ONE - On-device Neural Engine
Loading...
Searching...
No Matches
PALMaxPool2DCommon.h
Go to the documentation of this file.
1/*
2 * Copyright (c) 2023 Samsung Electronics Co., Ltd. All Rights Reserved
3 * Copyright 2020 The TensorFlow Authors. All Rights Reserved.
4 *
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at
8 *
9 * http://www.apache.org/licenses/LICENSE-2.0
10 *
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
16 */
17
18#ifndef LUCI_INTERPRETER_PAL_MAX_POOL_2D_COMMON_H
19#define LUCI_INTERPRETER_PAL_MAX_POOL_2D_COMMON_H
20
21#include "Params.h"
22#include "PALUtils.h"
23
25{
26
27inline void MaxPool(const PoolParams &params, const luci_interpreter::RuntimeShape &input_shape,
28 const float *input_data, const luci_interpreter::RuntimeShape &output_shape,
29 float *output_data)
30{
31 const int batches = input_shape.dims(0);
32 const int depth = output_shape.dims(3);
33 const int input_height = input_shape.dims(1);
34 const int input_width = input_shape.dims(2);
35 const int output_height = output_shape.dims(1);
36 const int output_width = output_shape.dims(2);
37 const int stride_height = params.stride_height;
38 const int stride_width = params.stride_width;
39 for (int batch = 0; batch < batches; ++batch)
40 {
41 for (int out_y = 0; out_y < output_height; ++out_y)
42 {
43 for (int out_x = 0; out_x < output_width; ++out_x)
44 {
45 for (int channel = 0; channel < depth; ++channel)
46 {
47 const int in_x_origin = (out_x * stride_width) - params.padding_values.width;
48 const int in_y_origin = (out_y * stride_height) - params.padding_values.height;
49 // Compute the boundaries of the filter region clamped so as to
50 // ensure that the filter window fits in the input array.
51 const int filter_x_start = std::max(0, -in_x_origin);
52 const int filter_x_end = std::min(params.filter_width, input_width - in_x_origin);
53 const int filter_y_start = std::max(0, -in_y_origin);
54 const int filter_y_end = std::min(params.filter_height, input_height - in_y_origin);
55 float max = std::numeric_limits<float>::lowest();
56 for (int filter_y = filter_y_start; filter_y < filter_y_end; ++filter_y)
57 {
58 for (int filter_x = filter_x_start; filter_x < filter_x_end; ++filter_x)
59 {
60 const int in_x = in_x_origin + filter_x;
61 const int in_y = in_y_origin + filter_y;
62
63 const int input_data_offset =
64 ((batch * input_shape.dims(1) + in_y) * input_shape.dims(2) + in_x) *
65 input_shape.dims(3) +
66 channel;
67
68 max = std::max(max, input_data[input_data_offset]);
69 }
70 }
71 const int output_data_offset =
72 ((batch * output_shape.dims(1) + out_y) * output_shape.dims(2) + out_x) *
74 channel;
75
76 output_data[output_data_offset] =
77 std::min(std::max(max, params.float_activation_min), params.float_activation_max);
78 }
79 }
80 }
81 }
82}
83
84} // namespace luci_interpreter_pal
85
86#endif // LUCI_INTERPRETER_PAL_MAX_POOL_2D_COMMON_H
int32_t dims(int i) const
Definition Tensor.h:108
const luci_interpreter::RuntimeShape output_shape
void MaxPool(const PoolParams &params, const luci_interpreter::RuntimeShape &input_shape, const uint8_t *input_data, const luci_interpreter::RuntimeShape &output_shape, uint8_t *output_data, luci_interpreter::DataType data_type)