Should we spawn n?

[abrown]

The definition of thread.spawn includes a value n specifying the number of parallel invocations to execute. The motivation for this addition is described here — in summary, generality and performance. I'm curious if there are strong opinions either way: should thread.spawn be able to spawn n invocations?

[conrad-watt]

No real objection, although since thread IDs in the MVP would need to be given out via a global lock (as sketched here), we may want to think about how much faster this would actually be.

EDIT: Sorry, I misunderstood the sketch above. I think the general point of "how does each thread get a distinct TID" is worth thinking about, but I think you're right that it can be solved with an atomically-incremented shared global.

[rossberg]

Personally, I think that we should start with the simplest primitive. Massively parallel hardware is worth considering, but the benefits perhaps sound a bit hypothetical at this point.

[rossberg]

FWIW, I also do think that having to fiddle back thread IDs through global state is going to be an unnecessary, significant complication for many use cases. So even if we have spwan_multi, we'd still want a simple spawn for most use cases.

[yamt]

i'd say "no" for now as the "multiple" version would make argument passing and error handling considerably more complex.