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_KERNELS_STRING_TO_HASH_BUCKET_OP_H_ 17 #define TENSORFLOW_CORE_KERNELS_STRING_TO_HASH_BUCKET_OP_H_ 18 19 #include <string> 20 21 #include "tensorflow/core/framework/op_kernel.h" 22 #include "tensorflow/core/framework/tensor.h" 23 #include "tensorflow/core/lib/core/errors.h" 24 #include "tensorflow/core/lib/core/status.h" 25 #include "tensorflow/core/platform/macros.h" 26 27 namespace tensorflow { 28 29 template <uint64 hash(const string&)> 30 class StringToHashBucketOp : public OpKernel { 31 public: 32 explicit StringToHashBucketOp(OpKernelConstruction* ctx) : OpKernel(ctx) { 33 OP_REQUIRES_OK(ctx, ctx->GetAttr("num_buckets", &num_buckets_)); 34 } 35 36 void Compute(OpKernelContext* context) override { 37 const Tensor* input_tensor; 38 OP_REQUIRES_OK(context, context->input("input", &input_tensor)); 39 const auto& input_flat = input_tensor->flat<string>(); 40 41 Tensor* output_tensor = nullptr; 42 OP_REQUIRES_OK(context, 43 context->allocate_output("output", input_tensor->shape(), 44 &output_tensor)); 45 auto output_flat = output_tensor->flat<int64>(); 46 47 typedef decltype(input_flat.size()) Index; 48 for (Index i = 0; i < input_flat.size(); ++i) { 49 const uint64 input_hash = hash(input_flat(i)); 50 const uint64 bucket_id = input_hash % num_buckets_; 51 // The number of buckets is always in the positive range of int64 so is 52 // the resulting bucket_id. Casting the bucket_id from uint64 to int64 is 53 // safe. 54 output_flat(i) = static_cast<int64>(bucket_id); 55 } 56 } 57 58 private: 59 int64 num_buckets_; 60 61 TF_DISALLOW_COPY_AND_ASSIGN(StringToHashBucketOp); 62 }; 63 64 template <uint64 hash(const uint64 (&)[2], const string&)> 65 class StringToKeyedHashBucketOp : public OpKernel { 66 public: 67 explicit StringToKeyedHashBucketOp(OpKernelConstruction* ctx) 68 : OpKernel(ctx) { 69 OP_REQUIRES_OK(ctx, ctx->GetAttr("num_buckets", &num_buckets_)); 70 71 std::vector<int64> key; 72 OP_REQUIRES_OK(ctx, ctx->GetAttr("key", &key)); 73 OP_REQUIRES(ctx, key.size() == 2, 74 errors::InvalidArgument("Key must have 2 elements")); 75 std::memcpy(key_, key.data(), sizeof(key_)); 76 } 77 78 void Compute(OpKernelContext* context) override { 79 const Tensor* input_tensor; 80 OP_REQUIRES_OK(context, context->input("input", &input_tensor)); 81 const auto& input_flat = input_tensor->flat<string>(); 82 83 Tensor* output_tensor = nullptr; 84 OP_REQUIRES_OK(context, 85 context->allocate_output("output", input_tensor->shape(), 86 &output_tensor)); 87 auto output_flat = output_tensor->flat<int64>(); 88 89 typedef decltype(input_flat.size()) Index; 90 for (Index i = 0; i < input_flat.size(); ++i) { 91 const uint64 input_hash = hash(key_, input_flat(i)); 92 const uint64 bucket_id = input_hash % num_buckets_; 93 // The number of buckets is always in the positive range of int64 so is 94 // the resulting bucket_id. Casting the bucket_id from uint64 to int64 is 95 // safe. 96 output_flat(i) = static_cast<int64>(bucket_id); 97 } 98 } 99 100 private: 101 int64 num_buckets_; 102 uint64 key_[2]; 103 104 TF_DISALLOW_COPY_AND_ASSIGN(StringToKeyedHashBucketOp); 105 }; 106 107 } // namespace tensorflow 108 109 #endif // TENSORFLOW_CORE_KERNELS_STRING_TO_HASH_BUCKET_OP_H_ 110