1 /* Copyright 2016 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 #ifndef TENSORFLOW_CORE_UTIL_STRIDED_SLICE_OP_H_ 16 #define TENSORFLOW_CORE_UTIL_STRIDED_SLICE_OP_H_ 17 18 #include "tensorflow/core/framework/tensor.h" 19 #include "tensorflow/core/framework/tensor_shape.h" 20 #include "tensorflow/core/framework/types.h" 21 #include "tensorflow/core/lib/core/status.h" 22 #include "tensorflow/core/lib/gtl/inlined_vector.h" 23 24 namespace tensorflow { 25 26 struct StridedSliceShapeSpec { 27 // Begin mask canonlized in dense form. 28 int32_t begin_dense_mask; 29 // End mask canonlized in dense form. 30 int32_t end_dense_mask; 31 // Shrink axis mask canonlized in dense form. 32 int32_t shrink_axis_dense_mask; 33 // output_to_sparse_mapping[i] represents output[i]'s the corresponding dim 34 // index in the begin_tensor. If 35 // output_to_sparse_mapping[i] is -1, it means the dimension doesn't show up 36 // in sparse_mapping. 37 gtl::InlinedVector<int64_t, 4> output_to_sparse_mapping; 38 // output_to_processing_mapping is similar to output_to_sparse_mapping, but 39 // for processing shape. 40 gtl::InlinedVector<int64_t, 4> output_to_processing_mapping; 41 // processing_to_sparse_mapping[i] represents input_shape[i]'s corresponding 42 // dim index in the begin_tensor. 43 gtl::InlinedVector<int64_t, 4> processing_to_sparse_mapping; 44 }; 45 46 // Runs validation on the strided slice op parameters. 47 // 48 // Is a separate translation unit from the kernel so that: 49 // 1. The op's shape function can use it. 50 // 2. The code size is reduced vs templating this on the kernel's type. 51 // 52 // Note that when input_shape is not fully specified, only <final_shape> and 53 // <processing_shape> are valid; <is_identity>, <is_simple_slice> and other 54 // output parameters will not be accurate. 55 // 56 // If <begin_tensor> or <end_tensor> are nullptr, <begin> and <end> will not be 57 // valid. In this case, <slice_dim0> and <is_identity> will be true only if a 58 // determination can be made based on the information given. A best effort is 59 // made to set <processing_shape> and <final_shape> based on <input_shape>, but 60 // some dimensions of <processing_shape> and/or <final_shape> may be unknown 61 // (-1). Any validation that can be done without complete information is 62 // performed. 63 // 64 Status ValidateStridedSliceOp( 65 const Tensor* begin_tensor, const Tensor* end_tensor, 66 const Tensor& strides_tensor, const PartialTensorShape& input_shape, 67 int32_t begin_mask_spec, int32_t end_mask_spec, const int32_t ellipsis_mask, 68 int32_t new_axis_mask, int32_t shrink_axis_mask, 69 PartialTensorShape* processing_shape, PartialTensorShape* final_shape, 70 bool* is_identity, bool* is_simple_slice, bool* slice_dim0, 71 gtl::InlinedVector<int64_t, 4>* begin, gtl::InlinedVector<int64_t, 4>* end, 72 gtl::InlinedVector<int64_t, 4>* strides, 73 StridedSliceShapeSpec* shape_spec = nullptr); 74 75 // Same as above, but the outputs are TensorShape, not PartialTensorShape 76 Status ValidateStridedSliceOp( 77 const Tensor* begin_tensor, const Tensor* end_tensor, 78 const Tensor& strides_tensor, const PartialTensorShape& input_shape, 79 int32_t begin_mask_spec, int32_t end_mask_spec, const int32_t ellipsis_mask, 80 int32_t new_axis_mask, int32_t shrink_axis_mask, 81 TensorShape* processing_shape, TensorShape* final_shape, bool* is_identity, 82 bool* is_simple_slice, bool* slice_dim0, 83 gtl::InlinedVector<int64_t, 4>* begin, gtl::InlinedVector<int64_t, 4>* end, 84 gtl::InlinedVector<int64_t, 4>* strides, 85 StridedSliceShapeSpec* shape_spec = nullptr); 86 87 // Simple class for determining if it is possible to broadcast a tensor to a 88 // strided slice. Modelled after tensorflow::BCast, but with a few key 89 // differences: 90 // - the input_shape must be broadcastable to output_shape 91 // (i.e. the slice shape does not grow). 92 // - does not allow reducing or flattening dimensions, since we cannot apply 93 // these simplications to the destination slice. 94 // - allows for remapping dimensions, required in order to associate the input 95 // with correct dimensions in the full (unsliced) destination tensor. 96 class StridedSliceAssignBCast { 97 public: 98 using Vec = gtl::InlinedVector<int64_t, 4>; 99 100 StridedSliceAssignBCast(const Vec& input_shape, const Vec& output_shape); 101 102 // Remaps broadcast, resize, and output dimensions via the provided map. 103 // Negative values in the map correspond to dimensions being removed. 104 // Unmapped dimensions are set to 1. 105 // 106 // This is to support remapping slice -> processing dimensions. To relate 107 // the sliced output dimensions back to processing dimensions (i.e. those 108 // relative to the original unsliced input), we need to remove any axes 109 // that were added via the `new_axis_mask`, and add back any axes that were 110 // removed via the `shrink_axis_mask`. For example, an expression like 111 // 112 // >>> t = tf.zeros([3, 3]) 113 // >>> t[2, tf.newaxis, 0:2, tf.newaxis] = tf.ones([1, 3, 1]) 114 // ^ ^ ^ ^ 115 // |__ shrink axis new axis __| | |__ new axis 116 // |_____ dim 1 of t 117 // 118 // would have `new_axis_mask = 0b1010` and `shrink_axis_mask = 0b0001`. The 119 // slice has shape [1, 3, 1], but the original input tensor `t` has shape 120 // [3, 3]. To remap the slice dimensions back to the input dimensions, the 121 // mapping would use `num_dims = 2`, `dimension_map = {-1, 1, -1}`. This 122 // removes the two new axes added for the slice, maps the middle slice 123 // dimension to input dimension 1, and leaves input dimension 0 to have a 124 // default size of 1 to add back the shrink axis. 125 // 126 // Returns false if the remapping fails. 127 bool RemapDimensions(int64_t num_dims, const Vec& dimension_map); 128 IsValid()129 bool IsValid() const { return valid_; } 130 IsBroadcastingRequired()131 bool IsBroadcastingRequired() const { return broadcasting_required_; } 132 reshape()133 const Vec& reshape() const { return reshape_; } 134 bcast()135 const Vec& bcast() const { return bcast_; } 136 result_shape()137 const Vec& result_shape() const { return result_shape_; } 138 139 private: 140 bool valid_ = true; 141 bool broadcasting_required_ = false; 142 Vec reshape_; 143 Vec bcast_; 144 Vec result_shape_; 145 }; 146 147 } // namespace tensorflow 148 149 #endif // TENSORFLOW_CORE_UTIL_STRIDED_SLICE_OP_H_ 150