AIR32F103(四) 27倍频216MHz,CoreMark跑分测试

2022-10-27 20:08 由 Milton 发表于 #操作系统

AIR32F103(一) 合宙AIR32F103CBT6开发板上手报告
AIR32F103(二) Linux环境和LibOpenCM3项目模板
AIR32F103(三) Linux环境基于标准外设库的项目模板
AIR32F103(四) 27倍频216MHz,CoreMark跑分测试
AIR32F103(五) FreeRTOSv202112核心库的集成和示例代码
AIR32F103(六) ADC,I2S,DMA和ADPCM实现的录音播放功能
AIR32F103(七) AIR32F103CBT6/CCT6启用96K内存
AIR32F103(八) 集成Helix MP3解码库播放MP3
AIR32F103(九) CAN总线的通信和ID过滤机制及实例
AIR32F103(十) 在无系统环境和FreeRTOS环境集成LVGL
AIR32F103(十一) 在AIR32F103上移植微雪墨水屏驱动
AIR32F103(十二) 搭载 AIR32F103CBT6 的Bluepill核心板

27倍频运行216MHz主频

合宙开发团队10月11日的提交中开源了AIR32F103的PLL倍频调节的代码, 使得在 Linux 下通过 GCC Arm 工具链也能编译运行216MHz.

代码示例

示例代码位于 Examples/NonFreeRTOS/RCC 下: https://gitee.com/iosetting/air32f103-template/tree/master/Examples/NonFreeRTOS/RCC

编译时的注意事项

编译时需要注意避免编译器对AIR_RCC_PLLConfig()这个函数的优化.

这个函数的源代码如下, 可以看到其中会对特定地址(例如 0x40016C00)进行连续的写操作, 编译时如果优化参数不是-O0, 就大概率会将这些写操作合并或调换位置.

uint32_t AIR_RCC_PLLConfig(uint32_t RCC_PLLSource, uint32_t RCC_PLLMul, uint8_t Latency)
{ 
  uint32_t sramsize = 0;
  uint32_t pllmul = 0;
  FunctionalState pwr_gating_state = 0;
  /* Check the parameters */
  assert_param(IS_RCC_PLL_SOURCE(RCC_PLLSource));
  assert_param(IS_RCC_PLL_MUL(RCC_PLLMul));
  
  *(uint32_t *)(0x400210F0) = BIT(0);//开启sys_cfg门控
  *(uint32_t *)(0x40016C00) = 0xa7d93a86;//解一、二、三级锁
  *(uint32_t *)(0x40016C00) = 0xab12dfcd;
  *(uint32_t *)(0x40016C00) = 0xcded3526;
  sramsize = *(uint32_t *)(0x40016C18);
  *(uint32_t *)(0x40016C18) = 0x200183FF;//配置sram大小, 将BOOT使用对sram打开 
  *(uint32_t *)(0x4002228C) = 0xa5a5a5a5;//QSPI解锁
  
  SysFreq_Set(RCC_PLLMul,Latency ,0,1);
  RCC->CFGR = (RCC->CFGR & ~0x00030000) | RCC_PLLSource;
  
  //恢复配置前状态
  *(uint32_t *)(0x40016C18) = sramsize;
  *(uint32_t *)(0x400210F0) = 0;//开启sys_cfg门控
  *(uint32_t *)(0x40016C00) = ~0xa7d93a86;//加一、二、三级锁
  *(uint32_t *)(0x40016C00) = ~0xab12dfcd;
  *(uint32_t *)(0x40016C00) = ~0xcded3526;
  *(uint32_t *)(0x4002228C) = ~0xa5a5a5a5;//QSPI解锁
  return 1;
}

解决的方法一, 是通过调整编译参数

在Keil5下, 可以对 air32f10x_rcc_ex.c 这个文件右键单独设置 AC6 编译选项. AC5可以使用注解, AC6不再支持文件内部单个函数的优化设置
在GCC Arm中, 可以通过 Makefile 对 air32f10x_rcc_ex.c 设置单独的-O0参数, 也可以在代码中增加屏障避免优化(例如在两行代码之间增加__NOP();), 还可以通过int foo(int i) __attribute__((optimize("-O3")));这样的形式, 参考GNU GCC文档

因此将库函数修改为

__attribute__((optimize("-O0"))) uint32_t AIR_RCC_PLLConfig(uint32_t RCC_PLLSource, uint32_t RCC_PLLMul, uint8_t Latency)

CoreMark跑分结果

示例中的 CoreMark_256MHz 项目, 可以将AIR32F103运行在最高256MHz主频下, 运行CoreMark性能测试. 以下是分别在 256MHz, 216MHz, 72MHz 不同编译器版本下的测试结果.

32倍频, 256MHz

编译器 GCC11.2.1

SYSCLK: 256000000, HCLK: 256000000, PCLK1: 128000000, PCLK2: 256000000, ADCCLK: 128000000␊
2K performance run parameters for coremark.␊
CoreMark Size    : 666␊
Total ticks      : 17054␊
Total time (secs): 17.054000␊
Iterations/Sec   : 586.372698␊
Iterations       : 10000␊
Compiler version : GCC11.2.1 20220111␊
Compiler flags   : -O3␊
Memory location  : STACK␊
seedcrc          : 0xe9f5␊
[0]crclist       : 0xe714␊
[0]crcmatrix     : 0x1fd7␊
[0]crcstate      : 0x8e3a␊
[0]crcfinal      : 0x988c␊
Correct operation validated. See readme.txt for run and reporting rules.␊
CoreMark 1.0 : 586.372698 / GCC11.2.1 20220111 -O3 / STACK␊
IR32F103 CoreMark␊

编译器 GCC11.3.1, 256MHz

SYSCLK: 256000000, HCLK: 256000000, PCLK1: 128000000, PCLK2: 256000000, ADCCLK: 128000000␊
2K performance run parameters for coremark.␊
CoreMark Size    : 666␊
Total ticks      : 17054␊
Total time (secs): 17.054000␊
Iterations/Sec   : 586.372698␊
Iterations       : 10000␊
Compiler version : GCC11.3.1 20220712␊
Compiler flags   : -O3␊
Memory location  : STACK␊
seedcrc          : 0xe9f5␊
[0]crclist       : 0xe714␊
[0]crcmatrix     : 0x1fd7␊
[0]crcstate      : 0x8e3a␊
[0]crcfinal      : 0x988c␊
Correct operation validated. See readme.txt for run and reporting rules.␊
CoreMark 1.0 : 586.372698 / GCC11.3.1 20220712 -O3 / STACK␊

编译器 GCC12.2.0 256MHz

SYSCLK: 256000000, HCLK: 256000000, PCLK1: 128000000, PCLK2: 256000000, ADCCLK: 128000000␊
2K performance run parameters for coremark.␊
CoreMark Size    : 666␊
Total ticks      : 15822␊
Total time (secs): 15.822000␊
Iterations/Sec   : 632.031349␊
Iterations       : 10000␊
Compiler version : GCC12.2.0␊
Compiler flags   : -O3␊
Memory location  : STACK␊
seedcrc          : 0xe9f5␊
[0]crclist       : 0xe714␊
[0]crcmatrix     : 0x1fd7␊
[0]crcstate      : 0x8e3a␊
[0]crcfinal      : 0x988c␊
Correct operation validated. See readme.txt for run and reporting rules.␊
CoreMark 1.0 : 632.031349 / GCC12.2.0 -O3 / STACK␊

27倍频, 216MHz

GCC11.2.1 216MHz

SYSCLK: 216000000, HCLK: 216000000, PCLK1: 108000000, PCLK2: 216000000, ADCCLK: 108000000␊
2K performance run parameters for coremark.␊
CoreMark Size    : 666␊
Total ticks      : 20213␊
Total time (secs): 20.213000␊
Iterations/Sec   : 494.731114␊
Iterations       : 10000␊
Compiler version : GCC11.2.1 20220111␊
Compiler flags   : -O3␊
Memory location  : STACK␊
seedcrc          : 0xe9f5␊
[0]crclist       : 0xe714␊
[0]crcmatrix     : 0x1fd7␊
[0]crcstate      : 0x8e3a␊
[0]crcfinal      : 0x988c␊
Correct operation validated. See readme.txt for run and reporting rules.␊
CoreMark 1.0 : 494.731114 / GCC11.2.1 20220111 -O3 / STACK␊

GCC11.3.1 216MHz

SYSCLK: 216000000, HCLK: 216000000, PCLK1: 108000000, PCLK2: 216000000, ADCCLK: 108000000␊
2K performance run parameters for coremark.␊
CoreMark Size    : 666␊
Total ticks      : 20213␊
Total time (secs): 20.213000␊
Iterations/Sec   : 494.731114␊
Iterations       : 10000␊
Compiler version : GCC11.3.1 20220712␊
Compiler flags   : -O3␊
Memory location  : STACK␊
seedcrc          : 0xe9f5␊
[0]crclist       : 0xe714␊
[0]crcmatrix     : 0x1fd7␊
[0]crcstate      : 0x8e3a␊
[0]crcfinal      : 0x988c␊
Correct operation validated. See readme.txt for run and reporting rules.␊
CoreMark 1.0 : 494.731114 / GCC11.3.1 20220712 -O3 / STACK␊

GCC12.2.0 216MHz

SYSCLK: 216000000, HCLK: 216000000, PCLK1: 108000000, PCLK2: 216000000, ADCCLK: 108000000␊
2K performance run parameters for coremark.␊
CoreMark Size    : 666␊
Total ticks      : 18753␊
Total time (secs): 18.753000␊
Iterations/Sec   : 533.248014␊
Iterations       : 10000␊
Compiler version : GCC12.2.0␊
Compiler flags   : -O3␊
Memory location  : STACK␊
seedcrc          : 0xe9f5␊
[0]crclist       : 0xe714␊
[0]crcmatrix     : 0x1fd7␊
[0]crcstate      : 0x8e3a␊
[0]crcfinal      : 0x988c␊
Correct operation validated. See readme.txt for run and reporting rules.␊
CoreMark 1.0 : 533.248014 / GCC12.2.0 -O3 / STACK␊

编译器 AC6 (GCCClang 15.0.0) 216MHz

SYSCLK: 216.0Mhz, HCLK: 216.0Mhz, PCLK1: 108.0Mhz, PCLK2: 216.0Mhz, ADCCLK: 108.0Mhz
2K performance run parameters for coremark.
CoreMark Size    : 666
Total ticks      : 16328
Total time (secs): 16.328000
Iterations/Sec   : 612.444880
Iterations       : 10000
Compiler version : GCCClang 15.0.0
Compiler flags   : -O3
Memory location  : STACK
seedcrc          : 0xe9f5
[0]crclist       : 0xe714
[0]crcmatrix     : 0x1fd7
[0]crcstate      : 0x8e3a
[0]crcfinal      : 0x988c
Correct operation validated. See readme.txt for run and reporting rules.
CoreMark 1.0 : 612.444880 / GCCClang 15.0.0 -O3 / STACK

9倍频, 72MHz

编译器 GCC11.2.1 72MHz

SYSCLK: 72000000, HCLK: 72000000, PCLK1: 36000000, PCLK2: 72000000, ADCCLK: 36000000␊
2K performance run parameters for coremark.␊
CoreMark Size    : 666␊
Total ticks      : 60677␊
Total time (secs): 60.677000␊
Iterations/Sec   : 164.807093␊
Iterations       : 10000␊
Compiler version : GCC11.2.1 20220111␊
Compiler flags   : -O3␊
Memory location  : STACK␊
seedcrc          : 0xe9f5␊
[0]crclist       : 0xe714␊
[0]crcmatrix     : 0x1fd7␊
[0]crcstate      : 0x8e3a␊
[0]crcfinal      : 0x988c␊
Correct operation validated. See readme.txt for run and reporting rules.␊
CoreMark 1.0 : 164.807093 / GCC11.2.1 20220111 -O3 / STACK␊

编译器 GCC11.3.1 72MHz

SYSCLK: 72000000, HCLK: 72000000, PCLK1: 36000000, PCLK2: 72000000, ADCCLK: 36000000␊
2K performance run parameters for coremark.␊
CoreMark Size    : 666␊
Total ticks      : 60677␊
Total time (secs): 60.677000␊
Iterations/Sec   : 164.807093␊
Iterations       : 10000␊
Compiler version : GCC11.3.1 20220712␊
Compiler flags   : -O3␊
Memory location  : STACK␊
seedcrc          : 0xe9f5␊
[0]crclist       : 0xe714␊
[0]crcmatrix     : 0x1fd7␊
[0]crcstate      : 0x8e3a␊
[0]crcfinal      : 0x988c␊
Correct operation validated. See readme.txt for run and reporting rules.␊
CoreMark 1.0 : 164.807093 / GCC11.3.1 20220712 -O3 / STACK␊

编译器 GCC12.2.0 72MHz

SYSCLK: 72000000, HCLK: 72000000, PCLK1: 36000000, PCLK2: 72000000, ADCCLK: 36000000␊
2K performance run parameters for coremark.␊
CoreMark Size    : 666␊
Total ticks      : 56293␊
Total time (secs): 56.293000␊
Iterations/Sec   : 177.641980␊
Iterations       : 10000␊
Compiler version : GCC12.2.0␊
Compiler flags   : -O3␊
Memory location  : STACK␊
seedcrc          : 0xe9f5␊
[0]crclist       : 0xe714␊
[0]crcmatrix     : 0x1fd7␊
[0]crcstate      : 0x8e3a␊
[0]crcfinal      : 0x988c␊
Correct operation validated. See readme.txt for run and reporting rules.␊
CoreMark 1.0 : 177.641980 / GCC12.2.0 -O3 / STACK␊

总结

可以看到, GCC11.2和GCC11.3是一样的, GCC12.2生成的二进制执行性能提升了接近8%, 但是性能最好的还是AC6, 比GCC12.2性能高了接近15%.