MCU：STM32F407
环境：VSCode + PlatformIO IDE + STM32CubeF4(v1.5.2)
编译器：7.2.1 20170904 (release) [ARM/embedded-7-branch revision 255204]
PlatformIO Core: version 5.1.1
STM32F4 Core Feature Description
问题描述
由于用的是STM32F4平台，在ST产品线中定位为高性能计算产品，带有FPU和DSP指令集，这不用来做数字信号处理属实浪费，刚好近期工作需求要用到FFT或者FIR，所以，万事俱备，开搞！！
1. DSP库在哪？文档？
看了一下PlatformIO自带的STM32CubeF4库，为了节约服务器带宽和用户下载时间，所以包很精简，可是...这用起来很不方便呀，没有文档怎么写代码？
image.png
于是去官网下了一个标准版本，1.6.1最新版，带了文档。然而后来才发现，原来CMSIS文档是有在线版的：>>> CMSIS DSP Software Library <<<
官网：STMicroelectronics/STM32CubeF4: STM32Cube MCU Full Package for the STM32F4 series - Github
CMSIS文档
2. 编译Demo
按理说三分钟出demo的事，结果一开始情况就不太妙...首先找到自带例程（为了要例程还是得下完整库）。拿一个FIR的例子开刀：
image.png
复制进自己的项目，该函数名，烧录运行？运行效果：
image.png
image.png
！！直接复制进来用会 遇到成吨报错轰炸！！
观察日志会发现基本全是CMSIS库中的变量缺失错误，以及类型缺失错误。定位错误可以发现：
image.png
类型错误很好解决，直接导入HAL的总入口.h文件
#include <stm32f4xx_hal.h>（加到math_helper.h顶部）。参考：#error "Compiler generates FPU instructions for a device without an FPU (check __FPU_PRESENT)"
3. arm_math.h:932:41: error: redefinition of '__SMLALD'
编译过后，错误少了一大筐：
image.png
但依然会有类似这种错误：
C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Include/cmsis_gcc.h:1652:31: note: previous definition of '__SHADD16' was here
1652 | __STATIC_FORCEINLINE uint32_t __SHADD16(uint32_t op1, uint32_t op2)
| ^~~~~~~~~
In file included from lib\arm_fir_example_f32\math_helper.h:45,
from lib\arm_fir_example_f32\arm_fir_example_f32.c:125:
C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\DSP\Include/arm_math.h:932:41: error: redefinition of '__SMLALD'
932 | CMSIS_INLINE __STATIC_INLINE uint64_t __SMLALD(
| ^~~~~~~~
In file included from C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Include/cmsis_compiler.h:48,
from C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Include/core_cm4.h:162,
from C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Device\ST\STM32F4xx\Include/stm32f407xx.h:167,
from C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Device\ST\STM32F4xx\Include/stm32f4xx.h:133,
from C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\STM32F4xx_HAL_Driver\Inc/stm32f4xx_hal_def.h:30,
from C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\STM32F4xx_HAL_Driver\Inc/stm32f4xx_hal_rcc.h:29,
from C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\STM32F4xx_HAL_Driver\Inc/stm32f4xx_hal_conf.h:281,
from C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\STM32F4xx_HAL_Driver\Inc/stm32f4xx_hal.h:30,
from lib\arm_fir_example_f32\math_helper.h:44,
from lib\arm_fir_example_f32\arm_fir_example_f32.c:125:
继续定位，发现缺失的变量在cmsis_iccarm.h文件有定义：
image.png
根据导入链，查看
arm_math.h文件，发现里面有这样一行宏定义，在VSCode中显示为灰色：
#if defined(ARM_MATH_CM7)
#include "core_cm7.h"
#define ARM_MATH_DSP
#elif defined (ARM_MATH_CM4)
#include "core_cm4.h"
#define ARM_MATH_DSP
这个解决办法就很明确了，直接在顶层include 文件添加一个宏定义，对应到自己的MCU平台即可。由于STM32F4系列对应于Cortex-M4架构，所以这里定义一个ARM_MATH_CM4。
math_helper.h头部：
#define ARM_MATH_CM4
#include <stm32f4xx_hal.h>
#include "arm_math.h"
编译：
4. arm_fir_example_f32.c:202: undefined reference to `arm_fir_f32'
Archiving .pio\build\black_f407zg\liba89\libarm_fir_example_f32.a
Archiving .pio\build\black_f407zg\libFrameworkCMSISDevice.a
Linking .pio\build\black_f407zg\firmware.elf
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: .pio\build\black_f407zg\liba89\libarm_fir_example_f32.a(arm_fir_example_f32.o): in function `fir_test':
C:\Documents\PlatformIO\Projects\STM32F4_ADC_Sampling_v2/lib\arm_fir_example_f32/arm_fir_example_f32.c:194: undefined reference to `arm_fir_init_f32'
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: C:\Documents\PlatformIO\Projects\STM32F4_ADC_Sampling_v2/lib\arm_fir_example_f32/arm_fir_example_f32.c:202: undefined reference to `arm_fir_f32'
collect2.exe: error: ld returned 1 exit status
*** [.pio\build\black_f407zg\firmware.elf] Error 1
image.png
经典的符号未定义问题，而且是在link阶段出现的，说明DSP库没有正确链接上。也就是说这个库对于编译器来说是不知道在哪个位置的，需要自己添加上。
定位库文件：
image.png
用的GCC编译器，进去发现问题来了，选哪个呢？
image.png
搜索了解后知道，l是指little endian，f是指带有硬件除法器，那么没有疑问，用带lf的那个，也就是
libarm_cortexM4lf_math.a。
5. 导入DSP静态库
问题又来了，怎么导入到PIO呢？
打开platformio.ini文件，在环境中添加一行：
build_flags =
-LC:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Lib\GCC
-larm_cortexM4lf_math
第N次编译，结果：
image.png
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: error: C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Lib\GCC\libarm_cortexM4lf_math.a(arm_fir_init_q15.o) uses VFP register arguments, .pio\build\black_f407zg\firmware.elf does not
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: failed to merge target specific data of file C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Lib\GCC\libarm_cortexM4lf_math.a(arm_fir_init_q15.o)
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: error: C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Lib\GCC\libarm_cortexM4lf_math.a(arm_fir_init_f32.o) uses VFP register arguments, .pio\build\black_f407zg\firmware.elf does not
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: failed to merge target specific data of file C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Lib\GCC\libarm_cortexM4lf_math.a(arm_fir_init_f32.o)
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: error: C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Lib\GCC\libarm_cortexM4lf_math.a(arm_fir_f32.o) uses VFP register arguments, .pio\build\black_f407zg\firmware.elf does not
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: failed to merge target specific data of file C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Lib\GCC\libarm_cortexM4lf_math.a(arm_fir_f32.o)
collect2.exe: error: ld returned 1 exit status
*** [.pio\build\black_f407zg\firmware.elf] Error 1
6. 启用编译器（MCU）浮点数扩展
这里理一下思路：我们要用F4的浮点数运算能力，但是浮点数运算需要用到VFP指令和寄存器，但是libarm_cortexM4lf_math.a使用了VFP寄存器，而.pio\build\black_f407zg\firmware.elf没有使用....
这个问题的解读就是，生成一个可执行文件elf时，其所有的中间对象*.o都要是同样的编译条件，至少对于VFP指令集的设置应该是相同的。但是目前的ELF文件没有启用这个选项。
那么问题就简单了，启用一下不就可以了，而方式应该就是通过对编译器传参，在platformIO中就对应于，设置platformio.ini文件中的build_flags。
查过以上资料后，发现我应该这么设置：
[env:black_f407zg]
build_type = debug
platform = ststm32
board = black_f407zg
framework = stm32cube
upload_protocol = stlink
debug_tool = stlink
monitor_speed = 115200
build_flags =
-LC:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Lib\ARM
-larm_cortexM4lf_math
-mthumb -mcpu=cortex-m4 -march=armv7e-m -mfloat-abi=hard -mfpu=fpv4-sp-d16
也就是添加一行5个flag
-mthumb -mcpu=cortex-m4 -march=armv7e-m -mfloat-abi=hard -mfpu=fpv4-sp-d16
使用过后，编译结果：
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: failed to merge target specific data of file .pio\build\black_f407zg\FrameworkHALDriver\Src\stm32f4xx_ll_fsmc.o
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: error: .pio\build\black_f407zg\src\system_stm32f4xx.o uses VFP register arguments, .pio\build\black_f407zg\firmware.elf does not
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: failed to merge target specific data of file .pio\build\black_f407zg\src\system_stm32f4xx.o
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: error: .pio\build\black_f407zg\liba89\libarm_fir_example_f32.a(arm_fir_example_f32.o) uses VFP register arguments, .pio\build\black_f407zg\firmware.elf does not
...
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: failed to merge target specific data of file .pio\build\black_f407zg\liba89\libarm_fir_example_f32.a(arm_fir_example_f32.o)
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: error: .pio\build\black_f407zg\liba89\libarm_fir_example_f32.a(fft.o) uses VFP register arguments, .pio\build\black_f407zg\firmware.elf does not
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: failed to merge target specific data of file .pio\build\black_f407zg\liba89\libarm_fir_example_f32.a(fft.o)
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: error: .pio\build\black_f407zg\liba89\libarm_fir_example_f32.a(math_helper.o) uses VFP register arguments, .pio\build\black_f407zg\firmware.elf does not
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: failed to merge target specific data of file .pio\build\black_f407zg\liba89\libarm_fir_example_f32.a(math_helper.o)
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: error: .pio\build\black_f407zg\liba89\libarm_fir_example_f32.a(arm_fir_data.o) uses VFP register arguments, .pio\build\black_f407zg\firmware.elf does not
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: failed to merge target specific data of file .pio\build\black_f407zg\liba89\libarm_fir_example_f32.a(arm_fir_data.o)
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: error: C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Lib\GCC\libarm_cortexM4lf_math.a(arm_fir_init_q15.o) uses VFP register arguments, .pio\build\black_f407zg\firmware.elf does not
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: failed to merge target specific data of file C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Lib\GCC\libarm_cortexM4lf_math.a(arm_fir_init_q15.o)
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: error: C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Lib\GCC\libarm_cortexM4lf_math.a(arm_fir_init_f32.o) uses VFP register arguments, .pio\build\black_f407zg\firmware.elf does not
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: failed to merge target specific data of file C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Lib\GCC\libarm_cortexM4lf_math.a(arm_fir_init_f32.o)
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: error: C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Lib\GCC\libarm_cortexM4lf_math.a(arm_fir_f32.o) uses VFP register arguments, .pio\build\black_f407zg\firmware.elf does not
c:/users/joy/.platformio/packages/toolchain-gccarmnoneeabi@2.00000.0/bin/../lib/gcc/arm-none-eabi/10.2.1/../../../../arm-none-eabi/bin/ld.exe: failed to merge target specific data of file C:\Users\JOY\.platformio\packages\framework-stm32cubef4\Drivers\CMSIS\Lib\GCC\libarm_cortexM4lf_math.a(arm_fir_f32.o)
collect2.exe: error: ld returned 1 exit status
*** [.pio\build\black_f407zg\firmware.elf] Error 1
使用过后，不但出现了前面的静态链接库链接不上的问题，其他所有文件都链接不上了。。。？？？
image.png
7. 大地的灾难
查查资料，发现众说纷纭，讲什么的都有，其中比较有代表性和参考价值的几篇：
Cortex-M4F: xxx.elf uses VFP register arguments, yyy.o does not · Issue #2660 · RIOT-OS/RIOT
这篇内容相当精炼，是用GCC+Makefile写的，编译参数透明，参考价值比较大。里面用到的编译参数是：CCFLAGS=-mcpu=cortex-m4 -mthumb -g -mfloat-abi=hard -fsingle-precision-constant -mfpu=fpv4-sp-d16 -I ../../CMSIS/CMSIS-master/CMSIS/Include -D ARM_MATH_CM4 -D __FPU_PRESENT=1Simple DSP with ARM CMSIS and GCC on the STM32F303
用Keil用DSP库是基本没太大问题的，至少不会有奇怪到难以解决的问题。Using cortex-m4 FPU with gcc toolchain
这篇是一个日本人写的，用谷歌翻译了一下，证明用STM32CubeIDE是可以直接用没问题的。如何在STM32 Cube IDE环境中使用CMSIS-DSP | ioloa.com
总结：这里查了一堆资料，然而并没有什么卵用。
8. 迟来的春天
这地方卡了我一整天，真的调到后面脑袋都大了。。。查了很多资料都没有解决，直到后来想到既然其他IDE没问题，会不会是PlatformIO的问题？然后看到这篇：
-
How to pass flags to final firmware.elf linkage? - PlatformIO Core - PlatformIO Community
这个标题似乎没有什么，但后来才反应过来这就是我想要的...不过里面提到的一个bug issue链接参考价值很大，： -
Flags specified in platformio.ini build_flags variable not applied to all necessary steps in the build process · Issue #594 · platformio/platformio-core
这里讲的是PlatformIO的一个bug，而且是2016年5月就有人提了issue，而且被官方人员关闭了。我当然会以为早已经修复好了，但是还是抱着试试的心态测试了一下，里面提到的方案是：
在platformio.ini文件中添加一行extra_script = update_link_flags.py：
[env:disco_f407vg]
platform = ststm32
framework = spl
board = disco_f407vg
build_flags = -mthumb -march=armv7e-m -mfloat-abi=hard -mfpu=fpv4-sp-d16
extra_script = update_link_flags.py
然后在项目根目录中添加一个文件：update_link_flags.py，内容为：
# Custom settings, as referred to as "extra_script" in platformio.ini
#
# See http://docs.platformio.org/en/latest/projectconf.html#extra-script
from SCons.Script import DefaultEnvironment
env = DefaultEnvironment()
env.Append(
LINKFLAGS=[
"-mthumb",
"-march=armv7e-m",
"-mfloat-abi=hard",
"-mfpu=fpv4-sp-d16"
]
)
9. 原因分析
总的来说，就是一颗老鼠屎，毁了整锅汤的案例重现，因为PlatformIO或者stm32插件的一个漏网的bug，导致后来成吨不明所以的编译错误。
表现为：
image.png
如图中描述的，编译过程的Flags不会对所有中间对象都生效，也就是说，这里的
firmware.elf文件，在编译过程中是没有拿到build_flags中的参数的，也就导致最终所有中间对象*.o文件无法链接到这个target中来。而解决办法就是，使用一个额外的脚本(extra_script)，用于在编译时，对每个编译对象重新添加一遍flags，从而规避了这个bug。
可是这都16年的bug了，怎么留到现在的？？？哭死o(╥﹏╥)o
其他解决方案
其实也有折中的办法，但是都不能完美解决问题。
- 使用softfp浮点数指令，这样可以解决，但是效率会成问题，尤其是DSP这类密集性计算算法。
- 使用
libarm_cortexM4l_math.a库文件，实测是可以正常编译且正常执行的。但是这种方案也是没有用到浮点数指令的，所以还是类似软件浮点数的方法，损失了效率。 - 换其他IDE，比如Keil或者STM32CubeIDE，似乎都不会遇到这个问题，但是代码编写体验不如VSCode。
总结
解决这个问题的过程是比较曲折的，甚至我试着按照PlatformIO格式更新了STM32Cube库到最新版，还有去ARM官网下载了新版的编译器，也没有效果。但是中间还是有比较多的收获，掌握了STM32 DSP库的用法，PIO插件的格式，以及查阅文档、检索资料的习惯等等。
