Runtime Dispatch

Every call runs on the CPU and dispatches to the best kernel of Device.cpu.capabilitiesEnabled, a Capabilities OptionSet:

import StringZilla

// `enabled` is what dispatch uses: detected on this CPU AND compiled into the binary.
let cpu = Device.cpu
print(try cpu.capabilitiesEnabled)                 // like "serial,neon,neonaes,neonsha"
print(try cpu.capabilitiesEnabled.contains(.neon)) // `Capabilities` is an `OptionSet`

// The two raw axes, when you specifically mean one of them:
let onThisCpu = try cpu.capabilitiesDetected
let inThisBinary = cpu.capabilitiesCompiled

// Prepare each thread that runs kernels:
try cpu.configureThread(cpu.capabilitiesEnabled)

• capabilitiesDetected is what the device can execute, from CPUID or HWCAP on the CPU.
• capabilitiesCompiled is what this binary contains for devices of its kind, from the ISA probes at build time.
• capabilitiesEnabled is what dispatch uses, both axes at once, and on the CPU always contains .serial.
• configureThread(_:) prepares the calling thread for the kernels of its argument, once per thread that runs them.

Reach for capabilitiesEnabled unless you specifically mean one of the raw axes. capabilitiesDetected describes the machine and says nothing about whether a kernel was compiled in, so a build whose ISA probes failed still reports your CPU’s full feature set while containing no SIMD kernels at all.

A Device is the host CPU or one GPU of a runtime, named by that runtime’s own ordinal:

let gpus = try Device.count(.metal)            // throws without a Metal device
let gpu = try Device(kind: .metal, ordinal: 0) // throws past the last one
print(try gpu.capabilitiesEnabled)             // like "metal"

configureThread(_:) throws a StringZilla.Error on a GPU, which keeps no thread state of its own: this package only reports GPU capabilities. .cpus, .gpus and .any group the CPU capabilities, the GPU ones, and all of them.