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 #include "tensorflow/core/util/saved_tensor_slice_util.h" 17 18 #include "tensorflow/core/lib/core/errors.h" 19 #include "tensorflow/core/lib/strings/ordered_code.h" 20 #include "tensorflow/core/lib/strings/str_util.h" 21 22 namespace tensorflow { 23 24 namespace checkpoint { 25 26 const char kSavedTensorSlicesKey[] = ""; 27 28 string EncodeTensorNameSlice(const string& name, const TensorSlice& slice) { 29 string buffer; 30 // All the tensor slice keys will start with a 0 31 tensorflow::strings::OrderedCode::WriteNumIncreasing(&buffer, 0); 32 tensorflow::strings::OrderedCode::WriteString(&buffer, name); 33 tensorflow::strings::OrderedCode::WriteNumIncreasing(&buffer, slice.dims()); 34 for (int d = 0; d < slice.dims(); ++d) { 35 // A trivial extent (meaning we take EVERYTHING) will default to -1 for both 36 // start and end. These will be properly parsed. 37 tensorflow::strings::OrderedCode::WriteSignedNumIncreasing(&buffer, 38 slice.start(d)); 39 tensorflow::strings::OrderedCode::WriteSignedNumIncreasing(&buffer, 40 slice.length(d)); 41 } 42 return buffer; 43 } 44 45 Status DecodeTensorNameSlice(const string& code, string* name, 46 tensorflow::TensorSlice* slice) { 47 StringPiece src(code); 48 uint64 x; 49 if (!tensorflow::strings::OrderedCode::ReadNumIncreasing(&src, &x)) { 50 return errors::Internal("Failed to parse the leading number: src = ", src); 51 } 52 if (x != 0) { 53 return errors::Internal( 54 "The leading number should always be 0 for any valid key: src = ", src); 55 } 56 if (!tensorflow::strings::OrderedCode::ReadString(&src, name)) { 57 return errors::Internal("Failed to parse the tensor name: src = ", src); 58 } 59 if (!tensorflow::strings::OrderedCode::ReadNumIncreasing(&src, &x)) { 60 return errors::Internal("Failed to parse the tensor rank: src = ", src); 61 } 62 if (x == 0) { 63 return errors::Internal("Expecting positive rank of the tensor, got ", x, 64 ", src = ", src); 65 } 66 if (x >= kint32max) { 67 return errors::Internal("Too many elements ", x); 68 } 69 slice->SetFullSlice(x); 70 for (int d = 0; d < static_cast<int32>(x); ++d) { 71 // We expected 2x integers 72 int64 start, length; 73 if (!tensorflow::strings::OrderedCode::ReadSignedNumIncreasing(&src, 74 &start)) { 75 return errors::Internal("Failed to parse start: src = ", src); 76 } 77 if (!tensorflow::strings::OrderedCode::ReadSignedNumIncreasing(&src, 78 &length)) { 79 return errors::Internal("Failed to parse length: src = ", src); 80 } 81 if (length >= 0) { 82 // a non-trivial extent 83 slice->set_start(d, start); 84 slice->set_length(d, length); 85 } 86 } 87 return Status::OK(); 88 } 89 90 Status ParseShapeAndSlice(const string& shape_and_slice, TensorShape* shape, 91 TensorSlice* slice, TensorShape* shape_slice) { 92 CHECK(!shape_and_slice.empty()); 93 // Syntax: dim0 dim1 dim2 ... <slice string> 94 // Where slice string is defined in core/framework/tensor_slice.h 95 std::vector<string> splits = str_util::Split(shape_and_slice, ' '); 96 97 // Must have at least 2 strings. 98 if (splits.size() < 2) { 99 return errors::InvalidArgument( 100 "Need least two elements in shape_and_slice specification: ", 101 shape_and_slice); 102 } 103 104 // The last split is the slice specification. 105 slice->Clear(); 106 auto status = slice->Parse(splits.back(), slice); 107 if (!status.ok()) return status; 108 109 // The first n-1 are the shape specification. 110 splits.pop_back(); 111 shape->Clear(); 112 for (const auto& s : splits) { 113 int64 dim; 114 if (!strings::safe_strto64(s, &dim)) { 115 return errors::InvalidArgument( 116 "Non numerical dimension in shape_and_slice: ", shape_and_slice); 117 } 118 shape->AddDim(dim); 119 } 120 121 // The specified slice must be compatible with the specified shape. 122 return slice->SliceTensorShape(*shape, shape_slice); 123 } 124 125 } // namespace checkpoint 126 127 } // namespace tensorflow 128
