• State TI Thinks Resolved
  • Locked Locked
  • Replies 11 replies
  • Answers 1 answer
  • Subscribers 83 subscribers
  • Views 388 views
  • Users 0 members are here
Support feedback
Options
Options
Similar topics

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.

TMS320F28069: TMS320F28069

Maida Farooq
Prodigy 180 points
Part Number: TMS320F28069
Other Parts Discussed in Thread: C2000WARE

Hi,

I am using TMS320F28069 for implementing a control algorithm. Can you guide me how many clock cycles does it take for a floating point division?

Thanks!

  • 0
    TI__Genius 11645 points

    Hi,

    You can find the cycle counts for different versions in the below document. The device you are referring to supports FPU32, so you can look at different routines and their cycles in the below document which is part of c2000ware.

    ...\ti\c2000\C2000Ware_3_03_00_00\libraries\math\FPUfastRTS\c28\docs\FASTRTS_SW_LIB_UG.pdf

  • 0 in reply to KARTHIKEYAN RAJAMANICKAM
    Prodigy 180 points

    I am trying to find the time for execution for a division of float variables because its taking about 5 us even with optimization. I have also included all the required libraries used in sample code 2806xfpu_hardware. In the document, I could find the time on page 46 of this document. These times are for a different device so do they apply here as well?

    Thanks,

  • 0 in reply to Maida Farooq
    TI__Expert 5410 points

    HI,

    The cycle count of the instructions will be consistent across devices. Can you share how many cycles the instructions take, tallied against the number in the doc?

    -Shantanu

  • 0
    Prodigy 180 points

    I am unsure of what number exactly should I consider from the doc. It is taking 2.7 us to compute one division operation even with optimizations. In the doc on page 46, the cycles for division are given to be 222 which translates to 2.5 us. However in the same document in previous pages, the cycles for division are 24. Could you please clarify which number is accurate?

  • 0 in reply to Maida Farooq
    TI__Expert 5410 points

    The cycles on page 47 are cycles for FID_ui64byui32 is 35, ui32byui32 is 23. THe max number of cycles is for ui64byui64 which is 53 (page 47). I am not sure where you are getting the 222 number from. Depending on the data type of each argument, you find different number of cycles (16 bit, 32 bit, 64 bit for divisor or dividend). 

  • 0
    Prodigy 180 points

    I got 222 from pg. 46 from executable file table function div_f32.out. If this is the case then I am not sure why its taking 220 cycles for performing 1 division calculation. 

  • 0 in reply to Maida Farooq
    TI__Expert 5410 points

    Hi,

    The 220 cycles number you are referring to is the __c28xabi_divf function from the standard RTS library which uses the Newton-Raphson method. This is done in software, which results in the high cycle count. If you use the FASTRTS library while enabling FPU32 in processor options and TMU support (as specified in the docs), you will see cycle counts under 42.'

    -Shantanu

  • 0 in reply to Shanty
    Prodigy 180 points

    I understand now thanks. However, with the optimization as well I am unable to reduce the execution time. Could you help me with that?

  • 0 in reply to Maida Farooq
    TI__Expert 5410 points

    Hi,

    Unless you rebuild the standard RTS library with optimization, you will not see any changes. If you are trying to optimize the approach with FPU, it will not yield any changes as the FPU instructions take a fixed number of cycles. 

  • 0 in reply to Shanty
    Prodigy 180 points

    Could you guide me on how to do that?

  • 0 in reply to Maida Farooq
    TI__Expert 5410 points

    You can experiment with the optimisations in build properties (o-1 to o-4). Refer https://www.ti.com/lit/spru514 for more information.