61759c51b5
research/determinism-cost.md §4.6, in the same record and by the same procedure: the 192 payloads decoded again on every platform with every kernel set it runs (15 new runs, 0 decodes differing from the record's), and the record's time and energy runs repeated in one session per host beside the record's own build and the float decoder. On x86_64 host A the integer decode on the narrow kernels took 144.8 ms and 12.43 J of marginal package energy at quality 6 against 210.7 ms and 14.03 J for the float decode beside it (the record's build: 728.0 ms, 92.14 J). On the build host its own CPU energy fell from 11.01 J to 6.43 J at quality 6 and is still above the float decode's 1.28 J by that count. Host B's integer decode was measured before and after, without a float decode beside it. The scripts that took the readings are in runs/; decode-throughput.md notes the new sets; the changelog records the change. Co-Authored-By: Claude Opus 5.5 <[email protected]>
28 lines
2.1 KiB
Shell
28 lines
2.1 KiB
Shell
# Time and energy on the build host, DET1's procedure (macos-energy.sh): the decoder's own CPU
|
|
# energy (per-task count) over each of 7 rounds of the 8 pictures, at q2 and q6.
|
|
# Variants: int-before (DET1's binary), int-after (this change, kernel set auto, or KSET), float (again).
|
|
set -euo pipefail
|
|
TAG=${TAG:?binary tag}
|
|
W=$RB_ROOT/det1/work; B=$RB_ROOT/idf/bin; PY=$RB_ROOT/venvs/sota-int/bin/python; O=$RB_ROOT/idf/energy; mkdir -p $O
|
|
IDS=$(python3 -c "import json;print(','.join(p['id'] for p in json.load(open('$W/pictures.json'))['pictures'][:8]))")
|
|
E="python3 -I wai/tools/wai_det_energy.py --rounds ${ROUNDS:-7} --k 1 --idle-s 5"
|
|
HW='"system": "macOS 27.0.1", "hardware": "arm64 desktop SoC, 24 performance and 8 efficiency cores, 80-core integrated GPU", "pictures": 8'
|
|
D=${MEASURED:-$(date +%F)}
|
|
for q in ${QS:-2 6}; do
|
|
for v in ${VARIANTS:-int-before int-after float}; do
|
|
case $v in
|
|
int-before) L=macos-int-cpu-q$q-before; CH="$RB_ROOT/det1/bin/wai_det_cost.native serve $W/jobs.energy.q$q.tsv"; M='"path": "integer", "fabric": "CPU", "threads": "all (32)", "kernels": "the record'"'"'s (DET1 build, kernel set wide)"' ;;
|
|
int-after) K=${KSET:-auto}; L=macos-int-cpu-q$q-after${KSET:+-$KSET}; CH="env WAI_INT_ISA=$K $B/native.$TAG serve $W/jobs.energy.q$q.tsv"; M='"path": "integer", "fabric": "CPU", "threads": "all (32)", "kernels": "narrow, kernel set '$K'"' ;;
|
|
float) L=macos-float-cpu-q$q-again; CH="$PY -I wai/tools/wai_determinism_cost.py fserve $W $q --device cpu --ids $IDS"; M='"path": "float", "fabric": "CPU", "threads": "the library default"' ;;
|
|
esac
|
|
$E --label $L --counter task --meta "{$HW, $M, \"quality\": $q, \"measured\": \"$D\"}" -- $CH > $O/$L.json
|
|
python3 - $O/$L.json <<'PY'
|
|
import json, sys, statistics as st
|
|
r = json.load(open(sys.argv[1]))
|
|
ms = [x["child_wall_ns"] / x["decodes"] / 1e6 for x in r["rounds"]]
|
|
j = [x["task"]["energy_nj"] / x["decodes"] / 1e9 for x in r["rounds"]]
|
|
la = [x["load_avg_before"][0] for x in r["rounds"]]
|
|
print(f'{r["label"]}: {st.median(ms):.1f} ms ({min(ms):.1f}-{max(ms):.1f}), task {st.median(j):.2f} J ({min(j):.2f}-{max(j):.2f}), load {min(la):.0f}-{max(la):.0f}')
|
|
PY
|
|
done
|
|
done
|