TMS320F28377D: how to set error flag for divide by zero

The LVF and LUF flags protect the algorithm if a numeric overflow or underflow respectively occurs while an appropriate FPU assembly instruction is being executed.  These are flags in the FPU status register, so they are only relevant to floating-point division.  Is that what you are doing?

If you are coding in C and performing a floating-point division the compiler will make a call to an RTS library function.  For a single-precision divide, the FS$$DIV function will be called.  That function will detect a division by zero and return the maximum floating point value expressible in single precision (i.e. single precision divide by zero will return  3.38953139e+38).

I have attached the code for that function, because this is explained in the comments near the top.  You can find this and other RTS source at:  C:\ti\ccsv7\tools\compiler\c2000_#.#.#\lib\src

Since the call is handling the exception itself, neither LVF nor LUF flags will be set.  If you want to trap this case you could modify the source to set those flags using the SETFLG instruction at the appropriate line.  This applies also to the double precision routine.

I have explained all this carefully because you asked about the LVF and LUF flags, which exist only on the FPU.  From your description you say you are seeing an ACC value of 0xFFFFFFFF, which would occur if you attempted an integer divide by zero.  Integer math instructions execute on the C28x core (not the FPU) so the LVF and LVF flags are not involved.

At this point I'd need mode clarity on what you want to do.  Could you send a code snippet to explain it please?

Regards,

Richard

fs_div28.asm

Richard,

Thanks. I will send this code by another mail.

> From your description you say you are seeing an ACC value of 0xFFFFFFFF, which would occur if you attempted an integer divide by zero. Integer math >instructions execute on >the C28x core (not the FPU) so the LVF and LVF flags are not involved.
-Yes, I think we use C28x core. So, typically, If we don't use FPU , is it able to set "exception /NMI" processing when divide by zero happen ?

Also, can we use ITRAP function for integer devide by zero case ?
because, it has NMI type and Interrupt_illegalOperationHandler() API function.

Yes, it is possible to do that, but the RTS assembly routine will have to be modified.

Executing the division with the integer types you sent me results in the LL$$DIV function being called. You will see the call in the dis-assembly window in CCS. To force a user interrupt on divide-by-zero you will have to modify the "ll_div28.asm" file to insert a TRAP instruction in the code. I have not done this, but it should be straightforward to compare the denominator argument against zero and insert a conditional branch. Then you insert and assembly instruction to generate the software interrupt, for example:
TRAP #20
...will force the USER1 interrupt.

Having modified the source, you will need to re-build the RTS library, which is described in section 8.4 od the C Compiler User's Guide:
http://www.ti.com/lit/ug/spru514m/spru514m.pdf

Having done this, in your C code you write an ISR (say, "TrapIsr()") and load the address to the PIE vector table as you would for any other ISR:
PieVectTable.USER1 = &TrapIsr;

You can then write the code in the ISR to handle the divide-by-zero event as you wish.

Regards,

Richard

Hi Richard,

Appricate your help.

But, I couldn't find ll_div28.asm file on my c2000ware.

C:\ti\c2000\C2000Ware_1_00_00_00.

Could you point out this location ?

Regards,

Kz777

You are welcome!

The file is not in C2000Ware.  It's part of the codegen tools and you can find it in the path in my first post:

C:\ti\ccsv7\tools\compiler\c2000_#.#.#\lib\src

...where "#.#.#" is the version of CGT you have installed.  For example, on my machine the full path name is:

C:\ti\ccsv7\tools\compiler\ti-cgt-c2000_16.9.3.LTS\lib\src

Regards,

Richard

Hi Richard,
Thanks for your information. I would like to one more thing .
As I explained above thread, we found such an API.
DriberLib provide C:\ti\c2000\C2000Ware_1_00_00_00\driverlib\f2837xd\driverlib\interrupt.c
Interrupt_illegalOperationHandler() API function
Even if we use above API, do we have to set above compiler re-setting ? Or Does this API not use for our integer divide use case ?

Regards,
Kz777

Yes, you will still need to insert the TRAP as Richard instructed. Interrupt_illegalOperationHandler() is just the default ISR provided by driverlib that gets called when an ITRAP occurs. If you want to use the same ISR for the inserted TRAP, you can do that by calling Interrupt_register() to put its address in the PIE vector table.

Interrupt_register(INT_USER1, &Interrupt_illegalOperationHandler);

Or you could do the same with a totally new ISR.

Whitney

Hi, Whitney,

Could you please let me know how to enable INT_USER1 in C-code?

main(){

   :
   :

   int64 s64_temp;
   Uint32 u32_temp;

   Interrupt_register(INT_USER1, &Interrupt_illegalOperationHandler);

   for(;;)
   {
      s64_temp = s64_temp / u32_temp;
      if(s64_temp < 0){
          // How should user set INT_USER1 for iLLEGAL interrupt?
      }
   }

}

static void Interrupt_illegalOperationHandler(void)
{
    //
    // Something has gone wrong.  The CPU has tried to execute an illegal
    // instruction, generating an illegal instruction trap (ITRAP).
    //
    ESTOP0;
    for(;;)
    {
        ;
    }
}

Regards,
Nagata.

Hi Richard,
Thanks.