切换到宽版
发帖 回复
返回列表
123
  • 8340阅读
  • 29回复

[ARM]挖尽GD32潜力,做逻辑分析仪（续）,采样速率提高至23M,60k存储深度,真正体现GD优越性 [复制链接]

 上一主题 下一主题
离线huaweiwx
数码4段
 

发帖
1372
M币
2167
专家
65
粉丝
297
只看楼主 倒序阅读 我要置顶 楼主  发表于: 2016-06-13
在我的帖子：挖尽GD32潜力,做逻辑分析仪,15M采样速率,60k存储深度,体现GD优越性 http://bbs.mydigit.cn/read.php?tid=1589478，近来通过进一步优化，将速度再次提高到23M，优化过程如下：



1 影响采样速度的是采样核心循环部分（原帖line 73～75）：
     for (i = BUFFERSIZE-1; i != 0; i--)  buffer [ i ]= GPIO_PORT_DATA_R

优化1 将指向端口的指针GPIO_PORT_DATA_R，这是一个常量，编译时指针放在代码段，在循环中需要读取这个存放在flash的地址；
现在我定义个指针变量：  volatile uint32_t *port = &GPIO_PORT_DATA_R
并将循环内的： buffer[ i ] = GPIO_PORT_DATA_R
改为：               buffer[ i ] = *port
port 为函数内局部变量，这样循环内部gcc优化时首先分配寄存器来存放局部变量，这样将port装入寄存器，就不需要每次去读取这个地址值了，节约了至少一个指令周期（可能还不止,视缓冲区情况而定）；

2 经测试，buffer [ i ] = *port  在循环内，生成的指令是：

80002a8:    68a2          ldr    r2, [r4, #8]
80002aa:    751a          strb    r2, [r3, #20]

核心部分读端口、写数组仅用两条指令就完成了，按平均1.2 mps/mhz 两个指令周期都不会到（stm 为三级流水线）考虑端口操作没有核心速度那么快，需要插入等待，写RAM也没有寄存器操作那么快（也需要插入等待）
因此如不考虑循环，这部分要使用4个左右的指令周期，即极限30Mhz的采样速率；
加上循环的i--和跳转；又会消耗4～6个指令周期，使实际速率只有15M（最高）；

3 从上可见，一半时间在处理循环变量和判断跳转，由于指令跳转还会导致指令高速缓冲区 ICACH的命中率降低，需要重复从flash中去读取指令，进一步降低了速率，因此改造这一步是关键：我将其写成：
  i= MAX_CAPTURE_SIZE;
  systick_disable();
  do{  
     logicdata[i--] =*port;  //1
     logicdata[i--] =*port;  //2;
     logicdata[i--] =*port;  //3
     logicdata[i--] =*port;  //4
     logicdata[i--] =*port;  //5
     logicdata[i--] =*port;  //6
     logicdata[i--] =*port;  //7
     logicdata[i--] =*port;  //8
     logicdata[i--] =*port;  //9
     logicdata[i--] =*port;  //10
     logicdata[i--] =*port;  //11
     logicdata[i--] =*port;  //12
     logicdata[i--] =*port;  //13
     logicdata[i--] =*port;  //14
     logicdata[i--] =*port;  //15
     logicdata[i--] =*port;  //16
     logicdata[i--] =*port;  //17
     logicdata[i--] =*port;  //18
     logicdata[i--] =*port;  //19
     logicdata[i--] =*port;  //20
  }while(i);
  systick_enable();
现在，我在循环内部，连续多次读取端口及写入数组，显然降低了循环所占时间的比例；经试验从2～20，读写整个60*1024个数据的总时间逐步降低，试验用24时间又明显升高（这和内部ICACH DCACH)大小和是否命中有关），再逐步增大，总时间又逐步减少；最终确定使用20次（这个数值需要保证能被数组大小MAX_CAPTURE_SIZE整除，否则会导致i=0,没有恰好到达while(i)来结束循环；
在进入循环前 我还关闭了systick的中断，防止干扰采样的时间准确性；最后一句是重新允许systick中断；
附 源程序 sllogiclogger_gd32.ino：
复制代码
  1. /*
  2.    SLLogicLogger - Simple logic analyser for Stellaris Launchpad
  3.    SLLogicLogger supports the SUMP protocol over UART. The UART is mapped
  4.    through the debug/flash controller on the Stellaris Launchpad to a virtual
  5.    com port. It samples signals on PORTB with samplerate 10 Mhz, buffersize
  6.    16kByte, sampling starts at any change on PORTB[0..7].
  9.    PB0 and PB1 are limited to 3.6 V! All other pins are 5 V tolerant.
  10.    The idea for this simple analyser comes from http://jjmz.free.fr/?p=148
  13.    Description of the SUMP protocol:
  14.    - http://www.sump.org/projects/analyzer/protocol/
  15.    - http://dangerousprototypes.com/docs/The_Logic_Sniffer%27s_extended_SUMP_protocol
  18.    A multiplatform client which supports the SUMP protocol is OLS:
  19.    http://www.lxtreme.nl/ols/
  20.    To install device support for OLS, copy ols.profile* file into the
  21.    ols/plugins folder.
  24.    Copyright (C) 2012 Thomas Fischl <tfischl@gmx.de> http://www.fischl.de
  25.    Last update: 2012-12-24
  28.    for gd32f103ret6 120M f_cpu samples signals on PORTA(0..7) with samplerate 23 Mhz, buffersize
  29.    60kByte. by huaweiwx@sina.com 2016.6.12
  30. */
  33. #define VERSION "SLLogicLogger v0.1"
  36. //boards
  37. #define LED  LED_BUILTIN  //LED_BUILTIN
  40. // buffer size
  41. #define MAX_CAPTURE_SIZE 60*1024
  42. #define MAX_SAMPLERATE  23000000
  45. // input port read regaddress & pins_arry
  48. #define GPIO_PORT_DATA_R GPIOA->regs->IDR
  51. uint8_t pin[8] = {PA0, PA1, PA2, PA3, PA4, PA5, PA6, PA7};
  54. // SUMP command defintions
  55. #define SUMP_RESET 0x00
  56. #define SUMP_ARM   0x01
  57. #define SUMP_QUERY 0x02
  58. #define SUMP_GET_METADATA 0x04
  61. // sampling buffer
  62. uint8_t logicdata[MAX_CAPTURE_SIZE+1];
  65. // send 32bit unsigned integer as SUMP metadata
  66. void sump_sendmeta_uint32(char type, unsigned int i) {
  67.   Serial1.write(type);
  68.   Serial1.write((i >> 24) & 0xff);
  69.   Serial1.write((i >> 16) & 0xff);
  70.   Serial1.write((i >> 8)  & 0xff);
  71.   Serial1.write(i & 0xff);
  72. }
  75. // send 8bit unsigned integer as SUMP metadata
  76. void sump_sendmeta_uint8(char type, unsigned char i) {
  77.   Serial1.write(type);
  78.   Serial1.write(i);
  79. }
  82. // wait for trigger and do sampling
  83. void dosampling() {
  84.   uint32_t i = 0xfffffff;
  85.   volatile uint32_t *port = &GPIO_PORT_DATA_R;
  86.   char v = GPIO_PORT_DATA_R;
  89. // get current gpio state an wait for change or timeout
  90.   while ((i-- != 0) && (GPIO_PORT_DATA_R == v));
  91.   digitalWrite(LED, LOW);  //for lowplus check
  92.   
  93.   i= MAX_CAPTURE_SIZE;
  94.   systick_disable();
  95.   
  96.   do{  
  97.      logicdata[i--] =*port;  //1
  98.      logicdata[i--] =*port;  //2;
  99.      logicdata[i--] =*port;  //3
  100.      logicdata[i--] =*port;  //4
  101.      logicdata[i--] =*port;  //5
  102.      logicdata[i--] =*port;  //6
  103.      logicdata[i--] =*port;  //7
  104.      logicdata[i--] =*port;  //8
  105.      logicdata[i--] =*port;  //9
  106.      logicdata[i--] =*port;  //10
  107.      logicdata[i--] =*port;  //11
  108.      logicdata[i--] =*port;  //12
  109.      logicdata[i--] =*port;  //13
  110.      logicdata[i--] =*port;  //14
  111.      logicdata[i--] =*port;  //15
  112.      logicdata[i--] =*port;  //16
  113.      logicdata[i--] =*port;  //17
  114.      logicdata[i--] =*port;  //18
  115.      logicdata[i--] =*port;  //19
  116.      logicdata[i--] =*port;  //20
  117.   }while(i);   //60/20=3*1024 cicules speed 23M
  120.   digitalWrite(LED, HIGH);
  121.   systick_enable();
  124. // send it over uart
  125.   for (i = MAX_CAPTURE_SIZE; i != 0; i--) Serial1.write(logicdata[i]); //output unsigned char 8bit data
  126. }
  129. //init
  130. void setup() {
  131.   pinMode(LED, OUTPUT);
  132.   for (uint32_t i = 0; i < 8; i++) pinMode(pin[i], INPUT); //for sample input
  135.   Serial1.begin(115200);
  138.   for (uint32_t i = 0; i < 7; i++) { //ready blink 3 times
  139.     digitalToggle(LED);
  140.     delay(250);
  141.   }
  142. }
  145. // main routine
  146. void loop () {
  147.   if (Serial1.available()) {
  148.     // get the new byte:
  149.     unsigned char cmd = Serial1.read();
  150.     switch (cmd) {
  151.       case SUMP_RESET:
  152.         break;
  153.       case SUMP_ARM:
  154.         dosampling();
  155.         break;
  156.       case SUMP_QUERY:
  157.         Serial1.print("1ALS");
  158.         break;
  159.       case SUMP_GET_METADATA:
  160.         // device name
  161.         Serial1.write(0x01);
  162.         Serial1.print(VERSION);
  163.         Serial1.write((uint8_t)0x00);
  166.         // amount of sample memory available (bytes)
  167.         sump_sendmeta_uint32(0x21, MAX_CAPTURE_SIZE);
  170.         // maximum sample rate (hz)
  171.         sump_sendmeta_uint32(0x23, MAX_SAMPLERATE);
  174.         // number of usable probes (short)
  175.         sump_sendmeta_uint8(0x40, 0x08);
  178.         // protocol version (short)
  179.         sump_sendmeta_uint8(0x41, 0x02);
  182.         // end of meta data
  183.         Serial1.write((uint8_t)0x00);
  186.         break;
  189.       // long commands.. consume bytes from uart
  190.       case 0xC0:
  191.       case 0xC4:
  192.       case 0xC8:
  193.       case 0xCC:
  194.       case 0xC1:
  195.       case 0xC5:
  196.       case 0xC9:
  197.       case 0xCD:
  198.       case 0xC2:
  199.       case 0xC6:
  200.       case 0xCA:
  201.       case 0xCE:
  202.       case 0x80:
  203.       case 0x81:
  204.       case 0x82:
  205.         Serial1.read();
  206.         Serial1.read();
  207.         Serial1.read();
  208.         Serial1.read();
  209.         break;
  210.     }
  211.   }
  212. }


SLLogicLogger配置文件ols.profile-SLLogicLogger.cfg：
复制代码
  1. # Configuration for SLLogicLogger
  4. # The short (single word) type of the device described in this profile
  5. device.type = SLLogicLogger
  6. # A longer description of the device
  7. device.description = SLLogicLogger (GD32RET6@120M)
  8. # The device interface, SERIAL only
  9. device.interface = SERIAL
  10. # The device's native clockspeed, in Hertz.
  11. #device.clockspeed = 80000000
  12. device.clockspeed = 120000000
  13. # Whether or not double-data-rate is supported by the device (also known as the "demux"-mode).
  14. device.supports_ddr = false
  15. # Supported sample rates in Hertz, separated by comma's
  16. #device.samplerates = 10000000
  17. device.samplerates =  23000000
  18. # The clockspeed used in the divider calculation, in Hertz. Defaults to 100MHz as most devices appear to use this.
  19. device.dividerClockspeed = 100000000
  20. # What capture clocks are supported
  21. device.captureclock = INTERNAL
  22. # The supported capture sizes, in bytes
  23. #device.capturesizes = 16384
  24. device.capturesizes = 61440
  25. # Whether or not the noise filter is supported
  26. device.feature.noisefilter = false
  27. # Whether or not Run-Length encoding is supported
  28. device.feature.rle = false
  29. # Whether or not a testing mode is supported
  30. device.feature.testmode = false
  31. # Whether or not triggers are supported
  32. device.feature.triggers = false
  33. # The number of trigger stages
  34. device.trigger.stages = 4
  35. # Whether or not "complex" triggers are supported
  36. device.trigger.complex = false
  39. # The total number of channels usable for capturing
  40. device.channel.count = 8
  41. # The number of channels groups, together with the channel count determines the channels per group
  42. device.channel.groups = 1
  43. # Whether the capture size is limited by the enabled channel groups
  44. device.capturesize.bound = true
  45. # Which numbering does the device support
  46. device.channel.numberingschemes = INSIDE, OUTSIDE
  49. # Is a delay after opening the port and device detection needed? (0 = no delay, >0 = delay in milliseconds)
  50. device.open.portdelay = 0
  51. # The receive timeout for the device (in milliseconds, 100 = default, <=0 = no timeout)
  52. device.receive.timeout = 100
  53. # Does the device need a high or low DTR-line to operate correctly? (high = true, low = false)
  54. device.open.portdtr = false
  55. # Which metadata keys correspond to this device profile? Value is a comma-separated list of (double quoted) names...
  56. device.metadata.keys = "SLLogicLogger"
  59. # In which order are samples sent back from the device? false = last sample first, true = first sample first
  60. device.samples.reverseOrder = true
  63. ###EOF###


上述内容+已编译的执行文件：













[ 此帖被huaweiwx在2016-06-13 16:43重新编辑 ]
本文内容包含图片或附件，获取更多资讯，请 登录 后查看；或者 注册 成为会员获得更多权限
本帖最近打赏记录：共12条打赏 ，M币+36
12
回复
举报
离线mosliu
数码肆段

发帖
29913
M币
14893
专家
2
粉丝
64
只看该作者 1楼 发表于: 2016-06-13
Re:挖尽GD32潜力,做逻辑分析仪（续）,采样速率提高至23M,60k存储深度,真正体现GD优 ..
请登录后查看
本帖最近打赏记录：共1条打赏 ，M币+2
回复
举报
离线yulijie
数码2段

发帖
398
M币
273
专家
1
粉丝
22
只看该作者 2楼 发表于: 2016-06-13
本帖最近打赏记录：共1条打赏 ，M币+2
回复
举报
离线huaweiwx
数码4段

发帖
1372
M币
2167
专家
65
粉丝
297
只看该作者 3楼 发表于: 2016-06-13
回 yulijie 的帖子
请登录后查看
本帖最近打赏记录：共1条打赏 ，M币+3
回复
举报
离线mindlwj
数码6级
发帖
47
M币
-370
专家
0
粉丝
2
只看该作者 4楼 发表于: 2016-06-13
Re:挖尽GD32潜力,做逻辑分析仪（续）,采样速率提高至23M,60k存储深度,真正体现GD优 ..
请登录后查看
本帖最近打赏记录：共1条打赏 ，M币+2
回复
举报
离线amo73
数码1级

发帖
882
M币
1363
专家
9
粉丝
34
只看该作者 5楼 发表于: 2016-06-13
Re:挖尽GD32潜力,做逻辑分析仪（续）,采样速率提高至23M,60k存储深度,真正体现GD优 ..
请登录后查看
本帖最近打赏记录：共1条打赏 ，M币+2
回复
举报
离线huaweiwx
数码4段

发帖
1372
M币
2167
专家
65
粉丝
297
只看该作者 6楼 发表于: 2016-06-13
Re:挖尽GD32潜力,做逻辑分析仪（续）,采样速率提高至23M,60k存储深度,真正体现GD优 ..
请登录后查看
本帖最近打赏记录：共1条打赏 ，M币+3
回复
举报
离线2545889167
数码之家帮帮团

发帖
13268
M币
20754
专家
302
粉丝
4860
只看该作者 7楼 发表于: 2016-06-13
Re:挖尽GD32潜力,做逻辑分析仪（续）,采样速率提高至23M,60k存储深度,真正体现GD优 ..
请登录后查看
本帖最近打赏记录：共1条打赏 ，M币+2
回复
举报
离线huaweiwx
数码4段

发帖
1372
M币
2167
专家
65
粉丝
297
只看该作者 8楼 发表于: 2016-06-13
回 2545889167 的帖子
请登录后查看
本帖最近打赏记录：共1条打赏 ，M币+3
回复
举报
离线2545889167
数码之家帮帮团

发帖
13268
M币
20754
专家
302
粉丝
4860
只看该作者 9楼 发表于: 2016-06-13
回 huaweiwx 的帖子
请登录后查看
回复
举报
发帖 回复
返回列表
123
快速回复
限80 字节
温馨提示：欢迎交流讨论，请勿发布纯表情、纯引用等灌水帖子；以免被删除
 
上一个 下一个