-
Notifications
You must be signed in to change notification settings - Fork 1.2k
/
CAddTable.scala
191 lines (169 loc) · 5.92 KB
/
CAddTable.scala
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
/*
* Copyright 2016 The BigDL Authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package com.intel.analytics.bigdl.nn
import javax.print.attribute.standard.MediaSize.Other
import com.intel.analytics.bigdl.nn.abstractnn.AbstractModule
import com.intel.analytics.bigdl.tensor.Tensor
import com.intel.analytics.bigdl.tensor.TensorNumericMath.TensorNumeric
import com.intel.analytics.bigdl.utils.{T, Table}
import com.intel.analytics.bigdl.utils.serializer.{DeserializeContext, ModuleSerializable}
import org.apache.hadoop.yarn.webapp.hamlet.HamletSpec.VAR
import scala.reflect._
/**
* Merge the input tensors in the input table by element wise adding them together. The input table
* is actually an array of tensor with same size.
* @param inplace reuse the input memory
* @param ev numeric operator
* @tparam T Numeric type. Only support float/double now
*/
@SerialVersionUID(7959261460060075605L)
class CAddTable[T: ClassTag, D: ClassTag](val inplace: Boolean = false)(
implicit ev: TensorNumeric[T], ev2: TensorNumeric[D])
extends AbstractModule[Table, Tensor[D], T] with MklInt8Convertible {
output = Tensor[D]()
@transient
private var bufferSumInput: Tensor[D] = null
@transient
private var bufferSumOutput: Tensor[D] = null
private def canExpand(inputSize: Array[Int], targetSize: Array[Int]): Boolean = {
var d = inputSize.length - 1
val diff = targetSize.length - inputSize.length
while(d >= 0) {
if (inputSize(d) != 1 && inputSize(d) != targetSize(d + diff)) {
return false
}
d -= 1
}
return true
}
private def sumAlongDims(tensor: Tensor[D], other: Tensor[D]): Tensor[D] = {
val diff = other.nDimension() - tensor.nDimension()
val size = tensor.size()
var target: Tensor[D] = other
if (bufferSumOutput == null) bufferSumOutput = Tensor[D]()
if (bufferSumInput == null) bufferSumInput = Tensor[D]()
var i = 0
while (i < other.nDimension()) {
if (i < diff) {
bufferSumOutput.sum(target, i + 1)
target = bufferSumInput.resizeAs(bufferSumOutput).copy(bufferSumOutput)
} else if (size(i - diff) == 1) {
bufferSumOutput.sum(target, i + 1)
target = bufferSumInput.resizeAs(bufferSumOutput).copy(bufferSumOutput)
}
i += 1
}
target
}
override def updateOutput(input: Table): Tensor[D] = {
var scalar = ev2.zero
var hasTensor = false
var hasScalar = false
var initTensor = false
var i = 1
while (i <= input.length()) {
val curTensor = input[Tensor[D]](i)
if (curTensor.isScalar) {
scalar = ev2.plus(scalar, curTensor.value())
hasScalar = true
} else if (curTensor.isTensor) {
if (initTensor) {
output = output.add(curTensor)
} else {
if (inplace) {
output.set(curTensor)
} else {
output.resizeAs(curTensor).copy(curTensor)
}
initTensor = true
}
hasTensor = true
}
i += 1
}
if (hasTensor && hasScalar) {
output.add(scalar)
} else if (hasScalar) {
if (inplace) {
output.set(input[Tensor[D]](1)).setValue(scalar)
} else {
output.resizeAs(input[Tensor[D]](1)).setValue(scalar)
}
}
output
}
override def updateGradInput(input: Table, gradOutput: Tensor[D]) : Table = {
var i = 1
var sum = ev2.zero
var calculateSum = false
while (i <= input.length()) {
if (i > gradInput.length) gradInput.insert(i, Tensor[T]().resizeAs(input(1)))
if (inplace) {
require(input[Tensor[D]](1).isSameSizeAs(gradOutput), "cannot use inplace for broadcast")
gradInput[Tensor[D]](i).set(gradOutput)
} else {
if (input[Tensor[D]](i).isSameSizeAs(gradOutput)) {
gradInput[Tensor[D]](i).resizeAs(gradOutput).copy(gradOutput)
} else if (canExpand(input[Tensor[D]](i).size(), gradOutput.size())) {
gradInput[Tensor[D]](i).resizeAs(input[Tensor[D]](i)).copy(
sumAlongDims(input[Tensor[D]](i), gradOutput))
} else {
require(input[Tensor[D]](i).isScalar, "Only support scalar broadcast backward now")
if (!calculateSum) {
sum = gradOutput.sum()
calculateSum = true
}
gradInput[Tensor[D]](i).resizeAs(input[Tensor[D]](i)).setValue(sum)
}
}
i += 1
}
i = input.length + 1
while (i <= gradInput.length) {
gradInput.remove(i)
}
gradInput
}
override def clearState(): this.type = {
if (!inplace) {
super.clearState()
}
this
}
override def getClassTagNumerics() : (Array[ClassTag[_]], Array[TensorNumeric[_]]) = {
(Array[ClassTag[_]](scala.reflect.classTag[T], scala.reflect.classTag[D]),
Array[TensorNumeric[_]](ev, ev2))
}
}
object CAddTable extends ModuleSerializable {
def apply[T: ClassTag](
inplace: Boolean = false)(implicit ev: TensorNumeric[T]) : CAddTable[T, T] = {
new CAddTable[T, T](inplace)
}
override def getTypes(context: DeserializeContext): (Array[ClassTag[_]],
Array[TensorNumeric[_]]) = {
var (tags, numerics) = super.getTypes(context)
val defaultTag = tags(0)
val defaultNumeric = numerics(0)
if (tags.size < 2) {
val extendedTags = Array[ClassTag[_]](defaultTag, defaultTag)
val extendNumerics = Array[TensorNumeric[_]](defaultNumeric, defaultNumeric)
(extendedTags, extendNumerics)
} else {
(tags, numerics)
}
}
}