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

stillwater-sc / universal / 43678966922
89%
master: 84%

Build:
Build:
LAST BUILD BRANCH: fix/fixpnt-quire-resolve-saturates
DEFAULT BRANCH: master
Ran 15 Nov 2026 05:18AM UTC
Jobs 1
Files 893
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

15 Nov 2026 04:57AM UTC coverage: 89.13% (-0.01%) from 89.141%
43678966922

push

github

web-flow
fix(takum): exact integer conversion, correctly rounded takum_log double conversion, defined integer casts (#2081)

Integers went into takum and takum_log through a double, which rounded every
integer above 2^53 first: takum<64,3>((1LL<<53)+1) was 2^53. Integer casts
cast a double, so NaR and every out-of-range value were undefined behaviour
(int(NaR) came out INT_MIN, (long long)maxpos LLONG_MIN). takum_log formed
l = 2 ln|x| and e^(l/2) in a double, ~|l| 2^-53 of absolute error, which
takum_log<64,3>, with 59 fraction bits of l, could not absorb.

- takum: integers convert exactly through the exact path's encoder
  (assign_wide), one rounding, at every width and in constant evaluation.
- takum_log: from a double or an integer, l is correctly rounded at every
  width. A double estimate 2 std::log(x) decides the encoding when both ends
  of its error interval encode alike, which keeps narrow layouts fast
  (conversion 14 -> 22 ns, addition 46 -> 57 ns for takum_log<16,3>). Otherwise
  l comes from takum_log_arithmetic's double-double log (~500 ns). The
  boundary-snap tolerance on the constexpr path is capped at half the layout's
  finest step.
- takum_log to double / float: correctly rounded wherever l = c + M/2^p is
  not exact in a double (takum_log<57,3> and up: 8 characteristic bits plus
  52 or more fraction bits), through the double-double exp, subnormals and
  overflow included. The other layouts keep std::exp, within an ulp.
- Integer casts in takum and takum_log follow posit's rule (#2073): truncate
  toward zero, saturate, NaR -> the type's minimum.
- operator=(char) on a signed char platform takes the signed path.
- Removed the unused convert(triple, takum&) stub, which returned finite
  values unchanged, and its triple.hpp include.
- table.hpp: the row count is 64-bit, with nbits <= 24 asserted
  (1u << nbits was UB for nbits >= 32).
- takum_wide::msb_index uses std::countl_zero instead of a linear scan. The
  128-step restoring division ... (continued)

119 of 128 new or added lines in 4 files covered. (92.97%)

14 existing lines in 1 file now uncovered.

63384 of 71114 relevant lines covered (89.13%)

12008065.61 hits per line

Uncovered Changes

Lines Coverage ∆ File
9
92.23
-6.93% include/sw/universal/number/takum/takum_log_impl.hpp

Coverage Regressions

Lines Coverage ∆ File
14
92.23
-6.93% include/sw/universal/number/takum/takum_log_impl.hpp
Jobs
ID Job ID Ran Files Coverage
1 43678966922.1 15 Nov 2026 05:18AM UTC 893
89.13
GitHub Action Run
Source Files on build 43678966922
  • Tree
  • List 893
  • Changed 6
  • Source Changed 5
  • Coverage Changed 6
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #43678966922
  • 3fc11593 on github
  • Prev Build on main (#43654278775)
  • Next Build on main (#43684754438)
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