benchmark.md

---
layout: post
title: Benchmark 数据
---

可以参考[benchmark_tools](../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
      * HUAWEI Mate10
 
* 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