可以参考benchmark_tools，推荐一键benchmark。
测试环境
- 测试模型
- fp32模型
- mobilenet_v1
- mobilenet_v2
- squeezenet_v1.1
- mnasnet
- shufflenet_v2
- int8模型
- mobilenet_v1
- mobilenet_v2
- resnet50
- 测试机器(android ndk ndk-r17c)
- 骁龙855
- xiaomi mi9, snapdragon 855
- 4xA76(1@2.84GHz + 3@2.4GHz) + 4xA55@1.78GHz
- 骁龙845
- xiaomi mi8, 845
- 2.8GHz（大四核），1.7GHz（小四核）
- 骁龙835
- xiaomi mix2, snapdragon 835
- 2.45GHz（大四核），1.9GHz（小四核）
- 骁龙625
- oppo R9s, snapdragon625
- A53 x 8, big core@2.0GHz
- 骁龙653
- 360 N5, snapdragon 653
- 4 x A73@2.0GHz + 4 x A53@1.4GHz
- 麒麟970
- 测试说明
- commit id: 12c129affaacd476e27a0a82b235a9d547d33f0f
- warmup=10, repeats=30，统计平均时间，单位是ms
- 当线程数为1时，
DeviceInfo::Global().SetRunMode设置LITE_POWER_HIGH，否者设置LITE_POWER_NO_BIND
- 模型的输入图像的维度是{1, 3, 224, 224}，输入图像的每一位数值是1
测试数据
fp32 模型测试数据
测试数据
fp32 模型测试数据
| 骁龙855 |
armv8 |
|
|
armv7 |
|
|
| num_threads |
1 |
2 |
4 |
1 |
2 |
4 |
| mobilenet_v1 |
31.64 |
18.98 |
10.67 |
33.17 |
19.55 |
11.43 |
| mobilenet_v2 |
25.54 |
13.80 |
8.75 |
29.25 |
15.19 |
9.65 |
| squeezenet_v1.1 |
26.81 |
14.39 |
8.92 |
28.63 |
15.37 |
9.53 |
| mnasnet |
25.39 |
13.89 |
9.63 |
28.97 |
15.54 |
10.10 |
| shufflenet_v2 |
13.85 |
7.81 |
5.87 |
14.64 |
8.35 |
6.14 |
| 骁龙845 |
armv8 |
|
|
armv7 |
|
|
| num_threads |
1 |
2 |
4 |
1 |
2 |
4 |
| mobilenet_v1 |
62.04 |
33.63 |
18.63 |
66.23 |
35.78 |
20.14 |
| mobilenet_v2 |
40.41 |
22.94 |
13.33 |
44.22 |
24.58 |
14.50 |
| squeezenet_v1.1 |
49.92 |
23.78 |
13.86 |
52.00 |
24.85 |
15.87 |
| mnasnet |
40.14 |
23.36 |
14.46 |
43.77 |
24.78 |
14.76 |
| shufflenet_v2 |
22.27 |
13.69 |
8.96 |
26.11 |
14.95 |
9.02 |
| 骁龙835 |
armv8 |
|
|
armv7 |
|
|
| num_threads |
1 |
2 |
4 |
1 |
2 |
4 |
| mobilenet_v1 |
89.57 |
50.88 |
27.62 |
96.11 |
53.18 |
31.99 |
| mobilenet_v2 |
59.92 |
33.93 |
20.91 |
64.04 |
36.85 |
23.10 |
| squeezenet_v1.1 |
65.25 |
37.92 |
23.40 |
74.87 |
40.96 |
23.69 |
| mnasnet |
60.97 |
35.04 |
22.40 |
64.88 |
37.90 |
24.53 |
| shufflenet_v2 |
30.87 |
19.33 |
12.78 |
31.71 |
19.52 |
13.25 |
| 骁龙625 |
armv8 |
|
|
armv7 |
|
|
| num_threads |
1 |
2 |
4 |
1 |
2 |
4 |
| mobilenet_v1 |
180.98 |
92.27 |
51.51 |
216.12 |
110.33 |
61.68 |
| mobilenet_v2 |
132.46 |
68.38 |
43.54 |
146.18 |
76.62 |
46.21 |
| squeezenet_v1.1 |
124.49 |
66.84 |
41.53 |
153.28 |
82.42 |
47.14 |
| mnasnet |
122.50 |
67.46 |
43.04 |
146.20 |
79.64 |
48.56 |
| shufflenet_v2 |
68.70 |
40.77 |
26.53 |
75.38 |
42.40 |
28.36 |
| 骁龙653 |
armv8 |
|
|
armv7 |
|
|
| num_threads |
1 |
2 |
4 |
1 |
2 |
4 |
| mobilenet_v1 |
121.27 |
59.36 |
34.06 |
126.55 |
64.96 |
39.23 |
| mobilenet_v2 |
79.48 |
46.17 |
27.81 |
87.93 |
48.28 |
31.87 |
| squeezenet_v1.1 |
81.10 |
42.66 |
42.07 |
82.29 |
45.88 |
28.84 |
| mnasnet |
75.60 |
44.22 |
30.16 |
82.99 |
49.07 |
32.34 |
| shufflenet_v2 |
39.18 |
23.54 |
16.73 |
40.12 |
24.76 |
17.68 |
| 麒麟970 |
armv8 |
|
|
armv7 |
|
|
| num_threads |
1 |
2 |
4 |
1 |
2 |
4 |
| mobilenet_v1 |
99.58 |
56.91 |
29.02 |
102.42 |
57.81 |
35.36 |
| mobilenet_v2 |
69.22 |
42.41 |
23.55 |
69.49 |
43.38 |
25.26 |
| squeezenet_v1.1 |
67.48 |
41.06 |
24.47 |
75.03 |
43.57 |
26.35 |
| mnasnet |
74.55 |
43.06 |
24.22 |
75.48 |
44.43 |
26.69 |
| shufflenet_v2 |
39.20 |
24.54 |
16.34 |
37.40 |
24.32 |
16.66 |
int8 模型测试数据
| 骁龙855 |
armv8 |
|
|
armv7 |
|
|
| num_threads |
1 |
2 |
4 |
1 |
2 |
4 |
| mobilenet_v1_int8 |
16.77 |
8.38 |
4.59 |
43.42 |
20.80 |
10.89 |
| mobilenet_v2_int8 |
22.81 |
13.71 |
10.43 |
29.65 |
20.09 |
13.99 |
| resnet50_int8 |
258.83 |
157.22 |
85.83 |
424.99 |
209.37 |
112.32 |
| 骁龙845 |
armv8 |
|
|
armv7 |
|
|
| num_threads |
1 |
2 |
4 |
1 |
2 |
4 |
| mobilenet_v1_int8 |
44.08 |
23.75 |
12.52 |
49.19 |
26.77 |
13.82 |
| mobilenet_v2_int8 |
36.61 |
22.70 |
15.29 |
40.51 |
25.84 |
17.89 |
| resnet50_int8 |
399.64 |
217.74 |
112.86 |
408.80 |
224.72 |
122.15 |
| 骁龙835 |
armv8 |
|
|
armv7 |
|
|
| num_threads |
1 |
2 |
4 |
1 |
2 |
4 |
| mobilenet_v1_int8 |
59.99 |
31.59 |
16.55 |
62.92 |
33.33 |
17.38 |
| mobilenet_v2_int8 |
50.68 |
31.25 |
21.62 |
52.56 |
33.88 |
24.31 |
| resnet50_int8 |
498.85 |
267.65 |
146.03 |
510.54 |
278.77 |
155.05 |
| 骁龙625 |
armv8 |
|
|
armv7 |
|
|
| num_threads |
1 |
2 |
4 |
1 |
2 |
4 |
| mobilenet_v1_int8 |
122.86 |
63.52 |
33.91 |
125.77 |
64.78 |
34.25 |
| mobilenet_v2_int8 |
110.71 |
67.76 |
49.85 |
114.63 |
71.74 |
51.73 |
| resnet50_int8 |
954.67 |
505.78 |
286.64 |
1016.64 |
532.84 |
305.20 |
| 骁龙653 |
armv8 |
|
|
armv7 |
|
|
| num_threads |
1 |
2 |
4 |
1 |
2 |
4 |
| mobilenet_v1_int8 |
81.46 |
42.99 |
31.69 |
81.20 |
42.46 |
23.47 |
| mobilenet_v2_int8 |
68.39 |
43.47 |
32.03 |
69.40 |
44.47 |
33.46 |
| resnet50_int8 |
687.59 |
369.70 |
208.99 |
684.55 |
369.04 |
208.42 |
| 麒麟970 |
armv8 |
|
|
armv7 |
|
|
| num_threads |
1 |
2 |
4 |
1 |
2 |
4 |
| mobilenet_v1_int8 |
64.27 |
35.48 |
18.76 |
64.63 |
37.67 |
20.70 |
| mobilenet_v2_int8 |
64.54 |
36.76 |
22.17 |
68.80 |
38.85 |
24.30 |
| resnet50_int8 |
509.94 |
268.95 |
276.13 |
520.57 |
281.92 |
157.82 |