DMA overhead

[ceciliac]

Hi, I have alot of data that I want to transfer from the eTPU via DMA (using MPC5746R). Which alternative is better performance wise: calling the DMARequest for every new data (each 50us) or buffering the data in the eTPU and transferring it in bigger chunks? Is there any overhead in calling the DMARequest more often?

Thanks,

Cecilia

[petervlna]

Hi,

This really depends on application needs.

Block transfers have better performance as the DMA don;t have to start for each transfer separately. (DMA engine start, Address reading, etc... are performed only once)

However then you have your slave on XBAR occupied for longer time. (depends on XBAR configuration).

Peter