1 // Tencent is pleased to support the open source community by making ncnn available.
2 //
3 // Copyright (C) 2019 THL A29 Limited, a Tencent company. All rights reserved.
4 //
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 "tanh_x86.h"
16
17 #include "x86_activation.h"
18
19 #include <math.h>
20
21 namespace ncnn {
22
TanH_x86()23 TanH_x86::TanH_x86()
24 {
25 #if __SSE2__
26 support_packing = true;
27 #endif // __SSE2__
28 }
29
forward_inplace(Mat & bottom_top_blob,const Option & opt) const30 int TanH_x86::forward_inplace(Mat& bottom_top_blob, const Option& opt) const
31 {
32 int w = bottom_top_blob.w;
33 int h = bottom_top_blob.h;
34 int channels = bottom_top_blob.c;
35 int size = w * h;
36 #if __SSE2__
37 int elempack = bottom_top_blob.elempack;
38
39 #if __AVX__
40 if (elempack == 8)
41 {
42 #pragma omp parallel for num_threads(opt.num_threads)
43 for (int q = 0; q < channels; q++)
44 {
45 float* ptr = bottom_top_blob.channel(q);
46
47 for (int i = 0; i < size; i++)
48 {
49 __m256 _p = _mm256_loadu_ps(ptr);
50 _p = tanh_avx(_p);
51 _mm256_storeu_ps(ptr, _p);
52 ptr += 8;
53 }
54 }
55
56 return 0;
57 }
58 #endif // __AVX__
59
60 if (elempack == 4)
61 {
62 #pragma omp parallel for num_threads(opt.num_threads)
63 for (int q = 0; q < channels; q++)
64 {
65 float* ptr = bottom_top_blob.channel(q);
66
67 for (int i = 0; i < size; i++)
68 {
69 __m128 _p = _mm_loadu_ps(ptr);
70 _p = tanh_sse(_p);
71 _mm_storeu_ps(ptr, _p);
72 ptr += 4;
73 }
74 }
75
76 return 0;
77 }
78 #endif // __SSE2__
79
80 #pragma omp parallel for num_threads(opt.num_threads)
81 for (int q = 0; q < channels; q++)
82 {
83 float* ptr = bottom_top_blob.channel(q);
84
85 int i = 0;
86 #if __SSE2__
87 #if __AVX__
88 for (; i + 7 < size; i += 8)
89 {
90 __m256 _p = _mm256_loadu_ps(ptr);
91 _p = tanh_avx(_p);
92 _mm256_storeu_ps(ptr, _p);
93 ptr += 8;
94 }
95 #endif // __AVX__
96 for (; i + 3 < size; i += 4)
97 {
98 __m128 _p = _mm_loadu_ps(ptr);
99 _p = tanh_sse(_p);
100 _mm_storeu_ps(ptr, _p);
101 ptr += 4;
102 }
103 #endif // __SSE2__
104 for (; i < size; i++)
105 {
106 *ptr = tanh(*ptr);
107 ptr++;
108 }
109 }
110
111 return 0;
112 }
113
114 } // namespace ncnn
115