Support mc.cores greater than number of physical cores

[benfasoli]

This adds a processes_per_node argument which is passed to parallel::mcmapply(mc.cores = processes_per_node). Specifying processes_per_node > cpus_per_node can more efficiently utilize nodes supporting hyperthreading.

The default behavior is processes_per_node = cpus_per_node for backwards compatibility.

Closes #56

[qdread]

Thanks @benfasoli ! I will take a look shortly.

[qdread]

Hey @benfasoli I'm testing out the PR but I think there is a bug still: check out for example line 173 in slurm_apply.R. There you need to pass the processes_per_node argument to whisker.render(). Otherwise, the slurm_run_R.txt template generates a slurm_run.R file with an empty argument for mc.cores. That would need to be fixed in a few places, I think. Do you mind going through and fixing that in the relevant places? thanks!

[benfasoli]

Nice catch. processes_per_node was being passed to the SLURM submission script template (where it is unused) and not to the R script template.

[qdread]

Wow that was quick! Thanks! 💯 ... I probably could have done it but I thought it would be fresher in your mind where all it was needed to fix.

[benfasoli]

Happy to help! Looks like this passes devtools::test() now.