USB Host: Polling Uses Lots of CPU

[samblenny]

Currently (10.0.0-beta.2 on Fruit Jam), it seems like polling for USB host device input events with usb.core.Device.read() uses a lot of CPU time compared to reading GPIO pins or I2C. I know this is vague. All I have at this point is anecdotal observations that USB polling in combination with picodvi and/or synthio appears to load down the RP2350 CPU (slow display updates, probability of dropped/stuck notes goes up at higher synthio sample rates, etc). For example, last year (around 9.1.4), when foamyguy ported my Pumpkin Toss game to use I2C gamepad input instead of USB gamepad input, the animation loop ended up running way faster.

I'd like to find a way to speed up USB polling for synth and game projects so that more CPU time is available for displayio and synthio background tasks. Maybe this could be improved by using asyncio or something?

I'm posting this issue at tannewt's request to split out one of the points I mentioned in USB Host Improvement Quest Umbrella Issue #10551. The point of this issue is to let people know I'm planning to work on this myself and to ask for comments and feedback.

[tannewt]

I suspect this is due to the 1000 hz polling rate of full and low speed USD devices. So we have to wait 1/1000 of a second to get the response when doing a read.

I'd be curious how pyusb on a linux device seems. I wonder if the OS is doing buffering and then the read just reads from the buffer.

[samblenny]

@tannewt For this one, I'd like to try profiling with GPIO pins and a logic analyzer. Do you have a favorite code snippet for quickly toggling pins from C?