Do multisession and multicore futures work on Apple Silicon (natively or in emulation)?

[mpodell]

My Late 2013 15" MBP is rapidly giving up the ghost, and I am keen to move into some new hardware digs. Given that native R on Apple Silicon is not fully baked (as discussed here), I'm left with the option of running R in emulation (Rosetta 2) on a new M1 13" MBP, moving to a Linux rig, or just living in the cloud. (A fourth option to drop$3K+ on an Intel 16" MBP doesn't seem to be a wise choice when Apple will be EOLing them this year--or perhaps R will shun ARM architectures so get an Intel rig while I can?)

To better inform a decision, does anyone know if multisession and multicore futures work in emulation on the M1 SoC? If so, will availableCores() see both the four high-performance cores and the four high-efficiency cores (i.e. eight cores), only one set or the other (only four cores), or none?

To be sure, futures running on Apple Silicon via emulation is not the only issue (there are many more), but any insights w/r to multisession and multicore would be much appreciated, as I have yet to find anything on the interwebs speaking to this issue.

Thanks.

[HenrikBengtsson]

I don't own Apple equipments but this is what I can say:

• multisession workers are just regular R sessions running in the background and taking commands over socket connection. Except for being set up via the parallelly package, they're basically parallel::makePSOCKcluster() workers. Because of this, I'd be extremely surprised if this would not work. I expect them to work on all OSes and architectures where R runs.

• multicore workers rely on forked parallel processing and use the parallel::mclapply() framework internally. Forked processing is quite low level and probably requires getting libraries and compilers correct. So, it might be that those will be unstable for a while until thoroughly tested, etc. I would ask about parallel::mclapply() on the R-devel mailing list. If you get a positive reply there, then multicore will work too.

[mpodell]

Thank you for your response. I will follow up with R-devel regarding parallel::mclapply() as you suggest and report back here.

[mpodell]

Asked and answered (mclapply expected to work just fine in emulation).