1 /* Copyright 2017 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 #define EIGEN_USE_THREADS 17 18 #include "tensorflow/core/framework/op_kernel.h" 19 #include "tensorflow/core/framework/register_types.h" 20 #include "tensorflow/core/framework/tensor.h" 21 #include "tensorflow/core/framework/types.h" 22 #include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor" 23 24 namespace tensorflow { 25 26 namespace { 27 template <typename T> 28 struct mod_op { 29 const T operator()(const T& a, const T& b) const { return a % b; } 30 }; 31 } // namespace 32 33 typedef Eigen::ThreadPoolDevice CPUDevice; 34 35 template <typename Tidx> 36 class UnravelIndexOp : public OpKernel { 37 public: 38 explicit UnravelIndexOp(OpKernelConstruction* ctx) : OpKernel(ctx) {} 39 40 void Compute(OpKernelContext* ctx) override { 41 const Tensor& indices_tensor = ctx->input(0); 42 OP_REQUIRES(ctx, 43 TensorShapeUtils::IsVector(indices_tensor.shape()) || 44 TensorShapeUtils::IsScalar(indices_tensor.shape()), 45 errors::InvalidArgument( 46 "The indices can only be scalar or vector, got \"", 47 indices_tensor.shape().DebugString(), "\"")); 48 49 const Tensor& dims_tensor = ctx->input(1); 50 OP_REQUIRES( 51 ctx, TensorShapeUtils::IsVector(dims_tensor.shape()), 52 errors::InvalidArgument("The indices can only be 1-D, got \"", 53 dims_tensor.shape().DebugString(), "\"")); 54 55 auto dims = dims_tensor.vec<Tidx>(); 56 57 Eigen::array<bool, 1> reverse({true}); 58 59 Tensor strides_tensor; 60 OP_REQUIRES_OK(ctx, 61 ctx->allocate_temp(DataTypeToEnum<Tidx>::value, 62 TensorShape({dims_tensor.NumElements()}), 63 &strides_tensor)); 64 65 auto strides = strides_tensor.vec<Tidx>(); 66 strides = dims.reverse(reverse) 67 .scan(0, Eigen::internal::ProdReducer<Tidx>(), false) 68 .reverse(reverse); 69 70 Tensor strides_shifted_tensor; 71 OP_REQUIRES_OK(ctx, 72 ctx->allocate_temp(DataTypeToEnum<Tidx>::value, 73 TensorShape({dims_tensor.NumElements()}), 74 &strides_shifted_tensor)); 75 76 auto strides_shifted = strides_shifted_tensor.vec<Tidx>(); 77 strides_shifted = dims.reverse(reverse) 78 .scan(0, Eigen::internal::ProdReducer<Tidx>(), true) 79 .reverse(reverse); 80 81 Tensor* output_tensor = nullptr; 82 if (TensorShapeUtils::IsScalar(indices_tensor.shape())) { 83 OP_REQUIRES_OK( 84 ctx, ctx->allocate_output(0, TensorShape({dims_tensor.NumElements()}), 85 &output_tensor)); 86 87 auto output = output_tensor->vec<Tidx>(); 88 89 output = output.constant(indices_tensor.scalar<Tidx>()()); 90 output = output.binaryExpr(strides, mod_op<Tidx>()) / strides_shifted; 91 } else { 92 OP_REQUIRES_OK( 93 ctx, ctx->allocate_output(0, 94 TensorShape({dims_tensor.NumElements(), 95 indices_tensor.NumElements()}), 96 &output_tensor)); 97 98 auto output = output_tensor->matrix<Tidx>(); 99 100 Eigen::array<int64, 2> reshape{{dims_tensor.NumElements(), 1}}; 101 Eigen::array<int64, 2> bcast({1, indices_tensor.NumElements()}); 102 Eigen::array<int64, 2> indices_reshape{{1, indices_tensor.NumElements()}}; 103 Eigen::array<int64, 2> indices_bcast({dims_tensor.NumElements(), 1}); 104 105 output = indices_tensor.vec<Tidx>() 106 .reshape(indices_reshape) 107 .broadcast(indices_bcast); 108 output = output.binaryExpr(strides.reshape(reshape).broadcast(bcast), 109 mod_op<Tidx>()) / 110 strides_shifted.reshape(reshape).broadcast(bcast); 111 } 112 } 113 }; 114 115 #define REGISTER_KERNEL(type) \ 116 REGISTER_KERNEL_BUILDER( \ 117 Name("UnravelIndex").Device(DEVICE_CPU).TypeConstraint<type>("Tidx"), \ 118 UnravelIndexOp<type>); 119 TF_CALL_int32(REGISTER_KERNEL) TF_CALL_int64(REGISTER_KERNEL) 120 #undef REGISTER_KERNEL 121 122 } // namespace tensorflow 123