; ++i) {
this->coeffRef(index+i) = values[i];
}
}