This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

division Operation on c64x+

Other Parts Discussed in Thread: MATHLIB

Hi

Some time back while searching for optimizing the division operation,  i got a very easy method to optimize division operation on TI DSPs in TI's e2e forum.  Same thing  I used before also on some algorithm development.

But now i am unable to recall it unfortunately. Neither i am getting link to refer back.

But what i could remember is, division was not carried by either using fastrts or IQMath or mathlib. It was just re organization of the statement/code line containing division operation. and division was applicable to any number but mostly integer.

Anyone could you please point me to that thread?

  • I don't recall a thread like that.  Are you sure you aren't talking about floating point division?  If you are, then this might be the thread.

    Thanks and regards,

    -George

  • Hi George,

    It was not this for sure. As I remember, it was not related to floating point division. It was just normal integer division where in, when we just re-organize  the division operation, timings reduces tremendously like for eg from around 5+ ms to just 200+ odd cycles. I tested it on c64x+ with CCS < 4, -o3 and algo was simple one like mean, std variation.  The thread was quite old may be it was between 2008 and 2011.

    Hope that helps. One who posted/answered that thread might help here. Really I do not remember their names.

  • Could you please help out?

  • I don't remember any such thread.  I have done a few searches and found nothing.  Try using the Advanced search with the link near the upper right corner.

    Thanks and regards,

    -George

  • Venkatesh a said:
    . It was just re organization of the statement/code line containing division operation. and division was applicable to any number but mostly integer.

    From your description, I doubt this is TI-specific.  There are lots of standard techniques for optimizing divisions, but I don't really have enough clues to identify the optimization you are looking for, particularly because you imply that it also works for floating point division, which has very different properties.  Are you dividing by a constant?  Any other details would be helpful when searching for this technique.

    The best optimization is to not do the division at all.  Why do you need the division?  Move divisions out of loops.  If you have a choice of divisor, divide by powers of two.  Try not to divide by variables.  I'm sure there are many more tips.