colstore.use_passive_openmp_wait

colstore.use_passive_openmp_wait()[source]

Opt into OMP_WAIT_POLICY=passive so idle OpenMP threads sleep.

The gather kernel runs short, bursty parallel regions; OpenMP’s default active wait makes idle worker threads busy-spin between them, burning CPU on otherwise-idle cores. passive makes them sleep instead. This is opt-in and never applied automatically: OMP_WAIT_POLICY is process-global – it governs every OpenMP runtime in the process (NumPy, the C++ kernel, …) – and the per-call gather thread cap already bounds colstore’s own spinning.

The policy is read once, when the OpenMP runtime first starts its thread pool (the first gather, or another OpenMP library’s first parallel region), so set it as early as possible – ideally before the first read. Returns True if this call set the variable, False if it was already set (and left untouched).

Return type:

bool