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 <string> |
17 | #include <unordered_set> |
18 | #include <utility> |
19 | |
20 | #include "tensorflow/core/framework/op_kernel.h" |
21 | #include "tensorflow/core/framework/register_types.h" |
22 | #include "tensorflow/core/framework/tensor.h" |
23 | #include "tensorflow/core/framework/tensor_shape.h" |
24 | #include "tensorflow/core/lib/core/status.h" |
25 | |
26 | namespace tensorflow { |
27 | template <typename T, typename Tidx> |
28 | class ListDiffOp : public OpKernel { |
29 | public: |
30 | explicit ListDiffOp(OpKernelConstruction* context) : OpKernel(context) { |
31 | const DataType dt = DataTypeToEnum<T>::v(); |
32 | const DataType dtidx = DataTypeToEnum<Tidx>::v(); |
33 | OP_REQUIRES_OK(context, context->MatchSignature({dt, dt}, {dt, dtidx})); |
34 | } |
35 | |
36 | void Compute(OpKernelContext* context) override { |
37 | const Tensor& x = context->input(0); |
38 | const Tensor& y = context->input(1); |
39 | |
40 | OP_REQUIRES(context, TensorShapeUtils::IsVector(x.shape()), |
41 | errors::InvalidArgument("x should be a 1D vector." )); |
42 | |
43 | OP_REQUIRES(context, TensorShapeUtils::IsVector(y.shape()), |
44 | errors::InvalidArgument("y should be a 1D vector." )); |
45 | |
46 | const auto Tx = x.vec<T>(); |
47 | const size_t x_size = Tx.size(); |
48 | const auto Ty = y.vec<T>(); |
49 | const size_t y_size = Ty.size(); |
50 | |
51 | OP_REQUIRES(context, x_size < std::numeric_limits<int32>::max(), |
52 | errors::InvalidArgument("x too large for int32 indexing" )); |
53 | |
54 | std::unordered_set<T> y_set; |
55 | y_set.reserve(y_size); |
56 | for (size_t i = 0; i < y_size; ++i) { |
57 | y_set.insert(Ty(i)); |
58 | } |
59 | |
60 | // Compute the size of the output. |
61 | |
62 | int64_t out_size = 0; |
63 | for (size_t i = 0; i < x_size; ++i) { |
64 | if (y_set.count(Tx(i)) == 0) { |
65 | ++out_size; |
66 | } |
67 | } |
68 | |
69 | // Allocate and populate outputs. |
70 | Tensor* out = nullptr; |
71 | OP_REQUIRES_OK(context, context->allocate_output(0, {out_size}, &out)); |
72 | auto Tout = out->vec<T>(); |
73 | |
74 | Tensor* indices = nullptr; |
75 | OP_REQUIRES_OK(context, context->allocate_output(1, {out_size}, &indices)); |
76 | auto Tindices = indices->vec<Tidx>(); |
77 | |
78 | for (Tidx i = 0, p = 0; i < static_cast<Tidx>(x_size); ++i) { |
79 | if (y_set.count(Tx(i)) == 0) { |
80 | OP_REQUIRES(context, p < out_size, |
81 | errors::InvalidArgument( |
82 | "Tried to set output index " , p, |
83 | " when output Tensor only had " , out_size, |
84 | " elements. Check that your " |
85 | "input tensors are not being concurrently mutated." )); |
86 | Tout(p) = Tx(i); |
87 | Tindices(p) = i; |
88 | p++; |
89 | } |
90 | } |
91 | } |
92 | }; |
93 | |
94 | #define REGISTER_LISTDIFF(type) \ |
95 | REGISTER_KERNEL_BUILDER(Name("ListDiff") \ |
96 | .Device(DEVICE_CPU) \ |
97 | .TypeConstraint<type>("T") \ |
98 | .TypeConstraint<int32>("out_idx"), \ |
99 | ListDiffOp<type, int32>) \ |
100 | REGISTER_KERNEL_BUILDER(Name("ListDiff") \ |
101 | .Device(DEVICE_CPU) \ |
102 | .TypeConstraint<type>("T") \ |
103 | .TypeConstraint<int64_t>("out_idx"), \ |
104 | ListDiffOp<type, int64>) |
105 | |
106 | TF_CALL_REAL_NUMBER_TYPES(REGISTER_LISTDIFF); |
107 | REGISTER_LISTDIFF(tstring); |
108 | #undef REGISTER_LISTDIFF |
109 | |
110 | } // namespace tensorflow |
111 | |