Norns on Raspberry Pi

[ngwese]

Just wanted to let you know that I managed to build maiden on my Pi3 last night. I fumbled my way through it, and now it seems to work fine!

…if you have details on which versions of the toolchain you used (npm, node, …etc.) i can add that info to the README.

[zebra]

sorry, maybe i was misleading

what i mean by “even” is, matron’s various threads, and scynth’s various threads are correctly assigned to different cores when appropriate (like when total usage cracks 60% or so.)

the single heavy thread is scsynth’s audio callback. audio processing is not multithreaded. (in supernova it is, but supernova doesn’t work on ARM yet.)

to demonstrate, i made flin.lua much heavier by making 10000 calls to math.pow() on each update, and otherwise same situation as you describe (full count of running voices.)
top output looks like this

i don’t get any xruns or other artifacts. i do get xruns / artifacts when scsynth main thread uses more than ~60%, which is a bummer, and would be great to improve this threshold somehow (larger buffer size doesn’t help as much as i’d expect.)

we actually had more problems with IO interrupts using realtime kernel, so dropped it.

(i’m not sure why it looks like scsynth’s audio thread takes up more cpu than any one core. but i assume it is an artifact or subtlety of top’s reporting / update cycle, not reality.)

its also possible to create dense enough OSC -> synth traffic that scsynth’s network thread becomes significant.

so yeah, it would be great if scsynth’s audio proc could be spread out, but for the near future it can’t.

[TheTechnobear]

thank you very much for this…

ok, from this, Im pretty much seeing the same as you then.
basically, im getting xruns when the main scsynth thread hits a bit over 60%.
(id hoped sc was properly threaded, as this is a painful issue on pd too on these platforms)

my matron threads look about the same, which is good considering this is where my new code is.

I think the only thing I can see is your scsynth is consuming less cpu, so either my supercollider build is not optimised, or im just running different/heavier tests than you.

did you just follow the normal SC build steps for 3.9, or did you add any extra flags?
(did you apply the timer fix… I forgot too )

anyway, thank you, this has definitely narrowed it down to SC, rather than ‘linux optimisation’.

on the positive side, whilst Im talking about improvements… really the system is performing very well, its smooth and at ‘normal usage’ , no xruns - oh I got maiden running yesterday.
when ive been playing with it (rather than coding ) , its proving to be a really fun platform.

one question (which apologies im sure you have already answered somewhere)
is there a way to create new synthdefs on the fly? im guessing not from maiden?
but i assume i can thru a remote sc lang.
but then can i interact with these thru lua? create an engine? add new parameters?

[shreeswifty]

there is a post related to converting existing synthdefs reltively rapidly by zebra i think it’s in norns help or dev its not exactly but it’s nice to know we are all getting to the lets make sounds part

[shreeswifty]

@shreeswifty i finally sat down and made a basic but extremely commented script using SoftCut engine (same as MLR.) maybe something like this can be integrated into a study. it took me longer than necessary cause i also got ambitious with making a barebones sampler engine demo; maybe will get around to finishing that.

github.com

catfact/dust-1/blob/softcut-sample/scripts/emb/softcut_sample.lua

--- softcut-sample
---
--- simple demonstration: play sample and randomly change rate

-- if the script sets this global variable,
-- norns will attempt to load the requested engine when launching the script.
-- 'SoftCut' is a polyphonic, tape-like record/play system with crossfading, used in MLR.
engine.name = 'SoftCut'

-- if the script sets this variable to a function, it will be executed when the engine is done loading.
function init()
   print("softcut_sample init")

   -- the global table 'engine' should now be populated with a lot of function-type fields,
   -- corresponding to "commands" declared in the engine's SC class.
   -- this will print them all (names and argument format) in the matron REPL:
   engine.list_commands()

   -- in SoftCut, all the voices share one large audio buffer.
   -- by default, they are assigned to read and write from adjacent regions.

This file has been truncated. show original

(2 asides:

• i’ve found that i can crash scsynth by flapping around past 0 with negative rates. dumb.
• dang, i’m not happy with the corners we cut for performance here. (linear resampling on playback, control-rate modulation of playback.) we’ve made some jack config adjustments, maybe we can squeeze the good-sounding stuff back in there.

these are related (ugen code) and i’ll put them on my list / on the GH issues.)

this may get you started

[TheTechnobear]

it’s nice to know we are all getting to the lets make sounds part

+1 to that

yeah, the lua side of using SC im familiar with (been playing with that after the first study)

what im quite interested in, is extending the SC side, so new SynthDefs, and i guess in Norns lingo Engines.

but im also interested, can be create these ‘on the fly’…
or what techniques can we explore in this area e.g. pattern defs
(and how is this support or not, by norns… or do we just use the convention sc gui?)

my main draw of SC is its dynamic nature, and I want to explore that, as well as the ‘UI’ interfacing elements of Lua.
(also im sure we at some point need to thing about whats more performant to do on the sc server rather than lua/sclang)

[shreeswifty]

Hey it just dawned on me that the pisound has two knobs and one button already so with some creative coding I am relatively sure with a few switches and routes I can solve the dev events stuff

[shreeswifty]

I want to do something similar because I want to load sctweets randomly with lua/buttons and encoder calls so it would require a similar vibe

[TheTechnobear]

pisound has two knobs and one button already

button yes , you can get to via gpio, or simply put script together that sends osc

knobs, are hardware gain… so they are not seen by the OS, so you can’t change their function. not a bad thing, input gain and output gain are very handy almost always

[Nordseele]

I’ve been tracking this thread and reading everything for a few days now, thanks for everything you’ve done !
I haven’t tried to install the Norns ecosystem on a RPI yet but I want to give it a try.
I have an old Monome 256 walnut, and also a Push 2, so I guess one of these options should work.

For the soundcard, I was planning to use an Expert Sleepers ES-8, do you think it would be possible to use an USB soundcard like that ? I use it with another RPi and PureData, but is it quite easy to « connect » the Norns environment and an external soundcard ?
The other option, or additional option I have is a Terminal Tedium Eurorack module, this one has three push buttons and 6 potentiometers, I wonder if Norns could live on this module, at the moment there’s a RPi3b in my build…

[lijnenspel]

Might wanna give this a try this weekend. I’m thinking of using a 2.4" SPI oled, launchpad mini, rme fireface ucx (in class compliant mode) and some ec11 encoders + simple buttons on a breadboard. I’m not great with linux, but okay with Lua

[shreeswifty]

true. i was hopeful on the input/output knobs – i wonder if i can get a mini arduino but i think piSound uses all the pin, So …

[TheTechnobear]

Ive done a full clean build - based on norns-image dev setup

and taken complete notes as I did it.

nornsinstall.txt

Note: I don’t take any credit for this, as the write-ups by @tehn & co are great, and cover 99% of it.
but there are a few things i changed on they way, and a few things to note, as we are not putting on norns hardware

UPDATE: I updated the instructions to point out that most can probably just use the stock kernel, Ive tested and it was fine… so Id only build your own kernel if you need to for a specific reason, and you can do it quickly (eg. by cross compiling.

limitations:

• I’m not repeating things (like how to build a kernel, and steps already in the norns documentation),
• its not specific , so you need to know a few things like how to specify your sound card in jackd
• ive not built the norns docs, so that not included

a request, if someone uses this, then please also take notes of any errors, and perhaps format the doc as you go along - so we can each improve it a little

(@tehn not sure if there is a better place to put it, Ive tried not to rehash whats in the other docs too much, rather refer, exception was where things were perhaps not entirely clear to me … you are of course free to do anything you wish with it)

anyway good news is, seems to be working perfectly with my push2.

---

some important things, that tripped me over,
and perhaps my be useful comments for norns devs (or not )

• setup.sh - config/interfaces will overwrite you wifi setup , and hardcode to a static ip address…
  I guess this is the norm for norns, as you can then use the screen to setup wifi (?) , but when you’ve no screen then thats going to be tricky, so id advise against overwriting
  (it caught me out, as i was missing a file required for push2 to launch properly )

• rasps-config - localisations = all - dont do it it takes ages, and you only really need your location…
  obviously norns hardware is international so needs the all , jic

• norns-image/config/norns-maiden.service
  this refers to maiden.arm, as presumably its being cross compiled, so you need to change this.

• i2cset I comment these out as they are norns hardware specific and might conflict with your hardware (but probably will be ok)

final note… this is not identical to the norns hardware release
this builds a dev image, which means its the latest (so unsupported?!) code, rather than released code,
the kernel is different/newer (4.14), and we may have different options.
so you cannot compare directly…

ive not tested things like usbsync

sleep , really does just sleep - it does not do the shutdown I expected.

I think for my personal needs, Im going to add a shutdown and probably restart option, hopefully this can just be a simple lua app, so I dont have to hack matron more than necessary.

anyway I hope this is all helpful to someone…
thanks to the norns team for all the effort,
as i type this i listening to “why?”, to which the answer is “because I can!”

enjoy
Mark

EDIT:
@tehn Ive noticed the wifi.lua and scripts dont appear to be working properly…
is this something worth investigating or a red-herring?
(I dont need it as I can just edit /etc/network/interfaces and all works)
first error is Failed to connect to non-global ctrl_ifname (nil) error: No such file or directory
so looks like ctrl_interface file?
(this is by running ~/norns/wifi.sh scan)
hotspot doesn’t appear to work from the UI, but does appear to work from wifi.sh hotspot

anyway, perhaps this is a red herring so wont waste your time, unless its an issue you hearing about elsewhere

[reijo]

nornsinstall.txt

wow that’s great!! thank you! gonna check this out with my pi 2 lying around…

[shreeswifty]

you might not be excited by the performance you get with a 2, but i assume you will know that

[reijo]

Yep I thought so…If the weather is going to be as bad as it announced this weekend I’ll give it a try.

Also I was thinking about getting a pi3 for other projects…so maybe its time now…

[TheTechnobear]

ok, not really rPI related … but I think @jasonw22 was asking about using a virtualbox?!

so the answer is yes, it can be made to work… Ive got it working under vmware

I actually did all the Push2 dev for matron on Vmware Fusion (mac) hosting Ubuntu 18.04 (64bit) , but I didnt bother trying the audio till this evening (for fun, over a glass of wine)

‘as expected’ jack and the ‘virtual audio driver’ did not work… (Ive logged a support request)

so I went a different route, VMware has excellent usb support,
so I plugged in a portable USB audio interface (Zoom H5 in AI mode) , then gave it to vmware to host, told jack to use this - and viola, the lovely dulcet tones of Norms, using my Push 2 to control it all.
(no xruns or anything, so thats a bonus!)

why? - why not?

no, actually, its useful for development purposes, Ive got better tools on desktop Ubuntu, so its easier for me to try things there, get them working … git push them, then bring them down to the rPI for testing.

so if you want to try norns out on the desktop give it a whirl, as ive mentioned previously in this thread the norns documentation is really good.

[jasonw22]

ok, not really rPI related … but I think @jasonw22 was asking about using a virtualbox?!

so the answer is yes, it can be made to work… Ive got it working under vmware

woohoo! I’m going to try to make time to give it a whirl this weekend.

[TheTechnobear]

key thing to note… if your building from ‘scratch’, the distros are still on supercollider 3.7 or 3.8,
and you need supercollider 3.9.3, but thats a really easy build (supercollider website gives steps)
apart from that, norns docs covers all thats needed, and theres a few extra hints in this thread (re go, and glide)