FFTC performance

Hi Angelo,

   There is no fixed answer to your  question as the choice between the two methods depends on your application requirements as well.

    Each push into the FFTC incurs a navigator overhead (on TX & RX side). This overhead is basically the time the navigator interface of the FFTC takes to stream data into the engine and out of the engine and the time it takes to decode the descriptor header.

The more blocks a packet has the less overhead you see.This is because of the pipe-lining in the FFTC wherein it can receive a block of data on the TX side, perform computation on a block and transmit a block of data out on the RX side simultaneously. However  when a packet pushed into the FFTC contains multiple payload, the output from the FFTC will be received (on the RX side) only after all the blocks in the packet have been processed.

  So on one hand if you have one block per packet the h/w overhead you will incur is higher but the application will be able to work on the output data as soon as FFTC completes processing that block while the FFTC is working on the next block/packet, whereas if you have multiple blocks you reduce the h/w overhead but the application would have to wait for the FFTC to complete processing all the blocks in the packet which could translate into latency. The additional advantage of multi-block packet is you will need fewer descriptors.

  Hope this helps you in making your decision.

Regards,

Sudhanand.