Crate fluxbench_core

Expand description

FluxBench Core - Worker Runtime

This crate provides the execution environment for benchmarks:

Bencher struct for iteration-based benchmarking
Timing: std::time::Instant per timed region, with the empty-timer overhead subtracted. Cycle counts come from RDTSCP (x86_64) or CNTVCT_EL0 (aarch64), and are 0 elsewhere
Global allocator interceptor for memory tracking
CPU affinity pinning for stable cycle-counter readings

The async feature (default) adds iter_async and drive_async_body on tokio. Without it, the crate has no tokio dependency.

Structs§

Bencher: The Bencher provides iteration control for benchmarks.
BenchmarkDef: Benchmark definition registered via #[flux::bench]
BenchmarkResult: Result of a single benchmark run
ChartDef: Chart definition registered via #[flux::chart].
CompareDef: Comparison group - groups multiple benchmarks for side-by-side comparison
GroupConfig: Run config a #[flux::group] sets for every benchmark in it and in its child groups.
GroupDef: Group definition registered via #[flux::group].
Instant: High-precision instant for benchmarking
MeasureConfig: Config for one benchmark measurement.
ReportDef: Report definition registered via #[flux::report].
Timer: Timer for measuring benchmark iterations
TrackingAllocator: Tracking allocator that wraps the system allocator
WorkerMain: Worker main loop

BatchSize: Inputs per timed batch for iter_batched and iter_batched_ref.
ChartType: Chart kind for #[flux::chart]
IterationMode: Mode of iteration for the benchmark
MeasureError: Error from one benchmark measurement.
Severity: Severity levels for CI integration
Throughput: Work one routine call processes.

DEFAULT_SAMPLE_COUNT: Default number of samples to collect (matches Criterion)
HAS_CYCLE_COUNTER: Whether this platform provides hardware cycle counters (x86_64 RDTSCP or AArch64 CNTVCT_EL0). When false, cycle counts are reported as 0 and only wall-clock nanoseconds are available. Whether this platform provides real cycle counters.
LARGE_INPUT_CHUNK: Most inputs a BatchSize::LargeInput batch holds.
MIN_SAMPLE_COUNT: Minimum sample count. Two samples give a defined variance.
SETUP_CHUNK: Most inputs a BatchSize::SmallInput batch holds. iter_with_setup uses it too.

current_allocation: Get current allocation statistics
cycles_overhead: Cycles one empty timer pair costs. It is 0 without a cycle counter.
drive_async_body: Run an async benchmark body to completion on rt. Its output is dropped.
reset_allocation_counter: Reset allocation counters (call before each iteration)
run_benchmark_loop: Run one benchmark body and return its result.
timer_overhead_ns: Nanoseconds one empty timer pair costs.