1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
|
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com>
//
// This Source Code Form is subject to the terms of the Mozilla
// Public License v. 2.0. If a copy of the MPL was not distributed
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
#ifndef EIGEN_CXX11_TENSOR_TENSOR_CUSTOM_OP_H
#define EIGEN_CXX11_TENSOR_TENSOR_CUSTOM_OP_H
namespace Eigen {
/** \class TensorCustomUnaryOp
* \ingroup CXX11_Tensor_Module
*
* \brief Tensor custom class.
*
*
*/
namespace internal {
template<typename CustomUnaryFunc, typename XprType>
struct traits<TensorCustomUnaryOp<CustomUnaryFunc, XprType> >
{
typedef typename XprType::Scalar Scalar;
typedef typename XprType::StorageKind StorageKind;
typedef typename XprType::Index Index;
typedef typename XprType::Nested Nested;
typedef typename remove_reference<Nested>::type _Nested;
static const int NumDimensions = traits<XprType>::NumDimensions;
static const int Layout = traits<XprType>::Layout;
};
template<typename CustomUnaryFunc, typename XprType>
struct eval<TensorCustomUnaryOp<CustomUnaryFunc, XprType>, Eigen::Dense>
{
typedef const TensorCustomUnaryOp<CustomUnaryFunc, XprType>& type;
};
template<typename CustomUnaryFunc, typename XprType>
struct nested<TensorCustomUnaryOp<CustomUnaryFunc, XprType> >
{
typedef TensorCustomUnaryOp<CustomUnaryFunc, XprType> type;
};
} // end namespace internal
template<typename CustomUnaryFunc, typename XprType>
class TensorCustomUnaryOp : public TensorBase<TensorCustomUnaryOp<CustomUnaryFunc, XprType>, ReadOnlyAccessors>
{
public:
typedef typename internal::traits<TensorCustomUnaryOp>::Scalar Scalar;
typedef typename Eigen::NumTraits<Scalar>::Real RealScalar;
typedef typename XprType::CoeffReturnType CoeffReturnType;
typedef typename internal::nested<TensorCustomUnaryOp>::type Nested;
typedef typename internal::traits<TensorCustomUnaryOp>::StorageKind StorageKind;
typedef typename internal::traits<TensorCustomUnaryOp>::Index Index;
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorCustomUnaryOp(const XprType& expr, const CustomUnaryFunc& func)
: m_expr(expr), m_func(func) {}
EIGEN_DEVICE_FUNC
const CustomUnaryFunc& func() const { return m_func; }
EIGEN_DEVICE_FUNC
const typename internal::remove_all<typename XprType::Nested>::type&
expression() const { return m_expr; }
protected:
typename XprType::Nested m_expr;
const CustomUnaryFunc m_func;
};
// Eval as rvalue
template<typename CustomUnaryFunc, typename XprType, typename Device>
struct TensorEvaluator<const TensorCustomUnaryOp<CustomUnaryFunc, XprType>, Device>
{
typedef TensorCustomUnaryOp<CustomUnaryFunc, XprType> ArgType;
typedef typename internal::traits<ArgType>::Index Index;
static const int NumDims = internal::traits<ArgType>::NumDimensions;
typedef DSizes<Index, NumDims> Dimensions;
typedef typename internal::remove_const<typename ArgType::Scalar>::type Scalar;
typedef typename internal::remove_const<typename XprType::CoeffReturnType>::type CoeffReturnType;
typedef typename PacketType<CoeffReturnType, Device>::type PacketReturnType;
static const int PacketSize = internal::unpacket_traits<PacketReturnType>::size;
enum {
IsAligned = false,
PacketAccess = (internal::packet_traits<Scalar>::size > 1),
BlockAccess = false,
Layout = TensorEvaluator<XprType, Device>::Layout,
CoordAccess = false, // to be implemented
RawAccess = false
};
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const ArgType& op, const Device& device)
: m_op(op), m_device(device), m_result(NULL)
{
m_dimensions = op.func().dimensions(op.expression());
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(CoeffReturnType* data) {
if (data) {
evalTo(data);
return false;
} else {
m_result = static_cast<CoeffReturnType*>(
m_device.allocate(dimensions().TotalSize() * sizeof(Scalar)));
evalTo(m_result);
return true;
}
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
if (m_result != NULL) {
m_device.deallocate(m_result);
m_result = NULL;
}
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType coeff(Index index) const {
return m_result[index];
}
template<int LoadMode>
EIGEN_DEVICE_FUNC PacketReturnType packet(Index index) const {
return internal::ploadt<PacketReturnType, LoadMode>(m_result + index);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const {
// TODO(rmlarsen): Extend CustomOp API to return its cost estimate.
return TensorOpCost(sizeof(CoeffReturnType), 0, 0, vectorized, PacketSize);
}
EIGEN_DEVICE_FUNC CoeffReturnType* data() const { return m_result; }
#ifdef EIGEN_USE_SYCL
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Device& device() const { return m_device; }
#endif
protected:
EIGEN_DEVICE_FUNC void evalTo(Scalar* data) {
TensorMap<Tensor<CoeffReturnType, NumDims, Layout, Index> > result(
data, m_dimensions);
m_op.func().eval(m_op.expression(), result, m_device);
}
Dimensions m_dimensions;
const ArgType m_op;
const Device& m_device;
CoeffReturnType* m_result;
};
/** \class TensorCustomBinaryOp
* \ingroup CXX11_Tensor_Module
*
* \brief Tensor custom class.
*
*
*/
namespace internal {
template<typename CustomBinaryFunc, typename LhsXprType, typename RhsXprType>
struct traits<TensorCustomBinaryOp<CustomBinaryFunc, LhsXprType, RhsXprType> >
{
typedef typename internal::promote_storage_type<typename LhsXprType::Scalar,
typename RhsXprType::Scalar>::ret Scalar;
typedef typename internal::promote_storage_type<typename LhsXprType::CoeffReturnType,
typename RhsXprType::CoeffReturnType>::ret CoeffReturnType;
typedef typename promote_storage_type<typename traits<LhsXprType>::StorageKind,
typename traits<RhsXprType>::StorageKind>::ret StorageKind;
typedef typename promote_index_type<typename traits<LhsXprType>::Index,
typename traits<RhsXprType>::Index>::type Index;
typedef typename LhsXprType::Nested LhsNested;
typedef typename RhsXprType::Nested RhsNested;
typedef typename remove_reference<LhsNested>::type _LhsNested;
typedef typename remove_reference<RhsNested>::type _RhsNested;
static const int NumDimensions = traits<LhsXprType>::NumDimensions;
static const int Layout = traits<LhsXprType>::Layout;
};
template<typename CustomBinaryFunc, typename LhsXprType, typename RhsXprType>
struct eval<TensorCustomBinaryOp<CustomBinaryFunc, LhsXprType, RhsXprType>, Eigen::Dense>
{
typedef const TensorCustomBinaryOp<CustomBinaryFunc, LhsXprType, RhsXprType>& type;
};
template<typename CustomBinaryFunc, typename LhsXprType, typename RhsXprType>
struct nested<TensorCustomBinaryOp<CustomBinaryFunc, LhsXprType, RhsXprType> >
{
typedef TensorCustomBinaryOp<CustomBinaryFunc, LhsXprType, RhsXprType> type;
};
} // end namespace internal
template<typename CustomBinaryFunc, typename LhsXprType, typename RhsXprType>
class TensorCustomBinaryOp : public TensorBase<TensorCustomBinaryOp<CustomBinaryFunc, LhsXprType, RhsXprType>, ReadOnlyAccessors>
{
public:
typedef typename internal::traits<TensorCustomBinaryOp>::Scalar Scalar;
typedef typename Eigen::NumTraits<Scalar>::Real RealScalar;
typedef typename internal::traits<TensorCustomBinaryOp>::CoeffReturnType CoeffReturnType;
typedef typename internal::nested<TensorCustomBinaryOp>::type Nested;
typedef typename internal::traits<TensorCustomBinaryOp>::StorageKind StorageKind;
typedef typename internal::traits<TensorCustomBinaryOp>::Index Index;
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorCustomBinaryOp(const LhsXprType& lhs, const RhsXprType& rhs, const CustomBinaryFunc& func)
: m_lhs_xpr(lhs), m_rhs_xpr(rhs), m_func(func) {}
EIGEN_DEVICE_FUNC
const CustomBinaryFunc& func() const { return m_func; }
EIGEN_DEVICE_FUNC
const typename internal::remove_all<typename LhsXprType::Nested>::type&
lhsExpression() const { return m_lhs_xpr; }
EIGEN_DEVICE_FUNC
const typename internal::remove_all<typename RhsXprType::Nested>::type&
rhsExpression() const { return m_rhs_xpr; }
protected:
typename LhsXprType::Nested m_lhs_xpr;
typename RhsXprType::Nested m_rhs_xpr;
const CustomBinaryFunc m_func;
};
// Eval as rvalue
template<typename CustomBinaryFunc, typename LhsXprType, typename RhsXprType, typename Device>
struct TensorEvaluator<const TensorCustomBinaryOp<CustomBinaryFunc, LhsXprType, RhsXprType>, Device>
{
typedef TensorCustomBinaryOp<CustomBinaryFunc, LhsXprType, RhsXprType> XprType;
typedef typename internal::traits<XprType>::Index Index;
static const int NumDims = internal::traits<XprType>::NumDimensions;
typedef DSizes<Index, NumDims> Dimensions;
typedef typename XprType::Scalar Scalar;
typedef typename internal::remove_const<typename XprType::CoeffReturnType>::type CoeffReturnType;
typedef typename PacketType<CoeffReturnType, Device>::type PacketReturnType;
static const int PacketSize = internal::unpacket_traits<PacketReturnType>::size;
enum {
IsAligned = false,
PacketAccess = (internal::packet_traits<Scalar>::size > 1),
BlockAccess = false,
Layout = TensorEvaluator<LhsXprType, Device>::Layout,
CoordAccess = false, // to be implemented
RawAccess = false
};
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
: m_op(op), m_device(device), m_result(NULL)
{
m_dimensions = op.func().dimensions(op.lhsExpression(), op.rhsExpression());
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(CoeffReturnType* data) {
if (data) {
evalTo(data);
return false;
} else {
m_result = static_cast<Scalar *>(m_device.allocate(dimensions().TotalSize() * sizeof(Scalar)));
evalTo(m_result);
return true;
}
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
if (m_result != NULL) {
m_device.deallocate(m_result);
m_result = NULL;
}
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType coeff(Index index) const {
return m_result[index];
}
template<int LoadMode>
EIGEN_DEVICE_FUNC PacketReturnType packet(Index index) const {
return internal::ploadt<PacketReturnType, LoadMode>(m_result + index);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const {
// TODO(rmlarsen): Extend CustomOp API to return its cost estimate.
return TensorOpCost(sizeof(CoeffReturnType), 0, 0, vectorized, PacketSize);
}
EIGEN_DEVICE_FUNC CoeffReturnType* data() const { return m_result; }
#ifdef EIGEN_USE_SYCL
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Device& device() const { return m_device; }
#endif
protected:
EIGEN_DEVICE_FUNC void evalTo(Scalar* data) {
TensorMap<Tensor<Scalar, NumDims, Layout> > result(data, m_dimensions);
m_op.func().eval(m_op.lhsExpression(), m_op.rhsExpression(), result, m_device);
}
Dimensions m_dimensions;
const XprType m_op;
const Device& m_device;
CoeffReturnType* m_result;
};
} // end namespace Eigen
#endif // EIGEN_CXX11_TENSOR_TENSOR_CUSTOM_OP_H
|
