1 /*
2 * Copyright (c) 2016-2022 Arm Limited.
3 *
4 * SPDX-License-Identifier: MIT
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to
8 * deal in the Software without restriction, including without limitation the
9 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10 * sell copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in all
14 * copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 */
24 #include "arm_compute/runtime/NEON/functions/NEScale.h"
25
26 #include "src/common/utils/Log.h"
27 #include "src/core/utils/ScaleUtils.h"
28 #include "src/cpu/operators/CpuScale.h"
29
30 namespace arm_compute
31 {
32 struct NEScale::Impl
33 {
34 const ITensor *src{ nullptr };
35 ITensor *dst{ nullptr };
36 Tensor dx{ nullptr }; /**< Element's distance between the X real coordinate and the smallest X following integer */
37 Tensor dy{ nullptr }; /**< Element's distance between the Y real coordinate and the smallest Y following integer */
38 Tensor offsets{ nullptr }; /**< Offset to access the element with NEAREST interpolation or the top-left element with BILINEAR interpolation in the input tensor */
39 std::unique_ptr<cpu::CpuScale> op{ nullptr };
40 };
41
NEScale()42 NEScale::NEScale()
43 : _impl(std::make_unique<Impl>())
44 {
45 }
46 NEScale::~NEScale() = default;
47
configure(ITensor * input,ITensor * output,const ScaleKernelInfo & info)48 void NEScale::configure(ITensor *input, ITensor *output, const ScaleKernelInfo &info)
49 {
50 ARM_COMPUTE_LOG_PARAMS(input, output, info);
51
52 _impl->src = input;
53 _impl->dst = output;
54 _impl->op = std::make_unique<cpu::CpuScale>();
55 _impl->op->configure(input->info(), output->info(), info);
56
57 // Configure for size of allocation of internal tensors
58 // Get data layout and width/height indices
59 const DataLayout data_layout = info.data_layout == DataLayout::UNKNOWN ? input->info()->data_layout() : info.data_layout;
60 const int idx_width = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH);
61 const int idx_height = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT);
62
63 // Compute the ratio between source width/height and destination width/height
64 const bool is_align_corners_used = info.align_corners && arm_compute::scale_utils::is_align_corners_allowed_sampling_policy(info.sampling_policy);
65 const auto wr = arm_compute::scale_utils::calculate_resize_ratio(input->info()->dimension(idx_width), output->info()->dimension(idx_width), is_align_corners_used);
66 const auto hr = arm_compute::scale_utils::calculate_resize_ratio(input->info()->dimension(idx_height), output->info()->dimension(idx_height), is_align_corners_used);
67
68 // Area interpolation behaves as Nearest Neighbour in case of up-sampling
69 InterpolationPolicy policy_to_use = (info.interpolation_policy == InterpolationPolicy::AREA && wr <= 1.f && hr <= 1.f) ? InterpolationPolicy::NEAREST_NEIGHBOR : info.interpolation_policy;
70
71 // Get the tensor shape
72 TensorShape shape(output->info()->dimension(idx_width));
73 shape.set(1, output->info()->dimension(idx_height), false);
74
75 bool precompute_indices_weights = arm_compute::scale_utils::is_precomputation_required(data_layout, input->info()->data_type(), policy_to_use, info.border_mode);
76
77 if(precompute_indices_weights)
78 {
79 const TensorInfo tensor_info_dxdy(shape, Format::F32);
80 const TensorInfo tensor_info_offsets(shape, Format::S32);
81
82 _impl->dx.allocator()->init(tensor_info_dxdy);
83 _impl->dy.allocator()->init(tensor_info_dxdy);
84 _impl->offsets.allocator()->init(tensor_info_offsets);
85 switch(policy_to_use)
86 {
87 case InterpolationPolicy::NEAREST_NEIGHBOR:
88 {
89 // Allocate once the configure methods have been called
90 _impl->offsets.allocator()->allocate();
91 break;
92 }
93 case InterpolationPolicy::BILINEAR:
94 {
95 // Allocate once the configure methods have been called
96 _impl->dx.allocator()->allocate();
97 _impl->dy.allocator()->allocate();
98 _impl->offsets.allocator()->allocate();
99 break;
100 }
101 case InterpolationPolicy::AREA:
102 {
103 break;
104 }
105 default:
106 ARM_COMPUTE_ERROR("Unsupported interpolation mode");
107 }
108 }
109 else
110 {
111 if(policy_to_use != InterpolationPolicy::NEAREST_NEIGHBOR && policy_to_use != InterpolationPolicy::BILINEAR && policy_to_use != InterpolationPolicy::AREA)
112 {
113 ARM_COMPUTE_ERROR("Unsupported interpolation mode");
114 }
115 }
116 }
117
validate(const ITensorInfo * input,const ITensorInfo * output,const ScaleKernelInfo & info)118 Status NEScale::validate(const ITensorInfo *input, const ITensorInfo *output, const ScaleKernelInfo &info)
119 {
120 return cpu::CpuScale::validate(input, output, info);
121 }
122
run()123 void NEScale::run()
124 {
125 ITensorPack pack;
126 pack.add_tensor(TensorType::ACL_SRC, _impl->src);
127 pack.add_tensor(TensorType::ACL_DST, _impl->dst);
128 pack.add_tensor(TensorType::ACL_INT_0, &_impl->dx);
129 pack.add_tensor(TensorType::ACL_INT_1, &_impl->dy);
130 pack.add_tensor(TensorType::ACL_INT_2, &_impl->offsets);
131 _impl->op->run(pack);
132 }
133 } // namespace arm_compute
134