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/framework/attr_value.pb.h" |
17 | #include "tensorflow/core/framework/common_shape_fns.h" |
18 | #include "tensorflow/core/framework/op.h" |
19 | #include "tensorflow/core/framework/shape_inference.h" |
20 | #include "tensorflow/core/framework/tensor_shape.pb.h" |
21 | #include "tensorflow/core/lib/core/errors.h" |
22 | |
23 | namespace tensorflow { |
24 | |
25 | REGISTER_SYSTEM_OP("_Arg" ) |
26 | .Output("output: T" ) |
27 | .Attr("T: type" ) |
28 | .Attr("index: int >= 0" ) |
29 | .SetIsStateful() |
30 | .SetShapeFn([](shape_inference::InferenceContext* context) { |
31 | const AttrValue* dtype_attr = context->GetAttr("T" ); |
32 | if (!dtype_attr) { |
33 | return errors::InvalidArgument( |
34 | "_Arg node does not have attribute \"T\"" ); |
35 | } |
36 | |
37 | const AttrValue* shape_attr = context->GetAttr("_output_shapes" ); |
38 | if (shape_attr && shape_attr->has_list()) { |
39 | if (shape_attr->list().shape().empty()) { |
40 | return errors::InvalidArgument( |
41 | "Invalid \"_output_shapes\" attribute value for _Arg node: " , |
42 | shape_attr->DebugString()); |
43 | } |
44 | const TensorShapeProto& shape_proto = shape_attr->list().shape(0); |
45 | shape_inference::ShapeHandle shape_handle; |
46 | TF_RETURN_IF_ERROR( |
47 | context->MakeShapeFromShapeProto(shape_proto, &shape_handle)); |
48 | context->set_output(0, shape_handle); |
49 | } else { |
50 | context->set_output(0, context->UnknownShape()); |
51 | } |
52 | |
53 | if (dtype_attr->type() != DT_RESOURCE) { |
54 | return OkStatus(); |
55 | } |
56 | |
57 | // If the argument is for a resource type, then also try to infer the |
58 | // type of the tensor store in the resource type. |
59 | dtype_attr = context->GetAttr("_handle_dtypes" ); |
60 | shape_attr = context->GetAttr("_handle_shapes" ); |
61 | // If either the shape or type attribute is not set then simply return |
62 | // with unknown output set above. |
63 | if (!dtype_attr || !shape_attr) { |
64 | return OkStatus(); |
65 | } |
66 | |
67 | if (dtype_attr->list().type().empty()) { |
68 | return errors::InvalidArgument( |
69 | "Invalid \"_handle_dtypes\" attribute value for _Arg node: " , |
70 | dtype_attr->DebugString()); |
71 | } |
72 | if (shape_attr->list().shape().empty()) { |
73 | return errors::InvalidArgument( |
74 | "Invalid \"_handle_shapes\" attribute value for _Arg node: " , |
75 | shape_attr->DebugString()); |
76 | } |
77 | DataType dtype = dtype_attr->list().type(0); |
78 | const TensorShapeProto& shape_proto = shape_attr->list().shape(0); |
79 | shape_inference::ShapeHandle shape_handle; |
80 | TF_RETURN_IF_ERROR( |
81 | context->MakeShapeFromShapeProto(shape_proto, &shape_handle)); |
82 | context->set_output_handle_shapes_and_types( |
83 | 0, std::vector<shape_inference::ShapeAndType>{{shape_handle, dtype}}); |
84 | return OkStatus(); |
85 | }) |
86 | .Doc(R"doc( |
87 | A graph node which represents an argument to a function. |
88 | |
89 | output: The argument. |
90 | index: This argument is the index-th argument of the function. |
91 | |
92 | Attributes for shape inference: |
93 | 1. _output_shapes: this attribute should contain a list of TensorShapeProto |
94 | describing the shape(s) of the tensor(s) this _Arg node will produce. If set, |
95 | _Arg node's shape inference function will use it as the node's output shapes. |
96 | 2. _handle_dtypes and _handle_shapes: these attributes can be set on an _Arg |
97 | node producing resource output(s). If set, value of _handle_dtypes should |
98 | contain the dtype(s) of the resource(s) and value of _handle_shapes should |
99 | contain the shape(s) of the resource(s). If both attributes are set, _Arg |
100 | node's shape inference function will use their values as the node's output |
101 | handle's type(s) and shape(s). |
102 | )doc" ); |
103 | |
104 | REGISTER_SYSTEM_OP("_DeviceArg" ) |
105 | .Output("output: T" ) |
106 | .Attr("T: type" ) |
107 | .Attr("index: int >= 0" ) |
108 | .SetIsStateful() |
109 | .SetShapeFn([](shape_inference::InferenceContext* context) { |
110 | context->set_output(0, context->UnknownShape()); |
111 | return OkStatus(); |
112 | }) |
113 | .Doc(R"doc( |
114 | A graph node which represents an argument to a function. |
115 | |
116 | output: The argument. |
117 | index: This argument is the index-th argument of the function. |
118 | )doc" ); |
119 | |
120 | REGISTER_SYSTEM_OP("_Retval" ) |
121 | .Input("input: T" ) |
122 | .Attr("T: type" ) |
123 | .Attr("index: int >= 0" ) |
124 | .SetIsStateful() |
125 | .SetShapeFn([](shape_inference::InferenceContext* context) { |
126 | return OkStatus(); |
127 | }) |
128 | .Doc(R"doc( |
129 | A graph node which represents a return value of a function. |
130 | |
131 | input: The return value. |
132 | index: This return value is the index-th return value of the function. |
133 | )doc" ); |
134 | |
135 | REGISTER_SYSTEM_OP("_DeviceRetval" ) |
136 | .Input("input: T" ) |
137 | .Attr("T: type" ) |
138 | .Attr("index: int >= 0" ) |
139 | .SetIsStateful() |
140 | .SetShapeFn([](shape_inference::InferenceContext* context) { |
141 | return OkStatus(); |
142 | }) |
143 | .Doc(R"doc( |
144 | A graph node which represents a return value of a function. |
145 | |
146 | input: The return value. |
147 | index: This return value is the index-th return value of the function. |
148 | )doc" ); |
149 | |
150 | REGISTER_SYSTEM_OP("_ListToArray" ) |
151 | .Input("input: Tin" ) |
152 | .Output("output: N * T" ) |
153 | .Attr("Tin: list(type)" ) |
154 | .Attr("T: type" ) |
155 | .Attr("N: int >= 1" ) |
156 | .SetShapeFn(shape_inference::UnknownShape) |
157 | .Doc(R"doc( |
158 | Converts a list of tensors to an array of tensors. |
159 | )doc" ); |
160 | |
161 | REGISTER_SYSTEM_OP("_ArrayToList" ) |
162 | .Input("input: N * T" ) |
163 | .Output("output: out_types" ) |
164 | .Attr("T: type" ) |
165 | .Attr("N: int >= 1" ) |
166 | .Attr("out_types: list(type)" ) |
167 | .SetShapeFn(shape_inference::UnknownShape) |
168 | .Doc(R"doc( |
169 | Converts an array of tensors to a list of tensors. |
170 | )doc" ); |
171 | |
172 | } // namespace tensorflow |
173 | |