• Home
  • Features
  • Pricing
  • Docs
  • Announcements
  • Sign In

stillwater-sc / universal / 39996255834
89%
master: 84%

Build:
Build:
LAST BUILD BRANCH: fix/issue-2113-followups
DEFAULT BRANCH: master
Ran 23 Oct 2026 05:46PM UTC
Jobs 1
Files 761
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst

23 Oct 2026 05:34PM UTC coverage: 88.073% (+0.001%) from 88.072%
39996255834

push

github

web-flow
perf(efloat): short products for a*b and the fixed-point engines, Mulders' mulhigh with a can-round fallback (#1808)

* perf(efloat): short products for a*b and the fixed-point engines (WIP, squashed at merge)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* perf(efloat): short products for a*b and the fixed-point engines, Mulders' mulhigh with a can-round fallback

efloat_mpn::mulhigh_n forms the high half of a product from every partial product a_i b_j with
i + j >= n - 1 (a basecase below 32 limbs, Mulders' recursion with a 0.6 n top product above),
so 0 <= a b - R B^(n-1) < n B^n: 0.63-0.75 of a full product from 16 limbs, 0.59-0.65 of a
square from 32.

- operator*: operands of 16 limbs or more first try the high half of the product, each padded
  with a zero limb below so that R is right to about 55 bits past the operands' length; where
  efloat_mpn::can_round (Ziv's test on limbs) passes, R rounds as the exact product does, and
  otherwise the full product decides, as before.
- efloat_mpn::mul_truncated, floor(a b / B^drop) or one unit below, serves the fixed-point
  engines (the series' powers and Horner products, exp's squarings, the versine's product and
  doublings, the arctangent and atanh engines) and pow's binary powering from 16 limbs; each such
  product counts as two truncations in the engines' error budgets, which stay inside their
  stated bounds. Below 16 limbs the same code as before runs.
- Karatsuba from 16 limbs (was 24) and the squaring basecase up to 48 (was 32), by measurement.

mpfr_gap, efloat / MPFR at 2048 / 8192 bits: a*b 1.8x 2.2x -> 1.6x 1.7x; pow(b,7) 1.3x 1.5x ->
1.1x 1.0x; exp 0.8x 1.1x -> 0.6x 0.8x; sin 1.0x 1.2x -> 0.8x 0.9x; log 0.7x 1.1x -> 0.6x 0.9x;
every function at 8192 bits at or below 1.0x. Below 1024 bits nothing changes.

Results are bit-identical to main, flags and carried precision included: 920,000 function
results across every family touched, 516,000 products and squares, and 960,000 inte... (continued)

141 of 154 new or added lines in 5 files covered. (91.56%)

1 existing line in 1 file now uncovered.

54115 of 61443 relevant lines covered (88.07%)

9554058.16 hits per line

Uncovered Changes

Lines Coverage ∆ File
13
90.84
-4.65% include/sw/universal/number/efloat/math/pow.hpp

Coverage Regressions

Lines Coverage ∆ File
1
97.04
0.0% include/sw/universal/number/cfloat/cfloat_impl.hpp
Jobs
ID Job ID Ran Files Coverage
1 39996255834.1 23 Oct 2026 05:46PM UTC 761
88.07
GitHub Action Run
Source Files on build 39996255834
  • Tree
  • List 761
  • Changed 7
  • Source Changed 6
  • Coverage Changed 7
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #39996255834
  • e0edfbd3 on github
  • Prev Build on main (#39982910469)
  • Next Build on main (#40025158222)
STATUS · Troubleshooting · Open an Issue · Sales · Support · CAREERS · ENTERPRISE · START FREE TRIAL · SCHEDULE DEMO
ANNOUNCEMENTS · TWITTER · TOS & SLA · Supported CI Services · What's a CI service? · Automated Testing

© 2026 Coveralls, Inc