search

pytorch / glow

Compiler for Neural Network hardware accelerators
Apache License 2.0
3.19k stars 683 forks source link

Adding a library for a new backend, CMSIS. #6101

Open fvalasiad opened 1 year ago

fvalasiad commented 1 year ago

Hello!

Currently implementing a new backend, CMSIS, which utilizes the CMSIS nn library, a collection of neural network kernels implemented using SIMD instructions to maximize performance on arm cortex processors.

How do I go about adding the library to be used by the backend? It is not really a library you are gonna find in your average linux distribution's repositories, unless the user manually builds it & installs it.

The way I see it we have three options:

  1. Require that the user has done so, that is, build & install it in his system, and add it as a dependency.
  2. Add it as a git submodule
  3. Copy paste it manually and have maintainers update it as needed occasionally.

Which one do you prefer? Have you got any alternatives to propose that I haven't thought of? Would love to know!

Expecting to hear from ya soon.

gianfranco95 commented 11 months ago

I would prefer the second option. Is there the possibility to contribute to CMSIS backend implementation?