Add support for putget to PIO

[horazont]

PIOs support changing the FIFO configuration to support random access instead of acting as a FIFO (see for example Section 11.4.8 and 11.4.9. of the reference manual, as well as FJOIN_RX_PUT / FJOIN_RX_GET in the SMx_SHIFTCTRL registers).

As far as I can tell, this is currently not supported by the RP235x HAL.

I was thinking to add support for that (as I am currently bypassing the HAL for one of my PIO uses because of that) and I'd like to get a quick gauge whether I'm heading the right way. Here's what I was planning to do:

1. Add typestate to PIOBuilder which tracks the configuration.
2. Add separate RxGet / RxPut / RxPutGet types which provide access to the "fifo" contents.
3. Depending on the typestate, change the output types of build() to use the correct Rx* form as mentioned in point 2.

This will increase the complexity of the builder somewhat, but I think this is the only typesafe way of doing it.

While we're at it, we could also move the effect of the buffers() call into typestate, so that the return type of build() wouldn't contain a tx if the tx has been merged into the rx fifo, and vice versa.

Opinions / thoughts?

[horazont]

As far as I can tell, we can avoid adding extra typestate to the StateMachine, if we add typestate to Rx instead of making different types. Then the StateMachine::uninit call can just take any variant of Rx instead of having to keep track which kind of Rx the StateMachine has to consume on uninit.

[thejpster]

Sounds reasonable on first glance.

What's the use case for random access?

[horazont]

I use it for a 64 channel LED matrix driver. Two FIFO slots for 32 channels each. The PIO drives the shift register and sets 8 output pins of the RP2350 on each shift register cycle.

…

On 4 June 2025 20:24:30 CEST, thejpster ***@***.***> wrote: thejpster left a comment (rp-rs/rp-hal#935) Sounds reasonable on first glance. What's the use case for random access? -- Reply to this email directly or view it on GitHub: #935 (comment) You are receiving this because you authored the thread. Message ID: ***@***.***>

[thejpster]

oh, nice.

[horazont]

Oha, looks like initial support for the feature was added in 8328792.

I would still consider the typestate method a valuable improvement. 8328792 does not introduce the necessary changes to the Rx struct and I don't think there's a safe way to do this without typestate.

(Also, we need to ensure that autopush is turned off, because that's UB with RxGet or RxPut enabled, according to 11.5.4 of the reference)

[horazont]

I opened #936, but there's an open question which we should discuss, preferably over there. It already works for my use case, but there's a few bits which need to be fixed.

I'd like to get some guidance on the open question before that, though :).