Processor counts

Processor counts.

Source: runtime/async/cpus.pep

logical_cpus

logical_cpus(): ulong

Returns the number of logical processors the operating system can run threads on: every hardware thread, so a core with hyperthreading (SMT) counts two or more times.

This is the upper bound on how many threads can run at the same moment, and a sensible default size for a pool of CPU-bound worker threads. Inside a virtual machine it counts the processors the guest sees. On Linux, a CPU quota on the process’s cgroup or any cgroup above it (cgroup v2’s cpu.max, or v1’s cpu.cfs_quota_us over cpu.cfs_period_us, as docker run --cpus sets) caps the count, rounded up to whole processors, as Go and the JVM do: a quota of 1.5 processors counts as 2.

Returns

• the number of logical processors, at least 1

See also

• physical_cpus

Example

import async;

main(): int {
    print(async.logical_cpus(), " logical processors\n");
    return 0;
}

physical_cpus

physical_cpus(): ulong

Returns the number of physical processor cores: hardware threads that share a core through hyperthreading (SMT) count once.

On Linux the count comes from the kernel’s CPU topology (/sys/devices/system/cpu), on macOS from hw.physicalcpu, and on Windows from GetLogicalProcessorInformationEx. Inside a virtual machine the guest cannot see the host’s hardware, so this counts the cores the hypervisor presents. Where the topology is not available, it returns logical_cpus().

Returns

• the number of physical cores, at least 1 and at most logical_cpus()

See also

• logical_cpus