• State TI Thinks Resolved
  • Locked Locked
  • Replies 4 replies
  • Answers 1 answer
  • Subscribers 82 subscribers
  • Views 679 views
  • Users 0 members are here
Support feedback
Options
Options
Related

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.

TMS320F28388S: FFT performance

HS FAE Germany
Mastermind 32400 points
Part Number: TMS320F28388S
Other Parts Discussed in Thread: C2000WARE

Hello,
customer uses C6746 for calculating a sliding DFT (256 pt.), harmonics extraction (up to 13th) and need around 90us for it. He wants to switch to the F28388S. Can he expect similar performance? Is there any benchmark between C2000 (including TMU) and C6746?

Regards, Holger

  • 0
    TI__Mastermind 26955 points

    Holger,

    Sorry for the delayed reply. The benchmarks chapter of the FPU SW user guide in C2000Ware will give you information on this. You can refer to \libraries\dsp\FPU\c28\docs.

    For example, a 256-pt 32-bit CFFT is benchmarked at 11024 cycles i.e. ~ 55us at 200MHz CPU clock.

    Thanks,

    Sira

  • 0 in reply to Sira Rao80
    TI__Mastermind 32400 points

    Hello Sira,
    thanks for hint.

    > For example, a 256-pt 32-bit CFFT is benchmarked at 11024 cycles i.e. ~ 55us at 200MHz CPU clock.
    would the TMU improve the performance for your example?

    Regards, Holger

  • 0 in reply to HS FAE Germany
    TI__Mastermind 26955 points

    Holger,

    I don't believe so. The reason is because the FFT is implemented in asm and uses the FPU as well as pre-computed look up tables for the twiddle factors (up to size 1024). Beyond this size, users will need to call a twiddle factor generating function, so here definitely the TMU should help.

    Let me confirm with our system architect Venki.

    Thanks,

    Sira

  • 0 in reply to Sira Rao80
    TI__Mastermind 26955 points

    Holger,

    Follow up

    - Twiddle factor computation is anyway outside the critical loop if we're generating tables, so even for larger sizes where we don't have pre-computed tables, the TMU can be used to generate them, but they would not affect the critical path.

    - For the core FFT computation (critical path), the TMU would not help. Confirmed with the system architect.

    Thanks,

    Sira