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
|
/* Copyright 2016 The TensorFlow Authors. All Rights Reserved.
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.
==============================================================================*/
#include "tensorflow/core/common_runtime/optimization_registry.h"
#include "tensorflow/core/graph/node_builder.h"
namespace tensorflow {
namespace {
Tensor make_zeros(const DataType& dtype, const TensorShapeProto& shape) {
Tensor tensor(dtype, TensorShape(shape));
// Conveniently, all numeric data types have 0x0 == zero. Otherwise we would
// need a giant switch statement here.
memset(const_cast<char*>(tensor.tensor_data().data()), 0,
tensor.tensor_data().size());
return tensor;
}
// Replaces occurrences of the "AccumulateNV2" stub operator with a graph of
// lower-level ops. The graph is equivalent (modulo certain corner cases)
// to the semantics of the original accumulate_n() Python op in math_ops.py.
// Implementing the op with a rewrite allows this new variant of accumulate_n
// to be differentiable.
//
// The binary code that generates AccumulateNV2 stub ops is located in a
// dynamic library built out of tensorflow/contrib/framework. Ideally, this
// class would also be in contrib, but calls to REGISTER_OPTIMIZATION() from
// third-party libraries aren't currently supported.
class AccumulateNV2RemovePass : public GraphOptimizationPass {
public:
Status Run(const GraphOptimizationPassOptions& options) override {
// TODO(freiss.oss@gmail.com): Substantial shared code with
// ParallelConcatRemovePass::Run(). Consider refactoring if someone makes
// a third similar rewrite.
if (options.graph == nullptr) {
// TODO(apassos) returning OK feels weird here as we can't do anything
// without a graph, but some tests require this.
return Status::OK();
}
Graph* g = options.graph->get();
if (g == nullptr) {
return errors::Internal(
"AccumulateNV2 removal should happen before partitioning and a "
"graph should be available.");
}
// Build up a todo list of ops to replace, *then* modify the graph
gtl::InlinedVector<Node*, 2> matches;
for (Node* n : g->op_nodes()) {
if (n->type_string() == "AccumulateNV2") {
matches.push_back(n);
}
}
for (Node* n : matches) {
TF_RETURN_IF_ERROR(rewriteNode(n, g));
}
return Status::OK();
}
Status rewriteNode(Node* n, Graph* g) {
AttrSlice n_attrs = n->attrs();
auto base_make_node = [n, g, &n_attrs](const string& op,
const string& name) {
NodeBuilder node_builder(name, op);
// The pieces of AccumulateNV2 should all be on the same node.
node_builder.Device(n->requested_device());
string colo;
if (GetNodeAttr(n_attrs, kColocationAttrName, &colo).ok()) {
node_builder.Attr(kColocationAttrName, colo);
}
return node_builder;
};
auto make_node = [n, g, &n_attrs, &base_make_node](string op) {
return base_make_node(
op, g->NewName(strings::StrCat(n->name(), "/Internal")));
};
DataType dtype;
TF_RETURN_IF_ERROR(GetNodeAttr(n_attrs, "T", &dtype));
TensorShapeProto shape;
TF_RETURN_IF_ERROR(GetNodeAttr(n_attrs, "shape", &shape));
std::vector<const Edge*> data_edges, control_edges;
for (const Edge* input_edge : n->in_edges()) {
if (input_edge->IsControlEdge()) {
control_edges.push_back(input_edge);
} else {
data_edges.push_back(input_edge);
}
}
// Create the following ops to replace the AccumulateNV2 placeholder:
Node* create_accumulator = nullptr; // TemporaryVariable op
Node* initial_val = nullptr; // Const op
Node* initialize_accumulator = nullptr; // Assign op
std::vector<Node*> add_values_to_accumulator; // AssignAdd ops
Node* clean_up_accumulator = nullptr; // DestroyTemporaryVariable
const string accumulator_name =
strings::StrCat(n->name(), "/Internal/Accumulator");
TensorShapeProto variable_shape;
variable_shape.add_dim()->set_size(0);
TF_RETURN_IF_ERROR(make_node("TemporaryVariable")
.Attr("shape", variable_shape)
.Attr("dtype", dtype)
.Attr("var_name", accumulator_name)
.Finalize(g, &create_accumulator));
if (PartialTensorShape(shape).IsFullyDefined()) {
// For fully defined shapes make a constant zero tensor.
TF_RETURN_IF_ERROR(make_node("Const")
.Attr("value", make_zeros(dtype, shape))
.Attr("dtype", dtype)
.Finalize(g, &initial_val));
} else {
// For partial shapes make a Fill operation to make a zero tensor with the
// shape of the first input.
Node* shape_node;
TF_RETURN_IF_ERROR(
make_node("Shape")
.Input(data_edges[0]->src(), data_edges[0]->src_output())
.Finalize(g, &shape_node));
Node* zero;
TF_RETURN_IF_ERROR(
make_node("Const")
.Attr("value", make_zeros(dtype, TensorShapeProto()))
.Attr("dtype", dtype)
.Finalize(g, &zero));
TF_RETURN_IF_ERROR(make_node("Fill")
.Input(shape_node)
.Input(zero)
.Finalize(g, &initial_val));
}
TF_RETURN_IF_ERROR(make_node("Assign")
.Attr("T", dtype)
.Input(create_accumulator) // ref: Ref(T)
.Input(initial_val) // value: T
.Attr("validate_shape", false)
.Finalize(g, &initialize_accumulator));
for (int i = 0; i < data_edges.size(); ++i) {
Node* assignAdd;
TF_RETURN_IF_ERROR(make_node("AssignAdd")
.Attr("T", dtype)
.Attr("use_locking", true)
.Input(initialize_accumulator) // ref: Ref(T)
.Input(data_edges[i]->src(),
data_edges[i]->src_output()) // value: T
.Finalize(g, &assignAdd));
add_values_to_accumulator.push_back(assignAdd);
}
// Note that we use the original placeholder op's name here
TF_RETURN_IF_ERROR(base_make_node("DestroyTemporaryVariable", n->name())
.Attr("T", dtype)
.Attr("var_name", accumulator_name)
.Input(initialize_accumulator)
.Finalize(g, &clean_up_accumulator));
// Add edges to the graph to ensure that operations occur in the right
// order:
// 1. Do anything that had a control edge to the AccumulateNV2 placeholder
// 2. Initialize accumulator
// 3. Add input values to accumulator (already handled by data edges
// added above)
// 4. Reclaim the buffer that held the accumulator
// 5. Do anything that depended on the AccumulateNV2 placeholder
for (const Edge* control_edge : control_edges) {
g->AddControlEdge(control_edge->src(), initialize_accumulator);
}
for (Node* assign_add : add_values_to_accumulator) {
g->AddControlEdge(assign_add, clean_up_accumulator);
}
for (const Edge* out_edge : n->out_edges()) {
if (out_edge->IsControlEdge()) {
g->AddControlEdge(clean_up_accumulator, out_edge->dst());
} else {
g->AddEdge(clean_up_accumulator, 0, out_edge->dst(),
out_edge->dst_input());
}
}
// Remove the original AccumulateNV2 placeholder op.
// This removal modifies the op and must happen after we have finished
// using its incoming/outgoing edge sets.
g->RemoveNode(n);
return Status::OK();
}
};
REGISTER_OPTIMIZATION(OptimizationPassRegistry::PRE_PLACEMENT, 0,
AccumulateNV2RemovePass);
} // namespace
} // namespace tensorflow
|
