Porting Espruino to a Cortex M0 platform

[espruino-discuss3]

Posted at 2015-02-08 by jose.angel.jimenez

Hi all,

I am an electronics engineer, with experience in several MCU and DSP architectures but with very limited knowledge of JS and the ARM Cortex architecture.

Trying to make a rough estimation about the effort it may take me to port Espruino to a platform based on ARM Cortex M0, 256KB Flash, 32KB RAM.

Thanks a lot for your comments and suggestions,
Jose Angel

[espruino-discuss3]

Posted at 2015-02-08 by @allObjects

@gfwilliams + Javascript ---> Bushfire ;-)

---

Posted at 2015-02-09 by @gfwilliams

Hi,

It depends on the make of chip. Nothing in the interpreter is particularly platform dependent, so getting it running should be trivial.

Accessing all the peripherals though? That's usually what takes a lot of time. If you use ST then at least a lot of that work has been done for you.

[ponyatov]

Most Cortex M0 chips I see around has a very small SRAM memory, some of them less then 8K.

At a first glance current Espruino execution model and JS itself wants at least 64K of SRAM to be more or less usable.

For such a small chips there is only two options:

• native C/C++ code (embedded C++ still wants some skills to keep it in footprint)
• low-level bytecode interpreter + some outside compiler runs on host or as a part of Espruino IDE runs in a browser
  • outside compiler will remove the most large components such as source-level interpreter and filesystem
  • single-byte opcode for stack machine can be smaller then the same code compiler into native MCU assembly

But I'm not sure it ever possible to AOT compile JavaScript at all due to its very dynamic nature.

As a variant for such small devices some sort of bytecoded FORTH/CLI can be used for REPL experimenting, but I also does not know anybody used FORTH for tasks more complex then LED blinking or nnCron task scheduling.

[fanoush]

Most Cortex M0 chips I see around has a very small SRAM memory, some of them less then 8K.

At a first glance current Espruino execution model and JS itself wants at least 64K of SRAM to be more or less usable.

nrf51 (the original Micro:bit) is M0, has 16K RAM and works with Espruino including the BLE stack (which takes away almost 8K of RAM!). It can run out of stack with some deeply nested javascript code with that remaining 8K of RAM but in general works pretty well for what it is. There is also 32K variant of that chip used in some smart bands and watches.

There are also some M0 based BLE chips from China even with 64K - like the PHY6220 used in very cheap thermometers, or the BlueX RF03 used in smart rings (and Amazfit Neo) has 208 KB, the old DA14585 from Dialog has 96K. The RP2040 is even dual core M0 and has 256K

On the other hand many STM32 chips are M3 and have limited RAM (32K,20K,16K,10K). I use Espruino with 20K variant of STM32F103 which is inside of those US$2 STLINK V2 debug dongles and it runs fine.

So to sum it up - the RAM size is unrelated to Cortex M variant and Espruino can run with 16K or RAM without major issues.