RTOS/TMS320C6678: Division by zero bit related

One probably has to recognize that there is no division instruction. This means that it ought to be a library function that performs the division. And in such case you are probably not in position to expect something overly specific. But in any case I can confirm that division by zero flag is set if RCPSP, single-precision reciprocal approximation, instruction is issued with 0 input...

This is the code I used for the example:

/*
 * main.c
 */

#include <stdio.h>
#include <xdc/std.h>
#include <xdc/cfg/global.h>
#include <xdc/runtime/System.h>


int main(void) {
    
	
	int numerator = 10;
	/*volatile*/ int denominator = 0;
	int result=0;
	result=numerator/denominator;
	
	System_printf("Result is %d\n", result);
	System_flush();

	return 0;
}

Divide_by_zero_test.cfg

This is my setup:

Code Composer Studio  Version: 6.2.0.00050

C compiler TI compiler v8.1.2

I compared all the visible register with a comparison tool. The only registers which changed are:

• Program Counter
• General purpose registers
• Timer registers
• TSR (it seems that an interrupt is blocked but we cannot understand which one)
• PCIE register

This is the export of all the registers before the division:

BeforeDivision.txt

This is the export of all the registers after the division:

AfterDivision.txt

We weren't able to find RCPSP register.

By the way, the result of the division is 31.

I still have the following question:

Which bit in which register should I check in order to verify that a division by zero happened?

Oh! It seems to be gross misunderstanding from my side. You originally mentioned FADCR, FAUCR, FMCR, and those are floating-point status/control registers. And I assumed that you are talking about floating-point operations. But presented main.c does integer division. It won't set no floating-point flags, sorry about confusion. No [computational] integer instruction would set any flag. And the only right thing to do under circumstances is to explicitly check that divisor is non-zero. I.e. you should do it in your code

if (denominator != 0) { result = numerator/denominator; }

Just in case, RCPSP is a floating-point instruction that is used in floating-point division subroutine, not a register.

Giovambattista Astorino said:

Correct me if I'm wrong, if I modify the main.c , changing the types (let me say I write float instead of integer), shoul I see any changes in the mentioned registers?

Even if you see changes it would be inappropriate to rely on it, because I don't think TI makes any promises about which flags does [floating-point] division library subroutine set. In other words if you want predictable behaviour, you have to write own division subroutine in assembly.

Looking from another angle. Let's say you aim to examine the flag. But how would it be different from checking divisor being zero prior division attempt? It's "if (something) do_something" in either case...