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

51
unsupported/Eigen/CXX11/src/Tensor/TensorBroadcasting.h
View File

@@ -101,6 +101,9 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
typedef DSizes<Index, NumDims> Dimensions;
typedef typename XprType::Scalar Scalar;
typedef typename TensorEvaluator<ArgType, Device>::Dimensions InputDimensions;
typedef typename XprType::CoeffReturnType CoeffReturnType;
typedef typename PacketType<CoeffReturnType, Device>::type PacketReturnType;
static const int PacketSize = internal::unpacket_traits<PacketReturnType>::size;
enum {
IsAligned = false,
@@ -140,9 +143,6 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
}
}
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*/) {
@@ -247,9 +247,8 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
template<int LoadMode>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PacketReturnType packetColMajor(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());
const Index originalIndex = index;
@@ -284,12 +283,12 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
// Todo: this could be extended to the second dimension if we're not
// broadcasting alongside the first dimension, and so on.
if (innermostLoc + packetSize <= m_impl.dimensions()[0]) {
if (innermostLoc + PacketSize <= m_impl.dimensions()[0]) {
return m_impl.template packet<Unaligned>(inputIndex);
} 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(inputIndex);
for (int i = 1; i < packetSize; ++i) {
for (int i = 1; i < PacketSize; ++i) {
values[i] = coeffColMajor(originalIndex+i);
}
PacketReturnType rslt = internal::pload<PacketReturnType>(values);
@@ -300,9 +299,8 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
template<int LoadMode>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PacketReturnType packetRowMajor(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());
const Index originalIndex = index;
@@ -337,12 +335,12 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
// Todo: this could be extended to the second dimension if we're not
// broadcasting alongside the first dimension, and so on.
if (innermostLoc + packetSize <= m_impl.dimensions()[NumDims-1]) {
if (innermostLoc + PacketSize <= m_impl.dimensions()[NumDims-1]) {
return m_impl.template packet<Unaligned>(inputIndex);
} 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(inputIndex);
for (int i = 1; i < packetSize; ++i) {
for (int i = 1; i < PacketSize; ++i) {
values[i] = coeffRowMajor(originalIndex+i);
}
PacketReturnType rslt = internal::pload<PacketReturnType>(values);
@@ -350,6 +348,29 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
}
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost
costPerCoeff(bool vectorized) const {
double compute_cost = TensorOpCost::AddCost<Index>();
if (NumDims > 0) {
for (int i = NumDims - 1; i > 0; --i) {
compute_cost += TensorOpCost::DivCost<Index>();
if (internal::index_statically_eq<Broadcast>()(i, 1)) {
compute_cost +=
TensorOpCost::MulCost<Index>() + TensorOpCost::AddCost<Index>();
} else {
if (!internal::index_statically_eq<InputDimensions>()(i, 1)) {
compute_cost += TensorOpCost::MulCost<Index>() +
TensorOpCost::ModCost<Index>() +
TensorOpCost::AddCost<Index>();
}
}
compute_cost +=
TensorOpCost::MulCost<Index>() + TensorOpCost::AddCost<Index>();
}
}
return m_impl.costPerCoeff(vectorized) +
TensorOpCost(0, 0, compute_cost, vectorized, PacketSize);
}
EIGEN_DEVICE_FUNC Scalar* data() const { return NULL; }