Module par 

Splitting work across threads.

This is the plain version the reference kernels use: scoped threads per call and a shared counter that hands out tasks in order. The pinned worker groups from spec/10-cpu.md replace it when the plan executor lands. Nothing a task computes may depend on which thread runs it, which is what keeps the kernels deterministic under any thread count.

Structs

Shared

A buffer many threads write into at once, each to elements no other thread touches.

Functions

available

The number of threads the machine offers.

for_each

Runs f(i) for every i in 0..n on up to threads threads.

map

(0..n).map(f).collect() on up to threads threads, in order.