1 /* Copyright 2015 The TensorFlow Authors. All Rights Reserved.
2
3 Licensed under the Apache License, Version 2.0 (the "License");
4 you may not use this file except in compliance with the License.
5 You may obtain a copy of the License at
6
7 http://www.apache.org/licenses/LICENSE-2.0
8
9 Unless required by applicable law or agreed to in writing, software
10 distributed under the License is distributed on an "AS IS" BASIS,
11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 See the License for the specific language governing permissions and
13 limitations under the License.
14 ==============================================================================*/
15
16 #ifndef TENSORFLOW_CORE_KERNELS_TYPED_QUEUE_H_
17 #define TENSORFLOW_CORE_KERNELS_TYPED_QUEUE_H_
18
19 #include <deque>
20 #include <queue>
21 #include <vector>
22
23 #include "tensorflow/core/framework/op_kernel.h"
24 #include "tensorflow/core/kernels/queue_base.h"
25 #include "tensorflow/core/platform/mutex.h"
26
27 namespace tensorflow {
28
29 // TypedQueue builds on QueueBase, with backing class (SubQueue)
30 // known and stored within. Shared methods that need to have access
31 // to the backed data sit in this class.
32 template <typename SubQueue>
33 class TypedQueue : public QueueBase {
34 public:
35 TypedQueue(const int32 capacity, const DataTypeVector& component_dtypes,
36 const std::vector<TensorShape>& component_shapes,
37 const string& name);
38
39 virtual Status Initialize(); // Must be called before any other method.
40
41 int64 MemoryUsed() const override;
42
43 protected:
44 std::vector<SubQueue> queues_ TF_GUARDED_BY(mu_);
45 }; // class TypedQueue
46
47 template <typename SubQueue>
TypedQueue(int32 capacity,const DataTypeVector & component_dtypes,const std::vector<TensorShape> & component_shapes,const string & name)48 TypedQueue<SubQueue>::TypedQueue(
49 int32 capacity, const DataTypeVector& component_dtypes,
50 const std::vector<TensorShape>& component_shapes, const string& name)
51 : QueueBase(capacity, component_dtypes, component_shapes, name) {}
52
53 template <typename SubQueue>
Initialize()54 Status TypedQueue<SubQueue>::Initialize() {
55 if (component_dtypes_.empty()) {
56 return errors::InvalidArgument("Empty component types for queue ", name_);
57 }
58 if (!component_shapes_.empty() &&
59 component_dtypes_.size() != component_shapes_.size()) {
60 return errors::InvalidArgument(
61 "Different number of component types. ",
62 "Types: ", DataTypeSliceString(component_dtypes_),
63 ", Shapes: ", ShapeListString(component_shapes_));
64 }
65
66 mutex_lock lock(mu_);
67 queues_.reserve(num_components());
68 for (int i = 0; i < num_components(); ++i) {
69 queues_.push_back(SubQueue());
70 }
71 return Status::OK();
72 }
73
74 namespace {
75
76 template <typename SubQueue>
SizeOf(const SubQueue & sq)77 int64 SizeOf(const SubQueue& sq) {
78 static_assert(sizeof(SubQueue) != sizeof(SubQueue), "SubQueue size unknown.");
79 return 0;
80 }
81
82 template <>
SizeOf(const std::deque<PersistentTensor> & sq)83 int64 SizeOf(const std::deque<PersistentTensor>& sq) {
84 if (sq.empty()) {
85 return 0;
86 }
87 return sq.size() * sq.front().AllocatedBytes();
88 }
89
90 template <>
SizeOf(const std::vector<PersistentTensor> & sq)91 int64 SizeOf(const std::vector<PersistentTensor>& sq) {
92 if (sq.empty()) {
93 return 0;
94 }
95 return sq.size() * sq.front().AllocatedBytes();
96 }
97
98 using TensorPair = std::pair<int64, PersistentTensor>;
99
100 template <typename U, typename V>
SizeOf(const std::priority_queue<TensorPair,U,V> & sq)101 int64 SizeOf(const std::priority_queue<TensorPair, U, V>& sq) {
102 if (sq.empty()) {
103 return 0;
104 }
105 return sq.size() * (sizeof(TensorPair) + sq.top().second.AllocatedBytes());
106 }
107
108 } // namespace
109
110 template <typename SubQueue>
MemoryUsed()111 int64 TypedQueue<SubQueue>::MemoryUsed() const {
112 int memory_size = 0;
113 mutex_lock l(mu_);
114 for (const auto& sq : queues_) {
115 memory_size += SizeOf(sq);
116 }
117 return memory_size;
118 }
119
120 } // namespace tensorflow
121
122 #endif // TENSORFLOW_CORE_KERNELS_TYPED_QUEUE_H_
123