I have an algorithm that's very easy to run in parallel (think finding a best path), extremely CPU i...

# coroutines

[Marc Plano-Lesay]

I have an algorithm that's very easy to run in parallel (think finding a best path), extremely CPU intensive, and would actually benefit from running only n branches at once for n CPU threads (once a solution is found, branches can be culled eagerly, so exploring a solution fully quicker is better than exploring more than n at once). Is there a dispatcher that would block new coroutines until some are finished, in such a way that only n would run for n CPU threads? Or is this too high level for dispatchers and I have to write my own orchestration logic?

[ephemient]

https://kotlinlang.org/api/kotlinx.coroutines/kotlinx-coroutines-core/kotlinx.coroutines/-coroutine-dispatcher/limited-parallelism.html

[Marc Plano-Lesay]

Ha nice, didn't find that one somehow - thanks!

[Nick Allen]

Dispatchers.Default is max(cpuCount, 2)

[Marc Plano-Lesay]

It's surprising, I get noticably different results by limiting Dispatchers.Default to Runtime.getRuntime().availableProcessors()

[rnett]

I'm guessing because using availableProcessors() ignores hyperthreading

[Marc Plano-Lesay]

It returns the number of logical processors, so in my case 8 (4*2), which is what I expected, and from my observations Dispatchers.Default was able to leverage all the logical processors as well