printf float problem STM32CubeIDE VSCode Clang-Compiler

[fe3]

Extension 3.7.0
st-arm-clang 19.1.6+st.10

Hello,

I am using CubeIDE für VSCode.

I write programs for the STM32F756ZGTx. I have set ST Arm Clang as the compiler in CubeMX. Redirecting the output (printf()) via the serial debug interface is possible without any problems. When I try to output float or double values with printf(), I get some weird results.

printf() with integer float/double yields correct results for integer float/double wrong
printf() with float/double integer always leads to incorrect results

I repeated the experiment for double variables. The same problem occurs here as well.

#include "main.h"
#include <stdio.h>

extern UART_HandleTypeDef huart3;

double floatVar = 1.1f;

void mainThread(void)
{	
	printf("\r\n\nHello from Nucleo F756 - float Test\r\n"
		 "-----------------------------------\r\n");

	for(;;){ 
		printf("Float value: %f - Integer %u\r\n", floatVar, (unsigned int) floatVar);
		printf("Integer %8u - Float value: %f\r\n", (unsigned int) floatVar, floatVar);
 HAL_Delay(500);
 floatVar += 1.0f;
	}
}

int __io_putchar(int ch) {
	HAL_UART_Transmit(&huart3, (uint8_t*) &ch, 1, 100);
	return ch;
}

 

 

Variable type float

Hello from Nucleo F756 - float Test
-----------------------------------
Float value: 1.887500 - Integer 2684354560
Integer 1 - Float value: -0.000000
Float value: 2.012500 - Integer 3221225472
Integer 2 - Float value: -2.000000
Float value: 2.137500 - Integer 3221225472
Integer 3 - Float value: -2.000000
Float value: 2.256250 - Integer 1610612736
Integer 4 - Float value: 36893490000000000000.000000
Float value: 2.318750 - Integer 1610612736
Integer 5 - Float value: 36893490000000000000.000000
Float value: 2.381250 - Integer 1610612736
Integer 6 - Float value: 36893490000000000000.000000
Float value: 2.443750 - Integer 1610612736
Integer 7 - Float value: 36893490000000000000.000000

Variable type double

Hello from Nucleo F756 - float Test
-----------------------------------
Float value: 0.000000 - Integer 2684354560
Integer 1 - Float value: -0.000000
Float value: 0.000000 - Integer 3489660928
Integer 2 - Float value: -8589935000.000000
Float value: 0.000000 - Integer 3489660928
Integer 3 - Float value: -8589935000.000000
Float value: 0.000000 - Integer 1744830464
Integer 4 - Float value: 2417851600000000000000000.000000
Float value: 0.000000 - Integer 1744830464
Integer 5 - Float value: 2417851600000000000000000.000000
Float value: 0.000000 - Integer 1744830464
Integer 6 - Float value: 2417851600000000000000000.000000
Float value: 0.000000 - Integer 1744830464
Integer 

Please find enclosed the project.

What am I doing wrong? Is a linker or compiler option missing?

Thanks

[fe3]

I found a solution for the correct output. To do this, I added a line to starm-clang.cmake. 

set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -Wl,--defsym=vfprintf=__d_vfprintf -Wl,--defsym=vfscanf=__d_vfscanf") 

Here again in the context of the file

 

if (STARM_TOOLCHAIN_CONFIG STREQUAL "STARM_HYBRID")
 set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} --gcc-specs=nano.specs")
 set(TOOLCHAIN_LINK_LIBRARIES "m")
elseif(STARM_TOOLCHAIN_CONFIG STREQUAL "STARM_NEWLIB")
 set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -lcrt0-nosys")
elseif(STARM_TOOLCHAIN_CONFIG STREQUAL "STARM_PICOLIBC")
 # added line to enable float support in printf/scanf
 set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -Wl,--defsym=vfprintf=__d_vfprintf -Wl,--defsym=vfscanf=__d_vfscanf") #<<<<
 set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -lcrt0-hosted -z norelro")

 After these changes, I can output both float and double values correctly. The combination with integer and float/double also works.

I would be very grateful if someone could tell me whether this is the approach planned by ST.

Thanks

[KnarfB]

This is an issue of the newlib nano or pico you are using. The don't support float formats by default. but supply a weak symbol which you may request linking like:

target_link_options(app PRIVATE 
 -specs=nano.specs 
 -Wl,-u,_printf_float 
 -lc 
 -lnosys 
)

or

clang --target=arm-none-eabi -mcpu=cortex-m4 -mthumb \
 -ffreestanding -nostdlib -nostartfiles \
 -I/path/to/picolibc/include \
 main.c \
 -L/path/to/picolibc/lib \
 -lc -lm \
 -Wl,-u,__printf_float \
 -o out.elf

 Haven't used it with clang (yet?), but with gcc. 

[fe3]

Sorry for the late reply. I tested it again. In my test program, -Wl,-u,__printf_float does not work for picolibc.