• DMA性能测试

        本程序主要用来计算DMA数据读写过程中所花费的总得时间周期,依据公式T=tStart+ceil(L/4)*2+ceil(L/256)*tTransform*2

    因为tTransform是一个常量（通常默认为11），因此只需根据debug文件分析出tStart即可。

         一开始我以为tStart是一个变量，是buffer2-2（读）的开始时刻与准备时刻最后一个buffer2-17（写）的差值，因此写了

    以下程序：

     1 #*************************************************************************
 2 
 3     # File Name: performance_testing.py
 4     # Author: jiangjing
 5     # Mail: jjdl@mail.ustc.edu.cn 
 6     # Created Time: Sat 15 Jul 2017 09:34:04 PM PDT
 7 #************************************************************************
 8 import math
 9 import sys
10 
11 def GetFirstBuffer2_2(path):
12     fDebug=open(path).read() 
13     index2_2=fDebug.index("[buffer2-2]")
14     index_space=fDebug.index(" ",index2_2+17)
15     return [index2_2,fDebug[index2_2+17:index_space]]
16 
17 def GetLastBuffer2_17(path,index2_2):
18     fDebug=open(path).read() 
19     index2_17=fDebug.rindex("[buffer2-17]",0,index2_2)
20     index_space=fDebug.index(" ",index2_17+18)
21     return [index2_17,fDebug[index2_17+18:index_space]]
22 
23 def GetTotalTime(L,path,tTransform=11): #the 'path' is the paht of debug file
24     buffer2_2=GetFirstBuffer2_2(path)
25     buffer2_17=GetLastBuffer2_17(path,buffer2_2[0])
26     return int(buffer2_2[1])-int(buffer2_17[1])+math.ceil(L/4)+math.ceil(L/256)*tTransform*2
27 
28 #the code just for debug,you can comment the following code
29 if __name__=='__main__':
30     if(len(sys.argv)>1):
31         print(GetTotalTime(int(sys.argv[1]),sys.argv[2]))
    View Code

         但师兄说其实tStart也是一个常量，所以又将程序简化了以下：

     1 #*************************************************************************
 2     # File Name: PerformanceTesting.py
 3     # Author: jiangjing
 4     # Mail: jjdl@mail.ustc.edu.cn 
 5     # Created Time: Sun 16 Jul 2017 06:12:22 AM PDT
 6 #************************************************************************
 7 import math
 8 
 9 tStart=2448
10 def GetTime(L,tTransform=11):
11     return tStart+math.ceil(L/4)+math.ceil(L/256)*tTransform*2
12 
13 def GetTotalTime(lList):
14     totalTime=0
15     for l in lList:
16         L=l[4]
17         totalTime+=GetTime(L)
18     return totalTime
    View Code

        下面的这个test.py为测试样例：

     1 #*************************************************************************
 2     # File Name: test.py
 3     # Author: jiangjing
 4     # Mail: jjdl@mail.ustc.edu.cn 
 5     # Created Time: Sun 16 Jul 2017 06:18:24 AM PDT
 6 #************************************************************************
 7 import PerformanceTesting as PT
 8 
 9 lList=[[0,0,0,0,65555],[0,0,0,0,672323]]
10 print(PT.GetTotalTime(lList))
    View Code

        考虑到测试的便捷性，师兄建议我采用c语言进行编写。

    　一下为C语言版本代码：

     1 #include"stdio.h"
 2 #include"math.h"
 3 #include"conv2_route_plan.c"
 4 #include"route_plan.h"
 5 #define uint32_t unsigned long
 6 
 7 uint32_t tStart=2448;
 8 uint32_t GetTime(uint32_t L){
 9     return tStart+ceil(L/4)*2+ceil(L/256)*11*2;
10 }
11 
12 uint32_t GetTotalTime(route_plan_t lList){
13     uint32_t totalTime=0;
14     uint32_t i;
15     for(i=0;i<lList.num_item;i++)
16     {
17         totalTime+=GetTime(lList.route_item_array[i].len);
18     }
19     return totalTime;
20 }
21 
22 int main()
23 {
24     printf("%d
",GetTotalTime(conv2_route_plan));
25     return 0;
26 }
    View Code

    　　其中“route_plan.h“是从师兄的项目中拷下来的，里面定义了输出数据的格式。

     1 #ifndef _ROUTE_PLAN_H_
 2 #define _ROUTE_PLAN_H_
 3 
 4 #include <stdint.h>
 5 
 6 typedef struct {
 7   uint32_t src_buffer_id;
 8   uint32_t src_buffer_offset;
 9   uint32_t dest_buffer_id;
10   uint32_t dest_buffer_offset;
11   size_t   len;
12 } route_item_t;
13 
14 typedef struct {
15   uint32_t num_item;
16   route_item_t* route_item_array;
17 } route_plan_t;
18 #endif // #ifndef _ROUTE_PLAN_H_
    View Code

    　　其中“conv2_route_plan.c”文件是main.py文件运行输出的结果之一，里面包含了计算计算clock必要的数据。通过改变dsp的数目，本文件中的数据也会发生变化。

    以下是dsp=32时的数据情况：

    #include "route_plan.h"

route_item_t conv2_route_item_array[] = {
  {0, 0.0, 32, 57856.0, 404992.0},
  {1, 0.0, 32, 462848.0, 57856.0},
  {0, 347136.0, 33, 0.0, 57856.0},
  {1, 0.0, 33, 57856.0, 404992.0},
  {2, 0.0, 33, 462848.0, 57856.0},
  {1, 347136.0, 34, 0.0, 57856.0},
  {2, 0.0, 34, 57856.0, 404992.0},
  {3, 0.0, 34, 462848.0, 57856.0},
  {2, 347136.0, 35, 0.0, 57856.0},
  {3, 0.0, 35, 57856.0, 404992.0},
  {4, 0.0, 35, 462848.0, 57856.0},
  {3, 347136.0, 36, 0.0, 57856.0},
  {4, 0.0, 36, 57856.0, 404992.0},
  {5, 0.0, 36, 462848.0, 57856.0},
  {4, 347136.0, 37, 0.0, 57856.0},
  {5, 0.0, 37, 57856.0, 404992.0},
  {6, 0.0, 37, 462848.0, 57856.0},
  {5, 347136.0, 38, 0.0, 57856.0},
  {6, 0.0, 38, 57856.0, 404992.0},
  {7, 0.0, 38, 462848.0, 57856.0},
  {6, 347136.0, 39, 0.0, 57856.0},
  {7, 0.0, 39, 57856.0, 404992.0},
  {8, 0.0, 39, 462848.0, 57856.0},
  {7, 347136.0, 40, 0.0, 57856.0},
  {8, 0.0, 40, 57856.0, 404992.0},
  {9, 0.0, 40, 462848.0, 57856.0},
  {8, 347136.0, 41, 0.0, 57856.0},
  {9, 0.0, 41, 57856.0, 404992.0},
  {10, 0.0, 41, 462848.0, 57856.0},
  {9, 347136.0, 42, 0.0, 57856.0},
  {10, 0.0, 42, 57856.0, 404992.0},
  {11, 0.0, 42, 462848.0, 57856.0},
  {10, 347136.0, 43, 0.0, 57856.0},
  {11, 0.0, 43, 57856.0, 404992.0},
  {12, 0.0, 43, 462848.0, 57856.0},
  {11, 347136.0, 44, 0.0, 57856.0},
  {12, 0.0, 44, 57856.0, 404992.0},
  {13, 0.0, 44, 462848.0, 57856.0},
  {12, 347136.0, 45, 0.0, 57856.0},
  {13, 0.0, 45, 57856.0, 404992.0},
  {14, 0.0, 45, 462848.0, 57856.0},
  {13, 347136.0, 46, 0.0, 57856.0},
  {14, 0.0, 46, 57856.0, 404992.0},
  {15, 0.0, 46, 462848.0, 57856.0},
  {14, 347136.0, 47, 0.0, 57856.0},
  {15, 0.0, 47, 57856.0, 404992.0},
  {16, 0.0, 47, 462848.0, 57856.0},
  {15, 347136.0, 48, 0.0, 57856.0},
  {16, 0.0, 48, 57856.0, 404992.0},
  {17, 0.0, 48, 462848.0, 57856.0},
  {16, 347136.0, 49, 0.0, 57856.0},
  {17, 0.0, 49, 57856.0, 404992.0},
  {18, 0.0, 49, 462848.0, 57856.0},
  {17, 347136.0, 50, 0.0, 57856.0},
  {18, 0.0, 50, 57856.0, 404992.0},
  {19, 0.0, 50, 462848.0, 57856.0},
  {18, 347136.0, 51, 0.0, 57856.0},
  {19, 0.0, 51, 57856.0, 404992.0},
  {20, 0.0, 51, 462848.0, 57856.0},
  {19, 347136.0, 52, 0.0, 57856.0},
  {20, 0.0, 52, 57856.0, 404992.0},
  {21, 0.0, 52, 462848.0, 57856.0},
  {20, 347136.0, 53, 0.0, 57856.0},
  {21, 0.0, 53, 57856.0, 404992.0},
  {22, 0.0, 53, 462848.0, 57856.0},
  {21, 347136.0, 54, 0.0, 57856.0},
  {22, 0.0, 54, 57856.0, 404992.0},
  {23, 0.0, 54, 462848.0, 57856.0},
  {22, 347136.0, 55, 0.0, 57856.0},
  {23, 0.0, 55, 57856.0, 404992.0},
  {24, 0.0, 55, 462848.0, 57856.0},
  {23, 347136.0, 56, 0.0, 57856.0},
  {24, 0.0, 56, 57856.0, 404992.0},
  {25, 0.0, 56, 462848.0, 57856.0},
  {24, 347136.0, 57, 0.0, 57856.0},
  {25, 0.0, 57, 57856.0, 404992.0},
  {26, 0.0, 57, 462848.0, 57856.0},
  {25, 347136.0, 58, 0.0, 57856.0},
  {26, 0.0, 58, 57856.0, 404992.0},
  {27, 0.0, 58, 462848.0, 57856.0},
  {26, 347136.0, 59, 0.0, 57856.0},
  {27, 0.0, 59, 57856.0, 404992.0},
  {28, 0.0, 59, 462848.0, 57856.0},
  {27, 347136.0, 60, 0.0, 57856.0},
  {28, 0.0, 60, 57856.0, 404992.0},
  {29, 0.0, 60, 462848.0, 57856.0},
  {28, 347136.0, 61, 0.0, 57856.0},
  {29, 0.0, 61, 57856.0, 404992.0},
  {30, 0.0, 61, 462848.0, 57856.0},
  {29, 347136.0, 62, 0.0, 57856.0},
  {30, 0.0, 62, 57856.0, 404992.0},
  {31, 0.0, 62, 462848.0, 57856.0},
  {30, 347136.0, 63, 0.0, 57856.0},
  {31, 0.0, 63, 57856.0, 404992.0},
};
route_plan_t conv2_route_plan = {
  .num_item = 94,
  .route_item_array = conv2_route_item_array
};
    View Code

    　　

    最后的测试结果如下表：
  • 相关阅读:
    git 通过 fork 解决代码冲突
    vue-cli3.0 多页面和单页面的配置以及相互之间的切换
    关于切换环境变量的问题（ 以vue-cli3.0 为例）
    vue-router 钩子
    Android eMMC 分区详解(转载)
    《PE总结 》– 重定位表(转载)
    Linux 文件系统
    爬虫登录，立FLAG
    ios tweak 开发
    ios app 砸壳
  • 原文地址：https://www.cnblogs.com/acm-jing/p/7193151.html
Copyright © 2020-2023  润新知