Expand description
§silicera-lab
Measurement laboratory: microbenchmarks across memory, integer, float, branch, and concurrency domains; cache-size workloads; tournaments; and a terminal UI for live visualization of real measurements only.
Numbers shown in the lab UI come from the measurement engine — never from hard-coded marketing placeholders.
Re-exports§
pub use arms::run_concurrency_arms;pub use arms::run_domain_harness;pub use arms::run_float_arms;pub use arms::run_full_harness;pub use arms::run_integer_arms;pub use arms::run_memory_arms;pub use arms::ArmsHarnessConfig;pub use arms::ArmsReport;pub use arms::HarnessSuiteReport;pub use bench::AlignmentBench;pub use bench::BranchBench;pub use bench::CacheTarget;pub use bench::ConcurrencyBench;pub use bench::FloatBench;pub use bench::HostIsaDot;pub use bench::HostIsaReduce;pub use bench::IntegerBench;pub use bench::MemOpBench;pub use bench::MemoryBench;pub use bench::WorkloadKind;pub use calm::run_calm_check;pub use calm::CalmCheckReport;pub use experiments::run_experiment;pub use experiments::ExperimentId;pub use experiments::ExperimentReport;pub use native_artifacts::default_artifact_kernels;pub use native_artifacts::find_workspace_root;pub use native_artifacts::run_native_artifact_compare;pub use native_artifacts::run_native_artifact_suite;pub use native_artifacts::suite_to_workload_entries;pub use native_artifacts::ArtifactArm;pub use native_artifacts::NativeArtifactReport;pub use native_artifacts::NativeArtifactSuite;pub use profile_gen::generate_profile;pub use profile_gen::measure_dispatch_overhead;pub use profile_gen::measure_dispatch_overhead_report;pub use profile_gen::measure_size_classes;pub use profile_gen::size_class_winner_story;pub use profile_gen::DispatchOverheadReport;pub use profile_gen::ProfileGenConfig;pub use profile_gen::SizeClassWinnerStory;pub use silicon_split::read_export;pub use silicon_split::sanitize_export;pub use silicon_split::split_report_paths;pub use silicon_split::split_report_profiles;pub use silicon_split::train_split_profile;pub use silicon_split::write_export;pub use silicon_split::write_machine_b_placeholder;pub use silicon_split::SiliconSplitProtocol;pub use single_machine::merge_artifacts_into_profile;pub use single_machine::run_single_machine_eval;pub use single_machine::write_eval_outputs;pub use single_machine::SingleMachineEval;pub use studies::measure_alignment;pub use studies::measure_placement;pub use studies::spot_check_profile;pub use studies::AlignmentReport;pub use studies::AlignmentSample;pub use studies::PlacementReport;pub use studies::PlacementSample;pub use studies::PlacementStrategy;pub use studies::SpotCheckItem;pub use studies::SpotCheckReport;pub use threads::default_thread_sweep;pub use threads::measure_thread_count_sweep;pub use threads::ThreadCountReport;pub use ui::run_lab_ui;pub use ui::summary_cv;pub use ui::LabSessionRow;pub use ui::LabSessionSummary;
Modules§
- arms
- Portable vs host-ISA vs Silicera selection harness.
- bench
- Microbenchmark workloads spanning memory / integer / float / branch / concurrency.
- calm
- Soft environment calm / noise probe (single machine).
- experiments
- Experiment domains for Silicera lab demos.
- native_
artifacts - Dual-artifact comparison: portable codegen vs
-C target-cpu=native. - profile_
gen - Profile generation from lab tournaments — including size-class memscan/copy and size-scaled float/integer targets.
- silicon_
split - Silicon Split — multi-machine measured strategy comparison.
- single_
machine - Single-Zen evaluation pack — everything one physical machine can prove.
- studies
- Research studies: alignment, placement, and spot-check verification.
- threads
- Thread-count / core-placement specialization experiments.
- ui
- Terminal lab UI — systems aesthetic, real measurements only.
Functions§
- run_
simple_ tournament - Run a named microbenchmark tournament with two simple variants.