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

Tensor block evaluation V2 support for unary/binary/broadcsting

Browse Source
This commit is contained in:
Eugene Zhulenev
2019-09-24 12:52:45 -07:00
parent efd9867ff0
commit ef9dfee7bd
40 changed files with 1162 additions and 169 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
unsupported/Eigen/CXX11/src/Tensor/TensorArgMax.h
View File

@@ -89,12 +89,17 @@ struct TensorEvaluator<const TensorIndexTupleOp<ArgType>, Device>
IsAligned = /*TensorEvaluator<ArgType, Device>::IsAligned*/ false,
PacketAccess = /*TensorEvaluator<ArgType, Device>::PacketAccess*/ false,
BlockAccess = false,
BlockAccessV2 = false,
PreferBlockAccess = false,
Layout = TensorEvaluator<ArgType, Device>::Layout,
CoordAccess = false, // to be implemented
RawAccess = false
};
//===- Tensor block evaluation strategy (see TensorBlock.h) -------------===//
typedef internal::TensorBlockNotImplemented TensorBlockV2;
//===--------------------------------------------------------------------===//
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
: m_impl(op.expression(), device) { }
@@ -226,12 +231,17 @@ struct TensorEvaluator<const TensorTupleReducerOp<ReduceOp, Dims, ArgType>, Devi
IsAligned = /*TensorEvaluator<ArgType, Device>::IsAligned*/ false,
PacketAccess = /*TensorEvaluator<ArgType, Device>::PacketAccess*/ false,
BlockAccess = false,
BlockAccessV2 = false,
PreferBlockAccess = false,
Layout = TensorEvaluator<const TensorReductionOp<ReduceOp, Dims, const TensorIndexTupleOp<ArgType> >, Device>::Layout,
CoordAccess = false, // to be implemented
RawAccess = false
};
//===- Tensor block evaluation strategy (see TensorBlock.h) -------------===//
typedef internal::TensorBlockNotImplemented TensorBlockV2;
//===--------------------------------------------------------------------===//
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
: m_orig_impl(op.expression(), device),
m_impl(op.expression().index_tuples().reduce(op.reduce_dims(), op.reduce_op()), device),