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
|
/*
* Copyright (c) 2019 Samsung Electronics Co., Ltd. All Rights Reserved
* 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/NEEmbeddingLookupKernel.h"
#include "arm_compute/core/Error.h"
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/ITensor.h"
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Validate.h"
#include "arm_compute/core/Window.h"
using namespace arm_compute;
NEEmbeddingLookupKernel::NEEmbeddingLookupKernel()
: _input(nullptr), _lookups(nullptr), _output(nullptr)
{
}
void NEEmbeddingLookupKernel::configure(const ITensor *input, ITensor *output,
const ITensor *lookups)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_ERROR_THROW_ON(validate(input->info(), output->info(), lookups->info()));
_input = input;
_output = output;
_lookups = lookups;
// Auto initialize output if not initialized
auto out_shape = input->info()->tensor_shape();
out_shape.set(out_shape.num_dimensions() - 1, lookups->info()->num_dimensions());
auto_init_if_empty(*output->info(), out_shape, 1, input->info()->data_type(),
input->info()->quantization_info());
INEKernel::configure(calculate_max_window(*output->info()));
}
Status NEEmbeddingLookupKernel::validate(const arm_compute::ITensorInfo *input,
const arm_compute::ITensorInfo *output,
const arm_compute::ITensorInfo *lookups)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output, lookups);
ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(
input, 1, DataType::U8, DataType::S8, DataType::QASYMM8, DataType::U16, DataType::S16,
DataType::U32, DataType::S32, DataType::F16, DataType::F32);
ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(lookups, 1, DataType::S32);
ARM_COMPUTE_ERROR_ON(input->num_dimensions() < 2 && input->num_dimensions() > 4);
ARM_COMPUTE_ERROR_ON(lookups->num_dimensions() > 1);
// Validate in case of configured output
if (output->total_size() > 0)
{
ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
ARM_COMPUTE_ERROR_ON(input->num_dimensions() != output->num_dimensions());
ARM_COMPUTE_ERROR_ON(output->dimension(output->num_dimensions() - 1) != lookups->dimension(0));
for (size_t i = 0; i < output->num_dimensions() - 1; ++i)
{
ARM_COMPUTE_ERROR_ON(input->dimension(i) != output->dimension(i));
}
}
return Status{};
}
void NEEmbeddingLookupKernel::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);
const size_t lookup_dim = _output->info()->num_dimensions() - 1;
Window output_window{window};
output_window.set(Window::DimX,
Window::Dimension(output_window.x().start(), output_window.x().end(),
_input->info()->dimension(0)));
Window out_slice = output_window.first_slice_window_4D();
do
{
Iterator output_it(_output, out_slice);
execute_window_loop(out_slice,
[&](const Coordinates &id) {
const int32_t lookup = *reinterpret_cast<int32_t *>(
_lookups->ptr_to_element(Coordinates{id[lookup_dim]}));
Coordinates input_id{id};
input_id.set(lookup_dim, lookup);
memcpy(output_it.ptr(), _input->ptr_to_element(input_id),
_output->info()->dimension(0) * _output->info()->element_size());
},
output_it);
} while (window.slide_window_slice_4D(out_slice));
}
|