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