sleep itself seems to take 9 ms

Toit

#sleep itself seems to take 9 ms

Thread channel in help

RobvanLopik 04/03/2023 02:08 PM

try this

main:
  t0 := Time.monotonic_us  
  1000.repeat:
    sleep --ms=1
  print (Time.monotonic_us - t0)

results in ~9800 instead of 1000. Increasing the sleep time makes the discrepancy go away.

sleep --ms=0

does not take extra time.
I found this because the stepper motor seemed much to slow

floitsch 04/03/2023 02:47 PM

There is a certain overhead for setting the sleep up. Is that what you are experiencing or is there something else happening for small numbers like 1?

RobvanLopik 04/03/2023 03:07 PM

I think 9 ms is a lot of overhead. In the stepper 1 ms is used, that effectively is 10 ms making it 10 times slower than expected

floitsch 04/03/2023 03:08 PM

I agree. Just making sure that's the actual issue.

RobvanLopik 04/03/2023 03:13 PM

When
I run on host there is no overhead! 50 microseconds turns into 133 us

floitsch 04/03/2023 03:14 PM

Might Reuther be the setup cost or using a timer that isn't precise enough. Both are different on the esp32

floitsch 04/03/2023 03:17 PM

Just fyi: most of us are on vacation for the next week, so there is likely nothing going to happen in that time.
1. Please remind us when we come back (just in case we forget).
2. As a workaround I would:

end := Time.monotonic_us + 10_000
while end > Time.monotonic_us: yield

RobvanLopik 04/03/2023 04:14 PM

Actually, when the host is Windows, the overhead is 15 ms instead of near 0 on Linux

bitphlipphar 04/03/2023 05:36 PM

It's a timer resolution issue, not really overhead per se.

bitphlipphar 04/03/2023 05:37 PM

On FreeRTOS, the smallest amount of time we can wait is a tick, which defaults to 10ms. Haven't looked at Windows lately.

bitphlipphar 04/03/2023 05:38 PM

We eventually hit a pthread_cond_wait: https://github.com/toitlang/toit/blob/995ecf05f0659ac0d3347bd84d181e54a4def311/src/os_win.cc#L99.

toit/os_win.cc at 995ecf05f0659ac0d3347bd84d181e54a4def311 · toitla...

Program your microcontrollers in a fast and robust high-level language. - toit/os_win.cc at 995ecf05f0659ac0d3347bd84d181e54a4def311 · toitlang/toit

bitphlipphar 04/03/2023 05:40 PM

It's called from here: https://github.com/toitlang/toit/blob/995ecf05f0659ac0d3347bd84d181e54a4def311/src/event_sources/timer.cc#L121.

toit/timer.cc at 995ecf05f0659ac0d3347bd84d181e54a4def311 · toitlan...

Program your microcontrollers in a fast and robust high-level language. - toit/timer.cc at 995ecf05f0659ac0d3347bd84d181e54a4def311 · toitlang/toit

bitphlipphar 04/03/2023 05:56 PM

If we can find a better primitive to support precise, cancelable sleep I'd be very supportive of using that instead of pthread_cond_wait with a timeout.

RobvanLopik 04/03/2023 06:04 PM

Thanks for the info. Maybe worth documenting, because the form sleep --ms= suggests milliseconds are meaningful. Anyway, this makes the steppermotor 10 times slower than expected

bitphlipphar 04/03/2023 06:05 PM

Yeah, that's a valid point. Maybe we can find a better way to support that than repeated sleeps. Thanks for bringing this up

RobvanLopik 04/03/2023 07:11 PM

For the stepper I think I have seen software that uses RMT. For the sleep you may need to roll your own; after all you have a microsecond clock in Time.monotonic_us

bitphlipphar 04/03/2023 07:20 PM

You can use RMT on Toit.

bitphlipphar 04/03/2023 07:22 PM

The one-wire package uses this: https://github.com/toitware/toit-1-wire/blob/main/src/one_wire.toit.(edited)

18 messages in total