File size: 3,195 Bytes
be903e2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
// Tencent is pleased to support the open source community by making ncnn available.
//
// Copyright (C) 2017 THL A29 Limited, a Tencent company. All rights reserved.
//
// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
// in compliance with the License. You may obtain a copy of the License at
//
// https://opensource.org/licenses/BSD-3-Clause
//
// Unless required by applicable law or agreed to in writing, software distributed
// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
// CONDITIONS OF ANY KIND, either express or implied. See the License for the
// specific language governing permissions and limitations under the License.

#include "absval_arm.h"

#if __ARM_NEON
#include <arm_neon.h>
#endif // __ARM_NEON

namespace ncnn {

AbsVal_arm::AbsVal_arm()
{
#if __ARM_NEON
    support_packing = true;
#endif // __ARM_NEON
}

int AbsVal_arm::forward_inplace(Mat& bottom_top_blob, const Option& opt) const
{
    int w = bottom_top_blob.w;
    int h = bottom_top_blob.h;
    int d = bottom_top_blob.d;
    int channels = bottom_top_blob.c;
    int elempack = bottom_top_blob.elempack;
    int size = w * h * d * elempack;

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

        int i = 0;
#if __ARM_NEON
        for (; i + 15 < size; i += 16)
        {
#if __aarch64__
            asm volatile(
                "prfm   pldl1keep, [%0, #512]   \n"
                "ld1    {v0.4s, v1.4s, v2.4s, v3.4s}, [%0] \n"
                "fabs   v0.4s, v0.4s            \n"
                "fabs   v1.4s, v1.4s            \n"
                "fabs   v2.4s, v2.4s            \n"
                "fabs   v3.4s, v3.4s            \n"
                "st1    {v0.4s, v1.4s, v2.4s, v3.4s}, [%0], #64 \n"
                : "=r"(ptr) // %0
                : "0"(ptr)
                : "memory", "v0", "v1", "v2", "v3");
#else  // __aarch64__
            asm volatile(
                "pld        [%0, #512]      \n"
                "vldm       %0, {d0-d7}     \n"
                "vabs.f32   q0, q0          \n"
                "vabs.f32   q1, q1          \n"
                "vabs.f32   q2, q2          \n"
                "vabs.f32   q3, q3          \n"
                "vstm       %0!, {d0-d7}    \n"
                : "=r"(ptr) // %0
                : "0"(ptr)
                : "memory", "q0", "q1", "q2", "q3");
#endif // __aarch64__
        }
        for (; i + 7 < size; i += 8)
        {
            float32x4_t _p0 = vld1q_f32(ptr);
            float32x4_t _p1 = vld1q_f32(ptr + 4);
            _p0 = vabsq_f32(_p0);
            _p1 = vabsq_f32(_p1);
            vst1q_f32(ptr, _p0);
            vst1q_f32(ptr + 4, _p1);
            ptr += 8;
        }
        for (; i + 3 < size; i += 4)
        {
            float32x4_t _p = vld1q_f32(ptr);
            _p = vabsq_f32(_p);
            vst1q_f32(ptr, _p);
            ptr += 4;
        }
#endif // __ARM_NEON
        for (; i < size; i++)
        {
            *ptr = *ptr > 0 ? *ptr : -*ptr;

            ptr++;
        }
    }

    return 0;
}

} // namespace ncnn