Avatar for the mootable user
mootable
decimal-scaled
BlogDocsChangelog

Performance History

Latest Results

ci(bench-trial): quote criterion filter so regex metachars survive The trial workflow interpolated `${{ inputs.filter }}` straight into the shell command, so a regex like `D38.*(tan|sin).*scale_19` tripped the bash parser on the unquoted parens before ever reaching cargo bench. Thread the filter through `$FILTER` so the value lands quoted.
main
6 minutes ago
ci(bench-trial): parameterise bench target + filter + features
main
5 hours ago
release: 0.4.1 — drop deprecated DecimalConsts alias
main
5 hours ago
ignore local bench artifact dirs
main
6 hours ago
docs(changelog): round out 0.4.0 entry
main
6 hours ago
docs(benchmarks): refresh D38 lib_cmp from GHA 26070046546
main
6 hours ago
docs(benchmarks): refresh 12 lib_cmp widths from GHA 26070046546 (D38 pending) Re-ingests Criterion artifacts from the bench-full lib_cmp matrix run (workflow 26070046546) for every width except D38, whose job is still grinding and will be patched in a follow-up. Adds scripts/lib_cmp_ingest.py to convert the per-width artifact tree (criterion-lib_cmp-D<N>/lib_cmp_<bitwidth>_s<scale>/<lib>_<op>/) into the medians.tsv format chart_gen.rs consumes, preserving rows for widths not represented in the artifact set (D38 / 128bit stays on the prior data). Re-renders the 77 affected per-op-per-width PNGs + the 6 summary PNGs under docs/figures/library_comparison/. The summary_128bit chart is not in the artifact set this run and is preserved as-is. The benchmarks.md prose carries qualitative claims only for the 12 widths covered here (the single numeric per-width table is the D38 I128 one, which is intentionally not touched).
main
6 hours ago
docs(benchmarks): add cross-version history section + chart_gen history mode Three new per-width PNGs under docs/figures/history/ chart the arithmetic + strict-transcendental cost trend across v0.2.5, v0.3.2, v0.3.3, and v0.4.0 (HEAD). One line per function (add / mul / div / sqrt / ln / sin), consistent colour scheme across the three width charts (D38 / D76 / D307) so a reader can scan vertically. chart_gen gains a 'history' subcommand that reads bench-history-results/<tag>/<group>/<width>/new/estimates.json and renders the three PNGs. v0.3.0 and v0.3.1 are intentionally excluded — they shipped with a D923 widen reference that won't compile under the standard feature set, per the bench-history scope note. The history PNGs are embedded just above the raw-data dump section in docs/benchmarks.md.
main
6 hours ago

Latest Branches

CodSpeed Performance Gauge
-13%
build(deps): update fastnum requirement from 0.4 to 0.7#11
8 hours ago
d91b7e9
dependabot/cargo/fastnum-0.7
CodSpeed Performance Gauge
-27%
build(deps): update criterion requirement from 0.5 to 0.8 in /bench-history#6
8 hours ago
64d25c3
dependabot/cargo/bench-history/criterion-0.8
CodSpeed Performance Gauge
-1%
build(deps): bump actions/setup-python from 5 to 6#9
8 hours ago
232d5fb
dependabot/github_actions/actions/setup-python-6
© 2026 CodSpeed Technology
Home Terms Privacy Docs