aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/NEStackLayerKernel.cpp
blob: d314fb1dd18f2193db2dc76644df84184cfe977c (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
/*
 * Copyright (c) 2018-2019 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.
 */
#include "arm_compute/core/NEON/kernels/NEStackLayerKernel.h"

#include "arm_compute/core/Error.h"
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/IAccessWindow.h"
#include "arm_compute/core/ITensor.h"
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/Validate.h"
#include "arm_compute/core/Window.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"

#include "support/ToolchainSupport.h"

using namespace arm_compute;
using namespace arm_compute::misc::shape_calculator;

namespace
{
Status validate_arguments(const ITensorInfo *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, const ITensorInfo *output)
{
    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::U8, DataType::S8,
                                                         DataType::U16, DataType::S16, DataType::U32, DataType::S32,
                                                         DataType::F16, DataType::F32);
    ARM_COMPUTE_RETURN_ERROR_ON(idx_input >= num_tensors);
    ARM_COMPUTE_RETURN_ERROR_ON(axis > input->num_dimensions());
    ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 4);

    if(output->total_size() != 0)
    {
        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), compute_stack_shape(*input, axis, num_tensors));
        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
    }

    return Status{};
}

std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, unsigned int axis, unsigned int num_tensors, ITensorInfo *output)
{
    // Output auto inizialitation if not yet initialized
    auto_init_if_empty(*output, input->clone()->set_tensor_shape(compute_stack_shape(*input, axis, num_tensors)));

    // Configure kernel window
    Window win = calculate_max_window(*input);

    return std::make_pair(Status{}, win);
}

inline Coordinates shift_from_axis_and_replace_coordinate(const Coordinates &id, unsigned int axis, unsigned int idx_input)
{
    constexpr int max_out_coord = 5; // Input shape is max a 4D shape, output is max 5D
    Coordinates   id_out        = id;
    for(unsigned int i = max_out_coord - 1; i > axis; --i)
    {
        id_out.set(i, id[i - 1]);
    }
    id_out.set(axis, idx_input);
    return id_out;
}
} // namespace

NEStackLayerKernel::NEStackLayerKernel()
    : _input(nullptr), _output(nullptr), _axis(), _idx_input(), _func(nullptr)
{
}

void NEStackLayerKernel::configure(const ITensor *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, ITensor *output)
{
    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), axis, idx_input, num_tensors, output->info()));

    _input     = input;
    _output    = output;
    _axis      = axis;
    _idx_input = idx_input;

    switch(input->info()->element_size())
    {
        case 1:
            _func = &NEStackLayerKernel::run_stack<uint8_t>;
            break;
        case 2:
            _func = &NEStackLayerKernel::run_stack<uint16_t>;
            break;
        case 4:
            _func = &NEStackLayerKernel::run_stack<uint32_t>;
            break;
        default:
            ARM_COMPUTE_ERROR("Element size not supported");
            break;
    }

    // Configure kernel window
    auto win_config = validate_and_configure_window(input->info(), axis, num_tensors, output->info());

    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
    INEKernel::configure(win_config.second);
}

Status NEStackLayerKernel::validate(const ITensorInfo *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, const ITensorInfo *output)
{
    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, axis, idx_input, num_tensors, output));
    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), axis, num_tensors, output->clone().get()).first);
    return Status{};
}

void NEStackLayerKernel::run(const Window &window, const ThreadInfo &info)
{
    ARM_COMPUTE_UNUSED(info);
    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);

    if(_func != nullptr)
    {
        (this->*_func)(window);
    }
}

template <typename T>
void NEStackLayerKernel::run_stack(const Window &window)
{
    Window window_out;
    window_out.use_tensor_dimensions(_output->info()->tensor_shape());

    Iterator input(_input, window);
    Iterator output(_output, window_out);

    const int stride_x = _output->info()->strides_in_bytes()[0];
    const int stride_y = _output->info()->num_dimensions() >= 1 ? _output->info()->strides_in_bytes()[1] : 0;
    const int stride_z = _output->info()->num_dimensions() >= 2 ? _output->info()->strides_in_bytes()[2] : 0;
    const int stride_w = _output->info()->num_dimensions() >= 3 ? _output->info()->strides_in_bytes()[3] : 0;
    const int stride_k = _output->info()->num_dimensions() >= 4 ? _output->info()->strides_in_bytes()[4] : 0;

    execute_window_loop(window, [&](const Coordinates & id)
    {
        Coordinates id_out                           = shift_from_axis_and_replace_coordinate(id, _axis, _idx_input);
        const int   idx                              = id_out[0] * stride_x + id_out[1] * stride_y + id_out[2] * stride_z + id_out[3] * stride_w + id_out[4] * stride_k;
        *(reinterpret_cast<T *>(output.ptr() + idx)) = *(reinterpret_cast<const T *>(input.ptr()));
    },
    input);
}