Hi,
I'm using FFTC accelerator on a TMS320TCI6618 with the driver provided in pdk_C6670_1_0_0_15 to perform a dft of size 2048.
The input queue uses monolithic descriptors while the output queue uses host descriptors with buffers in DDR3 memory.
Output queue is an high priority accumulation queue and the fftc driver is configured to use accumulator with a threshold of 1 packet, so a dsp interrupt is triggered every packet.
What I see is that the time delay between a push to the fftc and the coresponding interrupt on dsp is about 17us.
this time delay is correct with this configuration? is there any way to use FFTC in a more performant way?
I've also tried to use a queue with a dedicated queue pending signals wired directly into the chip level INTC0 insted of an high priority queue. But the delay reduces only to 14us.
Regards,
Luca.
