Function is not called if not declared as inline/static

[picatostas]

Hi I am starting to do some basics with register operations and came across with this situation.

The board I am using is a STM32L031k6 nucleo.

Code is the following

#include "stm32l031xx.h"
 
// Green led is at PB3
 
static void delay_ms(uint32_t delay);
 
int main(void)
{
	// Enable clock for GPIOB
	RCC->IOPENR |= RCC_IOPENR_GPIOBEN;
	// Set Gpio B3 as output, default mode is analog mode
	GPIOB->MODER = (GPIOB->MODER & ~(GPIO_MODER_MODE3)) | (GPIO_MODER_MODE3_0);
	while (1)
	{
		GPIOB->BSRR |= 0x8;
		delay_ms(20);
		GPIOB->BRR |= 0x8;
		delay_ms(20);
	}
}
 
void delay_ms(uint32_t delay)
{
	for (; delay > 0; delay--)
	{
		for (uint16_t i = 0; i < 2000; i++)
		{
			/* code */
		}
	}
 
}

if function is not declared as static, it never works.

If declared as static, it works most of the times that I do clean-build-transfer ( I don' t understand when it doesn' t)

if declared as __INLINE code takes like extra 60 B of space and works.

As IDE I am using VSCode with stm32-for-vscode extension, which sets the makefile based on a CubeMx proy.

Is the compiler doing some optimization and removing the function ? Is the MCU hanging for some reason ?

[TDK]

That's a bad delay function. The compiler can optimize it out entirely. It is probably doing so in certain cases.

A better (but still bad) delay would use volatile variables to prevent them from being optimized out. A good delay would be based on actual time, such as systick or timer counts.

[picatostas]

Thanks for the answer, what will change in this case calling it static or inline? I am aware that the way to go would've been as you said, timmer count or systick. I wanted something fast for the sake of blinking the led I will change it.

[Piranha]

For your quick test to work, add volatile qualifier for both - delay and i variables.

[picatostas]

Adding volatile to i and neither of static nor inline for the function declaration made it work. Thanks a lot.

[Nikita91]

For an active delay function based on the cycle counter have a look at:

https://community.st.com/s/question/0D53W00000BJ7KdSAL/is-haldelay1-quaranteed-to-be-close-to-1ms

[Tesla DeLorean]

You shouldn't need to RMW against BRR/BSRR, the whole point is you can do a single write at a bit level.

​

Review the code output by the compiler, and run in the debugger.

​

For the cases that dont work, inspect the listing file, and show that code.

​

U​se the counter in a freerunning TIM to mark time.

​

Use a scope to see signal​s faster than the human eye.