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
|
/* Copyright 2017 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 <deque>
#include "tensorflow/core/common_runtime/function.h"
#include "tensorflow/core/framework/partial_tensor_shape.h"
#include "tensorflow/core/framework/tensor.h"
#include "tensorflow/core/kernels/data/captured_function.h"
#include "tensorflow/core/kernels/data/dataset.h"
#include "tensorflow/core/kernels/data/parallel_map_iterator.h"
#include "tensorflow/core/lib/core/error_codes.pb.h"
#include "tensorflow/core/lib/random/random.h"
namespace tensorflow {
namespace {
// See documentation in ../ops/dataset_ops.cc for a high-level
// description of the following op.
class ParallelMapDatasetOp : public UnaryDatasetOpKernel {
public:
explicit ParallelMapDatasetOp(OpKernelConstruction* ctx)
: UnaryDatasetOpKernel(ctx),
graph_def_version_(ctx->graph_def_version()) {
OP_REQUIRES_OK(ctx, ctx->GetAttr("f", &func_));
OP_REQUIRES_OK(ctx, ctx->GetAttr("output_types", &output_types_));
OP_REQUIRES_OK(ctx, ctx->GetAttr("output_shapes", &output_shapes_));
}
protected:
void MakeDataset(OpKernelContext* ctx, DatasetBase* input,
DatasetBase** output) override {
OpInputList inputs;
OP_REQUIRES_OK(ctx, ctx->input_list("other_arguments", &inputs));
std::vector<Tensor> other_arguments;
other_arguments.reserve(inputs.size());
for (const Tensor& t : inputs) {
other_arguments.push_back(t);
}
int32 num_parallel_calls;
OP_REQUIRES_OK(ctx, ParseScalarArgument(ctx, "num_parallel_calls",
&num_parallel_calls));
OP_REQUIRES(ctx, num_parallel_calls > 0,
errors::InvalidArgument(
"num_parallel_calls must be greater than zero."));
std::unique_ptr<CapturedFunction> captured_func;
OP_REQUIRES_OK(ctx, CapturedFunction::Create(
func_, std::move(other_arguments), &captured_func));
*output = new Dataset(ctx, input, func_, num_parallel_calls, output_types_,
output_shapes_, std::move(captured_func));
}
private:
class Dataset : public DatasetBase {
public:
Dataset(OpKernelContext* ctx, const DatasetBase* input,
const NameAttrList& func, int32 num_parallel_calls,
const DataTypeVector& output_types,
const std::vector<PartialTensorShape>& output_shapes,
std::unique_ptr<CapturedFunction> captured_func)
: DatasetBase(DatasetContext(ctx)),
input_(input),
func_(func),
num_parallel_calls_(num_parallel_calls),
output_types_(output_types),
output_shapes_(output_shapes),
captured_func_(std::move(captured_func)) {
input_->Ref();
}
~Dataset() override { input_->Unref(); }
std::unique_ptr<IteratorBase> MakeIteratorInternal(
const string& prefix) const override {
auto init_func = [this](IteratorContext* ctx) {
return captured_func_->Instantiate(ctx);
};
auto map_func = [this](IteratorContext* ctx,
std::vector<Tensor> input_element,
std::vector<Tensor>* result, StatusCallback done) {
captured_func_->RunAsync(ctx, std::move(input_element), result,
std::move(done));
};
return NewParallelMapIterator(
{this, strings::StrCat(prefix, "::ParallelMap")}, input_,
std::move(init_func), std::move(map_func), num_parallel_calls_);
}
const DataTypeVector& output_dtypes() const override {
return output_types_;
}
const std::vector<PartialTensorShape>& output_shapes() const override {
return output_shapes_;
}
string DebugString() const override {
return "ParallelMapDatasetOp::Dataset";
}
protected:
Status AsGraphDefInternal(SerializationContext* ctx,
DatasetGraphDefBuilder* b,
Node** output) const override {
// Input: input_dataset
Node* input_graph_node = nullptr;
TF_RETURN_IF_ERROR(b->AddInputDataset(ctx, input_, &input_graph_node));
// Input: other_arguments
DataTypeVector other_arguments_types;
other_arguments_types.reserve(captured_func_->captured_inputs().size());
std::vector<Node*> other_arguments;
other_arguments.reserve(captured_func_->captured_inputs().size());
for (const Tensor& t : captured_func_->captured_inputs()) {
Node* node;
TF_RETURN_IF_ERROR(b->AddTensor(t, &node));
other_arguments.emplace_back(node);
other_arguments_types.emplace_back(t.dtype());
}
// Input: num_parallel_calls
Node* num_parallel_calls = nullptr;
TF_RETURN_IF_ERROR(
b->AddScalar(num_parallel_calls_, &num_parallel_calls));
// Attr: f
TF_RETURN_IF_ERROR(b->AddFunction(ctx, func_.name()));
AttrValue f;
b->BuildAttrValue(func_, &f);
// Attr: Targuments
AttrValue other_arguments_types_attr;
b->BuildAttrValue(other_arguments_types, &other_arguments_types_attr);
TF_RETURN_IF_ERROR(b->AddDataset(
this,
{std::make_pair(0, input_graph_node),
std::make_pair(2, num_parallel_calls)}, // Single tensor inputs.
{std::make_pair(1, other_arguments)}, // Tensor list inputs.
{std::make_pair("f", f),
std::make_pair("Targuments", other_arguments_types_attr)}, // Attrs
output));
return Status::OK();
}
private:
const DatasetBase* const input_;
const NameAttrList func_;
const int32 num_parallel_calls_;
const DataTypeVector output_types_;
const std::vector<PartialTensorShape> output_shapes_;
const std::unique_ptr<CapturedFunction> captured_func_;
};
const int graph_def_version_;
DataTypeVector output_types_;
std::vector<PartialTensorShape> output_shapes_;
NameAttrList func_;
};
REGISTER_KERNEL_BUILDER(Name("ParallelMapDataset").Device(DEVICE_CPU),
ParallelMapDatasetOp);
} // namespace
} // namespace tensorflow
|
