Erlang ARM32 JIT is born(grisp.org)
grisp.org
Erlang ARM32 JIT is born
https://www.grisp.org/blog/posts/2025-10-07-jit-arm32.3
19 comments
and 32-bit arm (nothing wrong with it; just like tcl and erlang, it's alive and well)
For a certain definitions of faster
Don't Erlang and Elixir have a lot of effort being put into them?
I don't have any experience with ARM, but from what I've seen people write, isn't 32-bit ARM discontinued after v7?
Their motivation is explained in the first post of the series[1]
[1] https://www.grisp.org/blog/posts/2025-06-23-jit-arm32.1#why-...
[1] https://www.grisp.org/blog/posts/2025-06-23-jit-arm32.1#why-...
For their real motivation[0], click on hardware at the top of the page.
Their existing hardware is aarch32. It really is that simple.
0. https://www.grisp.org/hardware
Their existing hardware is aarch32. It really is that simple.
0. https://www.grisp.org/hardware
loved them complaining about having "only" 16 registers
There's still a huge embedded market!
Plenty of microcontrollers have a single-digit number of Cortex-M cores and memory/flash counted in the megabytes. It'll be decades until that market reaches the multi-gigabyte point, so why bother wasting a whole bunch of memory on 64-bit pointers?
I'm not quite sure why you'd want to run Erlang on it, but the hardware exists.
Plenty of microcontrollers have a single-digit number of Cortex-M cores and memory/flash counted in the megabytes. It'll be decades until that market reaches the multi-gigabyte point, so why bother wasting a whole bunch of memory on 64-bit pointers?
I'm not quite sure why you'd want to run Erlang on it, but the hardware exists.
> I'm not quite sure why you'd want to run Erlang on it, but the hardware exists.
https://nerves-project.org/#features has a decent pitch for why. (Most of the features listed here aren't features of Nerves-the-Elixir-IoT-runtime-codebase per se, but rather benefits of Nerves-the-toolchain enabling you to easily build lean, embedded Erlang [on Linux] firmware images.)
https://nerves-project.org/#features has a decent pitch for why. (Most of the features listed here aren't features of Nerves-the-Elixir-IoT-runtime-codebase per se, but rather benefits of Nerves-the-toolchain enabling you to easily build lean, embedded Erlang [on Linux] firmware images.)
> I'm not quite sure why you'd want to run Erlang on it, but the hardware exists.
Erlang is invented before IoT was a thing to facilitate distributed computing for telecommunication in a highly reliable manner. It makes perfect sense to adapt it for driving fleets of cheap IoT devices.
Erlang is invented before IoT was a thing to facilitate distributed computing for telecommunication in a highly reliable manner. It makes perfect sense to adapt it for driving fleets of cheap IoT devices.
No, it's a supported ISA on most v8-a and I believe all v8-m implementations.
It's the only ISA on Cortex-A32, but not sure if any mainstream chips were ever produced with that core.
(Depending on course if you want to get specific about Arm/Thumb/Thumb2, I lumped them all together above).
It's the only ISA on Cortex-A32, but not sure if any mainstream chips were ever produced with that core.
(Depending on course if you want to get specific about Arm/Thumb/Thumb2, I lumped them all together above).
That does not mean ARM32 implementations and uses are stopping any time soon. Afaik arm hasn’t even obsoleted armv6, although Linux distributions are starting to drop it.
Doesn't mean that machines won't be built with other chips for a considerable time.
That said, if you're putting something like Erlang on a chip, aren't one likely to want the extra memory (and performance) of a slightly newer SoC.
That said, if you're putting something like Erlang on a chip, aren't one likely to want the extra memory (and performance) of a slightly newer SoC.
Take a look at their products. Seems like they run bare metal Erlang on embedded devices.
Cortex-M chips will still be made for decades.
Gah, misread that as esp32 JIT, which would be eye opening!
esp32 is now also RISC-V so I guess it wouldn't be completely out of the question. But I guess you meant this flavor
https://www.cadence.com/content/dam/cadence-www/global/en_US...
https://www.cadence.com/content/dam/cadence-www/global/en_US...
Either TBH, I imagined the main issue would be ram, even with psram. EQMX is used a lot for IOT and it'd be interesting seeing more heavy loads on the edge.
I miss working with Erlang especially, but it's also certainly kind of a niche thing. Other languages are faster and have more effort being put into them.