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

stillwater-sc / universal / 40151466553
89%
master: 84%

Build:
Build:
LAST BUILD BRANCH: main
DEFAULT BRANCH: master
Ran 25 Oct 2026 01:59AM UTC
Jobs 1
Files 762
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

25 Oct 2026 01:39AM UTC coverage: 88.222% (+0.06%) from 88.163%
40151466553

push

github

web-flow
perf(efloat): differences, short products and short divisions at small precisions, and a per-precision table against MPFR (#1827)

mpfr_bands, a new benchmark next to mpfr_gap, times efloat against MPFR at
20 precisions from 64 to 16384 bits, among them the half-steps between the
powers of 2 and odd numbers of 32-bit limbs. It found b - a at 2.1-3.4x
MPFR up to 1024 bits and one to three limbs at 1.6-2.9x, cells the four
whole-limb precisions of mpfr_gap never measure.

- Differences take #1822's full-length path, as mpfr_sub1sp does: Y shifted
  once, the limb it shifts out kept, a borrow from it. A gap of one bit can
  cancel any number, then exact, shifted up once or rounded as any other.
  The path takes odd 32-bit precisions too, and one to four limbs run in
  fixed-size code.
- a + b, a - b, a * b and a / b write into the result without copying a
  (assign_sum, assign_product, assign_quotient). The general frame, product
  and division are out of line, so their scratch frames stay off the short
  paths.
- Products of one to three limbs at the full precision: fixed-size
  schoolbook, rounded from the product's limbs.
- A one-limb divisor divides as an integer one. Its quotient, and the exact
  quotient below the divide-and-conquer threshold, carry the precision
  alone, and the remainder gives the next bit and the sticky bit, as
  mpfr_div_1 does. Quotients of one or two limbs by one or two divide by
  x86-64's divq in the assembly build (divrem_1_short, sb_div_qr_short:
  Knuth's Algorithm D).
- sqrt at whole 64-bit limbs up to 256 bits: the exact root of the
  precision's limbs, whose remainder R gives the next bit, set exactly when
  R > S.
- mpmath_gap guards b-a / a+b, a*b / a+b at 128 bits and sqrt / a*b at 512.

Against #1824, pinned, best of three: a+b 0.62-0.95x at every precision,
b-a 0.50-0.82x, a*b 0.39-0.78x up to 192 bits, sqrt 0.70-0.97x at whole
limbs up to 256 bits. In mpfr_gap, a+b at 128 bits went from 1.4x MPFR to
1.0x and a*b... (continued)

361 of 362 new or added lines in 3 files covered. (99.72%)

1 existing line in 1 file now uncovered.

54762 of 62073 relevant lines covered (88.22%)

9044075.37 hits per line

Uncovered Changes

Lines Coverage ∆ File
1
96.13
0.81% include/sw/universal/number/efloat/efloat_impl.hpp

Coverage Regressions

Lines Coverage ∆ File
1
97.04
-0.06% include/sw/universal/number/cfloat/cfloat_impl.hpp
Jobs
ID Job ID Ran Files Coverage
1 40151466553.1 25 Oct 2026 01:59AM UTC 762
88.22
GitHub Action Run
Source Files on build 40151466553
  • Tree
  • List 762
  • Changed 4
  • Source Changed 3
  • Coverage Changed 4
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #40151466553
  • f0782dca on github
  • Prev Build on main (#40136138324)
  • Next Build on main (#40168143550)
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