Expand description
Thread-count / core-placement specialization experiments.
Silicera-owned only: we vary the thread count passed into lab workloads and measure. We do not change global OS scheduling, affinity, or power plans. More threads is not assumed faster.
Structs§
- Thread
Count Point - One measured thread-count point.
- Thread
Count Report - Report for the thread-count sweep.
Functions§
- default_
thread_ sweep - Default thread counts to sweep (clamped to host logical CPUs).
- measure_
thread_ count_ sweep - Measure concurrency kernels across thread counts (Silicera-owned; no OS affinity).