search

csteinmetz1 / ronn

Randomized overdrive neural networks
https://csteinmetz1.github.io/ronn/
Apache License 2.0
133 stars 11 forks source link

arm64 guidance #7

Open markhanslip opened 7 months ago

markhanslip commented 7 months ago

Hiya,

This seems super cool and I really want to have a play with the real-time version, but I'm having a torrid time trying to get it to run on an M3 MacBook.

So far DAW-wise I've tried loading the pre-built plugins in Reaper (and in 2 other free DAWs that I'm too embarrassed to name) in 'Rosetta mode' and in ARM mode without joy. I've gotten closest to loading the plugin in a DAW with the AU version (the VST3 isn't scanning at all) in ARM mode but it still doesn't load fully.

I also tried building from scratch with various tweaks to the first cmake call (e.g. DCMAKE_OSX_ARCHITECTURES==x86_64;arm64 for a hopefully universal build), the OS version specified in CMakeLists.txt (e.g. set(CMAKE_OSX_ARCHITECTURES "x86_64;arm64")) and a more recent version of LibTorch and haven't managed to build a working version yet.

The Standalone builds and launches on M3 without changing the build process apart from obviously cmake --build build --target ronn_Standalone --config Release, but nothing much is happening sound-wise besides clicks and pops when changing parameters with bias selected, presumably because it was built for Intel.

If anyone has managed to get this working on an Apple ARM machine either in a DAW with the provided versions or by building from source then I'm all ears. As you probably tell, cmake and JUCE aren't exactly my wheelhouse and while I'm very happy to learn new stuff I'm also eager to get this working as a musician.

Thanks

markhanslip commented 7 months ago

It seems the files libtorch.dylib, libtorch_cpu.dylib and libc10.dylib are the culprits here, will update if I find a fix