DMA self-chaining

[DynamoBen]

I have a project where I want to do DMA self-chaining. The rp2 implementation doesn't allow for this, so I've used a second DMA channel to retrigger the first (CHx_AL3_READ_ADDR_TRIG). While this works it burns up a DMA channel. I've considered ring but the data I'm sending is an odd size so that doesn't work (514 words). Is there a good way to self-chain without the needing a second channel?

[GitHubsSilverBullet]

The 2040 hardware does not allow self chaining at all.
The only way would be to use an IRQ at the end of a transaction. If the target supports FIFOs then one loses almost no time (IIRC an IRQ happens n-elements before DMA finishes)

[DynamoBen]

OK thanks, I thought this was just a limitation in rp2 not in the rp2040 itself. I'll try IRQs, my concern was the performance impact of having the CPU servicing IRQs for up to 8 PIOs but I'll give it a try and see if it's actually an issue or not.

[GitHubsSilverBullet]

I thought this was just a limitation in rp2 not in the rp2040 itself.

Not sure what you mean by that.
The RP2040 can't self-trigger, the RP2350 can.

If your ›rp2‹ refers to Micropython, then you'd be amazed how little of the true capabilities are reflected in µPy.

[peterhinch]

I'm amazed that an effective high level language implementation of DMA exists.

[GitHubsSilverBullet]

Peter, I wasn't my intention to judge at all.
It's just that the current implementation barely touches the DMA systems true capabilities.
And that's OK, after all, FLASH space is limited on these devices.
If one needs more or refined functionality, one has to write the necessary code.
That's the same with ADC functionality in a parallel thread.

[DynamoBen]

I greatly appreciate whats there in µPy, having the current DMA high level implementation has saved me a lot of time! Yes the pico has some pretty amazing capabilities, while the pico isn't new to me micropython is I look forward to any new developments that might come in the future with this µC.