ncnn

1
// yala is pleased to support the open source community by making ncnn available.
2
//
3
//
4
// Copyright (C) 2022 yala <zhaojunchao@loongson.cn>;<junchao82@qq.com>. All rights reserved.
5
// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
6
// in compliance with the License. You may obtain a copy of the License at
7
//
8
// https://opensource.org/licenses/BSD-3-Clause
9
//
10
// Unless required by applicable law or agreed to in writing, software distributed
11
// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
12
// CONDITIONS OF ANY KIND, either express or implied. See the License for the
13
// specific language governing permissions and limitations under the License.
14

15
#include "hardsigmoid_loongarch.h"
16

17
#if __loongarch_sx
18
#include <lsxintrin.h>
19
#endif // __loongarch_sx
20

21
#include "loongarch_usability.h"
22

23
namespace ncnn {
24

25
HardSigmoid_loongarch::HardSigmoid_loongarch()
26
{
27
#if __loongarch_sx
28
    support_packing = true;
29
#endif // __loongarch_sx
30
}
31

32
int HardSigmoid_loongarch::forward_inplace(Mat& bottom_top_blob, const Option& opt) const
33
{
34
    int w = bottom_top_blob.w;
35
    int h = bottom_top_blob.h;
36
    int d = bottom_top_blob.d;
37
    int channels = bottom_top_blob.c;
38
    int elempack = bottom_top_blob.elempack;
39
    int size = w * h * d * elempack;
40

41
    #pragma omp parallel for num_threads(opt.num_threads)
42
    for (int q = 0; q < channels; q++)
43
    {
44
        float* ptr = bottom_top_blob.channel(q);
45

46
        int i = 0;
47
#if __loongarch_sx
48
        __m128 _zero = (__m128)__lsx_vreplgr2vr_w(0);
49
        __m128 _one = (__m128)__lsx_vreplfr2vr_s(1.f);
50
        __m128 _alpha = (__m128)__lsx_vreplfr2vr_s(alpha);
51
        __m128 _beta = (__m128)__lsx_vreplfr2vr_s(beta);
52
        for (; i + 3 < size; i += 4)
53
        {
54
            __builtin_prefetch(ptr + 16);
55
            __m128 _p = (__m128)__lsx_vld(ptr, 0);
56
            _p = __lsx_vfmadd_s(_alpha, _p, _beta);
57
            _p = __lsx_vfmax_s(_p, _zero);
58
            _p = __lsx_vfmin_s(_p, _one);
59
            __lsx_vst(_p, ptr, 0);
60

61
            ptr += 4;
62
        }
63
#endif // __loongarch_sx
64
        for (; i < size; i++)
65
        {
66
            if (*ptr < lower)
67
                *ptr = 0.f;
68
            else if (*ptr > upper)
69
                *ptr = 1.f;
70
            else
71
                *ptr = *ptr * alpha + beta;
72
            ++ptr;
73
        }
74
    }
75

76
    return 0;
77
}
78

79
} // namespace ncnn
80