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

stillwater-sc / universal / 38148623918
89%
master: 84%

Build:
Build:
LAST BUILD BRANCH: main
DEFAULT BRANCH: master
Ran 11 Oct 2026 03:33PM UTC
Jobs 1
Files 717
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

11 Oct 2026 02:47PM UTC coverage: 86.65% (+0.04%) from 86.609%
38148623918

push

github

web-flow
fix(multidouble): exact trunc, round, fmod, remainder; full-precision hypot, min/max, log1p (#1673)

dd, qd, dd_cascade, td_cascade and qd_cascade marshalled these through double (#1667):

- trunc and round rounded the argument to double first: above 2^53 the result was a
  different integer, below it a fraction carried in the low limbs was ignored.
- fmod and remainder took the remainder of the double roundings of x and y.
- hypot was worth 53 bits; min and max returned a double rounding of the winner.
- log1p was log(1 + a), which drops every limb of a below 1's last bit: log1p(1e-70) = 0.

A shared header, number/shared/multidouble_exact.hpp, now serves all five types:

- floor, ceil, trunc, round are exact: the integer is assembled limb by limb and confirmed
  against the exact difference x - n (the cascades' floor and ceil also renormalize now).
- fmod and remainder are exact: x - n y is reduced in a fixed-capacity Shewchuk expansion,
  25 quotient bits per step, so any x / y works; the result is rounded once.
- hypot scales by a power of two and takes sqrt(x^2 + y^2) in the type itself.
- log1p uses 2 atanh(a / (2 + a)) for |a| < 1/4.

The 20 truncate/fractional/hypot/minmax test files were MANUAL_TESTING stubs that checked
nothing. They now run test_suite_multidouble_exact.hpp, whose references are independent of
the code: libm bit for bit for double arguments across the whole range, x = k y + r with the
expected remainder formed exactly in ereal for multi-limb ones, and the defining properties
for the integer-valued functions. The logarithm suites gain a log1p check near 0. Every
file fails on the previous headers; all 118 multi-double tests pass.

Part of #1667; the hyperbolic, error and gamma functions follow.

BEHAVIOUR-CHANGE: dd/qd/dd_cascade/td_cascade/qd_cascade trunc, round, fmod, remainder, hypot, min, max and log1p return full-precision (and for the first four, exact) results instead of double roundings

Co-authored-by: Claude Opus 5... (continued)

301 of 324 new or added lines in 26 files covered. (92.9%)

8 existing lines in 1 file now uncovered.

48422 of 55882 relevant lines covered (86.65%)

9876278.54 hits per line

Uncovered Changes

Lines Coverage ∆ File
17
88.03
include/sw/universal/verification/test_suite_multidouble_exact.hpp
6
95.49
include/sw/universal/number/shared/multidouble_exact.hpp

Coverage Regressions

Lines Coverage ∆ File
8
96.36
-0.52% include/sw/universal/number/cfloat/cfloat_impl.hpp
Jobs
ID Job ID Ran Files Coverage
1 38148623918.1 11 Oct 2026 03:33PM UTC 717
86.65
GitHub Action Run
Source Files on build 38148623918
  • Tree
  • List 717
  • Changed 28
  • Source Changed 25
  • Coverage Changed 28
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #38148623918
  • daa612c2 on github
  • Prev Build on main (#38145667205)
  • Next Build on main (#38152437273)
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