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_LIB_IO_RECORD_WRITER_H_
17 #define TENSORFLOW_CORE_LIB_IO_RECORD_WRITER_H_
18
19 #include "tensorflow/core/lib/core/coding.h"
20 #include "tensorflow/core/lib/core/status.h"
21 #include "tensorflow/core/lib/core/stringpiece.h"
22 #include "tensorflow/core/lib/hash/crc32c.h"
23 #if !defined(IS_SLIM_BUILD)
24 #include "tensorflow/core/lib/io/zlib_compression_options.h"
25 #include "tensorflow/core/lib/io/zlib_outputbuffer.h"
26 #endif // IS_SLIM_BUILD
27 #include "tensorflow/core/platform/cord.h"
28 #include "tensorflow/core/platform/macros.h"
29 #include "tensorflow/core/platform/types.h"
30
31 namespace tensorflow {
32
33 class WritableFile;
34
35 namespace io {
36
37 class RecordWriterOptions {
38 public:
39 enum CompressionType { NONE = 0, ZLIB_COMPRESSION = 1 };
40 CompressionType compression_type = NONE;
41
42 static RecordWriterOptions CreateRecordWriterOptions(
43 const string& compression_type);
44
45 // Options specific to zlib compression.
46 #if !defined(IS_SLIM_BUILD)
47 tensorflow::io::ZlibCompressionOptions zlib_options;
48 #endif // IS_SLIM_BUILD
49 };
50
51 class RecordWriter {
52 public:
53 // Format of a single record:
54 // uint64 length
55 // uint32 masked crc of length
56 // byte data[length]
57 // uint32 masked crc of data
58 static const size_t kHeaderSize = sizeof(uint64) + sizeof(uint32);
59 static const size_t kFooterSize = sizeof(uint32);
60
61 // Create a writer that will append data to "*dest".
62 // "*dest" must be initially empty.
63 // "*dest" must remain live while this Writer is in use.
64 RecordWriter(WritableFile* dest,
65 const RecordWriterOptions& options = RecordWriterOptions());
66
67 // Calls Close() and logs if an error occurs.
68 //
69 // TODO(jhseu): Require that callers explicitly call Close() and remove the
70 // implicit Close() call in the destructor.
71 ~RecordWriter();
72
73 Status WriteRecord(StringPiece slice);
74
75 #if defined(PLATFORM_GOOGLE)
76 Status WriteRecord(const absl::Cord& data);
77 #endif
78
79 // Flushes any buffered data held by underlying containers of the
80 // RecordWriter to the WritableFile. Does *not* flush the
81 // WritableFile.
82 Status Flush();
83
84 // Writes all output to the file. Does *not* close the WritableFile.
85 //
86 // After calling Close(), any further calls to `WriteRecord()` or `Flush()`
87 // are invalid.
88 Status Close();
89
90 // Utility method to populate TFRecord headers. Populates record-header in
91 // "header[0,kHeaderSize-1]". The record-header is based on data[0, n-1].
92 inline static void PopulateHeader(char* header, const char* data, size_t n);
93
94 // Utility method to populate TFRecord footers. Populates record-footer in
95 // "footer[0,kFooterSize-1]". The record-footer is based on data[0, n-1].
96 inline static void PopulateFooter(char* footer, const char* data, size_t n);
97
98 #if defined(PLATFORM_GOOGLE)
99 inline static void PopulateHeader(char* header, const absl::Cord& data);
100 inline static void PopulateFooter(char* footer, const absl::Cord& data);
101 #endif
102
103 private:
104 WritableFile* dest_;
105 RecordWriterOptions options_;
106
MaskedCrc(const char * data,size_t n)107 inline static uint32 MaskedCrc(const char* data, size_t n) {
108 return crc32c::Mask(crc32c::Value(data, n));
109 }
110
111 #if defined(PLATFORM_GOOGLE)
MaskedCrc(const absl::Cord & data)112 inline static uint32 MaskedCrc(const absl::Cord& data) {
113 return crc32c::Mask(crc32c::Value(data));
114 }
115 #endif
116
117 TF_DISALLOW_COPY_AND_ASSIGN(RecordWriter);
118 };
119
PopulateHeader(char * header,const char * data,size_t n)120 void RecordWriter::PopulateHeader(char* header, const char* data, size_t n) {
121 core::EncodeFixed64(header + 0, n);
122 core::EncodeFixed32(header + sizeof(uint64),
123 MaskedCrc(header, sizeof(uint64)));
124 }
125
PopulateFooter(char * footer,const char * data,size_t n)126 void RecordWriter::PopulateFooter(char* footer, const char* data, size_t n) {
127 core::EncodeFixed32(footer, MaskedCrc(data, n));
128 }
129
130 #if defined(PLATFORM_GOOGLE)
PopulateHeader(char * header,const absl::Cord & data)131 void RecordWriter::PopulateHeader(char* header, const absl::Cord& data) {
132 core::EncodeFixed64(header + 0, data.size());
133 core::EncodeFixed32(header + sizeof(uint64),
134 MaskedCrc(header, sizeof(uint64)));
135 }
136
PopulateFooter(char * footer,const absl::Cord & data)137 void RecordWriter::PopulateFooter(char* footer, const absl::Cord& data) {
138 core::EncodeFixed32(footer, MaskedCrc(data));
139 }
140 #endif
141
142 } // namespace io
143 } // namespace tensorflow
144
145 #endif // TENSORFLOW_CORE_LIB_IO_RECORD_WRITER_H_
146