xref: /aosp_15_r20/external/ComputeLibrary/src/runtime/NEON/functions/NEReduceMean.cpp (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2018-2022 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/NEReduceMean.h"
25 
26 #include "arm_compute/core/Error.h"
27 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
28 #include "src/common/utils/Log.h"
29 #include "src/core/CPP/Validate.h"
30 #include "src/core/NEON/kernels/NEReductionOperationKernel.h"
31 #include "src/core/helpers/AutoConfiguration.h"
32 
33 namespace arm_compute
34 {
35 namespace
36 {
validate_config(const ITensorInfo * input,const Coordinates & reduction_axis,bool keep_dims,const ITensorInfo * output)37 Status validate_config(const ITensorInfo *input, const Coordinates &reduction_axis, bool keep_dims, const ITensorInfo *output)
38 {
39     ARM_COMPUTE_UNUSED(keep_dims);
40     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
41     ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input);
42     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8_SIGNED, DataType::QASYMM8, DataType::F16, DataType::F32);
43     ARM_COMPUTE_RETURN_ERROR_ON(reduction_axis.num_dimensions() < 1);
44     ARM_COMPUTE_RETURN_ERROR_ON(reduction_axis.num_dimensions() > input->num_dimensions());
45 
46     const unsigned int reduction_ops = reduction_axis.num_dimensions();
47     const int          input_dims    = input->num_dimensions();
48     Coordinates        axis_local    = reduction_axis;
49 
50     for(unsigned int i = 0; i < axis_local.num_dimensions(); ++i)
51     {
52         //axis: The dimensions to reduce. Must be in the range [-rank(input_tensor), rank(input_tensor)).
53         ARM_COMPUTE_RETURN_ERROR_ON(axis_local[i] < (-static_cast<int>(input->num_dimensions())));
54         ARM_COMPUTE_RETURN_ERROR_ON(axis_local[i] >= static_cast<int>(input->num_dimensions()));
55     }
56 
57     if(output->tensor_shape().total_size() != 0)
58     {
59         // Only validate if not using auto_init for the output tensor
60         TensorShape out_shape = input->tensor_shape();
61         // Validate output_shape only if not using auto_init
62         convert_negative_axis(axis_local, input_dims);
63         std::sort(axis_local.begin(), axis_local.begin() + reduction_ops);
64         for(unsigned int i = 0; i < reduction_ops; ++i)
65         {
66             ARM_COMPUTE_RETURN_ERROR_ON(axis_local[i] > 3);
67             ARM_COMPUTE_RETURN_ERROR_ON(static_cast<unsigned int>(axis_local[i]) > input->num_dimensions() - 1);
68             if(output->total_size() > 0 && keep_dims)
69             {
70                 ARM_COMPUTE_RETURN_ERROR_ON(output->dimension(axis_local[i]) != 1);
71             }
72             if(keep_dims)
73             {
74                 out_shape.set(axis_local[i], 1);
75             }
76             else
77             {
78                 ARM_COMPUTE_RETURN_ERROR_ON(i > static_cast<unsigned int>(axis_local[i]));
79                 const unsigned int remove_index = axis_local[i] - i;
80                 ARM_COMPUTE_RETURN_ERROR_ON(remove_index >= out_shape.num_dimensions());
81                 out_shape.remove_dimension(remove_index);
82             }
83         }
84         const TensorInfo out_info = input->clone()->set_tensor_shape(out_shape);
85         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(output, &out_info);
86     }
87     return Status{};
88 }
89 } // namespace
90 
91 NEReduceMean::~NEReduceMean() = default;
92 
NEReduceMean(std::shared_ptr<IMemoryManager> memory_manager)93 NEReduceMean::NEReduceMean(std::shared_ptr<IMemoryManager> memory_manager)
94     : _memory_group(std::move(memory_manager)), _reduction_kernels(), _reduced_outs(), _reshape(), _reduction_ops(), _keep_dims()
95 {
96 }
97 
validate(const ITensorInfo * input,const Coordinates & reduction_axis,bool keep_dims,const ITensorInfo * output)98 Status NEReduceMean::validate(const ITensorInfo *input, const Coordinates &reduction_axis, bool keep_dims, const ITensorInfo *output)
99 {
100     return validate_config(input, reduction_axis, keep_dims, output);
101 }
102 
configure(ITensor * input,const Coordinates & reduction_axis,bool keep_dims,ITensor * output)103 void NEReduceMean::configure(ITensor *input, const Coordinates &reduction_axis, bool keep_dims, ITensor *output)
104 {
105     ARM_COMPUTE_LOG_PARAMS(input, reduction_axis, keep_dims, output);
106 
107     // Perform validate step
108     ARM_COMPUTE_ERROR_THROW_ON(NEReduceMean::validate(input->info(), reduction_axis, keep_dims, output->info()));
109     // Output auto inizialitation if not yet initialized
110     const TensorShape output_shape = arm_compute::misc::shape_calculator::calculate_reduce_mean_shape(input->info(), reduction_axis, keep_dims);
111     auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(output_shape));
112 
113     _reduction_ops = reduction_axis.num_dimensions();
114     _reduction_kernels.resize(_reduction_ops);
115     _reduced_outs.resize(_reduction_ops - (keep_dims ? 1 : 0));
116     _keep_dims = keep_dims;
117 
118     ITensor *tmp_input  = input;
119     ITensor *tmp_output = output;
120 
121     Coordinates axis_local = reduction_axis;
122     const int   input_dims = tmp_input->info()->num_dimensions();
123 
124     convert_negative_axis(axis_local, input_dims);
125 
126     // Perform reduction for every axis
127     for(int i = 0; i < _reduction_ops; ++i)
128     {
129         TensorShape out_shape = i == 0 ? tmp_input->info()->tensor_shape() : (&_reduced_outs[i - 1])->info()->tensor_shape();
130         out_shape.set(axis_local[i], 1);
131         auto in = (i == 0) ? tmp_input : (&_reduced_outs[i - 1]);
132 
133         if(i == _reduction_ops - 1 && keep_dims)
134         {
135             _reduction_kernels[i].configure(in, tmp_output, axis_local[i], ReductionOperation::MEAN_SUM);
136         }
137         else
138         {
139             _reduced_outs[i].allocator()->init(TensorInfo(out_shape, tmp_output->info()->num_channels(), tmp_output->info()->data_type(), tmp_output->info()->quantization_info()));
140             _memory_group.manage(&_reduced_outs[i]);
141             _reduction_kernels[i].configure(in, &_reduced_outs[i], axis_local[i], ReductionOperation::MEAN_SUM);
142         }
143     }
144 
145     // Allocate intermediate tensors
146     for(int i = 0; i < _reduction_ops - (keep_dims ? 1 : 0); ++i)
147     {
148         _reduced_outs[i].allocator()->allocate();
149     }
150     // Configure reshape layer if we want to drop the dimensions
151     if(!keep_dims)
152     {
153         TensorShape out_shape = tmp_input->info()->tensor_shape();
154         // We have to sort the reduction axis vectors in order for remove_dimension
155         // to work properly
156         std::sort(axis_local.begin(), axis_local.begin() + _reduction_ops);
157         for(int i = 0; i < _reduction_ops; ++i)
158         {
159             out_shape.remove_dimension(axis_local[i] - i);
160         }
161         auto_init_if_empty(*tmp_output->info(), tmp_input->info()->clone()->set_tensor_shape(out_shape));
162         _reshape.configure(&_reduced_outs[_reduction_ops - 1], tmp_output);
163     }
164 }
165 
run()166 void NEReduceMean::run()
167 {
168     MemoryGroupResourceScope scope_mg(_memory_group);
169     for(auto &kernel : _reduction_kernels)
170     {
171         kernel.run();
172     }
173     if(!_keep_dims)
174     {
175         _reshape.run();
176     }
177 }
178 } // namespace arm_compute
179