xref: /aosp_15_r20/external/ComputeLibrary/src/core/NEON/kernels/NEL2NormalizeLayerKernel.cpp (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2017-2022 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 "src/core/NEON/kernels/NEL2NormalizeLayerKernel.h"
25 
26 #include "arm_compute/core/Error.h"
27 #include "arm_compute/core/Helpers.h"
28 #include "arm_compute/core/ITensor.h"
29 #include "arm_compute/core/TensorInfo.h"
30 #include "arm_compute/core/Utils.h"
31 #include "arm_compute/core/Validate.h"
32 #include "arm_compute/core/Window.h"
33 #include "src/common/cpuinfo/CpuIsaInfo.h"
34 #include "src/core/NEON/NEMath.h"
35 #include "src/core/common/Registrars.h"
36 #include "src/core/helpers/AutoConfiguration.h"
37 #include "src/core/helpers/WindowHelpers.h"
38 #include "src/cpu/kernels/l2normlayer/list.h"
39 
40 #include <arm_neon.h>
41 #include <cmath>
42 
43 namespace arm_compute
44 {
45 namespace
46 {
47 constexpr int max_input_tensor_dim = 3;
48 
49 struct L2NormalizeLayerSelectorData
50 {
51     DataType            dt;
52     unsigned int        actual_axis;
53     cpuinfo::CpuIsaInfo isa;
54 };
55 
56 using L2NormalizeLayerKernelSelctorPtr = std::add_pointer<bool(const L2NormalizeLayerSelectorData &data)>::type;
57 
58 using L2NormalizeLayerPtr = std::add_pointer<void(const ITensor *in, const ITensor *sum, ITensor *out, float epsilon, const Window &window, size_t axis)>::type;
59 
60 struct L2NormalizeLayerKernel
61 {
62     const char                            *name;
63     const L2NormalizeLayerKernelSelctorPtr is_selected;
64     L2NormalizeLayerPtr                    ukernel;
65 };
66 
67 static const L2NormalizeLayerKernel available_kernels[] =
68 {
69     {
70         "fp32_neon_l2normalize_x",
__anonedc45faa0202() 71         [](const L2NormalizeLayerSelectorData & data) { return data.dt == DataType::F32 && data.actual_axis == Window::DimX; },
72         REGISTER_FP32_NEON(arm_compute::cpu::neon_fp32_l2_normalize_x)
73     },
74     {
75         "fp32_neon_l2normalize_yz",
__anonedc45faa0302() 76         [](const L2NormalizeLayerSelectorData & data) { return data.dt == DataType::F32 && data.actual_axis != Window::DimX; },
77         REGISTER_FP32_NEON(arm_compute::cpu::neon_fp32_l2_normalize_yz)
78     },
79     {
80         "fp16_neon_l2normalize_x",
__anonedc45faa0402() 81         [](const L2NormalizeLayerSelectorData & data) { return data.dt == DataType::F16 && data.isa.fp16 && data.actual_axis == Window::DimX; },
82         REGISTER_FP16_NEON(arm_compute::cpu::neon_fp16_l2_normalize_x),
83     },
84     {
85         "fp16_neon_l2normalize_yz",
__anonedc45faa0502() 86         [](const L2NormalizeLayerSelectorData & data) { return data.dt == DataType::F16 && data.isa.fp16 && data.actual_axis != Window::DimX; },
87         REGISTER_FP16_NEON(arm_compute::cpu::neon_fp16_l2_normalize_yz),
88     },
89 };
90 
91 /** Micro-kernel selector
92  *
93  * @param[in] data Selection data passed to help pick the appropriate micro-kernel
94  *
95  * @return A matching micro-kernel else nullptr
96  */
get_implementation(const L2NormalizeLayerSelectorData & data)97 const L2NormalizeLayerKernel *get_implementation(const L2NormalizeLayerSelectorData &data)
98 {
99     for(const auto &uk : available_kernels)
100     {
101         if(uk.is_selected(data))
102         {
103             return &uk;
104         }
105     }
106     return nullptr;
107 }
108 
validate_arguments(const ITensorInfo * input,const ITensorInfo * sum,const ITensorInfo * output,int axis,float epsilon)109 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *sum, const ITensorInfo *output, int axis, float epsilon)
110 {
111     ARM_COMPUTE_UNUSED(epsilon);
112 
113     const uint32_t actual_axis = wrap_around(axis, max_input_tensor_dim);
114     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, sum, output);
115     ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, sum);
116     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F16, DataType::F32);
117     ARM_COMPUTE_RETURN_ERROR_ON_MSG(actual_axis > 2, "Actual axis greater than 2 is not supported");
118     ARM_COMPUTE_RETURN_ERROR_ON_MSG(actual_axis >= TensorShape::num_max_dimensions, "Actual normalization axis greater than max number of dimensions");
119 
120     // Reduce shape on axis
121     TensorShape sum_shape = input->tensor_shape();
122     sum_shape.set(actual_axis, 1);
123     ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(sum->tensor_shape(), sum_shape);
124 
125     if(output->total_size() != 0)
126     {
127         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
128         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
129         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(input->tensor_shape(), output->tensor_shape());
130         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, output);
131     }
132 
133     return Status{};
134 }
135 
validate_and_configure_window(ITensorInfo * input,ITensorInfo * output)136 std::tuple<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output)
137 {
138     Window win = calculate_max_window(*input, Steps());
139 
140     // Output auto initialization if not yet initialized
141     auto_init_if_empty(*output, input->tensor_shape(), 1, input->data_type());
142 
143     // NEL2NormalizeLayerKernel doesn't need padding so update_window_and_padding() can be skipped
144 
145     return std::make_tuple(Status{}, win);
146 }
147 } // namespace
148 
NEL2NormalizeLayerKernel()149 NEL2NormalizeLayerKernel::NEL2NormalizeLayerKernel()
150     : _input(nullptr), _sum(nullptr), _output(nullptr), _actual_axis(0), _epsilon(1e-12)
151 {
152 }
153 
configure(const ITensor * input,const ITensor * sum,ITensor * output,int axis,float epsilon)154 void NEL2NormalizeLayerKernel::configure(const ITensor *input, const ITensor *sum, ITensor *output, int axis, float epsilon)
155 {
156     ARM_COMPUTE_ERROR_ON_NULLPTR(input, sum, output);
157     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), sum->info(), output->info(), axis, epsilon));
158 
159     _input       = input;
160     _sum         = sum;
161     _output      = output;
162     _actual_axis = wrap_around(axis, max_input_tensor_dim);
163     _epsilon     = epsilon;
164 
165     // Configure kernel window
166     auto win_config = validate_and_configure_window(_input->info(), _output->info());
167     ARM_COMPUTE_ERROR_THROW_ON(std::get<0>(win_config));
168 
169     INEKernel::configure(std::get<1>(win_config));
170 }
171 
validate(const ITensorInfo * input,const ITensorInfo * sum,const ITensorInfo * output,int axis,float epsilon)172 Status NEL2NormalizeLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *sum, const ITensorInfo *output, int axis, float epsilon)
173 {
174     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, sum, output, axis, epsilon));
175     ARM_COMPUTE_RETURN_ON_ERROR(std::get<0>(validate_and_configure_window(input->clone().get(), output->clone().get())));
176 
177     return Status{};
178 }
179 
run(const Window & window,const ThreadInfo & info)180 void NEL2NormalizeLayerKernel::run(const Window &window, const ThreadInfo &info)
181 {
182     ARM_COMPUTE_UNUSED(info);
183     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
184     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
185 
186     if(_actual_axis > 2)
187     {
188         ARM_COMPUTE_ERROR("Unsupported normalization axis");
189     }
190 
191     const auto *uk = get_implementation(L2NormalizeLayerSelectorData{ _output->info()->data_type(), _actual_axis, CPUInfo::get().get_isa() });
192     ARM_COMPUTE_ERROR_ON(uk == nullptr);
193     ARM_COMPUTE_ERROR_ON(uk->ukernel == nullptr);
194 
195     uk->ukernel(_input, _sum, _output, _epsilon, window, _actual_axis);
196 }
197 } // namespace arm_compute
198