1 | /* Copyright 2021 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/kernels/tensor_to_hash_bucket_op.h" |
17 | |
18 | #include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor" |
19 | #include "tensorflow/core/framework/register_types.h" |
20 | |
21 | namespace tensorflow { |
22 | |
23 | typedef Eigen::ThreadPoolDevice CPUDevice; |
24 | typedef Eigen::GpuDevice GPUDevice; |
25 | |
26 | template <typename Device, typename T> |
27 | class TensorToHashBucketOp : public OpKernel { |
28 | public: |
29 | explicit TensorToHashBucketOp(OpKernelConstruction* ctx) : OpKernel(ctx) { |
30 | OP_REQUIRES_OK(ctx, ctx->GetAttr("num_buckets" , &num_buckets_)); |
31 | |
32 | DataType dtype; |
33 | OP_REQUIRES_OK(ctx, ctx->GetAttr("T" , &dtype)); |
34 | OP_REQUIRES(ctx, |
35 | dtype == DT_INT8 || dtype == DT_UINT8 || dtype == DT_INT16 || |
36 | dtype == DT_UINT16 || dtype == DT_INT32 || |
37 | dtype == DT_UINT32 || dtype == DT_INT64 || |
38 | dtype == DT_UINT64, |
39 | errors::InvalidArgument("TensorToHashBucketOp doesn't support " |
40 | "datatype " , |
41 | DataTypeString(dtype))); |
42 | } |
43 | |
44 | void Compute(OpKernelContext* context) override { |
45 | const Tensor* input_tensor; |
46 | OP_REQUIRES_OK(context, context->input("input" , &input_tensor)); |
47 | const auto& input_flat = input_tensor->flat<T>(); |
48 | |
49 | Tensor* output_tensor = nullptr; |
50 | OP_REQUIRES_OK(context, |
51 | context->allocate_output("output" , input_tensor->shape(), |
52 | &output_tensor)); |
53 | auto output_flat = output_tensor->flat<int64_t>(); |
54 | |
55 | functor::LaunchTensorToHashBucket<Device, T>()( |
56 | context, num_buckets_, input_flat.data(), input_tensor->NumElements(), |
57 | output_flat.data()); |
58 | } |
59 | |
60 | private: |
61 | int64_t num_buckets_; |
62 | |
63 | TF_DISALLOW_COPY_AND_ASSIGN(TensorToHashBucketOp); |
64 | }; |
65 | |
66 | #define REGISTER_CPU_KERNELS(type) \ |
67 | REGISTER_KERNEL_BUILDER(Name("_TensorToHashBucketFast") \ |
68 | .Device(DEVICE_CPU) \ |
69 | .TypeConstraint<type>("T"), \ |
70 | TensorToHashBucketOp<CPUDevice, type>); |
71 | |
72 | TF_CALL_INTEGRAL_TYPES(REGISTER_CPU_KERNELS); |
73 | |
74 | #undef REGISTER_CPU_KERNELS |
75 | |
76 | #if GOOGLE_CUDA |
77 | |
78 | #define REGISTER_GPU_KERNELS(type) \ |
79 | REGISTER_KERNEL_BUILDER(Name("_TensorToHashBucketFast") \ |
80 | .Device(DEVICE_GPU) \ |
81 | .TypeConstraint<type>("T"), \ |
82 | TensorToHashBucketOp<GPUDevice, type>); |
83 | |
84 | TF_CALL_INTEGRAL_TYPES(REGISTER_GPU_KERNELS); |
85 | |
86 | #undef REGISTER_GPU_KERNELS |
87 | |
88 | #endif // GOOGLE_CUDA |
89 | |
90 | } // namespace tensorflow |
91 | |