|
STAR
Cal STAR — firmware, DAQ server, and engine design
|
Classes | |
| class | _C |
Functions | |
| _install_instrumentation () | |
| _one (cfg_path, max_it, restarts, seed) | |
| _run_condition (mode, cfg_path, max_it, restarts, seed) | |
| _pc (mode, runs) | |
| main () | |
Variables | |
| ROOT = Path(__file__).resolve().parents[1] | |
| C = _C(); C.reset() | |
Two-way Layer-1 optimizer benchmark: plain Python vs the numba accelerator. Serial in-process, one subprocess per condition, warmup discarded. "python" sets ED_ACCEL=off so every candidate takes the authoritative Python path; "numba" runs the accelerator. The C backend this once compared against was deleted after the numba path overtook it. Timing here is noisy enough that a single run is not a measurement: identical work (seed pinned) has varied by 40% on a loaded box. Hence --reps and min/median. Run: .venv/bin/python scripts/bench_layer1_numba.py --max-iterations 20 --reps 5
|
protected |
Count accelerated evaluates and Python fallbacks.
Definition at line 29 of file bench_layer1_numba.py.
Referenced by _run_condition().
|
protected |
Definition at line 46 of file bench_layer1_numba.py.
Referenced by _run_condition().
|
protected |
Summarise N repetitions of one mode. A single wall time is not a measurement here: on a loaded box the SAME work (seed pinned, byte-identical candidate trajectory) was observed to vary 6.81s..11.59s, a 1.7x spread that dwarfs the C-vs-Numba difference. Report min (least interference) and median instead.
Definition at line 85 of file bench_layer1_numba.py.
Referenced by main().
|
protected |
Definition at line 70 of file bench_layer1_numba.py.
References _install_instrumentation(), and _one().
Referenced by main().
| bench_layer1_numba.main | ( | ) |
Definition at line 112 of file bench_layer1_numba.py.
References _pc(), _run_condition(), and main().
Referenced by main().
| bench_layer1_numba.C = _C(); C.reset() |
Definition at line 26 of file bench_layer1_numba.py.
| bench_layer1_numba.ROOT = Path(__file__).resolve().parents[1] |
Definition at line 19 of file bench_layer1_numba.py.