1/* Copyright 2021 The TensorFlow Authors. All Rights Reserved.
2
3Licensed under the Apache License, Version 2.0 (the "License");
4you may not use this file except in compliance with the License.
5You may obtain a copy of the License at
6
7 http://www.apache.org/licenses/LICENSE-2.0
8
9Unless required by applicable law or agreed to in writing, software
10distributed under the License is distributed on an "AS IS" BASIS,
11WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12See the License for the specific language governing permissions and
13limitations under the License.
14==============================================================================*/
15
16#ifndef TENSORFLOW_CORE_KERNELS_TENSOR_TO_HASH_BUCKET_OP_H_
17#define TENSORFLOW_CORE_KERNELS_TENSOR_TO_HASH_BUCKET_OP_H_
18
19#include <string>
20
21#include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor"
22#include "tensorflow/core/framework/op_kernel.h"
23#include "tensorflow/core/framework/tensor.h"
24#include "tensorflow/core/lib/core/errors.h"
25#include "tensorflow/core/lib/core/status.h"
26#include "tensorflow/core/lib/strings/stringprintf.h"
27#include "tensorflow/core/platform/fingerprint.h"
28#include "tensorflow/core/platform/macros.h"
29#include "tensorflow/core/platform/types.h"
30
31namespace tensorflow {
32
33namespace functor {
34
35template <typename Device, typename T>
36struct LaunchTensorToHashBucket {
37 void operator()(OpKernelContext* c, const int64_t num_buckets, const T* input,
38 const int num_elems, int64_t* output) {
39 string format = "%";
40 switch (DataTypeToEnum<T>::value) {
41 case DT_INT8:
42 case DT_INT16:
43 case DT_INT32:
44 strings::Appendf(&format, "d");
45 break;
46 case DT_INT64:
47 strings::Appendf(&format, "lld");
48 break;
49 default:
50 bool type_not_supported = true;
51 OP_REQUIRES(
52 c, !type_not_supported,
53 errors::InvalidArgument("Type not supported: ",
54 DataTypeString(DataTypeToEnum<T>::value)));
55 }
56
57 for (int i = 0; i < num_elems; ++i) {
58 string input_str = strings::Printf(format.c_str(), input[i]);
59 const uint64 input_hash = Fingerprint64(input_str);
60 const uint64 bucket_id = input_hash % num_buckets;
61 // The number of buckets is always in the positive range of int64 so is
62 // the resulting bucket_id. Casting the bucket_id from uint64 to int64 is
63 // safe.
64 output[i] = static_cast<int64_t>(bucket_id);
65 }
66 }
67};
68
69#if GOOGLE_CUDA
70template <typename T>
71struct LaunchTensorToHashBucket<Eigen::GpuDevice, T> {
72 void operator()(OpKernelContext* c, const int64_t num_buckets, const T* input,
73 const int num_elems, int64_t* output);
74};
75#endif // GOOGLE_CUDA
76} // namespace functor
77
78} // namespace tensorflow
79
80#endif // TENSORFLOW_CORE_KERNELS_TENSOR_TO_HASH_BUCKET_OP_H_
81