aboutsummaryrefslogtreecommitdiff
path: root/arm_compute/runtime/CPP/functions/CPPSplit.h
blob: 611cc41157112ef13568efb2bf6127ae0b613547 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
/*
 * Copyright (c) 2020 ARM Limited.
 *
 * SPDX-License-Identifier: MIT
 *
 * Permission is hereby granted, free of charge, to any person obtaining a copy
 * of this software and associated documentation files (the "Software"), to
 * deal in the Software without restriction, including without limitation the
 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
 * sell copies of the Software, and to permit persons to whom the Software is
 * furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in all
 * copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */
#ifndef ARM_COMPUTE_CPP_SPLIT_H
#define ARM_COMPUTE_CPP_SPLIT_H

#include "arm_compute/core/Error.h"
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Types.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"

#include "support/ToolchainSupport.h"

#include "arm_compute/runtime/IFunction.h"

namespace arm_compute
{
/** Basic function to split a tensor along a given axis */
template <typename SliceType, typename TensorInterfaceType = ITensor>
class CPPSplit : public IFunction
{
public:
    CPPSplit()
        : _outputs_vector(), _slice_functions(), _num_outputs(0)
    {
    }
    /** Static function to check if given info will lead to a valid configuration of @ref CPPSplit
     *
     * @param[in] input   The input tensor info. Data types supported: All.
     * @param[in] outputs A vector containing the output tensors' info. Data types supported: same as @p input.
     *                    The output tensors should match the input tensor dimensions for all shape dimensions apart
     *                    from the split dimension
     * @param[in] axis    Axis on which to split the input.
     *
     * @return a status
     */
    static Status validate(const ITensorInfo *input, const std::vector<ITensorInfo *> &outputs, unsigned int axis)
    {
        ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input);
        ARM_COMPUTE_RETURN_ERROR_ON(axis >= input->num_dimensions());
        ARM_COMPUTE_RETURN_ERROR_ON(outputs.size() < 2);

        // Get output shape
        TensorShape  output_shape{};
        unsigned int total_output_shape_size = 0;

        // Sum the output sizes and fall back to evenly-sized splits if any are zero
        const bool using_split_shapes = std::none_of(outputs.begin(), outputs.end(), [&total_output_shape_size](ITensorInfo * info)
        {
            unsigned int output_shape_size = info->tensor_shape().total_size();
            total_output_shape_size += output_shape_size;
            return output_shape_size == 0;
        });

        if(using_split_shapes)
        {
            ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape().total_size() != total_output_shape_size);
        }
        else
        {
            output_shape = arm_compute::misc::shape_calculator::compute_split_shape(input, axis, outputs.size());
            ARM_COMPUTE_RETURN_ERROR_ON(output_shape.total_size() == 0);
        }

        // Validate output tensors
        unsigned int axis_offset = 0;
        for(const auto &output : outputs)
        {
            ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output);
            if(using_split_shapes)
            {
                output_shape = output->tensor_shape();
                ARM_COMPUTE_RETURN_ERROR_ON(output_shape.total_size() == 0);
            }

            const size_t axis_split_step = output_shape[axis];

            // Start/End coordinates
            Coordinates start_coords;
            Coordinates end_coords;
            for(unsigned int d = 0; d < output_shape.num_dimensions(); ++d)
            {
                end_coords.set(d, -1);
            }

            // Output auto inizialitation if not yet initialized
            TensorInfo tmp_output_info = *output->clone();
            auto_init_if_empty(tmp_output_info, input->clone()->set_is_resizable(true).set_tensor_shape(output_shape));

            // Update coordinate on axis
            start_coords.set(axis, axis_offset);
            end_coords.set(axis, axis_offset + axis_split_step);

            ARM_COMPUTE_RETURN_ON_ERROR(SliceType::validate(input, output, start_coords, end_coords));
            axis_offset += axis_split_step;
        }

        return Status{};
    }

    /** Initialise the kernel's input and outputs.
     *
     * @param[in]  input   The input tensor. Data types supported: All
     * @param[out] outputs A vector containing the output tensors. Data types supported: Same as @p input.
     *                     The output tensors should match the input tensor dimensions for all shape dimensions apart
     *                     from the split dimension.
     * @param[in]  axis    Axis on which to split the input.
     */
    void configure(const TensorInterfaceType *input, const std::vector<TensorInterfaceType *> &outputs, unsigned int axis)
    {
        // Create Slice functions
        _num_outputs = outputs.size();
        _slice_functions.resize(_num_outputs);

        // Extract output tensor info
        std::vector<ITensorInfo *> outputs_info;
        for(auto &output : outputs)
        {
            ARM_COMPUTE_ERROR_ON_NULLPTR(output);
            outputs_info.emplace_back(output->info());
        }

        // If any of the outputs have a zero size, fall-back to using evenly-sized output splits
        const bool outputs_have_sizes = std::none_of(outputs_info.begin(), outputs_info.end(), [](ITensorInfo * info)
        {
            return info->tensor_shape().total_size() == 0;
        });

        // Validate
        ARM_COMPUTE_ERROR_THROW_ON(CPPSplit::validate(input->info(), outputs_info, axis));

        unsigned int axis_offset = 0;
        unsigned int i           = 0;

        for(const auto &output_info : outputs_info)
        {
            // Get output shape
            TensorShape output_shape = (outputs_have_sizes ?
                                        output_info->tensor_shape() :
                                        arm_compute::misc::shape_calculator::compute_split_shape(input->info(), axis, _num_outputs));

            const size_t axis_split_step = output_shape[axis];

            // Start/End coordinates
            Coordinates start_coords;
            Coordinates end_coords;

            for(unsigned int d = 0; d < output_shape.num_dimensions(); ++d)
            {
                end_coords.set(d, -1);
            }

            // Update coordinate on axis
            start_coords.set(axis, axis_offset);
            end_coords.set(axis, axis_offset + axis_split_step);

            // Configure slice function
            _slice_functions[i].configure(input, outputs[i], start_coords, end_coords);

            // Set valid region from shape
            outputs[i]->info()->set_valid_region(ValidRegion(Coordinates(), output_shape));

            // Update axis offset
            axis_offset += axis_split_step;
            ++i;
        }
    }

protected:
    std::vector<TensorInterfaceType *> _outputs_vector;
    std::vector<SliceType>             _slice_functions;
    unsigned int                       _num_outputs;
};

} // namespace arm_compute
#endif /* ARM_COMPUTE_CPP_SPLIT_H */