1/* Copyright 2015 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// See docs in ../ops/nn_ops.cc.
17
18#define EIGEN_USE_THREADS
19
20#include "tensorflow/core/kernels/l2loss_op.h"
21
22#include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor"
23#include "tensorflow/core/framework/numeric_op.h"
24#include "tensorflow/core/framework/op_kernel.h"
25#include "tensorflow/core/framework/register_types.h"
26#include "tensorflow/core/framework/tensor.h"
27
28namespace tensorflow {
29
30typedef Eigen::ThreadPoolDevice CPUDevice;
31
32template <typename T>
33class L2LossOp<CPUDevice, T> : public OpKernel {
34 public:
35 explicit L2LossOp(OpKernelConstruction* context) : OpKernel(context) {}
36
37 void Compute(OpKernelContext* context) override {
38 // The input tensor can be of any number of dimensions, even though it's
39 // 2D in most typical applications.
40 const Tensor& input = context->input(0);
41 // The output is a single number.
42 Tensor* output = nullptr;
43 OP_REQUIRES_OK(context,
44 context->allocate_output(0, TensorShape({}), &output));
45 const CPUDevice& d = context->eigen_device<CPUDevice>();
46 output->scalar<T>().device(d) =
47 (input.flat<T>().square() * static_cast<T>(0.5)).sum();
48 }
49};
50
51#define REGISTER_KERNEL(T) \
52 REGISTER_KERNEL_BUILDER( \
53 Name("L2Loss").Device(DEVICE_CPU).TypeConstraint<T>("T"), \
54 L2LossOp<CPUDevice, T>);
55
56REGISTER_KERNEL(float);
57REGISTER_KERNEL(double);
58REGISTER_KERNEL(Eigen::half);
59#ifdef INTEL_MKL
60// Since Eigen backend does not support bfloat16 ops, we are selectively
61// enabling them for MKL backend.
62REGISTER_KERNEL(bfloat16);
63#endif // INTEL_MKL
64#undef REGISTER_KERNEL
65
66} // namespace tensorflow
67