xref: /aosp_15_r20/external/ComputeLibrary/src/runtime/NEON/functions/NEL2NormalizeLayer.cpp (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2017-2021 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/NEL2NormalizeLayer.h"
25 
26 #include "arm_compute/core/Helpers.h"
27 #include "arm_compute/runtime/NEON/NEScheduler.h"
28 #include "src/common/utils/Log.h"
29 #include "src/core/NEON/kernels/NEL2NormalizeLayerKernel.h"
30 #include "src/core/NEON/kernels/NEReductionOperationKernel.h"
31 
32 namespace arm_compute
33 {
34 namespace
35 {
36 constexpr int max_input_tensor_dim = 3;
37 } // namespace
38 NEL2NormalizeLayer::~NEL2NormalizeLayer() = default;
39 
NEL2NormalizeLayer(std::shared_ptr<IMemoryManager> memory_manager)40 NEL2NormalizeLayer::NEL2NormalizeLayer(std::shared_ptr<IMemoryManager> memory_manager)
41     : _memory_group(std::move(memory_manager)), _reduce_func(), _normalize_kernel(), _sumsq()
42 {
43 }
44 
configure(ITensor * input,ITensor * output,int axis,float epsilon)45 void NEL2NormalizeLayer::configure(ITensor *input, ITensor *output, int axis, float epsilon)
46 {
47     ARM_COMPUTE_LOG_PARAMS(input, output, axis, epsilon);
48 
49     // Manage intermediate buffers
50     _memory_group.manage(&_sumsq);
51 
52     // Configure Kernels
53     const uint32_t actual_axis = wrap_around(axis, max_input_tensor_dim);
54     _reduce_func.configure(input, &_sumsq, actual_axis, ReductionOperation::SUM_SQUARE);
55     _normalize_kernel = std::make_unique<NEL2NormalizeLayerKernel>();
56     _normalize_kernel->configure(input, &_sumsq, output, axis, epsilon);
57 
58     // Allocate intermediate tensors
59     _sumsq.allocator()->allocate();
60 }
61 
validate(const ITensorInfo * input,const ITensorInfo * output,int axis,float epsilon)62 Status NEL2NormalizeLayer::validate(const ITensorInfo *input, const ITensorInfo *output, int axis, float epsilon)
63 {
64     TensorShape shape(input->tensor_shape());
65 
66     // Create intermediate tensor info
67     TensorInfo sum_sq;
68     sum_sq.set_data_type(input->data_type());
69     sum_sq.set_tensor_shape(shape);
70 
71     const uint32_t actual_axis = wrap_around(axis, max_input_tensor_dim);
72     ARM_COMPUTE_RETURN_ON_ERROR(NEReductionOperation::validate(input, &sum_sq, actual_axis, ReductionOperation::SUM_SQUARE));
73 
74     // Reduce shape on axis
75     shape.set(actual_axis, 1);
76     sum_sq.set_tensor_shape(shape);
77 
78     ARM_COMPUTE_RETURN_ON_ERROR(NEL2NormalizeLayerKernel::validate(input, &sum_sq, output, axis, epsilon));
79 
80     return Status{};
81 }
82 
run()83 void NEL2NormalizeLayer::run()
84 {
85     MemoryGroupResourceScope scope_mg(_memory_group);
86 
87     _reduce_func.run();
88     NEScheduler::get().schedule(_normalize_kernel.get(), Window::DimY);
89 }
90 } // namespace arm_compute
91