I'm wondering if anybody has a sense of when using a dma read or write operation is a win over transfers via polling assuming the spi clock is at or near the maximum for a given system clock speed.
Geoffrey
Breakeven point for DMA spi operations
Moderator: RoccoMarco
-
geoffrey.brown
- Posts: 122
- Joined: Thu May 07, 2015 9:47 pm
- Has thanked: 3 times
- Been thanked: 18 times
-
Giovanni
- Site Admin
- Posts: 14891
- Joined: Wed May 27, 2009 8:48 am
- Has thanked: 1202 times
- Been thanked: 996 times
Re: Breakeven point for DMA spi operations
Hi,
Polled operations are great when you need very low latency or exchanging just one or few frames at time. DMA operations are optimal for large transfer but have the extra cost of setting up DMA at begin, more latency.
Polled operations drawback is polling itself, the thread executing that is not yielding to other threads during operation, best used in low priority threads so it does not affect others.
Giovanni
Polled operations are great when you need very low latency or exchanging just one or few frames at time. DMA operations are optimal for large transfer but have the extra cost of setting up DMA at begin, more latency.
Polled operations drawback is polling itself, the thread executing that is not yielding to other threads during operation, best used in low priority threads so it does not affect others.
Giovanni