Compiler optimisation issue (remove used functions)[solved]

[MLE S.1]

Hi all,

On our card with a STM32L4 we use a siliconLabs BGM13P32 for the BLE.

The L4 communicate with the BLE SOC in NCP mode (see here for more details).

To do that i have integrated the SiliconLabs libs (BGLIB) to our project on stm32cubeIDE.

The project build fine, with no errors/warnings (default options) and i can communicate with the BLE SoC ...only if i use -O0 option in optimization options.

If i use for example -Og, the functions used in BGLIB (siliconLabs) is removed from the binary (checked on .map file). The project build without errors but the com with the BLE SoC fails !

For information our project is in C++. Join with this post the class of our Bluetooth Com. (bluetoothCom.h/cpp).

The whole project is build with -Og optimisation option, just bluetoothCom.cpp use -O0 option to have a correct communication.

Do you have an idea how to resolve this ?

[MLE S.1]

Thanks for your answers.

@Pavel A.​  i use source code (.c/.h) not a linked library.

@Community member​ , i suppose you talk about .list file generated by stm32cubeIDE ?

Under the result of this file for the method

void BluetoothCom::appHandleEvents(struct gecko_cmd_packet *evt)

With option -O0

With option -Og

In debug mode (disassembly )

code working with -O0

With -Og

All the siliconLabs function is declared as static inline

[MLE S.1]

If it can help, the callGraph with -O0 option

We can see the SiliconLabs functions

with -Og option ?!

[waclawek.jan]

> Code not executed with -Og

Did you verify that it should be executed, i.e. that evt != NULL and BGLIB_MSG_ID(evt->header)==gecko_evt_system_boot_id?

JW

[Tesla DeLorean]

It has flipped the logic such that the case's body code are after the return()

The mvn.w r2,#95 is a byte load of 160, it is filling the structures directly.

The use of static functions allows the compiler much freer reign to fold the functions within the single object (source).

[Pavel A.]

That's strange.

So what does -Og really mean: maximum optimization, that still allows debugging?

Why use -Og rather than -g -O0/O1?

-- pa

[MLE S.1]

@Community member​ , yes i verify that in step mode after a breakpoint, the code in the switch is not executed

@Pavel A.​ the -Og option reduce the code size but it let you to have a good debug experience. If i use the O1/2 ... option the result his the same it seems the code is removed !

[waclawek.jan]

> yes i verify that in step mode after a breakpoint, the code in the switch is not executed

I did not ask if the code *was* executed, I asked, whether there *was a reason* for it to be executed:

>> Did you verify that it should be executed, i.e. that evt != NULL and BGLIB_MSG_ID(evt->header)==gecko_evt_system_boot_id?

Observe the content of registers respective to the variables in those test.

JW

[MLE S.1]

Sorry you are right, only the case gecko_evt_le_connection_closed_id is triggered with breakpoint in "C++" code.

But if put a breakpoint in Disassembly side it is executed :

[MLE S.1]

If i add : -fno-inline in compilation options for bluetoothCom.cpp i see the function in map file and i can have a break point in c++ side when debugging. But it's not working yet ...

[MLE S.1 · Accepted solution]

I have found the issue ...

On stm32 side i use DMA to transfert the data from uart to a circular buffer. The problem is that the SiliconLabs rxcallBack is implemented to read directly on uart register ...

I just modified my code like this :

int32_t BluetoothCom::BGAPI_Rx(uint32_t len1, uint8_t* data1) {
	uint32_t length = len1;
	while (length) {
		if(uart2.bytesAvailable())//needed to be sure DMA have transfert data from uart...
		{
			*data1 = uart2.readc();
			length --;
			data1 ++;
		}
	}
	return len1;
}

added -fno-inline option to file compilation option and now all work fine !

Thanks all for your answer, i have learn a lot with compilation options !