when does ELL support micro:bit

[huoyunjie]

Hi, all, This project is very cool. I develop some products with ARM Cortex-M processor. I want to know when we can run ELL in micro:bit. Thanks for your hard work on this project!

[Tryljs]

我看了下生成的文件 应该是支持ARM端的

[huoyunjie]

多谢！我还没跑过，一直没有看到在micro:bit上运行的例子！

[Tryljs]

这个我是猜的 我也没有实现

[lovettchris]

We have not done it yet, and the nrf51822 chip on the microbit is very resource constrained (256kb flash and 32 kb RAM), so it would only be able to run a very small model, perhaps a ProtoNN based model might work. The following compile.exe command line might work:

ELL\build\bin\release\compile … --target custom --cpu cortex-m0 
--triple armv6m-unknown-none-eabi --features +armv6-m,+v6m --numBits 32 

But ELL models typically use a lot of floating point, so I would recommend instead you choose a board that has a Cortex-M4f processor at least, like the newer nRF52840 from Nordic. The following command line builds Cortex-M4F code, making use of the vectorized floating point instructions:

ELL\build\bin\release\compile … --target custom --numBits 32 
--cpu cortex-m4 --triple armv6m-gnueabi --features +vfp4-sp-d16,+soft-float

[lovettchris]

See also Building an ELL model for ARM Cortex-M4