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.

FFTc performances

Other Parts Discussed in Thread: TMS320C6670

Hello,

is there any figure about FFTc performances available?

I am interested in the total time duration starting from the push on the Tx FFTc queue and the push on the Rx queue by FFTc. Of course FFTc size is a parameter.

In particular: having FFTc input/output data buffer allocated into internal or external DSP memory may affect FFTc performances? If yes, to which extent?

Thanks

Regards

Stefano

  • The FFTc Throughput performance is detailed in the 'Throughput Performance Guide for C66x Keystone Devices' - App Note.  It can be found in the App Notes section of the product page -> TMS320C6670.

    The using Internal Memory and DDR3 should not impact the performance, which is dominated by the FFTc engine speed.

    Best Regards,

    Chad

  • Thanks for the good document.

    Any way the speed reported there is in a special condition in which the FFTc engine is continuously loaded and goes somehow in a steady-state in which pipelining is heavily involved.

    Is there any indication about how much the single-shot FFTc processing takes?

    I mean, what if I push one descriptor with one packet to process and I measure the time when the FFTc has done and written the result to internal memory location?

    I am asking because we are using FFTc like that and we measure about 14 us vs. the 4.8 us reported in steady-state by the doc.

    That's pretty a lot and we are wondering if we are missing something, or it is expected as it is the accelerator initial latency.

    We are using the FFTc driver provided in MCSDK 2.00.03.15, it could also be that we are badly using this one.

    Thanks

    Roberto

  • I'll ping the owner of this IP and see if he can comment on that. 

    Best Regards,

    Chad

  • Hi Stefano,

       The steady state numbers published are the core FFTC throughput results. What you are looking for is the overhead incurred due to the 'Multicore Navigator' on the input and output sides. The performance document referred  to by Chad contains a section for the throughput of the Navigator which should answer your question.

    Chad,

       You could forward this to the Navigator IP owner if the information in the performance document does not answer Stefano's question.

    Regards,

    Sudhanand.