Recent macOS Builds?

I noticed that the mod-host and mod-ui contain some macOS specific code. Has anybody done a recent build on macOS? If so, any tips/pointers/warnings?

Has anybody looked into any optimizations for M1 Macs? We likely have a Mac Studio in the pipeline and I’m curious if anybody’s started anything to take advantage of either the GPU or Neural Engine cores.

1 Like

While you can run those on macOS, you need a working JACK setup first.
And then you need to port the plugins too.

All doable, if someone wants to go and put in the effort for it.

Personally speaking, sounds much more interesting to get the plugins and host working on a browser (through web assembly) than porting to macOS. Performance is not comparable, but since the plugins we have were made for somewhat-lower-spec machines (compared to desktop pcs), it should mostly be fine…
My initial tests for this, which you can see in Emscripten-Generated Code are quite good. LV2 plugins on the browser :joy:

3 Likes

@falkTX : Thanks.

Some of it is a “More Power”/“What can I do with my new tool?”, the other part is that some folks have used ML to replicate effects and I’m curious if I could take one of the models (assuming it’s 16-bit clean), load it using Apple’s CoreML, and then have it routed to the Neural-Engine that’s on the new M1/M2 chips.

1 Like

Could always try a linux vm?

2 Likes

@dreamer : macOS is BSD-ish under the hood. Plus, doing a quick skim of what’s on GitHub reveals theres’s already macOS #ifdefs in the mod-host and mod-ui codebase, so the modules are already at least part way. there.

I dug into this. The ML-cores built into the M1/M2 chips are essentially optimized for post-training convolution (16-bit w/ a reduced operand set), so I’d have to down-sample anything I’d want to run through them.

That said, it would still be interesting to see how much throughput could be gained with generic parallelization, or even with full CoreML migration. I will likely finish the PiSOUND/Raspberry Pi4 build first.

3 Likes