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
|
/*
* Copyright (c) 2017 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/NEGEMMMatrixAccumulateBiasesKernel.h"
#include "arm_compute/core/AccessWindowStatic.h"
#include "arm_compute/core/Error.h"
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/ITensor.h"
#include "arm_compute/core/NEON/NEFixedPoint.h"
#include "arm_compute/core/Types.h"
#include "arm_compute/core/Validate.h"
#include "arm_compute/core/Window.h"
#include <arm_neon.h>
#include <cstddef>
#include <cstdint>
using namespace arm_compute;
NEGEMMMatrixAccumulateBiasesKernel::NEGEMMMatrixAccumulateBiasesKernel()
: _accum(nullptr), _biases(nullptr)
{
}
void NEGEMMMatrixAccumulateBiasesKernel::configure(ITensor *accum, const ITensor *biases)
{
ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(accum, 1, DataType::QS8, DataType::F32);
ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(biases, 1, DataType::QS8, DataType::F32);
ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(biases, accum);
ARM_COMPUTE_ERROR_ON(biases->info()->num_dimensions() != 1);
_biases = biases;
_accum = accum;
constexpr unsigned int num_elems_processed_per_iteration = 16;
// Configure kernel window
Window win = calculate_max_window(*accum->info(), Steps(num_elems_processed_per_iteration));
AccessWindowStatic biases_access(biases->info(), 0, 0, biases->info()->dimension(0), biases->info()->dimension(1));
update_window_and_padding(win,
AccessWindowHorizontal(accum->info(), 0, num_elems_processed_per_iteration),
biases_access);
AccessWindowHorizontal output_access(accum->info(), 0, num_elems_processed_per_iteration);
// Set the valid region for the accum tensor
Coordinates coord;
coord.set_num_dimensions(accum->info()->num_dimensions());
output_access.set_valid_region(win, ValidRegion(coord, accum->info()->tensor_shape()));
INEKernel::configure(win);
}
void NEGEMMMatrixAccumulateBiasesKernel::run(const Window &window)
{
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
Window win_biases;
win_biases.set(Window::DimX, Window::Dimension(window.x().start(), window.x().end(), window.x().step()));
win_biases.set(Window::DimY, Window::Dimension(0, 1, 1));
Iterator in0_out(_accum, window);
Iterator in1(_biases, win_biases);
switch(_accum->info()->data_type())
{
case DataType::F32:
{
execute_window_loop(window, [&](const Coordinates & id)
{
const float32x4x4_t accum = vld4q_f32(reinterpret_cast<const float *>(in0_out.ptr()));
const float32x4x4_t biases = vld4q_f32(reinterpret_cast<const float *>(in1.ptr()));
const float32x4x4_t res =
{
{
vaddq_f32(accum.val[0], biases.val[0]),
vaddq_f32(accum.val[1], biases.val[1]),
vaddq_f32(accum.val[2], biases.val[2]),
vaddq_f32(accum.val[3], biases.val[3])
}
};
vst4q_f32(reinterpret_cast<float *>(in0_out.ptr()), res);
},
in0_out, in1);
break;
}
case DataType::QS8:
{
execute_window_loop(window, [&](const Coordinates & id)
{
const qint8x16_t accum = vld1q_qs8(reinterpret_cast<const qint8_t *>(in0_out.ptr()));
const qint8x16_t biases = vld1q_qs8(reinterpret_cast<const qint8_t *>(in1.ptr()));
vst1q_qs8(reinterpret_cast<qint8_t *>(in0_out.ptr()), vqaddq_qs8(accum, biases));
},
in0_out, in1);
break;
}
default:
ARM_COMPUTE_ERROR("Data type not supported");
break;
}
}
|
