1 /**
2 * Copyright 2021 Huawei Technologies Co., Ltd
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17 #include "nnacl/infer/addn_infer.h"
18 #include "nnacl/infer/infer_register.h"
19 #include "nnacl/tensor_c_utils.h"
20
AddnInferShape(const TensorC * const * inputs,size_t inputs_size,TensorC ** outputs,size_t outputs_size,OpParameter * parameter)21 int AddnInferShape(const TensorC *const *inputs, size_t inputs_size, TensorC **outputs, size_t outputs_size,
22 OpParameter *parameter) {
23 int check_ret = CheckAugmentWithMinSize(inputs, inputs_size, outputs, outputs_size, parameter, 2, 1);
24 if (check_ret != NNACL_OK) {
25 return check_ret;
26 }
27
28 const TensorC *input = inputs[0];
29 TensorC *output = outputs[0];
30 if (inputs_size < 2) {
31 return NNACL_ERR;
32 }
33 SetDataTypeFormat(output, input);
34 if (!InferFlag(inputs, inputs_size)) {
35 return NNACL_INFER_INVALID;
36 }
37
38 size_t max_dims = input->shape_size_;
39 size_t max_dims_idx = 0;
40
41 // check zerp dimension
42 for (size_t i = 0; i < max_dims; i++) {
43 NNACL_CHECK_FALSE(input->shape_[i] == 0, NNACL_ERR);
44 }
45
46 // determine max_dims
47 for (size_t i = 1; i < inputs_size; ++i) {
48 if (inputs[i]->shape_size_ > max_dims) {
49 max_dims = inputs[i]->shape_size_;
50 max_dims_idx = i;
51 }
52 }
53 ShapeSet(output->shape_, &output->shape_size_, inputs[max_dims_idx]->shape_, inputs[max_dims_idx]->shape_size_);
54
55 // make sure all elements have the same size or 1 (broadcasting) in all dimensions
56 for (size_t i = 1; i < inputs_size; ++i) {
57 if ((inputs[i]->shape_size_ != max_dims) && (GetElementNum(inputs[i]) != GetElementNum(inputs[max_dims_idx]))) {
58 return NNACL_ERR;
59 }
60 if (inputs[i]->shape_size_ == max_dims) {
61 for (size_t j = 0; j < max_dims; j++) {
62 if (inputs[i]->shape_[j] != inputs[max_dims_idx]->shape_[j] && inputs[i]->shape_[j] != 1 &&
63 inputs[max_dims_idx]->shape_[j] != 1) {
64 return NNACL_ERR;
65 }
66 }
67 }
68 }
69
70 for (size_t d = 0; d < inputs[max_dims_idx]->shape_size_; ++d) {
71 size_t max_dim = 0;
72 for (size_t i = 0; i < inputs_size; ++i) {
73 size_t shift = max_dims - (size_t)(inputs[i]->shape_size_);
74 size_t dim = (i < shift) ? 1 : (size_t)(inputs[i]->shape_[d]);
75 if (dim > max_dim) {
76 max_dim = dim;
77 }
78 }
79 output->shape_[d] = (int)(max_dim); // set the biggest dimension in the output tensor
80 }
81
82 return NNACL_OK;
83 }
84
85 REG_INFER(AddN, PrimType_AddN, AddnInferShape)
86